| 1 | /* |
| 2 | * os-posix-lib.c |
| 3 | * |
| 4 | * Copyright (c) 2003-2008 Fabrice Bellard |
| 5 | * Copyright (c) 2010 Red Hat, Inc. |
| 6 | * |
| 7 | * QEMU library functions on POSIX which are shared between QEMU and |
| 8 | * the QEMU tools. |
| 9 | * |
| 10 | * Permission is hereby granted, free of charge, to any person obtaining a copy |
| 11 | * of this software and associated documentation files (the "Software"), to deal |
| 12 | * in the Software without restriction, including without limitation the rights |
| 13 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell |
| 14 | * copies of the Software, and to permit persons to whom the Software is |
| 15 | * furnished to do so, subject to the following conditions: |
| 16 | * |
| 17 | * The above copyright notice and this permission notice shall be included in |
| 18 | * all copies or substantial portions of the Software. |
| 19 | * |
| 20 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| 21 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 22 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
| 23 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
| 24 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
| 25 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN |
| 26 | * THE SOFTWARE. |
| 27 | */ |
| 28 | |
| 29 | #include "qemu/osdep.h" |
| 30 | #include <termios.h> |
| 31 | |
| 32 | #include <glib/gprintf.h> |
| 33 | |
| 34 | #include "system/system.h" |
| 35 | #include "trace.h" |
| 36 | #include "qapi/error.h" |
| 37 | #include "qemu/error-report.h" |
| 38 | #include "qemu/madvise.h" |
| 39 | #include "qemu/sockets.h" |
| 40 | #include "qemu/thread.h" |
| 41 | #include <libgen.h> |
| 42 | #include "qemu/cutils.h" |
| 43 | #include "qemu/units.h" |
| 44 | #include "qemu/thread-context.h" |
| 45 | #include "qemu/main-loop.h" |
| 46 | |
| 47 | #ifdef CONFIG_LINUX |
| 48 | #include <sys/syscall.h> |
| 49 | #endif |
| 50 | |
| 51 | #ifdef __FreeBSD__ |
| 52 | #include <sys/thr.h> |
| 53 | #include <sys/user.h> |
| 54 | #include <libutil.h> |
| 55 | #endif |
| 56 | |
| 57 | #ifdef __NetBSD__ |
| 58 | #include <lwp.h> |
| 59 | #endif |
| 60 | |
| 61 | #include "qemu/memalign.h" |
| 62 | #include "qemu/mmap-alloc.h" |
| 63 | |
| 64 | #define MAX_MEM_PREALLOC_THREAD_COUNT 32 |
| 65 | |
| 66 | struct MemsetThread; |
| 67 | |
| 68 | static QLIST_HEAD(, MemsetContext) memset_contexts = |
| 69 | QLIST_HEAD_INITIALIZER(memset_contexts); |
| 70 | |
| 71 | typedef struct MemsetContext { |
| 72 | bool all_threads_created; |
| 73 | bool any_thread_failed; |
| 74 | struct MemsetThread *threads; |
| 75 | int num_threads; |
| 76 | QLIST_ENTRY(MemsetContext) next; |
| 77 | } MemsetContext; |
| 78 | |
| 79 | struct MemsetThread { |
| 80 | char *addr; |
| 81 | size_t numpages; |
| 82 | size_t hpagesize; |
| 83 | QemuThread pgthread; |
| 84 | sigjmp_buf env; |
| 85 | MemsetContext *context; |
| 86 | }; |
| 87 | typedef struct MemsetThread MemsetThread; |
| 88 | |
| 89 | /* used by sigbus_handler() */ |
| 90 | static MemsetContext *sigbus_memset_context; |
| 91 | struct sigaction sigbus_oldact; |
| 92 | static QemuMutex sigbus_mutex; |
| 93 | |
| 94 | static QemuMutex page_mutex; |
| 95 | static QemuCond page_cond; |
| 96 | |
| 97 | int qemu_get_thread_id(void) |
| 98 | { |
| 99 | #if defined(__linux__) |
| 100 | return syscall(SYS_gettid); |
| 101 | #elif defined(__FreeBSD__) |
| 102 | /* thread id is up to INT_MAX */ |
| 103 | long tid; |
| 104 | thr_self(&tid); |
| 105 | return (int)tid; |
| 106 | #elif defined(__NetBSD__) |
| 107 | return _lwp_self(); |
| 108 | #elif defined(__OpenBSD__) |
| 109 | return getthrid(); |
| 110 | #else |
| 111 | return getpid(); |
| 112 | #endif |
| 113 | } |
| 114 | |
| 115 | int qemu_kill_thread(int tid, int sig) |
| 116 | { |
| 117 | #if defined(__linux__) |
| 118 | return syscall(__NR_tgkill, getpid(), tid, sig); |
| 119 | #elif defined(__FreeBSD__) |
| 120 | return thr_kill2(getpid(), tid, sig); |
| 121 | #elif defined(__NetBSD__) |
| 122 | return _lwp_kill(tid, sig); |
| 123 | #elif defined(__OpenBSD__) |
| 124 | return thrkill(tid, sig, NULL); |
| 125 | #else |
| 126 | return kill(tid, sig); |
| 127 | #endif |
| 128 | } |
| 129 | |
| 130 | int qemu_daemon(int nochdir, int noclose) |
| 131 | { |
| 132 | return daemon(nochdir, noclose); |
| 133 | } |
| 134 | |
| 135 | bool qemu_write_pidfile(const char *path, Error **errp) |
| 136 | { |
| 137 | int fd; |
| 138 | char pidstr[32]; |
| 139 | |
| 140 | while (1) { |
| 141 | struct stat a, b; |
| 142 | struct flock lock = { |
| 143 | .l_type = F_WRLCK, |
| 144 | .l_whence = SEEK_SET, |
| 145 | .l_len = 0, |
| 146 | }; |
| 147 | |
| 148 | fd = qemu_create(path, O_WRONLY, S_IRUSR | S_IWUSR, errp); |
| 149 | if (fd == -1) { |
| 150 | return false; |
| 151 | } |
| 152 | |
| 153 | if (fstat(fd, &b) < 0) { |
| 154 | error_setg_errno(errp, errno, "Cannot stat file"); |
| 155 | goto fail_close; |
| 156 | } |
| 157 | |
| 158 | if (fcntl(fd, F_SETLK, &lock)) { |
| 159 | error_setg_errno(errp, errno, "Cannot lock pid file"); |
| 160 | goto fail_close; |
| 161 | } |
| 162 | |
| 163 | /* |
| 164 | * Now make sure the path we locked is the same one that now |
| 165 | * exists on the filesystem. |
| 166 | */ |
| 167 | if (stat(path, &a) < 0) { |
| 168 | /* |
| 169 | * PID file disappeared, someone else must be racing with |
| 170 | * us, so try again. |
| 171 | */ |
| 172 | close(fd); |
| 173 | continue; |
| 174 | } |
| 175 | |
| 176 | if (a.st_ino == b.st_ino) { |
| 177 | break; |
| 178 | } |
| 179 | |
| 180 | /* |
| 181 | * PID file was recreated, someone else must be racing with |
| 182 | * us, so try again. |
| 183 | */ |
| 184 | close(fd); |
| 185 | } |
| 186 | |
| 187 | if (ftruncate(fd, 0) < 0) { |
| 188 | error_setg_errno(errp, errno, "Failed to truncate pid file"); |
| 189 | goto fail_unlink; |
| 190 | } |
| 191 | |
| 192 | snprintf(pidstr, sizeof(pidstr), FMT_pid "\n", getpid()); |
| 193 | if (qemu_write_full(fd, pidstr, strlen(pidstr)) != strlen(pidstr)) { |
| 194 | error_setg(errp, "Failed to write pid file"); |
| 195 | goto fail_unlink; |
| 196 | } |
| 197 | |
| 198 | return true; |
| 199 | |
| 200 | fail_unlink: |
| 201 | unlink(path); |
| 202 | fail_close: |
| 203 | close(fd); |
| 204 | return false; |
| 205 | } |
| 206 | |
| 207 | /* alloc shared memory pages */ |
| 208 | void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment, bool shared, |
| 209 | bool noreserve) |
| 210 | { |
| 211 | const uint32_t qemu_map_flags = (shared ? QEMU_MAP_SHARED : 0) | |
| 212 | (noreserve ? QEMU_MAP_NORESERVE : 0); |
| 213 | size_t align = QEMU_VMALLOC_ALIGN; |
| 214 | #ifndef EMSCRIPTEN |
| 215 | void *ptr = qemu_ram_mmap(-1, size, align, qemu_map_flags, 0); |
| 216 | |
| 217 | if (ptr == MAP_FAILED) { |
| 218 | return NULL; |
| 219 | } |
| 220 | #else |
| 221 | /* |
| 222 | * qemu_ram_mmap is not implemented for Emscripten. Use qemu_memalign |
| 223 | * for the anonymous allocation. noreserve is ignored as there is no swap |
| 224 | * space on Emscripten, and shared is ignored as there is no other |
| 225 | * processes on Emscripten. |
| 226 | */ |
| 227 | void *ptr = qemu_memalign(align, size); |
| 228 | #endif |
| 229 | |
| 230 | if (alignment) { |
| 231 | *alignment = align; |
| 232 | } |
| 233 | |
| 234 | trace_qemu_anon_ram_alloc(size, ptr); |
| 235 | return ptr; |
| 236 | } |
| 237 | |
| 238 | void qemu_anon_ram_free(void *ptr, size_t size) |
| 239 | { |
| 240 | trace_qemu_anon_ram_free(ptr, size); |
| 241 | #ifndef EMSCRIPTEN |
| 242 | qemu_ram_munmap(-1, ptr, size); |
| 243 | #else |
| 244 | /* |
| 245 | * qemu_ram_munmap is not implemented for Emscripten and qemu_memalign |
| 246 | * was used for the allocation. Use the corresponding freeing function |
| 247 | * here. |
| 248 | */ |
| 249 | qemu_vfree(ptr); |
| 250 | #endif |
| 251 | } |
| 252 | |
| 253 | bool qemu_set_blocking(int fd, bool block, Error **errp) |
| 254 | { |
| 255 | g_autoptr(GError) err = NULL; |
| 256 | |
| 257 | if (!g_unix_set_fd_nonblocking(fd, !block, &err)) { |
| 258 | error_setg_errno(errp, errno, |
| 259 | "Can't set file descriptor %d %s: %s", fd, |
| 260 | block ? "blocking" : "non-blocking", |
| 261 | err->message); |
| 262 | return false; |
| 263 | } |
| 264 | |
| 265 | return true; |
| 266 | } |
| 267 | |
| 268 | int socket_set_fast_reuse(int fd) |
| 269 | { |
| 270 | int val = 1, ret; |
| 271 | |
| 272 | ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, |
| 273 | (const char *)&val, sizeof(val)); |
| 274 | |
| 275 | assert(ret == 0); |
| 276 | |
| 277 | return ret; |
| 278 | } |
| 279 | |
| 280 | void qemu_set_cloexec(int fd) |
| 281 | { |
| 282 | int f; |
| 283 | f = fcntl(fd, F_GETFD); |
| 284 | assert(f != -1); |
| 285 | f = fcntl(fd, F_SETFD, f | FD_CLOEXEC); |
| 286 | assert(f != -1); |
| 287 | } |
| 288 | |
| 289 | int qemu_socketpair(int domain, int type, int protocol, int sv[2]) |
| 290 | { |
| 291 | int ret; |
| 292 | |
| 293 | #ifdef SOCK_CLOEXEC |
| 294 | ret = socketpair(domain, type | SOCK_CLOEXEC, protocol, sv); |
| 295 | if (ret != -1 || errno != EINVAL) { |
| 296 | return ret; |
| 297 | } |
| 298 | #endif |
| 299 | ret = socketpair(domain, type, protocol, sv); |
| 300 | if (ret == 0) { |
| 301 | qemu_set_cloexec(sv[0]); |
| 302 | qemu_set_cloexec(sv[1]); |
| 303 | } |
| 304 | |
| 305 | return ret; |
| 306 | } |
| 307 | |
| 308 | void qemu_clear_cloexec(int fd) |
| 309 | { |
| 310 | int f; |
| 311 | f = fcntl(fd, F_GETFD); |
| 312 | assert(f != -1); |
| 313 | f = fcntl(fd, F_SETFD, f & ~FD_CLOEXEC); |
| 314 | assert(f != -1); |
| 315 | } |
| 316 | |
| 317 | char * |
| 318 | qemu_get_local_state_dir(void) |
| 319 | { |
| 320 | return get_relocated_path(CONFIG_QEMU_LOCALSTATEDIR); |
| 321 | } |
| 322 | |
| 323 | void qemu_set_tty_echo(int fd, bool echo) |
| 324 | { |
| 325 | struct termios tty; |
| 326 | |
| 327 | tcgetattr(fd, &tty); |
| 328 | |
| 329 | if (echo) { |
| 330 | tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN; |
| 331 | } else { |
| 332 | tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN); |
| 333 | } |
| 334 | |
| 335 | tcsetattr(fd, TCSANOW, &tty); |
| 336 | } |
| 337 | |
| 338 | #ifdef CONFIG_LINUX |
| 339 | static void sigbus_handler(int signal, siginfo_t *siginfo, void *ctx) |
| 340 | #else /* CONFIG_LINUX */ |
| 341 | static void sigbus_handler(int signal) |
| 342 | #endif /* CONFIG_LINUX */ |
| 343 | { |
| 344 | int i; |
| 345 | |
| 346 | if (sigbus_memset_context) { |
| 347 | for (i = 0; i < sigbus_memset_context->num_threads; i++) { |
| 348 | MemsetThread *thread = &sigbus_memset_context->threads[i]; |
| 349 | |
| 350 | if (qemu_thread_is_self(&thread->pgthread)) { |
| 351 | siglongjmp(thread->env, 1); |
| 352 | } |
| 353 | } |
| 354 | } |
| 355 | |
| 356 | #ifdef CONFIG_LINUX |
| 357 | /* |
| 358 | * We assume that the MCE SIGBUS handler could have been registered. We |
| 359 | * should never receive BUS_MCEERR_AO on any of our threads, but only on |
| 360 | * the main thread registered for PR_MCE_KILL_EARLY. Further, we should not |
| 361 | * receive BUS_MCEERR_AR triggered by action of other threads on one of |
| 362 | * our threads. So, no need to check for unrelated SIGBUS when seeing one |
| 363 | * for our threads. |
| 364 | * |
| 365 | * We will forward to the MCE handler, which will either handle the SIGBUS |
| 366 | * or reinstall the default SIGBUS handler and reraise the SIGBUS. The |
| 367 | * default SIGBUS handler will crash the process, so we don't care. |
| 368 | */ |
| 369 | if (sigbus_oldact.sa_flags & SA_SIGINFO) { |
| 370 | sigbus_oldact.sa_sigaction(signal, siginfo, ctx); |
| 371 | return; |
| 372 | } |
| 373 | #endif /* CONFIG_LINUX */ |
| 374 | warn_report("qemu_prealloc_mem: unrelated SIGBUS detected and ignored"); |
| 375 | } |
| 376 | |
| 377 | static void *do_touch_pages(void *arg) |
| 378 | { |
| 379 | MemsetThread *memset_args = (MemsetThread *)arg; |
| 380 | sigset_t set, oldset; |
| 381 | int ret = 0; |
| 382 | |
| 383 | /* |
| 384 | * On Linux, the page faults from the loop below can cause mmap_sem |
| 385 | * contention with allocation of the thread stacks. Do not start |
| 386 | * clearing until all threads have been created. |
| 387 | */ |
| 388 | qemu_mutex_lock(&page_mutex); |
| 389 | while (!memset_args->context->all_threads_created) { |
| 390 | qemu_cond_wait(&page_cond, &page_mutex); |
| 391 | } |
| 392 | qemu_mutex_unlock(&page_mutex); |
| 393 | |
| 394 | /* unblock SIGBUS */ |
| 395 | sigemptyset(&set); |
| 396 | sigaddset(&set, SIGBUS); |
| 397 | pthread_sigmask(SIG_UNBLOCK, &set, &oldset); |
| 398 | |
| 399 | if (sigsetjmp(memset_args->env, 1)) { |
| 400 | ret = -EFAULT; |
| 401 | } else { |
| 402 | char *addr = memset_args->addr; |
| 403 | size_t numpages = memset_args->numpages; |
| 404 | size_t hpagesize = memset_args->hpagesize; |
| 405 | size_t i; |
| 406 | for (i = 0; i < numpages; i++) { |
| 407 | /* |
| 408 | * Read & write back the same value, so we don't |
| 409 | * corrupt existing user/app data that might be |
| 410 | * stored. |
| 411 | * |
| 412 | * 'volatile' to stop compiler optimizing this away |
| 413 | * to a no-op |
| 414 | */ |
| 415 | *(volatile char *)addr = *addr; |
| 416 | addr += hpagesize; |
| 417 | } |
| 418 | } |
| 419 | pthread_sigmask(SIG_SETMASK, &oldset, NULL); |
| 420 | return (void *)(uintptr_t)ret; |
| 421 | } |
| 422 | |
| 423 | static void *do_madv_populate_write_pages(void *arg) |
| 424 | { |
| 425 | MemsetThread *memset_args = (MemsetThread *)arg; |
| 426 | const size_t size = memset_args->numpages * memset_args->hpagesize; |
| 427 | char * const addr = memset_args->addr; |
| 428 | int ret = 0; |
| 429 | |
| 430 | /* See do_touch_pages(). */ |
| 431 | qemu_mutex_lock(&page_mutex); |
| 432 | while (!memset_args->context->all_threads_created) { |
| 433 | qemu_cond_wait(&page_cond, &page_mutex); |
| 434 | } |
| 435 | qemu_mutex_unlock(&page_mutex); |
| 436 | |
| 437 | if (size && qemu_madvise(addr, size, QEMU_MADV_POPULATE_WRITE)) { |
| 438 | ret = -errno; |
| 439 | } |
| 440 | return (void *)(uintptr_t)ret; |
| 441 | } |
| 442 | |
| 443 | static inline int get_memset_num_threads(size_t hpagesize, size_t numpages, |
| 444 | int max_threads) |
| 445 | { |
| 446 | long host_procs = sysconf(_SC_NPROCESSORS_ONLN); |
| 447 | int ret = 1; |
| 448 | |
| 449 | if (host_procs > 0) { |
| 450 | ret = MIN(MIN(host_procs, MAX_MEM_PREALLOC_THREAD_COUNT), max_threads); |
| 451 | } |
| 452 | |
| 453 | /* Especially with gigantic pages, don't create more threads than pages. */ |
| 454 | ret = MIN(ret, numpages); |
| 455 | /* Don't start threads to prealloc comparatively little memory. */ |
| 456 | ret = MIN(ret, MAX(1, hpagesize * numpages / (64 * MiB))); |
| 457 | |
| 458 | /* In case sysconf() fails, we fall back to single threaded */ |
| 459 | return ret; |
| 460 | } |
| 461 | |
| 462 | static int wait_and_free_mem_prealloc_context(MemsetContext *context) |
| 463 | { |
| 464 | int i, ret = 0, tmp; |
| 465 | |
| 466 | for (i = 0; i < context->num_threads; i++) { |
| 467 | tmp = (uintptr_t)qemu_thread_join(&context->threads[i].pgthread); |
| 468 | |
| 469 | if (tmp) { |
| 470 | ret = tmp; |
| 471 | } |
| 472 | } |
| 473 | g_free(context->threads); |
| 474 | g_free(context); |
| 475 | return ret; |
| 476 | } |
| 477 | |
| 478 | static int touch_all_pages(char *area, size_t hpagesize, size_t numpages, |
| 479 | int max_threads, ThreadContext *tc, bool async, |
| 480 | bool use_madv_populate_write) |
| 481 | { |
| 482 | static gsize initialized = 0; |
| 483 | MemsetContext *context = g_malloc0(sizeof(MemsetContext)); |
| 484 | size_t numpages_per_thread, leftover; |
| 485 | void *(*touch_fn)(void *); |
| 486 | int ret, i = 0; |
| 487 | char *addr = area; |
| 488 | |
| 489 | /* |
| 490 | * Asynchronous preallocation is only allowed when using MADV_POPULATE_WRITE |
| 491 | * and prealloc context for thread placement. |
| 492 | */ |
| 493 | if (!use_madv_populate_write || !tc) { |
| 494 | async = false; |
| 495 | } |
| 496 | |
| 497 | context->num_threads = |
| 498 | get_memset_num_threads(hpagesize, numpages, max_threads); |
| 499 | |
| 500 | if (g_once_init_enter(&initialized)) { |
| 501 | qemu_mutex_init(&page_mutex); |
| 502 | qemu_cond_init(&page_cond); |
| 503 | g_once_init_leave(&initialized, 1); |
| 504 | } |
| 505 | |
| 506 | if (use_madv_populate_write) { |
| 507 | /* |
| 508 | * Avoid creating a single thread for MADV_POPULATE_WRITE when |
| 509 | * preallocating synchronously. |
| 510 | */ |
| 511 | if (context->num_threads == 1 && !async) { |
| 512 | ret = 0; |
| 513 | if (qemu_madvise(area, hpagesize * numpages, |
| 514 | QEMU_MADV_POPULATE_WRITE)) { |
| 515 | ret = -errno; |
| 516 | } |
| 517 | g_free(context); |
| 518 | return ret; |
| 519 | } |
| 520 | touch_fn = do_madv_populate_write_pages; |
| 521 | } else { |
| 522 | touch_fn = do_touch_pages; |
| 523 | } |
| 524 | |
| 525 | context->threads = g_new0(MemsetThread, context->num_threads); |
| 526 | numpages_per_thread = numpages / context->num_threads; |
| 527 | leftover = numpages % context->num_threads; |
| 528 | for (i = 0; i < context->num_threads; i++) { |
| 529 | context->threads[i].addr = addr; |
| 530 | context->threads[i].numpages = numpages_per_thread + (i < leftover); |
| 531 | context->threads[i].hpagesize = hpagesize; |
| 532 | context->threads[i].context = context; |
| 533 | if (tc) { |
| 534 | thread_context_create_thread(tc, &context->threads[i].pgthread, |
| 535 | "touch_pages", |
| 536 | touch_fn, &context->threads[i], |
| 537 | QEMU_THREAD_JOINABLE); |
| 538 | } else { |
| 539 | qemu_thread_create(&context->threads[i].pgthread, "touch_pages", |
| 540 | touch_fn, &context->threads[i], |
| 541 | QEMU_THREAD_JOINABLE); |
| 542 | } |
| 543 | addr += context->threads[i].numpages * hpagesize; |
| 544 | } |
| 545 | |
| 546 | if (async) { |
| 547 | /* |
| 548 | * async requests currently require the BQL. Add it to the list and kick |
| 549 | * preallocation off during qemu_finish_async_prealloc_mem(). |
| 550 | */ |
| 551 | assert(bql_locked()); |
| 552 | QLIST_INSERT_HEAD(&memset_contexts, context, next); |
| 553 | return 0; |
| 554 | } |
| 555 | |
| 556 | if (!use_madv_populate_write) { |
| 557 | sigbus_memset_context = context; |
| 558 | } |
| 559 | |
| 560 | qemu_mutex_lock(&page_mutex); |
| 561 | context->all_threads_created = true; |
| 562 | qemu_cond_broadcast(&page_cond); |
| 563 | qemu_mutex_unlock(&page_mutex); |
| 564 | |
| 565 | ret = wait_and_free_mem_prealloc_context(context); |
| 566 | |
| 567 | if (!use_madv_populate_write) { |
| 568 | sigbus_memset_context = NULL; |
| 569 | } |
| 570 | return ret; |
| 571 | } |
| 572 | |
| 573 | bool qemu_finish_async_prealloc_mem(Error **errp) |
| 574 | { |
| 575 | int ret = 0, tmp; |
| 576 | MemsetContext *context, *next_context; |
| 577 | |
| 578 | /* Waiting for preallocation requires the BQL. */ |
| 579 | assert(bql_locked()); |
| 580 | if (QLIST_EMPTY(&memset_contexts)) { |
| 581 | return true; |
| 582 | } |
| 583 | |
| 584 | qemu_mutex_lock(&page_mutex); |
| 585 | QLIST_FOREACH(context, &memset_contexts, next) { |
| 586 | context->all_threads_created = true; |
| 587 | } |
| 588 | qemu_cond_broadcast(&page_cond); |
| 589 | qemu_mutex_unlock(&page_mutex); |
| 590 | |
| 591 | QLIST_FOREACH_SAFE(context, &memset_contexts, next, next_context) { |
| 592 | QLIST_REMOVE(context, next); |
| 593 | tmp = wait_and_free_mem_prealloc_context(context); |
| 594 | if (tmp) { |
| 595 | ret = tmp; |
| 596 | } |
| 597 | } |
| 598 | |
| 599 | if (ret) { |
| 600 | error_setg_errno(errp, -ret, |
| 601 | "qemu_prealloc_mem: preallocating memory failed"); |
| 602 | return false; |
| 603 | } |
| 604 | return true; |
| 605 | } |
| 606 | |
| 607 | static bool madv_populate_write_possible(char *area, size_t pagesize) |
| 608 | { |
| 609 | return !qemu_madvise(area, pagesize, QEMU_MADV_POPULATE_WRITE) || |
| 610 | errno != EINVAL; |
| 611 | } |
| 612 | |
| 613 | bool qemu_prealloc_mem(int fd, char *area, size_t sz, int max_threads, |
| 614 | ThreadContext *tc, bool async, Error **errp) |
| 615 | { |
| 616 | static gsize initialized; |
| 617 | int ret; |
| 618 | #ifndef EMSCRIPTEN |
| 619 | size_t hpagesize = qemu_fd_getpagesize(fd); |
| 620 | #else |
| 621 | /* |
| 622 | * mmap-alloc.c is excluded from Emscripten build, so qemu_fd_getpagesize |
| 623 | * is unavailable. Fallback to the lower level implementation. |
| 624 | */ |
| 625 | size_t hpagesize = qemu_real_host_page_size(); |
| 626 | #endif |
| 627 | size_t numpages = DIV_ROUND_UP(sz, hpagesize); |
| 628 | bool use_madv_populate_write; |
| 629 | struct sigaction act; |
| 630 | bool rv = true; |
| 631 | |
| 632 | /* |
| 633 | * Sense on every invocation, as MADV_POPULATE_WRITE cannot be used for |
| 634 | * some special mappings, such as mapping /dev/mem. |
| 635 | */ |
| 636 | use_madv_populate_write = madv_populate_write_possible(area, hpagesize); |
| 637 | |
| 638 | if (!use_madv_populate_write) { |
| 639 | if (g_once_init_enter(&initialized)) { |
| 640 | qemu_mutex_init(&sigbus_mutex); |
| 641 | g_once_init_leave(&initialized, 1); |
| 642 | } |
| 643 | |
| 644 | qemu_mutex_lock(&sigbus_mutex); |
| 645 | memset(&act, 0, sizeof(act)); |
| 646 | #ifdef CONFIG_LINUX |
| 647 | act.sa_sigaction = &sigbus_handler; |
| 648 | act.sa_flags = SA_SIGINFO; |
| 649 | #else /* CONFIG_LINUX */ |
| 650 | act.sa_handler = &sigbus_handler; |
| 651 | act.sa_flags = 0; |
| 652 | #endif /* CONFIG_LINUX */ |
| 653 | |
| 654 | ret = sigaction(SIGBUS, &act, &sigbus_oldact); |
| 655 | if (ret) { |
| 656 | qemu_mutex_unlock(&sigbus_mutex); |
| 657 | error_setg_errno(errp, errno, |
| 658 | "qemu_prealloc_mem: failed to install signal handler"); |
| 659 | return false; |
| 660 | } |
| 661 | } |
| 662 | |
| 663 | /* touch pages simultaneously */ |
| 664 | ret = touch_all_pages(area, hpagesize, numpages, max_threads, tc, async, |
| 665 | use_madv_populate_write); |
| 666 | if (ret) { |
| 667 | error_setg_errno(errp, -ret, |
| 668 | "qemu_prealloc_mem: preallocating memory failed"); |
| 669 | rv = false; |
| 670 | } |
| 671 | |
| 672 | if (!use_madv_populate_write) { |
| 673 | ret = sigaction(SIGBUS, &sigbus_oldact, NULL); |
| 674 | if (ret) { |
| 675 | /* Terminate QEMU since it can't recover from error */ |
| 676 | perror("qemu_prealloc_mem: failed to reinstall signal handler"); |
| 677 | exit(1); |
| 678 | } |
| 679 | qemu_mutex_unlock(&sigbus_mutex); |
| 680 | } |
| 681 | return rv; |
| 682 | } |
| 683 | |
| 684 | char *qemu_get_pid_name(pid_t pid) |
| 685 | { |
| 686 | char *name = NULL; |
| 687 | |
| 688 | #if defined(__FreeBSD__) |
| 689 | /* BSDs don't have /proc, but they provide a nice substitute */ |
| 690 | struct kinfo_proc *proc = kinfo_getproc(pid); |
| 691 | |
| 692 | if (proc) { |
| 693 | name = g_strdup(proc->ki_comm); |
| 694 | free(proc); |
| 695 | } |
| 696 | #else |
| 697 | /* Assume a system with reasonable procfs */ |
| 698 | char *pid_path; |
| 699 | size_t len; |
| 700 | |
| 701 | pid_path = g_strdup_printf("/proc/%d/cmdline", pid); |
| 702 | g_file_get_contents(pid_path, &name, &len, NULL); |
| 703 | g_free(pid_path); |
| 704 | #endif |
| 705 | |
| 706 | return name; |
| 707 | } |
| 708 | |
| 709 | |
| 710 | void *qemu_alloc_stack(size_t *sz) |
| 711 | { |
| 712 | void *ptr; |
| 713 | int flags; |
| 714 | #ifdef CONFIG_DEBUG_STACK_USAGE |
| 715 | void *ptr2; |
| 716 | #endif |
| 717 | size_t pagesz = qemu_real_host_page_size(); |
| 718 | #ifdef _SC_THREAD_STACK_MIN |
| 719 | /* avoid stacks smaller than _SC_THREAD_STACK_MIN */ |
| 720 | long min_stack_sz = sysconf(_SC_THREAD_STACK_MIN); |
| 721 | *sz = MAX(MAX(min_stack_sz, 0), *sz); |
| 722 | #endif |
| 723 | /* adjust stack size to a multiple of the page size */ |
| 724 | *sz = ROUND_UP(*sz, pagesz); |
| 725 | /* allocate one extra page for the guard page */ |
| 726 | *sz += pagesz; |
| 727 | |
| 728 | flags = MAP_PRIVATE | MAP_ANONYMOUS; |
| 729 | #if defined(MAP_STACK) && defined(__OpenBSD__) |
| 730 | /* Only enable MAP_STACK on OpenBSD. Other OS's such as |
| 731 | * Linux/FreeBSD/NetBSD have a flag with the same name |
| 732 | * but have differing functionality. OpenBSD will SEGV |
| 733 | * if it spots execution with a stack pointer pointing |
| 734 | * at memory that was not allocated with MAP_STACK. |
| 735 | */ |
| 736 | flags |= MAP_STACK; |
| 737 | #endif |
| 738 | |
| 739 | ptr = mmap(NULL, *sz, PROT_READ | PROT_WRITE, flags, -1, 0); |
| 740 | if (ptr == MAP_FAILED) { |
| 741 | perror("failed to allocate memory for stack"); |
| 742 | abort(); |
| 743 | } |
| 744 | |
| 745 | /* Stack grows down -- guard page at the bottom. */ |
| 746 | if (mprotect(ptr, pagesz, PROT_NONE) != 0) { |
| 747 | perror("failed to set up stack guard page"); |
| 748 | abort(); |
| 749 | } |
| 750 | |
| 751 | #ifdef CONFIG_DEBUG_STACK_USAGE |
| 752 | for (ptr2 = ptr + pagesz; ptr2 < ptr + *sz; ptr2 += sizeof(uint32_t)) { |
| 753 | *(uint32_t *)ptr2 = 0xdeadbeaf; |
| 754 | } |
| 755 | #endif |
| 756 | |
| 757 | return ptr; |
| 758 | } |
| 759 | |
| 760 | #ifdef CONFIG_DEBUG_STACK_USAGE |
| 761 | static __thread unsigned int max_stack_usage; |
| 762 | #endif |
| 763 | |
| 764 | void qemu_free_stack(void *stack, size_t sz) |
| 765 | { |
| 766 | #ifdef CONFIG_DEBUG_STACK_USAGE |
| 767 | unsigned int usage; |
| 768 | void *ptr; |
| 769 | |
| 770 | for (ptr = stack + qemu_real_host_page_size(); ptr < stack + sz; |
| 771 | ptr += sizeof(uint32_t)) { |
| 772 | if (*(uint32_t *)ptr != 0xdeadbeaf) { |
| 773 | break; |
| 774 | } |
| 775 | } |
| 776 | usage = sz - (uintptr_t) (ptr - stack); |
| 777 | if (usage > max_stack_usage) { |
| 778 | error_report("thread %d max stack usage increased from %u to %u", |
| 779 | qemu_get_thread_id(), max_stack_usage, usage); |
| 780 | max_stack_usage = usage; |
| 781 | } |
| 782 | #endif |
| 783 | |
| 784 | munmap(stack, sz); |
| 785 | } |
| 786 | |
| 787 | /* |
| 788 | * Disable CFI checks. |
| 789 | * We are going to call a signal handler directly. Such handler may or may not |
| 790 | * have been defined in our binary, so there's no guarantee that the pointer |
| 791 | * used to set the handler is a cfi-valid pointer. Since the handlers are |
| 792 | * stored in kernel memory, changing the handler to an attacker-defined |
| 793 | * function requires being able to call a sigaction() syscall, |
| 794 | * which is not as easy as overwriting a pointer in memory. |
| 795 | */ |
| 796 | QEMU_DISABLE_CFI |
| 797 | void sigaction_invoke(struct sigaction *action, |
| 798 | struct qemu_signalfd_siginfo *info) |
| 799 | { |
| 800 | siginfo_t si = {}; |
| 801 | si.si_signo = info->ssi_signo; |
| 802 | si.si_errno = info->ssi_errno; |
| 803 | si.si_code = info->ssi_code; |
| 804 | |
| 805 | /* Convert the minimal set of fields defined by POSIX. |
| 806 | * Positive si_code values are reserved for kernel-generated |
| 807 | * signals, where the valid siginfo fields are determined by |
| 808 | * the signal number. But according to POSIX, it is unspecified |
| 809 | * whether SI_USER and SI_QUEUE have values less than or equal to |
| 810 | * zero. |
| 811 | */ |
| 812 | if (info->ssi_code == SI_USER || info->ssi_code == SI_QUEUE || |
| 813 | info->ssi_code <= 0) { |
| 814 | /* SIGTERM, etc. */ |
| 815 | si.si_pid = info->ssi_pid; |
| 816 | si.si_uid = info->ssi_uid; |
| 817 | } else if (info->ssi_signo == SIGILL || info->ssi_signo == SIGFPE || |
| 818 | info->ssi_signo == SIGSEGV || info->ssi_signo == SIGBUS) { |
| 819 | si.si_addr = (void *)(uintptr_t)info->ssi_addr; |
| 820 | } else if (info->ssi_signo == SIGCHLD) { |
| 821 | si.si_pid = info->ssi_pid; |
| 822 | si.si_status = info->ssi_status; |
| 823 | si.si_uid = info->ssi_uid; |
| 824 | } |
| 825 | action->sa_sigaction(info->ssi_signo, &si, NULL); |
| 826 | } |
| 827 | |
| 828 | size_t qemu_get_host_physmem(void) |
| 829 | { |
| 830 | #ifdef _SC_PHYS_PAGES |
| 831 | long pages = sysconf(_SC_PHYS_PAGES); |
| 832 | if (pages > 0) { |
| 833 | if (pages > SIZE_MAX / qemu_real_host_page_size()) { |
| 834 | return SIZE_MAX; |
| 835 | } else { |
| 836 | return pages * qemu_real_host_page_size(); |
| 837 | } |
| 838 | } |
| 839 | #endif |
| 840 | return 0; |
| 841 | } |
| 842 | |
| 843 | int qemu_msync(void *addr, size_t length, int fd) |
| 844 | { |
| 845 | size_t align_mask = ~(qemu_real_host_page_size() - 1); |
| 846 | |
| 847 | /** |
| 848 | * There are no strict reqs as per the length of mapping |
| 849 | * to be synced. Still the length needs to follow the address |
| 850 | * alignment changes. Additionally - round the size to the multiple |
| 851 | * of PAGE_SIZE |
| 852 | */ |
| 853 | length += ((uintptr_t)addr & (qemu_real_host_page_size() - 1)); |
| 854 | length = (length + ~align_mask) & align_mask; |
| 855 | |
| 856 | addr = (void *)((uintptr_t)addr & align_mask); |
| 857 | |
| 858 | return msync(addr, length, MS_SYNC); |
| 859 | } |
| 860 | |
| 861 | static bool qemu_close_all_open_fd_proc(const int *skip, unsigned int nskip) |
| 862 | { |
| 863 | struct dirent *de; |
| 864 | int fd, dfd; |
| 865 | DIR *dir; |
| 866 | unsigned int skip_start = 0, skip_end = nskip; |
| 867 | |
| 868 | dir = opendir("/proc/self/fd"); |
| 869 | if (!dir) { |
| 870 | /* If /proc is not mounted, there is nothing that can be done. */ |
| 871 | return false; |
| 872 | } |
| 873 | /* Avoid closing the directory. */ |
| 874 | dfd = dirfd(dir); |
| 875 | |
| 876 | for (de = readdir(dir); de; de = readdir(dir)) { |
| 877 | bool close_fd = true; |
| 878 | |
| 879 | if (de->d_name[0] == '.') { |
| 880 | continue; |
| 881 | } |
| 882 | fd = atoi(de->d_name); |
| 883 | if (fd == dfd) { |
| 884 | continue; |
| 885 | } |
| 886 | |
| 887 | for (unsigned int i = skip_start; i < skip_end; i++) { |
| 888 | if (fd < skip[i]) { |
| 889 | /* We are below the next skipped fd, break */ |
| 890 | break; |
| 891 | } else if (fd == skip[i]) { |
| 892 | close_fd = false; |
| 893 | /* Restrict the range as we found fds matching start/end */ |
| 894 | if (i == skip_start) { |
| 895 | skip_start++; |
| 896 | } else if (i == skip_end) { |
| 897 | skip_end--; |
| 898 | } |
| 899 | break; |
| 900 | } |
| 901 | } |
| 902 | |
| 903 | if (close_fd) { |
| 904 | close(fd); |
| 905 | } |
| 906 | } |
| 907 | closedir(dir); |
| 908 | |
| 909 | return true; |
| 910 | } |
| 911 | |
| 912 | static bool qemu_close_all_open_fd_close_range(const int *skip, |
| 913 | unsigned int nskip, |
| 914 | int open_max) |
| 915 | { |
| 916 | #ifdef CONFIG_CLOSE_RANGE |
| 917 | int max_fd = open_max - 1; |
| 918 | int first = 0, last; |
| 919 | unsigned int cur_skip = 0; |
| 920 | int ret; |
| 921 | |
| 922 | do { |
| 923 | /* Find the start boundary of the range to close */ |
| 924 | while (cur_skip < nskip && first == skip[cur_skip]) { |
| 925 | cur_skip++; |
| 926 | first++; |
| 927 | } |
| 928 | |
| 929 | /* Find the upper boundary of the range to close */ |
| 930 | last = max_fd; |
| 931 | if (cur_skip < nskip) { |
| 932 | last = skip[cur_skip] - 1; |
| 933 | last = MIN(last, max_fd); |
| 934 | } |
| 935 | |
| 936 | /* With the adjustments to the range, we might be done. */ |
| 937 | if (first > last) { |
| 938 | break; |
| 939 | } |
| 940 | |
| 941 | ret = close_range(first, last, 0); |
| 942 | if (ret < 0) { |
| 943 | return false; |
| 944 | } |
| 945 | |
| 946 | first = last + 1; |
| 947 | } while (last < max_fd); |
| 948 | |
| 949 | return true; |
| 950 | #else |
| 951 | return false; |
| 952 | #endif |
| 953 | } |
| 954 | |
| 955 | static void qemu_close_all_open_fd_fallback(const int *skip, unsigned int nskip, |
| 956 | int open_max) |
| 957 | { |
| 958 | unsigned int cur_skip = 0; |
| 959 | |
| 960 | /* Fallback */ |
| 961 | for (int i = 0; i < open_max; i++) { |
| 962 | if (cur_skip < nskip && i == skip[cur_skip]) { |
| 963 | cur_skip++; |
| 964 | continue; |
| 965 | } |
| 966 | close(i); |
| 967 | } |
| 968 | } |
| 969 | |
| 970 | /* |
| 971 | * Close all open file descriptors. |
| 972 | */ |
| 973 | void qemu_close_all_open_fd(const int *skip, unsigned int nskip) |
| 974 | { |
| 975 | int open_max = sysconf(_SC_OPEN_MAX); |
| 976 | |
| 977 | assert(skip != NULL || nskip == 0); |
| 978 | |
| 979 | if (!qemu_close_all_open_fd_close_range(skip, nskip, open_max) && |
| 980 | !qemu_close_all_open_fd_proc(skip, nskip)) { |
| 981 | qemu_close_all_open_fd_fallback(skip, nskip, open_max); |
| 982 | } |
| 983 | } |
| 984 | |
| 985 | int qemu_shm_alloc(size_t size, Error **errp) |
| 986 | { |
| 987 | g_autoptr(GString) shm_name = g_string_new(NULL); |
| 988 | int fd, oflag, cur_sequence; |
| 989 | static int sequence; |
| 990 | mode_t mode; |
| 991 | |
| 992 | cur_sequence = qatomic_fetch_inc(&sequence); |
| 993 | |
| 994 | /* |
| 995 | * Let's use `mode = 0` because we don't want other processes to open our |
| 996 | * memory unless we share the file descriptor with them. |
| 997 | */ |
| 998 | mode = 0; |
| 999 | oflag = O_RDWR | O_CREAT | O_EXCL; |
| 1000 | |
| 1001 | /* |
| 1002 | * Some operating systems allow creating anonymous POSIX shared memory |
| 1003 | * objects (e.g. FreeBSD provides the SHM_ANON constant), but this is not |
| 1004 | * defined by POSIX, so let's create a unique name. |
| 1005 | * |
| 1006 | * From Linux's shm_open(3) man-page: |
| 1007 | * For portable use, a shared memory object should be identified |
| 1008 | * by a name of the form /somename;" |
| 1009 | */ |
| 1010 | g_string_printf(shm_name, "/qemu-" FMT_pid "-shm-%d", getpid(), |
| 1011 | cur_sequence); |
| 1012 | |
| 1013 | fd = shm_open(shm_name->str, oflag, mode); |
| 1014 | if (fd < 0) { |
| 1015 | error_setg_errno(errp, errno, |
| 1016 | "failed to create POSIX shared memory"); |
| 1017 | return -1; |
| 1018 | } |
| 1019 | |
| 1020 | /* |
| 1021 | * We have the file descriptor, so we no longer need to expose the |
| 1022 | * POSIX shared memory object. However it will remain allocated as long as |
| 1023 | * there are file descriptors pointing to it. |
| 1024 | */ |
| 1025 | shm_unlink(shm_name->str); |
| 1026 | |
| 1027 | if (ftruncate(fd, size) == -1) { |
| 1028 | error_setg_errno(errp, errno, |
| 1029 | "failed to resize POSIX shared memory to %zu", size); |
| 1030 | close(fd); |
| 1031 | return -1; |
| 1032 | } |
| 1033 | |
| 1034 | return fd; |
| 1035 | } |