path: root/src/cepis-http.c

/*
 * Transport primitives for the cep.is COBOL client.
 *
 * COBOL has no HTTP, and `curl_easy_setopt` is variadic, which a COBOL
 * CALL cannot express portably. Everything variadic therefore lives here,
 * behind four fixed-signature entry points that a COBOL CALL can reach:
 *
 *   cepis_http_get         one GET, no retrying -- the policy stays in COBOL
 *   cepis_sleep_ms         the backoff wait
 *   cepis_monotonic_ms     the clock CFG-TOTAL-TIMEOUT-MS is measured on
 *   cepis_http_date_delta  the RFC 9110 HTTP-date form of Retry-After
 *
 * All scalar arguments are `int` or `long long` passed BY REFERENCE, which
 * COBOL declares as PIC S9(9) COMP-5 and PIC S9(18) COMP-5. All string
 * arguments in are NUL-terminated by the caller. All string arguments out
 * are blank-filled to their capacity, so the COBOL field they land in is a
 * normal space-padded PIC X.
 */

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <strings.h>
#include <time.h>

#include <curl/curl.h>

#define CEPIS_OK_TRANSPORT 0
#define CEPIS_ERR_TRANSPORT 1
#define CEPIS_ERR_TOO_LARGE 2

struct cepis_body {
    char *buf;
    long cap;
    long len;
    int overflow;
};

struct cepis_header {
    char *buf;
    long cap;
    int seen;
};

static int cepis_curl_ready;

static void cepis_blank(char *dst, long cap)
{
    if (dst && cap > 0) {
        memset(dst, ' ', (size_t)cap);
    }
}

static void cepis_put(char *dst, long cap, const char *src)
{
    size_t n;

    cepis_blank(dst, cap);
    if (!dst || cap <= 0 || !src) {
        return;
    }
    n = strlen(src);
    if (n > (size_t)cap) {
        n = (size_t)cap;
    }
    memcpy(dst, src, n);
}

static size_t cepis_on_body(char *ptr, size_t size, size_t nmemb, void *userdata)
{
    struct cepis_body *sink = (struct cepis_body *)userdata;
    size_t n = size * nmemb;

    if (sink->len + (long)n > sink->cap) {
        sink->overflow = 1;
        return 0;
    }
    memcpy(sink->buf + sink->len, ptr, n);
    sink->len += (long)n;
    return n;
}

static size_t cepis_on_header(char *ptr, size_t size, size_t nmemb, void *userdata)
{
    struct cepis_header *hdr = (struct cepis_header *)userdata;
    size_t n = size * nmemb;
    static const char name[] = "retry-after:";
    size_t name_len = sizeof(name) - 1;
    size_t start, end, len;

    if (hdr->seen || n <= name_len || strncasecmp(ptr, name, name_len) != 0) {
        return n;
    }
    start = name_len;
    end = n;
    while (start < end && (ptr[start] == ' ' || ptr[start] == '\t')) {
        start++;
    }
    while (end > start && (ptr[end - 1] == '\r' || ptr[end - 1] == '\n' ||
                           ptr[end - 1] == ' ' || ptr[end - 1] == '\t')) {
        end--;
    }
    len = end - start;
    if (len > (size_t)hdr->cap) {
        len = (size_t)hdr->cap;
    }
    memcpy(hdr->buf, ptr + start, len);
    hdr->seen = 1;
    return n;
}

/*
 * Perform one GET. Returns 0 when the exchange completed (whatever the
 * status), 1 when it never completed, 2 when the body did not fit.
 */
int cepis_http_get(const char *url,
                   const char *user_agent,
                   const int *timeout_ms,
                   int *http_status,
                   char *retry_after,
                   const int *retry_after_cap,
                   char *body,
                   const int *body_cap,
                   int *body_len,
                   char *error_msg,
                   const int *error_cap)
{
    CURL *curl;
    CURLcode rc;
    struct curl_slist *headers = NULL;
    struct cepis_body sink;
    struct cepis_header hdr;
    char errbuf[CURL_ERROR_SIZE];
    long status = 0;
    int result = CEPIS_ERR_TRANSPORT;

    *http_status = 0;
    *body_len = 0;
    cepis_blank(retry_after, *retry_after_cap);
    cepis_blank(error_msg, *error_cap);

    if (!cepis_curl_ready) {
        if (curl_global_init(CURL_GLOBAL_DEFAULT) != CURLE_OK) {
            cepis_put(error_msg, *error_cap, "curl_global_init failed");
            return CEPIS_ERR_TRANSPORT;
        }
        cepis_curl_ready = 1;
    }

    curl = curl_easy_init();
    if (!curl) {
        cepis_put(error_msg, *error_cap, "curl_easy_init failed");
        return CEPIS_ERR_TRANSPORT;
    }

    sink.buf = body;
    sink.cap = *body_cap;
    sink.len = 0;
    sink.overflow = 0;

    hdr.buf = retry_after;
    hdr.cap = *retry_after_cap;
    hdr.seen = 0;

    errbuf[0] = '\0';

    headers = curl_slist_append(headers,
                                "Accept: application/json, text/csv;q=0.9, */*;q=0.1");
    headers = curl_slist_append(headers, "Accept-Encoding: identity");

    curl_easy_setopt(curl, CURLOPT_URL, url);
    curl_easy_setopt(curl, CURLOPT_HTTPGET, 1L);
    curl_easy_setopt(curl, CURLOPT_USERAGENT, user_agent);
    curl_easy_setopt(curl, CURLOPT_HTTPHEADER, headers);
    curl_easy_setopt(curl, CURLOPT_FOLLOWLOCATION, 1L);
    curl_easy_setopt(curl, CURLOPT_MAXREDIRS, 5L);
    curl_easy_setopt(curl, CURLOPT_NOSIGNAL, 1L);
    curl_easy_setopt(curl, CURLOPT_CONNECTTIMEOUT_MS, (long)*timeout_ms);
    curl_easy_setopt(curl, CURLOPT_TIMEOUT_MS, (long)*timeout_ms);
    curl_easy_setopt(curl, CURLOPT_WRITEFUNCTION, cepis_on_body);
    curl_easy_setopt(curl, CURLOPT_WRITEDATA, &sink);
    curl_easy_setopt(curl, CURLOPT_HEADERFUNCTION, cepis_on_header);
    curl_easy_setopt(curl, CURLOPT_HEADERDATA, &hdr);
    curl_easy_setopt(curl, CURLOPT_ERRORBUFFER, errbuf);

    rc = curl_easy_perform(curl);

    if (rc == CURLE_OK) {
        curl_easy_getinfo(curl, CURLINFO_RESPONSE_CODE, &status);
        *http_status = (int)status;
        *body_len = (int)sink.len;
        result = CEPIS_OK_TRANSPORT;
    } else if (sink.overflow) {
        cepis_put(error_msg, *error_cap, "response body exceeds the client buffer");
        result = CEPIS_ERR_TOO_LARGE;
    } else {
        cepis_put(error_msg, *error_cap,
                  errbuf[0] ? errbuf : curl_easy_strerror(rc));
        result = CEPIS_ERR_TRANSPORT;
    }

    curl_slist_free_all(headers);
    curl_easy_cleanup(curl);
    return result;
}

/* Sleep for the backoff. Negative and zero waits return at once. */
void cepis_sleep_ms(const int *ms)
{
    struct timespec req;

    if (*ms <= 0) {
        return;
    }
    req.tv_sec = *ms / 1000;
    req.tv_nsec = (long)(*ms % 1000) * 1000000L;
    while (nanosleep(&req, &req) == -1) {
        break;
    }
}

/*
 * Milliseconds off a monotonic clock. Wall time would let a clock step
 * during an NTP correction shorten or extend CFG-TOTAL-TIMEOUT-MS.
 */
void cepis_monotonic_ms(long long *out)
{
    struct timespec now;

    if (clock_gettime(CLOCK_MONOTONIC, &now) != 0) {
        *out = 0;
        return;
    }
    *out = (long long)now.tv_sec * 1000LL + (long long)(now.tv_nsec / 1000000L);
}

/*
 * Seconds from now until an HTTP-date, the second form Retry-After may
 * take. Returns 0 and sets *seconds when the date parses, 1 otherwise.
 * The numeric form is handled in COBOL, where it belongs.
 */
int cepis_http_date_delta(const char *value, int *seconds)
{
    time_t target;
    time_t now;
    double delta;

    *seconds = 0;
    target = curl_getdate(value, NULL);
    if (target == (time_t)-1) {
        return 1;
    }
    now = time(NULL);
    delta = difftime(target, now);
    if (delta < 0) {
        delta = 0;
    }
    if (delta > 2147483000.0) {
        delta = 2147483000.0;
    }
    *seconds = (int)delta;
    return 0;
}