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         pud_t *pud;
  22         pmd_t *pmd;
  23
  24         pgd = pgd_offset(mm, addr);
  25         if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
  26                 return NULL;
  27
  28         pud = pud_offset(pgd, addr);
  29         if (pud_none(*pud) || unlikely(pud_bad(*pud)))
  30                 return NULL;
  31
  32         pmd = pmd_offset(pud, addr);
  33         if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
  34                 return NULL;
  35
  36         return pte_offset_map(pmd, addr);
  37 }
  38
  39 static int __handle_fault(struct mm_struct *mm, unsigned long address,
  40                           int write_access)
  41 {
  42         struct vm_area_struct *vma;
  43         int ret = -EFAULT;
  44         int fault;
  45
  46         if (in_atomic())
  47                 return ret;
  48         down_read(&mm->mmap_sem);
  49         vma = find_vma(mm, address);
  50         if (unlikely(!vma))
  51                 goto out;
  52         if (unlikely(vma->vm_start > address)) {
  53                 if (!(vma->vm_flags & VM_GROWSDOWN))
  54                         goto out;
  55                 if (expand_stack(vma, address))
  56                         goto out;
  57         }
  58
  59         if (!write_access) {
  60                 /* page not present, check vm flags */
  61                 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
  62                         goto out;
  63         } else {
  64                 if (!(vma->vm_flags & VM_WRITE))
  65                         goto out;
  66         }
  67
  68 survive:
  69         fault = handle_mm_fault(mm, vma, address, write_access);
  70         if (unlikely(fault & VM_FAULT_ERROR)) {
  71                 if (fault & VM_FAULT_OOM)
  72                         goto out_of_memory;
  73                 else if (fault & VM_FAULT_SIGBUS)
  74                         goto out_sigbus;
  75                 BUG();
  76         }
  77         if (fault & VM_FAULT_MAJOR)
  78                 current->maj_flt++;
  79         else
  80                 current->min_flt++;
  81         ret = 0;
  82 out:
  83         up_read(&mm->mmap_sem);
  84         return ret;
  85
  86 out_of_memory:
  87         up_read(&mm->mmap_sem);
  88         if (is_global_init(current)) {
  89                 yield();
  90                 down_read(&mm->mmap_sem);
  91                 goto survive;
  92         }
  93         printk("VM: killing process %s\n", current->comm);
  94         return ret;
  95
  96 out_sigbus:
  97         up_read(&mm->mmap_sem);
  98         current->thread.prot_addr = address;
  99         current->thread.trap_no = 0x11;
 100         force_sig(SIGBUS, current);
 101         return ret;
 102 }
 103
 104 static size_t __user_copy_pt(unsigned long uaddr, void *kptr,
 105                              size_t n, int write_user)
 106 {
 107         struct mm_struct *mm = current->mm;
 108         unsigned long offset, pfn, done, size;
 109         pte_t *pte;
 110         void *from, *to;
 111
 112         done = 0;
 113 retry:
 114         spin_lock(&mm->page_table_lock);
 115         do {
 116                 pte = follow_table(mm, uaddr);
 117                 if (!pte || !pte_present(*pte) ||
 118                     (write_user && !pte_write(*pte)))
 119                         goto fault;
 120
 121                 pfn = pte_pfn(*pte);
 122
 123                 offset = uaddr & (PAGE_SIZE - 1);
 124                 size = min(n - done, PAGE_SIZE - offset);
 125                 if (write_user) {
 126                         to = (void *)((pfn << PAGE_SHIFT) + offset);
 127                         from = kptr + done;
 128                 } else {
 129                         from = (void *)((pfn << PAGE_SHIFT) + offset);
 130                         to = kptr + done;
 131                 }
 132                 memcpy(to, from, size);
 133                 done += size;
 134                 uaddr += size;
 135         } while (done < n);
 136         spin_unlock(&mm->page_table_lock);
 137         return n - done;
 138 fault:
 139         spin_unlock(&mm->page_table_lock);
 140         if (__handle_fault(mm, uaddr, write_user))
 141                 return n - done;
 142         goto retry;
 143 }
 144
 145 /*
 146  * Do DAT for user address by page table walk, return kernel address.
 147  * This function needs to be called with current->mm->page_table_lock held.
 148  */
 149 static unsigned long __dat_user_addr(unsigned long uaddr)
 150 {
 151         struct mm_struct *mm = current->mm;
 152         unsigned long pfn, ret;
 153         pte_t *pte;
 154         int rc;
 155
 156         ret = 0;
 157 retry:
 158         pte = follow_table(mm, uaddr);
 159         if (!pte || !pte_present(*pte))
 160                 goto fault;
 161
 162         pfn = pte_pfn(*pte);
 163         ret = (pfn << PAGE_SHIFT) + (uaddr & (PAGE_SIZE - 1));
 164 out:
 165         return ret;
 166 fault:
 167         spin_unlock(&mm->page_table_lock);
 168         rc = __handle_fault(mm, uaddr, 0);
 169         spin_lock(&mm->page_table_lock);
 170         if (rc)
 171                 goto out;
 172         goto retry;
 173 }
 174
 175 size_t copy_from_user_pt(size_t n, const void __user *from, void *to)
 176 {
 177         size_t rc;
 178
 179         if (segment_eq(get_fs(), KERNEL_DS)) {
 180                 memcpy(to, (void __kernel __force *) from, n);
 181                 return 0;
 182         }
 183         rc = __user_copy_pt((unsigned long) from, to, n, 0);
 184         if (unlikely(rc))
 185                 memset(to + n - rc, 0, rc);
 186         return rc;
 187 }
 188
 189 size_t copy_to_user_pt(size_t n, void __user *to, const void *from)
 190 {
 191         if (segment_eq(get_fs(), KERNEL_DS)) {
 192                 memcpy((void __kernel __force *) to, from, n);
 193                 return 0;
 194         }
 195         return __user_copy_pt((unsigned long) to, (void *) from, n, 1);
 196 }
 197
 198 static size_t clear_user_pt(size_t n, void __user *to)
 199 {
 200         long done, size, ret;
 201
 202         if (segment_eq(get_fs(), KERNEL_DS)) {
 203                 memset((void __kernel __force *) to, 0, n);
 204                 return 0;
 205         }
 206         done = 0;
 207         do {
 208                 if (n - done > PAGE_SIZE)
 209                         size = PAGE_SIZE;
 210                 else
 211                         size = n - done;
 212                 ret = __user_copy_pt((unsigned long) to + done,
 213                                       &empty_zero_page, size, 1);
 214                 done += size;
 215                 if (ret)
 216                         return ret + n - done;
 217         } while (done < n);
 218         return 0;
 219 }
 220
 221 static size_t strnlen_user_pt(size_t count, const char __user *src)
 222 {
 223         char *addr;
 224         unsigned long uaddr = (unsigned long) src;
 225         struct mm_struct *mm = current->mm;
 226         unsigned long offset, pfn, done, len;
 227         pte_t *pte;
 228         size_t len_str;
 229
 230         if (segment_eq(get_fs(), KERNEL_DS))
 231                 return strnlen((const char __kernel __force *) src, count) + 1;
 232         done = 0;
 233 retry:
 234         spin_lock(&mm->page_table_lock);
 235         do {
 236                 pte = follow_table(mm, uaddr);
 237                 if (!pte || !pte_present(*pte))
 238                         goto fault;
 239
 240                 pfn = pte_pfn(*pte);
 241                 offset = uaddr & (PAGE_SIZE-1);
 242                 addr = (char *)(pfn << PAGE_SHIFT) + offset;
 243                 len = min(count - done, PAGE_SIZE - offset);
 244                 len_str = strnlen(addr, len);
 245                 done += len_str;
 246                 uaddr += len_str;
 247         } while ((len_str == len) && (done < count));
 248         spin_unlock(&mm->page_table_lock);
 249         return done + 1;
 250 fault:
 251         spin_unlock(&mm->page_table_lock);
 252         if (__handle_fault(mm, uaddr, 0)) {
 253                 return 0;
 254         }
 255         goto retry;
 256 }
 257
 258 static size_t strncpy_from_user_pt(size_t count, const char __user *src,
 259                                    char *dst)
 260 {
 261         size_t n = strnlen_user_pt(count, src);
 262
 263         if (!n)
 264                 return -EFAULT;
 265         if (n > count)
 266                 n = count;
 267         if (segment_eq(get_fs(), KERNEL_DS)) {
 268                 memcpy(dst, (const char __kernel __force *) src, n);
 269                 if (dst[n-1] == '\0')
 270                         return n-1;
 271                 else
 272                         return n;
 273         }
 274         if (__user_copy_pt((unsigned long) src, dst, n, 0))
 275                 return -EFAULT;
 276         if (dst[n-1] == '\0')
 277                 return n-1;
 278         else
 279                 return n;
 280 }
 281
 282 static size_t copy_in_user_pt(size_t n, void __user *to,
 283                               const void __user *from)
 284 {
 285         struct mm_struct *mm = current->mm;
 286         unsigned long offset_from, offset_to, offset_max, pfn_from, pfn_to,
 287                       uaddr, done, size;
 288         unsigned long uaddr_from = (unsigned long) from;
 289         unsigned long uaddr_to = (unsigned long) to;
 290         pte_t *pte_from, *pte_to;
 291         int write_user;
 292
 293         if (segment_eq(get_fs(), KERNEL_DS)) {
 294                 memcpy((void __force *) to, (void __force *) from, n);
 295                 return 0;
 296         }
 297         done = 0;
 298 retry:
 299         spin_lock(&mm->page_table_lock);
 300         do {
 301                 pte_from = follow_table(mm, uaddr_from);
 302                 if (!pte_from || !pte_present(*pte_from)) {
 303                         uaddr = uaddr_from;
 304                         write_user = 0;
 305                         goto fault;
 306                 }
 307
 308                 pte_to = follow_table(mm, uaddr_to);
 309                 if (!pte_to || !pte_present(*pte_to) || !pte_write(*pte_to)) {
 310                         uaddr = uaddr_to;
 311                         write_user = 1;
 312                         goto fault;
 313                 }
 314
 315                 pfn_from = pte_pfn(*pte_from);
 316                 pfn_to = pte_pfn(*pte_to);
 317                 offset_from = uaddr_from & (PAGE_SIZE-1);
 318                 offset_to = uaddr_from & (PAGE_SIZE-1);
 319                 offset_max = max(offset_from, offset_to);
 320                 size = min(n - done, PAGE_SIZE - offset_max);
 321
 322                 memcpy((void *)(pfn_to << PAGE_SHIFT) + offset_to,
 323                        (void *)(pfn_from << PAGE_SHIFT) + offset_from, size);
 324                 done += size;
 325                 uaddr_from += size;
 326                 uaddr_to += size;
 327         } while (done < n);
 328         spin_unlock(&mm->page_table_lock);
 329         return n - done;
 330 fault:
 331         spin_unlock(&mm->page_table_lock);
 332         if (__handle_fault(mm, uaddr, write_user))
 333                 return n - done;
 334         goto retry;
 335 }
 336
 337 #define __futex_atomic_op(insn, ret, oldval, newval, uaddr, oparg)      \
 338         asm volatile("0: l   %1,0(%6)\n"                                \
 339                      "1: " insn                                         \
 340                      "2: cs  %1,%2,0(%6)\n"                             \
 341                      "3: jl  1b\n"                                      \
 342                      "   lhi %0,0\n"                                    \
 343                      "4:\n"                                             \
 344                      EX_TABLE(0b,4b) EX_TABLE(2b,4b) EX_TABLE(3b,4b)    \
 345                      : "=d" (ret), "=&d" (oldval), "=&d" (newval),      \
 346                        "=m" (*uaddr)                                    \
 347                      : "0" (-EFAULT), "d" (oparg), "a" (uaddr),         \
 348                        "m" (*uaddr) : "cc" );
 349
 350 static int __futex_atomic_op_pt(int op, int __user *uaddr, int oparg, int *old)
 351 {
 352         int oldval = 0, newval, ret;
 353
 354         switch (op) {
 355         case FUTEX_OP_SET:
 356                 __futex_atomic_op("lr %2,%5\n",
 357                                   ret, oldval, newval, uaddr, oparg);
 358                 break;
 359         case FUTEX_OP_ADD:
 360                 __futex_atomic_op("lr %2,%1\nar %2,%5\n",
 361                                   ret, oldval, newval, uaddr, oparg);
 362                 break;
 363         case FUTEX_OP_OR:
 364                 __futex_atomic_op("lr %2,%1\nor %2,%5\n",
 365                                   ret, oldval, newval, uaddr, oparg);
 366                 break;
 367         case FUTEX_OP_ANDN:
 368                 __futex_atomic_op("lr %2,%1\nnr %2,%5\n",
 369                                   ret, oldval, newval, uaddr, oparg);
 370                 break;
 371         case FUTEX_OP_XOR:
 372                 __futex_atomic_op("lr %2,%1\nxr %2,%5\n",
 373                                   ret, oldval, newval, uaddr, oparg);
 374                 break;
 375         default:
 376                 ret = -ENOSYS;
 377         }
 378         if (ret == 0)
 379                 *old = oldval;
 380         return ret;
 381 }
 382
 383 int futex_atomic_op_pt(int op, int __user *uaddr, int oparg, int *old)
 384 {
 385         int ret;
 386
 387         if (segment_eq(get_fs(), KERNEL_DS))
 388                 return __futex_atomic_op_pt(op, uaddr, oparg, old);
 389         spin_lock(&current->mm->page_table_lock);
 390         uaddr = (int __user *) __dat_user_addr((unsigned long) uaddr);
 391         if (!uaddr) {
 392                 spin_unlock(&current->mm->page_table_lock);
 393                 return -EFAULT;
 394         }
 395         get_page(virt_to_page(uaddr));
 396         spin_unlock(&current->mm->page_table_lock);
 397         ret = __futex_atomic_op_pt(op, uaddr, oparg, old);
 398         put_page(virt_to_page(uaddr));
 399         return ret;
 400 }
 401
 402 static int __futex_atomic_cmpxchg_pt(int __user *uaddr, int oldval, int newval)
 403 {
 404         int ret;
 405
 406         asm volatile("0: cs   %1,%4,0(%5)\n"
 407                      "1: lr   %0,%1\n"
 408                      "2:\n"
 409                      EX_TABLE(0b,2b) EX_TABLE(1b,2b)
 410                      : "=d" (ret), "+d" (oldval), "=m" (*uaddr)
 411                      : "0" (-EFAULT), "d" (newval), "a" (uaddr), "m" (*uaddr)
 412                      : "cc", "memory" );
 413         return ret;
 414 }
 415
 416 int futex_atomic_cmpxchg_pt(int __user *uaddr, int oldval, int newval)
 417 {
 418         int ret;
 419
 420         if (segment_eq(get_fs(), KERNEL_DS))
 421                 return __futex_atomic_cmpxchg_pt(uaddr, oldval, newval);
 422         spin_lock(&current->mm->page_table_lock);
 423         uaddr = (int __user *) __dat_user_addr((unsigned long) uaddr);
 424         if (!uaddr) {
 425                 spin_unlock(&current->mm->page_table_lock);
 426                 return -EFAULT;
 427         }
 428         get_page(virt_to_page(uaddr));
 429         spin_unlock(&current->mm->page_table_lock);
 430         ret = __futex_atomic_cmpxchg_pt(uaddr, oldval, newval);
 431         put_page(virt_to_page(uaddr));
 432         return ret;
 433 }
 434
 435 struct uaccess_ops uaccess_pt = {
 436         .copy_from_user         = copy_from_user_pt,
 437         .copy_from_user_small   = copy_from_user_pt,
 438         .copy_to_user           = copy_to_user_pt,
 439         .copy_to_user_small     = copy_to_user_pt,
 440         .copy_in_user           = copy_in_user_pt,
 441         .clear_user             = clear_user_pt,
 442         .strnlen_user           = strnlen_user_pt,
 443         .strncpy_from_user      = strncpy_from_user_pt,
 444         .futex_atomic_op        = futex_atomic_op_pt,
 445         .futex_atomic_cmpxchg   = futex_atomic_cmpxchg_pt,
 446 };