Line data Source code
1 : /*
2 : * Copyright (c) 2013 Juniper Networks, Inc. All rights reserved.
3 : */
4 :
5 : #include "bgp/peer_close_manager.h"
6 :
7 :
8 : #include <list>
9 : #include <map>
10 :
11 : #include <boost/foreach.hpp>
12 :
13 : #include "base/task_annotations.h"
14 : #include "bgp/bgp_log.h"
15 : #include "bgp/bgp_membership.h"
16 : #include "bgp/bgp_peer_types.h"
17 : #include "bgp/bgp_route.h"
18 : #include "bgp/bgp_server.h"
19 : #include "bgp/routing-instance/routing_instance.h"
20 : #include "net/community_type.h"
21 :
22 : #define PEER_CLOSE_MANAGER_LOG(msg) \
23 : BGP_LOG_PEER(Event, peer_close_->peer(), SandeshLevel::SYS_INFO, \
24 : BGP_LOG_FLAG_ALL, BGP_PEER_DIR_NA, \
25 : "PeerCloseManager: State " << GetStateName(state_) << \
26 : ", MembershipState: " << GetMembershipStateName(membership_state_) << \
27 : ", MembershipReqPending: " << membership_req_pending_ << \
28 : ", CloseAgain?: " << (close_again_ ? "Yes" : "No") << ": " << msg);
29 :
30 : #define PEER_CLOSE_MANAGER_TABLE_LOG(msg) \
31 : BGP_LOG_PEER_TABLE(peer_close_->peer(), SandeshLevel::SYS_INFO, \
32 : BGP_LOG_FLAG_ALL, table, \
33 : "PeerCloseManager: State " << GetStateName(state_) << \
34 : ", MembershipState: " << GetMembershipStateName(membership_state_) << \
35 : ", MembershipReqPending: " << membership_req_pending_ << \
36 : ", CloseAgain?: " << (close_again_ ? "Yes" : "No") << ": " << msg);
37 :
38 : #define MOVE_TO_STATE(state) \
39 : do { \
40 : assert(state_ != state); \
41 : PEER_CLOSE_MANAGER_LOG("Move to state " << GetStateName(state)); \
42 : state_ = state; \
43 : } while (false)
44 :
45 : // Create an instance of PeerCloseManager with back reference to parent IPeer
46 8064 : PeerCloseManager::PeerCloseManager(IPeerClose *peer_close,
47 8064 : boost::asio::io_context *io_service) :
48 8064 : peer_close_(peer_close), gr_timer_(NULL),
49 16128 : event_queue_(new WorkQueue<Event *>(
50 24192 : TaskScheduler::GetInstance()->GetTaskId(
51 8064 : peer_close_->GetTaskName()),
52 8064 : peer_close_->GetTaskInstance(),
53 8064 : boost::bind(&PeerCloseManager::EventCallback, this, _1))),
54 8064 : state_(NONE), close_again_(false), graceful_(true), gr_elapsed_(0),
55 16128 : llgr_elapsed_(0), membership_state_(MEMBERSHIP_NONE) {
56 8064 : stats_.init++;
57 8064 : membership_req_pending_ = 0;
58 8064 : gr_timer_ = TimerManager::CreateTimer(*io_service,
59 : "Graceful Restart Timer");
60 8064 : }
61 :
62 : // Create an instance of PeerCloseManager with back reference to parent IPeer
63 20413 : PeerCloseManager::PeerCloseManager(IPeerClose *peer_close) :
64 20413 : peer_close_(peer_close), gr_timer_(NULL),
65 40826 : event_queue_(new WorkQueue<Event *>(
66 61239 : TaskScheduler::GetInstance()->GetTaskId(
67 20413 : peer_close_->GetTaskName()),
68 20413 : peer_close_->GetTaskInstance(),
69 20413 : boost::bind(&PeerCloseManager::EventCallback, this, _1))),
70 20413 : state_(NONE), close_again_(false), graceful_(true), gr_elapsed_(0),
71 40826 : llgr_elapsed_(0), membership_state_(MEMBERSHIP_NONE) {
72 20413 : stats_.init++;
73 20413 : membership_req_pending_ = 0;
74 20413 : if (peer_close->peer() && peer_close->peer()->server()) {
75 16598 : gr_timer_ =
76 16598 : TimerManager::CreateTimer(*peer_close->peer()->server()->ioservice(),
77 : "Graceful Restart Timer");
78 : }
79 20413 : }
80 :
81 45865 : PeerCloseManager::~PeerCloseManager() {
82 28477 : event_queue_->Shutdown();
83 28477 : TimerManager::DeleteTimer(gr_timer_);
84 45865 : }
85 :
86 228607 : std::string PeerCloseManager::GetStateName(State state) const {
87 228607 : switch (state) {
88 32784 : case NONE:
89 32784 : return "NONE";
90 68739 : case GR_TIMER:
91 68739 : return "GR_TIMER";
92 20143 : case STALE:
93 20143 : return "STALE";
94 6102 : case LLGR_STALE:
95 6102 : return "LLGR_STALE";
96 7687 : case LLGR_TIMER:
97 7687 : return "LLGR_TIMER";
98 10913 : case SWEEP:
99 10913 : return "SWEEP";
100 82239 : case DELETE:
101 82239 : return "DELETE";
102 : }
103 0 : assert(false);
104 : return "";
105 : }
106 :
107 206438 : std::string PeerCloseManager::GetMembershipStateName(
108 : MembershipState state) const {
109 206438 : switch (state) {
110 93118 : case MEMBERSHIP_NONE:
111 93118 : return "NONE";
112 109296 : case MEMBERSHIP_IN_USE:
113 109296 : return "IN_USE";
114 4025 : case MEMBERSHIP_IN_WAIT:
115 4025 : return "IN_WAIT";
116 : }
117 0 : assert(false);
118 : return "";
119 : }
120 :
121 54141 : std::string PeerCloseManager::GetEventName(EventType eventType) const {
122 54141 : switch (eventType) {
123 0 : case EVENT_NONE:
124 0 : return "NONE";
125 10237 : case CLOSE:
126 10237 : return "CLOSE";
127 5626 : case EOR_RECEIVED:
128 5626 : return "EOR_RECEIVED";
129 9671 : case MEMBERSHIP_REQUEST:
130 9671 : return "MEMBERSHIP_REQUEST";
131 25896 : case MEMBERSHIP_REQUEST_COMPLETE_CALLBACK:
132 25896 : return "MEMBERSHIP_REQUEST_COMPLETE_CALLBACK";
133 2711 : case TIMER_CALLBACK:
134 2711 : return "TIMER_CALLBACK";
135 : }
136 :
137 0 : return "";
138 : }
139 :
140 143980 : void PeerCloseManager::EnqueueEvent(Event *event) {
141 164564 : PEER_CLOSE_MANAGER_LOG("Enqueued event " <<
142 : GetEventName(event->event_type) <<
143 : ", graceful " << event->graceful <<
144 : ", family " << Address::FamilyToString(event->family));
145 143998 : event_queue_->Enqueue(event);
146 144116 : }
147 :
148 : // Trigger closure of an IPeer
149 : //
150 : // Graceful close_state_: NONE
151 : // RibIn Stale Marking and Ribout deletion close_state_: STALE
152 : // StateMachine restart and GR timer start close_state_: GR_TIMER
153 : //
154 : // Peer IsReady() in GR timer callback (or via reception of all EoRs)
155 : // RibIn Sweep and Ribout Generation close_state_: SWEEP
156 : // MembershipRequestCallback close_state_: NONE
157 : //
158 : // Peer not IsReady() in GR timer callback
159 : // If LLGR supported close_state_: LLGR_STALE
160 : // RibIn Stale marking with LLGR_STALE community close_state_: LLGR_TIMER
161 : //
162 : // Peer not IsReady() in LLGR timer callback
163 : // RibIn Delete close_state_: DELETE
164 : // MembershipRequestCallback close_state_: NONE
165 : //
166 : // Peer IsReady() in LLGR timer callback (or via reception of all EoRs)
167 : // RibIn Sweep close_state_: SWEEP
168 : // MembershipRequestCallback close_state_: NONE
169 : //
170 : // If LLGR is not supported
171 : // RibIn Delete close_state_: DELETE
172 : // MembershipRequestCallback close_state_: NONE
173 : //
174 : // Close() call during any state other than NONE and DELETE
175 : // Cancel GR timer and restart GR Closure all over again
176 : //
177 : // NonGraceful close_state_ = * (except DELETE)
178 : // A. RibIn deletion and Ribout deletion close_state_ = DELETE
179 : // B. MembershipRequestCallback => Peers delete/StateMachine restart
180 : // close_state_ = NONE
181 : //
182 : // If Close is restarted, account for GR timer's elapsed time.
183 : //
184 : // Use graceful as true to close gracefully.
185 21035 : void PeerCloseManager::Close(bool graceful) {
186 21035 : EnqueueEvent(new Event(CLOSE, graceful));
187 21048 : }
188 :
189 21048 : void PeerCloseManager::Close(Event *event) {
190 : // Note down non-graceful close trigger. Once non-graceful closure is
191 : // triggered, it should remain so until close process is complete. Further
192 : // graceful closure calls until then should remain non-graceful.
193 21048 : graceful_ &= event->graceful;
194 21048 : CloseInternal();
195 21048 : }
196 :
197 21584 : void PeerCloseManager::CloseInternal() {
198 21584 : stats_.close++;
199 :
200 : // Ignore nested closures
201 21584 : if (close_again_) {
202 4 : PEER_CLOSE_MANAGER_LOG("Nested close calls ignored");
203 4 : return;
204 : }
205 :
206 21580 : switch (state_) {
207 20029 : case NONE:
208 20029 : stats_.ResetRouteStats();
209 20029 : ProcessClosure();
210 20029 : break;
211 :
212 340 : case GR_TIMER:
213 466 : PEER_CLOSE_MANAGER_LOG("Nested close: Restart GR");
214 340 : close_again_ = true;
215 340 : stats_.nested++;
216 340 : gr_elapsed_ += gr_timer_->GetElapsedTime();
217 340 : CloseComplete();
218 340 : break;
219 :
220 196 : case LLGR_TIMER:
221 196 : PEER_CLOSE_MANAGER_LOG("Nested close: Restart LLGR");
222 196 : close_again_ = true;
223 196 : stats_.nested++;
224 196 : llgr_elapsed_ += gr_timer_->GetElapsedTime();
225 196 : CloseComplete();
226 196 : break;
227 :
228 1015 : case STALE:
229 : case LLGR_STALE:
230 : case SWEEP:
231 : case DELETE:
232 1015 : PEER_CLOSE_MANAGER_LOG("Nested close");
233 1015 : close_again_ = true;
234 1015 : stats_.nested++;
235 1015 : break;
236 : }
237 : }
238 :
239 15148 : void PeerCloseManager::ProcessEORMarkerReceived(Address::Family family) {
240 15148 : EnqueueEvent(new Event(EOR_RECEIVED, family));
241 15149 : }
242 :
243 15150 : void PeerCloseManager::ProcessEORMarkerReceived(Event *event) {
244 15150 : if ((state_ == GR_TIMER || state_ == LLGR_TIMER) && !families_.empty()) {
245 1911 : if (event->family == Address::UNSPEC) {
246 40 : families_.clear();
247 : } else {
248 1871 : families_.erase(event->family);
249 : }
250 :
251 : // Start the timer if all EORs have been received.
252 1911 : if (families_.empty())
253 1289 : StartRestartTimer(0);
254 : }
255 15150 : }
256 :
257 : // Process RibIn during peer closure.
258 2810 : void PeerCloseManager::StartRestartTimer(int time) {
259 2810 : gr_timer_->Cancel();
260 3074 : PEER_CLOSE_MANAGER_LOG("GR Timer started to fire after " << time/1000 <<
261 : " seconds");
262 2810 : gr_timer_->Start(time,
263 : boost::bind(&PeerCloseManager::RestartTimerCallback, this));
264 2810 : }
265 :
266 2713 : bool PeerCloseManager::RestartTimerCallback() {
267 2713 : CHECK_CONCURRENCY("timer::TimerTask");
268 2715 : EnqueueEvent(new Event(TIMER_CALLBACK));
269 2715 : return false;
270 : }
271 :
272 2715 : void PeerCloseManager::RestartTimerCallback(Event *event) {
273 2715 : CHECK_CONCURRENCY(peer_close_->GetTaskName());
274 :
275 2825 : PEER_CLOSE_MANAGER_LOG("GR Timer callback started");
276 2715 : if (state_ != GR_TIMER && state_ != LLGR_TIMER)
277 960 : return;
278 :
279 1755 : if (peer_close_->IsReady() && !families_.empty()) {
280 : // Fake reception of all EORs.
281 204 : for (IPeerClose::Families::iterator i = families_.begin(), next = i;
282 408 : i != families_.end(); i = next) {
283 204 : next++;
284 216 : PEER_CLOSE_MANAGER_LOG("Simulate EoR reception for family " << *i);
285 204 : peer_close_->ReceiveEndOfRIB(*i);
286 : }
287 : } else {
288 1551 : ProcessClosure();
289 : }
290 : }
291 :
292 : // Route stale timer callback. If the peer has come back up, sweep routes for
293 : // those address families that are still active. Delete the rest
294 21580 : void PeerCloseManager::ProcessClosure() {
295 : // If the peer is back up and this address family is still supported,
296 : // sweep old paths which may not have come back in the new session
297 21580 : switch (state_) {
298 20029 : case NONE:
299 20029 : if (!graceful_ || !peer_close_->IsCloseGraceful()) {
300 19737 : MOVE_TO_STATE(DELETE);
301 18937 : stats_.deletes++;
302 : } else {
303 1287 : MOVE_TO_STATE(STALE);
304 1092 : stats_.stale++;
305 1092 : StaleNotify();
306 1092 : return;
307 : }
308 18937 : break;
309 1092 : case GR_TIMER:
310 1092 : if (peer_close_->IsReady()) {
311 768 : MOVE_TO_STATE(SWEEP);
312 728 : gr_elapsed_ = 0;
313 728 : llgr_elapsed_ = 0;
314 728 : stats_.sweep++;
315 728 : break;
316 : }
317 364 : if (peer_close_->IsCloseLongLivedGraceful()) {
318 291 : MOVE_TO_STATE(LLGR_STALE);
319 262 : stats_.llgr_stale++;
320 262 : peer_close_->LongLivedGracefulRestartStale();
321 262 : break;
322 : }
323 102 : MOVE_TO_STATE(DELETE);
324 102 : stats_.deletes++;
325 102 : break;
326 :
327 459 : case LLGR_TIMER:
328 459 : if (peer_close_->IsReady()) {
329 262 : MOVE_TO_STATE(SWEEP);
330 262 : gr_elapsed_ = 0;
331 262 : llgr_elapsed_ = 0;
332 262 : stats_.sweep++;
333 262 : break;
334 : }
335 226 : MOVE_TO_STATE(DELETE);
336 197 : stats_.deletes++;
337 197 : break;
338 :
339 0 : case STALE:
340 : case LLGR_STALE:
341 : case SWEEP:
342 : case DELETE:
343 0 : assert(false);
344 : return;
345 : }
346 :
347 20488 : if (state_ == DELETE)
348 19236 : peer_close_->CustomClose();
349 20488 : MembershipRequest();
350 : }
351 :
352 1439 : void PeerCloseManager::CloseComplete() {
353 1605 : MOVE_TO_STATE(NONE);
354 1439 : gr_timer_->Cancel();
355 1439 : families_.clear();
356 1439 : stats_.init++;
357 :
358 : // Nested closures trigger fresh GR
359 1439 : if (close_again_) {
360 536 : close_again_ = false;
361 536 : CloseInternal();
362 : }
363 1439 : }
364 :
365 1479 : bool PeerCloseManager::AssertSweepState(bool do_assert) {
366 1479 : bool check = (state_ == SWEEP);
367 1479 : if (do_assert)
368 711 : assert(check);
369 1479 : return check;
370 : }
371 :
372 95428 : bool PeerCloseManager::AssertMembershipManagerInUse(bool do_assert) {
373 95428 : bool check = false;
374 181951 : check |= (state_ == STALE || state_ == LLGR_STALE || state_ == SWEEP ||
375 86523 : state_ == DELETE);
376 95428 : check |= (membership_state_ == MEMBERSHIP_IN_USE);
377 95428 : check |= (membership_req_pending_ > 0);
378 95428 : if (do_assert)
379 94084 : assert(check);
380 95428 : return check;
381 : }
382 :
383 22988 : bool PeerCloseManager::AssertMembershipState(bool do_assert) {
384 22988 : bool check = (membership_state_ != MEMBERSHIP_IN_USE);
385 22988 : if (do_assert)
386 20502 : assert(check);
387 22988 : return check;
388 : }
389 :
390 21283 : bool PeerCloseManager::AssertMembershipReqCount(bool do_assert) {
391 21283 : bool check = !membership_req_pending_;
392 21283 : if (do_assert)
393 20438 : assert(check);
394 21283 : return check;
395 : }
396 :
397 1479 : void PeerCloseManager::TriggerSweepStateActions() {
398 1479 : CHECK_CONCURRENCY(peer_close_->GetTaskName());
399 1479 : if (!AssertSweepState())
400 576 : return;
401 :
402 : // Notify clients to trigger sweep as appropriate.
403 903 : peer_close_->GracefulRestartSweep();
404 :
405 : // Reset MembershipUse state after client has been notified above.
406 903 : set_membership_state(MEMBERSHIP_NONE);
407 903 : CloseComplete();
408 : }
409 :
410 : // Notify clients about entering Stale event.
411 1092 : void PeerCloseManager::StaleNotify() {
412 1092 : CHECK_CONCURRENCY(peer_close_->GetTaskName());
413 :
414 1092 : peer_close_->GracefulRestartStale();
415 1092 : if (!AssertMembershipState())
416 48 : return;
417 1044 : MembershipRequest(NULL);
418 : }
419 :
420 19554 : bool PeerCloseManager::CanUseMembershipManager() const {
421 19554 : return peer_close_->peer()->CanUseMembershipManager();
422 : }
423 :
424 20438 : void PeerCloseManager::GetRegisteredRibs(std::list<BgpTable *> *tables) {
425 20438 : BgpMembershipManager *mgr = peer_close_->peer()->server()->membership_mgr();
426 20438 : mgr->GetRegisteredRibs(peer_close_->peer(), tables);
427 20438 : }
428 :
429 83633 : bool PeerCloseManager::IsRegistered(BgpTable *table) const {
430 83633 : BgpMembershipManager *mgr = peer_close_->peer()->server()->membership_mgr();
431 83633 : return mgr->IsRegistered(peer_close_->peer(), table);
432 : }
433 :
434 73491 : void PeerCloseManager::Unregister(BgpTable *table) {
435 73491 : BgpMembershipManager *mgr = peer_close_->peer()->server()->membership_mgr();
436 73491 : mgr->Unregister(peer_close_->peer(), table);
437 73491 : }
438 :
439 3889 : void PeerCloseManager::WalkRibIn(BgpTable *table) {
440 3889 : BgpMembershipManager *mgr = peer_close_->peer()->server()->membership_mgr();
441 3889 : mgr->WalkRibIn(peer_close_->peer(), table);
442 3889 : }
443 :
444 4440 : void PeerCloseManager::UnregisterRibOut(BgpTable *table) {
445 4440 : BgpMembershipManager *mgr = peer_close_->peer()->server()->membership_mgr();
446 4440 : mgr->UnregisterRibOut(peer_close_->peer(), table);
447 4440 : }
448 :
449 4031 : bool PeerCloseManager::IsRibInRegistered(BgpTable *table) const {
450 4031 : BgpMembershipManager *mgr = peer_close_->peer()->server()->membership_mgr();
451 4031 : return mgr->IsRibInRegistered(peer_close_->peer(), table);
452 : }
453 :
454 1813 : void PeerCloseManager::UnregisterRibIn(BgpTable *table) {
455 1813 : BgpMembershipManager *mgr = peer_close_->peer()->server()->membership_mgr();
456 1813 : mgr->UnregisterRibIn(peer_close_->peer(), table);
457 1813 : }
458 :
459 21896 : void PeerCloseManager::MembershipRequest() {
460 21896 : if (!AssertMembershipState())
461 848 : return;
462 :
463 : // Pause if membership manager is not ready for usage.
464 21048 : if (!CanUseMembershipManager()) {
465 809 : set_membership_state(MEMBERSHIP_IN_WAIT);
466 844 : PEER_CLOSE_MANAGER_LOG("Wait for membership manager availability");
467 809 : return;
468 : }
469 20239 : set_membership_state(MEMBERSHIP_IN_USE);
470 20239 : EnqueueEvent(new Event(MEMBERSHIP_REQUEST));
471 : }
472 :
473 21283 : void PeerCloseManager::MembershipRequest(Event *evnet) {
474 21283 : CHECK_CONCURRENCY(peer_close_->GetTaskName());
475 :
476 21283 : set_membership_state(MEMBERSHIP_IN_USE);
477 21283 : if (!AssertMembershipReqCount())
478 10899 : return;
479 20835 : membership_req_pending_++;
480 20835 : std::list<BgpTable *> tables;
481 20835 : GetRegisteredRibs(&tables);
482 :
483 20835 : if (tables.empty()) {
484 10451 : assert(MembershipRequestCallback(NULL));
485 10451 : return;
486 : }
487 :
488 : // Account for extra increment above.
489 10384 : membership_req_pending_--;
490 178444 : BOOST_FOREACH(BgpTable *table, tables) {
491 84030 : membership_req_pending_++;
492 84030 : if (IsRegistered(table)) {
493 79803 : if (state_ == PeerCloseManager::DELETE) {
494 100256 : PEER_CLOSE_MANAGER_TABLE_LOG(
495 : "MembershipManager::Unregister");
496 73594 : Unregister(table);
497 6209 : } else if (state_ == PeerCloseManager::SWEEP) {
498 3797 : PEER_CLOSE_MANAGER_TABLE_LOG("MembershipManager::WalkRibIn");
499 1710 : WalkRibIn(table);
500 : } else {
501 11812 : PEER_CLOSE_MANAGER_TABLE_LOG(
502 : "MembershipManager::UnregisterRibOut");
503 4499 : UnregisterRibOut(table);
504 : }
505 : } else {
506 4227 : assert(IsRibInRegistered(table));
507 4227 : if (state_ == PeerCloseManager::DELETE) {
508 5024 : PEER_CLOSE_MANAGER_TABLE_LOG(
509 : "MembershipManager::UnregisterRibIn");
510 1913 : UnregisterRibIn(table);
511 : } else {
512 6118 : PEER_CLOSE_MANAGER_TABLE_LOG("MembershipManager::WalkRibIn");
513 2314 : WalkRibIn(table);
514 : }
515 : }
516 : }
517 20835 : }
518 :
519 84852 : void PeerCloseManager::MembershipRequestCallback() {
520 84852 : EnqueueEvent(new Event(MEMBERSHIP_REQUEST_COMPLETE_CALLBACK));
521 84964 : }
522 :
523 : // Close process for this peer in terms of walking RibIns and RibOuts are
524 : // complete. Do the final cleanups necessary and notify interested party
525 : //
526 : // Retrun true if we are done using membership manager, false otherwise.
527 95428 : bool PeerCloseManager::MembershipRequestCallback(Event *event) {
528 95428 : CHECK_CONCURRENCY(peer_close_->GetTaskName());
529 :
530 95428 : bool result = false;
531 114038 : PEER_CLOSE_MANAGER_LOG("MembershipRequestCallback");
532 :
533 95428 : if (!AssertMembershipManagerInUse())
534 0 : return result;
535 95428 : if (--membership_req_pending_)
536 73646 : return result;
537 :
538 : // Indicate to the caller that we are done using the membership manager.
539 21782 : result = true;
540 :
541 21782 : if (state_ == DELETE) {
542 19611 : MOVE_TO_STATE(NONE);
543 18782 : peer_close_->Delete();
544 18782 : gr_elapsed_ = 0;
545 18782 : llgr_elapsed_ = 0;
546 18782 : stats_.init++;
547 18782 : close_again_ = false;
548 18782 : graceful_ = true;
549 18782 : set_membership_state(MEMBERSHIP_NONE);
550 18782 : return result;
551 : }
552 :
553 : // Process nested closures.
554 3000 : if (close_again_) {
555 0 : set_membership_state(MEMBERSHIP_NONE);
556 0 : CloseComplete();
557 :
558 : // Nested closure can make membership manager to be in use again.
559 0 : return membership_state_ != MEMBERSHIP_IN_USE;
560 : }
561 :
562 : // If any GR stale timer has to be launched, then to wait for some time
563 : // hoping for the peer (and the paths) to come back up.
564 3000 : if (state_ == STALE) {
565 1140 : peer_close_->CloseComplete();
566 1335 : MOVE_TO_STATE(GR_TIMER);
567 1140 : peer_close_->GetGracefulRestartFamilies(&families_);
568 :
569 : // Offset restart time with elapsed time during nested closures.
570 1140 : int time = peer_close_->GetGracefulRestartTime() * 1000;
571 1140 : time -= gr_elapsed_;
572 1140 : if (time < 0)
573 0 : time = 0;
574 1140 : StartRestartTimer(time);
575 1140 : stats_.gr_timer++;
576 1140 : set_membership_state(MEMBERSHIP_NONE);
577 1140 : return result;
578 : }
579 :
580 : // From LLGR_STALE state, switch to LLGR_TIMER state. Typically this would
581 : // be a very long timer, and we expect to receive EORs before this timer
582 : // expires.
583 : // Just like for GRSTALE in the above case, before moving to LLGR_TIMER and
584 : // starting LLGR_TIMER, we can check if the peer can come back up.
585 1860 : if (state_ == LLGR_STALE) {
586 410 : MOVE_TO_STATE(LLGR_TIMER);
587 381 : peer_close_->CloseComplete();
588 381 : peer_close_->GetLongLivedGracefulRestartFamilies(&families_);
589 :
590 : // Offset restart time with elapsed time during nested closures.
591 381 : int time = peer_close_->GetLongLivedGracefulRestartTime() * 1000;
592 381 : time -= llgr_elapsed_;
593 381 : if (time < 0)
594 0 : time = 0;
595 381 : StartRestartTimer(time);
596 381 : stats_.llgr_timer++;
597 381 : set_membership_state(MEMBERSHIP_NONE);
598 381 : return result;
599 : }
600 :
601 1479 : TriggerSweepStateActions();
602 1479 : return result;
603 : }
604 :
605 64924 : void PeerCloseManager::FillRouteCloseInfo(PeerCloseInfo *close_info) const {
606 64924 : std::map<std::string, PeerCloseRouteInfo> route_stats;
607 :
608 714164 : for (int i = 0; i < Address::NUM_FAMILIES; i++) {
609 649240 : if (!stats_.route_stats[i].IsSet())
610 489637 : continue;
611 159603 : PeerCloseRouteInfo route_info;
612 159603 : route_info.set_staled(stats_.route_stats[i].staled);
613 159603 : route_info.set_llgr_staled(stats_.route_stats[i].llgr_staled);
614 159603 : route_info.set_refreshed(stats_.route_stats[i].refreshed);
615 159603 : route_info.set_fresh(stats_.route_stats[i].fresh);
616 159603 : route_info.set_deleted(stats_.route_stats[i].deleted);
617 159603 : route_stats[Address::FamilyToString(static_cast<Address::Family>(i))] = route_info;
618 159603 : }
619 :
620 64924 : if (!route_stats.empty())
621 63995 : close_info->set_route_stats(route_stats);
622 64924 : }
623 :
624 64924 : BgpNeighborResp *PeerCloseManager::FillCloseInfo(BgpNeighborResp *resp) const {
625 64924 : PeerCloseInfo peer_close_info;
626 64924 : peer_close_info.set_state(GetStateName(state_));
627 64924 : peer_close_info.set_membership_state(
628 129848 : GetMembershipStateName(membership_state_));
629 64924 : peer_close_info.set_close_again(close_again_);
630 64924 : peer_close_info.set_graceful(graceful_);
631 64924 : peer_close_info.set_init(stats_.init);
632 64924 : peer_close_info.set_close(stats_.close);
633 64924 : peer_close_info.set_nested(stats_.nested);
634 64924 : peer_close_info.set_deletes(stats_.deletes);
635 64924 : peer_close_info.set_stale(stats_.stale);
636 64924 : peer_close_info.set_llgr_stale(stats_.llgr_stale);
637 64924 : peer_close_info.set_sweep(stats_.sweep);
638 64924 : peer_close_info.set_gr_timer(stats_.gr_timer);
639 64924 : peer_close_info.set_llgr_timer(stats_.llgr_timer);
640 64924 : FillRouteCloseInfo(&peer_close_info);
641 :
642 64924 : resp->set_peer_close_info(peer_close_info);
643 :
644 64924 : return resp;
645 64924 : }
646 :
647 216932 : void PeerCloseManager::UpdateRouteStats(Address::Family family,
648 : const BgpPath *old_path, uint32_t path_flags) const {
649 216932 : if (state_ == NONE)
650 196793 : return;
651 :
652 20139 : if (!old_path)
653 88 : stats_.route_stats[family].fresh++;
654 20051 : else if (old_path->IsStale() && !(path_flags & BgpPath::Stale))
655 3519 : stats_.route_stats[family].refreshed++;
656 : }
657 :
658 100272 : bool PeerCloseManager::MembershipPathCallback(DBTablePartBase *root,
659 : BgpRoute *rt, BgpPath *path) {
660 100272 : CHECK_CONCURRENCY("db::DBTable");
661 100276 : DBRequest::DBOperation oper = DBRequest::DB_ENTRY_INVALID;
662 100276 : BgpAttrPtr attrs;
663 :
664 100276 : BgpTable *table = static_cast<BgpTable *>(root->parent());
665 100274 : assert(table);
666 :
667 100274 : uint32_t stale = 0;
668 :
669 100274 : switch (state_) {
670 0 : case NONE:
671 : case GR_TIMER:
672 : case LLGR_TIMER:
673 0 : return false;
674 :
675 6181 : case SWEEP:
676 :
677 : // Stale paths must be deleted.
678 6181 : if (!path->IsStale() && !path->IsLlgrStale())
679 3084 : return false;
680 3097 : if (path->IsStale()) {
681 3097 : path->ResetStale();
682 3098 : table->UpdateStalePathCount(-1);
683 : }
684 3100 : if (path->IsLlgrStale()) {
685 0 : path->ResetLlgrStale();
686 0 : table->UpdateLlgrStalePathCount(-1);
687 : }
688 3100 : oper = DBRequest::DB_ENTRY_DELETE;
689 3100 : attrs = NULL;
690 3099 : stats_.route_stats[table->family()].deleted++;
691 3101 : break;
692 :
693 75973 : case DELETE:
694 :
695 : // This path must be deleted. Hence attr is not required.
696 75973 : oper = DBRequest::DB_ENTRY_DELETE;
697 75973 : attrs = NULL;
698 75975 : stats_.route_stats[table->family()].deleted++;
699 76008 : break;
700 :
701 14893 : case STALE:
702 :
703 : // We do not support GR for multicast routes (yet).
704 29786 : if ((table->family() == Address::ERMVPN) ||
705 14893 : (table->family() == Address::MVPN)) {
706 0 : oper = DBRequest::DB_ENTRY_DELETE;
707 0 : attrs = NULL;
708 0 : stats_.route_stats[table->family()].deleted++;
709 0 : break;
710 : }
711 :
712 : // If path is already marked as stale, then there is no need to
713 : // process again. This can happen if the session flips while in
714 : // GR_TIMER state.
715 14893 : if (path->IsStale())
716 3342 : return false;
717 :
718 : // This path must be marked for staling. Update the local
719 : // preference and update the route accordingly.
720 11552 : oper = DBRequest::DB_ENTRY_ADD_CHANGE;
721 11552 : attrs = path->GetAttr();
722 11552 : stale = BgpPath::Stale;
723 11552 : stats_.route_stats[table->family()].staled++;
724 11554 : break;
725 :
726 3228 : case LLGR_STALE:
727 :
728 : // If the path has NO_LLGR community, DELETE it.
729 3228 : if (path->GetAttr()->community() &&
730 0 : path->GetAttr()->community()->ContainsValue(
731 : CommunityType::NoLlgr)) {
732 0 : oper = DBRequest::DB_ENTRY_DELETE;
733 0 : attrs = NULL;
734 0 : stats_.route_stats[table->family()].deleted++;
735 0 : break;
736 : }
737 :
738 : // If path is already marked as llgr_stale, then there is no
739 : // need to process again. This can happen if the session flips
740 : // while in LLGR_TIMER state.
741 3231 : if (path->IsLlgrStale())
742 0 : return false;
743 :
744 3230 : attrs = path->GetAttr();
745 3235 : stale = BgpPath::LlgrStale;
746 3235 : oper = DBRequest::DB_ENTRY_ADD_CHANGE;
747 3235 : stats_.route_stats[table->family()].llgr_staled++;
748 3236 : break;
749 : }
750 :
751 : // Feed the route modify/delete request to the table input process.
752 281664 : return table->InputCommon(root, rt, path, peer_close_->peer(), NULL, oper,
753 93884 : attrs, path->GetPathId(), path->GetFlags() | stale, path->GetLabel(),
754 93885 : path->GetL3Label());
755 100308 : }
756 :
757 : //
758 : // Handler for an Event.
759 : //
760 144129 : bool PeerCloseManager::EventCallback(Event *event) {
761 144129 : CHECK_CONCURRENCY(peer_close_->GetTaskName());
762 : bool result;
763 :
764 144129 : switch (event->event_type) {
765 0 : case EVENT_NONE:
766 0 : break;
767 21048 : case CLOSE:
768 21048 : Close(event);
769 21048 : break;
770 15150 : case EOR_RECEIVED:
771 15150 : ProcessEORMarkerReceived(event);
772 15150 : break;
773 20239 : case MEMBERSHIP_REQUEST:
774 20239 : MembershipRequest(event);
775 20239 : break;
776 84977 : case MEMBERSHIP_REQUEST_COMPLETE_CALLBACK:
777 84977 : result = MembershipRequestCallback(event);
778 :
779 : // Notify clients if we are no longer using the membership mgr.
780 84977 : if (result)
781 11331 : peer_close_->MembershipRequestCallbackComplete();
782 84977 : break;
783 2715 : case TIMER_CALLBACK:
784 2715 : RestartTimerCallback(event);
785 2715 : break;
786 : }
787 :
788 144129 : delete event;
789 144129 : return true;
790 : }
|
