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