yakpanel-core/YakPanel-server/backend/app/api/ssl.py

"""YakPanel - SSL/Domains API - Let's Encrypt via certbot"""
import os
import re
import shutil
import socket
import subprocess
import sys
import tempfile
from fastapi import APIRouter, Depends, HTTPException
from sqlalchemy.ext.asyncio import AsyncSession
from sqlalchemy import select
from pydantic import BaseModel
from typing import Optional

from app.core.database import get_db
from app.core.config import get_runtime_config
from app.core.utils import environment_with_system_path, exec_shell_sync, read_file, nginx_reload_all_known, nginx_binary_candidates
from app.api.auth import get_current_user
from app.models.user import User
from app.models.site import Site, Domain
from app.services.site_service import regenerate_site_vhost

router = APIRouter(prefix="/ssl", tags=["ssl"])

_CERTBOT_PATH_CANDIDATES = (
    "/usr/bin/certbot",
    "/usr/local/bin/certbot",
    "/snap/bin/certbot",
)


def _certbot_command() -> list[str] | None:
    """Resolve argv prefix to run certbot: [binary] or [python, -m, certbot]."""
    env = environment_with_system_path()
    path_var = env.get("PATH", "")

    exe = getattr(sys, "executable", None) or ""
    if exe and os.path.isfile(exe):
        try:
            r = subprocess.run(
                [exe, "-m", "certbot", "--version"],
                capture_output=True,
                text=True,
                timeout=20,
                env=env,
            )
            if r.returncode == 0:
                return [exe, "-m", "certbot"]
        except (FileNotFoundError, OSError, subprocess.TimeoutExpired):
            pass

    tried: list[str] = []
    w = shutil.which("certbot", path=path_var)
    if w and os.path.isfile(w):
        tried.append(w)

    for p in _CERTBOT_PATH_CANDIDATES:
        if p not in tried and os.path.isfile(p):
            tried.append(p)

    for exe in tried:
        try:
            r = subprocess.run(
                [exe, "--version"],
                capture_output=True,
                text=True,
                timeout=15,
                env=env,
            )
            if r.returncode == 0:
                return [exe]
        except (FileNotFoundError, OSError, subprocess.TimeoutExpired):
            continue

    for py_name in ("python3", "python"):
        py = shutil.which(py_name, path=path_var)
        if not py or not os.path.isfile(py):
            continue
        try:
            r = subprocess.run(
                [py, "-m", "certbot", "--version"],
                capture_output=True,
                text=True,
                timeout=20,
                env=env,
            )
            if r.returncode == 0:
                return [py, "-m", "certbot"]
        except (FileNotFoundError, OSError, subprocess.TimeoutExpired):
            continue

    return None


def _certbot_missing_message() -> str:
    return (
        "certbot is not installed or not reachable from the panel process. "
        "On the server, run one of: apt install certbot | dnf install certbot | yum install certbot | snap install certbot. "
        "Alternatively: pip install certbot (panel can use python3 -m certbot). "
        "If certbot is already installed, ensure /usr/bin is on PATH for the YakPanel service."
    )


async def _le_hostnames_for_domain_row(db: AsyncSession, dom_row: Optional[Domain], primary: str) -> list[str]:
    """All distinct hostnames for the site (for -d flags). Falls back to primary."""
    if not dom_row:
        return [primary] if primary else []
    result = await db.execute(select(Domain).where(Domain.pid == dom_row.pid).order_by(Domain.id))
    rows = result.scalars().all()
    seen: set[str] = set()
    out: list[str] = []
    for d in rows:
        n = (d.name or "").strip()
        if not n:
            continue
        key = n.lower()
        if key not in seen:
            seen.add(key)
            out.append(n)
    if primary and primary.lower() not in seen:
        out.insert(0, primary)
    return out if out else ([primary] if primary else [])


def _reload_panel_and_common_nginx() -> tuple[bool, str]:
    """Reload nginx so new vhost (ACME path) is live before certbot HTTP-01."""
    return nginx_reload_all_known(timeout=60)


def _localhost_accepts_tcp(port: int, timeout: float = 2.0) -> bool:
    """True if something accepts a TCP connection on this machine (checks IPv4 loopback)."""
    try:
        with socket.create_connection(("127.0.0.1", port), timeout=timeout):
            return True
    except OSError:
        return False


def _ss_reports_listen_443() -> bool | None:
    """Parse ss/netstat output; None if the probe could not run."""
    out, _ = exec_shell_sync("ss -tln 2>/dev/null || netstat -tln 2>/dev/null", timeout=5)
    if not out or not out.strip():
        return None
    return bool(re.search(r":443\b", out))


@router.get("/domains")
async def ssl_domains(
    current_user: User = Depends(get_current_user),
    db: AsyncSession = Depends(get_db),
):
    """List all domains from sites with site path for certbot webroot"""
    result = await db.execute(
        select(Domain, Site).join(Site, Domain.pid == Site.id).order_by(Domain.name)
    )
    rows = result.all()
    return [
        {
            "id": d.id,
            "name": d.name,
            "port": d.port,
            "site_id": s.id,
            "site_name": s.name,
            "site_path": s.path,
        }
        for d, s in rows
    ]


class RequestCertRequest(BaseModel):
    domain: str
    webroot: str
    email: str


@router.post("/request")
async def ssl_request_cert(
    body: RequestCertRequest,
    current_user: User = Depends(get_current_user),
    db: AsyncSession = Depends(get_db),
):
    """Request Let's Encrypt certificate via certbot (webroot challenge)."""
    if not body.domain or not body.webroot or not body.email:
        raise HTTPException(status_code=400, detail="domain, webroot and email required")
    if ".." in body.domain or ".." in body.webroot:
        raise HTTPException(status_code=400, detail="Invalid path")
    cfg = get_runtime_config()
    allowed = [os.path.abspath(cfg["www_root"]), os.path.abspath(cfg["setup_path"])]
    webroot_abs = os.path.abspath(body.webroot)
    if not any(webroot_abs.startswith(a + os.sep) or webroot_abs == a for a in allowed):
        raise HTTPException(status_code=400, detail="Webroot must be under www_root or setup_path")

    dom = body.domain.split(":")[0].strip()
    webroot_norm = webroot_abs.rstrip(os.sep)

    result_dom = await db.execute(select(Domain).where(Domain.name == dom).limit(1))
    dom_row = result_dom.scalar_one_or_none()
    if dom_row:
        regen_pre = await regenerate_site_vhost(db, dom_row.pid)
        if not regen_pre.get("status"):
            raise HTTPException(
                status_code=500,
                detail="Cannot refresh nginx vhost before certificate request: " + str(regen_pre.get("msg", "")),
            )
        ok_ngx, err_ngx = _reload_panel_and_common_nginx()
        if not ok_ngx:
            raise HTTPException(
                status_code=500,
                detail="Nginx test/reload failed before certificate request (fix config, then retry): " + err_ngx,
            )

    challenge_dir = os.path.join(webroot_norm, ".well-known", "acme-challenge")
    try:
        os.makedirs(challenge_dir, mode=0o755, exist_ok=True)
    except OSError as e:
        raise HTTPException(status_code=500, detail=f"Cannot create ACME webroot directory: {e}") from e

    prefix = _certbot_command()
    if not prefix:
        raise HTTPException(status_code=500, detail=_certbot_missing_message())

    hostnames = await _le_hostnames_for_domain_row(db, dom_row, dom)

    base_flags = [
        "--non-interactive",
        "--agree-tos",
        "--email",
        body.email,
        "--no-eff-email",
    ]

    cmd_webroot = prefix + ["certonly", "--webroot", "-w", webroot_norm, *base_flags]
    for h in hostnames:
        cmd_webroot.extend(["-d", h])
    cmd_webroot.extend(["--preferred-challenges", "http"])

    cmd_nginx = prefix + ["certonly", "--nginx", *base_flags]
    for h in hostnames:
        cmd_nginx.extend(["-d", h])

    env = environment_with_system_path()
    proc: subprocess.CompletedProcess[str] | None = None
    last_err = ""
    for cmd, label in ((cmd_webroot, "webroot"), (cmd_nginx, "nginx")):
        try:
            proc = subprocess.run(
                cmd,
                capture_output=True,
                text=True,
                timeout=300,
                env=env,
            )
        except FileNotFoundError:
            raise HTTPException(status_code=500, detail=_certbot_missing_message()) from None
        except subprocess.TimeoutExpired:
            raise HTTPException(status_code=500, detail="certbot timed out (300s)") from None
        if proc.returncode == 0:
            break
        chunk = (proc.stderr or proc.stdout or "").strip() or f"exit {proc.returncode}"
        last_err = f"[{label}] {chunk}"

    if proc is None or proc.returncode != 0:
        msg = last_err or "certbot failed"
        hint = (
            " Webroot and nginx plugins both failed. Check: "
            "DNS A/AAAA for every -d name points to this server; port 80 reaches the nginx that serves these hosts; "
            "site is enabled; install python3-certbot-nginx if the nginx method reports a missing plugin. "
            "If you use a CDN proxy, pause it or use DNS validation instead."
        )
        raise HTTPException(status_code=500, detail=(msg + hint)[:8000])

    row = dom_row
    if row:
        regen = await regenerate_site_vhost(db, row.pid)
        if not regen.get("status"):
            return {
                "status": True,
                "msg": "Certificate issued but nginx vhost update failed: " + str(regen.get("msg", "")),
                "output": (proc.stdout or "")[-2000:],
            }

    return {
        "status": True,
        "msg": "Certificate issued and nginx updated",
        "output": (proc.stdout or "")[-2000:],
    }


@router.get("/diagnostics")
async def ssl_diagnostics(current_user: User = Depends(get_current_user)):
    """
    Help debug HTTP vs HTTPS: compares panel-written vhosts with what nginx -T actually loads.
    ERR_CONNECTION_REFUSED on 443 usually means no listen 443 in the active nginx, or a firewall.
    """
    cfg = get_runtime_config()
    setup_abs = os.path.abspath((cfg.get("setup_path") or "").strip() or ".")
    vhost_dir = os.path.join(setup_abs, "panel", "vhost", "nginx")
    include_snippet = "include " + vhost_dir.replace(os.sep, "/") + "/*.conf;"

    vhost_summaries: list[dict] = []
    if os.path.isdir(vhost_dir):
        try:
            names = sorted(os.listdir(vhost_dir))
        except OSError:
            names = []
        for fn in names:
            if not fn.endswith(".conf") or fn.startswith("."):
                continue
            fp = os.path.join(vhost_dir, fn)
            if not os.path.isfile(fp):
                continue
            body = read_file(fp) or ""
            vhost_summaries.append({
                "file": fn,
                "has_listen_80": bool(re.search(r"\blisten\s+80\b", body)),
                "has_listen_443": bool(re.search(r"\blisten\s+.*443", body)),
                "has_ssl_directives": "ssl_certificate" in body,
            })

    any_vhost_443 = any(
        v.get("has_listen_443") and v.get("has_ssl_directives") for v in vhost_summaries
    )
    effective_listen_443 = False
    panel_include_in_effective_config = False
    nginx_t_errors: list[str] = []
    norm_vhost = vhost_dir.replace(os.sep, "/")
    env = environment_with_system_path()

    for ngx in nginx_binary_candidates():
        try:
            r = subprocess.run(
                [ngx, "-T"],
                capture_output=True,
                text=True,
                timeout=25,
                env=env,
            )
        except (FileNotFoundError, OSError, subprocess.TimeoutExpired) as e:
            nginx_t_errors.append(f"{ngx}: {e}")
            continue
        dump = (r.stdout or "") + (r.stderr or "")
        if r.returncode != 0:
            nginx_t_errors.append(f"{ngx}: " + (dump.strip()[:800] or f"-T exit {r.returncode}"))
            continue
        if re.search(r"\blisten\s+.*443", dump):
            effective_listen_443 = True
        if norm_vhost in dump or "panel/vhost/nginx" in dump:
            panel_include_in_effective_config = True

    hints: list[str] = []
    if not os.path.isdir(vhost_dir):
        hints.append(f"The panel vhost directory is missing ({vhost_dir}). Create a website in YakPanel first.")
    elif not vhost_summaries:
        hints.append("There are no .conf files under the panel nginx vhost directory.")

    le_live = "/etc/letsencrypt/live"
    le_present = False
    if os.path.isdir(le_live):
        try:
            le_present = any(
                n and not n.startswith(".")
                for n in os.listdir(le_live)
            )
        except OSError:
            le_present = False

    if le_present and vhost_summaries and not any_vhost_443:
        hints.append(
            "Let's Encrypt certs exist on this server but panel vhosts do not include an HTTPS (listen 443 ssl) block. "
            "Regenerate the vhost: edit the site and save, or use Request SSL again."
        )

    if any_vhost_443 and not effective_listen_443:
        hints.append(
            "Your panel .conf files define HTTPS, but nginx -T does not show any listen 443 — the daemon that handles traffic is not loading YakPanel vhosts. "
            "Add the include line below inside http { } for that nginx (e.g. /etc/nginx/nginx.conf), then nginx -t && reload."
        )
    elif vhost_summaries and not panel_include_in_effective_config:
        hints.append(
            "If http://domain shows the default 'Welcome to nginx' page, stock nginx is answering and likely does not include YakPanel vhosts. "
            "Add the include below (or symlink this directory into /etc/nginx/conf.d/)."
        )

    localhost_443_open = _localhost_accepts_tcp(443)
    ss_443 = _ss_reports_listen_443()

    if not localhost_443_open and not effective_listen_443:
        hints.append(
            "This server is not accepting TCP on 127.0.0.1:443 — nothing is listening on 443 yet. "
            "Fix nginx (listen 443 ssl + include panel vhosts) first; opening only the cloud firewall will not fix ERR_CONNECTION_REFUSED until nginx binds 443."
        )
    elif effective_listen_443 and localhost_443_open:
        hints.append(
            "Nginx loads HTTPS and 127.0.0.1:443 accepts connections on this host. "
            "If browsers off this machine still see connection refused, allow inbound TCP 443: "
            "sudo ufw allow 443/tcp && sudo ufw reload (or firewalld), and your VPS Security Group / provider firewall."
        )
    elif effective_listen_443 and not localhost_443_open:
        hints.append(
            "nginx -T reports listen 443, but connecting to 127.0.0.1:443 failed — check nginx error.log; nginx may have failed to bind (permission or address already in use)."
        )
    elif localhost_443_open and not effective_listen_443:
        hints.append(
            "127.0.0.1:443 accepts TCP, but nginx -T from panel binaries did not show listen 443 — another process may own 443; check ss -tlnp and which nginx serves port 80."
        )

    debian_sites = os.path.isdir("/etc/nginx/sites-available")
    rhel_conf = os.path.isdir("/etc/nginx/conf.d")
    layout = "unknown"
    if debian_sites:
        layout = "debian_sites_available"
    elif rhel_conf:
        layout = "rhel_conf_d"
    drop_deb = "/etc/nginx/sites-available/yakpanel-vhosts.conf"
    drop_rhel = "/etc/nginx/conf.d/yakpanel-vhosts.conf"
    nginx_wizard = {
        "detected_layout": layout,
        "include_snippet": include_snippet,
        "dropin_file_suggested": drop_deb if debian_sites else drop_rhel,
        "debian": {
            "sites_available_file": drop_deb,
            "sites_enabled_symlink": "/etc/nginx/sites-enabled/yakpanel-vhosts.conf",
            "steps": [
                f"printf '%s\\n' '{include_snippet}' | sudo tee {drop_deb}",
                f"sudo ln -sf {drop_deb} /etc/nginx/sites-enabled/yakpanel-vhosts.conf",
                "sudo nginx -t && sudo systemctl reload nginx",
            ],
        },
        "rhel": {
            "conf_d_file": drop_rhel,
            "steps": [
                f"printf '%s\\n' '{include_snippet}' | sudo tee {drop_rhel}",
                "sudo nginx -t && sudo systemctl reload nginx",
            ],
        },
        "note": "Run the steps for your distro as root. The include line must appear inside the main http { } context (conf.d files do automatically).",
    }

    return {
        "vhost_dir": vhost_dir,
        "include_snippet": include_snippet,
        "nginx_wizard": nginx_wizard,
        "vhosts": vhost_summaries,
        "any_vhost_listen_ssl": any_vhost_443,
        "nginx_effective_listen_443": effective_listen_443,
        "panel_vhost_path_in_nginx_t": panel_include_in_effective_config,
        "nginx_t_probe_errors": nginx_t_errors,
        "localhost_443_accepts_tcp": localhost_443_open,
        "ss_reports_443_listen": ss_443,
        "hints": hints,
    }


class DnsCertCloudflareRequest(BaseModel):
    domain: str
    email: str
    api_token: str


class DnsManualInstructionsRequest(BaseModel):
    domain: str


@router.post("/dns-request/cloudflare")
async def ssl_dns_cloudflare_cert(
    body: DnsCertCloudflareRequest,
    current_user: User = Depends(get_current_user),
    db: AsyncSession = Depends(get_db),
):
    """Request Let's Encrypt certificate using DNS-01 via Cloudflare (requires certbot-dns-cloudflare)."""
    dom = (body.domain or "").split(":")[0].strip()
    if not dom or ".." in dom or not body.email or not body.api_token:
        raise HTTPException(status_code=400, detail="domain, email, and api_token required")
    result_dom = await db.execute(select(Domain).where(Domain.name == dom).limit(1))
    dom_row = result_dom.scalar_one_or_none()
    if dom_row:
        regen_pre = await regenerate_site_vhost(db, dom_row.pid)
        if not regen_pre.get("status"):
            raise HTTPException(
                status_code=500,
                detail="Cannot refresh nginx vhost: " + str(regen_pre.get("msg", "")),
            )
        ok_ngx, err_ngx = _reload_panel_and_common_nginx()
        if not ok_ngx:
            raise HTTPException(
                status_code=500,
                detail="Nginx reload failed: " + err_ngx,
            )

    prefix = _certbot_command()
    if not prefix:
        raise HTTPException(status_code=500, detail=_certbot_missing_message())

    hostnames = await _le_hostnames_for_domain_row(db, dom_row, dom)
    if not hostnames:
        hostnames = [dom]
    cred_lines = f'dns_cloudflare_api_token = {body.api_token.strip()}\n'
    fd, cred_path = tempfile.mkstemp(suffix=".ini", prefix="yakpanel_cf_")
    try:
        os.write(fd, cred_lines.encode())
        os.close(fd)
        os.chmod(cred_path, 0o600)
    except OSError as e:
        try:
            os.close(fd)
        except OSError:
            pass
        raise HTTPException(status_code=500, detail=f"Cannot write credentials temp file: {e}") from e

    base_flags = [
        "--non-interactive",
        "--agree-tos",
        "--email",
        body.email.strip(),
        "--no-eff-email",
        "--dns-cloudflare",
        "--dns-cloudflare-credentials",
        cred_path,
    ]
    cmd = prefix + ["certonly"] + base_flags
    for h in hostnames:
        cmd.extend(["-d", h])
    env = environment_with_system_path()
    try:
        proc = subprocess.run(cmd, capture_output=True, text=True, timeout=600, env=env)
    except (FileNotFoundError, subprocess.TimeoutExpired) as e:
        try:
            os.unlink(cred_path)
        except OSError:
            pass
        raise HTTPException(status_code=500, detail=str(e)) from e
    finally:
        try:
            os.unlink(cred_path)
        except OSError:
            pass

    if proc.returncode != 0:
        err = (proc.stderr or proc.stdout or "").strip() or f"exit {proc.returncode}"
        raise HTTPException(
            status_code=500,
            detail="certbot DNS failed. Install certbot-dns-cloudflare (pip or OS package) if missing. " + err[:6000],
        )

    if dom_row:
        regen = await regenerate_site_vhost(db, dom_row.pid)
        if not regen.get("status"):
            return {
                "status": True,
                "msg": "Certificate issued but vhost regen failed: " + str(regen.get("msg", "")),
                "output": (proc.stdout or "")[-2000:],
            }

    return {
        "status": True,
        "msg": "Certificate issued via Cloudflare DNS-01",
        "output": (proc.stdout or "")[-2000:],
    }


@router.post("/dns-request/manual-instructions")
async def ssl_dns_manual_instructions(
    body: DnsManualInstructionsRequest,
    current_user: User = Depends(get_current_user),
):
    """Return TXT record host for ACME DNS-01 (user creates record then runs certbot --manual)."""
    d = (body.domain or "").split(":")[0].strip()
    if not d or ".." in d:
        raise HTTPException(status_code=400, detail="Invalid domain")
    return {
        "txt_record_name": f"_acme-challenge.{d}",
        "certbot_example": (
            f"sudo certbot certonly --manual --preferred-challenges dns --email you@example.com "
            f"--agree-tos -d {d}"
        ),
        "note": "Certbot will display the exact TXT value to create. After DNS propagates, continue in the terminal.",
    }


@router.get("/certificates")
async def ssl_list_certificates(current_user: User = Depends(get_current_user)):
    """List existing Let's Encrypt certificates"""
    live_dir = "/etc/letsencrypt/live"
    if not os.path.isdir(live_dir):
        return {"certificates": []}
    certs = []
    for name in os.listdir(live_dir):
        if name.startswith("."):
            continue
        path = os.path.join(live_dir, name)
        if os.path.isdir(path) and os.path.isfile(os.path.join(path, "fullchain.pem")):
            certs.append({"name": name, "path": path})
    return {"certificates": sorted(certs, key=lambda x: x["name"])}