xref: /openbmc/qemu/util/oslib-posix.c (revision 90f9e35b)
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 
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/error-report.h"
39 #include "qemu/madvise.h"
40 #include "qemu/sockets.h"
41 #include "qemu/thread.h"
42 #include <libgen.h>
43 #include "qemu/cutils.h"
44 #include "qemu/compiler.h"
45 #include "qemu/units.h"
46 
47 #ifdef CONFIG_LINUX
48 #include <sys/syscall.h>
49 #endif
50 
51 #ifdef __FreeBSD__
52 #include <sys/sysctl.h>
53 #include <sys/user.h>
54 #include <sys/thr.h>
55 #include <libutil.h>
56 #endif
57 
58 #ifdef __NetBSD__
59 #include <sys/sysctl.h>
60 #include <lwp.h>
61 #endif
62 
63 #ifdef __APPLE__
64 #include <mach-o/dyld.h>
65 #endif
66 
67 #ifdef __HAIKU__
68 #include <kernel/image.h>
69 #endif
70 
71 #include "qemu/mmap-alloc.h"
72 
73 #ifdef CONFIG_DEBUG_STACK_USAGE
74 #include "qemu/error-report.h"
75 #endif
76 
77 #define MAX_MEM_PREALLOC_THREAD_COUNT 16
78 
79 struct MemsetThread;
80 
81 typedef struct MemsetContext {
82     bool all_threads_created;
83     bool any_thread_failed;
84     struct MemsetThread *threads;
85     int num_threads;
86 } MemsetContext;
87 
88 struct MemsetThread {
89     char *addr;
90     size_t numpages;
91     size_t hpagesize;
92     QemuThread pgthread;
93     sigjmp_buf env;
94     MemsetContext *context;
95 };
96 typedef struct MemsetThread MemsetThread;
97 
98 /* used by sigbus_handler() */
99 static MemsetContext *sigbus_memset_context;
100 struct sigaction sigbus_oldact;
101 static QemuMutex sigbus_mutex;
102 
103 static QemuMutex page_mutex;
104 static QemuCond page_cond;
105 
106 int qemu_get_thread_id(void)
107 {
108 #if defined(__linux__)
109     return syscall(SYS_gettid);
110 #elif defined(__FreeBSD__)
111     /* thread id is up to INT_MAX */
112     long tid;
113     thr_self(&tid);
114     return (int)tid;
115 #elif defined(__NetBSD__)
116     return _lwp_self();
117 #elif defined(__OpenBSD__)
118     return getthrid();
119 #else
120     return getpid();
121 #endif
122 }
123 
124 int qemu_daemon(int nochdir, int noclose)
125 {
126     return daemon(nochdir, noclose);
127 }
128 
129 bool qemu_write_pidfile(const char *path, Error **errp)
130 {
131     int fd;
132     char pidstr[32];
133 
134     while (1) {
135         struct stat a, b;
136         struct flock lock = {
137             .l_type = F_WRLCK,
138             .l_whence = SEEK_SET,
139             .l_len = 0,
140         };
141 
142         fd = qemu_open_old(path, O_CREAT | O_WRONLY, S_IRUSR | S_IWUSR);
143         if (fd == -1) {
144             error_setg_errno(errp, errno, "Cannot open pid file");
145             return false;
146         }
147 
148         if (fstat(fd, &b) < 0) {
149             error_setg_errno(errp, errno, "Cannot stat file");
150             goto fail_close;
151         }
152 
153         if (fcntl(fd, F_SETLK, &lock)) {
154             error_setg_errno(errp, errno, "Cannot lock pid file");
155             goto fail_close;
156         }
157 
158         /*
159          * Now make sure the path we locked is the same one that now
160          * exists on the filesystem.
161          */
162         if (stat(path, &a) < 0) {
163             /*
164              * PID file disappeared, someone else must be racing with
165              * us, so try again.
166              */
167             close(fd);
168             continue;
169         }
170 
171         if (a.st_ino == b.st_ino) {
172             break;
173         }
174 
175         /*
176          * PID file was recreated, someone else must be racing with
177          * us, so try again.
178          */
179         close(fd);
180     }
181 
182     if (ftruncate(fd, 0) < 0) {
183         error_setg_errno(errp, errno, "Failed to truncate pid file");
184         goto fail_unlink;
185     }
186 
187     snprintf(pidstr, sizeof(pidstr), FMT_pid "\n", getpid());
188     if (write(fd, pidstr, strlen(pidstr)) != strlen(pidstr)) {
189         error_setg(errp, "Failed to write pid file");
190         goto fail_unlink;
191     }
192 
193     return true;
194 
195 fail_unlink:
196     unlink(path);
197 fail_close:
198     close(fd);
199     return false;
200 }
201 
202 void *qemu_oom_check(void *ptr)
203 {
204     if (ptr == NULL) {
205         fprintf(stderr, "Failed to allocate memory: %s\n", strerror(errno));
206         abort();
207     }
208     return ptr;
209 }
210 
211 void *qemu_try_memalign(size_t alignment, size_t size)
212 {
213     void *ptr;
214 
215     if (alignment < sizeof(void*)) {
216         alignment = sizeof(void*);
217     } else {
218         g_assert(is_power_of_2(alignment));
219     }
220 
221 #if defined(CONFIG_POSIX_MEMALIGN)
222     int ret;
223     ret = posix_memalign(&ptr, alignment, size);
224     if (ret != 0) {
225         errno = ret;
226         ptr = NULL;
227     }
228 #elif defined(CONFIG_BSD)
229     ptr = valloc(size);
230 #else
231     ptr = memalign(alignment, size);
232 #endif
233     trace_qemu_memalign(alignment, size, ptr);
234     return ptr;
235 }
236 
237 void *qemu_memalign(size_t alignment, size_t size)
238 {
239     return qemu_oom_check(qemu_try_memalign(alignment, size));
240 }
241 
242 /* alloc shared memory pages */
243 void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment, bool shared,
244                           bool noreserve)
245 {
246     const uint32_t qemu_map_flags = (shared ? QEMU_MAP_SHARED : 0) |
247                                     (noreserve ? QEMU_MAP_NORESERVE : 0);
248     size_t align = QEMU_VMALLOC_ALIGN;
249     void *ptr = qemu_ram_mmap(-1, size, align, qemu_map_flags, 0);
250 
251     if (ptr == MAP_FAILED) {
252         return NULL;
253     }
254 
255     if (alignment) {
256         *alignment = align;
257     }
258 
259     trace_qemu_anon_ram_alloc(size, ptr);
260     return ptr;
261 }
262 
263 void qemu_vfree(void *ptr)
264 {
265     trace_qemu_vfree(ptr);
266     free(ptr);
267 }
268 
269 void qemu_anon_ram_free(void *ptr, size_t size)
270 {
271     trace_qemu_anon_ram_free(ptr, size);
272     qemu_ram_munmap(-1, ptr, size);
273 }
274 
275 void qemu_set_block(int fd)
276 {
277     int f;
278     f = fcntl(fd, F_GETFL);
279     assert(f != -1);
280     f = fcntl(fd, F_SETFL, f & ~O_NONBLOCK);
281     assert(f != -1);
282 }
283 
284 int qemu_try_set_nonblock(int fd)
285 {
286     int f;
287     f = fcntl(fd, F_GETFL);
288     if (f == -1) {
289         return -errno;
290     }
291     if (fcntl(fd, F_SETFL, f | O_NONBLOCK) == -1) {
292         return -errno;
293     }
294     return 0;
295 }
296 
297 void qemu_set_nonblock(int fd)
298 {
299     int f;
300     f = qemu_try_set_nonblock(fd);
301     assert(f == 0);
302 }
303 
304 int socket_set_fast_reuse(int fd)
305 {
306     int val = 1, ret;
307 
308     ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
309                      (const char *)&val, sizeof(val));
310 
311     assert(ret == 0);
312 
313     return ret;
314 }
315 
316 void qemu_set_cloexec(int fd)
317 {
318     int f;
319     f = fcntl(fd, F_GETFD);
320     assert(f != -1);
321     f = fcntl(fd, F_SETFD, f | FD_CLOEXEC);
322     assert(f != -1);
323 }
324 
325 /*
326  * Creates a pipe with FD_CLOEXEC set on both file descriptors
327  */
328 int qemu_pipe(int pipefd[2])
329 {
330     int ret;
331 
332 #ifdef CONFIG_PIPE2
333     ret = pipe2(pipefd, O_CLOEXEC);
334     if (ret != -1 || errno != ENOSYS) {
335         return ret;
336     }
337 #endif
338     ret = pipe(pipefd);
339     if (ret == 0) {
340         qemu_set_cloexec(pipefd[0]);
341         qemu_set_cloexec(pipefd[1]);
342     }
343 
344     return ret;
345 }
346 
347 char *
348 qemu_get_local_state_pathname(const char *relative_pathname)
349 {
350     g_autofree char *dir = g_strdup_printf("%s/%s",
351                                            CONFIG_QEMU_LOCALSTATEDIR,
352                                            relative_pathname);
353     return get_relocated_path(dir);
354 }
355 
356 void qemu_set_tty_echo(int fd, bool echo)
357 {
358     struct termios tty;
359 
360     tcgetattr(fd, &tty);
361 
362     if (echo) {
363         tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN;
364     } else {
365         tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN);
366     }
367 
368     tcsetattr(fd, TCSANOW, &tty);
369 }
370 
371 static const char *exec_dir;
372 
373 void qemu_init_exec_dir(const char *argv0)
374 {
375     char *p = NULL;
376     char buf[PATH_MAX];
377 
378     if (exec_dir) {
379         return;
380     }
381 
382 #if defined(__linux__)
383     {
384         int len;
385         len = readlink("/proc/self/exe", buf, sizeof(buf) - 1);
386         if (len > 0) {
387             buf[len] = 0;
388             p = buf;
389         }
390     }
391 #elif defined(__FreeBSD__) \
392       || (defined(__NetBSD__) && defined(KERN_PROC_PATHNAME))
393     {
394 #if defined(__FreeBSD__)
395         static int mib[4] = {CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1};
396 #else
397         static int mib[4] = {CTL_KERN, KERN_PROC_ARGS, -1, KERN_PROC_PATHNAME};
398 #endif
399         size_t len = sizeof(buf) - 1;
400 
401         *buf = '\0';
402         if (!sysctl(mib, ARRAY_SIZE(mib), buf, &len, NULL, 0) &&
403             *buf) {
404             buf[sizeof(buf) - 1] = '\0';
405             p = buf;
406         }
407     }
408 #elif defined(__APPLE__)
409     {
410         char fpath[PATH_MAX];
411         uint32_t len = sizeof(fpath);
412         if (_NSGetExecutablePath(fpath, &len) == 0) {
413             p = realpath(fpath, buf);
414             if (!p) {
415                 return;
416             }
417         }
418     }
419 #elif defined(__HAIKU__)
420     {
421         image_info ii;
422         int32_t c = 0;
423 
424         *buf = '\0';
425         while (get_next_image_info(0, &c, &ii) == B_OK) {
426             if (ii.type == B_APP_IMAGE) {
427                 strncpy(buf, ii.name, sizeof(buf));
428                 buf[sizeof(buf) - 1] = 0;
429                 p = buf;
430                 break;
431             }
432         }
433     }
434 #endif
435     /* If we don't have any way of figuring out the actual executable
436        location then try argv[0].  */
437     if (!p && argv0) {
438         p = realpath(argv0, buf);
439     }
440     if (p) {
441         exec_dir = g_path_get_dirname(p);
442     } else {
443         exec_dir = CONFIG_BINDIR;
444     }
445 }
446 
447 const char *qemu_get_exec_dir(void)
448 {
449     return exec_dir;
450 }
451 
452 #ifdef CONFIG_LINUX
453 static void sigbus_handler(int signal, siginfo_t *siginfo, void *ctx)
454 #else /* CONFIG_LINUX */
455 static void sigbus_handler(int signal)
456 #endif /* CONFIG_LINUX */
457 {
458     int i;
459 
460     if (sigbus_memset_context) {
461         for (i = 0; i < sigbus_memset_context->num_threads; i++) {
462             MemsetThread *thread = &sigbus_memset_context->threads[i];
463 
464             if (qemu_thread_is_self(&thread->pgthread)) {
465                 siglongjmp(thread->env, 1);
466             }
467         }
468     }
469 
470 #ifdef CONFIG_LINUX
471     /*
472      * We assume that the MCE SIGBUS handler could have been registered. We
473      * should never receive BUS_MCEERR_AO on any of our threads, but only on
474      * the main thread registered for PR_MCE_KILL_EARLY. Further, we should not
475      * receive BUS_MCEERR_AR triggered by action of other threads on one of
476      * our threads. So, no need to check for unrelated SIGBUS when seeing one
477      * for our threads.
478      *
479      * We will forward to the MCE handler, which will either handle the SIGBUS
480      * or reinstall the default SIGBUS handler and reraise the SIGBUS. The
481      * default SIGBUS handler will crash the process, so we don't care.
482      */
483     if (sigbus_oldact.sa_flags & SA_SIGINFO) {
484         sigbus_oldact.sa_sigaction(signal, siginfo, ctx);
485         return;
486     }
487 #endif /* CONFIG_LINUX */
488     warn_report("os_mem_prealloc: unrelated SIGBUS detected and ignored");
489 }
490 
491 static void *do_touch_pages(void *arg)
492 {
493     MemsetThread *memset_args = (MemsetThread *)arg;
494     sigset_t set, oldset;
495     int ret = 0;
496 
497     /*
498      * On Linux, the page faults from the loop below can cause mmap_sem
499      * contention with allocation of the thread stacks.  Do not start
500      * clearing until all threads have been created.
501      */
502     qemu_mutex_lock(&page_mutex);
503     while (!memset_args->context->all_threads_created) {
504         qemu_cond_wait(&page_cond, &page_mutex);
505     }
506     qemu_mutex_unlock(&page_mutex);
507 
508     /* unblock SIGBUS */
509     sigemptyset(&set);
510     sigaddset(&set, SIGBUS);
511     pthread_sigmask(SIG_UNBLOCK, &set, &oldset);
512 
513     if (sigsetjmp(memset_args->env, 1)) {
514         ret = -EFAULT;
515     } else {
516         char *addr = memset_args->addr;
517         size_t numpages = memset_args->numpages;
518         size_t hpagesize = memset_args->hpagesize;
519         size_t i;
520         for (i = 0; i < numpages; i++) {
521             /*
522              * Read & write back the same value, so we don't
523              * corrupt existing user/app data that might be
524              * stored.
525              *
526              * 'volatile' to stop compiler optimizing this away
527              * to a no-op
528              */
529             *(volatile char *)addr = *addr;
530             addr += hpagesize;
531         }
532     }
533     pthread_sigmask(SIG_SETMASK, &oldset, NULL);
534     return (void *)(uintptr_t)ret;
535 }
536 
537 static void *do_madv_populate_write_pages(void *arg)
538 {
539     MemsetThread *memset_args = (MemsetThread *)arg;
540     const size_t size = memset_args->numpages * memset_args->hpagesize;
541     char * const addr = memset_args->addr;
542     int ret = 0;
543 
544     /* See do_touch_pages(). */
545     qemu_mutex_lock(&page_mutex);
546     while (!memset_args->context->all_threads_created) {
547         qemu_cond_wait(&page_cond, &page_mutex);
548     }
549     qemu_mutex_unlock(&page_mutex);
550 
551     if (size && qemu_madvise(addr, size, QEMU_MADV_POPULATE_WRITE)) {
552         ret = -errno;
553     }
554     return (void *)(uintptr_t)ret;
555 }
556 
557 static inline int get_memset_num_threads(size_t hpagesize, size_t numpages,
558                                          int smp_cpus)
559 {
560     long host_procs = sysconf(_SC_NPROCESSORS_ONLN);
561     int ret = 1;
562 
563     if (host_procs > 0) {
564         ret = MIN(MIN(host_procs, MAX_MEM_PREALLOC_THREAD_COUNT), smp_cpus);
565     }
566 
567     /* Especially with gigantic pages, don't create more threads than pages. */
568     ret = MIN(ret, numpages);
569     /* Don't start threads to prealloc comparatively little memory. */
570     ret = MIN(ret, MAX(1, hpagesize * numpages / (64 * MiB)));
571 
572     /* In case sysconf() fails, we fall back to single threaded */
573     return ret;
574 }
575 
576 static int touch_all_pages(char *area, size_t hpagesize, size_t numpages,
577                            int smp_cpus, bool use_madv_populate_write)
578 {
579     static gsize initialized = 0;
580     MemsetContext context = {
581         .num_threads = get_memset_num_threads(hpagesize, numpages, smp_cpus),
582     };
583     size_t numpages_per_thread, leftover;
584     void *(*touch_fn)(void *);
585     int ret = 0, i = 0;
586     char *addr = area;
587 
588     if (g_once_init_enter(&initialized)) {
589         qemu_mutex_init(&page_mutex);
590         qemu_cond_init(&page_cond);
591         g_once_init_leave(&initialized, 1);
592     }
593 
594     if (use_madv_populate_write) {
595         /* Avoid creating a single thread for MADV_POPULATE_WRITE */
596         if (context.num_threads == 1) {
597             if (qemu_madvise(area, hpagesize * numpages,
598                              QEMU_MADV_POPULATE_WRITE)) {
599                 return -errno;
600             }
601             return 0;
602         }
603         touch_fn = do_madv_populate_write_pages;
604     } else {
605         touch_fn = do_touch_pages;
606     }
607 
608     context.threads = g_new0(MemsetThread, context.num_threads);
609     numpages_per_thread = numpages / context.num_threads;
610     leftover = numpages % context.num_threads;
611     for (i = 0; i < context.num_threads; i++) {
612         context.threads[i].addr = addr;
613         context.threads[i].numpages = numpages_per_thread + (i < leftover);
614         context.threads[i].hpagesize = hpagesize;
615         context.threads[i].context = &context;
616         qemu_thread_create(&context.threads[i].pgthread, "touch_pages",
617                            touch_fn, &context.threads[i],
618                            QEMU_THREAD_JOINABLE);
619         addr += context.threads[i].numpages * hpagesize;
620     }
621 
622     if (!use_madv_populate_write) {
623         sigbus_memset_context = &context;
624     }
625 
626     qemu_mutex_lock(&page_mutex);
627     context.all_threads_created = true;
628     qemu_cond_broadcast(&page_cond);
629     qemu_mutex_unlock(&page_mutex);
630 
631     for (i = 0; i < context.num_threads; i++) {
632         int tmp = (uintptr_t)qemu_thread_join(&context.threads[i].pgthread);
633 
634         if (tmp) {
635             ret = tmp;
636         }
637     }
638 
639     if (!use_madv_populate_write) {
640         sigbus_memset_context = NULL;
641     }
642     g_free(context.threads);
643 
644     return ret;
645 }
646 
647 static bool madv_populate_write_possible(char *area, size_t pagesize)
648 {
649     return !qemu_madvise(area, pagesize, QEMU_MADV_POPULATE_WRITE) ||
650            errno != EINVAL;
651 }
652 
653 void os_mem_prealloc(int fd, char *area, size_t memory, int smp_cpus,
654                      Error **errp)
655 {
656     static gsize initialized;
657     int ret;
658     size_t hpagesize = qemu_fd_getpagesize(fd);
659     size_t numpages = DIV_ROUND_UP(memory, hpagesize);
660     bool use_madv_populate_write;
661     struct sigaction act;
662 
663     /*
664      * Sense on every invocation, as MADV_POPULATE_WRITE cannot be used for
665      * some special mappings, such as mapping /dev/mem.
666      */
667     use_madv_populate_write = madv_populate_write_possible(area, hpagesize);
668 
669     if (!use_madv_populate_write) {
670         if (g_once_init_enter(&initialized)) {
671             qemu_mutex_init(&sigbus_mutex);
672             g_once_init_leave(&initialized, 1);
673         }
674 
675         qemu_mutex_lock(&sigbus_mutex);
676         memset(&act, 0, sizeof(act));
677 #ifdef CONFIG_LINUX
678         act.sa_sigaction = &sigbus_handler;
679         act.sa_flags = SA_SIGINFO;
680 #else /* CONFIG_LINUX */
681         act.sa_handler = &sigbus_handler;
682         act.sa_flags = 0;
683 #endif /* CONFIG_LINUX */
684 
685         ret = sigaction(SIGBUS, &act, &sigbus_oldact);
686         if (ret) {
687             qemu_mutex_unlock(&sigbus_mutex);
688             error_setg_errno(errp, errno,
689                 "os_mem_prealloc: failed to install signal handler");
690             return;
691         }
692     }
693 
694     /* touch pages simultaneously */
695     ret = touch_all_pages(area, hpagesize, numpages, smp_cpus,
696                           use_madv_populate_write);
697     if (ret) {
698         error_setg_errno(errp, -ret,
699                          "os_mem_prealloc: preallocating memory failed");
700     }
701 
702     if (!use_madv_populate_write) {
703         ret = sigaction(SIGBUS, &sigbus_oldact, NULL);
704         if (ret) {
705             /* Terminate QEMU since it can't recover from error */
706             perror("os_mem_prealloc: failed to reinstall signal handler");
707             exit(1);
708         }
709         qemu_mutex_unlock(&sigbus_mutex);
710     }
711 }
712 
713 char *qemu_get_pid_name(pid_t pid)
714 {
715     char *name = NULL;
716 
717 #if defined(__FreeBSD__)
718     /* BSDs don't have /proc, but they provide a nice substitute */
719     struct kinfo_proc *proc = kinfo_getproc(pid);
720 
721     if (proc) {
722         name = g_strdup(proc->ki_comm);
723         free(proc);
724     }
725 #else
726     /* Assume a system with reasonable procfs */
727     char *pid_path;
728     size_t len;
729 
730     pid_path = g_strdup_printf("/proc/%d/cmdline", pid);
731     g_file_get_contents(pid_path, &name, &len, NULL);
732     g_free(pid_path);
733 #endif
734 
735     return name;
736 }
737 
738 
739 pid_t qemu_fork(Error **errp)
740 {
741     sigset_t oldmask, newmask;
742     struct sigaction sig_action;
743     int saved_errno;
744     pid_t pid;
745 
746     /*
747      * Need to block signals now, so that child process can safely
748      * kill off caller's signal handlers without a race.
749      */
750     sigfillset(&newmask);
751     if (pthread_sigmask(SIG_SETMASK, &newmask, &oldmask) != 0) {
752         error_setg_errno(errp, errno,
753                          "cannot block signals");
754         return -1;
755     }
756 
757     pid = fork();
758     saved_errno = errno;
759 
760     if (pid < 0) {
761         /* attempt to restore signal mask, but ignore failure, to
762          * avoid obscuring the fork failure */
763         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
764         error_setg_errno(errp, saved_errno,
765                          "cannot fork child process");
766         errno = saved_errno;
767         return -1;
768     } else if (pid) {
769         /* parent process */
770 
771         /* Restore our original signal mask now that the child is
772          * safely running. Only documented failures are EFAULT (not
773          * possible, since we are using just-grabbed mask) or EINVAL
774          * (not possible, since we are using correct arguments).  */
775         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
776     } else {
777         /* child process */
778         size_t i;
779 
780         /* Clear out all signal handlers from parent so nothing
781          * unexpected can happen in our child once we unblock
782          * signals */
783         sig_action.sa_handler = SIG_DFL;
784         sig_action.sa_flags = 0;
785         sigemptyset(&sig_action.sa_mask);
786 
787         for (i = 1; i < NSIG; i++) {
788             /* Only possible errors are EFAULT or EINVAL The former
789              * won't happen, the latter we expect, so no need to check
790              * return value */
791             (void)sigaction(i, &sig_action, NULL);
792         }
793 
794         /* Unmask all signals in child, since we've no idea what the
795          * caller's done with their signal mask and don't want to
796          * propagate that to children */
797         sigemptyset(&newmask);
798         if (pthread_sigmask(SIG_SETMASK, &newmask, NULL) != 0) {
799             Error *local_err = NULL;
800             error_setg_errno(&local_err, errno,
801                              "cannot unblock signals");
802             error_report_err(local_err);
803             _exit(1);
804         }
805     }
806     return pid;
807 }
808 
809 void *qemu_alloc_stack(size_t *sz)
810 {
811     void *ptr, *guardpage;
812     int flags;
813 #ifdef CONFIG_DEBUG_STACK_USAGE
814     void *ptr2;
815 #endif
816     size_t pagesz = qemu_real_host_page_size;
817 #ifdef _SC_THREAD_STACK_MIN
818     /* avoid stacks smaller than _SC_THREAD_STACK_MIN */
819     long min_stack_sz = sysconf(_SC_THREAD_STACK_MIN);
820     *sz = MAX(MAX(min_stack_sz, 0), *sz);
821 #endif
822     /* adjust stack size to a multiple of the page size */
823     *sz = ROUND_UP(*sz, pagesz);
824     /* allocate one extra page for the guard page */
825     *sz += pagesz;
826 
827     flags = MAP_PRIVATE | MAP_ANONYMOUS;
828 #if defined(MAP_STACK) && defined(__OpenBSD__)
829     /* Only enable MAP_STACK on OpenBSD. Other OS's such as
830      * Linux/FreeBSD/NetBSD have a flag with the same name
831      * but have differing functionality. OpenBSD will SEGV
832      * if it spots execution with a stack pointer pointing
833      * at memory that was not allocated with MAP_STACK.
834      */
835     flags |= MAP_STACK;
836 #endif
837 
838     ptr = mmap(NULL, *sz, PROT_READ | PROT_WRITE, flags, -1, 0);
839     if (ptr == MAP_FAILED) {
840         perror("failed to allocate memory for stack");
841         abort();
842     }
843 
844 #if defined(HOST_IA64)
845     /* separate register stack */
846     guardpage = ptr + (((*sz - pagesz) / 2) & ~pagesz);
847 #elif defined(HOST_HPPA)
848     /* stack grows up */
849     guardpage = ptr + *sz - pagesz;
850 #else
851     /* stack grows down */
852     guardpage = ptr;
853 #endif
854     if (mprotect(guardpage, pagesz, PROT_NONE) != 0) {
855         perror("failed to set up stack guard page");
856         abort();
857     }
858 
859 #ifdef CONFIG_DEBUG_STACK_USAGE
860     for (ptr2 = ptr + pagesz; ptr2 < ptr + *sz; ptr2 += sizeof(uint32_t)) {
861         *(uint32_t *)ptr2 = 0xdeadbeaf;
862     }
863 #endif
864 
865     return ptr;
866 }
867 
868 #ifdef CONFIG_DEBUG_STACK_USAGE
869 static __thread unsigned int max_stack_usage;
870 #endif
871 
872 void qemu_free_stack(void *stack, size_t sz)
873 {
874 #ifdef CONFIG_DEBUG_STACK_USAGE
875     unsigned int usage;
876     void *ptr;
877 
878     for (ptr = stack + qemu_real_host_page_size; ptr < stack + sz;
879          ptr += sizeof(uint32_t)) {
880         if (*(uint32_t *)ptr != 0xdeadbeaf) {
881             break;
882         }
883     }
884     usage = sz - (uintptr_t) (ptr - stack);
885     if (usage > max_stack_usage) {
886         error_report("thread %d max stack usage increased from %u to %u",
887                      qemu_get_thread_id(), max_stack_usage, usage);
888         max_stack_usage = usage;
889     }
890 #endif
891 
892     munmap(stack, sz);
893 }
894 
895 /*
896  * Disable CFI checks.
897  * We are going to call a signal hander directly. Such handler may or may not
898  * have been defined in our binary, so there's no guarantee that the pointer
899  * used to set the handler is a cfi-valid pointer. Since the handlers are
900  * stored in kernel memory, changing the handler to an attacker-defined
901  * function requires being able to call a sigaction() syscall,
902  * which is not as easy as overwriting a pointer in memory.
903  */
904 QEMU_DISABLE_CFI
905 void sigaction_invoke(struct sigaction *action,
906                       struct qemu_signalfd_siginfo *info)
907 {
908     siginfo_t si = {};
909     si.si_signo = info->ssi_signo;
910     si.si_errno = info->ssi_errno;
911     si.si_code = info->ssi_code;
912 
913     /* Convert the minimal set of fields defined by POSIX.
914      * Positive si_code values are reserved for kernel-generated
915      * signals, where the valid siginfo fields are determined by
916      * the signal number.  But according to POSIX, it is unspecified
917      * whether SI_USER and SI_QUEUE have values less than or equal to
918      * zero.
919      */
920     if (info->ssi_code == SI_USER || info->ssi_code == SI_QUEUE ||
921         info->ssi_code <= 0) {
922         /* SIGTERM, etc.  */
923         si.si_pid = info->ssi_pid;
924         si.si_uid = info->ssi_uid;
925     } else if (info->ssi_signo == SIGILL || info->ssi_signo == SIGFPE ||
926                info->ssi_signo == SIGSEGV || info->ssi_signo == SIGBUS) {
927         si.si_addr = (void *)(uintptr_t)info->ssi_addr;
928     } else if (info->ssi_signo == SIGCHLD) {
929         si.si_pid = info->ssi_pid;
930         si.si_status = info->ssi_status;
931         si.si_uid = info->ssi_uid;
932     }
933     action->sa_sigaction(info->ssi_signo, &si, NULL);
934 }
935 
936 #ifndef HOST_NAME_MAX
937 # ifdef _POSIX_HOST_NAME_MAX
938 #  define HOST_NAME_MAX _POSIX_HOST_NAME_MAX
939 # else
940 #  define HOST_NAME_MAX 255
941 # endif
942 #endif
943 
944 char *qemu_get_host_name(Error **errp)
945 {
946     long len = -1;
947     g_autofree char *hostname = NULL;
948 
949 #ifdef _SC_HOST_NAME_MAX
950     len = sysconf(_SC_HOST_NAME_MAX);
951 #endif /* _SC_HOST_NAME_MAX */
952 
953     if (len < 0) {
954         len = HOST_NAME_MAX;
955     }
956 
957     /* Unfortunately, gethostname() below does not guarantee a
958      * NULL terminated string. Therefore, allocate one byte more
959      * to be sure. */
960     hostname = g_new0(char, len + 1);
961 
962     if (gethostname(hostname, len) < 0) {
963         error_setg_errno(errp, errno,
964                          "cannot get hostname");
965         return NULL;
966     }
967 
968     return g_steal_pointer(&hostname);
969 }
970 
971 size_t qemu_get_host_physmem(void)
972 {
973 #ifdef _SC_PHYS_PAGES
974     long pages = sysconf(_SC_PHYS_PAGES);
975     if (pages > 0) {
976         if (pages > SIZE_MAX / qemu_real_host_page_size) {
977             return SIZE_MAX;
978         } else {
979             return pages * qemu_real_host_page_size;
980         }
981     }
982 #endif
983     return 0;
984 }
985