vatrog-vm-signaling/src/discovery/linux/host_probe.c

/* host_probe.c — the default Proxmox host-probe (see host_probe.h). The ONLY TU that knows
 * /etc/pve/qemu-server, the QMP socket path convention, and `info mtree`. Pure libc +
 * AF_UNIX + files; no vmie/vmctl. config() is cheap+local; live() does a bounded blocking
 * QMP round-trip (query-status + info mtree) and is fail-closed: anything it cannot confirm
 * leaves ok=0 (the VM is not brought up rather than guessed). */
#define _GNU_SOURCE
#include "host_probe.h"
#include "vmsig_event.h"          /* VMSIG_VM_* */
#include <stdlib.h>
#include <string.h>
#include <stdio.h>
#include <stdint.h>
#include <unistd.h>
#include <fcntl.h>
#include <sys/stat.h>
#include <sys/socket.h>
#include <sys/un.h>
#include <stddef.h>
#include <errno.h>

typedef struct {
    const char* watch_dir;   /* /dev/shm/vmsig            */
    const char* pve_conf;    /* /etc/pve/qemu-server      */
    const char* qmp_dir;     /* /var/run/qemu-server      */
} hp_cfg;

/* ---- /etc/pve config (stage 1) ----------------------------------------------- */

/* Read a whole small file into a heap buffer (NUL-terminated). NULL on error/oversize. */
static char* read_file(const char* path, size_t cap) {
    int fd = open(path, O_RDONLY | O_CLOEXEC);
    if (fd < 0) return NULL;
    char* buf = malloc(cap + 1);
    if (!buf) { close(fd); return NULL; }
    size_t got = 0;
    for (;;) {
        ssize_t n = read(fd, buf + got, cap - got);
        if (n < 0) { if (errno == EINTR) continue; free(buf); close(fd); return NULL; }
        if (n == 0) break;
        got += (size_t)n;
        if (got >= cap) break;
    }
    close(fd);
    buf[got] = 0;
    return buf;
}

/* Value of a top-level "key:" line (Proxmox ini), copied trimmed into out. 1 if found. */
static int conf_val(const char* conf, const char* key, char* out, size_t cap) {
    size_t klen = strlen(key);
    const char* p = conf;
    while (p && *p) {
        const char* line = p;
        const char* nl = strchr(p, '\n');
        size_t llen = nl ? (size_t)(nl - line) : strlen(line);
        if (llen > klen && strncmp(line, key, klen) == 0 && line[klen] == ':') {
            const char* v = line + klen + 1;
            while (*v == ' ' || *v == '\t') v++;
            size_t vlen = (size_t)((line + llen) - v);
            while (vlen && (v[vlen-1] == ' ' || v[vlen-1] == '\t' || v[vlen-1] == '\r')) vlen--;
            if (vlen >= cap) vlen = cap - 1;
            memcpy(out, v, vlen); out[vlen] = 0;
            return 1;
        }
        p = nl ? nl + 1 : NULL;
    }
    return 0;
}

static int hp_config(const struct vmsig_host_probe* p, uint32_t vmid, vmsig_host_facts* out) {
    const hp_cfg* c = p->ud;
    memset(out, 0, sizeof *out);
    out->vmid = vmid;

    snprintf(out->ram_path, sizeof out->ram_path, "%s/vm-%u-ram", c->watch_dir, vmid);
    snprintf(out->qmp_path, sizeof out->qmp_path, "%s/%u.qmp", c->qmp_dir, vmid);

    char path[VMSIG_HF_PATH_MAX + 32];
    snprintf(path, sizeof path, "%s/%u.conf", c->pve_conf, vmid);
    char* conf = read_file(path, 64 * 1024);
    if (!conf) { out->ok = 0; return 0; }     /* no host config => not a known VM */

    char tmp[VMSIG_HF_NAME_MAX];
    if (conf_val(conf, "name", out->name, sizeof out->name) == 0)
        snprintf(out->name, sizeof out->name, "vm-%u", vmid);
    if (conf_val(conf, "memory", tmp, sizeof tmp))
        out->cfg_ram_bytes = (uint64_t)strtoull(tmp, NULL, 10) * 1024ull * 1024ull;

    /* share=on is mandatory: without it the host mmap is a private copy, not guest RAM. */
    out->share_on = (strstr(conf, "share=on") != NULL) ? 1 : 0;
    free(conf);

    out->ok = out->share_on ? 1 : 0;          /* config-level pass; liveness is stage 2 */
    return 0;
}

/* ---- QMP liveness + mtree low (stage 2) -------------------------------------- */

static int qmp_connect(const char* path) {
    int fd = socket(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0);
    if (fd < 0) return -1;
    struct timeval tv = { .tv_sec = 0, .tv_usec = 250000 };   /* 250ms bound on each recv */
    setsockopt(fd, SOL_SOCKET, SO_RCVTIMEO, &tv, sizeof tv);
    setsockopt(fd, SOL_SOCKET, SO_SNDTIMEO, &tv, sizeof tv);

    struct sockaddr_un a;
    memset(&a, 0, sizeof a);
    a.sun_family = AF_UNIX;
    size_t n = strlen(path);
    socklen_t alen;
    if (path[0] == '@') {                       /* abstract namespace */
        if (n > sizeof a.sun_path) { close(fd); return -1; }
        a.sun_path[0] = 0;
        memcpy(a.sun_path + 1, path + 1, n - 1);
        alen = (socklen_t)(offsetof(struct sockaddr_un, sun_path) + n);
    } else {
        if (n >= sizeof a.sun_path) { close(fd); return -1; }
        memcpy(a.sun_path, path, n);
        alen = (socklen_t)(offsetof(struct sockaddr_un, sun_path) + n + 1);
    }
    if (connect(fd, (struct sockaddr*)&a, alen) < 0) { close(fd); return -1; }
    return fd;
}

/* Read ONE '\n'-terminated QMP message into buf (QMP frames each JSON object on a line;
 * an HMP string return keeps its newlines escaped, so it is still a single line). 1 / 0 / -1. */
static int qmp_read_line(int fd, char* buf, size_t cap, size_t* out_len) {
    size_t got = 0;
    while (got + 1 < cap) {
        ssize_t r = read(fd, buf + got, cap - 1 - got);
        if (r < 0) { if (errno == EINTR) continue; return -1; }   /* timeout/error */
        if (r == 0) return (got > 0) ? 1 : 0;
        got += (size_t)r;
        char* nl = memchr(buf, '\n', got);
        if (nl) { *out_len = got; buf[got] = 0; return 1; }
    }
    *out_len = got; buf[got] = 0;
    return 1;   /* line longer than cap: truncated but usable for our scans */
}

/* Read messages until one carries "return"/"error", skipping async "event"s. 1 if a return,
 * 0 if an error/closed, -1 on transport error. The matched message is left in buf. */
static int qmp_await_return(int fd, char* buf, size_t cap) {
    for (int i = 0; i < 64; i++) {
        size_t len = 0;
        int r = qmp_read_line(fd, buf, cap, &len);
        if (r <= 0) return r;
        if (strstr(buf, "\"error\"")) return 0;
        if (strstr(buf, "\"return\"")) return 1;
        /* greeting {"QMP":...} or async {"event":...} -> keep reading */
    }
    return -1;
}

static int qmp_cmd(int fd, const char* json, char* buf, size_t cap) {
    size_t n = strlen(json);
    if (write(fd, json, n) != (ssize_t)n) return -1;
    return qmp_await_return(fd, buf, cap);
}

/* Map a QEMU query-status "status" word to VMSIG_VM_*. Alive = running|paused. */
static int qmp_status_word(const char* buf) {
    const char* s = strstr(buf, "\"status\"");
    if (!s) return VMSIG_VM_UNKNOWN;
    s = strchr(s, ':'); if (!s) return VMSIG_VM_UNKNOWN;
    s = strchr(s, '"'); if (!s) return VMSIG_VM_UNKNOWN;
    s++;
    if (!strncmp(s, "running", 7))   return VMSIG_VM_RUNNING;
    if (!strncmp(s, "paused", 6))    return VMSIG_VM_PAUSED;
    if (!strncmp(s, "prelaunch", 9)) return VMSIG_VM_PAUSED;
    if (!strncmp(s, "shutdown", 8))  return VMSIG_VM_SHUTDOWN;
    if (!strncmp(s, "guest-panicked", 14) || !strncmp(s, "internal-error", 14))
        return VMSIG_VM_CRASHED;
    return VMSIG_VM_UNKNOWN;
}

/* Derive the below-4G split from `info mtree` text: the size of the RAM region whose guest
 * physical range starts at address 0. Standard QEMU split-RAM layout puts low RAM at
 * [0, low) and high RAM above 4G at file offset @low. FAIL-CLOSED: 0 if not found.
 * NOTE: parses HMP text (not a stable QMP schema) — verify against real `info mtree` output. */
static uint64_t mtree_low(const char* ret) {
    /* The return is a JSON string; lines inside are escaped "\n". Scan for the GPA-0 ram run:
     *   "    0000000000000000-<end16> (prio N, ram): ..." */
    const char* p = ret;
    while ((p = strstr(p, "0000000000000000-")) != NULL) {
        const char* end_hex = p + 17;          /* 16 zeros + '-' */
        char* stop = NULL;
        unsigned long long end = strtoull(end_hex, &stop, 16);
        /* the descriptor after the range must mark it RAM (not the i/o "system" root) */
        const char* tail = stop ? stop : end_hex;
        const char* nl = strstr(tail, "\\n");
        const char* lim = nl ? nl : (tail + 64);
        int is_ram = 0;
        for (const char* q = tail; q < lim && *q; q++)
            if (!strncmp(q, "ram)", 4)) { is_ram = 1; break; }
        if (is_ram && end > 0 && end != ~0ull) return end + 1ull;   /* [0, end] => low=end+1 */
        p = end_hex;
    }
    return 0;
}

static int hp_live(const struct vmsig_host_probe* p, vmsig_host_facts* io) {
    (void)p;
    io->retry = 0;
    int fd = qmp_connect(io->qmp_path);
    if (fd < 0) { io->retry = 1; io->ok = 0; return 0; }   /* QMP not up yet => transient */

    char* buf = malloc(256 * 1024);
    if (!buf) { close(fd); io->retry = 1; io->ok = 0; return 0; }

    int alive = 0;
    if (qmp_cmd(fd, "{\"execute\":\"qmp_capabilities\"}\n", buf, 256 * 1024) == 1 &&
        qmp_cmd(fd, "{\"execute\":\"query-status\"}\n",     buf, 256 * 1024) == 1) {
        io->vm_state = qmp_status_word(buf);
        alive = (io->vm_state == VMSIG_VM_RUNNING || io->vm_state == VMSIG_VM_PAUSED);
    } else {
        io->retry = 1;                          /* handshake failed mid-way => transient */
    }

    if (alive) {
        if (qmp_cmd(fd,
            "{\"execute\":\"human-monitor-command\","
            "\"arguments\":{\"command-line\":\"info mtree -f\"}}\n", buf, 256 * 1024) == 1) {
            io->low = mtree_low(buf);
        }
    }

    free(buf);
    close(fd);

    /* fail-closed: alive AND a parsed split => bring up; else not (stale / unparsable). */
    io->ok = (alive && io->low != 0) ? 1 : 0;
    return 0;
}

vmsig_host_probe host_probe_proxmox(const char* watch_dir, const char* pve_conf,
                                    const char* qmp_dir) {
    static hp_cfg cfg;   /* single daemon-wide probe; paths are process-lifetime strings */
    cfg.watch_dir = watch_dir;
    cfg.pve_conf  = pve_conf;
    cfg.qmp_dir   = qmp_dir;
    vmsig_host_probe p = { hp_config, hp_live, &cfg };
    return p;
}