ClickHouse/base/glibc-compatibility/musl/getauxval.c

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

206 lines
5.8 KiB
C
Raw Normal View History

#include "atomic.h"
Fix getauxval() by using procfs with fallback to environ getauxval() from glibc-compatibility did not work always correctly: - It does not work after setenv(), and this breaks vsyscalls, like sched_getcpu() [1] (and BaseDaemon.cpp always set TZ if timezone is defined, which is true for CI [2]). Also note, that fixing setenv() will not fix LSan, since the culprit is getauxval() [1]: https://bugzilla.redhat.com/show_bug.cgi?id=1163404 [2]: ClickHouse#32928 (comment) - Another think that is definitely broken is LSan (Leak Sanitizer), it relies on worked getauxval() but it does not work if __environ is not initialized yet (there is even a commit about this). And because of, at least, one leak had been introduced [3]: [3]: ClickHouse#33840 Fix this by using /proc/self/auxv with fallback to environ solution to make it compatible with environment that does not allow reading from auxv (or no procfs). v2: add fallback to environ solution v3: fix return value for __auxv_init_procfs() (cherry picked from commit f187c3499a9efe05af109aa8011f384f0b14297e) v4: more verbose message on errors, CI founds [1]: AUXV already has value (529267711) [1]: https://s3.amazonaws.com/clickhouse-test-reports/39103/2325f7e8442d1672ce5fb43b11039b6a8937e298/stress_test__memory__actions_.html v5: break at AT_NULL v6: ignore AT_IGNORE v7: suppress TSan and remove superior check to avoid abort() in case of race v8: proper suppressions (not inner function but itself) Refs: #33957 Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
2022-07-17 14:26:00 +00:00
#include <sys/auxv.h>
#include <fcntl.h> // open
#include <sys/stat.h> // O_RDONLY
#include <unistd.h> // read, close
#include <stdlib.h> // ssize_t
#include <stdio.h> // perror, fprintf
#include <link.h> // ElfW
2020-11-14 08:36:44 +00:00
#include <errno.h>
2022-11-01 14:33:52 +00:00
#include "syscall.h"
2022-11-08 16:25:03 +00:00
#if defined(__has_feature)
#if __has_feature(memory_sanitizer)
#include <sanitizer/msan_interface.h>
#endif
#endif
Fix getauxval() by using procfs with fallback to environ getauxval() from glibc-compatibility did not work always correctly: - It does not work after setenv(), and this breaks vsyscalls, like sched_getcpu() [1] (and BaseDaemon.cpp always set TZ if timezone is defined, which is true for CI [2]). Also note, that fixing setenv() will not fix LSan, since the culprit is getauxval() [1]: https://bugzilla.redhat.com/show_bug.cgi?id=1163404 [2]: ClickHouse#32928 (comment) - Another think that is definitely broken is LSan (Leak Sanitizer), it relies on worked getauxval() but it does not work if __environ is not initialized yet (there is even a commit about this). And because of, at least, one leak had been introduced [3]: [3]: ClickHouse#33840 Fix this by using /proc/self/auxv with fallback to environ solution to make it compatible with environment that does not allow reading from auxv (or no procfs). v2: add fallback to environ solution v3: fix return value for __auxv_init_procfs() (cherry picked from commit f187c3499a9efe05af109aa8011f384f0b14297e) v4: more verbose message on errors, CI founds [1]: AUXV already has value (529267711) [1]: https://s3.amazonaws.com/clickhouse-test-reports/39103/2325f7e8442d1672ce5fb43b11039b6a8937e298/stress_test__memory__actions_.html v5: break at AT_NULL v6: ignore AT_IGNORE v7: suppress TSan and remove superior check to avoid abort() in case of race v8: proper suppressions (not inner function but itself) Refs: #33957 Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
2022-07-17 14:26:00 +00:00
#define ARRAY_SIZE(a) sizeof((a))/sizeof((a[0]))
/// Suppress TSan since it is possible for this code to be called from multiple threads,
/// and initialization is safe to be done multiple times from multiple threads.
2024-03-03 23:11:55 +00:00
#define NO_SANITIZE_THREAD __attribute__((__no_sanitize__("thread")))
Fix getauxval() by using procfs with fallback to environ getauxval() from glibc-compatibility did not work always correctly: - It does not work after setenv(), and this breaks vsyscalls, like sched_getcpu() [1] (and BaseDaemon.cpp always set TZ if timezone is defined, which is true for CI [2]). Also note, that fixing setenv() will not fix LSan, since the culprit is getauxval() [1]: https://bugzilla.redhat.com/show_bug.cgi?id=1163404 [2]: ClickHouse#32928 (comment) - Another think that is definitely broken is LSan (Leak Sanitizer), it relies on worked getauxval() but it does not work if __environ is not initialized yet (there is even a commit about this). And because of, at least, one leak had been introduced [3]: [3]: ClickHouse#33840 Fix this by using /proc/self/auxv with fallback to environ solution to make it compatible with environment that does not allow reading from auxv (or no procfs). v2: add fallback to environ solution v3: fix return value for __auxv_init_procfs() (cherry picked from commit f187c3499a9efe05af109aa8011f384f0b14297e) v4: more verbose message on errors, CI founds [1]: AUXV already has value (529267711) [1]: https://s3.amazonaws.com/clickhouse-test-reports/39103/2325f7e8442d1672ce5fb43b11039b6a8937e298/stress_test__memory__actions_.html v5: break at AT_NULL v6: ignore AT_IGNORE v7: suppress TSan and remove superior check to avoid abort() in case of race v8: proper suppressions (not inner function but itself) Refs: #33957 Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
2022-07-17 14:26:00 +00:00
// We don't have libc struct available here.
// Compute aux vector manually (from /proc/self/auxv).
//
// Right now there is only 51 AT_* constants,
// so 64 should be enough until this implementation will be replaced with musl.
static unsigned long __auxv_procfs[64];
static unsigned long __auxv_secure = 0;
Fix getauxval() by using procfs with fallback to environ getauxval() from glibc-compatibility did not work always correctly: - It does not work after setenv(), and this breaks vsyscalls, like sched_getcpu() [1] (and BaseDaemon.cpp always set TZ if timezone is defined, which is true for CI [2]). Also note, that fixing setenv() will not fix LSan, since the culprit is getauxval() [1]: https://bugzilla.redhat.com/show_bug.cgi?id=1163404 [2]: ClickHouse#32928 (comment) - Another think that is definitely broken is LSan (Leak Sanitizer), it relies on worked getauxval() but it does not work if __environ is not initialized yet (there is even a commit about this). And because of, at least, one leak had been introduced [3]: [3]: ClickHouse#33840 Fix this by using /proc/self/auxv with fallback to environ solution to make it compatible with environment that does not allow reading from auxv (or no procfs). v2: add fallback to environ solution v3: fix return value for __auxv_init_procfs() (cherry picked from commit f187c3499a9efe05af109aa8011f384f0b14297e) v4: more verbose message on errors, CI founds [1]: AUXV already has value (529267711) [1]: https://s3.amazonaws.com/clickhouse-test-reports/39103/2325f7e8442d1672ce5fb43b11039b6a8937e298/stress_test__memory__actions_.html v5: break at AT_NULL v6: ignore AT_IGNORE v7: suppress TSan and remove superior check to avoid abort() in case of race v8: proper suppressions (not inner function but itself) Refs: #33957 Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
2022-07-17 14:26:00 +00:00
// Common
static unsigned long * __auxv_environ = NULL;
static void * volatile getauxval_func;
static unsigned long __auxv_init_environ(unsigned long type);
//
// auxv from procfs interface
//
ssize_t __retry_read(int fd, void * buf, size_t count)
{
for (;;)
{
2022-11-01 14:33:52 +00:00
// We cannot use the read syscall as it will be intercept by sanitizers, which aren't
// initialized yet. Emit syscall directly.
ssize_t ret = __syscall_ret(__syscall(SYS_read, fd, buf, count));
Fix getauxval() by using procfs with fallback to environ getauxval() from glibc-compatibility did not work always correctly: - It does not work after setenv(), and this breaks vsyscalls, like sched_getcpu() [1] (and BaseDaemon.cpp always set TZ if timezone is defined, which is true for CI [2]). Also note, that fixing setenv() will not fix LSan, since the culprit is getauxval() [1]: https://bugzilla.redhat.com/show_bug.cgi?id=1163404 [2]: ClickHouse#32928 (comment) - Another think that is definitely broken is LSan (Leak Sanitizer), it relies on worked getauxval() but it does not work if __environ is not initialized yet (there is even a commit about this). And because of, at least, one leak had been introduced [3]: [3]: ClickHouse#33840 Fix this by using /proc/self/auxv with fallback to environ solution to make it compatible with environment that does not allow reading from auxv (or no procfs). v2: add fallback to environ solution v3: fix return value for __auxv_init_procfs() (cherry picked from commit f187c3499a9efe05af109aa8011f384f0b14297e) v4: more verbose message on errors, CI founds [1]: AUXV already has value (529267711) [1]: https://s3.amazonaws.com/clickhouse-test-reports/39103/2325f7e8442d1672ce5fb43b11039b6a8937e298/stress_test__memory__actions_.html v5: break at AT_NULL v6: ignore AT_IGNORE v7: suppress TSan and remove superior check to avoid abort() in case of race v8: proper suppressions (not inner function but itself) Refs: #33957 Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
2022-07-17 14:26:00 +00:00
if (ret == -1)
{
if (errno == EINTR)
{
continue;
}
perror("Cannot read /proc/self/auxv");
abort();
}
return ret;
}
}
unsigned long NO_SANITIZE_THREAD __getauxval_procfs(unsigned long type)
{
if (type == AT_SECURE)
{
return __auxv_secure;
}
if (type >= ARRAY_SIZE(__auxv_procfs))
{
errno = ENOENT;
return 0;
}
return __auxv_procfs[type];
}
static unsigned long NO_SANITIZE_THREAD __auxv_init_procfs(unsigned long type)
{
// For debugging:
// - od -t dL /proc/self/auxv
// - LD_SHOW_AUX= ls
int fd = open("/proc/self/auxv", O_RDONLY);
// It is possible in case of:
// - no procfs mounted
// - on android you are not able to read it unless running from shell or debugging
// - some other issues
if (fd == -1)
{
// Fallback to environ.
a_cas_p(&getauxval_func, (void *)__auxv_init_procfs, (void *)__auxv_init_environ);
return __auxv_init_environ(type);
}
ElfW(auxv_t) aux;
/// NOTE: sizeof(aux) is very small (less then PAGE_SIZE), so partial read should not be possible.
_Static_assert(sizeof(aux) < 4096, "Unexpected sizeof(aux)");
while (__retry_read(fd, &aux, sizeof(aux)) == sizeof(aux))
{
2022-11-04 12:21:50 +00:00
#if defined(__has_feature)
#if __has_feature(memory_sanitizer)
__msan_unpoison(&aux, sizeof(aux));
#endif
#endif
Fix getauxval() by using procfs with fallback to environ getauxval() from glibc-compatibility did not work always correctly: - It does not work after setenv(), and this breaks vsyscalls, like sched_getcpu() [1] (and BaseDaemon.cpp always set TZ if timezone is defined, which is true for CI [2]). Also note, that fixing setenv() will not fix LSan, since the culprit is getauxval() [1]: https://bugzilla.redhat.com/show_bug.cgi?id=1163404 [2]: ClickHouse#32928 (comment) - Another think that is definitely broken is LSan (Leak Sanitizer), it relies on worked getauxval() but it does not work if __environ is not initialized yet (there is even a commit about this). And because of, at least, one leak had been introduced [3]: [3]: ClickHouse#33840 Fix this by using /proc/self/auxv with fallback to environ solution to make it compatible with environment that does not allow reading from auxv (or no procfs). v2: add fallback to environ solution v3: fix return value for __auxv_init_procfs() (cherry picked from commit f187c3499a9efe05af109aa8011f384f0b14297e) v4: more verbose message on errors, CI founds [1]: AUXV already has value (529267711) [1]: https://s3.amazonaws.com/clickhouse-test-reports/39103/2325f7e8442d1672ce5fb43b11039b6a8937e298/stress_test__memory__actions_.html v5: break at AT_NULL v6: ignore AT_IGNORE v7: suppress TSan and remove superior check to avoid abort() in case of race v8: proper suppressions (not inner function but itself) Refs: #33957 Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
2022-07-17 14:26:00 +00:00
if (aux.a_type == AT_NULL)
{
break;
}
if (aux.a_type == AT_IGNORE || aux.a_type == AT_IGNOREPPC)
{
continue;
}
2020-11-14 08:36:44 +00:00
Fix getauxval() by using procfs with fallback to environ getauxval() from glibc-compatibility did not work always correctly: - It does not work after setenv(), and this breaks vsyscalls, like sched_getcpu() [1] (and BaseDaemon.cpp always set TZ if timezone is defined, which is true for CI [2]). Also note, that fixing setenv() will not fix LSan, since the culprit is getauxval() [1]: https://bugzilla.redhat.com/show_bug.cgi?id=1163404 [2]: ClickHouse#32928 (comment) - Another think that is definitely broken is LSan (Leak Sanitizer), it relies on worked getauxval() but it does not work if __environ is not initialized yet (there is even a commit about this). And because of, at least, one leak had been introduced [3]: [3]: ClickHouse#33840 Fix this by using /proc/self/auxv with fallback to environ solution to make it compatible with environment that does not allow reading from auxv (or no procfs). v2: add fallback to environ solution v3: fix return value for __auxv_init_procfs() (cherry picked from commit f187c3499a9efe05af109aa8011f384f0b14297e) v4: more verbose message on errors, CI founds [1]: AUXV already has value (529267711) [1]: https://s3.amazonaws.com/clickhouse-test-reports/39103/2325f7e8442d1672ce5fb43b11039b6a8937e298/stress_test__memory__actions_.html v5: break at AT_NULL v6: ignore AT_IGNORE v7: suppress TSan and remove superior check to avoid abort() in case of race v8: proper suppressions (not inner function but itself) Refs: #33957 Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
2022-07-17 14:26:00 +00:00
if (aux.a_type >= ARRAY_SIZE(__auxv_procfs))
{
fprintf(stderr, "AT_* is out of range: %li (maximum allowed is %zu)\n", aux.a_type, ARRAY_SIZE(__auxv_procfs));
abort();
}
if (__auxv_procfs[aux.a_type])
{
/// It is possible due to race on initialization.
}
__auxv_procfs[aux.a_type] = aux.a_un.a_val;
}
close(fd);
__auxv_secure = __getauxval_procfs(AT_SECURE);
// Now we've initialized __auxv_procfs, next time getauxval() will only call __get_auxval().
a_cas_p(&getauxval_func, (void *)__auxv_init_procfs, (void *)__getauxval_procfs);
return __getauxval_procfs(type);
}
//
// auxv from environ interface
//
// NOTE: environ available only after static initializers,
// so you cannot rely on this if you need getauxval() before.
//
// Good example of such user is sanitizers, for example
// LSan will not work with __auxv_init_environ(),
// since it needs getauxval() before.
//
static size_t NO_SANITIZE_THREAD __find_auxv(unsigned long type)
2022-01-29 22:09:15 +00:00
{
size_t i;
Fix getauxval() by using procfs with fallback to environ getauxval() from glibc-compatibility did not work always correctly: - It does not work after setenv(), and this breaks vsyscalls, like sched_getcpu() [1] (and BaseDaemon.cpp always set TZ if timezone is defined, which is true for CI [2]). Also note, that fixing setenv() will not fix LSan, since the culprit is getauxval() [1]: https://bugzilla.redhat.com/show_bug.cgi?id=1163404 [2]: ClickHouse#32928 (comment) - Another think that is definitely broken is LSan (Leak Sanitizer), it relies on worked getauxval() but it does not work if __environ is not initialized yet (there is even a commit about this). And because of, at least, one leak had been introduced [3]: [3]: ClickHouse#33840 Fix this by using /proc/self/auxv with fallback to environ solution to make it compatible with environment that does not allow reading from auxv (or no procfs). v2: add fallback to environ solution v3: fix return value for __auxv_init_procfs() (cherry picked from commit f187c3499a9efe05af109aa8011f384f0b14297e) v4: more verbose message on errors, CI founds [1]: AUXV already has value (529267711) [1]: https://s3.amazonaws.com/clickhouse-test-reports/39103/2325f7e8442d1672ce5fb43b11039b6a8937e298/stress_test__memory__actions_.html v5: break at AT_NULL v6: ignore AT_IGNORE v7: suppress TSan and remove superior check to avoid abort() in case of race v8: proper suppressions (not inner function but itself) Refs: #33957 Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
2022-07-17 14:26:00 +00:00
for (i = 0; __auxv_environ[i]; i += 2)
2022-01-29 22:09:15 +00:00
{
Fix getauxval() by using procfs with fallback to environ getauxval() from glibc-compatibility did not work always correctly: - It does not work after setenv(), and this breaks vsyscalls, like sched_getcpu() [1] (and BaseDaemon.cpp always set TZ if timezone is defined, which is true for CI [2]). Also note, that fixing setenv() will not fix LSan, since the culprit is getauxval() [1]: https://bugzilla.redhat.com/show_bug.cgi?id=1163404 [2]: ClickHouse#32928 (comment) - Another think that is definitely broken is LSan (Leak Sanitizer), it relies on worked getauxval() but it does not work if __environ is not initialized yet (there is even a commit about this). And because of, at least, one leak had been introduced [3]: [3]: ClickHouse#33840 Fix this by using /proc/self/auxv with fallback to environ solution to make it compatible with environment that does not allow reading from auxv (or no procfs). v2: add fallback to environ solution v3: fix return value for __auxv_init_procfs() (cherry picked from commit f187c3499a9efe05af109aa8011f384f0b14297e) v4: more verbose message on errors, CI founds [1]: AUXV already has value (529267711) [1]: https://s3.amazonaws.com/clickhouse-test-reports/39103/2325f7e8442d1672ce5fb43b11039b6a8937e298/stress_test__memory__actions_.html v5: break at AT_NULL v6: ignore AT_IGNORE v7: suppress TSan and remove superior check to avoid abort() in case of race v8: proper suppressions (not inner function but itself) Refs: #33957 Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
2022-07-17 14:26:00 +00:00
if (__auxv_environ[i] == type)
{
2022-01-29 22:09:15 +00:00
return i + 1;
Fix getauxval() by using procfs with fallback to environ getauxval() from glibc-compatibility did not work always correctly: - It does not work after setenv(), and this breaks vsyscalls, like sched_getcpu() [1] (and BaseDaemon.cpp always set TZ if timezone is defined, which is true for CI [2]). Also note, that fixing setenv() will not fix LSan, since the culprit is getauxval() [1]: https://bugzilla.redhat.com/show_bug.cgi?id=1163404 [2]: ClickHouse#32928 (comment) - Another think that is definitely broken is LSan (Leak Sanitizer), it relies on worked getauxval() but it does not work if __environ is not initialized yet (there is even a commit about this). And because of, at least, one leak had been introduced [3]: [3]: ClickHouse#33840 Fix this by using /proc/self/auxv with fallback to environ solution to make it compatible with environment that does not allow reading from auxv (or no procfs). v2: add fallback to environ solution v3: fix return value for __auxv_init_procfs() (cherry picked from commit f187c3499a9efe05af109aa8011f384f0b14297e) v4: more verbose message on errors, CI founds [1]: AUXV already has value (529267711) [1]: https://s3.amazonaws.com/clickhouse-test-reports/39103/2325f7e8442d1672ce5fb43b11039b6a8937e298/stress_test__memory__actions_.html v5: break at AT_NULL v6: ignore AT_IGNORE v7: suppress TSan and remove superior check to avoid abort() in case of race v8: proper suppressions (not inner function but itself) Refs: #33957 Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
2022-07-17 14:26:00 +00:00
}
2022-01-29 22:09:15 +00:00
}
return (size_t) -1;
}
Fix getauxval() by using procfs with fallback to environ getauxval() from glibc-compatibility did not work always correctly: - It does not work after setenv(), and this breaks vsyscalls, like sched_getcpu() [1] (and BaseDaemon.cpp always set TZ if timezone is defined, which is true for CI [2]). Also note, that fixing setenv() will not fix LSan, since the culprit is getauxval() [1]: https://bugzilla.redhat.com/show_bug.cgi?id=1163404 [2]: ClickHouse#32928 (comment) - Another think that is definitely broken is LSan (Leak Sanitizer), it relies on worked getauxval() but it does not work if __environ is not initialized yet (there is even a commit about this). And because of, at least, one leak had been introduced [3]: [3]: ClickHouse#33840 Fix this by using /proc/self/auxv with fallback to environ solution to make it compatible with environment that does not allow reading from auxv (or no procfs). v2: add fallback to environ solution v3: fix return value for __auxv_init_procfs() (cherry picked from commit f187c3499a9efe05af109aa8011f384f0b14297e) v4: more verbose message on errors, CI founds [1]: AUXV already has value (529267711) [1]: https://s3.amazonaws.com/clickhouse-test-reports/39103/2325f7e8442d1672ce5fb43b11039b6a8937e298/stress_test__memory__actions_.html v5: break at AT_NULL v6: ignore AT_IGNORE v7: suppress TSan and remove superior check to avoid abort() in case of race v8: proper suppressions (not inner function but itself) Refs: #33957 Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
2022-07-17 14:26:00 +00:00
unsigned long NO_SANITIZE_THREAD __getauxval_environ(unsigned long type)
2020-11-14 08:36:44 +00:00
{
if (type == AT_SECURE)
return __auxv_secure;
Fix getauxval() by using procfs with fallback to environ getauxval() from glibc-compatibility did not work always correctly: - It does not work after setenv(), and this breaks vsyscalls, like sched_getcpu() [1] (and BaseDaemon.cpp always set TZ if timezone is defined, which is true for CI [2]). Also note, that fixing setenv() will not fix LSan, since the culprit is getauxval() [1]: https://bugzilla.redhat.com/show_bug.cgi?id=1163404 [2]: ClickHouse#32928 (comment) - Another think that is definitely broken is LSan (Leak Sanitizer), it relies on worked getauxval() but it does not work if __environ is not initialized yet (there is even a commit about this). And because of, at least, one leak had been introduced [3]: [3]: ClickHouse#33840 Fix this by using /proc/self/auxv with fallback to environ solution to make it compatible with environment that does not allow reading from auxv (or no procfs). v2: add fallback to environ solution v3: fix return value for __auxv_init_procfs() (cherry picked from commit f187c3499a9efe05af109aa8011f384f0b14297e) v4: more verbose message on errors, CI founds [1]: AUXV already has value (529267711) [1]: https://s3.amazonaws.com/clickhouse-test-reports/39103/2325f7e8442d1672ce5fb43b11039b6a8937e298/stress_test__memory__actions_.html v5: break at AT_NULL v6: ignore AT_IGNORE v7: suppress TSan and remove superior check to avoid abort() in case of race v8: proper suppressions (not inner function but itself) Refs: #33957 Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
2022-07-17 14:26:00 +00:00
if (__auxv_environ)
2020-11-14 08:36:44 +00:00
{
2022-01-29 22:09:15 +00:00
size_t index = __find_auxv(type);
if (index != ((size_t) -1))
Fix getauxval() by using procfs with fallback to environ getauxval() from glibc-compatibility did not work always correctly: - It does not work after setenv(), and this breaks vsyscalls, like sched_getcpu() [1] (and BaseDaemon.cpp always set TZ if timezone is defined, which is true for CI [2]). Also note, that fixing setenv() will not fix LSan, since the culprit is getauxval() [1]: https://bugzilla.redhat.com/show_bug.cgi?id=1163404 [2]: ClickHouse#32928 (comment) - Another think that is definitely broken is LSan (Leak Sanitizer), it relies on worked getauxval() but it does not work if __environ is not initialized yet (there is even a commit about this). And because of, at least, one leak had been introduced [3]: [3]: ClickHouse#33840 Fix this by using /proc/self/auxv with fallback to environ solution to make it compatible with environment that does not allow reading from auxv (or no procfs). v2: add fallback to environ solution v3: fix return value for __auxv_init_procfs() (cherry picked from commit f187c3499a9efe05af109aa8011f384f0b14297e) v4: more verbose message on errors, CI founds [1]: AUXV already has value (529267711) [1]: https://s3.amazonaws.com/clickhouse-test-reports/39103/2325f7e8442d1672ce5fb43b11039b6a8937e298/stress_test__memory__actions_.html v5: break at AT_NULL v6: ignore AT_IGNORE v7: suppress TSan and remove superior check to avoid abort() in case of race v8: proper suppressions (not inner function but itself) Refs: #33957 Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
2022-07-17 14:26:00 +00:00
return __auxv_environ[index];
2020-11-14 08:36:44 +00:00
}
2022-01-29 22:09:15 +00:00
errno = ENOENT;
return 0;
2020-11-14 08:36:44 +00:00
}
Fix getauxval() by using procfs with fallback to environ getauxval() from glibc-compatibility did not work always correctly: - It does not work after setenv(), and this breaks vsyscalls, like sched_getcpu() [1] (and BaseDaemon.cpp always set TZ if timezone is defined, which is true for CI [2]). Also note, that fixing setenv() will not fix LSan, since the culprit is getauxval() [1]: https://bugzilla.redhat.com/show_bug.cgi?id=1163404 [2]: ClickHouse#32928 (comment) - Another think that is definitely broken is LSan (Leak Sanitizer), it relies on worked getauxval() but it does not work if __environ is not initialized yet (there is even a commit about this). And because of, at least, one leak had been introduced [3]: [3]: ClickHouse#33840 Fix this by using /proc/self/auxv with fallback to environ solution to make it compatible with environment that does not allow reading from auxv (or no procfs). v2: add fallback to environ solution v3: fix return value for __auxv_init_procfs() (cherry picked from commit f187c3499a9efe05af109aa8011f384f0b14297e) v4: more verbose message on errors, CI founds [1]: AUXV already has value (529267711) [1]: https://s3.amazonaws.com/clickhouse-test-reports/39103/2325f7e8442d1672ce5fb43b11039b6a8937e298/stress_test__memory__actions_.html v5: break at AT_NULL v6: ignore AT_IGNORE v7: suppress TSan and remove superior check to avoid abort() in case of race v8: proper suppressions (not inner function but itself) Refs: #33957 Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
2022-07-17 14:26:00 +00:00
static unsigned long NO_SANITIZE_THREAD __auxv_init_environ(unsigned long type)
{
2022-01-29 22:09:15 +00:00
if (!__environ)
{
Fix getauxval() by using procfs with fallback to environ getauxval() from glibc-compatibility did not work always correctly: - It does not work after setenv(), and this breaks vsyscalls, like sched_getcpu() [1] (and BaseDaemon.cpp always set TZ if timezone is defined, which is true for CI [2]). Also note, that fixing setenv() will not fix LSan, since the culprit is getauxval() [1]: https://bugzilla.redhat.com/show_bug.cgi?id=1163404 [2]: ClickHouse#32928 (comment) - Another think that is definitely broken is LSan (Leak Sanitizer), it relies on worked getauxval() but it does not work if __environ is not initialized yet (there is even a commit about this). And because of, at least, one leak had been introduced [3]: [3]: ClickHouse#33840 Fix this by using /proc/self/auxv with fallback to environ solution to make it compatible with environment that does not allow reading from auxv (or no procfs). v2: add fallback to environ solution v3: fix return value for __auxv_init_procfs() (cherry picked from commit f187c3499a9efe05af109aa8011f384f0b14297e) v4: more verbose message on errors, CI founds [1]: AUXV already has value (529267711) [1]: https://s3.amazonaws.com/clickhouse-test-reports/39103/2325f7e8442d1672ce5fb43b11039b6a8937e298/stress_test__memory__actions_.html v5: break at AT_NULL v6: ignore AT_IGNORE v7: suppress TSan and remove superior check to avoid abort() in case of race v8: proper suppressions (not inner function but itself) Refs: #33957 Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
2022-07-17 14:26:00 +00:00
// __environ is not initialized yet so we can't initialize __auxv_environ right now.
2022-01-29 22:09:15 +00:00
// That's normally occurred only when getauxval() is called from some sanitizer's internal code.
errno = ENOENT;
return 0;
}
Fix getauxval() by using procfs with fallback to environ getauxval() from glibc-compatibility did not work always correctly: - It does not work after setenv(), and this breaks vsyscalls, like sched_getcpu() [1] (and BaseDaemon.cpp always set TZ if timezone is defined, which is true for CI [2]). Also note, that fixing setenv() will not fix LSan, since the culprit is getauxval() [1]: https://bugzilla.redhat.com/show_bug.cgi?id=1163404 [2]: ClickHouse#32928 (comment) - Another think that is definitely broken is LSan (Leak Sanitizer), it relies on worked getauxval() but it does not work if __environ is not initialized yet (there is even a commit about this). And because of, at least, one leak had been introduced [3]: [3]: ClickHouse#33840 Fix this by using /proc/self/auxv with fallback to environ solution to make it compatible with environment that does not allow reading from auxv (or no procfs). v2: add fallback to environ solution v3: fix return value for __auxv_init_procfs() (cherry picked from commit f187c3499a9efe05af109aa8011f384f0b14297e) v4: more verbose message on errors, CI founds [1]: AUXV already has value (529267711) [1]: https://s3.amazonaws.com/clickhouse-test-reports/39103/2325f7e8442d1672ce5fb43b11039b6a8937e298/stress_test__memory__actions_.html v5: break at AT_NULL v6: ignore AT_IGNORE v7: suppress TSan and remove superior check to avoid abort() in case of race v8: proper suppressions (not inner function but itself) Refs: #33957 Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
2022-07-17 14:26:00 +00:00
// Initialize __auxv_environ and __auxv_secure.
2022-01-29 22:09:15 +00:00
size_t i;
for (i = 0; __environ[i]; i++);
Fix getauxval() by using procfs with fallback to environ getauxval() from glibc-compatibility did not work always correctly: - It does not work after setenv(), and this breaks vsyscalls, like sched_getcpu() [1] (and BaseDaemon.cpp always set TZ if timezone is defined, which is true for CI [2]). Also note, that fixing setenv() will not fix LSan, since the culprit is getauxval() [1]: https://bugzilla.redhat.com/show_bug.cgi?id=1163404 [2]: ClickHouse#32928 (comment) - Another think that is definitely broken is LSan (Leak Sanitizer), it relies on worked getauxval() but it does not work if __environ is not initialized yet (there is even a commit about this). And because of, at least, one leak had been introduced [3]: [3]: ClickHouse#33840 Fix this by using /proc/self/auxv with fallback to environ solution to make it compatible with environment that does not allow reading from auxv (or no procfs). v2: add fallback to environ solution v3: fix return value for __auxv_init_procfs() (cherry picked from commit f187c3499a9efe05af109aa8011f384f0b14297e) v4: more verbose message on errors, CI founds [1]: AUXV already has value (529267711) [1]: https://s3.amazonaws.com/clickhouse-test-reports/39103/2325f7e8442d1672ce5fb43b11039b6a8937e298/stress_test__memory__actions_.html v5: break at AT_NULL v6: ignore AT_IGNORE v7: suppress TSan and remove superior check to avoid abort() in case of race v8: proper suppressions (not inner function but itself) Refs: #33957 Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
2022-07-17 14:26:00 +00:00
__auxv_environ = (unsigned long *) (__environ + i + 1);
2022-01-29 22:09:15 +00:00
size_t secure_idx = __find_auxv(AT_SECURE);
if (secure_idx != ((size_t) -1))
Fix getauxval() by using procfs with fallback to environ getauxval() from glibc-compatibility did not work always correctly: - It does not work after setenv(), and this breaks vsyscalls, like sched_getcpu() [1] (and BaseDaemon.cpp always set TZ if timezone is defined, which is true for CI [2]). Also note, that fixing setenv() will not fix LSan, since the culprit is getauxval() [1]: https://bugzilla.redhat.com/show_bug.cgi?id=1163404 [2]: ClickHouse#32928 (comment) - Another think that is definitely broken is LSan (Leak Sanitizer), it relies on worked getauxval() but it does not work if __environ is not initialized yet (there is even a commit about this). And because of, at least, one leak had been introduced [3]: [3]: ClickHouse#33840 Fix this by using /proc/self/auxv with fallback to environ solution to make it compatible with environment that does not allow reading from auxv (or no procfs). v2: add fallback to environ solution v3: fix return value for __auxv_init_procfs() (cherry picked from commit f187c3499a9efe05af109aa8011f384f0b14297e) v4: more verbose message on errors, CI founds [1]: AUXV already has value (529267711) [1]: https://s3.amazonaws.com/clickhouse-test-reports/39103/2325f7e8442d1672ce5fb43b11039b6a8937e298/stress_test__memory__actions_.html v5: break at AT_NULL v6: ignore AT_IGNORE v7: suppress TSan and remove superior check to avoid abort() in case of race v8: proper suppressions (not inner function but itself) Refs: #33957 Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
2022-07-17 14:26:00 +00:00
__auxv_secure = __auxv_environ[secure_idx];
Fix getauxval() by using procfs with fallback to environ getauxval() from glibc-compatibility did not work always correctly: - It does not work after setenv(), and this breaks vsyscalls, like sched_getcpu() [1] (and BaseDaemon.cpp always set TZ if timezone is defined, which is true for CI [2]). Also note, that fixing setenv() will not fix LSan, since the culprit is getauxval() [1]: https://bugzilla.redhat.com/show_bug.cgi?id=1163404 [2]: ClickHouse#32928 (comment) - Another think that is definitely broken is LSan (Leak Sanitizer), it relies on worked getauxval() but it does not work if __environ is not initialized yet (there is even a commit about this). And because of, at least, one leak had been introduced [3]: [3]: ClickHouse#33840 Fix this by using /proc/self/auxv with fallback to environ solution to make it compatible with environment that does not allow reading from auxv (or no procfs). v2: add fallback to environ solution v3: fix return value for __auxv_init_procfs() (cherry picked from commit f187c3499a9efe05af109aa8011f384f0b14297e) v4: more verbose message on errors, CI founds [1]: AUXV already has value (529267711) [1]: https://s3.amazonaws.com/clickhouse-test-reports/39103/2325f7e8442d1672ce5fb43b11039b6a8937e298/stress_test__memory__actions_.html v5: break at AT_NULL v6: ignore AT_IGNORE v7: suppress TSan and remove superior check to avoid abort() in case of race v8: proper suppressions (not inner function but itself) Refs: #33957 Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
2022-07-17 14:26:00 +00:00
// Now we need to switch to __getauxval_environ for all later calls, since
// everything is initialized.
a_cas_p(&getauxval_func, (void *)__auxv_init_environ, (void *)__getauxval_environ);
Fix getauxval() by using procfs with fallback to environ getauxval() from glibc-compatibility did not work always correctly: - It does not work after setenv(), and this breaks vsyscalls, like sched_getcpu() [1] (and BaseDaemon.cpp always set TZ if timezone is defined, which is true for CI [2]). Also note, that fixing setenv() will not fix LSan, since the culprit is getauxval() [1]: https://bugzilla.redhat.com/show_bug.cgi?id=1163404 [2]: ClickHouse#32928 (comment) - Another think that is definitely broken is LSan (Leak Sanitizer), it relies on worked getauxval() but it does not work if __environ is not initialized yet (there is even a commit about this). And because of, at least, one leak had been introduced [3]: [3]: ClickHouse#33840 Fix this by using /proc/self/auxv with fallback to environ solution to make it compatible with environment that does not allow reading from auxv (or no procfs). v2: add fallback to environ solution v3: fix return value for __auxv_init_procfs() (cherry picked from commit f187c3499a9efe05af109aa8011f384f0b14297e) v4: more verbose message on errors, CI founds [1]: AUXV already has value (529267711) [1]: https://s3.amazonaws.com/clickhouse-test-reports/39103/2325f7e8442d1672ce5fb43b11039b6a8937e298/stress_test__memory__actions_.html v5: break at AT_NULL v6: ignore AT_IGNORE v7: suppress TSan and remove superior check to avoid abort() in case of race v8: proper suppressions (not inner function but itself) Refs: #33957 Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
2022-07-17 14:26:00 +00:00
return __getauxval_environ(type);
}
Fix getauxval() by using procfs with fallback to environ getauxval() from glibc-compatibility did not work always correctly: - It does not work after setenv(), and this breaks vsyscalls, like sched_getcpu() [1] (and BaseDaemon.cpp always set TZ if timezone is defined, which is true for CI [2]). Also note, that fixing setenv() will not fix LSan, since the culprit is getauxval() [1]: https://bugzilla.redhat.com/show_bug.cgi?id=1163404 [2]: ClickHouse#32928 (comment) - Another think that is definitely broken is LSan (Leak Sanitizer), it relies on worked getauxval() but it does not work if __environ is not initialized yet (there is even a commit about this). And because of, at least, one leak had been introduced [3]: [3]: ClickHouse#33840 Fix this by using /proc/self/auxv with fallback to environ solution to make it compatible with environment that does not allow reading from auxv (or no procfs). v2: add fallback to environ solution v3: fix return value for __auxv_init_procfs() (cherry picked from commit f187c3499a9efe05af109aa8011f384f0b14297e) v4: more verbose message on errors, CI founds [1]: AUXV already has value (529267711) [1]: https://s3.amazonaws.com/clickhouse-test-reports/39103/2325f7e8442d1672ce5fb43b11039b6a8937e298/stress_test__memory__actions_.html v5: break at AT_NULL v6: ignore AT_IGNORE v7: suppress TSan and remove superior check to avoid abort() in case of race v8: proper suppressions (not inner function but itself) Refs: #33957 Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
2022-07-17 14:26:00 +00:00
// Callchain:
// - __auxv_init_procfs -> __getauxval_environ
// - __auxv_init_procfs -> __auxv_init_environ -> __getauxval_environ
static void * volatile getauxval_func = (void *)__auxv_init_procfs;
unsigned long getauxval(unsigned long type)
{
return ((unsigned long (*)(unsigned long))getauxval_func)(type);
}