1 xxxx-xx-xx Oracle, Inc.
3 * Support for networks:
4 socklnd - any kernel supported by Lustre,
5 qswlnd - Qsnet kernel modules 5.20 and later,
6 openiblnd - IbGold 1.8.2,
7 o2iblnd - OFED 1.3, 1.4.1, 1.4.2, 1.5.1 and 1.5.2
8 viblnd - Voltaire ibhost 3.4.5 and later,
9 ciblnd - Topspin 3.2.0,
10 iiblnd - Infiniserv 3.3 + PathBits patch,
11 gmlnd - GM 2.1.22 and later,
12 mxlnd - MX 1.2.10 or later,
13 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
17 Description: fix o2iblnd v2 regression of credit deadlock with v1 peers (bug 14425).
21 Description: lctl ping of 1.6.6 server from 1.8 router causes protocol error
23 -------------------------------------------------------------------------------
25 2010-10-29 Oracle, Inc.
27 * Support for networks:
28 socklnd - any kernel supported by Lustre,
29 qswlnd - Qsnet kernel modules 5.20 and later,
30 openiblnd - IbGold 1.8.2,
31 o2iblnd - OFED 1.3, 1.4.1, 1.4.2, 1.5.1 and 1.5.2
32 viblnd - Voltaire ibhost 3.4.5 and later,
33 ciblnd - Topspin 3.2.0,
34 iiblnd - Infiniserv 3.3 + PathBits patch,
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
41 Description: MDS node unresponsive
42 Details : improve hash distribution, doubled hash size for the lnet cookie
45 -------------------------------------------------------------------------------
47 2010-07-31 Oracle, Inc.
49 * Support for networks:
50 socklnd - any kernel supported by Lustre,
51 qswlnd - Qsnet kernel modules 5.20 and later,
52 openiblnd - IbGold 1.8.2,
53 o2iblnd - OFED 1.3, 1.4.1, 1.4.2 and 1.5.1
54 viblnd - Voltaire ibhost 3.4.5 and later,
55 ciblnd - Topspin 3.2.0,
56 iiblnd - Infiniserv 3.3 + PathBits patch,
57 gmlnd - GM 2.1.22 and later,
58 mxlnd - MX 1.2.10 or later,
59 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
63 Description: With peer health detection, o2iblnd makes only one attempt to
64 reconnect which is not enough with nodes running lustre 1.6
65 because of proto version mismatch. Fix o2iblnd to retry one more
68 Severity : enhancement
70 Description: Quiet some LNET messages
72 Severity : enhancement
74 Description: Add OFED 1.5.1 support
76 Severity : enhancement
78 Description: The peer health code lacked some important debugging info in
79 lnd_query code paths. We've added necessary debug prints,
80 not just for bug 21678, but also for future troubleshooting.
82 -------------------------------------------------------------------------------
84 2010-04-30 Oracle, Inc.
86 * Support for networks:
87 socklnd - any kernel supported by Lustre,
88 qswlnd - Qsnet kernel modules 5.20 and later,
89 openiblnd - IbGold 1.8.2,
90 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, 1.3, 1.4.1, and 1.4.2
91 viblnd - Voltaire ibhost 3.4.5 and later,
92 ciblnd - Topspin 3.2.0,
93 iiblnd - Infiniserv 3.3 + PathBits patch,
94 gmlnd - GM 2.1.22 and later,
95 mxlnd - MX 1.2.10 or later,
96 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
100 Description: lst: check # of remained RPCs before aborting
101 Details : lstcon_rpc_trans_postwait() calls lstcon_rpc_trans_abort() only
102 when the transaction is timeout, so if we got "end_session" to
103 interrupt waiting on transaction, then we can hit the assertion
104 failure ASSERTION(crpc->crp_stamp != 0)
108 Description: print more debug info for timedout ZC-req
109 Details : Print more information for timedout ZC-req and partial
110 received connection. Close connection for timedout ZC-req
111 Always send ZC_ACK on non-blocking connection(BULK_IN)
115 Description: Adding WIRE_ATTR attribute to LNET types
116 Details : LST nodes on different platforms might not communicate well
117 due to the lack of WIRE_ATTR attribute in some LNET structures
118 traversing network. The patch fixes the problem by adding
119 WIRE_ATTR where needed.
123 Description: hash MEs on RDMA portal
124 Details : RDMA portal can have very long ME list on client side, which
125 will trigger soft lockup because of long searching on list.
126 Hash MEs on RDMA portal can resolve this problem.
130 Description: fix for double release of ibc_lock in o2iblnd
131 Details : Re-acquire ibc_lock in kiblnd_post_tx_locked(). Add extra
132 reference to conn before calling kiblnd_post_tx_locked()
133 to avoid scenario when conn disappears inside
134 kiblnd_post_tx_locked().
136 -------------------------------------------------------------------------------
137 2010-01-29 Sun Microsystems, Inc.
139 * Support for networks:
140 socklnd - any kernel supported by Lustre,
141 qswlnd - Qsnet kernel modules 5.20 and later,
142 openiblnd - IbGold 1.8.2,
143 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, 1.3, 1.4.1, and 1.4.2
144 viblnd - Voltaire ibhost 3.4.5 and later,
145 ciblnd - Topspin 3.2.0,
146 iiblnd - Infiniserv 3.3 + PathBits patch,
147 gmlnd - GM 2.1.22 and later,
148 mxlnd - MX 1.2.10 or later,
149 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
153 Description: should update lp_alive for non-router peers.
155 Severity : enhancement
157 Description: LNet router shuffler.
159 Severity : enhancement
161 Description: LNet fine grain routing support.
165 Description: router checker stops working when system wall clock goes backward
166 Details : use monotonic timing source instead of system wall clock time.
168 Severity : enhancement
170 Description: avoid asymmetrical router failures
172 Severity : enhancement
174 Description: multiple-instance support for kptllnd
178 Description: ksocknal_close_conn_locked connection race
179 Details : A race was possible when ksocknal_create_conn calls
180 ksocknal_close_conn_locked for already closed conn.
182 Severity : enhancement
184 Description: port router pinger to userspace
188 Description: kptllnd HELLO protocol deadlock
189 Details : kptllnd HELLO protocol doesn't run to completion in finite time
193 Description: LNet selftest fixes and enhancements
195 Severity : enhancement
197 Description: allow a test node to be a member of multiple test groups
199 Severity : enhancement
201 Description: MXLND: eliminate hosts file, use arp for peer nic_id resolution
202 Details : an update from the upstream developer Scott Atchley.
205 -------------------------------------------------------------------------------
206 2009-07-31 Sun Microsystems, Inc.
208 * Support for networks:
209 socklnd - any kernel supported by Lustre,
210 qswlnd - Qsnet kernel modules 5.20 and later,
211 openiblnd - IbGold 1.8.2,
212 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, 1.3, and 1.4.1
213 viblnd - Voltaire ibhost 3.4.5 and later,
214 ciblnd - Topspin 3.2.0,
215 iiblnd - Infiniserv 3.3 + PathBits patch,
216 gmlnd - GM 2.1.22 and later,
217 mxlnd - MX 1.2.1 or later,
218 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
222 Description: router_proc.c is rewritten to use sysctl-interface for parameters
223 residing in /proc/sys/lnet
225 Severity : enhancement
227 Description: add a new LND optiion to control peer buffer credits on routers
231 Description: Fixing deadlock in usocklnd
232 Details : A deadlock was possible in usocklnd due to race condition while
233 tearing connection down. The problem resulted from erroneous
234 assumption that lnet_finalize() could have been called holding
235 some lnd-level locks.
238 Bugzilla : 13621, 15983
239 Description: Protocol V2 of o2iblnd
240 Details : o2iblnd V2 has several new features:
241 . map-on-demand: map-on-demand is disabled by default, it can
242 be enabled by using modparam "map_on_demand=@value@", @value@
243 should >= 0 and < 256, 0 will disable map-on-demand, any other
244 valid value will enable map-on-demand.
245 Oi2blnd will create FMR or physical MR for RDMA if fragments of
247 Enable map-on-demand will take less memory for new connection,
248 but a little more CPU for RDMA.
249 . iWARP : to support iWARP, please enable map-on-demand, 32 and 64
250 are recommanded value. iWARP will probably fail for value >=128.
251 . OOB NOOP message: to resolve deadlock on router.
252 . tunable peer_credits_hiw: (high water to return credits),
253 default value of peer_credits_hiw equals to (peer_credits -1),
254 user can change it between peer_credits/2 and (peer_credits - 1).
255 Lower value is recommended for high latency network.
256 . tunable message queue size: it always equals to peer_credits,
257 higher value is recommended for high latency network.
258 . It's compatible with earlier version of o2iblnd
262 Description: Fixing 'running out of ports' issue
263 Details : Add a delay before next reconnect attempt in ksocklnd in
264 the case of lost race. Limit the frequency of query-requests
265 in lnet. Improved handling of 'dead peer' notifications in
270 Description: Change ptllnd timeout and watchdog timers
271 Details : Add ptltrace_on_nal_failed and bump ptllnd timeout to match
272 Portals wire timeout.
276 Description: One down Lustre FS hangs ALL mounted Lustre filesystems
277 Details : Shared routing enhancements - peer health detection.
281 Description: IB path MTU mistakenly set to 1st path MTU when ib_mtu is off
282 Details : See comment 46 in bug 11245 for details - it's indeed a bug
283 introduced by the original 11245 fix.
287 Description: uptllnd credit overflow fix
288 Details : kptl_msg_t::ptlm_credits could be overflown by uptllnd since
293 Description: socklnd protocol version 3
294 Details : With current protocol V2, connections on router can be
295 blocked and can't receive any incoming messages when there is no
296 more router buffer, so ZC-ACK can't be handled (LNet message
297 can't be finalized) and will cause deadlock on router.
298 Protocol V3 has a dedicated connection for emergency messages
299 like ZC-ACK to router, messages on this dedicated connection
300 don't need any credit so will never be blocked. Also, V3 can send
301 keepalive ping in specified period for router healthy checking.
303 -------------------------------------------------------------------------------
304 12-31-2008 Sun Microsystems, Inc.
306 * Support for networks:
307 socklnd - any kernel supported by Lustre,
308 qswlnd - Qsnet kernel modules 5.20 and later,
309 openiblnd - IbGold 1.8.2,
310 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, and 1.3
311 viblnd - Voltaire ibhost 3.4.5 and later,
312 ciblnd - Topspin 3.2.0,
313 iiblnd - Infiniserv 3.3 + PathBits patch,
314 gmlnd - GM 2.1.22 and later,
315 mxlnd - MX 1.2.1 or later,
316 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
325 Description: workaround for OOM from o2iblnd
326 Details : OFED needs allocate big chunk of memory for QP while creating
327 connection for o2iblnd, OOM can happen if no such a contiguous
329 QP size is decided by concurrent_sends and max_fragments of
330 o2iblnd, now we permit user to specify smaller value for
331 concurrent_sends of o2iblnd(i.e: concurrent_sends=7), which
332 will decrease memory block size required by creating QP.
336 Description: Support Zerocopy receive of Chelsio device
337 Details : Chelsio driver can support zerocopy for iov[1] if it's
338 contiguous and large enough.
342 Description: fix credit flow deadlock in uptllnd
346 Description: finalize network operation in reasonable time
347 Details : conf-sanity test_32a couldn't stop ost and mds because it
348 tried to access non-existent peer and tcp connect took
349 quite long before timing out.
353 Description: Continuous recovery on 33 of 413 nodes after lustre oss failure
354 Details : Lost reference on conn prevents peer from being destroyed, which
355 could prevent new peer creation if peer count has reached upper
360 Description: LNET Selftest results in Soft lockup on OSS CPU
361 Details : only hits when 8 or more o2ib clients involved and a session is
362 torn down with 'lst end_session' without preceeding 'lst stop'.
366 Description: concurrent_sends in IB LNDs should not be changeable at run time
367 Details : concurrent_sends in IB LNDs should not be changeable at run time
371 Description: ptl_send_rpc hits LASSERT when ptl_send_buf fails
372 Details : only hits under out-of-memory situations
375 -------------------------------------------------------------------------------
377 2009-02-07 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
393 Description: workaround for OOM from o2iblnd
394 Details : OFED needs allocate big chunk of memory for QP while creating
395 connection for o2iblnd, OOM can happen if no such a contiguous
397 QP size is decided by concurrent_sends and max_fragments of
398 o2iblnd, now we permit user to specify smaller value for
399 concurrent_sends of o2iblnd(i.e: concurrent_sends=7), which
400 will decrease memory block size required by creating QP.
404 Description: Support Zerocopy receive of Chelsio device
405 Details : Chelsio driver can support zerocopy for iov[1] if it's
406 contiguous and large enough.
410 Description: fix credit flow deadlock in uptllnd
414 Description: finalize network operation in reasonable time
415 Details : conf-sanity test_32a couldn't stop ost and mds because it
416 tried to access non-existent peer and tcp connect took
417 quite long before timing out.
421 Description: Continuous recovery on 33 of 413 nodes after lustre oss failure
422 Details : Lost reference on conn prevents peer from being destroyed, which
423 could prevent new peer creation if peer count has reached upper
428 Description: LNET Selftest results in Soft lockup on OSS CPU
429 Details : only hits when 8 or more o2ib clients involved and a session is
430 torn down with 'lst end_session' without preceeding 'lst stop'.
434 Description: concurrent_sends in IB LNDs should not be changeable at run time
435 Details : concurrent_sends in IB LNDs should not be changeable at run time
437 -------------------------------------------------------------------------------
439 11-03-2008 Sun Microsystems, Inc.
441 * Support for networks:
442 socklnd - any kernel supported by Lustre,
443 qswlnd - Qsnet kernel modules 5.20 and later,
444 openiblnd - IbGold 1.8.2,
445 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, and 1.3
446 viblnd - Voltaire ibhost 3.4.5 and later,
447 ciblnd - Topspin 3.2.0,
448 iiblnd - Infiniserv 3.3 + PathBits patch,
449 gmlnd - GM 2.1.22 and later,
450 mxlnd - MX 1.2.1 or later,
451 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
455 Description: ptl_send_rpc hits LASSERT when ptl_send_buf fails
456 Details : only hits under out-of-memory situations
458 -------------------------------------------------------------------------------
461 04-26-2008 Sun Microsystems, Inc.
463 * Support for networks:
464 socklnd - any kernel supported by Lustre,
465 qswlnd - Qsnet kernel modules 5.20 and later,
466 openiblnd - IbGold 1.8.2,
467 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5
468 viblnd - Voltaire ibhost 3.4.5 and later,
469 ciblnd - Topspin 3.2.0,
470 iiblnd - Infiniserv 3.3 + PathBits patch,
471 gmlnd - GM 2.1.22 and later,
472 mxlnd - MX 1.2.1 or later,
473 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
477 Description: excessive debug information removed
478 Details : excessive debug information removed
482 Description: ksocknal_create_conn() hit ASSERTION during connection race
483 Details : ksocknal_create_conn() hit ASSERTION during connection race
487 Description: ksocknal_send_hello() hit ASSERTION while connecting race
488 Details : ksocknal_send_hello() hit ASSERTION while connecting race
492 Description: o2iblnd/ptllnd credit deadlock in a routed config.
493 Details : o2iblnd/ptllnd credit deadlock in a routed config.
497 Description: High load after starting lnet
498 Details : gmlnd should sleep in rx thread in interruptible way. Otherwise,
499 uptime utility reports high load that looks confusingly.
503 Description: ksocklnd fails to establish connection if accept_port is high
504 Details : PID remapping must not be done for active (outgoing) connections
506 --------------------------------------------------------------------------------
508 2008-01-11 Sun Microsystems, Inc.
510 * Support for networks:
511 socklnd - any kernel supported by Lustre,
512 qswlnd - Qsnet kernel modules 5.20 and later,
513 openiblnd - IbGold 1.8.2,
514 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5
515 viblnd - Voltaire ibhost 3.4.5 and later,
516 ciblnd - Topspin 3.2.0,
517 iiblnd - Infiniserv 3.3 + PathBits patch,
518 gmlnd - GM 2.1.22 and later,
519 mxlnd - MX 1.2.1 or later,
520 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
523 Description: liblustre network error
524 Details : liblustre clients should understand LNET_ACCEPT_PORT environment
525 variable even if they don't start lnet acceptor.
529 Description: Strange message from lnet (Ignoring prediction from the future)
530 Details : Incorrect calculation of peer's last_alive value in ksocklnd
532 --------------------------------------------------------------------------------
534 2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
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: ASSERTION(me == md->md_me) failed in lnet_match_md()
554 Description: increase send queue size for ciblnd/openiblnd
558 Description: new userspace socklnd
559 Details : Old userspace tcpnal that resided in lnet/ulnds/socklnd replaced
560 with new one - usocklnd.
562 Severity : enhancement
564 Description: Console message flood
565 Details : Make cdls ratelimiting more tunable by adding several tunable in
566 procfs /proc/sys/lnet/console_{min,max}_delay_centisecs and
567 /proc/sys/lnet/console_backoff.
569 --------------------------------------------------------------------------------
571 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
573 * Support for networks:
574 socklnd - any kernel supported by Lustre,
575 qswlnd - Qsnet kernel modules 5.20 and later,
576 openiblnd - IbGold 1.8.2,
577 o2iblnd - OFED 1.1 and 1.2,
578 viblnd - Voltaire ibhost 3.4.5 and later,
579 ciblnd - Topspin 3.2.0,
580 iiblnd - Infiniserv 3.3 + PathBits patch,
581 gmlnd - GM 2.1.22 and later,
582 mxlnd - MX 1.2.1 or later,
583 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
587 Description: /proc/sys/lnet has non-sysctl entries
588 Details : Updating dump_kernel/daemon_file/debug_mb to use sysctl variables
592 Description: TOE Kernel panic by ksocklnd
593 Details : offloaded sockets provide their own implementation of sendpage,
594 can't call tcp_sendpage() directly
598 Description: kibnal_shutdown() doesn't finish; lconf --cleanup hangs
599 Details : races between lnd_shutdown and peer creation prevent
600 lnd_shutdown from finishing.
604 Description: open files rlimit 1024 reached while liblustre testing
605 Details : ulnds/socklnd must close open socket after unsuccessful
610 Description: build error
611 Details : fix typos in gmlnd, ptllnd and viblnd
613 ------------------------------------------------------------------------------
615 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
617 * Support for networks:
618 socklnd - kernels up to 2.6.16,
619 qswlnd - Qsnet kernel modules 5.20 and later,
620 openiblnd - IbGold 1.8.2,
621 o2iblnd - OFED 1.1 and 1.2
622 viblnd - Voltaire ibhost 3.4.5 and later,
623 ciblnd - Topspin 3.2.0,
624 iiblnd - Infiniserv 3.3 + PathBits patch,
625 gmlnd - GM 2.1.22 and later,
626 mxlnd - MX 1.2.1 or later,
627 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
629 2007-06-21 Cluster File Systems, Inc. <info@clusterfs.com>
631 * Support for networks:
632 socklnd - kernels up to 2.6.16,
633 qswlnd - Qsnet kernel modules 5.20 and later,
634 openiblnd - IbGold 1.8.2,
636 viblnd - Voltaire ibhost 3.4.5 and later,
637 ciblnd - Topspin 3.2.0,
638 iiblnd - Infiniserv 3.3 + PathBits patch,
639 gmlnd - GM 2.1.22 and later,
640 mxlnd - MX 1.2.1 or later,
641 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
645 Description: Initialize cpumask before use
649 Description: ASSERTION failures when upgrading to the patchless zero-copy
651 Details : This bug affects "rolling upgrades", causing an inconsistent
652 protocol version negotiation and subsequent assertion failure
653 during rolling upgrades after the first wave of upgrades.
657 Details : Change "dropped message" CERRORs to D_NETERROR so they are
658 logged instead of creating "console chatter" when a lustre
659 timeout races with normal RPC completion.
662 Details : lnet_clear_peer_table can wait forever if user forgets to
666 Details : libcfs_id2str should check pid against LNET_PID_ANY.
670 Description: added LNET self test
671 Details : landing b_self_test
676 Description: cfs_duration_{u,n}sec() wrongly calculate nanosecond part of
678 Details : do_div() macro is used incorrectly.
680 2007-04-23 Cluster File Systems, Inc. <info@clusterfs.com>
684 Description: make panic on lbug configurable
688 Description: Add OFED1.2 support to o2iblnd
689 Details : o2iblnd depends on OFED's modules, if out-tree OFED's modules
690 are installed (other than kernel's in-tree infiniband), there
691 could be some problem while insmod o2iblnd (mismatch CRC of
693 If extra Module.symvers is supported in kernel (i.e, 2.6.17),
694 this link provides solution:
695 https://bugs.openfabrics.org/show_bug.cgi?id=355
696 if extra Module.symvers is not supported in kernel, we will
697 have to run the script in bug 12316 to update
698 $LINUX/module.symvers before building o2iblnd.
699 More details about this are in bug 12316.
701 ------------------------------------------------------------------------------
703 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
704 * version 1.4.10 / 1.6.0
705 * Support for networks:
706 socklnd - kernels up to 2.6.16,
707 qswlnd - Qsnet kernel modules 5.20 and later,
708 openiblnd - IbGold 1.8.2,
710 viblnd - Voltaire ibhost 3.4.5 and later,
711 ciblnd - Topspin 3.2.0,
712 iiblnd - Infiniserv 3.3 + PathBits patch,
713 gmlnd - GM 2.1.22 and later,
714 mxlnd - MX 1.2.1 or later,
715 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
719 Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
720 possibly accessed in kptllnd_shutdown. Ptllnd should init
721 kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
725 Description: gmlnd ignored some transmit errors when finalizing lnet messages.
729 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
733 Description: the_lnet.ln_finalizing was not set when the current thread is
734 about to complete messages. It only affects multi-threaded
740 Description: Changed the default kqswlnd ntxmsg=512
745 Description: Assertion failure in kernel ptllnd caused by posting passive
746 bulk buffers before connection establishment complete.
751 Description: A race in kernel ptllnd between deleting a peer and posting
752 new communications for it could hang communications -
753 manifesting as "Unexpectedly long timeout" messages.
758 Description: Kernel ptllnd lock ordering issue could hang a node.
763 Description: node crash on socket teardown race
766 Frequency : 'lctl peer_list' issued on a mx net
768 Description: Enable lctl's peer_list for MXLND
771 Frequency : after Ptllnd timeouts and portals congestion
773 Description: Credit overflows
774 Details : This was a bug in ptllnd connection establishment. The fix
775 implements better peer stamps to disambiguate connection
776 establishment and ensure both peers enter the credit flow
777 state machine consistently.
782 Description: kptllnd didn't propagate some network errors up to LNET
783 Details : This bug was spotted while investigating 11394. The fix
784 ensures network errors on sends and bulk transfers are
785 propagated to LNET/lustre correctly.
787 Severity : enhancement
789 Description: Fixed console chatter in case of -ETIMEDOUT.
791 Severity : enhancement
793 Description: Added D_NETTRACE for recording network packet history
794 (initially only for ptllnd). Also a separate userspace
795 ptllnd facility to gather history which should really be
796 covered by D_NETTRACE too, if only CDEBUG recorded history in
802 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
803 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
804 callback can occur before a connection has actually been
805 established. This caused an assertion failure previously.
807 Severity : enhancement
809 Description: Multiple instances for o2iblnd
810 Details : Allow multiple instances of o2iblnd to enable networking over
811 multiple HCAs and routing between them.
815 Description: lnet deadlock in router_checker
816 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
817 into BH locks to eliminate potential deadlock caused by
818 ksocknal_data_ready() preempting code holding these locks.
822 Description: Millions of failed socklnd connection attempts cause a very slow FS
823 Details : added a new route flag ksnr_scheduled to distinguish from
824 ksnr_connecting, so that a peer connection request is only turned
825 down for race concerns when an active connection to the same peer
826 is under progress (instead of just being scheduled).
828 ------------------------------------------------------------------------------
830 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
832 * Support for networks:
833 socklnd - kernels up to 2.6.16
834 qswlnd - Qsnet kernel modules 5.20 and later
835 openiblnd - IbGold 1.8.2
837 viblnd - Voltaire ibhost 3.4.5 and later
838 ciblnd - Topspin 3.2.0
839 iiblnd - Infiniserv 3.3 + PathBits patch
840 gmlnd - GM 2.1.22 and later
841 mxlnd - MX 1.2.1 or later
842 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
845 Severity : major on XT3
847 Description: libcfs overwrites /proc/sys/portals
848 Details : libcfs created a symlink from /proc/sys/portals to
849 /proc/sys/lnet for backwards compatibility. This is no
850 longer required and makes the Cray portals /proc variables
855 Description: OFED FMR API change
856 Details : This changes parameter usage to reflect a change in
857 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
858 that FMR support is only used in experimental versions of the
859 o2iblnd - this change does not affect standard usage at all.
861 Severity : enhancement
863 Description: new ko2iblnd module parameter: ib_mtu
864 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
865 HCAs. You can avoid this problem by setting the MTU to 1024
866 using this module parameter.
868 Severity : enhancement
869 Bugzilla : 11118/11620
870 Description: ptllnd small request message buffer alignment fix
871 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
872 Round up small message size on sends in case this option
873 is not supported. 11620 was a defect in the initial
874 implementation which effectively asserted all peers had to be
875 running the correct protocol version which was fixed by always
876 NAK-ing such requests and handling any misalignments they
881 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
882 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
883 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
885 Severity : enhancement
887 Description: Patchless ZC(zero copy) socklnd
888 Details : New protocol for socklnd, socklnd can support zero copy without
889 kernel patch, it's compatible with old socklnd. Checksum is
890 moved from tunables to modparams.
894 Description: When ksocknal_del_peer() is called upon a peer whose
895 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
896 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
899 Frequency : when ptlrpc is under heavy use and runs out of request buffer
901 Description: In lnet_match_blocked_msg(), md can be used without holding a
905 Frequency : very rarely
907 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
908 If connd connects a route which has been closed by
909 ksocknal_shutdown(), ksocknal_create_routes() may create new
910 routes which hold references on the peer, causing shutdown
911 process to wait for peer to disappear forever.
913 Severity : enhancement
915 Description: Dump XT3 portals traces on kptllnd timeout
916 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
917 dump Cray portals debug traces to a file. The kptllnd module
918 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
919 is the basename of the dump file.
922 Frequency : infrequent
924 Description: kernel ptllnd fix bug in connection re-establishment
925 Details : Kernel ptllnd could produce protocol errors e.g. illegal
926 matchbits and/or violate the credit flow protocol when trying
927 to re-establish a connection with a peer after an error or
930 Severity : enhancement
932 Description: Allow /proc/sys/lnet/debug to be set symbolically
933 Details : Allow debug and subsystem debug values to be read/set by name
934 in addition to numerically, for ease of use.
937 Frequency : only in configurations with LNET routers
939 Description: routes automatically marked down and recovered
940 Details : In configurations with LNET routers if a router fails routers
941 now actively try to recover routes that are down, unless they
942 are marked down by an administrator.
944 ------------------------------------------------------------------------------
946 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
949 Frequency : very rarely, in configurations with LNET routers and TCP
951 Description: incorrect data written to files on OSTs
952 Details : In certain high-load conditions incorrect data may be written
953 to files on the OST when using TCP networks.
955 ------------------------------------------------------------------------------
957 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
959 - rework CDEBUG messages rate-limiting mechanism b=10375
960 - add per-socket tunables for socklnd if the kernel is patched b=10327
962 ------------------------------------------------------------------------------
964 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
966 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
967 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
968 - quiet LNET startup LNI message for liblustre b=10128
969 - Better console error messages if 'ip2nets' can't match an IP address
970 - Fixed overflow/use-before-set bugs in linux-time.h
971 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
972 - LNET teardown failed an assertion about the route table being empty
973 - Fixed a crash in LNetEQPoll(<invalid handle>)
974 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
975 - improve debug message for liblustre/Catamount nodes (b=10116)
977 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
978 * Configuration change for the XT3
979 The PTLLND is now used to run Lustre over Portals on the XT3.
980 The configure option(s) --with-cray-portals are no longer
981 used. Rather --with-portals=<path-to-portals-includes> is
982 used to enable building on the XT3. In addition to enable
983 XT3 specific features the option --enable-cray-xt3 must be
986 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
987 * Portals has been removed, replaced by LNET.
988 LNET is new networking infrastructure for Lustre, it includes a
989 reorganized network configuration mode (see the user
990 documentation for full details) as well as support for routing
991 between different network fabrics. Lustre Networking Devices
992 (LNDS) for the supported network fabrics have also been created
993 for this new infrastructure.
995 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
1000 Frequency : rare (large Voltaire clusters only)
1002 Description: the default number of reserved transmit descriptors was too low
1003 for some large clusters
1004 Details : As a workaround, the number was increased. A proper fix includes
1007 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
1012 Frequency : occasional (large-scale events, cluster reboot, network failure)
1014 Description: too many error messages on console obscure actual problem and
1015 can slow down/panic server, or cause recovery to fail repeatedly
1016 Details : enable rate-limiting of console error messages, and some messages
1017 that were console errors now only go to the kernel log
1019 Severity : enhancement
1021 Description: add /proc/sys/portals/catastrophe entry which will report if
1022 that node has previously LBUGged
1024 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
1026 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
1028 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
1030 - handle error return code in kranal_check_fma_rx() (5915,6054)
1032 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
1034 - update vibnal (Voltaire IB NAL)
1035 - update gmnal (Myrinet NAL), gmnalid
1037 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
1039 * Landed portals:b_port_step as follows...
1041 - removed CFS_DECL_SPIN*
1042 just use 'spinlock_t' and initialise with spin_lock_init()
1044 - removed CFS_DECL_MUTEX*
1045 just use 'struct semaphore' and initialise with init_mutex()
1047 - removed CFS_DECL_RWSEM*
1048 just use 'struct rw_semaphore' and initialise with init_rwsem()
1050 - renamed cfs_sleep_chan -> cfs_waitq
1051 cfs_sleep_link -> cfs_waitlink
1053 - fixed race in linux version of arch-independent socknal
1054 (the ENOMEM/EAGAIN decision).
1056 - Didn't fix problems in Darwin version of arch-independent socknal
1057 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
1059 - removed libcfs types from non-socknal header files (only some types
1060 in the header files had been changed; the .c files hadn't been