]>
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" |
c905a368 | 43 | #include "qemu/compiler.h" |
89aec641 | 44 | #include "qemu/units.h" |
e2de2c49 | 45 | #include "qemu/thread-context.h" |
c1b0b93b | 46 | |
cbcfa041 PB |
47 | #ifdef CONFIG_LINUX |
48 | #include <sys/syscall.h> | |
49 | #endif | |
cbcfa041 | 50 | |
41975b26 | 51 | #ifdef __FreeBSD__ |
9548a891 | 52 | #include <sys/thr.h> |
06680b15 MAL |
53 | #include <sys/types.h> |
54 | #include <sys/user.h> | |
7dc9ae43 | 55 | #include <libutil.h> |
41975b26 AF |
56 | #endif |
57 | ||
094611b4 | 58 | #ifdef __NetBSD__ |
9548a891 | 59 | #include <lwp.h> |
094611b4 KR |
60 | #endif |
61 | ||
a9c94277 | 62 | #include "qemu/mmap-alloc.h" |
794e8f30 | 63 | |
7d992e4d PL |
64 | #ifdef CONFIG_DEBUG_STACK_USAGE |
65 | #include "qemu/error-report.h" | |
66 | #endif | |
67 | ||
dfd0dcc7 | 68 | #define MAX_MEM_PREALLOC_THREAD_COUNT 16 |
1e356fc1 | 69 | |
dba50678 DH |
70 | struct MemsetThread; |
71 | ||
72 | typedef struct MemsetContext { | |
73 | bool all_threads_created; | |
74 | bool any_thread_failed; | |
75 | struct MemsetThread *threads; | |
76 | int num_threads; | |
77 | } MemsetContext; | |
78 | ||
1e356fc1 JK |
79 | struct MemsetThread { |
80 | char *addr; | |
e947d47d SW |
81 | size_t numpages; |
82 | size_t hpagesize; | |
1e356fc1 JK |
83 | QemuThread pgthread; |
84 | sigjmp_buf env; | |
dba50678 | 85 | MemsetContext *context; |
1e356fc1 JK |
86 | }; |
87 | typedef struct MemsetThread MemsetThread; | |
88 | ||
dba50678 DH |
89 | /* used by sigbus_handler() */ |
90 | static MemsetContext *sigbus_memset_context; | |
29b838c0 | 91 | struct sigaction sigbus_oldact; |
a960d664 | 92 | static QemuMutex sigbus_mutex; |
1e356fc1 | 93 | |
037fb5eb | 94 | static QemuMutex page_mutex; |
95 | static QemuCond page_cond; | |
037fb5eb | 96 | |
cbcfa041 PB |
97 | int qemu_get_thread_id(void) |
98 | { | |
99 | #if defined(__linux__) | |
100 | return syscall(SYS_gettid); | |
9548a891 DC |
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(); | |
8edbca51 DC |
108 | #elif defined(__OpenBSD__) |
109 | return getthrid(); | |
cbcfa041 PB |
110 | #else |
111 | return getpid(); | |
112 | #endif | |
113 | } | |
f97742d0 AR |
114 | |
115 | int qemu_daemon(int nochdir, int noclose) | |
116 | { | |
117 | return daemon(nochdir, noclose); | |
118 | } | |
119 | ||
9e6bdef2 MAL |
120 | bool qemu_write_pidfile(const char *path, Error **errp) |
121 | { | |
122 | int fd; | |
123 | char pidstr[32]; | |
124 | ||
125 | while (1) { | |
126 | struct stat a, b; | |
35f7f3fb MAL |
127 | struct flock lock = { |
128 | .l_type = F_WRLCK, | |
129 | .l_whence = SEEK_SET, | |
130 | .l_len = 0, | |
131 | }; | |
9e6bdef2 | 132 | |
1b34d08f | 133 | fd = qemu_create(path, O_WRONLY, S_IRUSR | S_IWUSR, errp); |
9e6bdef2 | 134 | if (fd == -1) { |
9e6bdef2 MAL |
135 | return false; |
136 | } | |
137 | ||
138 | if (fstat(fd, &b) < 0) { | |
139 | error_setg_errno(errp, errno, "Cannot stat file"); | |
140 | goto fail_close; | |
141 | } | |
142 | ||
35f7f3fb | 143 | if (fcntl(fd, F_SETLK, &lock)) { |
9e6bdef2 MAL |
144 | error_setg_errno(errp, errno, "Cannot lock pid file"); |
145 | goto fail_close; | |
146 | } | |
147 | ||
148 | /* | |
149 | * Now make sure the path we locked is the same one that now | |
150 | * exists on the filesystem. | |
151 | */ | |
152 | if (stat(path, &a) < 0) { | |
153 | /* | |
154 | * PID file disappeared, someone else must be racing with | |
155 | * us, so try again. | |
156 | */ | |
157 | close(fd); | |
158 | continue; | |
159 | } | |
160 | ||
161 | if (a.st_ino == b.st_ino) { | |
162 | break; | |
163 | } | |
164 | ||
165 | /* | |
166 | * PID file was recreated, someone else must be racing with | |
167 | * us, so try again. | |
168 | */ | |
169 | close(fd); | |
170 | } | |
171 | ||
172 | if (ftruncate(fd, 0) < 0) { | |
173 | error_setg_errno(errp, errno, "Failed to truncate pid file"); | |
174 | goto fail_unlink; | |
175 | } | |
176 | ||
177 | snprintf(pidstr, sizeof(pidstr), FMT_pid "\n", getpid()); | |
96eb9b2b | 178 | if (qemu_write_full(fd, pidstr, strlen(pidstr)) != strlen(pidstr)) { |
9e6bdef2 MAL |
179 | error_setg(errp, "Failed to write pid file"); |
180 | goto fail_unlink; | |
181 | } | |
182 | ||
183 | return true; | |
184 | ||
185 | fail_unlink: | |
186 | unlink(path); | |
187 | fail_close: | |
188 | close(fd); | |
189 | return false; | |
190 | } | |
191 | ||
c1b0b93b | 192 | /* alloc shared memory pages */ |
8dbe22c6 DH |
193 | void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment, bool shared, |
194 | bool noreserve) | |
c1b0b93b | 195 | { |
8dbe22c6 DH |
196 | const uint32_t qemu_map_flags = (shared ? QEMU_MAP_SHARED : 0) | |
197 | (noreserve ? QEMU_MAP_NORESERVE : 0); | |
36b58628 | 198 | size_t align = QEMU_VMALLOC_ALIGN; |
b444f5c0 | 199 | void *ptr = qemu_ram_mmap(-1, size, align, qemu_map_flags, 0); |
36b58628 | 200 | |
7dda5dc8 | 201 | if (ptr == MAP_FAILED) { |
39228250 | 202 | return NULL; |
c2a8238a | 203 | } |
c2a8238a | 204 | |
a2b257d6 IM |
205 | if (alignment) { |
206 | *alignment = align; | |
207 | } | |
c2dfc5ba | 208 | |
6eebf958 | 209 | trace_qemu_anon_ram_alloc(size, ptr); |
c7f4111a | 210 | return ptr; |
c1b0b93b JS |
211 | } |
212 | ||
e7a09b92 PB |
213 | void qemu_anon_ram_free(void *ptr, size_t size) |
214 | { | |
215 | trace_qemu_anon_ram_free(ptr, size); | |
53adb9d4 | 216 | qemu_ram_munmap(-1, ptr, size); |
e7a09b92 PB |
217 | } |
218 | ||
ff5927ba | 219 | void qemu_socket_set_block(int fd) |
154b9a0c | 220 | { |
22e135fc | 221 | g_unix_set_fd_nonblocking(fd, false, NULL); |
154b9a0c PB |
222 | } |
223 | ||
ff5927ba | 224 | int qemu_socket_try_set_nonblock(int fd) |
9549e764 | 225 | { |
22e135fc | 226 | return g_unix_set_fd_nonblocking(fd, true, NULL) ? 0 : -errno; |
894022e6 LV |
227 | } |
228 | ||
ff5927ba | 229 | void qemu_socket_set_nonblock(int fd) |
894022e6 LV |
230 | { |
231 | int f; | |
ff5927ba | 232 | f = qemu_socket_try_set_nonblock(fd); |
894022e6 | 233 | assert(f == 0); |
9549e764 JS |
234 | } |
235 | ||
606600a1 SO |
236 | int socket_set_fast_reuse(int fd) |
237 | { | |
238 | int val = 1, ret; | |
239 | ||
240 | ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, | |
241 | (const char *)&val, sizeof(val)); | |
242 | ||
243 | assert(ret == 0); | |
244 | ||
245 | return ret; | |
246 | } | |
247 | ||
9549e764 JS |
248 | void qemu_set_cloexec(int fd) |
249 | { | |
250 | int f; | |
251 | f = fcntl(fd, F_GETFD); | |
7e6478e7 SS |
252 | assert(f != -1); |
253 | f = fcntl(fd, F_SETFD, f | FD_CLOEXEC); | |
254 | assert(f != -1); | |
9549e764 | 255 | } |
70e72ce4 | 256 | |
3c63b4e9 GT |
257 | int qemu_socketpair(int domain, int type, int protocol, int sv[2]) |
258 | { | |
259 | int ret; | |
260 | ||
261 | #ifdef SOCK_CLOEXEC | |
262 | ret = socketpair(domain, type | SOCK_CLOEXEC, protocol, sv); | |
263 | if (ret != -1 || errno != EINVAL) { | |
264 | return ret; | |
265 | } | |
266 | #endif | |
267 | ret = socketpair(domain, type, protocol, sv);; | |
268 | if (ret == 0) { | |
269 | qemu_set_cloexec(sv[0]); | |
270 | qemu_set_cloexec(sv[1]); | |
271 | } | |
272 | ||
273 | return ret; | |
274 | } | |
275 | ||
e2ea3515 | 276 | char * |
1fbf2665 | 277 | qemu_get_local_state_dir(void) |
e2ea3515 | 278 | { |
1fbf2665 | 279 | return get_relocated_path(CONFIG_QEMU_LOCALSTATEDIR); |
e2ea3515 | 280 | } |
13401ba0 SH |
281 | |
282 | void qemu_set_tty_echo(int fd, bool echo) | |
283 | { | |
284 | struct termios tty; | |
285 | ||
286 | tcgetattr(fd, &tty); | |
287 | ||
288 | if (echo) { | |
289 | tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN; | |
290 | } else { | |
291 | tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN); | |
292 | } | |
293 | ||
294 | tcsetattr(fd, TCSANOW, &tty); | |
295 | } | |
10f5bff6 | 296 | |
29b838c0 DH |
297 | #ifdef CONFIG_LINUX |
298 | static void sigbus_handler(int signal, siginfo_t *siginfo, void *ctx) | |
299 | #else /* CONFIG_LINUX */ | |
38183310 | 300 | static void sigbus_handler(int signal) |
29b838c0 | 301 | #endif /* CONFIG_LINUX */ |
38183310 | 302 | { |
1e356fc1 | 303 | int i; |
dba50678 DH |
304 | |
305 | if (sigbus_memset_context) { | |
306 | for (i = 0; i < sigbus_memset_context->num_threads; i++) { | |
307 | MemsetThread *thread = &sigbus_memset_context->threads[i]; | |
308 | ||
309 | if (qemu_thread_is_self(&thread->pgthread)) { | |
310 | siglongjmp(thread->env, 1); | |
1e356fc1 JK |
311 | } |
312 | } | |
313 | } | |
29b838c0 DH |
314 | |
315 | #ifdef CONFIG_LINUX | |
316 | /* | |
317 | * We assume that the MCE SIGBUS handler could have been registered. We | |
318 | * should never receive BUS_MCEERR_AO on any of our threads, but only on | |
319 | * the main thread registered for PR_MCE_KILL_EARLY. Further, we should not | |
320 | * receive BUS_MCEERR_AR triggered by action of other threads on one of | |
321 | * our threads. So, no need to check for unrelated SIGBUS when seeing one | |
322 | * for our threads. | |
323 | * | |
324 | * We will forward to the MCE handler, which will either handle the SIGBUS | |
325 | * or reinstall the default SIGBUS handler and reraise the SIGBUS. The | |
326 | * default SIGBUS handler will crash the process, so we don't care. | |
327 | */ | |
328 | if (sigbus_oldact.sa_flags & SA_SIGINFO) { | |
329 | sigbus_oldact.sa_sigaction(signal, siginfo, ctx); | |
330 | return; | |
331 | } | |
332 | #endif /* CONFIG_LINUX */ | |
6556aadc | 333 | warn_report("qemu_prealloc_mem: unrelated SIGBUS detected and ignored"); |
38183310 PB |
334 | } |
335 | ||
1e356fc1 JK |
336 | static void *do_touch_pages(void *arg) |
337 | { | |
338 | MemsetThread *memset_args = (MemsetThread *)arg; | |
1e356fc1 | 339 | sigset_t set, oldset; |
6c427ab9 | 340 | int ret = 0; |
1e356fc1 | 341 | |
037fb5eb | 342 | /* |
343 | * On Linux, the page faults from the loop below can cause mmap_sem | |
344 | * contention with allocation of the thread stacks. Do not start | |
345 | * clearing until all threads have been created. | |
346 | */ | |
347 | qemu_mutex_lock(&page_mutex); | |
dba50678 | 348 | while (!memset_args->context->all_threads_created) { |
037fb5eb | 349 | qemu_cond_wait(&page_cond, &page_mutex); |
350 | } | |
351 | qemu_mutex_unlock(&page_mutex); | |
352 | ||
1e356fc1 JK |
353 | /* unblock SIGBUS */ |
354 | sigemptyset(&set); | |
355 | sigaddset(&set, SIGBUS); | |
356 | pthread_sigmask(SIG_UNBLOCK, &set, &oldset); | |
357 | ||
358 | if (sigsetjmp(memset_args->env, 1)) { | |
6c427ab9 | 359 | ret = -EFAULT; |
1e356fc1 | 360 | } else { |
e947d47d SW |
361 | char *addr = memset_args->addr; |
362 | size_t numpages = memset_args->numpages; | |
363 | size_t hpagesize = memset_args->hpagesize; | |
364 | size_t i; | |
1e356fc1 | 365 | for (i = 0; i < numpages; i++) { |
9dc44aa5 DB |
366 | /* |
367 | * Read & write back the same value, so we don't | |
368 | * corrupt existing user/app data that might be | |
369 | * stored. | |
370 | * | |
371 | * 'volatile' to stop compiler optimizing this away | |
372 | * to a no-op | |
9dc44aa5 DB |
373 | */ |
374 | *(volatile char *)addr = *addr; | |
1e356fc1 JK |
375 | addr += hpagesize; |
376 | } | |
377 | } | |
378 | pthread_sigmask(SIG_SETMASK, &oldset, NULL); | |
6c427ab9 | 379 | return (void *)(uintptr_t)ret; |
1e356fc1 JK |
380 | } |
381 | ||
a384bfa3 DH |
382 | static void *do_madv_populate_write_pages(void *arg) |
383 | { | |
384 | MemsetThread *memset_args = (MemsetThread *)arg; | |
385 | const size_t size = memset_args->numpages * memset_args->hpagesize; | |
386 | char * const addr = memset_args->addr; | |
387 | int ret = 0; | |
388 | ||
389 | /* See do_touch_pages(). */ | |
390 | qemu_mutex_lock(&page_mutex); | |
dba50678 | 391 | while (!memset_args->context->all_threads_created) { |
a384bfa3 DH |
392 | qemu_cond_wait(&page_cond, &page_mutex); |
393 | } | |
394 | qemu_mutex_unlock(&page_mutex); | |
395 | ||
396 | if (size && qemu_madvise(addr, size, QEMU_MADV_POPULATE_WRITE)) { | |
397 | ret = -errno; | |
398 | } | |
399 | return (void *)(uintptr_t)ret; | |
400 | } | |
401 | ||
89aec641 | 402 | static inline int get_memset_num_threads(size_t hpagesize, size_t numpages, |
6556aadc | 403 | int max_threads) |
dfd0dcc7 JK |
404 | { |
405 | long host_procs = sysconf(_SC_NPROCESSORS_ONLN); | |
406 | int ret = 1; | |
407 | ||
408 | if (host_procs > 0) { | |
6556aadc | 409 | ret = MIN(MIN(host_procs, MAX_MEM_PREALLOC_THREAD_COUNT), max_threads); |
dfd0dcc7 | 410 | } |
89aec641 DH |
411 | |
412 | /* Especially with gigantic pages, don't create more threads than pages. */ | |
413 | ret = MIN(ret, numpages); | |
414 | /* Don't start threads to prealloc comparatively little memory. */ | |
415 | ret = MIN(ret, MAX(1, hpagesize * numpages / (64 * MiB))); | |
416 | ||
dfd0dcc7 JK |
417 | /* In case sysconf() fails, we fall back to single threaded */ |
418 | return ret; | |
419 | } | |
420 | ||
6c427ab9 | 421 | static int touch_all_pages(char *area, size_t hpagesize, size_t numpages, |
e04a34e5 DH |
422 | int max_threads, ThreadContext *tc, |
423 | bool use_madv_populate_write) | |
1e356fc1 | 424 | { |
78b3f67a | 425 | static gsize initialized = 0; |
dba50678 | 426 | MemsetContext context = { |
6556aadc | 427 | .num_threads = get_memset_num_threads(hpagesize, numpages, max_threads), |
dba50678 | 428 | }; |
037fb5eb | 429 | size_t numpages_per_thread, leftover; |
a384bfa3 | 430 | void *(*touch_fn)(void *); |
6c427ab9 | 431 | int ret = 0, i = 0; |
1e356fc1 | 432 | char *addr = area; |
1e356fc1 | 433 | |
78b3f67a PB |
434 | if (g_once_init_enter(&initialized)) { |
435 | qemu_mutex_init(&page_mutex); | |
436 | qemu_cond_init(&page_cond); | |
437 | g_once_init_leave(&initialized, 1); | |
438 | } | |
439 | ||
a384bfa3 | 440 | if (use_madv_populate_write) { |
ac86e5c3 DH |
441 | /* Avoid creating a single thread for MADV_POPULATE_WRITE */ |
442 | if (context.num_threads == 1) { | |
443 | if (qemu_madvise(area, hpagesize * numpages, | |
444 | QEMU_MADV_POPULATE_WRITE)) { | |
445 | return -errno; | |
446 | } | |
447 | return 0; | |
448 | } | |
a384bfa3 DH |
449 | touch_fn = do_madv_populate_write_pages; |
450 | } else { | |
451 | touch_fn = do_touch_pages; | |
452 | } | |
453 | ||
dba50678 DH |
454 | context.threads = g_new0(MemsetThread, context.num_threads); |
455 | numpages_per_thread = numpages / context.num_threads; | |
456 | leftover = numpages % context.num_threads; | |
457 | for (i = 0; i < context.num_threads; i++) { | |
458 | context.threads[i].addr = addr; | |
459 | context.threads[i].numpages = numpages_per_thread + (i < leftover); | |
460 | context.threads[i].hpagesize = hpagesize; | |
461 | context.threads[i].context = &context; | |
e04a34e5 DH |
462 | if (tc) { |
463 | thread_context_create_thread(tc, &context.threads[i].pgthread, | |
464 | "touch_pages", | |
465 | touch_fn, &context.threads[i], | |
466 | QEMU_THREAD_JOINABLE); | |
467 | } else { | |
468 | qemu_thread_create(&context.threads[i].pgthread, "touch_pages", | |
469 | touch_fn, &context.threads[i], | |
470 | QEMU_THREAD_JOINABLE); | |
471 | } | |
dba50678 DH |
472 | addr += context.threads[i].numpages * hpagesize; |
473 | } | |
474 | ||
475 | if (!use_madv_populate_write) { | |
476 | sigbus_memset_context = &context; | |
1e356fc1 | 477 | } |
278fb162 B |
478 | |
479 | qemu_mutex_lock(&page_mutex); | |
dba50678 | 480 | context.all_threads_created = true; |
037fb5eb | 481 | qemu_cond_broadcast(&page_cond); |
278fb162 | 482 | qemu_mutex_unlock(&page_mutex); |
037fb5eb | 483 | |
dba50678 DH |
484 | for (i = 0; i < context.num_threads; i++) { |
485 | int tmp = (uintptr_t)qemu_thread_join(&context.threads[i].pgthread); | |
6c427ab9 DH |
486 | |
487 | if (tmp) { | |
488 | ret = tmp; | |
489 | } | |
1e356fc1 | 490 | } |
dba50678 DH |
491 | |
492 | if (!use_madv_populate_write) { | |
493 | sigbus_memset_context = NULL; | |
494 | } | |
495 | g_free(context.threads); | |
1e356fc1 | 496 | |
6c427ab9 | 497 | return ret; |
1e356fc1 JK |
498 | } |
499 | ||
a384bfa3 DH |
500 | static bool madv_populate_write_possible(char *area, size_t pagesize) |
501 | { | |
502 | return !qemu_madvise(area, pagesize, QEMU_MADV_POPULATE_WRITE) || | |
503 | errno != EINVAL; | |
504 | } | |
505 | ||
6556aadc | 506 | void qemu_prealloc_mem(int fd, char *area, size_t sz, int max_threads, |
e04a34e5 | 507 | ThreadContext *tc, Error **errp) |
38183310 | 508 | { |
a960d664 | 509 | static gsize initialized; |
b7bf8f56 | 510 | int ret; |
1e356fc1 | 511 | size_t hpagesize = qemu_fd_getpagesize(fd); |
6556aadc | 512 | size_t numpages = DIV_ROUND_UP(sz, hpagesize); |
a384bfa3 | 513 | bool use_madv_populate_write; |
29b838c0 | 514 | struct sigaction act; |
38183310 | 515 | |
a384bfa3 DH |
516 | /* |
517 | * Sense on every invocation, as MADV_POPULATE_WRITE cannot be used for | |
518 | * some special mappings, such as mapping /dev/mem. | |
519 | */ | |
520 | use_madv_populate_write = madv_populate_write_possible(area, hpagesize); | |
521 | ||
522 | if (!use_madv_populate_write) { | |
a960d664 DH |
523 | if (g_once_init_enter(&initialized)) { |
524 | qemu_mutex_init(&sigbus_mutex); | |
525 | g_once_init_leave(&initialized, 1); | |
526 | } | |
527 | ||
528 | qemu_mutex_lock(&sigbus_mutex); | |
a384bfa3 | 529 | memset(&act, 0, sizeof(act)); |
29b838c0 DH |
530 | #ifdef CONFIG_LINUX |
531 | act.sa_sigaction = &sigbus_handler; | |
532 | act.sa_flags = SA_SIGINFO; | |
533 | #else /* CONFIG_LINUX */ | |
a384bfa3 DH |
534 | act.sa_handler = &sigbus_handler; |
535 | act.sa_flags = 0; | |
29b838c0 | 536 | #endif /* CONFIG_LINUX */ |
a384bfa3 | 537 | |
29b838c0 | 538 | ret = sigaction(SIGBUS, &act, &sigbus_oldact); |
a384bfa3 | 539 | if (ret) { |
dd4fc605 | 540 | qemu_mutex_unlock(&sigbus_mutex); |
a384bfa3 | 541 | error_setg_errno(errp, errno, |
6556aadc | 542 | "qemu_prealloc_mem: failed to install signal handler"); |
a384bfa3 DH |
543 | return; |
544 | } | |
38183310 PB |
545 | } |
546 | ||
1e356fc1 | 547 | /* touch pages simultaneously */ |
e04a34e5 | 548 | ret = touch_all_pages(area, hpagesize, numpages, max_threads, tc, |
a384bfa3 | 549 | use_madv_populate_write); |
6c427ab9 DH |
550 | if (ret) { |
551 | error_setg_errno(errp, -ret, | |
6556aadc | 552 | "qemu_prealloc_mem: preallocating memory failed"); |
056b68af | 553 | } |
38183310 | 554 | |
a384bfa3 | 555 | if (!use_madv_populate_write) { |
29b838c0 | 556 | ret = sigaction(SIGBUS, &sigbus_oldact, NULL); |
a384bfa3 DH |
557 | if (ret) { |
558 | /* Terminate QEMU since it can't recover from error */ | |
6556aadc | 559 | perror("qemu_prealloc_mem: failed to reinstall signal handler"); |
a384bfa3 DH |
560 | exit(1); |
561 | } | |
a960d664 | 562 | qemu_mutex_unlock(&sigbus_mutex); |
b7bf8f56 | 563 | } |
38183310 | 564 | } |
d57e4e48 | 565 | |
7dc9ae43 MP |
566 | char *qemu_get_pid_name(pid_t pid) |
567 | { | |
568 | char *name = NULL; | |
569 | ||
570 | #if defined(__FreeBSD__) | |
571 | /* BSDs don't have /proc, but they provide a nice substitute */ | |
572 | struct kinfo_proc *proc = kinfo_getproc(pid); | |
573 | ||
574 | if (proc) { | |
575 | name = g_strdup(proc->ki_comm); | |
576 | free(proc); | |
577 | } | |
578 | #else | |
579 | /* Assume a system with reasonable procfs */ | |
580 | char *pid_path; | |
581 | size_t len; | |
582 | ||
583 | pid_path = g_strdup_printf("/proc/%d/cmdline", pid); | |
584 | g_file_get_contents(pid_path, &name, &len, NULL); | |
585 | g_free(pid_path); | |
586 | #endif | |
587 | ||
588 | return name; | |
589 | } | |
590 | ||
591 | ||
57cb38b3 DB |
592 | pid_t qemu_fork(Error **errp) |
593 | { | |
594 | sigset_t oldmask, newmask; | |
595 | struct sigaction sig_action; | |
596 | int saved_errno; | |
597 | pid_t pid; | |
598 | ||
599 | /* | |
600 | * Need to block signals now, so that child process can safely | |
601 | * kill off caller's signal handlers without a race. | |
602 | */ | |
603 | sigfillset(&newmask); | |
604 | if (pthread_sigmask(SIG_SETMASK, &newmask, &oldmask) != 0) { | |
605 | error_setg_errno(errp, errno, | |
606 | "cannot block signals"); | |
607 | return -1; | |
608 | } | |
609 | ||
610 | pid = fork(); | |
611 | saved_errno = errno; | |
612 | ||
613 | if (pid < 0) { | |
614 | /* attempt to restore signal mask, but ignore failure, to | |
615 | * avoid obscuring the fork failure */ | |
616 | (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL); | |
617 | error_setg_errno(errp, saved_errno, | |
618 | "cannot fork child process"); | |
619 | errno = saved_errno; | |
620 | return -1; | |
621 | } else if (pid) { | |
622 | /* parent process */ | |
623 | ||
624 | /* Restore our original signal mask now that the child is | |
625 | * safely running. Only documented failures are EFAULT (not | |
626 | * possible, since we are using just-grabbed mask) or EINVAL | |
627 | * (not possible, since we are using correct arguments). */ | |
628 | (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL); | |
629 | } else { | |
630 | /* child process */ | |
631 | size_t i; | |
632 | ||
633 | /* Clear out all signal handlers from parent so nothing | |
634 | * unexpected can happen in our child once we unblock | |
635 | * signals */ | |
636 | sig_action.sa_handler = SIG_DFL; | |
637 | sig_action.sa_flags = 0; | |
638 | sigemptyset(&sig_action.sa_mask); | |
639 | ||
640 | for (i = 1; i < NSIG; i++) { | |
641 | /* Only possible errors are EFAULT or EINVAL The former | |
642 | * won't happen, the latter we expect, so no need to check | |
643 | * return value */ | |
644 | (void)sigaction(i, &sig_action, NULL); | |
645 | } | |
646 | ||
647 | /* Unmask all signals in child, since we've no idea what the | |
648 | * caller's done with their signal mask and don't want to | |
649 | * propagate that to children */ | |
650 | sigemptyset(&newmask); | |
651 | if (pthread_sigmask(SIG_SETMASK, &newmask, NULL) != 0) { | |
652 | Error *local_err = NULL; | |
653 | error_setg_errno(&local_err, errno, | |
654 | "cannot unblock signals"); | |
655 | error_report_err(local_err); | |
656 | _exit(1); | |
657 | } | |
658 | } | |
659 | return pid; | |
660 | } | |
8737d9e0 PL |
661 | |
662 | void *qemu_alloc_stack(size_t *sz) | |
663 | { | |
664 | void *ptr, *guardpage; | |
fc3d1bad | 665 | int flags; |
7d992e4d PL |
666 | #ifdef CONFIG_DEBUG_STACK_USAGE |
667 | void *ptr2; | |
668 | #endif | |
8e3b0cbb | 669 | size_t pagesz = qemu_real_host_page_size(); |
8737d9e0 PL |
670 | #ifdef _SC_THREAD_STACK_MIN |
671 | /* avoid stacks smaller than _SC_THREAD_STACK_MIN */ | |
672 | long min_stack_sz = sysconf(_SC_THREAD_STACK_MIN); | |
673 | *sz = MAX(MAX(min_stack_sz, 0), *sz); | |
674 | #endif | |
675 | /* adjust stack size to a multiple of the page size */ | |
676 | *sz = ROUND_UP(*sz, pagesz); | |
677 | /* allocate one extra page for the guard page */ | |
678 | *sz += pagesz; | |
679 | ||
fc3d1bad BS |
680 | flags = MAP_PRIVATE | MAP_ANONYMOUS; |
681 | #if defined(MAP_STACK) && defined(__OpenBSD__) | |
682 | /* Only enable MAP_STACK on OpenBSD. Other OS's such as | |
683 | * Linux/FreeBSD/NetBSD have a flag with the same name | |
684 | * but have differing functionality. OpenBSD will SEGV | |
685 | * if it spots execution with a stack pointer pointing | |
686 | * at memory that was not allocated with MAP_STACK. | |
687 | */ | |
688 | flags |= MAP_STACK; | |
689 | #endif | |
690 | ||
691 | ptr = mmap(NULL, *sz, PROT_READ | PROT_WRITE, flags, -1, 0); | |
8737d9e0 | 692 | if (ptr == MAP_FAILED) { |
e916a6e8 | 693 | perror("failed to allocate memory for stack"); |
8737d9e0 PL |
694 | abort(); |
695 | } | |
696 | ||
697 | #if defined(HOST_IA64) | |
698 | /* separate register stack */ | |
699 | guardpage = ptr + (((*sz - pagesz) / 2) & ~pagesz); | |
700 | #elif defined(HOST_HPPA) | |
701 | /* stack grows up */ | |
702 | guardpage = ptr + *sz - pagesz; | |
703 | #else | |
704 | /* stack grows down */ | |
705 | guardpage = ptr; | |
706 | #endif | |
707 | if (mprotect(guardpage, pagesz, PROT_NONE) != 0) { | |
e916a6e8 | 708 | perror("failed to set up stack guard page"); |
8737d9e0 PL |
709 | abort(); |
710 | } | |
711 | ||
7d992e4d PL |
712 | #ifdef CONFIG_DEBUG_STACK_USAGE |
713 | for (ptr2 = ptr + pagesz; ptr2 < ptr + *sz; ptr2 += sizeof(uint32_t)) { | |
714 | *(uint32_t *)ptr2 = 0xdeadbeaf; | |
715 | } | |
716 | #endif | |
717 | ||
8737d9e0 PL |
718 | return ptr; |
719 | } | |
720 | ||
7d992e4d PL |
721 | #ifdef CONFIG_DEBUG_STACK_USAGE |
722 | static __thread unsigned int max_stack_usage; | |
723 | #endif | |
724 | ||
8737d9e0 PL |
725 | void qemu_free_stack(void *stack, size_t sz) |
726 | { | |
7d992e4d PL |
727 | #ifdef CONFIG_DEBUG_STACK_USAGE |
728 | unsigned int usage; | |
729 | void *ptr; | |
730 | ||
8e3b0cbb | 731 | for (ptr = stack + qemu_real_host_page_size(); ptr < stack + sz; |
7d992e4d PL |
732 | ptr += sizeof(uint32_t)) { |
733 | if (*(uint32_t *)ptr != 0xdeadbeaf) { | |
734 | break; | |
735 | } | |
736 | } | |
737 | usage = sz - (uintptr_t) (ptr - stack); | |
738 | if (usage > max_stack_usage) { | |
739 | error_report("thread %d max stack usage increased from %u to %u", | |
740 | qemu_get_thread_id(), max_stack_usage, usage); | |
741 | max_stack_usage = usage; | |
742 | } | |
743 | #endif | |
744 | ||
8737d9e0 PL |
745 | munmap(stack, sz); |
746 | } | |
d98d4072 | 747 | |
c905a368 DB |
748 | /* |
749 | * Disable CFI checks. | |
750 | * We are going to call a signal hander directly. Such handler may or may not | |
751 | * have been defined in our binary, so there's no guarantee that the pointer | |
752 | * used to set the handler is a cfi-valid pointer. Since the handlers are | |
753 | * stored in kernel memory, changing the handler to an attacker-defined | |
754 | * function requires being able to call a sigaction() syscall, | |
755 | * which is not as easy as overwriting a pointer in memory. | |
756 | */ | |
757 | QEMU_DISABLE_CFI | |
d98d4072 PB |
758 | void sigaction_invoke(struct sigaction *action, |
759 | struct qemu_signalfd_siginfo *info) | |
760 | { | |
02ffa034 | 761 | siginfo_t si = {}; |
d98d4072 PB |
762 | si.si_signo = info->ssi_signo; |
763 | si.si_errno = info->ssi_errno; | |
764 | si.si_code = info->ssi_code; | |
765 | ||
766 | /* Convert the minimal set of fields defined by POSIX. | |
767 | * Positive si_code values are reserved for kernel-generated | |
768 | * signals, where the valid siginfo fields are determined by | |
769 | * the signal number. But according to POSIX, it is unspecified | |
770 | * whether SI_USER and SI_QUEUE have values less than or equal to | |
771 | * zero. | |
772 | */ | |
773 | if (info->ssi_code == SI_USER || info->ssi_code == SI_QUEUE || | |
774 | info->ssi_code <= 0) { | |
775 | /* SIGTERM, etc. */ | |
776 | si.si_pid = info->ssi_pid; | |
777 | si.si_uid = info->ssi_uid; | |
778 | } else if (info->ssi_signo == SIGILL || info->ssi_signo == SIGFPE || | |
779 | info->ssi_signo == SIGSEGV || info->ssi_signo == SIGBUS) { | |
780 | si.si_addr = (void *)(uintptr_t)info->ssi_addr; | |
781 | } else if (info->ssi_signo == SIGCHLD) { | |
782 | si.si_pid = info->ssi_pid; | |
783 | si.si_status = info->ssi_status; | |
784 | si.si_uid = info->ssi_uid; | |
d98d4072 PB |
785 | } |
786 | action->sa_sigaction(info->ssi_signo, &si, NULL); | |
787 | } | |
e47f4765 | 788 | |
ad06ef0e AB |
789 | size_t qemu_get_host_physmem(void) |
790 | { | |
791 | #ifdef _SC_PHYS_PAGES | |
792 | long pages = sysconf(_SC_PHYS_PAGES); | |
793 | if (pages > 0) { | |
8e3b0cbb | 794 | if (pages > SIZE_MAX / qemu_real_host_page_size()) { |
ad06ef0e AB |
795 | return SIZE_MAX; |
796 | } else { | |
8e3b0cbb | 797 | return pages * qemu_real_host_page_size(); |
ad06ef0e AB |
798 | } |
799 | } | |
800 | #endif | |
801 | return 0; | |
802 | } | |
e9c4e0a8 | 803 | |
73991a92 MAL |
804 | int qemu_msync(void *addr, size_t length, int fd) |
805 | { | |
806 | size_t align_mask = ~(qemu_real_host_page_size() - 1); | |
807 | ||
808 | /** | |
809 | * There are no strict reqs as per the length of mapping | |
810 | * to be synced. Still the length needs to follow the address | |
811 | * alignment changes. Additionally - round the size to the multiple | |
812 | * of PAGE_SIZE | |
813 | */ | |
814 | length += ((uintptr_t)addr & (qemu_real_host_page_size() - 1)); | |
815 | length = (length + ~align_mask) & align_mask; | |
816 | ||
817 | addr = (void *)((uintptr_t)addr & align_mask); | |
818 | ||
819 | return msync(addr, length, MS_SYNC); | |
820 | } |