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 int qemu_get_thread_id(void)
81 #if defined(__linux__)
82 return syscall(SYS_gettid
);
88 int qemu_daemon(int nochdir
, int noclose
)
90 return daemon(nochdir
, noclose
);
93 bool qemu_write_pidfile(const char *path
, Error
**errp
)
100 struct flock lock
= {
102 .l_whence
= SEEK_SET
,
106 fd
= qemu_open(path
, O_CREAT
| O_WRONLY
, S_IRUSR
| S_IWUSR
);
108 error_setg_errno(errp
, errno
, "Cannot open pid file");
112 if (fstat(fd
, &b
) < 0) {
113 error_setg_errno(errp
, errno
, "Cannot stat file");
117 if (fcntl(fd
, F_SETLK
, &lock
)) {
118 error_setg_errno(errp
, errno
, "Cannot lock pid file");
123 * Now make sure the path we locked is the same one that now
124 * exists on the filesystem.
126 if (stat(path
, &a
) < 0) {
128 * PID file disappeared, someone else must be racing with
135 if (a
.st_ino
== b
.st_ino
) {
140 * PID file was recreated, someone else must be racing with
146 if (ftruncate(fd
, 0) < 0) {
147 error_setg_errno(errp
, errno
, "Failed to truncate pid file");
151 snprintf(pidstr
, sizeof(pidstr
), FMT_pid
"\n", getpid());
152 if (write(fd
, pidstr
, strlen(pidstr
)) != strlen(pidstr
)) {
153 error_setg(errp
, "Failed to write pid file");
166 void *qemu_oom_check(void *ptr
)
169 fprintf(stderr
, "Failed to allocate memory: %s\n", strerror(errno
));
175 void *qemu_try_memalign(size_t alignment
, size_t size
)
179 if (alignment
< sizeof(void*)) {
180 alignment
= sizeof(void*);
183 #if defined(CONFIG_POSIX_MEMALIGN)
185 ret
= posix_memalign(&ptr
, alignment
, size
);
190 #elif defined(CONFIG_BSD)
193 ptr
= memalign(alignment
, size
);
195 trace_qemu_memalign(alignment
, size
, ptr
);
199 void *qemu_memalign(size_t alignment
, size_t size
)
201 return qemu_oom_check(qemu_try_memalign(alignment
, size
));
204 /* alloc shared memory pages */
205 void *qemu_anon_ram_alloc(size_t size
, uint64_t *alignment
, bool shared
)
207 size_t align
= QEMU_VMALLOC_ALIGN
;
208 void *ptr
= qemu_ram_mmap(-1, size
, align
, shared
, false);
210 if (ptr
== MAP_FAILED
) {
218 trace_qemu_anon_ram_alloc(size
, ptr
);
222 void qemu_vfree(void *ptr
)
224 trace_qemu_vfree(ptr
);
228 void qemu_anon_ram_free(void *ptr
, size_t size
)
230 trace_qemu_anon_ram_free(ptr
, size
);
231 qemu_ram_munmap(-1, ptr
, size
);
234 void qemu_set_block(int fd
)
237 f
= fcntl(fd
, F_GETFL
);
239 f
= fcntl(fd
, F_SETFL
, f
& ~O_NONBLOCK
);
243 void qemu_set_nonblock(int fd
)
246 f
= fcntl(fd
, F_GETFL
);
248 f
= fcntl(fd
, F_SETFL
, f
| O_NONBLOCK
);
252 * Previous to OpenBSD 6.3, fcntl(F_SETFL) is not permitted on
253 * memory devices and sets errno to ENODEV.
254 * It's OK if we fail to set O_NONBLOCK on devices like /dev/null,
255 * because they will never block anyway.
257 assert(errno
== ENODEV
);
264 int socket_set_fast_reuse(int fd
)
268 ret
= setsockopt(fd
, SOL_SOCKET
, SO_REUSEADDR
,
269 (const char *)&val
, sizeof(val
));
276 void qemu_set_cloexec(int fd
)
279 f
= fcntl(fd
, F_GETFD
);
281 f
= fcntl(fd
, F_SETFD
, f
| FD_CLOEXEC
);
286 * Creates a pipe with FD_CLOEXEC set on both file descriptors
288 int qemu_pipe(int pipefd
[2])
293 ret
= pipe2(pipefd
, O_CLOEXEC
);
294 if (ret
!= -1 || errno
!= ENOSYS
) {
300 qemu_set_cloexec(pipefd
[0]);
301 qemu_set_cloexec(pipefd
[1]);
308 qemu_get_local_state_pathname(const char *relative_pathname
)
310 return g_strdup_printf("%s/%s", CONFIG_QEMU_LOCALSTATEDIR
,
314 void qemu_set_tty_echo(int fd
, bool echo
)
321 tty
.c_lflag
|= ECHO
| ECHONL
| ICANON
| IEXTEN
;
323 tty
.c_lflag
&= ~(ECHO
| ECHONL
| ICANON
| IEXTEN
);
326 tcsetattr(fd
, TCSANOW
, &tty
);
329 static char exec_dir
[PATH_MAX
];
331 void qemu_init_exec_dir(const char *argv0
)
337 assert(!exec_dir
[0]);
339 #if defined(__linux__)
342 len
= readlink("/proc/self/exe", buf
, sizeof(buf
) - 1);
348 #elif defined(__FreeBSD__) \
349 || (defined(__NetBSD__) && defined(KERN_PROC_PATHNAME))
351 #if defined(__FreeBSD__)
352 static int mib
[4] = {CTL_KERN
, KERN_PROC
, KERN_PROC_PATHNAME
, -1};
354 static int mib
[4] = {CTL_KERN
, KERN_PROC_ARGS
, -1, KERN_PROC_PATHNAME
};
356 size_t len
= sizeof(buf
) - 1;
359 if (!sysctl(mib
, ARRAY_SIZE(mib
), buf
, &len
, NULL
, 0) &&
361 buf
[sizeof(buf
) - 1] = '\0';
366 /* If we don't have any way of figuring out the actual executable
367 location then try argv[0]. */
372 p
= realpath(argv0
, buf
);
377 dir
= g_path_get_dirname(p
);
379 pstrcpy(exec_dir
, sizeof(exec_dir
), dir
);
384 char *qemu_get_exec_dir(void)
386 return g_strdup(exec_dir
);
389 static void sigbus_handler(int signal
)
393 for (i
= 0; i
< memset_num_threads
; i
++) {
394 if (qemu_thread_is_self(&memset_thread
[i
].pgthread
)) {
395 siglongjmp(memset_thread
[i
].env
, 1);
401 static void *do_touch_pages(void *arg
)
403 MemsetThread
*memset_args
= (MemsetThread
*)arg
;
404 sigset_t set
, oldset
;
408 sigaddset(&set
, SIGBUS
);
409 pthread_sigmask(SIG_UNBLOCK
, &set
, &oldset
);
411 if (sigsetjmp(memset_args
->env
, 1)) {
412 memset_thread_failed
= true;
414 char *addr
= memset_args
->addr
;
415 size_t numpages
= memset_args
->numpages
;
416 size_t hpagesize
= memset_args
->hpagesize
;
418 for (i
= 0; i
< numpages
; i
++) {
420 * Read & write back the same value, so we don't
421 * corrupt existing user/app data that might be
424 * 'volatile' to stop compiler optimizing this away
427 * TODO: get a better solution from kernel so we
428 * don't need to write at all so we don't cause
429 * wear on the storage backing the region...
431 *(volatile char *)addr
= *addr
;
435 pthread_sigmask(SIG_SETMASK
, &oldset
, NULL
);
439 static inline int get_memset_num_threads(int smp_cpus
)
441 long host_procs
= sysconf(_SC_NPROCESSORS_ONLN
);
444 if (host_procs
> 0) {
445 ret
= MIN(MIN(host_procs
, MAX_MEM_PREALLOC_THREAD_COUNT
), smp_cpus
);
447 /* In case sysconf() fails, we fall back to single threaded */
451 static bool touch_all_pages(char *area
, size_t hpagesize
, size_t numpages
,
454 size_t numpages_per_thread
;
455 size_t size_per_thread
;
459 memset_thread_failed
= false;
460 memset_num_threads
= get_memset_num_threads(smp_cpus
);
461 memset_thread
= g_new0(MemsetThread
, memset_num_threads
);
462 numpages_per_thread
= (numpages
/ memset_num_threads
);
463 size_per_thread
= (hpagesize
* numpages_per_thread
);
464 for (i
= 0; i
< memset_num_threads
; i
++) {
465 memset_thread
[i
].addr
= addr
;
466 memset_thread
[i
].numpages
= (i
== (memset_num_threads
- 1)) ?
467 numpages
: numpages_per_thread
;
468 memset_thread
[i
].hpagesize
= hpagesize
;
469 qemu_thread_create(&memset_thread
[i
].pgthread
, "touch_pages",
470 do_touch_pages
, &memset_thread
[i
],
471 QEMU_THREAD_JOINABLE
);
472 addr
+= size_per_thread
;
473 numpages
-= numpages_per_thread
;
475 for (i
= 0; i
< memset_num_threads
; i
++) {
476 qemu_thread_join(&memset_thread
[i
].pgthread
);
478 g_free(memset_thread
);
479 memset_thread
= NULL
;
481 return memset_thread_failed
;
484 void os_mem_prealloc(int fd
, char *area
, size_t memory
, int smp_cpus
,
488 struct sigaction act
, oldact
;
489 size_t hpagesize
= qemu_fd_getpagesize(fd
);
490 size_t numpages
= DIV_ROUND_UP(memory
, hpagesize
);
492 memset(&act
, 0, sizeof(act
));
493 act
.sa_handler
= &sigbus_handler
;
496 ret
= sigaction(SIGBUS
, &act
, &oldact
);
498 error_setg_errno(errp
, errno
,
499 "os_mem_prealloc: failed to install signal handler");
503 /* touch pages simultaneously */
504 if (touch_all_pages(area
, hpagesize
, numpages
, smp_cpus
)) {
505 error_setg(errp
, "os_mem_prealloc: Insufficient free host memory "
506 "pages available to allocate guest RAM");
509 ret
= sigaction(SIGBUS
, &oldact
, NULL
);
511 /* Terminate QEMU since it can't recover from error */
512 perror("os_mem_prealloc: failed to reinstall signal handler");
517 char *qemu_get_pid_name(pid_t pid
)
521 #if defined(__FreeBSD__)
522 /* BSDs don't have /proc, but they provide a nice substitute */
523 struct kinfo_proc
*proc
= kinfo_getproc(pid
);
526 name
= g_strdup(proc
->ki_comm
);
530 /* Assume a system with reasonable procfs */
534 pid_path
= g_strdup_printf("/proc/%d/cmdline", pid
);
535 g_file_get_contents(pid_path
, &name
, &len
, NULL
);
543 pid_t
qemu_fork(Error
**errp
)
545 sigset_t oldmask
, newmask
;
546 struct sigaction sig_action
;
551 * Need to block signals now, so that child process can safely
552 * kill off caller's signal handlers without a race.
554 sigfillset(&newmask
);
555 if (pthread_sigmask(SIG_SETMASK
, &newmask
, &oldmask
) != 0) {
556 error_setg_errno(errp
, errno
,
557 "cannot block signals");
565 /* attempt to restore signal mask, but ignore failure, to
566 * avoid obscuring the fork failure */
567 (void)pthread_sigmask(SIG_SETMASK
, &oldmask
, NULL
);
568 error_setg_errno(errp
, saved_errno
,
569 "cannot fork child process");
575 /* Restore our original signal mask now that the child is
576 * safely running. Only documented failures are EFAULT (not
577 * possible, since we are using just-grabbed mask) or EINVAL
578 * (not possible, since we are using correct arguments). */
579 (void)pthread_sigmask(SIG_SETMASK
, &oldmask
, NULL
);
584 /* Clear out all signal handlers from parent so nothing
585 * unexpected can happen in our child once we unblock
587 sig_action
.sa_handler
= SIG_DFL
;
588 sig_action
.sa_flags
= 0;
589 sigemptyset(&sig_action
.sa_mask
);
591 for (i
= 1; i
< NSIG
; i
++) {
592 /* Only possible errors are EFAULT or EINVAL The former
593 * won't happen, the latter we expect, so no need to check
595 (void)sigaction(i
, &sig_action
, NULL
);
598 /* Unmask all signals in child, since we've no idea what the
599 * caller's done with their signal mask and don't want to
600 * propagate that to children */
601 sigemptyset(&newmask
);
602 if (pthread_sigmask(SIG_SETMASK
, &newmask
, NULL
) != 0) {
603 Error
*local_err
= NULL
;
604 error_setg_errno(&local_err
, errno
,
605 "cannot unblock signals");
606 error_report_err(local_err
);
613 void *qemu_alloc_stack(size_t *sz
)
615 void *ptr
, *guardpage
;
617 #ifdef CONFIG_DEBUG_STACK_USAGE
620 size_t pagesz
= qemu_real_host_page_size
;
621 #ifdef _SC_THREAD_STACK_MIN
622 /* avoid stacks smaller than _SC_THREAD_STACK_MIN */
623 long min_stack_sz
= sysconf(_SC_THREAD_STACK_MIN
);
624 *sz
= MAX(MAX(min_stack_sz
, 0), *sz
);
626 /* adjust stack size to a multiple of the page size */
627 *sz
= ROUND_UP(*sz
, pagesz
);
628 /* allocate one extra page for the guard page */
631 flags
= MAP_PRIVATE
| MAP_ANONYMOUS
;
632 #if defined(MAP_STACK) && defined(__OpenBSD__)
633 /* Only enable MAP_STACK on OpenBSD. Other OS's such as
634 * Linux/FreeBSD/NetBSD have a flag with the same name
635 * but have differing functionality. OpenBSD will SEGV
636 * if it spots execution with a stack pointer pointing
637 * at memory that was not allocated with MAP_STACK.
642 ptr
= mmap(NULL
, *sz
, PROT_READ
| PROT_WRITE
, flags
, -1, 0);
643 if (ptr
== MAP_FAILED
) {
644 perror("failed to allocate memory for stack");
648 #if defined(HOST_IA64)
649 /* separate register stack */
650 guardpage
= ptr
+ (((*sz
- pagesz
) / 2) & ~pagesz
);
651 #elif defined(HOST_HPPA)
653 guardpage
= ptr
+ *sz
- pagesz
;
655 /* stack grows down */
658 if (mprotect(guardpage
, pagesz
, PROT_NONE
) != 0) {
659 perror("failed to set up stack guard page");
663 #ifdef CONFIG_DEBUG_STACK_USAGE
664 for (ptr2
= ptr
+ pagesz
; ptr2
< ptr
+ *sz
; ptr2
+= sizeof(uint32_t)) {
665 *(uint32_t *)ptr2
= 0xdeadbeaf;
672 #ifdef CONFIG_DEBUG_STACK_USAGE
673 static __thread
unsigned int max_stack_usage
;
676 void qemu_free_stack(void *stack
, size_t sz
)
678 #ifdef CONFIG_DEBUG_STACK_USAGE
682 for (ptr
= stack
+ qemu_real_host_page_size
; ptr
< stack
+ sz
;
683 ptr
+= sizeof(uint32_t)) {
684 if (*(uint32_t *)ptr
!= 0xdeadbeaf) {
688 usage
= sz
- (uintptr_t) (ptr
- stack
);
689 if (usage
> max_stack_usage
) {
690 error_report("thread %d max stack usage increased from %u to %u",
691 qemu_get_thread_id(), max_stack_usage
, usage
);
692 max_stack_usage
= usage
;
699 void sigaction_invoke(struct sigaction
*action
,
700 struct qemu_signalfd_siginfo
*info
)
703 si
.si_signo
= info
->ssi_signo
;
704 si
.si_errno
= info
->ssi_errno
;
705 si
.si_code
= info
->ssi_code
;
707 /* Convert the minimal set of fields defined by POSIX.
708 * Positive si_code values are reserved for kernel-generated
709 * signals, where the valid siginfo fields are determined by
710 * the signal number. But according to POSIX, it is unspecified
711 * whether SI_USER and SI_QUEUE have values less than or equal to
714 if (info
->ssi_code
== SI_USER
|| info
->ssi_code
== SI_QUEUE
||
715 info
->ssi_code
<= 0) {
717 si
.si_pid
= info
->ssi_pid
;
718 si
.si_uid
= info
->ssi_uid
;
719 } else if (info
->ssi_signo
== SIGILL
|| info
->ssi_signo
== SIGFPE
||
720 info
->ssi_signo
== SIGSEGV
|| info
->ssi_signo
== SIGBUS
) {
721 si
.si_addr
= (void *)(uintptr_t)info
->ssi_addr
;
722 } else if (info
->ssi_signo
== SIGCHLD
) {
723 si
.si_pid
= info
->ssi_pid
;
724 si
.si_status
= info
->ssi_status
;
725 si
.si_uid
= info
->ssi_uid
;
727 action
->sa_sigaction(info
->ssi_signo
, &si
, NULL
);