linux-user/mmap.c

   1 /*
   2  *  mmap support for qemu
   3  *
   4  *  Copyright (c) 2003 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, write to the Free Software
  18  *  Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  19  */
  20 #include <stdlib.h>
  21 #include <stdio.h>
  22 #include <stdarg.h>
  23 #include <string.h>
  24 #include <unistd.h>
  25 #include <errno.h>
  26 #include <sys/mman.h>
  27
  28 #include "qemu.h"
  29 #include "qemu-common.h"
  30
  31 //#define DEBUG_MMAP
  32
  33 #if defined(USE_NPTL)
  34 pthread_mutex_t mmap_mutex;
  35 static int __thread mmap_lock_count;
  36
  37 void mmap_lock(void)
  38 {
  39     if (mmap_lock_count++ == 0) {
  40         pthread_mutex_lock(&mmap_mutex);
  41     }
  42 }
  43
  44 void mmap_unlock(void)
  45 {
  46     if (--mmap_lock_count == 0) {
  47         pthread_mutex_unlock(&mmap_mutex);
  48     }
  49 }
  50
  51 /* Grab lock to make sure things are in a consistent state after fork().  */
  52 void mmap_fork_start(void)
  53 {
  54     if (mmap_lock_count)
  55         abort();
  56     pthread_mutex_lock(&mmap_mutex);
  57 }
  58
  59 void mmap_fork_end(int child)
  60 {
  61     if (child)
  62         pthread_mutex_init(&mmap_mutex, NULL);
  63     else
  64         pthread_mutex_unlock(&mmap_mutex);
  65 }
  66 #else
  67 /* We aren't threadsafe to start with, so no need to worry about locking.  */
  68 void mmap_lock(void)
  69 {
  70 }
  71
  72 void mmap_unlock(void)
  73 {
  74 }
  75 #endif
  76
  77 void *qemu_vmalloc(size_t size)
  78 {
  79     void *p;
  80     unsigned long addr;
  81     mmap_lock();
  82     /* Use map and mark the pages as used.  */
  83     p = mmap(NULL, size, PROT_READ | PROT_WRITE,
  84              MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
  85
  86     addr = (unsigned long)p;
  87     if (addr == (target_ulong) addr) {
  88         /* Allocated region overlaps guest address space.
  89            This may recurse.  */
  90         page_set_flags(addr & TARGET_PAGE_MASK, TARGET_PAGE_ALIGN(addr + size),
  91                        PAGE_RESERVED);
  92     }
  93
  94     mmap_unlock();
  95     return p;
  96 }
  97
  98 void *qemu_malloc(size_t size)
  99 {
 100     char * p;
 101     size += 16;
 102     p = qemu_vmalloc(size);
 103     *(size_t *)p = size;
 104     return p + 16;
 105 }
 106
 107 /* We use map, which is always zero initialized.  */
 108 void * qemu_mallocz(size_t size)
 109 {
 110     return qemu_malloc(size);
 111 }
 112
 113 void qemu_free(void *ptr)
 114 {
 115     /* FIXME: We should unmark the reserved pages here.  However this gets
 116        complicated when one target page spans multiple host pages, so we
 117        don't bother.  */
 118     size_t *p;
 119     p = (size_t *)((char *)ptr - 16);
 120     munmap(p, *p);
 121 }
 122
 123 /* NOTE: all the constants are the HOST ones, but addresses are target. */
 124 int target_mprotect(abi_ulong start, abi_ulong len, int prot)
 125 {
 126     abi_ulong end, host_start, host_end, addr;
 127     int prot1, ret;
 128
 129 #ifdef DEBUG_MMAP
 130     printf("mprotect: start=0x" TARGET_FMT_lx
 131            "len=0x" TARGET_FMT_lx " prot=%c%c%c\n", start, len,
 132            prot & PROT_READ ? 'r' : '-',
 133            prot & PROT_WRITE ? 'w' : '-',
 134            prot & PROT_EXEC ? 'x' : '-');
 135 #endif
 136
 137     if ((start & ~TARGET_PAGE_MASK) != 0)
 138         return -EINVAL;
 139     len = TARGET_PAGE_ALIGN(len);
 140     end = start + len;
 141     if (end < start)
 142         return -EINVAL;
 143     prot &= PROT_READ | PROT_WRITE | PROT_EXEC;
 144     if (len == 0)
 145         return 0;
 146
 147     mmap_lock();
 148     host_start = start & qemu_host_page_mask;
 149     host_end = HOST_PAGE_ALIGN(end);
 150     if (start > host_start) {
 151         /* handle host page containing start */
 152         prot1 = prot;
 153         for(addr = host_start; addr < start; addr += TARGET_PAGE_SIZE) {
 154             prot1 |= page_get_flags(addr);
 155         }
 156         if (host_end == host_start + qemu_host_page_size) {
 157             for(addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) {
 158                 prot1 |= page_get_flags(addr);
 159             }
 160             end = host_end;
 161         }
 162         ret = mprotect(g2h(host_start), qemu_host_page_size, prot1 & PAGE_BITS);
 163         if (ret != 0)
 164             goto error;
 165         host_start += qemu_host_page_size;
 166     }
 167     if (end < host_end) {
 168         prot1 = prot;
 169         for(addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) {
 170             prot1 |= page_get_flags(addr);
 171         }
 172         ret = mprotect(g2h(host_end - qemu_host_page_size), qemu_host_page_size,
 173                        prot1 & PAGE_BITS);
 174         if (ret != 0)
 175             goto error;
 176         host_end -= qemu_host_page_size;
 177     }
 178
 179     /* handle the pages in the middle */
 180     if (host_start < host_end) {
 181         ret = mprotect(g2h(host_start), host_end - host_start, prot);
 182         if (ret != 0)
 183             goto error;
 184     }
 185     page_set_flags(start, start + len, prot | PAGE_VALID);
 186     mmap_unlock();
 187     return 0;
 188 error:
 189     mmap_unlock();
 190     return ret;
 191 }
 192
 193 /* map an incomplete host page */
 194 static int mmap_frag(abi_ulong real_start,
 195                      abi_ulong start, abi_ulong end,
 196                      int prot, int flags, int fd, abi_ulong offset)
 197 {
 198     abi_ulong real_end, addr;
 199     void *host_start;
 200     int prot1, prot_new;
 201
 202     real_end = real_start + qemu_host_page_size;
 203     host_start = g2h(real_start);
 204
 205     /* get the protection of the target pages outside the mapping */
 206     prot1 = 0;
 207     for(addr = real_start; addr < real_end; addr++) {
 208         if (addr < start || addr >= end)
 209             prot1 |= page_get_flags(addr);
 210     }
 211
 212     if (prot1 == 0) {
 213         /* no page was there, so we allocate one */
 214         void *p = mmap(host_start, qemu_host_page_size, prot,
 215                        flags | MAP_ANONYMOUS, -1, 0);
 216         if (p == MAP_FAILED)
 217             return -1;
 218         prot1 = prot;
 219     }
 220     prot1 &= PAGE_BITS;
 221
 222     prot_new = prot | prot1;
 223     if (!(flags & MAP_ANONYMOUS)) {
 224         /* msync() won't work here, so we return an error if write is
 225            possible while it is a shared mapping */
 226         if ((flags & MAP_TYPE) == MAP_SHARED &&
 227             (prot & PROT_WRITE))
 228             return -EINVAL;
 229
 230         /* adjust protection to be able to read */
 231         if (!(prot1 & PROT_WRITE))
 232             mprotect(host_start, qemu_host_page_size, prot1 | PROT_WRITE);
 233
 234         /* read the corresponding file data */
 235         pread(fd, g2h(start), end - start, offset);
 236
 237         /* put final protection */
 238         if (prot_new != (prot1 | PROT_WRITE))
 239             mprotect(host_start, qemu_host_page_size, prot_new);
 240     } else {
 241         /* just update the protection */
 242         if (prot_new != prot1) {
 243             mprotect(host_start, qemu_host_page_size, prot_new);
 244         }
 245     }
 246     return 0;
 247 }
 248
 249 #if defined(__CYGWIN__)
 250 /* Cygwin doesn't have a whole lot of address space.  */
 251 static abi_ulong mmap_next_start = 0x18000000;
 252 #else
 253 static abi_ulong mmap_next_start = 0x40000000;
 254 #endif
 255
 256 unsigned long last_brk;
 257
 258 /* find a free memory area of size 'size'. The search starts at
 259    'start'. If 'start' == 0, then a default start address is used.
 260    Return -1 if error.
 261 */
 262 /* page_init() marks pages used by the host as reserved to be sure not
 263    to use them. */
 264 static abi_ulong mmap_find_vma(abi_ulong start, abi_ulong size)
 265 {
 266     abi_ulong addr, addr1, addr_start;
 267     int prot;
 268     unsigned long new_brk;
 269
 270     new_brk = (unsigned long)sbrk(0);
 271     if (last_brk && last_brk < new_brk && last_brk == (target_ulong)last_brk) {
 272         /* This is a hack to catch the host allocating memory with brk().
 273            If it uses mmap then we loose.
 274            FIXME: We really want to avoid the host allocating memory in
 275            the first place, and maybe leave some slack to avoid switching
 276            to mmap.  */
 277         page_set_flags(last_brk & TARGET_PAGE_MASK,
 278                        TARGET_PAGE_ALIGN(new_brk),
 279                        PAGE_RESERVED);
 280     }
 281     last_brk = new_brk;
 282
 283     size = HOST_PAGE_ALIGN(size);
 284     start = start & qemu_host_page_mask;
 285     addr = start;
 286     if (addr == 0)
 287         addr = mmap_next_start;
 288     addr_start = addr;
 289     for(;;) {
 290         prot = 0;
 291         for(addr1 = addr; addr1 < (addr + size); addr1 += TARGET_PAGE_SIZE) {
 292             prot |= page_get_flags(addr1);
 293         }
 294         if (prot == 0)
 295             break;
 296         addr += qemu_host_page_size;
 297         /* we found nothing */
 298         if (addr == addr_start)
 299             return (abi_ulong)-1;
 300     }
 301     if (start == 0)
 302         mmap_next_start = addr + size;
 303     return addr;
 304 }
 305
 306 /* NOTE: all the constants are the HOST ones */
 307 abi_long target_mmap(abi_ulong start, abi_ulong len, int prot,
 308                      int flags, int fd, abi_ulong offset)
 309 {
 310     abi_ulong ret, end, real_start, real_end, retaddr, host_offset, host_len;
 311     unsigned long host_start;
 312
 313     mmap_lock();
 314 #ifdef DEBUG_MMAP
 315     {
 316         printf("mmap: start=0x" TARGET_FMT_lx
 317                " len=0x" TARGET_FMT_lx " prot=%c%c%c flags=",
 318                start, len,
 319                prot & PROT_READ ? 'r' : '-',
 320                prot & PROT_WRITE ? 'w' : '-',
 321                prot & PROT_EXEC ? 'x' : '-');
 322         if (flags & MAP_FIXED)
 323             printf("MAP_FIXED ");
 324         if (flags & MAP_ANONYMOUS)
 325             printf("MAP_ANON ");
 326         switch(flags & MAP_TYPE) {
 327         case MAP_PRIVATE:
 328             printf("MAP_PRIVATE ");
 329             break;
 330         case MAP_SHARED:
 331             printf("MAP_SHARED ");
 332             break;
 333         default:
 334             printf("[MAP_TYPE=0x%x] ", flags & MAP_TYPE);
 335             break;
 336         }
 337         printf("fd=%d offset=" TARGET_FMT_lx "\n", fd, offset);
 338     }
 339 #endif
 340
 341     if (offset & ~TARGET_PAGE_MASK) {
 342         errno = EINVAL;
 343         goto fail;
 344     }
 345
 346     len = TARGET_PAGE_ALIGN(len);
 347     if (len == 0)
 348         goto the_end;
 349     real_start = start & qemu_host_page_mask;
 350
 351     if (!(flags & MAP_FIXED)) {
 352         abi_ulong mmap_start;
 353         void *p;
 354         host_offset = offset & qemu_host_page_mask;
 355         host_len = len + offset - host_offset;
 356         host_len = HOST_PAGE_ALIGN(host_len);
 357         mmap_start = mmap_find_vma(real_start, host_len);
 358         if (mmap_start == (abi_ulong)-1) {
 359             errno = ENOMEM;
 360             goto fail;
 361         }
 362         /* Note: we prefer to control the mapping address. It is
 363            especially important if qemu_host_page_size >
 364            qemu_real_host_page_size */
 365         p = mmap(g2h(mmap_start),
 366                  host_len, prot, flags | MAP_FIXED, fd, host_offset);
 367         if (p == MAP_FAILED)
 368             goto fail;
 369         /* update start so that it points to the file position at 'offset' */
 370         host_start = (unsigned long)p;
 371         if (!(flags & MAP_ANONYMOUS))
 372             host_start += offset - host_offset;
 373         start = h2g(host_start);
 374     } else {
 375         int flg;
 376         target_ulong addr;
 377
 378         if (start & ~TARGET_PAGE_MASK) {
 379             errno = EINVAL;
 380             goto fail;
 381         }
 382         end = start + len;
 383         real_end = HOST_PAGE_ALIGN(end);
 384
 385         for(addr = real_start; addr < real_end; addr += TARGET_PAGE_SIZE) {
 386             flg = page_get_flags(addr);
 387             if (flg & PAGE_RESERVED) {
 388                 errno = ENXIO;
 389                 goto fail;
 390             }
 391         }
 392
 393         /* worst case: we cannot map the file because the offset is not
 394            aligned, so we read it */
 395         if (!(flags & MAP_ANONYMOUS) &&
 396             (offset & ~qemu_host_page_mask) != (start & ~qemu_host_page_mask)) {
 397             /* msync() won't work here, so we return an error if write is
 398                possible while it is a shared mapping */
 399             if ((flags & MAP_TYPE) == MAP_SHARED &&
 400                 (prot & PROT_WRITE)) {
 401                 errno = EINVAL;
 402                 goto fail;
 403             }
 404             retaddr = target_mmap(start, len, prot | PROT_WRITE,
 405                                   MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS,
 406                                   -1, 0);
 407             if (retaddr == -1)
 408                 goto fail;
 409             pread(fd, g2h(start), len, offset);
 410             if (!(prot & PROT_WRITE)) {
 411                 ret = target_mprotect(start, len, prot);
 412                 if (ret != 0) {
 413                     start = ret;
 414                     goto the_end;
 415                 }
 416             }
 417             goto the_end;
 418         }
 419
 420         /* handle the start of the mapping */
 421         if (start > real_start) {
 422             if (real_end == real_start + qemu_host_page_size) {
 423                 /* one single host page */
 424                 ret = mmap_frag(real_start, start, end,
 425                                 prot, flags, fd, offset);
 426                 if (ret == -1)
 427                     goto fail;
 428                 goto the_end1;
 429             }
 430             ret = mmap_frag(real_start, start, real_start + qemu_host_page_size,
 431                             prot, flags, fd, offset);
 432             if (ret == -1)
 433                 goto fail;
 434             real_start += qemu_host_page_size;
 435         }
 436         /* handle the end of the mapping */
 437         if (end < real_end) {
 438             ret = mmap_frag(real_end - qemu_host_page_size,
 439                             real_end - qemu_host_page_size, real_end,
 440                             prot, flags, fd,
 441                             offset + real_end - qemu_host_page_size - start);
 442             if (ret == -1)
 443                 goto fail;
 444             real_end -= qemu_host_page_size;
 445         }
 446
 447         /* map the middle (easier) */
 448         if (real_start < real_end) {
 449             void *p;
 450             unsigned long offset1;
 451             if (flags & MAP_ANONYMOUS)
 452                 offset1 = 0;
 453             else
 454                 offset1 = offset + real_start - start;
 455             p = mmap(g2h(real_start), real_end - real_start,
 456                      prot, flags, fd, offset1);
 457             if (p == MAP_FAILED)
 458                 goto fail;
 459         }
 460     }
 461  the_end1:
 462     page_set_flags(start, start + len, prot | PAGE_VALID);
 463  the_end:
 464 #ifdef DEBUG_MMAP
 465     printf("ret=0x" TARGET_FMT_lx "\n", start);
 466     page_dump(stdout);
 467     printf("\n");
 468 #endif
 469     mmap_unlock();
 470     return start;
 471 fail:
 472     mmap_unlock();
 473     return -1;
 474 }
 475
 476 int target_munmap(abi_ulong start, abi_ulong len)
 477 {
 478     abi_ulong end, real_start, real_end, addr;
 479     int prot, ret;
 480
 481 #ifdef DEBUG_MMAP
 482     printf("munmap: start=0x%lx len=0x%lx\n", start, len);
 483 #endif
 484     if (start & ~TARGET_PAGE_MASK)
 485         return -EINVAL;
 486     len = TARGET_PAGE_ALIGN(len);
 487     if (len == 0)
 488         return -EINVAL;
 489     mmap_lock();
 490     end = start + len;
 491     real_start = start & qemu_host_page_mask;
 492     real_end = HOST_PAGE_ALIGN(end);
 493
 494     if (start > real_start) {
 495         /* handle host page containing start */
 496         prot = 0;
 497         for(addr = real_start; addr < start; addr += TARGET_PAGE_SIZE) {
 498             prot |= page_get_flags(addr);
 499         }
 500         if (real_end == real_start + qemu_host_page_size) {
 501             for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
 502                 prot |= page_get_flags(addr);
 503             }
 504             end = real_end;
 505         }
 506         if (prot != 0)
 507             real_start += qemu_host_page_size;
 508     }
 509     if (end < real_end) {
 510         prot = 0;
 511         for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
 512             prot |= page_get_flags(addr);
 513         }
 514         if (prot != 0)
 515             real_end -= qemu_host_page_size;
 516     }
 517
 518     ret = 0;
 519     /* unmap what we can */
 520     if (real_start < real_end) {
 521         ret = munmap(g2h(real_start), real_end - real_start);
 522     }
 523
 524     if (ret == 0)
 525         page_set_flags(start, start + len, 0);
 526     mmap_unlock();
 527     return ret;
 528 }
 529
 530 /* XXX: currently, we only handle MAP_ANONYMOUS and not MAP_FIXED
 531    blocks which have been allocated starting on a host page */
 532 abi_long target_mremap(abi_ulong old_addr, abi_ulong old_size,
 533                        abi_ulong new_size, unsigned long flags,
 534                        abi_ulong new_addr)
 535 {
 536     int prot;
 537     unsigned long host_addr;
 538
 539     mmap_lock();
 540     /* XXX: use 5 args syscall */
 541     host_addr = (long)mremap(g2h(old_addr), old_size, new_size, flags);
 542     if (host_addr == -1) {
 543         new_addr = -1;
 544     } else {
 545         new_addr = h2g(host_addr);
 546         prot = page_get_flags(old_addr);
 547         page_set_flags(old_addr, old_addr + old_size, 0);
 548         page_set_flags(new_addr, new_addr + new_size, prot | PAGE_VALID);
 549     }
 550     mmap_unlock();
 551     return new_addr;
 552 }
 553
 554 int target_msync(abi_ulong start, abi_ulong len, int flags)
 555 {
 556     abi_ulong end;
 557
 558     if (start & ~TARGET_PAGE_MASK)
 559         return -EINVAL;
 560     len = TARGET_PAGE_ALIGN(len);
 561     end = start + len;
 562     if (end < start)
 563         return -EINVAL;
 564     if (end == start)
 565         return 0;
 566
 567     start &= qemu_host_page_mask;
 568     return msync(g2h(start), end - start, flags);
 569 }