container_apm.go 36 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208
  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. "go.opentelemetry.io/otel/attribute"
  26. semconv "go.opentelemetry.io/otel/semconv/v1.18.0"
  27. "inet.af/netaddr"
  28. )
  29. func (c *Container) getTrace(traceId uint64) (*tracing.Trace, bool) {
  30. trace, ok := c.traceMap[traceId]
  31. return trace, ok
  32. }
  33. func (c *Container) createTraceMap(traceId uint64, trace *tracing.Trace) {
  34. c.traceMap[traceId] = trace
  35. }
  36. // 查询或创建trace信息
  37. func (c *Container) getOrInitTrace(traceId uint64) (*tracing.Trace, error) {
  38. trace, ok := c.getTrace(traceId)
  39. if !ok {
  40. //new trace
  41. trace = tracing.NewTraceFromEvent(string(c.id))
  42. if trace == nil {
  43. // tracer 未初始化,返回错误
  44. return nil, fmt.Errorf("tracer is not initialized, cannot create trace")
  45. }
  46. //create TraceMap
  47. c.createTraceMap(traceId, trace)
  48. //create ParentSpan
  49. trace.CreateRootSpan(traceId)
  50. }
  51. return trace, nil
  52. }
  53. // getGrpcServerNetworkInfo 获取 gRPC server 的网络信息
  54. // 返回: IP地址, 端口号, 容器ID
  55. func (c *Container) getGrpcServerNetworkInfo() (string, uint16, string) {
  56. containerID := ""
  57. if c.cgroup != nil {
  58. containerID = c.cgroup.ContainerId
  59. }
  60. ipAddr := ""
  61. ifaces, err := net.Interfaces()
  62. if err == nil {
  63. for _, iface := range ifaces {
  64. if iface.Name == "eth0" {
  65. addrs, err := iface.Addrs()
  66. if err == nil {
  67. for _, addr := range addrs {
  68. var ipnet *net.IPNet
  69. switch v := addr.(type) {
  70. case *net.IPNet:
  71. ipnet = v
  72. case *net.IPAddr:
  73. ipnet = &net.IPNet{IP: v.IP, Mask: v.IP.DefaultMask()}
  74. }
  75. if ipnet != nil && ipnet.IP.To4() != nil {
  76. ipAddr = ipnet.IP.String()
  77. break
  78. }
  79. }
  80. }
  81. break
  82. }
  83. }
  84. }
  85. klog.Debugf("grpc server ip %s", ipAddr)
  86. // 本地端口尝试从AppInfo.Sport获取
  87. port := c.AppInfo.Sport
  88. klog.Debugf("grpc server port %d", port)
  89. return ipAddr, uint16(port), containerID
  90. }
  91. // Deprecated: InitTrace not used
  92. //func (c *Container) InitTrace(traceId uint64, r *l7.RequestData) error {
  93. // method, path, hostIp, port := l7.ParseHttpHost(r.Payload)
  94. // ip, err := netaddr.ParseIP(hostIp)
  95. // if err != nil {
  96. // //fmt.Println("host ip error")
  97. // hostIp = "127.0.0.1"
  98. // }
  99. // addr := netaddr.IPPortFrom(ip, port)
  100. // trace := tracing.NewTrace(string(c.id), addr)
  101. // if trace == nil {
  102. // return fmt.Errorf("OTEL_EXPORTER_OTLP_TRACES_ENDPOINT is null")
  103. // }
  104. // c.traceMap[traceId] = trace
  105. //
  106. // trace.TraceStart(method, path, r.Status, r.Duration)
  107. // return nil
  108. //}
  109. // 在任意阶段,r.TraceId 不等于0 则创建 traceMap && createParentSpan
  110. // 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
  111. func (c *Container) SendEvent(t *tracing.Trace, traceID uint64) {
  112. if t.AllEventReady(traceID) {
  113. t.SendEvent()
  114. klog.Debugf("SendEvent %d", traceID)
  115. //fmt.Println(t.GetSpan())
  116. //fmt.Println("===============")
  117. delete(c.traceMap, traceID)
  118. }
  119. }
  120. func (c *Container) valuableTrace(traceID uint64) bool {
  121. return traceID != 0
  122. }
  123. func (c *Container) onL7RequestApm(pid uint32, fd uint64, timestamp uint64, r *l7.RequestData) map[netaddr.IP]string {
  124. c.lock.Lock()
  125. defer c.lock.Unlock()
  126. if r.Protocol == l7.ProtocolDNS {
  127. ip2fqdn, _type, fqdn, ttl, ips := c.onDNSRequest(r)
  128. if c.l7Attach && c.valuableTrace(r.TraceId) {
  129. apmTrace, err := c.getOrInitTrace(r.TraceId)
  130. if err == nil {
  131. apmTrace.DNSTraceQueryEvent(r, _type, fqdn, ttl, ips)
  132. c.SendEvent(apmTrace, r.TraceId)
  133. }
  134. }
  135. return ip2fqdn
  136. }
  137. //if !c.valuableTrace(r.TraceId) {
  138. // return nil
  139. //}
  140. //klog.Infof("====ProtocolTrace+++++ start==== %d %d", pid, r.TraceId)
  141. // klog.Infof("====ProtocolTrace===== start==== %d %d", r.Protocol == l7.ProtocolTrace, c.l7Attach)
  142. if c.l7Attach && c.valuableTrace(r.TraceId) {
  143. // klog.Infof("====ProtocolTrace---- start==== %d %d", pid, r.TraceId)
  144. if r.TraceStart == TRACE_STATUS {
  145. // klog.Infof("====ProtocolTrace start==== %d %d", pid, r.TraceId)
  146. trace, err := c.getOrInitTrace(r.TraceId)
  147. if c.AppInfo.AppName != "" {
  148. klog.Debugf("->>> [%s] -> payload:[%s]", c.AppInfo.AppName, r.Payload)
  149. }
  150. if err == nil {
  151. switch r.Protocol {
  152. case l7.ProtocolHTTP:
  153. // cwother
  154. if !c.AppInfo.CodeType.IsGoCode() {
  155. r.SysvcFrom = l7.ParseHttpSysvcFrom(r.Payload)
  156. }
  157. method, requestURI, sn, sport, userAgent := l7.ParseHttpHostWithUserAgent(r.Payload, r.IsTls)
  158. // userAgent 可以在这里使用,例如传递给 trace.TraceStartEvent
  159. ip, _ := netaddr.ParseIP(sn)
  160. //codeType := c.GetCodeTypeFromCache(pid)
  161. container_id := ""
  162. if c.cgroup != nil {
  163. container_id = c.cgroup.ContainerId
  164. }
  165. trace.TraceStartEvent(method, requestURI, sn, userAgent, sport, r, netaddr.IPPortFrom(ip, sport), pid, c.GetAppInfo(), container_id)
  166. c.SendEvent(trace, r.TraceId)
  167. case l7.ProtocolGrpc:
  168. // gRPC
  169. ipAddr, port, containerID := c.getGrpcServerNetworkInfo()
  170. trace.GrpcServerTraceStartEvent(ipAddr, port, r, c.GetAppInfo(), containerID)
  171. c.SendEvent(trace, r.TraceId)
  172. case l7.ProtocolKafka:
  173. var sn string
  174. var sport uint16
  175. var container_id string
  176. conn := c.connectionsByPidFd[PidFd{Pid: pid, Fd: fd}]
  177. if conn != nil {
  178. sn = conn.ActualDest.IP().String()
  179. sport = conn.ActualDest.Port()
  180. }
  181. if c.cgroup != nil {
  182. container_id = c.cgroup.ContainerId
  183. }
  184. // MQ
  185. trace.MQConsumerTraceStartEvent(sn, sport, r, c.GetAppInfo(), container_id)
  186. c.SendEvent(trace, r.TraceId)
  187. }
  188. }
  189. return nil
  190. }
  191. if r.TraceEnd == TRACE_STATUS {
  192. klog.Debugf("====ProtocolTrace end==== %d %d", pid, r.TraceId)
  193. trace, err := c.getOrInitTrace(r.TraceId)
  194. if err == nil {
  195. trace.TraceEndEvent(r)
  196. c.SendEvent(trace, r.TraceId)
  197. }
  198. return nil
  199. }
  200. }
  201. /**
  202. * HTTP
  203. */
  204. if r.Protocol == l7.ProtocolHTTP {
  205. if c.l7Attach && c.valuableTrace(r.TraceId) {
  206. // 检查是否启用 Elasticsearch 检测
  207. if *flags.EnableElasticsearchDetection {
  208. // 解析 User-Agent 以检测 Elasticsearch 请求
  209. method, requestURI, sn, sport, userAgent := l7.ParseHttpHostWithUserAgent(r.Payload, r.IsTls)
  210. // 检查是否是 Elasticsearch 请求(通过 User-Agent)
  211. isElasticsearch := strings.Contains(strings.ToLower(userAgent), "elasticsearch")
  212. apmTrace, err := c.getOrInitTrace(r.TraceId)
  213. if err == nil {
  214. if isElasticsearch {
  215. r.Protocol = l7.ProtocolES
  216. // Elasticsearch 请求,按照 NoSQL 方式处理
  217. query := l7.ParseElasticsearch(r.Payload)
  218. if c.AppInfo.AppName != "" {
  219. klog.Debugf("[%s] ->>>>> Elasticsearch -> %s:%d Query:[%s]", c.AppInfo.AppName, sn, sport, query)
  220. }
  221. conn := c.connectionsByPidFd[PidFd{Pid: pid, Fd: fd}]
  222. if conn == nil {
  223. conn = &ActiveConnection{
  224. Dest: r.ComponentDAddr,
  225. ActualDest: r.ComponentDAddr,
  226. Timestamp: timestamp,
  227. }
  228. }
  229. apmTrace.NoSQLTraceQueryEvent(r.Protocol, semconv.DBSystemElasticsearch, method, query, r, conn.Src, conn.ActualDest)
  230. } else {
  231. // 普通 HTTP 请求
  232. apmTrace.HttpTraceRequestEvent(method, requestURI, sn, sport, r)
  233. }
  234. c.SendEvent(apmTrace, r.TraceId)
  235. }
  236. } else {
  237. // Elasticsearch 检测未启用,使用普通 HTTP 处理
  238. method, requestURI, sn, sport := l7.ParseHttpHost(r.Payload, r.IsTls)
  239. apmTrace, err := c.getOrInitTrace(r.TraceId)
  240. if err == nil {
  241. apmTrace.HttpTraceRequestEvent(method, requestURI, sn, sport, r)
  242. c.SendEvent(apmTrace, r.TraceId)
  243. }
  244. }
  245. }
  246. //return nil
  247. }
  248. /**
  249. * gRPC
  250. */
  251. if r.Protocol == l7.ProtocolGrpc {
  252. klog.Infoln("conn == nil r.Protocol == l7.ProtocolGrpc")
  253. klog.Infoln("enter the l7.ProtocolGrpc")
  254. if c.l7Attach && c.valuableTrace(r.TraceId) {
  255. apmTrace, err := c.getOrInitTrace(r.TraceId)
  256. if err == nil {
  257. apmTrace.GrpcClientTraceQueryEvent(r)
  258. c.SendEvent(apmTrace, r.TraceId)
  259. }
  260. }
  261. }
  262. conn := c.connectionsByPidFd[PidFd{Pid: pid, Fd: fd}]
  263. //fmt.Println("l7.connectionsByPidFd", conn, pid, fd)
  264. if conn == nil {
  265. conn = &ActiveConnection{
  266. Dest: r.ComponentDAddr,
  267. ActualDest: r.ComponentDAddr,
  268. Timestamp: timestamp,
  269. }
  270. //return nil
  271. }
  272. if timestamp != 0 && conn.Timestamp != timestamp {
  273. //if r.Protocol == l7.ProtocolGrpc {
  274. // klog.Infoln("timestamp != 0 && conn.Timestamp != timestamp r.Protocol == l7.ProtocolGrpc")
  275. // klog.Infoln("enter the l7.ProtocolGrpc")
  276. // if c.l7Attach && c.valuableTrace(r.TraceId) {
  277. // apmTrace, err := c.getOrInitTrace(r.TraceId)
  278. // if err == nil {
  279. // apmTrace.GrpcClientTraceQueryEvent(r)
  280. // c.SendEvent(apmTrace, r.TraceId)
  281. // }
  282. // }
  283. //}
  284. return nil
  285. }
  286. stats := c.l7Stats.get(r.Protocol, conn.Dest, conn.ActualDest)
  287. //trace := tracing.NewTrace(string(c.id), conn.ActualDest)
  288. switch r.Protocol {
  289. /**
  290. * HTTP
  291. */
  292. case l7.ProtocolHTTP:
  293. if c.AppInfo.AppName != "" {
  294. klog.Debugf("[%s] ->>>>> curl -> %s payload:[%s]", c.AppInfo.AppName, conn.ActualDest, r.Payload)
  295. }
  296. stats.observe(r.Status.Http(), "", r.Duration)
  297. /**
  298. * HTTP2
  299. */
  300. case l7.ProtocolHTTP2:
  301. if conn.http2Parser == nil {
  302. conn.http2Parser = l7.NewHttp2Parser()
  303. }
  304. requests := conn.http2Parser.Parse(r.Method, r.Payload, uint64(r.Duration))
  305. for _, req := range requests {
  306. stats.observe(req.Status.Http(), "", req.Duration)
  307. //trace.Http2Request(req.Method, req.Path, req.Scheme, req.Status, req.Duration)
  308. }
  309. /**
  310. * PostgreSQL
  311. */
  312. case l7.ProtocolPostgres:
  313. if r.Method != l7.MethodStatementClose {
  314. stats.observe(r.Status.String(), "", r.Duration)
  315. }
  316. //if conn.postgresParser == nil {
  317. // conn.postgresParser = l7.NewPostgresParser()
  318. //}
  319. //query := conn.postgresParser.Parse(r.Payload)
  320. //trace.PostgresQuery(query, r.Status.Error(), r.Duration)
  321. if c.l7Attach && c.valuableTrace(r.TraceId) {
  322. if conn.postgresParser == nil {
  323. conn.postgresParser = l7.NewPostgresParser()
  324. }
  325. query := conn.postgresParser.Parse(r.Payload)
  326. //trace.MysqlQuery(query, r.Status.Error(), r.Duration)
  327. if c.AppInfo.AppName != "" {
  328. klog.Debugf("[%s] ->>>>> %s -> %s payload:[%s]", c.AppInfo.AppName, r.Protocol.String(), conn.ActualDest, query)
  329. }
  330. //apmTrace, ok := c.getTrace(r.TraceId)
  331. apmTrace, err := c.getOrInitTrace(r.TraceId)
  332. //fmt.Println("mysql r.TraceId:", r.TraceId)
  333. //fmt.Println("ok:", ok)
  334. //fmt.Println("traceMap:", len(c.traceMap))
  335. if err == nil {
  336. //apmTrace.MysqlTraceQuery(query, r.Status.Error(), r.Duration, conn.ActualDest)
  337. //apmTrace.PostGreSqlTraceQueryEvent(query, r, conn.ActualDest)
  338. apmTrace.SQLTraceQueryEvent(r.Protocol, semconv.DBSystemPostgreSQL, query, r, conn.ActualDest)
  339. c.SendEvent(apmTrace, r.TraceId)
  340. }
  341. }
  342. /**
  343. * Mysql
  344. */
  345. case l7.ProtocolMysql:
  346. //fmt.Println("mysql r.TraceId:", r.TraceId)
  347. if r.Method != l7.MethodStatementClose {
  348. stats.observe(r.Status.String(), "", r.Duration)
  349. }
  350. if c.l7Attach && c.valuableTrace(r.TraceId) {
  351. if conn.mysqlParser == nil {
  352. conn.mysqlParser = l7.NewMysqlParser()
  353. }
  354. query := conn.mysqlParser.Parse(r.Payload, r.StatementId)
  355. //trace.MysqlQuery(query, r.Status.Error(), r.Duration)
  356. if c.AppInfo.AppName != "" {
  357. klog.Debugf("[%s] ->>>>> %s -> %s payload:[%s]", c.AppInfo.AppName, r.Protocol.String(), conn.ActualDest, query)
  358. }
  359. //apmTrace, ok := c.getTrace(r.TraceId)
  360. apmTrace, err := c.getOrInitTrace(r.TraceId)
  361. //fmt.Println("mysql r.TraceId:", r.TraceId)
  362. //fmt.Println("ok:", ok)
  363. //fmt.Println("traceMap:", len(c.traceMap))
  364. if err == nil {
  365. dbSystem := semconv.DBSystemMySQL
  366. // 根据端口白名单确定协议类型
  367. l7Type := flags.GetProtocolByPort(uint16(conn.ActualDest.Port()))
  368. if l7Type == l7.ProtocolMariaDB {
  369. dbSystem = semconv.DBSystemMariaDB
  370. } else if l7Type == l7.ProtocolTiDB {
  371. dbSystem = semconv.DBSystemTiDB
  372. }
  373. //apmTrace.MysqlTraceQuery(query, r.Status.Error(), r.Duration, conn.ActualDest)
  374. //apmTrace.MysqlTraceQueryEvent(query, r, conn.ActualDest)
  375. apmTrace.SQLTraceQueryEvent(l7Type, dbSystem, query, r, conn.ActualDest)
  376. c.SendEvent(apmTrace, r.TraceId)
  377. }
  378. }
  379. /**
  380. * DM (达梦数据库)
  381. */
  382. case l7.ProtocolDM:
  383. //统计dm的query次数
  384. stats.observe(r.Status.String(), "", r.Duration)
  385. //是否发送数据
  386. if c.l7Attach && c.valuableTrace(r.TraceId) {
  387. if conn.dmParser == nil {
  388. conn.dmParser = l7.NewDmParser()
  389. }
  390. query := conn.dmParser.Parse(r.Payload, r.StatementId)
  391. if c.AppInfo.AppName != "" {
  392. klog.Debugf("[%s] ->>>>> %s -> %s payload:[%s]", c.AppInfo.AppName, r.Protocol.String(), conn.ActualDest, query)
  393. }
  394. apmTrace, err := c.getOrInitTrace(r.TraceId)
  395. if err == nil {
  396. //apmTrace.DmTraceQueryEvent(query, r, conn.ActualDest)
  397. apmTrace.SQLTraceQueryEvent(r.Protocol, semconv.DBSystemDaMengDB, query, r, conn.ActualDest)
  398. c.SendEvent(apmTrace, r.TraceId)
  399. }
  400. }
  401. /**
  402. * Memcached
  403. */
  404. case l7.ProtocolMemcached:
  405. stats.observe(r.Status.String(), "", r.Duration)
  406. if c.l7Attach && c.valuableTrace(r.TraceId) {
  407. cmd, items := l7.ParseMemcached(r.Payload)
  408. if c.AppInfo.AppName != "" {
  409. klog.Debugf("[%s] ->>>>> %s -> %s payload:[%s]", c.AppInfo.AppName, r.Protocol.String(), conn.ActualDest, cmd+" "+strings.Join(items, " "))
  410. }
  411. apmTrace, err := c.getOrInitTrace(r.TraceId)
  412. if err == nil {
  413. statement := cmd
  414. if len(items) == 1 {
  415. statement += " " + items[0]
  416. } else if len(items) > 1 {
  417. joined := fmt.Sprintf("[%s]", strings.Join(items, " "))
  418. statement += " " + joined
  419. }
  420. apmTrace.NoSQLTraceQueryEvent(r.Protocol, semconv.DBSystemMemcached, cmd, statement, r, conn.Src, conn.ActualDest)
  421. c.SendEvent(apmTrace, r.TraceId)
  422. }
  423. }
  424. /**
  425. * Redis
  426. */
  427. case l7.ProtocolRedis:
  428. stats.observe(r.Status.String(), "", r.Duration)
  429. if c.l7Attach && c.valuableTrace(r.TraceId) {
  430. cmd, args := l7.ParseRedis(r.Payload)
  431. //fmt.Println("cmd", cmd)
  432. //fmt.Println("args", args)
  433. //apmTrace, ok := c.getTrace(r.TraceId)
  434. if c.AppInfo.AppName != "" {
  435. klog.Debugf("[%s] ->>>>> %s -> %s payload:[%s]", c.AppInfo.AppName, r.Protocol.String(), conn.ActualDest, cmd)
  436. }
  437. apmTrace, err := c.getOrInitTrace(r.TraceId)
  438. if err == nil {
  439. statement := cmd
  440. if args != "" {
  441. statement += " " + args
  442. }
  443. apmTrace.NoSQLTraceQueryEvent(r.Protocol, semconv.DBSystemRedis, cmd, statement, r, conn.Src, conn.ActualDest)
  444. c.SendEvent(apmTrace, r.TraceId)
  445. }
  446. }
  447. //trace.RedisQuery(cmd, args, r.Status.Error(), r.Duration)
  448. /**
  449. * gRPC
  450. */
  451. case l7.ProtocolGrpc:
  452. klog.Debugln("enter the l7.ProtocolGrpc")
  453. stats.observe(r.Status.String(), "", r.Duration)
  454. if c.l7Attach && c.valuableTrace(r.TraceId) {
  455. apmTrace, err := c.getOrInitTrace(r.TraceId)
  456. if err == nil {
  457. apmTrace.GrpcClientTraceQueryEvent(r)
  458. c.SendEvent(apmTrace, r.TraceId)
  459. }
  460. }
  461. /**
  462. * MongoDB
  463. */
  464. case l7.ProtocolMongo:
  465. stats.observe(r.Status.String(), "", r.Duration)
  466. if c.l7Attach && c.valuableTrace(r.TraceId) {
  467. query := l7.ParseMongo(r.Payload)
  468. if c.AppInfo.AppName != "" {
  469. klog.Debugf("[%s] ->>>>> MongoDB -> %s SQL:[%s]", c.AppInfo.AppName, conn.ActualDest, query)
  470. }
  471. apmTrace, err := c.getOrInitTrace(r.TraceId)
  472. if err == nil {
  473. // MongoDB query 格式通常是 JSON,如 {"insert":"users"} 或 {"find":"users","filter":{...}}
  474. apmTrace.NoSQLTraceQueryEvent(r.Protocol, semconv.DBSystemMongoDB, "", query, r, conn.Src, conn.ActualDest)
  475. c.SendEvent(apmTrace, r.TraceId)
  476. }
  477. }
  478. /**
  479. * Cassandra
  480. */
  481. case l7.ProtocolCassandra:
  482. stats.observe(r.Status.String(), "", r.Duration)
  483. if c.l7Attach && c.valuableTrace(r.TraceId) {
  484. if conn.cassandraParser == nil {
  485. conn.cassandraParser = l7.NewCassandraParser()
  486. }
  487. var query string
  488. query = string(r.Payload)
  489. //query := conn.cassandraParser.Parse(r.Payload)
  490. if c.AppInfo.AppName != "" {
  491. klog.Debugf("[%s] ->>>>> %s -> %s payload:[%s]", c.AppInfo.AppName, r.Protocol.String(), conn.ActualDest, query)
  492. }
  493. apmTrace, err := c.getOrInitTrace(r.TraceId)
  494. if err == nil {
  495. apmTrace.NoSQLTraceQueryEvent(r.Protocol, semconv.DBSystemCassandra, "", query, r, conn.Src, conn.ActualDest)
  496. c.SendEvent(apmTrace, r.TraceId)
  497. }
  498. }
  499. /**
  500. * BuntDB
  501. */
  502. case l7.ProtocolBuntDB:
  503. stats.observe(r.Status.String(), "", r.Duration)
  504. if c.l7Attach && c.valuableTrace(r.TraceId) {
  505. query := string(r.Payload)
  506. if c.AppInfo.AppName != "" {
  507. klog.Debugf("[%s] ->>>>> %s -> %s payload:[%s]", c.AppInfo.AppName, r.Protocol.String(), conn.ActualDest, query)
  508. }
  509. apmTrace, err := c.getOrInitTrace(r.TraceId)
  510. if err == nil {
  511. // BuntDB 是键值数据库,使用 NoSQLTraceQueryEvent,operation 从 payload 中提取(如 "Set: key_name")
  512. operation := ""
  513. statement := query
  514. dbSystem := attribute.String("db.system", "buntdb")
  515. defaultLocalAddr := netaddr.IPPortFrom(netaddr.MustParseIP("127.0.0.1"), 0)
  516. apmTrace.NoSQLTraceQueryEvent(r.Protocol, dbSystem, operation, statement, r, defaultLocalAddr, defaultLocalAddr)
  517. c.SendEvent(apmTrace, r.TraceId)
  518. }
  519. }
  520. /**
  521. * LevelDB
  522. */
  523. case l7.ProtocolLevelDB:
  524. stats.observe(r.Status.String(), "", r.Duration)
  525. if c.l7Attach && c.valuableTrace(r.TraceId) {
  526. query := string(r.Payload)
  527. if c.AppInfo.AppName != "" {
  528. klog.Debugf("[%s] ->>>>> %s -> %s payload:[%s]", c.AppInfo.AppName, r.Protocol.String(), conn.ActualDest, query)
  529. }
  530. apmTrace, err := c.getOrInitTrace(r.TraceId)
  531. if err == nil {
  532. operation := ""
  533. statement := query
  534. dbSystem := attribute.String("db.system", "leveldb")
  535. defaultLocalAddr := netaddr.IPPortFrom(netaddr.MustParseIP("127.0.0.1"), 0)
  536. apmTrace.NoSQLTraceQueryEvent(r.Protocol, dbSystem, operation, statement, r, defaultLocalAddr, defaultLocalAddr)
  537. c.SendEvent(apmTrace, r.TraceId)
  538. }
  539. }
  540. /**
  541. * Kafka
  542. */
  543. case l7.ProtocolKafka:
  544. stats.observe(r.Status.String(), "", r.Duration)
  545. if c.l7Attach && c.valuableTrace(r.TraceId) {
  546. if c.AppInfo.AppName != "" {
  547. klog.Debugf("[%s] ->>>>> %s -> %s payload:[%s]", c.AppInfo.AppName, r.Protocol.String(), conn.ActualDest, r.DestAddrString)
  548. }
  549. apmTrace, err := c.getOrInitTrace(r.TraceId)
  550. if err == nil {
  551. apmTrace.MQTraceQueryEvent(r.Protocol, semconv.MessagingKafkaClientID("kafka"), "", "", r, conn.Src, conn.ActualDest)
  552. c.SendEvent(apmTrace, r.TraceId)
  553. }
  554. }
  555. /**
  556. * RabbitMQ / NATS
  557. */
  558. case l7.ProtocolRabbitmq, l7.ProtocolNats:
  559. stats.observe(r.Status.String(), r.Method.String(), 0)
  560. if c.l7Attach && c.valuableTrace(r.TraceId) {
  561. }
  562. /**
  563. * Dubbo2
  564. */
  565. case l7.ProtocolDubbo2:
  566. stats.observe(r.Status.String(), "", r.Duration)
  567. if c.l7Attach && c.valuableTrace(r.TraceId) {
  568. }
  569. }
  570. return nil
  571. }
  572. func (c *Container) buildIDs(pid uint32) bool {
  573. c.lock.Lock()
  574. defer c.lock.Unlock()
  575. p := c.processes[pid]
  576. if p != nil {
  577. p.cmdline = string(proc.GetRealCmdline(pid))
  578. }
  579. var sns []string
  580. var sport uint16
  581. for address, val := range c.getListens() {
  582. if val == 1 {
  583. ip := address.IP()
  584. if ip.Is4() && !ip.IsLoopback() {
  585. // 获取端口号
  586. sport = address.Port()
  587. sns = append(sns, fmt.Sprintf("%s:%d", ip, sport))
  588. ////c.instanceID.IntVal, c.instanceID.HashtVal, _ =
  589. //c.AppInfo.Sn = ip.String()
  590. //c.AppInfo.Sport = int(port)
  591. //strInstanceID := utils.BuildInt64ID(fmt.Sprintf("%s:%d", ip.String(), port))
  592. //fmt.Println(port)
  593. ////os.Exit(1)
  594. //c.AppInfo.InstanceIdHash.IntVal, _ = strInstanceID.ToInt64()
  595. //c.AppInfo.InstanceIdHash.HashtVal = strInstanceID.ToHashByte()
  596. ////c.AppInfo.InstanceId = c.instanceID.IntVal
  597. //strAgentID := utils.BuildInt64ID(fmt.Sprintf("%s:%s", strInstanceID, string(proc.GetExe(pid))))
  598. //c.AppInfo.AgentId, _ = strAgentID.ToInt64()
  599. //c.AppInfo.CodeType = c.GetCodeTypeFromCache(pid)
  600. //return true
  601. }
  602. }
  603. }
  604. if len(sns) > 0 {
  605. //c.instanceID.IntVal, c.instanceID.HashtVal, _ =
  606. snsStr := strings.Join(sns, ",")
  607. c.AppInfo.Sn = snsStr
  608. c.AppInfo.Sport = int(sport)
  609. strInstanceID := utils.BuildInt64ID(fmt.Sprintf("%s:%d", c.AppInfo.Sn, sport))
  610. c.AppInfo.InstanceIdHash.IntVal, _ = strInstanceID.ToInt64()
  611. c.AppInfo.InstanceIdHash.HashtVal = strInstanceID.ToHashByte()
  612. // strAgentID := utils.BuildInt64ID(fmt.Sprintf("%s:%s", utils.GetHostIP(), string(proc.GetExe(pid))))
  613. // c.AppInfo.AgentId, _ = strAgentID.ToInt64()
  614. // c.AppInfo.CodeType = c.GetCodeTypeFromCache(pid)
  615. return true
  616. }
  617. return false
  618. }
  619. func (c *Container) ReBuildIds(pid uint32) {
  620. c.lock.Lock()
  621. defer c.lock.Unlock()
  622. var sns []string
  623. var sport uint16
  624. for address, val := range c.getListens() {
  625. if val == 1 {
  626. ip := address.IP()
  627. if ip.Is4() && !ip.IsLoopback() {
  628. // 获取端口号
  629. sport = address.Port()
  630. sns = append(sns, fmt.Sprintf("%s:%d", ip, sport))
  631. }
  632. }
  633. }
  634. if len(sns) > 0 {
  635. snsStr := strings.Join(sns, ",")
  636. c.AppInfo.Sn = snsStr
  637. c.AppInfo.Sport = int(sport)
  638. strInstanceID := utils.BuildInt64ID(fmt.Sprintf("%s:%d", c.AppInfo.Sn, sport))
  639. c.AppInfo.InstanceIdHash.IntVal, _ = strInstanceID.ToInt64()
  640. c.AppInfo.InstanceIdHash.HashtVal = strInstanceID.ToHashByte()
  641. // strAgentID := utils.BuildInt64ID(fmt.Sprintf("%s:%s", utils.GetHostIP(), string(proc.GetExe(pid))))
  642. strAgentID := utils.BuildInt64ID(fmt.Sprintf("%s:%d", utils.GetHostIP(), c.AppInfo.AppIdHash.IntVal))
  643. c.AppInfo.AgentId, _ = strAgentID.ToInt64()
  644. c.AppInfo.CodeType = c.GetCodeTypeFromCache(pid)
  645. }
  646. }
  647. func (c *Container) StackProcess(event ebpftracer.StackEvent, tracer *ebpftracer.Tracer) {
  648. c.lock.Lock()
  649. defer c.lock.Unlock()
  650. // get the associated uprobe
  651. uprobe, err := c.GetUprobe(event, tracer)
  652. if err != nil {
  653. //fmt.Println("GetUprobeGetUprobe errer: %v", err)
  654. klog.Errorf("failed to get uprobe for event %+v: %+v", event, err)
  655. return
  656. }
  657. if event.TraceId <= 0 {
  658. //fmt.Println("StackProcess TraceId id 0")
  659. klog.Errorf("failed to get uprobe(traceId is <= 0) for event %+v", event)
  660. return
  661. }
  662. // 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)
  663. stackFun := ebpftracer.StackFunEvent{}
  664. stackFun.Uprobe = &uprobe
  665. stackFun.StackEvent = event
  666. apmTrace, ok := c.getTrace(event.TraceId)
  667. if ok {
  668. apmTrace.FunAdd(stackFun)
  669. }
  670. }
  671. func byteExtractString(nameString [100]byte) string {
  672. n := bytes.IndexFunc(nameString[:], func(r rune) bool {
  673. return r == 0 || r < 32 || r > 126 // 截取到第一个零值或非打印字符
  674. })
  675. if n == -1 {
  676. n = len(nameString) // 没找到零值或非打印字符,使用数组长度
  677. }
  678. return string(nameString[:n])
  679. }
  680. func (c *Container) StackProcess2(event ebpftracer.StackEvent, tracer *ebpftracer.Tracer) {
  681. c.lock.Lock()
  682. defer c.lock.Unlock()
  683. // get the associated uprobe
  684. switch event.Location {
  685. case 0: // ret
  686. Funcname := ""
  687. if event.Type != uint64(CodeTypeJava) {
  688. uprobe, err := c.GetUprobe(event, tracer)
  689. if err != nil {
  690. //fmt.Println("GetUprobeGetUprobe errer: %v", err)
  691. klog.Errorf("failed to get uprobe for event %+v: %+v", event, err)
  692. return
  693. }
  694. Funcname = uprobe.Funcname
  695. } else {
  696. ClassName := byteExtractString(event.ClassName)
  697. MethedName := byteExtractString(event.MethedName)
  698. Funcname = ClassName + "." + MethedName
  699. }
  700. if event.TraceId <= 0 {
  701. //fmt.Println("StackProcess TraceId id 0")
  702. klog.Errorf("failed to get uprobe(traceId is <= 0) for event %+v", event)
  703. return
  704. }
  705. //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)
  706. apmTrace, err := c.getOrInitTrace(event.TraceId)
  707. if err == nil {
  708. //fmt.Println("append FuncTraceQuery fun:", event.TraceId, uprobe.Funcname, event.Pid)
  709. duration := event.TimeNsEnd - event.TimeNsStart
  710. apmTrace.FuncTraceQuery(Funcname, time.Duration(duration), event.TimeNsStart, event.TimeNsEnd)
  711. c.SendEvent(apmTrace, event.TraceId)
  712. }
  713. }
  714. }
  715. // ResolveAddress returns the symbol(s) and offset of the given address.
  716. func (c *Container) ResolveAddress(addr uint64, symbols []elf.Symbol) (syms []elf.Symbol, offset uint, err error) {
  717. if addr == 0 {
  718. // err = errors.Wrapf(SymbolNotFoundError, "0")
  719. return
  720. }
  721. // symbols, _, err := e.Symbols()
  722. if err != nil {
  723. return
  724. }
  725. idx := sort.Search(len(symbols), func(i int) bool { return symbols[i].Value > addr })
  726. if idx == 0 {
  727. // err = errors.Wrap(SymbolNotFoundError, fmt.Sprintf("%x", addr))
  728. return
  729. }
  730. // why diff symbol may contains the same addr?
  731. sym := symbols[idx-1]
  732. for i := idx - 1; i >= 0 && symbols[i].Value == sym.Value; i-- {
  733. syms = append(syms, symbols[i])
  734. }
  735. for i := idx; i < len(symbols) && symbols[i].Value == sym.Value; i++ {
  736. syms = append(syms, symbols[i])
  737. }
  738. return syms, uint(addr - sym.Value), nil
  739. }
  740. type MemoryMap struct {
  741. Start, End uint64
  742. }
  743. // ReadFirstLineOfMapsFile reads the first line of /proc/<pid>/maps file and return the memory map as a MemoryMap struct
  744. func ReadFirstLineOfMapsFile(pid string) (*MemoryMap, error) {
  745. file, err := os.Open(fmt.Sprintf("/proc/%s/maps", pid))
  746. if err != nil {
  747. return nil, err
  748. }
  749. defer file.Close()
  750. scanner := bufio.NewScanner(file)
  751. if scanner.Scan() {
  752. fields := strings.Fields(scanner.Text())
  753. addresses := strings.Split(fields[0], "-")
  754. if len(addresses) != 2 {
  755. return nil, errors.New("unexpected format in /proc/<pid>/maps")
  756. }
  757. start, err := strconv.ParseUint(addresses[0], 16, 64)
  758. if err != nil {
  759. return nil, err
  760. }
  761. end, err := strconv.ParseUint(addresses[1], 16, 64)
  762. if err != nil {
  763. return nil, err
  764. }
  765. return &MemoryMap{
  766. Start: start,
  767. End: end,
  768. }, nil
  769. }
  770. if err := scanner.Err(); err != nil {
  771. return nil, err
  772. }
  773. return nil, errors.New("empty /proc/<pid>/maps")
  774. }
  775. func (c *Container) GetUprobe(event ebpftracer.StackEvent, tracer *ebpftracer.Tracer) (uprobe tracer.Uprobe, err error) {
  776. //fmt.Println("GetUprobe entory:")
  777. memoryMap, _ := ReadFirstLineOfMapsFile(strconv.Itoa(int(event.Pid)))
  778. Address := event.Ip - memoryMap.Start
  779. // fmt.Printf("memoryMap.Start: %x, event.Ip: %x, Address: %x\n", memoryMap.Start, event.Ip, Address)
  780. for _, fun := range c.UprobesMap {
  781. funAddress := fun.Address + fun.AbsOffset
  782. // fmt.Printf("GetUprobeGetUprobeGetUprobe:fun.Address %x, fun.AbsOffset: %x\n", fun.Address, fun.AbsOffset)
  783. if funAddress == Address {
  784. // fmt.Printf("---GetUprobeGetUprobeGetUprobe: %x, event.Ip: %x ---- %s--%x\n", memoryMap.Start, event.Ip, fun.Funcname, fun.Address)
  785. return fun, nil
  786. }
  787. }
  788. syms, _, err := c.ResolveAddress(event.Ip, tracer.Symbols)
  789. if err != nil {
  790. return
  791. }
  792. for _, sym := range syms {
  793. //fmt.Println("GetUprobeGetUprobeGetUprobe: %s+%d", sym.Name, offset)
  794. uprobe, ok := tracer.UprobesMap[fmt.Sprintf("%s-%s", sym.Name, sym.Value)]
  795. if ok {
  796. return uprobe, nil
  797. }
  798. }
  799. err = errors.New("uprobe not found")
  800. return
  801. }
  802. func (c *Container) GetAppInfo() AppInfo {
  803. return c.AppInfo
  804. }
  805. // 可注入前置
  806. func (c *Container) checkEventReady() bool {
  807. c.lock.Lock()
  808. defer c.lock.Unlock()
  809. return c.l7EventReady
  810. }
  811. func (c *Container) eventReady() {
  812. c.lock.Lock()
  813. defer c.lock.Unlock()
  814. c.l7EventReady = true
  815. }
  816. // uprobe前置
  817. func (c *Container) Isl7AttachSuccess() bool {
  818. c.lock.Lock()
  819. defer c.lock.Unlock()
  820. return c.l7Attach
  821. }
  822. func (c *Container) l7AttachSuccess() {
  823. c.lock.Lock()
  824. defer c.lock.Unlock()
  825. c.l7Attach = true
  826. }
  827. func (c *Container) IsLicenseFuse() bool {
  828. c.lock.Lock()
  829. defer c.lock.Unlock()
  830. return c.licenseFuse
  831. }
  832. func (c *Container) LicenseFuse() {
  833. c.lock.Lock()
  834. defer c.lock.Unlock()
  835. c.licenseFuse = true
  836. }
  837. func (c *Container) ClearLicenseFuse() {
  838. c.lock.Lock()
  839. defer c.lock.Unlock()
  840. c.licenseFuse = false
  841. }
  842. func (c *Container) ctrlStack(r *Registry, pid uint32) {
  843. resp, err := c.GetCodeSetting(r)
  844. if err != nil {
  845. klog.WithField("pid", pid).WithError(err).Error("[ctrlStack] GetCodeSetting failed.")
  846. return
  847. }
  848. if resp.BlackWhiteSettings.CollectStack == OPEN_STACK {
  849. // 有黑白名单规则 &&
  850. // 之前有注入 先卸载再注入
  851. // 之前没注入 直接注入
  852. // 没有有黑白名单 直接卸载
  853. if c.hasStackRule(resp) {
  854. if c.stackRuleUpdate(resp) {
  855. // 重新注入
  856. err = c.DetachStack(pid, APP_UNINSTALL)
  857. if err != nil {
  858. klog.WithError(err).Errorf("[ctrlStack][end] Failed detach stack trace!")
  859. }
  860. }
  861. klog.WithField("pid", pid).Infoln("[ctrlStack] Attach app stack.")
  862. c.saveWhiteStackSettingInfo(resp)
  863. err = c.AttachStack(r.tracer, pid)
  864. if err != nil {
  865. c.AppInfo.SetAppStackError()
  866. klog.WithField("pid", pid).WithError(err).Errorf("[ctrlStack][end] Failed attach stack trace!")
  867. }
  868. } else {
  869. if c.noOrigRule() {
  870. return
  871. }
  872. c.saveWhiteStackSettingInfo(resp)
  873. // 关闭堆栈
  874. err = c.DetachStack(pid, APP_UNINSTALL)
  875. if err != nil {
  876. klog.WithError(err).Errorf("[ctrlStack][end] Failed detach stack trace!")
  877. }
  878. }
  879. } else {
  880. if c.noOrigRule() {
  881. return
  882. }
  883. c.saveWhiteStackSettingInfo(resp)
  884. // 关闭堆栈
  885. err = c.DetachStack(pid, APP_UNINSTALL)
  886. if err != nil {
  887. klog.WithError(err).Errorf("[ctrlStack][end] Failed detach stack trace!")
  888. }
  889. }
  890. }
  891. func (c *Container) verifyAttachConditions(r *Registry, pid uint32) (bool, int) {
  892. p := c.processes[pid]
  893. if p != nil && c.checkEventReady() {
  894. codeType := c.GetCodeTypeFromCache(pid)
  895. if codeType.IsUnknownCode() {
  896. klog.WithField("pid", pid).Debug("[verify] unknown language.")
  897. return false, 0
  898. }
  899. cmdline := p.GetCmdline()
  900. if len(cmdline) == 0 {
  901. return false, 0
  902. }
  903. //whiteListByCode := r.getWhiteListByCodeType(codeType)
  904. whiteListByCode := r.getWhiteListAll()
  905. //klog.WithField("pid", pid).WithField("codeType", codeType.String()).
  906. // Infof("[verify] white list %v", utils.ToString(whiteListByCode))
  907. // 当前语言的白名单规则
  908. for _, setting := range whiteListByCode {
  909. ruleVal := setting.Filters
  910. if ruleVal == "" {
  911. continue
  912. }
  913. // 判断规则
  914. if strings.Contains(cmdline, ruleVal) {
  915. //if !codeType.IsJvmCode() {
  916. // klog.WithField("pid", pid).Warning("[verify] This agent version only supports JVM applications.")
  917. // return false, 0
  918. //}
  919. c.WhiteSettingInfo.AppName = setting.AppName
  920. c.WhiteSettingInfo.Filters = setting.Filters
  921. klog.WithField("pid", pid).
  922. WithField("codeType", codeType.String()).
  923. WithField("ruleVal", ruleVal).
  924. WithField("cmdline", cmdline).
  925. //WithField("stack", setting.OpenStack).
  926. WithField("white list", utils.ToString(whiteListByCode)).
  927. Infoln("[verify] check successful.")
  928. return true, 0
  929. }
  930. }
  931. }
  932. return false, 0
  933. }
  934. // 1.卸载入口
  935. func (c *Container) Detach(tracer *ebpftracer.Tracer, pid uint32, detachType APP_TYPE) {
  936. c.lock.Lock()
  937. defer c.lock.Unlock()
  938. if p := c.processes[pid]; p != nil {
  939. err := c.DetachUprobes(tracer, pid, detachType)
  940. if err != nil {
  941. klog.WithError(err).Errorln("DetachUprobes Error.")
  942. }
  943. err = c.DetachStack(pid, detachType)
  944. if err != nil {
  945. klog.WithError(err).Errorln("DetachStack Error.")
  946. }
  947. // 关闭7层监控
  948. c.l7Attach = false
  949. // 变更应用状态
  950. if err != nil {
  951. detachType = detachType.Error()
  952. }
  953. c.AppInfo.SetAppStatus(detachType)
  954. }
  955. }
  956. // 1.1卸载uprobe
  957. func (c *Container) DetachUprobes(tracer *ebpftracer.Tracer, pid uint32, detachType APP_TYPE) error {
  958. // close uprobe
  959. if p := c.processes[pid]; p != nil {
  960. for _, u := range p.uprobes {
  961. err := u.Close()
  962. if err != nil {
  963. return err
  964. }
  965. }
  966. p.uprobes = []link.Link{}
  967. switch detachType {
  968. case APP_UNINSTALL, APP_FUSE, APP_LICENSE_FUSE:
  969. codeType := c.GetCodeTypeFromCache(pid)
  970. switch codeType {
  971. case CodeTypeJava:
  972. p.jvmAttachOnce = false
  973. case CodeTypeGo:
  974. p.goTlsUprobesChecked = false
  975. p.openSslUprobesChecked = false
  976. default:
  977. }
  978. case APP_UPROBE_ERROR:
  979. klog.Infof("[DetachUprobes] ERROR_DETACH for pid %d", pid)
  980. default:
  981. }
  982. //delete the proc info form proc_info_map(for kernel) when the uprobe detached
  983. if err := tracer.DelKProcInfo(pid); err != nil {
  984. return fmt.Errorf("[DetachUprobes] failed to delete KProcInfo for pid %d, detach type is:%s", pid, detachType)
  985. } else {
  986. klog.Infof("[DetachUprobes] delete KProcInfo success for pid %d,detachType:%s", pid, detachType.String())
  987. c.AppInfo.EBPFProcInfo = nil
  988. }
  989. } else {
  990. return fmt.Errorf("[DetachUprobes] cannot find uprobe for pid %d", pid)
  991. }
  992. return nil
  993. }
  994. // 1.2卸载堆栈
  995. func (c *Container) DetachStack(pid uint32, detachType APP_TYPE) error {
  996. if p := c.processes[pid]; p != nil {
  997. var err error
  998. codeType := c.GetCodeTypeFromCache(pid)
  999. switch codeType {
  1000. // 1.2.1 卸载 jvm堆栈
  1001. case CodeTypeJava:
  1002. err = c.detachJvmStack(pid)
  1003. default:
  1004. err = p.closeStackUprobes()
  1005. }
  1006. if err != nil {
  1007. klog.WithError(err).Errorln("[detachStack] failed to detach stack")
  1008. return err
  1009. }
  1010. p.stackAttachOnce = false
  1011. } else {
  1012. return fmt.Errorf("[DetachStack] cannot find uprobe for pid %d", pid)
  1013. }
  1014. return nil
  1015. }
  1016. // 1.2.1 卸载 jvm堆栈
  1017. func (c *Container) detachJvmStack(pid uint32) error {
  1018. if p := c.processes[pid]; p != nil {
  1019. //if p.stackStatus.IsStackUprobesSuccess() || len(p.stackUprobes) > 0 {
  1020. //}
  1021. // 卸载 JavaAgent
  1022. var err error
  1023. if p.stackStatus.IsJattachSuccess() {
  1024. // 卸载堆栈probes
  1025. err = p.closeStackUprobes()
  1026. if err != nil {
  1027. klog.WithError(err).Errorf("[detachJvmStack] closeStackUprobes")
  1028. }
  1029. err = p.uninstallJavaAgent()
  1030. if err != nil {
  1031. klog.WithError(err).Errorf("[detachJvmStack] uninstallJavaAgent")
  1032. }
  1033. }
  1034. return err
  1035. }
  1036. return nil
  1037. }
  1038. func (c *Container) getRootfs() string {
  1039. if c.metadata != nil && c.metadata.rootfs != "" {
  1040. return path.Join(*flags.HostDirPathPrefix, c.metadata.rootfs)
  1041. }
  1042. return ""
  1043. }
  1044. func (c *Container) BuildActiveApps(runtimeApps map[uint32]AppStatusInfo, pid uint32) {
  1045. if c == nil {
  1046. //klog.WithField("pid", pid).Warningln("[BuildActiveApps] container_apm is nil.")
  1047. return
  1048. }
  1049. if c.AppInfo.AppName == "" {
  1050. return
  1051. }
  1052. klog.WithField("pid", pid).WithField("appname", c.AppInfo.AppName).Infof("[BuildActiveApps] container %s is running.", c.AppInfo.AppName)
  1053. detail := AppStatusInfo{
  1054. Pid: pid,
  1055. ProcName: c.containerName,
  1056. AppName: c.AppInfo.AppName,
  1057. Language: c.AppInfo.CodeType.String(),
  1058. LanguageVersion: c.AppInfo.Version,
  1059. AppID: c.AppInfo.AppIdHash.IntVal,
  1060. AgentID: c.AppInfo.AgentId,
  1061. InstanceID: c.AppInfo.InstanceIdHash.IntVal,
  1062. Sn: c.AppInfo.Sn,
  1063. Sport: c.AppInfo.Sport,
  1064. RegisterAt: time.Unix(c.AppInfo.RegisterAt, 0).Format("060102 15:04:05"),
  1065. PreStatus: c.AppInfo.PreStatus,
  1066. Status: c.AppInfo.Status,
  1067. Rule: c.WhiteSettingInfo.Filters,
  1068. Container: string(c.id),
  1069. }
  1070. detail.Rule = fmt.Sprintf("%s|%d", c.WhiteSettingInfo.Filters, c.WhiteSettingInfo.WhiteStackSettingInfo.OpenStack)
  1071. if c.AppInfo.UpdateAt != 0 {
  1072. detail.UpdateAt = time.Unix(c.AppInfo.UpdateAt, 0).Format("060102 15:04:05")
  1073. }
  1074. p := c.processes[pid]
  1075. if p != nil {
  1076. detail.StackStatus = p.stackStatus.String()
  1077. v := 0
  1078. if !p.versionFailed {
  1079. v = 1
  1080. }
  1081. detail.StackStatus += fmt.Sprintf("V=%d", v)
  1082. }
  1083. runtimeApps[pid] = detail
  1084. }
  1085. func (c *Container) AgentCtrl(r *Registry, pid uint32) {
  1086. if c == nil {
  1087. //klog.WithField("pid", pid).Warningln("[AgentCtrl] cannot find container.")
  1088. return
  1089. }
  1090. var err error
  1091. verifyAttachConditions, _ := c.verifyAttachConditions(r, pid)
  1092. // License fuse UNINSTALL(仅在开启 License 校验时)
  1093. if *flags.EnableLicenseCheck && c.IsLicenseFuse() {
  1094. if c.Isl7AttachSuccess() {
  1095. c.Detach(r.tracer, pid, APP_LICENSE_FUSE)
  1096. }
  1097. klog.WithField("pid", pid).Infoln("[AgentCtrl] License fusing.")
  1098. return
  1099. }
  1100. // fusing UNINSTALL
  1101. if r.isFusing && c.Isl7AttachSuccess() {
  1102. c.Detach(r.tracer, pid, APP_FUSE)
  1103. klog.WithField("pid", pid).Infoln("[AgentCtrl] fusing")
  1104. return
  1105. }
  1106. // verify UNINSTALL
  1107. if !verifyAttachConditions && c.Isl7AttachSuccess() {
  1108. c.Detach(r.tracer, pid, APP_UNINSTALL)
  1109. klog.WithField("pid", pid).Infoln("[AgentCtrl] rule uninstall.")
  1110. return
  1111. }
  1112. if verifyAttachConditions {
  1113. err = c.RegisterAppInfo(r, pid)
  1114. if err != nil {
  1115. klog.WithError(err).Errorf("[AgentCtrl] Failed registerAppInfo.")
  1116. return
  1117. }
  1118. klog.WithField("pid", pid).Infoln("[AgentCtrl] Attach uprobes.")
  1119. err = c.AttachUprobes(r.tracer, pid, "Agentctrl")
  1120. if err != nil {
  1121. klog.WithField("pid", pid).WithError(err).Errorf("[AgentCtrl] Failed attach uprobes error!")
  1122. return
  1123. } else {
  1124. klog.WithField("pid", pid).Infoln("[AgentCtrl] Attach uprobes success!")
  1125. }
  1126. // 堆栈控制
  1127. c.ctrlStack(r, pid)
  1128. }
  1129. }