drivers/xen/xenfs/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/sched.h>
  11 #include <linux/slab.h>
  12 #include <linux/string.h>
  13 #include <linux/errno.h>
  14 #include <linux/mm.h>
  15 #include <linux/mman.h>
  16 #include <linux/uaccess.h>
  17 #include <linux/swap.h>
  18 #include <linux/highmem.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/seq_file.h>
  21
  22 #include <asm/pgalloc.h>
  23 #include <asm/pgtable.h>
  24 #include <asm/tlb.h>
  25 #include <asm/xen/hypervisor.h>
  26 #include <asm/xen/hypercall.h>
  27
  28 #include <xen/xen.h>
  29 #include <xen/privcmd.h>
  30 #include <xen/interface/xen.h>
  31 #include <xen/features.h>
  32 #include <xen/page.h>
  33 #include <xen/xen-ops.h>
  34
  35 #ifndef HAVE_ARCH_PRIVCMD_MMAP
  36 static int privcmd_enforce_singleshot_mapping(struct vm_area_struct *vma);
  37 #endif
  38
  39 static long privcmd_ioctl_hypercall(void __user *udata)
  40 {
  41         struct privcmd_hypercall hypercall;
  42         long ret;
  43
  44         if (copy_from_user(&hypercall, udata, sizeof(hypercall)))
  45                 return -EFAULT;
  46
  47         ret = privcmd_call(hypercall.op,
  48                            hypercall.arg[0], hypercall.arg[1],
  49                            hypercall.arg[2], hypercall.arg[3],
  50                            hypercall.arg[4]);
  51
  52         return ret;
  53 }
  54
  55 static void free_page_list(struct list_head *pages)
  56 {
  57         struct page *p, *n;
  58
  59         list_for_each_entry_safe(p, n, pages, lru)
  60                 __free_page(p);
  61
  62         INIT_LIST_HEAD(pages);
  63 }
  64
  65 /*
  66  * Given an array of items in userspace, return a list of pages
  67  * containing the data.  If copying fails, either because of memory
  68  * allocation failure or a problem reading user memory, return an
  69  * error code; its up to the caller to dispose of any partial list.
  70  */
  71 static int gather_array(struct list_head *pagelist,
  72                         unsigned nelem, size_t size,
  73                         void __user *data)
  74 {
  75         unsigned pageidx;
  76         void *pagedata;
  77         int ret;
  78
  79         if (size > PAGE_SIZE)
  80                 return 0;
  81
  82         pageidx = PAGE_SIZE;
  83         pagedata = NULL;        /* quiet, gcc */
  84         while (nelem--) {
  85                 if (pageidx > PAGE_SIZE-size) {
  86                         struct page *page = alloc_page(GFP_KERNEL);
  87
  88                         ret = -ENOMEM;
  89                         if (page == NULL)
  90                                 goto fail;
  91
  92                         pagedata = page_address(page);
  93
  94                         list_add_tail(&page->lru, pagelist);
  95                         pageidx = 0;
  96                 }
  97
  98                 ret = -EFAULT;
  99                 if (copy_from_user(pagedata + pageidx, data, size))
 100                         goto fail;
 101
 102                 data += size;
 103                 pageidx += size;
 104         }
 105
 106         ret = 0;
 107
 108 fail:
 109         return ret;
 110 }
 111
 112 /*
 113  * Call function "fn" on each element of the array fragmented
 114  * over a list of pages.
 115  */
 116 static int traverse_pages(unsigned nelem, size_t size,
 117                           struct list_head *pos,
 118                           int (*fn)(void *data, void *state),
 119                           void *state)
 120 {
 121         void *pagedata;
 122         unsigned pageidx;
 123         int ret = 0;
 124
 125         BUG_ON(size > PAGE_SIZE);
 126
 127         pageidx = PAGE_SIZE;
 128         pagedata = NULL;        /* hush, gcc */
 129
 130         while (nelem--) {
 131                 if (pageidx > PAGE_SIZE-size) {
 132                         struct page *page;
 133                         pos = pos->next;
 134                         page = list_entry(pos, struct page, lru);
 135                         pagedata = page_address(page);
 136                         pageidx = 0;
 137                 }
 138
 139                 ret = (*fn)(pagedata + pageidx, state);
 140                 if (ret)
 141                         break;
 142                 pageidx += size;
 143         }
 144
 145         return ret;
 146 }
 147
 148 struct mmap_mfn_state {
 149         unsigned long va;
 150         struct vm_area_struct *vma;
 151         domid_t domain;
 152 };
 153
 154 static int mmap_mfn_range(void *data, void *state)
 155 {
 156         struct privcmd_mmap_entry *msg = data;
 157         struct mmap_mfn_state *st = state;
 158         struct vm_area_struct *vma = st->vma;
 159         int rc;
 160
 161         /* Do not allow range to wrap the address space. */
 162         if ((msg->npages > (LONG_MAX >> PAGE_SHIFT)) ||
 163             ((unsigned long)(msg->npages << PAGE_SHIFT) >= -st->va))
 164                 return -EINVAL;
 165
 166         /* Range chunks must be contiguous in va space. */
 167         if ((msg->va != st->va) ||
 168             ((msg->va+(msg->npages<<PAGE_SHIFT)) > vma->vm_end))
 169                 return -EINVAL;
 170
 171         rc = xen_remap_domain_mfn_range(vma,
 172                                         msg->va & PAGE_MASK,
 173                                         msg->mfn, msg->npages,
 174                                         vma->vm_page_prot,
 175                                         st->domain);
 176         if (rc < 0)
 177                 return rc;
 178
 179         st->va += msg->npages << PAGE_SHIFT;
 180
 181         return 0;
 182 }
 183
 184 static long privcmd_ioctl_mmap(void __user *udata)
 185 {
 186         struct privcmd_mmap mmapcmd;
 187         struct mm_struct *mm = current->mm;
 188         struct vm_area_struct *vma;
 189         int rc;
 190         LIST_HEAD(pagelist);
 191         struct mmap_mfn_state state;
 192
 193         if (!xen_initial_domain())
 194                 return -EPERM;
 195
 196         if (copy_from_user(&mmapcmd, udata, sizeof(mmapcmd)))
 197                 return -EFAULT;
 198
 199         rc = gather_array(&pagelist,
 200                           mmapcmd.num, sizeof(struct privcmd_mmap_entry),
 201                           mmapcmd.entry);
 202
 203         if (rc || list_empty(&pagelist))
 204                 goto out;
 205
 206         down_write(&mm->mmap_sem);
 207
 208         {
 209                 struct page *page = list_first_entry(&pagelist,
 210                                                      struct page, lru);
 211                 struct privcmd_mmap_entry *msg = page_address(page);
 212
 213                 vma = find_vma(mm, msg->va);
 214                 rc = -EINVAL;
 215
 216                 if (!vma || (msg->va != vma->vm_start) ||
 217                     !privcmd_enforce_singleshot_mapping(vma))
 218                         goto out_up;
 219         }
 220
 221         state.va = vma->vm_start;
 222         state.vma = vma;
 223         state.domain = mmapcmd.dom;
 224
 225         rc = traverse_pages(mmapcmd.num, sizeof(struct privcmd_mmap_entry),
 226                             &pagelist,
 227                             mmap_mfn_range, &state);
 228
 229
 230 out_up:
 231         up_write(&mm->mmap_sem);
 232
 233 out:
 234         free_page_list(&pagelist);
 235
 236         return rc;
 237 }
 238
 239 struct mmap_batch_state {
 240         domid_t domain;
 241         unsigned long va;
 242         struct vm_area_struct *vma;
 243         int err;
 244
 245         xen_pfn_t __user *user;
 246 };
 247
 248 static int mmap_batch_fn(void *data, void *state)
 249 {
 250         xen_pfn_t *mfnp = data;
 251         struct mmap_batch_state *st = state;
 252
 253         if (xen_remap_domain_mfn_range(st->vma, st->va & PAGE_MASK, *mfnp, 1,
 254                                        st->vma->vm_page_prot, st->domain) < 0) {
 255                 *mfnp |= 0xf0000000U;
 256                 st->err++;
 257         }
 258         st->va += PAGE_SIZE;
 259
 260         return 0;
 261 }
 262
 263 static int mmap_return_errors(void *data, void *state)
 264 {
 265         xen_pfn_t *mfnp = data;
 266         struct mmap_batch_state *st = state;
 267
 268         put_user(*mfnp, st->user++);
 269
 270         return 0;
 271 }
 272
 273 static struct vm_operations_struct privcmd_vm_ops;
 274
 275 static long privcmd_ioctl_mmap_batch(void __user *udata)
 276 {
 277         int ret;
 278         struct privcmd_mmapbatch m;
 279         struct mm_struct *mm = current->mm;
 280         struct vm_area_struct *vma;
 281         unsigned long nr_pages;
 282         LIST_HEAD(pagelist);
 283         struct mmap_batch_state state;
 284
 285         if (!xen_initial_domain())
 286                 return -EPERM;
 287
 288         if (copy_from_user(&m, udata, sizeof(m)))
 289                 return -EFAULT;
 290
 291         nr_pages = m.num;
 292         if ((m.num <= 0) || (nr_pages > (LONG_MAX >> PAGE_SHIFT)))
 293                 return -EINVAL;
 294
 295         ret = gather_array(&pagelist, m.num, sizeof(xen_pfn_t),
 296                            m.arr);
 297
 298         if (ret || list_empty(&pagelist))
 299                 goto out;
 300
 301         down_write(&mm->mmap_sem);
 302
 303         vma = find_vma(mm, m.addr);
 304         ret = -EINVAL;
 305         if (!vma ||
 306             vma->vm_ops != &privcmd_vm_ops ||
 307             (m.addr != vma->vm_start) ||
 308             ((m.addr + (nr_pages << PAGE_SHIFT)) != vma->vm_end) ||
 309             !privcmd_enforce_singleshot_mapping(vma)) {
 310                 up_write(&mm->mmap_sem);
 311                 goto out;
 312         }
 313
 314         state.domain = m.dom;
 315         state.vma = vma;
 316         state.va = m.addr;
 317         state.err = 0;
 318
 319         ret = traverse_pages(m.num, sizeof(xen_pfn_t),
 320                              &pagelist, mmap_batch_fn, &state);
 321
 322         up_write(&mm->mmap_sem);
 323
 324         if (state.err > 0) {
 325                 ret = 0;
 326
 327                 state.user = m.arr;
 328                 traverse_pages(m.num, sizeof(xen_pfn_t),
 329                                &pagelist,
 330                                mmap_return_errors, &state);
 331         }
 332
 333 out:
 334         free_page_list(&pagelist);
 335
 336         return ret;
 337 }
 338
 339 static long privcmd_ioctl(struct file *file,
 340                           unsigned int cmd, unsigned long data)
 341 {
 342         int ret = -ENOSYS;
 343         void __user *udata = (void __user *) data;
 344
 345         switch (cmd) {
 346         case IOCTL_PRIVCMD_HYPERCALL:
 347                 ret = privcmd_ioctl_hypercall(udata);
 348                 break;
 349
 350         case IOCTL_PRIVCMD_MMAP:
 351                 ret = privcmd_ioctl_mmap(udata);
 352                 break;
 353
 354         case IOCTL_PRIVCMD_MMAPBATCH:
 355                 ret = privcmd_ioctl_mmap_batch(udata);
 356                 break;
 357
 358         default:
 359                 ret = -EINVAL;
 360                 break;
 361         }
 362
 363         return ret;
 364 }
 365
 366 #ifndef HAVE_ARCH_PRIVCMD_MMAP
 367 static int privcmd_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 368 {
 369         printk(KERN_DEBUG "privcmd_fault: vma=%p %lx-%lx, pgoff=%lx, uv=%p\n",
 370                vma, vma->vm_start, vma->vm_end,
 371                vmf->pgoff, vmf->virtual_address);
 372
 373         return VM_FAULT_SIGBUS;
 374 }
 375
 376 static struct vm_operations_struct privcmd_vm_ops = {
 377         .fault = privcmd_fault
 378 };
 379
 380 static int privcmd_mmap(struct file *file, struct vm_area_struct *vma)
 381 {
 382         /* Unsupported for auto-translate guests. */
 383         if (xen_feature(XENFEAT_auto_translated_physmap))
 384                 return -ENOSYS;
 385
 386         /* DONTCOPY is essential for Xen as copy_page_range is broken. */
 387         vma->vm_flags |= VM_RESERVED | VM_IO | VM_DONTCOPY;
 388         vma->vm_ops = &privcmd_vm_ops;
 389         vma->vm_private_data = NULL;
 390
 391         return 0;
 392 }
 393
 394 static int privcmd_enforce_singleshot_mapping(struct vm_area_struct *vma)
 395 {
 396         return (xchg(&vma->vm_private_data, (void *)1) == NULL);
 397 }
 398 #endif
 399
 400 const struct file_operations privcmd_file_ops = {
 401         .unlocked_ioctl = privcmd_ioctl,
 402         .mmap = privcmd_mmap,
 403 };