mmap-alloc: fix hugetlbfs misaligned length in ppc64
[qemu/kevin.git] / util / oslib-posix.c
blob37c5854b9c8214cedb58fc0cc38169cd9d46fac6
1 /*
2 * os-posix-lib.c
4 * Copyright (c) 2003-2008 Fabrice Bellard
5 * Copyright (c) 2010 Red Hat, Inc.
7 * QEMU library functions on POSIX which are shared between QEMU and
8 * the QEMU tools.
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:
17 * The above copyright notice and this permission notice shall be included in
18 * all copies or substantial portions of the Software.
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.
29 #include "qemu/osdep.h"
30 #include <termios.h>
32 #include <glib/gprintf.h>
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"
42 #ifdef CONFIG_LINUX
43 #include <sys/syscall.h>
44 #endif
46 #ifdef __FreeBSD__
47 #include <sys/sysctl.h>
48 #include <sys/user.h>
49 #include <libutil.h>
50 #endif
52 #ifdef __NetBSD__
53 #include <sys/sysctl.h>
54 #endif
56 #include "qemu/mmap-alloc.h"
58 #ifdef CONFIG_DEBUG_STACK_USAGE
59 #include "qemu/error-report.h"
60 #endif
62 #define MAX_MEM_PREALLOC_THREAD_COUNT 16
64 struct MemsetThread {
65 char *addr;
66 size_t numpages;
67 size_t hpagesize;
68 QemuThread pgthread;
69 sigjmp_buf env;
71 typedef struct MemsetThread MemsetThread;
73 static MemsetThread *memset_thread;
74 static int memset_num_threads;
75 static bool memset_thread_failed;
77 int qemu_get_thread_id(void)
79 #if defined(__linux__)
80 return syscall(SYS_gettid);
81 #else
82 return getpid();
83 #endif
86 int qemu_daemon(int nochdir, int noclose)
88 return daemon(nochdir, noclose);
91 bool qemu_write_pidfile(const char *path, Error **errp)
93 int fd;
94 char pidstr[32];
96 while (1) {
97 struct stat a, b;
98 struct flock lock = {
99 .l_type = F_WRLCK,
100 .l_whence = SEEK_SET,
101 .l_len = 0,
104 fd = qemu_open(path, O_CREAT | O_WRONLY, S_IRUSR | S_IWUSR);
105 if (fd == -1) {
106 error_setg_errno(errp, errno, "Cannot open pid file");
107 return false;
110 if (fstat(fd, &b) < 0) {
111 error_setg_errno(errp, errno, "Cannot stat file");
112 goto fail_close;
115 if (fcntl(fd, F_SETLK, &lock)) {
116 error_setg_errno(errp, errno, "Cannot lock pid file");
117 goto fail_close;
121 * Now make sure the path we locked is the same one that now
122 * exists on the filesystem.
124 if (stat(path, &a) < 0) {
126 * PID file disappeared, someone else must be racing with
127 * us, so try again.
129 close(fd);
130 continue;
133 if (a.st_ino == b.st_ino) {
134 break;
138 * PID file was recreated, someone else must be racing with
139 * us, so try again.
141 close(fd);
144 if (ftruncate(fd, 0) < 0) {
145 error_setg_errno(errp, errno, "Failed to truncate pid file");
146 goto fail_unlink;
149 snprintf(pidstr, sizeof(pidstr), FMT_pid "\n", getpid());
150 if (write(fd, pidstr, strlen(pidstr)) != strlen(pidstr)) {
151 error_setg(errp, "Failed to write pid file");
152 goto fail_unlink;
155 return true;
157 fail_unlink:
158 unlink(path);
159 fail_close:
160 close(fd);
161 return false;
164 void *qemu_oom_check(void *ptr)
166 if (ptr == NULL) {
167 fprintf(stderr, "Failed to allocate memory: %s\n", strerror(errno));
168 abort();
170 return ptr;
173 void *qemu_try_memalign(size_t alignment, size_t size)
175 void *ptr;
177 if (alignment < sizeof(void*)) {
178 alignment = sizeof(void*);
181 #if defined(CONFIG_POSIX_MEMALIGN)
182 int ret;
183 ret = posix_memalign(&ptr, alignment, size);
184 if (ret != 0) {
185 errno = ret;
186 ptr = NULL;
188 #elif defined(CONFIG_BSD)
189 ptr = valloc(size);
190 #else
191 ptr = memalign(alignment, size);
192 #endif
193 trace_qemu_memalign(alignment, size, ptr);
194 return ptr;
197 void *qemu_memalign(size_t alignment, size_t size)
199 return qemu_oom_check(qemu_try_memalign(alignment, size));
202 /* alloc shared memory pages */
203 void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment, bool shared)
205 size_t align = QEMU_VMALLOC_ALIGN;
206 void *ptr = qemu_ram_mmap(-1, size, align, shared);
208 if (ptr == MAP_FAILED) {
209 return NULL;
212 if (alignment) {
213 *alignment = align;
216 trace_qemu_anon_ram_alloc(size, ptr);
217 return ptr;
220 void qemu_vfree(void *ptr)
222 trace_qemu_vfree(ptr);
223 free(ptr);
226 void qemu_anon_ram_free(void *ptr, size_t size)
228 trace_qemu_anon_ram_free(ptr, size);
229 qemu_ram_munmap(-1, ptr, size);
232 void qemu_set_block(int fd)
234 int f;
235 f = fcntl(fd, F_GETFL);
236 assert(f != -1);
237 f = fcntl(fd, F_SETFL, f & ~O_NONBLOCK);
238 assert(f != -1);
241 void qemu_set_nonblock(int fd)
243 int f;
244 f = fcntl(fd, F_GETFL);
245 assert(f != -1);
246 f = fcntl(fd, F_SETFL, f | O_NONBLOCK);
247 assert(f != -1);
250 int socket_set_fast_reuse(int fd)
252 int val = 1, ret;
254 ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
255 (const char *)&val, sizeof(val));
257 assert(ret == 0);
259 return ret;
262 void qemu_set_cloexec(int fd)
264 int f;
265 f = fcntl(fd, F_GETFD);
266 assert(f != -1);
267 f = fcntl(fd, F_SETFD, f | FD_CLOEXEC);
268 assert(f != -1);
272 * Creates a pipe with FD_CLOEXEC set on both file descriptors
274 int qemu_pipe(int pipefd[2])
276 int ret;
278 #ifdef CONFIG_PIPE2
279 ret = pipe2(pipefd, O_CLOEXEC);
280 if (ret != -1 || errno != ENOSYS) {
281 return ret;
283 #endif
284 ret = pipe(pipefd);
285 if (ret == 0) {
286 qemu_set_cloexec(pipefd[0]);
287 qemu_set_cloexec(pipefd[1]);
290 return ret;
293 char *
294 qemu_get_local_state_pathname(const char *relative_pathname)
296 return g_strdup_printf("%s/%s", CONFIG_QEMU_LOCALSTATEDIR,
297 relative_pathname);
300 void qemu_set_tty_echo(int fd, bool echo)
302 struct termios tty;
304 tcgetattr(fd, &tty);
306 if (echo) {
307 tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN;
308 } else {
309 tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN);
312 tcsetattr(fd, TCSANOW, &tty);
315 static char exec_dir[PATH_MAX];
317 void qemu_init_exec_dir(const char *argv0)
319 char *dir;
320 char *p = NULL;
321 char buf[PATH_MAX];
323 assert(!exec_dir[0]);
325 #if defined(__linux__)
327 int len;
328 len = readlink("/proc/self/exe", buf, sizeof(buf) - 1);
329 if (len > 0) {
330 buf[len] = 0;
331 p = buf;
334 #elif defined(__FreeBSD__) \
335 || (defined(__NetBSD__) && defined(KERN_PROC_PATHNAME))
337 #if defined(__FreeBSD__)
338 static int mib[4] = {CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1};
339 #else
340 static int mib[4] = {CTL_KERN, KERN_PROC_ARGS, -1, KERN_PROC_PATHNAME};
341 #endif
342 size_t len = sizeof(buf) - 1;
344 *buf = '\0';
345 if (!sysctl(mib, ARRAY_SIZE(mib), buf, &len, NULL, 0) &&
346 *buf) {
347 buf[sizeof(buf) - 1] = '\0';
348 p = buf;
351 #endif
352 /* If we don't have any way of figuring out the actual executable
353 location then try argv[0]. */
354 if (!p) {
355 if (!argv0) {
356 return;
358 p = realpath(argv0, buf);
359 if (!p) {
360 return;
363 dir = g_path_get_dirname(p);
365 pstrcpy(exec_dir, sizeof(exec_dir), dir);
367 g_free(dir);
370 char *qemu_get_exec_dir(void)
372 return g_strdup(exec_dir);
375 static void sigbus_handler(int signal)
377 int i;
378 if (memset_thread) {
379 for (i = 0; i < memset_num_threads; i++) {
380 if (qemu_thread_is_self(&memset_thread[i].pgthread)) {
381 siglongjmp(memset_thread[i].env, 1);
387 static void *do_touch_pages(void *arg)
389 MemsetThread *memset_args = (MemsetThread *)arg;
390 sigset_t set, oldset;
392 /* unblock SIGBUS */
393 sigemptyset(&set);
394 sigaddset(&set, SIGBUS);
395 pthread_sigmask(SIG_UNBLOCK, &set, &oldset);
397 if (sigsetjmp(memset_args->env, 1)) {
398 memset_thread_failed = true;
399 } else {
400 char *addr = memset_args->addr;
401 size_t numpages = memset_args->numpages;
402 size_t hpagesize = memset_args->hpagesize;
403 size_t i;
404 for (i = 0; i < numpages; i++) {
406 * Read & write back the same value, so we don't
407 * corrupt existing user/app data that might be
408 * stored.
410 * 'volatile' to stop compiler optimizing this away
411 * to a no-op
413 * TODO: get a better solution from kernel so we
414 * don't need to write at all so we don't cause
415 * wear on the storage backing the region...
417 *(volatile char *)addr = *addr;
418 addr += hpagesize;
421 pthread_sigmask(SIG_SETMASK, &oldset, NULL);
422 return NULL;
425 static inline int get_memset_num_threads(int smp_cpus)
427 long host_procs = sysconf(_SC_NPROCESSORS_ONLN);
428 int ret = 1;
430 if (host_procs > 0) {
431 ret = MIN(MIN(host_procs, MAX_MEM_PREALLOC_THREAD_COUNT), smp_cpus);
433 /* In case sysconf() fails, we fall back to single threaded */
434 return ret;
437 static bool touch_all_pages(char *area, size_t hpagesize, size_t numpages,
438 int smp_cpus)
440 size_t numpages_per_thread;
441 size_t size_per_thread;
442 char *addr = area;
443 int i = 0;
445 memset_thread_failed = false;
446 memset_num_threads = get_memset_num_threads(smp_cpus);
447 memset_thread = g_new0(MemsetThread, memset_num_threads);
448 numpages_per_thread = (numpages / memset_num_threads);
449 size_per_thread = (hpagesize * numpages_per_thread);
450 for (i = 0; i < memset_num_threads; i++) {
451 memset_thread[i].addr = addr;
452 memset_thread[i].numpages = (i == (memset_num_threads - 1)) ?
453 numpages : numpages_per_thread;
454 memset_thread[i].hpagesize = hpagesize;
455 qemu_thread_create(&memset_thread[i].pgthread, "touch_pages",
456 do_touch_pages, &memset_thread[i],
457 QEMU_THREAD_JOINABLE);
458 addr += size_per_thread;
459 numpages -= numpages_per_thread;
461 for (i = 0; i < memset_num_threads; i++) {
462 qemu_thread_join(&memset_thread[i].pgthread);
464 g_free(memset_thread);
465 memset_thread = NULL;
467 return memset_thread_failed;
470 void os_mem_prealloc(int fd, char *area, size_t memory, int smp_cpus,
471 Error **errp)
473 int ret;
474 struct sigaction act, oldact;
475 size_t hpagesize = qemu_fd_getpagesize(fd);
476 size_t numpages = DIV_ROUND_UP(memory, hpagesize);
478 memset(&act, 0, sizeof(act));
479 act.sa_handler = &sigbus_handler;
480 act.sa_flags = 0;
482 ret = sigaction(SIGBUS, &act, &oldact);
483 if (ret) {
484 error_setg_errno(errp, errno,
485 "os_mem_prealloc: failed to install signal handler");
486 return;
489 /* touch pages simultaneously */
490 if (touch_all_pages(area, hpagesize, numpages, smp_cpus)) {
491 error_setg(errp, "os_mem_prealloc: Insufficient free host memory "
492 "pages available to allocate guest RAM");
495 ret = sigaction(SIGBUS, &oldact, NULL);
496 if (ret) {
497 /* Terminate QEMU since it can't recover from error */
498 perror("os_mem_prealloc: failed to reinstall signal handler");
499 exit(1);
504 char *qemu_get_pid_name(pid_t pid)
506 char *name = NULL;
508 #if defined(__FreeBSD__)
509 /* BSDs don't have /proc, but they provide a nice substitute */
510 struct kinfo_proc *proc = kinfo_getproc(pid);
512 if (proc) {
513 name = g_strdup(proc->ki_comm);
514 free(proc);
516 #else
517 /* Assume a system with reasonable procfs */
518 char *pid_path;
519 size_t len;
521 pid_path = g_strdup_printf("/proc/%d/cmdline", pid);
522 g_file_get_contents(pid_path, &name, &len, NULL);
523 g_free(pid_path);
524 #endif
526 return name;
530 pid_t qemu_fork(Error **errp)
532 sigset_t oldmask, newmask;
533 struct sigaction sig_action;
534 int saved_errno;
535 pid_t pid;
538 * Need to block signals now, so that child process can safely
539 * kill off caller's signal handlers without a race.
541 sigfillset(&newmask);
542 if (pthread_sigmask(SIG_SETMASK, &newmask, &oldmask) != 0) {
543 error_setg_errno(errp, errno,
544 "cannot block signals");
545 return -1;
548 pid = fork();
549 saved_errno = errno;
551 if (pid < 0) {
552 /* attempt to restore signal mask, but ignore failure, to
553 * avoid obscuring the fork failure */
554 (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
555 error_setg_errno(errp, saved_errno,
556 "cannot fork child process");
557 errno = saved_errno;
558 return -1;
559 } else if (pid) {
560 /* parent process */
562 /* Restore our original signal mask now that the child is
563 * safely running. Only documented failures are EFAULT (not
564 * possible, since we are using just-grabbed mask) or EINVAL
565 * (not possible, since we are using correct arguments). */
566 (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
567 } else {
568 /* child process */
569 size_t i;
571 /* Clear out all signal handlers from parent so nothing
572 * unexpected can happen in our child once we unblock
573 * signals */
574 sig_action.sa_handler = SIG_DFL;
575 sig_action.sa_flags = 0;
576 sigemptyset(&sig_action.sa_mask);
578 for (i = 1; i < NSIG; i++) {
579 /* Only possible errors are EFAULT or EINVAL The former
580 * won't happen, the latter we expect, so no need to check
581 * return value */
582 (void)sigaction(i, &sig_action, NULL);
585 /* Unmask all signals in child, since we've no idea what the
586 * caller's done with their signal mask and don't want to
587 * propagate that to children */
588 sigemptyset(&newmask);
589 if (pthread_sigmask(SIG_SETMASK, &newmask, NULL) != 0) {
590 Error *local_err = NULL;
591 error_setg_errno(&local_err, errno,
592 "cannot unblock signals");
593 error_report_err(local_err);
594 _exit(1);
597 return pid;
600 void *qemu_alloc_stack(size_t *sz)
602 void *ptr, *guardpage;
603 int flags;
604 #ifdef CONFIG_DEBUG_STACK_USAGE
605 void *ptr2;
606 #endif
607 size_t pagesz = getpagesize();
608 #ifdef _SC_THREAD_STACK_MIN
609 /* avoid stacks smaller than _SC_THREAD_STACK_MIN */
610 long min_stack_sz = sysconf(_SC_THREAD_STACK_MIN);
611 *sz = MAX(MAX(min_stack_sz, 0), *sz);
612 #endif
613 /* adjust stack size to a multiple of the page size */
614 *sz = ROUND_UP(*sz, pagesz);
615 /* allocate one extra page for the guard page */
616 *sz += pagesz;
618 flags = MAP_PRIVATE | MAP_ANONYMOUS;
619 #if defined(MAP_STACK) && defined(__OpenBSD__)
620 /* Only enable MAP_STACK on OpenBSD. Other OS's such as
621 * Linux/FreeBSD/NetBSD have a flag with the same name
622 * but have differing functionality. OpenBSD will SEGV
623 * if it spots execution with a stack pointer pointing
624 * at memory that was not allocated with MAP_STACK.
626 flags |= MAP_STACK;
627 #endif
629 ptr = mmap(NULL, *sz, PROT_READ | PROT_WRITE, flags, -1, 0);
630 if (ptr == MAP_FAILED) {
631 perror("failed to allocate memory for stack");
632 abort();
635 #if defined(HOST_IA64)
636 /* separate register stack */
637 guardpage = ptr + (((*sz - pagesz) / 2) & ~pagesz);
638 #elif defined(HOST_HPPA)
639 /* stack grows up */
640 guardpage = ptr + *sz - pagesz;
641 #else
642 /* stack grows down */
643 guardpage = ptr;
644 #endif
645 if (mprotect(guardpage, pagesz, PROT_NONE) != 0) {
646 perror("failed to set up stack guard page");
647 abort();
650 #ifdef CONFIG_DEBUG_STACK_USAGE
651 for (ptr2 = ptr + pagesz; ptr2 < ptr + *sz; ptr2 += sizeof(uint32_t)) {
652 *(uint32_t *)ptr2 = 0xdeadbeaf;
654 #endif
656 return ptr;
659 #ifdef CONFIG_DEBUG_STACK_USAGE
660 static __thread unsigned int max_stack_usage;
661 #endif
663 void qemu_free_stack(void *stack, size_t sz)
665 #ifdef CONFIG_DEBUG_STACK_USAGE
666 unsigned int usage;
667 void *ptr;
669 for (ptr = stack + getpagesize(); ptr < stack + sz;
670 ptr += sizeof(uint32_t)) {
671 if (*(uint32_t *)ptr != 0xdeadbeaf) {
672 break;
675 usage = sz - (uintptr_t) (ptr - stack);
676 if (usage > max_stack_usage) {
677 error_report("thread %d max stack usage increased from %u to %u",
678 qemu_get_thread_id(), max_stack_usage, usage);
679 max_stack_usage = usage;
681 #endif
683 munmap(stack, sz);
686 void sigaction_invoke(struct sigaction *action,
687 struct qemu_signalfd_siginfo *info)
689 siginfo_t si = {};
690 si.si_signo = info->ssi_signo;
691 si.si_errno = info->ssi_errno;
692 si.si_code = info->ssi_code;
694 /* Convert the minimal set of fields defined by POSIX.
695 * Positive si_code values are reserved for kernel-generated
696 * signals, where the valid siginfo fields are determined by
697 * the signal number. But according to POSIX, it is unspecified
698 * whether SI_USER and SI_QUEUE have values less than or equal to
699 * zero.
701 if (info->ssi_code == SI_USER || info->ssi_code == SI_QUEUE ||
702 info->ssi_code <= 0) {
703 /* SIGTERM, etc. */
704 si.si_pid = info->ssi_pid;
705 si.si_uid = info->ssi_uid;
706 } else if (info->ssi_signo == SIGILL || info->ssi_signo == SIGFPE ||
707 info->ssi_signo == SIGSEGV || info->ssi_signo == SIGBUS) {
708 si.si_addr = (void *)(uintptr_t)info->ssi_addr;
709 } else if (info->ssi_signo == SIGCHLD) {
710 si.si_pid = info->ssi_pid;
711 si.si_status = info->ssi_status;
712 si.si_uid = info->ssi_uid;
714 action->sa_sigaction(info->ssi_signo, &si, NULL);