]>
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 | ||
a8d25326 | 34 | #include "qemu-common.h" |
9c17d615 | 35 | #include "sysemu/sysemu.h" |
c1b0b93b | 36 | #include "trace.h" |
da34e65c | 37 | #include "qapi/error.h" |
1de7afc9 | 38 | #include "qemu/sockets.h" |
db725815 | 39 | #include "qemu/thread.h" |
10f5bff6 | 40 | #include <libgen.h> |
f348b6d1 | 41 | #include "qemu/cutils.h" |
c905a368 | 42 | #include "qemu/compiler.h" |
c1b0b93b | 43 | |
cbcfa041 PB |
44 | #ifdef CONFIG_LINUX |
45 | #include <sys/syscall.h> | |
46 | #endif | |
cbcfa041 | 47 | |
41975b26 AF |
48 | #ifdef __FreeBSD__ |
49 | #include <sys/sysctl.h> | |
a7764f15 | 50 | #include <sys/user.h> |
9548a891 | 51 | #include <sys/thr.h> |
7dc9ae43 | 52 | #include <libutil.h> |
41975b26 AF |
53 | #endif |
54 | ||
094611b4 KR |
55 | #ifdef __NetBSD__ |
56 | #include <sys/sysctl.h> | |
9548a891 | 57 | #include <lwp.h> |
094611b4 KR |
58 | #endif |
59 | ||
2032e243 DC |
60 | #ifdef __APPLE__ |
61 | #include <mach-o/dyld.h> | |
62 | #endif | |
63 | ||
2b9b9e70 DC |
64 | #ifdef __HAIKU__ |
65 | #include <kernel/image.h> | |
66 | #endif | |
67 | ||
a9c94277 | 68 | #include "qemu/mmap-alloc.h" |
794e8f30 | 69 | |
7d992e4d PL |
70 | #ifdef CONFIG_DEBUG_STACK_USAGE |
71 | #include "qemu/error-report.h" | |
72 | #endif | |
73 | ||
dfd0dcc7 | 74 | #define MAX_MEM_PREALLOC_THREAD_COUNT 16 |
1e356fc1 JK |
75 | |
76 | struct MemsetThread { | |
77 | char *addr; | |
e947d47d SW |
78 | size_t numpages; |
79 | size_t hpagesize; | |
1e356fc1 JK |
80 | QemuThread pgthread; |
81 | sigjmp_buf env; | |
82 | }; | |
83 | typedef struct MemsetThread MemsetThread; | |
84 | ||
85 | static MemsetThread *memset_thread; | |
86 | static int memset_num_threads; | |
87 | static bool memset_thread_failed; | |
88 | ||
037fb5eb | 89 | static QemuMutex page_mutex; |
90 | static QemuCond page_cond; | |
91 | static bool threads_created_flag; | |
92 | ||
cbcfa041 PB |
93 | int qemu_get_thread_id(void) |
94 | { | |
95 | #if defined(__linux__) | |
96 | return syscall(SYS_gettid); | |
9548a891 DC |
97 | #elif defined(__FreeBSD__) |
98 | /* thread id is up to INT_MAX */ | |
99 | long tid; | |
100 | thr_self(&tid); | |
101 | return (int)tid; | |
102 | #elif defined(__NetBSD__) | |
103 | return _lwp_self(); | |
8edbca51 DC |
104 | #elif defined(__OpenBSD__) |
105 | return getthrid(); | |
cbcfa041 PB |
106 | #else |
107 | return getpid(); | |
108 | #endif | |
109 | } | |
f97742d0 AR |
110 | |
111 | int qemu_daemon(int nochdir, int noclose) | |
112 | { | |
113 | return daemon(nochdir, noclose); | |
114 | } | |
115 | ||
9e6bdef2 MAL |
116 | bool qemu_write_pidfile(const char *path, Error **errp) |
117 | { | |
118 | int fd; | |
119 | char pidstr[32]; | |
120 | ||
121 | while (1) { | |
122 | struct stat a, b; | |
35f7f3fb MAL |
123 | struct flock lock = { |
124 | .l_type = F_WRLCK, | |
125 | .l_whence = SEEK_SET, | |
126 | .l_len = 0, | |
127 | }; | |
9e6bdef2 | 128 | |
448058aa | 129 | fd = qemu_open_old(path, O_CREAT | O_WRONLY, S_IRUSR | S_IWUSR); |
9e6bdef2 MAL |
130 | if (fd == -1) { |
131 | error_setg_errno(errp, errno, "Cannot open pid file"); | |
132 | return false; | |
133 | } | |
134 | ||
135 | if (fstat(fd, &b) < 0) { | |
136 | error_setg_errno(errp, errno, "Cannot stat file"); | |
137 | goto fail_close; | |
138 | } | |
139 | ||
35f7f3fb | 140 | if (fcntl(fd, F_SETLK, &lock)) { |
9e6bdef2 MAL |
141 | error_setg_errno(errp, errno, "Cannot lock pid file"); |
142 | goto fail_close; | |
143 | } | |
144 | ||
145 | /* | |
146 | * Now make sure the path we locked is the same one that now | |
147 | * exists on the filesystem. | |
148 | */ | |
149 | if (stat(path, &a) < 0) { | |
150 | /* | |
151 | * PID file disappeared, someone else must be racing with | |
152 | * us, so try again. | |
153 | */ | |
154 | close(fd); | |
155 | continue; | |
156 | } | |
157 | ||
158 | if (a.st_ino == b.st_ino) { | |
159 | break; | |
160 | } | |
161 | ||
162 | /* | |
163 | * PID file was recreated, someone else must be racing with | |
164 | * us, so try again. | |
165 | */ | |
166 | close(fd); | |
167 | } | |
168 | ||
169 | if (ftruncate(fd, 0) < 0) { | |
170 | error_setg_errno(errp, errno, "Failed to truncate pid file"); | |
171 | goto fail_unlink; | |
172 | } | |
173 | ||
174 | snprintf(pidstr, sizeof(pidstr), FMT_pid "\n", getpid()); | |
175 | if (write(fd, pidstr, strlen(pidstr)) != strlen(pidstr)) { | |
176 | error_setg(errp, "Failed to write pid file"); | |
177 | goto fail_unlink; | |
178 | } | |
179 | ||
180 | return true; | |
181 | ||
182 | fail_unlink: | |
183 | unlink(path); | |
184 | fail_close: | |
185 | close(fd); | |
186 | return false; | |
187 | } | |
188 | ||
b152aa84 | 189 | void *qemu_oom_check(void *ptr) |
c1b0b93b JS |
190 | { |
191 | if (ptr == NULL) { | |
192 | fprintf(stderr, "Failed to allocate memory: %s\n", strerror(errno)); | |
193 | abort(); | |
194 | } | |
195 | return ptr; | |
196 | } | |
c1b0b93b | 197 | |
7d2a35cc | 198 | void *qemu_try_memalign(size_t alignment, size_t size) |
c1b0b93b JS |
199 | { |
200 | void *ptr; | |
e5354657 KW |
201 | |
202 | if (alignment < sizeof(void*)) { | |
203 | alignment = sizeof(void*); | |
ed6f53f9 PMD |
204 | } else { |
205 | g_assert(is_power_of_2(alignment)); | |
e5354657 KW |
206 | } |
207 | ||
9bc5a719 | 208 | #if defined(CONFIG_POSIX_MEMALIGN) |
c1b0b93b JS |
209 | int ret; |
210 | ret = posix_memalign(&ptr, alignment, size); | |
211 | if (ret != 0) { | |
7d2a35cc KW |
212 | errno = ret; |
213 | ptr = NULL; | |
c1b0b93b JS |
214 | } |
215 | #elif defined(CONFIG_BSD) | |
7d2a35cc | 216 | ptr = valloc(size); |
c1b0b93b | 217 | #else |
7d2a35cc | 218 | ptr = memalign(alignment, size); |
c1b0b93b JS |
219 | #endif |
220 | trace_qemu_memalign(alignment, size, ptr); | |
221 | return ptr; | |
222 | } | |
223 | ||
7d2a35cc KW |
224 | void *qemu_memalign(size_t alignment, size_t size) |
225 | { | |
226 | return qemu_oom_check(qemu_try_memalign(alignment, size)); | |
227 | } | |
228 | ||
c1b0b93b | 229 | /* alloc shared memory pages */ |
06329cce | 230 | void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment, bool shared) |
c1b0b93b | 231 | { |
36b58628 | 232 | size_t align = QEMU_VMALLOC_ALIGN; |
44a4ff31 | 233 | void *ptr = qemu_ram_mmap(-1, size, align, false, shared, false, 0); |
36b58628 | 234 | |
7dda5dc8 | 235 | if (ptr == MAP_FAILED) { |
39228250 | 236 | return NULL; |
c2a8238a | 237 | } |
c2a8238a | 238 | |
a2b257d6 IM |
239 | if (alignment) { |
240 | *alignment = align; | |
241 | } | |
c2dfc5ba | 242 | |
6eebf958 | 243 | trace_qemu_anon_ram_alloc(size, ptr); |
c7f4111a | 244 | return ptr; |
c1b0b93b JS |
245 | } |
246 | ||
247 | void qemu_vfree(void *ptr) | |
248 | { | |
249 | trace_qemu_vfree(ptr); | |
250 | free(ptr); | |
251 | } | |
9549e764 | 252 | |
e7a09b92 PB |
253 | void qemu_anon_ram_free(void *ptr, size_t size) |
254 | { | |
255 | trace_qemu_anon_ram_free(ptr, size); | |
53adb9d4 | 256 | qemu_ram_munmap(-1, ptr, size); |
e7a09b92 PB |
257 | } |
258 | ||
f9e8cacc | 259 | void qemu_set_block(int fd) |
154b9a0c PB |
260 | { |
261 | int f; | |
262 | f = fcntl(fd, F_GETFL); | |
da93b820 LQ |
263 | assert(f != -1); |
264 | f = fcntl(fd, F_SETFL, f & ~O_NONBLOCK); | |
265 | assert(f != -1); | |
154b9a0c PB |
266 | } |
267 | ||
894022e6 | 268 | int qemu_try_set_nonblock(int fd) |
9549e764 JS |
269 | { |
270 | int f; | |
271 | f = fcntl(fd, F_GETFL); | |
02cdcc96 | 272 | if (f == -1) { |
894022e6 LV |
273 | return -errno; |
274 | } | |
275 | if (fcntl(fd, F_SETFL, f | O_NONBLOCK) == -1) { | |
276 | #ifdef __OpenBSD__ | |
02cdcc96 PMD |
277 | /* |
278 | * Previous to OpenBSD 6.3, fcntl(F_SETFL) is not permitted on | |
279 | * memory devices and sets errno to ENODEV. | |
280 | * It's OK if we fail to set O_NONBLOCK on devices like /dev/null, | |
281 | * because they will never block anyway. | |
282 | */ | |
894022e6 LV |
283 | if (errno == ENODEV) { |
284 | return 0; | |
285 | } | |
02cdcc96 | 286 | #endif |
894022e6 LV |
287 | return -errno; |
288 | } | |
289 | return 0; | |
290 | } | |
291 | ||
292 | void qemu_set_nonblock(int fd) | |
293 | { | |
294 | int f; | |
295 | f = qemu_try_set_nonblock(fd); | |
296 | assert(f == 0); | |
9549e764 JS |
297 | } |
298 | ||
606600a1 SO |
299 | int socket_set_fast_reuse(int fd) |
300 | { | |
301 | int val = 1, ret; | |
302 | ||
303 | ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, | |
304 | (const char *)&val, sizeof(val)); | |
305 | ||
306 | assert(ret == 0); | |
307 | ||
308 | return ret; | |
309 | } | |
310 | ||
9549e764 JS |
311 | void qemu_set_cloexec(int fd) |
312 | { | |
313 | int f; | |
314 | f = fcntl(fd, F_GETFD); | |
7e6478e7 SS |
315 | assert(f != -1); |
316 | f = fcntl(fd, F_SETFD, f | FD_CLOEXEC); | |
317 | assert(f != -1); | |
9549e764 | 318 | } |
70e72ce4 JS |
319 | |
320 | /* | |
321 | * Creates a pipe with FD_CLOEXEC set on both file descriptors | |
322 | */ | |
323 | int qemu_pipe(int pipefd[2]) | |
324 | { | |
325 | int ret; | |
326 | ||
327 | #ifdef CONFIG_PIPE2 | |
328 | ret = pipe2(pipefd, O_CLOEXEC); | |
329 | if (ret != -1 || errno != ENOSYS) { | |
330 | return ret; | |
331 | } | |
332 | #endif | |
333 | ret = pipe(pipefd); | |
334 | if (ret == 0) { | |
335 | qemu_set_cloexec(pipefd[0]); | |
336 | qemu_set_cloexec(pipefd[1]); | |
337 | } | |
338 | ||
339 | return ret; | |
340 | } | |
38671423 | 341 | |
e2ea3515 LE |
342 | char * |
343 | qemu_get_local_state_pathname(const char *relative_pathname) | |
344 | { | |
fcb4f59c PB |
345 | g_autofree char *dir = g_strdup_printf("%s/%s", |
346 | CONFIG_QEMU_LOCALSTATEDIR, | |
347 | relative_pathname); | |
348 | return get_relocated_path(dir); | |
e2ea3515 | 349 | } |
13401ba0 SH |
350 | |
351 | void qemu_set_tty_echo(int fd, bool echo) | |
352 | { | |
353 | struct termios tty; | |
354 | ||
355 | tcgetattr(fd, &tty); | |
356 | ||
357 | if (echo) { | |
358 | tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN; | |
359 | } else { | |
360 | tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN); | |
361 | } | |
362 | ||
363 | tcsetattr(fd, TCSANOW, &tty); | |
364 | } | |
10f5bff6 | 365 | |
9386a4a7 | 366 | static const char *exec_dir; |
10f5bff6 FZ |
367 | |
368 | void qemu_init_exec_dir(const char *argv0) | |
369 | { | |
10f5bff6 FZ |
370 | char *p = NULL; |
371 | char buf[PATH_MAX]; | |
372 | ||
9386a4a7 | 373 | if (exec_dir) { |
a4c13869 PB |
374 | return; |
375 | } | |
10f5bff6 FZ |
376 | |
377 | #if defined(__linux__) | |
378 | { | |
379 | int len; | |
380 | len = readlink("/proc/self/exe", buf, sizeof(buf) - 1); | |
381 | if (len > 0) { | |
382 | buf[len] = 0; | |
383 | p = buf; | |
384 | } | |
385 | } | |
094611b4 KR |
386 | #elif defined(__FreeBSD__) \ |
387 | || (defined(__NetBSD__) && defined(KERN_PROC_PATHNAME)) | |
10f5bff6 | 388 | { |
094611b4 | 389 | #if defined(__FreeBSD__) |
10f5bff6 | 390 | static int mib[4] = {CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1}; |
094611b4 KR |
391 | #else |
392 | static int mib[4] = {CTL_KERN, KERN_PROC_ARGS, -1, KERN_PROC_PATHNAME}; | |
393 | #endif | |
10f5bff6 FZ |
394 | size_t len = sizeof(buf) - 1; |
395 | ||
396 | *buf = '\0'; | |
397 | if (!sysctl(mib, ARRAY_SIZE(mib), buf, &len, NULL, 0) && | |
398 | *buf) { | |
399 | buf[sizeof(buf) - 1] = '\0'; | |
400 | p = buf; | |
401 | } | |
402 | } | |
2032e243 DC |
403 | #elif defined(__APPLE__) |
404 | { | |
405 | char fpath[PATH_MAX]; | |
406 | uint32_t len = sizeof(fpath); | |
407 | if (_NSGetExecutablePath(fpath, &len) == 0) { | |
408 | p = realpath(fpath, buf); | |
409 | if (!p) { | |
410 | return; | |
411 | } | |
412 | } | |
413 | } | |
2b9b9e70 DC |
414 | #elif defined(__HAIKU__) |
415 | { | |
416 | image_info ii; | |
417 | int32_t c = 0; | |
418 | ||
419 | *buf = '\0'; | |
420 | while (get_next_image_info(0, &c, &ii) == B_OK) { | |
421 | if (ii.type == B_APP_IMAGE) { | |
422 | strncpy(buf, ii.name, sizeof(buf)); | |
423 | buf[sizeof(buf) - 1] = 0; | |
424 | p = buf; | |
425 | break; | |
426 | } | |
427 | } | |
428 | } | |
10f5bff6 FZ |
429 | #endif |
430 | /* If we don't have any way of figuring out the actual executable | |
431 | location then try argv[0]. */ | |
9386a4a7 | 432 | if (!p && argv0) { |
10f5bff6 | 433 | p = realpath(argv0, buf); |
10f5bff6 | 434 | } |
9386a4a7 PB |
435 | if (p) { |
436 | exec_dir = g_path_get_dirname(p); | |
437 | } else { | |
438 | exec_dir = CONFIG_BINDIR; | |
439 | } | |
10f5bff6 FZ |
440 | } |
441 | ||
a4c13869 | 442 | const char *qemu_get_exec_dir(void) |
10f5bff6 | 443 | { |
a4c13869 | 444 | return exec_dir; |
10f5bff6 | 445 | } |
38183310 | 446 | |
38183310 PB |
447 | static void sigbus_handler(int signal) |
448 | { | |
1e356fc1 JK |
449 | int i; |
450 | if (memset_thread) { | |
451 | for (i = 0; i < memset_num_threads; i++) { | |
452 | if (qemu_thread_is_self(&memset_thread[i].pgthread)) { | |
453 | siglongjmp(memset_thread[i].env, 1); | |
454 | } | |
455 | } | |
456 | } | |
38183310 PB |
457 | } |
458 | ||
1e356fc1 JK |
459 | static void *do_touch_pages(void *arg) |
460 | { | |
461 | MemsetThread *memset_args = (MemsetThread *)arg; | |
1e356fc1 | 462 | sigset_t set, oldset; |
1e356fc1 | 463 | |
037fb5eb | 464 | /* |
465 | * On Linux, the page faults from the loop below can cause mmap_sem | |
466 | * contention with allocation of the thread stacks. Do not start | |
467 | * clearing until all threads have been created. | |
468 | */ | |
469 | qemu_mutex_lock(&page_mutex); | |
470 | while(!threads_created_flag){ | |
471 | qemu_cond_wait(&page_cond, &page_mutex); | |
472 | } | |
473 | qemu_mutex_unlock(&page_mutex); | |
474 | ||
1e356fc1 JK |
475 | /* unblock SIGBUS */ |
476 | sigemptyset(&set); | |
477 | sigaddset(&set, SIGBUS); | |
478 | pthread_sigmask(SIG_UNBLOCK, &set, &oldset); | |
479 | ||
480 | if (sigsetjmp(memset_args->env, 1)) { | |
481 | memset_thread_failed = true; | |
482 | } else { | |
e947d47d SW |
483 | char *addr = memset_args->addr; |
484 | size_t numpages = memset_args->numpages; | |
485 | size_t hpagesize = memset_args->hpagesize; | |
486 | size_t i; | |
1e356fc1 | 487 | for (i = 0; i < numpages; i++) { |
9dc44aa5 DB |
488 | /* |
489 | * Read & write back the same value, so we don't | |
490 | * corrupt existing user/app data that might be | |
491 | * stored. | |
492 | * | |
493 | * 'volatile' to stop compiler optimizing this away | |
494 | * to a no-op | |
495 | * | |
496 | * TODO: get a better solution from kernel so we | |
497 | * don't need to write at all so we don't cause | |
498 | * wear on the storage backing the region... | |
499 | */ | |
500 | *(volatile char *)addr = *addr; | |
1e356fc1 JK |
501 | addr += hpagesize; |
502 | } | |
503 | } | |
504 | pthread_sigmask(SIG_SETMASK, &oldset, NULL); | |
505 | return NULL; | |
506 | } | |
507 | ||
dfd0dcc7 JK |
508 | static inline int get_memset_num_threads(int smp_cpus) |
509 | { | |
510 | long host_procs = sysconf(_SC_NPROCESSORS_ONLN); | |
511 | int ret = 1; | |
512 | ||
513 | if (host_procs > 0) { | |
514 | ret = MIN(MIN(host_procs, MAX_MEM_PREALLOC_THREAD_COUNT), smp_cpus); | |
515 | } | |
516 | /* In case sysconf() fails, we fall back to single threaded */ | |
517 | return ret; | |
518 | } | |
519 | ||
1e356fc1 JK |
520 | static bool touch_all_pages(char *area, size_t hpagesize, size_t numpages, |
521 | int smp_cpus) | |
522 | { | |
78b3f67a | 523 | static gsize initialized = 0; |
037fb5eb | 524 | size_t numpages_per_thread, leftover; |
1e356fc1 JK |
525 | char *addr = area; |
526 | int i = 0; | |
527 | ||
78b3f67a PB |
528 | if (g_once_init_enter(&initialized)) { |
529 | qemu_mutex_init(&page_mutex); | |
530 | qemu_cond_init(&page_cond); | |
531 | g_once_init_leave(&initialized, 1); | |
532 | } | |
533 | ||
1e356fc1 | 534 | memset_thread_failed = false; |
037fb5eb | 535 | threads_created_flag = false; |
dfd0dcc7 | 536 | memset_num_threads = get_memset_num_threads(smp_cpus); |
1e356fc1 | 537 | memset_thread = g_new0(MemsetThread, memset_num_threads); |
037fb5eb | 538 | numpages_per_thread = numpages / memset_num_threads; |
539 | leftover = numpages % memset_num_threads; | |
1e356fc1 JK |
540 | for (i = 0; i < memset_num_threads; i++) { |
541 | memset_thread[i].addr = addr; | |
037fb5eb | 542 | memset_thread[i].numpages = numpages_per_thread + (i < leftover); |
1e356fc1 JK |
543 | memset_thread[i].hpagesize = hpagesize; |
544 | qemu_thread_create(&memset_thread[i].pgthread, "touch_pages", | |
545 | do_touch_pages, &memset_thread[i], | |
546 | QEMU_THREAD_JOINABLE); | |
037fb5eb | 547 | addr += memset_thread[i].numpages * hpagesize; |
1e356fc1 | 548 | } |
278fb162 B |
549 | |
550 | qemu_mutex_lock(&page_mutex); | |
037fb5eb | 551 | threads_created_flag = true; |
552 | qemu_cond_broadcast(&page_cond); | |
278fb162 | 553 | qemu_mutex_unlock(&page_mutex); |
037fb5eb | 554 | |
1e356fc1 JK |
555 | for (i = 0; i < memset_num_threads; i++) { |
556 | qemu_thread_join(&memset_thread[i].pgthread); | |
557 | } | |
558 | g_free(memset_thread); | |
559 | memset_thread = NULL; | |
560 | ||
561 | return memset_thread_failed; | |
562 | } | |
563 | ||
564 | void os_mem_prealloc(int fd, char *area, size_t memory, int smp_cpus, | |
565 | Error **errp) | |
38183310 | 566 | { |
b7bf8f56 | 567 | int ret; |
38183310 | 568 | struct sigaction act, oldact; |
1e356fc1 JK |
569 | size_t hpagesize = qemu_fd_getpagesize(fd); |
570 | size_t numpages = DIV_ROUND_UP(memory, hpagesize); | |
38183310 PB |
571 | |
572 | memset(&act, 0, sizeof(act)); | |
573 | act.sa_handler = &sigbus_handler; | |
574 | act.sa_flags = 0; | |
575 | ||
576 | ret = sigaction(SIGBUS, &act, &oldact); | |
577 | if (ret) { | |
056b68af IM |
578 | error_setg_errno(errp, errno, |
579 | "os_mem_prealloc: failed to install signal handler"); | |
580 | return; | |
38183310 PB |
581 | } |
582 | ||
1e356fc1 JK |
583 | /* touch pages simultaneously */ |
584 | if (touch_all_pages(area, hpagesize, numpages, smp_cpus)) { | |
056b68af | 585 | error_setg(errp, "os_mem_prealloc: Insufficient free host memory " |
462e5d50 | 586 | "pages available to allocate guest RAM"); |
056b68af | 587 | } |
38183310 | 588 | |
056b68af IM |
589 | ret = sigaction(SIGBUS, &oldact, NULL); |
590 | if (ret) { | |
591 | /* Terminate QEMU since it can't recover from error */ | |
592 | perror("os_mem_prealloc: failed to reinstall signal handler"); | |
593 | exit(1); | |
b7bf8f56 | 594 | } |
38183310 | 595 | } |
d57e4e48 | 596 | |
7dc9ae43 MP |
597 | char *qemu_get_pid_name(pid_t pid) |
598 | { | |
599 | char *name = NULL; | |
600 | ||
601 | #if defined(__FreeBSD__) | |
602 | /* BSDs don't have /proc, but they provide a nice substitute */ | |
603 | struct kinfo_proc *proc = kinfo_getproc(pid); | |
604 | ||
605 | if (proc) { | |
606 | name = g_strdup(proc->ki_comm); | |
607 | free(proc); | |
608 | } | |
609 | #else | |
610 | /* Assume a system with reasonable procfs */ | |
611 | char *pid_path; | |
612 | size_t len; | |
613 | ||
614 | pid_path = g_strdup_printf("/proc/%d/cmdline", pid); | |
615 | g_file_get_contents(pid_path, &name, &len, NULL); | |
616 | g_free(pid_path); | |
617 | #endif | |
618 | ||
619 | return name; | |
620 | } | |
621 | ||
622 | ||
57cb38b3 DB |
623 | pid_t qemu_fork(Error **errp) |
624 | { | |
625 | sigset_t oldmask, newmask; | |
626 | struct sigaction sig_action; | |
627 | int saved_errno; | |
628 | pid_t pid; | |
629 | ||
630 | /* | |
631 | * Need to block signals now, so that child process can safely | |
632 | * kill off caller's signal handlers without a race. | |
633 | */ | |
634 | sigfillset(&newmask); | |
635 | if (pthread_sigmask(SIG_SETMASK, &newmask, &oldmask) != 0) { | |
636 | error_setg_errno(errp, errno, | |
637 | "cannot block signals"); | |
638 | return -1; | |
639 | } | |
640 | ||
641 | pid = fork(); | |
642 | saved_errno = errno; | |
643 | ||
644 | if (pid < 0) { | |
645 | /* attempt to restore signal mask, but ignore failure, to | |
646 | * avoid obscuring the fork failure */ | |
647 | (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL); | |
648 | error_setg_errno(errp, saved_errno, | |
649 | "cannot fork child process"); | |
650 | errno = saved_errno; | |
651 | return -1; | |
652 | } else if (pid) { | |
653 | /* parent process */ | |
654 | ||
655 | /* Restore our original signal mask now that the child is | |
656 | * safely running. Only documented failures are EFAULT (not | |
657 | * possible, since we are using just-grabbed mask) or EINVAL | |
658 | * (not possible, since we are using correct arguments). */ | |
659 | (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL); | |
660 | } else { | |
661 | /* child process */ | |
662 | size_t i; | |
663 | ||
664 | /* Clear out all signal handlers from parent so nothing | |
665 | * unexpected can happen in our child once we unblock | |
666 | * signals */ | |
667 | sig_action.sa_handler = SIG_DFL; | |
668 | sig_action.sa_flags = 0; | |
669 | sigemptyset(&sig_action.sa_mask); | |
670 | ||
671 | for (i = 1; i < NSIG; i++) { | |
672 | /* Only possible errors are EFAULT or EINVAL The former | |
673 | * won't happen, the latter we expect, so no need to check | |
674 | * return value */ | |
675 | (void)sigaction(i, &sig_action, NULL); | |
676 | } | |
677 | ||
678 | /* Unmask all signals in child, since we've no idea what the | |
679 | * caller's done with their signal mask and don't want to | |
680 | * propagate that to children */ | |
681 | sigemptyset(&newmask); | |
682 | if (pthread_sigmask(SIG_SETMASK, &newmask, NULL) != 0) { | |
683 | Error *local_err = NULL; | |
684 | error_setg_errno(&local_err, errno, | |
685 | "cannot unblock signals"); | |
686 | error_report_err(local_err); | |
687 | _exit(1); | |
688 | } | |
689 | } | |
690 | return pid; | |
691 | } | |
8737d9e0 PL |
692 | |
693 | void *qemu_alloc_stack(size_t *sz) | |
694 | { | |
695 | void *ptr, *guardpage; | |
fc3d1bad | 696 | int flags; |
7d992e4d PL |
697 | #ifdef CONFIG_DEBUG_STACK_USAGE |
698 | void *ptr2; | |
699 | #endif | |
038adc2f | 700 | size_t pagesz = qemu_real_host_page_size; |
8737d9e0 PL |
701 | #ifdef _SC_THREAD_STACK_MIN |
702 | /* avoid stacks smaller than _SC_THREAD_STACK_MIN */ | |
703 | long min_stack_sz = sysconf(_SC_THREAD_STACK_MIN); | |
704 | *sz = MAX(MAX(min_stack_sz, 0), *sz); | |
705 | #endif | |
706 | /* adjust stack size to a multiple of the page size */ | |
707 | *sz = ROUND_UP(*sz, pagesz); | |
708 | /* allocate one extra page for the guard page */ | |
709 | *sz += pagesz; | |
710 | ||
fc3d1bad BS |
711 | flags = MAP_PRIVATE | MAP_ANONYMOUS; |
712 | #if defined(MAP_STACK) && defined(__OpenBSD__) | |
713 | /* Only enable MAP_STACK on OpenBSD. Other OS's such as | |
714 | * Linux/FreeBSD/NetBSD have a flag with the same name | |
715 | * but have differing functionality. OpenBSD will SEGV | |
716 | * if it spots execution with a stack pointer pointing | |
717 | * at memory that was not allocated with MAP_STACK. | |
718 | */ | |
719 | flags |= MAP_STACK; | |
720 | #endif | |
721 | ||
722 | ptr = mmap(NULL, *sz, PROT_READ | PROT_WRITE, flags, -1, 0); | |
8737d9e0 | 723 | if (ptr == MAP_FAILED) { |
e916a6e8 | 724 | perror("failed to allocate memory for stack"); |
8737d9e0 PL |
725 | abort(); |
726 | } | |
727 | ||
728 | #if defined(HOST_IA64) | |
729 | /* separate register stack */ | |
730 | guardpage = ptr + (((*sz - pagesz) / 2) & ~pagesz); | |
731 | #elif defined(HOST_HPPA) | |
732 | /* stack grows up */ | |
733 | guardpage = ptr + *sz - pagesz; | |
734 | #else | |
735 | /* stack grows down */ | |
736 | guardpage = ptr; | |
737 | #endif | |
738 | if (mprotect(guardpage, pagesz, PROT_NONE) != 0) { | |
e916a6e8 | 739 | perror("failed to set up stack guard page"); |
8737d9e0 PL |
740 | abort(); |
741 | } | |
742 | ||
7d992e4d PL |
743 | #ifdef CONFIG_DEBUG_STACK_USAGE |
744 | for (ptr2 = ptr + pagesz; ptr2 < ptr + *sz; ptr2 += sizeof(uint32_t)) { | |
745 | *(uint32_t *)ptr2 = 0xdeadbeaf; | |
746 | } | |
747 | #endif | |
748 | ||
8737d9e0 PL |
749 | return ptr; |
750 | } | |
751 | ||
7d992e4d PL |
752 | #ifdef CONFIG_DEBUG_STACK_USAGE |
753 | static __thread unsigned int max_stack_usage; | |
754 | #endif | |
755 | ||
8737d9e0 PL |
756 | void qemu_free_stack(void *stack, size_t sz) |
757 | { | |
7d992e4d PL |
758 | #ifdef CONFIG_DEBUG_STACK_USAGE |
759 | unsigned int usage; | |
760 | void *ptr; | |
761 | ||
038adc2f | 762 | for (ptr = stack + qemu_real_host_page_size; ptr < stack + sz; |
7d992e4d PL |
763 | ptr += sizeof(uint32_t)) { |
764 | if (*(uint32_t *)ptr != 0xdeadbeaf) { | |
765 | break; | |
766 | } | |
767 | } | |
768 | usage = sz - (uintptr_t) (ptr - stack); | |
769 | if (usage > max_stack_usage) { | |
770 | error_report("thread %d max stack usage increased from %u to %u", | |
771 | qemu_get_thread_id(), max_stack_usage, usage); | |
772 | max_stack_usage = usage; | |
773 | } | |
774 | #endif | |
775 | ||
8737d9e0 PL |
776 | munmap(stack, sz); |
777 | } | |
d98d4072 | 778 | |
c905a368 DB |
779 | /* |
780 | * Disable CFI checks. | |
781 | * We are going to call a signal hander directly. Such handler may or may not | |
782 | * have been defined in our binary, so there's no guarantee that the pointer | |
783 | * used to set the handler is a cfi-valid pointer. Since the handlers are | |
784 | * stored in kernel memory, changing the handler to an attacker-defined | |
785 | * function requires being able to call a sigaction() syscall, | |
786 | * which is not as easy as overwriting a pointer in memory. | |
787 | */ | |
788 | QEMU_DISABLE_CFI | |
d98d4072 PB |
789 | void sigaction_invoke(struct sigaction *action, |
790 | struct qemu_signalfd_siginfo *info) | |
791 | { | |
02ffa034 | 792 | siginfo_t si = {}; |
d98d4072 PB |
793 | si.si_signo = info->ssi_signo; |
794 | si.si_errno = info->ssi_errno; | |
795 | si.si_code = info->ssi_code; | |
796 | ||
797 | /* Convert the minimal set of fields defined by POSIX. | |
798 | * Positive si_code values are reserved for kernel-generated | |
799 | * signals, where the valid siginfo fields are determined by | |
800 | * the signal number. But according to POSIX, it is unspecified | |
801 | * whether SI_USER and SI_QUEUE have values less than or equal to | |
802 | * zero. | |
803 | */ | |
804 | if (info->ssi_code == SI_USER || info->ssi_code == SI_QUEUE || | |
805 | info->ssi_code <= 0) { | |
806 | /* SIGTERM, etc. */ | |
807 | si.si_pid = info->ssi_pid; | |
808 | si.si_uid = info->ssi_uid; | |
809 | } else if (info->ssi_signo == SIGILL || info->ssi_signo == SIGFPE || | |
810 | info->ssi_signo == SIGSEGV || info->ssi_signo == SIGBUS) { | |
811 | si.si_addr = (void *)(uintptr_t)info->ssi_addr; | |
812 | } else if (info->ssi_signo == SIGCHLD) { | |
813 | si.si_pid = info->ssi_pid; | |
814 | si.si_status = info->ssi_status; | |
815 | si.si_uid = info->ssi_uid; | |
d98d4072 PB |
816 | } |
817 | action->sa_sigaction(info->ssi_signo, &si, NULL); | |
818 | } | |
e47f4765 MP |
819 | |
820 | #ifndef HOST_NAME_MAX | |
821 | # ifdef _POSIX_HOST_NAME_MAX | |
822 | # define HOST_NAME_MAX _POSIX_HOST_NAME_MAX | |
823 | # else | |
824 | # define HOST_NAME_MAX 255 | |
825 | # endif | |
826 | #endif | |
827 | ||
828 | char *qemu_get_host_name(Error **errp) | |
829 | { | |
830 | long len = -1; | |
831 | g_autofree char *hostname = NULL; | |
832 | ||
833 | #ifdef _SC_HOST_NAME_MAX | |
834 | len = sysconf(_SC_HOST_NAME_MAX); | |
835 | #endif /* _SC_HOST_NAME_MAX */ | |
836 | ||
837 | if (len < 0) { | |
838 | len = HOST_NAME_MAX; | |
839 | } | |
840 | ||
841 | /* Unfortunately, gethostname() below does not guarantee a | |
842 | * NULL terminated string. Therefore, allocate one byte more | |
843 | * to be sure. */ | |
844 | hostname = g_new0(char, len + 1); | |
845 | ||
846 | if (gethostname(hostname, len) < 0) { | |
847 | error_setg_errno(errp, errno, | |
848 | "cannot get hostname"); | |
849 | return NULL; | |
850 | } | |
851 | ||
852 | return g_steal_pointer(&hostname); | |
853 | } | |
ad06ef0e AB |
854 | |
855 | size_t qemu_get_host_physmem(void) | |
856 | { | |
857 | #ifdef _SC_PHYS_PAGES | |
858 | long pages = sysconf(_SC_PHYS_PAGES); | |
859 | if (pages > 0) { | |
860 | if (pages > SIZE_MAX / qemu_real_host_page_size) { | |
861 | return SIZE_MAX; | |
862 | } else { | |
863 | return pages * qemu_real_host_page_size; | |
864 | } | |
865 | } | |
866 | #endif | |
867 | return 0; | |
868 | } |