container_apm.go 27 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892
  1. package containers
  2. import (
  3. "bufio"
  4. "bytes"
  5. "debug/elf"
  6. "fmt"
  7. "os"
  8. "path"
  9. "sort"
  10. "strconv"
  11. "strings"
  12. "time"
  13. "github.com/cilium/ebpf/link"
  14. "github.com/coroot/coroot-node-agent/flags"
  15. "github.com/coroot/coroot-node-agent/ebpftracer"
  16. "github.com/coroot/coroot-node-agent/ebpftracer/l7"
  17. "github.com/coroot/coroot-node-agent/ebpftracer/tracer"
  18. "github.com/coroot/coroot-node-agent/proc"
  19. "github.com/coroot/coroot-node-agent/tracing"
  20. "github.com/coroot/coroot-node-agent/utils"
  21. . "github.com/coroot/coroot-node-agent/utils/modelse"
  22. "github.com/pkg/errors"
  23. klog "github.com/sirupsen/logrus"
  24. semconv "go.opentelemetry.io/otel/semconv/v1.18.0"
  25. "inet.af/netaddr"
  26. )
  27. const (
  28. TRACE_STATUS = 1
  29. )
  30. func (c *Container) getTrace(traceId uint64) (*tracing.Trace, bool) {
  31. trace, ok := c.traceMap[traceId]
  32. return trace, ok
  33. }
  34. func (c *Container) createTraceMap(traceId uint64, trace *tracing.Trace) {
  35. c.traceMap[traceId] = trace
  36. }
  37. // 查询或创建trace信息
  38. func (c *Container) getOrInitTrace(traceId uint64) (*tracing.Trace, error) {
  39. trace, ok := c.getTrace(traceId)
  40. if !ok {
  41. //new trace
  42. trace = tracing.NewTraceFromEvent(string(c.id))
  43. //create TraceMap
  44. c.createTraceMap(traceId, trace)
  45. //create ParentSpan
  46. trace.CreateRootSpan(traceId)
  47. }
  48. return trace, nil
  49. }
  50. func (c *Container) InitTrace(traceId uint64, r *l7.RequestData) error {
  51. method, path, hostIp, port := l7.ParseHttpHost(r.Payload)
  52. ip, err := netaddr.ParseIP(hostIp)
  53. if err != nil {
  54. //fmt.Println("host ip error")
  55. hostIp = "127.0.0.1"
  56. }
  57. addr := netaddr.IPPortFrom(ip, port)
  58. trace := tracing.NewTrace(string(c.id), addr)
  59. if trace == nil {
  60. return fmt.Errorf("OTEL_EXPORTER_OTLP_TRACES_ENDPOINT is null")
  61. }
  62. c.traceMap[traceId] = trace
  63. trace.TraceStart(method, path, r.Status, r.Duration)
  64. return nil
  65. }
  66. // 在任意阶段,r.TraceId 不等于0 则创建 traceMap && createParentSpan
  67. // 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
  68. func (c *Container) SendEvent(t *tracing.Trace, traceID uint64) {
  69. if t.AllEventReady(traceID) {
  70. t.SendEvent()
  71. klog.Debugf("SendEvent %d", traceID)
  72. //fmt.Println(t.GetSpan())
  73. //fmt.Println("===============")
  74. delete(c.traceMap, traceID)
  75. }
  76. }
  77. func (c *Container) valuableTrace(traceID uint64) bool {
  78. return traceID != 0
  79. }
  80. func (c *Container) onL7RequestApm(pid uint32, fd uint64, timestamp uint64, r *l7.RequestData) map[netaddr.IP]string {
  81. c.lock.Lock()
  82. defer c.lock.Unlock()
  83. if r.Protocol == l7.ProtocolDNS {
  84. ip2fqdn, _type, fqdn, ttl, ips := c.onDNSRequest(r)
  85. if c.l7Attach && c.valuableTrace(r.TraceId) {
  86. apmTrace, err := c.getOrInitTrace(r.TraceId)
  87. if err == nil {
  88. apmTrace.DNSTraceQueryEvent(r, _type, fqdn, ttl, ips)
  89. c.SendEvent(apmTrace, r.TraceId)
  90. }
  91. }
  92. return ip2fqdn
  93. }
  94. //if !c.valuableTrace(r.TraceId) {
  95. // return nil
  96. //}
  97. //klog.Infof("====ProtocolTrace+++++ start==== %d %d", pid, r.TraceId)
  98. // klog.Infof("====ProtocolTrace===== start==== %d %d", r.Protocol == l7.ProtocolTrace, c.l7Attach)
  99. if r.Protocol == l7.ProtocolTrace && c.l7Attach && c.valuableTrace(r.TraceId) {
  100. // klog.Infof("====ProtocolTrace---- start==== %d %d", pid, r.TraceId)
  101. if r.TraceStart == TRACE_STATUS {
  102. // klog.Infof("====ProtocolTrace start==== %d %d", pid, r.TraceId)
  103. trace, err := c.getOrInitTrace(r.TraceId)
  104. if c.AppInfo.AppName != "" {
  105. klog.Debugf("->>> [%s] -> payload:[%s]", c.AppInfo.AppName, r.Payload)
  106. }
  107. if err == nil {
  108. method, requestURI, sn, sport := l7.ParseHttpHost(r.Payload)
  109. ip, _ := netaddr.ParseIP(sn)
  110. //codeType := c.GetCodeTypeFromCache(pid)
  111. trace.TraceStartEvent(method, requestURI, sn, sport, r.Status, netaddr.IPPortFrom(ip, sport), pid, c.GetAppInfo())
  112. c.SendEvent(trace, r.TraceId)
  113. }
  114. return nil
  115. }
  116. if r.TraceEnd == TRACE_STATUS {
  117. klog.Debugf("====ProtocolTrace end==== %d %d", pid, r.TraceId)
  118. trace, err := c.getOrInitTrace(r.TraceId)
  119. if err == nil {
  120. trace.TraceEndEvent(r)
  121. c.SendEvent(trace, r.TraceId)
  122. }
  123. return nil
  124. }
  125. }
  126. if r.Protocol == l7.ProtocolHTTP {
  127. if c.l7Attach && c.valuableTrace(r.TraceId) {
  128. method, requestURI, sn, sport := l7.ParseHttpHost(r.Payload)
  129. apmTrace, err := c.getOrInitTrace(r.TraceId)
  130. //fmt.Println("ProtocolHTTP-----", r.TraceId, err)
  131. if err == nil {
  132. apmTrace.HttpTraceRequestEvent(method, requestURI, sn, sport, r)
  133. c.SendEvent(apmTrace, r.TraceId)
  134. }
  135. }
  136. //return nil
  137. }
  138. conn := c.connectionsByPidFd[PidFd{Pid: pid, Fd: fd}]
  139. //fmt.Println("l7.connectionsByPidFd", conn, pid, fd)
  140. if conn == nil {
  141. return nil
  142. }
  143. if timestamp != 0 && conn.Timestamp != timestamp {
  144. return nil
  145. }
  146. stats := c.l7Stats.get(r.Protocol, conn.Dest, conn.ActualDest)
  147. //trace := tracing.NewTrace(string(c.id), conn.ActualDest)
  148. switch r.Protocol {
  149. case l7.ProtocolHTTP:
  150. if c.AppInfo.AppName != "" {
  151. klog.Debugf("[%s] ->>>>> curl -> %s payload:[%s]", c.AppInfo.AppName, conn.ActualDest, r.Payload)
  152. }
  153. stats.observe(r.Status.Http(), "", r.Duration)
  154. case l7.ProtocolHTTP2:
  155. if conn.http2Parser == nil {
  156. conn.http2Parser = l7.NewHttp2Parser()
  157. }
  158. requests := conn.http2Parser.Parse(r.Method, r.Payload, uint64(r.Duration))
  159. for _, req := range requests {
  160. stats.observe(req.Status.Http(), "", req.Duration)
  161. //trace.Http2Request(req.Method, req.Path, req.Scheme, req.Status, req.Duration)
  162. }
  163. case l7.ProtocolPostgres:
  164. if r.Method != l7.MethodStatementClose {
  165. stats.observe(r.Status.String(), "", r.Duration)
  166. }
  167. //if conn.postgresParser == nil {
  168. // conn.postgresParser = l7.NewPostgresParser()
  169. //}
  170. //query := conn.postgresParser.Parse(r.Payload)
  171. //trace.PostgresQuery(query, r.Status.Error(), r.Duration)
  172. if c.l7Attach && c.valuableTrace(r.TraceId) {
  173. if conn.postgresParser == nil {
  174. conn.postgresParser = l7.NewPostgresParser()
  175. }
  176. query := conn.postgresParser.Parse(r.Payload)
  177. //trace.MysqlQuery(query, r.Status.Error(), r.Duration)
  178. if c.AppInfo.AppName != "" {
  179. klog.Debugf("[%s] ->>>>> Postgres -> %s payload:[%s]", c.AppInfo.AppName, conn.ActualDest, query)
  180. }
  181. //apmTrace, ok := c.getTrace(r.TraceId)
  182. apmTrace, err := c.getOrInitTrace(r.TraceId)
  183. //fmt.Println("mysql r.TraceId:", r.TraceId)
  184. //fmt.Println("ok:", ok)
  185. //fmt.Println("traceMap:", len(c.traceMap))
  186. if err == nil {
  187. //apmTrace.MysqlTraceQuery(query, r.Status.Error(), r.Duration, conn.ActualDest)
  188. //apmTrace.PostGreSqlTraceQueryEvent(query, r, conn.ActualDest)
  189. apmTrace.SQLTraceQueryEvent(l7.ProtocolPostgres, semconv.DBSystemPostgreSQL, query, r, conn.ActualDest)
  190. c.SendEvent(apmTrace, r.TraceId)
  191. }
  192. }
  193. case l7.ProtocolMysql:
  194. if r.Method != l7.MethodStatementClose {
  195. stats.observe(r.Status.String(), "", r.Duration)
  196. }
  197. if c.l7Attach && c.valuableTrace(r.TraceId) {
  198. if conn.mysqlParser == nil {
  199. conn.mysqlParser = l7.NewMysqlParser()
  200. }
  201. query := conn.mysqlParser.Parse(r.Payload, r.StatementId)
  202. //trace.MysqlQuery(query, r.Status.Error(), r.Duration)
  203. if c.AppInfo.AppName != "" {
  204. klog.Debugf("[%s] ->>>>> Mysql -> %s payload:[%s]", c.AppInfo.AppName, conn.ActualDest, query)
  205. }
  206. //apmTrace, ok := c.getTrace(r.TraceId)
  207. apmTrace, err := c.getOrInitTrace(r.TraceId)
  208. //fmt.Println("mysql r.TraceId:", r.TraceId)
  209. //fmt.Println("ok:", ok)
  210. //fmt.Println("traceMap:", len(c.traceMap))
  211. if err == nil {
  212. //apmTrace.MysqlTraceQuery(query, r.Status.Error(), r.Duration, conn.ActualDest)
  213. //apmTrace.MysqlTraceQueryEvent(query, r, conn.ActualDest)
  214. apmTrace.SQLTraceQueryEvent(l7.ProtocolMysql, semconv.DBSystemMySQL, query, r, conn.ActualDest)
  215. c.SendEvent(apmTrace, r.TraceId)
  216. }
  217. }
  218. case l7.ProtocolDM:
  219. //统计dm的query次数
  220. stats.observe(r.Status.String(), "", r.Duration)
  221. //是否发送数据
  222. if c.l7Attach && c.valuableTrace(r.TraceId) {
  223. if conn.dmParser == nil {
  224. conn.dmParser = l7.NewDmParser()
  225. }
  226. query := conn.dmParser.Parse(r.Payload, r.StatementId)
  227. if c.AppInfo.AppName != "" {
  228. klog.Debugf("[%s] ->>>>> Mysql -> %s DMSQL:[%s]", c.AppInfo.AppName, conn.ActualDest, query)
  229. }
  230. apmTrace, err := c.getOrInitTrace(r.TraceId)
  231. if err == nil {
  232. //apmTrace.DmTraceQueryEvent(query, r, conn.ActualDest)
  233. apmTrace.SQLTraceQueryEvent(l7.ProtocolDM, semconv.DBSystemDaMengDB, query, r, conn.ActualDest)
  234. c.SendEvent(apmTrace, r.TraceId)
  235. }
  236. }
  237. case l7.ProtocolMemcached:
  238. stats.observe(r.Status.String(), "", r.Duration)
  239. if c.l7Attach && c.valuableTrace(r.TraceId) {
  240. }
  241. //cmd, items := l7.ParseMemcached(r.Payload)
  242. //trace.MemcachedQuery(cmd, items, r.Status.Error(), r.Duration)
  243. case l7.ProtocolRedis:
  244. stats.observe(r.Status.String(), "", r.Duration)
  245. if c.l7Attach && c.valuableTrace(r.TraceId) {
  246. cmd, args := l7.ParseRedis(r.Payload)
  247. //fmt.Println("cmd", cmd)
  248. //fmt.Println("args", args)
  249. //apmTrace, ok := c.getTrace(r.TraceId)
  250. if c.AppInfo.AppName != "" {
  251. klog.Debugf("[%s] ->>>>> Redis -> %s DMSQL:[%s]", c.AppInfo.AppName, conn.ActualDest, cmd)
  252. }
  253. apmTrace, err := c.getOrInitTrace(r.TraceId)
  254. if err == nil {
  255. //apmTrace.RedisTraceQuery(cmd, args, r.Status.Error(), r.Duration)
  256. apmTrace.RedisTraceQueryEvent(cmd, args, r, conn.ActualDest)
  257. c.SendEvent(apmTrace, r.TraceId)
  258. }
  259. }
  260. //trace.RedisQuery(cmd, args, r.Status.Error(), r.Duration)
  261. case l7.ProtocolMongo:
  262. stats.observe(r.Status.String(), "", r.Duration)
  263. if c.l7Attach && c.valuableTrace(r.TraceId) {
  264. query := l7.ParseMongo(r.Payload)
  265. if c.AppInfo.AppName != "" {
  266. klog.Debugf("[%s] ->>>>> MongoDB -> %s SQL:[%s]", c.AppInfo.AppName, conn.ActualDest, query)
  267. }
  268. apmTrace, err := c.getOrInitTrace(r.TraceId)
  269. if err == nil {
  270. //apmTrace.RedisTraceQuery(cmd, args, r.Status.Error(), r.Duration)
  271. apmTrace.MongoTraceQueryEvent(query, r, conn.ActualDest)
  272. c.SendEvent(apmTrace, r.TraceId)
  273. }
  274. }
  275. case l7.ProtocolKafka, l7.ProtocolCassandra:
  276. stats.observe(r.Status.String(), "", r.Duration)
  277. if c.l7Attach && c.valuableTrace(r.TraceId) {
  278. }
  279. case l7.ProtocolRabbitmq, l7.ProtocolNats:
  280. stats.observe(r.Status.String(), r.Method.String(), 0)
  281. if c.l7Attach && c.valuableTrace(r.TraceId) {
  282. }
  283. case l7.ProtocolDubbo2:
  284. stats.observe(r.Status.String(), "", r.Duration)
  285. if c.l7Attach && c.valuableTrace(r.TraceId) {
  286. }
  287. }
  288. return nil
  289. }
  290. func (c *Container) buildIDs(pid uint32) bool {
  291. c.lock.Lock()
  292. defer c.lock.Unlock()
  293. p := c.processes[pid]
  294. if p != nil {
  295. p.cmdline = string(proc.GetRealCmdline(pid))
  296. }
  297. var sns []string
  298. var sport uint16
  299. for address, val := range c.getListens() {
  300. if val == 1 {
  301. ip := address.IP()
  302. if ip.Is4() && !ip.IsLoopback() {
  303. // 获取端口号
  304. sport = address.Port()
  305. sns = append(sns, fmt.Sprintf("%s:%d", ip, sport))
  306. ////c.instanceID.IntVal, c.instanceID.HashtVal, _ =
  307. //c.AppInfo.Sn = ip.String()
  308. //c.AppInfo.Sport = int(port)
  309. //strInstanceID := utils.BuildInt64ID(fmt.Sprintf("%s:%d", ip.String(), port))
  310. //fmt.Println(port)
  311. ////os.Exit(1)
  312. //c.AppInfo.InstanceIdHash.IntVal, _ = strInstanceID.ToInt64()
  313. //c.AppInfo.InstanceIdHash.HashtVal = strInstanceID.ToHashByte()
  314. ////c.AppInfo.InstanceId = c.instanceID.IntVal
  315. //strAgentID := utils.BuildInt64ID(fmt.Sprintf("%s:%s", strInstanceID, string(proc.GetExe(pid))))
  316. //c.AppInfo.AgentId, _ = strAgentID.ToInt64()
  317. //c.AppInfo.CodeType = c.GetCodeTypeFromCache(pid)
  318. //return true
  319. }
  320. }
  321. }
  322. if len(sns) > 0 {
  323. //c.instanceID.IntVal, c.instanceID.HashtVal, _ =
  324. snsStr := strings.Join(sns, ",")
  325. c.AppInfo.Sn = snsStr
  326. c.AppInfo.Sport = int(sport)
  327. strInstanceID := utils.BuildInt64ID(fmt.Sprintf("%s:%d", c.AppInfo.Sn, sport))
  328. c.AppInfo.InstanceIdHash.IntVal, _ = strInstanceID.ToInt64()
  329. c.AppInfo.InstanceIdHash.HashtVal = strInstanceID.ToHashByte()
  330. // strAgentID := utils.BuildInt64ID(fmt.Sprintf("%s:%s", utils.GetHostIP(), string(proc.GetExe(pid))))
  331. // c.AppInfo.AgentId, _ = strAgentID.ToInt64()
  332. c.AppInfo.CodeType = c.GetCodeTypeFromCache(pid)
  333. return true
  334. }
  335. return false
  336. }
  337. func (c *Container) ReBuildIds(pid uint32) {
  338. c.lock.Lock()
  339. defer c.lock.Unlock()
  340. var sns []string
  341. var sport uint16
  342. for address, val := range c.getListens() {
  343. if val == 1 {
  344. ip := address.IP()
  345. if ip.Is4() && !ip.IsLoopback() {
  346. // 获取端口号
  347. sport = address.Port()
  348. sns = append(sns, fmt.Sprintf("%s:%d", ip, sport))
  349. }
  350. }
  351. }
  352. if len(sns) > 0 {
  353. snsStr := strings.Join(sns, ",")
  354. c.AppInfo.Sn = snsStr
  355. c.AppInfo.Sport = int(sport)
  356. strInstanceID := utils.BuildInt64ID(fmt.Sprintf("%s:%d", c.AppInfo.Sn, sport))
  357. c.AppInfo.InstanceIdHash.IntVal, _ = strInstanceID.ToInt64()
  358. c.AppInfo.InstanceIdHash.HashtVal = strInstanceID.ToHashByte()
  359. // strAgentID := utils.BuildInt64ID(fmt.Sprintf("%s:%s", utils.GetHostIP(), string(proc.GetExe(pid))))
  360. strAgentID := utils.BuildInt64ID(fmt.Sprintf("%s:%d", utils.GetHostIP(), c.AppInfo.AppIdHash.IntVal))
  361. c.AppInfo.AgentId, _ = strAgentID.ToInt64()
  362. c.AppInfo.CodeType = c.GetCodeTypeFromCache(pid)
  363. }
  364. }
  365. func (c *Container) StackProcess(event ebpftracer.StackEvent, tracer *ebpftracer.Tracer) {
  366. c.lock.Lock()
  367. defer c.lock.Unlock()
  368. // get the associated uprobe
  369. uprobe, err := c.GetUprobe(event, tracer)
  370. if err != nil {
  371. //fmt.Println("GetUprobeGetUprobe errer: %v", err)
  372. klog.Errorf("failed to get uprobe for event %+v: %+v", event, err)
  373. return
  374. }
  375. if event.TraceId <= 0 {
  376. //fmt.Println("StackProcess TraceId id 0")
  377. klog.Errorf("failed to get uprobe(traceId is <= 0) for event %+v", event)
  378. return
  379. }
  380. // 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)
  381. stackFun := ebpftracer.StackFunEvent{}
  382. stackFun.Uprobe = &uprobe
  383. stackFun.StackEvent = event
  384. apmTrace, ok := c.getTrace(event.TraceId)
  385. if ok {
  386. apmTrace.FunAdd(stackFun)
  387. }
  388. }
  389. func byteExtractString(nameString [100]byte) string {
  390. n := bytes.IndexFunc(nameString[:], func(r rune) bool {
  391. return r == 0 || r < 32 || r > 126 // 截取到第一个零值或非打印字符
  392. })
  393. if n == -1 {
  394. n = len(nameString) // 没找到零值或非打印字符,使用数组长度
  395. }
  396. return string(nameString[:n])
  397. }
  398. func (c *Container) StackProcess2(event ebpftracer.StackEvent, tracer *ebpftracer.Tracer) {
  399. c.lock.Lock()
  400. defer c.lock.Unlock()
  401. // get the associated uprobe
  402. switch event.Location {
  403. case 0: // ret
  404. Funcname := ""
  405. if event.Type != uint64(CodeTypeJava) {
  406. uprobe, err := c.GetUprobe(event, tracer)
  407. if err != nil {
  408. //fmt.Println("GetUprobeGetUprobe errer: %v", err)
  409. klog.Errorf("failed to get uprobe for event %+v: %+v", event, err)
  410. return
  411. }
  412. Funcname = uprobe.Funcname
  413. } else {
  414. ClassName := byteExtractString(event.ClassName)
  415. MethedName := byteExtractString(event.MethedName)
  416. Funcname = ClassName + "." + MethedName
  417. }
  418. if event.TraceId <= 0 {
  419. //fmt.Println("StackProcess TraceId id 0")
  420. klog.Errorf("failed to get uprobe(traceId is <= 0) for event %+v", event)
  421. return
  422. }
  423. //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)
  424. apmTrace, err := c.getOrInitTrace(event.TraceId)
  425. if err == nil {
  426. //fmt.Println("append FuncTraceQuery fun:", event.TraceId, uprobe.Funcname, event.Pid)
  427. duration := event.TimeNsEnd - event.TimeNsStart
  428. apmTrace.FuncTraceQuery(Funcname, time.Duration(duration), event.TimeNsStart, event.TimeNsEnd)
  429. c.SendEvent(apmTrace, event.TraceId)
  430. }
  431. }
  432. }
  433. // ResolveAddress returns the symbol(s) and offset of the given address.
  434. func (c *Container) ResolveAddress(addr uint64, symbols []elf.Symbol) (syms []elf.Symbol, offset uint, err error) {
  435. if addr == 0 {
  436. // err = errors.Wrapf(SymbolNotFoundError, "0")
  437. return
  438. }
  439. // symbols, _, err := e.Symbols()
  440. if err != nil {
  441. return
  442. }
  443. idx := sort.Search(len(symbols), func(i int) bool { return symbols[i].Value > addr })
  444. if idx == 0 {
  445. // err = errors.Wrap(SymbolNotFoundError, fmt.Sprintf("%x", addr))
  446. return
  447. }
  448. // why diff symbol may contains the same addr?
  449. sym := symbols[idx-1]
  450. for i := idx - 1; i >= 0 && symbols[i].Value == sym.Value; i-- {
  451. syms = append(syms, symbols[i])
  452. }
  453. for i := idx; i < len(symbols) && symbols[i].Value == sym.Value; i++ {
  454. syms = append(syms, symbols[i])
  455. }
  456. return syms, uint(addr - sym.Value), nil
  457. }
  458. type MemoryMap struct {
  459. Start, End uint64
  460. }
  461. // ReadFirstLineOfMapsFile reads the first line of /proc/<pid>/maps file and return the memory map as a MemoryMap struct
  462. func ReadFirstLineOfMapsFile(pid string) (*MemoryMap, error) {
  463. file, err := os.Open(fmt.Sprintf("/proc/%s/maps", pid))
  464. if err != nil {
  465. return nil, err
  466. }
  467. defer file.Close()
  468. scanner := bufio.NewScanner(file)
  469. if scanner.Scan() {
  470. fields := strings.Fields(scanner.Text())
  471. addresses := strings.Split(fields[0], "-")
  472. if len(addresses) != 2 {
  473. return nil, errors.New("unexpected format in /proc/<pid>/maps")
  474. }
  475. start, err := strconv.ParseUint(addresses[0], 16, 64)
  476. if err != nil {
  477. return nil, err
  478. }
  479. end, err := strconv.ParseUint(addresses[1], 16, 64)
  480. if err != nil {
  481. return nil, err
  482. }
  483. return &MemoryMap{
  484. Start: start,
  485. End: end,
  486. }, nil
  487. }
  488. if err := scanner.Err(); err != nil {
  489. return nil, err
  490. }
  491. return nil, errors.New("empty /proc/<pid>/maps")
  492. }
  493. func (c *Container) GetUprobe(event ebpftracer.StackEvent, tracer *ebpftracer.Tracer) (uprobe tracer.Uprobe, err error) {
  494. //fmt.Println("GetUprobe entory:")
  495. memoryMap, _ := ReadFirstLineOfMapsFile(strconv.Itoa(int(event.Pid)))
  496. Address := event.Ip - memoryMap.Start
  497. // fmt.Printf("memoryMap.Start: %x, event.Ip: %x, Address: %x\n", memoryMap.Start, event.Ip, Address)
  498. for _, fun := range c.UprobesMap {
  499. funAddress := fun.Address + fun.AbsOffset
  500. // fmt.Printf("GetUprobeGetUprobeGetUprobe:fun.Address %x, fun.AbsOffset: %x\n", fun.Address, fun.AbsOffset)
  501. if funAddress == Address {
  502. // fmt.Printf("---GetUprobeGetUprobeGetUprobe: %x, event.Ip: %x ---- %s--%x\n", memoryMap.Start, event.Ip, fun.Funcname, fun.Address)
  503. return fun, nil
  504. }
  505. }
  506. syms, _, err := c.ResolveAddress(event.Ip, tracer.Symbols)
  507. if err != nil {
  508. return
  509. }
  510. for _, sym := range syms {
  511. //fmt.Println("GetUprobeGetUprobeGetUprobe: %s+%d", sym.Name, offset)
  512. uprobe, ok := tracer.UprobesMap[fmt.Sprintf("%s-%s", sym.Name, sym.Value)]
  513. if ok {
  514. return uprobe, nil
  515. }
  516. }
  517. err = errors.New("uprobe not found")
  518. return
  519. }
  520. func (c *Container) GetAppInfo() AppInfo {
  521. return c.AppInfo
  522. }
  523. // 可注入前置
  524. func (c *Container) checkEventReady() bool {
  525. c.lock.Lock()
  526. defer c.lock.Unlock()
  527. return c.l7EventReady
  528. }
  529. func (c *Container) eventReady() {
  530. c.lock.Lock()
  531. defer c.lock.Unlock()
  532. c.l7EventReady = true
  533. }
  534. // uprobe前置
  535. func (c *Container) Isl7AttachSuccess() bool {
  536. c.lock.Lock()
  537. defer c.lock.Unlock()
  538. return c.l7Attach
  539. }
  540. func (c *Container) l7AttachSuccess() {
  541. c.lock.Lock()
  542. defer c.lock.Unlock()
  543. c.l7Attach = true
  544. }
  545. func (c *Container) ctrlStack(r *Registry, pid uint32) {
  546. resp, err := c.GetCodeSetting(r)
  547. if err != nil {
  548. klog.WithField("pid", pid).WithError(err).Error("[ctrlStack] GetCodeSetting failed.")
  549. return
  550. }
  551. if resp.BlackWhiteSettings.CollectStack == OPEN_STACK {
  552. // 有黑白名单规则 &&
  553. // 之前有注入 先卸载再注入
  554. // 之前没注入 直接注入
  555. // 没有有黑白名单 直接卸载
  556. if c.hasStackRule(resp) {
  557. if c.stackRuleUpdate(resp) {
  558. // 重新注入
  559. err = c.DetachStack(pid, APP_UNINSTALL)
  560. if err != nil {
  561. klog.WithError(err).Errorf("[ctrlStack][end] Failed detach stack trace!")
  562. }
  563. }
  564. klog.WithField("pid", pid).Infoln("[ctrlStack] Attach app stack.")
  565. c.saveWhiteStackSettingInfo(resp)
  566. err = c.AttachStack(r.tracer, pid)
  567. if err != nil {
  568. c.AppInfo.SetAppStackError()
  569. klog.WithField("pid", pid).WithError(err).Errorf("[ctrlStack][end] Failed attach stack trace!")
  570. }
  571. } else {
  572. if c.noOrigRule() {
  573. return
  574. }
  575. c.saveWhiteStackSettingInfo(resp)
  576. // 关闭堆栈
  577. err = c.DetachStack(pid, APP_UNINSTALL)
  578. if err != nil {
  579. klog.WithError(err).Errorf("[ctrlStack][end] Failed detach stack trace!")
  580. }
  581. }
  582. } else {
  583. if c.noOrigRule() {
  584. return
  585. }
  586. c.saveWhiteStackSettingInfo(resp)
  587. // 关闭堆栈
  588. err = c.DetachStack(pid, APP_UNINSTALL)
  589. if err != nil {
  590. klog.WithError(err).Errorf("[ctrlStack][end] Failed detach stack trace!")
  591. }
  592. }
  593. }
  594. func (c *Container) verifyAttachConditions(r *Registry, pid uint32) (bool, int) {
  595. p := c.processes[pid]
  596. if p != nil && c.checkEventReady() {
  597. codeType := c.GetCodeTypeFromCache(pid)
  598. if codeType.IsUnknownCode() {
  599. klog.WithField("pid", pid).Debug("[verify] unknown language.")
  600. return false, 0
  601. }
  602. cmdline := p.GetCmdline()
  603. if len(cmdline) == 0 {
  604. return false, 0
  605. }
  606. //whiteListByCode := r.getWhiteListByCodeType(codeType)
  607. whiteListByCode := r.getWhiteListAll()
  608. //klog.WithField("pid", pid).WithField("codeType", codeType.String()).
  609. // Infof("[verify] white list %v", utils.ToString(whiteListByCode))
  610. // 当前语言的白名单规则
  611. for _, setting := range whiteListByCode {
  612. ruleVal := setting.Filters
  613. if ruleVal == "" {
  614. continue
  615. }
  616. // 判断规则
  617. if strings.Contains(cmdline, ruleVal) {
  618. //if !codeType.IsJvmCode() {
  619. // klog.WithField("pid", pid).Warning("[verify] This agent version only supports JVM applications.")
  620. // return false, 0
  621. //}
  622. c.WhiteSettingInfo.AppName = setting.AppName
  623. c.WhiteSettingInfo.Filters = setting.Filters
  624. klog.WithField("pid", pid).
  625. WithField("codeType", codeType.String()).
  626. WithField("ruleVal", ruleVal).
  627. WithField("cmdline", cmdline).
  628. //WithField("stack", setting.OpenStack).
  629. WithField("white list", utils.ToString(whiteListByCode)).
  630. Infoln("[verify] check successful.")
  631. return true, 0
  632. }
  633. }
  634. }
  635. return false, 0
  636. }
  637. // 1.卸载入口
  638. func (c *Container) Detach(tracer *ebpftracer.Tracer, pid uint32, detachType APP_TYPE) {
  639. c.lock.Lock()
  640. defer c.lock.Unlock()
  641. if p := c.processes[pid]; p != nil {
  642. err := c.DetachUprobes(tracer, pid, detachType)
  643. if err != nil {
  644. klog.WithError(err).Errorln("DetachUprobes Error.")
  645. }
  646. err = c.DetachStack(pid, detachType)
  647. if err != nil {
  648. klog.WithError(err).Errorln("DetachStack Error.")
  649. }
  650. // 关闭7层监控
  651. c.l7Attach = false
  652. // 变更应用状态
  653. if err != nil {
  654. detachType = detachType.Error()
  655. }
  656. c.AppInfo.SetAppStatus(detachType)
  657. }
  658. }
  659. // 1.1卸载uprobe
  660. func (c *Container) DetachUprobes(tracer *ebpftracer.Tracer, pid uint32, detachType APP_TYPE) error {
  661. // close uprobe
  662. if p := c.processes[pid]; p != nil {
  663. for _, u := range p.uprobes {
  664. err := u.Close()
  665. if err != nil {
  666. return err
  667. }
  668. }
  669. p.uprobes = []link.Link{}
  670. switch detachType {
  671. case APP_UNINSTALL, APP_FUSE:
  672. codeType := c.GetCodeTypeFromCache(pid)
  673. switch codeType {
  674. case CodeTypeJava:
  675. p.jvmAttachOnce = false
  676. case CodeTypeGo:
  677. p.goTlsUprobesChecked = false
  678. p.openSslUprobesChecked = false
  679. default:
  680. }
  681. case APP_UPROBE_ERROR:
  682. klog.Infof("[DetachUprobes] ERROR_DETACH for pid %d", pid)
  683. default:
  684. }
  685. //delete the proc info form proc_info_map(for kernel) when the uprobe detached
  686. if err := tracer.DelKProcInfo(pid); err != nil {
  687. return fmt.Errorf("[DetachUprobes] failed to delete KProcInfo for pid %d, detach type is:%s", pid, detachType)
  688. } else {
  689. klog.Infof("[DetachUprobes] delete KProcInfo success for pid %d,detachType:%d", pid, detachType)
  690. c.AppInfo.EBPFProcInfo = nil
  691. }
  692. } else {
  693. return fmt.Errorf("[DetachUprobes] cannot find uprobe for pid %d", pid)
  694. }
  695. return nil
  696. }
  697. // 1.2卸载堆栈
  698. func (c *Container) DetachStack(pid uint32, detachType APP_TYPE) error {
  699. if p := c.processes[pid]; p != nil {
  700. var err error
  701. codeType := c.GetCodeTypeFromCache(pid)
  702. switch codeType {
  703. // 1.2.1 卸载 jvm堆栈
  704. case CodeTypeJava:
  705. err = c.detachJvmStack(pid)
  706. default:
  707. err = p.closeStackUprobes()
  708. }
  709. if err != nil {
  710. klog.WithError(err).Errorln("[detachStack] failed to detach stack")
  711. return err
  712. }
  713. p.stackAttachOnce = false
  714. } else {
  715. return fmt.Errorf("[DetachStack] cannot find uprobe for pid %d", pid)
  716. }
  717. return nil
  718. }
  719. // 1.2.1 卸载 jvm堆栈
  720. func (c *Container) detachJvmStack(pid uint32) error {
  721. if p := c.processes[pid]; p != nil {
  722. //if p.stackStatus.IsStackUprobesSuccess() || len(p.stackUprobes) > 0 {
  723. //}
  724. // 卸载 JavaAgent
  725. var err error
  726. if p.stackStatus.IsJattachSuccess() {
  727. // 卸载堆栈probes
  728. err = p.closeStackUprobes()
  729. if err != nil {
  730. klog.WithError(err).Errorf("[detachJvmStack] closeStackUprobes")
  731. }
  732. err = p.uninstallJavaAgent()
  733. if err != nil {
  734. klog.WithError(err).Errorf("[detachJvmStack] uninstallJavaAgent")
  735. }
  736. }
  737. return err
  738. }
  739. return nil
  740. }
  741. func (c *Container) getRootfs() string {
  742. if c.metadata != nil && c.metadata.rootfs != "" {
  743. return path.Join(*flags.HostDirPathPrefix, c.metadata.rootfs)
  744. }
  745. return ""
  746. }
  747. func (c *Container) BuildActiveApps(runtimeApps map[uint32]AppStatusInfo, pid uint32) {
  748. if c == nil {
  749. //klog.WithField("pid", pid).Warningln("[BuildActiveApps] container_apm is nil.")
  750. return
  751. }
  752. if c.AppInfo.AppName == "" {
  753. return
  754. }
  755. klog.WithField("pid", pid).WithField("appname", c.AppInfo.AppName).Infof("[BuildActiveApps] container %s is running.", c.AppInfo.AppName)
  756. detail := AppStatusInfo{
  757. Pid: pid,
  758. ProcName: c.containerName,
  759. AppName: c.AppInfo.AppName,
  760. Language: c.AppInfo.CodeType.String(),
  761. AppID: c.AppInfo.AppIdHash.IntVal,
  762. AgentID: c.AppInfo.AgentId,
  763. InstanceID: c.AppInfo.InstanceIdHash.IntVal,
  764. Sn: c.AppInfo.Sn,
  765. Sport: c.AppInfo.Sport,
  766. RegisterAt: time.Unix(c.AppInfo.RegisterAt, 0).Format("060102 15:04:05"),
  767. PreStatus: c.AppInfo.PreStatus,
  768. Status: c.AppInfo.Status,
  769. Rule: c.WhiteSettingInfo.Filters,
  770. Container: string(c.id),
  771. }
  772. detail.Rule = fmt.Sprintf("%s|%d", c.WhiteSettingInfo.Filters, c.WhiteSettingInfo.WhiteStackSettingInfo.OpenStack)
  773. if c.AppInfo.UpdateAt != 0 {
  774. detail.UpdateAt = time.Unix(c.AppInfo.UpdateAt, 0).Format("060102 15:04:05")
  775. }
  776. p := c.processes[pid]
  777. if p != nil {
  778. detail.StackStatus = p.stackStatus.String()
  779. v := 0
  780. if !p.versionFailed {
  781. v = 1
  782. }
  783. detail.StackStatus += fmt.Sprintf("V=%d", v)
  784. }
  785. runtimeApps[pid] = detail
  786. }
  787. func (c *Container) AgentCtrl(r *Registry, pid uint32) {
  788. if c == nil {
  789. //klog.WithField("pid", pid).Warningln("[AgentCtrl] cannot find container.")
  790. return
  791. }
  792. var err error
  793. verifyAttachConditions, _ := c.verifyAttachConditions(r, pid)
  794. // fusing UNINSTALL
  795. if r.isFusing && c.Isl7AttachSuccess() {
  796. c.Detach(r.tracer, pid, APP_FUSE)
  797. klog.WithField("pid", pid).Infoln("[AgentCtrl] fusing")
  798. return
  799. }
  800. // verify UNINSTALL
  801. if !verifyAttachConditions && c.Isl7AttachSuccess() {
  802. c.Detach(r.tracer, pid, APP_UNINSTALL)
  803. klog.WithField("pid", pid).Infoln("[AgentCtrl] rule uninstall.")
  804. return
  805. }
  806. if verifyAttachConditions {
  807. err = c.RegisterAppInfo(r, pid)
  808. if err != nil {
  809. klog.WithError(err).Errorf("[AgentCtrl] Failed registerAppInfo.")
  810. return
  811. }
  812. klog.WithField("pid", pid).Infoln("[AgentCtrl] Attach uprobes.")
  813. err = c.AttachUprobes(r.tracer, pid, "Agentctrl")
  814. if err != nil {
  815. klog.WithField("pid", pid).WithError(err).Errorf("[AgentCtrl] Failed attach uprobes error!")
  816. return
  817. } else {
  818. klog.WithField("pid", pid).Infoln("[AgentCtrl] Attach uprobes success!")
  819. }
  820. // 堆栈控制
  821. c.ctrlStack(r, pid)
  822. }
  823. }