xref: /qemu/util/oslib-posix.c (revision 99dbfd1d) 
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 int qemu_get_thread_id(void)
59 {
60 #if defined(__linux__)
61     return syscall(SYS_gettid);
62 #else
63     return getpid();
64 #endif
65 }
66 
67 int qemu_daemon(int nochdir, int noclose)
68 {
69     return daemon(nochdir, noclose);
70 }
71 
72 void *qemu_oom_check(void *ptr)
73 {
74     if (ptr == NULL) {
75         fprintf(stderr, "Failed to allocate memory: %s\n", strerror(errno));
76         abort();
77     }
78     return ptr;
79 }
80 
81 void *qemu_try_memalign(size_t alignment, size_t size)
82 {
83     void *ptr;
84 
85     if (alignment < sizeof(void*)) {
86         alignment = sizeof(void*);
87     }
88 
89 #if defined(_POSIX_C_SOURCE) && !defined(__sun__)
90     int ret;
91     ret = posix_memalign(&ptr, alignment, size);
92     if (ret != 0) {
93         errno = ret;
94         ptr = NULL;
95     }
96 #elif defined(CONFIG_BSD)
97     ptr = valloc(size);
98 #else
99     ptr = memalign(alignment, size);
100 #endif
101     trace_qemu_memalign(alignment, size, ptr);
102     return ptr;
103 }
104 
105 void *qemu_memalign(size_t alignment, size_t size)
106 {
107     return qemu_oom_check(qemu_try_memalign(alignment, size));
108 }
109 
110 /* alloc shared memory pages */
111 void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment)
112 {
113     size_t align = QEMU_VMALLOC_ALIGN;
114     void *ptr = qemu_ram_mmap(-1, size, align, false);
115 
116     if (ptr == MAP_FAILED) {
117         return NULL;
118     }
119 
120     if (alignment) {
121         *alignment = align;
122     }
123 
124     trace_qemu_anon_ram_alloc(size, ptr);
125     return ptr;
126 }
127 
128 void qemu_vfree(void *ptr)
129 {
130     trace_qemu_vfree(ptr);
131     free(ptr);
132 }
133 
134 void qemu_anon_ram_free(void *ptr, size_t size)
135 {
136     trace_qemu_anon_ram_free(ptr, size);
137     qemu_ram_munmap(ptr, size);
138 }
139 
140 void qemu_set_block(int fd)
141 {
142     int f;
143     f = fcntl(fd, F_GETFL);
144     fcntl(fd, F_SETFL, f & ~O_NONBLOCK);
145 }
146 
147 void qemu_set_nonblock(int fd)
148 {
149     int f;
150     f = fcntl(fd, F_GETFL);
151     fcntl(fd, F_SETFL, f | O_NONBLOCK);
152 }
153 
154 int socket_set_fast_reuse(int fd)
155 {
156     int val = 1, ret;
157 
158     ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
159                      (const char *)&val, sizeof(val));
160 
161     assert(ret == 0);
162 
163     return ret;
164 }
165 
166 void qemu_set_cloexec(int fd)
167 {
168     int f;
169     f = fcntl(fd, F_GETFD);
170     fcntl(fd, F_SETFD, f | FD_CLOEXEC);
171 }
172 
173 /*
174  * Creates a pipe with FD_CLOEXEC set on both file descriptors
175  */
176 int qemu_pipe(int pipefd[2])
177 {
178     int ret;
179 
180 #ifdef CONFIG_PIPE2
181     ret = pipe2(pipefd, O_CLOEXEC);
182     if (ret != -1 || errno != ENOSYS) {
183         return ret;
184     }
185 #endif
186     ret = pipe(pipefd);
187     if (ret == 0) {
188         qemu_set_cloexec(pipefd[0]);
189         qemu_set_cloexec(pipefd[1]);
190     }
191 
192     return ret;
193 }
194 
195 int qemu_utimens(const char *path, const struct timespec *times)
196 {
197     struct timeval tv[2], tv_now;
198     struct stat st;
199     int i;
200 #ifdef CONFIG_UTIMENSAT
201     int ret;
202 
203     ret = utimensat(AT_FDCWD, path, times, AT_SYMLINK_NOFOLLOW);
204     if (ret != -1 || errno != ENOSYS) {
205         return ret;
206     }
207 #endif
208     /* Fallback: use utimes() instead of utimensat() */
209 
210     /* happy if special cases */
211     if (times[0].tv_nsec == UTIME_OMIT && times[1].tv_nsec == UTIME_OMIT) {
212         return 0;
213     }
214     if (times[0].tv_nsec == UTIME_NOW && times[1].tv_nsec == UTIME_NOW) {
215         return utimes(path, NULL);
216     }
217 
218     /* prepare for hard cases */
219     if (times[0].tv_nsec == UTIME_NOW || times[1].tv_nsec == UTIME_NOW) {
220         gettimeofday(&tv_now, NULL);
221     }
222     if (times[0].tv_nsec == UTIME_OMIT || times[1].tv_nsec == UTIME_OMIT) {
223         stat(path, &st);
224     }
225 
226     for (i = 0; i < 2; i++) {
227         if (times[i].tv_nsec == UTIME_NOW) {
228             tv[i].tv_sec = tv_now.tv_sec;
229             tv[i].tv_usec = tv_now.tv_usec;
230         } else if (times[i].tv_nsec == UTIME_OMIT) {
231             tv[i].tv_sec = (i == 0) ? st.st_atime : st.st_mtime;
232             tv[i].tv_usec = 0;
233         } else {
234             tv[i].tv_sec = times[i].tv_sec;
235             tv[i].tv_usec = times[i].tv_nsec / 1000;
236         }
237     }
238 
239     return utimes(path, &tv[0]);
240 }
241 
242 char *
243 qemu_get_local_state_pathname(const char *relative_pathname)
244 {
245     return g_strdup_printf("%s/%s", CONFIG_QEMU_LOCALSTATEDIR,
246                            relative_pathname);
247 }
248 
249 void qemu_set_tty_echo(int fd, bool echo)
250 {
251     struct termios tty;
252 
253     tcgetattr(fd, &tty);
254 
255     if (echo) {
256         tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN;
257     } else {
258         tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN);
259     }
260 
261     tcsetattr(fd, TCSANOW, &tty);
262 }
263 
264 static char exec_dir[PATH_MAX];
265 
266 void qemu_init_exec_dir(const char *argv0)
267 {
268     char *dir;
269     char *p = NULL;
270     char buf[PATH_MAX];
271 
272     assert(!exec_dir[0]);
273 
274 #if defined(__linux__)
275     {
276         int len;
277         len = readlink("/proc/self/exe", buf, sizeof(buf) - 1);
278         if (len > 0) {
279             buf[len] = 0;
280             p = buf;
281         }
282     }
283 #elif defined(__FreeBSD__)
284     {
285         static int mib[4] = {CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1};
286         size_t len = sizeof(buf) - 1;
287 
288         *buf = '\0';
289         if (!sysctl(mib, ARRAY_SIZE(mib), buf, &len, NULL, 0) &&
290             *buf) {
291             buf[sizeof(buf) - 1] = '\0';
292             p = buf;
293         }
294     }
295 #endif
296     /* If we don't have any way of figuring out the actual executable
297        location then try argv[0].  */
298     if (!p) {
299         if (!argv0) {
300             return;
301         }
302         p = realpath(argv0, buf);
303         if (!p) {
304             return;
305         }
306     }
307     dir = g_path_get_dirname(p);
308 
309     pstrcpy(exec_dir, sizeof(exec_dir), dir);
310 
311     g_free(dir);
312 }
313 
314 char *qemu_get_exec_dir(void)
315 {
316     return g_strdup(exec_dir);
317 }
318 
319 static sigjmp_buf sigjump;
320 
321 static void sigbus_handler(int signal)
322 {
323     siglongjmp(sigjump, 1);
324 }
325 
326 void os_mem_prealloc(int fd, char *area, size_t memory, Error **errp)
327 {
328     int ret;
329     struct sigaction act, oldact;
330     sigset_t set, oldset;
331 
332     memset(&act, 0, sizeof(act));
333     act.sa_handler = &sigbus_handler;
334     act.sa_flags = 0;
335 
336     ret = sigaction(SIGBUS, &act, &oldact);
337     if (ret) {
338         error_setg_errno(errp, errno,
339             "os_mem_prealloc: failed to install signal handler");
340         return;
341     }
342 
343     /* unblock SIGBUS */
344     sigemptyset(&set);
345     sigaddset(&set, SIGBUS);
346     pthread_sigmask(SIG_UNBLOCK, &set, &oldset);
347 
348     if (sigsetjmp(sigjump, 1)) {
349         error_setg(errp, "os_mem_prealloc: Insufficient free host memory "
350             "pages available to allocate guest RAM\n");
351     } else {
352         int i;
353         size_t hpagesize = qemu_fd_getpagesize(fd);
354         size_t numpages = DIV_ROUND_UP(memory, hpagesize);
355 
356         /* MAP_POPULATE silently ignores failures */
357         for (i = 0; i < numpages; i++) {
358             memset(area + (hpagesize * i), 0, 1);
359         }
360     }
361 
362     ret = sigaction(SIGBUS, &oldact, NULL);
363     if (ret) {
364         /* Terminate QEMU since it can't recover from error */
365         perror("os_mem_prealloc: failed to reinstall signal handler");
366         exit(1);
367     }
368     pthread_sigmask(SIG_SETMASK, &oldset, NULL);
369 }
370 
371 
372 static struct termios oldtty;
373 
374 static void term_exit(void)
375 {
376     tcsetattr(0, TCSANOW, &oldtty);
377 }
378 
379 static void term_init(void)
380 {
381     struct termios tty;
382 
383     tcgetattr(0, &tty);
384     oldtty = tty;
385 
386     tty.c_iflag &= ~(IGNBRK|BRKINT|PARMRK|ISTRIP
387                           |INLCR|IGNCR|ICRNL|IXON);
388     tty.c_oflag |= OPOST;
389     tty.c_lflag &= ~(ECHO|ECHONL|ICANON|IEXTEN);
390     tty.c_cflag &= ~(CSIZE|PARENB);
391     tty.c_cflag |= CS8;
392     tty.c_cc[VMIN] = 1;
393     tty.c_cc[VTIME] = 0;
394 
395     tcsetattr(0, TCSANOW, &tty);
396 
397     atexit(term_exit);
398 }
399 
400 int qemu_read_password(char *buf, int buf_size)
401 {
402     uint8_t ch;
403     int i, ret;
404 
405     printf("password: ");
406     fflush(stdout);
407     term_init();
408     i = 0;
409     for (;;) {
410         ret = read(0, &ch, 1);
411         if (ret == -1) {
412             if (errno == EAGAIN || errno == EINTR) {
413                 continue;
414             } else {
415                 break;
416             }
417         } else if (ret == 0) {
418             ret = -1;
419             break;
420         } else {
421             if (ch == '\r' ||
422                 ch == '\n') {
423                 ret = 0;
424                 break;
425             }
426             if (i < (buf_size - 1)) {
427                 buf[i++] = ch;
428             }
429         }
430     }
431     term_exit();
432     buf[i] = '\0';
433     printf("\n");
434     return ret;
435 }
436 
437 
438 char *qemu_get_pid_name(pid_t pid)
439 {
440     char *name = NULL;
441 
442 #if defined(__FreeBSD__)
443     /* BSDs don't have /proc, but they provide a nice substitute */
444     struct kinfo_proc *proc = kinfo_getproc(pid);
445 
446     if (proc) {
447         name = g_strdup(proc->ki_comm);
448         free(proc);
449     }
450 #else
451     /* Assume a system with reasonable procfs */
452     char *pid_path;
453     size_t len;
454 
455     pid_path = g_strdup_printf("/proc/%d/cmdline", pid);
456     g_file_get_contents(pid_path, &name, &len, NULL);
457     g_free(pid_path);
458 #endif
459 
460     return name;
461 }
462 
463 
464 pid_t qemu_fork(Error **errp)
465 {
466     sigset_t oldmask, newmask;
467     struct sigaction sig_action;
468     int saved_errno;
469     pid_t pid;
470 
471     /*
472      * Need to block signals now, so that child process can safely
473      * kill off caller's signal handlers without a race.
474      */
475     sigfillset(&newmask);
476     if (pthread_sigmask(SIG_SETMASK, &newmask, &oldmask) != 0) {
477         error_setg_errno(errp, errno,
478                          "cannot block signals");
479         return -1;
480     }
481 
482     pid = fork();
483     saved_errno = errno;
484 
485     if (pid < 0) {
486         /* attempt to restore signal mask, but ignore failure, to
487          * avoid obscuring the fork failure */
488         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
489         error_setg_errno(errp, saved_errno,
490                          "cannot fork child process");
491         errno = saved_errno;
492         return -1;
493     } else if (pid) {
494         /* parent process */
495 
496         /* Restore our original signal mask now that the child is
497          * safely running. Only documented failures are EFAULT (not
498          * possible, since we are using just-grabbed mask) or EINVAL
499          * (not possible, since we are using correct arguments).  */
500         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
501     } else {
502         /* child process */
503         size_t i;
504 
505         /* Clear out all signal handlers from parent so nothing
506          * unexpected can happen in our child once we unblock
507          * signals */
508         sig_action.sa_handler = SIG_DFL;
509         sig_action.sa_flags = 0;
510         sigemptyset(&sig_action.sa_mask);
511 
512         for (i = 1; i < NSIG; i++) {
513             /* Only possible errors are EFAULT or EINVAL The former
514              * won't happen, the latter we expect, so no need to check
515              * return value */
516             (void)sigaction(i, &sig_action, NULL);
517         }
518 
519         /* Unmask all signals in child, since we've no idea what the
520          * caller's done with their signal mask and don't want to
521          * propagate that to children */
522         sigemptyset(&newmask);
523         if (pthread_sigmask(SIG_SETMASK, &newmask, NULL) != 0) {
524             Error *local_err = NULL;
525             error_setg_errno(&local_err, errno,
526                              "cannot unblock signals");
527             error_report_err(local_err);
528             _exit(1);
529         }
530     }
531     return pid;
532 }
533 
534 void *qemu_alloc_stack(size_t *sz)
535 {
536     void *ptr, *guardpage;
537 #ifdef CONFIG_DEBUG_STACK_USAGE
538     void *ptr2;
539 #endif
540     size_t pagesz = getpagesize();
541 #ifdef _SC_THREAD_STACK_MIN
542     /* avoid stacks smaller than _SC_THREAD_STACK_MIN */
543     long min_stack_sz = sysconf(_SC_THREAD_STACK_MIN);
544     *sz = MAX(MAX(min_stack_sz, 0), *sz);
545 #endif
546     /* adjust stack size to a multiple of the page size */
547     *sz = ROUND_UP(*sz, pagesz);
548     /* allocate one extra page for the guard page */
549     *sz += pagesz;
550 
551     ptr = mmap(NULL, *sz, PROT_READ | PROT_WRITE,
552                MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
553     if (ptr == MAP_FAILED) {
554         abort();
555     }
556 
557 #if defined(HOST_IA64)
558     /* separate register stack */
559     guardpage = ptr + (((*sz - pagesz) / 2) & ~pagesz);
560 #elif defined(HOST_HPPA)
561     /* stack grows up */
562     guardpage = ptr + *sz - pagesz;
563 #else
564     /* stack grows down */
565     guardpage = ptr;
566 #endif
567     if (mprotect(guardpage, pagesz, PROT_NONE) != 0) {
568         abort();
569     }
570 
571 #ifdef CONFIG_DEBUG_STACK_USAGE
572     for (ptr2 = ptr + pagesz; ptr2 < ptr + *sz; ptr2 += sizeof(uint32_t)) {
573         *(uint32_t *)ptr2 = 0xdeadbeaf;
574     }
575 #endif
576 
577     return ptr;
578 }
579 
580 #ifdef CONFIG_DEBUG_STACK_USAGE
581 static __thread unsigned int max_stack_usage;
582 #endif
583 
584 void qemu_free_stack(void *stack, size_t sz)
585 {
586 #ifdef CONFIG_DEBUG_STACK_USAGE
587     unsigned int usage;
588     void *ptr;
589 
590     for (ptr = stack + getpagesize(); ptr < stack + sz;
591          ptr += sizeof(uint32_t)) {
592         if (*(uint32_t *)ptr != 0xdeadbeaf) {
593             break;
594         }
595     }
596     usage = sz - (uintptr_t) (ptr - stack);
597     if (usage > max_stack_usage) {
598         error_report("thread %d max stack usage increased from %u to %u",
599                      qemu_get_thread_id(), max_stack_usage, usage);
600         max_stack_usage = usage;
601     }
602 #endif
603 
604     munmap(stack, sz);
605 }
606 
607 void sigaction_invoke(struct sigaction *action,
608                       struct qemu_signalfd_siginfo *info)
609 {
610     siginfo_t si = { 0 };
611     si.si_signo = info->ssi_signo;
612     si.si_errno = info->ssi_errno;
613     si.si_code = info->ssi_code;
614 
615     /* Convert the minimal set of fields defined by POSIX.
616      * Positive si_code values are reserved for kernel-generated
617      * signals, where the valid siginfo fields are determined by
618      * the signal number.  But according to POSIX, it is unspecified
619      * whether SI_USER and SI_QUEUE have values less than or equal to
620      * zero.
621      */
622     if (info->ssi_code == SI_USER || info->ssi_code == SI_QUEUE ||
623         info->ssi_code <= 0) {
624         /* SIGTERM, etc.  */
625         si.si_pid = info->ssi_pid;
626         si.si_uid = info->ssi_uid;
627     } else if (info->ssi_signo == SIGILL || info->ssi_signo == SIGFPE ||
628                info->ssi_signo == SIGSEGV || info->ssi_signo == SIGBUS) {
629         si.si_addr = (void *)(uintptr_t)info->ssi_addr;
630     } else if (info->ssi_signo == SIGCHLD) {
631         si.si_pid = info->ssi_pid;
632         si.si_status = info->ssi_status;
633         si.si_uid = info->ssi_uid;
634     } else if (info->ssi_signo == SIGIO) {
635         si.si_band = info->ssi_band;
636     }
637     action->sa_sigaction(info->ssi_signo, &si, NULL);
638 }
639