container_apm.go 27 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883
  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. }
  265. //query := l7.ParseMongo(r.Payload)
  266. //trace.MongoQuery(query, r.Status.Error(), r.Duration)
  267. case l7.ProtocolKafka, l7.ProtocolCassandra:
  268. stats.observe(r.Status.String(), "", r.Duration)
  269. if c.l7Attach && c.valuableTrace(r.TraceId) {
  270. }
  271. case l7.ProtocolRabbitmq, l7.ProtocolNats:
  272. stats.observe(r.Status.String(), r.Method.String(), 0)
  273. if c.l7Attach && c.valuableTrace(r.TraceId) {
  274. }
  275. case l7.ProtocolDubbo2:
  276. stats.observe(r.Status.String(), "", r.Duration)
  277. if c.l7Attach && c.valuableTrace(r.TraceId) {
  278. }
  279. }
  280. return nil
  281. }
  282. func (c *Container) buildIDs(pid uint32) bool {
  283. c.lock.Lock()
  284. defer c.lock.Unlock()
  285. p := c.processes[pid]
  286. if p != nil {
  287. p.cmdline = string(proc.GetRealCmdline(pid))
  288. }
  289. var sns []string
  290. var sport uint16
  291. for address, val := range c.getListens() {
  292. if val == 1 {
  293. ip := address.IP()
  294. if ip.Is4() && !ip.IsLoopback() {
  295. // 获取端口号
  296. sport = address.Port()
  297. sns = append(sns, fmt.Sprintf("%s:%d", ip, sport))
  298. ////c.instanceID.IntVal, c.instanceID.HashtVal, _ =
  299. //c.AppInfo.Sn = ip.String()
  300. //c.AppInfo.Sport = int(port)
  301. //strInstanceID := utils.BuildInt64ID(fmt.Sprintf("%s:%d", ip.String(), port))
  302. //fmt.Println(port)
  303. ////os.Exit(1)
  304. //c.AppInfo.InstanceIdHash.IntVal, _ = strInstanceID.ToInt64()
  305. //c.AppInfo.InstanceIdHash.HashtVal = strInstanceID.ToHashByte()
  306. ////c.AppInfo.InstanceId = c.instanceID.IntVal
  307. //strAgentID := utils.BuildInt64ID(fmt.Sprintf("%s:%s", strInstanceID, string(proc.GetExe(pid))))
  308. //c.AppInfo.AgentId, _ = strAgentID.ToInt64()
  309. //c.AppInfo.CodeType = c.GetCodeTypeFromCache(pid)
  310. //return true
  311. }
  312. }
  313. }
  314. if len(sns) > 0 {
  315. //c.instanceID.IntVal, c.instanceID.HashtVal, _ =
  316. snsStr := strings.Join(sns, ",")
  317. c.AppInfo.Sn = snsStr
  318. c.AppInfo.Sport = int(sport)
  319. strInstanceID := utils.BuildInt64ID(fmt.Sprintf("%s:%d", c.AppInfo.Sn, sport))
  320. c.AppInfo.InstanceIdHash.IntVal, _ = strInstanceID.ToInt64()
  321. c.AppInfo.InstanceIdHash.HashtVal = strInstanceID.ToHashByte()
  322. // strAgentID := utils.BuildInt64ID(fmt.Sprintf("%s:%s", utils.GetHostIP(), string(proc.GetExe(pid))))
  323. // c.AppInfo.AgentId, _ = strAgentID.ToInt64()
  324. c.AppInfo.CodeType = c.GetCodeTypeFromCache(pid)
  325. return true
  326. }
  327. return false
  328. }
  329. func (c *Container) ReBuildIds(pid uint32) {
  330. c.lock.Lock()
  331. defer c.lock.Unlock()
  332. var sns []string
  333. var sport uint16
  334. for address, val := range c.getListens() {
  335. if val == 1 {
  336. ip := address.IP()
  337. if ip.Is4() && !ip.IsLoopback() {
  338. // 获取端口号
  339. sport = address.Port()
  340. sns = append(sns, fmt.Sprintf("%s:%d", ip, sport))
  341. }
  342. }
  343. }
  344. if len(sns) > 0 {
  345. snsStr := strings.Join(sns, ",")
  346. c.AppInfo.Sn = snsStr
  347. c.AppInfo.Sport = int(sport)
  348. strInstanceID := utils.BuildInt64ID(fmt.Sprintf("%s:%d", c.AppInfo.Sn, sport))
  349. c.AppInfo.InstanceIdHash.IntVal, _ = strInstanceID.ToInt64()
  350. c.AppInfo.InstanceIdHash.HashtVal = strInstanceID.ToHashByte()
  351. // strAgentID := utils.BuildInt64ID(fmt.Sprintf("%s:%s", utils.GetHostIP(), string(proc.GetExe(pid))))
  352. strAgentID := utils.BuildInt64ID(fmt.Sprintf("%s:%d", utils.GetHostIP(), c.AppInfo.AppIdHash.IntVal))
  353. c.AppInfo.AgentId, _ = strAgentID.ToInt64()
  354. c.AppInfo.CodeType = c.GetCodeTypeFromCache(pid)
  355. }
  356. }
  357. func (c *Container) StackProcess(event ebpftracer.StackEvent, tracer *ebpftracer.Tracer) {
  358. c.lock.Lock()
  359. defer c.lock.Unlock()
  360. // get the associated uprobe
  361. uprobe, err := c.GetUprobe(event, tracer)
  362. if err != nil {
  363. //fmt.Println("GetUprobeGetUprobe errer: %v", err)
  364. klog.Errorf("failed to get uprobe for event %+v: %+v", event, err)
  365. return
  366. }
  367. if event.TraceId <= 0 {
  368. //fmt.Println("StackProcess TraceId id 0")
  369. klog.Errorf("failed to get uprobe(traceId is <= 0) for event %+v", event)
  370. return
  371. }
  372. // 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)
  373. stackFun := ebpftracer.StackFunEvent{}
  374. stackFun.Uprobe = &uprobe
  375. stackFun.StackEvent = event
  376. apmTrace, ok := c.getTrace(event.TraceId)
  377. if ok {
  378. apmTrace.FunAdd(stackFun)
  379. }
  380. }
  381. func byteExtractString(nameString [100]byte) string {
  382. n := bytes.IndexFunc(nameString[:], func(r rune) bool {
  383. return r == 0 || r < 32 || r > 126 // 截取到第一个零值或非打印字符
  384. })
  385. if n == -1 {
  386. n = len(nameString) // 没找到零值或非打印字符,使用数组长度
  387. }
  388. return string(nameString[:n])
  389. }
  390. func (c *Container) StackProcess2(event ebpftracer.StackEvent, tracer *ebpftracer.Tracer) {
  391. c.lock.Lock()
  392. defer c.lock.Unlock()
  393. // get the associated uprobe
  394. switch event.Location {
  395. case 0: // ret
  396. Funcname := ""
  397. if event.Type != uint64(CodeTypeJava) {
  398. uprobe, err := c.GetUprobe(event, tracer)
  399. if err != nil {
  400. //fmt.Println("GetUprobeGetUprobe errer: %v", err)
  401. klog.Errorf("failed to get uprobe for event %+v: %+v", event, err)
  402. return
  403. }
  404. Funcname = uprobe.Funcname
  405. } else {
  406. ClassName := byteExtractString(event.ClassName)
  407. MethedName := byteExtractString(event.MethedName)
  408. Funcname = ClassName + "." + MethedName
  409. }
  410. if event.TraceId <= 0 {
  411. //fmt.Println("StackProcess TraceId id 0")
  412. klog.Errorf("failed to get uprobe(traceId is <= 0) for event %+v", event)
  413. return
  414. }
  415. //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)
  416. apmTrace, err := c.getOrInitTrace(event.TraceId)
  417. if err == nil {
  418. //fmt.Println("append FuncTraceQuery fun:", event.TraceId, uprobe.Funcname, event.Pid)
  419. duration := event.TimeNsEnd - event.TimeNsStart
  420. apmTrace.FuncTraceQuery(Funcname, time.Duration(duration), event.TimeNsStart, event.TimeNsEnd)
  421. c.SendEvent(apmTrace, event.TraceId)
  422. }
  423. }
  424. }
  425. // ResolveAddress returns the symbol(s) and offset of the given address.
  426. func (c *Container) ResolveAddress(addr uint64, symbols []elf.Symbol) (syms []elf.Symbol, offset uint, err error) {
  427. if addr == 0 {
  428. // err = errors.Wrapf(SymbolNotFoundError, "0")
  429. return
  430. }
  431. // symbols, _, err := e.Symbols()
  432. if err != nil {
  433. return
  434. }
  435. idx := sort.Search(len(symbols), func(i int) bool { return symbols[i].Value > addr })
  436. if idx == 0 {
  437. // err = errors.Wrap(SymbolNotFoundError, fmt.Sprintf("%x", addr))
  438. return
  439. }
  440. // why diff symbol may contains the same addr?
  441. sym := symbols[idx-1]
  442. for i := idx - 1; i >= 0 && symbols[i].Value == sym.Value; i-- {
  443. syms = append(syms, symbols[i])
  444. }
  445. for i := idx; i < len(symbols) && symbols[i].Value == sym.Value; i++ {
  446. syms = append(syms, symbols[i])
  447. }
  448. return syms, uint(addr - sym.Value), nil
  449. }
  450. type MemoryMap struct {
  451. Start, End uint64
  452. }
  453. // ReadFirstLineOfMapsFile reads the first line of /proc/<pid>/maps file and return the memory map as a MemoryMap struct
  454. func ReadFirstLineOfMapsFile(pid string) (*MemoryMap, error) {
  455. file, err := os.Open(fmt.Sprintf("/proc/%s/maps", pid))
  456. if err != nil {
  457. return nil, err
  458. }
  459. defer file.Close()
  460. scanner := bufio.NewScanner(file)
  461. if scanner.Scan() {
  462. fields := strings.Fields(scanner.Text())
  463. addresses := strings.Split(fields[0], "-")
  464. if len(addresses) != 2 {
  465. return nil, errors.New("unexpected format in /proc/<pid>/maps")
  466. }
  467. start, err := strconv.ParseUint(addresses[0], 16, 64)
  468. if err != nil {
  469. return nil, err
  470. }
  471. end, err := strconv.ParseUint(addresses[1], 16, 64)
  472. if err != nil {
  473. return nil, err
  474. }
  475. return &MemoryMap{
  476. Start: start,
  477. End: end,
  478. }, nil
  479. }
  480. if err := scanner.Err(); err != nil {
  481. return nil, err
  482. }
  483. return nil, errors.New("empty /proc/<pid>/maps")
  484. }
  485. func (c *Container) GetUprobe(event ebpftracer.StackEvent, tracer *ebpftracer.Tracer) (uprobe tracer.Uprobe, err error) {
  486. //fmt.Println("GetUprobe entory:")
  487. memoryMap, _ := ReadFirstLineOfMapsFile(strconv.Itoa(int(event.Pid)))
  488. Address := event.Ip - memoryMap.Start
  489. // fmt.Printf("memoryMap.Start: %x, event.Ip: %x, Address: %x\n", memoryMap.Start, event.Ip, Address)
  490. for _, fun := range c.UprobesMap {
  491. funAddress := fun.Address + fun.AbsOffset
  492. // fmt.Printf("GetUprobeGetUprobeGetUprobe:fun.Address %x, fun.AbsOffset: %x\n", fun.Address, fun.AbsOffset)
  493. if funAddress == Address {
  494. // fmt.Printf("---GetUprobeGetUprobeGetUprobe: %x, event.Ip: %x ---- %s--%x\n", memoryMap.Start, event.Ip, fun.Funcname, fun.Address)
  495. return fun, nil
  496. }
  497. }
  498. syms, _, err := c.ResolveAddress(event.Ip, tracer.Symbols)
  499. if err != nil {
  500. return
  501. }
  502. for _, sym := range syms {
  503. //fmt.Println("GetUprobeGetUprobeGetUprobe: %s+%d", sym.Name, offset)
  504. uprobe, ok := tracer.UprobesMap[fmt.Sprintf("%s-%s", sym.Name, sym.Value)]
  505. if ok {
  506. return uprobe, nil
  507. }
  508. }
  509. err = errors.New("uprobe not found")
  510. return
  511. }
  512. func (c *Container) GetAppInfo() AppInfo {
  513. return c.AppInfo
  514. }
  515. // 可注入前置
  516. func (c *Container) checkEventReady() bool {
  517. c.lock.Lock()
  518. defer c.lock.Unlock()
  519. return c.l7EventReady
  520. }
  521. func (c *Container) eventReady() {
  522. c.lock.Lock()
  523. defer c.lock.Unlock()
  524. c.l7EventReady = true
  525. }
  526. // uprobe前置
  527. func (c *Container) Isl7AttachSuccess() bool {
  528. c.lock.Lock()
  529. defer c.lock.Unlock()
  530. return c.l7Attach
  531. }
  532. func (c *Container) l7AttachSuccess() {
  533. c.lock.Lock()
  534. defer c.lock.Unlock()
  535. c.l7Attach = true
  536. }
  537. func (c *Container) ctrlStack(r *Registry, pid uint32) {
  538. resp, err := c.GetCodeSetting(r)
  539. if err != nil {
  540. klog.WithField("pid", pid).WithError(err).Error("[ctrlStack] GetCodeSetting failed.")
  541. return
  542. }
  543. if resp.BlackWhiteSettings.CollectStack == OPEN_STACK {
  544. // 有黑白名单规则 &&
  545. // 之前有注入 先卸载再注入
  546. // 之前没注入 直接注入
  547. // 没有有黑白名单 直接卸载
  548. if c.hasStackRule(resp) {
  549. if c.stackRuleUpdate(resp) {
  550. // 重新注入
  551. err = c.DetachStack(pid, APP_UNINSTALL)
  552. if err != nil {
  553. klog.WithError(err).Errorf("[ctrlStack][end] Failed detach stack trace!")
  554. }
  555. }
  556. klog.WithField("pid", pid).Infoln("[ctrlStack] Attach app stack.")
  557. c.saveWhiteStackSettingInfo(resp)
  558. err = c.AttachStack(r.tracer, pid)
  559. if err != nil {
  560. c.AppInfo.SetAppStackError()
  561. klog.WithField("pid", pid).WithError(err).Errorf("[ctrlStack][end] Failed attach stack trace!")
  562. }
  563. } else {
  564. if c.noOrigRule() {
  565. return
  566. }
  567. c.saveWhiteStackSettingInfo(resp)
  568. // 关闭堆栈
  569. err = c.DetachStack(pid, APP_UNINSTALL)
  570. if err != nil {
  571. klog.WithError(err).Errorf("[ctrlStack][end] Failed detach stack trace!")
  572. }
  573. }
  574. } else {
  575. if c.noOrigRule() {
  576. return
  577. }
  578. c.saveWhiteStackSettingInfo(resp)
  579. // 关闭堆栈
  580. err = c.DetachStack(pid, APP_UNINSTALL)
  581. if err != nil {
  582. klog.WithError(err).Errorf("[ctrlStack][end] Failed detach stack trace!")
  583. }
  584. }
  585. }
  586. func (c *Container) verifyAttachConditions(r *Registry, pid uint32) (bool, int) {
  587. p := c.processes[pid]
  588. if p != nil && c.checkEventReady() {
  589. codeType := c.GetCodeTypeFromCache(pid)
  590. if codeType.IsUnknownCode() {
  591. klog.WithField("pid", pid).Debug("[verify] unknown language.")
  592. return false, 0
  593. }
  594. cmdline := p.GetCmdline()
  595. if len(cmdline) == 0 {
  596. return false, 0
  597. }
  598. //whiteListByCode := r.getWhiteListByCodeType(codeType)
  599. whiteListByCode := r.getWhiteListAll()
  600. //klog.WithField("pid", pid).WithField("codeType", codeType.String()).
  601. // Infof("[verify] white list %v", utils.ToString(whiteListByCode))
  602. // 当前语言的白名单规则
  603. for _, setting := range whiteListByCode {
  604. ruleVal := setting.Filters
  605. if ruleVal == "" {
  606. continue
  607. }
  608. // 判断规则
  609. if strings.Contains(cmdline, ruleVal) {
  610. //if !codeType.IsJvmCode() {
  611. // klog.WithField("pid", pid).Warning("[verify] This agent version only supports JVM applications.")
  612. // return false, 0
  613. //}
  614. c.WhiteSettingInfo.AppName = setting.AppName
  615. c.WhiteSettingInfo.Filters = setting.Filters
  616. klog.WithField("pid", pid).
  617. WithField("codeType", codeType.String()).
  618. WithField("ruleVal", ruleVal).
  619. WithField("cmdline", cmdline).
  620. //WithField("stack", setting.OpenStack).
  621. WithField("white list", utils.ToString(whiteListByCode)).
  622. Infoln("[verify] check successful.")
  623. return true, 0
  624. }
  625. }
  626. }
  627. return false, 0
  628. }
  629. // 1.卸载入口
  630. func (c *Container) Detach(tracer *ebpftracer.Tracer, pid uint32, detachType APP_TYPE) {
  631. c.lock.Lock()
  632. defer c.lock.Unlock()
  633. if p := c.processes[pid]; p != nil {
  634. err := c.DetachUprobes(tracer, pid, detachType)
  635. if err != nil {
  636. klog.WithError(err).Errorln("DetachUprobes Error.")
  637. }
  638. err = c.DetachStack(pid, detachType)
  639. if err != nil {
  640. klog.WithError(err).Errorln("DetachStack Error.")
  641. }
  642. // 关闭7层监控
  643. c.l7Attach = false
  644. // 变更应用状态
  645. if err != nil {
  646. detachType = detachType.Error()
  647. }
  648. c.AppInfo.SetAppStatus(detachType)
  649. }
  650. }
  651. // 1.1卸载uprobe
  652. func (c *Container) DetachUprobes(tracer *ebpftracer.Tracer, pid uint32, detachType APP_TYPE) error {
  653. // close uprobe
  654. if p := c.processes[pid]; p != nil {
  655. for _, u := range p.uprobes {
  656. err := u.Close()
  657. if err != nil {
  658. return err
  659. }
  660. }
  661. p.uprobes = []link.Link{}
  662. switch detachType {
  663. case APP_UNINSTALL, APP_FUSE:
  664. codeType := c.GetCodeTypeFromCache(pid)
  665. switch codeType {
  666. case CodeTypeJava:
  667. p.jvmAttachOnce = false
  668. case CodeTypeGo:
  669. p.goTlsUprobesChecked = false
  670. p.openSslUprobesChecked = false
  671. default:
  672. }
  673. case APP_UPROBE_ERROR:
  674. klog.Infof("[DetachUprobes] ERROR_DETACH for pid %d", pid)
  675. default:
  676. }
  677. //delete the proc info form proc_info_map(for kernel) when the uprobe detached
  678. if err := tracer.DelKProcInfo(pid); err != nil {
  679. return fmt.Errorf("[DetachUprobes] failed to delete KProcInfo for pid %d, detach type is:%s", pid, detachType)
  680. } else {
  681. klog.Infof("[DetachUprobes] delete KProcInfo success for pid %d,detachType:%d", pid, detachType)
  682. c.AppInfo.EBPFProcInfo = nil
  683. }
  684. } else {
  685. return fmt.Errorf("[DetachUprobes] cannot find uprobe for pid %d", pid)
  686. }
  687. return nil
  688. }
  689. // 1.2卸载堆栈
  690. func (c *Container) DetachStack(pid uint32, detachType APP_TYPE) error {
  691. if p := c.processes[pid]; p != nil {
  692. var err error
  693. codeType := c.GetCodeTypeFromCache(pid)
  694. switch codeType {
  695. // 1.2.1 卸载 jvm堆栈
  696. case CodeTypeJava:
  697. err = c.detachJvmStack(pid)
  698. default:
  699. err = p.closeStackUprobes()
  700. }
  701. if err != nil {
  702. klog.WithError(err).Errorln("[detachStack] failed to detach stack")
  703. return err
  704. }
  705. p.stackAttachOnce = false
  706. } else {
  707. return fmt.Errorf("[DetachStack] cannot find uprobe for pid %d", pid)
  708. }
  709. return nil
  710. }
  711. // 1.2.1 卸载 jvm堆栈
  712. func (c *Container) detachJvmStack(pid uint32) error {
  713. if p := c.processes[pid]; p != nil {
  714. //if p.stackStatus.IsStackUprobesSuccess() || len(p.stackUprobes) > 0 {
  715. //}
  716. // 卸载 JavaAgent
  717. var err error
  718. if p.stackStatus.IsJattachSuccess() {
  719. // 卸载堆栈probes
  720. err = p.closeStackUprobes()
  721. if err != nil {
  722. klog.WithError(err).Errorf("[detachJvmStack] closeStackUprobes")
  723. }
  724. err = p.uninstallJavaAgent()
  725. if err != nil {
  726. klog.WithError(err).Errorf("[detachJvmStack] uninstallJavaAgent")
  727. }
  728. }
  729. return err
  730. }
  731. return nil
  732. }
  733. func (c *Container) getRootfs() string {
  734. if c.metadata != nil && c.metadata.rootfs != "" {
  735. return path.Join(*flags.HostDirPathPrefix, c.metadata.rootfs)
  736. }
  737. return ""
  738. }
  739. func (c *Container) BuildActiveApps(runtimeApps map[uint32]AppStatusInfo, pid uint32) {
  740. if c == nil {
  741. //klog.WithField("pid", pid).Warningln("[BuildActiveApps] container_apm is nil.")
  742. return
  743. }
  744. if c.AppInfo.AppName == "" {
  745. return
  746. }
  747. klog.WithField("pid", pid).WithField("appname", c.AppInfo.AppName).Infof("[BuildActiveApps] container %s is running.", c.AppInfo.AppName)
  748. detail := AppStatusInfo{
  749. Pid: pid,
  750. ProcName: c.containerName,
  751. AppName: c.AppInfo.AppName,
  752. Language: c.AppInfo.CodeType.String(),
  753. AppID: c.AppInfo.AppIdHash.IntVal,
  754. AgentID: c.AppInfo.AgentId,
  755. InstanceID: c.AppInfo.InstanceIdHash.IntVal,
  756. Sn: c.AppInfo.Sn,
  757. Sport: c.AppInfo.Sport,
  758. RegisterAt: time.Unix(c.AppInfo.RegisterAt, 0).Format("060102 15:04:05"),
  759. PreStatus: c.AppInfo.PreStatus,
  760. Status: c.AppInfo.Status,
  761. Rule: c.WhiteSettingInfo.Filters,
  762. Container: string(c.id),
  763. }
  764. detail.Rule = fmt.Sprintf("%s|%d", c.WhiteSettingInfo.Filters, c.WhiteSettingInfo.WhiteStackSettingInfo.OpenStack)
  765. if c.AppInfo.UpdateAt != 0 {
  766. detail.UpdateAt = time.Unix(c.AppInfo.UpdateAt, 0).Format("060102 15:04:05")
  767. }
  768. p := c.processes[pid]
  769. if p != nil {
  770. detail.StackStatus = p.stackStatus.String()
  771. v := 0
  772. if !p.versionFailed {
  773. v = 1
  774. }
  775. detail.StackStatus += fmt.Sprintf("V=%d", v)
  776. }
  777. runtimeApps[pid] = detail
  778. }
  779. func (c *Container) AgentCtrl(r *Registry, pid uint32) {
  780. if c == nil {
  781. //klog.WithField("pid", pid).Warningln("[AgentCtrl] cannot find container.")
  782. return
  783. }
  784. var err error
  785. verifyAttachConditions, _ := c.verifyAttachConditions(r, pid)
  786. // fusing UNINSTALL
  787. if r.isFusing && c.Isl7AttachSuccess() {
  788. c.Detach(r.tracer, pid, APP_FUSE)
  789. klog.WithField("pid", pid).Infoln("[AgentCtrl] fusing")
  790. return
  791. }
  792. // verify UNINSTALL
  793. if !verifyAttachConditions && c.Isl7AttachSuccess() {
  794. c.Detach(r.tracer, pid, APP_UNINSTALL)
  795. klog.WithField("pid", pid).Infoln("[AgentCtrl] rule uninstall.")
  796. return
  797. }
  798. if verifyAttachConditions {
  799. err = c.RegisterAppInfo(r, pid)
  800. if err != nil {
  801. klog.WithError(err).Errorf("[AgentCtrl] Failed registerAppInfo.")
  802. return
  803. }
  804. klog.WithField("pid", pid).Infoln("[AgentCtrl] Attach uprobes.")
  805. err = c.AttachUprobes(r.tracer, pid, "Agentctrl")
  806. if err != nil {
  807. klog.WithField("pid", pid).WithError(err).Errorf("[AgentCtrl] Failed attach uprobes error!")
  808. return
  809. } else {
  810. klog.WithField("pid", pid).Infoln("[AgentCtrl] Attach uprobes success!")
  811. }
  812. // 堆栈控制
  813. c.ctrlStack(r, pid)
  814. }
  815. }