err.no Git - linux-2.6/blob - arch/s390/lib/uaccess_pt.c

   1 /*
   2  *  arch/s390/lib/uaccess_pt.c
   3  *
   4  *  User access functions based on page table walks for enhanced
   5  *  system layout without hardware support.
   6  *
   7  *    Copyright IBM Corp. 2006
   8  *    Author(s): Gerald Schaefer (gerald.schaefer@de.ibm.com)
   9  */
  10
  11 #include <linux/errno.h>
  12 #include <linux/hardirq.h>
  13 #include <linux/mm.h>
  14 #include <asm/uaccess.h>
  15 #include <asm/futex.h>
  16 #include "uaccess.h"
  17
  18 static inline pte_t *follow_table(struct mm_struct *mm, unsigned long addr)
  19 {
  20         pgd_t *pgd;
  21         pmd_t *pmd;
  22
  23         pgd = pgd_offset(mm, addr);
  24         if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
  25                 return NULL;
  26
  27         pmd = pmd_offset(pgd, addr);
  28         if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
  29                 return NULL;
  30
  31         return pte_offset_map(pmd, addr);
  32 }
  33
  34 static int __handle_fault(struct mm_struct *mm, unsigned long address,
  35                           int write_access)
  36 {
  37         struct vm_area_struct *vma;
  38         int ret = -EFAULT;
  39         int fault;
  40
  41         if (in_atomic())
  42                 return ret;
  43         down_read(&mm->mmap_sem);
  44         vma = find_vma(mm, address);
  45         if (unlikely(!vma))
  46                 goto out;
  47         if (unlikely(vma->vm_start > address)) {
  48                 if (!(vma->vm_flags & VM_GROWSDOWN))
  49                         goto out;
  50                 if (expand_stack(vma, address))
  51                         goto out;
  52         }
  53
  54         if (!write_access) {
  55                 /* page not present, check vm flags */
  56                 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
  57                         goto out;
  58         } else {
  59                 if (!(vma->vm_flags & VM_WRITE))
  60                         goto out;
  61         }
  62
  63 survive:
  64         fault = handle_mm_fault(mm, vma, address, write_access);
  65         if (unlikely(fault & VM_FAULT_ERROR)) {
  66                 if (fault & VM_FAULT_OOM)
  67                         goto out_of_memory;
  68                 else if (fault & VM_FAULT_SIGBUS)
  69                         goto out_sigbus;
  70                 BUG();
  71         }
  72         if (fault & VM_FAULT_MAJOR)
  73                 current->maj_flt++;
  74         else
  75                 current->min_flt++;
  76         ret = 0;
  77 out:
  78         up_read(&mm->mmap_sem);
  79         return ret;
  80
  81 out_of_memory:
  82         up_read(&mm->mmap_sem);
  83         if (is_global_init(current)) {
  84                 yield();
  85                 down_read(&mm->mmap_sem);
  86                 goto survive;
  87         }
  88         printk("VM: killing process %s\n", current->comm);
  89         return ret;
  90
  91 out_sigbus:
  92         up_read(&mm->mmap_sem);
  93         current->thread.prot_addr = address;
  94         current->thread.trap_no = 0x11;
  95         force_sig(SIGBUS, current);
  96         return ret;
  97 }
  98
  99 static size_t __user_copy_pt(unsigned long uaddr, void *kptr,
 100                              size_t n, int write_user)
 101 {
 102         struct mm_struct *mm = current->mm;
 103         unsigned long offset, pfn, done, size;
 104         pte_t *pte;
 105         void *from, *to;
 106
 107         done = 0;
 108 retry:
 109         spin_lock(&mm->page_table_lock);
 110         do {
 111                 pte = follow_table(mm, uaddr);
 112                 if (!pte || !pte_present(*pte) ||
 113                     (write_user && !pte_write(*pte)))
 114                         goto fault;
 115
 116                 pfn = pte_pfn(*pte);
 117                 if (!pfn_valid(pfn))
 118                         goto out;
 119
 120                 offset = uaddr & (PAGE_SIZE - 1);
 121                 size = min(n - done, PAGE_SIZE - offset);
 122                 if (write_user) {
 123                         to = (void *)((pfn << PAGE_SHIFT) + offset);
 124                         from = kptr + done;
 125                 } else {
 126                         from = (void *)((pfn << PAGE_SHIFT) + offset);
 127                         to = kptr + done;
 128                 }
 129                 memcpy(to, from, size);
 130                 done += size;
 131                 uaddr += size;
 132         } while (done < n);
 133 out:
 134         spin_unlock(&mm->page_table_lock);
 135         return n - done;
 136 fault:
 137         spin_unlock(&mm->page_table_lock);
 138         if (__handle_fault(mm, uaddr, write_user))
 139                 return n - done;
 140         goto retry;
 141 }
 142
 143 /*
 144  * Do DAT for user address by page table walk, return kernel address.
 145  * This function needs to be called with current->mm->page_table_lock held.
 146  */
 147 static unsigned long __dat_user_addr(unsigned long uaddr)
 148 {
 149         struct mm_struct *mm = current->mm;
 150         unsigned long pfn, ret;
 151         pte_t *pte;
 152         int rc;
 153
 154         ret = 0;
 155 retry:
 156         pte = follow_table(mm, uaddr);
 157         if (!pte || !pte_present(*pte))
 158                 goto fault;
 159
 160         pfn = pte_pfn(*pte);
 161         if (!pfn_valid(pfn))
 162                 goto out;
 163
 164         ret = (pfn << PAGE_SHIFT) + (uaddr & (PAGE_SIZE - 1));
 165 out:
 166         return ret;
 167 fault:
 168         spin_unlock(&mm->page_table_lock);
 169         rc = __handle_fault(mm, uaddr, 0);
 170         spin_lock(&mm->page_table_lock);
 171         if (rc)
 172                 goto out;
 173         goto retry;
 174 }
 175
 176 size_t copy_from_user_pt(size_t n, const void __user *from, void *to)
 177 {
 178         size_t rc;
 179
 180         if (segment_eq(get_fs(), KERNEL_DS)) {
 181                 memcpy(to, (void __kernel __force *) from, n);
 182                 return 0;
 183         }
 184         rc = __user_copy_pt((unsigned long) from, to, n, 0);
 185         if (unlikely(rc))
 186                 memset(to + n - rc, 0, rc);
 187         return rc;
 188 }
 189
 190 size_t copy_to_user_pt(size_t n, void __user *to, const void *from)
 191 {
 192         if (segment_eq(get_fs(), KERNEL_DS)) {
 193                 memcpy((void __kernel __force *) to, from, n);
 194                 return 0;
 195         }
 196         return __user_copy_pt((unsigned long) to, (void *) from, n, 1);
 197 }
 198
 199 static size_t clear_user_pt(size_t n, void __user *to)
 200 {
 201         long done, size, ret;
 202
 203         if (segment_eq(get_fs(), KERNEL_DS)) {
 204                 memset((void __kernel __force *) to, 0, n);
 205                 return 0;
 206         }
 207         done = 0;
 208         do {
 209                 if (n - done > PAGE_SIZE)
 210                         size = PAGE_SIZE;
 211                 else
 212                         size = n - done;
 213                 ret = __user_copy_pt((unsigned long) to + done,
 214                                       &empty_zero_page, size, 1);
 215                 done += size;
 216                 if (ret)
 217                         return ret + n - done;
 218         } while (done < n);
 219         return 0;
 220 }
 221
 222 static size_t strnlen_user_pt(size_t count, const char __user *src)
 223 {
 224         char *addr;
 225         unsigned long uaddr = (unsigned long) src;
 226         struct mm_struct *mm = current->mm;
 227         unsigned long offset, pfn, done, len;
 228         pte_t *pte;
 229         size_t len_str;
 230
 231         if (segment_eq(get_fs(), KERNEL_DS))
 232                 return strnlen((const char __kernel __force *) src, count) + 1;
 233         done = 0;
 234 retry:
 235         spin_lock(&mm->page_table_lock);
 236         do {
 237                 pte = follow_table(mm, uaddr);
 238                 if (!pte || !pte_present(*pte))
 239                         goto fault;
 240
 241                 pfn = pte_pfn(*pte);
 242                 if (!pfn_valid(pfn)) {
 243                         done = -1;
 244                         goto out;
 245                 }
 246
 247                 offset = uaddr & (PAGE_SIZE-1);
 248                 addr = (char *)(pfn << PAGE_SHIFT) + offset;
 249                 len = min(count - done, PAGE_SIZE - offset);
 250                 len_str = strnlen(addr, len);
 251                 done += len_str;
 252                 uaddr += len_str;
 253         } while ((len_str == len) && (done < count));
 254 out:
 255         spin_unlock(&mm->page_table_lock);
 256         return done + 1;
 257 fault:
 258         spin_unlock(&mm->page_table_lock);
 259         if (__handle_fault(mm, uaddr, 0)) {
 260                 return 0;
 261         }
 262         goto retry;
 263 }
 264
 265 static size_t strncpy_from_user_pt(size_t count, const char __user *src,
 266                                    char *dst)
 267 {
 268         size_t n = strnlen_user_pt(count, src);
 269
 270         if (!n)
 271                 return -EFAULT;
 272         if (n > count)
 273                 n = count;
 274         if (segment_eq(get_fs(), KERNEL_DS)) {
 275                 memcpy(dst, (const char __kernel __force *) src, n);
 276                 if (dst[n-1] == '\0')
 277                         return n-1;
 278                 else
 279                         return n;
 280         }
 281         if (__user_copy_pt((unsigned long) src, dst, n, 0))
 282                 return -EFAULT;
 283         if (dst[n-1] == '\0')
 284                 return n-1;
 285         else
 286                 return n;
 287 }
 288
 289 static size_t copy_in_user_pt(size_t n, void __user *to,
 290                               const void __user *from)
 291 {
 292         struct mm_struct *mm = current->mm;
 293         unsigned long offset_from, offset_to, offset_max, pfn_from, pfn_to,
 294                       uaddr, done, size;
 295         unsigned long uaddr_from = (unsigned long) from;
 296         unsigned long uaddr_to = (unsigned long) to;
 297         pte_t *pte_from, *pte_to;
 298         int write_user;
 299
 300         done = 0;
 301 retry:
 302         spin_lock(&mm->page_table_lock);
 303         do {
 304                 pte_from = follow_table(mm, uaddr_from);
 305                 if (!pte_from || !pte_present(*pte_from)) {
 306                         uaddr = uaddr_from;
 307                         write_user = 0;
 308                         goto fault;
 309                 }
 310
 311                 pte_to = follow_table(mm, uaddr_to);
 312                 if (!pte_to || !pte_present(*pte_to) || !pte_write(*pte_to)) {
 313                         uaddr = uaddr_to;
 314                         write_user = 1;
 315                         goto fault;
 316                 }
 317
 318                 pfn_from = pte_pfn(*pte_from);
 319                 if (!pfn_valid(pfn_from))
 320                         goto out;
 321                 pfn_to = pte_pfn(*pte_to);
 322                 if (!pfn_valid(pfn_to))
 323                         goto out;
 324
 325                 offset_from = uaddr_from & (PAGE_SIZE-1);
 326                 offset_to = uaddr_from & (PAGE_SIZE-1);
 327                 offset_max = max(offset_from, offset_to);
 328                 size = min(n - done, PAGE_SIZE - offset_max);
 329
 330                 memcpy((void *)(pfn_to << PAGE_SHIFT) + offset_to,
 331                        (void *)(pfn_from << PAGE_SHIFT) + offset_from, size);
 332                 done += size;
 333                 uaddr_from += size;
 334                 uaddr_to += size;
 335         } while (done < n);
 336 out:
 337         spin_unlock(&mm->page_table_lock);
 338         return n - done;
 339 fault:
 340         spin_unlock(&mm->page_table_lock);
 341         if (__handle_fault(mm, uaddr, write_user))
 342                 return n - done;
 343         goto retry;
 344 }
 345
 346 #define __futex_atomic_op(insn, ret, oldval, newval, uaddr, oparg)      \
 347         asm volatile("0: l   %1,0(%6)\n"                                \
 348                      "1: " insn                                         \
 349                      "2: cs  %1,%2,0(%6)\n"                             \
 350                      "3: jl  1b\n"                                      \
 351                      "   lhi %0,0\n"                                    \
 352                      "4:\n"                                             \
 353                      EX_TABLE(0b,4b) EX_TABLE(2b,4b) EX_TABLE(3b,4b)    \
 354                      : "=d" (ret), "=&d" (oldval), "=&d" (newval),      \
 355                        "=m" (*uaddr)                                    \
 356                      : "0" (-EFAULT), "d" (oparg), "a" (uaddr),         \
 357                        "m" (*uaddr) : "cc" );
 358
 359 int futex_atomic_op_pt(int op, int __user *uaddr, int oparg, int *old)
 360 {
 361         int oldval = 0, newval, ret;
 362
 363         spin_lock(&current->mm->page_table_lock);
 364         uaddr = (int __user *) __dat_user_addr((unsigned long) uaddr);
 365         if (!uaddr) {
 366                 spin_unlock(&current->mm->page_table_lock);
 367                 return -EFAULT;
 368         }
 369         get_page(virt_to_page(uaddr));
 370         spin_unlock(&current->mm->page_table_lock);
 371         switch (op) {
 372         case FUTEX_OP_SET:
 373                 __futex_atomic_op("lr %2,%5\n",
 374                                   ret, oldval, newval, uaddr, oparg);
 375                 break;
 376         case FUTEX_OP_ADD:
 377                 __futex_atomic_op("lr %2,%1\nar %2,%5\n",
 378                                   ret, oldval, newval, uaddr, oparg);
 379                 break;
 380         case FUTEX_OP_OR:
 381                 __futex_atomic_op("lr %2,%1\nor %2,%5\n",
 382                                   ret, oldval, newval, uaddr, oparg);
 383                 break;
 384         case FUTEX_OP_ANDN:
 385                 __futex_atomic_op("lr %2,%1\nnr %2,%5\n",
 386                                   ret, oldval, newval, uaddr, oparg);
 387                 break;
 388         case FUTEX_OP_XOR:
 389                 __futex_atomic_op("lr %2,%1\nxr %2,%5\n",
 390                                   ret, oldval, newval, uaddr, oparg);
 391                 break;
 392         default:
 393                 ret = -ENOSYS;
 394         }
 395         put_page(virt_to_page(uaddr));
 396         *old = oldval;
 397         return ret;
 398 }
 399
 400 int futex_atomic_cmpxchg_pt(int __user *uaddr, int oldval, int newval)
 401 {
 402         int ret;
 403
 404         spin_lock(&current->mm->page_table_lock);
 405         uaddr = (int __user *) __dat_user_addr((unsigned long) uaddr);
 406         if (!uaddr) {
 407                 spin_unlock(&current->mm->page_table_lock);
 408                 return -EFAULT;
 409         }
 410         get_page(virt_to_page(uaddr));
 411         spin_unlock(&current->mm->page_table_lock);
 412         asm volatile("   cs   %1,%4,0(%5)\n"
 413                      "0: lr   %0,%1\n"
 414                      "1:\n"
 415                      EX_TABLE(0b,1b)
 416                      : "=d" (ret), "+d" (oldval), "=m" (*uaddr)
 417                      : "0" (-EFAULT), "d" (newval), "a" (uaddr), "m" (*uaddr)
 418                      : "cc", "memory" );
 419         put_page(virt_to_page(uaddr));
 420         return ret;
 421 }
 422
 423 struct uaccess_ops uaccess_pt = {
 424         .copy_from_user         = copy_from_user_pt,
 425         .copy_from_user_small   = copy_from_user_pt,
 426         .copy_to_user           = copy_to_user_pt,
 427         .copy_to_user_small     = copy_to_user_pt,
 428         .copy_in_user           = copy_in_user_pt,
 429         .clear_user             = clear_user_pt,
 430         .strnlen_user           = strnlen_user_pt,
 431         .strncpy_from_user      = strncpy_from_user_pt,
 432         .futex_atomic_op        = futex_atomic_op_pt,
 433         .futex_atomic_cmpxchg   = futex_atomic_cmpxchg_pt,
 434 };