diff options
Diffstat (limited to 'xmppcb.c')
| -rw-r--r-- | xmppcb.c | 1190 |
1 files changed, 1190 insertions, 0 deletions
diff --git a/xmppcb.c b/xmppcb.c new file mode 100644 index 0000000..c934550 --- /dev/null +++ b/xmppcb.c @@ -0,0 +1,1190 @@ +/* xmppcb - a lean XMPP AI chatbot in C. + * + * Joins MUC rooms, logs messages, and answers @bot commands via an + * OpenAI-compatible LLM HTTP API. Transport is TLS (STARTTLS for XMPP, + * HTTPS for the LLM); there is no end-to-end / OMEMO encryption. + * + * One dependency: OpenSSL. Everything else is hand-rolled. + */ + +#ifndef _POSIX_C_SOURCE +#define _POSIX_C_SOURCE 200809L +#endif + +#include <ctype.h> +#include <errno.h> +#include <netdb.h> +#include <poll.h> +#include <signal.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <strings.h> +#include <time.h> +#include <unistd.h> +#include <sys/socket.h> +#include <sys/stat.h> +#include <sys/types.h> + +#include <openssl/err.h> +#include <openssl/ssl.h> +#include <openssl/x509v3.h> + +#include "config.h" + +#define NROOMS ((int)(sizeof(ROOMS) / sizeof(ROOMS[0]))) + +/* buffer sizes */ +#define INBUF 65536 +#define STANZABUF 66000 +#define BODYBUF 16384 +#define CTXBUF 24576 +#define PROMPTBUF 32768 +#define INSTRBUF 8192 +#define ESCBUF 98304 +#define JSONBUF 131072 +#define REQBUF 140000 +#define HTTPBUF 98304 +#define LLMOUT 8192 +#define CTX_FILEBUF 131072 +#define CTX_MAXLINES 4000 + +/* run_session() return codes */ +enum { SESS_RETRY = 0, SESS_AUTHFAIL = 1 }; + +/* ------------------------------------------------------------------ */ +/* globals */ +/* ------------------------------------------------------------------ */ + +static SSL_CTX *g_ctx; +static const char *g_password; +static const char *g_apikey; +static char g_domain[256]; +static char g_localpart[128]; +static char g_instruction[INSTRBUF]; + +static char inbuf[INBUF]; +static size_t inlen; +static char stanza[STANZABUF]; + +typedef struct { int fd; SSL *ssl; } Conn; + +static void die(const char *msg) { + fprintf(stderr, "xmppcb: %s\n", msg); + exit(1); +} + +/* ------------------------------------------------------------------ */ +/* small utilities */ +/* ------------------------------------------------------------------ */ + +/* case-insensitive substring search */ +static const char *ci_strstr(const char *hay, const char *needle) { + size_t nl = strlen(needle); + if (!nl) return hay; + for (; *hay; hay++) + if (strncasecmp(hay, needle, nl) == 0) + return hay; + return NULL; +} + +static size_t utf8_encode(char *d, size_t dsz, unsigned cp) { + if (cp < 0x80) { + if (dsz < 1) return 0; + d[0] = (char)cp; + return 1; + } + if (cp < 0x800) { + if (dsz < 2) return 0; + d[0] = (char)(0xC0 | (cp >> 6)); + d[1] = (char)(0x80 | (cp & 0x3F)); + return 2; + } + if (cp < 0x10000) { + if (dsz < 3) return 0; + d[0] = (char)(0xE0 | (cp >> 12)); + d[1] = (char)(0x80 | ((cp >> 6) & 0x3F)); + d[2] = (char)(0x80 | (cp & 0x3F)); + return 3; + } + if (dsz < 4) return 0; + d[0] = (char)(0xF0 | (cp >> 18)); + d[1] = (char)(0x80 | ((cp >> 12) & 0x3F)); + d[2] = (char)(0x80 | ((cp >> 6) & 0x3F)); + d[3] = (char)(0x80 | (cp & 0x3F)); + return 4; +} + +static const char B64[] = + "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/"; + +static size_t b64encode(char *dst, const unsigned char *src, size_t n) { + size_t o = 0, i = 0; + while (i + 3 <= n) { + unsigned v = (src[i] << 16) | (src[i + 1] << 8) | src[i + 2]; + dst[o++] = B64[(v >> 18) & 63]; + dst[o++] = B64[(v >> 12) & 63]; + dst[o++] = B64[(v >> 6) & 63]; + dst[o++] = B64[v & 63]; + i += 3; + } + if (n - i == 1) { + unsigned v = src[i] << 16; + dst[o++] = B64[(v >> 18) & 63]; + dst[o++] = B64[(v >> 12) & 63]; + dst[o++] = '='; + dst[o++] = '='; + } else if (n - i == 2) { + unsigned v = (src[i] << 16) | (src[i + 1] << 8); + dst[o++] = B64[(v >> 18) & 63]; + dst[o++] = B64[(v >> 12) & 63]; + dst[o++] = B64[(v >> 6) & 63]; + dst[o++] = '='; + } + dst[o] = '\0'; + return o; +} + +/* escape text for inclusion in XML element/attribute content */ +static void xml_escape(char *dst, size_t dsz, const char *src) { + size_t o = 0; + for (; *src; src++) { + const char *r = NULL; + switch (*src) { + case '&': r = "&"; break; + case '<': r = "<"; break; + case '>': r = ">"; break; + case '"': r = """; break; + case '\'': r = "'"; break; + } + if (r) { + size_t l = strlen(r); + if (o + l >= dsz) break; + memcpy(dst + o, r, l); + o += l; + } else { + if (o + 1 >= dsz) break; + dst[o++] = *src; + } + } + dst[o] = '\0'; +} + +/* resolve XML entities in src into dst */ +static void xml_unescape(char *dst, size_t dsz, const char *src) { + size_t o = 0; + while (*src && o + 1 < dsz) { + if (*src == '&') { + if (!strncmp(src, "&", 5)) { dst[o++] = '&'; src += 5; continue; } + if (!strncmp(src, "<", 4)) { dst[o++] = '<'; src += 4; continue; } + if (!strncmp(src, ">", 4)) { dst[o++] = '>'; src += 4; continue; } + if (!strncmp(src, """, 6)) { dst[o++] = '"'; src += 6; continue; } + if (!strncmp(src, "'", 6)) { dst[o++] = '\''; src += 6; continue; } + if (src[1] == '#') { + int base = 10; + const char *q = src + 2; + if (*q == 'x' || *q == 'X') { base = 16; q++; } + char *endp; + long cp = strtol(q, &endp, base); + if (endp != q && *endp == ';' && cp > 0) { + o += utf8_encode(dst + o, dsz - 1 - o, (unsigned)cp); + src = endp + 1; + continue; + } + } + } + dst[o++] = *src++; + } + dst[o] = '\0'; +} + +/* escape a string for use as a JSON string value */ +static void json_escape(char *dst, size_t dsz, const char *src) { + size_t o = 0; + for (; *src; src++) { + unsigned char ch = (unsigned char)*src; + char tmp[8]; + const char *r; + size_t rl; + switch (ch) { + case '"': r = "\\\""; rl = 2; break; + case '\\': r = "\\\\"; rl = 2; break; + case '\n': r = "\\n"; rl = 2; break; + case '\r': r = "\\r"; rl = 2; break; + case '\t': r = "\\t"; rl = 2; break; + case '\b': r = "\\b"; rl = 2; break; + case '\f': r = "\\f"; rl = 2; break; + default: + if (ch < 0x20) { + snprintf(tmp, sizeof tmp, "\\u%04x", ch); + r = tmp; + rl = 6; + } else { + tmp[0] = (char)ch; + r = tmp; + rl = 1; + } + } + if (o + rl >= dsz) break; + memcpy(dst + o, r, rl); + o += rl; + } + dst[o] = '\0'; +} + +/* extract a JSON string value for "key" from json into out */ +static int json_get_string(const char *json, const char *key, + char *out, size_t osz) { + char pat[64]; + snprintf(pat, sizeof pat, "\"%s\"", key); + const char *p = strstr(json, pat); + if (!p) return 0; + p += strlen(pat); + while (*p && *p != ':') p++; + if (*p != ':') return 0; + p++; + while (*p == ' ' || *p == '\t' || *p == '\n' || *p == '\r') p++; + if (*p != '"') return 0; + p++; + size_t o = 0; + while (*p && *p != '"') { + if (*p == '\\') { + p++; + if (!*p) break; + switch (*p) { + case 'n': if (o + 1 < osz) out[o++] = '\n'; break; + case 't': if (o + 1 < osz) out[o++] = '\t'; break; + case 'r': if (o + 1 < osz) out[o++] = '\r'; break; + case 'b': if (o + 1 < osz) out[o++] = '\b'; break; + case 'f': if (o + 1 < osz) out[o++] = '\f'; break; + case '/': if (o + 1 < osz) out[o++] = '/'; break; + case '"': if (o + 1 < osz) out[o++] = '"'; break; + case '\\': if (o + 1 < osz) out[o++] = '\\'; break; + case 'u': { + unsigned cp = 0; + for (int i = 1; i <= 4 && p[i]; i++) { + char c = p[i]; + cp <<= 4; + if (c >= '0' && c <= '9') cp |= (unsigned)(c - '0'); + else if (c >= 'a' && c <= 'f') cp |= (unsigned)(c - 'a' + 10); + else if (c >= 'A' && c <= 'F') cp |= (unsigned)(c - 'A' + 10); + } + p += 4; + if (cp >= 0xD800 && cp <= 0xDBFF && + p[1] == '\\' && p[2] == 'u') { + unsigned lo = 0; + for (int i = 3; i <= 6 && p[i]; i++) { + char c = p[i]; + lo <<= 4; + if (c >= '0' && c <= '9') lo |= (unsigned)(c - '0'); + else if (c >= 'a' && c <= 'f') lo |= (unsigned)(c - 'a' + 10); + else if (c >= 'A' && c <= 'F') lo |= (unsigned)(c - 'A' + 10); + } + cp = 0x10000 + ((cp - 0xD800) << 10) + (lo - 0xDC00); + p += 6; + } + o += utf8_encode(out + o, osz - 1 - o, cp); + break; + } + default: + if (o + 1 < osz) out[o++] = *p; + } + p++; + } else { + if (o + 1 < osz) out[o++] = *p; + p++; + } + } + out[o] = '\0'; + return 1; +} + +/* decode HTTP chunked transfer encoding */ +static void dechunk(const char *src, char *dst, size_t dsz) { + size_t o = 0; + while (*src) { + char *end; + long sz = strtol(src, &end, 16); + if (end == src || sz <= 0) break; + src = strchr(end, '\n'); + if (!src) break; + src++; + for (long i = 0; i < sz && *src && o + 1 < dsz; i++) + dst[o++] = *src++; + if (*src == '\r') src++; + if (*src == '\n') src++; + } + dst[o] = '\0'; +} + +static void rtrim(char *s) { + size_t n = strlen(s); + while (n && isspace((unsigned char)s[n - 1])) s[--n] = '\0'; +} + +/* ------------------------------------------------------------------ */ +/* minimal XML stanza parsing */ +/* ------------------------------------------------------------------ */ + +/* Length of the first complete XML chunk in b[0..n): an <?xml?> decl, + * the <stream:stream> open tag, or a full top-level stanza. 0 if more + * data is needed. */ +static size_t xml_chunk_len(const char *b, size_t n) { + size_t i = 0; + while (i < n && (b[i] == ' ' || b[i] == '\t' || + b[i] == '\r' || b[i] == '\n')) + i++; + if (i >= n || b[i] != '<') return 0; + + if (i + 1 < n && b[i + 1] == '?') { /* <?xml ... ?> */ + for (size_t j = i + 2; j + 1 < n; j++) + if (b[j] == '?' && b[j + 1] == '>') + return j + 2; + return 0; + } + + int depth = 0; + size_t j = i; + while (j < n) { + if (b[j] != '<') { j++; continue; } + int closing = (j + 1 < n && b[j + 1] == '/'); + char q = 0; + size_t k = j + 1; + while (k < n) { + char ch = b[k]; + if (q) { if (ch == q) q = 0; } + else if (ch == '\'' || ch == '"') q = ch; + else if (ch == '>') break; + k++; + } + if (k >= n) return 0; /* tag incomplete */ + int selfclose = (b[k - 1] == '/'); + if (depth == 0 && !closing && !selfclose && + k - (j + 1) >= 13 && + memcmp(b + j + 1, "stream:stream", 13) == 0) + return k + 1; /* stream open */ + if (closing) depth--; + else if (!selfclose) depth++; + j = k + 1; + if (depth <= 0) return j; /* stanza complete */ + } + return 0; +} + +/* name of the root element of s into out */ +static void root_tag(const char *s, char *out, size_t osz) { + size_t o = 0; + while (*s && *s != '<') s++; + if (*s == '<') s++; + while (*s && *s != ' ' && *s != '\t' && *s != '\r' && + *s != '\n' && *s != '>' && *s != '/' && o + 1 < osz) + out[o++] = *s++; + out[o] = '\0'; +} + +/* value of attribute `name` from the root tag of stanza */ +static int attr_get(const char *st, const char *name, + char *out, size_t osz) { + out[0] = '\0'; + const char *end = st; + char q = 0; + while (*end) { /* end of root open tag */ + if (q) { if (*end == q) q = 0; } + else if (*end == '\'' || *end == '"') q = *end; + else if (*end == '>') break; + end++; + } + size_t nl = strlen(name); + for (const char *p = st; p < end; p++) { + if ((*p == ' ' || *p == '\t' || *p == '\r' || *p == '\n') && + p + 1 + nl < end && + strncmp(p + 1, name, nl) == 0) { + const char *a = p + 1 + nl; + while (a < end && (*a == ' ' || *a == '\t')) a++; + if (a < end && *a == '=') { + a++; + while (a < end && (*a == ' ' || *a == '\t')) a++; + if (a < end && (*a == '\'' || *a == '"')) { + char qq = *a++; + size_t o = 0; + while (a < end && *a != qq && o + 1 < osz) + out[o++] = *a++; + out[o] = '\0'; + return 1; + } + } + } + } + return 0; +} + +/* text content of the first <name>...</name> child element */ +static int elem_text(const char *st, const char *name, + char *out, size_t osz) { + out[0] = '\0'; + size_t nl = strlen(name); + const char *p = st; + while ((p = strchr(p, '<'))) { + if (strncmp(p + 1, name, nl) == 0) { + char after = p[1 + nl]; + if (after == ' ' || after == '\t' || after == '\r' || + after == '\n' || after == '>' || after == '/') { + const char *gt = p + 1; + char q = 0; + while (*gt) { + if (q) { if (*gt == q) q = 0; } + else if (*gt == '\'' || *gt == '"') q = *gt; + else if (*gt == '>') break; + gt++; + } + if (*gt != '>') return 0; + if (gt > p && gt[-1] == '/') return 1; /* empty */ + const char *txt = gt + 1; + char close[64]; + snprintf(close, sizeof close, "</%s", name); + const char *ce = strstr(txt, close); + if (!ce) return 0; + size_t len = (size_t)(ce - txt); + if (len >= osz) len = osz - 1; + memcpy(out, txt, len); + out[len] = '\0'; + return 1; + } + } + p++; + } + return 0; +} + +/* ------------------------------------------------------------------ */ +/* connection: TCP + optional TLS */ +/* ------------------------------------------------------------------ */ + +static int conn_write(Conn *c, const char *buf, size_t len) { + size_t off = 0; + while (off < len) { + int n = c->ssl + ? SSL_write(c->ssl, buf + off, (int)(len - off)) + : (int)write(c->fd, buf + off, len - off); + if (n <= 0) return -1; + off += (size_t)n; + } + return 0; +} + +static int conn_puts(Conn *c, const char *s) { + return conn_write(c, s, strlen(s)); +} + +static int conn_read_some(Conn *c, char *buf, size_t len) { + return c->ssl ? SSL_read(c->ssl, buf, (int)len) + : (int)read(c->fd, buf, len); +} + +static int tcp_connect(const char *host, int port) { + char portstr[16]; + snprintf(portstr, sizeof portstr, "%d", port); + struct addrinfo hints, *res, *rp; + memset(&hints, 0, sizeof hints); + hints.ai_family = AF_UNSPEC; + hints.ai_socktype = SOCK_STREAM; + if (getaddrinfo(host, portstr, &hints, &res) != 0) + return -1; + int fd = -1; + for (rp = res; rp; rp = rp->ai_next) { + fd = socket(rp->ai_family, rp->ai_socktype, rp->ai_protocol); + if (fd < 0) continue; + if (connect(fd, rp->ai_addr, rp->ai_addrlen) == 0) break; + close(fd); + fd = -1; + } + freeaddrinfo(res); + if (fd >= 0) { + struct timeval tv = { 60, 0 }; + setsockopt(fd, SOL_SOCKET, SO_RCVTIMEO, &tv, sizeof tv); + } + return fd; +} + +static void tls_init(void) { + g_ctx = SSL_CTX_new(TLS_client_method()); + if (!g_ctx) die("SSL_CTX_new failed"); + SSL_CTX_set_min_proto_version(g_ctx, TLS1_2_VERSION); +#if TLS_VERIFY + if (!SSL_CTX_set_default_verify_paths(g_ctx)) + fprintf(stderr, "xmppcb: warning: no system CA certificates\n"); +#endif +} + +static SSL *tls_wrap(int fd, const char *host) { + SSL *ssl = SSL_new(g_ctx); + if (!ssl) return NULL; + SSL_set_fd(ssl, fd); + SSL_set_tlsext_host_name(ssl, host); +#if TLS_VERIFY + SSL_set_verify(ssl, SSL_VERIFY_PEER, NULL); + X509_VERIFY_PARAM *vp = SSL_get0_param(ssl); + X509_VERIFY_PARAM_set_hostflags(vp, X509_CHECK_FLAG_NO_PARTIAL_WILDCARDS); + X509_VERIFY_PARAM_set1_host(vp, host, 0); +#endif + if (SSL_connect(ssl) != 1) { + fprintf(stderr, "xmppcb: TLS handshake with %s failed\n", host); + ERR_print_errors_fp(stderr); + SSL_free(ssl); + return NULL; + } + return ssl; +} + +/* ------------------------------------------------------------------ */ +/* stanza stream reader */ +/* ------------------------------------------------------------------ */ + +/* Pull one complete XML chunk out of inbuf into out. Returns its + * length, or 0 if inbuf does not yet hold a full chunk. */ +static size_t extract_chunk(char *out, size_t osz) { + size_t ws = 0; + while (ws < inlen && (inbuf[ws] == ' ' || inbuf[ws] == '\t' || + inbuf[ws] == '\r' || inbuf[ws] == '\n')) + ws++; + if (ws) { + memmove(inbuf, inbuf + ws, inlen - ws); + inlen -= ws; + } + size_t L = xml_chunk_len(inbuf, inlen); + if (L == 0 || L >= osz) return 0; + memcpy(out, inbuf, L); + out[L] = '\0'; + memmove(inbuf, inbuf + L, inlen - L); + inlen -= L; + return L; +} + +/* Blocking: read the next XML chunk. Used during login. */ +static int next_chunk(Conn *c, char *out, size_t osz) { + for (;;) { + size_t L = extract_chunk(out, osz); + if (L) return (int)L; + if (inlen >= sizeof inbuf) return -1; + int n = conn_read_some(c, inbuf + inlen, sizeof inbuf - inlen); + if (n <= 0) return -1; + inlen += (size_t)n; + } +} + +/* Read chunks until one whose root element is `want`. */ +static int await_tag(Conn *c, const char *want) { + char tag[64]; + for (int i = 0; i < 50; i++) { + if (next_chunk(c, stanza, sizeof stanza) < 0) return 0; + root_tag(stanza, tag, sizeof tag); + if (strcmp(tag, want) == 0) return 1; + if (strcmp(tag, "failure") == 0 || + strcmp(tag, "stream:error") == 0) return 0; + } + return 0; +} + +/* ------------------------------------------------------------------ */ +/* per-room chat history */ +/* ------------------------------------------------------------------ */ + +static void history_path(const char *room, char *out, size_t osz) { + snprintf(out, osz, "%s/%s.log", HISTORY_DIR, room); + for (char *p = out + strlen(HISTORY_DIR) + 1; *p; p++) + if (*p == '/') *p = '_'; +} + +static void history_append(const char *room, const char *sender, + const char *body) { + char path[512]; + history_path(room, path, sizeof path); + FILE *f = fopen(path, "a"); + if (!f) return; + fprintf(f, "%ld\t%s\t", (long)time(NULL), sender); + for (const char *p = body; *p; p++) + fputc((*p == '\n' || *p == '\r' || *p == '\t') ? ' ' : *p, f); + fputc('\n', f); + fclose(f); +} + +/* Build conversation context for the LLM. If cutoff != 0, includes + * messages at or after that time; otherwise the last HISTORY_CONTEXT + * messages. Returns 1 if any messages were found. */ +static int history_context(const char *room, time_t cutoff, + char *out, size_t osz) { + static char fb[CTX_FILEBUF]; + static char *lines[CTX_MAXLINES]; + + char path[512]; + history_path(room, path, sizeof path); + FILE *f = fopen(path, "rb"); + if (!f) { out[0] = '\0'; return 0; } + + fseek(f, 0, SEEK_END); + long fsz = ftell(f); + long start = 0; + if (fsz > CTX_FILEBUF - 1) + start = fsz - (CTX_FILEBUF - 1); + fseek(f, start, SEEK_SET); + size_t got = fread(fb, 1, CTX_FILEBUF - 1, f); + fclose(f); + fb[got] = '\0'; + + char *lp = fb; + if (start > 0) { /* drop partial first line */ + char *nl = strchr(fb, '\n'); + if (nl) lp = nl + 1; + } + + int nl = 0; + for (char *p = lp; *p && nl < CTX_MAXLINES; ) { + lines[nl++] = p; + char *e = strchr(p, '\n'); + if (!e) break; + *e = '\0'; + p = e + 1; + } + + int from; + if (cutoff == 0) { + from = nl > HISTORY_CONTEXT ? nl - HISTORY_CONTEXT : 0; + } else { + from = nl; + for (int i = 0; i < nl; i++) { + if (atol(lines[i]) >= (long)cutoff) { from = i; break; } + } + if (nl - from > 600) from = nl - 600; + } + if (nl - from <= 0) { out[0] = '\0'; return 0; } + + size_t o = 0; + if (cutoff == 0) { + o += (size_t)snprintf(out + o, osz - o, + "(last %d messages)\n", nl - from); + } else { + char tb[32]; + time_t cc = cutoff; + strftime(tb, sizeof tb, "%Y-%m-%d %H:%M", localtime(&cc)); + o += (size_t)snprintf(out + o, osz - o, "(since %s)\n", tb); + } + + for (int i = from; i < nl && o + 1 < osz; i++) { + char *l = lines[i]; + char *t1 = strchr(l, '\t'); + if (!t1) continue; + *t1 = '\0'; + char *sender = t1 + 1; + char *t2 = strchr(sender, '\t'); + if (!t2) continue; + *t2 = '\0'; + char *body = t2 + 1; + time_t ts = (time_t)atol(l); + char tb[24]; + strftime(tb, sizeof tb, "%Y-%m-%d %H:%M", localtime(&ts)); + int w = snprintf(out + o, osz - o, "[%s] %s: %s\n", + tb, sender, body); + if (w < 0 || (size_t)w >= osz - o) { out[o] = '\0'; break; } + o += (size_t)w; + } + out[o] = '\0'; + return 1; +} + +/* ------------------------------------------------------------------ */ +/* LLM HTTP call */ +/* ------------------------------------------------------------------ */ + +/* POST a prompt to the LLM API. On success returns 1 and the reply is + * in out. On failure returns 0 and out holds an error message. */ +static int llm_call(const char *user_prompt, char *out, size_t osz) { + static char esc_instr[ESCBUF]; + static char esc_prompt[ESCBUF]; + static char jsonbody[JSONBUF]; + static char req[REQBUF]; + static char resp[HTTPBUF]; + static char httpbody[HTTPBUF]; + + json_escape(esc_instr, sizeof esc_instr, g_instruction); + json_escape(esc_prompt, sizeof esc_prompt, user_prompt); + + int blen = snprintf(jsonbody, sizeof jsonbody, + "{\"model\":\"%s\",\"messages\":[" + "{\"role\":\"system\",\"content\":\"%s\"}," + "{\"role\":\"user\",\"content\":\"%s\"}]," + "\"max_completion_tokens\":%d}", + LLM_MODEL, esc_instr, esc_prompt, LLM_MAX_TOKENS); + if (blen < 0 || blen >= (int)sizeof jsonbody) { + snprintf(out, osz, "Error: request too large"); + return 0; + } + + int fd = tcp_connect(LLM_HOST, LLM_PORT); + if (fd < 0) { + snprintf(out, osz, "Error: cannot reach %s", LLM_HOST); + return 0; + } + SSL *ssl = tls_wrap(fd, LLM_HOST); + if (!ssl) { + close(fd); + snprintf(out, osz, "Error: TLS handshake failed"); + return 0; + } + Conn lc = { fd, ssl }; + + int rlen = snprintf(req, sizeof req, + "POST %s HTTP/1.1\r\n" + "Host: %s\r\n" + "Authorization: Bearer %s\r\n" + "Content-Type: application/json\r\n" + "Content-Length: %d\r\n" + "Connection: close\r\n" + "\r\n%s", + LLM_PATH, LLM_HOST, g_apikey, blen, jsonbody); + if (rlen < 0 || rlen >= (int)sizeof req || + conn_write(&lc, req, (size_t)rlen) < 0) { + SSL_free(ssl); + close(fd); + snprintf(out, osz, "Error: failed to send request"); + return 0; + } + + size_t rl = 0; + int n; + while (rl < sizeof resp - 1 && + (n = conn_read_some(&lc, resp + rl, sizeof resp - 1 - rl)) > 0) + rl += (size_t)n; + resp[rl] = '\0'; + SSL_shutdown(ssl); + SSL_free(ssl); + close(fd); + + char *hdrend = strstr(resp, "\r\n\r\n"); + if (!hdrend) { + snprintf(out, osz, "Error: bad HTTP response"); + return 0; + } + *hdrend = '\0'; + char *bodyp = hdrend + 4; + int status = 0; + sscanf(resp, "%*s %d", &status); + + const char *bd; + if (ci_strstr(resp, "transfer-encoding: chunked")) { + dechunk(bodyp, httpbody, sizeof httpbody); + bd = httpbody; + } else { + bd = bodyp; + } + + if (status != 200) { + char detail[512]; + if (!json_get_string(bd, "message", detail, sizeof detail)) + snprintf(detail, sizeof detail, "HTTP %d", status); + snprintf(out, osz, "Error: LLM API: %s", detail); + return 0; + } + + const char *m = strstr(bd, "\"message\""); + if (!json_get_string(m ? m : bd, "content", out, osz) || !out[0]) { + snprintf(out, osz, "Error: could not parse LLM response"); + return 0; + } + return 1; +} + +/* ------------------------------------------------------------------ */ +/* commands */ +/* ------------------------------------------------------------------ */ + +static char g_ctxbuf[CTXBUF]; +static char g_prompt[PROMPTBUF]; +static char g_reply[LLMOUT]; +static char g_out[LLMOUT + 64]; + +/* Parse a period token (1h/1d/2w/1m). Returns a cutoff time, or 0. */ +static time_t parse_period(const char *tok) { + if (!tok || !*tok) return 0; + long num = 0; + int any = 0; + const char *p = tok; + while (*p >= '0' && *p <= '9') { num = num * 10 + (*p - '0'); p++; any = 1; } + if (!any || !p[0] || p[1]) return 0; + long secs; + switch (*p) { + case 'h': secs = 3600; break; + case 'd': secs = 86400; break; + case 'w': secs = 604800; break; + case 'm': secs = 86400L * 30; break; + default: return 0; + } + return time(NULL) - num * secs; +} + +static const char *help_text(void) { + return + "Commands:\n" + " " CMD_PREFIX " summarize [period] - summarise the chat\n" + " " CMD_PREFIX " tasks [period] - extract action items\n" + " " CMD_PREFIX " <question> - ask anything\n" + "\n" + " Period: 1h 1d 2d 1w 2w 1m (default: last 50 messages)"; +} + +static void muc_send(Conn *c, const char *room, const char *text) { + static char esc[ESCBUF]; + static char msg[ESCBUF + 256]; + xml_escape(esc, sizeof esc, text); + snprintf(msg, sizeof msg, + "<message to='%s' type='groupchat'><body>%s</body></message>", + room, esc); + conn_puts(c, msg); +} + +static void cmd_summarize(Conn *c, const char *room, time_t cutoff) { + if (!history_context(room, cutoff, g_ctxbuf, sizeof g_ctxbuf)) { + muc_send(c, room, "No messages found for that time period."); + return; + } + snprintf(g_prompt, sizeof g_prompt, + "Summarize this group chat concisely. Include key points " + "and decisions:\n\n%s", g_ctxbuf); + if (llm_call(g_prompt, g_reply, sizeof g_reply)) { + snprintf(g_out, sizeof g_out, "Summary:\n%s", g_reply); + muc_send(c, room, g_out); + } else { + muc_send(c, room, g_reply); + } +} + +static void cmd_tasks(Conn *c, const char *room, time_t cutoff) { + if (!history_context(room, cutoff, g_ctxbuf, sizeof g_ctxbuf)) { + muc_send(c, room, "No messages found for that time period."); + return; + } + snprintf(g_prompt, sizeof g_prompt, + "Extract all action items and tasks from this conversation. " + "For each task include: who is responsible (if mentioned), " + "what needs to be done, and any deadlines mentioned. Be " + "concise:\n\n%s", g_ctxbuf); + if (llm_call(g_prompt, g_reply, sizeof g_reply)) { + snprintf(g_out, sizeof g_out, "Tasks:\n%s", g_reply); + muc_send(c, room, g_out); + } else { + muc_send(c, room, g_reply); + } +} + +static void cmd_ask(Conn *c, const char *room, const char *question) { + if (!question[0]) { + muc_send(c, room, "What would you like to ask?"); + return; + } + history_context(room, 0, g_ctxbuf, sizeof g_ctxbuf); + snprintf(g_prompt, sizeof g_prompt, + "You are a helpful assistant in a group chat.\n\n" + "Recent conversation:\n%s\n\n" + "Question: %s\n\n" + "Answer concisely and helpfully. If the answer is in the " + "conversation, reference it.", g_ctxbuf, question); + if (llm_call(g_prompt, g_reply, sizeof g_reply)) + muc_send(c, room, g_reply); + else + muc_send(c, room, g_reply); +} + +/* Parse a @bot command out of a chat message and run it. */ +static void handle_command(Conn *c, const char *room, const char *body) { + while (*body == ' ' || *body == '\t') body++; + size_t pl = strlen(CMD_PREFIX); + if (strncasecmp(body, CMD_PREFIX, pl) != 0) return; + + char rest[4096]; + snprintf(rest, sizeof rest, "%s", body + pl); + char *r = rest; + while (*r == ' ' || *r == '\t') r++; + memmove(rest, r, strlen(r) + 1); + rtrim(rest); + if (!rest[0]) { muc_send(c, room, help_text()); return; } + + char low[4096]; + size_t i; + for (i = 0; rest[i]; i++) + low[i] = (char)tolower((unsigned char)rest[i]); + low[i] = '\0'; + + /* last whitespace-delimited token of `low` */ + int e = (int)strlen(low); + int s = e; + while (s > 0 && !isspace((unsigned char)low[s - 1])) s--; + char lasttok[64]; + int ltl = e - s; + if (ltl > 63) ltl = 63; + memcpy(lasttok, low + s, (size_t)ltl); + lasttok[ltl] = '\0'; + time_t cutoff = parse_period(lasttok); + + char cmd[4096]; + if (cutoff) { + int ce = s; + while (ce > 0 && isspace((unsigned char)low[ce - 1])) ce--; + memcpy(cmd, low, (size_t)ce); + cmd[ce] = '\0'; + } else { + snprintf(cmd, sizeof cmd, "%s", low); + } + + if (!cmd[0] && !cutoff) { muc_send(c, room, help_text()); return; } + + if (!strcmp(cmd, "summarize") || !strcmp(cmd, "summary") || + !strcmp(cmd, "s")) { + cmd_summarize(c, room, cutoff); + } else if (!strcmp(cmd, "tasks") || !strcmp(cmd, "task") || + !strcmp(cmd, "t")) { + cmd_tasks(c, room, cutoff); + } else if (!strcmp(cmd, "help")) { + muc_send(c, room, help_text()); + } else if (cmd[0]) { + char q[4096]; + if (cutoff) { /* drop the period token */ + int re = (int)strlen(rest); + int rs = re; + while (rs > 0 && !isspace((unsigned char)rest[rs - 1])) rs--; + while (rs > 0 && isspace((unsigned char)rest[rs - 1])) rs--; + memcpy(q, rest, (size_t)rs); + q[rs] = '\0'; + } else { + snprintf(q, sizeof q, "%s", rest); + } + cmd_ask(c, room, q); + } else { + muc_send(c, room, help_text()); + } +} + +/* ------------------------------------------------------------------ */ +/* stanza dispatch */ +/* ------------------------------------------------------------------ */ + +static void dispatch(Conn *c, const char *st) { + char tag[64]; + root_tag(st, tag, sizeof tag); + + if (strcmp(tag, "message") == 0) { + char type[32]; + attr_get(st, "type", type, sizeof type); + if (strcmp(type, "groupchat") != 0) return; + if (strstr(st, "<delay")) return; /* history replay */ + + char from[384]; + if (!attr_get(st, "from", from, sizeof from)) return; + char *slash = strchr(from, '/'); + if (!slash || !slash[1]) return; /* room-level msg */ + *slash = '\0'; + const char *room = from; + const char *nick = slash + 1; + if (strcmp(nick, XMPP_NICK) == 0) return; /* our own message */ + + char raw[BODYBUF]; + if (!elem_text(st, "body", raw, sizeof raw) || !raw[0]) return; + char bdy[BODYBUF]; + xml_unescape(bdy, sizeof bdy, raw); + + history_append(room, nick, bdy); + handle_command(c, room, bdy); + + } else if (strcmp(tag, "iq") == 0) { + if (strstr(st, "urn:xmpp:ping")) { /* answer pings */ + char id[64], from[384]; + attr_get(st, "id", id, sizeof id); + attr_get(st, "from", from, sizeof from); + char reply[512]; + snprintf(reply, sizeof reply, + "<iq type='result' to='%s' id='%s'/>", from, id); + conn_puts(c, reply); + } + } +} + +/* ------------------------------------------------------------------ */ +/* XMPP session */ +/* ------------------------------------------------------------------ */ + +static void send_stream_header(Conn *c) { + char buf[512]; + snprintf(buf, sizeof buf, + "<?xml version='1.0'?>" + "<stream:stream xmlns='jabber:client' " + "xmlns:stream='http://etherx.jabber.org/streams' " + "to='%s' version='1.0'>", g_domain); + conn_puts(c, buf); +} + +static void send_auth(Conn *c) { + char plain[1024]; + size_t pl = 0; + plain[pl++] = '\0'; + size_t ll = strlen(g_localpart), pwl = strlen(g_password); + memcpy(plain + pl, g_localpart, ll); pl += ll; + plain[pl++] = '\0'; + memcpy(plain + pl, g_password, pwl); pl += pwl; + + char b64[2048]; + b64encode(b64, (unsigned char *)plain, pl); + char msg[2400]; + snprintf(msg, sizeof msg, + "<auth xmlns='urn:ietf:params:xml:ns:xmpp-sasl' " + "mechanism='PLAIN'>%s</auth>", b64); + conn_puts(c, msg); +} + +/* Run one full XMPP session. Returns a SESS_* code. */ +static int run_session(void) { + inlen = 0; + + int fd = tcp_connect(XMPP_HOST, XMPP_PORT); + if (fd < 0) { + fprintf(stderr, "xmppcb: connect to %s:%d failed\n", + XMPP_HOST, XMPP_PORT); + return SESS_RETRY; + } + Conn c = { fd, NULL }; + + /* stream + STARTTLS negotiation (plaintext) */ + send_stream_header(&c); + if (!await_tag(&c, "stream:features")) goto drop; + if (!strstr(stanza, "starttls")) { + fprintf(stderr, "xmppcb: server did not offer STARTTLS\n"); + goto drop; + } + conn_puts(&c, "<starttls xmlns='urn:ietf:params:xml:ns:xmpp-tls'/>"); + if (!await_tag(&c, "proceed")) goto drop; + + /* upgrade to TLS */ + inlen = 0; + c.ssl = tls_wrap(fd, g_domain); + if (!c.ssl) goto drop; + + /* restart stream, then SASL PLAIN */ + send_stream_header(&c); + if (!await_tag(&c, "stream:features")) goto drop_tls; + if (!strstr(stanza, "PLAIN")) { + fprintf(stderr, "xmppcb: server did not offer SASL PLAIN\n"); + goto drop_tls; + } + send_auth(&c); + if (!await_tag(&c, "success")) { + fprintf(stderr, "xmppcb: authentication failed\n"); + SSL_free(c.ssl); + close(fd); + return SESS_AUTHFAIL; + } + + /* restart stream, then resource bind */ + send_stream_header(&c); + if (!await_tag(&c, "stream:features")) goto drop_tls; + conn_puts(&c, + "<iq type='set' id='bind1'>" + "<bind xmlns='urn:ietf:params:xml:ns:xmpp-bind'>" + "<resource>xmppcb</resource></bind></iq>"); + if (!await_tag(&c, "iq")) goto drop_tls; + + /* online, join rooms */ + conn_puts(&c, "<presence/>"); + for (int i = 0; i < NROOMS; i++) { + char p[512]; + snprintf(p, sizeof p, + "<presence to='%s/%s'>" + "<x xmlns='http://jabber.org/protocol/muc'>" + "<history maxstanzas='0'/></x></presence>", + ROOMS[i], XMPP_NICK); + conn_puts(&c, p); + fprintf(stderr, "xmppcb: joining %s\n", ROOMS[i]); + } + fprintf(stderr, "xmppcb: online as %s\n", XMPP_JID); + + /* main loop */ + struct pollfd pfd = { fd, POLLIN, 0 }; + for (;;) { + while (extract_chunk(stanza, sizeof stanza) > 0) { + if (strncmp(stanza, "</", 2) == 0) goto drop_tls; + char tag[64]; + root_tag(stanza, tag, sizeof tag); + if (strcmp(tag, "stream:error") == 0) goto drop_tls; + dispatch(&c, stanza); + } + int pr = poll(&pfd, 1, KEEPALIVE_SEC * 1000); + if (pr < 0) { + if (errno == EINTR) continue; + goto drop_tls; + } + if (pr == 0) { /* keepalive */ + if (conn_write(&c, " ", 1) < 0) goto drop_tls; + continue; + } + if (inlen >= sizeof inbuf) goto drop_tls; + int n = conn_read_some(&c, inbuf + inlen, sizeof inbuf - inlen); + if (n <= 0) goto drop_tls; + inlen += (size_t)n; + while (SSL_pending(c.ssl) > 0 && inlen < sizeof inbuf) { + n = conn_read_some(&c, inbuf + inlen, sizeof inbuf - inlen); + if (n <= 0) break; + inlen += (size_t)n; + } + } + +drop_tls: + if (c.ssl) { SSL_shutdown(c.ssl); SSL_free(c.ssl); } + close(fd); + return SESS_RETRY; +drop: + close(fd); + return SESS_RETRY; +} + +/* ------------------------------------------------------------------ */ +/* main */ +/* ------------------------------------------------------------------ */ + +int main(void) { + signal(SIGPIPE, SIG_IGN); + + const char *at = strchr(XMPP_JID, '@'); + if (!at || !at[1]) die("XMPP_JID must be user@domain"); + size_t ll = (size_t)(at - XMPP_JID); + if (ll == 0) die("XMPP_JID must be user@domain"); + if (ll >= sizeof g_localpart) die("JID localpart too long"); + memcpy(g_localpart, XMPP_JID, ll); + g_localpart[ll] = '\0'; + snprintf(g_domain, sizeof g_domain, "%s", at + 1); + + g_password = getenv("XMPP_PASSWORD"); + g_apikey = getenv("LLM_API_KEY"); + if (!g_password || !*g_password) + die("set the XMPP_PASSWORD environment variable"); + if (!g_apikey || !*g_apikey) + die("set the LLM_API_KEY environment variable"); + + FILE *f = fopen(INSTRUCTION_FILE, "rb"); + if (f) { + size_t n = fread(g_instruction, 1, sizeof g_instruction - 1, f); + g_instruction[n] = '\0'; + fclose(f); + } else { + fprintf(stderr, "xmppcb: warning: %s not found, " + "using empty system prompt\n", INSTRUCTION_FILE); + g_instruction[0] = '\0'; + } + + if (mkdir(HISTORY_DIR, 0700) != 0 && errno != EEXIST) + die("cannot create history directory"); + + tls_init(); + + for (;;) { + int r = run_session(); + if (r == SESS_AUTHFAIL) return 1; + fprintf(stderr, "xmppcb: disconnected, reconnecting in 5s\n"); + sleep(5); + } +} |