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
   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 <asm/uaccess.h>
  13 #include <asm/futex.h>
  14 #include "uaccess.h"
  15
  16 static inline pte_t *follow_table(struct mm_struct *mm, unsigned long addr)
  17 {
  18         pgd_t *pgd;
  19         pud_t *pud;
  20         pmd_t *pmd;
  21
  22         pgd = pgd_offset(mm, addr);
  23         if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
  24                 return (pte_t *) 0x3a;
  25
  26         pud = pud_offset(pgd, addr);
  27         if (pud_none(*pud) || unlikely(pud_bad(*pud)))
  28                 return (pte_t *) 0x3b;
  29
  30         pmd = pmd_offset(pud, addr);
  31         if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
  32                 return (pte_t *) 0x10;
  33
  34         return pte_offset_map(pmd, addr);
  35 }
  36
  37 static __always_inline size_t __user_copy_pt(unsigned long uaddr, void *kptr,
  38                                              size_t n, int write_user)
  39 {
  40         struct mm_struct *mm = current->mm;
  41         unsigned long offset, pfn, done, size;
  42         pte_t *pte;
  43         void *from, *to;
  44
  45         done = 0;
  46 retry:
  47         spin_lock(&mm->page_table_lock);
  48         do {
  49                 pte = follow_table(mm, uaddr);
  50                 if ((unsigned long) pte < 0x1000)
  51                         goto fault;
  52                 if (!pte_present(*pte)) {
  53                         pte = (pte_t *) 0x11;
  54                         goto fault;
  55                 } else if (write_user && !pte_write(*pte)) {
  56                         pte = (pte_t *) 0x04;
  57                         goto fault;
  58                 }
  59
  60                 pfn = pte_pfn(*pte);
  61                 offset = uaddr & (PAGE_SIZE - 1);
  62                 size = min(n - done, PAGE_SIZE - offset);
  63                 if (write_user) {
  64                         to = (void *)((pfn << PAGE_SHIFT) + offset);
  65                         from = kptr + done;
  66                 } else {
  67                         from = (void *)((pfn << PAGE_SHIFT) + offset);
  68                         to = kptr + done;
  69                 }
  70                 memcpy(to, from, size);
  71                 done += size;
  72                 uaddr += size;
  73         } while (done < n);
  74         spin_unlock(&mm->page_table_lock);
  75         return n - done;
  76 fault:
  77         spin_unlock(&mm->page_table_lock);
  78         if (__handle_fault(uaddr, (unsigned long) pte, write_user))
  79                 return n - done;
  80         goto retry;
  81 }
  82
  83 /*
  84  * Do DAT for user address by page table walk, return kernel address.
  85  * This function needs to be called with current->mm->page_table_lock held.
  86  */
  87 static __always_inline unsigned long __dat_user_addr(unsigned long uaddr)
  88 {
  89         struct mm_struct *mm = current->mm;
  90         unsigned long pfn;
  91         pte_t *pte;
  92         int rc;
  93
  94 retry:
  95         pte = follow_table(mm, uaddr);
  96         if ((unsigned long) pte < 0x1000)
  97                 goto fault;
  98         if (!pte_present(*pte)) {
  99                 pte = (pte_t *) 0x11;
 100                 goto fault;
 101         }
 102
 103         pfn = pte_pfn(*pte);
 104         return (pfn << PAGE_SHIFT) + (uaddr & (PAGE_SIZE - 1));
 105 fault:
 106         spin_unlock(&mm->page_table_lock);
 107         rc = __handle_fault(uaddr, (unsigned long) pte, 0);
 108         spin_lock(&mm->page_table_lock);
 109         if (!rc)
 110                 goto retry;
 111         return 0;
 112 }
 113
 114 size_t copy_from_user_pt(size_t n, const void __user *from, void *to)
 115 {
 116         size_t rc;
 117
 118         if (segment_eq(get_fs(), KERNEL_DS)) {
 119                 memcpy(to, (void __kernel __force *) from, n);
 120                 return 0;
 121         }
 122         rc = __user_copy_pt((unsigned long) from, to, n, 0);
 123         if (unlikely(rc))
 124                 memset(to + n - rc, 0, rc);
 125         return rc;
 126 }
 127
 128 size_t copy_to_user_pt(size_t n, void __user *to, const void *from)
 129 {
 130         if (segment_eq(get_fs(), KERNEL_DS)) {
 131                 memcpy((void __kernel __force *) to, from, n);
 132                 return 0;
 133         }
 134         return __user_copy_pt((unsigned long) to, (void *) from, n, 1);
 135 }
 136
 137 static size_t clear_user_pt(size_t n, void __user *to)
 138 {
 139         long done, size, ret;
 140
 141         if (segment_eq(get_fs(), KERNEL_DS)) {
 142                 memset((void __kernel __force *) to, 0, n);
 143                 return 0;
 144         }
 145         done = 0;
 146         do {
 147                 if (n - done > PAGE_SIZE)
 148                         size = PAGE_SIZE;
 149                 else
 150                         size = n - done;
 151                 ret = __user_copy_pt((unsigned long) to + done,
 152                                       &empty_zero_page, size, 1);
 153                 done += size;
 154                 if (ret)
 155                         return ret + n - done;
 156         } while (done < n);
 157         return 0;
 158 }
 159
 160 static size_t strnlen_user_pt(size_t count, const char __user *src)
 161 {
 162         char *addr;
 163         unsigned long uaddr = (unsigned long) src;
 164         struct mm_struct *mm = current->mm;
 165         unsigned long offset, pfn, done, len;
 166         pte_t *pte;
 167         size_t len_str;
 168
 169         if (segment_eq(get_fs(), KERNEL_DS))
 170                 return strnlen((const char __kernel __force *) src, count) + 1;
 171         done = 0;
 172 retry:
 173         spin_lock(&mm->page_table_lock);
 174         do {
 175                 pte = follow_table(mm, uaddr);
 176                 if ((unsigned long) pte < 0x1000)
 177                         goto fault;
 178                 if (!pte_present(*pte)) {
 179                         pte = (pte_t *) 0x11;
 180                         goto fault;
 181                 }
 182
 183                 pfn = pte_pfn(*pte);
 184                 offset = uaddr & (PAGE_SIZE-1);
 185                 addr = (char *)(pfn << PAGE_SHIFT) + offset;
 186                 len = min(count - done, PAGE_SIZE - offset);
 187                 len_str = strnlen(addr, len);
 188                 done += len_str;
 189                 uaddr += len_str;
 190         } while ((len_str == len) && (done < count));
 191         spin_unlock(&mm->page_table_lock);
 192         return done + 1;
 193 fault:
 194         spin_unlock(&mm->page_table_lock);
 195         if (__handle_fault(uaddr, (unsigned long) pte, 0))
 196                 return 0;
 197         goto retry;
 198 }
 199
 200 static size_t strncpy_from_user_pt(size_t count, const char __user *src,
 201                                    char *dst)
 202 {
 203         size_t n = strnlen_user_pt(count, src);
 204
 205         if (!n)
 206                 return -EFAULT;
 207         if (n > count)
 208                 n = count;
 209         if (segment_eq(get_fs(), KERNEL_DS)) {
 210                 memcpy(dst, (const char __kernel __force *) src, n);
 211                 if (dst[n-1] == '\0')
 212                         return n-1;
 213                 else
 214                         return n;
 215         }
 216         if (__user_copy_pt((unsigned long) src, dst, n, 0))
 217                 return -EFAULT;
 218         if (dst[n-1] == '\0')
 219                 return n-1;
 220         else
 221                 return n;
 222 }
 223
 224 static size_t copy_in_user_pt(size_t n, void __user *to,
 225                               const void __user *from)
 226 {
 227         struct mm_struct *mm = current->mm;
 228         unsigned long offset_from, offset_to, offset_max, pfn_from, pfn_to,
 229                       uaddr, done, size, error_code;
 230         unsigned long uaddr_from = (unsigned long) from;
 231         unsigned long uaddr_to = (unsigned long) to;
 232         pte_t *pte_from, *pte_to;
 233         int write_user;
 234
 235         if (segment_eq(get_fs(), KERNEL_DS)) {
 236                 memcpy((void __force *) to, (void __force *) from, n);
 237                 return 0;
 238         }
 239         done = 0;
 240 retry:
 241         spin_lock(&mm->page_table_lock);
 242         do {
 243                 write_user = 0;
 244                 uaddr = uaddr_from;
 245                 pte_from = follow_table(mm, uaddr_from);
 246                 error_code = (unsigned long) pte_from;
 247                 if (error_code < 0x1000)
 248                         goto fault;
 249                 if (!pte_present(*pte_from)) {
 250                         error_code = 0x11;
 251                         goto fault;
 252                 }
 253
 254                 write_user = 1;
 255                 uaddr = uaddr_to;
 256                 pte_to = follow_table(mm, uaddr_to);
 257                 error_code = (unsigned long) pte_to;
 258                 if (error_code < 0x1000)
 259                         goto fault;
 260                 if (!pte_present(*pte_to)) {
 261                         error_code = 0x11;
 262                         goto fault;
 263                 } else if (!pte_write(*pte_to)) {
 264                         error_code = 0x04;
 265                         goto fault;
 266                 }
 267
 268                 pfn_from = pte_pfn(*pte_from);
 269                 pfn_to = pte_pfn(*pte_to);
 270                 offset_from = uaddr_from & (PAGE_SIZE-1);
 271                 offset_to = uaddr_from & (PAGE_SIZE-1);
 272                 offset_max = max(offset_from, offset_to);
 273                 size = min(n - done, PAGE_SIZE - offset_max);
 274
 275                 memcpy((void *)(pfn_to << PAGE_SHIFT) + offset_to,
 276                        (void *)(pfn_from << PAGE_SHIFT) + offset_from, size);
 277                 done += size;
 278                 uaddr_from += size;
 279                 uaddr_to += size;
 280         } while (done < n);
 281         spin_unlock(&mm->page_table_lock);
 282         return n - done;
 283 fault:
 284         spin_unlock(&mm->page_table_lock);
 285         if (__handle_fault(uaddr, error_code, write_user))
 286                 return n - done;
 287         goto retry;
 288 }
 289
 290 #define __futex_atomic_op(insn, ret, oldval, newval, uaddr, oparg)      \
 291         asm volatile("0: l   %1,0(%6)\n"                                \
 292                      "1: " insn                                         \
 293                      "2: cs  %1,%2,0(%6)\n"                             \
 294                      "3: jl  1b\n"                                      \
 295                      "   lhi %0,0\n"                                    \
 296                      "4:\n"                                             \
 297                      EX_TABLE(0b,4b) EX_TABLE(2b,4b) EX_TABLE(3b,4b)    \
 298                      : "=d" (ret), "=&d" (oldval), "=&d" (newval),      \
 299                        "=m" (*uaddr)                                    \
 300                      : "0" (-EFAULT), "d" (oparg), "a" (uaddr),         \
 301                        "m" (*uaddr) : "cc" );
 302
 303 static int __futex_atomic_op_pt(int op, u32 __user *uaddr, int oparg, int *old)
 304 {
 305         int oldval = 0, newval, ret;
 306
 307         switch (op) {
 308         case FUTEX_OP_SET:
 309                 __futex_atomic_op("lr %2,%5\n",
 310                                   ret, oldval, newval, uaddr, oparg);
 311                 break;
 312         case FUTEX_OP_ADD:
 313                 __futex_atomic_op("lr %2,%1\nar %2,%5\n",
 314                                   ret, oldval, newval, uaddr, oparg);
 315                 break;
 316         case FUTEX_OP_OR:
 317                 __futex_atomic_op("lr %2,%1\nor %2,%5\n",
 318                                   ret, oldval, newval, uaddr, oparg);
 319                 break;
 320         case FUTEX_OP_ANDN:
 321                 __futex_atomic_op("lr %2,%1\nnr %2,%5\n",
 322                                   ret, oldval, newval, uaddr, oparg);
 323                 break;
 324         case FUTEX_OP_XOR:
 325                 __futex_atomic_op("lr %2,%1\nxr %2,%5\n",
 326                                   ret, oldval, newval, uaddr, oparg);
 327                 break;
 328         default:
 329                 ret = -ENOSYS;
 330         }
 331         if (ret == 0)
 332                 *old = oldval;
 333         return ret;
 334 }
 335
 336 int futex_atomic_op_pt(int op, u32 __user *uaddr, int oparg, int *old)
 337 {
 338         int ret;
 339
 340         if (segment_eq(get_fs(), KERNEL_DS))
 341                 return __futex_atomic_op_pt(op, uaddr, oparg, old);
 342         spin_lock(&current->mm->page_table_lock);
 343         uaddr = (u32 __force __user *)
 344                 __dat_user_addr((__force unsigned long) uaddr);
 345         if (!uaddr) {
 346                 spin_unlock(&current->mm->page_table_lock);
 347                 return -EFAULT;
 348         }
 349         get_page(virt_to_page(uaddr));
 350         spin_unlock(&current->mm->page_table_lock);
 351         ret = __futex_atomic_op_pt(op, uaddr, oparg, old);
 352         put_page(virt_to_page(uaddr));
 353         return ret;
 354 }
 355
 356 static int __futex_atomic_cmpxchg_pt(u32 *uval, u32 __user *uaddr,
 357                                      u32 oldval, u32 newval)
 358 {
 359         int ret;
 360
 361         asm volatile("0: cs   %1,%4,0(%5)\n"
 362                      "1: la   %0,0\n"
 363                      "2:\n"
 364                      EX_TABLE(0b,2b) EX_TABLE(1b,2b)
 365                      : "=d" (ret), "+d" (oldval), "=m" (*uaddr)
 366                      : "0" (-EFAULT), "d" (newval), "a" (uaddr), "m" (*uaddr)
 367                      : "cc", "memory" );
 368         *uval = oldval;
 369         return ret;
 370 }
 371
 372 int futex_atomic_cmpxchg_pt(u32 *uval, u32 __user *uaddr,
 373                             u32 oldval, u32 newval)
 374 {
 375         int ret;
 376
 377         if (segment_eq(get_fs(), KERNEL_DS))
 378                 return __futex_atomic_cmpxchg_pt(uval, uaddr, oldval, newval);
 379         spin_lock(&current->mm->page_table_lock);
 380         uaddr = (u32 __force __user *)
 381                 __dat_user_addr((__force unsigned long) uaddr);
 382         if (!uaddr) {
 383                 spin_unlock(&current->mm->page_table_lock);
 384                 return -EFAULT;
 385         }
 386         get_page(virt_to_page(uaddr));
 387         spin_unlock(&current->mm->page_table_lock);
 388         ret = __futex_atomic_cmpxchg_pt(uval, uaddr, oldval, newval);
 389         put_page(virt_to_page(uaddr));
 390         return ret;
 391 }
 392
 393 struct uaccess_ops uaccess_pt = {
 394         .copy_from_user         = copy_from_user_pt,
 395         .copy_from_user_small   = copy_from_user_pt,
 396         .copy_to_user           = copy_to_user_pt,
 397         .copy_to_user_small     = copy_to_user_pt,
 398         .copy_in_user           = copy_in_user_pt,
 399         .clear_user             = clear_user_pt,
 400         .strnlen_user           = strnlen_user_pt,
 401         .strncpy_from_user      = strncpy_from_user_pt,
 402         .futex_atomic_op        = futex_atomic_op_pt,
 403         .futex_atomic_cmpxchg   = futex_atomic_cmpxchg_pt,
 404 };