2 * OS includes and handling of OS dependencies
4 * This header exists to pull in some common system headers that
5 * most code in QEMU will want, and to fix up some possible issues with
6 * it (missing defines, Windows weirdness, and so on).
8 * To avoid getting into possible circular include dependencies, this
9 * file should not include any other QEMU headers, with the exceptions
10 * of config-host.h, config-target.h, qemu/compiler.h,
11 * sysemu/os-posix.h, sysemu/os-win32.h, glib-compat.h and
12 * qemu/typedefs.h, all of which are doing a similar job to this file
13 * and are under similar constraints.
15 * This header also contains prototypes for functions defined in
16 * os-*.c and util/oslib-*.c; those would probably be better split
17 * out into separate header files.
19 * In an ideal world this header would contain only:
20 * (1) things which everybody needs
21 * (2) things without which code would work on most platforms but
22 * fail to compile or misbehave on a minority of host OSes
24 * This work is licensed under the terms of the GNU GPL, version 2 or later.
25 * See the COPYING file in the top-level directory.
30 #include "config-host.h"
32 #include CONFIG_TARGET
34 #include "exec/poison.h"
37 #include "qemu/compiler.h"
39 /* Older versions of C++ don't get definitions of various macros from
40 * stdlib.h unless we define these macros before first inclusion of
43 #ifndef __STDC_CONSTANT_MACROS
44 #define __STDC_CONSTANT_MACROS
46 #ifndef __STDC_LIMIT_MACROS
47 #define __STDC_LIMIT_MACROS
49 #ifndef __STDC_FORMAT_MACROS
50 #define __STDC_FORMAT_MACROS
53 /* The following block of code temporarily renames the daemon() function so the
54 * compiler does not see the warning associated with it in stdlib.h on OSX
57 #define daemon qemu_fake_daemon_function
60 QEMU_EXTERN_C int daemon(int, int);
64 /* as defined in sdkddkver.h */
66 #define _WIN32_WINNT 0x0600 /* Vista */
68 /* reduces the number of implicitly included headers */
69 #ifndef WIN32_LEAN_AND_MEAN
70 #define WIN32_LEAN_AND_MEAN
74 /* enable C99/POSIX format strings (needs mingw32-runtime 3.15 or later) */
76 #define __USE_MINGW_ANSI_STDIO 1
83 #include <sys/types.h>
91 /* Put unistd.h before time.h as that triggers localtime_r/gmtime_r
92 * function availability on recentish Mingw-w64 platforms. */
100 #include <sys/time.h>
102 /* setjmp must be declared before sysemu/os-win32.h
103 * because it is redefined there. */
111 #if defined(__linux__) && defined(__sparc__)
112 /* The SPARC definition of QEMU_VMALLOC_ALIGN needs SHMLBA */
117 #include <sys/wait.h>
119 #define WIFEXITED(x) 1
120 #define WEXITSTATUS(x) (x)
124 #include <AvailabilityMacros.h>
128 * This is somewhat like a system header; it must be outside any extern "C"
129 * block because it includes system headers itself, including glib.h,
130 * which will not compile if inside an extern "C" block.
132 #include "glib-compat.h"
135 #include "sysemu/os-win32.h"
139 #include "sysemu/os-posix.h"
146 #include "qemu/typedefs.h"
149 * For mingw, as of v6.0.0, the function implementing the assert macro is
150 * not marked as noreturn, so the compiler cannot delete code following an
151 * assert(false) as unused. We rely on this within the code base to delete
152 * code that is unreachable when features are disabled.
153 * All supported versions of Glib's g_assert() satisfy this requirement.
157 #define assert(x) g_assert(x)
161 * According to waitpid man page:
163 * This macro is not specified in POSIX.1-2001 and is not
164 * available on some UNIX implementations (e.g., AIX, SunOS).
165 * Therefore, enclose its use inside #ifdef WCOREDUMP ... #endif.
168 #define WCOREDUMP(status) 0
171 * We have a lot of unaudited code that may fail in strange ways, or
172 * even be a security risk during migration, if you disable assertions
173 * at compile-time. You may comment out these safety checks if you
174 * absolutely want to disable assertion overhead, but it is not
175 * supported upstream so the risk is all yours. Meanwhile, please
176 * submit patches to remove any side-effects inside an assertion, or
177 * fixing error handling that should use Error instead of assert.
180 #error building with NDEBUG is not supported
182 #ifdef G_DISABLE_ASSERT
183 #error building with G_DISABLE_ASSERT is not supported
187 #define O_LARGEFILE 0
192 #ifndef MAP_ANONYMOUS
193 #define MAP_ANONYMOUS MAP_ANON
195 #ifndef MAP_FIXED_NOREPLACE
196 #define MAP_FIXED_NOREPLACE 0
198 #ifndef MAP_NORESERVE
199 #define MAP_NORESERVE 0
202 #define ENOMEDIUM ENODEV
204 #if !defined(ENOTSUP)
207 #if !defined(ECANCELED)
208 #define ECANCELED 4097
210 #if !defined(EMEDIUMTYPE)
211 #define EMEDIUMTYPE 4098
213 #if !defined(ESHUTDOWN)
214 #define ESHUTDOWN 4099
217 /* time_t may be either 32 or 64 bits depending on the host OS, and
218 * can be either signed or unsigned, so we can't just hardcode a
219 * specific maximum value. This is not a C preprocessor constant,
220 * so you can't use TIME_MAX in an #ifdef, but for our purposes
221 * this isn't a problem.
224 /* The macros TYPE_SIGNED, TYPE_WIDTH, and TYPE_MAXIMUM are from
225 * Gnulib, and are under the LGPL v2.1 or (at your option) any
229 /* True if the real type T is signed. */
230 #define TYPE_SIGNED(t) (!((t)0 < (t)-1))
232 /* The width in bits of the integer type or expression T.
233 * Padding bits are not supported.
235 #define TYPE_WIDTH(t) (sizeof(t) * CHAR_BIT)
237 /* The maximum and minimum values for the integer type T. */
238 #define TYPE_MAXIMUM(t) \
239 ((t) (!TYPE_SIGNED(t) \
241 : ((((t)1 << (TYPE_WIDTH(t) - 2)) - 1) * 2 + 1)))
244 #define TIME_MAX TYPE_MAXIMUM(time_t)
247 /* HOST_LONG_BITS is the size of a native pointer in bits. */
248 #if UINTPTR_MAX == UINT32_MAX
249 # define HOST_LONG_BITS 32
250 #elif UINTPTR_MAX == UINT64_MAX
251 # define HOST_LONG_BITS 64
253 # error Unknown pointer size
256 /* Mac OSX has a <stdint.h> bug that incorrectly defines SIZE_MAX with
257 * the wrong type. Our replacement isn't usable in preprocessor
258 * expressions, but it is sufficient for our needs. */
259 #ifdef HAVE_BROKEN_SIZE_MAX
261 #define SIZE_MAX ((size_t)-1)
265 * Two variations of MIN/MAX macros. The first is for runtime use, and
266 * evaluates arguments only once (so it is safe even with side
267 * effects), but will not work in constant contexts (such as array
268 * size declarations) because of the '{}'. The second is for constant
269 * expression use, where evaluating arguments twice is safe because
270 * the result is going to be constant anyway, but will not work in a
271 * runtime context because of a void expression where a value is
272 * expected. Thus, both gcc and clang will fail to compile if you use
273 * the wrong macro (even if the error may seem a bit cryptic).
275 * Note that neither form is usable as an #if condition; if you truly
276 * need to write conditional code that depends on a minimum or maximum
277 * determined by the pre-processor instead of the compiler, you'll
278 * have to open-code it. Sadly, Coverity is severely confused by the
279 * constant variants, so we have to dumb things down there.
284 typeof(1 ? (a) : (b)) _a = (a), _b = (b); \
290 typeof(1 ? (a) : (b)) _a = (a), _b = (b); \
295 # define MIN_CONST(a, b) ((a) < (b) ? (a) : (b))
296 # define MAX_CONST(a, b) ((a) > (b) ? (a) : (b))
298 # define MIN_CONST(a, b) \
299 __builtin_choose_expr( \
300 __builtin_constant_p(a) && __builtin_constant_p(b), \
301 (a) < (b) ? (a) : (b), \
303 # define MAX_CONST(a, b) \
304 __builtin_choose_expr( \
305 __builtin_constant_p(a) && __builtin_constant_p(b), \
306 (a) > (b) ? (a) : (b), \
311 * Minimum function that returns zero only if both values are zero.
312 * Intended for use with unsigned values only.
315 #define MIN_NON_ZERO(a, b) \
317 typeof(1 ? (a) : (b)) _a = (a), _b = (b); \
318 _a == 0 ? _b : (_b == 0 || _b > _a) ? _a : _b; \
323 * Round number down to multiple. Safe when m is not a power of 2 (see
324 * ROUND_DOWN for a faster version when a power of 2 is guaranteed).
326 #define QEMU_ALIGN_DOWN(n, m) ((n) / (m) * (m))
329 * Round number up to multiple. Safe when m is not a power of 2 (see
330 * ROUND_UP for a faster version when a power of 2 is guaranteed).
332 #define QEMU_ALIGN_UP(n, m) QEMU_ALIGN_DOWN((n) + (m) - 1, (m))
334 /* Check if n is a multiple of m */
335 #define QEMU_IS_ALIGNED(n, m) (((n) % (m)) == 0)
337 /* n-byte align pointer down */
338 #define QEMU_ALIGN_PTR_DOWN(p, n) \
339 ((typeof(p))QEMU_ALIGN_DOWN((uintptr_t)(p), (n)))
341 /* n-byte align pointer up */
342 #define QEMU_ALIGN_PTR_UP(p, n) \
343 ((typeof(p))QEMU_ALIGN_UP((uintptr_t)(p), (n)))
345 /* Check if pointer p is n-bytes aligned */
346 #define QEMU_PTR_IS_ALIGNED(p, n) QEMU_IS_ALIGNED((uintptr_t)(p), (n))
349 * Round number down to multiple. Requires that d be a power of 2 (see
350 * QEMU_ALIGN_UP for a safer but slower version on arbitrary
351 * numbers); works even if d is a smaller type than n.
354 #define ROUND_DOWN(n, d) ((n) & -(0 ? (n) : (d)))
358 * Round number up to multiple. Requires that d be a power of 2 (see
359 * QEMU_ALIGN_UP for a safer but slower version on arbitrary
360 * numbers); works even if d is a smaller type than n.
363 #define ROUND_UP(n, d) ROUND_DOWN((n) + (d) - 1, (d))
367 #define DIV_ROUND_UP(n, d) (((n) + (d) - 1) / (d))
371 * &(x)[0] is always a pointer - if it's same type as x then the argument is a
372 * pointer, not an array.
374 #define QEMU_IS_ARRAY(x) (!__builtin_types_compatible_p(typeof(x), \
377 #define ARRAY_SIZE(x) ((sizeof(x) / sizeof((x)[0])) + \
378 QEMU_BUILD_BUG_ON_ZERO(!QEMU_IS_ARRAY(x)))
381 int qemu_daemon(int nochdir, int noclose);
382 void *qemu_try_memalign(size_t alignment, size_t size);
383 void *qemu_memalign(size_t alignment, size_t size);
384 void *qemu_anon_ram_alloc(size_t size, uint64_t *align, bool shared,
386 void qemu_vfree(void *ptr);
387 void qemu_anon_ram_free(void *ptr, size_t size);
390 * It's an analog of GLIB's g_autoptr_cleanup_generic_gfree(), used to define
393 static inline void qemu_cleanup_generic_vfree(void *p)
395 void **pp = (void **)p;
400 * Analog of g_autofree, but qemu_vfree is called on cleanup instead of g_free.
402 #define QEMU_AUTO_VFREE __attribute__((cleanup(qemu_cleanup_generic_vfree)))
405 #define HAVE_CHARDEV_SERIAL 1
406 #elif defined(__linux__) || defined(__sun__) || defined(__FreeBSD__) \
407 || defined(__NetBSD__) || defined(__OpenBSD__) || defined(__DragonFly__) \
408 || defined(__GLIBC__) || defined(__APPLE__)
409 #define HAVE_CHARDEV_SERIAL 1
412 #if defined(__linux__) || defined(__FreeBSD__) || \
413 defined(__FreeBSD_kernel__) || defined(__DragonFly__)
414 #define HAVE_CHARDEV_PARPORT 1
417 #if defined(__HAIKU__)
418 #define SIGIO SIGPOLL
421 #if defined(CONFIG_LINUX)
422 #ifndef BUS_MCEERR_AR
423 #define BUS_MCEERR_AR 4
425 #ifndef BUS_MCEERR_AO
426 #define BUS_MCEERR_AO 5
430 #if defined(__linux__) && \
431 (defined(__x86_64__) || defined(__arm__) || defined(__aarch64__) \
432 || defined(__powerpc64__))
433 /* Use 2 MiB alignment so transparent hugepages can be used by KVM.
434 Valgrind does not support alignments larger than 1 MiB,
435 therefore we need special code which handles running on Valgrind. */
436 # define QEMU_VMALLOC_ALIGN (512 * 4096)
437 #elif defined(__linux__) && defined(__s390x__)
438 /* Use 1 MiB (segment size) alignment so gmap can be used by KVM. */
439 # define QEMU_VMALLOC_ALIGN (256 * 4096)
440 #elif defined(__linux__) && defined(__sparc__)
441 # define QEMU_VMALLOC_ALIGN MAX(qemu_real_host_page_size, SHMLBA)
443 # define QEMU_VMALLOC_ALIGN qemu_real_host_page_size
447 struct qemu_signalfd_siginfo {
448 uint32_t ssi_signo; /* Signal number */
449 int32_t ssi_errno; /* Error number (unused) */
450 int32_t ssi_code; /* Signal code */
451 uint32_t ssi_pid; /* PID of sender */
452 uint32_t ssi_uid; /* Real UID of sender */
453 int32_t ssi_fd; /* File descriptor (SIGIO) */
454 uint32_t ssi_tid; /* Kernel timer ID (POSIX timers) */
455 uint32_t ssi_band; /* Band event (SIGIO) */
456 uint32_t ssi_overrun; /* POSIX timer overrun count */
457 uint32_t ssi_trapno; /* Trap number that caused signal */
458 int32_t ssi_status; /* Exit status or signal (SIGCHLD) */
459 int32_t ssi_int; /* Integer sent by sigqueue(2) */
460 uint64_t ssi_ptr; /* Pointer sent by sigqueue(2) */
461 uint64_t ssi_utime; /* User CPU time consumed (SIGCHLD) */
462 uint64_t ssi_stime; /* System CPU time consumed (SIGCHLD) */
463 uint64_t ssi_addr; /* Address that generated signal
464 (for hardware-generated signals) */
465 uint8_t pad[48]; /* Pad size to 128 bytes (allow for
466 additional fields in the future) */
469 int qemu_signalfd(const sigset_t *mask);
470 void sigaction_invoke(struct sigaction *action,
471 struct qemu_signalfd_siginfo *info);
475 * Don't introduce new usage of this function, prefer the following
476 * qemu_open/qemu_create that take an "Error **errp"
478 int qemu_open_old(const char *name, int flags, ...);
479 int qemu_open(const char *name, int flags, Error **errp);
480 int qemu_create(const char *name, int flags, mode_t mode, Error **errp);
481 int qemu_close(int fd);
482 int qemu_unlink(const char *name);
484 int qemu_dup_flags(int fd, int flags);
485 int qemu_dup(int fd);
486 int qemu_lock_fd(int fd, int64_t start, int64_t len, bool exclusive);
487 int qemu_unlock_fd(int fd, int64_t start, int64_t len);
488 int qemu_lock_fd_test(int fd, int64_t start, int64_t len, bool exclusive);
489 bool qemu_has_ofd_lock(void);
492 #if defined(__HAIKU__) && defined(__i386__)
493 #define FMT_pid "%ld"
495 #define FMT_pid "%" PRId64
500 bool qemu_write_pidfile(const char *pidfile, Error **errp);
502 int qemu_get_thread_id(void);
510 * Use the same value as Linux for now.
514 ssize_t readv(int fd, const struct iovec *iov, int iov_cnt);
515 ssize_t writev(int fd, const struct iovec *iov, int iov_cnt);
519 static inline void qemu_timersub(const struct timeval *val1,
520 const struct timeval *val2,
523 res->tv_sec = val1->tv_sec - val2->tv_sec;
524 if (val1->tv_usec < val2->tv_usec) {
526 res->tv_usec = val1->tv_usec - val2->tv_usec + 1000 * 1000;
528 res->tv_usec = val1->tv_usec - val2->tv_usec;
532 #define qemu_timersub timersub
535 void qemu_set_cloexec(int fd);
537 /* Starting on QEMU 2.5, qemu_hw_version() returns "2.5+" by default
538 * instead of QEMU_VERSION, so setting hw_version on MachineClass
539 * is no longer mandatory.
541 * Do NOT change this string, or it will break compatibility on all
542 * machine classes that don't set hw_version.
544 #define QEMU_HW_VERSION "2.5+"
546 /* QEMU "hardware version" setting. Used to replace code that exposed
547 * QEMU_VERSION to guests in the past and need to keep compatibility.
548 * Do not use qemu_hw_version() in new code.
550 void qemu_set_hw_version(const char *);
551 const char *qemu_hw_version(void);
553 void fips_set_state(bool requested);
554 bool fips_get_state(void);
556 /* Return a dynamically allocated pathname denoting a file or directory that is
557 * appropriate for storing local state.
559 * @relative_pathname need not start with a directory separator; one will be
560 * added automatically.
562 * The caller is responsible for releasing the value returned with g_free()
565 char *qemu_get_local_state_pathname(const char *relative_pathname);
567 /* Find program directory, and save it for later usage with
568 * qemu_get_exec_dir().
569 * Try OS specific API first, if not working, parse from argv0. */
570 void qemu_init_exec_dir(const char *argv0);
572 /* Get the saved exec dir. */
573 const char *qemu_get_exec_dir(void);
577 * @type: the auxiliary vector key to lookup
579 * Search the auxiliary vector for @type, returning the value
580 * or 0 if @type is not present.
582 unsigned long qemu_getauxval(unsigned long type);
584 void qemu_set_tty_echo(int fd, bool echo);
586 void os_mem_prealloc(int fd, char *area, size_t sz, int smp_cpus,
591 * @pid: pid of a process
593 * For given @pid fetch its name. Caller is responsible for
594 * freeing the string when no longer needed.
595 * Returns allocated string on success, NULL on failure.
597 char *qemu_get_pid_name(pid_t pid);
602 * A version of fork that avoids signal handler race
603 * conditions that can lead to child process getting
604 * signals that are otherwise only expected by the
605 * parent. It also resets all signal handlers to the
608 * Returns 0 to child process, pid number to parent
611 pid_t qemu_fork(Error **errp);
613 /* Using intptr_t ensures that qemu_*_page_mask is sign-extended even
614 * when intptr_t is 32-bit and we are aligning a long long.
616 extern uintptr_t qemu_real_host_page_size;
617 extern intptr_t qemu_real_host_page_mask;
620 * After using getopt or getopt_long, if you need to parse another set
621 * of options, then you must reset optind. Unfortunately the way to
622 * do this varies between implementations of getopt.
624 static inline void qemu_reset_optind(void)
635 * qemu_get_host_name:
636 * @errp: Error object
638 * Operating system agnostic way of querying host name.
640 * Returns allocated hostname (caller should free), NULL on failure.
642 char *qemu_get_host_name(Error **errp);
645 * qemu_get_host_physmem:
647 * Operating system agnostic way of querying host memory.
649 * Returns amount of physical memory on the system. This is purely
650 * advisery and may return 0 if we can't work it out. At the other
651 * end we saturate to SIZE_MAX if you are lucky enough to have that
654 size_t qemu_get_host_physmem(void);
657 * Toggle write/execute on the pages marked MAP_JIT
658 * for the current thread.
660 #if defined(MAC_OS_VERSION_11_0) && \
661 MAC_OS_X_VERSION_MAX_ALLOWED >= MAC_OS_VERSION_11_0
662 static inline void qemu_thread_jit_execute(void)
664 if (__builtin_available(macOS 11.0, *)) {
665 pthread_jit_write_protect_np(true);
669 static inline void qemu_thread_jit_write(void)
671 if (__builtin_available(macOS 11.0, *)) {
672 pthread_jit_write_protect_np(false);
676 static inline void qemu_thread_jit_write(void) {}
677 static inline void qemu_thread_jit_execute(void) {}
681 * Platforms which do not support system() return ENOSYS
683 #ifndef HAVE_SYSTEM_FUNCTION
684 #define system platform_does_not_support_system
685 static inline int platform_does_not_support_system(const char *command)
690 #endif /* !HAVE_SYSTEM_FUNCTION */
693 * Duplicate directory entry @dent.
695 * It is highly recommended to use this function instead of open coding
696 * duplication of @c dirent objects, because the actual @c struct @c dirent
697 * size may be bigger or shorter than @c sizeof(struct dirent) and correct
698 * handling is platform specific (see gitlab issue #841).
700 * @dent - original directory entry to be duplicated
701 * @returns duplicated directory entry which should be freed with g_free()
703 struct dirent *qemu_dirent_dup(struct dirent *dent);