bsd-user/mmap.c

   1 /*
   2  *  mmap support for qemu
   3  *
   4  *  Copyright (c) 2003 - 2008 Fabrice Bellard
   5  *
   6  *  This program is free software; you can redistribute it and/or modify
   7  *  it under the terms of the GNU General Public License as published by
   8  *  the Free Software Foundation; either version 2 of the License, or
   9  *  (at your option) any later version.
  10  *
  11  *  This program is distributed in the hope that it will be useful,
  12  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  *  GNU General Public License for more details.
  15  *
  16  *  You should have received a copy of the GNU General Public License
  17  *  along with this program; if not, see <http://www.gnu.org/licenses/>.
  18  */
  19 #include "qemu/osdep.h"
  20
  21 #include "qemu.h"
  22 #include "qemu-common.h"
  23 #include "bsd-mman.h"
  24 #include "exec/exec-all.h"
  25
  26 //#define DEBUG_MMAP
  27
  28 static pthread_mutex_t mmap_mutex = PTHREAD_MUTEX_INITIALIZER;
  29 static __thread int mmap_lock_count;
  30
  31 void mmap_lock(void)
  32 {
  33     if (mmap_lock_count++ == 0) {
  34         pthread_mutex_lock(&mmap_mutex);
  35     }
  36 }
  37
  38 void mmap_unlock(void)
  39 {
  40     if (--mmap_lock_count == 0) {
  41         pthread_mutex_unlock(&mmap_mutex);
  42     }
  43 }
  44
  45 bool have_mmap_lock(void)
  46 {
  47     return mmap_lock_count > 0 ? true : false;
  48 }
  49
  50 /* Grab lock to make sure things are in a consistent state after fork().  */
  51 void mmap_fork_start(void)
  52 {
  53     if (mmap_lock_count)
  54         abort();
  55     pthread_mutex_lock(&mmap_mutex);
  56 }
  57
  58 void mmap_fork_end(int child)
  59 {
  60     if (child)
  61         pthread_mutex_init(&mmap_mutex, NULL);
  62     else
  63         pthread_mutex_unlock(&mmap_mutex);
  64 }
  65
  66 /* NOTE: all the constants are the HOST ones, but addresses are target. */
  67 int target_mprotect(abi_ulong start, abi_ulong len, int prot)
  68 {
  69     abi_ulong end, host_start, host_end, addr;
  70     int prot1, ret;
  71
  72 #ifdef DEBUG_MMAP
  73     printf("mprotect: start=0x" TARGET_FMT_lx
  74            " len=0x" TARGET_FMT_lx " prot=%c%c%c\n", start, len,
  75            prot & PROT_READ ? 'r' : '-',
  76            prot & PROT_WRITE ? 'w' : '-',
  77            prot & PROT_EXEC ? 'x' : '-');
  78 #endif
  79
  80     if ((start & ~TARGET_PAGE_MASK) != 0)
  81         return -EINVAL;
  82     len = TARGET_PAGE_ALIGN(len);
  83     end = start + len;
  84     if (end < start)
  85         return -EINVAL;
  86     prot &= PROT_READ | PROT_WRITE | PROT_EXEC;
  87     if (len == 0)
  88         return 0;
  89
  90     mmap_lock();
  91     host_start = start & qemu_host_page_mask;
  92     host_end = HOST_PAGE_ALIGN(end);
  93     if (start > host_start) {
  94         /* handle host page containing start */
  95         prot1 = prot;
  96         for (addr = host_start; addr < start; addr += TARGET_PAGE_SIZE) {
  97             prot1 |= page_get_flags(addr);
  98         }
  99         if (host_end == host_start + qemu_host_page_size) {
 100             for (addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) {
 101                 prot1 |= page_get_flags(addr);
 102             }
 103             end = host_end;
 104         }
 105         ret = mprotect(g2h_untagged(host_start),
 106                        qemu_host_page_size, prot1 & PAGE_BITS);
 107         if (ret != 0)
 108             goto error;
 109         host_start += qemu_host_page_size;
 110     }
 111     if (end < host_end) {
 112         prot1 = prot;
 113         for (addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) {
 114             prot1 |= page_get_flags(addr);
 115         }
 116         ret = mprotect(g2h_untagged(host_end - qemu_host_page_size),
 117                        qemu_host_page_size, prot1 & PAGE_BITS);
 118         if (ret != 0)
 119             goto error;
 120         host_end -= qemu_host_page_size;
 121     }
 122
 123     /* handle the pages in the middle */
 124     if (host_start < host_end) {
 125         ret = mprotect(g2h_untagged(host_start), host_end - host_start, prot);
 126         if (ret != 0)
 127             goto error;
 128     }
 129     page_set_flags(start, start + len, prot | PAGE_VALID);
 130     mmap_unlock();
 131     return 0;
 132 error:
 133     mmap_unlock();
 134     return ret;
 135 }
 136
 137 /* map an incomplete host page */
 138 static int mmap_frag(abi_ulong real_start,
 139                      abi_ulong start, abi_ulong end,
 140                      int prot, int flags, int fd, abi_ulong offset)
 141 {
 142     abi_ulong real_end, addr;
 143     void *host_start;
 144     int prot1, prot_new;
 145
 146     real_end = real_start + qemu_host_page_size;
 147     host_start = g2h_untagged(real_start);
 148
 149     /* get the protection of the target pages outside the mapping */
 150     prot1 = 0;
 151     for (addr = real_start; addr < real_end; addr++) {
 152         if (addr < start || addr >= end)
 153             prot1 |= page_get_flags(addr);
 154     }
 155
 156     if (prot1 == 0) {
 157         /* no page was there, so we allocate one */
 158         void *p = mmap(host_start, qemu_host_page_size, prot,
 159                        flags | MAP_ANON, -1, 0);
 160         if (p == MAP_FAILED)
 161             return -1;
 162         prot1 = prot;
 163     }
 164     prot1 &= PAGE_BITS;
 165
 166     prot_new = prot | prot1;
 167     if (!(flags & MAP_ANON)) {
 168         /* msync() won't work here, so we return an error if write is
 169            possible while it is a shared mapping */
 170         if ((flags & TARGET_BSD_MAP_FLAGMASK) == MAP_SHARED &&
 171             (prot & PROT_WRITE))
 172             return -1;
 173
 174         /* adjust protection to be able to read */
 175         if (!(prot1 & PROT_WRITE))
 176             mprotect(host_start, qemu_host_page_size, prot1 | PROT_WRITE);
 177
 178         /* read the corresponding file data */
 179         pread(fd, g2h_untagged(start), end - start, offset);
 180
 181         /* put final protection */
 182         if (prot_new != (prot1 | PROT_WRITE))
 183             mprotect(host_start, qemu_host_page_size, prot_new);
 184     } else {
 185         /* just update the protection */
 186         if (prot_new != prot1) {
 187             mprotect(host_start, qemu_host_page_size, prot_new);
 188         }
 189     }
 190     return 0;
 191 }
 192
 193 static abi_ulong mmap_next_start = 0x40000000;
 194
 195 unsigned long last_brk;
 196
 197 /* find a free memory area of size 'size'. The search starts at
 198    'start'. If 'start' == 0, then a default start address is used.
 199    Return -1 if error.
 200 */
 201 /* page_init() marks pages used by the host as reserved to be sure not
 202    to use them. */
 203 static abi_ulong mmap_find_vma(abi_ulong start, abi_ulong size)
 204 {
 205     abi_ulong addr, addr1, addr_start;
 206     int prot;
 207     unsigned long new_brk;
 208
 209     new_brk = (unsigned long)sbrk(0);
 210     if (last_brk && last_brk < new_brk && last_brk == (target_ulong)last_brk) {
 211         /* This is a hack to catch the host allocating memory with brk().
 212            If it uses mmap then we loose.
 213            FIXME: We really want to avoid the host allocating memory in
 214            the first place, and maybe leave some slack to avoid switching
 215            to mmap.  */
 216         page_set_flags(last_brk & TARGET_PAGE_MASK,
 217                        TARGET_PAGE_ALIGN(new_brk),
 218                        PAGE_RESERVED);
 219     }
 220     last_brk = new_brk;
 221
 222     size = HOST_PAGE_ALIGN(size);
 223     start = start & qemu_host_page_mask;
 224     addr = start;
 225     if (addr == 0)
 226         addr = mmap_next_start;
 227     addr_start = addr;
 228     for (;;) {
 229         prot = 0;
 230         for (addr1 = addr; addr1 < (addr + size); addr1 += TARGET_PAGE_SIZE) {
 231             prot |= page_get_flags(addr1);
 232         }
 233         if (prot == 0)
 234             break;
 235         addr += qemu_host_page_size;
 236         /* we found nothing */
 237         if (addr == addr_start)
 238             return (abi_ulong)-1;
 239     }
 240     if (start == 0)
 241         mmap_next_start = addr + size;
 242     return addr;
 243 }
 244
 245 /* NOTE: all the constants are the HOST ones */
 246 abi_long target_mmap(abi_ulong start, abi_ulong len, int prot,
 247                      int flags, int fd, abi_ulong offset)
 248 {
 249     abi_ulong ret, end, real_start, real_end, retaddr, host_offset, host_len;
 250     unsigned long host_start;
 251
 252     mmap_lock();
 253 #ifdef DEBUG_MMAP
 254     {
 255         printf("mmap: start=0x" TARGET_FMT_lx
 256                " len=0x" TARGET_FMT_lx " prot=%c%c%c flags=",
 257                start, len,
 258                prot & PROT_READ ? 'r' : '-',
 259                prot & PROT_WRITE ? 'w' : '-',
 260                prot & PROT_EXEC ? 'x' : '-');
 261         if (flags & MAP_FIXED)
 262             printf("MAP_FIXED ");
 263         if (flags & MAP_ANON)
 264             printf("MAP_ANON ");
 265         switch (flags & TARGET_BSD_MAP_FLAGMASK) {
 266         case MAP_PRIVATE:
 267             printf("MAP_PRIVATE ");
 268             break;
 269         case MAP_SHARED:
 270             printf("MAP_SHARED ");
 271             break;
 272         default:
 273             printf("[MAP_FLAGMASK=0x%x] ", flags & TARGET_BSD_MAP_FLAGMASK);
 274             break;
 275         }
 276         printf("fd=%d offset=" TARGET_FMT_lx "\n", fd, offset);
 277     }
 278 #endif
 279
 280     if (offset & ~TARGET_PAGE_MASK) {
 281         errno = EINVAL;
 282         goto fail;
 283     }
 284
 285     len = TARGET_PAGE_ALIGN(len);
 286     if (len == 0)
 287         goto the_end;
 288     real_start = start & qemu_host_page_mask;
 289
 290     if (!(flags & MAP_FIXED)) {
 291         abi_ulong mmap_start;
 292         void *p;
 293         host_offset = offset & qemu_host_page_mask;
 294         host_len = len + offset - host_offset;
 295         host_len = HOST_PAGE_ALIGN(host_len);
 296         mmap_start = mmap_find_vma(real_start, host_len);
 297         if (mmap_start == (abi_ulong)-1) {
 298             errno = ENOMEM;
 299             goto fail;
 300         }
 301         /* Note: we prefer to control the mapping address. It is
 302            especially important if qemu_host_page_size >
 303            qemu_real_host_page_size */
 304         p = mmap(g2h_untagged(mmap_start),
 305                  host_len, prot, flags | MAP_FIXED, fd, host_offset);
 306         if (p == MAP_FAILED)
 307             goto fail;
 308         /* update start so that it points to the file position at 'offset' */
 309         host_start = (unsigned long)p;
 310         if (!(flags & MAP_ANON))
 311             host_start += offset - host_offset;
 312         start = h2g(host_start);
 313     } else {
 314         int flg;
 315         target_ulong addr;
 316
 317         if (start & ~TARGET_PAGE_MASK) {
 318             errno = EINVAL;
 319             goto fail;
 320         }
 321         end = start + len;
 322         real_end = HOST_PAGE_ALIGN(end);
 323
 324         for (addr = real_start; addr < real_end; addr += TARGET_PAGE_SIZE) {
 325             flg = page_get_flags(addr);
 326             if (flg & PAGE_RESERVED) {
 327                 errno = ENXIO;
 328                 goto fail;
 329             }
 330         }
 331
 332         /* worst case: we cannot map the file because the offset is not
 333            aligned, so we read it */
 334         if (!(flags & MAP_ANON) &&
 335             (offset & ~qemu_host_page_mask) != (start & ~qemu_host_page_mask)) {
 336             /* msync() won't work here, so we return an error if write is
 337                possible while it is a shared mapping */
 338             if ((flags & TARGET_BSD_MAP_FLAGMASK) == MAP_SHARED &&
 339                 (prot & PROT_WRITE)) {
 340                 errno = EINVAL;
 341                 goto fail;
 342             }
 343             retaddr = target_mmap(start, len, prot | PROT_WRITE,
 344                                   MAP_FIXED | MAP_PRIVATE | MAP_ANON,
 345                                   -1, 0);
 346             if (retaddr == -1)
 347                 goto fail;
 348             pread(fd, g2h_untagged(start), len, offset);
 349             if (!(prot & PROT_WRITE)) {
 350                 ret = target_mprotect(start, len, prot);
 351                 if (ret != 0) {
 352                     start = ret;
 353                     goto the_end;
 354                 }
 355             }
 356             goto the_end;
 357         }
 358
 359         /* handle the start of the mapping */
 360         if (start > real_start) {
 361             if (real_end == real_start + qemu_host_page_size) {
 362                 /* one single host page */
 363                 ret = mmap_frag(real_start, start, end,
 364                                 prot, flags, fd, offset);
 365                 if (ret == -1)
 366                     goto fail;
 367                 goto the_end1;
 368             }
 369             ret = mmap_frag(real_start, start, real_start + qemu_host_page_size,
 370                             prot, flags, fd, offset);
 371             if (ret == -1)
 372                 goto fail;
 373             real_start += qemu_host_page_size;
 374         }
 375         /* handle the end of the mapping */
 376         if (end < real_end) {
 377             ret = mmap_frag(real_end - qemu_host_page_size,
 378                             real_end - qemu_host_page_size, real_end,
 379                             prot, flags, fd,
 380                             offset + real_end - qemu_host_page_size - start);
 381             if (ret == -1)
 382                 goto fail;
 383             real_end -= qemu_host_page_size;
 384         }
 385
 386         /* map the middle (easier) */
 387         if (real_start < real_end) {
 388             void *p;
 389             unsigned long offset1;
 390             if (flags & MAP_ANON)
 391                 offset1 = 0;
 392             else
 393                 offset1 = offset + real_start - start;
 394             p = mmap(g2h_untagged(real_start), real_end - real_start,
 395                      prot, flags, fd, offset1);
 396             if (p == MAP_FAILED)
 397                 goto fail;
 398         }
 399     }
 400  the_end1:
 401     page_set_flags(start, start + len, prot | PAGE_VALID);
 402  the_end:
 403 #ifdef DEBUG_MMAP
 404     printf("ret=0x" TARGET_FMT_lx "\n", start);
 405     page_dump(stdout);
 406     printf("\n");
 407 #endif
 408     mmap_unlock();
 409     return start;
 410 fail:
 411     mmap_unlock();
 412     return -1;
 413 }
 414
 415 int target_munmap(abi_ulong start, abi_ulong len)
 416 {
 417     abi_ulong end, real_start, real_end, addr;
 418     int prot, ret;
 419
 420 #ifdef DEBUG_MMAP
 421     printf("munmap: start=0x%lx len=0x%lx\n", start, len);
 422 #endif
 423     if (start & ~TARGET_PAGE_MASK)
 424         return -EINVAL;
 425     len = TARGET_PAGE_ALIGN(len);
 426     if (len == 0)
 427         return -EINVAL;
 428     mmap_lock();
 429     end = start + len;
 430     real_start = start & qemu_host_page_mask;
 431     real_end = HOST_PAGE_ALIGN(end);
 432
 433     if (start > real_start) {
 434         /* handle host page containing start */
 435         prot = 0;
 436         for (addr = real_start; addr < start; addr += TARGET_PAGE_SIZE) {
 437             prot |= page_get_flags(addr);
 438         }
 439         if (real_end == real_start + qemu_host_page_size) {
 440             for (addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
 441                 prot |= page_get_flags(addr);
 442             }
 443             end = real_end;
 444         }
 445         if (prot != 0)
 446             real_start += qemu_host_page_size;
 447     }
 448     if (end < real_end) {
 449         prot = 0;
 450         for (addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
 451             prot |= page_get_flags(addr);
 452         }
 453         if (prot != 0)
 454             real_end -= qemu_host_page_size;
 455     }
 456
 457     ret = 0;
 458     /* unmap what we can */
 459     if (real_start < real_end) {
 460         ret = munmap(g2h_untagged(real_start), real_end - real_start);
 461     }
 462
 463     if (ret == 0)
 464         page_set_flags(start, start + len, 0);
 465     mmap_unlock();
 466     return ret;
 467 }
 468
 469 int target_msync(abi_ulong start, abi_ulong len, int flags)
 470 {
 471     abi_ulong end;
 472
 473     if (start & ~TARGET_PAGE_MASK)
 474         return -EINVAL;
 475     len = TARGET_PAGE_ALIGN(len);
 476     end = start + len;
 477     if (end < start)
 478         return -EINVAL;
 479     if (end == start)
 480         return 0;
 481
 482     start &= qemu_host_page_mask;
 483     return msync(g2h_untagged(start), end - start, flags);
 484 }