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: ptlrpcd stuck in lnet allocations under memory pressure
18 Details : Set PF_MEMALLOC on outgoing path to prevent deadlock on memory
23 Description: fix o2iblnd v2 regression of credit deadlock with v1 peers
28 Description: lctl ping of 1.6.6 server from 1.8 router causes protocol error
30 -------------------------------------------------------------------------------
32 2010-10-29 Oracle, Inc.
34 * Support for networks:
35 socklnd - any kernel supported by Lustre,
36 qswlnd - Qsnet kernel modules 5.20 and later,
37 openiblnd - IbGold 1.8.2,
38 o2iblnd - OFED 1.3, 1.4.1, 1.4.2, 1.5.1 and 1.5.2
39 viblnd - Voltaire ibhost 3.4.5 and later,
40 ciblnd - Topspin 3.2.0,
41 iiblnd - Infiniserv 3.3 + PathBits patch,
42 gmlnd - GM 2.1.22 and later,
43 mxlnd - MX 1.2.10 or later,
44 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
48 Description: MDS node unresponsive
49 Details : improve hash distribution, doubled hash size for the lnet cookie
52 -------------------------------------------------------------------------------
54 2010-07-31 Oracle, Inc.
56 * Support for networks:
57 socklnd - any kernel supported by Lustre,
58 qswlnd - Qsnet kernel modules 5.20 and later,
59 openiblnd - IbGold 1.8.2,
60 o2iblnd - OFED 1.3, 1.4.1, 1.4.2 and 1.5.1
61 viblnd - Voltaire ibhost 3.4.5 and later,
62 ciblnd - Topspin 3.2.0,
63 iiblnd - Infiniserv 3.3 + PathBits patch,
64 gmlnd - GM 2.1.22 and later,
65 mxlnd - MX 1.2.10 or later,
66 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
70 Description: With peer health detection, o2iblnd makes only one attempt to
71 reconnect which is not enough with nodes running lustre 1.6
72 because of proto version mismatch. Fix o2iblnd to retry one more
75 Severity : enhancement
77 Description: Quiet some LNET messages
79 Severity : enhancement
81 Description: Add OFED 1.5.1 support
83 Severity : enhancement
85 Description: The peer health code lacked some important debugging info in
86 lnd_query code paths. We've added necessary debug prints,
87 not just for bug 21678, but also for future troubleshooting.
89 -------------------------------------------------------------------------------
91 2010-04-30 Oracle, Inc.
93 * Support for networks:
94 socklnd - any kernel supported by Lustre,
95 qswlnd - Qsnet kernel modules 5.20 and later,
96 openiblnd - IbGold 1.8.2,
97 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, 1.3, 1.4.1, and 1.4.2
98 viblnd - Voltaire ibhost 3.4.5 and later,
99 ciblnd - Topspin 3.2.0,
100 iiblnd - Infiniserv 3.3 + PathBits patch,
101 gmlnd - GM 2.1.22 and later,
102 mxlnd - MX 1.2.10 or later,
103 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
107 Description: lst: check # of remained RPCs before aborting
108 Details : lstcon_rpc_trans_postwait() calls lstcon_rpc_trans_abort() only
109 when the transaction is timeout, so if we got "end_session" to
110 interrupt waiting on transaction, then we can hit the assertion
111 failure ASSERTION(crpc->crp_stamp != 0)
115 Description: print more debug info for timedout ZC-req
116 Details : Print more information for timedout ZC-req and partial
117 received connection. Close connection for timedout ZC-req
118 Always send ZC_ACK on non-blocking connection(BULK_IN)
122 Description: Adding WIRE_ATTR attribute to LNET types
123 Details : LST nodes on different platforms might not communicate well
124 due to the lack of WIRE_ATTR attribute in some LNET structures
125 traversing network. The patch fixes the problem by adding
126 WIRE_ATTR where needed.
130 Description: hash MEs on RDMA portal
131 Details : RDMA portal can have very long ME list on client side, which
132 will trigger soft lockup because of long searching on list.
133 Hash MEs on RDMA portal can resolve this problem.
137 Description: fix for double release of ibc_lock in o2iblnd
138 Details : Re-acquire ibc_lock in kiblnd_post_tx_locked(). Add extra
139 reference to conn before calling kiblnd_post_tx_locked()
140 to avoid scenario when conn disappears inside
141 kiblnd_post_tx_locked().
143 -------------------------------------------------------------------------------
144 2010-01-29 Sun Microsystems, Inc.
146 * Support for networks:
147 socklnd - any kernel supported by Lustre,
148 qswlnd - Qsnet kernel modules 5.20 and later,
149 openiblnd - IbGold 1.8.2,
150 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, 1.3, 1.4.1, and 1.4.2
151 viblnd - Voltaire ibhost 3.4.5 and later,
152 ciblnd - Topspin 3.2.0,
153 iiblnd - Infiniserv 3.3 + PathBits patch,
154 gmlnd - GM 2.1.22 and later,
155 mxlnd - MX 1.2.10 or later,
156 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
160 Description: should update lp_alive for non-router peers.
162 Severity : enhancement
164 Description: LNet router shuffler.
166 Severity : enhancement
168 Description: LNet fine grain routing support.
172 Description: router checker stops working when system wall clock goes backward
173 Details : use monotonic timing source instead of system wall clock time.
175 Severity : enhancement
177 Description: avoid asymmetrical router failures
179 Severity : enhancement
181 Description: multiple-instance support for kptllnd
185 Description: ksocknal_close_conn_locked connection race
186 Details : A race was possible when ksocknal_create_conn calls
187 ksocknal_close_conn_locked for already closed conn.
189 Severity : enhancement
191 Description: port router pinger to userspace
195 Description: kptllnd HELLO protocol deadlock
196 Details : kptllnd HELLO protocol doesn't run to completion in finite time
200 Description: LNet selftest fixes and enhancements
202 Severity : enhancement
204 Description: allow a test node to be a member of multiple test groups
206 Severity : enhancement
208 Description: MXLND: eliminate hosts file, use arp for peer nic_id resolution
209 Details : an update from the upstream developer Scott Atchley.
212 -------------------------------------------------------------------------------
213 2009-07-31 Sun Microsystems, Inc.
215 * Support for networks:
216 socklnd - any kernel supported by Lustre,
217 qswlnd - Qsnet kernel modules 5.20 and later,
218 openiblnd - IbGold 1.8.2,
219 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, 1.3, and 1.4.1
220 viblnd - Voltaire ibhost 3.4.5 and later,
221 ciblnd - Topspin 3.2.0,
222 iiblnd - Infiniserv 3.3 + PathBits patch,
223 gmlnd - GM 2.1.22 and later,
224 mxlnd - MX 1.2.1 or later,
225 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
229 Description: router_proc.c is rewritten to use sysctl-interface for parameters
230 residing in /proc/sys/lnet
232 Severity : enhancement
234 Description: add a new LND optiion to control peer buffer credits on routers
238 Description: Fixing deadlock in usocklnd
239 Details : A deadlock was possible in usocklnd due to race condition while
240 tearing connection down. The problem resulted from erroneous
241 assumption that lnet_finalize() could have been called holding
242 some lnd-level locks.
245 Bugzilla : 13621, 15983
246 Description: Protocol V2 of o2iblnd
247 Details : o2iblnd V2 has several new features:
248 . map-on-demand: map-on-demand is disabled by default, it can
249 be enabled by using modparam "map_on_demand=@value@", @value@
250 should >= 0 and < 256, 0 will disable map-on-demand, any other
251 valid value will enable map-on-demand.
252 Oi2blnd will create FMR or physical MR for RDMA if fragments of
254 Enable map-on-demand will take less memory for new connection,
255 but a little more CPU for RDMA.
256 . iWARP : to support iWARP, please enable map-on-demand, 32 and 64
257 are recommanded value. iWARP will probably fail for value >=128.
258 . OOB NOOP message: to resolve deadlock on router.
259 . tunable peer_credits_hiw: (high water to return credits),
260 default value of peer_credits_hiw equals to (peer_credits -1),
261 user can change it between peer_credits/2 and (peer_credits - 1).
262 Lower value is recommended for high latency network.
263 . tunable message queue size: it always equals to peer_credits,
264 higher value is recommended for high latency network.
265 . It's compatible with earlier version of o2iblnd
269 Description: Fixing 'running out of ports' issue
270 Details : Add a delay before next reconnect attempt in ksocklnd in
271 the case of lost race. Limit the frequency of query-requests
272 in lnet. Improved handling of 'dead peer' notifications in
277 Description: Change ptllnd timeout and watchdog timers
278 Details : Add ptltrace_on_nal_failed and bump ptllnd timeout to match
279 Portals wire timeout.
283 Description: One down Lustre FS hangs ALL mounted Lustre filesystems
284 Details : Shared routing enhancements - peer health detection.
288 Description: IB path MTU mistakenly set to 1st path MTU when ib_mtu is off
289 Details : See comment 46 in bug 11245 for details - it's indeed a bug
290 introduced by the original 11245 fix.
294 Description: uptllnd credit overflow fix
295 Details : kptl_msg_t::ptlm_credits could be overflown by uptllnd since
300 Description: socklnd protocol version 3
301 Details : With current protocol V2, connections on router can be
302 blocked and can't receive any incoming messages when there is no
303 more router buffer, so ZC-ACK can't be handled (LNet message
304 can't be finalized) and will cause deadlock on router.
305 Protocol V3 has a dedicated connection for emergency messages
306 like ZC-ACK to router, messages on this dedicated connection
307 don't need any credit so will never be blocked. Also, V3 can send
308 keepalive ping in specified period for router healthy checking.
310 -------------------------------------------------------------------------------
311 12-31-2008 Sun Microsystems, Inc.
313 * Support for networks:
314 socklnd - any kernel supported by Lustre,
315 qswlnd - Qsnet kernel modules 5.20 and later,
316 openiblnd - IbGold 1.8.2,
317 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, and 1.3
318 viblnd - Voltaire ibhost 3.4.5 and later,
319 ciblnd - Topspin 3.2.0,
320 iiblnd - Infiniserv 3.3 + PathBits patch,
321 gmlnd - GM 2.1.22 and later,
322 mxlnd - MX 1.2.1 or later,
323 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
332 Description: workaround for OOM from o2iblnd
333 Details : OFED needs allocate big chunk of memory for QP while creating
334 connection for o2iblnd, OOM can happen if no such a contiguous
336 QP size is decided by concurrent_sends and max_fragments of
337 o2iblnd, now we permit user to specify smaller value for
338 concurrent_sends of o2iblnd(i.e: concurrent_sends=7), which
339 will decrease memory block size required by creating QP.
343 Description: Support Zerocopy receive of Chelsio device
344 Details : Chelsio driver can support zerocopy for iov[1] if it's
345 contiguous and large enough.
349 Description: fix credit flow deadlock in uptllnd
353 Description: finalize network operation in reasonable time
354 Details : conf-sanity test_32a couldn't stop ost and mds because it
355 tried to access non-existent peer and tcp connect took
356 quite long before timing out.
360 Description: Continuous recovery on 33 of 413 nodes after lustre oss failure
361 Details : Lost reference on conn prevents peer from being destroyed, which
362 could prevent new peer creation if peer count has reached upper
367 Description: LNET Selftest results in Soft lockup on OSS CPU
368 Details : only hits when 8 or more o2ib clients involved and a session is
369 torn down with 'lst end_session' without preceeding 'lst stop'.
373 Description: concurrent_sends in IB LNDs should not be changeable at run time
374 Details : concurrent_sends in IB LNDs should not be changeable at run time
378 Description: ptl_send_rpc hits LASSERT when ptl_send_buf fails
379 Details : only hits under out-of-memory situations
382 -------------------------------------------------------------------------------
384 2009-02-07 Sun Microsystems, Inc.
386 * Support for networks:
387 socklnd - any kernel supported by Lustre,
388 qswlnd - Qsnet kernel modules 5.20 and later,
389 openiblnd - IbGold 1.8.2,
390 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, and 1.3
391 viblnd - Voltaire ibhost 3.4.5 and later,
392 ciblnd - Topspin 3.2.0,
393 iiblnd - Infiniserv 3.3 + PathBits patch,
394 gmlnd - GM 2.1.22 and later,
395 mxlnd - MX 1.2.1 or later,
396 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
400 Description: workaround for OOM from o2iblnd
401 Details : OFED needs allocate big chunk of memory for QP while creating
402 connection for o2iblnd, OOM can happen if no such a contiguous
404 QP size is decided by concurrent_sends and max_fragments of
405 o2iblnd, now we permit user to specify smaller value for
406 concurrent_sends of o2iblnd(i.e: concurrent_sends=7), which
407 will decrease memory block size required by creating QP.
411 Description: Support Zerocopy receive of Chelsio device
412 Details : Chelsio driver can support zerocopy for iov[1] if it's
413 contiguous and large enough.
417 Description: fix credit flow deadlock in uptllnd
421 Description: finalize network operation in reasonable time
422 Details : conf-sanity test_32a couldn't stop ost and mds because it
423 tried to access non-existent peer and tcp connect took
424 quite long before timing out.
428 Description: Continuous recovery on 33 of 413 nodes after lustre oss failure
429 Details : Lost reference on conn prevents peer from being destroyed, which
430 could prevent new peer creation if peer count has reached upper
435 Description: LNET Selftest results in Soft lockup on OSS CPU
436 Details : only hits when 8 or more o2ib clients involved and a session is
437 torn down with 'lst end_session' without preceeding 'lst stop'.
441 Description: concurrent_sends in IB LNDs should not be changeable at run time
442 Details : concurrent_sends in IB LNDs should not be changeable at run time
444 -------------------------------------------------------------------------------
446 11-03-2008 Sun Microsystems, Inc.
448 * Support for networks:
449 socklnd - any kernel supported by Lustre,
450 qswlnd - Qsnet kernel modules 5.20 and later,
451 openiblnd - IbGold 1.8.2,
452 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, and 1.3
453 viblnd - Voltaire ibhost 3.4.5 and later,
454 ciblnd - Topspin 3.2.0,
455 iiblnd - Infiniserv 3.3 + PathBits patch,
456 gmlnd - GM 2.1.22 and later,
457 mxlnd - MX 1.2.1 or later,
458 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
462 Description: ptl_send_rpc hits LASSERT when ptl_send_buf fails
463 Details : only hits under out-of-memory situations
465 -------------------------------------------------------------------------------
468 04-26-2008 Sun Microsystems, Inc.
470 * Support for networks:
471 socklnd - any kernel supported by Lustre,
472 qswlnd - Qsnet kernel modules 5.20 and later,
473 openiblnd - IbGold 1.8.2,
474 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5
475 viblnd - Voltaire ibhost 3.4.5 and later,
476 ciblnd - Topspin 3.2.0,
477 iiblnd - Infiniserv 3.3 + PathBits patch,
478 gmlnd - GM 2.1.22 and later,
479 mxlnd - MX 1.2.1 or later,
480 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
484 Description: excessive debug information removed
485 Details : excessive debug information removed
489 Description: ksocknal_create_conn() hit ASSERTION during connection race
490 Details : ksocknal_create_conn() hit ASSERTION during connection race
494 Description: ksocknal_send_hello() hit ASSERTION while connecting race
495 Details : ksocknal_send_hello() hit ASSERTION while connecting race
499 Description: o2iblnd/ptllnd credit deadlock in a routed config.
500 Details : o2iblnd/ptllnd credit deadlock in a routed config.
504 Description: High load after starting lnet
505 Details : gmlnd should sleep in rx thread in interruptible way. Otherwise,
506 uptime utility reports high load that looks confusingly.
510 Description: ksocklnd fails to establish connection if accept_port is high
511 Details : PID remapping must not be done for active (outgoing) connections
513 --------------------------------------------------------------------------------
515 2008-01-11 Sun Microsystems, Inc.
517 * Support for networks:
518 socklnd - any kernel supported by Lustre,
519 qswlnd - Qsnet kernel modules 5.20 and later,
520 openiblnd - IbGold 1.8.2,
521 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5
522 viblnd - Voltaire ibhost 3.4.5 and later,
523 ciblnd - Topspin 3.2.0,
524 iiblnd - Infiniserv 3.3 + PathBits patch,
525 gmlnd - GM 2.1.22 and later,
526 mxlnd - MX 1.2.1 or later,
527 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
530 Description: liblustre network error
531 Details : liblustre clients should understand LNET_ACCEPT_PORT environment
532 variable even if they don't start lnet acceptor.
536 Description: Strange message from lnet (Ignoring prediction from the future)
537 Details : Incorrect calculation of peer's last_alive value in ksocklnd
539 --------------------------------------------------------------------------------
541 2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
543 * Support for networks:
544 socklnd - any kernel supported by Lustre,
545 qswlnd - Qsnet kernel modules 5.20 and later,
546 openiblnd - IbGold 1.8.2,
547 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5.
548 viblnd - Voltaire ibhost 3.4.5 and later,
549 ciblnd - Topspin 3.2.0,
550 iiblnd - Infiniserv 3.3 + PathBits patch,
551 gmlnd - GM 2.1.22 and later,
552 mxlnd - MX 1.2.1 or later,
553 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
557 Description: ASSERTION(me == md->md_me) failed in lnet_match_md()
561 Description: increase send queue size for ciblnd/openiblnd
565 Description: new userspace socklnd
566 Details : Old userspace tcpnal that resided in lnet/ulnds/socklnd replaced
567 with new one - usocklnd.
569 Severity : enhancement
571 Description: Console message flood
572 Details : Make cdls ratelimiting more tunable by adding several tunable in
573 procfs /proc/sys/lnet/console_{min,max}_delay_centisecs and
574 /proc/sys/lnet/console_backoff.
576 --------------------------------------------------------------------------------
578 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
580 * Support for networks:
581 socklnd - any kernel supported by Lustre,
582 qswlnd - Qsnet kernel modules 5.20 and later,
583 openiblnd - IbGold 1.8.2,
584 o2iblnd - OFED 1.1 and 1.2,
585 viblnd - Voltaire ibhost 3.4.5 and later,
586 ciblnd - Topspin 3.2.0,
587 iiblnd - Infiniserv 3.3 + PathBits patch,
588 gmlnd - GM 2.1.22 and later,
589 mxlnd - MX 1.2.1 or later,
590 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
594 Description: /proc/sys/lnet has non-sysctl entries
595 Details : Updating dump_kernel/daemon_file/debug_mb to use sysctl variables
599 Description: TOE Kernel panic by ksocklnd
600 Details : offloaded sockets provide their own implementation of sendpage,
601 can't call tcp_sendpage() directly
605 Description: kibnal_shutdown() doesn't finish; lconf --cleanup hangs
606 Details : races between lnd_shutdown and peer creation prevent
607 lnd_shutdown from finishing.
611 Description: open files rlimit 1024 reached while liblustre testing
612 Details : ulnds/socklnd must close open socket after unsuccessful
617 Description: build error
618 Details : fix typos in gmlnd, ptllnd and viblnd
620 ------------------------------------------------------------------------------
622 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
624 * Support for networks:
625 socklnd - kernels up to 2.6.16,
626 qswlnd - Qsnet kernel modules 5.20 and later,
627 openiblnd - IbGold 1.8.2,
628 o2iblnd - OFED 1.1 and 1.2
629 viblnd - Voltaire ibhost 3.4.5 and later,
630 ciblnd - Topspin 3.2.0,
631 iiblnd - Infiniserv 3.3 + PathBits patch,
632 gmlnd - GM 2.1.22 and later,
633 mxlnd - MX 1.2.1 or later,
634 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
636 2007-06-21 Cluster File Systems, Inc. <info@clusterfs.com>
638 * Support for networks:
639 socklnd - kernels up to 2.6.16,
640 qswlnd - Qsnet kernel modules 5.20 and later,
641 openiblnd - IbGold 1.8.2,
643 viblnd - Voltaire ibhost 3.4.5 and later,
644 ciblnd - Topspin 3.2.0,
645 iiblnd - Infiniserv 3.3 + PathBits patch,
646 gmlnd - GM 2.1.22 and later,
647 mxlnd - MX 1.2.1 or later,
648 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
652 Description: Initialize cpumask before use
656 Description: ASSERTION failures when upgrading to the patchless zero-copy
658 Details : This bug affects "rolling upgrades", causing an inconsistent
659 protocol version negotiation and subsequent assertion failure
660 during rolling upgrades after the first wave of upgrades.
664 Details : Change "dropped message" CERRORs to D_NETERROR so they are
665 logged instead of creating "console chatter" when a lustre
666 timeout races with normal RPC completion.
669 Details : lnet_clear_peer_table can wait forever if user forgets to
673 Details : libcfs_id2str should check pid against LNET_PID_ANY.
677 Description: added LNET self test
678 Details : landing b_self_test
683 Description: cfs_duration_{u,n}sec() wrongly calculate nanosecond part of
685 Details : do_div() macro is used incorrectly.
687 2007-04-23 Cluster File Systems, Inc. <info@clusterfs.com>
691 Description: make panic on lbug configurable
695 Description: Add OFED1.2 support to o2iblnd
696 Details : o2iblnd depends on OFED's modules, if out-tree OFED's modules
697 are installed (other than kernel's in-tree infiniband), there
698 could be some problem while insmod o2iblnd (mismatch CRC of
700 If extra Module.symvers is supported in kernel (i.e, 2.6.17),
701 this link provides solution:
702 https://bugs.openfabrics.org/show_bug.cgi?id=355
703 if extra Module.symvers is not supported in kernel, we will
704 have to run the script in bug 12316 to update
705 $LINUX/module.symvers before building o2iblnd.
706 More details about this are in bug 12316.
708 ------------------------------------------------------------------------------
710 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
711 * version 1.4.10 / 1.6.0
712 * Support for networks:
713 socklnd - kernels up to 2.6.16,
714 qswlnd - Qsnet kernel modules 5.20 and later,
715 openiblnd - IbGold 1.8.2,
717 viblnd - Voltaire ibhost 3.4.5 and later,
718 ciblnd - Topspin 3.2.0,
719 iiblnd - Infiniserv 3.3 + PathBits patch,
720 gmlnd - GM 2.1.22 and later,
721 mxlnd - MX 1.2.1 or later,
722 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
726 Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
727 possibly accessed in kptllnd_shutdown. Ptllnd should init
728 kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
732 Description: gmlnd ignored some transmit errors when finalizing lnet messages.
736 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
740 Description: the_lnet.ln_finalizing was not set when the current thread is
741 about to complete messages. It only affects multi-threaded
747 Description: Changed the default kqswlnd ntxmsg=512
752 Description: Assertion failure in kernel ptllnd caused by posting passive
753 bulk buffers before connection establishment complete.
758 Description: A race in kernel ptllnd between deleting a peer and posting
759 new communications for it could hang communications -
760 manifesting as "Unexpectedly long timeout" messages.
765 Description: Kernel ptllnd lock ordering issue could hang a node.
770 Description: node crash on socket teardown race
773 Frequency : 'lctl peer_list' issued on a mx net
775 Description: Enable lctl's peer_list for MXLND
778 Frequency : after Ptllnd timeouts and portals congestion
780 Description: Credit overflows
781 Details : This was a bug in ptllnd connection establishment. The fix
782 implements better peer stamps to disambiguate connection
783 establishment and ensure both peers enter the credit flow
784 state machine consistently.
789 Description: kptllnd didn't propagate some network errors up to LNET
790 Details : This bug was spotted while investigating 11394. The fix
791 ensures network errors on sends and bulk transfers are
792 propagated to LNET/lustre correctly.
794 Severity : enhancement
796 Description: Fixed console chatter in case of -ETIMEDOUT.
798 Severity : enhancement
800 Description: Added D_NETTRACE for recording network packet history
801 (initially only for ptllnd). Also a separate userspace
802 ptllnd facility to gather history which should really be
803 covered by D_NETTRACE too, if only CDEBUG recorded history in
809 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
810 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
811 callback can occur before a connection has actually been
812 established. This caused an assertion failure previously.
814 Severity : enhancement
816 Description: Multiple instances for o2iblnd
817 Details : Allow multiple instances of o2iblnd to enable networking over
818 multiple HCAs and routing between them.
822 Description: lnet deadlock in router_checker
823 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
824 into BH locks to eliminate potential deadlock caused by
825 ksocknal_data_ready() preempting code holding these locks.
829 Description: Millions of failed socklnd connection attempts cause a very slow FS
830 Details : added a new route flag ksnr_scheduled to distinguish from
831 ksnr_connecting, so that a peer connection request is only turned
832 down for race concerns when an active connection to the same peer
833 is under progress (instead of just being scheduled).
835 ------------------------------------------------------------------------------
837 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
839 * Support for networks:
840 socklnd - kernels up to 2.6.16
841 qswlnd - Qsnet kernel modules 5.20 and later
842 openiblnd - IbGold 1.8.2
844 viblnd - Voltaire ibhost 3.4.5 and later
845 ciblnd - Topspin 3.2.0
846 iiblnd - Infiniserv 3.3 + PathBits patch
847 gmlnd - GM 2.1.22 and later
848 mxlnd - MX 1.2.1 or later
849 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
852 Severity : major on XT3
854 Description: libcfs overwrites /proc/sys/portals
855 Details : libcfs created a symlink from /proc/sys/portals to
856 /proc/sys/lnet for backwards compatibility. This is no
857 longer required and makes the Cray portals /proc variables
862 Description: OFED FMR API change
863 Details : This changes parameter usage to reflect a change in
864 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
865 that FMR support is only used in experimental versions of the
866 o2iblnd - this change does not affect standard usage at all.
868 Severity : enhancement
870 Description: new ko2iblnd module parameter: ib_mtu
871 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
872 HCAs. You can avoid this problem by setting the MTU to 1024
873 using this module parameter.
875 Severity : enhancement
876 Bugzilla : 11118/11620
877 Description: ptllnd small request message buffer alignment fix
878 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
879 Round up small message size on sends in case this option
880 is not supported. 11620 was a defect in the initial
881 implementation which effectively asserted all peers had to be
882 running the correct protocol version which was fixed by always
883 NAK-ing such requests and handling any misalignments they
888 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
889 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
890 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
892 Severity : enhancement
894 Description: Patchless ZC(zero copy) socklnd
895 Details : New protocol for socklnd, socklnd can support zero copy without
896 kernel patch, it's compatible with old socklnd. Checksum is
897 moved from tunables to modparams.
901 Description: When ksocknal_del_peer() is called upon a peer whose
902 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
903 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
906 Frequency : when ptlrpc is under heavy use and runs out of request buffer
908 Description: In lnet_match_blocked_msg(), md can be used without holding a
912 Frequency : very rarely
914 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
915 If connd connects a route which has been closed by
916 ksocknal_shutdown(), ksocknal_create_routes() may create new
917 routes which hold references on the peer, causing shutdown
918 process to wait for peer to disappear forever.
920 Severity : enhancement
922 Description: Dump XT3 portals traces on kptllnd timeout
923 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
924 dump Cray portals debug traces to a file. The kptllnd module
925 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
926 is the basename of the dump file.
929 Frequency : infrequent
931 Description: kernel ptllnd fix bug in connection re-establishment
932 Details : Kernel ptllnd could produce protocol errors e.g. illegal
933 matchbits and/or violate the credit flow protocol when trying
934 to re-establish a connection with a peer after an error or
937 Severity : enhancement
939 Description: Allow /proc/sys/lnet/debug to be set symbolically
940 Details : Allow debug and subsystem debug values to be read/set by name
941 in addition to numerically, for ease of use.
944 Frequency : only in configurations with LNET routers
946 Description: routes automatically marked down and recovered
947 Details : In configurations with LNET routers if a router fails routers
948 now actively try to recover routes that are down, unless they
949 are marked down by an administrator.
951 ------------------------------------------------------------------------------
953 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
956 Frequency : very rarely, in configurations with LNET routers and TCP
958 Description: incorrect data written to files on OSTs
959 Details : In certain high-load conditions incorrect data may be written
960 to files on the OST when using TCP networks.
962 ------------------------------------------------------------------------------
964 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
966 - rework CDEBUG messages rate-limiting mechanism b=10375
967 - add per-socket tunables for socklnd if the kernel is patched b=10327
969 ------------------------------------------------------------------------------
971 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
973 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
974 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
975 - quiet LNET startup LNI message for liblustre b=10128
976 - Better console error messages if 'ip2nets' can't match an IP address
977 - Fixed overflow/use-before-set bugs in linux-time.h
978 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
979 - LNET teardown failed an assertion about the route table being empty
980 - Fixed a crash in LNetEQPoll(<invalid handle>)
981 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
982 - improve debug message for liblustre/Catamount nodes (b=10116)
984 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
985 * Configuration change for the XT3
986 The PTLLND is now used to run Lustre over Portals on the XT3.
987 The configure option(s) --with-cray-portals are no longer
988 used. Rather --with-portals=<path-to-portals-includes> is
989 used to enable building on the XT3. In addition to enable
990 XT3 specific features the option --enable-cray-xt3 must be
993 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
994 * Portals has been removed, replaced by LNET.
995 LNET is new networking infrastructure for Lustre, it includes a
996 reorganized network configuration mode (see the user
997 documentation for full details) as well as support for routing
998 between different network fabrics. Lustre Networking Devices
999 (LNDS) for the supported network fabrics have also been created
1000 for this new infrastructure.
1002 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
1007 Frequency : rare (large Voltaire clusters only)
1009 Description: the default number of reserved transmit descriptors was too low
1010 for some large clusters
1011 Details : As a workaround, the number was increased. A proper fix includes
1014 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
1019 Frequency : occasional (large-scale events, cluster reboot, network failure)
1021 Description: too many error messages on console obscure actual problem and
1022 can slow down/panic server, or cause recovery to fail repeatedly
1023 Details : enable rate-limiting of console error messages, and some messages
1024 that were console errors now only go to the kernel log
1026 Severity : enhancement
1028 Description: add /proc/sys/portals/catastrophe entry which will report if
1029 that node has previously LBUGged
1031 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
1033 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
1035 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
1037 - handle error return code in kranal_check_fma_rx() (5915,6054)
1039 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
1041 - update vibnal (Voltaire IB NAL)
1042 - update gmnal (Myrinet NAL), gmnalid
1044 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
1046 * Landed portals:b_port_step as follows...
1048 - removed CFS_DECL_SPIN*
1049 just use 'spinlock_t' and initialise with spin_lock_init()
1051 - removed CFS_DECL_MUTEX*
1052 just use 'struct semaphore' and initialise with init_mutex()
1054 - removed CFS_DECL_RWSEM*
1055 just use 'struct rw_semaphore' and initialise with init_rwsem()
1057 - renamed cfs_sleep_chan -> cfs_waitq
1058 cfs_sleep_link -> cfs_waitlink
1060 - fixed race in linux version of arch-independent socknal
1061 (the ENOMEM/EAGAIN decision).
1063 - Didn't fix problems in Darwin version of arch-independent socknal
1064 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
1066 - removed libcfs types from non-socknal header files (only some types
1067 in the header files had been changed; the .c files hadn't been