1 2012-xx-xx Whamcloud, Inc.
3 * Support for networks:
4 socklnd - any kernel supported by Lustre,
5 qswlnd - Qsnet kernel modules 5.20 and later,
6 o2iblnd - OFED 1.3, 1.4.1, 1.4.2, 1.5.1, 1.5.2, 1.5.3.1 and 1.5.3.2
7 gmlnd - GM 2.1.22 and later,
8 mxlnd - MX 1.2.10 or later,
9 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
11 See http://wiki.whamcloud.com/display/PUB/Changelog+1.8 for more information.
13 -------------------------------------------------------------------------------
15 2012-05-31 Whamcloud, Inc.
17 * Support for networks:
18 socklnd - any kernel supported by Lustre,
19 qswlnd - Qsnet kernel modules 5.20 and later,
20 o2iblnd - OFED 1.3, 1.4.1, 1.4.2, 1.5.1, 1.5.2, 1.5.3.1 and 1.5.3.2
21 gmlnd - GM 2.1.22 and later,
22 mxlnd - MX 1.2.10 or later,
23 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
25 See http://wiki.whamcloud.com/display/PUB/Changelog+1.8 for more information.
27 -------------------------------------------------------------------------------
29 2011-10-31 Whamcloud, Inc.
31 * Support for networks:
32 socklnd - any kernel supported by Lustre,
33 qswlnd - Qsnet kernel modules 5.20 and later,
34 o2iblnd - OFED 1.3, 1.4.1, 1.4.2, 1.5.1, 1.5.2, 1.5.3.1 and 1.5.3.2
35 gmlnd - GM 2.1.22 and later,
36 mxlnd - MX 1.2.10 or later,
37 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
39 See http://wiki.whamcloud.com/display/PUB/Changelog+1.8 for more information.
41 -------------------------------------------------------------------------------
43 2011-06-30 Whamcloud, Inc.
45 * Support for networks:
46 socklnd - any kernel supported by Lustre,
47 qswlnd - Qsnet kernel modules 5.20 and later,
48 o2iblnd - OFED 1.3, 1.4.1, 1.4.2, 1.5.1, 1.5.2 and 1.5.3.1,
49 gmlnd - GM 2.1.22 and later,
50 mxlnd - MX 1.2.10 or later,
51 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
54 Jira : LU-388, LU-303, TT-121
55 Description: Add support for OFED 1.5.3.1
57 -------------------------------------------------------------------------------
59 2011-05-12 Oracle, Inc.
61 * Support for networks:
62 socklnd - any kernel supported by Lustre,
63 qswlnd - Qsnet kernel modules 5.20 and later,
64 openiblnd - IbGold 1.8.2,
65 o2iblnd - OFED 1.3, 1.4.1, 1.4.2, 1.5.1 and 1.5.2
66 viblnd - Voltaire ibhost 3.4.5 and later,
67 ciblnd - Topspin 3.2.0,
68 iiblnd - Infiniserv 3.3 + PathBits patch,
69 gmlnd - GM 2.1.22 and later,
70 mxlnd - MX 1.2.10 or later,
71 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
75 Description: fix contention on ksock_tx_t
79 Description: dynamically grow/shrink connd threads pool
83 Description: ptlrpcd stuck in lnet allocations under memory pressure
84 Details : Set PF_MEMALLOC on outgoing path to prevent deadlock on memory
89 Description: fix o2iblnd v2 regression of credit deadlock with v1 peers
94 Description: lctl ping of 1.6.6 server from 1.8 router causes protocol error
96 -------------------------------------------------------------------------------
98 2010-10-29 Oracle, Inc.
100 * Support for networks:
101 socklnd - any kernel supported by Lustre,
102 qswlnd - Qsnet kernel modules 5.20 and later,
103 openiblnd - IbGold 1.8.2,
104 o2iblnd - OFED 1.3, 1.4.1, 1.4.2, 1.5.1 and 1.5.2
105 viblnd - Voltaire ibhost 3.4.5 and later,
106 ciblnd - Topspin 3.2.0,
107 iiblnd - Infiniserv 3.3 + PathBits patch,
108 gmlnd - GM 2.1.22 and later,
109 mxlnd - MX 1.2.10 or later,
110 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
114 Description: MDS node unresponsive
115 Details : improve hash distribution, doubled hash size for the lnet cookie
118 -------------------------------------------------------------------------------
120 2010-07-31 Oracle, Inc.
122 * Support for networks:
123 socklnd - any kernel supported by Lustre,
124 qswlnd - Qsnet kernel modules 5.20 and later,
125 openiblnd - IbGold 1.8.2,
126 o2iblnd - OFED 1.3, 1.4.1, 1.4.2 and 1.5.1
127 viblnd - Voltaire ibhost 3.4.5 and later,
128 ciblnd - Topspin 3.2.0,
129 iiblnd - Infiniserv 3.3 + PathBits patch,
130 gmlnd - GM 2.1.22 and later,
131 mxlnd - MX 1.2.10 or later,
132 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
136 Description: With peer health detection, o2iblnd makes only one attempt to
137 reconnect which is not enough with nodes running lustre 1.6
138 because of proto version mismatch. Fix o2iblnd to retry one more
141 Severity : enhancement
143 Description: Quiet some LNET messages
145 Severity : enhancement
147 Description: Add OFED 1.5.1 support
149 Severity : enhancement
151 Description: The peer health code lacked some important debugging info in
152 lnd_query code paths. We've added necessary debug prints,
153 not just for bug 21678, but also for future troubleshooting.
155 -------------------------------------------------------------------------------
157 2010-04-30 Oracle, Inc.
159 * Support for networks:
160 socklnd - any kernel supported by Lustre,
161 qswlnd - Qsnet kernel modules 5.20 and later,
162 openiblnd - IbGold 1.8.2,
163 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, 1.3, 1.4.1, and 1.4.2
164 viblnd - Voltaire ibhost 3.4.5 and later,
165 ciblnd - Topspin 3.2.0,
166 iiblnd - Infiniserv 3.3 + PathBits patch,
167 gmlnd - GM 2.1.22 and later,
168 mxlnd - MX 1.2.10 or later,
169 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
173 Description: lst: check # of remained RPCs before aborting
174 Details : lstcon_rpc_trans_postwait() calls lstcon_rpc_trans_abort() only
175 when the transaction is timeout, so if we got "end_session" to
176 interrupt waiting on transaction, then we can hit the assertion
177 failure ASSERTION(crpc->crp_stamp != 0)
181 Description: print more debug info for timedout ZC-req
182 Details : Print more information for timedout ZC-req and partial
183 received connection. Close connection for timedout ZC-req
184 Always send ZC_ACK on non-blocking connection(BULK_IN)
188 Description: Adding WIRE_ATTR attribute to LNET types
189 Details : LST nodes on different platforms might not communicate well
190 due to the lack of WIRE_ATTR attribute in some LNET structures
191 traversing network. The patch fixes the problem by adding
192 WIRE_ATTR where needed.
196 Description: hash MEs on RDMA portal
197 Details : RDMA portal can have very long ME list on client side, which
198 will trigger soft lockup because of long searching on list.
199 Hash MEs on RDMA portal can resolve this problem.
203 Description: fix for double release of ibc_lock in o2iblnd
204 Details : Re-acquire ibc_lock in kiblnd_post_tx_locked(). Add extra
205 reference to conn before calling kiblnd_post_tx_locked()
206 to avoid scenario when conn disappears inside
207 kiblnd_post_tx_locked().
209 -------------------------------------------------------------------------------
210 2010-01-29 Sun Microsystems, Inc.
212 * Support for networks:
213 socklnd - any kernel supported by Lustre,
214 qswlnd - Qsnet kernel modules 5.20 and later,
215 openiblnd - IbGold 1.8.2,
216 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, 1.3, 1.4.1, and 1.4.2
217 viblnd - Voltaire ibhost 3.4.5 and later,
218 ciblnd - Topspin 3.2.0,
219 iiblnd - Infiniserv 3.3 + PathBits patch,
220 gmlnd - GM 2.1.22 and later,
221 mxlnd - MX 1.2.10 or later,
222 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
226 Description: should update lp_alive for non-router peers.
228 Severity : enhancement
230 Description: LNet router shuffler.
232 Severity : enhancement
234 Description: LNet fine grain routing support.
238 Description: router checker stops working when system wall clock goes backward
239 Details : use monotonic timing source instead of system wall clock time.
241 Severity : enhancement
243 Description: avoid asymmetrical router failures
245 Severity : enhancement
247 Description: multiple-instance support for kptllnd
251 Description: ksocknal_close_conn_locked connection race
252 Details : A race was possible when ksocknal_create_conn calls
253 ksocknal_close_conn_locked for already closed conn.
255 Severity : enhancement
257 Description: port router pinger to userspace
261 Description: kptllnd HELLO protocol deadlock
262 Details : kptllnd HELLO protocol doesn't run to completion in finite time
266 Description: LNet selftest fixes and enhancements
268 Severity : enhancement
270 Description: allow a test node to be a member of multiple test groups
272 Severity : enhancement
274 Description: MXLND: eliminate hosts file, use arp for peer nic_id resolution
275 Details : an update from the upstream developer Scott Atchley.
278 -------------------------------------------------------------------------------
279 2009-07-31 Sun Microsystems, Inc.
281 * Support for networks:
282 socklnd - any kernel supported by Lustre,
283 qswlnd - Qsnet kernel modules 5.20 and later,
284 openiblnd - IbGold 1.8.2,
285 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, 1.3, and 1.4.1
286 viblnd - Voltaire ibhost 3.4.5 and later,
287 ciblnd - Topspin 3.2.0,
288 iiblnd - Infiniserv 3.3 + PathBits patch,
289 gmlnd - GM 2.1.22 and later,
290 mxlnd - MX 1.2.1 or later,
291 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
295 Description: router_proc.c is rewritten to use sysctl-interface for parameters
296 residing in /proc/sys/lnet
298 Severity : enhancement
300 Description: add a new LND optiion to control peer buffer credits on routers
304 Description: Fixing deadlock in usocklnd
305 Details : A deadlock was possible in usocklnd due to race condition while
306 tearing connection down. The problem resulted from erroneous
307 assumption that lnet_finalize() could have been called holding
308 some lnd-level locks.
311 Bugzilla : 13621, 15983
312 Description: Protocol V2 of o2iblnd
313 Details : o2iblnd V2 has several new features:
314 . map-on-demand: map-on-demand is disabled by default, it can
315 be enabled by using modparam "map_on_demand=@value@", @value@
316 should >= 0 and < 256, 0 will disable map-on-demand, any other
317 valid value will enable map-on-demand.
318 Oi2blnd will create FMR or physical MR for RDMA if fragments of
320 Enable map-on-demand will take less memory for new connection,
321 but a little more CPU for RDMA.
322 . iWARP : to support iWARP, please enable map-on-demand, 32 and 64
323 are recommanded value. iWARP will probably fail for value >=128.
324 . OOB NOOP message: to resolve deadlock on router.
325 . tunable peer_credits_hiw: (high water to return credits),
326 default value of peer_credits_hiw equals to (peer_credits -1),
327 user can change it between peer_credits/2 and (peer_credits - 1).
328 Lower value is recommended for high latency network.
329 . tunable message queue size: it always equals to peer_credits,
330 higher value is recommended for high latency network.
331 . It's compatible with earlier version of o2iblnd
335 Description: Fixing 'running out of ports' issue
336 Details : Add a delay before next reconnect attempt in ksocklnd in
337 the case of lost race. Limit the frequency of query-requests
338 in lnet. Improved handling of 'dead peer' notifications in
343 Description: Change ptllnd timeout and watchdog timers
344 Details : Add ptltrace_on_nal_failed and bump ptllnd timeout to match
345 Portals wire timeout.
349 Description: One down Lustre FS hangs ALL mounted Lustre filesystems
350 Details : Shared routing enhancements - peer health detection.
354 Description: IB path MTU mistakenly set to 1st path MTU when ib_mtu is off
355 Details : See comment 46 in bug 11245 for details - it's indeed a bug
356 introduced by the original 11245 fix.
360 Description: uptllnd credit overflow fix
361 Details : kptl_msg_t::ptlm_credits could be overflown by uptllnd since
366 Description: socklnd protocol version 3
367 Details : With current protocol V2, connections on router can be
368 blocked and can't receive any incoming messages when there is no
369 more router buffer, so ZC-ACK can't be handled (LNet message
370 can't be finalized) and will cause deadlock on router.
371 Protocol V3 has a dedicated connection for emergency messages
372 like ZC-ACK to router, messages on this dedicated connection
373 don't need any credit so will never be blocked. Also, V3 can send
374 keepalive ping in specified period for router healthy checking.
376 -------------------------------------------------------------------------------
377 12-31-2008 Sun Microsystems, Inc.
379 * Support for networks:
380 socklnd - any kernel supported by Lustre,
381 qswlnd - Qsnet kernel modules 5.20 and later,
382 openiblnd - IbGold 1.8.2,
383 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, and 1.3
384 viblnd - Voltaire ibhost 3.4.5 and later,
385 ciblnd - Topspin 3.2.0,
386 iiblnd - Infiniserv 3.3 + PathBits patch,
387 gmlnd - GM 2.1.22 and later,
388 mxlnd - MX 1.2.1 or later,
389 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
398 Description: workaround for OOM from o2iblnd
399 Details : OFED needs allocate big chunk of memory for QP while creating
400 connection for o2iblnd, OOM can happen if no such a contiguous
402 QP size is decided by concurrent_sends and max_fragments of
403 o2iblnd, now we permit user to specify smaller value for
404 concurrent_sends of o2iblnd(i.e: concurrent_sends=7), which
405 will decrease memory block size required by creating QP.
409 Description: Support Zerocopy receive of Chelsio device
410 Details : Chelsio driver can support zerocopy for iov[1] if it's
411 contiguous and large enough.
415 Description: fix credit flow deadlock in uptllnd
419 Description: finalize network operation in reasonable time
420 Details : conf-sanity test_32a couldn't stop ost and mds because it
421 tried to access non-existent peer and tcp connect took
422 quite long before timing out.
426 Description: Continuous recovery on 33 of 413 nodes after lustre oss failure
427 Details : Lost reference on conn prevents peer from being destroyed, which
428 could prevent new peer creation if peer count has reached upper
433 Description: LNET Selftest results in Soft lockup on OSS CPU
434 Details : only hits when 8 or more o2ib clients involved and a session is
435 torn down with 'lst end_session' without preceeding 'lst stop'.
439 Description: concurrent_sends in IB LNDs should not be changeable at run time
440 Details : concurrent_sends in IB LNDs should not be changeable at run time
444 Description: ptl_send_rpc hits LASSERT when ptl_send_buf fails
445 Details : only hits under out-of-memory situations
448 -------------------------------------------------------------------------------
450 2009-02-07 Sun Microsystems, Inc.
452 * Support for networks:
453 socklnd - any kernel supported by Lustre,
454 qswlnd - Qsnet kernel modules 5.20 and later,
455 openiblnd - IbGold 1.8.2,
456 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, and 1.3
457 viblnd - Voltaire ibhost 3.4.5 and later,
458 ciblnd - Topspin 3.2.0,
459 iiblnd - Infiniserv 3.3 + PathBits patch,
460 gmlnd - GM 2.1.22 and later,
461 mxlnd - MX 1.2.1 or later,
462 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
466 Description: workaround for OOM from o2iblnd
467 Details : OFED needs allocate big chunk of memory for QP while creating
468 connection for o2iblnd, OOM can happen if no such a contiguous
470 QP size is decided by concurrent_sends and max_fragments of
471 o2iblnd, now we permit user to specify smaller value for
472 concurrent_sends of o2iblnd(i.e: concurrent_sends=7), which
473 will decrease memory block size required by creating QP.
477 Description: Support Zerocopy receive of Chelsio device
478 Details : Chelsio driver can support zerocopy for iov[1] if it's
479 contiguous and large enough.
483 Description: fix credit flow deadlock in uptllnd
487 Description: finalize network operation in reasonable time
488 Details : conf-sanity test_32a couldn't stop ost and mds because it
489 tried to access non-existent peer and tcp connect took
490 quite long before timing out.
494 Description: Continuous recovery on 33 of 413 nodes after lustre oss failure
495 Details : Lost reference on conn prevents peer from being destroyed, which
496 could prevent new peer creation if peer count has reached upper
501 Description: LNET Selftest results in Soft lockup on OSS CPU
502 Details : only hits when 8 or more o2ib clients involved and a session is
503 torn down with 'lst end_session' without preceeding 'lst stop'.
507 Description: concurrent_sends in IB LNDs should not be changeable at run time
508 Details : concurrent_sends in IB LNDs should not be changeable at run time
510 -------------------------------------------------------------------------------
512 11-03-2008 Sun Microsystems, Inc.
514 * Support for networks:
515 socklnd - any kernel supported by Lustre,
516 qswlnd - Qsnet kernel modules 5.20 and later,
517 openiblnd - IbGold 1.8.2,
518 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, and 1.3
519 viblnd - Voltaire ibhost 3.4.5 and later,
520 ciblnd - Topspin 3.2.0,
521 iiblnd - Infiniserv 3.3 + PathBits patch,
522 gmlnd - GM 2.1.22 and later,
523 mxlnd - MX 1.2.1 or later,
524 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
528 Description: ptl_send_rpc hits LASSERT when ptl_send_buf fails
529 Details : only hits under out-of-memory situations
531 -------------------------------------------------------------------------------
534 04-26-2008 Sun Microsystems, Inc.
536 * Support for networks:
537 socklnd - any kernel supported by Lustre,
538 qswlnd - Qsnet kernel modules 5.20 and later,
539 openiblnd - IbGold 1.8.2,
540 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5
541 viblnd - Voltaire ibhost 3.4.5 and later,
542 ciblnd - Topspin 3.2.0,
543 iiblnd - Infiniserv 3.3 + PathBits patch,
544 gmlnd - GM 2.1.22 and later,
545 mxlnd - MX 1.2.1 or later,
546 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
550 Description: excessive debug information removed
551 Details : excessive debug information removed
555 Description: ksocknal_create_conn() hit ASSERTION during connection race
556 Details : ksocknal_create_conn() hit ASSERTION during connection race
560 Description: ksocknal_send_hello() hit ASSERTION while connecting race
561 Details : ksocknal_send_hello() hit ASSERTION while connecting race
565 Description: o2iblnd/ptllnd credit deadlock in a routed config.
566 Details : o2iblnd/ptllnd credit deadlock in a routed config.
570 Description: High load after starting lnet
571 Details : gmlnd should sleep in rx thread in interruptible way. Otherwise,
572 uptime utility reports high load that looks confusingly.
576 Description: ksocklnd fails to establish connection if accept_port is high
577 Details : PID remapping must not be done for active (outgoing) connections
579 --------------------------------------------------------------------------------
581 2008-01-11 Sun Microsystems, Inc.
583 * Support for networks:
584 socklnd - any kernel supported by Lustre,
585 qswlnd - Qsnet kernel modules 5.20 and later,
586 openiblnd - IbGold 1.8.2,
587 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5
588 viblnd - Voltaire ibhost 3.4.5 and later,
589 ciblnd - Topspin 3.2.0,
590 iiblnd - Infiniserv 3.3 + PathBits patch,
591 gmlnd - GM 2.1.22 and later,
592 mxlnd - MX 1.2.1 or later,
593 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
596 Description: liblustre network error
597 Details : liblustre clients should understand LNET_ACCEPT_PORT environment
598 variable even if they don't start lnet acceptor.
602 Description: Strange message from lnet (Ignoring prediction from the future)
603 Details : Incorrect calculation of peer's last_alive value in ksocklnd
605 --------------------------------------------------------------------------------
607 2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
609 * Support for networks:
610 socklnd - any kernel supported by Lustre,
611 qswlnd - Qsnet kernel modules 5.20 and later,
612 openiblnd - IbGold 1.8.2,
613 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5.
614 viblnd - Voltaire ibhost 3.4.5 and later,
615 ciblnd - Topspin 3.2.0,
616 iiblnd - Infiniserv 3.3 + PathBits patch,
617 gmlnd - GM 2.1.22 and later,
618 mxlnd - MX 1.2.1 or later,
619 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
623 Description: ASSERTION(me == md->md_me) failed in lnet_match_md()
627 Description: increase send queue size for ciblnd/openiblnd
631 Description: new userspace socklnd
632 Details : Old userspace tcpnal that resided in lnet/ulnds/socklnd replaced
633 with new one - usocklnd.
635 Severity : enhancement
637 Description: Console message flood
638 Details : Make cdls ratelimiting more tunable by adding several tunable in
639 procfs /proc/sys/lnet/console_{min,max}_delay_centisecs and
640 /proc/sys/lnet/console_backoff.
642 --------------------------------------------------------------------------------
644 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
646 * Support for networks:
647 socklnd - any kernel supported by Lustre,
648 qswlnd - Qsnet kernel modules 5.20 and later,
649 openiblnd - IbGold 1.8.2,
650 o2iblnd - OFED 1.1 and 1.2,
651 viblnd - Voltaire ibhost 3.4.5 and later,
652 ciblnd - Topspin 3.2.0,
653 iiblnd - Infiniserv 3.3 + PathBits patch,
654 gmlnd - GM 2.1.22 and later,
655 mxlnd - MX 1.2.1 or later,
656 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
660 Description: /proc/sys/lnet has non-sysctl entries
661 Details : Updating dump_kernel/daemon_file/debug_mb to use sysctl variables
665 Description: TOE Kernel panic by ksocklnd
666 Details : offloaded sockets provide their own implementation of sendpage,
667 can't call tcp_sendpage() directly
671 Description: kibnal_shutdown() doesn't finish; lconf --cleanup hangs
672 Details : races between lnd_shutdown and peer creation prevent
673 lnd_shutdown from finishing.
677 Description: open files rlimit 1024 reached while liblustre testing
678 Details : ulnds/socklnd must close open socket after unsuccessful
683 Description: build error
684 Details : fix typos in gmlnd, ptllnd and viblnd
686 ------------------------------------------------------------------------------
688 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
690 * Support for networks:
691 socklnd - kernels up to 2.6.16,
692 qswlnd - Qsnet kernel modules 5.20 and later,
693 openiblnd - IbGold 1.8.2,
694 o2iblnd - OFED 1.1 and 1.2
695 viblnd - Voltaire ibhost 3.4.5 and later,
696 ciblnd - Topspin 3.2.0,
697 iiblnd - Infiniserv 3.3 + PathBits patch,
698 gmlnd - GM 2.1.22 and later,
699 mxlnd - MX 1.2.1 or later,
700 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
702 2007-06-21 Cluster File Systems, Inc. <info@clusterfs.com>
704 * Support for networks:
705 socklnd - kernels up to 2.6.16,
706 qswlnd - Qsnet kernel modules 5.20 and later,
707 openiblnd - IbGold 1.8.2,
709 viblnd - Voltaire ibhost 3.4.5 and later,
710 ciblnd - Topspin 3.2.0,
711 iiblnd - Infiniserv 3.3 + PathBits patch,
712 gmlnd - GM 2.1.22 and later,
713 mxlnd - MX 1.2.1 or later,
714 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
718 Description: Initialize cpumask before use
722 Description: ASSERTION failures when upgrading to the patchless zero-copy
724 Details : This bug affects "rolling upgrades", causing an inconsistent
725 protocol version negotiation and subsequent assertion failure
726 during rolling upgrades after the first wave of upgrades.
730 Details : Change "dropped message" CERRORs to D_NETERROR so they are
731 logged instead of creating "console chatter" when a lustre
732 timeout races with normal RPC completion.
735 Details : lnet_clear_peer_table can wait forever if user forgets to
739 Details : libcfs_id2str should check pid against LNET_PID_ANY.
743 Description: added LNET self test
744 Details : landing b_self_test
749 Description: cfs_duration_{u,n}sec() wrongly calculate nanosecond part of
751 Details : do_div() macro is used incorrectly.
753 2007-04-23 Cluster File Systems, Inc. <info@clusterfs.com>
757 Description: make panic on lbug configurable
761 Description: Add OFED1.2 support to o2iblnd
762 Details : o2iblnd depends on OFED's modules, if out-tree OFED's modules
763 are installed (other than kernel's in-tree infiniband), there
764 could be some problem while insmod o2iblnd (mismatch CRC of
766 If extra Module.symvers is supported in kernel (i.e, 2.6.17),
767 this link provides solution:
768 https://bugs.openfabrics.org/show_bug.cgi?id=355
769 if extra Module.symvers is not supported in kernel, we will
770 have to run the script in bug 12316 to update
771 $LINUX/module.symvers before building o2iblnd.
772 More details about this are in bug 12316.
774 ------------------------------------------------------------------------------
776 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
777 * version 1.4.10 / 1.6.0
778 * Support for networks:
779 socklnd - kernels up to 2.6.16,
780 qswlnd - Qsnet kernel modules 5.20 and later,
781 openiblnd - IbGold 1.8.2,
783 viblnd - Voltaire ibhost 3.4.5 and later,
784 ciblnd - Topspin 3.2.0,
785 iiblnd - Infiniserv 3.3 + PathBits patch,
786 gmlnd - GM 2.1.22 and later,
787 mxlnd - MX 1.2.1 or later,
788 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
792 Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
793 possibly accessed in kptllnd_shutdown. Ptllnd should init
794 kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
798 Description: gmlnd ignored some transmit errors when finalizing lnet messages.
802 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
806 Description: the_lnet.ln_finalizing was not set when the current thread is
807 about to complete messages. It only affects multi-threaded
813 Description: Changed the default kqswlnd ntxmsg=512
818 Description: Assertion failure in kernel ptllnd caused by posting passive
819 bulk buffers before connection establishment complete.
824 Description: A race in kernel ptllnd between deleting a peer and posting
825 new communications for it could hang communications -
826 manifesting as "Unexpectedly long timeout" messages.
831 Description: Kernel ptllnd lock ordering issue could hang a node.
836 Description: node crash on socket teardown race
839 Frequency : 'lctl peer_list' issued on a mx net
841 Description: Enable lctl's peer_list for MXLND
844 Frequency : after Ptllnd timeouts and portals congestion
846 Description: Credit overflows
847 Details : This was a bug in ptllnd connection establishment. The fix
848 implements better peer stamps to disambiguate connection
849 establishment and ensure both peers enter the credit flow
850 state machine consistently.
855 Description: kptllnd didn't propagate some network errors up to LNET
856 Details : This bug was spotted while investigating 11394. The fix
857 ensures network errors on sends and bulk transfers are
858 propagated to LNET/lustre correctly.
860 Severity : enhancement
862 Description: Fixed console chatter in case of -ETIMEDOUT.
864 Severity : enhancement
866 Description: Added D_NETTRACE for recording network packet history
867 (initially only for ptllnd). Also a separate userspace
868 ptllnd facility to gather history which should really be
869 covered by D_NETTRACE too, if only CDEBUG recorded history in
875 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
876 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
877 callback can occur before a connection has actually been
878 established. This caused an assertion failure previously.
880 Severity : enhancement
882 Description: Multiple instances for o2iblnd
883 Details : Allow multiple instances of o2iblnd to enable networking over
884 multiple HCAs and routing between them.
888 Description: lnet deadlock in router_checker
889 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
890 into BH locks to eliminate potential deadlock caused by
891 ksocknal_data_ready() preempting code holding these locks.
895 Description: Millions of failed socklnd connection attempts cause a very slow FS
896 Details : added a new route flag ksnr_scheduled to distinguish from
897 ksnr_connecting, so that a peer connection request is only turned
898 down for race concerns when an active connection to the same peer
899 is under progress (instead of just being scheduled).
901 ------------------------------------------------------------------------------
903 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
905 * Support for networks:
906 socklnd - kernels up to 2.6.16
907 qswlnd - Qsnet kernel modules 5.20 and later
908 openiblnd - IbGold 1.8.2
910 viblnd - Voltaire ibhost 3.4.5 and later
911 ciblnd - Topspin 3.2.0
912 iiblnd - Infiniserv 3.3 + PathBits patch
913 gmlnd - GM 2.1.22 and later
914 mxlnd - MX 1.2.1 or later
915 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
918 Severity : major on XT3
920 Description: libcfs overwrites /proc/sys/portals
921 Details : libcfs created a symlink from /proc/sys/portals to
922 /proc/sys/lnet for backwards compatibility. This is no
923 longer required and makes the Cray portals /proc variables
928 Description: OFED FMR API change
929 Details : This changes parameter usage to reflect a change in
930 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
931 that FMR support is only used in experimental versions of the
932 o2iblnd - this change does not affect standard usage at all.
934 Severity : enhancement
936 Description: new ko2iblnd module parameter: ib_mtu
937 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
938 HCAs. You can avoid this problem by setting the MTU to 1024
939 using this module parameter.
941 Severity : enhancement
942 Bugzilla : 11118/11620
943 Description: ptllnd small request message buffer alignment fix
944 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
945 Round up small message size on sends in case this option
946 is not supported. 11620 was a defect in the initial
947 implementation which effectively asserted all peers had to be
948 running the correct protocol version which was fixed by always
949 NAK-ing such requests and handling any misalignments they
954 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
955 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
956 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
958 Severity : enhancement
960 Description: Patchless ZC(zero copy) socklnd
961 Details : New protocol for socklnd, socklnd can support zero copy without
962 kernel patch, it's compatible with old socklnd. Checksum is
963 moved from tunables to modparams.
967 Description: When ksocknal_del_peer() is called upon a peer whose
968 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
969 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
972 Frequency : when ptlrpc is under heavy use and runs out of request buffer
974 Description: In lnet_match_blocked_msg(), md can be used without holding a
978 Frequency : very rarely
980 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
981 If connd connects a route which has been closed by
982 ksocknal_shutdown(), ksocknal_create_routes() may create new
983 routes which hold references on the peer, causing shutdown
984 process to wait for peer to disappear forever.
986 Severity : enhancement
988 Description: Dump XT3 portals traces on kptllnd timeout
989 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
990 dump Cray portals debug traces to a file. The kptllnd module
991 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
992 is the basename of the dump file.
995 Frequency : infrequent
997 Description: kernel ptllnd fix bug in connection re-establishment
998 Details : Kernel ptllnd could produce protocol errors e.g. illegal
999 matchbits and/or violate the credit flow protocol when trying
1000 to re-establish a connection with a peer after an error or
1003 Severity : enhancement
1005 Description: Allow /proc/sys/lnet/debug to be set symbolically
1006 Details : Allow debug and subsystem debug values to be read/set by name
1007 in addition to numerically, for ease of use.
1010 Frequency : only in configurations with LNET routers
1012 Description: routes automatically marked down and recovered
1013 Details : In configurations with LNET routers if a router fails routers
1014 now actively try to recover routes that are down, unless they
1015 are marked down by an administrator.
1017 ------------------------------------------------------------------------------
1019 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1022 Frequency : very rarely, in configurations with LNET routers and TCP
1024 Description: incorrect data written to files on OSTs
1025 Details : In certain high-load conditions incorrect data may be written
1026 to files on the OST when using TCP networks.
1028 ------------------------------------------------------------------------------
1030 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
1032 - rework CDEBUG messages rate-limiting mechanism b=10375
1033 - add per-socket tunables for socklnd if the kernel is patched b=10327
1035 ------------------------------------------------------------------------------
1037 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
1039 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
1040 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
1041 - quiet LNET startup LNI message for liblustre b=10128
1042 - Better console error messages if 'ip2nets' can't match an IP address
1043 - Fixed overflow/use-before-set bugs in linux-time.h
1044 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
1045 - LNET teardown failed an assertion about the route table being empty
1046 - Fixed a crash in LNetEQPoll(<invalid handle>)
1047 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
1048 - improve debug message for liblustre/Catamount nodes (b=10116)
1050 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
1051 * Configuration change for the XT3
1052 The PTLLND is now used to run Lustre over Portals on the XT3.
1053 The configure option(s) --with-cray-portals are no longer
1054 used. Rather --with-portals=<path-to-portals-includes> is
1055 used to enable building on the XT3. In addition to enable
1056 XT3 specific features the option --enable-cray-xt3 must be
1059 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
1060 * Portals has been removed, replaced by LNET.
1061 LNET is new networking infrastructure for Lustre, it includes a
1062 reorganized network configuration mode (see the user
1063 documentation for full details) as well as support for routing
1064 between different network fabrics. Lustre Networking Devices
1065 (LNDS) for the supported network fabrics have also been created
1066 for this new infrastructure.
1068 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
1073 Frequency : rare (large Voltaire clusters only)
1075 Description: the default number of reserved transmit descriptors was too low
1076 for some large clusters
1077 Details : As a workaround, the number was increased. A proper fix includes
1080 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
1085 Frequency : occasional (large-scale events, cluster reboot, network failure)
1087 Description: too many error messages on console obscure actual problem and
1088 can slow down/panic server, or cause recovery to fail repeatedly
1089 Details : enable rate-limiting of console error messages, and some messages
1090 that were console errors now only go to the kernel log
1092 Severity : enhancement
1094 Description: add /proc/sys/portals/catastrophe entry which will report if
1095 that node has previously LBUGged
1097 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
1099 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
1101 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
1103 - handle error return code in kranal_check_fma_rx() (5915,6054)
1105 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
1107 - update vibnal (Voltaire IB NAL)
1108 - update gmnal (Myrinet NAL), gmnalid
1110 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
1112 * Landed portals:b_port_step as follows...
1114 - removed CFS_DECL_SPIN*
1115 just use 'spinlock_t' and initialise with spin_lock_init()
1117 - removed CFS_DECL_MUTEX*
1118 just use 'struct semaphore' and initialise with init_mutex()
1120 - removed CFS_DECL_RWSEM*
1121 just use 'struct rw_semaphore' and initialise with init_rwsem()
1123 - renamed cfs_sleep_chan -> cfs_waitq
1124 cfs_sleep_link -> cfs_waitlink
1126 - fixed race in linux version of arch-independent socknal
1127 (the ENOMEM/EAGAIN decision).
1129 - Didn't fix problems in Darwin version of arch-independent socknal
1130 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
1132 - removed libcfs types from non-socknal header files (only some types
1133 in the header files had been changed; the .c files hadn't been