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
  26 #include <asm/pgalloc.h>
  27 #include <asm/pgtable.h>
  28 #include <asm/tlb.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/features.h>
  36 #include <xen/page.h>
  37 #include <xen/xen-ops.h>
  38 #include <xen/balloon.h>
  39
  40 #include "privcmd.h"
  41
  42 MODULE_LICENSE("GPL");
  43
  44 #define PRIV_VMA_LOCKED ((void *)1)
  45
  46 static int privcmd_vma_range_is_mapped(
  47                struct vm_area_struct *vma,
  48                unsigned long addr,
  49                unsigned long nr_pages);
  50
  51 static long privcmd_ioctl_hypercall(void __user *udata)
  52 {
  53         struct privcmd_hypercall hypercall;
  54         long ret;
  55
  56         if (copy_from_user(&hypercall, udata, sizeof(hypercall)))
  57                 return -EFAULT;
  58
  59         xen_preemptible_hcall_begin();
  60         ret = privcmd_call(hypercall.op,
  61                            hypercall.arg[0], hypercall.arg[1],
  62                            hypercall.arg[2], hypercall.arg[3],
  63                            hypercall.arg[4]);
  64         xen_preemptible_hcall_end();
  65
  66         return ret;
  67 }
  68
  69 static void free_page_list(struct list_head *pages)
  70 {
  71         struct page *p, *n;
  72
  73         list_for_each_entry_safe(p, n, pages, lru)
  74                 __free_page(p);
  75
  76         INIT_LIST_HEAD(pages);
  77 }
  78
  79 /*
  80  * Given an array of items in userspace, return a list of pages
  81  * containing the data.  If copying fails, either because of memory
  82  * allocation failure or a problem reading user memory, return an
  83  * error code; its up to the caller to dispose of any partial list.
  84  */
  85 static int gather_array(struct list_head *pagelist,
  86                         unsigned nelem, size_t size,
  87                         const void __user *data)
  88 {
  89         unsigned pageidx;
  90         void *pagedata;
  91         int ret;
  92
  93         if (size > PAGE_SIZE)
  94                 return 0;
  95
  96         pageidx = PAGE_SIZE;
  97         pagedata = NULL;        /* quiet, gcc */
  98         while (nelem--) {
  99                 if (pageidx > PAGE_SIZE-size) {
 100                         struct page *page = alloc_page(GFP_KERNEL);
 101
 102                         ret = -ENOMEM;
 103                         if (page == NULL)
 104                                 goto fail;
 105
 106                         pagedata = page_address(page);
 107
 108                         list_add_tail(&page->lru, pagelist);
 109                         pageidx = 0;
 110                 }
 111
 112                 ret = -EFAULT;
 113                 if (copy_from_user(pagedata + pageidx, data, size))
 114                         goto fail;
 115
 116                 data += size;
 117                 pageidx += size;
 118         }
 119
 120         ret = 0;
 121
 122 fail:
 123         return ret;
 124 }
 125
 126 /*
 127  * Call function "fn" on each element of the array fragmented
 128  * over a list of pages.
 129  */
 130 static int traverse_pages(unsigned nelem, size_t size,
 131                           struct list_head *pos,
 132                           int (*fn)(void *data, void *state),
 133                           void *state)
 134 {
 135         void *pagedata;
 136         unsigned pageidx;
 137         int ret = 0;
 138
 139         BUG_ON(size > PAGE_SIZE);
 140
 141         pageidx = PAGE_SIZE;
 142         pagedata = NULL;        /* hush, gcc */
 143
 144         while (nelem--) {
 145                 if (pageidx > PAGE_SIZE-size) {
 146                         struct page *page;
 147                         pos = pos->next;
 148                         page = list_entry(pos, struct page, lru);
 149                         pagedata = page_address(page);
 150                         pageidx = 0;
 151                 }
 152
 153                 ret = (*fn)(pagedata + pageidx, state);
 154                 if (ret)
 155                         break;
 156                 pageidx += size;
 157         }
 158
 159         return ret;
 160 }
 161
 162 struct mmap_mfn_state {
 163         unsigned long va;
 164         struct vm_area_struct *vma;
 165         domid_t domain;
 166 };
 167
 168 static int mmap_mfn_range(void *data, void *state)
 169 {
 170         struct privcmd_mmap_entry *msg = data;
 171         struct mmap_mfn_state *st = state;
 172         struct vm_area_struct *vma = st->vma;
 173         int rc;
 174
 175         /* Do not allow range to wrap the address space. */
 176         if ((msg->npages > (LONG_MAX >> PAGE_SHIFT)) ||
 177             ((unsigned long)(msg->npages << PAGE_SHIFT) >= -st->va))
 178                 return -EINVAL;
 179
 180         /* Range chunks must be contiguous in va space. */
 181         if ((msg->va != st->va) ||
 182             ((msg->va+(msg->npages<<PAGE_SHIFT)) > vma->vm_end))
 183                 return -EINVAL;
 184
 185         rc = xen_remap_domain_mfn_range(vma,
 186                                         msg->va & PAGE_MASK,
 187                                         msg->mfn, msg->npages,
 188                                         vma->vm_page_prot,
 189                                         st->domain, NULL);
 190         if (rc < 0)
 191                 return rc;
 192
 193         st->va += msg->npages << PAGE_SHIFT;
 194
 195         return 0;
 196 }
 197
 198 static long privcmd_ioctl_mmap(void __user *udata)
 199 {
 200         struct privcmd_mmap mmapcmd;
 201         struct mm_struct *mm = current->mm;
 202         struct vm_area_struct *vma;
 203         int rc;
 204         LIST_HEAD(pagelist);
 205         struct mmap_mfn_state state;
 206
 207         /* We only support privcmd_ioctl_mmap_batch for auto translated. */
 208         if (xen_feature(XENFEAT_auto_translated_physmap))
 209                 return -ENOSYS;
 210
 211         if (copy_from_user(&mmapcmd, udata, sizeof(mmapcmd)))
 212                 return -EFAULT;
 213
 214         rc = gather_array(&pagelist,
 215                           mmapcmd.num, sizeof(struct privcmd_mmap_entry),
 216                           mmapcmd.entry);
 217
 218         if (rc || list_empty(&pagelist))
 219                 goto out;
 220
 221         down_write(&mm->mmap_sem);
 222
 223         {
 224                 struct page *page = list_first_entry(&pagelist,
 225                                                      struct page, lru);
 226                 struct privcmd_mmap_entry *msg = page_address(page);
 227
 228                 vma = find_vma(mm, msg->va);
 229                 rc = -EINVAL;
 230
 231                 if (!vma || (msg->va != vma->vm_start) || vma->vm_private_data)
 232                         goto out_up;
 233                 vma->vm_private_data = PRIV_VMA_LOCKED;
 234         }
 235
 236         state.va = vma->vm_start;
 237         state.vma = vma;
 238         state.domain = mmapcmd.dom;
 239
 240         rc = traverse_pages(mmapcmd.num, sizeof(struct privcmd_mmap_entry),
 241                             &pagelist,
 242                             mmap_mfn_range, &state);
 243
 244
 245 out_up:
 246         up_write(&mm->mmap_sem);
 247
 248 out:
 249         free_page_list(&pagelist);
 250
 251         return rc;
 252 }
 253
 254 struct mmap_batch_state {
 255         domid_t domain;
 256         unsigned long va;
 257         struct vm_area_struct *vma;
 258         int index;
 259         /* A tristate:
 260          *      0 for no errors
 261          *      1 if at least one error has happened (and no
 262          *          -ENOENT errors have happened)
 263          *      -ENOENT if at least 1 -ENOENT has happened.
 264          */
 265         int global_error;
 266         int version;
 267
 268         /* User-space mfn array to store errors in the second pass for V1. */
 269         xen_pfn_t __user *user_mfn;
 270         /* User-space int array to store errors in the second pass for V2. */
 271         int __user *user_err;
 272 };
 273
 274 /* auto translated dom0 note: if domU being created is PV, then mfn is
 275  * mfn(addr on bus). If it's auto xlated, then mfn is pfn (input to HAP).
 276  */
 277 static int mmap_batch_fn(void *data, void *state)
 278 {
 279         xen_pfn_t *mfnp = data;
 280         struct mmap_batch_state *st = state;
 281         struct vm_area_struct *vma = st->vma;
 282         struct page **pages = vma->vm_private_data;
 283         struct page *cur_page = NULL;
 284         int ret;
 285
 286         if (xen_feature(XENFEAT_auto_translated_physmap))
 287                 cur_page = pages[st->index++];
 288
 289         ret = xen_remap_domain_mfn_range(st->vma, st->va & PAGE_MASK, *mfnp, 1,
 290                                          st->vma->vm_page_prot, st->domain,
 291                                          &cur_page);
 292
 293         /* Store error code for second pass. */
 294         if (st->version == 1) {
 295                 if (ret < 0) {
 296                         /*
 297                          * V1 encodes the error codes in the 32bit top nibble of the
 298                          * mfn (with its known limitations vis-a-vis 64 bit callers).
 299                          */
 300                         *mfnp |= (ret == -ENOENT) ?
 301                                                 PRIVCMD_MMAPBATCH_PAGED_ERROR :
 302                                                 PRIVCMD_MMAPBATCH_MFN_ERROR;
 303                 }
 304         } else { /* st->version == 2 */
 305                 *((int *) mfnp) = ret;
 306         }
 307
 308         /* And see if it affects the global_error. */
 309         if (ret < 0) {
 310                 if (ret == -ENOENT)
 311                         st->global_error = -ENOENT;
 312                 else {
 313                         /* Record that at least one error has happened. */
 314                         if (st->global_error == 0)
 315                                 st->global_error = 1;
 316                 }
 317         }
 318         st->va += PAGE_SIZE;
 319
 320         return 0;
 321 }
 322
 323 static int mmap_return_errors(void *data, void *state)
 324 {
 325         struct mmap_batch_state *st = state;
 326
 327         if (st->version == 1) {
 328                 xen_pfn_t mfnp = *((xen_pfn_t *) data);
 329                 if (mfnp & PRIVCMD_MMAPBATCH_MFN_ERROR)
 330                         return __put_user(mfnp, st->user_mfn++);
 331                 else
 332                         st->user_mfn++;
 333         } else { /* st->version == 2 */
 334                 int err = *((int *) data);
 335                 if (err)
 336                         return __put_user(err, st->user_err++);
 337                 else
 338                         st->user_err++;
 339         }
 340
 341         return 0;
 342 }
 343
 344 /* Allocate pfns that are then mapped with gmfns from foreign domid. Update
 345  * the vma with the page info to use later.
 346  * Returns: 0 if success, otherwise -errno
 347  */
 348 static int alloc_empty_pages(struct vm_area_struct *vma, int numpgs)
 349 {
 350         int rc;
 351         struct page **pages;
 352
 353         pages = kcalloc(numpgs, sizeof(pages[0]), GFP_KERNEL);
 354         if (pages == NULL)
 355                 return -ENOMEM;
 356
 357         rc = alloc_xenballooned_pages(numpgs, pages, 0);
 358         if (rc != 0) {
 359                 pr_warn("%s Could not alloc %d pfns rc:%d\n", __func__,
 360                         numpgs, rc);
 361                 kfree(pages);
 362                 return -ENOMEM;
 363         }
 364         BUG_ON(vma->vm_private_data != NULL);
 365         vma->vm_private_data = pages;
 366
 367         return 0;
 368 }
 369
 370 static struct vm_operations_struct privcmd_vm_ops;
 371
 372 static long privcmd_ioctl_mmap_batch(void __user *udata, int version)
 373 {
 374         int ret;
 375         struct privcmd_mmapbatch_v2 m;
 376         struct mm_struct *mm = current->mm;
 377         struct vm_area_struct *vma;
 378         unsigned long nr_pages;
 379         LIST_HEAD(pagelist);
 380         struct mmap_batch_state state;
 381
 382         switch (version) {
 383         case 1:
 384                 if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch)))
 385                         return -EFAULT;
 386                 /* Returns per-frame error in m.arr. */
 387                 m.err = NULL;
 388                 if (!access_ok(VERIFY_WRITE, m.arr, m.num * sizeof(*m.arr)))
 389                         return -EFAULT;
 390                 break;
 391         case 2:
 392                 if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch_v2)))
 393                         return -EFAULT;
 394                 /* Returns per-frame error code in m.err. */
 395                 if (!access_ok(VERIFY_WRITE, m.err, m.num * (sizeof(*m.err))))
 396                         return -EFAULT;
 397                 break;
 398         default:
 399                 return -EINVAL;
 400         }
 401
 402         nr_pages = m.num;
 403         if ((m.num <= 0) || (nr_pages > (LONG_MAX >> PAGE_SHIFT)))
 404                 return -EINVAL;
 405
 406         ret = gather_array(&pagelist, m.num, sizeof(xen_pfn_t), m.arr);
 407
 408         if (ret)
 409                 goto out;
 410         if (list_empty(&pagelist)) {
 411                 ret = -EINVAL;
 412                 goto out;
 413         }
 414
 415         if (version == 2) {
 416                 /* Zero error array now to only copy back actual errors. */
 417                 if (clear_user(m.err, sizeof(int) * m.num)) {
 418                         ret = -EFAULT;
 419                         goto out;
 420                 }
 421         }
 422
 423         down_write(&mm->mmap_sem);
 424
 425         vma = find_vma(mm, m.addr);
 426         if (!vma ||
 427             vma->vm_ops != &privcmd_vm_ops) {
 428                 ret = -EINVAL;
 429                 goto out_unlock;
 430         }
 431
 432         /*
 433          * Caller must either:
 434          *
 435          * Map the whole VMA range, which will also allocate all the
 436          * pages required for the auto_translated_physmap case.
 437          *
 438          * Or
 439          *
 440          * Map unmapped holes left from a previous map attempt (e.g.,
 441          * because those foreign frames were previously paged out).
 442          */
 443         if (vma->vm_private_data == NULL) {
 444                 if (m.addr != vma->vm_start ||
 445                     m.addr + (nr_pages << PAGE_SHIFT) != vma->vm_end) {
 446                         ret = -EINVAL;
 447                         goto out_unlock;
 448                 }
 449                 if (xen_feature(XENFEAT_auto_translated_physmap)) {
 450                         ret = alloc_empty_pages(vma, m.num);
 451                         if (ret < 0)
 452                                 goto out_unlock;
 453                 } else
 454                         vma->vm_private_data = PRIV_VMA_LOCKED;
 455         } else {
 456                 if (m.addr < vma->vm_start ||
 457                     m.addr + (nr_pages << PAGE_SHIFT) > vma->vm_end) {
 458                         ret = -EINVAL;
 459                         goto out_unlock;
 460                 }
 461                 if (privcmd_vma_range_is_mapped(vma, m.addr, nr_pages)) {
 462                         ret = -EINVAL;
 463                         goto out_unlock;
 464                 }
 465         }
 466
 467         state.domain        = m.dom;
 468         state.vma           = vma;
 469         state.va            = m.addr;
 470         state.index         = 0;
 471         state.global_error  = 0;
 472         state.version       = version;
 473
 474         /* mmap_batch_fn guarantees ret == 0 */
 475         BUG_ON(traverse_pages(m.num, sizeof(xen_pfn_t),
 476                              &pagelist, mmap_batch_fn, &state));
 477
 478         up_write(&mm->mmap_sem);
 479
 480         if (state.global_error) {
 481                 /* Write back errors in second pass. */
 482                 state.user_mfn = (xen_pfn_t *)m.arr;
 483                 state.user_err = m.err;
 484                 ret = traverse_pages(m.num, sizeof(xen_pfn_t),
 485                                                          &pagelist, mmap_return_errors, &state);
 486         } else
 487                 ret = 0;
 488
 489         /* If we have not had any EFAULT-like global errors then set the global
 490          * error to -ENOENT if necessary. */
 491         if ((ret == 0) && (state.global_error == -ENOENT))
 492                 ret = -ENOENT;
 493
 494 out:
 495         free_page_list(&pagelist);
 496         return ret;
 497
 498 out_unlock:
 499         up_write(&mm->mmap_sem);
 500         goto out;
 501 }
 502
 503 static long privcmd_ioctl(struct file *file,
 504                           unsigned int cmd, unsigned long data)
 505 {
 506         int ret = -ENOSYS;
 507         void __user *udata = (void __user *) data;
 508
 509         switch (cmd) {
 510         case IOCTL_PRIVCMD_HYPERCALL:
 511                 ret = privcmd_ioctl_hypercall(udata);
 512                 break;
 513
 514         case IOCTL_PRIVCMD_MMAP:
 515                 ret = privcmd_ioctl_mmap(udata);
 516                 break;
 517
 518         case IOCTL_PRIVCMD_MMAPBATCH:
 519                 ret = privcmd_ioctl_mmap_batch(udata, 1);
 520                 break;
 521
 522         case IOCTL_PRIVCMD_MMAPBATCH_V2:
 523                 ret = privcmd_ioctl_mmap_batch(udata, 2);
 524                 break;
 525
 526         default:
 527                 ret = -EINVAL;
 528                 break;
 529         }
 530
 531         return ret;
 532 }
 533
 534 static void privcmd_close(struct vm_area_struct *vma)
 535 {
 536         struct page **pages = vma->vm_private_data;
 537         int numpgs = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
 538         int rc;
 539
 540         if (!xen_feature(XENFEAT_auto_translated_physmap) || !numpgs || !pages)
 541                 return;
 542
 543         rc = xen_unmap_domain_mfn_range(vma, numpgs, pages);
 544         if (rc == 0)
 545                 free_xenballooned_pages(numpgs, pages);
 546         else
 547                 pr_crit("unable to unmap MFN range: leaking %d pages. rc=%d\n",
 548                         numpgs, rc);
 549         kfree(pages);
 550 }
 551
 552 static int privcmd_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 553 {
 554         printk(KERN_DEBUG "privcmd_fault: vma=%p %lx-%lx, pgoff=%lx, uv=%p\n",
 555                vma, vma->vm_start, vma->vm_end,
 556                vmf->pgoff, vmf->virtual_address);
 557
 558         return VM_FAULT_SIGBUS;
 559 }
 560
 561 static struct vm_operations_struct privcmd_vm_ops = {
 562         .close = privcmd_close,
 563         .fault = privcmd_fault
 564 };
 565
 566 static int privcmd_mmap(struct file *file, struct vm_area_struct *vma)
 567 {
 568         /* DONTCOPY is essential for Xen because copy_page_range doesn't know
 569          * how to recreate these mappings */
 570         vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTCOPY |
 571                          VM_DONTEXPAND | VM_DONTDUMP;
 572         vma->vm_ops = &privcmd_vm_ops;
 573         vma->vm_private_data = NULL;
 574
 575         return 0;
 576 }
 577
 578 /*
 579  * For MMAPBATCH*. This allows asserting the singleshot mapping
 580  * on a per pfn/pte basis. Mapping calls that fail with ENOENT
 581  * can be then retried until success.
 582  */
 583 static int is_mapped_fn(pte_t *pte, struct page *pmd_page,
 584                         unsigned long addr, void *data)
 585 {
 586         return pte_none(*pte) ? 0 : -EBUSY;
 587 }
 588
 589 static int privcmd_vma_range_is_mapped(
 590                    struct vm_area_struct *vma,
 591                    unsigned long addr,
 592                    unsigned long nr_pages)
 593 {
 594         return apply_to_page_range(vma->vm_mm, addr, nr_pages << PAGE_SHIFT,
 595                                    is_mapped_fn, NULL) != 0;
 596 }
 597
 598 const struct file_operations xen_privcmd_fops = {
 599         .owner = THIS_MODULE,
 600         .unlocked_ioctl = privcmd_ioctl,
 601         .mmap = privcmd_mmap,
 602 };
 603 EXPORT_SYMBOL_GPL(xen_privcmd_fops);
 604
 605 static struct miscdevice privcmd_dev = {
 606         .minor = MISC_DYNAMIC_MINOR,
 607         .name = "xen/privcmd",
 608         .fops = &xen_privcmd_fops,
 609 };
 610
 611 static int __init privcmd_init(void)
 612 {
 613         int err;
 614
 615         if (!xen_domain())
 616                 return -ENODEV;
 617
 618         err = misc_register(&privcmd_dev);
 619         if (err != 0) {
 620                 pr_err("Could not register Xen privcmd device\n");
 621                 return err;
 622         }
 623         return 0;
 624 }
 625
 626 static void __exit privcmd_exit(void)
 627 {
 628         misc_deregister(&privcmd_dev);
 629 }
 630
 631 module_init(privcmd_init);
 632 module_exit(privcmd_exit);