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 #include <linux/kernel.h>
  10 #include <linux/module.h>
  11 #include <linux/sched.h>
  12 #include <linux/slab.h>
  13 #include <linux/string.h>
  14 #include <linux/errno.h>
  15 #include <linux/mm.h>
  16 #include <linux/mman.h>
  17 #include <linux/uaccess.h>
  18 #include <linux/swap.h>
  19 #include <linux/highmem.h>
  20 #include <linux/pagemap.h>
  21 #include <linux/seq_file.h>
  22 #include <linux/miscdevice.h>
  23
  24 #include <asm/pgalloc.h>
  25 #include <asm/pgtable.h>
  26 #include <asm/tlb.h>
  27 #include <asm/xen/hypervisor.h>
  28 #include <asm/xen/hypercall.h>
  29
  30 #include <xen/xen.h>
  31 #include <xen/privcmd.h>
  32 #include <xen/interface/xen.h>
  33 #include <xen/features.h>
  34 #include <xen/page.h>
  35 #include <xen/xen-ops.h>
  36 #include <xen/balloon.h>
  37
  38 #include "privcmd.h"
  39
  40 MODULE_LICENSE("GPL");
  41
  42 #define PRIV_VMA_LOCKED ((void *)1)
  43
  44 #ifndef HAVE_ARCH_PRIVCMD_MMAP
  45 static int privcmd_enforce_singleshot_mapping(struct vm_area_struct *vma);
  46 #endif
  47
  48 static long privcmd_ioctl_hypercall(void __user *udata)
  49 {
  50         struct privcmd_hypercall hypercall;
  51         long ret;
  52
  53         if (copy_from_user(&hypercall, udata, sizeof(hypercall)))
  54                 return -EFAULT;
  55
  56         ret = privcmd_call(hypercall.op,
  57                            hypercall.arg[0], hypercall.arg[1],
  58                            hypercall.arg[2], hypercall.arg[3],
  59                            hypercall.arg[4]);
  60
  61         return ret;
  62 }
  63
  64 static void free_page_list(struct list_head *pages)
  65 {
  66         struct page *p, *n;
  67
  68         list_for_each_entry_safe(p, n, pages, lru)
  69                 __free_page(p);
  70
  71         INIT_LIST_HEAD(pages);
  72 }
  73
  74 /*
  75  * Given an array of items in userspace, return a list of pages
  76  * containing the data.  If copying fails, either because of memory
  77  * allocation failure or a problem reading user memory, return an
  78  * error code; its up to the caller to dispose of any partial list.
  79  */
  80 static int gather_array(struct list_head *pagelist,
  81                         unsigned nelem, size_t size,
  82                         const void __user *data)
  83 {
  84         unsigned pageidx;
  85         void *pagedata;
  86         int ret;
  87
  88         if (size > PAGE_SIZE)
  89                 return 0;
  90
  91         pageidx = PAGE_SIZE;
  92         pagedata = NULL;        /* quiet, gcc */
  93         while (nelem--) {
  94                 if (pageidx > PAGE_SIZE-size) {
  95                         struct page *page = alloc_page(GFP_KERNEL);
  96
  97                         ret = -ENOMEM;
  98                         if (page == NULL)
  99                                 goto fail;
 100
 101                         pagedata = page_address(page);
 102
 103                         list_add_tail(&page->lru, pagelist);
 104                         pageidx = 0;
 105                 }
 106
 107                 ret = -EFAULT;
 108                 if (copy_from_user(pagedata + pageidx, data, size))
 109                         goto fail;
 110
 111                 data += size;
 112                 pageidx += size;
 113         }
 114
 115         ret = 0;
 116
 117 fail:
 118         return ret;
 119 }
 120
 121 /*
 122  * Call function "fn" on each element of the array fragmented
 123  * over a list of pages.
 124  */
 125 static int traverse_pages(unsigned nelem, size_t size,
 126                           struct list_head *pos,
 127                           int (*fn)(void *data, void *state),
 128                           void *state)
 129 {
 130         void *pagedata;
 131         unsigned pageidx;
 132         int ret = 0;
 133
 134         BUG_ON(size > PAGE_SIZE);
 135
 136         pageidx = PAGE_SIZE;
 137         pagedata = NULL;        /* hush, gcc */
 138
 139         while (nelem--) {
 140                 if (pageidx > PAGE_SIZE-size) {
 141                         struct page *page;
 142                         pos = pos->next;
 143                         page = list_entry(pos, struct page, lru);
 144                         pagedata = page_address(page);
 145                         pageidx = 0;
 146                 }
 147
 148                 ret = (*fn)(pagedata + pageidx, state);
 149                 if (ret)
 150                         break;
 151                 pageidx += size;
 152         }
 153
 154         return ret;
 155 }
 156
 157 struct mmap_mfn_state {
 158         unsigned long va;
 159         struct vm_area_struct *vma;
 160         domid_t domain;
 161 };
 162
 163 static int mmap_mfn_range(void *data, void *state)
 164 {
 165         struct privcmd_mmap_entry *msg = data;
 166         struct mmap_mfn_state *st = state;
 167         struct vm_area_struct *vma = st->vma;
 168         int rc;
 169
 170         /* Do not allow range to wrap the address space. */
 171         if ((msg->npages > (LONG_MAX >> PAGE_SHIFT)) ||
 172             ((unsigned long)(msg->npages << PAGE_SHIFT) >= -st->va))
 173                 return -EINVAL;
 174
 175         /* Range chunks must be contiguous in va space. */
 176         if ((msg->va != st->va) ||
 177             ((msg->va+(msg->npages<<PAGE_SHIFT)) > vma->vm_end))
 178                 return -EINVAL;
 179
 180         rc = xen_remap_domain_mfn_range(vma,
 181                                         msg->va & PAGE_MASK,
 182                                         msg->mfn, msg->npages,
 183                                         vma->vm_page_prot,
 184                                         st->domain, NULL);
 185         if (rc < 0)
 186                 return rc;
 187
 188         st->va += msg->npages << PAGE_SHIFT;
 189
 190         return 0;
 191 }
 192
 193 static long privcmd_ioctl_mmap(void __user *udata)
 194 {
 195         struct privcmd_mmap mmapcmd;
 196         struct mm_struct *mm = current->mm;
 197         struct vm_area_struct *vma;
 198         int rc;
 199         LIST_HEAD(pagelist);
 200         struct mmap_mfn_state state;
 201
 202         /* We only support privcmd_ioctl_mmap_batch for auto translated. */
 203         if (xen_feature(XENFEAT_auto_translated_physmap))
 204                 return -ENOSYS;
 205
 206         if (copy_from_user(&mmapcmd, udata, sizeof(mmapcmd)))
 207                 return -EFAULT;
 208
 209         rc = gather_array(&pagelist,
 210                           mmapcmd.num, sizeof(struct privcmd_mmap_entry),
 211                           mmapcmd.entry);
 212
 213         if (rc || list_empty(&pagelist))
 214                 goto out;
 215
 216         down_write(&mm->mmap_sem);
 217
 218         {
 219                 struct page *page = list_first_entry(&pagelist,
 220                                                      struct page, lru);
 221                 struct privcmd_mmap_entry *msg = page_address(page);
 222
 223                 vma = find_vma(mm, msg->va);
 224                 rc = -EINVAL;
 225
 226                 if (!vma || (msg->va != vma->vm_start) ||
 227                     !privcmd_enforce_singleshot_mapping(vma))
 228                         goto out_up;
 229         }
 230
 231         state.va = vma->vm_start;
 232         state.vma = vma;
 233         state.domain = mmapcmd.dom;
 234
 235         rc = traverse_pages(mmapcmd.num, sizeof(struct privcmd_mmap_entry),
 236                             &pagelist,
 237                             mmap_mfn_range, &state);
 238
 239
 240 out_up:
 241         up_write(&mm->mmap_sem);
 242
 243 out:
 244         free_page_list(&pagelist);
 245
 246         return rc;
 247 }
 248
 249 struct mmap_batch_state {
 250         domid_t domain;
 251         unsigned long va;
 252         struct vm_area_struct *vma;
 253         int index;
 254         /* A tristate:
 255          *      0 for no errors
 256          *      1 if at least one error has happened (and no
 257          *          -ENOENT errors have happened)
 258          *      -ENOENT if at least 1 -ENOENT has happened.
 259          */
 260         int global_error;
 261         int version;
 262
 263         /* User-space mfn array to store errors in the second pass for V1. */
 264         xen_pfn_t __user *user_mfn;
 265         /* User-space int array to store errors in the second pass for V2. */
 266         int __user *user_err;
 267 };
 268
 269 /* auto translated dom0 note: if domU being created is PV, then mfn is
 270  * mfn(addr on bus). If it's auto xlated, then mfn is pfn (input to HAP).
 271  */
 272 static int mmap_batch_fn(void *data, void *state)
 273 {
 274         xen_pfn_t *mfnp = data;
 275         struct mmap_batch_state *st = state;
 276         struct vm_area_struct *vma = st->vma;
 277         struct page **pages = vma->vm_private_data;
 278         struct page *cur_page = NULL;
 279         int ret;
 280
 281         if (xen_feature(XENFEAT_auto_translated_physmap))
 282                 cur_page = pages[st->index++];
 283
 284         ret = xen_remap_domain_mfn_range(st->vma, st->va & PAGE_MASK, *mfnp, 1,
 285                                          st->vma->vm_page_prot, st->domain,
 286                                          &cur_page);
 287
 288         /* Store error code for second pass. */
 289         if (st->version == 1) {
 290                 if (ret < 0) {
 291                         /*
 292                          * V1 encodes the error codes in the 32bit top nibble of the
 293                          * mfn (with its known limitations vis-a-vis 64 bit callers).
 294                          */
 295                         *mfnp |= (ret == -ENOENT) ?
 296                                                 PRIVCMD_MMAPBATCH_PAGED_ERROR :
 297                                                 PRIVCMD_MMAPBATCH_MFN_ERROR;
 298                 }
 299         } else { /* st->version == 2 */
 300                 *((int *) mfnp) = ret;
 301         }
 302
 303         /* And see if it affects the global_error. */
 304         if (ret < 0) {
 305                 if (ret == -ENOENT)
 306                         st->global_error = -ENOENT;
 307                 else {
 308                         /* Record that at least one error has happened. */
 309                         if (st->global_error == 0)
 310                                 st->global_error = 1;
 311                 }
 312         }
 313         st->va += PAGE_SIZE;
 314
 315         return 0;
 316 }
 317
 318 static int mmap_return_errors(void *data, void *state)
 319 {
 320         struct mmap_batch_state *st = state;
 321
 322         if (st->version == 1) {
 323                 xen_pfn_t mfnp = *((xen_pfn_t *) data);
 324                 if (mfnp & PRIVCMD_MMAPBATCH_MFN_ERROR)
 325                         return __put_user(mfnp, st->user_mfn++);
 326                 else
 327                         st->user_mfn++;
 328         } else { /* st->version == 2 */
 329                 int err = *((int *) data);
 330                 if (err)
 331                         return __put_user(err, st->user_err++);
 332                 else
 333                         st->user_err++;
 334         }
 335
 336         return 0;
 337 }
 338
 339 /* Allocate pfns that are then mapped with gmfns from foreign domid. Update
 340  * the vma with the page info to use later.
 341  * Returns: 0 if success, otherwise -errno
 342  */
 343 static int alloc_empty_pages(struct vm_area_struct *vma, int numpgs)
 344 {
 345         int rc;
 346         struct page **pages;
 347
 348         pages = kcalloc(numpgs, sizeof(pages[0]), GFP_KERNEL);
 349         if (pages == NULL)
 350                 return -ENOMEM;
 351
 352         rc = alloc_xenballooned_pages(numpgs, pages, 0);
 353         if (rc != 0) {
 354                 pr_warn("%s Could not alloc %d pfns rc:%d\n", __func__,
 355                         numpgs, rc);
 356                 kfree(pages);
 357                 return -ENOMEM;
 358         }
 359         BUG_ON(vma->vm_private_data != PRIV_VMA_LOCKED);
 360         vma->vm_private_data = pages;
 361
 362         return 0;
 363 }
 364
 365 static struct vm_operations_struct privcmd_vm_ops;
 366
 367 static long privcmd_ioctl_mmap_batch(void __user *udata, int version)
 368 {
 369         int ret;
 370         struct privcmd_mmapbatch_v2 m;
 371         struct mm_struct *mm = current->mm;
 372         struct vm_area_struct *vma;
 373         unsigned long nr_pages;
 374         LIST_HEAD(pagelist);
 375         struct mmap_batch_state state;
 376
 377         switch (version) {
 378         case 1:
 379                 if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch)))
 380                         return -EFAULT;
 381                 /* Returns per-frame error in m.arr. */
 382                 m.err = NULL;
 383                 if (!access_ok(VERIFY_WRITE, m.arr, m.num * sizeof(*m.arr)))
 384                         return -EFAULT;
 385                 break;
 386         case 2:
 387                 if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch_v2)))
 388                         return -EFAULT;
 389                 /* Returns per-frame error code in m.err. */
 390                 if (!access_ok(VERIFY_WRITE, m.err, m.num * (sizeof(*m.err))))
 391                         return -EFAULT;
 392                 break;
 393         default:
 394                 return -EINVAL;
 395         }
 396
 397         nr_pages = m.num;
 398         if ((m.num <= 0) || (nr_pages > (LONG_MAX >> PAGE_SHIFT)))
 399                 return -EINVAL;
 400
 401         ret = gather_array(&pagelist, m.num, sizeof(xen_pfn_t), m.arr);
 402
 403         if (ret)
 404                 goto out;
 405         if (list_empty(&pagelist)) {
 406                 ret = -EINVAL;
 407                 goto out;
 408         }
 409
 410         if (version == 2) {
 411                 /* Zero error array now to only copy back actual errors. */
 412                 if (clear_user(m.err, sizeof(int) * m.num)) {
 413                         ret = -EFAULT;
 414                         goto out;
 415                 }
 416         }
 417
 418         down_write(&mm->mmap_sem);
 419
 420         vma = find_vma(mm, m.addr);
 421         if (!vma ||
 422             vma->vm_ops != &privcmd_vm_ops ||
 423             (m.addr != vma->vm_start) ||
 424             ((m.addr + (nr_pages << PAGE_SHIFT)) != vma->vm_end) ||
 425             !privcmd_enforce_singleshot_mapping(vma)) {
 426                 up_write(&mm->mmap_sem);
 427                 ret = -EINVAL;
 428                 goto out;
 429         }
 430         if (xen_feature(XENFEAT_auto_translated_physmap)) {
 431                 ret = alloc_empty_pages(vma, m.num);
 432                 if (ret < 0) {
 433                         up_write(&mm->mmap_sem);
 434                         goto out;
 435                 }
 436         }
 437
 438         state.domain        = m.dom;
 439         state.vma           = vma;
 440         state.va            = m.addr;
 441         state.index         = 0;
 442         state.global_error  = 0;
 443         state.version       = version;
 444
 445         /* mmap_batch_fn guarantees ret == 0 */
 446         BUG_ON(traverse_pages(m.num, sizeof(xen_pfn_t),
 447                              &pagelist, mmap_batch_fn, &state));
 448
 449         up_write(&mm->mmap_sem);
 450
 451         if (state.global_error) {
 452                 /* Write back errors in second pass. */
 453                 state.user_mfn = (xen_pfn_t *)m.arr;
 454                 state.user_err = m.err;
 455                 ret = traverse_pages(m.num, sizeof(xen_pfn_t),
 456                                                          &pagelist, mmap_return_errors, &state);
 457         } else
 458                 ret = 0;
 459
 460         /* If we have not had any EFAULT-like global errors then set the global
 461          * error to -ENOENT if necessary. */
 462         if ((ret == 0) && (state.global_error == -ENOENT))
 463                 ret = -ENOENT;
 464
 465 out:
 466         free_page_list(&pagelist);
 467
 468         return ret;
 469 }
 470
 471 static long privcmd_ioctl(struct file *file,
 472                           unsigned int cmd, unsigned long data)
 473 {
 474         int ret = -ENOSYS;
 475         void __user *udata = (void __user *) data;
 476
 477         switch (cmd) {
 478         case IOCTL_PRIVCMD_HYPERCALL:
 479                 ret = privcmd_ioctl_hypercall(udata);
 480                 break;
 481
 482         case IOCTL_PRIVCMD_MMAP:
 483                 ret = privcmd_ioctl_mmap(udata);
 484                 break;
 485
 486         case IOCTL_PRIVCMD_MMAPBATCH:
 487                 ret = privcmd_ioctl_mmap_batch(udata, 1);
 488                 break;
 489
 490         case IOCTL_PRIVCMD_MMAPBATCH_V2:
 491                 ret = privcmd_ioctl_mmap_batch(udata, 2);
 492                 break;
 493
 494         default:
 495                 ret = -EINVAL;
 496                 break;
 497         }
 498
 499         return ret;
 500 }
 501
 502 static void privcmd_close(struct vm_area_struct *vma)
 503 {
 504         struct page **pages = vma->vm_private_data;
 505         int numpgs = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
 506
 507         if (!xen_feature(XENFEAT_auto_translated_physmap) || !numpgs || !pages)
 508                 return;
 509
 510         xen_unmap_domain_mfn_range(vma, numpgs, pages);
 511         free_xenballooned_pages(numpgs, pages);
 512         kfree(pages);
 513 }
 514
 515 static int privcmd_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 516 {
 517         printk(KERN_DEBUG "privcmd_fault: vma=%p %lx-%lx, pgoff=%lx, uv=%p\n",
 518                vma, vma->vm_start, vma->vm_end,
 519                vmf->pgoff, vmf->virtual_address);
 520
 521         return VM_FAULT_SIGBUS;
 522 }
 523
 524 static struct vm_operations_struct privcmd_vm_ops = {
 525         .close = privcmd_close,
 526         .fault = privcmd_fault
 527 };
 528
 529 static int privcmd_mmap(struct file *file, struct vm_area_struct *vma)
 530 {
 531         /* DONTCOPY is essential for Xen because copy_page_range doesn't know
 532          * how to recreate these mappings */
 533         vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTCOPY |
 534                          VM_DONTEXPAND | VM_DONTDUMP;
 535         vma->vm_ops = &privcmd_vm_ops;
 536         vma->vm_private_data = NULL;
 537
 538         return 0;
 539 }
 540
 541 static int privcmd_enforce_singleshot_mapping(struct vm_area_struct *vma)
 542 {
 543         return !cmpxchg(&vma->vm_private_data, NULL, PRIV_VMA_LOCKED);
 544 }
 545
 546 const struct file_operations xen_privcmd_fops = {
 547         .owner = THIS_MODULE,
 548         .unlocked_ioctl = privcmd_ioctl,
 549         .mmap = privcmd_mmap,
 550 };
 551 EXPORT_SYMBOL_GPL(xen_privcmd_fops);
 552
 553 static struct miscdevice privcmd_dev = {
 554         .minor = MISC_DYNAMIC_MINOR,
 555         .name = "xen/privcmd",
 556         .fops = &xen_privcmd_fops,
 557 };
 558
 559 static int __init privcmd_init(void)
 560 {
 561         int err;
 562
 563         if (!xen_domain())
 564                 return -ENODEV;
 565
 566         err = misc_register(&privcmd_dev);
 567         if (err != 0) {
 568                 printk(KERN_ERR "Could not register Xen privcmd device\n");
 569                 return err;
 570         }
 571         return 0;
 572 }
 573
 574 static void __exit privcmd_exit(void)
 575 {
 576         misc_deregister(&privcmd_dev);
 577 }
 578
 579 module_init(privcmd_init);
 580 module_exit(privcmd_exit);