fix(assets): ip 字段拒收主机名,并告诉调用方该怎么改

上一次修复让非法 ip 不再打挂企业归属重算,但脏数据仍在持续产生。这次从源头
拦住:assets.ip 会被转成 inet 参与网段归属,存主机名的资产对所有 IP/CIDR
范围规则都是隐形的。

- db 层新增 ValidateAssetIP + ErrAssetIPInvalid,在 UpsertIP /
  UpsertHTTPService / UpsertOtherService / UpsertEndpoint 四个入口校验。
  放在 db 层是因为 Agent 工具和资产 API 都经由它们,不必写两遍;内部调用方
  (enrich 的解析结果、task_assets 的 ParseCIDR 输出、RecordValue 已 ParseIP
  过滤的分支)传的本就是合法地址,不受影响。
- 错误信息给的是改法而不只是「非法」:带上收到的原值,并指出主机名应改用
  type=subdomain 填 domain,或先解析 A/AAAA 记录再填 ip。insert_assets 的
  错误是逐条返回的({index, error}),同批其余资产照常入库,Agent 拿到后能
  定位到具体那条并在下一轮自行修正。
- 同步收紧 insert_assets 的工具描述与 ip 字段说明,在提示层先挡一道。
- 空 ip 仍然合法:service / endpoint 的 ip 是可选字段。

验证: 新增 db 层四类型拒收测试与 agent 层「一批混入一条非法、其余照常入库」
测试,均连 PG 实跑通过;干净库 go test -p 1 ./... 无新增失败(仅剩既有的
TestCompanyScopeInvalid)。

CHANGELOG.md

@@ -18,6 +18,10 @@
 - 无法解析的 `ip` 不再静默跳过：企业范围保存后会明确提示有多少条资产的 `ip` 不是合法 IP（含具体 id 和取值），提醒这些资产不会被 IP/CIDR 规则归属；同一信息也写入服务端日志，覆盖没有前端响应的路径（删除企业、scopesentry 同步、Agent 写入）。
 - 任务测试范围的 IP/CIDR 匹配同步改用 `try_inet()`，替换原先只检查字符集的正则守卫（`abc.def` 这类全 hex 字母的主机名此前仍会漏过并触发同样的报错）。
 
+#### 修改的功能
+
+- `ip`、`service`、`endpoint` 资产的 `ip` 字段不再接受主机名：`insert_assets` 和资产 API 会逐条返回带 `index` 的错误，并直接给出改正方法（改用 `type=subdomain` 填 `domain`，或先解析 A/AAAA 记录），Agent 可据此自行修正后重插。同批次中的其他资产不受影响，照常入库。
+
 ## [0.3.3] - 2026-08-23
 
 ### Worker

agent/insert_assets_test.go

@@ -3,6 +3,7 @@ package agent
 import (
 	"context"
 	"encoding/json"
+	"strings"
 	"testing"
 
 	"github.com/Autumn-27/artex/db"
@@ -394,3 +395,56 @@ func TestInsertAssetsDedup(t *testing.T) {
 		t.Errorf("dedup: want same ID on double insert, got %v vs %v", id1, id2)
 	}
 }
+
+// =====================================================================
+// TestInsertAssetsRejectsHostnameIPPerItem
+// 一批里混入 ip 填了主机名的一条 → 只有那条失败，其余照常入库，
+// 且错误里带得上 index 和改正方法，Agent 下一轮能自己修好。
+// =====================================================================
+func TestInsertAssetsRejectsHostnameIPPerItem(t *testing.T) {
+	d := testDB(t)
+	defer d.Close()
+
+	ts := NewToolSet(nil, "")
+	ts.SetAssetStore(d.Assets(), d.Companies())
+	defer d.Exec(`DELETE FROM assets WHERE ip IN ('198.51.100.23','cdn.badip-test.com') OR domain='badip-test.com'`)
+
+	out := callInsertAssets(t, ts, map[string]any{
+		"assets": []any{
+			map[string]any{"type": "root_domain", "domain": "badip-test.com"},
+			map[string]any{"type": "ip", "ip": "cdn.badip-test.com"},
+			map[string]any{"type": "ip", "ip": "198.51.100.23"},
+		},
+	})
+
+	// The two valid entries must survive the bad one — a whole-batch failure
+	// would make the agent re-send assets that were already fine.
+	results, _ := out["results"].([]any)
+	if len(results) != 2 {
+		t.Fatalf("results=%v, want the 2 valid assets", out["results"])
+	}
+
+	errsRaw, _ := out["errors"].([]any)
+	if len(errsRaw) != 1 {
+		t.Fatalf("errors=%v, want exactly the invalid entry", out["errors"])
+	}
+	entry, _ := errsRaw[0].(map[string]any)
+	if index, _ := entry["index"].(float64); int(index) != 1 {
+		t.Fatalf("error index=%v, want 1", entry["index"])
+	}
+	message, _ := entry["error"].(string)
+	for _, want := range []string{"cdn.badip-test.com", "type=subdomain", "A/AAAA"} {
+		if !strings.Contains(message, want) {
+			t.Fatalf("error message %q lacks %q — agent cannot act on it", message, want)
+		}
+	}
+
+	// The rejected value must not have reached the table.
+	var stored int
+	if err := d.QueryRow(`SELECT count(*) FROM assets WHERE ip='cdn.badip-test.com'`).Scan(&stored); err != nil {
+		t.Fatal(err)
+	}
+	if stored != 0 {
+		t.Fatalf("rejected hostname still stored in assets.ip (%d rows)", stored)
+	}
+}

agent/tools_insert.go

@@ -79,7 +79,7 @@ func (t *ToolSet) insertAssets() actool.CoreTool {
 		"发现新资产时立即批量插入资产到资产库。支持一次提交多种资产类型：root_domain/ip/subdomain/app/service/endpoint。\n"+
 			"每条资产的字段说明：\n"+
 			"• root_domain：domain(根域名, 必填)、icp(备案, 可选)\n"+
-			"• ip：ip(IP地址, 必填)、bound_domains([域名])、open_ports([{port,service}])\n"+
+			"• ip：ip(IP地址, 必填, 必须是 IPv4/IPv6 地址而非主机名)、bound_domains([域名])、open_ports([{port,service}])\n"+
 			"• subdomain：domain(子域名, 必填)、record_type(A/AAAA/CNAME等)、record_value(记录值)、icp(可选)\n"+
 			"• app：app_name(应用名, 必填)、bundle_id(bundle id, 可选)、category、description、app_icp、company_id(归属企业 id, 可选)\n"+
 			"• service(http)：url(必填)、technologies([指纹])、status_code、content_length、page_title、favicon_mmh3、auth([{type,username,password,...}])、service_ip\n"+
@@ -108,7 +108,7 @@ func (t *ToolSet) insertAssets() actool.CoreTool {
 						"description": "DNS 解析值列表（subdomain 可选，如 [\"1.2.3.4\",\"2.3.4.5\"]）",
 					},
 					// ip
-					"ip": str("IP 地址（ip 类型必填；service/endpoint 类型可填，用于关联 IP）"),
+					"ip": str("IP 地址，必须是 IPv4/IPv6 地址，不能填主机名（主机名请用 type=subdomain 的 domain 字段）；ip 类型必填；service/endpoint 类型可填，用于关联 IP"),
 					"bound_domains": map[string]any{
 						"type":        "array",
 						"items":       map[string]any{"type": "string"},

db/assets.go

@@ -3,6 +3,7 @@ package db
 import (
 	"database/sql"
 	"encoding/json"
+	"errors"
 	"fmt"
 	"net"
 	"net/url"
@@ -300,6 +301,27 @@ RETURNING id`, domain, icpVal, companyID, taskIDs).Scan(&id)
 	return id, err
 }
 
+// ErrAssetIPInvalid marks a non-address value in an asset's ip field. Both
+// insert_assets and the asset API report it per item with the item index, so one
+// bad entry never costs the rest of the batch.
+var ErrAssetIPInvalid = errors.New("invalid asset ip")
+
+// ValidateAssetIP keeps hostnames out of assets.ip. Network attribution casts
+// that column to inet (see try_inet in schema.sql), so a hostname stored here is
+// silently invisible to every IP/CIDR scope rule — the asset simply never gets
+// attributed and nobody can tell why. The message states the fix rather than
+// just the fault, so an agent can correct the item on its next turn. An empty
+// value is accepted: the ip field is optional for service and endpoint assets.
+func ValidateAssetIP(value string) error {
+	if value == "" || net.ParseIP(value) != nil {
+		return nil
+	}
+	return fmt.Errorf(
+		"%w: ip 必须是 IPv4/IPv6 地址，收到 %q。若这是主机名，请改用 type=subdomain 并填 domain 字段；"+
+			"若确实要登记地址，请先解析出 A/AAAA 记录，再用解析出的地址填 ip",
+		ErrAssetIPInvalid, value)
+}
+
 // =====================================================================
 // UpsertIP
 // =====================================================================
@@ -317,6 +339,9 @@ func (s *AssetStore) UpsertIP(req UpsertIPReq) (int64, error) {
 	if req.IP == "" {
 		return 0, fmt.Errorf("ip is required")
 	}
+	if err := ValidateAssetIP(req.IP); err != nil {
+		return 0, err
+	}
 	if s.tx == nil {
 		return s.withCompanyScopeMutation(func(scoped *AssetStore) (int64, error) {
 			return scoped.UpsertIP(req)
@@ -659,6 +684,9 @@ func (s *AssetStore) UpsertHTTPService(req UpsertHTTPServiceReq) (int64, error)
 	if req.URL == "" {
 		return 0, fmt.Errorf("url is required")
 	}
+	if err := ValidateAssetIP(req.IP); err != nil {
+		return 0, err
+	}
 	if s.tx == nil {
 		return s.withCompanyScopeMutation(func(scoped *AssetStore) (int64, error) {
 			return scoped.UpsertHTTPService(req)
@@ -793,6 +821,9 @@ func (s *AssetStore) UpsertOtherService(req UpsertOtherServiceReq) (int64, error
 	if req.Domain == "" && req.IP == "" {
 		return 0, fmt.Errorf("domain or ip is required")
 	}
+	if err := ValidateAssetIP(req.IP); err != nil {
+		return 0, err
+	}
 	if req.Port == 0 {
 		return 0, fmt.Errorf("port is required")
 	}
@@ -936,6 +967,9 @@ func (s *AssetStore) UpsertEndpoint(req UpsertEndpointReq) (int64, error) {
 	if req.Method == "" {
 		return 0, fmt.Errorf("method is required")
 	}
+	if err := ValidateAssetIP(req.IP); err != nil {
+		return 0, err
+	}
 	if s.tx == nil {
 		return s.withCompanyScopeMutation(func(scoped *AssetStore) (int64, error) {
 			return scoped.UpsertEndpoint(req)

db/assets_test.go

@@ -1,8 +1,10 @@
 package db
 
 import (
+	"errors"
 	"fmt"
 	"slices"
+	"strings"
 	"testing"
 	"time"
 )
@@ -946,3 +948,54 @@ func TestAssetPaginationUsesStableIDTieBreaker(t *testing.T) {
 		t.Fatalf("DSL pagination order=%v want=%v", got, want)
 	}
 }
+
+// TestAssetIPRejectsHostname pins the write-side guard added after a hostname in
+// assets.ip broke company attribution: every asset type that persists an ip must
+// refuse a non-address, and the message must tell the caller how to fix it.
+func TestAssetIPRejectsHostname(t *testing.T) {
+	d, av2, _ := testSetup(t)
+	defer d.Close()
+
+	hostname := "d63cd476.cdn.ucloud.com.cn"
+	cases := []struct {
+		name string
+		call func() (int64, error)
+	}{
+		{"ip", func() (int64, error) {
+			return av2.UpsertIP(UpsertIPReq{IP: hostname})
+		}},
+		{"http service", func() (int64, error) {
+			return av2.UpsertHTTPService(UpsertHTTPServiceReq{URL: "https://example.com/", IP: hostname})
+		}},
+		{"other service", func() (int64, error) {
+			return av2.UpsertOtherService(UpsertOtherServiceReq{IP: hostname, Port: 22, ServiceName: "ssh"})
+		}},
+		{"endpoint", func() (int64, error) {
+			return av2.UpsertEndpoint(UpsertEndpointReq{URL: "https://example.com/a", Method: "GET", IP: hostname})
+		}},
+	}
+	for _, tc := range cases {
+		id, err := tc.call()
+		if !errors.Is(err, ErrAssetIPInvalid) {
+			deleteAsset(d, id)
+			t.Fatalf("%s: err=%v, want %v", tc.name, err, ErrAssetIPInvalid)
+		}
+		// The caller has to learn the remedy, not just that something was wrong.
+		if !strings.Contains(err.Error(), hostname) || !strings.Contains(err.Error(), "type=subdomain") {
+			t.Fatalf("%s: message is not actionable: %v", tc.name, err)
+		}
+	}
+
+	// An empty ip stays legal for the types where it is optional, and a real
+	// address is still accepted.
+	id, err := av2.UpsertHTTPService(UpsertHTTPServiceReq{URL: "https://ip-guard.example.com/"})
+	if err != nil {
+		t.Fatalf("empty ip rejected: %v", err)
+	}
+	deleteAsset(d, id)
+	id, err = av2.UpsertIP(UpsertIPReq{IP: "203.0.113.7"})
+	if err != nil {
+		t.Fatalf("valid ip rejected: %v", err)
+	}
+	deleteAsset(d, id)
+}