1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29#include "qemu/osdep.h"
30#include <termios.h>
31
32#include <glib/gprintf.h>
33
34#include "qemu-common.h"
35#include "sysemu/sysemu.h"
36#include "trace.h"
37#include "qapi/error.h"
38#include "qemu/sockets.h"
39#include "qemu/thread.h"
40#include <libgen.h>
41#include "qemu/cutils.h"
42#include "qemu/compiler.h"
43
44#ifdef CONFIG_LINUX
45#include <sys/syscall.h>
46#endif
47
48#ifdef __FreeBSD__
49#include <sys/sysctl.h>
50#include <sys/user.h>
51#include <sys/thr.h>
52#include <libutil.h>
53#endif
54
55#ifdef __NetBSD__
56#include <sys/sysctl.h>
57#include <lwp.h>
58#endif
59
60#ifdef __APPLE__
61#include <mach-o/dyld.h>
62#endif
63
64#ifdef __HAIKU__
65#include <kernel/image.h>
66#endif
67
68#include "qemu/mmap-alloc.h"
69
70#ifdef CONFIG_DEBUG_STACK_USAGE
71#include "qemu/error-report.h"
72#endif
73
74#define MAX_MEM_PREALLOC_THREAD_COUNT 16
75
76struct MemsetThread {
77 char *addr;
78 size_t numpages;
79 size_t hpagesize;
80 QemuThread pgthread;
81 sigjmp_buf env;
82};
83typedef struct MemsetThread MemsetThread;
84
85static MemsetThread *memset_thread;
86static int memset_num_threads;
87static bool memset_thread_failed;
88
89static QemuMutex page_mutex;
90static QemuCond page_cond;
91static bool threads_created_flag;
92
93int qemu_get_thread_id(void)
94{
95#if defined(__linux__)
96 return syscall(SYS_gettid);
97#elif defined(__FreeBSD__)
98
99 long tid;
100 thr_self(&tid);
101 return (int)tid;
102#elif defined(__NetBSD__)
103 return _lwp_self();
104#elif defined(__OpenBSD__)
105 return getthrid();
106#else
107 return getpid();
108#endif
109}
110
111int qemu_daemon(int nochdir, int noclose)
112{
113 return daemon(nochdir, noclose);
114}
115
116bool 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;
123 struct flock lock = {
124 .l_type = F_WRLCK,
125 .l_whence = SEEK_SET,
126 .l_len = 0,
127 };
128
129 fd = qemu_open_old(path, O_CREAT | O_WRONLY, S_IRUSR | S_IWUSR);
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
140 if (fcntl(fd, F_SETLK, &lock)) {
141 error_setg_errno(errp, errno, "Cannot lock pid file");
142 goto fail_close;
143 }
144
145
146
147
148
149 if (stat(path, &a) < 0) {
150
151
152
153
154 close(fd);
155 continue;
156 }
157
158 if (a.st_ino == b.st_ino) {
159 break;
160 }
161
162
163
164
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
182fail_unlink:
183 unlink(path);
184fail_close:
185 close(fd);
186 return false;
187}
188
189void *qemu_oom_check(void *ptr)
190{
191 if (ptr == NULL) {
192 fprintf(stderr, "Failed to allocate memory: %s\n", strerror(errno));
193 abort();
194 }
195 return ptr;
196}
197
198void *qemu_try_memalign(size_t alignment, size_t size)
199{
200 void *ptr;
201
202 if (alignment < sizeof(void*)) {
203 alignment = sizeof(void*);
204 } else {
205 g_assert(is_power_of_2(alignment));
206 }
207
208#if defined(CONFIG_POSIX_MEMALIGN)
209 int ret;
210 ret = posix_memalign(&ptr, alignment, size);
211 if (ret != 0) {
212 errno = ret;
213 ptr = NULL;
214 }
215#elif defined(CONFIG_BSD)
216 ptr = valloc(size);
217#else
218 ptr = memalign(alignment, size);
219#endif
220 trace_qemu_memalign(alignment, size, ptr);
221 return ptr;
222}
223
224void *qemu_memalign(size_t alignment, size_t size)
225{
226 return qemu_oom_check(qemu_try_memalign(alignment, size));
227}
228
229
230void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment, bool shared)
231{
232 size_t align = QEMU_VMALLOC_ALIGN;
233 void *ptr = qemu_ram_mmap(-1, size, align, false, shared, false, 0);
234
235 if (ptr == MAP_FAILED) {
236 return NULL;
237 }
238
239 if (alignment) {
240 *alignment = align;
241 }
242
243 trace_qemu_anon_ram_alloc(size, ptr);
244 return ptr;
245}
246
247void qemu_vfree(void *ptr)
248{
249 trace_qemu_vfree(ptr);
250 free(ptr);
251}
252
253void qemu_anon_ram_free(void *ptr, size_t size)
254{
255 trace_qemu_anon_ram_free(ptr, size);
256 qemu_ram_munmap(-1, ptr, size);
257}
258
259void qemu_set_block(int fd)
260{
261 int f;
262 f = fcntl(fd, F_GETFL);
263 assert(f != -1);
264 f = fcntl(fd, F_SETFL, f & ~O_NONBLOCK);
265 assert(f != -1);
266}
267
268int qemu_try_set_nonblock(int fd)
269{
270 int f;
271 f = fcntl(fd, F_GETFL);
272 if (f == -1) {
273 return -errno;
274 }
275 if (fcntl(fd, F_SETFL, f | O_NONBLOCK) == -1) {
276#ifdef __OpenBSD__
277
278
279
280
281
282
283 if (errno == ENODEV) {
284 return 0;
285 }
286#endif
287 return -errno;
288 }
289 return 0;
290}
291
292void qemu_set_nonblock(int fd)
293{
294 int f;
295 f = qemu_try_set_nonblock(fd);
296 assert(f == 0);
297}
298
299int 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
311void qemu_set_cloexec(int fd)
312{
313 int f;
314 f = fcntl(fd, F_GETFD);
315 assert(f != -1);
316 f = fcntl(fd, F_SETFD, f | FD_CLOEXEC);
317 assert(f != -1);
318}
319
320
321
322
323int 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}
341
342char *
343qemu_get_local_state_pathname(const char *relative_pathname)
344{
345 g_autofree char *dir = g_strdup_printf("%s/%s",
346 CONFIG_QEMU_LOCALSTATEDIR,
347 relative_pathname);
348 return get_relocated_path(dir);
349}
350
351void 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}
365
366static const char *exec_dir;
367
368void qemu_init_exec_dir(const char *argv0)
369{
370 char *p = NULL;
371 char buf[PATH_MAX];
372
373 if (exec_dir) {
374 return;
375 }
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 }
386#elif defined(__FreeBSD__) \
387 || (defined(__NetBSD__) && defined(KERN_PROC_PATHNAME))
388 {
389#if defined(__FreeBSD__)
390 static int mib[4] = {CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1};
391#else
392 static int mib[4] = {CTL_KERN, KERN_PROC_ARGS, -1, KERN_PROC_PATHNAME};
393#endif
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 }
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 }
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 }
429#endif
430
431
432 if (!p && argv0) {
433 p = realpath(argv0, buf);
434 }
435 if (p) {
436 exec_dir = g_path_get_dirname(p);
437 } else {
438 exec_dir = CONFIG_BINDIR;
439 }
440}
441
442const char *qemu_get_exec_dir(void)
443{
444 return exec_dir;
445}
446
447static void sigbus_handler(int signal)
448{
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 }
457}
458
459static void *do_touch_pages(void *arg)
460{
461 MemsetThread *memset_args = (MemsetThread *)arg;
462 sigset_t set, oldset;
463
464
465
466
467
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
475
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 {
483 char *addr = memset_args->addr;
484 size_t numpages = memset_args->numpages;
485 size_t hpagesize = memset_args->hpagesize;
486 size_t i;
487 for (i = 0; i < numpages; i++) {
488
489
490
491
492
493
494
495
496
497
498
499
500 *(volatile char *)addr = *addr;
501 addr += hpagesize;
502 }
503 }
504 pthread_sigmask(SIG_SETMASK, &oldset, NULL);
505 return NULL;
506}
507
508static 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
517 return ret;
518}
519
520static bool touch_all_pages(char *area, size_t hpagesize, size_t numpages,
521 int smp_cpus)
522{
523 static gsize initialized = 0;
524 size_t numpages_per_thread, leftover;
525 char *addr = area;
526 int i = 0;
527
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
534 memset_thread_failed = false;
535 threads_created_flag = false;
536 memset_num_threads = get_memset_num_threads(smp_cpus);
537 memset_thread = g_new0(MemsetThread, memset_num_threads);
538 numpages_per_thread = numpages / memset_num_threads;
539 leftover = numpages % memset_num_threads;
540 for (i = 0; i < memset_num_threads; i++) {
541 memset_thread[i].addr = addr;
542 memset_thread[i].numpages = numpages_per_thread + (i < leftover);
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);
547 addr += memset_thread[i].numpages * hpagesize;
548 }
549
550 qemu_mutex_lock(&page_mutex);
551 threads_created_flag = true;
552 qemu_cond_broadcast(&page_cond);
553 qemu_mutex_unlock(&page_mutex);
554
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
564void os_mem_prealloc(int fd, char *area, size_t memory, int smp_cpus,
565 Error **errp)
566{
567 int ret;
568 struct sigaction act, oldact;
569 size_t hpagesize = qemu_fd_getpagesize(fd);
570 size_t numpages = DIV_ROUND_UP(memory, hpagesize);
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) {
578 error_setg_errno(errp, errno,
579 "os_mem_prealloc: failed to install signal handler");
580 return;
581 }
582
583
584 if (touch_all_pages(area, hpagesize, numpages, smp_cpus)) {
585 error_setg(errp, "os_mem_prealloc: Insufficient free host memory "
586 "pages available to allocate guest RAM");
587 }
588
589 ret = sigaction(SIGBUS, &oldact, NULL);
590 if (ret) {
591
592 perror("os_mem_prealloc: failed to reinstall signal handler");
593 exit(1);
594 }
595}
596
597char *qemu_get_pid_name(pid_t pid)
598{
599 char *name = NULL;
600
601#if defined(__FreeBSD__)
602
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
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
623pid_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
632
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
646
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
654
655
656
657
658
659 (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
660 } else {
661
662 size_t i;
663
664
665
666
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
673
674
675 (void)sigaction(i, &sig_action, NULL);
676 }
677
678
679
680
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}
692
693void *qemu_alloc_stack(size_t *sz)
694{
695 void *ptr, *guardpage;
696 int flags;
697#ifdef CONFIG_DEBUG_STACK_USAGE
698 void *ptr2;
699#endif
700 size_t pagesz = qemu_real_host_page_size;
701#ifdef _SC_THREAD_STACK_MIN
702
703 long min_stack_sz = sysconf(_SC_THREAD_STACK_MIN);
704 *sz = MAX(MAX(min_stack_sz, 0), *sz);
705#endif
706
707 *sz = ROUND_UP(*sz, pagesz);
708
709 *sz += pagesz;
710
711 flags = MAP_PRIVATE | MAP_ANONYMOUS;
712#if defined(MAP_STACK) && defined(__OpenBSD__)
713
714
715
716
717
718
719 flags |= MAP_STACK;
720#endif
721
722 ptr = mmap(NULL, *sz, PROT_READ | PROT_WRITE, flags, -1, 0);
723 if (ptr == MAP_FAILED) {
724 perror("failed to allocate memory for stack");
725 abort();
726 }
727
728#if defined(HOST_IA64)
729
730 guardpage = ptr + (((*sz - pagesz) / 2) & ~pagesz);
731#elif defined(HOST_HPPA)
732
733 guardpage = ptr + *sz - pagesz;
734#else
735
736 guardpage = ptr;
737#endif
738 if (mprotect(guardpage, pagesz, PROT_NONE) != 0) {
739 perror("failed to set up stack guard page");
740 abort();
741 }
742
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
749 return ptr;
750}
751
752#ifdef CONFIG_DEBUG_STACK_USAGE
753static __thread unsigned int max_stack_usage;
754#endif
755
756void qemu_free_stack(void *stack, size_t sz)
757{
758#ifdef CONFIG_DEBUG_STACK_USAGE
759 unsigned int usage;
760 void *ptr;
761
762 for (ptr = stack + qemu_real_host_page_size; ptr < stack + sz;
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
776 munmap(stack, sz);
777}
778
779
780
781
782
783
784
785
786
787
788QEMU_DISABLE_CFI
789void sigaction_invoke(struct sigaction *action,
790 struct qemu_signalfd_siginfo *info)
791{
792 siginfo_t si = {};
793 si.si_signo = info->ssi_signo;
794 si.si_errno = info->ssi_errno;
795 si.si_code = info->ssi_code;
796
797
798
799
800
801
802
803
804 if (info->ssi_code == SI_USER || info->ssi_code == SI_QUEUE ||
805 info->ssi_code <= 0) {
806
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;
816 }
817 action->sa_sigaction(info->ssi_signo, &si, NULL);
818}
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
828char *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
836
837 if (len < 0) {
838 len = HOST_NAME_MAX;
839 }
840
841
842
843
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}
854
855size_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}
869