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