drivers/xen/privcmd.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /******************************************************************************
   3  * privcmd.c
   4  *
   5  * Interface to privileged domain-0 commands.
   6  *
   7  * Copyright (c) 2002-2004, K A Fraser, B Dragovic
   8  */
   9
  10 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
  11
  12 #include <linux/kernel.h>
  13 #include <linux/module.h>
  14 #include <linux/sched.h>
  15 #include <linux/slab.h>
  16 #include <linux/string.h>
  17 #include <linux/errno.h>
  18 #include <linux/mm.h>
  19 #include <linux/mman.h>
  20 #include <linux/uaccess.h>
  21 #include <linux/swap.h>
  22 #include <linux/highmem.h>
  23 #include <linux/pagemap.h>
  24 #include <linux/seq_file.h>
  25 #include <linux/miscdevice.h>
  26 #include <linux/moduleparam.h>
  27
  28 #include <asm/pgalloc.h>
  29 #include <asm/xen/hypervisor.h>
  30 #include <asm/xen/hypercall.h>
  31
  32 #include <xen/xen.h>
  33 #include <xen/privcmd.h>
  34 #include <xen/interface/xen.h>
  35 #include <xen/interface/memory.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         int ret = 0;
 195
 196         BUG_ON(size > PAGE_SIZE);
 197
 198         while (nelem) {
 199                 int nr = (PAGE_SIZE/size);
 200                 struct page *page;
 201                 if (nr > nelem)
 202                         nr = nelem;
 203                 pos = pos->next;
 204                 page = list_entry(pos, struct page, lru);
 205                 pagedata = page_address(page);
 206                 ret = (*fn)(pagedata, nr, state);
 207                 if (ret)
 208                         break;
 209                 nelem -= nr;
 210         }
 211
 212         return ret;
 213 }
 214
 215 struct mmap_gfn_state {
 216         unsigned long va;
 217         struct vm_area_struct *vma;
 218         domid_t domain;
 219 };
 220
 221 static int mmap_gfn_range(void *data, void *state)
 222 {
 223         struct privcmd_mmap_entry *msg = data;
 224         struct mmap_gfn_state *st = state;
 225         struct vm_area_struct *vma = st->vma;
 226         int rc;
 227
 228         /* Do not allow range to wrap the address space. */
 229         if ((msg->npages > (LONG_MAX >> PAGE_SHIFT)) ||
 230             ((unsigned long)(msg->npages << PAGE_SHIFT) >= -st->va))
 231                 return -EINVAL;
 232
 233         /* Range chunks must be contiguous in va space. */
 234         if ((msg->va != st->va) ||
 235             ((msg->va+(msg->npages<<PAGE_SHIFT)) > vma->vm_end))
 236                 return -EINVAL;
 237
 238         rc = xen_remap_domain_gfn_range(vma,
 239                                         msg->va & PAGE_MASK,
 240                                         msg->mfn, msg->npages,
 241                                         vma->vm_page_prot,
 242                                         st->domain, NULL);
 243         if (rc < 0)
 244                 return rc;
 245
 246         st->va += msg->npages << PAGE_SHIFT;
 247
 248         return 0;
 249 }
 250
 251 static long privcmd_ioctl_mmap(struct file *file, void __user *udata)
 252 {
 253         struct privcmd_data *data = file->private_data;
 254         struct privcmd_mmap mmapcmd;
 255         struct mm_struct *mm = current->mm;
 256         struct vm_area_struct *vma;
 257         int rc;
 258         LIST_HEAD(pagelist);
 259         struct mmap_gfn_state state;
 260
 261         /* We only support privcmd_ioctl_mmap_batch for auto translated. */
 262         if (xen_feature(XENFEAT_auto_translated_physmap))
 263                 return -ENOSYS;
 264
 265         if (copy_from_user(&mmapcmd, udata, sizeof(mmapcmd)))
 266                 return -EFAULT;
 267
 268         /* If restriction is in place, check the domid matches */
 269         if (data->domid != DOMID_INVALID && data->domid != mmapcmd.dom)
 270                 return -EPERM;
 271
 272         rc = gather_array(&pagelist,
 273                           mmapcmd.num, sizeof(struct privcmd_mmap_entry),
 274                           mmapcmd.entry);
 275
 276         if (rc || list_empty(&pagelist))
 277                 goto out;
 278
 279         mmap_write_lock(mm);
 280
 281         {
 282                 struct page *page = list_first_entry(&pagelist,
 283                                                      struct page, lru);
 284                 struct privcmd_mmap_entry *msg = page_address(page);
 285
 286                 vma = find_vma(mm, msg->va);
 287                 rc = -EINVAL;
 288
 289                 if (!vma || (msg->va != vma->vm_start) || vma->vm_private_data)
 290                         goto out_up;
 291                 vma->vm_private_data = PRIV_VMA_LOCKED;
 292         }
 293
 294         state.va = vma->vm_start;
 295         state.vma = vma;
 296         state.domain = mmapcmd.dom;
 297
 298         rc = traverse_pages(mmapcmd.num, sizeof(struct privcmd_mmap_entry),
 299                             &pagelist,
 300                             mmap_gfn_range, &state);
 301
 302
 303 out_up:
 304         mmap_write_unlock(mm);
 305
 306 out:
 307         free_page_list(&pagelist);
 308
 309         return rc;
 310 }
 311
 312 struct mmap_batch_state {
 313         domid_t domain;
 314         unsigned long va;
 315         struct vm_area_struct *vma;
 316         int index;
 317         /* A tristate:
 318          *      0 for no errors
 319          *      1 if at least one error has happened (and no
 320          *          -ENOENT errors have happened)
 321          *      -ENOENT if at least 1 -ENOENT has happened.
 322          */
 323         int global_error;
 324         int version;
 325
 326         /* User-space gfn array to store errors in the second pass for V1. */
 327         xen_pfn_t __user *user_gfn;
 328         /* User-space int array to store errors in the second pass for V2. */
 329         int __user *user_err;
 330 };
 331
 332 /* auto translated dom0 note: if domU being created is PV, then gfn is
 333  * mfn(addr on bus). If it's auto xlated, then gfn is pfn (input to HAP).
 334  */
 335 static int mmap_batch_fn(void *data, int nr, void *state)
 336 {
 337         xen_pfn_t *gfnp = data;
 338         struct mmap_batch_state *st = state;
 339         struct vm_area_struct *vma = st->vma;
 340         struct page **pages = vma->vm_private_data;
 341         struct page **cur_pages = NULL;
 342         int ret;
 343
 344         if (xen_feature(XENFEAT_auto_translated_physmap))
 345                 cur_pages = &pages[st->index];
 346
 347         BUG_ON(nr < 0);
 348         ret = xen_remap_domain_gfn_array(st->vma, st->va & PAGE_MASK, gfnp, nr,
 349                                          (int *)gfnp, st->vma->vm_page_prot,
 350                                          st->domain, cur_pages);
 351
 352         /* Adjust the global_error? */
 353         if (ret != nr) {
 354                 if (ret == -ENOENT)
 355                         st->global_error = -ENOENT;
 356                 else {
 357                         /* Record that at least one error has happened. */
 358                         if (st->global_error == 0)
 359                                 st->global_error = 1;
 360                 }
 361         }
 362         st->va += XEN_PAGE_SIZE * nr;
 363         st->index += nr / XEN_PFN_PER_PAGE;
 364
 365         return 0;
 366 }
 367
 368 static int mmap_return_error(int err, struct mmap_batch_state *st)
 369 {
 370         int ret;
 371
 372         if (st->version == 1) {
 373                 if (err) {
 374                         xen_pfn_t gfn;
 375
 376                         ret = get_user(gfn, st->user_gfn);
 377                         if (ret < 0)
 378                                 return ret;
 379                         /*
 380                          * V1 encodes the error codes in the 32bit top
 381                          * nibble of the gfn (with its known
 382                          * limitations vis-a-vis 64 bit callers).
 383                          */
 384                         gfn |= (err == -ENOENT) ?
 385                                 PRIVCMD_MMAPBATCH_PAGED_ERROR :
 386                                 PRIVCMD_MMAPBATCH_MFN_ERROR;
 387                         return __put_user(gfn, st->user_gfn++);
 388                 } else
 389                         st->user_gfn++;
 390         } else { /* st->version == 2 */
 391                 if (err)
 392                         return __put_user(err, st->user_err++);
 393                 else
 394                         st->user_err++;
 395         }
 396
 397         return 0;
 398 }
 399
 400 static int mmap_return_errors(void *data, int nr, void *state)
 401 {
 402         struct mmap_batch_state *st = state;
 403         int *errs = data;
 404         int i;
 405         int ret;
 406
 407         for (i = 0; i < nr; i++) {
 408                 ret = mmap_return_error(errs[i], st);
 409                 if (ret < 0)
 410                         return ret;
 411         }
 412         return 0;
 413 }
 414
 415 /* Allocate pfns that are then mapped with gfns from foreign domid. Update
 416  * the vma with the page info to use later.
 417  * Returns: 0 if success, otherwise -errno
 418  */
 419 static int alloc_empty_pages(struct vm_area_struct *vma, int numpgs)
 420 {
 421         int rc;
 422         struct page **pages;
 423
 424         pages = kcalloc(numpgs, sizeof(pages[0]), GFP_KERNEL);
 425         if (pages == NULL)
 426                 return -ENOMEM;
 427
 428         rc = alloc_xenballooned_pages(numpgs, pages);
 429         if (rc != 0) {
 430                 pr_warn("%s Could not alloc %d pfns rc:%d\n", __func__,
 431                         numpgs, rc);
 432                 kfree(pages);
 433                 return -ENOMEM;
 434         }
 435         BUG_ON(vma->vm_private_data != NULL);
 436         vma->vm_private_data = pages;
 437
 438         return 0;
 439 }
 440
 441 static const struct vm_operations_struct privcmd_vm_ops;
 442
 443 static long privcmd_ioctl_mmap_batch(
 444         struct file *file, void __user *udata, int version)
 445 {
 446         struct privcmd_data *data = file->private_data;
 447         int ret;
 448         struct privcmd_mmapbatch_v2 m;
 449         struct mm_struct *mm = current->mm;
 450         struct vm_area_struct *vma;
 451         unsigned long nr_pages;
 452         LIST_HEAD(pagelist);
 453         struct mmap_batch_state state;
 454
 455         switch (version) {
 456         case 1:
 457                 if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch)))
 458                         return -EFAULT;
 459                 /* Returns per-frame error in m.arr. */
 460                 m.err = NULL;
 461                 if (!access_ok(m.arr, m.num * sizeof(*m.arr)))
 462                         return -EFAULT;
 463                 break;
 464         case 2:
 465                 if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch_v2)))
 466                         return -EFAULT;
 467                 /* Returns per-frame error code in m.err. */
 468                 if (!access_ok(m.err, m.num * (sizeof(*m.err))))
 469                         return -EFAULT;
 470                 break;
 471         default:
 472                 return -EINVAL;
 473         }
 474
 475         /* If restriction is in place, check the domid matches */
 476         if (data->domid != DOMID_INVALID && data->domid != m.dom)
 477                 return -EPERM;
 478
 479         nr_pages = DIV_ROUND_UP(m.num, XEN_PFN_PER_PAGE);
 480         if ((m.num <= 0) || (nr_pages > (LONG_MAX >> PAGE_SHIFT)))
 481                 return -EINVAL;
 482
 483         ret = gather_array(&pagelist, m.num, sizeof(xen_pfn_t), m.arr);
 484
 485         if (ret)
 486                 goto out;
 487         if (list_empty(&pagelist)) {
 488                 ret = -EINVAL;
 489                 goto out;
 490         }
 491
 492         if (version == 2) {
 493                 /* Zero error array now to only copy back actual errors. */
 494                 if (clear_user(m.err, sizeof(int) * m.num)) {
 495                         ret = -EFAULT;
 496                         goto out;
 497                 }
 498         }
 499
 500         mmap_write_lock(mm);
 501
 502         vma = find_vma(mm, m.addr);
 503         if (!vma ||
 504             vma->vm_ops != &privcmd_vm_ops) {
 505                 ret = -EINVAL;
 506                 goto out_unlock;
 507         }
 508
 509         /*
 510          * Caller must either:
 511          *
 512          * Map the whole VMA range, which will also allocate all the
 513          * pages required for the auto_translated_physmap case.
 514          *
 515          * Or
 516          *
 517          * Map unmapped holes left from a previous map attempt (e.g.,
 518          * because those foreign frames were previously paged out).
 519          */
 520         if (vma->vm_private_data == NULL) {
 521                 if (m.addr != vma->vm_start ||
 522                     m.addr + (nr_pages << PAGE_SHIFT) != vma->vm_end) {
 523                         ret = -EINVAL;
 524                         goto out_unlock;
 525                 }
 526                 if (xen_feature(XENFEAT_auto_translated_physmap)) {
 527                         ret = alloc_empty_pages(vma, nr_pages);
 528                         if (ret < 0)
 529                                 goto out_unlock;
 530                 } else
 531                         vma->vm_private_data = PRIV_VMA_LOCKED;
 532         } else {
 533                 if (m.addr < vma->vm_start ||
 534                     m.addr + (nr_pages << PAGE_SHIFT) > vma->vm_end) {
 535                         ret = -EINVAL;
 536                         goto out_unlock;
 537                 }
 538                 if (privcmd_vma_range_is_mapped(vma, m.addr, nr_pages)) {
 539                         ret = -EINVAL;
 540                         goto out_unlock;
 541                 }
 542         }
 543
 544         state.domain        = m.dom;
 545         state.vma           = vma;
 546         state.va            = m.addr;
 547         state.index         = 0;
 548         state.global_error  = 0;
 549         state.version       = version;
 550
 551         BUILD_BUG_ON(((PAGE_SIZE / sizeof(xen_pfn_t)) % XEN_PFN_PER_PAGE) != 0);
 552         /* mmap_batch_fn guarantees ret == 0 */
 553         BUG_ON(traverse_pages_block(m.num, sizeof(xen_pfn_t),
 554                                     &pagelist, mmap_batch_fn, &state));
 555
 556         mmap_write_unlock(mm);
 557
 558         if (state.global_error) {
 559                 /* Write back errors in second pass. */
 560                 state.user_gfn = (xen_pfn_t *)m.arr;
 561                 state.user_err = m.err;
 562                 ret = traverse_pages_block(m.num, sizeof(xen_pfn_t),
 563                                            &pagelist, mmap_return_errors, &state);
 564         } else
 565                 ret = 0;
 566
 567         /* If we have not had any EFAULT-like global errors then set the global
 568          * error to -ENOENT if necessary. */
 569         if ((ret == 0) && (state.global_error == -ENOENT))
 570                 ret = -ENOENT;
 571
 572 out:
 573         free_page_list(&pagelist);
 574         return ret;
 575
 576 out_unlock:
 577         mmap_write_unlock(mm);
 578         goto out;
 579 }
 580
 581 static int lock_pages(
 582         struct privcmd_dm_op_buf kbufs[], unsigned int num,
 583         struct page *pages[], unsigned int nr_pages)
 584 {
 585         unsigned int i;
 586
 587         for (i = 0; i < num; i++) {
 588                 unsigned int requested;
 589                 int pinned;
 590
 591                 requested = DIV_ROUND_UP(
 592                         offset_in_page(kbufs[i].uptr) + kbufs[i].size,
 593                         PAGE_SIZE);
 594                 if (requested > nr_pages)
 595                         return -ENOSPC;
 596
 597                 pinned = get_user_pages_fast(
 598                         (unsigned long) kbufs[i].uptr,
 599                         requested, FOLL_WRITE, pages);
 600                 if (pinned < 0)
 601                         return pinned;
 602
 603                 nr_pages -= pinned;
 604                 pages += pinned;
 605         }
 606
 607         return 0;
 608 }
 609
 610 static void unlock_pages(struct page *pages[], unsigned int nr_pages)
 611 {
 612         unsigned int i;
 613
 614         if (!pages)
 615                 return;
 616
 617         for (i = 0; i < nr_pages; i++) {
 618                 if (pages[i])
 619                         put_page(pages[i]);
 620         }
 621 }
 622
 623 static long privcmd_ioctl_dm_op(struct file *file, void __user *udata)
 624 {
 625         struct privcmd_data *data = file->private_data;
 626         struct privcmd_dm_op kdata;
 627         struct privcmd_dm_op_buf *kbufs;
 628         unsigned int nr_pages = 0;
 629         struct page **pages = NULL;
 630         struct xen_dm_op_buf *xbufs = NULL;
 631         unsigned int i;
 632         long rc;
 633
 634         if (copy_from_user(&kdata, udata, sizeof(kdata)))
 635                 return -EFAULT;
 636
 637         /* If restriction is in place, check the domid matches */
 638         if (data->domid != DOMID_INVALID && data->domid != kdata.dom)
 639                 return -EPERM;
 640
 641         if (kdata.num == 0)
 642                 return 0;
 643
 644         if (kdata.num > privcmd_dm_op_max_num)
 645                 return -E2BIG;
 646
 647         kbufs = kcalloc(kdata.num, sizeof(*kbufs), GFP_KERNEL);
 648         if (!kbufs)
 649                 return -ENOMEM;
 650
 651         if (copy_from_user(kbufs, kdata.ubufs,
 652                            sizeof(*kbufs) * kdata.num)) {
 653                 rc = -EFAULT;
 654                 goto out;
 655         }
 656
 657         for (i = 0; i < kdata.num; i++) {
 658                 if (kbufs[i].size > privcmd_dm_op_buf_max_size) {
 659                         rc = -E2BIG;
 660                         goto out;
 661                 }
 662
 663                 if (!access_ok(kbufs[i].uptr,
 664                                kbufs[i].size)) {
 665                         rc = -EFAULT;
 666                         goto out;
 667                 }
 668
 669                 nr_pages += DIV_ROUND_UP(
 670                         offset_in_page(kbufs[i].uptr) + kbufs[i].size,
 671                         PAGE_SIZE);
 672         }
 673
 674         pages = kcalloc(nr_pages, sizeof(*pages), GFP_KERNEL);
 675         if (!pages) {
 676                 rc = -ENOMEM;
 677                 goto out;
 678         }
 679
 680         xbufs = kcalloc(kdata.num, sizeof(*xbufs), GFP_KERNEL);
 681         if (!xbufs) {
 682                 rc = -ENOMEM;
 683                 goto out;
 684         }
 685
 686         rc = lock_pages(kbufs, kdata.num, pages, nr_pages);
 687         if (rc)
 688                 goto out;
 689
 690         for (i = 0; i < kdata.num; i++) {
 691                 set_xen_guest_handle(xbufs[i].h, kbufs[i].uptr);
 692                 xbufs[i].size = kbufs[i].size;
 693         }
 694
 695         xen_preemptible_hcall_begin();
 696         rc = HYPERVISOR_dm_op(kdata.dom, kdata.num, xbufs);
 697         xen_preemptible_hcall_end();
 698
 699 out:
 700         unlock_pages(pages, nr_pages);
 701         kfree(xbufs);
 702         kfree(pages);
 703         kfree(kbufs);
 704
 705         return rc;
 706 }
 707
 708 static long privcmd_ioctl_restrict(struct file *file, void __user *udata)
 709 {
 710         struct privcmd_data *data = file->private_data;
 711         domid_t dom;
 712
 713         if (copy_from_user(&dom, udata, sizeof(dom)))
 714                 return -EFAULT;
 715
 716         /* Set restriction to the specified domain, or check it matches */
 717         if (data->domid == DOMID_INVALID)
 718                 data->domid = dom;
 719         else if (data->domid != dom)
 720                 return -EINVAL;
 721
 722         return 0;
 723 }
 724
 725 static long privcmd_ioctl_mmap_resource(struct file *file, void __user *udata)
 726 {
 727         struct privcmd_data *data = file->private_data;
 728         struct mm_struct *mm = current->mm;
 729         struct vm_area_struct *vma;
 730         struct privcmd_mmap_resource kdata;
 731         xen_pfn_t *pfns = NULL;
 732         struct xen_mem_acquire_resource xdata;
 733         int rc;
 734
 735         if (copy_from_user(&kdata, udata, sizeof(kdata)))
 736                 return -EFAULT;
 737
 738         /* If restriction is in place, check the domid matches */
 739         if (data->domid != DOMID_INVALID && data->domid != kdata.dom)
 740                 return -EPERM;
 741
 742         mmap_write_lock(mm);
 743
 744         vma = find_vma(mm, kdata.addr);
 745         if (!vma || vma->vm_ops != &privcmd_vm_ops) {
 746                 rc = -EINVAL;
 747                 goto out;
 748         }
 749
 750         pfns = kcalloc(kdata.num, sizeof(*pfns), GFP_KERNEL);
 751         if (!pfns) {
 752                 rc = -ENOMEM;
 753                 goto out;
 754         }
 755
 756         if (IS_ENABLED(CONFIG_XEN_AUTO_XLATE) &&
 757             xen_feature(XENFEAT_auto_translated_physmap)) {
 758                 unsigned int nr = DIV_ROUND_UP(kdata.num, XEN_PFN_PER_PAGE);
 759                 struct page **pages;
 760                 unsigned int i;
 761
 762                 rc = alloc_empty_pages(vma, nr);
 763                 if (rc < 0)
 764                         goto out;
 765
 766                 pages = vma->vm_private_data;
 767                 for (i = 0; i < kdata.num; i++) {
 768                         xen_pfn_t pfn =
 769                                 page_to_xen_pfn(pages[i / XEN_PFN_PER_PAGE]);
 770
 771                         pfns[i] = pfn + (i % XEN_PFN_PER_PAGE);
 772                 }
 773         } else
 774                 vma->vm_private_data = PRIV_VMA_LOCKED;
 775
 776         memset(&xdata, 0, sizeof(xdata));
 777         xdata.domid = kdata.dom;
 778         xdata.type = kdata.type;
 779         xdata.id = kdata.id;
 780         xdata.frame = kdata.idx;
 781         xdata.nr_frames = kdata.num;
 782         set_xen_guest_handle(xdata.frame_list, pfns);
 783
 784         xen_preemptible_hcall_begin();
 785         rc = HYPERVISOR_memory_op(XENMEM_acquire_resource, &xdata);
 786         xen_preemptible_hcall_end();
 787
 788         if (rc)
 789                 goto out;
 790
 791         if (IS_ENABLED(CONFIG_XEN_AUTO_XLATE) &&
 792             xen_feature(XENFEAT_auto_translated_physmap)) {
 793                 rc = xen_remap_vma_range(vma, kdata.addr, kdata.num << PAGE_SHIFT);
 794         } else {
 795                 unsigned int domid =
 796                         (xdata.flags & XENMEM_rsrc_acq_caller_owned) ?
 797                         DOMID_SELF : kdata.dom;
 798                 int num;
 799
 800                 num = xen_remap_domain_mfn_array(vma,
 801                                                  kdata.addr & PAGE_MASK,
 802                                                  pfns, kdata.num, (int *)pfns,
 803                                                  vma->vm_page_prot,
 804                                                  domid,
 805                                                  vma->vm_private_data);
 806                 if (num < 0)
 807                         rc = num;
 808                 else if (num != kdata.num) {
 809                         unsigned int i;
 810
 811                         for (i = 0; i < num; i++) {
 812                                 rc = pfns[i];
 813                                 if (rc < 0)
 814                                         break;
 815                         }
 816                 } else
 817                         rc = 0;
 818         }
 819
 820 out:
 821         mmap_write_unlock(mm);
 822         kfree(pfns);
 823
 824         return rc;
 825 }
 826
 827 static long privcmd_ioctl(struct file *file,
 828                           unsigned int cmd, unsigned long data)
 829 {
 830         int ret = -ENOTTY;
 831         void __user *udata = (void __user *) data;
 832
 833         switch (cmd) {
 834         case IOCTL_PRIVCMD_HYPERCALL:
 835                 ret = privcmd_ioctl_hypercall(file, udata);
 836                 break;
 837
 838         case IOCTL_PRIVCMD_MMAP:
 839                 ret = privcmd_ioctl_mmap(file, udata);
 840                 break;
 841
 842         case IOCTL_PRIVCMD_MMAPBATCH:
 843                 ret = privcmd_ioctl_mmap_batch(file, udata, 1);
 844                 break;
 845
 846         case IOCTL_PRIVCMD_MMAPBATCH_V2:
 847                 ret = privcmd_ioctl_mmap_batch(file, udata, 2);
 848                 break;
 849
 850         case IOCTL_PRIVCMD_DM_OP:
 851                 ret = privcmd_ioctl_dm_op(file, udata);
 852                 break;
 853
 854         case IOCTL_PRIVCMD_RESTRICT:
 855                 ret = privcmd_ioctl_restrict(file, udata);
 856                 break;
 857
 858         case IOCTL_PRIVCMD_MMAP_RESOURCE:
 859                 ret = privcmd_ioctl_mmap_resource(file, udata);
 860                 break;
 861
 862         default:
 863                 break;
 864         }
 865
 866         return ret;
 867 }
 868
 869 static int privcmd_open(struct inode *ino, struct file *file)
 870 {
 871         struct privcmd_data *data = kzalloc(sizeof(*data), GFP_KERNEL);
 872
 873         if (!data)
 874                 return -ENOMEM;
 875
 876         /* DOMID_INVALID implies no restriction */
 877         data->domid = DOMID_INVALID;
 878
 879         file->private_data = data;
 880         return 0;
 881 }
 882
 883 static int privcmd_release(struct inode *ino, struct file *file)
 884 {
 885         struct privcmd_data *data = file->private_data;
 886
 887         kfree(data);
 888         return 0;
 889 }
 890
 891 static void privcmd_close(struct vm_area_struct *vma)
 892 {
 893         struct page **pages = vma->vm_private_data;
 894         int numpgs = vma_pages(vma);
 895         int numgfns = (vma->vm_end - vma->vm_start) >> XEN_PAGE_SHIFT;
 896         int rc;
 897
 898         if (!xen_feature(XENFEAT_auto_translated_physmap) || !numpgs || !pages)
 899                 return;
 900
 901         rc = xen_unmap_domain_gfn_range(vma, numgfns, pages);
 902         if (rc == 0)
 903                 free_xenballooned_pages(numpgs, pages);
 904         else
 905                 pr_crit("unable to unmap MFN range: leaking %d pages. rc=%d\n",
 906                         numpgs, rc);
 907         kfree(pages);
 908 }
 909
 910 static vm_fault_t privcmd_fault(struct vm_fault *vmf)
 911 {
 912         printk(KERN_DEBUG "privcmd_fault: vma=%p %lx-%lx, pgoff=%lx, uv=%p\n",
 913                vmf->vma, vmf->vma->vm_start, vmf->vma->vm_end,
 914                vmf->pgoff, (void *)vmf->address);
 915
 916         return VM_FAULT_SIGBUS;
 917 }
 918
 919 static const struct vm_operations_struct privcmd_vm_ops = {
 920         .close = privcmd_close,
 921         .fault = privcmd_fault
 922 };
 923
 924 static int privcmd_mmap(struct file *file, struct vm_area_struct *vma)
 925 {
 926         /* DONTCOPY is essential for Xen because copy_page_range doesn't know
 927          * how to recreate these mappings */
 928         vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTCOPY |
 929                          VM_DONTEXPAND | VM_DONTDUMP;
 930         vma->vm_ops = &privcmd_vm_ops;
 931         vma->vm_private_data = NULL;
 932
 933         return 0;
 934 }
 935
 936 /*
 937  * For MMAPBATCH*. This allows asserting the singleshot mapping
 938  * on a per pfn/pte basis. Mapping calls that fail with ENOENT
 939  * can be then retried until success.
 940  */
 941 static int is_mapped_fn(pte_t *pte, unsigned long addr, void *data)
 942 {
 943         return pte_none(*pte) ? 0 : -EBUSY;
 944 }
 945
 946 static int privcmd_vma_range_is_mapped(
 947                    struct vm_area_struct *vma,
 948                    unsigned long addr,
 949                    unsigned long nr_pages)
 950 {
 951         return apply_to_page_range(vma->vm_mm, addr, nr_pages << PAGE_SHIFT,
 952                                    is_mapped_fn, NULL) != 0;
 953 }
 954
 955 const struct file_operations xen_privcmd_fops = {
 956         .owner = THIS_MODULE,
 957         .unlocked_ioctl = privcmd_ioctl,
 958         .open = privcmd_open,
 959         .release = privcmd_release,
 960         .mmap = privcmd_mmap,
 961 };
 962 EXPORT_SYMBOL_GPL(xen_privcmd_fops);
 963
 964 static struct miscdevice privcmd_dev = {
 965         .minor = MISC_DYNAMIC_MINOR,
 966         .name = "xen/privcmd",
 967         .fops = &xen_privcmd_fops,
 968 };
 969
 970 static int __init privcmd_init(void)
 971 {
 972         int err;
 973
 974         if (!xen_domain())
 975                 return -ENODEV;
 976
 977         err = misc_register(&privcmd_dev);
 978         if (err != 0) {
 979                 pr_err("Could not register Xen privcmd device\n");
 980                 return err;
 981         }
 982
 983         err = misc_register(&xen_privcmdbuf_dev);
 984         if (err != 0) {
 985                 pr_err("Could not register Xen hypercall-buf device\n");
 986                 misc_deregister(&privcmd_dev);
 987                 return err;
 988         }
 989
 990         return 0;
 991 }
 992
 993 static void __exit privcmd_exit(void)
 994 {
 995         misc_deregister(&privcmd_dev);
 996         misc_deregister(&xen_privcmdbuf_dev);
 997 }
 998
 999 module_init(privcmd_init);
1000 module_exit(privcmd_exit);