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 20403 : PeerCloseManager::PeerCloseManager(IPeerClose *peer_close) :
64 20403 : peer_close_(peer_close), gr_timer_(NULL),
65 40806 : event_queue_(new WorkQueue<Event *>(
66 61209 : TaskScheduler::GetInstance()->GetTaskId(
67 20403 : peer_close_->GetTaskName()),
68 20403 : peer_close_->GetTaskInstance(),
69 20403 : boost::bind(&PeerCloseManager::EventCallback, this, _1))),
70 20403 : state_(NONE), close_again_(false), graceful_(true), gr_elapsed_(0),
71 40806 : llgr_elapsed_(0), membership_state_(MEMBERSHIP_NONE) {
72 20403 : stats_.init++;
73 20403 : membership_req_pending_ = 0;
74 20403 : if (peer_close->peer() && peer_close->peer()->server()) {
75 16588 : gr_timer_ =
76 16588 : TimerManager::CreateTimer(*peer_close->peer()->server()->ioservice(),
77 : "Graceful Restart Timer");
78 : }
79 20403 : }
80 :
81 45856 : PeerCloseManager::~PeerCloseManager() {
82 28467 : event_queue_->Shutdown();
83 28467 : TimerManager::DeleteTimer(gr_timer_);
84 45856 : }
85 :
86 223986 : std::string PeerCloseManager::GetStateName(State state) const {
87 223986 : switch (state) {
88 32238 : case NONE:
89 32238 : return "NONE";
90 67746 : case GR_TIMER:
91 67746 : return "GR_TIMER";
92 19107 : case STALE:
93 19107 : return "STALE";
94 5989 : case LLGR_STALE:
95 5989 : return "LLGR_STALE";
96 6502 : case LLGR_TIMER:
97 6502 : return "LLGR_TIMER";
98 10156 : case SWEEP:
99 10156 : return "SWEEP";
100 82248 : case DELETE:
101 82248 : return "DELETE";
102 : }
103 0 : assert(false);
104 : return "";
105 : }
106 :
107 202296 : std::string PeerCloseManager::GetMembershipStateName(
108 : MembershipState state) const {
109 202296 : switch (state) {
110 90473 : case MEMBERSHIP_NONE:
111 90473 : return "NONE";
112 107483 : case MEMBERSHIP_IN_USE:
113 107483 : return "IN_USE";
114 4340 : case MEMBERSHIP_IN_WAIT:
115 4340 : return "IN_WAIT";
116 : }
117 0 : assert(false);
118 : return "";
119 : }
120 :
121 53350 : std::string PeerCloseManager::GetEventName(EventType eventType) const {
122 53350 : switch (eventType) {
123 0 : case EVENT_NONE:
124 0 : return "NONE";
125 10066 : case CLOSE:
126 10066 : return "CLOSE";
127 5435 : case EOR_RECEIVED:
128 5435 : return "EOR_RECEIVED";
129 9592 : case MEMBERSHIP_REQUEST:
130 9592 : return "MEMBERSHIP_REQUEST";
131 25529 : case MEMBERSHIP_REQUEST_COMPLETE_CALLBACK:
132 25529 : return "MEMBERSHIP_REQUEST_COMPLETE_CALLBACK";
133 2728 : case TIMER_CALLBACK:
134 2728 : return "TIMER_CALLBACK";
135 : }
136 :
137 0 : return "";
138 : }
139 :
140 142809 : void PeerCloseManager::EnqueueEvent(Event *event) {
141 163335 : PEER_CLOSE_MANAGER_LOG("Enqueued event " <<
142 : GetEventName(event->event_type) <<
143 : ", graceful " << event->graceful <<
144 : ", family " << Address::FamilyToString(event->family));
145 142829 : event_queue_->Enqueue(event);
146 143022 : }
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 20840 : void PeerCloseManager::Close(bool graceful) {
186 20840 : EnqueueEvent(new Event(CLOSE, graceful));
187 20848 : }
188 :
189 20848 : 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 20848 : graceful_ &= event->graceful;
194 20848 : CloseInternal();
195 20848 : }
196 :
197 21381 : void PeerCloseManager::CloseInternal() {
198 21381 : stats_.close++;
199 :
200 : // Ignore nested closures
201 21381 : if (close_again_) {
202 4 : PEER_CLOSE_MANAGER_LOG("Nested close calls ignored");
203 4 : return;
204 : }
205 :
206 21377 : switch (state_) {
207 19824 : case NONE:
208 19824 : stats_.ResetRouteStats();
209 19824 : ProcessClosure();
210 19824 : break;
211 :
212 337 : case GR_TIMER:
213 463 : PEER_CLOSE_MANAGER_LOG("Nested close: Restart GR");
214 337 : close_again_ = true;
215 337 : stats_.nested++;
216 337 : gr_elapsed_ += gr_timer_->GetElapsedTime();
217 337 : CloseComplete();
218 337 : 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 1020 : case STALE:
229 : case LLGR_STALE:
230 : case SWEEP:
231 : case DELETE:
232 1020 : PEER_CLOSE_MANAGER_LOG("Nested close");
233 1020 : close_again_ = true;
234 1020 : stats_.nested++;
235 1020 : break;
236 : }
237 : }
238 :
239 14721 : void PeerCloseManager::ProcessEORMarkerReceived(Address::Family family) {
240 14721 : EnqueueEvent(new Event(EOR_RECEIVED, family));
241 14725 : }
242 :
243 14725 : void PeerCloseManager::ProcessEORMarkerReceived(Event *event) {
244 14725 : if ((state_ == GR_TIMER || state_ == LLGR_TIMER) && !families_.empty()) {
245 1745 : if (event->family == Address::UNSPEC) {
246 40 : families_.clear();
247 : } else {
248 1705 : families_.erase(event->family);
249 : }
250 :
251 : // Start the timer if all EORs have been received.
252 1745 : if (families_.empty())
253 1134 : StartRestartTimer(0);
254 : }
255 14725 : }
256 :
257 : // Process RibIn during peer closure.
258 2498 : void PeerCloseManager::StartRestartTimer(int time) {
259 2498 : gr_timer_->Cancel();
260 2762 : PEER_CLOSE_MANAGER_LOG("GR Timer started to fire after " << time/1000 <<
261 : " seconds");
262 2498 : gr_timer_->Start(time,
263 : boost::bind(&PeerCloseManager::RestartTimerCallback, this));
264 2498 : }
265 :
266 2732 : bool PeerCloseManager::RestartTimerCallback() {
267 2732 : CHECK_CONCURRENCY("timer::TimerTask");
268 2731 : EnqueueEvent(new Event(TIMER_CALLBACK));
269 2732 : return false;
270 : }
271 :
272 2732 : void PeerCloseManager::RestartTimerCallback(Event *event) {
273 2732 : CHECK_CONCURRENCY(peer_close_->GetTaskName());
274 :
275 2842 : PEER_CLOSE_MANAGER_LOG("GR Timer callback started");
276 2732 : if (state_ != GR_TIMER && state_ != LLGR_TIMER)
277 960 : return;
278 :
279 1772 : if (peer_close_->IsReady() && !families_.empty()) {
280 : // Fake reception of all EORs.
281 205 : for (IPeerClose::Families::iterator i = families_.begin(), next = i;
282 412 : i != families_.end(); i = next) {
283 207 : next++;
284 219 : PEER_CLOSE_MANAGER_LOG("Simulate EoR reception for family " << *i);
285 207 : peer_close_->ReceiveEndOfRIB(*i);
286 : }
287 : } else {
288 1567 : 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 21391 : 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 21391 : switch (state_) {
298 19824 : case NONE:
299 19824 : if (!graceful_ || !peer_close_->IsCloseGraceful()) {
300 19688 : MOVE_TO_STATE(DELETE);
301 18889 : stats_.deletes++;
302 : } else {
303 1130 : MOVE_TO_STATE(STALE);
304 935 : stats_.stale++;
305 935 : StaleNotify();
306 935 : return;
307 : }
308 18889 : break;
309 1010 : case GR_TIMER:
310 1010 : if (peer_close_->IsReady()) {
311 651 : MOVE_TO_STATE(SWEEP);
312 611 : gr_elapsed_ = 0;
313 611 : llgr_elapsed_ = 0;
314 611 : stats_.sweep++;
315 611 : break;
316 : }
317 399 : if (peer_close_->IsCloseLongLivedGraceful()) {
318 311 : MOVE_TO_STATE(LLGR_STALE);
319 282 : stats_.llgr_stale++;
320 282 : peer_close_->LongLivedGracefulRestartStale();
321 282 : break;
322 : }
323 117 : MOVE_TO_STATE(DELETE);
324 117 : stats_.deletes++;
325 117 : break;
326 :
327 557 : case LLGR_TIMER:
328 557 : if (peer_close_->IsReady()) {
329 316 : MOVE_TO_STATE(SWEEP);
330 316 : gr_elapsed_ = 0;
331 316 : llgr_elapsed_ = 0;
332 316 : stats_.sweep++;
333 316 : break;
334 : }
335 270 : MOVE_TO_STATE(DELETE);
336 241 : stats_.deletes++;
337 241 : 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 20456 : if (state_ == DELETE)
348 19247 : peer_close_->CustomClose();
349 20456 : MembershipRequest();
350 : }
351 :
352 1282 : void PeerCloseManager::CloseComplete() {
353 1448 : MOVE_TO_STATE(NONE);
354 1282 : gr_timer_->Cancel();
355 1282 : families_.clear();
356 1282 : stats_.init++;
357 :
358 : // Nested closures trigger fresh GR
359 1282 : if (close_again_) {
360 533 : close_again_ = false;
361 533 : CloseInternal();
362 : }
363 1282 : }
364 :
365 1325 : bool PeerCloseManager::AssertSweepState(bool do_assert) {
366 1325 : bool check = (state_ == SWEEP);
367 1325 : if (do_assert)
368 557 : assert(check);
369 1325 : return check;
370 : }
371 :
372 95019 : bool PeerCloseManager::AssertMembershipManagerInUse(bool do_assert) {
373 95019 : bool check = false;
374 181468 : check |= (state_ == STALE || state_ == LLGR_STALE || state_ == SWEEP ||
375 86449 : state_ == DELETE);
376 95019 : check |= (membership_state_ == MEMBERSHIP_IN_USE);
377 95019 : check |= (membership_req_pending_ > 0);
378 95019 : if (do_assert)
379 93675 : assert(check);
380 95019 : return check;
381 : }
382 :
383 22787 : bool PeerCloseManager::AssertMembershipState(bool do_assert) {
384 22787 : bool check = (membership_state_ != MEMBERSHIP_IN_USE);
385 22787 : if (do_assert)
386 20131 : assert(check);
387 22787 : return check;
388 : }
389 :
390 21011 : bool PeerCloseManager::AssertMembershipReqCount(bool do_assert) {
391 21011 : bool check = !membership_req_pending_;
392 21011 : if (do_assert)
393 20079 : assert(check);
394 21011 : return check;
395 : }
396 :
397 1325 : void PeerCloseManager::TriggerSweepStateActions() {
398 1325 : CHECK_CONCURRENCY(peer_close_->GetTaskName());
399 1325 : if (!AssertSweepState())
400 576 : return;
401 :
402 : // Notify clients to trigger sweep as appropriate.
403 749 : peer_close_->GracefulRestartSweep();
404 :
405 : // Reset MembershipUse state after client has been notified above.
406 749 : set_membership_state(MEMBERSHIP_NONE);
407 749 : CloseComplete();
408 : }
409 :
410 : // Notify clients about entering Stale event.
411 935 : void PeerCloseManager::StaleNotify() {
412 935 : CHECK_CONCURRENCY(peer_close_->GetTaskName());
413 :
414 935 : peer_close_->GracefulRestartStale();
415 935 : if (!AssertMembershipState())
416 48 : return;
417 887 : MembershipRequest(NULL);
418 : }
419 :
420 19340 : bool PeerCloseManager::CanUseMembershipManager() const {
421 19340 : return peer_close_->peer()->CanUseMembershipManager();
422 : }
423 :
424 20079 : void PeerCloseManager::GetRegisteredRibs(std::list<BgpTable *> *tables) {
425 20079 : BgpMembershipManager *mgr = peer_close_->peer()->server()->membership_mgr();
426 20079 : mgr->GetRegisteredRibs(peer_close_->peer(), tables);
427 20079 : }
428 :
429 83263 : bool PeerCloseManager::IsRegistered(BgpTable *table) const {
430 83263 : BgpMembershipManager *mgr = peer_close_->peer()->server()->membership_mgr();
431 83263 : return mgr->IsRegistered(peer_close_->peer(), table);
432 : }
433 :
434 73475 : void PeerCloseManager::Unregister(BgpTable *table) {
435 73475 : BgpMembershipManager *mgr = peer_close_->peer()->server()->membership_mgr();
436 73475 : mgr->Unregister(peer_close_->peer(), table);
437 73475 : }
438 :
439 3703 : void PeerCloseManager::WalkRibIn(BgpTable *table) {
440 3703 : BgpMembershipManager *mgr = peer_close_->peer()->server()->membership_mgr();
441 3703 : mgr->WalkRibIn(peer_close_->peer(), table);
442 3703 : }
443 :
444 4291 : void PeerCloseManager::UnregisterRibOut(BgpTable *table) {
445 4291 : BgpMembershipManager *mgr = peer_close_->peer()->server()->membership_mgr();
446 4291 : mgr->UnregisterRibOut(peer_close_->peer(), table);
447 4291 : }
448 :
449 3992 : bool PeerCloseManager::IsRibInRegistered(BgpTable *table) const {
450 3992 : BgpMembershipManager *mgr = peer_close_->peer()->server()->membership_mgr();
451 3992 : return mgr->IsRibInRegistered(peer_close_->peer(), table);
452 : }
453 :
454 1794 : void PeerCloseManager::UnregisterRibIn(BgpTable *table) {
455 1794 : BgpMembershipManager *mgr = peer_close_->peer()->server()->membership_mgr();
456 1794 : mgr->UnregisterRibIn(peer_close_->peer(), table);
457 1794 : }
458 :
459 21852 : void PeerCloseManager::MembershipRequest() {
460 21852 : if (!AssertMembershipState())
461 848 : return;
462 :
463 : // Pause if membership manager is not ready for usage.
464 21004 : if (!CanUseMembershipManager()) {
465 880 : set_membership_state(MEMBERSHIP_IN_WAIT);
466 909 : PEER_CLOSE_MANAGER_LOG("Wait for membership manager availability");
467 880 : return;
468 : }
469 20124 : set_membership_state(MEMBERSHIP_IN_USE);
470 20124 : EnqueueEvent(new Event(MEMBERSHIP_REQUEST));
471 : }
472 :
473 21011 : void PeerCloseManager::MembershipRequest(Event *evnet) {
474 21011 : CHECK_CONCURRENCY(peer_close_->GetTaskName());
475 :
476 21011 : set_membership_state(MEMBERSHIP_IN_USE);
477 21011 : if (!AssertMembershipReqCount())
478 10860 : return;
479 20563 : membership_req_pending_++;
480 20563 : std::list<BgpTable *> tables;
481 20563 : GetRegisteredRibs(&tables);
482 :
483 20563 : if (tables.empty()) {
484 10412 : assert(MembershipRequestCallback(NULL));
485 10412 : return;
486 : }
487 :
488 : // Account for extra increment above.
489 10151 : membership_req_pending_--;
490 177645 : BOOST_FOREACH(BgpTable *table, tables) {
491 83747 : membership_req_pending_++;
492 83747 : if (IsRegistered(table)) {
493 79512 : if (state_ == PeerCloseManager::DELETE) {
494 100209 : PEER_CLOSE_MANAGER_TABLE_LOG(
495 : "MembershipManager::Unregister");
496 73593 : Unregister(table);
497 5919 : } else if (state_ == PeerCloseManager::SWEEP) {
498 3487 : PEER_CLOSE_MANAGER_TABLE_LOG("MembershipManager::WalkRibIn");
499 1566 : WalkRibIn(table);
500 : } else {
501 11517 : PEER_CLOSE_MANAGER_TABLE_LOG(
502 : "MembershipManager::UnregisterRibOut");
503 4353 : UnregisterRibOut(table);
504 : }
505 : } else {
506 4235 : assert(IsRibInRegistered(table));
507 4235 : if (state_ == PeerCloseManager::DELETE) {
508 5024 : PEER_CLOSE_MANAGER_TABLE_LOG(
509 : "MembershipManager::UnregisterRibIn");
510 1911 : UnregisterRibIn(table);
511 : } else {
512 6108 : PEER_CLOSE_MANAGER_TABLE_LOG("MembershipManager::WalkRibIn");
513 2324 : WalkRibIn(table);
514 : }
515 : }
516 : }
517 20563 : }
518 :
519 84381 : void PeerCloseManager::MembershipRequestCallback() {
520 84381 : EnqueueEvent(new Event(MEMBERSHIP_REQUEST_COMPLETE_CALLBACK));
521 84592 : }
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 95019 : bool PeerCloseManager::MembershipRequestCallback(Event *event) {
528 95019 : CHECK_CONCURRENCY(peer_close_->GetTaskName());
529 :
530 95019 : bool result = false;
531 113618 : PEER_CLOSE_MANAGER_LOG("MembershipRequestCallback");
532 :
533 95019 : if (!AssertMembershipManagerInUse())
534 0 : return result;
535 95019 : if (--membership_req_pending_)
536 73596 : return result;
537 :
538 : // Indicate to the caller that we are done using the membership manager.
539 21423 : result = true;
540 :
541 21423 : if (state_ == DELETE) {
542 19562 : MOVE_TO_STATE(NONE);
543 18734 : peer_close_->Delete();
544 18734 : gr_elapsed_ = 0;
545 18734 : llgr_elapsed_ = 0;
546 18734 : stats_.init++;
547 18734 : close_again_ = false;
548 18734 : graceful_ = true;
549 18734 : set_membership_state(MEMBERSHIP_NONE);
550 18734 : return result;
551 : }
552 :
553 : // Process nested closures.
554 2689 : 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 2689 : if (state_ == STALE) {
565 983 : peer_close_->CloseComplete();
566 1178 : MOVE_TO_STATE(GR_TIMER);
567 983 : peer_close_->GetGracefulRestartFamilies(&families_);
568 :
569 : // Offset restart time with elapsed time during nested closures.
570 983 : int time = peer_close_->GetGracefulRestartTime() * 1000;
571 983 : time -= gr_elapsed_;
572 983 : if (time < 0)
573 0 : time = 0;
574 983 : StartRestartTimer(time);
575 983 : stats_.gr_timer++;
576 983 : set_membership_state(MEMBERSHIP_NONE);
577 983 : 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 1706 : 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 1325 : TriggerSweepStateActions();
602 1325 : return result;
603 : }
604 :
605 62931 : void PeerCloseManager::FillRouteCloseInfo(PeerCloseInfo *close_info) const {
606 62931 : std::map<std::string, PeerCloseRouteInfo> route_stats;
607 :
608 692241 : for (int i = 0; i < Address::NUM_FAMILIES; i++) {
609 629310 : if (!stats_.route_stats[i].IsSet())
610 474292 : continue;
611 155018 : PeerCloseRouteInfo route_info;
612 155018 : route_info.set_staled(stats_.route_stats[i].staled);
613 155018 : route_info.set_llgr_staled(stats_.route_stats[i].llgr_staled);
614 155018 : route_info.set_refreshed(stats_.route_stats[i].refreshed);
615 155018 : route_info.set_fresh(stats_.route_stats[i].fresh);
616 155018 : route_info.set_deleted(stats_.route_stats[i].deleted);
617 155018 : route_stats[Address::FamilyToString(static_cast<Address::Family>(i))] = route_info;
618 155018 : }
619 :
620 62931 : if (!route_stats.empty())
621 62101 : close_info->set_route_stats(route_stats);
622 62931 : }
623 :
624 62931 : BgpNeighborResp *PeerCloseManager::FillCloseInfo(BgpNeighborResp *resp) const {
625 62931 : PeerCloseInfo peer_close_info;
626 62931 : peer_close_info.set_state(GetStateName(state_));
627 62931 : peer_close_info.set_membership_state(
628 125862 : GetMembershipStateName(membership_state_));
629 62931 : peer_close_info.set_close_again(close_again_);
630 62931 : peer_close_info.set_graceful(graceful_);
631 62931 : peer_close_info.set_init(stats_.init);
632 62931 : peer_close_info.set_close(stats_.close);
633 62931 : peer_close_info.set_nested(stats_.nested);
634 62931 : peer_close_info.set_deletes(stats_.deletes);
635 62931 : peer_close_info.set_stale(stats_.stale);
636 62931 : peer_close_info.set_llgr_stale(stats_.llgr_stale);
637 62931 : peer_close_info.set_sweep(stats_.sweep);
638 62931 : peer_close_info.set_gr_timer(stats_.gr_timer);
639 62931 : peer_close_info.set_llgr_timer(stats_.llgr_timer);
640 62931 : FillRouteCloseInfo(&peer_close_info);
641 :
642 62931 : resp->set_peer_close_info(peer_close_info);
643 :
644 62931 : return resp;
645 62931 : }
646 :
647 216519 : void PeerCloseManager::UpdateRouteStats(Address::Family family,
648 : const BgpPath *old_path, uint32_t path_flags) const {
649 216519 : if (state_ == NONE)
650 197156 : return;
651 :
652 19363 : if (!old_path)
653 70 : stats_.route_stats[family].fresh++;
654 19293 : else if (old_path->IsStale() && !(path_flags & BgpPath::Stale))
655 3108 : stats_.route_stats[family].refreshed++;
656 : }
657 :
658 100329 : bool PeerCloseManager::MembershipPathCallback(DBTablePartBase *root,
659 : BgpRoute *rt, BgpPath *path) {
660 100329 : CHECK_CONCURRENCY("db::DBTable");
661 100338 : DBRequest::DBOperation oper = DBRequest::DB_ENTRY_INVALID;
662 100338 : BgpAttrPtr attrs;
663 :
664 100338 : BgpTable *table = static_cast<BgpTable *>(root->parent());
665 100339 : assert(table);
666 :
667 100339 : uint32_t stale = 0;
668 :
669 100339 : switch (state_) {
670 0 : case NONE:
671 : case GR_TIMER:
672 : case LLGR_TIMER:
673 0 : return false;
674 :
675 5766 : case SWEEP:
676 :
677 : // Stale paths must be deleted.
678 5766 : if (!path->IsStale() && !path->IsLlgrStale())
679 2640 : return false;
680 3128 : if (path->IsStale()) {
681 3128 : path->ResetStale();
682 3128 : table->UpdateStalePathCount(-1);
683 : }
684 3130 : if (path->IsLlgrStale()) {
685 0 : path->ResetLlgrStale();
686 0 : table->UpdateLlgrStalePathCount(-1);
687 : }
688 3130 : oper = DBRequest::DB_ENTRY_DELETE;
689 3130 : attrs = NULL;
690 3130 : stats_.route_stats[table->family()].deleted++;
691 3130 : break;
692 :
693 76854 : case DELETE:
694 :
695 : // This path must be deleted. Hence attr is not required.
696 76854 : oper = DBRequest::DB_ENTRY_DELETE;
697 76854 : attrs = NULL;
698 76856 : stats_.route_stats[table->family()].deleted++;
699 76890 : break;
700 :
701 14481 : case STALE:
702 :
703 : // We do not support GR for multicast routes (yet).
704 28962 : if ((table->family() == Address::ERMVPN) ||
705 14481 : (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 14481 : if (path->IsStale())
716 3288 : return false;
717 :
718 : // This path must be marked for staling. Update the local
719 : // preference and update the route accordingly.
720 11194 : oper = DBRequest::DB_ENTRY_ADD_CHANGE;
721 11194 : attrs = path->GetAttr();
722 11197 : stale = BgpPath::Stale;
723 11197 : stats_.route_stats[table->family()].staled++;
724 11197 : break;
725 :
726 3234 : case LLGR_STALE:
727 :
728 : // If the path has NO_LLGR community, DELETE it.
729 3234 : 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 3234 : if (path->IsLlgrStale())
742 0 : return false;
743 :
744 3232 : attrs = path->GetAttr();
745 3233 : stale = BgpPath::LlgrStale;
746 3233 : oper = DBRequest::DB_ENTRY_ADD_CHANGE;
747 3233 : stats_.route_stats[table->family()].llgr_staled++;
748 3235 : break;
749 : }
750 :
751 : // Feed the route modify/delete request to the table input process.
752 283341 : return table->InputCommon(root, rt, path, peer_close_->peer(), NULL, oper,
753 94446 : attrs, path->GetPathId(), path->GetFlags() | stale, path->GetLabel(),
754 94445 : path->GetL3Label());
755 100372 : }
756 :
757 : //
758 : // Handler for an Event.
759 : //
760 143036 : bool PeerCloseManager::EventCallback(Event *event) {
761 143036 : CHECK_CONCURRENCY(peer_close_->GetTaskName());
762 : bool result;
763 :
764 143036 : switch (event->event_type) {
765 0 : case EVENT_NONE:
766 0 : break;
767 20848 : case CLOSE:
768 20848 : Close(event);
769 20848 : break;
770 14725 : case EOR_RECEIVED:
771 14725 : ProcessEORMarkerReceived(event);
772 14725 : break;
773 20124 : case MEMBERSHIP_REQUEST:
774 20124 : MembershipRequest(event);
775 20124 : break;
776 84607 : case MEMBERSHIP_REQUEST_COMPLETE_CALLBACK:
777 84607 : result = MembershipRequestCallback(event);
778 :
779 : // Notify clients if we are no longer using the membership mgr.
780 84607 : if (result)
781 11011 : peer_close_->MembershipRequestCallbackComplete();
782 84607 : break;
783 2732 : case TIMER_CALLBACK:
784 2732 : RestartTimerCallback(event);
785 2732 : break;
786 : }
787 :
788 143036 : delete event;
789 143036 : return true;
790 : }
|
