path: root/collectors/ebpf.plugin/ebpf_cachestat.c

// SPDX-License-Identifier: GPL-3.0-or-later

#include "ebpf.h"
#include "ebpf_cachestat.h"

static ebpf_data_t cachestat_data;
netdata_publish_cachestat_t **cachestat_pid;

static struct bpf_link **probe_links = NULL;
static struct bpf_object *objects = NULL;

static char *cachestat_counter_dimension_name[NETDATA_CACHESTAT_END] = { "ratio", "dirty", "hit",
                                                                         "miss" };
static netdata_syscall_stat_t cachestat_counter_aggregated_data[NETDATA_CACHESTAT_END];
static netdata_publish_syscall_t cachestat_counter_publish_aggregated[NETDATA_CACHESTAT_END];

netdata_cachestat_pid_t *cachestat_vector = NULL;

static netdata_idx_t *cachestat_hash_values = NULL;

static int read_thread_closed = 1;

struct netdata_static_thread cachestat_threads = {"CACHESTAT KERNEL",
                                                  NULL, NULL, 1, NULL,
                                                  NULL,  NULL};

static ebpf_local_maps_t cachestat_maps[] = {{.name = "cstat_pid", .internal_input = ND_EBPF_DEFAULT_PID_SIZE,
                                             .user_input = 0},
                                             {.name = NULL, .internal_input = 0, .user_input = 0}};

static int *map_fd = NULL;

struct config cachestat_config = { .first_section = NULL,
    .last_section = NULL,
    .mutex = NETDATA_MUTEX_INITIALIZER,
    .index = { .avl_tree = { .root = NULL, .compar = appconfig_section_compare },
        .rwlock = AVL_LOCK_INITIALIZER } };

/*****************************************************************
 *
 *  FUNCTIONS TO CLOSE THE THREAD
 *
 *****************************************************************/

/**
 * Clean PID structures
 *
 * Clean the allocated structures.
 */
void clean_cachestat_pid_structures() {
    struct pid_stat *pids = root_of_pids;
    while (pids) {
        freez(cachestat_pid[pids->pid]);

        pids = pids->next;
    }
}

/**
 * Clean up the main thread.
 *
 * @param ptr thread data.
 */
static void ebpf_cachestat_cleanup(void *ptr)
{
    ebpf_module_t *em = (ebpf_module_t *)ptr;
    if (!em->enabled)
        return;

    heartbeat_t hb;
    heartbeat_init(&hb);
    uint32_t tick = 2*USEC_PER_MS;
    while (!read_thread_closed) {
        usec_t dt = heartbeat_next(&hb, tick);
        UNUSED(dt);
    }

    ebpf_cleanup_publish_syscall(cachestat_counter_publish_aggregated);

    freez(cachestat_vector);
    freez(cachestat_hash_values);

    struct bpf_program *prog;
    size_t i = 0 ;
    bpf_object__for_each_program(prog, objects) {
        bpf_link__destroy(probe_links[i]);
        i++;
    }
    bpf_object__close(objects);
}

/*****************************************************************
 *
 *  COMMON FUNCTIONS
 *
 *****************************************************************/

/**
 * Update publish
 *
 * Update publish values before to write dimension.
 *
 * @param out  strcuture that will receive data.
 * @param mpa  calls for mark_page_accessed during the last second.
 * @param mbd  calls for mark_buffer_dirty during the last second.
 * @param apcl calls for add_to_page_cache_lru during the last second.
 * @param apd  calls for account_page_dirtied during the last second.
 */
void cachestat_update_publish(netdata_publish_cachestat_t *out, uint64_t mpa, uint64_t mbd,
                              uint64_t apcl, uint64_t apd)
{
    // Adapted algorithm from https://github.com/iovisor/bcc/blob/master/tools/cachestat.py#L126-L138
    calculated_number total = (calculated_number) (((long long)mpa) - ((long long)mbd));
    if (total < 0)
        total = 0;

    calculated_number misses = (calculated_number) ( ((long long) apcl) - ((long long) apd) );
    if (misses < 0)
        misses = 0;

    // If hits are < 0, then its possible misses are overestimate due to possibly page cache read ahead adding
    // more pages than needed. In this case just assume misses as total and reset hits.
    calculated_number hits = total - misses;
    if (hits < 0 ) {
        misses = total;
        hits = 0;
    }

    calculated_number ratio = (total > 0) ? hits/total : 1;

    out->ratio = (long long )(ratio*100);
    out->hit = (long long)hits;
    out->miss = (long long)misses;
}

/**
 * Save previous values
 *
 * Save values used this time.
 *
 * @param publish
 */
static void save_previous_values(netdata_publish_cachestat_t *publish) {
    publish->prev.mark_page_accessed = cachestat_hash_values[NETDATA_KEY_CALLS_MARK_PAGE_ACCESSED];
    publish->prev.account_page_dirtied = cachestat_hash_values[NETDATA_KEY_CALLS_ACCOUNT_PAGE_DIRTIED];
    publish->prev.add_to_page_cache_lru = cachestat_hash_values[NETDATA_KEY_CALLS_ADD_TO_PAGE_CACHE_LRU];
    publish->prev.mark_buffer_dirty = cachestat_hash_values[NETDATA_KEY_CALLS_MARK_BUFFER_DIRTY];
}

/**
 * Calculate statistics
 *
 * @param publish the structure where we will store the data.
 */
static void calculate_stats(netdata_publish_cachestat_t *publish) {
    if (!publish->prev.mark_page_accessed) {
        save_previous_values(publish);
        return;
    }

    uint64_t mpa = cachestat_hash_values[NETDATA_KEY_CALLS_MARK_PAGE_ACCESSED] - publish->prev.mark_page_accessed;
    uint64_t mbd = cachestat_hash_values[NETDATA_KEY_CALLS_MARK_BUFFER_DIRTY] - publish->prev.mark_buffer_dirty;
    uint64_t apcl = cachestat_hash_values[NETDATA_KEY_CALLS_ADD_TO_PAGE_CACHE_LRU] - publish->prev.add_to_page_cache_lru;
    uint64_t apd = cachestat_hash_values[NETDATA_KEY_CALLS_ACCOUNT_PAGE_DIRTIED] - publish->prev.account_page_dirtied;

    save_previous_values(publish);

    // We are changing the original algorithm to have a smooth ratio.
    cachestat_update_publish(publish, mpa, mbd, apcl, apd);
}


/*****************************************************************
 *
 *  APPS
 *
 *****************************************************************/

/**
 * Apps Accumulator
 *
 * Sum all values read from kernel and store in the first address.
 *
 * @param out the vector with read values.
 */
static void cachestat_apps_accumulator(netdata_cachestat_pid_t *out)
{
    int i, end = (running_on_kernel >= NETDATA_KERNEL_V4_15) ? ebpf_nprocs : 1;
    netdata_cachestat_pid_t *total = &out[0];
    for (i = 1; i < end; i++) {
        netdata_cachestat_pid_t *w = &out[i];
        total->account_page_dirtied += w->account_page_dirtied;
        total->add_to_page_cache_lru