blueswir1 | 8477850 | 2008-10-26 20:33:16 +0000 | [diff] [blame] | 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 |
Blue Swirl | 8167ee8 | 2009-07-16 20:47:01 +0000 | [diff] [blame] | 17 | * along with this program; if not, see <http://www.gnu.org/licenses/>. |
blueswir1 | 8477850 | 2008-10-26 20:33:16 +0000 | [diff] [blame] | 18 | */ |
Peter Maydell | 2231197 | 2016-01-29 17:49:53 +0000 | [diff] [blame] | 19 | #include "qemu/osdep.h" |
blueswir1 | 8477850 | 2008-10-26 20:33:16 +0000 | [diff] [blame] | 20 | |
| 21 | #include "qemu.h" |
| 22 | #include "qemu-common.h" |
blueswir1 | 6c173b3 | 2008-11-29 14:05:16 +0000 | [diff] [blame] | 23 | #include "bsd-mman.h" |
Paolo Bonzini | 05757c5 | 2018-06-01 14:15:01 +0200 | [diff] [blame] | 24 | #include "exec/exec-all.h" |
blueswir1 | 8477850 | 2008-10-26 20:33:16 +0000 | [diff] [blame] | 25 | |
| 26 | //#define DEBUG_MMAP |
| 27 | |
Alex Bennée | 95992b6 | 2017-03-20 14:36:47 +0000 | [diff] [blame] | 28 | static pthread_mutex_t mmap_mutex = PTHREAD_MUTEX_INITIALIZER; |
Peter Maydell | 06943a6 | 2017-07-18 17:26:31 +0100 | [diff] [blame] | 29 | static __thread int mmap_lock_count; |
blueswir1 | 8477850 | 2008-10-26 20:33:16 +0000 | [diff] [blame] | 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 | |
Alex Bennée | 301e40e | 2016-10-27 16:10:00 +0100 | [diff] [blame] | 45 | bool have_mmap_lock(void) |
| 46 | { |
| 47 | return mmap_lock_count > 0 ? true : false; |
| 48 | } |
| 49 | |
blueswir1 | 8477850 | 2008-10-26 20:33:16 +0000 | [diff] [blame] | 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 | } |
blueswir1 | 8477850 | 2008-10-26 20:33:16 +0000 | [diff] [blame] | 65 | |
blueswir1 | 8477850 | 2008-10-26 20:33:16 +0000 | [diff] [blame] | 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(host_start), qemu_host_page_size, prot1 & PAGE_BITS); |
| 106 | if (ret != 0) |
| 107 | goto error; |
| 108 | host_start += qemu_host_page_size; |
| 109 | } |
| 110 | if (end < host_end) { |
| 111 | prot1 = prot; |
| 112 | for(addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) { |
| 113 | prot1 |= page_get_flags(addr); |
| 114 | } |
| 115 | ret = mprotect(g2h(host_end - qemu_host_page_size), qemu_host_page_size, |
| 116 | prot1 & PAGE_BITS); |
| 117 | if (ret != 0) |
| 118 | goto error; |
| 119 | host_end -= qemu_host_page_size; |
| 120 | } |
| 121 | |
| 122 | /* handle the pages in the middle */ |
| 123 | if (host_start < host_end) { |
| 124 | ret = mprotect(g2h(host_start), host_end - host_start, prot); |
| 125 | if (ret != 0) |
| 126 | goto error; |
| 127 | } |
| 128 | page_set_flags(start, start + len, prot | PAGE_VALID); |
| 129 | mmap_unlock(); |
| 130 | return 0; |
| 131 | error: |
| 132 | mmap_unlock(); |
| 133 | return ret; |
| 134 | } |
| 135 | |
| 136 | /* map an incomplete host page */ |
| 137 | static int mmap_frag(abi_ulong real_start, |
| 138 | abi_ulong start, abi_ulong end, |
| 139 | int prot, int flags, int fd, abi_ulong offset) |
| 140 | { |
| 141 | abi_ulong real_end, addr; |
| 142 | void *host_start; |
| 143 | int prot1, prot_new; |
| 144 | |
| 145 | real_end = real_start + qemu_host_page_size; |
| 146 | host_start = g2h(real_start); |
| 147 | |
| 148 | /* get the protection of the target pages outside the mapping */ |
| 149 | prot1 = 0; |
| 150 | for(addr = real_start; addr < real_end; addr++) { |
| 151 | if (addr < start || addr >= end) |
| 152 | prot1 |= page_get_flags(addr); |
| 153 | } |
| 154 | |
| 155 | if (prot1 == 0) { |
| 156 | /* no page was there, so we allocate one */ |
| 157 | void *p = mmap(host_start, qemu_host_page_size, prot, |
| 158 | flags | MAP_ANON, -1, 0); |
| 159 | if (p == MAP_FAILED) |
| 160 | return -1; |
| 161 | prot1 = prot; |
| 162 | } |
| 163 | prot1 &= PAGE_BITS; |
| 164 | |
| 165 | prot_new = prot | prot1; |
| 166 | if (!(flags & MAP_ANON)) { |
| 167 | /* msync() won't work here, so we return an error if write is |
| 168 | possible while it is a shared mapping */ |
blueswir1 | 6c173b3 | 2008-11-29 14:05:16 +0000 | [diff] [blame] | 169 | if ((flags & TARGET_BSD_MAP_FLAGMASK) == MAP_SHARED && |
blueswir1 | 8477850 | 2008-10-26 20:33:16 +0000 | [diff] [blame] | 170 | (prot & PROT_WRITE)) |
Blue Swirl | 059bca4 | 2010-01-31 13:41:07 +0000 | [diff] [blame] | 171 | return -1; |
blueswir1 | 8477850 | 2008-10-26 20:33:16 +0000 | [diff] [blame] | 172 | |
| 173 | /* adjust protection to be able to read */ |
| 174 | if (!(prot1 & PROT_WRITE)) |
| 175 | mprotect(host_start, qemu_host_page_size, prot1 | PROT_WRITE); |
| 176 | |
| 177 | /* read the corresponding file data */ |
| 178 | pread(fd, g2h(start), end - start, offset); |
| 179 | |
| 180 | /* put final protection */ |
| 181 | if (prot_new != (prot1 | PROT_WRITE)) |
| 182 | mprotect(host_start, qemu_host_page_size, prot_new); |
| 183 | } else { |
| 184 | /* just update the protection */ |
| 185 | if (prot_new != prot1) { |
| 186 | mprotect(host_start, qemu_host_page_size, prot_new); |
| 187 | } |
| 188 | } |
| 189 | return 0; |
| 190 | } |
| 191 | |
blueswir1 | 8477850 | 2008-10-26 20:33:16 +0000 | [diff] [blame] | 192 | static abi_ulong mmap_next_start = 0x40000000; |
blueswir1 | 8477850 | 2008-10-26 20:33:16 +0000 | [diff] [blame] | 193 | |
| 194 | unsigned long last_brk; |
| 195 | |
| 196 | /* find a free memory area of size 'size'. The search starts at |
| 197 | 'start'. If 'start' == 0, then a default start address is used. |
| 198 | Return -1 if error. |
| 199 | */ |
| 200 | /* page_init() marks pages used by the host as reserved to be sure not |
| 201 | to use them. */ |
| 202 | static abi_ulong mmap_find_vma(abi_ulong start, abi_ulong size) |
| 203 | { |
| 204 | abi_ulong addr, addr1, addr_start; |
| 205 | int prot; |
| 206 | unsigned long new_brk; |
| 207 | |
| 208 | new_brk = (unsigned long)sbrk(0); |
| 209 | if (last_brk && last_brk < new_brk && last_brk == (target_ulong)last_brk) { |
| 210 | /* This is a hack to catch the host allocating memory with brk(). |
| 211 | If it uses mmap then we loose. |
| 212 | FIXME: We really want to avoid the host allocating memory in |
| 213 | the first place, and maybe leave some slack to avoid switching |
| 214 | to mmap. */ |
| 215 | page_set_flags(last_brk & TARGET_PAGE_MASK, |
| 216 | TARGET_PAGE_ALIGN(new_brk), |
| 217 | PAGE_RESERVED); |
| 218 | } |
| 219 | last_brk = new_brk; |
| 220 | |
| 221 | size = HOST_PAGE_ALIGN(size); |
| 222 | start = start & qemu_host_page_mask; |
| 223 | addr = start; |
| 224 | if (addr == 0) |
| 225 | addr = mmap_next_start; |
| 226 | addr_start = addr; |
| 227 | for(;;) { |
| 228 | prot = 0; |
| 229 | for(addr1 = addr; addr1 < (addr + size); addr1 += TARGET_PAGE_SIZE) { |
| 230 | prot |= page_get_flags(addr1); |
| 231 | } |
| 232 | if (prot == 0) |
| 233 | break; |
| 234 | addr += qemu_host_page_size; |
| 235 | /* we found nothing */ |
| 236 | if (addr == addr_start) |
| 237 | return (abi_ulong)-1; |
| 238 | } |
| 239 | if (start == 0) |
| 240 | mmap_next_start = addr + size; |
| 241 | return addr; |
| 242 | } |
| 243 | |
| 244 | /* NOTE: all the constants are the HOST ones */ |
| 245 | abi_long target_mmap(abi_ulong start, abi_ulong len, int prot, |
| 246 | int flags, int fd, abi_ulong offset) |
| 247 | { |
| 248 | abi_ulong ret, end, real_start, real_end, retaddr, host_offset, host_len; |
| 249 | unsigned long host_start; |
| 250 | |
| 251 | mmap_lock(); |
| 252 | #ifdef DEBUG_MMAP |
| 253 | { |
| 254 | printf("mmap: start=0x" TARGET_FMT_lx |
| 255 | " len=0x" TARGET_FMT_lx " prot=%c%c%c flags=", |
| 256 | start, len, |
| 257 | prot & PROT_READ ? 'r' : '-', |
| 258 | prot & PROT_WRITE ? 'w' : '-', |
| 259 | prot & PROT_EXEC ? 'x' : '-'); |
| 260 | if (flags & MAP_FIXED) |
| 261 | printf("MAP_FIXED "); |
| 262 | if (flags & MAP_ANON) |
| 263 | printf("MAP_ANON "); |
blueswir1 | 6c173b3 | 2008-11-29 14:05:16 +0000 | [diff] [blame] | 264 | switch(flags & TARGET_BSD_MAP_FLAGMASK) { |
blueswir1 | 8477850 | 2008-10-26 20:33:16 +0000 | [diff] [blame] | 265 | case MAP_PRIVATE: |
| 266 | printf("MAP_PRIVATE "); |
| 267 | break; |
| 268 | case MAP_SHARED: |
| 269 | printf("MAP_SHARED "); |
| 270 | break; |
| 271 | default: |
blueswir1 | 6c173b3 | 2008-11-29 14:05:16 +0000 | [diff] [blame] | 272 | printf("[MAP_FLAGMASK=0x%x] ", flags & TARGET_BSD_MAP_FLAGMASK); |
blueswir1 | 8477850 | 2008-10-26 20:33:16 +0000 | [diff] [blame] | 273 | break; |
| 274 | } |
| 275 | printf("fd=%d offset=" TARGET_FMT_lx "\n", fd, offset); |
| 276 | } |
| 277 | #endif |
| 278 | |
| 279 | if (offset & ~TARGET_PAGE_MASK) { |
| 280 | errno = EINVAL; |
| 281 | goto fail; |
| 282 | } |
| 283 | |
| 284 | len = TARGET_PAGE_ALIGN(len); |
| 285 | if (len == 0) |
| 286 | goto the_end; |
| 287 | real_start = start & qemu_host_page_mask; |
| 288 | |
| 289 | if (!(flags & MAP_FIXED)) { |
| 290 | abi_ulong mmap_start; |
| 291 | void *p; |
| 292 | host_offset = offset & qemu_host_page_mask; |
| 293 | host_len = len + offset - host_offset; |
| 294 | host_len = HOST_PAGE_ALIGN(host_len); |
| 295 | mmap_start = mmap_find_vma(real_start, host_len); |
| 296 | if (mmap_start == (abi_ulong)-1) { |
| 297 | errno = ENOMEM; |
| 298 | goto fail; |
| 299 | } |
| 300 | /* Note: we prefer to control the mapping address. It is |
| 301 | especially important if qemu_host_page_size > |
| 302 | qemu_real_host_page_size */ |
| 303 | p = mmap(g2h(mmap_start), |
| 304 | host_len, prot, flags | MAP_FIXED, fd, host_offset); |
| 305 | if (p == MAP_FAILED) |
| 306 | goto fail; |
| 307 | /* update start so that it points to the file position at 'offset' */ |
| 308 | host_start = (unsigned long)p; |
| 309 | if (!(flags & MAP_ANON)) |
| 310 | host_start += offset - host_offset; |
| 311 | start = h2g(host_start); |
| 312 | } else { |
| 313 | int flg; |
| 314 | target_ulong addr; |
| 315 | |
| 316 | if (start & ~TARGET_PAGE_MASK) { |
| 317 | errno = EINVAL; |
| 318 | goto fail; |
| 319 | } |
| 320 | end = start + len; |
| 321 | real_end = HOST_PAGE_ALIGN(end); |
| 322 | |
| 323 | for(addr = real_start; addr < real_end; addr += TARGET_PAGE_SIZE) { |
| 324 | flg = page_get_flags(addr); |
| 325 | if (flg & PAGE_RESERVED) { |
| 326 | errno = ENXIO; |
| 327 | goto fail; |
| 328 | } |
| 329 | } |
| 330 | |
| 331 | /* worst case: we cannot map the file because the offset is not |
| 332 | aligned, so we read it */ |
| 333 | if (!(flags & MAP_ANON) && |
| 334 | (offset & ~qemu_host_page_mask) != (start & ~qemu_host_page_mask)) { |
| 335 | /* msync() won't work here, so we return an error if write is |
| 336 | possible while it is a shared mapping */ |
blueswir1 | 6c173b3 | 2008-11-29 14:05:16 +0000 | [diff] [blame] | 337 | if ((flags & TARGET_BSD_MAP_FLAGMASK) == MAP_SHARED && |
blueswir1 | 8477850 | 2008-10-26 20:33:16 +0000 | [diff] [blame] | 338 | (prot & PROT_WRITE)) { |
| 339 | errno = EINVAL; |
| 340 | goto fail; |
| 341 | } |
| 342 | retaddr = target_mmap(start, len, prot | PROT_WRITE, |
| 343 | MAP_FIXED | MAP_PRIVATE | MAP_ANON, |
| 344 | -1, 0); |
| 345 | if (retaddr == -1) |
| 346 | goto fail; |
| 347 | pread(fd, g2h(start), len, offset); |
| 348 | if (!(prot & PROT_WRITE)) { |
| 349 | ret = target_mprotect(start, len, prot); |
| 350 | if (ret != 0) { |
| 351 | start = ret; |
| 352 | goto the_end; |
| 353 | } |
| 354 | } |
| 355 | goto the_end; |
| 356 | } |
| 357 | |
| 358 | /* handle the start of the mapping */ |
| 359 | if (start > real_start) { |
| 360 | if (real_end == real_start + qemu_host_page_size) { |
| 361 | /* one single host page */ |
| 362 | ret = mmap_frag(real_start, start, end, |
| 363 | prot, flags, fd, offset); |
| 364 | if (ret == -1) |
| 365 | goto fail; |
| 366 | goto the_end1; |
| 367 | } |
| 368 | ret = mmap_frag(real_start, start, real_start + qemu_host_page_size, |
| 369 | prot, flags, fd, offset); |
| 370 | if (ret == -1) |
| 371 | goto fail; |
| 372 | real_start += qemu_host_page_size; |
| 373 | } |
| 374 | /* handle the end of the mapping */ |
| 375 | if (end < real_end) { |
| 376 | ret = mmap_frag(real_end - qemu_host_page_size, |
| 377 | real_end - qemu_host_page_size, real_end, |
| 378 | prot, flags, fd, |
| 379 | offset + real_end - qemu_host_page_size - start); |
| 380 | if (ret == -1) |
| 381 | goto fail; |
| 382 | real_end -= qemu_host_page_size; |
| 383 | } |
| 384 | |
| 385 | /* map the middle (easier) */ |
| 386 | if (real_start < real_end) { |
| 387 | void *p; |
| 388 | unsigned long offset1; |
| 389 | if (flags & MAP_ANON) |
| 390 | offset1 = 0; |
| 391 | else |
| 392 | offset1 = offset + real_start - start; |
| 393 | p = mmap(g2h(real_start), real_end - real_start, |
| 394 | prot, flags, fd, offset1); |
| 395 | if (p == MAP_FAILED) |
| 396 | goto fail; |
| 397 | } |
| 398 | } |
| 399 | the_end1: |
| 400 | page_set_flags(start, start + len, prot | PAGE_VALID); |
| 401 | the_end: |
| 402 | #ifdef DEBUG_MMAP |
| 403 | printf("ret=0x" TARGET_FMT_lx "\n", start); |
| 404 | page_dump(stdout); |
| 405 | printf("\n"); |
| 406 | #endif |
| 407 | mmap_unlock(); |
| 408 | return start; |
| 409 | fail: |
| 410 | mmap_unlock(); |
| 411 | return -1; |
| 412 | } |
| 413 | |
| 414 | int target_munmap(abi_ulong start, abi_ulong len) |
| 415 | { |
| 416 | abi_ulong end, real_start, real_end, addr; |
| 417 | int prot, ret; |
| 418 | |
| 419 | #ifdef DEBUG_MMAP |
| 420 | printf("munmap: start=0x%lx len=0x%lx\n", start, len); |
| 421 | #endif |
| 422 | if (start & ~TARGET_PAGE_MASK) |
| 423 | return -EINVAL; |
| 424 | len = TARGET_PAGE_ALIGN(len); |
| 425 | if (len == 0) |
| 426 | return -EINVAL; |
| 427 | mmap_lock(); |
| 428 | end = start + len; |
| 429 | real_start = start & qemu_host_page_mask; |
| 430 | real_end = HOST_PAGE_ALIGN(end); |
| 431 | |
| 432 | if (start > real_start) { |
| 433 | /* handle host page containing start */ |
| 434 | prot = 0; |
| 435 | for(addr = real_start; addr < start; addr += TARGET_PAGE_SIZE) { |
| 436 | prot |= page_get_flags(addr); |
| 437 | } |
| 438 | if (real_end == real_start + qemu_host_page_size) { |
| 439 | for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) { |
| 440 | prot |= page_get_flags(addr); |
| 441 | } |
| 442 | end = real_end; |
| 443 | } |
| 444 | if (prot != 0) |
| 445 | real_start += qemu_host_page_size; |
| 446 | } |
| 447 | if (end < real_end) { |
| 448 | prot = 0; |
| 449 | for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) { |
| 450 | prot |= page_get_flags(addr); |
| 451 | } |
| 452 | if (prot != 0) |
| 453 | real_end -= qemu_host_page_size; |
| 454 | } |
| 455 | |
| 456 | ret = 0; |
| 457 | /* unmap what we can */ |
| 458 | if (real_start < real_end) { |
| 459 | ret = munmap(g2h(real_start), real_end - real_start); |
| 460 | } |
| 461 | |
| 462 | if (ret == 0) |
| 463 | page_set_flags(start, start + len, 0); |
| 464 | mmap_unlock(); |
| 465 | return ret; |
| 466 | } |
| 467 | |
| 468 | int target_msync(abi_ulong start, abi_ulong len, int flags) |
| 469 | { |
| 470 | abi_ulong end; |
| 471 | |
| 472 | if (start & ~TARGET_PAGE_MASK) |
| 473 | return -EINVAL; |
| 474 | len = TARGET_PAGE_ALIGN(len); |
| 475 | end = start + len; |
| 476 | if (end < start) |
| 477 | return -EINVAL; |
| 478 | if (end == start) |
| 479 | return 0; |
| 480 | |
| 481 | start &= qemu_host_page_mask; |
| 482 | return msync(g2h(start), end - start, flags); |
| 483 | } |