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 <stdlib.h> 20 #include <stdio.h> 21 #include <stdarg.h> 22 #include <string.h> 23 #include <unistd.h> 24 #include <errno.h> 25 #include <sys/mman.h> 26 27 #include "qemu.h" 28 #include "qemu-common.h" 29 #include "bsd-mman.h" 30 31 //#define DEBUG_MMAP 32 33 #if defined(CONFIG_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 /* NOTE: all the constants are the HOST ones, but addresses are target. */ 78 int target_mprotect(abi_ulong start, abi_ulong len, int prot) 79 { 80 abi_ulong end, host_start, host_end, addr; 81 int prot1, ret; 82 83 #ifdef DEBUG_MMAP 84 printf("mprotect: start=0x" TARGET_FMT_lx 85 " len=0x" TARGET_FMT_lx " prot=%c%c%c\n", start, len, 86 prot & PROT_READ ? 'r' : '-', 87 prot & PROT_WRITE ? 'w' : '-', 88 prot & PROT_EXEC ? 'x' : '-'); 89 #endif 90 91 if ((start & ~TARGET_PAGE_MASK) != 0) 92 return -EINVAL; 93 len = TARGET_PAGE_ALIGN(len); 94 end = start + len; 95 if (end < start) 96 return -EINVAL; 97 prot &= PROT_READ | PROT_WRITE | PROT_EXEC; 98 if (len == 0) 99 return 0; 100 101 mmap_lock(); 102 host_start = start & qemu_host_page_mask; 103 host_end = HOST_PAGE_ALIGN(end); 104 if (start > host_start) { 105 /* handle host page containing start */ 106 prot1 = prot; 107 for(addr = host_start; addr < start; addr += TARGET_PAGE_SIZE) { 108 prot1 |= page_get_flags(addr); 109 } 110 if (host_end == host_start + qemu_host_page_size) { 111 for(addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) { 112 prot1 |= page_get_flags(addr); 113 } 114 end = host_end; 115 } 116 ret = mprotect(g2h(host_start), qemu_host_page_size, prot1 & PAGE_BITS); 117 if (ret != 0) 118 goto error; 119 host_start += qemu_host_page_size; 120 } 121 if (end < host_end) { 122 prot1 = prot; 123 for(addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) { 124 prot1 |= page_get_flags(addr); 125 } 126 ret = mprotect(g2h(host_end - qemu_host_page_size), qemu_host_page_size, 127 prot1 & PAGE_BITS); 128 if (ret != 0) 129 goto error; 130 host_end -= qemu_host_page_size; 131 } 132 133 /* handle the pages in the middle */ 134 if (host_start < host_end) { 135 ret = mprotect(g2h(host_start), host_end - host_start, prot); 136 if (ret != 0) 137 goto error; 138 } 139 page_set_flags(start, start + len, prot | PAGE_VALID); 140 mmap_unlock(); 141 return 0; 142 error: 143 mmap_unlock(); 144 return ret; 145 } 146 147 /* map an incomplete host page */ 148 static int mmap_frag(abi_ulong real_start, 149 abi_ulong start, abi_ulong end, 150 int prot, int flags, int fd, abi_ulong offset) 151 { 152 abi_ulong real_end, addr; 153 void *host_start; 154 int prot1, prot_new; 155 156 real_end = real_start + qemu_host_page_size; 157 host_start = g2h(real_start); 158 159 /* get the protection of the target pages outside the mapping */ 160 prot1 = 0; 161 for(addr = real_start; addr < real_end; addr++) { 162 if (addr < start || addr >= end) 163 prot1 |= page_get_flags(addr); 164 } 165 166 if (prot1 == 0) { 167 /* no page was there, so we allocate one */ 168 void *p = mmap(host_start, qemu_host_page_size, prot, 169 flags | MAP_ANON, -1, 0); 170 if (p == MAP_FAILED) 171 return -1; 172 prot1 = prot; 173 } 174 prot1 &= PAGE_BITS; 175 176 prot_new = prot | prot1; 177 if (!(flags & MAP_ANON)) { 178 /* msync() won't work here, so we return an error if write is 179 possible while it is a shared mapping */ 180 if ((flags & TARGET_BSD_MAP_FLAGMASK) == MAP_SHARED && 181 (prot & PROT_WRITE)) 182 return -1; 183 184 /* adjust protection to be able to read */ 185 if (!(prot1 & PROT_WRITE)) 186 mprotect(host_start, qemu_host_page_size, prot1 | PROT_WRITE); 187 188 /* read the corresponding file data */ 189 pread(fd, g2h(start), end - start, offset); 190 191 /* put final protection */ 192 if (prot_new != (prot1 | PROT_WRITE)) 193 mprotect(host_start, qemu_host_page_size, prot_new); 194 } else { 195 /* just update the protection */ 196 if (prot_new != prot1) { 197 mprotect(host_start, qemu_host_page_size, prot_new); 198 } 199 } 200 return 0; 201 } 202 203 #if defined(__CYGWIN__) 204 /* Cygwin doesn't have a whole lot of address space. */ 205 static abi_ulong mmap_next_start = 0x18000000; 206 #else 207 static abi_ulong mmap_next_start = 0x40000000; 208 #endif 209 210 unsigned long last_brk; 211 212 /* find a free memory area of size 'size'. The search starts at 213 'start'. If 'start' == 0, then a default start address is used. 214 Return -1 if error. 215 */ 216 /* page_init() marks pages used by the host as reserved to be sure not 217 to use them. */ 218 static abi_ulong mmap_find_vma(abi_ulong start, abi_ulong size) 219 { 220 abi_ulong addr, addr1, addr_start; 221 int prot; 222 unsigned long new_brk; 223 224 new_brk = (unsigned long)sbrk(0); 225 if (last_brk && last_brk < new_brk && last_brk == (target_ulong)last_brk) { 226 /* This is a hack to catch the host allocating memory with brk(). 227 If it uses mmap then we loose. 228 FIXME: We really want to avoid the host allocating memory in 229 the first place, and maybe leave some slack to avoid switching 230 to mmap. */ 231 page_set_flags(last_brk & TARGET_PAGE_MASK, 232 TARGET_PAGE_ALIGN(new_brk), 233 PAGE_RESERVED); 234 } 235 last_brk = new_brk; 236 237 size = HOST_PAGE_ALIGN(size); 238 start = start & qemu_host_page_mask; 239 addr = start; 240 if (addr == 0) 241 addr = mmap_next_start; 242 addr_start = addr; 243 for(;;) { 244 prot = 0; 245 for(addr1 = addr; addr1 < (addr + size); addr1 += TARGET_PAGE_SIZE) { 246 prot |= page_get_flags(addr1); 247 } 248 if (prot == 0) 249 break; 250 addr += qemu_host_page_size; 251 /* we found nothing */ 252 if (addr == addr_start) 253 return (abi_ulong)-1; 254 } 255 if (start == 0) 256 mmap_next_start = addr + size; 257 return addr; 258 } 259 260 /* NOTE: all the constants are the HOST ones */ 261 abi_long target_mmap(abi_ulong start, abi_ulong len, int prot, 262 int flags, int fd, abi_ulong offset) 263 { 264 abi_ulong ret, end, real_start, real_end, retaddr, host_offset, host_len; 265 unsigned long host_start; 266 267 mmap_lock(); 268 #ifdef DEBUG_MMAP 269 { 270 printf("mmap: start=0x" TARGET_FMT_lx 271 " len=0x" TARGET_FMT_lx " prot=%c%c%c flags=", 272 start, len, 273 prot & PROT_READ ? 'r' : '-', 274 prot & PROT_WRITE ? 'w' : '-', 275 prot & PROT_EXEC ? 'x' : '-'); 276 if (flags & MAP_FIXED) 277 printf("MAP_FIXED "); 278 if (flags & MAP_ANON) 279 printf("MAP_ANON "); 280 switch(flags & TARGET_BSD_MAP_FLAGMASK) { 281 case MAP_PRIVATE: 282 printf("MAP_PRIVATE "); 283 break; 284 case MAP_SHARED: 285 printf("MAP_SHARED "); 286 break; 287 default: 288 printf("[MAP_FLAGMASK=0x%x] ", flags & TARGET_BSD_MAP_FLAGMASK); 289 break; 290 } 291 printf("fd=%d offset=" TARGET_FMT_lx "\n", fd, offset); 292 } 293 #endif 294 295 if (offset & ~TARGET_PAGE_MASK) { 296 errno = EINVAL; 297 goto fail; 298 } 299 300 len = TARGET_PAGE_ALIGN(len); 301 if (len == 0) 302 goto the_end; 303 real_start = start & qemu_host_page_mask; 304 305 if (!(flags & MAP_FIXED)) { 306 abi_ulong mmap_start; 307 void *p; 308 host_offset = offset & qemu_host_page_mask; 309 host_len = len + offset - host_offset; 310 host_len = HOST_PAGE_ALIGN(host_len); 311 mmap_start = mmap_find_vma(real_start, host_len); 312 if (mmap_start == (abi_ulong)-1) { 313 errno = ENOMEM; 314 goto fail; 315 } 316 /* Note: we prefer to control the mapping address. It is 317 especially important if qemu_host_page_size > 318 qemu_real_host_page_size */ 319 p = mmap(g2h(mmap_start), 320 host_len, prot, flags | MAP_FIXED, fd, host_offset); 321 if (p == MAP_FAILED) 322 goto fail; 323 /* update start so that it points to the file position at 'offset' */ 324 host_start = (unsigned long)p; 325 if (!(flags & MAP_ANON)) 326 host_start += offset - host_offset; 327 start = h2g(host_start); 328 } else { 329 int flg; 330 target_ulong addr; 331 332 if (start & ~TARGET_PAGE_MASK) { 333 errno = EINVAL; 334 goto fail; 335 } 336 end = start + len; 337 real_end = HOST_PAGE_ALIGN(end); 338 339 for(addr = real_start; addr < real_end; addr += TARGET_PAGE_SIZE) { 340 flg = page_get_flags(addr); 341 if (flg & PAGE_RESERVED) { 342 errno = ENXIO; 343 goto fail; 344 } 345 } 346 347 /* worst case: we cannot map the file because the offset is not 348 aligned, so we read it */ 349 if (!(flags & MAP_ANON) && 350 (offset & ~qemu_host_page_mask) != (start & ~qemu_host_page_mask)) { 351 /* msync() won't work here, so we return an error if write is 352 possible while it is a shared mapping */ 353 if ((flags & TARGET_BSD_MAP_FLAGMASK) == MAP_SHARED && 354 (prot & PROT_WRITE)) { 355 errno = EINVAL; 356 goto fail; 357 } 358 retaddr = target_mmap(start, len, prot | PROT_WRITE, 359 MAP_FIXED | MAP_PRIVATE | MAP_ANON, 360 -1, 0); 361 if (retaddr == -1) 362 goto fail; 363 pread(fd, g2h(start), len, offset); 364 if (!(prot & PROT_WRITE)) { 365 ret = target_mprotect(start, len, prot); 366 if (ret != 0) { 367 start = ret; 368 goto the_end; 369 } 370 } 371 goto the_end; 372 } 373 374 /* handle the start of the mapping */ 375 if (start > real_start) { 376 if (real_end == real_start + qemu_host_page_size) { 377 /* one single host page */ 378 ret = mmap_frag(real_start, start, end, 379 prot, flags, fd, offset); 380 if (ret == -1) 381 goto fail; 382 goto the_end1; 383 } 384 ret = mmap_frag(real_start, start, real_start + qemu_host_page_size, 385 prot, flags, fd, offset); 386 if (ret == -1) 387 goto fail; 388 real_start += qemu_host_page_size; 389 } 390 /* handle the end of the mapping */ 391 if (end < real_end) { 392 ret = mmap_frag(real_end - qemu_host_page_size, 393 real_end - qemu_host_page_size, real_end, 394 prot, flags, fd, 395 offset + real_end - qemu_host_page_size - start); 396 if (ret == -1) 397 goto fail; 398 real_end -= qemu_host_page_size; 399 } 400 401 /* map the middle (easier) */ 402 if (real_start < real_end) { 403 void *p; 404 unsigned long offset1; 405 if (flags & MAP_ANON) 406 offset1 = 0; 407 else 408 offset1 = offset + real_start - start; 409 p = mmap(g2h(real_start), real_end - real_start, 410 prot, flags, fd, offset1); 411 if (p == MAP_FAILED) 412 goto fail; 413 } 414 } 415 the_end1: 416 page_set_flags(start, start + len, prot | PAGE_VALID); 417 the_end: 418 #ifdef DEBUG_MMAP 419 printf("ret=0x" TARGET_FMT_lx "\n", start); 420 page_dump(stdout); 421 printf("\n"); 422 #endif 423 mmap_unlock(); 424 return start; 425 fail: 426 mmap_unlock(); 427 return -1; 428 } 429 430 int target_munmap(abi_ulong start, abi_ulong len) 431 { 432 abi_ulong end, real_start, real_end, addr; 433 int prot, ret; 434 435 #ifdef DEBUG_MMAP 436 printf("munmap: start=0x%lx len=0x%lx\n", start, len); 437 #endif 438 if (start & ~TARGET_PAGE_MASK) 439 return -EINVAL; 440 len = TARGET_PAGE_ALIGN(len); 441 if (len == 0) 442 return -EINVAL; 443 mmap_lock(); 444 end = start + len; 445 real_start = start & qemu_host_page_mask; 446 real_end = HOST_PAGE_ALIGN(end); 447 448 if (start > real_start) { 449 /* handle host page containing start */ 450 prot = 0; 451 for(addr = real_start; addr < start; addr += TARGET_PAGE_SIZE) { 452 prot |= page_get_flags(addr); 453 } 454 if (real_end == real_start + qemu_host_page_size) { 455 for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) { 456 prot |= page_get_flags(addr); 457 } 458 end = real_end; 459 } 460 if (prot != 0) 461 real_start += qemu_host_page_size; 462 } 463 if (end < real_end) { 464 prot = 0; 465 for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) { 466 prot |= page_get_flags(addr); 467 } 468 if (prot != 0) 469 real_end -= qemu_host_page_size; 470 } 471 472 ret = 0; 473 /* unmap what we can */ 474 if (real_start < real_end) { 475 ret = munmap(g2h(real_start), real_end - real_start); 476 } 477 478 if (ret == 0) 479 page_set_flags(start, start + len, 0); 480 mmap_unlock(); 481 return ret; 482 } 483 484 int target_msync(abi_ulong start, abi_ulong len, int flags) 485 { 486 abi_ulong end; 487 488 if (start & ~TARGET_PAGE_MASK) 489 return -EINVAL; 490 len = TARGET_PAGE_ALIGN(len); 491 end = start + len; 492 if (end < start) 493 return -EINVAL; 494 if (end == start) 495 return 0; 496 497 start &= qemu_host_page_mask; 498 return msync(g2h(start), end - start, flags); 499 } 500