1 2011-06-30 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 and 1.5.3.1,
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
12 Jira : LU-388, LU-303, TT-121
13 Description: Add support for OFED 1.5.3.1
15 -------------------------------------------------------------------------------
17 2011-05-12 Oracle, Inc.
19 * Support for networks:
20 socklnd - any kernel supported by Lustre,
21 qswlnd - Qsnet kernel modules 5.20 and later,
22 openiblnd - IbGold 1.8.2,
23 o2iblnd - OFED 1.3, 1.4.1, 1.4.2, 1.5.1 and 1.5.2
24 viblnd - Voltaire ibhost 3.4.5 and later,
25 ciblnd - Topspin 3.2.0,
26 iiblnd - Infiniserv 3.3 + PathBits patch,
27 gmlnd - GM 2.1.22 and later,
28 mxlnd - MX 1.2.10 or later,
29 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
33 Description: fix contention on ksock_tx_t
37 Description: dynamically grow/shrink connd threads pool
41 Description: ptlrpcd stuck in lnet allocations under memory pressure
42 Details : Set PF_MEMALLOC on outgoing path to prevent deadlock on memory
47 Description: fix o2iblnd v2 regression of credit deadlock with v1 peers
52 Description: lctl ping of 1.6.6 server from 1.8 router causes protocol error
54 -------------------------------------------------------------------------------
56 2010-10-29 Oracle, Inc.
58 * Support for networks:
59 socklnd - any kernel supported by Lustre,
60 qswlnd - Qsnet kernel modules 5.20 and later,
61 openiblnd - IbGold 1.8.2,
62 o2iblnd - OFED 1.3, 1.4.1, 1.4.2, 1.5.1 and 1.5.2
63 viblnd - Voltaire ibhost 3.4.5 and later,
64 ciblnd - Topspin 3.2.0,
65 iiblnd - Infiniserv 3.3 + PathBits patch,
66 gmlnd - GM 2.1.22 and later,
67 mxlnd - MX 1.2.10 or later,
68 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
72 Description: MDS node unresponsive
73 Details : improve hash distribution, doubled hash size for the lnet cookie
76 -------------------------------------------------------------------------------
78 2010-07-31 Oracle, Inc.
80 * Support for networks:
81 socklnd - any kernel supported by Lustre,
82 qswlnd - Qsnet kernel modules 5.20 and later,
83 openiblnd - IbGold 1.8.2,
84 o2iblnd - OFED 1.3, 1.4.1, 1.4.2 and 1.5.1
85 viblnd - Voltaire ibhost 3.4.5 and later,
86 ciblnd - Topspin 3.2.0,
87 iiblnd - Infiniserv 3.3 + PathBits patch,
88 gmlnd - GM 2.1.22 and later,
89 mxlnd - MX 1.2.10 or later,
90 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
94 Description: With peer health detection, o2iblnd makes only one attempt to
95 reconnect which is not enough with nodes running lustre 1.6
96 because of proto version mismatch. Fix o2iblnd to retry one more
99 Severity : enhancement
101 Description: Quiet some LNET messages
103 Severity : enhancement
105 Description: Add OFED 1.5.1 support
107 Severity : enhancement
109 Description: The peer health code lacked some important debugging info in
110 lnd_query code paths. We've added necessary debug prints,
111 not just for bug 21678, but also for future troubleshooting.
113 -------------------------------------------------------------------------------
115 2010-04-30 Oracle, Inc.
117 * Support for networks:
118 socklnd - any kernel supported by Lustre,
119 qswlnd - Qsnet kernel modules 5.20 and later,
120 openiblnd - IbGold 1.8.2,
121 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, 1.3, 1.4.1, and 1.4.2
122 viblnd - Voltaire ibhost 3.4.5 and later,
123 ciblnd - Topspin 3.2.0,
124 iiblnd - Infiniserv 3.3 + PathBits patch,
125 gmlnd - GM 2.1.22 and later,
126 mxlnd - MX 1.2.10 or later,
127 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
131 Description: lst: check # of remained RPCs before aborting
132 Details : lstcon_rpc_trans_postwait() calls lstcon_rpc_trans_abort() only
133 when the transaction is timeout, so if we got "end_session" to
134 interrupt waiting on transaction, then we can hit the assertion
135 failure ASSERTION(crpc->crp_stamp != 0)
139 Description: print more debug info for timedout ZC-req
140 Details : Print more information for timedout ZC-req and partial
141 received connection. Close connection for timedout ZC-req
142 Always send ZC_ACK on non-blocking connection(BULK_IN)
146 Description: Adding WIRE_ATTR attribute to LNET types
147 Details : LST nodes on different platforms might not communicate well
148 due to the lack of WIRE_ATTR attribute in some LNET structures
149 traversing network. The patch fixes the problem by adding
150 WIRE_ATTR where needed.
154 Description: hash MEs on RDMA portal
155 Details : RDMA portal can have very long ME list on client side, which
156 will trigger soft lockup because of long searching on list.
157 Hash MEs on RDMA portal can resolve this problem.
161 Description: fix for double release of ibc_lock in o2iblnd
162 Details : Re-acquire ibc_lock in kiblnd_post_tx_locked(). Add extra
163 reference to conn before calling kiblnd_post_tx_locked()
164 to avoid scenario when conn disappears inside
165 kiblnd_post_tx_locked().
167 -------------------------------------------------------------------------------
168 2010-01-29 Sun Microsystems, Inc.
170 * Support for networks:
171 socklnd - any kernel supported by Lustre,
172 qswlnd - Qsnet kernel modules 5.20 and later,
173 openiblnd - IbGold 1.8.2,
174 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, 1.3, 1.4.1, and 1.4.2
175 viblnd - Voltaire ibhost 3.4.5 and later,
176 ciblnd - Topspin 3.2.0,
177 iiblnd - Infiniserv 3.3 + PathBits patch,
178 gmlnd - GM 2.1.22 and later,
179 mxlnd - MX 1.2.10 or later,
180 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
184 Description: should update lp_alive for non-router peers.
186 Severity : enhancement
188 Description: LNet router shuffler.
190 Severity : enhancement
192 Description: LNet fine grain routing support.
196 Description: router checker stops working when system wall clock goes backward
197 Details : use monotonic timing source instead of system wall clock time.
199 Severity : enhancement
201 Description: avoid asymmetrical router failures
203 Severity : enhancement
205 Description: multiple-instance support for kptllnd
209 Description: ksocknal_close_conn_locked connection race
210 Details : A race was possible when ksocknal_create_conn calls
211 ksocknal_close_conn_locked for already closed conn.
213 Severity : enhancement
215 Description: port router pinger to userspace
219 Description: kptllnd HELLO protocol deadlock
220 Details : kptllnd HELLO protocol doesn't run to completion in finite time
224 Description: LNet selftest fixes and enhancements
226 Severity : enhancement
228 Description: allow a test node to be a member of multiple test groups
230 Severity : enhancement
232 Description: MXLND: eliminate hosts file, use arp for peer nic_id resolution
233 Details : an update from the upstream developer Scott Atchley.
236 -------------------------------------------------------------------------------
237 2009-07-31 Sun Microsystems, Inc.
239 * Support for networks:
240 socklnd - any kernel supported by Lustre,
241 qswlnd - Qsnet kernel modules 5.20 and later,
242 openiblnd - IbGold 1.8.2,
243 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, 1.3, and 1.4.1
244 viblnd - Voltaire ibhost 3.4.5 and later,
245 ciblnd - Topspin 3.2.0,
246 iiblnd - Infiniserv 3.3 + PathBits patch,
247 gmlnd - GM 2.1.22 and later,
248 mxlnd - MX 1.2.1 or later,
249 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
253 Description: router_proc.c is rewritten to use sysctl-interface for parameters
254 residing in /proc/sys/lnet
256 Severity : enhancement
258 Description: add a new LND optiion to control peer buffer credits on routers
262 Description: Fixing deadlock in usocklnd
263 Details : A deadlock was possible in usocklnd due to race condition while
264 tearing connection down. The problem resulted from erroneous
265 assumption that lnet_finalize() could have been called holding
266 some lnd-level locks.
269 Bugzilla : 13621, 15983
270 Description: Protocol V2 of o2iblnd
271 Details : o2iblnd V2 has several new features:
272 . map-on-demand: map-on-demand is disabled by default, it can
273 be enabled by using modparam "map_on_demand=@value@", @value@
274 should >= 0 and < 256, 0 will disable map-on-demand, any other
275 valid value will enable map-on-demand.
276 Oi2blnd will create FMR or physical MR for RDMA if fragments of
278 Enable map-on-demand will take less memory for new connection,
279 but a little more CPU for RDMA.
280 . iWARP : to support iWARP, please enable map-on-demand, 32 and 64
281 are recommanded value. iWARP will probably fail for value >=128.
282 . OOB NOOP message: to resolve deadlock on router.
283 . tunable peer_credits_hiw: (high water to return credits),
284 default value of peer_credits_hiw equals to (peer_credits -1),
285 user can change it between peer_credits/2 and (peer_credits - 1).
286 Lower value is recommended for high latency network.
287 . tunable message queue size: it always equals to peer_credits,
288 higher value is recommended for high latency network.
289 . It's compatible with earlier version of o2iblnd
293 Description: Fixing 'running out of ports' issue
294 Details : Add a delay before next reconnect attempt in ksocklnd in
295 the case of lost race. Limit the frequency of query-requests
296 in lnet. Improved handling of 'dead peer' notifications in
301 Description: Change ptllnd timeout and watchdog timers
302 Details : Add ptltrace_on_nal_failed and bump ptllnd timeout to match
303 Portals wire timeout.
307 Description: One down Lustre FS hangs ALL mounted Lustre filesystems
308 Details : Shared routing enhancements - peer health detection.
312 Description: IB path MTU mistakenly set to 1st path MTU when ib_mtu is off
313 Details : See comment 46 in bug 11245 for details - it's indeed a bug
314 introduced by the original 11245 fix.
318 Description: uptllnd credit overflow fix
319 Details : kptl_msg_t::ptlm_credits could be overflown by uptllnd since
324 Description: socklnd protocol version 3
325 Details : With current protocol V2, connections on router can be
326 blocked and can't receive any incoming messages when there is no
327 more router buffer, so ZC-ACK can't be handled (LNet message
328 can't be finalized) and will cause deadlock on router.
329 Protocol V3 has a dedicated connection for emergency messages
330 like ZC-ACK to router, messages on this dedicated connection
331 don't need any credit so will never be blocked. Also, V3 can send
332 keepalive ping in specified period for router healthy checking.
334 -------------------------------------------------------------------------------
335 12-31-2008 Sun Microsystems, Inc.
337 * Support for networks:
338 socklnd - any kernel supported by Lustre,
339 qswlnd - Qsnet kernel modules 5.20 and later,
340 openiblnd - IbGold 1.8.2,
341 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, and 1.3
342 viblnd - Voltaire ibhost 3.4.5 and later,
343 ciblnd - Topspin 3.2.0,
344 iiblnd - Infiniserv 3.3 + PathBits patch,
345 gmlnd - GM 2.1.22 and later,
346 mxlnd - MX 1.2.1 or later,
347 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
356 Description: workaround for OOM from o2iblnd
357 Details : OFED needs allocate big chunk of memory for QP while creating
358 connection for o2iblnd, OOM can happen if no such a contiguous
360 QP size is decided by concurrent_sends and max_fragments of
361 o2iblnd, now we permit user to specify smaller value for
362 concurrent_sends of o2iblnd(i.e: concurrent_sends=7), which
363 will decrease memory block size required by creating QP.
367 Description: Support Zerocopy receive of Chelsio device
368 Details : Chelsio driver can support zerocopy for iov[1] if it's
369 contiguous and large enough.
373 Description: fix credit flow deadlock in uptllnd
377 Description: finalize network operation in reasonable time
378 Details : conf-sanity test_32a couldn't stop ost and mds because it
379 tried to access non-existent peer and tcp connect took
380 quite long before timing out.
384 Description: Continuous recovery on 33 of 413 nodes after lustre oss failure
385 Details : Lost reference on conn prevents peer from being destroyed, which
386 could prevent new peer creation if peer count has reached upper
391 Description: LNET Selftest results in Soft lockup on OSS CPU
392 Details : only hits when 8 or more o2ib clients involved and a session is
393 torn down with 'lst end_session' without preceeding 'lst stop'.
397 Description: concurrent_sends in IB LNDs should not be changeable at run time
398 Details : concurrent_sends in IB LNDs should not be changeable at run time
402 Description: ptl_send_rpc hits LASSERT when ptl_send_buf fails
403 Details : only hits under out-of-memory situations
406 -------------------------------------------------------------------------------
408 2009-02-07 Sun Microsystems, Inc.
410 * Support for networks:
411 socklnd - any kernel supported by Lustre,
412 qswlnd - Qsnet kernel modules 5.20 and later,
413 openiblnd - IbGold 1.8.2,
414 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, and 1.3
415 viblnd - Voltaire ibhost 3.4.5 and later,
416 ciblnd - Topspin 3.2.0,
417 iiblnd - Infiniserv 3.3 + PathBits patch,
418 gmlnd - GM 2.1.22 and later,
419 mxlnd - MX 1.2.1 or later,
420 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
424 Description: workaround for OOM from o2iblnd
425 Details : OFED needs allocate big chunk of memory for QP while creating
426 connection for o2iblnd, OOM can happen if no such a contiguous
428 QP size is decided by concurrent_sends and max_fragments of
429 o2iblnd, now we permit user to specify smaller value for
430 concurrent_sends of o2iblnd(i.e: concurrent_sends=7), which
431 will decrease memory block size required by creating QP.
435 Description: Support Zerocopy receive of Chelsio device
436 Details : Chelsio driver can support zerocopy for iov[1] if it's
437 contiguous and large enough.
441 Description: fix credit flow deadlock in uptllnd
445 Description: finalize network operation in reasonable time
446 Details : conf-sanity test_32a couldn't stop ost and mds because it
447 tried to access non-existent peer and tcp connect took
448 quite long before timing out.
452 Description: Continuous recovery on 33 of 413 nodes after lustre oss failure
453 Details : Lost reference on conn prevents peer from being destroyed, which
454 could prevent new peer creation if peer count has reached upper
459 Description: LNET Selftest results in Soft lockup on OSS CPU
460 Details : only hits when 8 or more o2ib clients involved and a session is
461 torn down with 'lst end_session' without preceeding 'lst stop'.
465 Description: concurrent_sends in IB LNDs should not be changeable at run time
466 Details : concurrent_sends in IB LNDs should not be changeable at run time
468 -------------------------------------------------------------------------------
470 11-03-2008 Sun Microsystems, Inc.
472 * Support for networks:
473 socklnd - any kernel supported by Lustre,
474 qswlnd - Qsnet kernel modules 5.20 and later,
475 openiblnd - IbGold 1.8.2,
476 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, and 1.3
477 viblnd - Voltaire ibhost 3.4.5 and later,
478 ciblnd - Topspin 3.2.0,
479 iiblnd - Infiniserv 3.3 + PathBits patch,
480 gmlnd - GM 2.1.22 and later,
481 mxlnd - MX 1.2.1 or later,
482 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
486 Description: ptl_send_rpc hits LASSERT when ptl_send_buf fails
487 Details : only hits under out-of-memory situations
489 -------------------------------------------------------------------------------
492 04-26-2008 Sun Microsystems, Inc.
494 * Support for networks:
495 socklnd - any kernel supported by Lustre,
496 qswlnd - Qsnet kernel modules 5.20 and later,
497 openiblnd - IbGold 1.8.2,
498 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5
499 viblnd - Voltaire ibhost 3.4.5 and later,
500 ciblnd - Topspin 3.2.0,
501 iiblnd - Infiniserv 3.3 + PathBits patch,
502 gmlnd - GM 2.1.22 and later,
503 mxlnd - MX 1.2.1 or later,
504 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
508 Description: excessive debug information removed
509 Details : excessive debug information removed
513 Description: ksocknal_create_conn() hit ASSERTION during connection race
514 Details : ksocknal_create_conn() hit ASSERTION during connection race
518 Description: ksocknal_send_hello() hit ASSERTION while connecting race
519 Details : ksocknal_send_hello() hit ASSERTION while connecting race
523 Description: o2iblnd/ptllnd credit deadlock in a routed config.
524 Details : o2iblnd/ptllnd credit deadlock in a routed config.
528 Description: High load after starting lnet
529 Details : gmlnd should sleep in rx thread in interruptible way. Otherwise,
530 uptime utility reports high load that looks confusingly.
534 Description: ksocklnd fails to establish connection if accept_port is high
535 Details : PID remapping must not be done for active (outgoing) connections
537 --------------------------------------------------------------------------------
539 2008-01-11 Sun Microsystems, Inc.
541 * Support for networks:
542 socklnd - any kernel supported by Lustre,
543 qswlnd - Qsnet kernel modules 5.20 and later,
544 openiblnd - IbGold 1.8.2,
545 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5
546 viblnd - Voltaire ibhost 3.4.5 and later,
547 ciblnd - Topspin 3.2.0,
548 iiblnd - Infiniserv 3.3 + PathBits patch,
549 gmlnd - GM 2.1.22 and later,
550 mxlnd - MX 1.2.1 or later,
551 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
554 Description: liblustre network error
555 Details : liblustre clients should understand LNET_ACCEPT_PORT environment
556 variable even if they don't start lnet acceptor.
560 Description: Strange message from lnet (Ignoring prediction from the future)
561 Details : Incorrect calculation of peer's last_alive value in ksocklnd
563 --------------------------------------------------------------------------------
565 2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
567 * Support for networks:
568 socklnd - any kernel supported by Lustre,
569 qswlnd - Qsnet kernel modules 5.20 and later,
570 openiblnd - IbGold 1.8.2,
571 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5.
572 viblnd - Voltaire ibhost 3.4.5 and later,
573 ciblnd - Topspin 3.2.0,
574 iiblnd - Infiniserv 3.3 + PathBits patch,
575 gmlnd - GM 2.1.22 and later,
576 mxlnd - MX 1.2.1 or later,
577 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
581 Description: ASSERTION(me == md->md_me) failed in lnet_match_md()
585 Description: increase send queue size for ciblnd/openiblnd
589 Description: new userspace socklnd
590 Details : Old userspace tcpnal that resided in lnet/ulnds/socklnd replaced
591 with new one - usocklnd.
593 Severity : enhancement
595 Description: Console message flood
596 Details : Make cdls ratelimiting more tunable by adding several tunable in
597 procfs /proc/sys/lnet/console_{min,max}_delay_centisecs and
598 /proc/sys/lnet/console_backoff.
600 --------------------------------------------------------------------------------
602 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
604 * Support for networks:
605 socklnd - any kernel supported by Lustre,
606 qswlnd - Qsnet kernel modules 5.20 and later,
607 openiblnd - IbGold 1.8.2,
608 o2iblnd - OFED 1.1 and 1.2,
609 viblnd - Voltaire ibhost 3.4.5 and later,
610 ciblnd - Topspin 3.2.0,
611 iiblnd - Infiniserv 3.3 + PathBits patch,
612 gmlnd - GM 2.1.22 and later,
613 mxlnd - MX 1.2.1 or later,
614 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
618 Description: /proc/sys/lnet has non-sysctl entries
619 Details : Updating dump_kernel/daemon_file/debug_mb to use sysctl variables
623 Description: TOE Kernel panic by ksocklnd
624 Details : offloaded sockets provide their own implementation of sendpage,
625 can't call tcp_sendpage() directly
629 Description: kibnal_shutdown() doesn't finish; lconf --cleanup hangs
630 Details : races between lnd_shutdown and peer creation prevent
631 lnd_shutdown from finishing.
635 Description: open files rlimit 1024 reached while liblustre testing
636 Details : ulnds/socklnd must close open socket after unsuccessful
641 Description: build error
642 Details : fix typos in gmlnd, ptllnd and viblnd
644 ------------------------------------------------------------------------------
646 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
648 * Support for networks:
649 socklnd - kernels up to 2.6.16,
650 qswlnd - Qsnet kernel modules 5.20 and later,
651 openiblnd - IbGold 1.8.2,
652 o2iblnd - OFED 1.1 and 1.2
653 viblnd - Voltaire ibhost 3.4.5 and later,
654 ciblnd - Topspin 3.2.0,
655 iiblnd - Infiniserv 3.3 + PathBits patch,
656 gmlnd - GM 2.1.22 and later,
657 mxlnd - MX 1.2.1 or later,
658 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
660 2007-06-21 Cluster File Systems, Inc. <info@clusterfs.com>
662 * Support for networks:
663 socklnd - kernels up to 2.6.16,
664 qswlnd - Qsnet kernel modules 5.20 and later,
665 openiblnd - IbGold 1.8.2,
667 viblnd - Voltaire ibhost 3.4.5 and later,
668 ciblnd - Topspin 3.2.0,
669 iiblnd - Infiniserv 3.3 + PathBits patch,
670 gmlnd - GM 2.1.22 and later,
671 mxlnd - MX 1.2.1 or later,
672 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
676 Description: Initialize cpumask before use
680 Description: ASSERTION failures when upgrading to the patchless zero-copy
682 Details : This bug affects "rolling upgrades", causing an inconsistent
683 protocol version negotiation and subsequent assertion failure
684 during rolling upgrades after the first wave of upgrades.
688 Details : Change "dropped message" CERRORs to D_NETERROR so they are
689 logged instead of creating "console chatter" when a lustre
690 timeout races with normal RPC completion.
693 Details : lnet_clear_peer_table can wait forever if user forgets to
697 Details : libcfs_id2str should check pid against LNET_PID_ANY.
701 Description: added LNET self test
702 Details : landing b_self_test
707 Description: cfs_duration_{u,n}sec() wrongly calculate nanosecond part of
709 Details : do_div() macro is used incorrectly.
711 2007-04-23 Cluster File Systems, Inc. <info@clusterfs.com>
715 Description: make panic on lbug configurable
719 Description: Add OFED1.2 support to o2iblnd
720 Details : o2iblnd depends on OFED's modules, if out-tree OFED's modules
721 are installed (other than kernel's in-tree infiniband), there
722 could be some problem while insmod o2iblnd (mismatch CRC of
724 If extra Module.symvers is supported in kernel (i.e, 2.6.17),
725 this link provides solution:
726 https://bugs.openfabrics.org/show_bug.cgi?id=355
727 if extra Module.symvers is not supported in kernel, we will
728 have to run the script in bug 12316 to update
729 $LINUX/module.symvers before building o2iblnd.
730 More details about this are in bug 12316.
732 ------------------------------------------------------------------------------
734 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
735 * version 1.4.10 / 1.6.0
736 * Support for networks:
737 socklnd - kernels up to 2.6.16,
738 qswlnd - Qsnet kernel modules 5.20 and later,
739 openiblnd - IbGold 1.8.2,
741 viblnd - Voltaire ibhost 3.4.5 and later,
742 ciblnd - Topspin 3.2.0,
743 iiblnd - Infiniserv 3.3 + PathBits patch,
744 gmlnd - GM 2.1.22 and later,
745 mxlnd - MX 1.2.1 or later,
746 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
750 Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
751 possibly accessed in kptllnd_shutdown. Ptllnd should init
752 kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
756 Description: gmlnd ignored some transmit errors when finalizing lnet messages.
760 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
764 Description: the_lnet.ln_finalizing was not set when the current thread is
765 about to complete messages. It only affects multi-threaded
771 Description: Changed the default kqswlnd ntxmsg=512
776 Description: Assertion failure in kernel ptllnd caused by posting passive
777 bulk buffers before connection establishment complete.
782 Description: A race in kernel ptllnd between deleting a peer and posting
783 new communications for it could hang communications -
784 manifesting as "Unexpectedly long timeout" messages.
789 Description: Kernel ptllnd lock ordering issue could hang a node.
794 Description: node crash on socket teardown race
797 Frequency : 'lctl peer_list' issued on a mx net
799 Description: Enable lctl's peer_list for MXLND
802 Frequency : after Ptllnd timeouts and portals congestion
804 Description: Credit overflows
805 Details : This was a bug in ptllnd connection establishment. The fix
806 implements better peer stamps to disambiguate connection
807 establishment and ensure both peers enter the credit flow
808 state machine consistently.
813 Description: kptllnd didn't propagate some network errors up to LNET
814 Details : This bug was spotted while investigating 11394. The fix
815 ensures network errors on sends and bulk transfers are
816 propagated to LNET/lustre correctly.
818 Severity : enhancement
820 Description: Fixed console chatter in case of -ETIMEDOUT.
822 Severity : enhancement
824 Description: Added D_NETTRACE for recording network packet history
825 (initially only for ptllnd). Also a separate userspace
826 ptllnd facility to gather history which should really be
827 covered by D_NETTRACE too, if only CDEBUG recorded history in
833 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
834 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
835 callback can occur before a connection has actually been
836 established. This caused an assertion failure previously.
838 Severity : enhancement
840 Description: Multiple instances for o2iblnd
841 Details : Allow multiple instances of o2iblnd to enable networking over
842 multiple HCAs and routing between them.
846 Description: lnet deadlock in router_checker
847 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
848 into BH locks to eliminate potential deadlock caused by
849 ksocknal_data_ready() preempting code holding these locks.
853 Description: Millions of failed socklnd connection attempts cause a very slow FS
854 Details : added a new route flag ksnr_scheduled to distinguish from
855 ksnr_connecting, so that a peer connection request is only turned
856 down for race concerns when an active connection to the same peer
857 is under progress (instead of just being scheduled).
859 ------------------------------------------------------------------------------
861 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
863 * Support for networks:
864 socklnd - kernels up to 2.6.16
865 qswlnd - Qsnet kernel modules 5.20 and later
866 openiblnd - IbGold 1.8.2
868 viblnd - Voltaire ibhost 3.4.5 and later
869 ciblnd - Topspin 3.2.0
870 iiblnd - Infiniserv 3.3 + PathBits patch
871 gmlnd - GM 2.1.22 and later
872 mxlnd - MX 1.2.1 or later
873 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
876 Severity : major on XT3
878 Description: libcfs overwrites /proc/sys/portals
879 Details : libcfs created a symlink from /proc/sys/portals to
880 /proc/sys/lnet for backwards compatibility. This is no
881 longer required and makes the Cray portals /proc variables
886 Description: OFED FMR API change
887 Details : This changes parameter usage to reflect a change in
888 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
889 that FMR support is only used in experimental versions of the
890 o2iblnd - this change does not affect standard usage at all.
892 Severity : enhancement
894 Description: new ko2iblnd module parameter: ib_mtu
895 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
896 HCAs. You can avoid this problem by setting the MTU to 1024
897 using this module parameter.
899 Severity : enhancement
900 Bugzilla : 11118/11620
901 Description: ptllnd small request message buffer alignment fix
902 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
903 Round up small message size on sends in case this option
904 is not supported. 11620 was a defect in the initial
905 implementation which effectively asserted all peers had to be
906 running the correct protocol version which was fixed by always
907 NAK-ing such requests and handling any misalignments they
912 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
913 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
914 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
916 Severity : enhancement
918 Description: Patchless ZC(zero copy) socklnd
919 Details : New protocol for socklnd, socklnd can support zero copy without
920 kernel patch, it's compatible with old socklnd. Checksum is
921 moved from tunables to modparams.
925 Description: When ksocknal_del_peer() is called upon a peer whose
926 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
927 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
930 Frequency : when ptlrpc is under heavy use and runs out of request buffer
932 Description: In lnet_match_blocked_msg(), md can be used without holding a
936 Frequency : very rarely
938 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
939 If connd connects a route which has been closed by
940 ksocknal_shutdown(), ksocknal_create_routes() may create new
941 routes which hold references on the peer, causing shutdown
942 process to wait for peer to disappear forever.
944 Severity : enhancement
946 Description: Dump XT3 portals traces on kptllnd timeout
947 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
948 dump Cray portals debug traces to a file. The kptllnd module
949 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
950 is the basename of the dump file.
953 Frequency : infrequent
955 Description: kernel ptllnd fix bug in connection re-establishment
956 Details : Kernel ptllnd could produce protocol errors e.g. illegal
957 matchbits and/or violate the credit flow protocol when trying
958 to re-establish a connection with a peer after an error or
961 Severity : enhancement
963 Description: Allow /proc/sys/lnet/debug to be set symbolically
964 Details : Allow debug and subsystem debug values to be read/set by name
965 in addition to numerically, for ease of use.
968 Frequency : only in configurations with LNET routers
970 Description: routes automatically marked down and recovered
971 Details : In configurations with LNET routers if a router fails routers
972 now actively try to recover routes that are down, unless they
973 are marked down by an administrator.
975 ------------------------------------------------------------------------------
977 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
980 Frequency : very rarely, in configurations with LNET routers and TCP
982 Description: incorrect data written to files on OSTs
983 Details : In certain high-load conditions incorrect data may be written
984 to files on the OST when using TCP networks.
986 ------------------------------------------------------------------------------
988 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
990 - rework CDEBUG messages rate-limiting mechanism b=10375
991 - add per-socket tunables for socklnd if the kernel is patched b=10327
993 ------------------------------------------------------------------------------
995 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
997 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
998 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
999 - quiet LNET startup LNI message for liblustre b=10128
1000 - Better console error messages if 'ip2nets' can't match an IP address
1001 - Fixed overflow/use-before-set bugs in linux-time.h
1002 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
1003 - LNET teardown failed an assertion about the route table being empty
1004 - Fixed a crash in LNetEQPoll(<invalid handle>)
1005 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
1006 - improve debug message for liblustre/Catamount nodes (b=10116)
1008 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
1009 * Configuration change for the XT3
1010 The PTLLND is now used to run Lustre over Portals on the XT3.
1011 The configure option(s) --with-cray-portals are no longer
1012 used. Rather --with-portals=<path-to-portals-includes> is
1013 used to enable building on the XT3. In addition to enable
1014 XT3 specific features the option --enable-cray-xt3 must be
1017 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
1018 * Portals has been removed, replaced by LNET.
1019 LNET is new networking infrastructure for Lustre, it includes a
1020 reorganized network configuration mode (see the user
1021 documentation for full details) as well as support for routing
1022 between different network fabrics. Lustre Networking Devices
1023 (LNDS) for the supported network fabrics have also been created
1024 for this new infrastructure.
1026 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
1031 Frequency : rare (large Voltaire clusters only)
1033 Description: the default number of reserved transmit descriptors was too low
1034 for some large clusters
1035 Details : As a workaround, the number was increased. A proper fix includes
1038 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
1043 Frequency : occasional (large-scale events, cluster reboot, network failure)
1045 Description: too many error messages on console obscure actual problem and
1046 can slow down/panic server, or cause recovery to fail repeatedly
1047 Details : enable rate-limiting of console error messages, and some messages
1048 that were console errors now only go to the kernel log
1050 Severity : enhancement
1052 Description: add /proc/sys/portals/catastrophe entry which will report if
1053 that node has previously LBUGged
1055 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
1057 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
1059 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
1061 - handle error return code in kranal_check_fma_rx() (5915,6054)
1063 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
1065 - update vibnal (Voltaire IB NAL)
1066 - update gmnal (Myrinet NAL), gmnalid
1068 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
1070 * Landed portals:b_port_step as follows...
1072 - removed CFS_DECL_SPIN*
1073 just use 'spinlock_t' and initialise with spin_lock_init()
1075 - removed CFS_DECL_MUTEX*
1076 just use 'struct semaphore' and initialise with init_mutex()
1078 - removed CFS_DECL_RWSEM*
1079 just use 'struct rw_semaphore' and initialise with init_rwsem()
1081 - renamed cfs_sleep_chan -> cfs_waitq
1082 cfs_sleep_link -> cfs_waitlink
1084 - fixed race in linux version of arch-independent socknal
1085 (the ENOMEM/EAGAIN decision).
1087 - Didn't fix problems in Darwin version of arch-independent socknal
1088 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
1090 - removed libcfs types from non-socknal header files (only some types
1091 in the header files had been changed; the .c files hadn't been