container_apm.go 31 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040
  1. package containers
  2. import (
  3. "bufio"
  4. "bytes"
  5. "debug/elf"
  6. "fmt"
  7. "net"
  8. "os"
  9. "path"
  10. "sort"
  11. "strconv"
  12. "strings"
  13. "time"
  14. "github.com/cilium/ebpf/link"
  15. "github.com/coroot/coroot-node-agent/flags"
  16. "github.com/coroot/coroot-node-agent/ebpftracer"
  17. "github.com/coroot/coroot-node-agent/ebpftracer/l7"
  18. "github.com/coroot/coroot-node-agent/ebpftracer/tracer"
  19. "github.com/coroot/coroot-node-agent/proc"
  20. "github.com/coroot/coroot-node-agent/tracing"
  21. "github.com/coroot/coroot-node-agent/utils"
  22. . "github.com/coroot/coroot-node-agent/utils/modelse"
  23. "github.com/pkg/errors"
  24. klog "github.com/sirupsen/logrus"
  25. semconv "go.opentelemetry.io/otel/semconv/v1.18.0"
  26. "inet.af/netaddr"
  27. )
  28. const (
  29. TRACE_STATUS = 1
  30. )
  31. func (c *Container) getTrace(traceId uint64) (*tracing.Trace, bool) {
  32. trace, ok := c.traceMap[traceId]
  33. return trace, ok
  34. }
  35. func (c *Container) createTraceMap(traceId uint64, trace *tracing.Trace) {
  36. c.traceMap[traceId] = trace
  37. }
  38. // 查询或创建trace信息
  39. func (c *Container) getOrInitTrace(traceId uint64) (*tracing.Trace, error) {
  40. trace, ok := c.getTrace(traceId)
  41. if !ok {
  42. //new trace
  43. trace = tracing.NewTraceFromEvent(string(c.id))
  44. //create TraceMap
  45. c.createTraceMap(traceId, trace)
  46. //create ParentSpan
  47. trace.CreateRootSpan(traceId)
  48. }
  49. return trace, nil
  50. }
  51. // getGrpcServerNetworkInfo 获取 gRPC server 的网络信息
  52. // 返回: IP地址, 端口号, 容器ID
  53. func (c *Container) getGrpcServerNetworkInfo() (string, uint16, string) {
  54. containerID := ""
  55. if c.cgroup != nil {
  56. containerID = c.cgroup.ContainerId
  57. }
  58. ipAddr := ""
  59. ifaces, err := net.Interfaces()
  60. if err == nil {
  61. for _, iface := range ifaces {
  62. if iface.Name == "eth0" {
  63. addrs, err := iface.Addrs()
  64. if err == nil {
  65. for _, addr := range addrs {
  66. var ipnet *net.IPNet
  67. switch v := addr.(type) {
  68. case *net.IPNet:
  69. ipnet = v
  70. case *net.IPAddr:
  71. ipnet = &net.IPNet{IP: v.IP, Mask: v.IP.DefaultMask()}
  72. }
  73. if ipnet != nil && ipnet.IP.To4() != nil {
  74. ipAddr = ipnet.IP.String()
  75. break
  76. }
  77. }
  78. }
  79. break
  80. }
  81. }
  82. }
  83. klog.Debugf("grpc server ip %s", ipAddr)
  84. // 本地端口尝试从AppInfo.Sport获取
  85. port := c.AppInfo.Sport
  86. klog.Debugf("grpc server port %d", port)
  87. return ipAddr, uint16(port), containerID
  88. }
  89. // Deprecated: InitTrace not used
  90. //func (c *Container) InitTrace(traceId uint64, r *l7.RequestData) error {
  91. // method, path, hostIp, port := l7.ParseHttpHost(r.Payload)
  92. // ip, err := netaddr.ParseIP(hostIp)
  93. // if err != nil {
  94. // //fmt.Println("host ip error")
  95. // hostIp = "127.0.0.1"
  96. // }
  97. // addr := netaddr.IPPortFrom(ip, port)
  98. // trace := tracing.NewTrace(string(c.id), addr)
  99. // if trace == nil {
  100. // return fmt.Errorf("OTEL_EXPORTER_OTLP_TRACES_ENDPOINT is null")
  101. // }
  102. // c.traceMap[traceId] = trace
  103. //
  104. // trace.TraceStart(method, path, r.Status, r.Duration)
  105. // return nil
  106. //}
  107. // 在任意阶段,r.TraceId 不等于0 则创建 traceMap && createParentSpan
  108. // 更新 createTraceSpan 机制,更新触发traceEnd机制,当事件个数满足时,任意event均可触发end
  109. func (c *Container) SendEvent(t *tracing.Trace, traceID uint64) {
  110. if t.AllEventReady(traceID) {
  111. t.SendEvent()
  112. klog.Debugf("SendEvent %d", traceID)
  113. //fmt.Println(t.GetSpan())
  114. //fmt.Println("===============")
  115. delete(c.traceMap, traceID)
  116. }
  117. }
  118. func (c *Container) valuableTrace(traceID uint64) bool {
  119. return traceID != 0
  120. }
  121. func (c *Container) onL7RequestApm(pid uint32, fd uint64, timestamp uint64, r *l7.RequestData) map[netaddr.IP]string {
  122. c.lock.Lock()
  123. defer c.lock.Unlock()
  124. if r.Protocol == l7.ProtocolDNS {
  125. ip2fqdn, _type, fqdn, ttl, ips := c.onDNSRequest(r)
  126. if c.l7Attach && c.valuableTrace(r.TraceId) {
  127. apmTrace, err := c.getOrInitTrace(r.TraceId)
  128. if err == nil {
  129. apmTrace.DNSTraceQueryEvent(r, _type, fqdn, ttl, ips)
  130. c.SendEvent(apmTrace, r.TraceId)
  131. }
  132. }
  133. return ip2fqdn
  134. }
  135. //if !c.valuableTrace(r.TraceId) {
  136. // return nil
  137. //}
  138. //klog.Infof("====ProtocolTrace+++++ start==== %d %d", pid, r.TraceId)
  139. // klog.Infof("====ProtocolTrace===== start==== %d %d", r.Protocol == l7.ProtocolTrace, c.l7Attach)
  140. if r.Protocol == l7.ProtocolTrace && c.l7Attach && c.valuableTrace(r.TraceId) {
  141. // klog.Infof("====ProtocolTrace---- start==== %d %d", pid, r.TraceId)
  142. if r.TraceStart == TRACE_STATUS {
  143. // klog.Infof("====ProtocolTrace start==== %d %d", pid, r.TraceId)
  144. trace, err := c.getOrInitTrace(r.TraceId)
  145. if c.AppInfo.AppName != "" {
  146. klog.Debugf("->>> [%s] -> payload:[%s]", c.AppInfo.AppName, r.Payload)
  147. }
  148. if err == nil {
  149. if r.TraceType == 0 {
  150. method, requestURI, sn, sport := l7.ParseHttpHost(r.Payload, r.IsTls)
  151. ip, _ := netaddr.ParseIP(sn)
  152. //codeType := c.GetCodeTypeFromCache(pid)
  153. container_id := ""
  154. if c.cgroup != nil {
  155. container_id = c.cgroup.ContainerId
  156. }
  157. trace.TraceStartEvent(method, requestURI, sn, sport, r.Status, netaddr.IPPortFrom(ip, sport), pid, c.GetAppInfo(), container_id)
  158. c.SendEvent(trace, r.TraceId)
  159. } else if r.TraceType == 1 {
  160. ipAddr, port, containerID := c.getGrpcServerNetworkInfo()
  161. trace.GrpcServerTraceStartEvent(ipAddr, port, r, c.GetAppInfo(), containerID)
  162. c.SendEvent(trace, r.TraceId)
  163. // apmTrace, err := c.getOrInitTrace(r.TraceId)
  164. // if err == nil {
  165. // apmTrace.GrpcServerTraceQueryEvent(r, c.GetAppInfo())
  166. // c.SendEvent(apmTrace, r.TraceId)
  167. // }
  168. }
  169. }
  170. return nil
  171. }
  172. if r.TraceEnd == TRACE_STATUS {
  173. klog.Debugf("====ProtocolTrace end==== %d %d", pid, r.TraceId)
  174. trace, err := c.getOrInitTrace(r.TraceId)
  175. if err == nil {
  176. trace.TraceEndEvent(r)
  177. c.SendEvent(trace, r.TraceId)
  178. }
  179. return nil
  180. }
  181. }
  182. if r.Protocol == l7.ProtocolHTTP {
  183. if c.l7Attach && c.valuableTrace(r.TraceId) {
  184. method, requestURI, sn, sport := l7.ParseHttpHost(r.Payload, r.IsTls)
  185. apmTrace, err := c.getOrInitTrace(r.TraceId)
  186. //fmt.Println("ProtocolHTTP-----", r.TraceId, err)
  187. if err == nil {
  188. apmTrace.HttpTraceRequestEvent(method, requestURI, sn, sport, r)
  189. c.SendEvent(apmTrace, r.TraceId)
  190. }
  191. }
  192. //return nil
  193. }
  194. conn := c.connectionsByPidFd[PidFd{Pid: pid, Fd: fd}]
  195. //fmt.Println("l7.connectionsByPidFd", conn, pid, fd)
  196. if conn == nil {
  197. if r.Protocol == l7.ProtocolGrpc {
  198. klog.Infoln("conn == nil r.Protocol == l7.ProtocolGrpc")
  199. klog.Infoln("enter the l7.ProtocolGrpc")
  200. if c.l7Attach && c.valuableTrace(r.TraceId) {
  201. apmTrace, err := c.getOrInitTrace(r.TraceId)
  202. if err == nil {
  203. apmTrace.GrpcClientTraceQueryEvent(r)
  204. c.SendEvent(apmTrace, r.TraceId)
  205. }
  206. }
  207. }
  208. return nil
  209. }
  210. if timestamp != 0 && conn.Timestamp != timestamp {
  211. if r.Protocol == l7.ProtocolGrpc {
  212. klog.Infoln("timestamp != 0 && conn.Timestamp != timestamp r.Protocol == l7.ProtocolGrpc")
  213. klog.Infoln("enter the l7.ProtocolGrpc")
  214. if c.l7Attach && c.valuableTrace(r.TraceId) {
  215. apmTrace, err := c.getOrInitTrace(r.TraceId)
  216. if err == nil {
  217. apmTrace.GrpcClientTraceQueryEvent(r)
  218. c.SendEvent(apmTrace, r.TraceId)
  219. }
  220. }
  221. }
  222. return nil
  223. }
  224. stats := c.l7Stats.get(r.Protocol, conn.Dest, conn.ActualDest)
  225. //trace := tracing.NewTrace(string(c.id), conn.ActualDest)
  226. switch r.Protocol {
  227. /**
  228. * HTTP
  229. */
  230. case l7.ProtocolHTTP:
  231. if c.AppInfo.AppName != "" {
  232. klog.Debugf("[%s] ->>>>> curl -> %s payload:[%s]", c.AppInfo.AppName, conn.ActualDest, r.Payload)
  233. }
  234. stats.observe(r.Status.Http(), "", r.Duration)
  235. /**
  236. * HTTP2
  237. */
  238. case l7.ProtocolHTTP2:
  239. if conn.http2Parser == nil {
  240. conn.http2Parser = l7.NewHttp2Parser()
  241. }
  242. requests := conn.http2Parser.Parse(r.Method, r.Payload, uint64(r.Duration))
  243. for _, req := range requests {
  244. stats.observe(req.Status.Http(), "", req.Duration)
  245. //trace.Http2Request(req.Method, req.Path, req.Scheme, req.Status, req.Duration)
  246. }
  247. /**
  248. * PostgreSQL
  249. */
  250. case l7.ProtocolPostgres:
  251. if r.Method != l7.MethodStatementClose {
  252. stats.observe(r.Status.String(), "", r.Duration)
  253. }
  254. //if conn.postgresParser == nil {
  255. // conn.postgresParser = l7.NewPostgresParser()
  256. //}
  257. //query := conn.postgresParser.Parse(r.Payload)
  258. //trace.PostgresQuery(query, r.Status.Error(), r.Duration)
  259. if c.l7Attach && c.valuableTrace(r.TraceId) {
  260. if conn.postgresParser == nil {
  261. conn.postgresParser = l7.NewPostgresParser()
  262. }
  263. query := conn.postgresParser.Parse(r.Payload)
  264. //trace.MysqlQuery(query, r.Status.Error(), r.Duration)
  265. if c.AppInfo.AppName != "" {
  266. klog.Debugf("[%s] ->>>>> %s -> %s payload:[%s]", c.AppInfo.AppName, r.Protocol.String(), conn.ActualDest, query)
  267. }
  268. //apmTrace, ok := c.getTrace(r.TraceId)
  269. apmTrace, err := c.getOrInitTrace(r.TraceId)
  270. //fmt.Println("mysql r.TraceId:", r.TraceId)
  271. //fmt.Println("ok:", ok)
  272. //fmt.Println("traceMap:", len(c.traceMap))
  273. if err == nil {
  274. //apmTrace.MysqlTraceQuery(query, r.Status.Error(), r.Duration, conn.ActualDest)
  275. //apmTrace.PostGreSqlTraceQueryEvent(query, r, conn.ActualDest)
  276. apmTrace.SQLTraceQueryEvent(r.Protocol, semconv.DBSystemPostgreSQL, query, r, conn.ActualDest)
  277. c.SendEvent(apmTrace, r.TraceId)
  278. }
  279. }
  280. /**
  281. * Mysql
  282. */
  283. case l7.ProtocolMysql:
  284. if r.Method != l7.MethodStatementClose {
  285. stats.observe(r.Status.String(), "", r.Duration)
  286. }
  287. if c.l7Attach && c.valuableTrace(r.TraceId) {
  288. if conn.mysqlParser == nil {
  289. conn.mysqlParser = l7.NewMysqlParser()
  290. }
  291. query := conn.mysqlParser.Parse(r.Payload, r.StatementId)
  292. //trace.MysqlQuery(query, r.Status.Error(), r.Duration)
  293. if c.AppInfo.AppName != "" {
  294. klog.Debugf("[%s] ->>>>> %s -> %s payload:[%s]", c.AppInfo.AppName, r.Protocol.String(), conn.ActualDest, query)
  295. }
  296. //apmTrace, ok := c.getTrace(r.TraceId)
  297. apmTrace, err := c.getOrInitTrace(r.TraceId)
  298. //fmt.Println("mysql r.TraceId:", r.TraceId)
  299. //fmt.Println("ok:", ok)
  300. //fmt.Println("traceMap:", len(c.traceMap))
  301. if err == nil {
  302. dbSystem := semconv.DBSystemMySQL
  303. // 根据端口白名单确定协议类型
  304. l7Type := flags.GetProtocolByPort(uint16(conn.ActualDest.Port()))
  305. if l7Type == l7.ProtocolMariaDB {
  306. dbSystem = semconv.DBSystemMariaDB
  307. }
  308. //apmTrace.MysqlTraceQuery(query, r.Status.Error(), r.Duration, conn.ActualDest)
  309. //apmTrace.MysqlTraceQueryEvent(query, r, conn.ActualDest)
  310. apmTrace.SQLTraceQueryEvent(l7Type, dbSystem, query, r, conn.ActualDest)
  311. c.SendEvent(apmTrace, r.TraceId)
  312. }
  313. }
  314. /**
  315. * DM (达梦数据库)
  316. */
  317. case l7.ProtocolDM:
  318. //统计dm的query次数
  319. stats.observe(r.Status.String(), "", r.Duration)
  320. //是否发送数据
  321. if c.l7Attach && c.valuableTrace(r.TraceId) {
  322. if conn.dmParser == nil {
  323. conn.dmParser = l7.NewDmParser()
  324. }
  325. query := conn.dmParser.Parse(r.Payload, r.StatementId)
  326. if c.AppInfo.AppName != "" {
  327. klog.Debugf("[%s] ->>>>> %s -> %s payload:[%s]", c.AppInfo.AppName, r.Protocol.String(), conn.ActualDest, query)
  328. }
  329. apmTrace, err := c.getOrInitTrace(r.TraceId)
  330. if err == nil {
  331. //apmTrace.DmTraceQueryEvent(query, r, conn.ActualDest)
  332. apmTrace.SQLTraceQueryEvent(r.Protocol, semconv.DBSystemDaMengDB, query, r, conn.ActualDest)
  333. c.SendEvent(apmTrace, r.TraceId)
  334. }
  335. }
  336. /**
  337. * Memcached
  338. */
  339. case l7.ProtocolMemcached:
  340. stats.observe(r.Status.String(), "", r.Duration)
  341. if c.l7Attach && c.valuableTrace(r.TraceId) {
  342. cmd, items := l7.ParseMemcached(r.Payload)
  343. if c.AppInfo.AppName != "" {
  344. klog.Debugf("[%s] ->>>>> %s -> %s payload:[%s]", c.AppInfo.AppName, r.Protocol.String(), conn.ActualDest, cmd+" "+strings.Join(items, " "))
  345. }
  346. apmTrace, err := c.getOrInitTrace(r.TraceId)
  347. if err == nil {
  348. statement := cmd
  349. if len(items) == 1 {
  350. statement += " " + items[0]
  351. } else if len(items) > 1 {
  352. joined := fmt.Sprintf("[%s]", strings.Join(items, " "))
  353. statement += " " + joined
  354. }
  355. apmTrace.NoSQLTraceQueryEvent(r.Protocol, semconv.DBSystemMemcached, cmd, statement, r, conn.ActualDest)
  356. c.SendEvent(apmTrace, r.TraceId)
  357. }
  358. }
  359. /**
  360. * Redis
  361. */
  362. case l7.ProtocolRedis:
  363. stats.observe(r.Status.String(), "", r.Duration)
  364. if c.l7Attach && c.valuableTrace(r.TraceId) {
  365. cmd, args := l7.ParseRedis(r.Payload)
  366. //fmt.Println("cmd", cmd)
  367. //fmt.Println("args", args)
  368. //apmTrace, ok := c.getTrace(r.TraceId)
  369. if c.AppInfo.AppName != "" {
  370. klog.Debugf("[%s] ->>>>> %s -> %s payload:[%s]", c.AppInfo.AppName, r.Protocol.String(), conn.ActualDest, cmd)
  371. }
  372. apmTrace, err := c.getOrInitTrace(r.TraceId)
  373. if err == nil {
  374. statement := cmd
  375. if args != "" {
  376. statement += " " + args
  377. }
  378. apmTrace.NoSQLTraceQueryEvent(r.Protocol, semconv.DBSystemRedis, cmd, statement, r, conn.ActualDest)
  379. c.SendEvent(apmTrace, r.TraceId)
  380. }
  381. }
  382. //trace.RedisQuery(cmd, args, r.Status.Error(), r.Duration)
  383. /**
  384. * gRPC
  385. */
  386. case l7.ProtocolGrpc:
  387. klog.Debugln("enter the l7.ProtocolGrpc")
  388. stats.observe(r.Status.String(), "", r.Duration)
  389. if c.l7Attach && c.valuableTrace(r.TraceId) {
  390. apmTrace, err := c.getOrInitTrace(r.TraceId)
  391. if err == nil {
  392. apmTrace.GrpcClientTraceQueryEvent(r)
  393. c.SendEvent(apmTrace, r.TraceId)
  394. }
  395. }
  396. /**
  397. * MongoDB
  398. */
  399. case l7.ProtocolMongo:
  400. stats.observe(r.Status.String(), "", r.Duration)
  401. if c.l7Attach && c.valuableTrace(r.TraceId) {
  402. query := l7.ParseMongo(r.Payload)
  403. if c.AppInfo.AppName != "" {
  404. klog.Debugf("[%s] ->>>>> MongoDB -> %s SQL:[%s]", c.AppInfo.AppName, conn.ActualDest, query)
  405. }
  406. apmTrace, err := c.getOrInitTrace(r.TraceId)
  407. if err == nil {
  408. // MongoDB query 格式通常是 JSON,如 {"insert":"users"} 或 {"find":"users","filter":{...}}
  409. apmTrace.NoSQLTraceQueryEvent(r.Protocol, semconv.DBSystemMongoDB, "", query, r, conn.ActualDest)
  410. c.SendEvent(apmTrace, r.TraceId)
  411. }
  412. }
  413. /**
  414. * Kafka / Cassandra
  415. */
  416. case l7.ProtocolKafka, l7.ProtocolCassandra:
  417. stats.observe(r.Status.String(), "", r.Duration)
  418. if c.l7Attach && c.valuableTrace(r.TraceId) {
  419. }
  420. /**
  421. * RabbitMQ / NATS
  422. */
  423. case l7.ProtocolRabbitmq, l7.ProtocolNats:
  424. stats.observe(r.Status.String(), r.Method.String(), 0)
  425. if c.l7Attach && c.valuableTrace(r.TraceId) {
  426. }
  427. /**
  428. * Dubbo2
  429. */
  430. case l7.ProtocolDubbo2:
  431. stats.observe(r.Status.String(), "", r.Duration)
  432. if c.l7Attach && c.valuableTrace(r.TraceId) {
  433. }
  434. }
  435. return nil
  436. }
  437. func (c *Container) buildIDs(pid uint32) bool {
  438. c.lock.Lock()
  439. defer c.lock.Unlock()
  440. p := c.processes[pid]
  441. if p != nil {
  442. p.cmdline = string(proc.GetRealCmdline(pid))
  443. }
  444. var sns []string
  445. var sport uint16
  446. for address, val := range c.getListens() {
  447. if val == 1 {
  448. ip := address.IP()
  449. if ip.Is4() && !ip.IsLoopback() {
  450. // 获取端口号
  451. sport = address.Port()
  452. sns = append(sns, fmt.Sprintf("%s:%d", ip, sport))
  453. ////c.instanceID.IntVal, c.instanceID.HashtVal, _ =
  454. //c.AppInfo.Sn = ip.String()
  455. //c.AppInfo.Sport = int(port)
  456. //strInstanceID := utils.BuildInt64ID(fmt.Sprintf("%s:%d", ip.String(), port))
  457. //fmt.Println(port)
  458. ////os.Exit(1)
  459. //c.AppInfo.InstanceIdHash.IntVal, _ = strInstanceID.ToInt64()
  460. //c.AppInfo.InstanceIdHash.HashtVal = strInstanceID.ToHashByte()
  461. ////c.AppInfo.InstanceId = c.instanceID.IntVal
  462. //strAgentID := utils.BuildInt64ID(fmt.Sprintf("%s:%s", strInstanceID, string(proc.GetExe(pid))))
  463. //c.AppInfo.AgentId, _ = strAgentID.ToInt64()
  464. //c.AppInfo.CodeType = c.GetCodeTypeFromCache(pid)
  465. //return true
  466. }
  467. }
  468. }
  469. if len(sns) > 0 {
  470. //c.instanceID.IntVal, c.instanceID.HashtVal, _ =
  471. snsStr := strings.Join(sns, ",")
  472. c.AppInfo.Sn = snsStr
  473. c.AppInfo.Sport = int(sport)
  474. strInstanceID := utils.BuildInt64ID(fmt.Sprintf("%s:%d", c.AppInfo.Sn, sport))
  475. c.AppInfo.InstanceIdHash.IntVal, _ = strInstanceID.ToInt64()
  476. c.AppInfo.InstanceIdHash.HashtVal = strInstanceID.ToHashByte()
  477. // strAgentID := utils.BuildInt64ID(fmt.Sprintf("%s:%s", utils.GetHostIP(), string(proc.GetExe(pid))))
  478. // c.AppInfo.AgentId, _ = strAgentID.ToInt64()
  479. // c.AppInfo.CodeType = c.GetCodeTypeFromCache(pid)
  480. return true
  481. }
  482. return false
  483. }
  484. func (c *Container) ReBuildIds(pid uint32) {
  485. c.lock.Lock()
  486. defer c.lock.Unlock()
  487. var sns []string
  488. var sport uint16
  489. for address, val := range c.getListens() {
  490. if val == 1 {
  491. ip := address.IP()
  492. if ip.Is4() && !ip.IsLoopback() {
  493. // 获取端口号
  494. sport = address.Port()
  495. sns = append(sns, fmt.Sprintf("%s:%d", ip, sport))
  496. }
  497. }
  498. }
  499. if len(sns) > 0 {
  500. snsStr := strings.Join(sns, ",")
  501. c.AppInfo.Sn = snsStr
  502. c.AppInfo.Sport = int(sport)
  503. strInstanceID := utils.BuildInt64ID(fmt.Sprintf("%s:%d", c.AppInfo.Sn, sport))
  504. c.AppInfo.InstanceIdHash.IntVal, _ = strInstanceID.ToInt64()
  505. c.AppInfo.InstanceIdHash.HashtVal = strInstanceID.ToHashByte()
  506. // strAgentID := utils.BuildInt64ID(fmt.Sprintf("%s:%s", utils.GetHostIP(), string(proc.GetExe(pid))))
  507. strAgentID := utils.BuildInt64ID(fmt.Sprintf("%s:%d", utils.GetHostIP(), c.AppInfo.AppIdHash.IntVal))
  508. c.AppInfo.AgentId, _ = strAgentID.ToInt64()
  509. c.AppInfo.CodeType = c.GetCodeTypeFromCache(pid)
  510. }
  511. }
  512. func (c *Container) StackProcess(event ebpftracer.StackEvent, tracer *ebpftracer.Tracer) {
  513. c.lock.Lock()
  514. defer c.lock.Unlock()
  515. // get the associated uprobe
  516. uprobe, err := c.GetUprobe(event, tracer)
  517. if err != nil {
  518. //fmt.Println("GetUprobeGetUprobe errer: %v", err)
  519. klog.Errorf("failed to get uprobe for event %+v: %+v", event, err)
  520. return
  521. }
  522. if event.TraceId <= 0 {
  523. //fmt.Println("StackProcess TraceId id 0")
  524. klog.Errorf("failed to get uprobe(traceId is <= 0) for event %+v", event)
  525. return
  526. }
  527. // fmt.Printf("StackProcess 函数入口开始处理 fun:TraceId:%lld, Funcname:%s, time: %lld\n", event.TraceId, uprobe.Funcname, event.TimeNsEnd-event.TimeNsStart)
  528. stackFun := ebpftracer.StackFunEvent{}
  529. stackFun.Uprobe = &uprobe
  530. stackFun.StackEvent = event
  531. apmTrace, ok := c.getTrace(event.TraceId)
  532. if ok {
  533. apmTrace.FunAdd(stackFun)
  534. }
  535. }
  536. func byteExtractString(nameString [100]byte) string {
  537. n := bytes.IndexFunc(nameString[:], func(r rune) bool {
  538. return r == 0 || r < 32 || r > 126 // 截取到第一个零值或非打印字符
  539. })
  540. if n == -1 {
  541. n = len(nameString) // 没找到零值或非打印字符,使用数组长度
  542. }
  543. return string(nameString[:n])
  544. }
  545. func (c *Container) StackProcess2(event ebpftracer.StackEvent, tracer *ebpftracer.Tracer) {
  546. c.lock.Lock()
  547. defer c.lock.Unlock()
  548. // get the associated uprobe
  549. switch event.Location {
  550. case 0: // ret
  551. Funcname := ""
  552. if event.Type != uint64(CodeTypeJava) {
  553. uprobe, err := c.GetUprobe(event, tracer)
  554. if err != nil {
  555. //fmt.Println("GetUprobeGetUprobe errer: %v", err)
  556. klog.Errorf("failed to get uprobe for event %+v: %+v", event, err)
  557. return
  558. }
  559. Funcname = uprobe.Funcname
  560. } else {
  561. ClassName := byteExtractString(event.ClassName)
  562. MethedName := byteExtractString(event.MethedName)
  563. Funcname = ClassName + "." + MethedName
  564. }
  565. if event.TraceId <= 0 {
  566. //fmt.Println("StackProcess TraceId id 0")
  567. klog.Errorf("failed to get uprobe(traceId is <= 0) for event %+v", event)
  568. return
  569. }
  570. //fmt.Printf("StackProcess 函数入口开始处理 fun:TraceId:%lld, Funcname:%s, time: %lld\n", event.TraceId, uprobe.Funcname, event.TimeNsEnd-event.TimeNsStart)
  571. apmTrace, err := c.getOrInitTrace(event.TraceId)
  572. if err == nil {
  573. //fmt.Println("append FuncTraceQuery fun:", event.TraceId, uprobe.Funcname, event.Pid)
  574. duration := event.TimeNsEnd - event.TimeNsStart
  575. apmTrace.FuncTraceQuery(Funcname, time.Duration(duration), event.TimeNsStart, event.TimeNsEnd)
  576. c.SendEvent(apmTrace, event.TraceId)
  577. }
  578. }
  579. }
  580. // ResolveAddress returns the symbol(s) and offset of the given address.
  581. func (c *Container) ResolveAddress(addr uint64, symbols []elf.Symbol) (syms []elf.Symbol, offset uint, err error) {
  582. if addr == 0 {
  583. // err = errors.Wrapf(SymbolNotFoundError, "0")
  584. return
  585. }
  586. // symbols, _, err := e.Symbols()
  587. if err != nil {
  588. return
  589. }
  590. idx := sort.Search(len(symbols), func(i int) bool { return symbols[i].Value > addr })
  591. if idx == 0 {
  592. // err = errors.Wrap(SymbolNotFoundError, fmt.Sprintf("%x", addr))
  593. return
  594. }
  595. // why diff symbol may contains the same addr?
  596. sym := symbols[idx-1]
  597. for i := idx - 1; i >= 0 && symbols[i].Value == sym.Value; i-- {
  598. syms = append(syms, symbols[i])
  599. }
  600. for i := idx; i < len(symbols) && symbols[i].Value == sym.Value; i++ {
  601. syms = append(syms, symbols[i])
  602. }
  603. return syms, uint(addr - sym.Value), nil
  604. }
  605. type MemoryMap struct {
  606. Start, End uint64
  607. }
  608. // ReadFirstLineOfMapsFile reads the first line of /proc/<pid>/maps file and return the memory map as a MemoryMap struct
  609. func ReadFirstLineOfMapsFile(pid string) (*MemoryMap, error) {
  610. file, err := os.Open(fmt.Sprintf("/proc/%s/maps", pid))
  611. if err != nil {
  612. return nil, err
  613. }
  614. defer file.Close()
  615. scanner := bufio.NewScanner(file)
  616. if scanner.Scan() {
  617. fields := strings.Fields(scanner.Text())
  618. addresses := strings.Split(fields[0], "-")
  619. if len(addresses) != 2 {
  620. return nil, errors.New("unexpected format in /proc/<pid>/maps")
  621. }
  622. start, err := strconv.ParseUint(addresses[0], 16, 64)
  623. if err != nil {
  624. return nil, err
  625. }
  626. end, err := strconv.ParseUint(addresses[1], 16, 64)
  627. if err != nil {
  628. return nil, err
  629. }
  630. return &MemoryMap{
  631. Start: start,
  632. End: end,
  633. }, nil
  634. }
  635. if err := scanner.Err(); err != nil {
  636. return nil, err
  637. }
  638. return nil, errors.New("empty /proc/<pid>/maps")
  639. }
  640. func (c *Container) GetUprobe(event ebpftracer.StackEvent, tracer *ebpftracer.Tracer) (uprobe tracer.Uprobe, err error) {
  641. //fmt.Println("GetUprobe entory:")
  642. memoryMap, _ := ReadFirstLineOfMapsFile(strconv.Itoa(int(event.Pid)))
  643. Address := event.Ip - memoryMap.Start
  644. // fmt.Printf("memoryMap.Start: %x, event.Ip: %x, Address: %x\n", memoryMap.Start, event.Ip, Address)
  645. for _, fun := range c.UprobesMap {
  646. funAddress := fun.Address + fun.AbsOffset
  647. // fmt.Printf("GetUprobeGetUprobeGetUprobe:fun.Address %x, fun.AbsOffset: %x\n", fun.Address, fun.AbsOffset)
  648. if funAddress == Address {
  649. // fmt.Printf("---GetUprobeGetUprobeGetUprobe: %x, event.Ip: %x ---- %s--%x\n", memoryMap.Start, event.Ip, fun.Funcname, fun.Address)
  650. return fun, nil
  651. }
  652. }
  653. syms, _, err := c.ResolveAddress(event.Ip, tracer.Symbols)
  654. if err != nil {
  655. return
  656. }
  657. for _, sym := range syms {
  658. //fmt.Println("GetUprobeGetUprobeGetUprobe: %s+%d", sym.Name, offset)
  659. uprobe, ok := tracer.UprobesMap[fmt.Sprintf("%s-%s", sym.Name, sym.Value)]
  660. if ok {
  661. return uprobe, nil
  662. }
  663. }
  664. err = errors.New("uprobe not found")
  665. return
  666. }
  667. func (c *Container) GetAppInfo() AppInfo {
  668. return c.AppInfo
  669. }
  670. // 可注入前置
  671. func (c *Container) checkEventReady() bool {
  672. c.lock.Lock()
  673. defer c.lock.Unlock()
  674. return c.l7EventReady
  675. }
  676. func (c *Container) eventReady() {
  677. c.lock.Lock()
  678. defer c.lock.Unlock()
  679. c.l7EventReady = true
  680. }
  681. // uprobe前置
  682. func (c *Container) Isl7AttachSuccess() bool {
  683. c.lock.Lock()
  684. defer c.lock.Unlock()
  685. return c.l7Attach
  686. }
  687. func (c *Container) l7AttachSuccess() {
  688. c.lock.Lock()
  689. defer c.lock.Unlock()
  690. c.l7Attach = true
  691. }
  692. func (c *Container) ctrlStack(r *Registry, pid uint32) {
  693. resp, err := c.GetCodeSetting(r)
  694. if err != nil {
  695. klog.WithField("pid", pid).WithError(err).Error("[ctrlStack] GetCodeSetting failed.")
  696. return
  697. }
  698. if resp.BlackWhiteSettings.CollectStack == OPEN_STACK {
  699. // 有黑白名单规则 &&
  700. // 之前有注入 先卸载再注入
  701. // 之前没注入 直接注入
  702. // 没有有黑白名单 直接卸载
  703. if c.hasStackRule(resp) {
  704. if c.stackRuleUpdate(resp) {
  705. // 重新注入
  706. err = c.DetachStack(pid, APP_UNINSTALL)
  707. if err != nil {
  708. klog.WithError(err).Errorf("[ctrlStack][end] Failed detach stack trace!")
  709. }
  710. }
  711. klog.WithField("pid", pid).Infoln("[ctrlStack] Attach app stack.")
  712. c.saveWhiteStackSettingInfo(resp)
  713. err = c.AttachStack(r.tracer, pid)
  714. if err != nil {
  715. c.AppInfo.SetAppStackError()
  716. klog.WithField("pid", pid).WithError(err).Errorf("[ctrlStack][end] Failed attach stack trace!")
  717. }
  718. } else {
  719. if c.noOrigRule() {
  720. return
  721. }
  722. c.saveWhiteStackSettingInfo(resp)
  723. // 关闭堆栈
  724. err = c.DetachStack(pid, APP_UNINSTALL)
  725. if err != nil {
  726. klog.WithError(err).Errorf("[ctrlStack][end] Failed detach stack trace!")
  727. }
  728. }
  729. } else {
  730. if c.noOrigRule() {
  731. return
  732. }
  733. c.saveWhiteStackSettingInfo(resp)
  734. // 关闭堆栈
  735. err = c.DetachStack(pid, APP_UNINSTALL)
  736. if err != nil {
  737. klog.WithError(err).Errorf("[ctrlStack][end] Failed detach stack trace!")
  738. }
  739. }
  740. }
  741. func (c *Container) verifyAttachConditions(r *Registry, pid uint32) (bool, int) {
  742. p := c.processes[pid]
  743. if p != nil && c.checkEventReady() {
  744. codeType := c.GetCodeTypeFromCache(pid)
  745. if codeType.IsUnknownCode() {
  746. klog.WithField("pid", pid).Debug("[verify] unknown language.")
  747. return false, 0
  748. }
  749. cmdline := p.GetCmdline()
  750. if len(cmdline) == 0 {
  751. return false, 0
  752. }
  753. //whiteListByCode := r.getWhiteListByCodeType(codeType)
  754. whiteListByCode := r.getWhiteListAll()
  755. //klog.WithField("pid", pid).WithField("codeType", codeType.String()).
  756. // Infof("[verify] white list %v", utils.ToString(whiteListByCode))
  757. // 当前语言的白名单规则
  758. for _, setting := range whiteListByCode {
  759. ruleVal := setting.Filters
  760. if ruleVal == "" {
  761. continue
  762. }
  763. // 判断规则
  764. if strings.Contains(cmdline, ruleVal) {
  765. //if !codeType.IsJvmCode() {
  766. // klog.WithField("pid", pid).Warning("[verify] This agent version only supports JVM applications.")
  767. // return false, 0
  768. //}
  769. c.WhiteSettingInfo.AppName = setting.AppName
  770. c.WhiteSettingInfo.Filters = setting.Filters
  771. klog.WithField("pid", pid).
  772. WithField("codeType", codeType.String()).
  773. WithField("ruleVal", ruleVal).
  774. WithField("cmdline", cmdline).
  775. //WithField("stack", setting.OpenStack).
  776. WithField("white list", utils.ToString(whiteListByCode)).
  777. Infoln("[verify] check successful.")
  778. return true, 0
  779. }
  780. }
  781. }
  782. return false, 0
  783. }
  784. // 1.卸载入口
  785. func (c *Container) Detach(tracer *ebpftracer.Tracer, pid uint32, detachType APP_TYPE) {
  786. c.lock.Lock()
  787. defer c.lock.Unlock()
  788. if p := c.processes[pid]; p != nil {
  789. err := c.DetachUprobes(tracer, pid, detachType)
  790. if err != nil {
  791. klog.WithError(err).Errorln("DetachUprobes Error.")
  792. }
  793. err = c.DetachStack(pid, detachType)
  794. if err != nil {
  795. klog.WithError(err).Errorln("DetachStack Error.")
  796. }
  797. // 关闭7层监控
  798. c.l7Attach = false
  799. // 变更应用状态
  800. if err != nil {
  801. detachType = detachType.Error()
  802. }
  803. c.AppInfo.SetAppStatus(detachType)
  804. }
  805. }
  806. // 1.1卸载uprobe
  807. func (c *Container) DetachUprobes(tracer *ebpftracer.Tracer, pid uint32, detachType APP_TYPE) error {
  808. // close uprobe
  809. if p := c.processes[pid]; p != nil {
  810. for _, u := range p.uprobes {
  811. err := u.Close()
  812. if err != nil {
  813. return err
  814. }
  815. }
  816. p.uprobes = []link.Link{}
  817. switch detachType {
  818. case APP_UNINSTALL, APP_FUSE:
  819. codeType := c.GetCodeTypeFromCache(pid)
  820. switch codeType {
  821. case CodeTypeJava:
  822. p.jvmAttachOnce = false
  823. case CodeTypeGo:
  824. p.goTlsUprobesChecked = false
  825. p.openSslUprobesChecked = false
  826. default:
  827. }
  828. case APP_UPROBE_ERROR:
  829. klog.Infof("[DetachUprobes] ERROR_DETACH for pid %d", pid)
  830. default:
  831. }
  832. //delete the proc info form proc_info_map(for kernel) when the uprobe detached
  833. if err := tracer.DelKProcInfo(pid); err != nil {
  834. return fmt.Errorf("[DetachUprobes] failed to delete KProcInfo for pid %d, detach type is:%s", pid, detachType)
  835. } else {
  836. klog.Infof("[DetachUprobes] delete KProcInfo success for pid %d,detachType:%d", pid, detachType)
  837. c.AppInfo.EBPFProcInfo = nil
  838. }
  839. } else {
  840. return fmt.Errorf("[DetachUprobes] cannot find uprobe for pid %d", pid)
  841. }
  842. return nil
  843. }
  844. // 1.2卸载堆栈
  845. func (c *Container) DetachStack(pid uint32, detachType APP_TYPE) error {
  846. if p := c.processes[pid]; p != nil {
  847. var err error
  848. codeType := c.GetCodeTypeFromCache(pid)
  849. switch codeType {
  850. // 1.2.1 卸载 jvm堆栈
  851. case CodeTypeJava:
  852. err = c.detachJvmStack(pid)
  853. default:
  854. err = p.closeStackUprobes()
  855. }
  856. if err != nil {
  857. klog.WithError(err).Errorln("[detachStack] failed to detach stack")
  858. return err
  859. }
  860. p.stackAttachOnce = false
  861. } else {
  862. return fmt.Errorf("[DetachStack] cannot find uprobe for pid %d", pid)
  863. }
  864. return nil
  865. }
  866. // 1.2.1 卸载 jvm堆栈
  867. func (c *Container) detachJvmStack(pid uint32) error {
  868. if p := c.processes[pid]; p != nil {
  869. //if p.stackStatus.IsStackUprobesSuccess() || len(p.stackUprobes) > 0 {
  870. //}
  871. // 卸载 JavaAgent
  872. var err error
  873. if p.stackStatus.IsJattachSuccess() {
  874. // 卸载堆栈probes
  875. err = p.closeStackUprobes()
  876. if err != nil {
  877. klog.WithError(err).Errorf("[detachJvmStack] closeStackUprobes")
  878. }
  879. err = p.uninstallJavaAgent()
  880. if err != nil {
  881. klog.WithError(err).Errorf("[detachJvmStack] uninstallJavaAgent")
  882. }
  883. }
  884. return err
  885. }
  886. return nil
  887. }
  888. func (c *Container) getRootfs() string {
  889. if c.metadata != nil && c.metadata.rootfs != "" {
  890. return path.Join(*flags.HostDirPathPrefix, c.metadata.rootfs)
  891. }
  892. return ""
  893. }
  894. func (c *Container) BuildActiveApps(runtimeApps map[uint32]AppStatusInfo, pid uint32) {
  895. if c == nil {
  896. //klog.WithField("pid", pid).Warningln("[BuildActiveApps] container_apm is nil.")
  897. return
  898. }
  899. if c.AppInfo.AppName == "" {
  900. return
  901. }
  902. klog.WithField("pid", pid).WithField("appname", c.AppInfo.AppName).Infof("[BuildActiveApps] container %s is running.", c.AppInfo.AppName)
  903. detail := AppStatusInfo{
  904. Pid: pid,
  905. ProcName: c.containerName,
  906. AppName: c.AppInfo.AppName,
  907. Language: c.AppInfo.CodeType.String(),
  908. AppID: c.AppInfo.AppIdHash.IntVal,
  909. AgentID: c.AppInfo.AgentId,
  910. InstanceID: c.AppInfo.InstanceIdHash.IntVal,
  911. Sn: c.AppInfo.Sn,
  912. Sport: c.AppInfo.Sport,
  913. RegisterAt: time.Unix(c.AppInfo.RegisterAt, 0).Format("060102 15:04:05"),
  914. PreStatus: c.AppInfo.PreStatus,
  915. Status: c.AppInfo.Status,
  916. Rule: c.WhiteSettingInfo.Filters,
  917. Container: string(c.id),
  918. }
  919. detail.Rule = fmt.Sprintf("%s|%d", c.WhiteSettingInfo.Filters, c.WhiteSettingInfo.WhiteStackSettingInfo.OpenStack)
  920. if c.AppInfo.UpdateAt != 0 {
  921. detail.UpdateAt = time.Unix(c.AppInfo.UpdateAt, 0).Format("060102 15:04:05")
  922. }
  923. p := c.processes[pid]
  924. if p != nil {
  925. detail.StackStatus = p.stackStatus.String()
  926. v := 0
  927. if !p.versionFailed {
  928. v = 1
  929. }
  930. detail.StackStatus += fmt.Sprintf("V=%d", v)
  931. }
  932. runtimeApps[pid] = detail
  933. }
  934. func (c *Container) AgentCtrl(r *Registry, pid uint32) {
  935. if c == nil {
  936. //klog.WithField("pid", pid).Warningln("[AgentCtrl] cannot find container.")
  937. return
  938. }
  939. var err error
  940. verifyAttachConditions, _ := c.verifyAttachConditions(r, pid)
  941. // fusing UNINSTALL
  942. if r.isFusing && c.Isl7AttachSuccess() {
  943. c.Detach(r.tracer, pid, APP_FUSE)
  944. klog.WithField("pid", pid).Infoln("[AgentCtrl] fusing")
  945. return
  946. }
  947. // verify UNINSTALL
  948. if !verifyAttachConditions && c.Isl7AttachSuccess() {
  949. c.Detach(r.tracer, pid, APP_UNINSTALL)
  950. klog.WithField("pid", pid).Infoln("[AgentCtrl] rule uninstall.")
  951. return
  952. }
  953. if verifyAttachConditions {
  954. err = c.RegisterAppInfo(r, pid)
  955. if err != nil {
  956. klog.WithError(err).Errorf("[AgentCtrl] Failed registerAppInfo.")
  957. return
  958. }
  959. klog.WithField("pid", pid).Infoln("[AgentCtrl] Attach uprobes.")
  960. err = c.AttachUprobes(r.tracer, pid, "Agentctrl")
  961. if err != nil {
  962. klog.WithField("pid", pid).WithError(err).Errorf("[AgentCtrl] Failed attach uprobes error!")
  963. return
  964. } else {
  965. klog.WithField("pid", pid).Infoln("[AgentCtrl] Attach uprobes success!")
  966. }
  967. // 堆栈控制
  968. c.ctrlStack(r, pid)
  969. }
  970. }