]>
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" |
89aec641 | 43 | #include "qemu/units.h" |
c1b0b93b | 44 | |
cbcfa041 PB |
45 | #ifdef CONFIG_LINUX |
46 | #include <sys/syscall.h> | |
47 | #endif | |
cbcfa041 | 48 | |
41975b26 AF |
49 | #ifdef __FreeBSD__ |
50 | #include <sys/sysctl.h> | |
a7764f15 | 51 | #include <sys/user.h> |
9548a891 | 52 | #include <sys/thr.h> |
7dc9ae43 | 53 | #include <libutil.h> |
41975b26 AF |
54 | #endif |
55 | ||
094611b4 KR |
56 | #ifdef __NetBSD__ |
57 | #include <sys/sysctl.h> | |
9548a891 | 58 | #include <lwp.h> |
094611b4 KR |
59 | #endif |
60 | ||
2032e243 DC |
61 | #ifdef __APPLE__ |
62 | #include <mach-o/dyld.h> | |
63 | #endif | |
64 | ||
2b9b9e70 DC |
65 | #ifdef __HAIKU__ |
66 | #include <kernel/image.h> | |
67 | #endif | |
68 | ||
a9c94277 | 69 | #include "qemu/mmap-alloc.h" |
794e8f30 | 70 | |
7d992e4d PL |
71 | #ifdef CONFIG_DEBUG_STACK_USAGE |
72 | #include "qemu/error-report.h" | |
73 | #endif | |
74 | ||
dfd0dcc7 | 75 | #define MAX_MEM_PREALLOC_THREAD_COUNT 16 |
1e356fc1 | 76 | |
dba50678 DH |
77 | struct MemsetThread; |
78 | ||
79 | typedef struct MemsetContext { | |
80 | bool all_threads_created; | |
81 | bool any_thread_failed; | |
82 | struct MemsetThread *threads; | |
83 | int num_threads; | |
84 | } MemsetContext; | |
85 | ||
1e356fc1 JK |
86 | struct MemsetThread { |
87 | char *addr; | |
e947d47d SW |
88 | size_t numpages; |
89 | size_t hpagesize; | |
1e356fc1 JK |
90 | QemuThread pgthread; |
91 | sigjmp_buf env; | |
dba50678 | 92 | MemsetContext *context; |
1e356fc1 JK |
93 | }; |
94 | typedef struct MemsetThread MemsetThread; | |
95 | ||
dba50678 DH |
96 | /* used by sigbus_handler() */ |
97 | static MemsetContext *sigbus_memset_context; | |
1e356fc1 | 98 | |
037fb5eb | 99 | static QemuMutex page_mutex; |
100 | static QemuCond page_cond; | |
037fb5eb | 101 | |
cbcfa041 PB |
102 | int qemu_get_thread_id(void) |
103 | { | |
104 | #if defined(__linux__) | |
105 | return syscall(SYS_gettid); | |
9548a891 DC |
106 | #elif defined(__FreeBSD__) |
107 | /* thread id is up to INT_MAX */ | |
108 | long tid; | |
109 | thr_self(&tid); | |
110 | return (int)tid; | |
111 | #elif defined(__NetBSD__) | |
112 | return _lwp_self(); | |
8edbca51 DC |
113 | #elif defined(__OpenBSD__) |
114 | return getthrid(); | |
cbcfa041 PB |
115 | #else |
116 | return getpid(); | |
117 | #endif | |
118 | } | |
f97742d0 AR |
119 | |
120 | int qemu_daemon(int nochdir, int noclose) | |
121 | { | |
122 | return daemon(nochdir, noclose); | |
123 | } | |
124 | ||
9e6bdef2 MAL |
125 | bool qemu_write_pidfile(const char *path, Error **errp) |
126 | { | |
127 | int fd; | |
128 | char pidstr[32]; | |
129 | ||
130 | while (1) { | |
131 | struct stat a, b; | |
35f7f3fb MAL |
132 | struct flock lock = { |
133 | .l_type = F_WRLCK, | |
134 | .l_whence = SEEK_SET, | |
135 | .l_len = 0, | |
136 | }; | |
9e6bdef2 | 137 | |
448058aa | 138 | fd = qemu_open_old(path, O_CREAT | O_WRONLY, S_IRUSR | S_IWUSR); |
9e6bdef2 MAL |
139 | if (fd == -1) { |
140 | error_setg_errno(errp, errno, "Cannot open pid file"); | |
141 | return false; | |
142 | } | |
143 | ||
144 | if (fstat(fd, &b) < 0) { | |
145 | error_setg_errno(errp, errno, "Cannot stat file"); | |
146 | goto fail_close; | |
147 | } | |
148 | ||
35f7f3fb | 149 | if (fcntl(fd, F_SETLK, &lock)) { |
9e6bdef2 MAL |
150 | error_setg_errno(errp, errno, "Cannot lock pid file"); |
151 | goto fail_close; | |
152 | } | |
153 | ||
154 | /* | |
155 | * Now make sure the path we locked is the same one that now | |
156 | * exists on the filesystem. | |
157 | */ | |
158 | if (stat(path, &a) < 0) { | |
159 | /* | |
160 | * PID file disappeared, someone else must be racing with | |
161 | * us, so try again. | |
162 | */ | |
163 | close(fd); | |
164 | continue; | |
165 | } | |
166 | ||
167 | if (a.st_ino == b.st_ino) { | |
168 | break; | |
169 | } | |
170 | ||
171 | /* | |
172 | * PID file was recreated, someone else must be racing with | |
173 | * us, so try again. | |
174 | */ | |
175 | close(fd); | |
176 | } | |
177 | ||
178 | if (ftruncate(fd, 0) < 0) { | |
179 | error_setg_errno(errp, errno, "Failed to truncate pid file"); | |
180 | goto fail_unlink; | |
181 | } | |
182 | ||
183 | snprintf(pidstr, sizeof(pidstr), FMT_pid "\n", getpid()); | |
184 | if (write(fd, pidstr, strlen(pidstr)) != strlen(pidstr)) { | |
185 | error_setg(errp, "Failed to write pid file"); | |
186 | goto fail_unlink; | |
187 | } | |
188 | ||
189 | return true; | |
190 | ||
191 | fail_unlink: | |
192 | unlink(path); | |
193 | fail_close: | |
194 | close(fd); | |
195 | return false; | |
196 | } | |
197 | ||
b152aa84 | 198 | void *qemu_oom_check(void *ptr) |
c1b0b93b JS |
199 | { |
200 | if (ptr == NULL) { | |
201 | fprintf(stderr, "Failed to allocate memory: %s\n", strerror(errno)); | |
202 | abort(); | |
203 | } | |
204 | return ptr; | |
205 | } | |
c1b0b93b | 206 | |
7d2a35cc | 207 | void *qemu_try_memalign(size_t alignment, size_t size) |
c1b0b93b JS |
208 | { |
209 | void *ptr; | |
e5354657 KW |
210 | |
211 | if (alignment < sizeof(void*)) { | |
212 | alignment = sizeof(void*); | |
ed6f53f9 PMD |
213 | } else { |
214 | g_assert(is_power_of_2(alignment)); | |
e5354657 KW |
215 | } |
216 | ||
9bc5a719 | 217 | #if defined(CONFIG_POSIX_MEMALIGN) |
c1b0b93b JS |
218 | int ret; |
219 | ret = posix_memalign(&ptr, alignment, size); | |
220 | if (ret != 0) { | |
7d2a35cc KW |
221 | errno = ret; |
222 | ptr = NULL; | |
c1b0b93b JS |
223 | } |
224 | #elif defined(CONFIG_BSD) | |
7d2a35cc | 225 | ptr = valloc(size); |
c1b0b93b | 226 | #else |
7d2a35cc | 227 | ptr = memalign(alignment, size); |
c1b0b93b JS |
228 | #endif |
229 | trace_qemu_memalign(alignment, size, ptr); | |
230 | return ptr; | |
231 | } | |
232 | ||
7d2a35cc KW |
233 | void *qemu_memalign(size_t alignment, size_t size) |
234 | { | |
235 | return qemu_oom_check(qemu_try_memalign(alignment, size)); | |
236 | } | |
237 | ||
c1b0b93b | 238 | /* alloc shared memory pages */ |
8dbe22c6 DH |
239 | void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment, bool shared, |
240 | bool noreserve) | |
c1b0b93b | 241 | { |
8dbe22c6 DH |
242 | const uint32_t qemu_map_flags = (shared ? QEMU_MAP_SHARED : 0) | |
243 | (noreserve ? QEMU_MAP_NORESERVE : 0); | |
36b58628 | 244 | size_t align = QEMU_VMALLOC_ALIGN; |
b444f5c0 | 245 | void *ptr = qemu_ram_mmap(-1, size, align, qemu_map_flags, 0); |
36b58628 | 246 | |
7dda5dc8 | 247 | if (ptr == MAP_FAILED) { |
39228250 | 248 | return NULL; |
c2a8238a | 249 | } |
c2a8238a | 250 | |
a2b257d6 IM |
251 | if (alignment) { |
252 | *alignment = align; | |
253 | } | |
c2dfc5ba | 254 | |
6eebf958 | 255 | trace_qemu_anon_ram_alloc(size, ptr); |
c7f4111a | 256 | return ptr; |
c1b0b93b JS |
257 | } |
258 | ||
259 | void qemu_vfree(void *ptr) | |
260 | { | |
261 | trace_qemu_vfree(ptr); | |
262 | free(ptr); | |
263 | } | |
9549e764 | 264 | |
e7a09b92 PB |
265 | void qemu_anon_ram_free(void *ptr, size_t size) |
266 | { | |
267 | trace_qemu_anon_ram_free(ptr, size); | |
53adb9d4 | 268 | qemu_ram_munmap(-1, ptr, size); |
e7a09b92 PB |
269 | } |
270 | ||
f9e8cacc | 271 | void qemu_set_block(int fd) |
154b9a0c PB |
272 | { |
273 | int f; | |
274 | f = fcntl(fd, F_GETFL); | |
da93b820 LQ |
275 | assert(f != -1); |
276 | f = fcntl(fd, F_SETFL, f & ~O_NONBLOCK); | |
277 | assert(f != -1); | |
154b9a0c PB |
278 | } |
279 | ||
894022e6 | 280 | int qemu_try_set_nonblock(int fd) |
9549e764 JS |
281 | { |
282 | int f; | |
283 | f = fcntl(fd, F_GETFL); | |
02cdcc96 | 284 | if (f == -1) { |
894022e6 LV |
285 | return -errno; |
286 | } | |
287 | if (fcntl(fd, F_SETFL, f | O_NONBLOCK) == -1) { | |
894022e6 LV |
288 | return -errno; |
289 | } | |
290 | return 0; | |
291 | } | |
292 | ||
293 | void qemu_set_nonblock(int fd) | |
294 | { | |
295 | int f; | |
296 | f = qemu_try_set_nonblock(fd); | |
297 | assert(f == 0); | |
9549e764 JS |
298 | } |
299 | ||
606600a1 SO |
300 | int socket_set_fast_reuse(int fd) |
301 | { | |
302 | int val = 1, ret; | |
303 | ||
304 | ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, | |
305 | (const char *)&val, sizeof(val)); | |
306 | ||
307 | assert(ret == 0); | |
308 | ||
309 | return ret; | |
310 | } | |
311 | ||
9549e764 JS |
312 | void qemu_set_cloexec(int fd) |
313 | { | |
314 | int f; | |
315 | f = fcntl(fd, F_GETFD); | |
7e6478e7 SS |
316 | assert(f != -1); |
317 | f = fcntl(fd, F_SETFD, f | FD_CLOEXEC); | |
318 | assert(f != -1); | |
9549e764 | 319 | } |
70e72ce4 JS |
320 | |
321 | /* | |
322 | * Creates a pipe with FD_CLOEXEC set on both file descriptors | |
323 | */ | |
324 | int qemu_pipe(int pipefd[2]) | |
325 | { | |
326 | int ret; | |
327 | ||
328 | #ifdef CONFIG_PIPE2 | |
329 | ret = pipe2(pipefd, O_CLOEXEC); | |
330 | if (ret != -1 || errno != ENOSYS) { | |
331 | return ret; | |
332 | } | |
333 | #endif | |
334 | ret = pipe(pipefd); | |
335 | if (ret == 0) { | |
336 | qemu_set_cloexec(pipefd[0]); | |
337 | qemu_set_cloexec(pipefd[1]); | |
338 | } | |
339 | ||
340 | return ret; | |
341 | } | |
38671423 | 342 | |
e2ea3515 LE |
343 | char * |
344 | qemu_get_local_state_pathname(const char *relative_pathname) | |
345 | { | |
fcb4f59c PB |
346 | g_autofree char *dir = g_strdup_printf("%s/%s", |
347 | CONFIG_QEMU_LOCALSTATEDIR, | |
348 | relative_pathname); | |
349 | return get_relocated_path(dir); | |
e2ea3515 | 350 | } |
13401ba0 SH |
351 | |
352 | void qemu_set_tty_echo(int fd, bool echo) | |
353 | { | |
354 | struct termios tty; | |
355 | ||
356 | tcgetattr(fd, &tty); | |
357 | ||
358 | if (echo) { | |
359 | tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN; | |
360 | } else { | |
361 | tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN); | |
362 | } | |
363 | ||
364 | tcsetattr(fd, TCSANOW, &tty); | |
365 | } | |
10f5bff6 | 366 | |
9386a4a7 | 367 | static const char *exec_dir; |
10f5bff6 FZ |
368 | |
369 | void qemu_init_exec_dir(const char *argv0) | |
370 | { | |
10f5bff6 FZ |
371 | char *p = NULL; |
372 | char buf[PATH_MAX]; | |
373 | ||
9386a4a7 | 374 | if (exec_dir) { |
a4c13869 PB |
375 | return; |
376 | } | |
10f5bff6 FZ |
377 | |
378 | #if defined(__linux__) | |
379 | { | |
380 | int len; | |
381 | len = readlink("/proc/self/exe", buf, sizeof(buf) - 1); | |
382 | if (len > 0) { | |
383 | buf[len] = 0; | |
384 | p = buf; | |
385 | } | |
386 | } | |
094611b4 KR |
387 | #elif defined(__FreeBSD__) \ |
388 | || (defined(__NetBSD__) && defined(KERN_PROC_PATHNAME)) | |
10f5bff6 | 389 | { |
094611b4 | 390 | #if defined(__FreeBSD__) |
10f5bff6 | 391 | static int mib[4] = {CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1}; |
094611b4 KR |
392 | #else |
393 | static int mib[4] = {CTL_KERN, KERN_PROC_ARGS, -1, KERN_PROC_PATHNAME}; | |
394 | #endif | |
10f5bff6 FZ |
395 | size_t len = sizeof(buf) - 1; |
396 | ||
397 | *buf = '\0'; | |
398 | if (!sysctl(mib, ARRAY_SIZE(mib), buf, &len, NULL, 0) && | |
399 | *buf) { | |
400 | buf[sizeof(buf) - 1] = '\0'; | |
401 | p = buf; | |
402 | } | |
403 | } | |
2032e243 DC |
404 | #elif defined(__APPLE__) |
405 | { | |
406 | char fpath[PATH_MAX]; | |
407 | uint32_t len = sizeof(fpath); | |
408 | if (_NSGetExecutablePath(fpath, &len) == 0) { | |
409 | p = realpath(fpath, buf); | |
410 | if (!p) { | |
411 | return; | |
412 | } | |
413 | } | |
414 | } | |
2b9b9e70 DC |
415 | #elif defined(__HAIKU__) |
416 | { | |
417 | image_info ii; | |
418 | int32_t c = 0; | |
419 | ||
420 | *buf = '\0'; | |
421 | while (get_next_image_info(0, &c, &ii) == B_OK) { | |
422 | if (ii.type == B_APP_IMAGE) { | |
423 | strncpy(buf, ii.name, sizeof(buf)); | |
424 | buf[sizeof(buf) - 1] = 0; | |
425 | p = buf; | |
426 | break; | |
427 | } | |
428 | } | |
429 | } | |
10f5bff6 FZ |
430 | #endif |
431 | /* If we don't have any way of figuring out the actual executable | |
432 | location then try argv[0]. */ | |
9386a4a7 | 433 | if (!p && argv0) { |
10f5bff6 | 434 | p = realpath(argv0, buf); |
10f5bff6 | 435 | } |
9386a4a7 PB |
436 | if (p) { |
437 | exec_dir = g_path_get_dirname(p); | |
438 | } else { | |
439 | exec_dir = CONFIG_BINDIR; | |
440 | } | |
10f5bff6 FZ |
441 | } |
442 | ||
a4c13869 | 443 | const char *qemu_get_exec_dir(void) |
10f5bff6 | 444 | { |
a4c13869 | 445 | return exec_dir; |
10f5bff6 | 446 | } |
38183310 | 447 | |
38183310 PB |
448 | static void sigbus_handler(int signal) |
449 | { | |
1e356fc1 | 450 | int i; |
dba50678 DH |
451 | |
452 | if (sigbus_memset_context) { | |
453 | for (i = 0; i < sigbus_memset_context->num_threads; i++) { | |
454 | MemsetThread *thread = &sigbus_memset_context->threads[i]; | |
455 | ||
456 | if (qemu_thread_is_self(&thread->pgthread)) { | |
457 | siglongjmp(thread->env, 1); | |
1e356fc1 JK |
458 | } |
459 | } | |
460 | } | |
38183310 PB |
461 | } |
462 | ||
1e356fc1 JK |
463 | static void *do_touch_pages(void *arg) |
464 | { | |
465 | MemsetThread *memset_args = (MemsetThread *)arg; | |
1e356fc1 | 466 | sigset_t set, oldset; |
6c427ab9 | 467 | int ret = 0; |
1e356fc1 | 468 | |
037fb5eb | 469 | /* |
470 | * On Linux, the page faults from the loop below can cause mmap_sem | |
471 | * contention with allocation of the thread stacks. Do not start | |
472 | * clearing until all threads have been created. | |
473 | */ | |
474 | qemu_mutex_lock(&page_mutex); | |
dba50678 | 475 | while (!memset_args->context->all_threads_created) { |
037fb5eb | 476 | qemu_cond_wait(&page_cond, &page_mutex); |
477 | } | |
478 | qemu_mutex_unlock(&page_mutex); | |
479 | ||
1e356fc1 JK |
480 | /* unblock SIGBUS */ |
481 | sigemptyset(&set); | |
482 | sigaddset(&set, SIGBUS); | |
483 | pthread_sigmask(SIG_UNBLOCK, &set, &oldset); | |
484 | ||
485 | if (sigsetjmp(memset_args->env, 1)) { | |
6c427ab9 | 486 | ret = -EFAULT; |
1e356fc1 | 487 | } else { |
e947d47d SW |
488 | char *addr = memset_args->addr; |
489 | size_t numpages = memset_args->numpages; | |
490 | size_t hpagesize = memset_args->hpagesize; | |
491 | size_t i; | |
1e356fc1 | 492 | for (i = 0; i < numpages; i++) { |
9dc44aa5 DB |
493 | /* |
494 | * Read & write back the same value, so we don't | |
495 | * corrupt existing user/app data that might be | |
496 | * stored. | |
497 | * | |
498 | * 'volatile' to stop compiler optimizing this away | |
499 | * to a no-op | |
9dc44aa5 DB |
500 | */ |
501 | *(volatile char *)addr = *addr; | |
1e356fc1 JK |
502 | addr += hpagesize; |
503 | } | |
504 | } | |
505 | pthread_sigmask(SIG_SETMASK, &oldset, NULL); | |
6c427ab9 | 506 | return (void *)(uintptr_t)ret; |
1e356fc1 JK |
507 | } |
508 | ||
a384bfa3 DH |
509 | static void *do_madv_populate_write_pages(void *arg) |
510 | { | |
511 | MemsetThread *memset_args = (MemsetThread *)arg; | |
512 | const size_t size = memset_args->numpages * memset_args->hpagesize; | |
513 | char * const addr = memset_args->addr; | |
514 | int ret = 0; | |
515 | ||
516 | /* See do_touch_pages(). */ | |
517 | qemu_mutex_lock(&page_mutex); | |
dba50678 | 518 | while (!memset_args->context->all_threads_created) { |
a384bfa3 DH |
519 | qemu_cond_wait(&page_cond, &page_mutex); |
520 | } | |
521 | qemu_mutex_unlock(&page_mutex); | |
522 | ||
523 | if (size && qemu_madvise(addr, size, QEMU_MADV_POPULATE_WRITE)) { | |
524 | ret = -errno; | |
525 | } | |
526 | return (void *)(uintptr_t)ret; | |
527 | } | |
528 | ||
89aec641 DH |
529 | static inline int get_memset_num_threads(size_t hpagesize, size_t numpages, |
530 | int smp_cpus) | |
dfd0dcc7 JK |
531 | { |
532 | long host_procs = sysconf(_SC_NPROCESSORS_ONLN); | |
533 | int ret = 1; | |
534 | ||
535 | if (host_procs > 0) { | |
536 | ret = MIN(MIN(host_procs, MAX_MEM_PREALLOC_THREAD_COUNT), smp_cpus); | |
537 | } | |
89aec641 DH |
538 | |
539 | /* Especially with gigantic pages, don't create more threads than pages. */ | |
540 | ret = MIN(ret, numpages); | |
541 | /* Don't start threads to prealloc comparatively little memory. */ | |
542 | ret = MIN(ret, MAX(1, hpagesize * numpages / (64 * MiB))); | |
543 | ||
dfd0dcc7 JK |
544 | /* In case sysconf() fails, we fall back to single threaded */ |
545 | return ret; | |
546 | } | |
547 | ||
6c427ab9 | 548 | static int touch_all_pages(char *area, size_t hpagesize, size_t numpages, |
a384bfa3 | 549 | int smp_cpus, bool use_madv_populate_write) |
1e356fc1 | 550 | { |
78b3f67a | 551 | static gsize initialized = 0; |
dba50678 | 552 | MemsetContext context = { |
89aec641 | 553 | .num_threads = get_memset_num_threads(hpagesize, numpages, smp_cpus), |
dba50678 | 554 | }; |
037fb5eb | 555 | size_t numpages_per_thread, leftover; |
a384bfa3 | 556 | void *(*touch_fn)(void *); |
6c427ab9 | 557 | int ret = 0, i = 0; |
1e356fc1 | 558 | char *addr = area; |
1e356fc1 | 559 | |
78b3f67a PB |
560 | if (g_once_init_enter(&initialized)) { |
561 | qemu_mutex_init(&page_mutex); | |
562 | qemu_cond_init(&page_cond); | |
563 | g_once_init_leave(&initialized, 1); | |
564 | } | |
565 | ||
a384bfa3 DH |
566 | if (use_madv_populate_write) { |
567 | touch_fn = do_madv_populate_write_pages; | |
568 | } else { | |
569 | touch_fn = do_touch_pages; | |
570 | } | |
571 | ||
dba50678 DH |
572 | context.threads = g_new0(MemsetThread, context.num_threads); |
573 | numpages_per_thread = numpages / context.num_threads; | |
574 | leftover = numpages % context.num_threads; | |
575 | for (i = 0; i < context.num_threads; i++) { | |
576 | context.threads[i].addr = addr; | |
577 | context.threads[i].numpages = numpages_per_thread + (i < leftover); | |
578 | context.threads[i].hpagesize = hpagesize; | |
579 | context.threads[i].context = &context; | |
580 | qemu_thread_create(&context.threads[i].pgthread, "touch_pages", | |
581 | touch_fn, &context.threads[i], | |
1e356fc1 | 582 | QEMU_THREAD_JOINABLE); |
dba50678 DH |
583 | addr += context.threads[i].numpages * hpagesize; |
584 | } | |
585 | ||
586 | if (!use_madv_populate_write) { | |
587 | sigbus_memset_context = &context; | |
1e356fc1 | 588 | } |
278fb162 B |
589 | |
590 | qemu_mutex_lock(&page_mutex); | |
dba50678 | 591 | context.all_threads_created = true; |
037fb5eb | 592 | qemu_cond_broadcast(&page_cond); |
278fb162 | 593 | qemu_mutex_unlock(&page_mutex); |
037fb5eb | 594 | |
dba50678 DH |
595 | for (i = 0; i < context.num_threads; i++) { |
596 | int tmp = (uintptr_t)qemu_thread_join(&context.threads[i].pgthread); | |
6c427ab9 DH |
597 | |
598 | if (tmp) { | |
599 | ret = tmp; | |
600 | } | |
1e356fc1 | 601 | } |
dba50678 DH |
602 | |
603 | if (!use_madv_populate_write) { | |
604 | sigbus_memset_context = NULL; | |
605 | } | |
606 | g_free(context.threads); | |
1e356fc1 | 607 | |
6c427ab9 | 608 | return ret; |
1e356fc1 JK |
609 | } |
610 | ||
a384bfa3 DH |
611 | static bool madv_populate_write_possible(char *area, size_t pagesize) |
612 | { | |
613 | return !qemu_madvise(area, pagesize, QEMU_MADV_POPULATE_WRITE) || | |
614 | errno != EINVAL; | |
615 | } | |
616 | ||
1e356fc1 JK |
617 | void os_mem_prealloc(int fd, char *area, size_t memory, int smp_cpus, |
618 | Error **errp) | |
38183310 | 619 | { |
b7bf8f56 | 620 | int ret; |
38183310 | 621 | struct sigaction act, oldact; |
1e356fc1 JK |
622 | size_t hpagesize = qemu_fd_getpagesize(fd); |
623 | size_t numpages = DIV_ROUND_UP(memory, hpagesize); | |
a384bfa3 | 624 | bool use_madv_populate_write; |
38183310 | 625 | |
a384bfa3 DH |
626 | /* |
627 | * Sense on every invocation, as MADV_POPULATE_WRITE cannot be used for | |
628 | * some special mappings, such as mapping /dev/mem. | |
629 | */ | |
630 | use_madv_populate_write = madv_populate_write_possible(area, hpagesize); | |
631 | ||
632 | if (!use_madv_populate_write) { | |
633 | memset(&act, 0, sizeof(act)); | |
634 | act.sa_handler = &sigbus_handler; | |
635 | act.sa_flags = 0; | |
636 | ||
637 | ret = sigaction(SIGBUS, &act, &oldact); | |
638 | if (ret) { | |
639 | error_setg_errno(errp, errno, | |
640 | "os_mem_prealloc: failed to install signal handler"); | |
641 | return; | |
642 | } | |
38183310 PB |
643 | } |
644 | ||
1e356fc1 | 645 | /* touch pages simultaneously */ |
a384bfa3 DH |
646 | ret = touch_all_pages(area, hpagesize, numpages, smp_cpus, |
647 | use_madv_populate_write); | |
6c427ab9 DH |
648 | if (ret) { |
649 | error_setg_errno(errp, -ret, | |
650 | "os_mem_prealloc: preallocating memory failed"); | |
056b68af | 651 | } |
38183310 | 652 | |
a384bfa3 DH |
653 | if (!use_madv_populate_write) { |
654 | ret = sigaction(SIGBUS, &oldact, NULL); | |
655 | if (ret) { | |
656 | /* Terminate QEMU since it can't recover from error */ | |
657 | perror("os_mem_prealloc: failed to reinstall signal handler"); | |
658 | exit(1); | |
659 | } | |
b7bf8f56 | 660 | } |
38183310 | 661 | } |
d57e4e48 | 662 | |
7dc9ae43 MP |
663 | char *qemu_get_pid_name(pid_t pid) |
664 | { | |
665 | char *name = NULL; | |
666 | ||
667 | #if defined(__FreeBSD__) | |
668 | /* BSDs don't have /proc, but they provide a nice substitute */ | |
669 | struct kinfo_proc *proc = kinfo_getproc(pid); | |
670 | ||
671 | if (proc) { | |
672 | name = g_strdup(proc->ki_comm); | |
673 | free(proc); | |
674 | } | |
675 | #else | |
676 | /* Assume a system with reasonable procfs */ | |
677 | char *pid_path; | |
678 | size_t len; | |
679 | ||
680 | pid_path = g_strdup_printf("/proc/%d/cmdline", pid); | |
681 | g_file_get_contents(pid_path, &name, &len, NULL); | |
682 | g_free(pid_path); | |
683 | #endif | |
684 | ||
685 | return name; | |
686 | } | |
687 | ||
688 | ||
57cb38b3 DB |
689 | pid_t qemu_fork(Error **errp) |
690 | { | |
691 | sigset_t oldmask, newmask; | |
692 | struct sigaction sig_action; | |
693 | int saved_errno; | |
694 | pid_t pid; | |
695 | ||
696 | /* | |
697 | * Need to block signals now, so that child process can safely | |
698 | * kill off caller's signal handlers without a race. | |
699 | */ | |
700 | sigfillset(&newmask); | |
701 | if (pthread_sigmask(SIG_SETMASK, &newmask, &oldmask) != 0) { | |
702 | error_setg_errno(errp, errno, | |
703 | "cannot block signals"); | |
704 | return -1; | |
705 | } | |
706 | ||
707 | pid = fork(); | |
708 | saved_errno = errno; | |
709 | ||
710 | if (pid < 0) { | |
711 | /* attempt to restore signal mask, but ignore failure, to | |
712 | * avoid obscuring the fork failure */ | |
713 | (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL); | |
714 | error_setg_errno(errp, saved_errno, | |
715 | "cannot fork child process"); | |
716 | errno = saved_errno; | |
717 | return -1; | |
718 | } else if (pid) { | |
719 | /* parent process */ | |
720 | ||
721 | /* Restore our original signal mask now that the child is | |
722 | * safely running. Only documented failures are EFAULT (not | |
723 | * possible, since we are using just-grabbed mask) or EINVAL | |
724 | * (not possible, since we are using correct arguments). */ | |
725 | (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL); | |
726 | } else { | |
727 | /* child process */ | |
728 | size_t i; | |
729 | ||
730 | /* Clear out all signal handlers from parent so nothing | |
731 | * unexpected can happen in our child once we unblock | |
732 | * signals */ | |
733 | sig_action.sa_handler = SIG_DFL; | |
734 | sig_action.sa_flags = 0; | |
735 | sigemptyset(&sig_action.sa_mask); | |
736 | ||
737 | for (i = 1; i < NSIG; i++) { | |
738 | /* Only possible errors are EFAULT or EINVAL The former | |
739 | * won't happen, the latter we expect, so no need to check | |
740 | * return value */ | |
741 | (void)sigaction(i, &sig_action, NULL); | |
742 | } | |
743 | ||
744 | /* Unmask all signals in child, since we've no idea what the | |
745 | * caller's done with their signal mask and don't want to | |
746 | * propagate that to children */ | |
747 | sigemptyset(&newmask); | |
748 | if (pthread_sigmask(SIG_SETMASK, &newmask, NULL) != 0) { | |
749 | Error *local_err = NULL; | |
750 | error_setg_errno(&local_err, errno, | |
751 | "cannot unblock signals"); | |
752 | error_report_err(local_err); | |
753 | _exit(1); | |
754 | } | |
755 | } | |
756 | return pid; | |
757 | } | |
8737d9e0 PL |
758 | |
759 | void *qemu_alloc_stack(size_t *sz) | |
760 | { | |
761 | void *ptr, *guardpage; | |
fc3d1bad | 762 | int flags; |
7d992e4d PL |
763 | #ifdef CONFIG_DEBUG_STACK_USAGE |
764 | void *ptr2; | |
765 | #endif | |
038adc2f | 766 | size_t pagesz = qemu_real_host_page_size; |
8737d9e0 PL |
767 | #ifdef _SC_THREAD_STACK_MIN |
768 | /* avoid stacks smaller than _SC_THREAD_STACK_MIN */ | |
769 | long min_stack_sz = sysconf(_SC_THREAD_STACK_MIN); | |
770 | *sz = MAX(MAX(min_stack_sz, 0), *sz); | |
771 | #endif | |
772 | /* adjust stack size to a multiple of the page size */ | |
773 | *sz = ROUND_UP(*sz, pagesz); | |
774 | /* allocate one extra page for the guard page */ | |
775 | *sz += pagesz; | |
776 | ||
fc3d1bad BS |
777 | flags = MAP_PRIVATE | MAP_ANONYMOUS; |
778 | #if defined(MAP_STACK) && defined(__OpenBSD__) | |
779 | /* Only enable MAP_STACK on OpenBSD. Other OS's such as | |
780 | * Linux/FreeBSD/NetBSD have a flag with the same name | |
781 | * but have differing functionality. OpenBSD will SEGV | |
782 | * if it spots execution with a stack pointer pointing | |
783 | * at memory that was not allocated with MAP_STACK. | |
784 | */ | |
785 | flags |= MAP_STACK; | |
786 | #endif | |
787 | ||
788 | ptr = mmap(NULL, *sz, PROT_READ | PROT_WRITE, flags, -1, 0); | |
8737d9e0 | 789 | if (ptr == MAP_FAILED) { |
e916a6e8 | 790 | perror("failed to allocate memory for stack"); |
8737d9e0 PL |
791 | abort(); |
792 | } | |
793 | ||
794 | #if defined(HOST_IA64) | |
795 | /* separate register stack */ | |
796 | guardpage = ptr + (((*sz - pagesz) / 2) & ~pagesz); | |
797 | #elif defined(HOST_HPPA) | |
798 | /* stack grows up */ | |
799 | guardpage = ptr + *sz - pagesz; | |
800 | #else | |
801 | /* stack grows down */ | |
802 | guardpage = ptr; | |
803 | #endif | |
804 | if (mprotect(guardpage, pagesz, PROT_NONE) != 0) { | |
e916a6e8 | 805 | perror("failed to set up stack guard page"); |
8737d9e0 PL |
806 | abort(); |
807 | } | |
808 | ||
7d992e4d PL |
809 | #ifdef CONFIG_DEBUG_STACK_USAGE |
810 | for (ptr2 = ptr + pagesz; ptr2 < ptr + *sz; ptr2 += sizeof(uint32_t)) { | |
811 | *(uint32_t *)ptr2 = 0xdeadbeaf; | |
812 | } | |
813 | #endif | |
814 | ||
8737d9e0 PL |
815 | return ptr; |
816 | } | |
817 | ||
7d992e4d PL |
818 | #ifdef CONFIG_DEBUG_STACK_USAGE |
819 | static __thread unsigned int max_stack_usage; | |
820 | #endif | |
821 | ||
8737d9e0 PL |
822 | void qemu_free_stack(void *stack, size_t sz) |
823 | { | |
7d992e4d PL |
824 | #ifdef CONFIG_DEBUG_STACK_USAGE |
825 | unsigned int usage; | |
826 | void *ptr; | |
827 | ||
038adc2f | 828 | for (ptr = stack + qemu_real_host_page_size; ptr < stack + sz; |
7d992e4d PL |
829 | ptr += sizeof(uint32_t)) { |
830 | if (*(uint32_t *)ptr != 0xdeadbeaf) { | |
831 | break; | |
832 | } | |
833 | } | |
834 | usage = sz - (uintptr_t) (ptr - stack); | |
835 | if (usage > max_stack_usage) { | |
836 | error_report("thread %d max stack usage increased from %u to %u", | |
837 | qemu_get_thread_id(), max_stack_usage, usage); | |
838 | max_stack_usage = usage; | |
839 | } | |
840 | #endif | |
841 | ||
8737d9e0 PL |
842 | munmap(stack, sz); |
843 | } | |
d98d4072 | 844 | |
c905a368 DB |
845 | /* |
846 | * Disable CFI checks. | |
847 | * We are going to call a signal hander directly. Such handler may or may not | |
848 | * have been defined in our binary, so there's no guarantee that the pointer | |
849 | * used to set the handler is a cfi-valid pointer. Since the handlers are | |
850 | * stored in kernel memory, changing the handler to an attacker-defined | |
851 | * function requires being able to call a sigaction() syscall, | |
852 | * which is not as easy as overwriting a pointer in memory. | |
853 | */ | |
854 | QEMU_DISABLE_CFI | |
d98d4072 PB |
855 | void sigaction_invoke(struct sigaction *action, |
856 | struct qemu_signalfd_siginfo *info) | |
857 | { | |
02ffa034 | 858 | siginfo_t si = {}; |
d98d4072 PB |
859 | si.si_signo = info->ssi_signo; |
860 | si.si_errno = info->ssi_errno; | |
861 | si.si_code = info->ssi_code; | |
862 | ||
863 | /* Convert the minimal set of fields defined by POSIX. | |
864 | * Positive si_code values are reserved for kernel-generated | |
865 | * signals, where the valid siginfo fields are determined by | |
866 | * the signal number. But according to POSIX, it is unspecified | |
867 | * whether SI_USER and SI_QUEUE have values less than or equal to | |
868 | * zero. | |
869 | */ | |
870 | if (info->ssi_code == SI_USER || info->ssi_code == SI_QUEUE || | |
871 | info->ssi_code <= 0) { | |
872 | /* SIGTERM, etc. */ | |
873 | si.si_pid = info->ssi_pid; | |
874 | si.si_uid = info->ssi_uid; | |
875 | } else if (info->ssi_signo == SIGILL || info->ssi_signo == SIGFPE || | |
876 | info->ssi_signo == SIGSEGV || info->ssi_signo == SIGBUS) { | |
877 | si.si_addr = (void *)(uintptr_t)info->ssi_addr; | |
878 | } else if (info->ssi_signo == SIGCHLD) { | |
879 | si.si_pid = info->ssi_pid; | |
880 | si.si_status = info->ssi_status; | |
881 | si.si_uid = info->ssi_uid; | |
d98d4072 PB |
882 | } |
883 | action->sa_sigaction(info->ssi_signo, &si, NULL); | |
884 | } | |
e47f4765 MP |
885 | |
886 | #ifndef HOST_NAME_MAX | |
887 | # ifdef _POSIX_HOST_NAME_MAX | |
888 | # define HOST_NAME_MAX _POSIX_HOST_NAME_MAX | |
889 | # else | |
890 | # define HOST_NAME_MAX 255 | |
891 | # endif | |
892 | #endif | |
893 | ||
894 | char *qemu_get_host_name(Error **errp) | |
895 | { | |
896 | long len = -1; | |
897 | g_autofree char *hostname = NULL; | |
898 | ||
899 | #ifdef _SC_HOST_NAME_MAX | |
900 | len = sysconf(_SC_HOST_NAME_MAX); | |
901 | #endif /* _SC_HOST_NAME_MAX */ | |
902 | ||
903 | if (len < 0) { | |
904 | len = HOST_NAME_MAX; | |
905 | } | |
906 | ||
907 | /* Unfortunately, gethostname() below does not guarantee a | |
908 | * NULL terminated string. Therefore, allocate one byte more | |
909 | * to be sure. */ | |
910 | hostname = g_new0(char, len + 1); | |
911 | ||
912 | if (gethostname(hostname, len) < 0) { | |
913 | error_setg_errno(errp, errno, | |
914 | "cannot get hostname"); | |
915 | return NULL; | |
916 | } | |
917 | ||
918 | return g_steal_pointer(&hostname); | |
919 | } | |
ad06ef0e AB |
920 | |
921 | size_t qemu_get_host_physmem(void) | |
922 | { | |
923 | #ifdef _SC_PHYS_PAGES | |
924 | long pages = sysconf(_SC_PHYS_PAGES); | |
925 | if (pages > 0) { | |
926 | if (pages > SIZE_MAX / qemu_real_host_page_size) { | |
927 | return SIZE_MAX; | |
928 | } else { | |
929 | return pages * qemu_real_host_page_size; | |
930 | } | |
931 | } | |
932 | #endif | |
933 | return 0; | |
934 | } |