xref: /openbmc/qemu/util/oslib-posix.c (revision 795c40b8)
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 "sysemu/sysemu.h"
35 #include "trace.h"
36 #include "qapi/error.h"
37 #include "qemu/sockets.h"
38 #include <libgen.h>
39 #include <sys/signal.h>
40 #include "qemu/cutils.h"
41 
42 #ifdef CONFIG_LINUX
43 #include <sys/syscall.h>
44 #endif
45 
46 #ifdef __FreeBSD__
47 #include <sys/sysctl.h>
48 #include <sys/user.h>
49 #include <libutil.h>
50 #endif
51 
52 #include "qemu/mmap-alloc.h"
53 
54 #ifdef CONFIG_DEBUG_STACK_USAGE
55 #include "qemu/error-report.h"
56 #endif
57 
58 #define MAX_MEM_PREALLOC_THREAD_COUNT 16
59 
60 struct MemsetThread {
61     char *addr;
62     uint64_t numpages;
63     uint64_t hpagesize;
64     QemuThread pgthread;
65     sigjmp_buf env;
66 };
67 typedef struct MemsetThread MemsetThread;
68 
69 static MemsetThread *memset_thread;
70 static int memset_num_threads;
71 static bool memset_thread_failed;
72 
73 int qemu_get_thread_id(void)
74 {
75 #if defined(__linux__)
76     return syscall(SYS_gettid);
77 #else
78     return getpid();
79 #endif
80 }
81 
82 int qemu_daemon(int nochdir, int noclose)
83 {
84     return daemon(nochdir, noclose);
85 }
86 
87 void *qemu_oom_check(void *ptr)
88 {
89     if (ptr == NULL) {
90         fprintf(stderr, "Failed to allocate memory: %s\n", strerror(errno));
91         abort();
92     }
93     return ptr;
94 }
95 
96 void *qemu_try_memalign(size_t alignment, size_t size)
97 {
98     void *ptr;
99 
100     if (alignment < sizeof(void*)) {
101         alignment = sizeof(void*);
102     }
103 
104 #if defined(_POSIX_C_SOURCE) && !defined(__sun__)
105     int ret;
106     ret = posix_memalign(&ptr, alignment, size);
107     if (ret != 0) {
108         errno = ret;
109         ptr = NULL;
110     }
111 #elif defined(CONFIG_BSD)
112     ptr = valloc(size);
113 #else
114     ptr = memalign(alignment, size);
115 #endif
116     trace_qemu_memalign(alignment, size, ptr);
117     return ptr;
118 }
119 
120 void *qemu_memalign(size_t alignment, size_t size)
121 {
122     return qemu_oom_check(qemu_try_memalign(alignment, size));
123 }
124 
125 /* alloc shared memory pages */
126 void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment)
127 {
128     size_t align = QEMU_VMALLOC_ALIGN;
129     void *ptr = qemu_ram_mmap(-1, size, align, false);
130 
131     if (ptr == MAP_FAILED) {
132         return NULL;
133     }
134 
135     if (alignment) {
136         *alignment = align;
137     }
138 
139     trace_qemu_anon_ram_alloc(size, ptr);
140     return ptr;
141 }
142 
143 void qemu_vfree(void *ptr)
144 {
145     trace_qemu_vfree(ptr);
146     free(ptr);
147 }
148 
149 void qemu_anon_ram_free(void *ptr, size_t size)
150 {
151     trace_qemu_anon_ram_free(ptr, size);
152     qemu_ram_munmap(ptr, size);
153 }
154 
155 void qemu_set_block(int fd)
156 {
157     int f;
158     f = fcntl(fd, F_GETFL);
159     fcntl(fd, F_SETFL, f & ~O_NONBLOCK);
160 }
161 
162 void qemu_set_nonblock(int fd)
163 {
164     int f;
165     f = fcntl(fd, F_GETFL);
166     fcntl(fd, F_SETFL, f | O_NONBLOCK);
167 }
168 
169 int socket_set_fast_reuse(int fd)
170 {
171     int val = 1, ret;
172 
173     ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
174                      (const char *)&val, sizeof(val));
175 
176     assert(ret == 0);
177 
178     return ret;
179 }
180 
181 void qemu_set_cloexec(int fd)
182 {
183     int f;
184     f = fcntl(fd, F_GETFD);
185     fcntl(fd, F_SETFD, f | FD_CLOEXEC);
186 }
187 
188 /*
189  * Creates a pipe with FD_CLOEXEC set on both file descriptors
190  */
191 int qemu_pipe(int pipefd[2])
192 {
193     int ret;
194 
195 #ifdef CONFIG_PIPE2
196     ret = pipe2(pipefd, O_CLOEXEC);
197     if (ret != -1 || errno != ENOSYS) {
198         return ret;
199     }
200 #endif
201     ret = pipe(pipefd);
202     if (ret == 0) {
203         qemu_set_cloexec(pipefd[0]);
204         qemu_set_cloexec(pipefd[1]);
205     }
206 
207     return ret;
208 }
209 
210 int qemu_utimens(const char *path, const struct timespec *times)
211 {
212     struct timeval tv[2], tv_now;
213     struct stat st;
214     int i;
215 #ifdef CONFIG_UTIMENSAT
216     int ret;
217 
218     ret = utimensat(AT_FDCWD, path, times, AT_SYMLINK_NOFOLLOW);
219     if (ret != -1 || errno != ENOSYS) {
220         return ret;
221     }
222 #endif
223     /* Fallback: use utimes() instead of utimensat() */
224 
225     /* happy if special cases */
226     if (times[0].tv_nsec == UTIME_OMIT && times[1].tv_nsec == UTIME_OMIT) {
227         return 0;
228     }
229     if (times[0].tv_nsec == UTIME_NOW && times[1].tv_nsec == UTIME_NOW) {
230         return utimes(path, NULL);
231     }
232 
233     /* prepare for hard cases */
234     if (times[0].tv_nsec == UTIME_NOW || times[1].tv_nsec == UTIME_NOW) {
235         gettimeofday(&tv_now, NULL);
236     }
237     if (times[0].tv_nsec == UTIME_OMIT || times[1].tv_nsec == UTIME_OMIT) {
238         stat(path, &st);
239     }
240 
241     for (i = 0; i < 2; i++) {
242         if (times[i].tv_nsec == UTIME_NOW) {
243             tv[i].tv_sec = tv_now.tv_sec;
244             tv[i].tv_usec = tv_now.tv_usec;
245         } else if (times[i].tv_nsec == UTIME_OMIT) {
246             tv[i].tv_sec = (i == 0) ? st.st_atime : st.st_mtime;
247             tv[i].tv_usec = 0;
248         } else {
249             tv[i].tv_sec = times[i].tv_sec;
250             tv[i].tv_usec = times[i].tv_nsec / 1000;
251         }
252     }
253 
254     return utimes(path, &tv[0]);
255 }
256 
257 char *
258 qemu_get_local_state_pathname(const char *relative_pathname)
259 {
260     return g_strdup_printf("%s/%s", CONFIG_QEMU_LOCALSTATEDIR,
261                            relative_pathname);
262 }
263 
264 void qemu_set_tty_echo(int fd, bool echo)
265 {
266     struct termios tty;
267 
268     tcgetattr(fd, &tty);
269 
270     if (echo) {
271         tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN;
272     } else {
273         tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN);
274     }
275 
276     tcsetattr(fd, TCSANOW, &tty);
277 }
278 
279 static char exec_dir[PATH_MAX];
280 
281 void qemu_init_exec_dir(const char *argv0)
282 {
283     char *dir;
284     char *p = NULL;
285     char buf[PATH_MAX];
286 
287     assert(!exec_dir[0]);
288 
289 #if defined(__linux__)
290     {
291         int len;
292         len = readlink("/proc/self/exe", buf, sizeof(buf) - 1);
293         if (len > 0) {
294             buf[len] = 0;
295             p = buf;
296         }
297     }
298 #elif defined(__FreeBSD__)
299     {
300         static int mib[4] = {CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1};
301         size_t len = sizeof(buf) - 1;
302 
303         *buf = '\0';
304         if (!sysctl(mib, ARRAY_SIZE(mib), buf, &len, NULL, 0) &&
305             *buf) {
306             buf[sizeof(buf) - 1] = '\0';
307             p = buf;
308         }
309     }
310 #endif
311     /* If we don't have any way of figuring out the actual executable
312        location then try argv[0].  */
313     if (!p) {
314         if (!argv0) {
315             return;
316         }
317         p = realpath(argv0, buf);
318         if (!p) {
319             return;
320         }
321     }
322     dir = g_path_get_dirname(p);
323 
324     pstrcpy(exec_dir, sizeof(exec_dir), dir);
325 
326     g_free(dir);
327 }
328 
329 char *qemu_get_exec_dir(void)
330 {
331     return g_strdup(exec_dir);
332 }
333 
334 static void sigbus_handler(int signal)
335 {
336     int i;
337     if (memset_thread) {
338         for (i = 0; i < memset_num_threads; i++) {
339             if (qemu_thread_is_self(&memset_thread[i].pgthread)) {
340                 siglongjmp(memset_thread[i].env, 1);
341             }
342         }
343     }
344 }
345 
346 static void *do_touch_pages(void *arg)
347 {
348     MemsetThread *memset_args = (MemsetThread *)arg;
349     char *addr = memset_args->addr;
350     uint64_t numpages = memset_args->numpages;
351     uint64_t hpagesize = memset_args->hpagesize;
352     sigset_t set, oldset;
353     int i = 0;
354 
355     /* unblock SIGBUS */
356     sigemptyset(&set);
357     sigaddset(&set, SIGBUS);
358     pthread_sigmask(SIG_UNBLOCK, &set, &oldset);
359 
360     if (sigsetjmp(memset_args->env, 1)) {
361         memset_thread_failed = true;
362     } else {
363         for (i = 0; i < numpages; i++) {
364             /*
365              * Read & write back the same value, so we don't
366              * corrupt existing user/app data that might be
367              * stored.
368              *
369              * 'volatile' to stop compiler optimizing this away
370              * to a no-op
371              *
372              * TODO: get a better solution from kernel so we
373              * don't need to write at all so we don't cause
374              * wear on the storage backing the region...
375              */
376             *(volatile char *)addr = *addr;
377             addr += hpagesize;
378         }
379     }
380     pthread_sigmask(SIG_SETMASK, &oldset, NULL);
381     return NULL;
382 }
383 
384 static inline int get_memset_num_threads(int smp_cpus)
385 {
386     long host_procs = sysconf(_SC_NPROCESSORS_ONLN);
387     int ret = 1;
388 
389     if (host_procs > 0) {
390         ret = MIN(MIN(host_procs, MAX_MEM_PREALLOC_THREAD_COUNT), smp_cpus);
391     }
392     /* In case sysconf() fails, we fall back to single threaded */
393     return ret;
394 }
395 
396 static bool touch_all_pages(char *area, size_t hpagesize, size_t numpages,
397                             int smp_cpus)
398 {
399     uint64_t numpages_per_thread, size_per_thread;
400     char *addr = area;
401     int i = 0;
402 
403     memset_thread_failed = false;
404     memset_num_threads = get_memset_num_threads(smp_cpus);
405     memset_thread = g_new0(MemsetThread, memset_num_threads);
406     numpages_per_thread = (numpages / memset_num_threads);
407     size_per_thread = (hpagesize * numpages_per_thread);
408     for (i = 0; i < memset_num_threads; i++) {
409         memset_thread[i].addr = addr;
410         memset_thread[i].numpages = (i == (memset_num_threads - 1)) ?
411                                     numpages : numpages_per_thread;
412         memset_thread[i].hpagesize = hpagesize;
413         qemu_thread_create(&memset_thread[i].pgthread, "touch_pages",
414                            do_touch_pages, &memset_thread[i],
415                            QEMU_THREAD_JOINABLE);
416         addr += size_per_thread;
417         numpages -= numpages_per_thread;
418     }
419     for (i = 0; i < memset_num_threads; i++) {
420         qemu_thread_join(&memset_thread[i].pgthread);
421     }
422     g_free(memset_thread);
423     memset_thread = NULL;
424 
425     return memset_thread_failed;
426 }
427 
428 void os_mem_prealloc(int fd, char *area, size_t memory, int smp_cpus,
429                      Error **errp)
430 {
431     int ret;
432     struct sigaction act, oldact;
433     size_t hpagesize = qemu_fd_getpagesize(fd);
434     size_t numpages = DIV_ROUND_UP(memory, hpagesize);
435 
436     memset(&act, 0, sizeof(act));
437     act.sa_handler = &sigbus_handler;
438     act.sa_flags = 0;
439 
440     ret = sigaction(SIGBUS, &act, &oldact);
441     if (ret) {
442         error_setg_errno(errp, errno,
443             "os_mem_prealloc: failed to install signal handler");
444         return;
445     }
446 
447     /* touch pages simultaneously */
448     if (touch_all_pages(area, hpagesize, numpages, smp_cpus)) {
449         error_setg(errp, "os_mem_prealloc: Insufficient free host memory "
450             "pages available to allocate guest RAM\n");
451     }
452 
453     ret = sigaction(SIGBUS, &oldact, NULL);
454     if (ret) {
455         /* Terminate QEMU since it can't recover from error */
456         perror("os_mem_prealloc: failed to reinstall signal handler");
457         exit(1);
458     }
459 }
460 
461 
462 static struct termios oldtty;
463 
464 static void term_exit(void)
465 {
466     tcsetattr(0, TCSANOW, &oldtty);
467 }
468 
469 static void term_init(void)
470 {
471     struct termios tty;
472 
473     tcgetattr(0, &tty);
474     oldtty = tty;
475 
476     tty.c_iflag &= ~(IGNBRK|BRKINT|PARMRK|ISTRIP
477                           |INLCR|IGNCR|ICRNL|IXON);
478     tty.c_oflag |= OPOST;
479     tty.c_lflag &= ~(ECHO|ECHONL|ICANON|IEXTEN);
480     tty.c_cflag &= ~(CSIZE|PARENB);
481     tty.c_cflag |= CS8;
482     tty.c_cc[VMIN] = 1;
483     tty.c_cc[VTIME] = 0;
484 
485     tcsetattr(0, TCSANOW, &tty);
486 
487     atexit(term_exit);
488 }
489 
490 int qemu_read_password(char *buf, int buf_size)
491 {
492     uint8_t ch;
493     int i, ret;
494 
495     printf("password: ");
496     fflush(stdout);
497     term_init();
498     i = 0;
499     for (;;) {
500         ret = read(0, &ch, 1);
501         if (ret == -1) {
502             if (errno == EAGAIN || errno == EINTR) {
503                 continue;
504             } else {
505                 break;
506             }
507         } else if (ret == 0) {
508             ret = -1;
509             break;
510         } else {
511             if (ch == '\r' ||
512                 ch == '\n') {
513                 ret = 0;
514                 break;
515             }
516             if (i < (buf_size - 1)) {
517                 buf[i++] = ch;
518             }
519         }
520     }
521     term_exit();
522     buf[i] = '\0';
523     printf("\n");
524     return ret;
525 }
526 
527 
528 char *qemu_get_pid_name(pid_t pid)
529 {
530     char *name = NULL;
531 
532 #if defined(__FreeBSD__)
533     /* BSDs don't have /proc, but they provide a nice substitute */
534     struct kinfo_proc *proc = kinfo_getproc(pid);
535 
536     if (proc) {
537         name = g_strdup(proc->ki_comm);
538         free(proc);
539     }
540 #else
541     /* Assume a system with reasonable procfs */
542     char *pid_path;
543     size_t len;
544 
545     pid_path = g_strdup_printf("/proc/%d/cmdline", pid);
546     g_file_get_contents(pid_path, &name, &len, NULL);
547     g_free(pid_path);
548 #endif
549 
550     return name;
551 }
552 
553 
554 pid_t qemu_fork(Error **errp)
555 {
556     sigset_t oldmask, newmask;
557     struct sigaction sig_action;
558     int saved_errno;
559     pid_t pid;
560 
561     /*
562      * Need to block signals now, so that child process can safely
563      * kill off caller's signal handlers without a race.
564      */
565     sigfillset(&newmask);
566     if (pthread_sigmask(SIG_SETMASK, &newmask, &oldmask) != 0) {
567         error_setg_errno(errp, errno,
568                          "cannot block signals");
569         return -1;
570     }
571 
572     pid = fork();
573     saved_errno = errno;
574 
575     if (pid < 0) {
576         /* attempt to restore signal mask, but ignore failure, to
577          * avoid obscuring the fork failure */
578         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
579         error_setg_errno(errp, saved_errno,
580                          "cannot fork child process");
581         errno = saved_errno;
582         return -1;
583     } else if (pid) {
584         /* parent process */
585 
586         /* Restore our original signal mask now that the child is
587          * safely running. Only documented failures are EFAULT (not
588          * possible, since we are using just-grabbed mask) or EINVAL
589          * (not possible, since we are using correct arguments).  */
590         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
591     } else {
592         /* child process */
593         size_t i;
594 
595         /* Clear out all signal handlers from parent so nothing
596          * unexpected can happen in our child once we unblock
597          * signals */
598         sig_action.sa_handler = SIG_DFL;
599         sig_action.sa_flags = 0;
600         sigemptyset(&sig_action.sa_mask);
601 
602         for (i = 1; i < NSIG; i++) {
603             /* Only possible errors are EFAULT or EINVAL The former
604              * won't happen, the latter we expect, so no need to check
605              * return value */
606             (void)sigaction(i, &sig_action, NULL);
607         }
608 
609         /* Unmask all signals in child, since we've no idea what the
610          * caller's done with their signal mask and don't want to
611          * propagate that to children */
612         sigemptyset(&newmask);
613         if (pthread_sigmask(SIG_SETMASK, &newmask, NULL) != 0) {
614             Error *local_err = NULL;
615             error_setg_errno(&local_err, errno,
616                              "cannot unblock signals");
617             error_report_err(local_err);
618             _exit(1);
619         }
620     }
621     return pid;
622 }
623 
624 void *qemu_alloc_stack(size_t *sz)
625 {
626     void *ptr, *guardpage;
627 #ifdef CONFIG_DEBUG_STACK_USAGE
628     void *ptr2;
629 #endif
630     size_t pagesz = getpagesize();
631 #ifdef _SC_THREAD_STACK_MIN
632     /* avoid stacks smaller than _SC_THREAD_STACK_MIN */
633     long min_stack_sz = sysconf(_SC_THREAD_STACK_MIN);
634     *sz = MAX(MAX(min_stack_sz, 0), *sz);
635 #endif
636     /* adjust stack size to a multiple of the page size */
637     *sz = ROUND_UP(*sz, pagesz);
638     /* allocate one extra page for the guard page */
639     *sz += pagesz;
640 
641     ptr = mmap(NULL, *sz, PROT_READ | PROT_WRITE,
642                MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
643     if (ptr == MAP_FAILED) {
644         abort();
645     }
646 
647 #if defined(HOST_IA64)
648     /* separate register stack */
649     guardpage = ptr + (((*sz - pagesz) / 2) & ~pagesz);
650 #elif defined(HOST_HPPA)
651     /* stack grows up */
652     guardpage = ptr + *sz - pagesz;
653 #else
654     /* stack grows down */
655     guardpage = ptr;
656 #endif
657     if (mprotect(guardpage, pagesz, PROT_NONE) != 0) {
658         abort();
659     }
660 
661 #ifdef CONFIG_DEBUG_STACK_USAGE
662     for (ptr2 = ptr + pagesz; ptr2 < ptr + *sz; ptr2 += sizeof(uint32_t)) {
663         *(uint32_t *)ptr2 = 0xdeadbeaf;
664     }
665 #endif
666 
667     return ptr;
668 }
669 
670 #ifdef CONFIG_DEBUG_STACK_USAGE
671 static __thread unsigned int max_stack_usage;
672 #endif
673 
674 void qemu_free_stack(void *stack, size_t sz)
675 {
676 #ifdef CONFIG_DEBUG_STACK_USAGE
677     unsigned int usage;
678     void *ptr;
679 
680     for (ptr = stack + getpagesize(); ptr < stack + sz;
681          ptr += sizeof(uint32_t)) {
682         if (*(uint32_t *)ptr != 0xdeadbeaf) {
683             break;
684         }
685     }
686     usage = sz - (uintptr_t) (ptr - stack);
687     if (usage > max_stack_usage) {
688         error_report("thread %d max stack usage increased from %u to %u",
689                      qemu_get_thread_id(), max_stack_usage, usage);
690         max_stack_usage = usage;
691     }
692 #endif
693 
694     munmap(stack, sz);
695 }
696 
697 void sigaction_invoke(struct sigaction *action,
698                       struct qemu_signalfd_siginfo *info)
699 {
700     siginfo_t si = { 0 };
701     si.si_signo = info->ssi_signo;
702     si.si_errno = info->ssi_errno;
703     si.si_code = info->ssi_code;
704 
705     /* Convert the minimal set of fields defined by POSIX.
706      * Positive si_code values are reserved for kernel-generated
707      * signals, where the valid siginfo fields are determined by
708      * the signal number.  But according to POSIX, it is unspecified
709      * whether SI_USER and SI_QUEUE have values less than or equal to
710      * zero.
711      */
712     if (info->ssi_code == SI_USER || info->ssi_code == SI_QUEUE ||
713         info->ssi_code <= 0) {
714         /* SIGTERM, etc.  */
715         si.si_pid = info->ssi_pid;
716         si.si_uid = info->ssi_uid;
717     } else if (info->ssi_signo == SIGILL || info->ssi_signo == SIGFPE ||
718                info->ssi_signo == SIGSEGV || info->ssi_signo == SIGBUS) {
719         si.si_addr = (void *)(uintptr_t)info->ssi_addr;
720     } else if (info->ssi_signo == SIGCHLD) {
721         si.si_pid = info->ssi_pid;
722         si.si_status = info->ssi_status;
723         si.si_uid = info->ssi_uid;
724     }
725     action->sa_sigaction(info->ssi_signo, &si, NULL);
726 }
727