review修复: SSRF重定向绕过+无价歧义标注

- fetch逐跳校验重定向(每跳过域名+DNS私有IP双校验,防redirect到内网)
- 文档未列价的模型落0时description显式标注(避免0=免费歧义)
- 清理无用import

pipeline_platform/platform_ability.py

@@ -15,6 +15,7 @@ pipeline-service: platform_ability — 平台内部 agent 能力包（pipeline_i
      （llm_vendor.endpoints + llm_api_profile 模板 + llm_model 含四价）
 """
 
+import asyncio
 import json
 import logging
 import re
@@ -105,6 +106,7 @@ async def _h_platform_llm_status(sor, params, ctx):
 # ────────────────────── 工具 2：抓取模型 API 文档 ──────────────────────
 
 _MAX_DOC_CHARS = 60000
+_MAX_REDIRECTS = 5
 
 
 def _validate_doc_url(url: str) -> str:
@@ -133,25 +135,69 @@ def _html_to_text(html: str) -> str:
     return txt.strip()
 
 
+def _is_private_ip(ip: str) -> bool:
+    """DNS 解析后的二次防线：解析到私有/环回/链路本地地址一律拒绝（DNS rebinding 防护）。"""
+    import ipaddress
+    try:
+        a = ipaddress.ip_address(ip)
+    except ValueError:
+        return True  # 解析不出就当私有（拒绝）
+    return (a.is_private or a.is_loopback or a.is_link_local
+            or a.is_reserved or a.is_multicast or a.is_unspecified)
+
+
+async def _fetch_url_safe(url: str, max_redirects: int = _MAX_REDIRECTS):
+    """带重定向逐跳校验的抓取（每一跳都过域名+DNS双重校验）。
+
+    返回 (text, content_type)；失败抛 ValueError。
+    """
+    import aiohttp
+    current = url
+    for _ in range(max_redirects + 1):
+        verr = _validate_doc_url(current)
+        if verr:
+            raise ValueError(verr)
+        host = re.sub(r"^https?://", "", current).split("/")[0].split(":")[0].lower()
+        # DNS 解析校验（同步阻塞短调用，可接受；防 DNS rebinding 指向内网）
+        try:
+            infos = await asyncio.get_event_loop().getaddrinfo(host, None)
+        except Exception as e:
+            raise ValueError("文档域名无法解析：%s" % str(e)[:120])
+        for info in infos:
+            ip = str(info[4][0])
+            if _is_private_ip(ip):
+                raise ValueError("文档域名解析到内网地址（%s），拒绝访问" % ip)
+        timeout = aiohttp.ClientTimeout(total=30)
+        async with aiohttp.ClientSession(timeout=timeout) as sess:
+            async with sess.get(current, headers={"User-Agent": "Mozilla/5.0"},
+                                allow_redirects=False, ssl=False) as resp:
+                if resp.status in (301, 302, 303, 307, 308):
+                    loc = resp.headers.get("Location", "")
+                    if not loc:
+                        raise ValueError("重定向缺少 Location")
+                    if loc.startswith("/"):
+                        scheme = "https" if current.startswith("https") else "http"
+                        loc = "%s://%s%s" % (scheme, host, loc)
+                    current = loc
+                    continue
+                if resp.status != 200:
+                    raise ValueError("抓取失败：HTTP %d" % resp.status)
+                ctype = resp.headers.get("Content-Type", "")
+                raw = await resp.text(errors="replace")
+                return raw, ctype
+    raise ValueError("重定向次数超限（>%d）" % max_redirects)
+
+
 async def _h_fetch_model_doc(sor, params, ctx):
     """抓取模型 API 文档页面 → 纯文本（供 LLM 提取配置规格）。"""
     err = await _require_owner(sor, ctx)
     if err:
         return err
     url = (params.get("url") or "").strip()
-    verr = _validate_doc_url(url)
-    if verr:
-        return verr
-    import aiohttp
     try:
-        timeout = aiohttp.ClientTimeout(total=30)
+        raw, ctype = await _fetch_url_safe(url)
-        async with aiohttp.ClientSession(timeout=timeout) as sess:
+    except ValueError as e:
-            async with sess.get(url, headers={"User-Agent": "Mozilla/5.0"},
+        return str(e)
-                                allow_redirects=True, ssl=False) as resp:
-                if resp.status != 200:
-                    return "抓取失败：HTTP %d" % resp.status
-                ctype = resp.headers.get("Content-Type", "")
-                raw = await resp.text(errors="replace")
     except Exception as e:
         return "抓取失败：%s" % str(e)[:200]
     if "html" in ctype.lower() or raw.lstrip()[:15].lower().startswith(("<!doctype", "<html")):
@@ -332,9 +378,14 @@ async def _h_apply_llm_config(sor, params, ctx):
             skipped.append("(缺 vendor_model_id)")
             continue
         cap = m.get("capability") or "t2t"
-        price_in, price_out = _f(m.get("price_input")), _f(m.get("price_output"))
+        raw_pi, raw_po = m.get("price_input"), m.get("price_output")
+        price_in, price_out = _f(raw_pi), _f(raw_po)
         cost_in = _f(m.get("cost_input"), price_in)
         cost_out = _f(m.get("cost_output"), price_out)
+        desc = (m.get("description", "") or "").strip()
+        # 文档未给价（null）→ 落 0 并显式标注，避免「0=免费」歧义
+        if raw_pi is None or raw_po is None:
+            desc = "⚠️文档未列定价（暂落0，需人工补录）。" + desc
         recs = await sor.sqlExe(
             "SELECT id FROM llm_model WHERE vendor_id=${v}$ AND vendor_model_id=${m}$",
             {"v": vendor_id, "m": vmid})
@@ -346,7 +397,7 @@ async def _h_apply_llm_config(sor, params, ctx):
                 "cost_input=${ci}$, cost_output=${co}$, description=${d}$, updated_at=NOW() "
                 "WHERE id=${i}$",
                 {"pi": price_in, "po": price_out, "ci": cost_in, "co": cost_out,
-                 "d": m.get("description", "") or "", "i": mid})
+                 "d": desc, "i": mid})
             await sor.sqlExe("COMMIT", {})
             updated.append(vmid)
             continue
@@ -358,7 +409,7 @@ async def _h_apply_llm_config(sor, params, ctx):
             "ppid": "", "price_input": price_in, "price_output": price_out,
             "cost_input": cost_in, "cost_output": cost_out,
             "default_params": "{}", "status": "active",
-            "description": m.get("description", "") or "", "org_id": "0"})
+            "description": desc, "org_id": "0"})
         created.append(vmid)
 
     rt_note = ""