acrn-hypervisor/tools/acrn-crashlog/acrnprobe/android_events.c
Liu, Xinwu 41b39c5e1f tools: acrn-crashlog: Defer the vm events processing when failed
In the original design, acrnprobe marked all handled VMs'events as "synced"
in file vmrecordid(this patch changes the name to VM_eventsID.log).
Currently, the Android log events are not logged if the first attempt at
reading collecting them from the VM fails. This patch changes the logic
so that the acrn-crashlog tool will retry continuously.

This patch defines different tags for handled VMs'events, and only marks
VMs'events "synced" after it returns successfully.

Signed-off-by: Liu, Xinwu <xinwu.liu@intel.com>
Reviewed-by: xiaojin2 <xiaojing.liu@intel.com>
Reviewed-by: Jin Zhi <zhi.jin@intel.com>
Acked-by: Geoffroy Van Cutsem <geoffroy.vancutsem@intel.com>
Acked-by: Chen gang <gang.c.chen@intel.com>
2018-06-29 15:23:18 +08:00

679 lines
14 KiB
C

/*
* Copyright (C) 2018 Intel Corporation
* SPDX-License-Identifier: BSD-3-Clause
*/
#include <stdio.h>
#include <unistd.h>
#include <string.h>
#include <malloc.h>
#include <openssl/sha.h>
#include <signal.h>
#include <limits.h>
#include <stdlib.h>
#include "android_events.h"
#include "strutils.h"
#include "cmdutils.h"
#include "log_sys.h"
#include "fsutils.h"
#include "history.h"
#define VM_WARNING_LINES 2000
#define LOOP_DEV_MAX 8
#define ANDROID_EVT_KEY_LEN 20
static const char *android_img = "/data/android/android.img";
char *loop_dev;
/* Find the head of str, caller must guarantee that 'str' is not in
* the first line.
*/
static char *line_head(const char *str)
{
while (*str != '\n')
str--;
return (char *)(str + 1);
}
/* Find the next event that needs to be synced.
* There is a history_event file in UOS side, it records UOS's events in
* real-time. Generally, the cursor point to the first unsynchronized line.
*/
static char *vm_next_event_to_sync(const char *cursor, const struct vm_t *vm)
{
char *line_to_sync = (char *)~(0);
char *syncevent;
char *p;
char *type;
char *subtype;
char *target;
int ret;
int id;
if (!cursor || !vm)
return NULL;
/* find all syncing types start from cursor,
* focus the event with smaller address.
*/
for_each_syncevent_vm(id, syncevent, vm) {
if (!syncevent)
continue;
ret = asprintf(&type, "\n%s ", syncevent);
if (ret < 0) {
LOGE("calculate vm event failed, out of memory\n");
return NULL;
}
/* a sync event may configured as type/subtype */
p = strchr(type, '/');
if (p) {
*p = 0;
subtype = p + 1;
} else {
subtype = NULL;
}
if (subtype)
p = strstr(cursor, subtype);
else
p = strstr(cursor, type);
if (p) {
target = line_head(p);
line_to_sync = MIN(line_to_sync, target);
}
free(type);
}
if (line_to_sync == (char *)~(0))
return NULL;
return line_to_sync;
}
#define VMRECORD_HEAD_LINES 6
#define VMRECORD_TAG_LEN 9
#define VMRECORD_TAG_WAITING_SYNC " <=="
#define VMRECORD_TAG_NOT_FOUND "NOT_FOUND"
#define VMRECORD_TAG_SUCCESS " "
static int generate_log_vmrecord(const char *path)
{
const char * const head =
"/* DONT EDIT!\n"
" * This file records VM id synced or about to be synched,\n"
" * the tag \"<==\" indicates event waiting to sync.\n"
" * the tag \"NOT_FOUND\" indicates event not found in UOS.\n"
" */\n\n";
LOGD("Generate (%s)\n", path);
return overwrite_file(path, head);
}
enum stage1_refresh_type_t {
MM_ONLY,
MM_FILE
};
/**
* There are 2 stages in vm events sync.
* Stage1: record events to log_vmrecordid file.
* Stage2: call sender's callback for each recorded events.
*
* The design reason is to give UOS some time to log to storage.
*/
static int refresh_key_synced_stage1(const struct sender_t *sender,
struct vm_t *vm,
const char *key,
enum stage1_refresh_type_t type)
{
char log_new[64];
char *log_vmrecordid;
int sid;
if (!key || !sender || !vm)
return -1;
sid = sender_id(sender);
if (sid == -1)
return -1;
log_vmrecordid = sender->log_vmrecordid;
/* the length of key must be 20, and its value can not be
* 00000000000000000000.
*/
if ((strlen(key) == ANDROID_EVT_KEY_LEN) &&
strcmp(key, "00000000000000000000")) {
sprintf(vm->last_synced_line_key[sid],
"%s", key);
if (type == MM_ONLY)
return 0;
/* create a log file, so we can locate
* the right place in case of reboot
*/
if (!file_exists(log_vmrecordid))
generate_log_vmrecord(log_vmrecordid);
sprintf(log_new, "%s %s %s\n", vm->name, key,
VMRECORD_TAG_WAITING_SYNC);
append_file(log_vmrecordid, log_new);
return 0;
}
LOGE("try to record an invalid key (%s) for (%s)\n",
key, vm->name);
return -1;
}
enum stage2_refresh_type_t {
SUCCESS,
NOT_FOUND
};
static int refresh_key_synced_stage2(const struct mm_file_t *m_vm_records,
const char *key,
enum stage2_refresh_type_t type)
{
char *lhead, *ltail;
char *tag;
if (!key || strlen(key) != ANDROID_EVT_KEY_LEN || !m_vm_records ||
m_vm_records->size <= 0)
return -1;
lhead = strstr(m_vm_records->begin, key);
if (!lhead)
return -1;
ltail = strchr(lhead, '\n');
if (!ltail)
return -1;
tag = strstr(lhead, VMRECORD_TAG_WAITING_SYNC);
if (!tag || tag >= ltail)
return -1;
/* re-mark symbol "<==" for synced key */
if (type == SUCCESS)
memcpy(tag, VMRECORD_TAG_SUCCESS, VMRECORD_TAG_LEN);
else if (type == NOT_FOUND)
memcpy(tag, VMRECORD_TAG_NOT_FOUND, VMRECORD_TAG_LEN);
else
return -1;
return 0;
}
static int get_vm_history(struct vm_t *vm, const struct sender_t *sender,
void **data)
{
char vm_history[PATH_MAX];
unsigned long size;
int ret;
int sid;
if (!vm || !sender)
return -1;
sid = sender_id(sender);
if (sid == -1)
return -1;
snprintf(vm_history, sizeof(vm_history), "/tmp/%s_%s",
"vm_hist", vm->name);
if (get_file_size(vm_history) == (int)vm->history_size[sid])
return 0;
if (*data)
free(*data);
ret = read_full_binary_file(vm_history, &size, data);
if (ret) {
LOGE("read (%s) with error (%s)\n", vm_history,
strerror(errno));
*data = NULL;
return ret;
}
ret = strcnt(*data, '\n');
if (ret > VM_WARNING_LINES) {
LOGW("File too large, (%d) lines in (%s)\n",
ret, vm_history);
}
vm->history_size[sid] = size;
return size;
}
static void sync_lines_stage1(const struct sender_t *sender, const void *data[])
{
int id, sid;
int ret;
struct vm_t *vm;
char *start;
char *line_to_sync;
char vmkey[ANDROID_WORD_LEN];
const char * const vm_format =
IGN_ONEWORD ANDROID_KEY_FMT IGN_RESTS;
sid = sender_id(sender);
if (sid == -1)
return;
for_each_vm(id, vm, conf) {
if (!vm)
continue;
if (!vm->online)
continue;
if (vm->last_synced_line_key[sid][0]) {
start = strstr(data[id],
vm->last_synced_line_key[sid]);
if (start == NULL) {
LOGW("no synced id (%s), sync from head\n",
vm->last_synced_line_key[sid]);
start = (char *)data[id];
} else {
start = next_line(start);
}
} else {
start = (char *)data[id];
}
while ((line_to_sync = vm_next_event_to_sync(start, vm))) {
/* It's possible that log's content isn't ready
* at this moment, so we postpone the fn until
* the next loop
*/
//fn(line_to_sync, vm);
vmkey[0] = 0;
ret = sscanf(line_to_sync, vm_format, vmkey);
if (ret != 1) {
LOGE("get an invalid line from (%s), skip\n",
vm->name);
start = next_line(line_to_sync);
continue;
}
LOGD("stage1 %s\n", vmkey);
if (vmkey[0])
refresh_key_synced_stage1(sender, vm,
vmkey, MM_FILE);
start = next_line(line_to_sync);
}
}
}
static void sync_lines_stage2(const struct sender_t *sender, const void *data[],
int (*fn)(const char*, const struct vm_t *))
{
struct mm_file_t *m_vm_records;
char *line;
char *cursor;
const char * const record_fmt =
VM_NAME_FMT ANDROID_KEY_FMT IGN_RESTS;
char vm_name[32];
char vmkey[ANDROID_WORD_LEN];
int id;
struct vm_t *vm;
int ret;
m_vm_records = mmap_file(sender->log_vmrecordid);
if (!m_vm_records) {
LOGE("mmap %s failed, strerror(%s)\n", sender->log_vmrecordid,
strerror(errno));
return;
}
if (!m_vm_records->size ||
mm_count_lines(m_vm_records) < VMRECORD_HEAD_LINES) {
LOGE("(%s) invalid\n", sender->log_vmrecordid);
goto out;
}
cursor = strstr(m_vm_records->begin, " " VMRECORD_TAG_WAITING_SYNC);
if (!cursor)
goto out;
line = line_head(cursor);
while (line) {
char *vm_hist_line;
vmkey[0] = 0;
/* VMNAME xxxxxxxxxxxxxxxxxxxx <== */
ret = sscanf(line, record_fmt, vm_name, vmkey);
if (ret != 2) {
LOGE("parse vm record failed\n");
goto out;
}
for_each_vm(id, vm, conf) {
if (!vm)
continue;
if (!vm->online)
continue;
if (strcmp(vm->name, vm_name))
continue;
vm_hist_line = strstr(data[id], vmkey);
if (!vm_hist_line) {
LOGE("mark vmevent(%s) as unfound,", vmkey);
LOGE("history_event in UOS was recreated?\n");
refresh_key_synced_stage2(m_vm_records, vmkey,
NOT_FOUND);
break;
}
ret = fn(line_head(vm_hist_line), vm);
if (!ret)
refresh_key_synced_stage2(m_vm_records, vmkey,
SUCCESS);
}
cursor = next_line(line);
if (!cursor)
break;
line = strstr(cursor, VMRECORD_TAG_WAITING_SYNC);
if (!line)
break;
line = line_head(line);
}
out:
unmap_file(m_vm_records);
}
/* This function only for initialization */
static void get_last_line_synced(const struct sender_t *sender)
{
int id;
int sid;
int ret;
struct vm_t *vm;
char vmkey[ANDROID_WORD_LEN];
char vm_name[32];
if (!sender)
return;
sid = sender_id(sender);
if (sid == -1)
return;
for_each_vm(id, vm, conf) {
if (!vm)
continue;
if (!vm->online)
continue;
/* generally only exec for each vm once */
if (vm->last_synced_line_key[sid][0])
continue;
snprintf(vm_name, sizeof(vm_name), "%s ", vm->name);
ret = file_read_key_value_r(sender->log_vmrecordid, vm_name,
sizeof(vmkey), vmkey);
if (ret == -ENOENT) {
LOGD("no (%s), will generate\n",
sender->log_vmrecordid);
generate_log_vmrecord(sender->log_vmrecordid);
continue;
} else if (ret == -ENOMSG) {
LOGD("no vm record id for (%s)\n", vm->name);
continue;
} else if (ret < 0) {
LOGE("read key-value in (%s) for (%s), error (%s)\n",
sender->log_vmrecordid, vm->name,
strerror(errno));
continue;
}
ret = refresh_key_synced_stage1(sender, vm, vmkey, MM_ONLY);
if (ret < 0) {
LOGE("get a non-key vm event (%s) for (%s)\n",
vmkey, vm->name);
continue;
}
}
}
static int is_data_partition(char *loop_dev)
{
int ret = 0;
char *out = exec_out2mem("debugfs -R \"ls\" %s", loop_dev);
if (!out) {
LOGE("debugfs -R ls %s failed, error (%s)\n", loop_dev,
strerror(errno));
return 0;
}
if (strstr(out, "app-lib") &&
strstr(out, "tombstones") &&
strstr(out, "dalvik-cache")) {
ret = 1;
}
free(out);
return ret;
}
static char *setup_loop_dev(void)
{
/* Currently UOS image(/data/android/android.img) mounted by
* launch_UOS.sh, we need mount its data partition to loop device
*/
char *out;
char *end;
char loop_dev_tmp[32];
int i;
if (!file_exists(android_img)) {
LOGW("img(%s) is not available\n", android_img);
return NULL;
}
loop_dev = exec_out2mem("losetup -f");
if (!loop_dev) {
/* get cmd out fail */
LOGE("losetup -f failed, error (%s)\n",
strerror(errno));
return NULL;
} else if (strncmp(loop_dev, "/dev/loop", strlen("/dev/loop"))) {
/* it's not a loop dev */
LOGE("get error loop dev (%s)\n", loop_dev);
goto free_loop_dev;
} else if (strncmp(loop_dev, "/dev/loop0", strlen("/dev/loop0"))) {
/* it's not loop0, was img mounted? */
for (i = 0; i < LOOP_DEV_MAX; i++) {
snprintf(loop_dev_tmp, ARRAY_SIZE(loop_dev_tmp),
"/dev/loop%d", i);
if (is_data_partition(loop_dev_tmp)) {
free(loop_dev);
loop_dev = strdup(loop_dev_tmp);
if (!loop_dev) {
LOGE("out of memory\n");
return NULL;
}
return loop_dev;
}
}
}
end = strchr(loop_dev, '\n');
if (end)
*end = 0;
out = exec_out2mem("fdisk -lu %s", android_img);
if (!out) {
LOGE("fdisk -lu %s failed, error (%s)\n", android_img,
strerror(errno));
goto free_loop_dev;
}
/* find data partition, sector unit = 512 bytes */
char partition_start[32];
char sectors[32];
unsigned long pstart;
char *partition;
const char * const partition_fmt =
IGN_ONEWORD "%31[^ ]" IGN_SPACES
IGN_ONEWORD "%31[^ ]" IGN_RESTS;
char *cursor = out;
int ret;
while (cursor &&
(partition = strstr(cursor, android_img))) {
cursor = strchr(partition, '\n');
ret = sscanf(partition, partition_fmt,
partition_start, sectors);
if (ret != 2)
continue;
LOGD("start (%s) sectors(%s)\n", partition_start, sectors);
/* size < 1G */
if (atoi(sectors) < 1 * 2 * 1024 * 1024)
continue;
pstart = atol(partition_start) * 512;
if (pstart == 0)
continue;
ret = exec_out2file(NULL, "losetup -o %lu %s %s",
pstart, loop_dev, android_img);
/* if error occurs, is_data_partition will return false,
* only print error message here.
*/
if (ret != 0)
LOGE("(losetup -o %lu %s %s) failed, return %d\n",
pstart, loop_dev, android_img, ret);
if (is_data_partition(loop_dev)) {
goto success;
} else {
ret = exec_out2file(NULL, "losetup -d %s", loop_dev);
/* may lose a loop dev */
if (ret != 0)
LOGE("(losetup -d %s) failed, return %d\n",
loop_dev, ret);
sleep(1);
}
}
free(out);
free_loop_dev:
free(loop_dev);
return NULL;
success:
free(out);
return loop_dev;
}
static int ping_vm_fs(char *loop_dev)
{
int id;
int res;
int count = 0;
struct vm_t *vm;
struct mm_file_t *vm_hist;
char vm_history[PATH_MAX];
char cmd[512];
const char prefix[] = "#V1.0 CURRENTUPTIME";
/* ensure history_event in uos available */
for_each_vm(id, vm, conf) {
if (!vm)
continue;
snprintf(vm_history, sizeof(vm_history), "/tmp/%s_%s",
"vm_hist", vm->name);
snprintf(cmd, sizeof(cmd),
"dump logs/history_event %s", vm_history);
res = remove(vm_history);
if (res == -1 && errno != ENOENT)
LOGE("remove %s failed\n", vm_history);
res = debugfs_cmd(loop_dev, cmd, NULL);
if (res) {
vm->online = 0;
continue;
}
vm_hist = mmap_file(vm_history);
if (vm_hist == NULL) {
vm->online = 0;
LOGE("%s(%s) unavailable\n", vm->name, vm_history);
continue;
}
if (vm_hist->size &&
!strncmp(vm_hist->begin, prefix, strlen(prefix))) {
vm->online = 1;
count++;
} else {
vm->online = 0;
LOGE("%s(%s) unavailable\n", vm->name, vm_history);
}
unmap_file(vm_hist);
}
return count;
}
/* This function searches all android vms' new events and call the fn for
* each event.
*
* Note that: fn should return 0 to indicate event has been handled,
* or fn will be called in a time loop until it returns 0.
*/
void refresh_vm_history(const struct sender_t *sender,
int (*fn)(const char*, const struct vm_t *))
{
int ret;
int id;
struct vm_t *vm;
static void *data[VM_MAX];
if (!loop_dev) {
loop_dev = setup_loop_dev();
if (!loop_dev)
return;
LOGI("setup loop dev successful\n");
}
if (sender_id(sender) == 0) {
ret = ping_vm_fs(loop_dev);
if (!ret)
return;
}
get_last_line_synced(sender);
for_each_vm(id, vm, conf) {
if (!vm)
continue;
if (!vm->online)
continue;
get_vm_history(vm, sender, &data[id]);
}
sync_lines_stage2(sender, (const void **)data, fn);
sync_lines_stage1(sender, (const void **)data);
}