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
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
29 #include "qemu/osdep.h"
32 #include <glib/gprintf.h>
34 #include "qemu-common.h"
35 #include "sysemu/sysemu.h"
37 #include "qapi/error.h"
38 #include "qemu/sockets.h"
39 #include "qemu/thread.h"
41 #include <sys/signal.h>
42 #include "qemu/cutils.h"
45 #include <sys/syscall.h>
49 #include <sys/sysctl.h>
55 #include <sys/sysctl.h>
58 #include "qemu/mmap-alloc.h"
60 #ifdef CONFIG_DEBUG_STACK_USAGE
61 #include "qemu/error-report.h"
64 #define MAX_MEM_PREALLOC_THREAD_COUNT 16
73 typedef struct MemsetThread MemsetThread
;
75 static MemsetThread
*memset_thread
;
76 static int memset_num_threads
;
77 static bool memset_thread_failed
;
79 static QemuMutex page_mutex
;
80 static QemuCond page_cond
;
81 static bool threads_created_flag
;
83 int qemu_get_thread_id(void)
85 #if defined(__linux__)
86 return syscall(SYS_gettid
);
92 int qemu_daemon(int nochdir
, int noclose
)
94 return daemon(nochdir
, noclose
);
97 bool qemu_write_pidfile(const char *path
, Error
**errp
)
104 struct flock lock
= {
106 .l_whence
= SEEK_SET
,
110 fd
= qemu_open(path
, O_CREAT
| O_WRONLY
, S_IRUSR
| S_IWUSR
);
112 error_setg_errno(errp
, errno
, "Cannot open pid file");
116 if (fstat(fd
, &b
) < 0) {
117 error_setg_errno(errp
, errno
, "Cannot stat file");
121 if (fcntl(fd
, F_SETLK
, &lock
)) {
122 error_setg_errno(errp
, errno
, "Cannot lock pid file");
127 * Now make sure the path we locked is the same one that now
128 * exists on the filesystem.
130 if (stat(path
, &a
) < 0) {
132 * PID file disappeared, someone else must be racing with
139 if (a
.st_ino
== b
.st_ino
) {
144 * PID file was recreated, someone else must be racing with
150 if (ftruncate(fd
, 0) < 0) {
151 error_setg_errno(errp
, errno
, "Failed to truncate pid file");
155 snprintf(pidstr
, sizeof(pidstr
), FMT_pid
"\n", getpid());
156 if (write(fd
, pidstr
, strlen(pidstr
)) != strlen(pidstr
)) {
157 error_setg(errp
, "Failed to write pid file");
170 void *qemu_oom_check(void *ptr
)
173 fprintf(stderr
, "Failed to allocate memory: %s\n", strerror(errno
));
179 void *qemu_try_memalign(size_t alignment
, size_t size
)
183 if (alignment
< sizeof(void*)) {
184 alignment
= sizeof(void*);
187 #if defined(CONFIG_POSIX_MEMALIGN)
189 ret
= posix_memalign(&ptr
, alignment
, size
);
194 #elif defined(CONFIG_BSD)
197 ptr
= memalign(alignment
, size
);
199 trace_qemu_memalign(alignment
, size
, ptr
);
203 void *qemu_memalign(size_t alignment
, size_t size
)
205 return qemu_oom_check(qemu_try_memalign(alignment
, size
));
208 /* alloc shared memory pages */
209 void *qemu_anon_ram_alloc(size_t size
, uint64_t *alignment
, bool shared
)
211 size_t align
= QEMU_VMALLOC_ALIGN
;
212 void *ptr
= qemu_ram_mmap(-1, size
, align
, shared
, false);
214 if (ptr
== MAP_FAILED
) {
222 trace_qemu_anon_ram_alloc(size
, ptr
);
226 void qemu_vfree(void *ptr
)
228 trace_qemu_vfree(ptr
);
232 void qemu_anon_ram_free(void *ptr
, size_t size
)
234 trace_qemu_anon_ram_free(ptr
, size
);
235 qemu_ram_munmap(-1, ptr
, size
);
238 void qemu_set_block(int fd
)
241 f
= fcntl(fd
, F_GETFL
);
243 f
= fcntl(fd
, F_SETFL
, f
& ~O_NONBLOCK
);
247 void qemu_set_nonblock(int fd
)
250 f
= fcntl(fd
, F_GETFL
);
252 f
= fcntl(fd
, F_SETFL
, f
| O_NONBLOCK
);
256 * Previous to OpenBSD 6.3, fcntl(F_SETFL) is not permitted on
257 * memory devices and sets errno to ENODEV.
258 * It's OK if we fail to set O_NONBLOCK on devices like /dev/null,
259 * because they will never block anyway.
261 assert(errno
== ENODEV
);
268 int socket_set_fast_reuse(int fd
)
272 ret
= setsockopt(fd
, SOL_SOCKET
, SO_REUSEADDR
,
273 (const char *)&val
, sizeof(val
));
280 void qemu_set_cloexec(int fd
)
283 f
= fcntl(fd
, F_GETFD
);
285 f
= fcntl(fd
, F_SETFD
, f
| FD_CLOEXEC
);
290 * Creates a pipe with FD_CLOEXEC set on both file descriptors
292 int qemu_pipe(int pipefd
[2])
297 ret
= pipe2(pipefd
, O_CLOEXEC
);
298 if (ret
!= -1 || errno
!= ENOSYS
) {
304 qemu_set_cloexec(pipefd
[0]);
305 qemu_set_cloexec(pipefd
[1]);
312 qemu_get_local_state_pathname(const char *relative_pathname
)
314 return g_strdup_printf("%s/%s", CONFIG_QEMU_LOCALSTATEDIR
,
318 void qemu_set_tty_echo(int fd
, bool echo
)
325 tty
.c_lflag
|= ECHO
| ECHONL
| ICANON
| IEXTEN
;
327 tty
.c_lflag
&= ~(ECHO
| ECHONL
| ICANON
| IEXTEN
);
330 tcsetattr(fd
, TCSANOW
, &tty
);
333 static char exec_dir
[PATH_MAX
];
335 void qemu_init_exec_dir(const char *argv0
)
341 assert(!exec_dir
[0]);
343 #if defined(__linux__)
346 len
= readlink("/proc/self/exe", buf
, sizeof(buf
) - 1);
352 #elif defined(__FreeBSD__) \
353 || (defined(__NetBSD__) && defined(KERN_PROC_PATHNAME))
355 #if defined(__FreeBSD__)
356 static int mib
[4] = {CTL_KERN
, KERN_PROC
, KERN_PROC_PATHNAME
, -1};
358 static int mib
[4] = {CTL_KERN
, KERN_PROC_ARGS
, -1, KERN_PROC_PATHNAME
};
360 size_t len
= sizeof(buf
) - 1;
363 if (!sysctl(mib
, ARRAY_SIZE(mib
), buf
, &len
, NULL
, 0) &&
365 buf
[sizeof(buf
) - 1] = '\0';
370 /* If we don't have any way of figuring out the actual executable
371 location then try argv[0]. */
376 p
= realpath(argv0
, buf
);
381 dir
= g_path_get_dirname(p
);
383 pstrcpy(exec_dir
, sizeof(exec_dir
), dir
);
388 char *qemu_get_exec_dir(void)
390 return g_strdup(exec_dir
);
393 static void sigbus_handler(int signal
)
397 for (i
= 0; i
< memset_num_threads
; i
++) {
398 if (qemu_thread_is_self(&memset_thread
[i
].pgthread
)) {
399 siglongjmp(memset_thread
[i
].env
, 1);
405 static void *do_touch_pages(void *arg
)
407 MemsetThread
*memset_args
= (MemsetThread
*)arg
;
408 sigset_t set
, oldset
;
411 * On Linux, the page faults from the loop below can cause mmap_sem
412 * contention with allocation of the thread stacks. Do not start
413 * clearing until all threads have been created.
415 qemu_mutex_lock(&page_mutex
);
416 while(!threads_created_flag
){
417 qemu_cond_wait(&page_cond
, &page_mutex
);
419 qemu_mutex_unlock(&page_mutex
);
423 sigaddset(&set
, SIGBUS
);
424 pthread_sigmask(SIG_UNBLOCK
, &set
, &oldset
);
426 if (sigsetjmp(memset_args
->env
, 1)) {
427 memset_thread_failed
= true;
429 char *addr
= memset_args
->addr
;
430 size_t numpages
= memset_args
->numpages
;
431 size_t hpagesize
= memset_args
->hpagesize
;
433 for (i
= 0; i
< numpages
; i
++) {
435 * Read & write back the same value, so we don't
436 * corrupt existing user/app data that might be
439 * 'volatile' to stop compiler optimizing this away
442 * TODO: get a better solution from kernel so we
443 * don't need to write at all so we don't cause
444 * wear on the storage backing the region...
446 *(volatile char *)addr
= *addr
;
450 pthread_sigmask(SIG_SETMASK
, &oldset
, NULL
);
454 static inline int get_memset_num_threads(int smp_cpus
)
456 long host_procs
= sysconf(_SC_NPROCESSORS_ONLN
);
459 if (host_procs
> 0) {
460 ret
= MIN(MIN(host_procs
, MAX_MEM_PREALLOC_THREAD_COUNT
), smp_cpus
);
462 /* In case sysconf() fails, we fall back to single threaded */
466 static bool touch_all_pages(char *area
, size_t hpagesize
, size_t numpages
,
469 static gsize initialized
= 0;
470 size_t numpages_per_thread
, leftover
;
474 if (g_once_init_enter(&initialized
)) {
475 qemu_mutex_init(&page_mutex
);
476 qemu_cond_init(&page_cond
);
477 g_once_init_leave(&initialized
, 1);
480 memset_thread_failed
= false;
481 threads_created_flag
= false;
482 memset_num_threads
= get_memset_num_threads(smp_cpus
);
483 memset_thread
= g_new0(MemsetThread
, memset_num_threads
);
484 numpages_per_thread
= numpages
/ memset_num_threads
;
485 leftover
= numpages
% memset_num_threads
;
486 for (i
= 0; i
< memset_num_threads
; i
++) {
487 memset_thread
[i
].addr
= addr
;
488 memset_thread
[i
].numpages
= numpages_per_thread
+ (i
< leftover
);
489 memset_thread
[i
].hpagesize
= hpagesize
;
490 qemu_thread_create(&memset_thread
[i
].pgthread
, "touch_pages",
491 do_touch_pages
, &memset_thread
[i
],
492 QEMU_THREAD_JOINABLE
);
493 addr
+= memset_thread
[i
].numpages
* hpagesize
;
495 threads_created_flag
= true;
496 qemu_cond_broadcast(&page_cond
);
498 for (i
= 0; i
< memset_num_threads
; i
++) {
499 qemu_thread_join(&memset_thread
[i
].pgthread
);
501 g_free(memset_thread
);
502 memset_thread
= NULL
;
504 return memset_thread_failed
;
507 void os_mem_prealloc(int fd
, char *area
, size_t memory
, int smp_cpus
,
511 struct sigaction act
, oldact
;
512 size_t hpagesize
= qemu_fd_getpagesize(fd
);
513 size_t numpages
= DIV_ROUND_UP(memory
, hpagesize
);
515 memset(&act
, 0, sizeof(act
));
516 act
.sa_handler
= &sigbus_handler
;
519 ret
= sigaction(SIGBUS
, &act
, &oldact
);
521 error_setg_errno(errp
, errno
,
522 "os_mem_prealloc: failed to install signal handler");
526 /* touch pages simultaneously */
527 if (touch_all_pages(area
, hpagesize
, numpages
, smp_cpus
)) {
528 error_setg(errp
, "os_mem_prealloc: Insufficient free host memory "
529 "pages available to allocate guest RAM");
532 ret
= sigaction(SIGBUS
, &oldact
, NULL
);
534 /* Terminate QEMU since it can't recover from error */
535 perror("os_mem_prealloc: failed to reinstall signal handler");
540 char *qemu_get_pid_name(pid_t pid
)
544 #if defined(__FreeBSD__)
545 /* BSDs don't have /proc, but they provide a nice substitute */
546 struct kinfo_proc
*proc
= kinfo_getproc(pid
);
549 name
= g_strdup(proc
->ki_comm
);
553 /* Assume a system with reasonable procfs */
557 pid_path
= g_strdup_printf("/proc/%d/cmdline", pid
);
558 g_file_get_contents(pid_path
, &name
, &len
, NULL
);
566 pid_t
qemu_fork(Error
**errp
)
568 sigset_t oldmask
, newmask
;
569 struct sigaction sig_action
;
574 * Need to block signals now, so that child process can safely
575 * kill off caller's signal handlers without a race.
577 sigfillset(&newmask
);
578 if (pthread_sigmask(SIG_SETMASK
, &newmask
, &oldmask
) != 0) {
579 error_setg_errno(errp
, errno
,
580 "cannot block signals");
588 /* attempt to restore signal mask, but ignore failure, to
589 * avoid obscuring the fork failure */
590 (void)pthread_sigmask(SIG_SETMASK
, &oldmask
, NULL
);
591 error_setg_errno(errp
, saved_errno
,
592 "cannot fork child process");
598 /* Restore our original signal mask now that the child is
599 * safely running. Only documented failures are EFAULT (not
600 * possible, since we are using just-grabbed mask) or EINVAL
601 * (not possible, since we are using correct arguments). */
602 (void)pthread_sigmask(SIG_SETMASK
, &oldmask
, NULL
);
607 /* Clear out all signal handlers from parent so nothing
608 * unexpected can happen in our child once we unblock
610 sig_action
.sa_handler
= SIG_DFL
;
611 sig_action
.sa_flags
= 0;
612 sigemptyset(&sig_action
.sa_mask
);
614 for (i
= 1; i
< NSIG
; i
++) {
615 /* Only possible errors are EFAULT or EINVAL The former
616 * won't happen, the latter we expect, so no need to check
618 (void)sigaction(i
, &sig_action
, NULL
);
621 /* Unmask all signals in child, since we've no idea what the
622 * caller's done with their signal mask and don't want to
623 * propagate that to children */
624 sigemptyset(&newmask
);
625 if (pthread_sigmask(SIG_SETMASK
, &newmask
, NULL
) != 0) {
626 Error
*local_err
= NULL
;
627 error_setg_errno(&local_err
, errno
,
628 "cannot unblock signals");
629 error_report_err(local_err
);
636 void *qemu_alloc_stack(size_t *sz
)
638 void *ptr
, *guardpage
;
640 #ifdef CONFIG_DEBUG_STACK_USAGE
643 size_t pagesz
= qemu_real_host_page_size
;
644 #ifdef _SC_THREAD_STACK_MIN
645 /* avoid stacks smaller than _SC_THREAD_STACK_MIN */
646 long min_stack_sz
= sysconf(_SC_THREAD_STACK_MIN
);
647 *sz
= MAX(MAX(min_stack_sz
, 0), *sz
);
649 /* adjust stack size to a multiple of the page size */
650 *sz
= ROUND_UP(*sz
, pagesz
);
651 /* allocate one extra page for the guard page */
654 flags
= MAP_PRIVATE
| MAP_ANONYMOUS
;
655 #if defined(MAP_STACK) && defined(__OpenBSD__)
656 /* Only enable MAP_STACK on OpenBSD. Other OS's such as
657 * Linux/FreeBSD/NetBSD have a flag with the same name
658 * but have differing functionality. OpenBSD will SEGV
659 * if it spots execution with a stack pointer pointing
660 * at memory that was not allocated with MAP_STACK.
665 ptr
= mmap(NULL
, *sz
, PROT_READ
| PROT_WRITE
, flags
, -1, 0);
666 if (ptr
== MAP_FAILED
) {
667 perror("failed to allocate memory for stack");
671 #if defined(HOST_IA64)
672 /* separate register stack */
673 guardpage
= ptr
+ (((*sz
- pagesz
) / 2) & ~pagesz
);
674 #elif defined(HOST_HPPA)
676 guardpage
= ptr
+ *sz
- pagesz
;
678 /* stack grows down */
681 if (mprotect(guardpage
, pagesz
, PROT_NONE
) != 0) {
682 perror("failed to set up stack guard page");
686 #ifdef CONFIG_DEBUG_STACK_USAGE
687 for (ptr2
= ptr
+ pagesz
; ptr2
< ptr
+ *sz
; ptr2
+= sizeof(uint32_t)) {
688 *(uint32_t *)ptr2
= 0xdeadbeaf;
695 #ifdef CONFIG_DEBUG_STACK_USAGE
696 static __thread
unsigned int max_stack_usage
;
699 void qemu_free_stack(void *stack
, size_t sz
)
701 #ifdef CONFIG_DEBUG_STACK_USAGE
705 for (ptr
= stack
+ qemu_real_host_page_size
; ptr
< stack
+ sz
;
706 ptr
+= sizeof(uint32_t)) {
707 if (*(uint32_t *)ptr
!= 0xdeadbeaf) {
711 usage
= sz
- (uintptr_t) (ptr
- stack
);
712 if (usage
> max_stack_usage
) {
713 error_report("thread %d max stack usage increased from %u to %u",
714 qemu_get_thread_id(), max_stack_usage
, usage
);
715 max_stack_usage
= usage
;
722 void sigaction_invoke(struct sigaction
*action
,
723 struct qemu_signalfd_siginfo
*info
)
726 si
.si_signo
= info
->ssi_signo
;
727 si
.si_errno
= info
->ssi_errno
;
728 si
.si_code
= info
->ssi_code
;
730 /* Convert the minimal set of fields defined by POSIX.
731 * Positive si_code values are reserved for kernel-generated
732 * signals, where the valid siginfo fields are determined by
733 * the signal number. But according to POSIX, it is unspecified
734 * whether SI_USER and SI_QUEUE have values less than or equal to
737 if (info
->ssi_code
== SI_USER
|| info
->ssi_code
== SI_QUEUE
||
738 info
->ssi_code
<= 0) {
740 si
.si_pid
= info
->ssi_pid
;
741 si
.si_uid
= info
->ssi_uid
;
742 } else if (info
->ssi_signo
== SIGILL
|| info
->ssi_signo
== SIGFPE
||
743 info
->ssi_signo
== SIGSEGV
|| info
->ssi_signo
== SIGBUS
) {
744 si
.si_addr
= (void *)(uintptr_t)info
->ssi_addr
;
745 } else if (info
->ssi_signo
== SIGCHLD
) {
746 si
.si_pid
= info
->ssi_pid
;
747 si
.si_status
= info
->ssi_status
;
748 si
.si_uid
= info
->ssi_uid
;
750 action
->sa_sigaction(info
->ssi_signo
, &si
, NULL
);