/*
* Copyright (C) 2018, The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include "statsd_writer.h"
#include <cutils/fs.h>
#include <cutils/sockets.h>
#include <cutils/threads.h>
#include <errno.h>
#include <fcntl.h>
#include <inttypes.h>
#include <poll.h>
#include <private/android_filesystem_config.h>
#include <private/android_logger.h>
#include <stdarg.h>
#include <stdatomic.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <sys/uio.h>
#include <sys/un.h>
#include <time.h>
#include <unistd.h>
/* branchless on many architectures. */
#define min(x, y) ((y) ^ (((x) ^ (y)) & -((x) < (y))))
#ifndef htole32
#if __BYTE_ORDER == __LITTLE_ENDIAN
#define htole32(x) (x)
#else
#define htole32(x) __bswap_32(x)
#endif
#endif
#ifndef htole64
#if __BYTE_ORDER == __LITTLE_ENDIAN
#define htole64(x) (x)
#else
#define htole64(x) __bswap_64(x)
#endif
#endif
static pthread_mutex_t log_init_lock = PTHREAD_MUTEX_INITIALIZER;
static atomic_int dropped = 0;
static atomic_int log_error = 0;
static atomic_int atom_tag = 0;
void statsd_writer_init_lock() {
/*
* If we trigger a signal handler in the middle of locked activity and the
* signal handler logs a message, we could get into a deadlock state.
*/
pthread_mutex_lock(&log_init_lock);
}
int statd_writer_trylock() {
return pthread_mutex_trylock(&log_init_lock);
}
void statsd_writer_init_unlock() {
pthread_mutex_unlock(&log_init_lock);
}
static int statsdAvailable();
static int statsdOpen();
static void statsdClose();
static int statsdWrite(struct timespec* ts, struct iovec* vec, size_t nr);
static void statsdNoteDrop();
struct android_log_transport_write statsdLoggerWrite = {
.name = "statsd",
.sock = -EBADF,
.available = statsdAvailable,
.open = statsdOpen,
.close = statsdClose,
.write = statsdWrite,
.noteDrop = statsdNoteDrop,
};
/* log_init_lock assumed */
static int statsdOpen() {
int i, ret = 0;
i = atomic_load(&statsdLoggerWrite.sock);
if (i < 0) {
int flags = SOCK_DGRAM;
#ifdef SOCK_CLOEXEC
flags |= SOCK_CLOEXEC;
#endif
#ifdef SOCK_NONBLOCK
flags |= SOCK_NONBLOCK;
#endif
int sock = TEMP_FAILURE_RETRY(socket(PF_UNIX, flags, 0));
if (sock < 0) {
ret = -errno;
} else {
int sndbuf = 1 * 1024 * 1024; // set max send buffer size 1MB
socklen_t bufLen = sizeof(sndbuf);
// SO_RCVBUF does not have an effect on unix domain socket, but SO_SNDBUF does.
// Proceed to connect even setsockopt fails.
setsockopt(sock, SOL_SOCKET, SO_SNDBUF, &sndbuf, bufLen);
struct sockaddr_un un;
memset(&un, 0, sizeof(struct sockaddr_un));
un.sun_family = AF_UNIX;
strcpy(un.sun_path, "/dev/socket/statsdw");
if (TEMP_FAILURE_RETRY(
connect(sock, (struct sockaddr*)&un, sizeof(struct sockaddr_un))) < 0) {
ret = -errno;
switch (ret) {
case -ENOTCONN:
case -ECONNREFUSED:
case -ENOENT:
i = atomic_exchange(&statsdLoggerWrite.sock, ret);
/* FALLTHRU */
default:
break;
}
close(sock);
} else {
ret = atomic_exchange(&statsdLoggerWrite.sock, sock);
if ((ret >= 0) && (ret != sock)) {
close(ret);
}
ret = 0;
}
}
}
return ret;
}
static void __statsdClose(int negative_errno) {
int sock = atomic_exchange(&statsdLoggerWrite.sock, negative_errno);
if (sock >= 0) {
close(sock);
}
}
static void statsdClose() {
__statsdClose(-EBADF);
}
static int statsdAvailable() {
if (atomic_load(&statsdLoggerWrite.sock) < 0) {
if (access("/dev/socket/statsdw", W_OK) == 0) {
return 0;
}
return -EBADF;
}
return 1;
}
static void statsdNoteDrop(int error, int tag) {
atomic_fetch_add_explicit(&dropped, 1, memory_order_relaxed);
atomic_exchange_explicit(&log_error, error, memory_order_relaxed);
atomic_exchange_explicit(&atom_tag, tag, memory_order_relaxed);
}
static int statsdWrite(struct timespec* ts, struct iovec* vec, size_t nr) {
ssize_t ret;
int sock;
static const unsigned headerLength = 1;
struct iovec newVec[nr + headerLength];
android_log_header_t header;
size_t i, payloadSize;
sock = atomic_load(&statsdLoggerWrite.sock);
if (sock < 0) switch (sock) {
case -ENOTCONN:
case -ECONNREFUSED:
case -ENOENT:
break;
default:
return -EBADF;
}
/*
* struct {
* // what we provide to socket
* android_log_header_t header;
* // caller provides
* union {
* struct {
* char prio;
* char payload[];
* } string;
* struct {
* uint32_t tag
* char payload[];
* } binary;
* };
* };
*/
header.tid = gettid();
header.realtime.tv_sec = ts->tv_sec;
header.realtime.tv_nsec = ts->tv_nsec;
newVec[0].iov_base = (unsigned char*)&header;
newVec[0].iov_len = sizeof(header);
// If we dropped events before, try to tell statsd.
if (sock >= 0) {
int32_t snapshot = atomic_exchange_explicit(&dropped, 0, memory_order_relaxed);
if (snapshot) {
android_log_event_long_t buffer;
header.id = LOG_ID_STATS;
// store the last log error in the tag field. This tag field is not used by statsd.
buffer.header.tag = htole32(atomic_load(&log_error));
buffer.payload.type = EVENT_TYPE_LONG;
// format:
// |atom_tag|dropped_count|
int64_t composed_long = atomic_load(&atom_tag);
// Send 2 int32's via an int64.
composed_long = ((composed_long << 32) | ((int64_t)snapshot));
buffer.payload.data = htole64(composed_long);
newVec[headerLength].iov_base = &buffer;
newVec[headerLength].iov_len = sizeof(buffer);
ret = TEMP_FAILURE_RETRY(writev(sock, newVec, 2));
if (ret != (ssize_t)(sizeof(header) + sizeof(buffer))) {
atomic_fetch_add_explicit(&dropped, snapshot, memory_order_relaxed);
}
}
}
header.id = LOG_ID_STATS;
for (payloadSize = 0, i = headerLength; i < nr + headerLength; i++) {
newVec[i].iov_base = vec[i - headerLength].iov_base;
payloadSize += newVec[i].iov_len = vec[i - headerLength].iov_len;
if (payloadSize > LOGGER_ENTRY_MAX_PAYLOAD) {
newVec[i].iov_len -= payloadSize - LOGGER_ENTRY_MAX_PAYLOAD;
if (newVec[i].iov_len) {
++i;
}
break;
}
}
/*
* The write below could be lost, but will never block.
*
* ENOTCONN occurs if statsd has died.
* ENOENT occurs if statsd is not running and socket is missing.
* ECONNREFUSED occurs if we can not reconnect to statsd.
* EAGAIN occurs if statsd is overloaded.
*/
if (sock < 0) {
ret = sock;
} else {
ret = TEMP_FAILURE_RETRY(writev(sock, newVec, i));
if (ret < 0) {
ret = -errno;
}
}
switch (ret) {
case -ENOTCONN:
case -ECONNREFUSED:
case -ENOENT:
if (statd_writer_trylock()) {
return ret; /* in a signal handler? try again when less stressed
*/
}
__statsdClose(ret);
ret = statsdOpen();
statsd_writer_init_unlock();
if (ret < 0) {
return ret;
}
ret = TEMP_FAILURE_RETRY(writev(atomic_load(&statsdLoggerWrite.sock), newVec, i));
if (ret < 0) {
ret = -errno;
}
/* FALLTHRU */
default:
break;
}
if (ret > (ssize_t)sizeof(header)) {
ret -= sizeof(header);
}
return ret;
}