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 "sysemu/sysemu.h"
36 #include "qapi/error.h"
37 #include "qemu/error-report.h"
38 #include "qemu/madvise.h"
39 #include "qemu/sockets.h"
40 #include "qemu/thread.h"
42 #include "qemu/cutils.h"
43 #include "qemu/units.h"
44 #include "qemu/thread-context.h"
47 #include <sys/syscall.h>
60 #include "qemu/mmap-alloc.h"
62 #define MAX_MEM_PREALLOC_THREAD_COUNT 16
66 typedef struct MemsetContext
{
67 bool all_threads_created
;
68 bool any_thread_failed
;
69 struct MemsetThread
*threads
;
79 MemsetContext
*context
;
81 typedef struct MemsetThread MemsetThread
;
83 /* used by sigbus_handler() */
84 static MemsetContext
*sigbus_memset_context
;
85 struct sigaction sigbus_oldact
;
86 static QemuMutex sigbus_mutex
;
88 static QemuMutex page_mutex
;
89 static QemuCond page_cond
;
91 int qemu_get_thread_id(void)
93 #if defined(__linux__)
94 return syscall(SYS_gettid
);
95 #elif defined(__FreeBSD__)
96 /* thread id is up to INT_MAX */
100 #elif defined(__NetBSD__)
102 #elif defined(__OpenBSD__)
109 int qemu_daemon(int nochdir
, int noclose
)
111 return daemon(nochdir
, noclose
);
114 bool qemu_write_pidfile(const char *path
, Error
**errp
)
121 struct flock lock
= {
123 .l_whence
= SEEK_SET
,
127 fd
= qemu_create(path
, O_WRONLY
, S_IRUSR
| S_IWUSR
, errp
);
132 if (fstat(fd
, &b
) < 0) {
133 error_setg_errno(errp
, errno
, "Cannot stat file");
137 if (fcntl(fd
, F_SETLK
, &lock
)) {
138 error_setg_errno(errp
, errno
, "Cannot lock pid file");
143 * Now make sure the path we locked is the same one that now
144 * exists on the filesystem.
146 if (stat(path
, &a
) < 0) {
148 * PID file disappeared, someone else must be racing with
155 if (a
.st_ino
== b
.st_ino
) {
160 * PID file was recreated, someone else must be racing with
166 if (ftruncate(fd
, 0) < 0) {
167 error_setg_errno(errp
, errno
, "Failed to truncate pid file");
171 snprintf(pidstr
, sizeof(pidstr
), FMT_pid
"\n", getpid());
172 if (qemu_write_full(fd
, pidstr
, strlen(pidstr
)) != strlen(pidstr
)) {
173 error_setg(errp
, "Failed to write pid file");
186 /* alloc shared memory pages */
187 void *qemu_anon_ram_alloc(size_t size
, uint64_t *alignment
, bool shared
,
190 const uint32_t qemu_map_flags
= (shared
? QEMU_MAP_SHARED
: 0) |
191 (noreserve
? QEMU_MAP_NORESERVE
: 0);
192 size_t align
= QEMU_VMALLOC_ALIGN
;
193 void *ptr
= qemu_ram_mmap(-1, size
, align
, qemu_map_flags
, 0);
195 if (ptr
== MAP_FAILED
) {
203 trace_qemu_anon_ram_alloc(size
, ptr
);
207 void qemu_anon_ram_free(void *ptr
, size_t size
)
209 trace_qemu_anon_ram_free(ptr
, size
);
210 qemu_ram_munmap(-1, ptr
, size
);
213 void qemu_socket_set_block(int fd
)
215 g_unix_set_fd_nonblocking(fd
, false, NULL
);
218 int qemu_socket_try_set_nonblock(int fd
)
220 return g_unix_set_fd_nonblocking(fd
, true, NULL
) ? 0 : -errno
;
223 void qemu_socket_set_nonblock(int fd
)
226 f
= qemu_socket_try_set_nonblock(fd
);
230 int socket_set_fast_reuse(int fd
)
234 ret
= setsockopt(fd
, SOL_SOCKET
, SO_REUSEADDR
,
235 (const char *)&val
, sizeof(val
));
242 void qemu_set_cloexec(int fd
)
245 f
= fcntl(fd
, F_GETFD
);
247 f
= fcntl(fd
, F_SETFD
, f
| FD_CLOEXEC
);
251 int qemu_socketpair(int domain
, int type
, int protocol
, int sv
[2])
256 ret
= socketpair(domain
, type
| SOCK_CLOEXEC
, protocol
, sv
);
257 if (ret
!= -1 || errno
!= EINVAL
) {
261 ret
= socketpair(domain
, type
, protocol
, sv
);;
263 qemu_set_cloexec(sv
[0]);
264 qemu_set_cloexec(sv
[1]);
271 qemu_get_local_state_dir(void)
273 return get_relocated_path(CONFIG_QEMU_LOCALSTATEDIR
);
276 void qemu_set_tty_echo(int fd
, bool echo
)
283 tty
.c_lflag
|= ECHO
| ECHONL
| ICANON
| IEXTEN
;
285 tty
.c_lflag
&= ~(ECHO
| ECHONL
| ICANON
| IEXTEN
);
288 tcsetattr(fd
, TCSANOW
, &tty
);
292 static void sigbus_handler(int signal
, siginfo_t
*siginfo
, void *ctx
)
293 #else /* CONFIG_LINUX */
294 static void sigbus_handler(int signal
)
295 #endif /* CONFIG_LINUX */
299 if (sigbus_memset_context
) {
300 for (i
= 0; i
< sigbus_memset_context
->num_threads
; i
++) {
301 MemsetThread
*thread
= &sigbus_memset_context
->threads
[i
];
303 if (qemu_thread_is_self(&thread
->pgthread
)) {
304 siglongjmp(thread
->env
, 1);
311 * We assume that the MCE SIGBUS handler could have been registered. We
312 * should never receive BUS_MCEERR_AO on any of our threads, but only on
313 * the main thread registered for PR_MCE_KILL_EARLY. Further, we should not
314 * receive BUS_MCEERR_AR triggered by action of other threads on one of
315 * our threads. So, no need to check for unrelated SIGBUS when seeing one
318 * We will forward to the MCE handler, which will either handle the SIGBUS
319 * or reinstall the default SIGBUS handler and reraise the SIGBUS. The
320 * default SIGBUS handler will crash the process, so we don't care.
322 if (sigbus_oldact
.sa_flags
& SA_SIGINFO
) {
323 sigbus_oldact
.sa_sigaction(signal
, siginfo
, ctx
);
326 #endif /* CONFIG_LINUX */
327 warn_report("qemu_prealloc_mem: unrelated SIGBUS detected and ignored");
330 static void *do_touch_pages(void *arg
)
332 MemsetThread
*memset_args
= (MemsetThread
*)arg
;
333 sigset_t set
, oldset
;
337 * On Linux, the page faults from the loop below can cause mmap_sem
338 * contention with allocation of the thread stacks. Do not start
339 * clearing until all threads have been created.
341 qemu_mutex_lock(&page_mutex
);
342 while (!memset_args
->context
->all_threads_created
) {
343 qemu_cond_wait(&page_cond
, &page_mutex
);
345 qemu_mutex_unlock(&page_mutex
);
349 sigaddset(&set
, SIGBUS
);
350 pthread_sigmask(SIG_UNBLOCK
, &set
, &oldset
);
352 if (sigsetjmp(memset_args
->env
, 1)) {
355 char *addr
= memset_args
->addr
;
356 size_t numpages
= memset_args
->numpages
;
357 size_t hpagesize
= memset_args
->hpagesize
;
359 for (i
= 0; i
< numpages
; i
++) {
361 * Read & write back the same value, so we don't
362 * corrupt existing user/app data that might be
365 * 'volatile' to stop compiler optimizing this away
368 *(volatile char *)addr
= *addr
;
372 pthread_sigmask(SIG_SETMASK
, &oldset
, NULL
);
373 return (void *)(uintptr_t)ret
;
376 static void *do_madv_populate_write_pages(void *arg
)
378 MemsetThread
*memset_args
= (MemsetThread
*)arg
;
379 const size_t size
= memset_args
->numpages
* memset_args
->hpagesize
;
380 char * const addr
= memset_args
->addr
;
383 /* See do_touch_pages(). */
384 qemu_mutex_lock(&page_mutex
);
385 while (!memset_args
->context
->all_threads_created
) {
386 qemu_cond_wait(&page_cond
, &page_mutex
);
388 qemu_mutex_unlock(&page_mutex
);
390 if (size
&& qemu_madvise(addr
, size
, QEMU_MADV_POPULATE_WRITE
)) {
393 return (void *)(uintptr_t)ret
;
396 static inline int get_memset_num_threads(size_t hpagesize
, size_t numpages
,
399 long host_procs
= sysconf(_SC_NPROCESSORS_ONLN
);
402 if (host_procs
> 0) {
403 ret
= MIN(MIN(host_procs
, MAX_MEM_PREALLOC_THREAD_COUNT
), max_threads
);
406 /* Especially with gigantic pages, don't create more threads than pages. */
407 ret
= MIN(ret
, numpages
);
408 /* Don't start threads to prealloc comparatively little memory. */
409 ret
= MIN(ret
, MAX(1, hpagesize
* numpages
/ (64 * MiB
)));
411 /* In case sysconf() fails, we fall back to single threaded */
415 static int touch_all_pages(char *area
, size_t hpagesize
, size_t numpages
,
416 int max_threads
, ThreadContext
*tc
,
417 bool use_madv_populate_write
)
419 static gsize initialized
= 0;
420 MemsetContext context
= {
421 .num_threads
= get_memset_num_threads(hpagesize
, numpages
, max_threads
),
423 size_t numpages_per_thread
, leftover
;
424 void *(*touch_fn
)(void *);
428 if (g_once_init_enter(&initialized
)) {
429 qemu_mutex_init(&page_mutex
);
430 qemu_cond_init(&page_cond
);
431 g_once_init_leave(&initialized
, 1);
434 if (use_madv_populate_write
) {
435 /* Avoid creating a single thread for MADV_POPULATE_WRITE */
436 if (context
.num_threads
== 1) {
437 if (qemu_madvise(area
, hpagesize
* numpages
,
438 QEMU_MADV_POPULATE_WRITE
)) {
443 touch_fn
= do_madv_populate_write_pages
;
445 touch_fn
= do_touch_pages
;
448 context
.threads
= g_new0(MemsetThread
, context
.num_threads
);
449 numpages_per_thread
= numpages
/ context
.num_threads
;
450 leftover
= numpages
% context
.num_threads
;
451 for (i
= 0; i
< context
.num_threads
; i
++) {
452 context
.threads
[i
].addr
= addr
;
453 context
.threads
[i
].numpages
= numpages_per_thread
+ (i
< leftover
);
454 context
.threads
[i
].hpagesize
= hpagesize
;
455 context
.threads
[i
].context
= &context
;
457 thread_context_create_thread(tc
, &context
.threads
[i
].pgthread
,
459 touch_fn
, &context
.threads
[i
],
460 QEMU_THREAD_JOINABLE
);
462 qemu_thread_create(&context
.threads
[i
].pgthread
, "touch_pages",
463 touch_fn
, &context
.threads
[i
],
464 QEMU_THREAD_JOINABLE
);
466 addr
+= context
.threads
[i
].numpages
* hpagesize
;
469 if (!use_madv_populate_write
) {
470 sigbus_memset_context
= &context
;
473 qemu_mutex_lock(&page_mutex
);
474 context
.all_threads_created
= true;
475 qemu_cond_broadcast(&page_cond
);
476 qemu_mutex_unlock(&page_mutex
);
478 for (i
= 0; i
< context
.num_threads
; i
++) {
479 int tmp
= (uintptr_t)qemu_thread_join(&context
.threads
[i
].pgthread
);
486 if (!use_madv_populate_write
) {
487 sigbus_memset_context
= NULL
;
489 g_free(context
.threads
);
494 static bool madv_populate_write_possible(char *area
, size_t pagesize
)
496 return !qemu_madvise(area
, pagesize
, QEMU_MADV_POPULATE_WRITE
) ||
500 bool qemu_prealloc_mem(int fd
, char *area
, size_t sz
, int max_threads
,
501 ThreadContext
*tc
, Error
**errp
)
503 static gsize initialized
;
505 size_t hpagesize
= qemu_fd_getpagesize(fd
);
506 size_t numpages
= DIV_ROUND_UP(sz
, hpagesize
);
507 bool use_madv_populate_write
;
508 struct sigaction act
;
512 * Sense on every invocation, as MADV_POPULATE_WRITE cannot be used for
513 * some special mappings, such as mapping /dev/mem.
515 use_madv_populate_write
= madv_populate_write_possible(area
, hpagesize
);
517 if (!use_madv_populate_write
) {
518 if (g_once_init_enter(&initialized
)) {
519 qemu_mutex_init(&sigbus_mutex
);
520 g_once_init_leave(&initialized
, 1);
523 qemu_mutex_lock(&sigbus_mutex
);
524 memset(&act
, 0, sizeof(act
));
526 act
.sa_sigaction
= &sigbus_handler
;
527 act
.sa_flags
= SA_SIGINFO
;
528 #else /* CONFIG_LINUX */
529 act
.sa_handler
= &sigbus_handler
;
531 #endif /* CONFIG_LINUX */
533 ret
= sigaction(SIGBUS
, &act
, &sigbus_oldact
);
535 qemu_mutex_unlock(&sigbus_mutex
);
536 error_setg_errno(errp
, errno
,
537 "qemu_prealloc_mem: failed to install signal handler");
542 /* touch pages simultaneously */
543 ret
= touch_all_pages(area
, hpagesize
, numpages
, max_threads
, tc
,
544 use_madv_populate_write
);
546 error_setg_errno(errp
, -ret
,
547 "qemu_prealloc_mem: preallocating memory failed");
551 if (!use_madv_populate_write
) {
552 ret
= sigaction(SIGBUS
, &sigbus_oldact
, NULL
);
554 /* Terminate QEMU since it can't recover from error */
555 perror("qemu_prealloc_mem: failed to reinstall signal handler");
558 qemu_mutex_unlock(&sigbus_mutex
);
563 char *qemu_get_pid_name(pid_t pid
)
567 #if defined(__FreeBSD__)
568 /* BSDs don't have /proc, but they provide a nice substitute */
569 struct kinfo_proc
*proc
= kinfo_getproc(pid
);
572 name
= g_strdup(proc
->ki_comm
);
576 /* Assume a system with reasonable procfs */
580 pid_path
= g_strdup_printf("/proc/%d/cmdline", pid
);
581 g_file_get_contents(pid_path
, &name
, &len
, NULL
);
589 void *qemu_alloc_stack(size_t *sz
)
593 #ifdef CONFIG_DEBUG_STACK_USAGE
596 size_t pagesz
= qemu_real_host_page_size();
597 #ifdef _SC_THREAD_STACK_MIN
598 /* avoid stacks smaller than _SC_THREAD_STACK_MIN */
599 long min_stack_sz
= sysconf(_SC_THREAD_STACK_MIN
);
600 *sz
= MAX(MAX(min_stack_sz
, 0), *sz
);
602 /* adjust stack size to a multiple of the page size */
603 *sz
= ROUND_UP(*sz
, pagesz
);
604 /* allocate one extra page for the guard page */
607 flags
= MAP_PRIVATE
| MAP_ANONYMOUS
;
608 #if defined(MAP_STACK) && defined(__OpenBSD__)
609 /* Only enable MAP_STACK on OpenBSD. Other OS's such as
610 * Linux/FreeBSD/NetBSD have a flag with the same name
611 * but have differing functionality. OpenBSD will SEGV
612 * if it spots execution with a stack pointer pointing
613 * at memory that was not allocated with MAP_STACK.
618 ptr
= mmap(NULL
, *sz
, PROT_READ
| PROT_WRITE
, flags
, -1, 0);
619 if (ptr
== MAP_FAILED
) {
620 perror("failed to allocate memory for stack");
624 /* Stack grows down -- guard page at the bottom. */
625 if (mprotect(ptr
, pagesz
, PROT_NONE
) != 0) {
626 perror("failed to set up stack guard page");
630 #ifdef CONFIG_DEBUG_STACK_USAGE
631 for (ptr2
= ptr
+ pagesz
; ptr2
< ptr
+ *sz
; ptr2
+= sizeof(uint32_t)) {
632 *(uint32_t *)ptr2
= 0xdeadbeaf;
639 #ifdef CONFIG_DEBUG_STACK_USAGE
640 static __thread
unsigned int max_stack_usage
;
643 void qemu_free_stack(void *stack
, size_t sz
)
645 #ifdef CONFIG_DEBUG_STACK_USAGE
649 for (ptr
= stack
+ qemu_real_host_page_size(); ptr
< stack
+ sz
;
650 ptr
+= sizeof(uint32_t)) {
651 if (*(uint32_t *)ptr
!= 0xdeadbeaf) {
655 usage
= sz
- (uintptr_t) (ptr
- stack
);
656 if (usage
> max_stack_usage
) {
657 error_report("thread %d max stack usage increased from %u to %u",
658 qemu_get_thread_id(), max_stack_usage
, usage
);
659 max_stack_usage
= usage
;
667 * Disable CFI checks.
668 * We are going to call a signal handler directly. Such handler may or may not
669 * have been defined in our binary, so there's no guarantee that the pointer
670 * used to set the handler is a cfi-valid pointer. Since the handlers are
671 * stored in kernel memory, changing the handler to an attacker-defined
672 * function requires being able to call a sigaction() syscall,
673 * which is not as easy as overwriting a pointer in memory.
676 void sigaction_invoke(struct sigaction
*action
,
677 struct qemu_signalfd_siginfo
*info
)
680 si
.si_signo
= info
->ssi_signo
;
681 si
.si_errno
= info
->ssi_errno
;
682 si
.si_code
= info
->ssi_code
;
684 /* Convert the minimal set of fields defined by POSIX.
685 * Positive si_code values are reserved for kernel-generated
686 * signals, where the valid siginfo fields are determined by
687 * the signal number. But according to POSIX, it is unspecified
688 * whether SI_USER and SI_QUEUE have values less than or equal to
691 if (info
->ssi_code
== SI_USER
|| info
->ssi_code
== SI_QUEUE
||
692 info
->ssi_code
<= 0) {
694 si
.si_pid
= info
->ssi_pid
;
695 si
.si_uid
= info
->ssi_uid
;
696 } else if (info
->ssi_signo
== SIGILL
|| info
->ssi_signo
== SIGFPE
||
697 info
->ssi_signo
== SIGSEGV
|| info
->ssi_signo
== SIGBUS
) {
698 si
.si_addr
= (void *)(uintptr_t)info
->ssi_addr
;
699 } else if (info
->ssi_signo
== SIGCHLD
) {
700 si
.si_pid
= info
->ssi_pid
;
701 si
.si_status
= info
->ssi_status
;
702 si
.si_uid
= info
->ssi_uid
;
704 action
->sa_sigaction(info
->ssi_signo
, &si
, NULL
);
707 size_t qemu_get_host_physmem(void)
709 #ifdef _SC_PHYS_PAGES
710 long pages
= sysconf(_SC_PHYS_PAGES
);
712 if (pages
> SIZE_MAX
/ qemu_real_host_page_size()) {
715 return pages
* qemu_real_host_page_size();
722 int qemu_msync(void *addr
, size_t length
, int fd
)
724 size_t align_mask
= ~(qemu_real_host_page_size() - 1);
727 * There are no strict reqs as per the length of mapping
728 * to be synced. Still the length needs to follow the address
729 * alignment changes. Additionally - round the size to the multiple
732 length
+= ((uintptr_t)addr
& (qemu_real_host_page_size() - 1));
733 length
= (length
+ ~align_mask
) & align_mask
;
735 addr
= (void *)((uintptr_t)addr
& align_mask
);
737 return msync(addr
, length
, MS_SYNC
);