arch/s390/lib/uaccess_pt.c

   1 /*
   2  *  User access functions based on page table walks for enhanced
   3  *  system layout without hardware support.
   4  *
   5  *    Copyright IBM Corp. 2006, 2012
   6  *    Author(s): Gerald Schaefer (gerald.schaefer@de.ibm.com)
   7  */
   8
   9 #include <linux/errno.h>
  10 #include <linux/hardirq.h>
  11 #include <linux/mm.h>
  12 #include <linux/hugetlb.h>
  13 #include <asm/uaccess.h>
  14 #include <asm/futex.h>
  15 #include "uaccess.h"
  16
  17 #ifndef CONFIG_64BIT
  18 #define AHI     "ahi"
  19 #define SLR     "slr"
  20 #else
  21 #define AHI     "aghi"
  22 #define SLR     "slgr"
  23 #endif
  24
  25 static size_t strnlen_kernel(size_t count, const char __user *src)
  26 {
  27         register unsigned long reg0 asm("0") = 0UL;
  28         unsigned long tmp1, tmp2;
  29
  30         asm volatile(
  31                 "   la    %2,0(%1)\n"
  32                 "   la    %3,0(%0,%1)\n"
  33                 "  "SLR"  %0,%0\n"
  34                 "0: srst  %3,%2\n"
  35                 "   jo    0b\n"
  36                 "   la    %0,1(%3)\n"   /* strnlen_kernel results includes \0 */
  37                 "  "SLR"  %0,%1\n"
  38                 "1:\n"
  39                 EX_TABLE(0b,1b)
  40                 : "+a" (count), "+a" (src), "=a" (tmp1), "=a" (tmp2)
  41                 : "d" (reg0) : "cc", "memory");
  42         return count;
  43 }
  44
  45 static size_t copy_in_kernel(size_t count, void __user *to,
  46                              const void __user *from)
  47 {
  48         unsigned long tmp1;
  49
  50         asm volatile(
  51                 "  "AHI"  %0,-1\n"
  52                 "   jo    5f\n"
  53                 "   bras  %3,3f\n"
  54                 "0:"AHI"  %0,257\n"
  55                 "1: mvc   0(1,%1),0(%2)\n"
  56                 "   la    %1,1(%1)\n"
  57                 "   la    %2,1(%2)\n"
  58                 "  "AHI"  %0,-1\n"
  59                 "   jnz   1b\n"
  60                 "   j     5f\n"
  61                 "2: mvc   0(256,%1),0(%2)\n"
  62                 "   la    %1,256(%1)\n"
  63                 "   la    %2,256(%2)\n"
  64                 "3:"AHI"  %0,-256\n"
  65                 "   jnm   2b\n"
  66                 "4: ex    %0,1b-0b(%3)\n"
  67                 "5:"SLR"  %0,%0\n"
  68                 "6:\n"
  69                 EX_TABLE(1b,6b) EX_TABLE(2b,0b) EX_TABLE(4b,0b)
  70                 : "+a" (count), "+a" (to), "+a" (from), "=a" (tmp1)
  71                 : : "cc", "memory");
  72         return count;
  73 }
  74
  75 /*
  76  * Returns kernel address for user virtual address. If the returned address is
  77  * >= -4095 (IS_ERR_VALUE(x) returns true), a fault has occurred and the
  78  * address contains the (negative) exception code.
  79  */
  80 #ifdef CONFIG_64BIT
  81
  82 static unsigned long follow_table(struct mm_struct *mm,
  83                                   unsigned long address, int write)
  84 {
  85         unsigned long *table = (unsigned long *)__pa(mm->pgd);
  86
  87         if (unlikely(address > mm->context.asce_limit - 1))
  88                 return -0x38UL;
  89         switch (mm->context.asce_bits & _ASCE_TYPE_MASK) {
  90         case _ASCE_TYPE_REGION1:
  91                 table = table + ((address >> 53) & 0x7ff);
  92                 if (unlikely(*table & _REGION_ENTRY_INVALID))
  93                         return -0x39UL;
  94                 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
  95                 /* fallthrough */
  96         case _ASCE_TYPE_REGION2:
  97                 table = table + ((address >> 42) & 0x7ff);
  98                 if (unlikely(*table & _REGION_ENTRY_INVALID))
  99                         return -0x3aUL;
 100                 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
 101                 /* fallthrough */
 102         case _ASCE_TYPE_REGION3:
 103                 table = table + ((address >> 31) & 0x7ff);
 104                 if (unlikely(*table & _REGION_ENTRY_INVALID))
 105                         return -0x3bUL;
 106                 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
 107                 /* fallthrough */
 108         case _ASCE_TYPE_SEGMENT:
 109                 table = table + ((address >> 20) & 0x7ff);
 110                 if (unlikely(*table & _SEGMENT_ENTRY_INVALID))
 111                         return -0x10UL;
 112                 if (unlikely(*table & _SEGMENT_ENTRY_LARGE)) {
 113                         if (write && (*table & _SEGMENT_ENTRY_PROTECT))
 114                                 return -0x04UL;
 115                         return (*table & _SEGMENT_ENTRY_ORIGIN_LARGE) +
 116                                 (address & ~_SEGMENT_ENTRY_ORIGIN_LARGE);
 117                 }
 118                 table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
 119         }
 120         table = table + ((address >> 12) & 0xff);
 121         if (unlikely(*table & _PAGE_INVALID))
 122                 return -0x11UL;
 123         if (write && (*table & _PAGE_PROTECT))
 124                 return -0x04UL;
 125         return (*table & PAGE_MASK) + (address & ~PAGE_MASK);
 126 }
 127
 128 #else /* CONFIG_64BIT */
 129
 130 static unsigned long follow_table(struct mm_struct *mm,
 131                                   unsigned long address, int write)
 132 {
 133         unsigned long *table = (unsigned long *)__pa(mm->pgd);
 134
 135         table = table + ((address >> 20) & 0x7ff);
 136         if (unlikely(*table & _SEGMENT_ENTRY_INVALID))
 137                 return -0x10UL;
 138         table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
 139         table = table + ((address >> 12) & 0xff);
 140         if (unlikely(*table & _PAGE_INVALID))
 141                 return -0x11UL;
 142         if (write && (*table & _PAGE_PROTECT))
 143                 return -0x04UL;
 144         return (*table & PAGE_MASK) + (address & ~PAGE_MASK);
 145 }
 146
 147 #endif /* CONFIG_64BIT */
 148
 149 static __always_inline size_t __user_copy_pt(unsigned long uaddr, void *kptr,
 150                                              size_t n, int write_user)
 151 {
 152         struct mm_struct *mm = current->mm;
 153         unsigned long offset, done, size, kaddr;
 154         void *from, *to;
 155
 156         if (!mm)
 157                 return n;
 158         done = 0;
 159 retry:
 160         spin_lock(&mm->page_table_lock);
 161         do {
 162                 kaddr = follow_table(mm, uaddr, write_user);
 163                 if (IS_ERR_VALUE(kaddr))
 164                         goto fault;
 165
 166                 offset = uaddr & ~PAGE_MASK;
 167                 size = min(n - done, PAGE_SIZE - offset);
 168                 if (write_user) {
 169                         to = (void *) kaddr;
 170                         from = kptr + done;
 171                 } else {
 172                         from = (void *) kaddr;
 173                         to = kptr + done;
 174                 }
 175                 memcpy(to, from, size);
 176                 done += size;
 177                 uaddr += size;
 178         } while (done < n);
 179         spin_unlock(&mm->page_table_lock);
 180         return n - done;
 181 fault:
 182         spin_unlock(&mm->page_table_lock);
 183         if (__handle_fault(uaddr, -kaddr, write_user))
 184                 return n - done;
 185         goto retry;
 186 }
 187
 188 /*
 189  * Do DAT for user address by page table walk, return kernel address.
 190  * This function needs to be called with current->mm->page_table_lock held.
 191  */
 192 static __always_inline unsigned long __dat_user_addr(unsigned long uaddr,
 193                                                      int write)
 194 {
 195         struct mm_struct *mm = current->mm;
 196         unsigned long kaddr;
 197         int rc;
 198
 199 retry:
 200         kaddr = follow_table(mm, uaddr, write);
 201         if (IS_ERR_VALUE(kaddr))
 202                 goto fault;
 203
 204         return kaddr;
 205 fault:
 206         spin_unlock(&mm->page_table_lock);
 207         rc = __handle_fault(uaddr, -kaddr, write);
 208         spin_lock(&mm->page_table_lock);
 209         if (!rc)
 210                 goto retry;
 211         return 0;
 212 }
 213
 214 size_t copy_from_user_pt(size_t n, const void __user *from, void *to)
 215 {
 216         size_t rc;
 217
 218         if (segment_eq(get_fs(), KERNEL_DS))
 219                 return copy_in_kernel(n, (void __user *) to, from);
 220         rc = __user_copy_pt((unsigned long) from, to, n, 0);
 221         if (unlikely(rc))
 222                 memset(to + n - rc, 0, rc);
 223         return rc;
 224 }
 225
 226 size_t copy_to_user_pt(size_t n, void __user *to, const void *from)
 227 {
 228         if (segment_eq(get_fs(), KERNEL_DS))
 229                 return copy_in_kernel(n, to, (void __user *) from);
 230         return __user_copy_pt((unsigned long) to, (void *) from, n, 1);
 231 }
 232
 233 static size_t clear_user_pt(size_t n, void __user *to)
 234 {
 235         void *zpage = (void *) empty_zero_page;
 236         long done, size, ret;
 237
 238         done = 0;
 239         do {
 240                 if (n - done > PAGE_SIZE)
 241                         size = PAGE_SIZE;
 242                 else
 243                         size = n - done;
 244                 if (segment_eq(get_fs(), KERNEL_DS))
 245                         ret = copy_in_kernel(n, to, (void __user *) zpage);
 246                 else
 247                         ret = __user_copy_pt((unsigned long) to, zpage, size, 1);
 248                 done += size;
 249                 to += size;
 250                 if (ret)
 251                         return ret + n - done;
 252         } while (done < n);
 253         return 0;
 254 }
 255
 256 static size_t strnlen_user_pt(size_t count, const char __user *src)
 257 {
 258         unsigned long uaddr = (unsigned long) src;
 259         struct mm_struct *mm = current->mm;
 260         unsigned long offset, done, len, kaddr;
 261         size_t len_str;
 262
 263         if (unlikely(!count))
 264                 return 0;
 265         if (segment_eq(get_fs(), KERNEL_DS))
 266                 return strnlen_kernel(count, src);
 267         if (!mm)
 268                 return 0;
 269         done = 0;
 270 retry:
 271         spin_lock(&mm->page_table_lock);
 272         do {
 273                 kaddr = follow_table(mm, uaddr, 0);
 274                 if (IS_ERR_VALUE(kaddr))
 275                         goto fault;
 276
 277                 offset = uaddr & ~PAGE_MASK;
 278                 len = min(count - done, PAGE_SIZE - offset);
 279                 len_str = strnlen((char *) kaddr, len);
 280                 done += len_str;
 281                 uaddr += len_str;
 282         } while ((len_str == len) && (done < count));
 283         spin_unlock(&mm->page_table_lock);
 284         return done + 1;
 285 fault:
 286         spin_unlock(&mm->page_table_lock);
 287         if (__handle_fault(uaddr, -kaddr, 0))
 288                 return 0;
 289         goto retry;
 290 }
 291
 292 static size_t strncpy_from_user_pt(size_t count, const char __user *src,
 293                                    char *dst)
 294 {
 295         size_t done, len, offset, len_str;
 296
 297         if (unlikely(!count))
 298                 return 0;
 299         done = 0;
 300         do {
 301                 offset = (size_t)src & ~PAGE_MASK;
 302                 len = min(count - done, PAGE_SIZE - offset);
 303                 if (segment_eq(get_fs(), KERNEL_DS)) {
 304                         if (copy_in_kernel(len, (void __user *) dst, src))
 305                                 return -EFAULT;
 306                 } else {
 307                         if (__user_copy_pt((unsigned long) src, dst, len, 0))
 308                                 return -EFAULT;
 309                 }
 310                 len_str = strnlen(dst, len);
 311                 done += len_str;
 312                 src += len_str;
 313                 dst += len_str;
 314         } while ((len_str == len) && (done < count));
 315         return done;
 316 }
 317
 318 static size_t copy_in_user_pt(size_t n, void __user *to,
 319                               const void __user *from)
 320 {
 321         struct mm_struct *mm = current->mm;
 322         unsigned long offset_max, uaddr, done, size, error_code;
 323         unsigned long uaddr_from = (unsigned long) from;
 324         unsigned long uaddr_to = (unsigned long) to;
 325         unsigned long kaddr_to, kaddr_from;
 326         int write_user;
 327
 328         if (segment_eq(get_fs(), KERNEL_DS))
 329                 return copy_in_kernel(n, to, from);
 330         if (!mm)
 331                 return n;
 332         done = 0;
 333 retry:
 334         spin_lock(&mm->page_table_lock);
 335         do {
 336                 write_user = 0;
 337                 uaddr = uaddr_from;
 338                 kaddr_from = follow_table(mm, uaddr_from, 0);
 339                 error_code = kaddr_from;
 340                 if (IS_ERR_VALUE(error_code))
 341                         goto fault;
 342
 343                 write_user = 1;
 344                 uaddr = uaddr_to;
 345                 kaddr_to = follow_table(mm, uaddr_to, 1);
 346                 error_code = (unsigned long) kaddr_to;
 347                 if (IS_ERR_VALUE(error_code))
 348                         goto fault;
 349
 350                 offset_max = max(uaddr_from & ~PAGE_MASK,
 351                                  uaddr_to & ~PAGE_MASK);
 352                 size = min(n - done, PAGE_SIZE - offset_max);
 353
 354                 memcpy((void *) kaddr_to, (void *) kaddr_from, size);
 355                 done += size;
 356                 uaddr_from += size;
 357                 uaddr_to += size;
 358         } while (done < n);
 359         spin_unlock(&mm->page_table_lock);
 360         return n - done;
 361 fault:
 362         spin_unlock(&mm->page_table_lock);
 363         if (__handle_fault(uaddr, -error_code, write_user))
 364                 return n - done;
 365         goto retry;
 366 }
 367
 368 #define __futex_atomic_op(insn, ret, oldval, newval, uaddr, oparg)      \
 369         asm volatile("0: l   %1,0(%6)\n"                                \
 370                      "1: " insn                                         \
 371                      "2: cs  %1,%2,0(%6)\n"                             \
 372                      "3: jl  1b\n"                                      \
 373                      "   lhi %0,0\n"                                    \
 374                      "4:\n"                                             \
 375                      EX_TABLE(0b,4b) EX_TABLE(2b,4b) EX_TABLE(3b,4b)    \
 376                      : "=d" (ret), "=&d" (oldval), "=&d" (newval),      \
 377                        "=m" (*uaddr)                                    \
 378                      : "0" (-EFAULT), "d" (oparg), "a" (uaddr),         \
 379                        "m" (*uaddr) : "cc" );
 380
 381 static int __futex_atomic_op_pt(int op, u32 __user *uaddr, int oparg, int *old)
 382 {
 383         int oldval = 0, newval, ret;
 384
 385         switch (op) {
 386         case FUTEX_OP_SET:
 387                 __futex_atomic_op("lr %2,%5\n",
 388                                   ret, oldval, newval, uaddr, oparg);
 389                 break;
 390         case FUTEX_OP_ADD:
 391                 __futex_atomic_op("lr %2,%1\nar %2,%5\n",
 392                                   ret, oldval, newval, uaddr, oparg);
 393                 break;
 394         case FUTEX_OP_OR:
 395                 __futex_atomic_op("lr %2,%1\nor %2,%5\n",
 396                                   ret, oldval, newval, uaddr, oparg);
 397                 break;
 398         case FUTEX_OP_ANDN:
 399                 __futex_atomic_op("lr %2,%1\nnr %2,%5\n",
 400                                   ret, oldval, newval, uaddr, oparg);
 401                 break;
 402         case FUTEX_OP_XOR:
 403                 __futex_atomic_op("lr %2,%1\nxr %2,%5\n",
 404                                   ret, oldval, newval, uaddr, oparg);
 405                 break;
 406         default:
 407                 ret = -ENOSYS;
 408         }
 409         if (ret == 0)
 410                 *old = oldval;
 411         return ret;
 412 }
 413
 414 int futex_atomic_op_pt(int op, u32 __user *uaddr, int oparg, int *old)
 415 {
 416         int ret;
 417
 418         if (segment_eq(get_fs(), KERNEL_DS))
 419                 return __futex_atomic_op_pt(op, uaddr, oparg, old);
 420         if (unlikely(!current->mm))
 421                 return -EFAULT;
 422         spin_lock(&current->mm->page_table_lock);
 423         uaddr = (u32 __force __user *)
 424                 __dat_user_addr((__force unsigned long) uaddr, 1);
 425         if (!uaddr) {
 426                 spin_unlock(&current->mm->page_table_lock);
 427                 return -EFAULT;
 428         }
 429         get_page(virt_to_page(uaddr));
 430         spin_unlock(&current->mm->page_table_lock);
 431         ret = __futex_atomic_op_pt(op, uaddr, oparg, old);
 432         put_page(virt_to_page(uaddr));
 433         return ret;
 434 }
 435
 436 static int __futex_atomic_cmpxchg_pt(u32 *uval, u32 __user *uaddr,
 437                                      u32 oldval, u32 newval)
 438 {
 439         int ret;
 440
 441         asm volatile("0: cs   %1,%4,0(%5)\n"
 442                      "1: la   %0,0\n"
 443                      "2:\n"
 444                      EX_TABLE(0b,2b) EX_TABLE(1b,2b)
 445                      : "=d" (ret), "+d" (oldval), "=m" (*uaddr)
 446                      : "0" (-EFAULT), "d" (newval), "a" (uaddr), "m" (*uaddr)
 447                      : "cc", "memory" );
 448         *uval = oldval;
 449         return ret;
 450 }
 451
 452 int futex_atomic_cmpxchg_pt(u32 *uval, u32 __user *uaddr,
 453                             u32 oldval, u32 newval)
 454 {
 455         int ret;
 456
 457         if (segment_eq(get_fs(), KERNEL_DS))
 458                 return __futex_atomic_cmpxchg_pt(uval, uaddr, oldval, newval);
 459         if (unlikely(!current->mm))
 460                 return -EFAULT;
 461         spin_lock(&current->mm->page_table_lock);
 462         uaddr = (u32 __force __user *)
 463                 __dat_user_addr((__force unsigned long) uaddr, 1);
 464         if (!uaddr) {
 465                 spin_unlock(&current->mm->page_table_lock);
 466                 return -EFAULT;
 467         }
 468         get_page(virt_to_page(uaddr));
 469         spin_unlock(&current->mm->page_table_lock);
 470         ret = __futex_atomic_cmpxchg_pt(uval, uaddr, oldval, newval);
 471         put_page(virt_to_page(uaddr));
 472         return ret;
 473 }
 474
 475 struct uaccess_ops uaccess_pt = {
 476         .copy_from_user         = copy_from_user_pt,
 477         .copy_to_user           = copy_to_user_pt,
 478         .copy_in_user           = copy_in_user_pt,
 479         .clear_user             = clear_user_pt,
 480         .strnlen_user           = strnlen_user_pt,
 481         .strncpy_from_user      = strncpy_from_user_pt,
 482         .futex_atomic_op        = futex_atomic_op_pt,
 483         .futex_atomic_cmpxchg   = futex_atomic_cmpxchg_pt,
 484 };