linux/drivers/xen/privcmd.c
<<
>>
Prefs
   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/xen/hypervisor.h>
  29#include <asm/xen/hypercall.h>
  30
  31#include <xen/xen.h>
  32#include <xen/privcmd.h>
  33#include <xen/interface/xen.h>
  34#include <xen/interface/memory.h>
  35#include <xen/interface/hvm/dm_op.h>
  36#include <xen/features.h>
  37#include <xen/page.h>
  38#include <xen/xen-ops.h>
  39#include <xen/balloon.h>
  40
  41#include "privcmd.h"
  42
  43MODULE_LICENSE("GPL");
  44
  45#define PRIV_VMA_LOCKED ((void *)1)
  46
  47static unsigned int privcmd_dm_op_max_num = 16;
  48module_param_named(dm_op_max_nr_bufs, privcmd_dm_op_max_num, uint, 0644);
  49MODULE_PARM_DESC(dm_op_max_nr_bufs,
  50                 "Maximum number of buffers per dm_op hypercall");
  51
  52static unsigned int privcmd_dm_op_buf_max_size = 4096;
  53module_param_named(dm_op_buf_max_size, privcmd_dm_op_buf_max_size, uint,
  54                   0644);
  55MODULE_PARM_DESC(dm_op_buf_max_size,
  56                 "Maximum size of a dm_op hypercall buffer");
  57
  58struct privcmd_data {
  59        domid_t domid;
  60};
  61
  62static int privcmd_vma_range_is_mapped(
  63               struct vm_area_struct *vma,
  64               unsigned long addr,
  65               unsigned long nr_pages);
  66
  67static long privcmd_ioctl_hypercall(struct file *file, void __user *udata)
  68{
  69        struct privcmd_data *data = file->private_data;
  70        struct privcmd_hypercall hypercall;
  71        long ret;
  72
  73        /* Disallow arbitrary hypercalls if restricted */
  74        if (data->domid != DOMID_INVALID)
  75                return -EPERM;
  76
  77        if (copy_from_user(&hypercall, udata, sizeof(hypercall)))
  78                return -EFAULT;
  79
  80        xen_preemptible_hcall_begin();
  81        ret = privcmd_call(hypercall.op,
  82                           hypercall.arg[0], hypercall.arg[1],
  83                           hypercall.arg[2], hypercall.arg[3],
  84                           hypercall.arg[4]);
  85        xen_preemptible_hcall_end();
  86
  87        return ret;
  88}
  89
  90static void free_page_list(struct list_head *pages)
  91{
  92        struct page *p, *n;
  93
  94        list_for_each_entry_safe(p, n, pages, lru)
  95                __free_page(p);
  96
  97        INIT_LIST_HEAD(pages);
  98}
  99
 100/*
 101 * Given an array of items in userspace, return a list of pages
 102 * containing the data.  If copying fails, either because of memory
 103 * allocation failure or a problem reading user memory, return an
 104 * error code; its up to the caller to dispose of any partial list.
 105 */
 106static int gather_array(struct list_head *pagelist,
 107                        unsigned nelem, size_t size,
 108                        const void __user *data)
 109{
 110        unsigned pageidx;
 111        void *pagedata;
 112        int ret;
 113
 114        if (size > PAGE_SIZE)
 115                return 0;
 116
 117        pageidx = PAGE_SIZE;
 118        pagedata = NULL;        /* quiet, gcc */
 119        while (nelem--) {
 120                if (pageidx > PAGE_SIZE-size) {
 121                        struct page *page = alloc_page(GFP_KERNEL);
 122
 123                        ret = -ENOMEM;
 124                        if (page == NULL)
 125                                goto fail;
 126
 127                        pagedata = page_address(page);
 128
 129                        list_add_tail(&page->lru, pagelist);
 130                        pageidx = 0;
 131                }
 132
 133                ret = -EFAULT;
 134                if (copy_from_user(pagedata + pageidx, data, size))
 135                        goto fail;
 136
 137                data += size;
 138                pageidx += size;
 139        }
 140
 141        ret = 0;
 142
 143fail:
 144        return ret;
 145}
 146
 147/*
 148 * Call function "fn" on each element of the array fragmented
 149 * over a list of pages.
 150 */
 151static int traverse_pages(unsigned nelem, size_t size,
 152                          struct list_head *pos,
 153                          int (*fn)(void *data, void *state),
 154                          void *state)
 155{
 156        void *pagedata;
 157        unsigned pageidx;
 158        int ret = 0;
 159
 160        BUG_ON(size > PAGE_SIZE);
 161
 162        pageidx = PAGE_SIZE;
 163        pagedata = NULL;        /* hush, gcc */
 164
 165        while (nelem--) {
 166                if (pageidx > PAGE_SIZE-size) {
 167                        struct page *page;
 168                        pos = pos->next;
 169                        page = list_entry(pos, struct page, lru);
 170                        pagedata = page_address(page);
 171                        pageidx = 0;
 172                }
 173
 174                ret = (*fn)(pagedata + pageidx, state);
 175                if (ret)
 176                        break;
 177                pageidx += size;
 178        }
 179
 180        return ret;
 181}
 182
 183/*
 184 * Similar to traverse_pages, but use each page as a "block" of
 185 * data to be processed as one unit.
 186 */
 187static int traverse_pages_block(unsigned nelem, size_t size,
 188                                struct list_head *pos,
 189                                int (*fn)(void *data, int nr, void *state),
 190                                void *state)
 191{
 192        void *pagedata;
 193        int ret = 0;
 194
 195        BUG_ON(size > PAGE_SIZE);
 196
 197        while (nelem) {
 198                int nr = (PAGE_SIZE/size);
 199                struct page *page;
 200                if (nr > nelem)
 201                        nr = nelem;
 202                pos = pos->next;
 203                page = list_entry(pos, struct page, lru);
 204                pagedata = page_address(page);
 205                ret = (*fn)(pagedata, nr, state);
 206                if (ret)
 207                        break;
 208                nelem -= nr;
 209        }
 210
 211        return ret;
 212}
 213
 214struct mmap_gfn_state {
 215        unsigned long va;
 216        struct vm_area_struct *vma;
 217        domid_t domain;
 218};
 219
 220static int mmap_gfn_range(void *data, void *state)
 221{
 222        struct privcmd_mmap_entry *msg = data;
 223        struct mmap_gfn_state *st = state;
 224        struct vm_area_struct *vma = st->vma;
 225        int rc;
 226
 227        /* Do not allow range to wrap the address space. */
 228        if ((msg->npages > (LONG_MAX >> PAGE_SHIFT)) ||
 229            ((unsigned long)(msg->npages << PAGE_SHIFT) >= -st->va))
 230                return -EINVAL;
 231
 232        /* Range chunks must be contiguous in va space. */
 233        if ((msg->va != st->va) ||
 234            ((msg->va+(msg->npages<<PAGE_SHIFT)) > vma->vm_end))
 235                return -EINVAL;
 236
 237        rc = xen_remap_domain_gfn_range(vma,
 238                                        msg->va & PAGE_MASK,
 239                                        msg->mfn, msg->npages,
 240                                        vma->vm_page_prot,
 241                                        st->domain, NULL);
 242        if (rc < 0)
 243                return rc;
 244
 245        st->va += msg->npages << PAGE_SHIFT;
 246
 247        return 0;
 248}
 249
 250static long privcmd_ioctl_mmap(struct file *file, void __user *udata)
 251{
 252        struct privcmd_data *data = file->private_data;
 253        struct privcmd_mmap mmapcmd;
 254        struct mm_struct *mm = current->mm;
 255        struct vm_area_struct *vma;
 256        int rc;
 257        LIST_HEAD(pagelist);
 258        struct mmap_gfn_state state;
 259
 260        /* We only support privcmd_ioctl_mmap_batch for auto translated. */
 261        if (xen_feature(XENFEAT_auto_translated_physmap))
 262                return -ENOSYS;
 263
 264        if (copy_from_user(&mmapcmd, udata, sizeof(mmapcmd)))
 265                return -EFAULT;
 266
 267        /* If restriction is in place, check the domid matches */
 268        if (data->domid != DOMID_INVALID && data->domid != mmapcmd.dom)
 269                return -EPERM;
 270
 271        rc = gather_array(&pagelist,
 272                          mmapcmd.num, sizeof(struct privcmd_mmap_entry),
 273                          mmapcmd.entry);
 274
 275        if (rc || list_empty(&pagelist))
 276                goto out;
 277
 278        mmap_write_lock(mm);
 279
 280        {
 281                struct page *page = list_first_entry(&pagelist,
 282                                                     struct page, lru);
 283                struct privcmd_mmap_entry *msg = page_address(page);
 284
 285                vma = find_vma(mm, msg->va);
 286                rc = -EINVAL;
 287
 288                if (!vma || (msg->va != vma->vm_start) || vma->vm_private_data)
 289                        goto out_up;
 290                vma->vm_private_data = PRIV_VMA_LOCKED;
 291        }
 292
 293        state.va = vma->vm_start;
 294        state.vma = vma;
 295        state.domain = mmapcmd.dom;
 296
 297        rc = traverse_pages(mmapcmd.num, sizeof(struct privcmd_mmap_entry),
 298                            &pagelist,
 299                            mmap_gfn_range, &state);
 300
 301
 302out_up:
 303        mmap_write_unlock(mm);
 304
 305out:
 306        free_page_list(&pagelist);
 307
 308        return rc;
 309}
 310
 311struct mmap_batch_state {
 312        domid_t domain;
 313        unsigned long va;
 314        struct vm_area_struct *vma;
 315        int index;
 316        /* A tristate:
 317         *      0 for no errors
 318         *      1 if at least one error has happened (and no
 319         *          -ENOENT errors have happened)
 320         *      -ENOENT if at least 1 -ENOENT has happened.
 321         */
 322        int global_error;
 323        int version;
 324
 325        /* User-space gfn array to store errors in the second pass for V1. */
 326        xen_pfn_t __user *user_gfn;
 327        /* User-space int array to store errors in the second pass for V2. */
 328        int __user *user_err;
 329};
 330
 331/* auto translated dom0 note: if domU being created is PV, then gfn is
 332 * mfn(addr on bus). If it's auto xlated, then gfn is pfn (input to HAP).
 333 */
 334static int mmap_batch_fn(void *data, int nr, void *state)
 335{
 336        xen_pfn_t *gfnp = data;
 337        struct mmap_batch_state *st = state;
 338        struct vm_area_struct *vma = st->vma;
 339        struct page **pages = vma->vm_private_data;
 340        struct page **cur_pages = NULL;
 341        int ret;
 342
 343        if (xen_feature(XENFEAT_auto_translated_physmap))
 344                cur_pages = &pages[st->index];
 345
 346        BUG_ON(nr < 0);
 347        ret = xen_remap_domain_gfn_array(st->vma, st->va & PAGE_MASK, gfnp, nr,
 348                                         (int *)gfnp, st->vma->vm_page_prot,
 349                                         st->domain, cur_pages);
 350
 351        /* Adjust the global_error? */
 352        if (ret != nr) {
 353                if (ret == -ENOENT)
 354                        st->global_error = -ENOENT;
 355                else {
 356                        /* Record that at least one error has happened. */
 357                        if (st->global_error == 0)
 358                                st->global_error = 1;
 359                }
 360        }
 361        st->va += XEN_PAGE_SIZE * nr;
 362        st->index += nr / XEN_PFN_PER_PAGE;
 363
 364        return 0;
 365}
 366
 367static int mmap_return_error(int err, struct mmap_batch_state *st)
 368{
 369        int ret;
 370
 371        if (st->version == 1) {
 372                if (err) {
 373                        xen_pfn_t gfn;
 374
 375                        ret = get_user(gfn, st->user_gfn);
 376                        if (ret < 0)
 377                                return ret;
 378                        /*
 379                         * V1 encodes the error codes in the 32bit top
 380                         * nibble of the gfn (with its known
 381                         * limitations vis-a-vis 64 bit callers).
 382                         */
 383                        gfn |= (err == -ENOENT) ?
 384                                PRIVCMD_MMAPBATCH_PAGED_ERROR :
 385                                PRIVCMD_MMAPBATCH_MFN_ERROR;
 386                        return __put_user(gfn, st->user_gfn++);
 387                } else
 388                        st->user_gfn++;
 389        } else { /* st->version == 2 */
 390                if (err)
 391                        return __put_user(err, st->user_err++);
 392                else
 393                        st->user_err++;
 394        }
 395
 396        return 0;
 397}
 398
 399static int mmap_return_errors(void *data, int nr, void *state)
 400{
 401        struct mmap_batch_state *st = state;
 402        int *errs = data;
 403        int i;
 404        int ret;
 405
 406        for (i = 0; i < nr; i++) {
 407                ret = mmap_return_error(errs[i], st);
 408                if (ret < 0)
 409                        return ret;
 410        }
 411        return 0;
 412}
 413
 414/* Allocate pfns that are then mapped with gfns from foreign domid. Update
 415 * the vma with the page info to use later.
 416 * Returns: 0 if success, otherwise -errno
 417 */
 418static int alloc_empty_pages(struct vm_area_struct *vma, int numpgs)
 419{
 420        int rc;
 421        struct page **pages;
 422
 423        pages = kcalloc(numpgs, sizeof(pages[0]), GFP_KERNEL);
 424        if (pages == NULL)
 425                return -ENOMEM;
 426
 427        rc = xen_alloc_unpopulated_pages(numpgs, pages);
 428        if (rc != 0) {
 429                pr_warn("%s Could not alloc %d pfns rc:%d\n", __func__,
 430                        numpgs, rc);
 431                kfree(pages);
 432                return -ENOMEM;
 433        }
 434        BUG_ON(vma->vm_private_data != NULL);
 435        vma->vm_private_data = pages;
 436
 437        return 0;
 438}
 439
 440static const struct vm_operations_struct privcmd_vm_ops;
 441
 442static long privcmd_ioctl_mmap_batch(
 443        struct file *file, void __user *udata, int version)
 444{
 445        struct privcmd_data *data = file->private_data;
 446        int ret;
 447        struct privcmd_mmapbatch_v2 m;
 448        struct mm_struct *mm = current->mm;
 449        struct vm_area_struct *vma;
 450        unsigned long nr_pages;
 451        LIST_HEAD(pagelist);
 452        struct mmap_batch_state state;
 453
 454        switch (version) {
 455        case 1:
 456                if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch)))
 457                        return -EFAULT;
 458                /* Returns per-frame error in m.arr. */
 459                m.err = NULL;
 460                if (!access_ok(m.arr, m.num * sizeof(*m.arr)))
 461                        return -EFAULT;
 462                break;
 463        case 2:
 464                if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch_v2)))
 465                        return -EFAULT;
 466                /* Returns per-frame error code in m.err. */
 467                if (!access_ok(m.err, m.num * (sizeof(*m.err))))
 468                        return -EFAULT;
 469                break;
 470        default:
 471                return -EINVAL;
 472        }
 473
 474        /* If restriction is in place, check the domid matches */
 475        if (data->domid != DOMID_INVALID && data->domid != m.dom)
 476                return -EPERM;
 477
 478        nr_pages = DIV_ROUND_UP(m.num, XEN_PFN_PER_PAGE);
 479        if ((m.num <= 0) || (nr_pages > (LONG_MAX >> PAGE_SHIFT)))
 480                return -EINVAL;
 481
 482        ret = gather_array(&pagelist, m.num, sizeof(xen_pfn_t), m.arr);
 483
 484        if (ret)
 485                goto out;
 486        if (list_empty(&pagelist)) {
 487                ret = -EINVAL;
 488                goto out;
 489        }
 490
 491        if (version == 2) {
 492                /* Zero error array now to only copy back actual errors. */
 493                if (clear_user(m.err, sizeof(int) * m.num)) {
 494                        ret = -EFAULT;
 495                        goto out;
 496                }
 497        }
 498
 499        mmap_write_lock(mm);
 500
 501        vma = find_vma(mm, m.addr);
 502        if (!vma ||
 503            vma->vm_ops != &privcmd_vm_ops) {
 504                ret = -EINVAL;
 505                goto out_unlock;
 506        }
 507
 508        /*
 509         * Caller must either:
 510         *
 511         * Map the whole VMA range, which will also allocate all the
 512         * pages required for the auto_translated_physmap case.
 513         *
 514         * Or
 515         *
 516         * Map unmapped holes left from a previous map attempt (e.g.,
 517         * because those foreign frames were previously paged out).
 518         */
 519        if (vma->vm_private_data == NULL) {
 520                if (m.addr != vma->vm_start ||
 521                    m.addr + (nr_pages << PAGE_SHIFT) != vma->vm_end) {
 522                        ret = -EINVAL;
 523                        goto out_unlock;
 524                }
 525                if (xen_feature(XENFEAT_auto_translated_physmap)) {
 526                        ret = alloc_empty_pages(vma, nr_pages);
 527                        if (ret < 0)
 528                                goto out_unlock;
 529                } else
 530                        vma->vm_private_data = PRIV_VMA_LOCKED;
 531        } else {
 532                if (m.addr < vma->vm_start ||
 533                    m.addr + (nr_pages << PAGE_SHIFT) > vma->vm_end) {
 534                        ret = -EINVAL;
 535                        goto out_unlock;
 536                }
 537                if (privcmd_vma_range_is_mapped(vma, m.addr, nr_pages)) {
 538                        ret = -EINVAL;
 539                        goto out_unlock;
 540                }
 541        }
 542
 543        state.domain        = m.dom;
 544        state.vma           = vma;
 545        state.va            = m.addr;
 546        state.index         = 0;
 547        state.global_error  = 0;
 548        state.version       = version;
 549
 550        BUILD_BUG_ON(((PAGE_SIZE / sizeof(xen_pfn_t)) % XEN_PFN_PER_PAGE) != 0);
 551        /* mmap_batch_fn guarantees ret == 0 */
 552        BUG_ON(traverse_pages_block(m.num, sizeof(xen_pfn_t),
 553                                    &pagelist, mmap_batch_fn, &state));
 554
 555        mmap_write_unlock(mm);
 556
 557        if (state.global_error) {
 558                /* Write back errors in second pass. */
 559                state.user_gfn = (xen_pfn_t *)m.arr;
 560                state.user_err = m.err;
 561                ret = traverse_pages_block(m.num, sizeof(xen_pfn_t),
 562                                           &pagelist, mmap_return_errors, &state);
 563        } else
 564                ret = 0;
 565
 566        /* If we have not had any EFAULT-like global errors then set the global
 567         * error to -ENOENT if necessary. */
 568        if ((ret == 0) && (state.global_error == -ENOENT))
 569                ret = -ENOENT;
 570
 571out:
 572        free_page_list(&pagelist);
 573        return ret;
 574
 575out_unlock:
 576        mmap_write_unlock(mm);
 577        goto out;
 578}
 579
 580static int lock_pages(
 581        struct privcmd_dm_op_buf kbufs[], unsigned int num,
 582        struct page *pages[], unsigned int nr_pages, unsigned int *pinned)
 583{
 584        unsigned int i;
 585
 586        for (i = 0; i < num; i++) {
 587                unsigned int requested;
 588                int page_count;
 589
 590                requested = DIV_ROUND_UP(
 591                        offset_in_page(kbufs[i].uptr) + kbufs[i].size,
 592                        PAGE_SIZE);
 593                if (requested > nr_pages)
 594                        return -ENOSPC;
 595
 596                page_count = pin_user_pages_fast(
 597                        (unsigned long) kbufs[i].uptr,
 598                        requested, FOLL_WRITE, pages);
 599                if (page_count < 0)
 600                        return page_count;
 601
 602                *pinned += page_count;
 603                nr_pages -= page_count;
 604                pages += page_count;
 605        }
 606
 607        return 0;
 608}
 609
 610static void unlock_pages(struct page *pages[], unsigned int nr_pages)
 611{
 612        unpin_user_pages_dirty_lock(pages, nr_pages, true);
 613}
 614
 615static long privcmd_ioctl_dm_op(struct file *file, void __user *udata)
 616{
 617        struct privcmd_data *data = file->private_data;
 618        struct privcmd_dm_op kdata;
 619        struct privcmd_dm_op_buf *kbufs;
 620        unsigned int nr_pages = 0;
 621        struct page **pages = NULL;
 622        struct xen_dm_op_buf *xbufs = NULL;
 623        unsigned int i;
 624        long rc;
 625        unsigned int pinned = 0;
 626
 627        if (copy_from_user(&kdata, udata, sizeof(kdata)))
 628                return -EFAULT;
 629
 630        /* If restriction is in place, check the domid matches */
 631        if (data->domid != DOMID_INVALID && data->domid != kdata.dom)
 632                return -EPERM;
 633
 634        if (kdata.num == 0)
 635                return 0;
 636
 637        if (kdata.num > privcmd_dm_op_max_num)
 638                return -E2BIG;
 639
 640        kbufs = kcalloc(kdata.num, sizeof(*kbufs), GFP_KERNEL);
 641        if (!kbufs)
 642                return -ENOMEM;
 643
 644        if (copy_from_user(kbufs, kdata.ubufs,
 645                           sizeof(*kbufs) * kdata.num)) {
 646                rc = -EFAULT;
 647                goto out;
 648        }
 649
 650        for (i = 0; i < kdata.num; i++) {
 651                if (kbufs[i].size > privcmd_dm_op_buf_max_size) {
 652                        rc = -E2BIG;
 653                        goto out;
 654                }
 655
 656                if (!access_ok(kbufs[i].uptr,
 657                               kbufs[i].size)) {
 658                        rc = -EFAULT;
 659                        goto out;
 660                }
 661
 662                nr_pages += DIV_ROUND_UP(
 663                        offset_in_page(kbufs[i].uptr) + kbufs[i].size,
 664                        PAGE_SIZE);
 665        }
 666
 667        pages = kcalloc(nr_pages, sizeof(*pages), GFP_KERNEL);
 668        if (!pages) {
 669                rc = -ENOMEM;
 670                goto out;
 671        }
 672
 673        xbufs = kcalloc(kdata.num, sizeof(*xbufs), GFP_KERNEL);
 674        if (!xbufs) {
 675                rc = -ENOMEM;
 676                goto out;
 677        }
 678
 679        rc = lock_pages(kbufs, kdata.num, pages, nr_pages, &pinned);
 680        if (rc < 0) {
 681                nr_pages = pinned;
 682                goto out;
 683        }
 684
 685        for (i = 0; i < kdata.num; i++) {
 686                set_xen_guest_handle(xbufs[i].h, kbufs[i].uptr);
 687                xbufs[i].size = kbufs[i].size;
 688        }
 689
 690        xen_preemptible_hcall_begin();
 691        rc = HYPERVISOR_dm_op(kdata.dom, kdata.num, xbufs);
 692        xen_preemptible_hcall_end();
 693
 694out:
 695        unlock_pages(pages, nr_pages);
 696        kfree(xbufs);
 697        kfree(pages);
 698        kfree(kbufs);
 699
 700        return rc;
 701}
 702
 703static long privcmd_ioctl_restrict(struct file *file, void __user *udata)
 704{
 705        struct privcmd_data *data = file->private_data;
 706        domid_t dom;
 707
 708        if (copy_from_user(&dom, udata, sizeof(dom)))
 709                return -EFAULT;
 710
 711        /* Set restriction to the specified domain, or check it matches */
 712        if (data->domid == DOMID_INVALID)
 713                data->domid = dom;
 714        else if (data->domid != dom)
 715                return -EINVAL;
 716
 717        return 0;
 718}
 719
 720static long privcmd_ioctl_mmap_resource(struct file *file,
 721                                struct privcmd_mmap_resource __user *udata)
 722{
 723        struct privcmd_data *data = file->private_data;
 724        struct mm_struct *mm = current->mm;
 725        struct vm_area_struct *vma;
 726        struct privcmd_mmap_resource kdata;
 727        xen_pfn_t *pfns = NULL;
 728        struct xen_mem_acquire_resource xdata = { };
 729        int rc;
 730
 731        if (copy_from_user(&kdata, udata, sizeof(kdata)))
 732                return -EFAULT;
 733
 734        /* If restriction is in place, check the domid matches */
 735        if (data->domid != DOMID_INVALID && data->domid != kdata.dom)
 736                return -EPERM;
 737
 738        /* Both fields must be set or unset */
 739        if (!!kdata.addr != !!kdata.num)
 740                return -EINVAL;
 741
 742        xdata.domid = kdata.dom;
 743        xdata.type = kdata.type;
 744        xdata.id = kdata.id;
 745
 746        if (!kdata.addr && !kdata.num) {
 747                /* Query the size of the resource. */
 748                rc = HYPERVISOR_memory_op(XENMEM_acquire_resource, &xdata);
 749                if (rc)
 750                        return rc;
 751                return __put_user(xdata.nr_frames, &udata->num);
 752        }
 753
 754        mmap_write_lock(mm);
 755
 756        vma = find_vma(mm, kdata.addr);
 757        if (!vma || vma->vm_ops != &privcmd_vm_ops) {
 758                rc = -EINVAL;
 759                goto out;
 760        }
 761
 762        pfns = kcalloc(kdata.num, sizeof(*pfns), GFP_KERNEL);
 763        if (!pfns) {
 764                rc = -ENOMEM;
 765                goto out;
 766        }
 767
 768        if (IS_ENABLED(CONFIG_XEN_AUTO_XLATE) &&
 769            xen_feature(XENFEAT_auto_translated_physmap)) {
 770                unsigned int nr = DIV_ROUND_UP(kdata.num, XEN_PFN_PER_PAGE);
 771                struct page **pages;
 772                unsigned int i;
 773
 774                rc = alloc_empty_pages(vma, nr);
 775                if (rc < 0)
 776                        goto out;
 777
 778                pages = vma->vm_private_data;
 779                for (i = 0; i < kdata.num; i++) {
 780                        xen_pfn_t pfn =
 781                                page_to_xen_pfn(pages[i / XEN_PFN_PER_PAGE]);
 782
 783                        pfns[i] = pfn + (i % XEN_PFN_PER_PAGE);
 784                }
 785        } else
 786                vma->vm_private_data = PRIV_VMA_LOCKED;
 787
 788        xdata.frame = kdata.idx;
 789        xdata.nr_frames = kdata.num;
 790        set_xen_guest_handle(xdata.frame_list, pfns);
 791
 792        xen_preemptible_hcall_begin();
 793        rc = HYPERVISOR_memory_op(XENMEM_acquire_resource, &xdata);
 794        xen_preemptible_hcall_end();
 795
 796        if (rc)
 797                goto out;
 798
 799        if (IS_ENABLED(CONFIG_XEN_AUTO_XLATE) &&
 800            xen_feature(XENFEAT_auto_translated_physmap)) {
 801                rc = xen_remap_vma_range(vma, kdata.addr, kdata.num << PAGE_SHIFT);
 802        } else {
 803                unsigned int domid =
 804                        (xdata.flags & XENMEM_rsrc_acq_caller_owned) ?
 805                        DOMID_SELF : kdata.dom;
 806                int num;
 807
 808                num = xen_remap_domain_mfn_array(vma,
 809                                                 kdata.addr & PAGE_MASK,
 810                                                 pfns, kdata.num, (int *)pfns,
 811                                                 vma->vm_page_prot,
 812                                                 domid,
 813                                                 vma->vm_private_data);
 814                if (num < 0)
 815                        rc = num;
 816                else if (num != kdata.num) {
 817                        unsigned int i;
 818
 819                        for (i = 0; i < num; i++) {
 820                                rc = pfns[i];
 821                                if (rc < 0)
 822                                        break;
 823                        }
 824                } else
 825                        rc = 0;
 826        }
 827
 828out:
 829        mmap_write_unlock(mm);
 830        kfree(pfns);
 831
 832        return rc;
 833}
 834
 835static long privcmd_ioctl(struct file *file,
 836                          unsigned int cmd, unsigned long data)
 837{
 838        int ret = -ENOTTY;
 839        void __user *udata = (void __user *) data;
 840
 841        switch (cmd) {
 842        case IOCTL_PRIVCMD_HYPERCALL:
 843                ret = privcmd_ioctl_hypercall(file, udata);
 844                break;
 845
 846        case IOCTL_PRIVCMD_MMAP:
 847                ret = privcmd_ioctl_mmap(file, udata);
 848                break;
 849
 850        case IOCTL_PRIVCMD_MMAPBATCH:
 851                ret = privcmd_ioctl_mmap_batch(file, udata, 1);
 852                break;
 853
 854        case IOCTL_PRIVCMD_MMAPBATCH_V2:
 855                ret = privcmd_ioctl_mmap_batch(file, udata, 2);
 856                break;
 857
 858        case IOCTL_PRIVCMD_DM_OP:
 859                ret = privcmd_ioctl_dm_op(file, udata);
 860                break;
 861
 862        case IOCTL_PRIVCMD_RESTRICT:
 863                ret = privcmd_ioctl_restrict(file, udata);
 864                break;
 865
 866        case IOCTL_PRIVCMD_MMAP_RESOURCE:
 867                ret = privcmd_ioctl_mmap_resource(file, udata);
 868                break;
 869
 870        default:
 871                break;
 872        }
 873
 874        return ret;
 875}
 876
 877static int privcmd_open(struct inode *ino, struct file *file)
 878{
 879        struct privcmd_data *data = kzalloc(sizeof(*data), GFP_KERNEL);
 880
 881        if (!data)
 882                return -ENOMEM;
 883
 884        /* DOMID_INVALID implies no restriction */
 885        data->domid = DOMID_INVALID;
 886
 887        file->private_data = data;
 888        return 0;
 889}
 890
 891static int privcmd_release(struct inode *ino, struct file *file)
 892{
 893        struct privcmd_data *data = file->private_data;
 894
 895        kfree(data);
 896        return 0;
 897}
 898
 899static void privcmd_close(struct vm_area_struct *vma)
 900{
 901        struct page **pages = vma->vm_private_data;
 902        int numpgs = vma_pages(vma);
 903        int numgfns = (vma->vm_end - vma->vm_start) >> XEN_PAGE_SHIFT;
 904        int rc;
 905
 906        if (!xen_feature(XENFEAT_auto_translated_physmap) || !numpgs || !pages)
 907                return;
 908
 909        rc = xen_unmap_domain_gfn_range(vma, numgfns, pages);
 910        if (rc == 0)
 911                xen_free_unpopulated_pages(numpgs, pages);
 912        else
 913                pr_crit("unable to unmap MFN range: leaking %d pages. rc=%d\n",
 914                        numpgs, rc);
 915        kfree(pages);
 916}
 917
 918static vm_fault_t privcmd_fault(struct vm_fault *vmf)
 919{
 920        printk(KERN_DEBUG "privcmd_fault: vma=%p %lx-%lx, pgoff=%lx, uv=%p\n",
 921               vmf->vma, vmf->vma->vm_start, vmf->vma->vm_end,
 922               vmf->pgoff, (void *)vmf->address);
 923
 924        return VM_FAULT_SIGBUS;
 925}
 926
 927static const struct vm_operations_struct privcmd_vm_ops = {
 928        .close = privcmd_close,
 929        .fault = privcmd_fault
 930};
 931
 932static int privcmd_mmap(struct file *file, struct vm_area_struct *vma)
 933{
 934        /* DONTCOPY is essential for Xen because copy_page_range doesn't know
 935         * how to recreate these mappings */
 936        vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTCOPY |
 937                         VM_DONTEXPAND | VM_DONTDUMP;
 938        vma->vm_ops = &privcmd_vm_ops;
 939        vma->vm_private_data = NULL;
 940
 941        return 0;
 942}
 943
 944/*
 945 * For MMAPBATCH*. This allows asserting the singleshot mapping
 946 * on a per pfn/pte basis. Mapping calls that fail with ENOENT
 947 * can be then retried until success.
 948 */
 949static int is_mapped_fn(pte_t *pte, unsigned long addr, void *data)
 950{
 951        return pte_none(*pte) ? 0 : -EBUSY;
 952}
 953
 954static int privcmd_vma_range_is_mapped(
 955                   struct vm_area_struct *vma,
 956                   unsigned long addr,
 957                   unsigned long nr_pages)
 958{
 959        return apply_to_page_range(vma->vm_mm, addr, nr_pages << PAGE_SHIFT,
 960                                   is_mapped_fn, NULL) != 0;
 961}
 962
 963const struct file_operations xen_privcmd_fops = {
 964        .owner = THIS_MODULE,
 965        .unlocked_ioctl = privcmd_ioctl,
 966        .open = privcmd_open,
 967        .release = privcmd_release,
 968        .mmap = privcmd_mmap,
 969};
 970EXPORT_SYMBOL_GPL(xen_privcmd_fops);
 971
 972static struct miscdevice privcmd_dev = {
 973        .minor = MISC_DYNAMIC_MINOR,
 974        .name = "xen/privcmd",
 975        .fops = &xen_privcmd_fops,
 976};
 977
 978static int __init privcmd_init(void)
 979{
 980        int err;
 981
 982        if (!xen_domain())
 983                return -ENODEV;
 984
 985        err = misc_register(&privcmd_dev);
 986        if (err != 0) {
 987                pr_err("Could not register Xen privcmd device\n");
 988                return err;
 989        }
 990
 991        err = misc_register(&xen_privcmdbuf_dev);
 992        if (err != 0) {
 993                pr_err("Could not register Xen hypercall-buf device\n");
 994                misc_deregister(&privcmd_dev);
 995                return err;
 996        }
 997
 998        return 0;
 999}
1000
1001static void __exit privcmd_exit(void)
1002{
1003        misc_deregister(&privcmd_dev);
1004        misc_deregister(&xen_privcmdbuf_dev);
1005}
1006
1007module_init(privcmd_init);
1008module_exit(privcmd_exit);
1009