Adding mechanism to all the backends to switch the logging formats at runtime while leveraging the function pointer table based upon the Kconfigs which also cleans up the logging backend design. Also demonstrate the working API with the changes to syst sample. Clean up Kconfig for backends and add a standard template to generate the Kconfigs for logging formats to be used by all backends. Signed-off-by: Aastha Grover <aastha.grover@intel.com>
496 lines
9.4 KiB
C
496 lines
9.4 KiB
C
/*
|
|
* Copyright (c) 2020 Nordic Semiconductor ASA
|
|
*
|
|
* SPDX-License-Identifier: Apache-2.0
|
|
*/
|
|
|
|
#include <stdio.h>
|
|
#include <stdlib.h>
|
|
#include <logging/log_backend.h>
|
|
#include <logging/log_output_dict.h>
|
|
#include <logging/log_backend_std.h>
|
|
#include <assert.h>
|
|
#include <fs/fs.h>
|
|
|
|
#define MAX_PATH_LEN 256
|
|
#define MAX_FLASH_WRITE_SIZE 256
|
|
#define LOG_PREFIX_LEN (sizeof(CONFIG_LOG_BACKEND_FS_FILE_PREFIX) - 1)
|
|
#define MAX_FILE_NUMERAL 9999
|
|
#define FILE_NUMERAL_LEN 4
|
|
|
|
enum backend_fs_state {
|
|
BACKEND_FS_NOT_INITIALIZED = 0,
|
|
BACKEND_FS_CORRUPTED,
|
|
BACKEND_FS_OK
|
|
};
|
|
|
|
static struct fs_file_t file;
|
|
static enum backend_fs_state backend_state = BACKEND_FS_NOT_INITIALIZED;
|
|
static int file_ctr, newest, oldest;
|
|
|
|
static int allocate_new_file(struct fs_file_t *file);
|
|
static int del_oldest_log(void);
|
|
static int get_log_file_id(struct fs_dirent *ent);
|
|
#ifndef CONFIG_LOG_BACKEND_FS_TESTSUITE
|
|
static uint32_t log_format_current = CONFIG_LOG_BACKEND_FS_OUTPUT_DEFAULT;
|
|
#endif
|
|
|
|
static int check_log_volumen_available(void)
|
|
{
|
|
int index = 0;
|
|
char const *name;
|
|
int rc = 0;
|
|
|
|
while (rc == 0) {
|
|
rc = fs_readmount(&index, &name);
|
|
if (rc == 0) {
|
|
if (strncmp(CONFIG_LOG_BACKEND_FS_DIR,
|
|
name,
|
|
strlen(name))
|
|
== 0) {
|
|
return 0;
|
|
}
|
|
}
|
|
}
|
|
|
|
return -ENOENT;
|
|
}
|
|
|
|
static int create_log_dir(const char *path)
|
|
{
|
|
const char *next;
|
|
const char *last = path + (strlen(path) - 1);
|
|
char w_path[MAX_PATH_LEN];
|
|
int rc, len;
|
|
struct fs_dir_t dir;
|
|
|
|
fs_dir_t_init(&dir);
|
|
|
|
/* the fist directory name is the mount point*/
|
|
/* the firs path's letter might be meaningless `/`, let's skip it */
|
|
next = strchr(path + 1, '/');
|
|
if (!next) {
|
|
return 0;
|
|
}
|
|
|
|
while (true) {
|
|
next++;
|
|
if (next > last) {
|
|
return 0;
|
|
}
|
|
next = strchr(next, '/');
|
|
if (!next) {
|
|
next = last;
|
|
len = last - path + 1;
|
|
} else {
|
|
len = next - path;
|
|
}
|
|
|
|
memcpy(w_path, path, len);
|
|
w_path[len] = 0;
|
|
|
|
rc = fs_opendir(&dir, w_path);
|
|
if (rc) {
|
|
/* assume directory doesn't exist */
|
|
rc = fs_mkdir(w_path);
|
|
if (rc) {
|
|
break;
|
|
}
|
|
}
|
|
rc = fs_closedir(&dir);
|
|
if (rc) {
|
|
break;
|
|
}
|
|
}
|
|
|
|
return rc;
|
|
|
|
}
|
|
|
|
static int check_log_file_exist(int num)
|
|
{
|
|
struct fs_dir_t dir;
|
|
struct fs_dirent ent;
|
|
int rc;
|
|
|
|
fs_dir_t_init(&dir);
|
|
|
|
rc = fs_opendir(&dir, CONFIG_LOG_BACKEND_FS_DIR);
|
|
if (rc) {
|
|
return -EIO;
|
|
}
|
|
|
|
while (true) {
|
|
rc = fs_readdir(&dir, &ent);
|
|
if (rc < 0) {
|
|
rc = -EIO;
|
|
goto close_dir;
|
|
}
|
|
if (ent.name[0] == 0) {
|
|
break;
|
|
}
|
|
|
|
rc = get_log_file_id(&ent);
|
|
|
|
if (rc == num) {
|
|
rc = 1;
|
|
goto close_dir;
|
|
}
|
|
}
|
|
|
|
rc = 0;
|
|
|
|
close_dir:
|
|
(void) fs_closedir(&dir);
|
|
|
|
return rc;
|
|
}
|
|
|
|
int write_log_to_file(uint8_t *data, size_t length, void *ctx)
|
|
{
|
|
int rc;
|
|
struct fs_file_t *f = &file;
|
|
|
|
if (backend_state == BACKEND_FS_NOT_INITIALIZED) {
|
|
if (check_log_volumen_available()) {
|
|
return length;
|
|
}
|
|
rc = create_log_dir(CONFIG_LOG_BACKEND_FS_DIR);
|
|
if (!rc) {
|
|
rc = allocate_new_file(&file);
|
|
}
|
|
backend_state = (rc ? BACKEND_FS_CORRUPTED : BACKEND_FS_OK);
|
|
}
|
|
|
|
if (backend_state == BACKEND_FS_OK) {
|
|
|
|
/* Check if new data overwrites max file size.
|
|
* If so, create new log file.
|
|
*/
|
|
int size = fs_tell(f);
|
|
|
|
if (size < 0) {
|
|
backend_state = BACKEND_FS_CORRUPTED;
|
|
|
|
return length;
|
|
} else if ((size + length) > CONFIG_LOG_BACKEND_FS_FILE_SIZE) {
|
|
rc = allocate_new_file(f);
|
|
|
|
if (rc < 0) {
|
|
goto on_error;
|
|
}
|
|
}
|
|
|
|
rc = fs_write(f, data, length);
|
|
if (rc >= 0) {
|
|
if (IS_ENABLED(CONFIG_LOG_BACKEND_FS_OVERWRITE) &&
|
|
(rc != length)) {
|
|
del_oldest_log();
|
|
|
|
return 0;
|
|
}
|
|
/* If overwrite is disabled, full memory
|
|
* cause the log record abandonment.
|
|
*/
|
|
length = rc;
|
|
} else {
|
|
rc = check_log_file_exist(newest);
|
|
if (rc == 0) {
|
|
/* file was lost somehow
|
|
* try to get a new one
|
|
*/
|
|
file_ctr--;
|
|
rc = allocate_new_file(f);
|
|
if (rc < 0) {
|
|
goto on_error;
|
|
}
|
|
} else if (rc < 0) {
|
|
/* fs is corrupted*/
|
|
goto on_error;
|
|
}
|
|
length = 0;
|
|
}
|
|
|
|
rc = fs_sync(f);
|
|
if (rc < 0) {
|
|
/* Something is wrong */
|
|
goto on_error;
|
|
}
|
|
}
|
|
|
|
return length;
|
|
|
|
on_error:
|
|
backend_state = BACKEND_FS_CORRUPTED;
|
|
return length;
|
|
}
|
|
|
|
static int get_log_file_id(struct fs_dirent *ent)
|
|
{
|
|
size_t len;
|
|
int num;
|
|
|
|
if (ent->type != FS_DIR_ENTRY_FILE) {
|
|
return -1;
|
|
}
|
|
|
|
len = strlen(ent->name);
|
|
|
|
if (len != LOG_PREFIX_LEN + FILE_NUMERAL_LEN) {
|
|
return -1;
|
|
}
|
|
|
|
if (memcmp(ent->name, CONFIG_LOG_BACKEND_FS_FILE_PREFIX, LOG_PREFIX_LEN) != 0) {
|
|
return -1;
|
|
}
|
|
|
|
num = atoi(ent->name + LOG_PREFIX_LEN);
|
|
|
|
if (num <= MAX_FILE_NUMERAL && num >= 0) {
|
|
return num;
|
|
}
|
|
|
|
return -1;
|
|
}
|
|
|
|
static int allocate_new_file(struct fs_file_t *file)
|
|
{
|
|
/* In case of no log file or current file fills up
|
|
* create new log file.
|
|
*/
|
|
int rc;
|
|
struct fs_statvfs stat;
|
|
int curr_file_num;
|
|
struct fs_dirent ent;
|
|
char fname[MAX_PATH_LEN];
|
|
|
|
assert(file);
|
|
|
|
if (backend_state == BACKEND_FS_NOT_INITIALIZED) {
|
|
/* Search for the last used log number. */
|
|
struct fs_dir_t dir;
|
|
int file_num = 0;
|
|
|
|
fs_dir_t_init(&dir);
|
|
curr_file_num = 0;
|
|
int max = 0, min = MAX_FILE_NUMERAL;
|
|
|
|
rc = fs_opendir(&dir, CONFIG_LOG_BACKEND_FS_DIR);
|
|
|
|
while (rc >= 0) {
|
|
rc = fs_readdir(&dir, &ent);
|
|
if ((rc < 0) || (ent.name[0] == 0)) {
|
|
break;
|
|
}
|
|
|
|
file_num = get_log_file_id(&ent);
|
|
if (file_num >= 0) {
|
|
|
|
if (file_num > max) {
|
|
max = file_num;
|
|
}
|
|
|
|
if (file_num < min) {
|
|
min = file_num;
|
|
}
|
|
++file_ctr;
|
|
}
|
|
}
|
|
|
|
oldest = min;
|
|
|
|
if ((file_ctr > 1) &&
|
|
((max - min) >
|
|
2 * CONFIG_LOG_BACKEND_FS_FILES_LIMIT)) {
|
|
/* oldest log is in the range around the min */
|
|
newest = min;
|
|
oldest = max;
|
|
(void)fs_closedir(&dir);
|
|
rc = fs_opendir(&dir, CONFIG_LOG_BACKEND_FS_DIR);
|
|
|
|
while (rc == 0) {
|
|
rc = fs_readdir(&dir, &ent);
|
|
if ((rc < 0) || (ent.name[0] == 0)) {
|
|
break;
|
|
}
|
|
|
|
file_num = get_log_file_id(&ent);
|
|
if (file_num < min + CONFIG_LOG_BACKEND_FS_FILES_LIMIT) {
|
|
if (newest < file_num) {
|
|
newest = file_num;
|
|
}
|
|
}
|
|
|
|
if (file_num > max - CONFIG_LOG_BACKEND_FS_FILES_LIMIT) {
|
|
if (oldest > file_num) {
|
|
oldest = file_num;
|
|
}
|
|
}
|
|
}
|
|
} else {
|
|
newest = max;
|
|
oldest = min;
|
|
}
|
|
|
|
(void)fs_closedir(&dir);
|
|
if (rc < 0) {
|
|
goto out;
|
|
}
|
|
|
|
curr_file_num = newest;
|
|
|
|
if (file_ctr >= 1) {
|
|
curr_file_num++;
|
|
if (curr_file_num > MAX_FILE_NUMERAL) {
|
|
curr_file_num = 0;
|
|
}
|
|
}
|
|
|
|
backend_state = BACKEND_FS_OK;
|
|
} else {
|
|
fs_close(file);
|
|
|
|
curr_file_num = newest;
|
|
curr_file_num++;
|
|
if (curr_file_num > MAX_FILE_NUMERAL) {
|
|
curr_file_num = 0;
|
|
}
|
|
}
|
|
|
|
rc = fs_statvfs(CONFIG_LOG_BACKEND_FS_DIR, &stat);
|
|
|
|
/* Check if there is enough space to write file or max files number
|
|
* is not exceeded.
|
|
*/
|
|
while ((file_ctr >= CONFIG_LOG_BACKEND_FS_FILES_LIMIT) ||
|
|
((stat.f_bfree * stat.f_frsize) <=
|
|
CONFIG_LOG_BACKEND_FS_FILE_SIZE)) {
|
|
|
|
if (IS_ENABLED(CONFIG_LOG_BACKEND_FS_OVERWRITE)) {
|
|
rc = del_oldest_log();
|
|
if (rc < 0) {
|
|
goto out;
|
|
}
|
|
|
|
rc = fs_statvfs(CONFIG_LOG_BACKEND_FS_DIR,
|
|
&stat);
|
|
if (rc < 0) {
|
|
goto out;
|
|
}
|
|
} else {
|
|
return -ENOSPC;
|
|
}
|
|
}
|
|
|
|
snprintf(fname, sizeof(fname), "%s/%s%04d",
|
|
CONFIG_LOG_BACKEND_FS_DIR,
|
|
CONFIG_LOG_BACKEND_FS_FILE_PREFIX, curr_file_num);
|
|
|
|
rc = fs_open(file, fname, FS_O_CREATE | FS_O_WRITE);
|
|
if (rc < 0) {
|
|
goto out;
|
|
}
|
|
++file_ctr;
|
|
newest = curr_file_num;
|
|
|
|
out:
|
|
return rc;
|
|
}
|
|
|
|
static int del_oldest_log(void)
|
|
{
|
|
int rc;
|
|
static char dellname[MAX_PATH_LEN];
|
|
|
|
while (true) {
|
|
snprintf(dellname, sizeof(dellname), "%s/%s%04d",
|
|
CONFIG_LOG_BACKEND_FS_DIR,
|
|
CONFIG_LOG_BACKEND_FS_FILE_PREFIX, oldest);
|
|
rc = fs_unlink(dellname);
|
|
|
|
if ((rc == 0) || (rc == -ENOENT)) {
|
|
oldest++;
|
|
if (oldest > MAX_FILE_NUMERAL) {
|
|
oldest = 0;
|
|
}
|
|
|
|
if (rc == 0) {
|
|
--file_ctr;
|
|
break;
|
|
}
|
|
} else {
|
|
break;
|
|
}
|
|
}
|
|
|
|
return rc;
|
|
}
|
|
|
|
BUILD_ASSERT(!IS_ENABLED(CONFIG_LOG_MODE_IMMEDIATE),
|
|
"Immediate logging is not supported by LOG FS backend.");
|
|
|
|
#ifndef CONFIG_LOG_BACKEND_FS_TESTSUITE
|
|
|
|
static uint8_t __aligned(4) buf[MAX_FLASH_WRITE_SIZE];
|
|
LOG_OUTPUT_DEFINE(log_output, write_log_to_file, buf, MAX_FLASH_WRITE_SIZE);
|
|
|
|
static void put(const struct log_backend *const backend,
|
|
struct log_msg *msg)
|
|
{
|
|
log_backend_std_put(&log_output, 0, msg);
|
|
}
|
|
|
|
static void log_backend_fs_init(const struct log_backend *const backend)
|
|
{
|
|
}
|
|
|
|
static void panic(struct log_backend const *const backend)
|
|
{
|
|
/* In case of panic deinitialize backend. It is better to keep
|
|
* current data rather than log new and risk of failure.
|
|
*/
|
|
log_backend_deactivate(backend);
|
|
}
|
|
|
|
static void dropped(const struct log_backend *const backend, uint32_t cnt)
|
|
{
|
|
ARG_UNUSED(backend);
|
|
|
|
if (IS_ENABLED(CONFIG_LOG_BACKEND_FS_OUTPUT_DICTIONARY)) {
|
|
log_dict_output_dropped_process(&log_output, cnt);
|
|
} else {
|
|
log_backend_std_dropped(&log_output, cnt);
|
|
}
|
|
}
|
|
|
|
static void process(const struct log_backend *const backend,
|
|
union log_msg2_generic *msg)
|
|
{
|
|
uint32_t flags = log_backend_std_get_flags();
|
|
|
|
log_format_func_t log_output_func = log_format_func_t_get(log_format_current);
|
|
|
|
log_output_func(&log_output, &msg->log, flags);
|
|
}
|
|
|
|
static int format_set(const struct log_backend *const backend, uint32_t log_type)
|
|
{
|
|
log_format_current = log_type;
|
|
return 0;
|
|
}
|
|
|
|
static const struct log_backend_api log_backend_fs_api = {
|
|
.process = IS_ENABLED(CONFIG_LOG2) ? process : NULL,
|
|
.put = put,
|
|
.put_sync_string = NULL,
|
|
.put_sync_hexdump = NULL,
|
|
.panic = panic,
|
|
.init = log_backend_fs_init,
|
|
.dropped = dropped,
|
|
.format_set = IS_ENABLED(CONFIG_LOG1) ? NULL : format_set,
|
|
};
|
|
|
|
|
|
LOG_BACKEND_DEFINE(log_backend_fs, log_backend_fs_api, true);
|
|
#endif
|