]>
Commit | Line | Data |
---|---|---|
c1b0b93b JS |
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 | ||
aafd7584 | 29 | #include "qemu/osdep.h" |
13401ba0 | 30 | #include <termios.h> |
13401ba0 | 31 | |
e2ea3515 LE |
32 | #include <glib/gprintf.h> |
33 | ||
9c17d615 | 34 | #include "sysemu/sysemu.h" |
c1b0b93b | 35 | #include "trace.h" |
da34e65c | 36 | #include "qapi/error.h" |
29b838c0 | 37 | #include "qemu/error-report.h" |
b85ea5fa | 38 | #include "qemu/madvise.h" |
1de7afc9 | 39 | #include "qemu/sockets.h" |
db725815 | 40 | #include "qemu/thread.h" |
10f5bff6 | 41 | #include <libgen.h> |
f348b6d1 | 42 | #include "qemu/cutils.h" |
89aec641 | 43 | #include "qemu/units.h" |
e2de2c49 | 44 | #include "qemu/thread-context.h" |
c1b0b93b | 45 | |
cbcfa041 PB |
46 | #ifdef CONFIG_LINUX |
47 | #include <sys/syscall.h> | |
48 | #endif | |
cbcfa041 | 49 | |
41975b26 | 50 | #ifdef __FreeBSD__ |
9548a891 | 51 | #include <sys/thr.h> |
06680b15 | 52 | #include <sys/user.h> |
7dc9ae43 | 53 | #include <libutil.h> |
41975b26 AF |
54 | #endif |
55 | ||
094611b4 | 56 | #ifdef __NetBSD__ |
9548a891 | 57 | #include <lwp.h> |
094611b4 KR |
58 | #endif |
59 | ||
a9c94277 | 60 | #include "qemu/mmap-alloc.h" |
794e8f30 | 61 | |
dfd0dcc7 | 62 | #define MAX_MEM_PREALLOC_THREAD_COUNT 16 |
1e356fc1 | 63 | |
dba50678 DH |
64 | struct MemsetThread; |
65 | ||
66 | typedef struct MemsetContext { | |
67 | bool all_threads_created; | |
68 | bool any_thread_failed; | |
69 | struct MemsetThread *threads; | |
70 | int num_threads; | |
71 | } MemsetContext; | |
72 | ||
1e356fc1 JK |
73 | struct MemsetThread { |
74 | char *addr; | |
e947d47d SW |
75 | size_t numpages; |
76 | size_t hpagesize; | |
1e356fc1 JK |
77 | QemuThread pgthread; |
78 | sigjmp_buf env; | |
dba50678 | 79 | MemsetContext *context; |
1e356fc1 JK |
80 | }; |
81 | typedef struct MemsetThread MemsetThread; | |
82 | ||
dba50678 DH |
83 | /* used by sigbus_handler() */ |
84 | static MemsetContext *sigbus_memset_context; | |
29b838c0 | 85 | struct sigaction sigbus_oldact; |
a960d664 | 86 | static QemuMutex sigbus_mutex; |
1e356fc1 | 87 | |
037fb5eb | 88 | static QemuMutex page_mutex; |
89 | static QemuCond page_cond; | |
037fb5eb | 90 | |
cbcfa041 PB |
91 | int qemu_get_thread_id(void) |
92 | { | |
93 | #if defined(__linux__) | |
94 | return syscall(SYS_gettid); | |
9548a891 DC |
95 | #elif defined(__FreeBSD__) |
96 | /* thread id is up to INT_MAX */ | |
97 | long tid; | |
98 | thr_self(&tid); | |
99 | return (int)tid; | |
100 | #elif defined(__NetBSD__) | |
101 | return _lwp_self(); | |
8edbca51 DC |
102 | #elif defined(__OpenBSD__) |
103 | return getthrid(); | |
cbcfa041 PB |
104 | #else |
105 | return getpid(); | |
106 | #endif | |
107 | } | |
f97742d0 AR |
108 | |
109 | int qemu_daemon(int nochdir, int noclose) | |
110 | { | |
111 | return daemon(nochdir, noclose); | |
112 | } | |
113 | ||
9e6bdef2 MAL |
114 | bool qemu_write_pidfile(const char *path, Error **errp) |
115 | { | |
116 | int fd; | |
117 | char pidstr[32]; | |
118 | ||
119 | while (1) { | |
120 | struct stat a, b; | |
35f7f3fb MAL |
121 | struct flock lock = { |
122 | .l_type = F_WRLCK, | |
123 | .l_whence = SEEK_SET, | |
124 | .l_len = 0, | |
125 | }; | |
9e6bdef2 | 126 | |
1b34d08f | 127 | fd = qemu_create(path, O_WRONLY, S_IRUSR | S_IWUSR, errp); |
9e6bdef2 | 128 | if (fd == -1) { |
9e6bdef2 MAL |
129 | return false; |
130 | } | |
131 | ||
132 | if (fstat(fd, &b) < 0) { | |
133 | error_setg_errno(errp, errno, "Cannot stat file"); | |
134 | goto fail_close; | |
135 | } | |
136 | ||
35f7f3fb | 137 | if (fcntl(fd, F_SETLK, &lock)) { |
9e6bdef2 MAL |
138 | error_setg_errno(errp, errno, "Cannot lock pid file"); |
139 | goto fail_close; | |
140 | } | |
141 | ||
142 | /* | |
143 | * Now make sure the path we locked is the same one that now | |
144 | * exists on the filesystem. | |
145 | */ | |
146 | if (stat(path, &a) < 0) { | |
147 | /* | |
148 | * PID file disappeared, someone else must be racing with | |
149 | * us, so try again. | |
150 | */ | |
151 | close(fd); | |
152 | continue; | |
153 | } | |
154 | ||
155 | if (a.st_ino == b.st_ino) { | |
156 | break; | |
157 | } | |
158 | ||
159 | /* | |
160 | * PID file was recreated, someone else must be racing with | |
161 | * us, so try again. | |
162 | */ | |
163 | close(fd); | |
164 | } | |
165 | ||
166 | if (ftruncate(fd, 0) < 0) { | |
167 | error_setg_errno(errp, errno, "Failed to truncate pid file"); | |
168 | goto fail_unlink; | |
169 | } | |
170 | ||
171 | snprintf(pidstr, sizeof(pidstr), FMT_pid "\n", getpid()); | |
96eb9b2b | 172 | if (qemu_write_full(fd, pidstr, strlen(pidstr)) != strlen(pidstr)) { |
9e6bdef2 MAL |
173 | error_setg(errp, "Failed to write pid file"); |
174 | goto fail_unlink; | |
175 | } | |
176 | ||
177 | return true; | |
178 | ||
179 | fail_unlink: | |
180 | unlink(path); | |
181 | fail_close: | |
182 | close(fd); | |
183 | return false; | |
184 | } | |
185 | ||
c1b0b93b | 186 | /* alloc shared memory pages */ |
8dbe22c6 DH |
187 | void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment, bool shared, |
188 | bool noreserve) | |
c1b0b93b | 189 | { |
8dbe22c6 DH |
190 | const uint32_t qemu_map_flags = (shared ? QEMU_MAP_SHARED : 0) | |
191 | (noreserve ? QEMU_MAP_NORESERVE : 0); | |
36b58628 | 192 | size_t align = QEMU_VMALLOC_ALIGN; |
b444f5c0 | 193 | void *ptr = qemu_ram_mmap(-1, size, align, qemu_map_flags, 0); |
36b58628 | 194 | |
7dda5dc8 | 195 | if (ptr == MAP_FAILED) { |
39228250 | 196 | return NULL; |
c2a8238a | 197 | } |
c2a8238a | 198 | |
a2b257d6 IM |
199 | if (alignment) { |
200 | *alignment = align; | |
201 | } | |
c2dfc5ba | 202 | |
6eebf958 | 203 | trace_qemu_anon_ram_alloc(size, ptr); |
c7f4111a | 204 | return ptr; |
c1b0b93b JS |
205 | } |
206 | ||
e7a09b92 PB |
207 | void qemu_anon_ram_free(void *ptr, size_t size) |
208 | { | |
209 | trace_qemu_anon_ram_free(ptr, size); | |
53adb9d4 | 210 | qemu_ram_munmap(-1, ptr, size); |
e7a09b92 PB |
211 | } |
212 | ||
ff5927ba | 213 | void qemu_socket_set_block(int fd) |
154b9a0c | 214 | { |
22e135fc | 215 | g_unix_set_fd_nonblocking(fd, false, NULL); |
154b9a0c PB |
216 | } |
217 | ||
ff5927ba | 218 | int qemu_socket_try_set_nonblock(int fd) |
9549e764 | 219 | { |
22e135fc | 220 | return g_unix_set_fd_nonblocking(fd, true, NULL) ? 0 : -errno; |
894022e6 LV |
221 | } |
222 | ||
ff5927ba | 223 | void qemu_socket_set_nonblock(int fd) |
894022e6 LV |
224 | { |
225 | int f; | |
ff5927ba | 226 | f = qemu_socket_try_set_nonblock(fd); |
894022e6 | 227 | assert(f == 0); |
9549e764 JS |
228 | } |
229 | ||
606600a1 SO |
230 | int socket_set_fast_reuse(int fd) |
231 | { | |
232 | int val = 1, ret; | |
233 | ||
234 | ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, | |
235 | (const char *)&val, sizeof(val)); | |
236 | ||
237 | assert(ret == 0); | |
238 | ||
239 | return ret; | |
240 | } | |
241 | ||
9549e764 JS |
242 | void qemu_set_cloexec(int fd) |
243 | { | |
244 | int f; | |
245 | f = fcntl(fd, F_GETFD); | |
7e6478e7 SS |
246 | assert(f != -1); |
247 | f = fcntl(fd, F_SETFD, f | FD_CLOEXEC); | |
248 | assert(f != -1); | |
9549e764 | 249 | } |
70e72ce4 | 250 | |
3c63b4e9 GT |
251 | int qemu_socketpair(int domain, int type, int protocol, int sv[2]) |
252 | { | |
253 | int ret; | |
254 | ||
255 | #ifdef SOCK_CLOEXEC | |
256 | ret = socketpair(domain, type | SOCK_CLOEXEC, protocol, sv); | |
257 | if (ret != -1 || errno != EINVAL) { | |
258 | return ret; | |
259 | } | |
260 | #endif | |
261 | ret = socketpair(domain, type, protocol, sv);; | |
262 | if (ret == 0) { | |
263 | qemu_set_cloexec(sv[0]); | |
264 | qemu_set_cloexec(sv[1]); | |
265 | } | |
266 | ||
267 | return ret; | |
268 | } | |
269 | ||
e2ea3515 | 270 | char * |
1fbf2665 | 271 | qemu_get_local_state_dir(void) |
e2ea3515 | 272 | { |
1fbf2665 | 273 | return get_relocated_path(CONFIG_QEMU_LOCALSTATEDIR); |
e2ea3515 | 274 | } |
13401ba0 SH |
275 | |
276 | void qemu_set_tty_echo(int fd, bool echo) | |
277 | { | |
278 | struct termios tty; | |
279 | ||
280 | tcgetattr(fd, &tty); | |
281 | ||
282 | if (echo) { | |
283 | tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN; | |
284 | } else { | |
285 | tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN); | |
286 | } | |
287 | ||
288 | tcsetattr(fd, TCSANOW, &tty); | |
289 | } | |
10f5bff6 | 290 | |
29b838c0 DH |
291 | #ifdef CONFIG_LINUX |
292 | static void sigbus_handler(int signal, siginfo_t *siginfo, void *ctx) | |
293 | #else /* CONFIG_LINUX */ | |
38183310 | 294 | static void sigbus_handler(int signal) |
29b838c0 | 295 | #endif /* CONFIG_LINUX */ |
38183310 | 296 | { |
1e356fc1 | 297 | int i; |
dba50678 DH |
298 | |
299 | if (sigbus_memset_context) { | |
300 | for (i = 0; i < sigbus_memset_context->num_threads; i++) { | |
301 | MemsetThread *thread = &sigbus_memset_context->threads[i]; | |
302 | ||
303 | if (qemu_thread_is_self(&thread->pgthread)) { | |
304 | siglongjmp(thread->env, 1); | |
1e356fc1 JK |
305 | } |
306 | } | |
307 | } | |
29b838c0 DH |
308 | |
309 | #ifdef CONFIG_LINUX | |
310 | /* | |
311 | * We assume that the MCE SIGBUS handler could have been registered. We | |
312 | * should never receive BUS_MCEERR_AO on any of our threads, but only on | |
313 | * the main thread registered for PR_MCE_KILL_EARLY. Further, we should not | |
314 | * receive BUS_MCEERR_AR triggered by action of other threads on one of | |
315 | * our threads. So, no need to check for unrelated SIGBUS when seeing one | |
316 | * for our threads. | |
317 | * | |
318 | * We will forward to the MCE handler, which will either handle the SIGBUS | |
319 | * or reinstall the default SIGBUS handler and reraise the SIGBUS. The | |
320 | * default SIGBUS handler will crash the process, so we don't care. | |
321 | */ | |
322 | if (sigbus_oldact.sa_flags & SA_SIGINFO) { | |
323 | sigbus_oldact.sa_sigaction(signal, siginfo, ctx); | |
324 | return; | |
325 | } | |
326 | #endif /* CONFIG_LINUX */ | |
6556aadc | 327 | warn_report("qemu_prealloc_mem: unrelated SIGBUS detected and ignored"); |
38183310 PB |
328 | } |
329 | ||
1e356fc1 JK |
330 | static void *do_touch_pages(void *arg) |
331 | { | |
332 | MemsetThread *memset_args = (MemsetThread *)arg; | |
1e356fc1 | 333 | sigset_t set, oldset; |
6c427ab9 | 334 | int ret = 0; |
1e356fc1 | 335 | |
037fb5eb | 336 | /* |
337 | * On Linux, the page faults from the loop below can cause mmap_sem | |
338 | * contention with allocation of the thread stacks. Do not start | |
339 | * clearing until all threads have been created. | |
340 | */ | |
341 | qemu_mutex_lock(&page_mutex); | |
dba50678 | 342 | while (!memset_args->context->all_threads_created) { |
037fb5eb | 343 | qemu_cond_wait(&page_cond, &page_mutex); |
344 | } | |
345 | qemu_mutex_unlock(&page_mutex); | |
346 | ||
1e356fc1 JK |
347 | /* unblock SIGBUS */ |
348 | sigemptyset(&set); | |
349 | sigaddset(&set, SIGBUS); | |
350 | pthread_sigmask(SIG_UNBLOCK, &set, &oldset); | |
351 | ||
352 | if (sigsetjmp(memset_args->env, 1)) { | |
6c427ab9 | 353 | ret = -EFAULT; |
1e356fc1 | 354 | } else { |
e947d47d SW |
355 | char *addr = memset_args->addr; |
356 | size_t numpages = memset_args->numpages; | |
357 | size_t hpagesize = memset_args->hpagesize; | |
358 | size_t i; | |
1e356fc1 | 359 | for (i = 0; i < numpages; i++) { |
9dc44aa5 DB |
360 | /* |
361 | * Read & write back the same value, so we don't | |
362 | * corrupt existing user/app data that might be | |
363 | * stored. | |
364 | * | |
365 | * 'volatile' to stop compiler optimizing this away | |
366 | * to a no-op | |
9dc44aa5 DB |
367 | */ |
368 | *(volatile char *)addr = *addr; | |
1e356fc1 JK |
369 | addr += hpagesize; |
370 | } | |
371 | } | |
372 | pthread_sigmask(SIG_SETMASK, &oldset, NULL); | |
6c427ab9 | 373 | return (void *)(uintptr_t)ret; |
1e356fc1 JK |
374 | } |
375 | ||
a384bfa3 DH |
376 | static void *do_madv_populate_write_pages(void *arg) |
377 | { | |
378 | MemsetThread *memset_args = (MemsetThread *)arg; | |
379 | const size_t size = memset_args->numpages * memset_args->hpagesize; | |
380 | char * const addr = memset_args->addr; | |
381 | int ret = 0; | |
382 | ||
383 | /* See do_touch_pages(). */ | |
384 | qemu_mutex_lock(&page_mutex); | |
dba50678 | 385 | while (!memset_args->context->all_threads_created) { |
a384bfa3 DH |
386 | qemu_cond_wait(&page_cond, &page_mutex); |
387 | } | |
388 | qemu_mutex_unlock(&page_mutex); | |
389 | ||
390 | if (size && qemu_madvise(addr, size, QEMU_MADV_POPULATE_WRITE)) { | |
391 | ret = -errno; | |
392 | } | |
393 | return (void *)(uintptr_t)ret; | |
394 | } | |
395 | ||
89aec641 | 396 | static inline int get_memset_num_threads(size_t hpagesize, size_t numpages, |
6556aadc | 397 | int max_threads) |
dfd0dcc7 JK |
398 | { |
399 | long host_procs = sysconf(_SC_NPROCESSORS_ONLN); | |
400 | int ret = 1; | |
401 | ||
402 | if (host_procs > 0) { | |
6556aadc | 403 | ret = MIN(MIN(host_procs, MAX_MEM_PREALLOC_THREAD_COUNT), max_threads); |
dfd0dcc7 | 404 | } |
89aec641 DH |
405 | |
406 | /* Especially with gigantic pages, don't create more threads than pages. */ | |
407 | ret = MIN(ret, numpages); | |
408 | /* Don't start threads to prealloc comparatively little memory. */ | |
409 | ret = MIN(ret, MAX(1, hpagesize * numpages / (64 * MiB))); | |
410 | ||
dfd0dcc7 JK |
411 | /* In case sysconf() fails, we fall back to single threaded */ |
412 | return ret; | |
413 | } | |
414 | ||
6c427ab9 | 415 | static int touch_all_pages(char *area, size_t hpagesize, size_t numpages, |
e04a34e5 DH |
416 | int max_threads, ThreadContext *tc, |
417 | bool use_madv_populate_write) | |
1e356fc1 | 418 | { |
78b3f67a | 419 | static gsize initialized = 0; |
dba50678 | 420 | MemsetContext context = { |
6556aadc | 421 | .num_threads = get_memset_num_threads(hpagesize, numpages, max_threads), |
dba50678 | 422 | }; |
037fb5eb | 423 | size_t numpages_per_thread, leftover; |
a384bfa3 | 424 | void *(*touch_fn)(void *); |
6c427ab9 | 425 | int ret = 0, i = 0; |
1e356fc1 | 426 | char *addr = area; |
1e356fc1 | 427 | |
78b3f67a PB |
428 | if (g_once_init_enter(&initialized)) { |
429 | qemu_mutex_init(&page_mutex); | |
430 | qemu_cond_init(&page_cond); | |
431 | g_once_init_leave(&initialized, 1); | |
432 | } | |
433 | ||
a384bfa3 | 434 | if (use_madv_populate_write) { |
ac86e5c3 DH |
435 | /* Avoid creating a single thread for MADV_POPULATE_WRITE */ |
436 | if (context.num_threads == 1) { | |
437 | if (qemu_madvise(area, hpagesize * numpages, | |
438 | QEMU_MADV_POPULATE_WRITE)) { | |
439 | return -errno; | |
440 | } | |
441 | return 0; | |
442 | } | |
a384bfa3 DH |
443 | touch_fn = do_madv_populate_write_pages; |
444 | } else { | |
445 | touch_fn = do_touch_pages; | |
446 | } | |
447 | ||
dba50678 DH |
448 | context.threads = g_new0(MemsetThread, context.num_threads); |
449 | numpages_per_thread = numpages / context.num_threads; | |
450 | leftover = numpages % context.num_threads; | |
451 | for (i = 0; i < context.num_threads; i++) { | |
452 | context.threads[i].addr = addr; | |
453 | context.threads[i].numpages = numpages_per_thread + (i < leftover); | |
454 | context.threads[i].hpagesize = hpagesize; | |
455 | context.threads[i].context = &context; | |
e04a34e5 DH |
456 | if (tc) { |
457 | thread_context_create_thread(tc, &context.threads[i].pgthread, | |
458 | "touch_pages", | |
459 | touch_fn, &context.threads[i], | |
460 | QEMU_THREAD_JOINABLE); | |
461 | } else { | |
462 | qemu_thread_create(&context.threads[i].pgthread, "touch_pages", | |
463 | touch_fn, &context.threads[i], | |
464 | QEMU_THREAD_JOINABLE); | |
465 | } | |
dba50678 DH |
466 | addr += context.threads[i].numpages * hpagesize; |
467 | } | |
468 | ||
469 | if (!use_madv_populate_write) { | |
470 | sigbus_memset_context = &context; | |
1e356fc1 | 471 | } |
278fb162 B |
472 | |
473 | qemu_mutex_lock(&page_mutex); | |
dba50678 | 474 | context.all_threads_created = true; |
037fb5eb | 475 | qemu_cond_broadcast(&page_cond); |
278fb162 | 476 | qemu_mutex_unlock(&page_mutex); |
037fb5eb | 477 | |
dba50678 DH |
478 | for (i = 0; i < context.num_threads; i++) { |
479 | int tmp = (uintptr_t)qemu_thread_join(&context.threads[i].pgthread); | |
6c427ab9 DH |
480 | |
481 | if (tmp) { | |
482 | ret = tmp; | |
483 | } | |
1e356fc1 | 484 | } |
dba50678 DH |
485 | |
486 | if (!use_madv_populate_write) { | |
487 | sigbus_memset_context = NULL; | |
488 | } | |
489 | g_free(context.threads); | |
1e356fc1 | 490 | |
6c427ab9 | 491 | return ret; |
1e356fc1 JK |
492 | } |
493 | ||
a384bfa3 DH |
494 | static bool madv_populate_write_possible(char *area, size_t pagesize) |
495 | { | |
496 | return !qemu_madvise(area, pagesize, QEMU_MADV_POPULATE_WRITE) || | |
497 | errno != EINVAL; | |
498 | } | |
499 | ||
6556aadc | 500 | void qemu_prealloc_mem(int fd, char *area, size_t sz, int max_threads, |
e04a34e5 | 501 | ThreadContext *tc, Error **errp) |
38183310 | 502 | { |
a960d664 | 503 | static gsize initialized; |
b7bf8f56 | 504 | int ret; |
1e356fc1 | 505 | size_t hpagesize = qemu_fd_getpagesize(fd); |
6556aadc | 506 | size_t numpages = DIV_ROUND_UP(sz, hpagesize); |
a384bfa3 | 507 | bool use_madv_populate_write; |
29b838c0 | 508 | struct sigaction act; |
38183310 | 509 | |
a384bfa3 DH |
510 | /* |
511 | * Sense on every invocation, as MADV_POPULATE_WRITE cannot be used for | |
512 | * some special mappings, such as mapping /dev/mem. | |
513 | */ | |
514 | use_madv_populate_write = madv_populate_write_possible(area, hpagesize); | |
515 | ||
516 | if (!use_madv_populate_write) { | |
a960d664 DH |
517 | if (g_once_init_enter(&initialized)) { |
518 | qemu_mutex_init(&sigbus_mutex); | |
519 | g_once_init_leave(&initialized, 1); | |
520 | } | |
521 | ||
522 | qemu_mutex_lock(&sigbus_mutex); | |
a384bfa3 | 523 | memset(&act, 0, sizeof(act)); |
29b838c0 DH |
524 | #ifdef CONFIG_LINUX |
525 | act.sa_sigaction = &sigbus_handler; | |
526 | act.sa_flags = SA_SIGINFO; | |
527 | #else /* CONFIG_LINUX */ | |
a384bfa3 DH |
528 | act.sa_handler = &sigbus_handler; |
529 | act.sa_flags = 0; | |
29b838c0 | 530 | #endif /* CONFIG_LINUX */ |
a384bfa3 | 531 | |
29b838c0 | 532 | ret = sigaction(SIGBUS, &act, &sigbus_oldact); |
a384bfa3 | 533 | if (ret) { |
dd4fc605 | 534 | qemu_mutex_unlock(&sigbus_mutex); |
a384bfa3 | 535 | error_setg_errno(errp, errno, |
6556aadc | 536 | "qemu_prealloc_mem: failed to install signal handler"); |
a384bfa3 DH |
537 | return; |
538 | } | |
38183310 PB |
539 | } |
540 | ||
1e356fc1 | 541 | /* touch pages simultaneously */ |
e04a34e5 | 542 | ret = touch_all_pages(area, hpagesize, numpages, max_threads, tc, |
a384bfa3 | 543 | use_madv_populate_write); |
6c427ab9 DH |
544 | if (ret) { |
545 | error_setg_errno(errp, -ret, | |
6556aadc | 546 | "qemu_prealloc_mem: preallocating memory failed"); |
056b68af | 547 | } |
38183310 | 548 | |
a384bfa3 | 549 | if (!use_madv_populate_write) { |
29b838c0 | 550 | ret = sigaction(SIGBUS, &sigbus_oldact, NULL); |
a384bfa3 DH |
551 | if (ret) { |
552 | /* Terminate QEMU since it can't recover from error */ | |
6556aadc | 553 | perror("qemu_prealloc_mem: failed to reinstall signal handler"); |
a384bfa3 DH |
554 | exit(1); |
555 | } | |
a960d664 | 556 | qemu_mutex_unlock(&sigbus_mutex); |
b7bf8f56 | 557 | } |
38183310 | 558 | } |
d57e4e48 | 559 | |
7dc9ae43 MP |
560 | char *qemu_get_pid_name(pid_t pid) |
561 | { | |
562 | char *name = NULL; | |
563 | ||
564 | #if defined(__FreeBSD__) | |
565 | /* BSDs don't have /proc, but they provide a nice substitute */ | |
566 | struct kinfo_proc *proc = kinfo_getproc(pid); | |
567 | ||
568 | if (proc) { | |
569 | name = g_strdup(proc->ki_comm); | |
570 | free(proc); | |
571 | } | |
572 | #else | |
573 | /* Assume a system with reasonable procfs */ | |
574 | char *pid_path; | |
575 | size_t len; | |
576 | ||
577 | pid_path = g_strdup_printf("/proc/%d/cmdline", pid); | |
578 | g_file_get_contents(pid_path, &name, &len, NULL); | |
579 | g_free(pid_path); | |
580 | #endif | |
581 | ||
582 | return name; | |
583 | } | |
584 | ||
585 | ||
8737d9e0 PL |
586 | void *qemu_alloc_stack(size_t *sz) |
587 | { | |
588 | void *ptr, *guardpage; | |
fc3d1bad | 589 | int flags; |
7d992e4d PL |
590 | #ifdef CONFIG_DEBUG_STACK_USAGE |
591 | void *ptr2; | |
592 | #endif | |
8e3b0cbb | 593 | size_t pagesz = qemu_real_host_page_size(); |
8737d9e0 PL |
594 | #ifdef _SC_THREAD_STACK_MIN |
595 | /* avoid stacks smaller than _SC_THREAD_STACK_MIN */ | |
596 | long min_stack_sz = sysconf(_SC_THREAD_STACK_MIN); | |
597 | *sz = MAX(MAX(min_stack_sz, 0), *sz); | |
598 | #endif | |
599 | /* adjust stack size to a multiple of the page size */ | |
600 | *sz = ROUND_UP(*sz, pagesz); | |
601 | /* allocate one extra page for the guard page */ | |
602 | *sz += pagesz; | |
603 | ||
fc3d1bad BS |
604 | flags = MAP_PRIVATE | MAP_ANONYMOUS; |
605 | #if defined(MAP_STACK) && defined(__OpenBSD__) | |
606 | /* Only enable MAP_STACK on OpenBSD. Other OS's such as | |
607 | * Linux/FreeBSD/NetBSD have a flag with the same name | |
608 | * but have differing functionality. OpenBSD will SEGV | |
609 | * if it spots execution with a stack pointer pointing | |
610 | * at memory that was not allocated with MAP_STACK. | |
611 | */ | |
612 | flags |= MAP_STACK; | |
613 | #endif | |
614 | ||
615 | ptr = mmap(NULL, *sz, PROT_READ | PROT_WRITE, flags, -1, 0); | |
8737d9e0 | 616 | if (ptr == MAP_FAILED) { |
e916a6e8 | 617 | perror("failed to allocate memory for stack"); |
8737d9e0 PL |
618 | abort(); |
619 | } | |
620 | ||
621 | #if defined(HOST_IA64) | |
622 | /* separate register stack */ | |
623 | guardpage = ptr + (((*sz - pagesz) / 2) & ~pagesz); | |
624 | #elif defined(HOST_HPPA) | |
625 | /* stack grows up */ | |
626 | guardpage = ptr + *sz - pagesz; | |
627 | #else | |
628 | /* stack grows down */ | |
629 | guardpage = ptr; | |
630 | #endif | |
631 | if (mprotect(guardpage, pagesz, PROT_NONE) != 0) { | |
e916a6e8 | 632 | perror("failed to set up stack guard page"); |
8737d9e0 PL |
633 | abort(); |
634 | } | |
635 | ||
7d992e4d PL |
636 | #ifdef CONFIG_DEBUG_STACK_USAGE |
637 | for (ptr2 = ptr + pagesz; ptr2 < ptr + *sz; ptr2 += sizeof(uint32_t)) { | |
638 | *(uint32_t *)ptr2 = 0xdeadbeaf; | |
639 | } | |
640 | #endif | |
641 | ||
8737d9e0 PL |
642 | return ptr; |
643 | } | |
644 | ||
7d992e4d PL |
645 | #ifdef CONFIG_DEBUG_STACK_USAGE |
646 | static __thread unsigned int max_stack_usage; | |
647 | #endif | |
648 | ||
8737d9e0 PL |
649 | void qemu_free_stack(void *stack, size_t sz) |
650 | { | |
7d992e4d PL |
651 | #ifdef CONFIG_DEBUG_STACK_USAGE |
652 | unsigned int usage; | |
653 | void *ptr; | |
654 | ||
8e3b0cbb | 655 | for (ptr = stack + qemu_real_host_page_size(); ptr < stack + sz; |
7d992e4d PL |
656 | ptr += sizeof(uint32_t)) { |
657 | if (*(uint32_t *)ptr != 0xdeadbeaf) { | |
658 | break; | |
659 | } | |
660 | } | |
661 | usage = sz - (uintptr_t) (ptr - stack); | |
662 | if (usage > max_stack_usage) { | |
663 | error_report("thread %d max stack usage increased from %u to %u", | |
664 | qemu_get_thread_id(), max_stack_usage, usage); | |
665 | max_stack_usage = usage; | |
666 | } | |
667 | #endif | |
668 | ||
8737d9e0 PL |
669 | munmap(stack, sz); |
670 | } | |
d98d4072 | 671 | |
c905a368 DB |
672 | /* |
673 | * Disable CFI checks. | |
d02d06f8 | 674 | * We are going to call a signal handler directly. Such handler may or may not |
c905a368 DB |
675 | * have been defined in our binary, so there's no guarantee that the pointer |
676 | * used to set the handler is a cfi-valid pointer. Since the handlers are | |
677 | * stored in kernel memory, changing the handler to an attacker-defined | |
678 | * function requires being able to call a sigaction() syscall, | |
679 | * which is not as easy as overwriting a pointer in memory. | |
680 | */ | |
681 | QEMU_DISABLE_CFI | |
d98d4072 PB |
682 | void sigaction_invoke(struct sigaction *action, |
683 | struct qemu_signalfd_siginfo *info) | |
684 | { | |
02ffa034 | 685 | siginfo_t si = {}; |
d98d4072 PB |
686 | si.si_signo = info->ssi_signo; |
687 | si.si_errno = info->ssi_errno; | |
688 | si.si_code = info->ssi_code; | |
689 | ||
690 | /* Convert the minimal set of fields defined by POSIX. | |
691 | * Positive si_code values are reserved for kernel-generated | |
692 | * signals, where the valid siginfo fields are determined by | |
693 | * the signal number. But according to POSIX, it is unspecified | |
694 | * whether SI_USER and SI_QUEUE have values less than or equal to | |
695 | * zero. | |
696 | */ | |
697 | if (info->ssi_code == SI_USER || info->ssi_code == SI_QUEUE || | |
698 | info->ssi_code <= 0) { | |
699 | /* SIGTERM, etc. */ | |
700 | si.si_pid = info->ssi_pid; | |
701 | si.si_uid = info->ssi_uid; | |
702 | } else if (info->ssi_signo == SIGILL || info->ssi_signo == SIGFPE || | |
703 | info->ssi_signo == SIGSEGV || info->ssi_signo == SIGBUS) { | |
704 | si.si_addr = (void *)(uintptr_t)info->ssi_addr; | |
705 | } else if (info->ssi_signo == SIGCHLD) { | |
706 | si.si_pid = info->ssi_pid; | |
707 | si.si_status = info->ssi_status; | |
708 | si.si_uid = info->ssi_uid; | |
d98d4072 PB |
709 | } |
710 | action->sa_sigaction(info->ssi_signo, &si, NULL); | |
711 | } | |
e47f4765 | 712 | |
ad06ef0e AB |
713 | size_t qemu_get_host_physmem(void) |
714 | { | |
715 | #ifdef _SC_PHYS_PAGES | |
716 | long pages = sysconf(_SC_PHYS_PAGES); | |
717 | if (pages > 0) { | |
8e3b0cbb | 718 | if (pages > SIZE_MAX / qemu_real_host_page_size()) { |
ad06ef0e AB |
719 | return SIZE_MAX; |
720 | } else { | |
8e3b0cbb | 721 | return pages * qemu_real_host_page_size(); |
ad06ef0e AB |
722 | } |
723 | } | |
724 | #endif | |
725 | return 0; | |
726 | } | |
e9c4e0a8 | 727 | |
73991a92 MAL |
728 | int qemu_msync(void *addr, size_t length, int fd) |
729 | { | |
730 | size_t align_mask = ~(qemu_real_host_page_size() - 1); | |
731 | ||
732 | /** | |
733 | * There are no strict reqs as per the length of mapping | |
734 | * to be synced. Still the length needs to follow the address | |
735 | * alignment changes. Additionally - round the size to the multiple | |
736 | * of PAGE_SIZE | |
737 | */ | |
738 | length += ((uintptr_t)addr & (qemu_real_host_page_size() - 1)); | |
739 | length = (length + ~align_mask) & align_mask; | |
740 | ||
741 | addr = (void *)((uintptr_t)addr & align_mask); | |
742 | ||
743 | return msync(addr, length, MS_SYNC); | |
744 | } |