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 <libgen.h>
40#include <sys/signal.h>
41#include "qemu/cutils.h"
42
43#ifdef CONFIG_LINUX
44#include <sys/syscall.h>
45#endif
46
47#ifdef __FreeBSD__
48#include <sys/sysctl.h>
49#include <sys/user.h>
50#include <libutil.h>
51#endif
52
53#ifdef __NetBSD__
54#include <sys/sysctl.h>
55#endif
56
57#include "qemu/mmap-alloc.h"
58
59#ifdef CONFIG_DEBUG_STACK_USAGE
60#include "qemu/error-report.h"
61#endif
62
63#define MAX_MEM_PREALLOC_THREAD_COUNT 16
64
65struct MemsetThread {
66 char *addr;
67 size_t numpages;
68 size_t hpagesize;
69 QemuThread pgthread;
70 sigjmp_buf env;
71};
72typedef struct MemsetThread MemsetThread;
73
74static MemsetThread *memset_thread;
75static int memset_num_threads;
76static bool memset_thread_failed;
77
78int qemu_get_thread_id(void)
79{
80#if defined(__linux__)
81 return syscall(SYS_gettid);
82#else
83 return getpid();
84#endif
85}
86
87int qemu_daemon(int nochdir, int noclose)
88{
89 return daemon(nochdir, noclose);
90}
91
92bool qemu_write_pidfile(const char *path, Error **errp)
93{
94 int fd;
95 char pidstr[32];
96
97 while (1) {
98 struct stat a, b;
99 struct flock lock = {
100 .l_type = F_WRLCK,
101 .l_whence = SEEK_SET,
102 .l_len = 0,
103 };
104
105 fd = qemu_open(path, O_CREAT | O_WRONLY, S_IRUSR | S_IWUSR);
106 if (fd == -1) {
107 error_setg_errno(errp, errno, "Cannot open pid file");
108 return false;
109 }
110
111 if (fstat(fd, &b) < 0) {
112 error_setg_errno(errp, errno, "Cannot stat file");
113 goto fail_close;
114 }
115
116 if (fcntl(fd, F_SETLK, &lock)) {
117 error_setg_errno(errp, errno, "Cannot lock pid file");
118 goto fail_close;
119 }
120
121
122
123
124
125 if (stat(path, &a) < 0) {
126
127
128
129
130 close(fd);
131 continue;
132 }
133
134 if (a.st_ino == b.st_ino) {
135 break;
136 }
137
138
139
140
141
142 close(fd);
143 }
144
145 if (ftruncate(fd, 0) < 0) {
146 error_setg_errno(errp, errno, "Failed to truncate pid file");
147 goto fail_unlink;
148 }
149
150 snprintf(pidstr, sizeof(pidstr), FMT_pid "\n", getpid());
151 if (write(fd, pidstr, strlen(pidstr)) != strlen(pidstr)) {
152 error_setg(errp, "Failed to write pid file");
153 goto fail_unlink;
154 }
155
156 return true;
157
158fail_unlink:
159 unlink(path);
160fail_close:
161 close(fd);
162 return false;
163}
164
165void *qemu_oom_check(void *ptr)
166{
167 if (ptr == NULL) {
168 fprintf(stderr, "Failed to allocate memory: %s\n", strerror(errno));
169 abort();
170 }
171 return ptr;
172}
173
174void *qemu_try_memalign(size_t alignment, size_t size)
175{
176 void *ptr;
177
178 if (alignment < sizeof(void*)) {
179 alignment = sizeof(void*);
180 }
181
182#if defined(CONFIG_POSIX_MEMALIGN)
183 int ret;
184 ret = posix_memalign(&ptr, alignment, size);
185 if (ret != 0) {
186 errno = ret;
187 ptr = NULL;
188 }
189#elif defined(CONFIG_BSD)
190 ptr = valloc(size);
191#else
192 ptr = memalign(alignment, size);
193#endif
194 trace_qemu_memalign(alignment, size, ptr);
195 return ptr;
196}
197
198void *qemu_memalign(size_t alignment, size_t size)
199{
200 return qemu_oom_check(qemu_try_memalign(alignment, size));
201}
202
203
204void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment, bool shared)
205{
206 size_t align = QEMU_VMALLOC_ALIGN;
207 void *ptr = qemu_ram_mmap(-1, size, align, shared, false);
208
209 if (ptr == MAP_FAILED) {
210 return NULL;
211 }
212
213 if (alignment) {
214 *alignment = align;
215 }
216
217 trace_qemu_anon_ram_alloc(size, ptr);
218 return ptr;
219}
220
221void qemu_vfree(void *ptr)
222{
223 trace_qemu_vfree(ptr);
224 free(ptr);
225}
226
227void qemu_anon_ram_free(void *ptr, size_t size)
228{
229 trace_qemu_anon_ram_free(ptr, size);
230 qemu_ram_munmap(-1, ptr, size);
231}
232
233void qemu_set_block(int fd)
234{
235 int f;
236 f = fcntl(fd, F_GETFL);
237 assert(f != -1);
238 f = fcntl(fd, F_SETFL, f & ~O_NONBLOCK);
239 assert(f != -1);
240}
241
242void qemu_set_nonblock(int fd)
243{
244 int f;
245 f = fcntl(fd, F_GETFL);
246 assert(f != -1);
247 f = fcntl(fd, F_SETFL, f | O_NONBLOCK);
248#ifdef __OpenBSD__
249 if (f == -1) {
250
251
252
253
254
255
256 assert(errno == ENODEV);
257 }
258#else
259 assert(f != -1);
260#endif
261}
262
263int socket_set_fast_reuse(int fd)
264{
265 int val = 1, ret;
266
267 ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
268 (const char *)&val, sizeof(val));
269
270 assert(ret == 0);
271
272 return ret;
273}
274
275void qemu_set_cloexec(int fd)
276{
277 int f;
278 f = fcntl(fd, F_GETFD);
279 assert(f != -1);
280 f = fcntl(fd, F_SETFD, f | FD_CLOEXEC);
281 assert(f != -1);
282}
283
284
285
286
287int qemu_pipe(int pipefd[2])
288{
289 int ret;
290
291#ifdef CONFIG_PIPE2
292 ret = pipe2(pipefd, O_CLOEXEC);
293 if (ret != -1 || errno != ENOSYS) {
294 return ret;
295 }
296#endif
297 ret = pipe(pipefd);
298 if (ret == 0) {
299 qemu_set_cloexec(pipefd[0]);
300 qemu_set_cloexec(pipefd[1]);
301 }
302
303 return ret;
304}
305
306char *
307qemu_get_local_state_pathname(const char *relative_pathname)
308{
309 return g_strdup_printf("%s/%s", CONFIG_QEMU_LOCALSTATEDIR,
310 relative_pathname);
311}
312
313void qemu_set_tty_echo(int fd, bool echo)
314{
315 struct termios tty;
316
317 tcgetattr(fd, &tty);
318
319 if (echo) {
320 tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN;
321 } else {
322 tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN);
323 }
324
325 tcsetattr(fd, TCSANOW, &tty);
326}
327
328static char exec_dir[PATH_MAX];
329
330void qemu_init_exec_dir(const char *argv0)
331{
332 char *dir;
333 char *p = NULL;
334 char buf[PATH_MAX];
335
336 assert(!exec_dir[0]);
337
338#if defined(__linux__)
339 {
340 int len;
341 len = readlink("/proc/self/exe", buf, sizeof(buf) - 1);
342 if (len > 0) {
343 buf[len] = 0;
344 p = buf;
345 }
346 }
347#elif defined(__FreeBSD__) \
348 || (defined(__NetBSD__) && defined(KERN_PROC_PATHNAME))
349 {
350#if defined(__FreeBSD__)
351 static int mib[4] = {CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1};
352#else
353 static int mib[4] = {CTL_KERN, KERN_PROC_ARGS, -1, KERN_PROC_PATHNAME};
354#endif
355 size_t len = sizeof(buf) - 1;
356
357 *buf = '\0';
358 if (!sysctl(mib, ARRAY_SIZE(mib), buf, &len, NULL, 0) &&
359 *buf) {
360 buf[sizeof(buf) - 1] = '\0';
361 p = buf;
362 }
363 }
364#endif
365
366
367 if (!p) {
368 if (!argv0) {
369 return;
370 }
371 p = realpath(argv0, buf);
372 if (!p) {
373 return;
374 }
375 }
376 dir = g_path_get_dirname(p);
377
378 pstrcpy(exec_dir, sizeof(exec_dir), dir);
379
380 g_free(dir);
381}
382
383char *qemu_get_exec_dir(void)
384{
385 return g_strdup(exec_dir);
386}
387
388static void sigbus_handler(int signal)
389{
390 int i;
391 if (memset_thread) {
392 for (i = 0; i < memset_num_threads; i++) {
393 if (qemu_thread_is_self(&memset_thread[i].pgthread)) {
394 siglongjmp(memset_thread[i].env, 1);
395 }
396 }
397 }
398}
399
400static void *do_touch_pages(void *arg)
401{
402 MemsetThread *memset_args = (MemsetThread *)arg;
403 sigset_t set, oldset;
404
405
406 sigemptyset(&set);
407 sigaddset(&set, SIGBUS);
408 pthread_sigmask(SIG_UNBLOCK, &set, &oldset);
409
410 if (sigsetjmp(memset_args->env, 1)) {
411 memset_thread_failed = true;
412 } else {
413 char *addr = memset_args->addr;
414 size_t numpages = memset_args->numpages;
415 size_t hpagesize = memset_args->hpagesize;
416 size_t i;
417 for (i = 0; i < numpages; i++) {
418
419
420
421
422
423
424
425
426
427
428
429
430 *(volatile char *)addr = *addr;
431 addr += hpagesize;
432 }
433 }
434 pthread_sigmask(SIG_SETMASK, &oldset, NULL);
435 return NULL;
436}
437
438static inline int get_memset_num_threads(int smp_cpus)
439{
440 long host_procs = sysconf(_SC_NPROCESSORS_ONLN);
441 int ret = 1;
442
443 if (host_procs > 0) {
444 ret = MIN(MIN(host_procs, MAX_MEM_PREALLOC_THREAD_COUNT), smp_cpus);
445 }
446
447 return ret;
448}
449
450static bool touch_all_pages(char *area, size_t hpagesize, size_t numpages,
451 int smp_cpus)
452{
453 size_t numpages_per_thread;
454 size_t size_per_thread;
455 char *addr = area;
456 int i = 0;
457
458 memset_thread_failed = false;
459 memset_num_threads = get_memset_num_threads(smp_cpus);
460 memset_thread = g_new0(MemsetThread, memset_num_threads);
461 numpages_per_thread = (numpages / memset_num_threads);
462 size_per_thread = (hpagesize * numpages_per_thread);
463 for (i = 0; i < memset_num_threads; i++) {
464 memset_thread[i].addr = addr;
465 memset_thread[i].numpages = (i == (memset_num_threads - 1)) ?
466 numpages : numpages_per_thread;
467 memset_thread[i].hpagesize = hpagesize;
468 qemu_thread_create(&memset_thread[i].pgthread, "touch_pages",
469 do_touch_pages, &memset_thread[i],
470 QEMU_THREAD_JOINABLE);
471 addr += size_per_thread;
472 numpages -= numpages_per_thread;
473 }
474 for (i = 0; i < memset_num_threads; i++) {
475 qemu_thread_join(&memset_thread[i].pgthread);
476 }
477 g_free(memset_thread);
478 memset_thread = NULL;
479
480 return memset_thread_failed;
481}
482
483void os_mem_prealloc(int fd, char *area, size_t memory, int smp_cpus,
484 Error **errp)
485{
486 int ret;
487 struct sigaction act, oldact;
488 size_t hpagesize = qemu_fd_getpagesize(fd);
489 size_t numpages = DIV_ROUND_UP(memory, hpagesize);
490
491 memset(&act, 0, sizeof(act));
492 act.sa_handler = &sigbus_handler;
493 act.sa_flags = 0;
494
495 ret = sigaction(SIGBUS, &act, &oldact);
496 if (ret) {
497 error_setg_errno(errp, errno,
498 "os_mem_prealloc: failed to install signal handler");
499 return;
500 }
501
502
503 if (touch_all_pages(area, hpagesize, numpages, smp_cpus)) {
504 error_setg(errp, "os_mem_prealloc: Insufficient free host memory "
505 "pages available to allocate guest RAM");
506 }
507
508 ret = sigaction(SIGBUS, &oldact, NULL);
509 if (ret) {
510
511 perror("os_mem_prealloc: failed to reinstall signal handler");
512 exit(1);
513 }
514}
515
516uint64_t qemu_get_pmem_size(const char *filename, Error **errp)
517{
518 struct stat st;
519
520 if (stat(filename, &st) < 0) {
521 error_setg(errp, "unable to stat pmem file \"%s\"", filename);
522 return 0;
523 }
524
525#if defined(__linux__)
526
527 if (S_ISCHR(st.st_mode)) {
528 char *subsystem_path = NULL;
529 char *subsystem = NULL;
530 char *size_path = NULL;
531 char *size_str = NULL;
532 uint64_t ret = 0;
533
534 subsystem_path = g_strdup_printf("/sys/dev/char/%d:%d/subsystem",
535 major(st.st_rdev), minor(st.st_rdev));
536 subsystem = g_file_read_link(subsystem_path, NULL);
537 if (!subsystem) {
538 error_setg(errp, "unable to read subsystem for pmem file \"%s\"",
539 filename);
540 goto devdax_err;
541 }
542
543 if (!g_str_has_suffix(subsystem, "/dax")) {
544 error_setg(errp, "pmem file \"%s\" is not a dax device", filename);
545 goto devdax_err;
546 }
547
548 size_path = g_strdup_printf("/sys/dev/char/%d:%d/size",
549 major(st.st_rdev), minor(st.st_rdev));
550 if (!g_file_get_contents(size_path, &size_str, NULL, NULL)) {
551 error_setg(errp, "unable to read size for pmem file \"%s\"",
552 size_path);
553 goto devdax_err;
554 }
555
556 ret = g_ascii_strtoull(size_str, NULL, 0);
557
558devdax_err:
559 g_free(size_str);
560 g_free(size_path);
561 g_free(subsystem);
562 g_free(subsystem_path);
563 return ret;
564 }
565#endif
566
567 return st.st_size;
568}
569
570char *qemu_get_pid_name(pid_t pid)
571{
572 char *name = NULL;
573
574#if defined(__FreeBSD__)
575
576 struct kinfo_proc *proc = kinfo_getproc(pid);
577
578 if (proc) {
579 name = g_strdup(proc->ki_comm);
580 free(proc);
581 }
582#else
583
584 char *pid_path;
585 size_t len;
586
587 pid_path = g_strdup_printf("/proc/%d/cmdline", pid);
588 g_file_get_contents(pid_path, &name, &len, NULL);
589 g_free(pid_path);
590#endif
591
592 return name;
593}
594
595
596pid_t qemu_fork(Error **errp)
597{
598 sigset_t oldmask, newmask;
599 struct sigaction sig_action;
600 int saved_errno;
601 pid_t pid;
602
603
604
605
606
607 sigfillset(&newmask);
608 if (pthread_sigmask(SIG_SETMASK, &newmask, &oldmask) != 0) {
609 error_setg_errno(errp, errno,
610 "cannot block signals");
611 return -1;
612 }
613
614 pid = fork();
615 saved_errno = errno;
616
617 if (pid < 0) {
618
619
620 (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
621 error_setg_errno(errp, saved_errno,
622 "cannot fork child process");
623 errno = saved_errno;
624 return -1;
625 } else if (pid) {
626
627
628
629
630
631
632 (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
633 } else {
634
635 size_t i;
636
637
638
639
640 sig_action.sa_handler = SIG_DFL;
641 sig_action.sa_flags = 0;
642 sigemptyset(&sig_action.sa_mask);
643
644 for (i = 1; i < NSIG; i++) {
645
646
647
648 (void)sigaction(i, &sig_action, NULL);
649 }
650
651
652
653
654 sigemptyset(&newmask);
655 if (pthread_sigmask(SIG_SETMASK, &newmask, NULL) != 0) {
656 Error *local_err = NULL;
657 error_setg_errno(&local_err, errno,
658 "cannot unblock signals");
659 error_report_err(local_err);
660 _exit(1);
661 }
662 }
663 return pid;
664}
665
666void *qemu_alloc_stack(size_t *sz)
667{
668 void *ptr, *guardpage;
669 int flags;
670#ifdef CONFIG_DEBUG_STACK_USAGE
671 void *ptr2;
672#endif
673 size_t pagesz = getpagesize();
674#ifdef _SC_THREAD_STACK_MIN
675
676 long min_stack_sz = sysconf(_SC_THREAD_STACK_MIN);
677 *sz = MAX(MAX(min_stack_sz, 0), *sz);
678#endif
679
680 *sz = ROUND_UP(*sz, pagesz);
681
682 *sz += pagesz;
683
684 flags = MAP_PRIVATE | MAP_ANONYMOUS;
685#if defined(MAP_STACK) && defined(__OpenBSD__)
686
687
688
689
690
691
692 flags |= MAP_STACK;
693#endif
694
695 ptr = mmap(NULL, *sz, PROT_READ | PROT_WRITE, flags, -1, 0);
696 if (ptr == MAP_FAILED) {
697 perror("failed to allocate memory for stack");
698 abort();
699 }
700
701#if defined(HOST_IA64)
702
703 guardpage = ptr + (((*sz - pagesz) / 2) & ~pagesz);
704#elif defined(HOST_HPPA)
705
706 guardpage = ptr + *sz - pagesz;
707#else
708
709 guardpage = ptr;
710#endif
711 if (mprotect(guardpage, pagesz, PROT_NONE) != 0) {
712 perror("failed to set up stack guard page");
713 abort();
714 }
715
716#ifdef CONFIG_DEBUG_STACK_USAGE
717 for (ptr2 = ptr + pagesz; ptr2 < ptr + *sz; ptr2 += sizeof(uint32_t)) {
718 *(uint32_t *)ptr2 = 0xdeadbeaf;
719 }
720#endif
721
722 return ptr;
723}
724
725#ifdef CONFIG_DEBUG_STACK_USAGE
726static __thread unsigned int max_stack_usage;
727#endif
728
729void qemu_free_stack(void *stack, size_t sz)
730{
731#ifdef CONFIG_DEBUG_STACK_USAGE
732 unsigned int usage;
733 void *ptr;
734
735 for (ptr = stack + getpagesize(); ptr < stack + sz;
736 ptr += sizeof(uint32_t)) {
737 if (*(uint32_t *)ptr != 0xdeadbeaf) {
738 break;
739 }
740 }
741 usage = sz - (uintptr_t) (ptr - stack);
742 if (usage > max_stack_usage) {
743 error_report("thread %d max stack usage increased from %u to %u",
744 qemu_get_thread_id(), max_stack_usage, usage);
745 max_stack_usage = usage;
746 }
747#endif
748
749 munmap(stack, sz);
750}
751
752void sigaction_invoke(struct sigaction *action,
753 struct qemu_signalfd_siginfo *info)
754{
755 siginfo_t si = {};
756 si.si_signo = info->ssi_signo;
757 si.si_errno = info->ssi_errno;
758 si.si_code = info->ssi_code;
759
760
761
762
763
764
765
766
767 if (info->ssi_code == SI_USER || info->ssi_code == SI_QUEUE ||
768 info->ssi_code <= 0) {
769
770 si.si_pid = info->ssi_pid;
771 si.si_uid = info->ssi_uid;
772 } else if (info->ssi_signo == SIGILL || info->ssi_signo == SIGFPE ||
773 info->ssi_signo == SIGSEGV || info->ssi_signo == SIGBUS) {
774 si.si_addr = (void *)(uintptr_t)info->ssi_addr;
775 } else if (info->ssi_signo == SIGCHLD) {
776 si.si_pid = info->ssi_pid;
777 si.si_status = info->ssi_status;
778 si.si_uid = info->ssi_uid;
779 }
780 action->sa_sigaction(info->ssi_signo, &si, NULL);
781}
782