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., 51 Franklin Street - Fifth Floor, Boston,
  19  *  MA 02110-1301, USA.
  20  */
  21 #include <stdlib.h>
  22 #include <stdio.h>
  23 #include <stdarg.h>
  24 #include <string.h>
  25 #include <unistd.h>
  26 #include <errno.h>
  27 #include <sys/mman.h>
  28 #include <linux/mman.h>
  29 #include <linux/unistd.h>
  30
  31 #include "qemu.h"
  32 #include "qemu-common.h"
  33
  34 //#define DEBUG_MMAP
  35
  36 #if defined(USE_NPTL)
  37 pthread_mutex_t mmap_mutex;
  38 static int __thread mmap_lock_count;
  39
  40 void mmap_lock(void)
  41 {
  42     if (mmap_lock_count++ == 0) {
  43         pthread_mutex_lock(&mmap_mutex);
  44     }
  45 }
  46
  47 void mmap_unlock(void)
  48 {
  49     if (--mmap_lock_count == 0) {
  50         pthread_mutex_unlock(&mmap_mutex);
  51     }
  52 }
  53
  54 /* Grab lock to make sure things are in a consistent state after fork().  */
  55 void mmap_fork_start(void)
  56 {
  57     if (mmap_lock_count)
  58         abort();
  59     pthread_mutex_lock(&mmap_mutex);
  60 }
  61
  62 void mmap_fork_end(int child)
  63 {
  64     if (child)
  65         pthread_mutex_init(&mmap_mutex, NULL);
  66     else
  67         pthread_mutex_unlock(&mmap_mutex);
  68 }
  69 #else
  70 /* We aren't threadsafe to start with, so no need to worry about locking.  */
  71 void mmap_lock(void)
  72 {
  73 }
  74
  75 void mmap_unlock(void)
  76 {
  77 }
  78 #endif
  79
  80 void *qemu_vmalloc(size_t size)
  81 {
  82     void *p;
  83     unsigned long addr;
  84     mmap_lock();
  85     /* Use map and mark the pages as used.  */
  86     p = mmap(NULL, size, PROT_READ | PROT_WRITE,
  87              MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
  88
  89     addr = (unsigned long)p;
  90     if (addr == (target_ulong) addr) {
  91         /* Allocated region overlaps guest address space.
  92            This may recurse.  */
  93         page_set_flags(addr & TARGET_PAGE_MASK, TARGET_PAGE_ALIGN(addr + size),
  94                        PAGE_RESERVED);
  95     }
  96
  97     mmap_unlock();
  98     return p;
  99 }
 100
 101 void *qemu_malloc(size_t size)
 102 {
 103     char * p;
 104     size += 16;
 105     p = qemu_vmalloc(size);
 106     *(size_t *)p = size;
 107     return p + 16;
 108 }
 109
 110 /* We use map, which is always zero initialized.  */
 111 void * qemu_mallocz(size_t size)
 112 {
 113     return qemu_malloc(size);
 114 }
 115
 116 void qemu_free(void *ptr)
 117 {
 118     /* FIXME: We should unmark the reserved pages here.  However this gets
 119        complicated when one target page spans multiple host pages, so we
 120        don't bother.  */
 121     size_t *p;
 122     p = (size_t *)((char *)ptr - 16);
 123     munmap(p, *p);
 124 }
 125
 126 /* NOTE: all the constants are the HOST ones, but addresses are target. */
 127 int target_mprotect(abi_ulong start, abi_ulong len, int prot)
 128 {
 129     abi_ulong end, host_start, host_end, addr;
 130     int prot1, ret;
 131
 132 #ifdef DEBUG_MMAP
 133     printf("mprotect: start=0x" TARGET_FMT_lx
 134            "len=0x" TARGET_FMT_lx " prot=%c%c%c\n", start, len,
 135            prot & PROT_READ ? 'r' : '-',
 136            prot & PROT_WRITE ? 'w' : '-',
 137            prot & PROT_EXEC ? 'x' : '-');
 138 #endif
 139
 140     if ((start & ~TARGET_PAGE_MASK) != 0)
 141         return -EINVAL;
 142     len = TARGET_PAGE_ALIGN(len);
 143     end = start + len;
 144     if (end < start)
 145         return -EINVAL;
 146     prot &= PROT_READ | PROT_WRITE | PROT_EXEC;
 147     if (len == 0)
 148         return 0;
 149
 150     mmap_lock();
 151     host_start = start & qemu_host_page_mask;
 152     host_end = HOST_PAGE_ALIGN(end);
 153     if (start > host_start) {
 154         /* handle host page containing start */
 155         prot1 = prot;
 156         for(addr = host_start; addr < start; addr += TARGET_PAGE_SIZE) {
 157             prot1 |= page_get_flags(addr);
 158         }
 159         if (host_end == host_start + qemu_host_page_size) {
 160             for(addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) {
 161                 prot1 |= page_get_flags(addr);
 162             }
 163             end = host_end;
 164         }
 165         ret = mprotect(g2h(host_start), qemu_host_page_size, prot1 & PAGE_BITS);
 166         if (ret != 0)
 167             goto error;
 168         host_start += qemu_host_page_size;
 169     }
 170     if (end < host_end) {
 171         prot1 = prot;
 172         for(addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) {
 173             prot1 |= page_get_flags(addr);
 174         }
 175         ret = mprotect(g2h(host_end - qemu_host_page_size), qemu_host_page_size,
 176                        prot1 & PAGE_BITS);
 177         if (ret != 0)
 178             goto error;
 179         host_end -= qemu_host_page_size;
 180     }
 181
 182     /* handle the pages in the middle */
 183     if (host_start < host_end) {
 184         ret = mprotect(g2h(host_start), host_end - host_start, prot);
 185         if (ret != 0)
 186             goto error;
 187     }
 188     page_set_flags(start, start + len, prot | PAGE_VALID);
 189     mmap_unlock();
 190     return 0;
 191 error:
 192     mmap_unlock();
 193     return ret;
 194 }
 195
 196 /* map an incomplete host page */
 197 static int mmap_frag(abi_ulong real_start,
 198                      abi_ulong start, abi_ulong end,
 199                      int prot, int flags, int fd, abi_ulong offset)
 200 {
 201     abi_ulong real_end, addr;
 202     void *host_start;
 203     int prot1, prot_new;
 204
 205     real_end = real_start + qemu_host_page_size;
 206     host_start = g2h(real_start);
 207
 208     /* get the protection of the target pages outside the mapping */
 209     prot1 = 0;
 210     for(addr = real_start; addr < real_end; addr++) {
 211         if (addr < start || addr >= end)
 212             prot1 |= page_get_flags(addr);
 213     }
 214
 215     if (prot1 == 0) {
 216         /* no page was there, so we allocate one */
 217         void *p = mmap(host_start, qemu_host_page_size, prot,
 218                        flags | MAP_ANONYMOUS, -1, 0);
 219         if (p == MAP_FAILED)
 220             return -1;
 221         prot1 = prot;
 222     }
 223     prot1 &= PAGE_BITS;
 224
 225     prot_new = prot | prot1;
 226     if (!(flags & MAP_ANONYMOUS)) {
 227         /* msync() won't work here, so we return an error if write is
 228            possible while it is a shared mapping */
 229         if ((flags & MAP_TYPE) == MAP_SHARED &&
 230             (prot & PROT_WRITE))
 231             return -EINVAL;
 232
 233         /* adjust protection to be able to read */
 234         if (!(prot1 & PROT_WRITE))
 235             mprotect(host_start, qemu_host_page_size, prot1 | PROT_WRITE);
 236
 237         /* read the corresponding file data */
 238         pread(fd, g2h(start), end - start, offset);
 239
 240         /* put final protection */
 241         if (prot_new != (prot1 | PROT_WRITE))
 242             mprotect(host_start, qemu_host_page_size, prot_new);
 243     } else {
 244         /* just update the protection */
 245         if (prot_new != prot1) {
 246             mprotect(host_start, qemu_host_page_size, prot_new);
 247         }
 248     }
 249     return 0;
 250 }
 251
 252 #if defined(__CYGWIN__)
 253 /* Cygwin doesn't have a whole lot of address space.  */
 254 static abi_ulong mmap_next_start = 0x18000000;
 255 #else
 256 static abi_ulong mmap_next_start = 0x40000000;
 257 #endif
 258
 259 unsigned long last_brk;
 260
 261 /* find a free memory area of size 'size'. The search starts at
 262    'start'. If 'start' == 0, then a default start address is used.
 263    Return -1 if error.
 264 */
 265 /* page_init() marks pages used by the host as reserved to be sure not
 266    to use them. */
 267 static abi_ulong mmap_find_vma(abi_ulong start, abi_ulong size)
 268 {
 269     abi_ulong addr, addr1, addr_start;
 270     int prot;
 271     unsigned long new_brk;
 272
 273     new_brk = (unsigned long)sbrk(0);
 274     if (last_brk && last_brk < new_brk && last_brk == (target_ulong)last_brk) {
 275         /* This is a hack to catch the host allocating memory with brk().
 276            If it uses mmap then we loose.
 277            FIXME: We really want to avoid the host allocating memory in
 278            the first place, and maybe leave some slack to avoid switching
 279            to mmap.  */
 280         page_set_flags(last_brk & TARGET_PAGE_MASK,
 281                        TARGET_PAGE_ALIGN(new_brk),
 282                        PAGE_RESERVED);
 283     }
 284     last_brk = new_brk;
 285
 286     size = HOST_PAGE_ALIGN(size);
 287     start = start & qemu_host_page_mask;
 288     addr = start;
 289     if (addr == 0)
 290         addr = mmap_next_start;
 291     addr_start = addr;
 292     for(;;) {
 293         prot = 0;
 294         for(addr1 = addr; addr1 < (addr + size); addr1 += TARGET_PAGE_SIZE) {
 295             prot |= page_get_flags(addr1);
 296         }
 297         if (prot == 0)
 298             break;
 299         addr += qemu_host_page_size;
 300         /* we found nothing */
 301         if (addr == addr_start)
 302             return (abi_ulong)-1;
 303     }
 304     if (start == 0)
 305         mmap_next_start = addr + size;
 306     return addr;
 307 }
 308
 309 /* NOTE: all the constants are the HOST ones */
 310 abi_long target_mmap(abi_ulong start, abi_ulong len, int prot,
 311                      int flags, int fd, abi_ulong offset)
 312 {
 313     abi_ulong ret, end, real_start, real_end, retaddr, host_offset, host_len;
 314     unsigned long host_start;
 315
 316     mmap_lock();
 317 #ifdef DEBUG_MMAP
 318     {
 319         printf("mmap: start=0x" TARGET_FMT_lx
 320                " len=0x" TARGET_FMT_lx " prot=%c%c%c flags=",
 321                start, len,
 322                prot & PROT_READ ? 'r' : '-',
 323                prot & PROT_WRITE ? 'w' : '-',
 324                prot & PROT_EXEC ? 'x' : '-');
 325         if (flags & MAP_FIXED)
 326             printf("MAP_FIXED ");
 327         if (flags & MAP_ANONYMOUS)
 328             printf("MAP_ANON ");
 329         switch(flags & MAP_TYPE) {
 330         case MAP_PRIVATE:
 331             printf("MAP_PRIVATE ");
 332             break;
 333         case MAP_SHARED:
 334             printf("MAP_SHARED ");
 335             break;
 336         default:
 337             printf("[MAP_TYPE=0x%x] ", flags & MAP_TYPE);
 338             break;
 339         }
 340         printf("fd=%d offset=" TARGET_FMT_lx "\n", fd, offset);
 341     }
 342 #endif
 343
 344     if (offset & ~TARGET_PAGE_MASK) {
 345         errno = EINVAL;
 346         goto fail;
 347     }
 348
 349     len = TARGET_PAGE_ALIGN(len);
 350     if (len == 0)
 351         goto the_end;
 352     real_start = start & qemu_host_page_mask;
 353
 354     if (!(flags & MAP_FIXED)) {
 355         abi_ulong mmap_start;
 356         void *p;
 357         host_offset = offset & qemu_host_page_mask;
 358         host_len = len + offset - host_offset;
 359         host_len = HOST_PAGE_ALIGN(host_len);
 360         mmap_start = mmap_find_vma(real_start, host_len);
 361         if (mmap_start == (abi_ulong)-1) {
 362             errno = ENOMEM;
 363             goto fail;
 364         }
 365         /* Note: we prefer to control the mapping address. It is
 366            especially important if qemu_host_page_size >
 367            qemu_real_host_page_size */
 368         p = mmap(g2h(mmap_start),
 369                  host_len, prot, flags | MAP_FIXED, fd, host_offset);
 370         if (p == MAP_FAILED)
 371             goto fail;
 372         /* update start so that it points to the file position at 'offset' */
 373         host_start = (unsigned long)p;
 374         if (!(flags & MAP_ANONYMOUS))
 375             host_start += offset - host_offset;
 376         start = h2g(host_start);
 377     } else {
 378         int flg;
 379         target_ulong addr;
 380
 381         if (start & ~TARGET_PAGE_MASK) {
 382             errno = EINVAL;
 383             goto fail;
 384         }
 385         end = start + len;
 386         real_end = HOST_PAGE_ALIGN(end);
 387
 388         /*
 389          * Test if requested memory area fits target address space
 390          * It can fail only on 64-bit host with 32-bit target.
 391          * On any other target/host host mmap() handles this error correctly.
 392          */
 393         if ((unsigned long)start + len - 1 > (abi_ulong) -1) {
 394             errno = EINVAL;
 395             goto fail;
 396         }
 397
 398         for(addr = real_start; addr < real_end; addr += TARGET_PAGE_SIZE) {
 399             flg = page_get_flags(addr);
 400             if (flg & PAGE_RESERVED) {
 401                 errno = ENXIO;
 402                 goto fail;
 403             }
 404         }
 405
 406         /* worst case: we cannot map the file because the offset is not
 407            aligned, so we read it */
 408         if (!(flags & MAP_ANONYMOUS) &&
 409             (offset & ~qemu_host_page_mask) != (start & ~qemu_host_page_mask)) {
 410             /* msync() won't work here, so we return an error if write is
 411                possible while it is a shared mapping */
 412             if ((flags & MAP_TYPE) == MAP_SHARED &&
 413                 (prot & PROT_WRITE)) {
 414                 errno = EINVAL;
 415                 goto fail;
 416             }
 417             retaddr = target_mmap(start, len, prot | PROT_WRITE,
 418                                   MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS,
 419                                   -1, 0);
 420             if (retaddr == -1)
 421                 goto fail;
 422             pread(fd, g2h(start), len, offset);
 423             if (!(prot & PROT_WRITE)) {
 424                 ret = target_mprotect(start, len, prot);
 425                 if (ret != 0) {
 426                     start = ret;
 427                     goto the_end;
 428                 }
 429             }
 430             goto the_end;
 431         }
 432
 433         /* handle the start of the mapping */
 434         if (start > real_start) {
 435             if (real_end == real_start + qemu_host_page_size) {
 436                 /* one single host page */
 437                 ret = mmap_frag(real_start, start, end,
 438                                 prot, flags, fd, offset);
 439                 if (ret == -1)
 440                     goto fail;
 441                 goto the_end1;
 442             }
 443             ret = mmap_frag(real_start, start, real_start + qemu_host_page_size,
 444                             prot, flags, fd, offset);
 445             if (ret == -1)
 446                 goto fail;
 447             real_start += qemu_host_page_size;
 448         }
 449         /* handle the end of the mapping */
 450         if (end < real_end) {
 451             ret = mmap_frag(real_end - qemu_host_page_size,
 452                             real_end - qemu_host_page_size, real_end,
 453                             prot, flags, fd,
 454                             offset + real_end - qemu_host_page_size - start);
 455             if (ret == -1)
 456                 goto fail;
 457             real_end -= qemu_host_page_size;
 458         }
 459
 460         /* map the middle (easier) */
 461         if (real_start < real_end) {
 462             void *p;
 463             unsigned long offset1;
 464             if (flags & MAP_ANONYMOUS)
 465                 offset1 = 0;
 466             else
 467                 offset1 = offset + real_start - start;
 468             p = mmap(g2h(real_start), real_end - real_start,
 469                      prot, flags, fd, offset1);
 470             if (p == MAP_FAILED)
 471                 goto fail;
 472         }
 473     }
 474  the_end1:
 475     page_set_flags(start, start + len, prot | PAGE_VALID);
 476  the_end:
 477 #ifdef DEBUG_MMAP
 478     printf("ret=0x" TARGET_FMT_lx "\n", start);
 479     page_dump(stdout);
 480     printf("\n");
 481 #endif
 482     mmap_unlock();
 483     return start;
 484 fail:
 485     mmap_unlock();
 486     return -1;
 487 }
 488
 489 int target_munmap(abi_ulong start, abi_ulong len)
 490 {
 491     abi_ulong end, real_start, real_end, addr;
 492     int prot, ret;
 493
 494 #ifdef DEBUG_MMAP
 495     printf("munmap: start=0x%lx len=0x%lx\n", start, len);
 496 #endif
 497     if (start & ~TARGET_PAGE_MASK)
 498         return -EINVAL;
 499     len = TARGET_PAGE_ALIGN(len);
 500     if (len == 0)
 501         return -EINVAL;
 502     mmap_lock();
 503     end = start + len;
 504     real_start = start & qemu_host_page_mask;
 505     real_end = HOST_PAGE_ALIGN(end);
 506
 507     if (start > real_start) {
 508         /* handle host page containing start */
 509         prot = 0;
 510         for(addr = real_start; addr < start; addr += TARGET_PAGE_SIZE) {
 511             prot |= page_get_flags(addr);
 512         }
 513         if (real_end == real_start + qemu_host_page_size) {
 514             for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
 515                 prot |= page_get_flags(addr);
 516             }
 517             end = real_end;
 518         }
 519         if (prot != 0)
 520             real_start += qemu_host_page_size;
 521     }
 522     if (end < real_end) {
 523         prot = 0;
 524         for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
 525             prot |= page_get_flags(addr);
 526         }
 527         if (prot != 0)
 528             real_end -= qemu_host_page_size;
 529     }
 530
 531     ret = 0;
 532     /* unmap what we can */
 533     if (real_start < real_end) {
 534         ret = munmap(g2h(real_start), real_end - real_start);
 535     }
 536
 537     if (ret == 0)
 538         page_set_flags(start, start + len, 0);
 539     mmap_unlock();
 540     return ret;
 541 }
 542
 543 abi_long target_mremap(abi_ulong old_addr, abi_ulong old_size,
 544                        abi_ulong new_size, unsigned long flags,
 545                        abi_ulong new_addr)
 546 {
 547     int prot;
 548     void *host_addr;
 549
 550     mmap_lock();
 551
 552     if (flags & MREMAP_FIXED)
 553         host_addr = (void *) syscall(__NR_mremap, g2h(old_addr),
 554                                      old_size, new_size,
 555                                      flags,
 556                                      new_addr);
 557     else if (flags & MREMAP_MAYMOVE) {
 558         abi_ulong mmap_start;
 559
 560         mmap_start = mmap_find_vma(0, new_size);
 561
 562         if (mmap_start == -1) {
 563             errno = ENOMEM;
 564             host_addr = MAP_FAILED;
 565         } else
 566             host_addr = (void *) syscall(__NR_mremap, g2h(old_addr),
 567                                          old_size, new_size,
 568                                          flags | MREMAP_FIXED,
 569                                          g2h(mmap_start));
 570     } else {
 571         host_addr = mremap(g2h(old_addr), old_size, new_size, flags);
 572         /* Check if address fits target address space */
 573         if ((unsigned long)host_addr + new_size > (abi_ulong)-1) {
 574             /* Revert mremap() changes */
 575             host_addr = mremap(g2h(old_addr), new_size, old_size, flags);
 576             errno = ENOMEM;
 577             host_addr = MAP_FAILED;
 578         }
 579     }
 580
 581     if (host_addr == MAP_FAILED) {
 582         new_addr = -1;
 583     } else {
 584         new_addr = h2g(host_addr);
 585         prot = page_get_flags(old_addr);
 586         page_set_flags(old_addr, old_addr + old_size, 0);
 587         page_set_flags(new_addr, new_addr + new_size, prot | PAGE_VALID);
 588     }
 589     mmap_unlock();
 590     return new_addr;
 591 }
 592
 593 int target_msync(abi_ulong start, abi_ulong len, int flags)
 594 {
 595     abi_ulong end;
 596
 597     if (start & ~TARGET_PAGE_MASK)
 598         return -EINVAL;
 599     len = TARGET_PAGE_ALIGN(len);
 600     end = start + len;
 601     if (end < start)
 602         return -EINVAL;
 603     if (end == start)
 604         return 0;
 605
 606     start &= qemu_host_page_mask;
 607     return msync(g2h(start), end - start, flags);
 608 }