Merge remote-tracking branch 'remotes/vivier2/tags/linux-user-for-4.1-pull-request...
[qemu/ar7.git] / util / oslib-posix.c
blobfe0309ce7ae5d3a5e40ccb89bf9b1df249930b1b
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 "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"
43 #ifdef CONFIG_LINUX
44 #include <sys/syscall.h>
45 #endif
47 #ifdef __FreeBSD__
48 #include <sys/sysctl.h>
49 #include <sys/user.h>
50 #include <libutil.h>
51 #endif
53 #ifdef __NetBSD__
54 #include <sys/sysctl.h>
55 #endif
57 #include "qemu/mmap-alloc.h"
59 #ifdef CONFIG_DEBUG_STACK_USAGE
60 #include "qemu/error-report.h"
61 #endif
63 #define MAX_MEM_PREALLOC_THREAD_COUNT 16
65 struct MemsetThread {
66 char *addr;
67 size_t numpages;
68 size_t hpagesize;
69 QemuThread pgthread;
70 sigjmp_buf env;
72 typedef struct MemsetThread MemsetThread;
74 static MemsetThread *memset_thread;
75 static int memset_num_threads;
76 static bool memset_thread_failed;
78 int qemu_get_thread_id(void)
80 #if defined(__linux__)
81 return syscall(SYS_gettid);
82 #else
83 return getpid();
84 #endif
87 int qemu_daemon(int nochdir, int noclose)
89 return daemon(nochdir, noclose);
92 bool qemu_write_pidfile(const char *path, Error **errp)
94 int fd;
95 char pidstr[32];
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,
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;
111 if (fstat(fd, &b) < 0) {
112 error_setg_errno(errp, errno, "Cannot stat file");
113 goto fail_close;
116 if (fcntl(fd, F_SETLK, &lock)) {
117 error_setg_errno(errp, errno, "Cannot lock pid file");
118 goto fail_close;
122 * Now make sure the path we locked is the same one that now
123 * exists on the filesystem.
125 if (stat(path, &a) < 0) {
127 * PID file disappeared, someone else must be racing with
128 * us, so try again.
130 close(fd);
131 continue;
134 if (a.st_ino == b.st_ino) {
135 break;
139 * PID file was recreated, someone else must be racing with
140 * us, so try again.
142 close(fd);
145 if (ftruncate(fd, 0) < 0) {
146 error_setg_errno(errp, errno, "Failed to truncate pid file");
147 goto fail_unlink;
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;
156 return true;
158 fail_unlink:
159 unlink(path);
160 fail_close:
161 close(fd);
162 return false;
165 void *qemu_oom_check(void *ptr)
167 if (ptr == NULL) {
168 fprintf(stderr, "Failed to allocate memory: %s\n", strerror(errno));
169 abort();
171 return ptr;
174 void *qemu_try_memalign(size_t alignment, size_t size)
176 void *ptr;
178 if (alignment < sizeof(void*)) {
179 alignment = sizeof(void*);
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;
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;
198 void *qemu_memalign(size_t alignment, size_t size)
200 return qemu_oom_check(qemu_try_memalign(alignment, size));
203 /* alloc shared memory pages */
204 void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment, bool shared)
206 size_t align = QEMU_VMALLOC_ALIGN;
207 void *ptr = qemu_ram_mmap(-1, size, align, shared, false);
209 if (ptr == MAP_FAILED) {
210 return NULL;
213 if (alignment) {
214 *alignment = align;
217 trace_qemu_anon_ram_alloc(size, ptr);
218 return ptr;
221 void qemu_vfree(void *ptr)
223 trace_qemu_vfree(ptr);
224 free(ptr);
227 void qemu_anon_ram_free(void *ptr, size_t size)
229 trace_qemu_anon_ram_free(ptr, size);
230 qemu_ram_munmap(-1, ptr, size);
233 void qemu_set_block(int fd)
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);
242 void qemu_set_nonblock(int fd)
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) {
251 * Previous to OpenBSD 6.3, fcntl(F_SETFL) is not permitted on
252 * memory devices and sets errno to ENODEV.
253 * It's OK if we fail to set O_NONBLOCK on devices like /dev/null,
254 * because they will never block anyway.
256 assert(errno == ENODEV);
258 #else
259 assert(f != -1);
260 #endif
263 int socket_set_fast_reuse(int fd)
265 int val = 1, ret;
267 ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
268 (const char *)&val, sizeof(val));
270 assert(ret == 0);
272 return ret;
275 void qemu_set_cloexec(int fd)
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);
285 * Creates a pipe with FD_CLOEXEC set on both file descriptors
287 int qemu_pipe(int pipefd[2])
289 int ret;
291 #ifdef CONFIG_PIPE2
292 ret = pipe2(pipefd, O_CLOEXEC);
293 if (ret != -1 || errno != ENOSYS) {
294 return ret;
296 #endif
297 ret = pipe(pipefd);
298 if (ret == 0) {
299 qemu_set_cloexec(pipefd[0]);
300 qemu_set_cloexec(pipefd[1]);
303 return ret;
306 char *
307 qemu_get_local_state_pathname(const char *relative_pathname)
309 return g_strdup_printf("%s/%s", CONFIG_QEMU_LOCALSTATEDIR,
310 relative_pathname);
313 void qemu_set_tty_echo(int fd, bool echo)
315 struct termios tty;
317 tcgetattr(fd, &tty);
319 if (echo) {
320 tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN;
321 } else {
322 tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN);
325 tcsetattr(fd, TCSANOW, &tty);
328 static char exec_dir[PATH_MAX];
330 void qemu_init_exec_dir(const char *argv0)
332 char *dir;
333 char *p = NULL;
334 char buf[PATH_MAX];
336 assert(!exec_dir[0]);
338 #if defined(__linux__)
340 int len;
341 len = readlink("/proc/self/exe", buf, sizeof(buf) - 1);
342 if (len > 0) {
343 buf[len] = 0;
344 p = buf;
347 #elif defined(__FreeBSD__) \
348 || (defined(__NetBSD__) && defined(KERN_PROC_PATHNAME))
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;
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;
364 #endif
365 /* If we don't have any way of figuring out the actual executable
366 location then try argv[0]. */
367 if (!p) {
368 if (!argv0) {
369 return;
371 p = realpath(argv0, buf);
372 if (!p) {
373 return;
376 dir = g_path_get_dirname(p);
378 pstrcpy(exec_dir, sizeof(exec_dir), dir);
380 g_free(dir);
383 char *qemu_get_exec_dir(void)
385 return g_strdup(exec_dir);
388 static void sigbus_handler(int signal)
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);
400 static void *do_touch_pages(void *arg)
402 MemsetThread *memset_args = (MemsetThread *)arg;
403 sigset_t set, oldset;
405 /* unblock SIGBUS */
406 sigemptyset(&set);
407 sigaddset(&set, SIGBUS);
408 pthread_sigmask(SIG_UNBLOCK, &set, &oldset);
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++) {
419 * Read & write back the same value, so we don't
420 * corrupt existing user/app data that might be
421 * stored.
423 * 'volatile' to stop compiler optimizing this away
424 * to a no-op
426 * TODO: get a better solution from kernel so we
427 * don't need to write at all so we don't cause
428 * wear on the storage backing the region...
430 *(volatile char *)addr = *addr;
431 addr += hpagesize;
434 pthread_sigmask(SIG_SETMASK, &oldset, NULL);
435 return NULL;
438 static inline int get_memset_num_threads(int smp_cpus)
440 long host_procs = sysconf(_SC_NPROCESSORS_ONLN);
441 int ret = 1;
443 if (host_procs > 0) {
444 ret = MIN(MIN(host_procs, MAX_MEM_PREALLOC_THREAD_COUNT), smp_cpus);
446 /* In case sysconf() fails, we fall back to single threaded */
447 return ret;
450 static bool touch_all_pages(char *area, size_t hpagesize, size_t numpages,
451 int smp_cpus)
453 size_t numpages_per_thread;
454 size_t size_per_thread;
455 char *addr = area;
456 int i = 0;
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;
474 for (i = 0; i < memset_num_threads; i++) {
475 qemu_thread_join(&memset_thread[i].pgthread);
477 g_free(memset_thread);
478 memset_thread = NULL;
480 return memset_thread_failed;
483 void os_mem_prealloc(int fd, char *area, size_t memory, int smp_cpus,
484 Error **errp)
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);
491 memset(&act, 0, sizeof(act));
492 act.sa_handler = &sigbus_handler;
493 act.sa_flags = 0;
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;
502 /* touch pages simultaneously */
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");
508 ret = sigaction(SIGBUS, &oldact, NULL);
509 if (ret) {
510 /* Terminate QEMU since it can't recover from error */
511 perror("os_mem_prealloc: failed to reinstall signal handler");
512 exit(1);
516 uint64_t qemu_get_pmem_size(const char *filename, Error **errp)
518 struct stat st;
520 if (stat(filename, &st) < 0) {
521 error_setg(errp, "unable to stat pmem file \"%s\"", filename);
522 return 0;
525 #if defined(__linux__)
526 /* Special handling for devdax character devices */
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;
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;
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;
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;
556 ret = g_ascii_strtoull(size_str, NULL, 0);
558 devdax_err:
559 g_free(size_str);
560 g_free(size_path);
561 g_free(subsystem);
562 g_free(subsystem_path);
563 return ret;
565 #endif /* defined(__linux__) */
567 return st.st_size;
570 char *qemu_get_pid_name(pid_t pid)
572 char *name = NULL;
574 #if defined(__FreeBSD__)
575 /* BSDs don't have /proc, but they provide a nice substitute */
576 struct kinfo_proc *proc = kinfo_getproc(pid);
578 if (proc) {
579 name = g_strdup(proc->ki_comm);
580 free(proc);
582 #else
583 /* Assume a system with reasonable procfs */
584 char *pid_path;
585 size_t len;
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
592 return name;
596 pid_t qemu_fork(Error **errp)
598 sigset_t oldmask, newmask;
599 struct sigaction sig_action;
600 int saved_errno;
601 pid_t pid;
604 * Need to block signals now, so that child process can safely
605 * kill off caller's signal handlers without a race.
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;
614 pid = fork();
615 saved_errno = errno;
617 if (pid < 0) {
618 /* attempt to restore signal mask, but ignore failure, to
619 * avoid obscuring the fork failure */
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 /* parent process */
628 /* Restore our original signal mask now that the child is
629 * safely running. Only documented failures are EFAULT (not
630 * possible, since we are using just-grabbed mask) or EINVAL
631 * (not possible, since we are using correct arguments). */
632 (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
633 } else {
634 /* child process */
635 size_t i;
637 /* Clear out all signal handlers from parent so nothing
638 * unexpected can happen in our child once we unblock
639 * signals */
640 sig_action.sa_handler = SIG_DFL;
641 sig_action.sa_flags = 0;
642 sigemptyset(&sig_action.sa_mask);
644 for (i = 1; i < NSIG; i++) {
645 /* Only possible errors are EFAULT or EINVAL The former
646 * won't happen, the latter we expect, so no need to check
647 * return value */
648 (void)sigaction(i, &sig_action, NULL);
651 /* Unmask all signals in child, since we've no idea what the
652 * caller's done with their signal mask and don't want to
653 * propagate that to children */
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);
663 return pid;
666 void *qemu_alloc_stack(size_t *sz)
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 /* avoid stacks smaller than _SC_THREAD_STACK_MIN */
676 long min_stack_sz = sysconf(_SC_THREAD_STACK_MIN);
677 *sz = MAX(MAX(min_stack_sz, 0), *sz);
678 #endif
679 /* adjust stack size to a multiple of the page size */
680 *sz = ROUND_UP(*sz, pagesz);
681 /* allocate one extra page for the guard page */
682 *sz += pagesz;
684 flags = MAP_PRIVATE | MAP_ANONYMOUS;
685 #if defined(MAP_STACK) && defined(__OpenBSD__)
686 /* Only enable MAP_STACK on OpenBSD. Other OS's such as
687 * Linux/FreeBSD/NetBSD have a flag with the same name
688 * but have differing functionality. OpenBSD will SEGV
689 * if it spots execution with a stack pointer pointing
690 * at memory that was not allocated with MAP_STACK.
692 flags |= MAP_STACK;
693 #endif
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();
701 #if defined(HOST_IA64)
702 /* separate register stack */
703 guardpage = ptr + (((*sz - pagesz) / 2) & ~pagesz);
704 #elif defined(HOST_HPPA)
705 /* stack grows up */
706 guardpage = ptr + *sz - pagesz;
707 #else
708 /* stack grows down */
709 guardpage = ptr;
710 #endif
711 if (mprotect(guardpage, pagesz, PROT_NONE) != 0) {
712 perror("failed to set up stack guard page");
713 abort();
716 #ifdef CONFIG_DEBUG_STACK_USAGE
717 for (ptr2 = ptr + pagesz; ptr2 < ptr + *sz; ptr2 += sizeof(uint32_t)) {
718 *(uint32_t *)ptr2 = 0xdeadbeaf;
720 #endif
722 return ptr;
725 #ifdef CONFIG_DEBUG_STACK_USAGE
726 static __thread unsigned int max_stack_usage;
727 #endif
729 void qemu_free_stack(void *stack, size_t sz)
731 #ifdef CONFIG_DEBUG_STACK_USAGE
732 unsigned int usage;
733 void *ptr;
735 for (ptr = stack + getpagesize(); ptr < stack + sz;
736 ptr += sizeof(uint32_t)) {
737 if (*(uint32_t *)ptr != 0xdeadbeaf) {
738 break;
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;
747 #endif
749 munmap(stack, sz);
752 void sigaction_invoke(struct sigaction *action,
753 struct qemu_signalfd_siginfo *info)
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;
760 /* Convert the minimal set of fields defined by POSIX.
761 * Positive si_code values are reserved for kernel-generated
762 * signals, where the valid siginfo fields are determined by
763 * the signal number. But according to POSIX, it is unspecified
764 * whether SI_USER and SI_QUEUE have values less than or equal to
765 * zero.
767 if (info->ssi_code == SI_USER || info->ssi_code == SI_QUEUE ||
768 info->ssi_code <= 0) {
769 /* SIGTERM, etc. */
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;
780 action->sa_sigaction(info->ssi_signo, &si, NULL);