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         if (!xen_initial_domain())
 203                 return -EPERM;
 204
 205         /* We only support privcmd_ioctl_mmap_batch for auto translated. */
 206         if (xen_feature(XENFEAT_auto_translated_physmap))
 207                 return -ENOSYS;
 208
 209         if (copy_from_user(&mmapcmd, udata, sizeof(mmapcmd)))
 210                 return -EFAULT;
 211
 212         rc = gather_array(&pagelist,
 213                           mmapcmd.num, sizeof(struct privcmd_mmap_entry),
 214                           mmapcmd.entry);
 215
 216         if (rc || list_empty(&pagelist))
 217                 goto out;
 218
 219         down_write(&mm->mmap_sem);
 220
 221         {
 222                 struct page *page = list_first_entry(&pagelist,
 223                                                      struct page, lru);
 224                 struct privcmd_mmap_entry *msg = page_address(page);
 225
 226                 vma = find_vma(mm, msg->va);
 227                 rc = -EINVAL;
 228
 229                 if (!vma || (msg->va != vma->vm_start) ||
 230                     !privcmd_enforce_singleshot_mapping(vma))
 231                         goto out_up;
 232         }
 233
 234         state.va = vma->vm_start;
 235         state.vma = vma;
 236         state.domain = mmapcmd.dom;
 237
 238         rc = traverse_pages(mmapcmd.num, sizeof(struct privcmd_mmap_entry),
 239                             &pagelist,
 240                             mmap_mfn_range, &state);
 241
 242
 243 out_up:
 244         up_write(&mm->mmap_sem);
 245
 246 out:
 247         free_page_list(&pagelist);
 248
 249         return rc;
 250 }
 251
 252 struct mmap_batch_state {
 253         domid_t domain;
 254         unsigned long va;
 255         struct vm_area_struct *vma;
 256         int index;
 257         /* A tristate:
 258          *      0 for no errors
 259          *      1 if at least one error has happened (and no
 260          *          -ENOENT errors have happened)
 261          *      -ENOENT if at least 1 -ENOENT has happened.
 262          */
 263         int global_error;
 264         /* An array for individual errors */
 265         int *err;
 266
 267         /* User-space mfn array to store errors in the second pass for V1. */
 268         xen_pfn_t __user *user_mfn;
 269 };
 270
 271 /* auto translated dom0 note: if domU being created is PV, then mfn is
 272  * mfn(addr on bus). If it's auto xlated, then mfn is pfn (input to HAP).
 273  */
 274 static int mmap_batch_fn(void *data, void *state)
 275 {
 276         xen_pfn_t *mfnp = data;
 277         struct mmap_batch_state *st = state;
 278         struct vm_area_struct *vma = st->vma;
 279         struct page **pages = vma->vm_private_data;
 280         struct page *cur_page = NULL;
 281         int ret;
 282
 283         if (xen_feature(XENFEAT_auto_translated_physmap))
 284                 cur_page = pages[st->index++];
 285
 286         ret = xen_remap_domain_mfn_range(st->vma, st->va & PAGE_MASK, *mfnp, 1,
 287                                          st->vma->vm_page_prot, st->domain,
 288                                          &cur_page);
 289
 290         /* Store error code for second pass. */
 291         *(st->err++) = ret;
 292
 293         /* And see if it affects the global_error. */
 294         if (ret < 0) {
 295                 if (ret == -ENOENT)
 296                         st->global_error = -ENOENT;
 297                 else {
 298                         /* Record that at least one error has happened. */
 299                         if (st->global_error == 0)
 300                                 st->global_error = 1;
 301                 }
 302         }
 303         st->va += PAGE_SIZE;
 304
 305         return 0;
 306 }
 307
 308 static int mmap_return_errors_v1(void *data, void *state)
 309 {
 310         xen_pfn_t *mfnp = data;
 311         struct mmap_batch_state *st = state;
 312         int err = *(st->err++);
 313
 314         /*
 315          * V1 encodes the error codes in the 32bit top nibble of the
 316          * mfn (with its known limitations vis-a-vis 64 bit callers).
 317          */
 318         *mfnp |= (err == -ENOENT) ?
 319                                 PRIVCMD_MMAPBATCH_PAGED_ERROR :
 320                                 PRIVCMD_MMAPBATCH_MFN_ERROR;
 321         return __put_user(*mfnp, st->user_mfn++);
 322 }
 323
 324 /* Allocate pfns that are then mapped with gmfns from foreign domid. Update
 325  * the vma with the page info to use later.
 326  * Returns: 0 if success, otherwise -errno
 327  */
 328 static int alloc_empty_pages(struct vm_area_struct *vma, int numpgs)
 329 {
 330         int rc;
 331         struct page **pages;
 332
 333         pages = kcalloc(numpgs, sizeof(pages[0]), GFP_KERNEL);
 334         if (pages == NULL)
 335                 return -ENOMEM;
 336
 337         rc = alloc_xenballooned_pages(numpgs, pages, 0);
 338         if (rc != 0) {
 339                 pr_warn("%s Could not alloc %d pfns rc:%d\n", __func__,
 340                         numpgs, rc);
 341                 kfree(pages);
 342                 return -ENOMEM;
 343         }
 344         BUG_ON(vma->vm_private_data != PRIV_VMA_LOCKED);
 345         vma->vm_private_data = pages;
 346
 347         return 0;
 348 }
 349
 350 static struct vm_operations_struct privcmd_vm_ops;
 351
 352 static long privcmd_ioctl_mmap_batch(void __user *udata, int version)
 353 {
 354         int ret;
 355         struct privcmd_mmapbatch_v2 m;
 356         struct mm_struct *mm = current->mm;
 357         struct vm_area_struct *vma;
 358         unsigned long nr_pages;
 359         LIST_HEAD(pagelist);
 360         int *err_array = NULL;
 361         struct mmap_batch_state state;
 362
 363         if (!xen_initial_domain())
 364                 return -EPERM;
 365
 366         switch (version) {
 367         case 1:
 368                 if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch)))
 369                         return -EFAULT;
 370                 /* Returns per-frame error in m.arr. */
 371                 m.err = NULL;
 372                 if (!access_ok(VERIFY_WRITE, m.arr, m.num * sizeof(*m.arr)))
 373                         return -EFAULT;
 374                 break;
 375         case 2:
 376                 if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch_v2)))
 377                         return -EFAULT;
 378                 /* Returns per-frame error code in m.err. */
 379                 if (!access_ok(VERIFY_WRITE, m.err, m.num * (sizeof(*m.err))))
 380                         return -EFAULT;
 381                 break;
 382         default:
 383                 return -EINVAL;
 384         }
 385
 386         nr_pages = m.num;
 387         if ((m.num <= 0) || (nr_pages > (LONG_MAX >> PAGE_SHIFT)))
 388                 return -EINVAL;
 389
 390         ret = gather_array(&pagelist, m.num, sizeof(xen_pfn_t), m.arr);
 391
 392         if (ret)
 393                 goto out;
 394         if (list_empty(&pagelist)) {
 395                 ret = -EINVAL;
 396                 goto out;
 397         }
 398
 399         err_array = kcalloc(m.num, sizeof(int), GFP_KERNEL);
 400         if (err_array == NULL) {
 401                 ret = -ENOMEM;
 402                 goto out;
 403         }
 404
 405         down_write(&mm->mmap_sem);
 406
 407         vma = find_vma(mm, m.addr);
 408         if (!vma ||
 409             vma->vm_ops != &privcmd_vm_ops ||
 410             (m.addr != vma->vm_start) ||
 411             ((m.addr + (nr_pages << PAGE_SHIFT)) != vma->vm_end) ||
 412             !privcmd_enforce_singleshot_mapping(vma)) {
 413                 up_write(&mm->mmap_sem);
 414                 ret = -EINVAL;
 415                 goto out;
 416         }
 417         if (xen_feature(XENFEAT_auto_translated_physmap)) {
 418                 ret = alloc_empty_pages(vma, m.num);
 419                 if (ret < 0) {
 420                         up_write(&mm->mmap_sem);
 421                         goto out;
 422                 }
 423         }
 424
 425         state.domain        = m.dom;
 426         state.vma           = vma;
 427         state.va            = m.addr;
 428         state.index         = 0;
 429         state.global_error  = 0;
 430         state.err           = err_array;
 431
 432         /* mmap_batch_fn guarantees ret == 0 */
 433         BUG_ON(traverse_pages(m.num, sizeof(xen_pfn_t),
 434                              &pagelist, mmap_batch_fn, &state));
 435
 436         up_write(&mm->mmap_sem);
 437
 438         if (version == 1) {
 439                 if (state.global_error) {
 440                         /* Write back errors in second pass. */
 441                         state.user_mfn = (xen_pfn_t *)m.arr;
 442                         state.err      = err_array;
 443                         ret = traverse_pages(m.num, sizeof(xen_pfn_t),
 444                                              &pagelist, mmap_return_errors_v1, &state);
 445                 } else
 446                         ret = 0;
 447
 448         } else if (version == 2) {
 449                 ret = __copy_to_user(m.err, err_array, m.num * sizeof(int));
 450                 if (ret)
 451                         ret = -EFAULT;
 452         }
 453
 454         /* If we have not had any EFAULT-like global errors then set the global
 455          * error to -ENOENT if necessary. */
 456         if ((ret == 0) && (state.global_error == -ENOENT))
 457                 ret = -ENOENT;
 458
 459 out:
 460         kfree(err_array);
 461         free_page_list(&pagelist);
 462
 463         return ret;
 464 }
 465
 466 static long privcmd_ioctl(struct file *file,
 467                           unsigned int cmd, unsigned long data)
 468 {
 469         int ret = -ENOSYS;
 470         void __user *udata = (void __user *) data;
 471
 472         switch (cmd) {
 473         case IOCTL_PRIVCMD_HYPERCALL:
 474                 ret = privcmd_ioctl_hypercall(udata);
 475                 break;
 476
 477         case IOCTL_PRIVCMD_MMAP:
 478                 ret = privcmd_ioctl_mmap(udata);
 479                 break;
 480
 481         case IOCTL_PRIVCMD_MMAPBATCH:
 482                 ret = privcmd_ioctl_mmap_batch(udata, 1);
 483                 break;
 484
 485         case IOCTL_PRIVCMD_MMAPBATCH_V2:
 486                 ret = privcmd_ioctl_mmap_batch(udata, 2);
 487                 break;
 488
 489         default:
 490                 ret = -EINVAL;
 491                 break;
 492         }
 493
 494         return ret;
 495 }
 496
 497 static void privcmd_close(struct vm_area_struct *vma)
 498 {
 499         struct page **pages = vma->vm_private_data;
 500         int numpgs = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
 501
 502         if (!xen_feature(XENFEAT_auto_translated_physmap || !numpgs || !pages))
 503                 return;
 504
 505         xen_unmap_domain_mfn_range(vma, numpgs, pages);
 506         free_xenballooned_pages(numpgs, pages);
 507         kfree(pages);
 508 }
 509
 510 static int privcmd_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 511 {
 512         printk(KERN_DEBUG "privcmd_fault: vma=%p %lx-%lx, pgoff=%lx, uv=%p\n",
 513                vma, vma->vm_start, vma->vm_end,
 514                vmf->pgoff, vmf->virtual_address);
 515
 516         return VM_FAULT_SIGBUS;
 517 }
 518
 519 static struct vm_operations_struct privcmd_vm_ops = {
 520         .close = privcmd_close,
 521         .fault = privcmd_fault
 522 };
 523
 524 static int privcmd_mmap(struct file *file, struct vm_area_struct *vma)
 525 {
 526         /* DONTCOPY is essential for Xen because copy_page_range doesn't know
 527          * how to recreate these mappings */
 528         vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTCOPY |
 529                          VM_DONTEXPAND | VM_DONTDUMP;
 530         vma->vm_ops = &privcmd_vm_ops;
 531         vma->vm_private_data = NULL;
 532
 533         return 0;
 534 }
 535
 536 static int privcmd_enforce_singleshot_mapping(struct vm_area_struct *vma)
 537 {
 538         return !cmpxchg(&vma->vm_private_data, NULL, PRIV_VMA_LOCKED);
 539 }
 540
 541 const struct file_operations xen_privcmd_fops = {
 542         .owner = THIS_MODULE,
 543         .unlocked_ioctl = privcmd_ioctl,
 544         .mmap = privcmd_mmap,
 545 };
 546 EXPORT_SYMBOL_GPL(xen_privcmd_fops);
 547
 548 static struct miscdevice privcmd_dev = {
 549         .minor = MISC_DYNAMIC_MINOR,
 550         .name = "xen/privcmd",
 551         .fops = &xen_privcmd_fops,
 552 };
 553
 554 static int __init privcmd_init(void)
 555 {
 556         int err;
 557
 558         if (!xen_domain())
 559                 return -ENODEV;
 560
 561         err = misc_register(&privcmd_dev);
 562         if (err != 0) {
 563                 printk(KERN_ERR "Could not register Xen privcmd device\n");
 564                 return err;
 565         }
 566         return 0;
 567 }
 568
 569 static void __exit privcmd_exit(void)
 570 {
 571         misc_deregister(&privcmd_dev);
 572 }
 573
 574 module_init(privcmd_init);
 575 module_exit(privcmd_exit);