1 2010-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 and 1.5.1
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
16 -------------------------------------------------------------------------------
18 2010-07-31 Oracle, Inc.
20 * Support for networks:
21 socklnd - any kernel supported by Lustre,
22 qswlnd - Qsnet kernel modules 5.20 and later,
23 openiblnd - IbGold 1.8.2,
24 o2iblnd - OFED 1.3, 1.4.1, 1.4.2 and 1.5.1
25 viblnd - Voltaire ibhost 3.4.5 and later,
26 ciblnd - Topspin 3.2.0,
27 iiblnd - Infiniserv 3.3 + PathBits patch,
28 gmlnd - GM 2.1.22 and later,
29 mxlnd - MX 1.2.10 or later,
30 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
34 Description: With peer health detection, o2iblnd makes only one attempt to
35 reconnect which is not enough with nodes running lustre 1.6
36 because of proto version mismatch. Fix o2iblnd to retry one more
39 Severity : enhancement
41 Description: Quiet some LNET messages
43 Severity : enhancement
45 Description: Add OFED 1.5.1 support
47 Severity : enhancement
49 Description: The peer health code lacked some important debugging info in
50 lnd_query code paths. We've added necessary debug prints,
51 not just for bug 21678, but also for future troubleshooting.
53 -------------------------------------------------------------------------------
55 2010-04-30 Oracle, Inc.
57 * Support for networks:
58 socklnd - any kernel supported by Lustre,
59 qswlnd - Qsnet kernel modules 5.20 and later,
60 openiblnd - IbGold 1.8.2,
61 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, 1.3, 1.4.1, and 1.4.2
62 viblnd - Voltaire ibhost 3.4.5 and later,
63 ciblnd - Topspin 3.2.0,
64 iiblnd - Infiniserv 3.3 + PathBits patch,
65 gmlnd - GM 2.1.22 and later,
66 mxlnd - MX 1.2.10 or later,
67 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
71 Description: lst: check # of remained RPCs before aborting
72 Details : lstcon_rpc_trans_postwait() calls lstcon_rpc_trans_abort() only
73 when the transaction is timeout, so if we got "end_session" to
74 interrupt waiting on transaction, then we can hit the assertion
75 failure ASSERTION(crpc->crp_stamp != 0)
79 Description: print more debug info for timedout ZC-req
80 Details : Print more information for timedout ZC-req and partial
81 received connection. Close connection for timedout ZC-req
82 Always send ZC_ACK on non-blocking connection(BULK_IN)
86 Description: Adding WIRE_ATTR attribute to LNET types
87 Details : LST nodes on different platforms might not communicate well
88 due to the lack of WIRE_ATTR attribute in some LNET structures
89 traversing network. The patch fixes the problem by adding
90 WIRE_ATTR where needed.
94 Description: hash MEs on RDMA portal
95 Details : RDMA portal can have very long ME list on client side, which
96 will trigger soft lockup because of long searching on list.
97 Hash MEs on RDMA portal can resolve this problem.
101 Description: fix for double release of ibc_lock in o2iblnd
102 Details : Re-acquire ibc_lock in kiblnd_post_tx_locked(). Add extra
103 reference to conn before calling kiblnd_post_tx_locked()
104 to avoid scenario when conn disappears inside
105 kiblnd_post_tx_locked().
107 -------------------------------------------------------------------------------
108 2010-01-29 Sun Microsystems, Inc.
110 * Support for networks:
111 socklnd - any kernel supported by Lustre,
112 qswlnd - Qsnet kernel modules 5.20 and later,
113 openiblnd - IbGold 1.8.2,
114 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, 1.3, 1.4.1, and 1.4.2
115 viblnd - Voltaire ibhost 3.4.5 and later,
116 ciblnd - Topspin 3.2.0,
117 iiblnd - Infiniserv 3.3 + PathBits patch,
118 gmlnd - GM 2.1.22 and later,
119 mxlnd - MX 1.2.10 or later,
120 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
124 Description: should update lp_alive for non-router peers.
126 Severity : enhancement
128 Description: LNet router shuffler.
130 Severity : enhancement
132 Description: LNet fine grain routing support.
136 Description: router checker stops working when system wall clock goes backward
137 Details : use monotonic timing source instead of system wall clock time.
139 Severity : enhancement
141 Description: avoid asymmetrical router failures
143 Severity : enhancement
145 Description: multiple-instance support for kptllnd
149 Description: ksocknal_close_conn_locked connection race
150 Details : A race was possible when ksocknal_create_conn calls
151 ksocknal_close_conn_locked for already closed conn.
153 Severity : enhancement
155 Description: port router pinger to userspace
159 Description: kptllnd HELLO protocol deadlock
160 Details : kptllnd HELLO protocol doesn't run to completion in finite time
164 Description: LNet selftest fixes and enhancements
166 Severity : enhancement
168 Description: allow a test node to be a member of multiple test groups
170 Severity : enhancement
172 Description: MXLND: eliminate hosts file, use arp for peer nic_id resolution
173 Details : an update from the upstream developer Scott Atchley.
176 -------------------------------------------------------------------------------
177 2009-07-31 Sun Microsystems, Inc.
179 * Support for networks:
180 socklnd - any kernel supported by Lustre,
181 qswlnd - Qsnet kernel modules 5.20 and later,
182 openiblnd - IbGold 1.8.2,
183 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, 1.3, and 1.4.1
184 viblnd - Voltaire ibhost 3.4.5 and later,
185 ciblnd - Topspin 3.2.0,
186 iiblnd - Infiniserv 3.3 + PathBits patch,
187 gmlnd - GM 2.1.22 and later,
188 mxlnd - MX 1.2.1 or later,
189 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
193 Description: router_proc.c is rewritten to use sysctl-interface for parameters
194 residing in /proc/sys/lnet
196 Severity : enhancement
198 Description: add a new LND optiion to control peer buffer credits on routers
202 Description: Fixing deadlock in usocklnd
203 Details : A deadlock was possible in usocklnd due to race condition while
204 tearing connection down. The problem resulted from erroneous
205 assumption that lnet_finalize() could have been called holding
206 some lnd-level locks.
209 Bugzilla : 13621, 15983
210 Description: Protocol V2 of o2iblnd
211 Details : o2iblnd V2 has several new features:
212 . map-on-demand: map-on-demand is disabled by default, it can
213 be enabled by using modparam "map_on_demand=@value@", @value@
214 should >= 0 and < 256, 0 will disable map-on-demand, any other
215 valid value will enable map-on-demand.
216 Oi2blnd will create FMR or physical MR for RDMA if fragments of
218 Enable map-on-demand will take less memory for new connection,
219 but a little more CPU for RDMA.
220 . iWARP : to support iWARP, please enable map-on-demand, 32 and 64
221 are recommanded value. iWARP will probably fail for value >=128.
222 . OOB NOOP message: to resolve deadlock on router.
223 . tunable peer_credits_hiw: (high water to return credits),
224 default value of peer_credits_hiw equals to (peer_credits -1),
225 user can change it between peer_credits/2 and (peer_credits - 1).
226 Lower value is recommended for high latency network.
227 . tunable message queue size: it always equals to peer_credits,
228 higher value is recommended for high latency network.
229 . It's compatible with earlier version of o2iblnd
233 Description: Fixing 'running out of ports' issue
234 Details : Add a delay before next reconnect attempt in ksocklnd in
235 the case of lost race. Limit the frequency of query-requests
236 in lnet. Improved handling of 'dead peer' notifications in
241 Description: Change ptllnd timeout and watchdog timers
242 Details : Add ptltrace_on_nal_failed and bump ptllnd timeout to match
243 Portals wire timeout.
247 Description: One down Lustre FS hangs ALL mounted Lustre filesystems
248 Details : Shared routing enhancements - peer health detection.
252 Description: IB path MTU mistakenly set to 1st path MTU when ib_mtu is off
253 Details : See comment 46 in bug 11245 for details - it's indeed a bug
254 introduced by the original 11245 fix.
258 Description: uptllnd credit overflow fix
259 Details : kptl_msg_t::ptlm_credits could be overflown by uptllnd since
264 Description: socklnd protocol version 3
265 Details : With current protocol V2, connections on router can be
266 blocked and can't receive any incoming messages when there is no
267 more router buffer, so ZC-ACK can't be handled (LNet message
268 can't be finalized) and will cause deadlock on router.
269 Protocol V3 has a dedicated connection for emergency messages
270 like ZC-ACK to router, messages on this dedicated connection
271 don't need any credit so will never be blocked. Also, V3 can send
272 keepalive ping in specified period for router healthy checking.
274 -------------------------------------------------------------------------------
275 12-31-2008 Sun Microsystems, Inc.
277 * Support for networks:
278 socklnd - any kernel supported by Lustre,
279 qswlnd - Qsnet kernel modules 5.20 and later,
280 openiblnd - IbGold 1.8.2,
281 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, and 1.3
282 viblnd - Voltaire ibhost 3.4.5 and later,
283 ciblnd - Topspin 3.2.0,
284 iiblnd - Infiniserv 3.3 + PathBits patch,
285 gmlnd - GM 2.1.22 and later,
286 mxlnd - MX 1.2.1 or later,
287 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
296 Description: workaround for OOM from o2iblnd
297 Details : OFED needs allocate big chunk of memory for QP while creating
298 connection for o2iblnd, OOM can happen if no such a contiguous
300 QP size is decided by concurrent_sends and max_fragments of
301 o2iblnd, now we permit user to specify smaller value for
302 concurrent_sends of o2iblnd(i.e: concurrent_sends=7), which
303 will decrease memory block size required by creating QP.
307 Description: Support Zerocopy receive of Chelsio device
308 Details : Chelsio driver can support zerocopy for iov[1] if it's
309 contiguous and large enough.
313 Description: fix credit flow deadlock in uptllnd
317 Description: finalize network operation in reasonable time
318 Details : conf-sanity test_32a couldn't stop ost and mds because it
319 tried to access non-existent peer and tcp connect took
320 quite long before timing out.
324 Description: Continuous recovery on 33 of 413 nodes after lustre oss failure
325 Details : Lost reference on conn prevents peer from being destroyed, which
326 could prevent new peer creation if peer count has reached upper
331 Description: LNET Selftest results in Soft lockup on OSS CPU
332 Details : only hits when 8 or more o2ib clients involved and a session is
333 torn down with 'lst end_session' without preceeding 'lst stop'.
337 Description: concurrent_sends in IB LNDs should not be changeable at run time
338 Details : concurrent_sends in IB LNDs should not be changeable at run time
342 Description: ptl_send_rpc hits LASSERT when ptl_send_buf fails
343 Details : only hits under out-of-memory situations
346 -------------------------------------------------------------------------------
348 2009-02-07 Sun Microsystems, Inc.
350 * Support for networks:
351 socklnd - any kernel supported by Lustre,
352 qswlnd - Qsnet kernel modules 5.20 and later,
353 openiblnd - IbGold 1.8.2,
354 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, and 1.3
355 viblnd - Voltaire ibhost 3.4.5 and later,
356 ciblnd - Topspin 3.2.0,
357 iiblnd - Infiniserv 3.3 + PathBits patch,
358 gmlnd - GM 2.1.22 and later,
359 mxlnd - MX 1.2.1 or later,
360 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
364 Description: workaround for OOM from o2iblnd
365 Details : OFED needs allocate big chunk of memory for QP while creating
366 connection for o2iblnd, OOM can happen if no such a contiguous
368 QP size is decided by concurrent_sends and max_fragments of
369 o2iblnd, now we permit user to specify smaller value for
370 concurrent_sends of o2iblnd(i.e: concurrent_sends=7), which
371 will decrease memory block size required by creating QP.
375 Description: Support Zerocopy receive of Chelsio device
376 Details : Chelsio driver can support zerocopy for iov[1] if it's
377 contiguous and large enough.
381 Description: fix credit flow deadlock in uptllnd
385 Description: finalize network operation in reasonable time
386 Details : conf-sanity test_32a couldn't stop ost and mds because it
387 tried to access non-existent peer and tcp connect took
388 quite long before timing out.
392 Description: Continuous recovery on 33 of 413 nodes after lustre oss failure
393 Details : Lost reference on conn prevents peer from being destroyed, which
394 could prevent new peer creation if peer count has reached upper
399 Description: LNET Selftest results in Soft lockup on OSS CPU
400 Details : only hits when 8 or more o2ib clients involved and a session is
401 torn down with 'lst end_session' without preceeding 'lst stop'.
405 Description: concurrent_sends in IB LNDs should not be changeable at run time
406 Details : concurrent_sends in IB LNDs should not be changeable at run time
408 -------------------------------------------------------------------------------
410 11-03-2008 Sun Microsystems, Inc.
412 * Support for networks:
413 socklnd - any kernel supported by Lustre,
414 qswlnd - Qsnet kernel modules 5.20 and later,
415 openiblnd - IbGold 1.8.2,
416 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, and 1.3
417 viblnd - Voltaire ibhost 3.4.5 and later,
418 ciblnd - Topspin 3.2.0,
419 iiblnd - Infiniserv 3.3 + PathBits patch,
420 gmlnd - GM 2.1.22 and later,
421 mxlnd - MX 1.2.1 or later,
422 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
426 Description: ptl_send_rpc hits LASSERT when ptl_send_buf fails
427 Details : only hits under out-of-memory situations
429 -------------------------------------------------------------------------------
432 04-26-2008 Sun Microsystems, Inc.
434 * Support for networks:
435 socklnd - any kernel supported by Lustre,
436 qswlnd - Qsnet kernel modules 5.20 and later,
437 openiblnd - IbGold 1.8.2,
438 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5
439 viblnd - Voltaire ibhost 3.4.5 and later,
440 ciblnd - Topspin 3.2.0,
441 iiblnd - Infiniserv 3.3 + PathBits patch,
442 gmlnd - GM 2.1.22 and later,
443 mxlnd - MX 1.2.1 or later,
444 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
448 Description: excessive debug information removed
449 Details : excessive debug information removed
453 Description: ksocknal_create_conn() hit ASSERTION during connection race
454 Details : ksocknal_create_conn() hit ASSERTION during connection race
458 Description: ksocknal_send_hello() hit ASSERTION while connecting race
459 Details : ksocknal_send_hello() hit ASSERTION while connecting race
463 Description: o2iblnd/ptllnd credit deadlock in a routed config.
464 Details : o2iblnd/ptllnd credit deadlock in a routed config.
468 Description: High load after starting lnet
469 Details : gmlnd should sleep in rx thread in interruptible way. Otherwise,
470 uptime utility reports high load that looks confusingly.
474 Description: ksocklnd fails to establish connection if accept_port is high
475 Details : PID remapping must not be done for active (outgoing) connections
477 --------------------------------------------------------------------------------
479 2008-01-11 Sun Microsystems, Inc.
481 * Support for networks:
482 socklnd - any kernel supported by Lustre,
483 qswlnd - Qsnet kernel modules 5.20 and later,
484 openiblnd - IbGold 1.8.2,
485 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5
486 viblnd - Voltaire ibhost 3.4.5 and later,
487 ciblnd - Topspin 3.2.0,
488 iiblnd - Infiniserv 3.3 + PathBits patch,
489 gmlnd - GM 2.1.22 and later,
490 mxlnd - MX 1.2.1 or later,
491 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
494 Description: liblustre network error
495 Details : liblustre clients should understand LNET_ACCEPT_PORT environment
496 variable even if they don't start lnet acceptor.
500 Description: Strange message from lnet (Ignoring prediction from the future)
501 Details : Incorrect calculation of peer's last_alive value in ksocklnd
503 --------------------------------------------------------------------------------
505 2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
507 * Support for networks:
508 socklnd - any kernel supported by Lustre,
509 qswlnd - Qsnet kernel modules 5.20 and later,
510 openiblnd - IbGold 1.8.2,
511 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5.
512 viblnd - Voltaire ibhost 3.4.5 and later,
513 ciblnd - Topspin 3.2.0,
514 iiblnd - Infiniserv 3.3 + PathBits patch,
515 gmlnd - GM 2.1.22 and later,
516 mxlnd - MX 1.2.1 or later,
517 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
521 Description: ASSERTION(me == md->md_me) failed in lnet_match_md()
525 Description: increase send queue size for ciblnd/openiblnd
529 Description: new userspace socklnd
530 Details : Old userspace tcpnal that resided in lnet/ulnds/socklnd replaced
531 with new one - usocklnd.
533 Severity : enhancement
535 Description: Console message flood
536 Details : Make cdls ratelimiting more tunable by adding several tunable in
537 procfs /proc/sys/lnet/console_{min,max}_delay_centisecs and
538 /proc/sys/lnet/console_backoff.
540 --------------------------------------------------------------------------------
542 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
544 * Support for networks:
545 socklnd - any kernel supported by Lustre,
546 qswlnd - Qsnet kernel modules 5.20 and later,
547 openiblnd - IbGold 1.8.2,
548 o2iblnd - OFED 1.1 and 1.2,
549 viblnd - Voltaire ibhost 3.4.5 and later,
550 ciblnd - Topspin 3.2.0,
551 iiblnd - Infiniserv 3.3 + PathBits patch,
552 gmlnd - GM 2.1.22 and later,
553 mxlnd - MX 1.2.1 or later,
554 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
558 Description: /proc/sys/lnet has non-sysctl entries
559 Details : Updating dump_kernel/daemon_file/debug_mb to use sysctl variables
563 Description: TOE Kernel panic by ksocklnd
564 Details : offloaded sockets provide their own implementation of sendpage,
565 can't call tcp_sendpage() directly
569 Description: kibnal_shutdown() doesn't finish; lconf --cleanup hangs
570 Details : races between lnd_shutdown and peer creation prevent
571 lnd_shutdown from finishing.
575 Description: open files rlimit 1024 reached while liblustre testing
576 Details : ulnds/socklnd must close open socket after unsuccessful
581 Description: build error
582 Details : fix typos in gmlnd, ptllnd and viblnd
584 ------------------------------------------------------------------------------
586 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
588 * Support for networks:
589 socklnd - kernels up to 2.6.16,
590 qswlnd - Qsnet kernel modules 5.20 and later,
591 openiblnd - IbGold 1.8.2,
592 o2iblnd - OFED 1.1 and 1.2
593 viblnd - Voltaire ibhost 3.4.5 and later,
594 ciblnd - Topspin 3.2.0,
595 iiblnd - Infiniserv 3.3 + PathBits patch,
596 gmlnd - GM 2.1.22 and later,
597 mxlnd - MX 1.2.1 or later,
598 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
600 2007-06-21 Cluster File Systems, Inc. <info@clusterfs.com>
602 * Support for networks:
603 socklnd - kernels up to 2.6.16,
604 qswlnd - Qsnet kernel modules 5.20 and later,
605 openiblnd - IbGold 1.8.2,
607 viblnd - Voltaire ibhost 3.4.5 and later,
608 ciblnd - Topspin 3.2.0,
609 iiblnd - Infiniserv 3.3 + PathBits patch,
610 gmlnd - GM 2.1.22 and later,
611 mxlnd - MX 1.2.1 or later,
612 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
616 Description: Initialize cpumask before use
620 Description: ASSERTION failures when upgrading to the patchless zero-copy
622 Details : This bug affects "rolling upgrades", causing an inconsistent
623 protocol version negotiation and subsequent assertion failure
624 during rolling upgrades after the first wave of upgrades.
628 Details : Change "dropped message" CERRORs to D_NETERROR so they are
629 logged instead of creating "console chatter" when a lustre
630 timeout races with normal RPC completion.
633 Details : lnet_clear_peer_table can wait forever if user forgets to
637 Details : libcfs_id2str should check pid against LNET_PID_ANY.
641 Description: added LNET self test
642 Details : landing b_self_test
647 Description: cfs_duration_{u,n}sec() wrongly calculate nanosecond part of
649 Details : do_div() macro is used incorrectly.
651 2007-04-23 Cluster File Systems, Inc. <info@clusterfs.com>
655 Description: make panic on lbug configurable
659 Description: Add OFED1.2 support to o2iblnd
660 Details : o2iblnd depends on OFED's modules, if out-tree OFED's modules
661 are installed (other than kernel's in-tree infiniband), there
662 could be some problem while insmod o2iblnd (mismatch CRC of
664 If extra Module.symvers is supported in kernel (i.e, 2.6.17),
665 this link provides solution:
666 https://bugs.openfabrics.org/show_bug.cgi?id=355
667 if extra Module.symvers is not supported in kernel, we will
668 have to run the script in bug 12316 to update
669 $LINUX/module.symvers before building o2iblnd.
670 More details about this are in bug 12316.
672 ------------------------------------------------------------------------------
674 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
675 * version 1.4.10 / 1.6.0
676 * Support for networks:
677 socklnd - kernels up to 2.6.16,
678 qswlnd - Qsnet kernel modules 5.20 and later,
679 openiblnd - IbGold 1.8.2,
681 viblnd - Voltaire ibhost 3.4.5 and later,
682 ciblnd - Topspin 3.2.0,
683 iiblnd - Infiniserv 3.3 + PathBits patch,
684 gmlnd - GM 2.1.22 and later,
685 mxlnd - MX 1.2.1 or later,
686 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
690 Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
691 possibly accessed in kptllnd_shutdown. Ptllnd should init
692 kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
696 Description: gmlnd ignored some transmit errors when finalizing lnet messages.
700 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
704 Description: the_lnet.ln_finalizing was not set when the current thread is
705 about to complete messages. It only affects multi-threaded
711 Description: Changed the default kqswlnd ntxmsg=512
716 Description: Assertion failure in kernel ptllnd caused by posting passive
717 bulk buffers before connection establishment complete.
722 Description: A race in kernel ptllnd between deleting a peer and posting
723 new communications for it could hang communications -
724 manifesting as "Unexpectedly long timeout" messages.
729 Description: Kernel ptllnd lock ordering issue could hang a node.
734 Description: node crash on socket teardown race
737 Frequency : 'lctl peer_list' issued on a mx net
739 Description: Enable lctl's peer_list for MXLND
742 Frequency : after Ptllnd timeouts and portals congestion
744 Description: Credit overflows
745 Details : This was a bug in ptllnd connection establishment. The fix
746 implements better peer stamps to disambiguate connection
747 establishment and ensure both peers enter the credit flow
748 state machine consistently.
753 Description: kptllnd didn't propagate some network errors up to LNET
754 Details : This bug was spotted while investigating 11394. The fix
755 ensures network errors on sends and bulk transfers are
756 propagated to LNET/lustre correctly.
758 Severity : enhancement
760 Description: Fixed console chatter in case of -ETIMEDOUT.
762 Severity : enhancement
764 Description: Added D_NETTRACE for recording network packet history
765 (initially only for ptllnd). Also a separate userspace
766 ptllnd facility to gather history which should really be
767 covered by D_NETTRACE too, if only CDEBUG recorded history in
773 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
774 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
775 callback can occur before a connection has actually been
776 established. This caused an assertion failure previously.
778 Severity : enhancement
780 Description: Multiple instances for o2iblnd
781 Details : Allow multiple instances of o2iblnd to enable networking over
782 multiple HCAs and routing between them.
786 Description: lnet deadlock in router_checker
787 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
788 into BH locks to eliminate potential deadlock caused by
789 ksocknal_data_ready() preempting code holding these locks.
793 Description: Millions of failed socklnd connection attempts cause a very slow FS
794 Details : added a new route flag ksnr_scheduled to distinguish from
795 ksnr_connecting, so that a peer connection request is only turned
796 down for race concerns when an active connection to the same peer
797 is under progress (instead of just being scheduled).
799 ------------------------------------------------------------------------------
801 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
803 * Support for networks:
804 socklnd - kernels up to 2.6.16
805 qswlnd - Qsnet kernel modules 5.20 and later
806 openiblnd - IbGold 1.8.2
808 viblnd - Voltaire ibhost 3.4.5 and later
809 ciblnd - Topspin 3.2.0
810 iiblnd - Infiniserv 3.3 + PathBits patch
811 gmlnd - GM 2.1.22 and later
812 mxlnd - MX 1.2.1 or later
813 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
816 Severity : major on XT3
818 Description: libcfs overwrites /proc/sys/portals
819 Details : libcfs created a symlink from /proc/sys/portals to
820 /proc/sys/lnet for backwards compatibility. This is no
821 longer required and makes the Cray portals /proc variables
826 Description: OFED FMR API change
827 Details : This changes parameter usage to reflect a change in
828 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
829 that FMR support is only used in experimental versions of the
830 o2iblnd - this change does not affect standard usage at all.
832 Severity : enhancement
834 Description: new ko2iblnd module parameter: ib_mtu
835 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
836 HCAs. You can avoid this problem by setting the MTU to 1024
837 using this module parameter.
839 Severity : enhancement
840 Bugzilla : 11118/11620
841 Description: ptllnd small request message buffer alignment fix
842 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
843 Round up small message size on sends in case this option
844 is not supported. 11620 was a defect in the initial
845 implementation which effectively asserted all peers had to be
846 running the correct protocol version which was fixed by always
847 NAK-ing such requests and handling any misalignments they
852 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
853 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
854 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
856 Severity : enhancement
858 Description: Patchless ZC(zero copy) socklnd
859 Details : New protocol for socklnd, socklnd can support zero copy without
860 kernel patch, it's compatible with old socklnd. Checksum is
861 moved from tunables to modparams.
865 Description: When ksocknal_del_peer() is called upon a peer whose
866 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
867 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
870 Frequency : when ptlrpc is under heavy use and runs out of request buffer
872 Description: In lnet_match_blocked_msg(), md can be used without holding a
876 Frequency : very rarely
878 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
879 If connd connects a route which has been closed by
880 ksocknal_shutdown(), ksocknal_create_routes() may create new
881 routes which hold references on the peer, causing shutdown
882 process to wait for peer to disappear forever.
884 Severity : enhancement
886 Description: Dump XT3 portals traces on kptllnd timeout
887 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
888 dump Cray portals debug traces to a file. The kptllnd module
889 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
890 is the basename of the dump file.
893 Frequency : infrequent
895 Description: kernel ptllnd fix bug in connection re-establishment
896 Details : Kernel ptllnd could produce protocol errors e.g. illegal
897 matchbits and/or violate the credit flow protocol when trying
898 to re-establish a connection with a peer after an error or
901 Severity : enhancement
903 Description: Allow /proc/sys/lnet/debug to be set symbolically
904 Details : Allow debug and subsystem debug values to be read/set by name
905 in addition to numerically, for ease of use.
908 Frequency : only in configurations with LNET routers
910 Description: routes automatically marked down and recovered
911 Details : In configurations with LNET routers if a router fails routers
912 now actively try to recover routes that are down, unless they
913 are marked down by an administrator.
915 ------------------------------------------------------------------------------
917 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
920 Frequency : very rarely, in configurations with LNET routers and TCP
922 Description: incorrect data written to files on OSTs
923 Details : In certain high-load conditions incorrect data may be written
924 to files on the OST when using TCP networks.
926 ------------------------------------------------------------------------------
928 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
930 - rework CDEBUG messages rate-limiting mechanism b=10375
931 - add per-socket tunables for socklnd if the kernel is patched b=10327
933 ------------------------------------------------------------------------------
935 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
937 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
938 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
939 - quiet LNET startup LNI message for liblustre b=10128
940 - Better console error messages if 'ip2nets' can't match an IP address
941 - Fixed overflow/use-before-set bugs in linux-time.h
942 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
943 - LNET teardown failed an assertion about the route table being empty
944 - Fixed a crash in LNetEQPoll(<invalid handle>)
945 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
946 - improve debug message for liblustre/Catamount nodes (b=10116)
948 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
949 * Configuration change for the XT3
950 The PTLLND is now used to run Lustre over Portals on the XT3.
951 The configure option(s) --with-cray-portals are no longer
952 used. Rather --with-portals=<path-to-portals-includes> is
953 used to enable building on the XT3. In addition to enable
954 XT3 specific features the option --enable-cray-xt3 must be
957 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
958 * Portals has been removed, replaced by LNET.
959 LNET is new networking infrastructure for Lustre, it includes a
960 reorganized network configuration mode (see the user
961 documentation for full details) as well as support for routing
962 between different network fabrics. Lustre Networking Devices
963 (LNDS) for the supported network fabrics have also been created
964 for this new infrastructure.
966 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
971 Frequency : rare (large Voltaire clusters only)
973 Description: the default number of reserved transmit descriptors was too low
974 for some large clusters
975 Details : As a workaround, the number was increased. A proper fix includes
978 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
983 Frequency : occasional (large-scale events, cluster reboot, network failure)
985 Description: too many error messages on console obscure actual problem and
986 can slow down/panic server, or cause recovery to fail repeatedly
987 Details : enable rate-limiting of console error messages, and some messages
988 that were console errors now only go to the kernel log
990 Severity : enhancement
992 Description: add /proc/sys/portals/catastrophe entry which will report if
993 that node has previously LBUGged
995 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
997 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
999 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
1001 - handle error return code in kranal_check_fma_rx() (5915,6054)
1003 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
1005 - update vibnal (Voltaire IB NAL)
1006 - update gmnal (Myrinet NAL), gmnalid
1008 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
1010 * Landed portals:b_port_step as follows...
1012 - removed CFS_DECL_SPIN*
1013 just use 'spinlock_t' and initialise with spin_lock_init()
1015 - removed CFS_DECL_MUTEX*
1016 just use 'struct semaphore' and initialise with init_mutex()
1018 - removed CFS_DECL_RWSEM*
1019 just use 'struct rw_semaphore' and initialise with init_rwsem()
1021 - renamed cfs_sleep_chan -> cfs_waitq
1022 cfs_sleep_link -> cfs_waitlink
1024 - fixed race in linux version of arch-independent socknal
1025 (the ENOMEM/EAGAIN decision).
1027 - Didn't fix problems in Darwin version of arch-independent socknal
1028 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
1030 - removed libcfs types from non-socknal header files (only some types
1031 in the header files had been changed; the .c files hadn't been