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
15 -------------------------------------------------------------------------------
17 2010-10-29 Oracle, Inc.
19 * Support for networks:
20 socklnd - any kernel supported by Lustre,
21 qswlnd - Qsnet kernel modules 5.20 and later,
22 openiblnd - IbGold 1.8.2,
23 o2iblnd - OFED 1.3, 1.4.1, 1.4.2, 1.5.1 and 1.5.2
24 viblnd - Voltaire ibhost 3.4.5 and later,
25 ciblnd - Topspin 3.2.0,
26 iiblnd - Infiniserv 3.3 + PathBits patch,
27 gmlnd - GM 2.1.22 and later,
28 mxlnd - MX 1.2.10 or later,
29 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
33 Description: MDS node unresponsive
34 Details : improve hash distribution, doubled hash size for the lnet cookie
37 -------------------------------------------------------------------------------
39 2010-07-31 Oracle, Inc.
41 * Support for networks:
42 socklnd - any kernel supported by Lustre,
43 qswlnd - Qsnet kernel modules 5.20 and later,
44 openiblnd - IbGold 1.8.2,
45 o2iblnd - OFED 1.3, 1.4.1, 1.4.2 and 1.5.1
46 viblnd - Voltaire ibhost 3.4.5 and later,
47 ciblnd - Topspin 3.2.0,
48 iiblnd - Infiniserv 3.3 + PathBits patch,
49 gmlnd - GM 2.1.22 and later,
50 mxlnd - MX 1.2.10 or later,
51 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
55 Description: With peer health detection, o2iblnd makes only one attempt to
56 reconnect which is not enough with nodes running lustre 1.6
57 because of proto version mismatch. Fix o2iblnd to retry one more
60 Severity : enhancement
62 Description: Quiet some LNET messages
64 Severity : enhancement
66 Description: Add OFED 1.5.1 support
68 Severity : enhancement
70 Description: The peer health code lacked some important debugging info in
71 lnd_query code paths. We've added necessary debug prints,
72 not just for bug 21678, but also for future troubleshooting.
74 -------------------------------------------------------------------------------
76 2010-04-30 Oracle, Inc.
78 * Support for networks:
79 socklnd - any kernel supported by Lustre,
80 qswlnd - Qsnet kernel modules 5.20 and later,
81 openiblnd - IbGold 1.8.2,
82 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, 1.3, 1.4.1, and 1.4.2
83 viblnd - Voltaire ibhost 3.4.5 and later,
84 ciblnd - Topspin 3.2.0,
85 iiblnd - Infiniserv 3.3 + PathBits patch,
86 gmlnd - GM 2.1.22 and later,
87 mxlnd - MX 1.2.10 or later,
88 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
92 Description: lst: check # of remained RPCs before aborting
93 Details : lstcon_rpc_trans_postwait() calls lstcon_rpc_trans_abort() only
94 when the transaction is timeout, so if we got "end_session" to
95 interrupt waiting on transaction, then we can hit the assertion
96 failure ASSERTION(crpc->crp_stamp != 0)
100 Description: print more debug info for timedout ZC-req
101 Details : Print more information for timedout ZC-req and partial
102 received connection. Close connection for timedout ZC-req
103 Always send ZC_ACK on non-blocking connection(BULK_IN)
107 Description: Adding WIRE_ATTR attribute to LNET types
108 Details : LST nodes on different platforms might not communicate well
109 due to the lack of WIRE_ATTR attribute in some LNET structures
110 traversing network. The patch fixes the problem by adding
111 WIRE_ATTR where needed.
115 Description: hash MEs on RDMA portal
116 Details : RDMA portal can have very long ME list on client side, which
117 will trigger soft lockup because of long searching on list.
118 Hash MEs on RDMA portal can resolve this problem.
122 Description: fix for double release of ibc_lock in o2iblnd
123 Details : Re-acquire ibc_lock in kiblnd_post_tx_locked(). Add extra
124 reference to conn before calling kiblnd_post_tx_locked()
125 to avoid scenario when conn disappears inside
126 kiblnd_post_tx_locked().
128 -------------------------------------------------------------------------------
129 2010-01-29 Sun Microsystems, Inc.
131 * Support for networks:
132 socklnd - any kernel supported by Lustre,
133 qswlnd - Qsnet kernel modules 5.20 and later,
134 openiblnd - IbGold 1.8.2,
135 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, 1.3, 1.4.1, and 1.4.2
136 viblnd - Voltaire ibhost 3.4.5 and later,
137 ciblnd - Topspin 3.2.0,
138 iiblnd - Infiniserv 3.3 + PathBits patch,
139 gmlnd - GM 2.1.22 and later,
140 mxlnd - MX 1.2.10 or later,
141 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
145 Description: should update lp_alive for non-router peers.
147 Severity : enhancement
149 Description: LNet router shuffler.
151 Severity : enhancement
153 Description: LNet fine grain routing support.
157 Description: router checker stops working when system wall clock goes backward
158 Details : use monotonic timing source instead of system wall clock time.
160 Severity : enhancement
162 Description: avoid asymmetrical router failures
164 Severity : enhancement
166 Description: multiple-instance support for kptllnd
170 Description: ksocknal_close_conn_locked connection race
171 Details : A race was possible when ksocknal_create_conn calls
172 ksocknal_close_conn_locked for already closed conn.
174 Severity : enhancement
176 Description: port router pinger to userspace
180 Description: kptllnd HELLO protocol deadlock
181 Details : kptllnd HELLO protocol doesn't run to completion in finite time
185 Description: LNet selftest fixes and enhancements
187 Severity : enhancement
189 Description: allow a test node to be a member of multiple test groups
191 Severity : enhancement
193 Description: MXLND: eliminate hosts file, use arp for peer nic_id resolution
194 Details : an update from the upstream developer Scott Atchley.
197 -------------------------------------------------------------------------------
198 2009-07-31 Sun Microsystems, Inc.
200 * Support for networks:
201 socklnd - any kernel supported by Lustre,
202 qswlnd - Qsnet kernel modules 5.20 and later,
203 openiblnd - IbGold 1.8.2,
204 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, 1.3, and 1.4.1
205 viblnd - Voltaire ibhost 3.4.5 and later,
206 ciblnd - Topspin 3.2.0,
207 iiblnd - Infiniserv 3.3 + PathBits patch,
208 gmlnd - GM 2.1.22 and later,
209 mxlnd - MX 1.2.1 or later,
210 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
214 Description: router_proc.c is rewritten to use sysctl-interface for parameters
215 residing in /proc/sys/lnet
217 Severity : enhancement
219 Description: add a new LND optiion to control peer buffer credits on routers
223 Description: Fixing deadlock in usocklnd
224 Details : A deadlock was possible in usocklnd due to race condition while
225 tearing connection down. The problem resulted from erroneous
226 assumption that lnet_finalize() could have been called holding
227 some lnd-level locks.
230 Bugzilla : 13621, 15983
231 Description: Protocol V2 of o2iblnd
232 Details : o2iblnd V2 has several new features:
233 . map-on-demand: map-on-demand is disabled by default, it can
234 be enabled by using modparam "map_on_demand=@value@", @value@
235 should >= 0 and < 256, 0 will disable map-on-demand, any other
236 valid value will enable map-on-demand.
237 Oi2blnd will create FMR or physical MR for RDMA if fragments of
239 Enable map-on-demand will take less memory for new connection,
240 but a little more CPU for RDMA.
241 . iWARP : to support iWARP, please enable map-on-demand, 32 and 64
242 are recommanded value. iWARP will probably fail for value >=128.
243 . OOB NOOP message: to resolve deadlock on router.
244 . tunable peer_credits_hiw: (high water to return credits),
245 default value of peer_credits_hiw equals to (peer_credits -1),
246 user can change it between peer_credits/2 and (peer_credits - 1).
247 Lower value is recommended for high latency network.
248 . tunable message queue size: it always equals to peer_credits,
249 higher value is recommended for high latency network.
250 . It's compatible with earlier version of o2iblnd
254 Description: Fixing 'running out of ports' issue
255 Details : Add a delay before next reconnect attempt in ksocklnd in
256 the case of lost race. Limit the frequency of query-requests
257 in lnet. Improved handling of 'dead peer' notifications in
262 Description: Change ptllnd timeout and watchdog timers
263 Details : Add ptltrace_on_nal_failed and bump ptllnd timeout to match
264 Portals wire timeout.
268 Description: One down Lustre FS hangs ALL mounted Lustre filesystems
269 Details : Shared routing enhancements - peer health detection.
273 Description: IB path MTU mistakenly set to 1st path MTU when ib_mtu is off
274 Details : See comment 46 in bug 11245 for details - it's indeed a bug
275 introduced by the original 11245 fix.
279 Description: uptllnd credit overflow fix
280 Details : kptl_msg_t::ptlm_credits could be overflown by uptllnd since
285 Description: socklnd protocol version 3
286 Details : With current protocol V2, connections on router can be
287 blocked and can't receive any incoming messages when there is no
288 more router buffer, so ZC-ACK can't be handled (LNet message
289 can't be finalized) and will cause deadlock on router.
290 Protocol V3 has a dedicated connection for emergency messages
291 like ZC-ACK to router, messages on this dedicated connection
292 don't need any credit so will never be blocked. Also, V3 can send
293 keepalive ping in specified period for router healthy checking.
295 -------------------------------------------------------------------------------
296 12-31-2008 Sun Microsystems, Inc.
298 * Support for networks:
299 socklnd - any kernel supported by Lustre,
300 qswlnd - Qsnet kernel modules 5.20 and later,
301 openiblnd - IbGold 1.8.2,
302 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, and 1.3
303 viblnd - Voltaire ibhost 3.4.5 and later,
304 ciblnd - Topspin 3.2.0,
305 iiblnd - Infiniserv 3.3 + PathBits patch,
306 gmlnd - GM 2.1.22 and later,
307 mxlnd - MX 1.2.1 or later,
308 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
317 Description: workaround for OOM from o2iblnd
318 Details : OFED needs allocate big chunk of memory for QP while creating
319 connection for o2iblnd, OOM can happen if no such a contiguous
321 QP size is decided by concurrent_sends and max_fragments of
322 o2iblnd, now we permit user to specify smaller value for
323 concurrent_sends of o2iblnd(i.e: concurrent_sends=7), which
324 will decrease memory block size required by creating QP.
328 Description: Support Zerocopy receive of Chelsio device
329 Details : Chelsio driver can support zerocopy for iov[1] if it's
330 contiguous and large enough.
334 Description: fix credit flow deadlock in uptllnd
338 Description: finalize network operation in reasonable time
339 Details : conf-sanity test_32a couldn't stop ost and mds because it
340 tried to access non-existent peer and tcp connect took
341 quite long before timing out.
345 Description: Continuous recovery on 33 of 413 nodes after lustre oss failure
346 Details : Lost reference on conn prevents peer from being destroyed, which
347 could prevent new peer creation if peer count has reached upper
352 Description: LNET Selftest results in Soft lockup on OSS CPU
353 Details : only hits when 8 or more o2ib clients involved and a session is
354 torn down with 'lst end_session' without preceeding 'lst stop'.
358 Description: concurrent_sends in IB LNDs should not be changeable at run time
359 Details : concurrent_sends in IB LNDs should not be changeable at run time
363 Description: ptl_send_rpc hits LASSERT when ptl_send_buf fails
364 Details : only hits under out-of-memory situations
367 -------------------------------------------------------------------------------
369 2009-02-07 Sun Microsystems, Inc.
371 * Support for networks:
372 socklnd - any kernel supported by Lustre,
373 qswlnd - Qsnet kernel modules 5.20 and later,
374 openiblnd - IbGold 1.8.2,
375 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, and 1.3
376 viblnd - Voltaire ibhost 3.4.5 and later,
377 ciblnd - Topspin 3.2.0,
378 iiblnd - Infiniserv 3.3 + PathBits patch,
379 gmlnd - GM 2.1.22 and later,
380 mxlnd - MX 1.2.1 or later,
381 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
385 Description: workaround for OOM from o2iblnd
386 Details : OFED needs allocate big chunk of memory for QP while creating
387 connection for o2iblnd, OOM can happen if no such a contiguous
389 QP size is decided by concurrent_sends and max_fragments of
390 o2iblnd, now we permit user to specify smaller value for
391 concurrent_sends of o2iblnd(i.e: concurrent_sends=7), which
392 will decrease memory block size required by creating QP.
396 Description: Support Zerocopy receive of Chelsio device
397 Details : Chelsio driver can support zerocopy for iov[1] if it's
398 contiguous and large enough.
402 Description: fix credit flow deadlock in uptllnd
406 Description: finalize network operation in reasonable time
407 Details : conf-sanity test_32a couldn't stop ost and mds because it
408 tried to access non-existent peer and tcp connect took
409 quite long before timing out.
413 Description: Continuous recovery on 33 of 413 nodes after lustre oss failure
414 Details : Lost reference on conn prevents peer from being destroyed, which
415 could prevent new peer creation if peer count has reached upper
420 Description: LNET Selftest results in Soft lockup on OSS CPU
421 Details : only hits when 8 or more o2ib clients involved and a session is
422 torn down with 'lst end_session' without preceeding 'lst stop'.
426 Description: concurrent_sends in IB LNDs should not be changeable at run time
427 Details : concurrent_sends in IB LNDs should not be changeable at run time
429 -------------------------------------------------------------------------------
431 11-03-2008 Sun Microsystems, Inc.
433 * Support for networks:
434 socklnd - any kernel supported by Lustre,
435 qswlnd - Qsnet kernel modules 5.20 and later,
436 openiblnd - IbGold 1.8.2,
437 o2iblnd - OFED 1.1, 1.2.0, 1.2.5, and 1.3
438 viblnd - Voltaire ibhost 3.4.5 and later,
439 ciblnd - Topspin 3.2.0,
440 iiblnd - Infiniserv 3.3 + PathBits patch,
441 gmlnd - GM 2.1.22 and later,
442 mxlnd - MX 1.2.1 or later,
443 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
447 Description: ptl_send_rpc hits LASSERT when ptl_send_buf fails
448 Details : only hits under out-of-memory situations
450 -------------------------------------------------------------------------------
453 04-26-2008 Sun Microsystems, Inc.
455 * Support for networks:
456 socklnd - any kernel supported by Lustre,
457 qswlnd - Qsnet kernel modules 5.20 and later,
458 openiblnd - IbGold 1.8.2,
459 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5
460 viblnd - Voltaire ibhost 3.4.5 and later,
461 ciblnd - Topspin 3.2.0,
462 iiblnd - Infiniserv 3.3 + PathBits patch,
463 gmlnd - GM 2.1.22 and later,
464 mxlnd - MX 1.2.1 or later,
465 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
469 Description: excessive debug information removed
470 Details : excessive debug information removed
474 Description: ksocknal_create_conn() hit ASSERTION during connection race
475 Details : ksocknal_create_conn() hit ASSERTION during connection race
479 Description: ksocknal_send_hello() hit ASSERTION while connecting race
480 Details : ksocknal_send_hello() hit ASSERTION while connecting race
484 Description: o2iblnd/ptllnd credit deadlock in a routed config.
485 Details : o2iblnd/ptllnd credit deadlock in a routed config.
489 Description: High load after starting lnet
490 Details : gmlnd should sleep in rx thread in interruptible way. Otherwise,
491 uptime utility reports high load that looks confusingly.
495 Description: ksocklnd fails to establish connection if accept_port is high
496 Details : PID remapping must not be done for active (outgoing) connections
498 --------------------------------------------------------------------------------
500 2008-01-11 Sun Microsystems, Inc.
502 * Support for networks:
503 socklnd - any kernel supported by Lustre,
504 qswlnd - Qsnet kernel modules 5.20 and later,
505 openiblnd - IbGold 1.8.2,
506 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5
507 viblnd - Voltaire ibhost 3.4.5 and later,
508 ciblnd - Topspin 3.2.0,
509 iiblnd - Infiniserv 3.3 + PathBits patch,
510 gmlnd - GM 2.1.22 and later,
511 mxlnd - MX 1.2.1 or later,
512 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
515 Description: liblustre network error
516 Details : liblustre clients should understand LNET_ACCEPT_PORT environment
517 variable even if they don't start lnet acceptor.
521 Description: Strange message from lnet (Ignoring prediction from the future)
522 Details : Incorrect calculation of peer's last_alive value in ksocklnd
524 --------------------------------------------------------------------------------
526 2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
528 * Support for networks:
529 socklnd - any kernel supported by Lustre,
530 qswlnd - Qsnet kernel modules 5.20 and later,
531 openiblnd - IbGold 1.8.2,
532 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5.
533 viblnd - Voltaire ibhost 3.4.5 and later,
534 ciblnd - Topspin 3.2.0,
535 iiblnd - Infiniserv 3.3 + PathBits patch,
536 gmlnd - GM 2.1.22 and later,
537 mxlnd - MX 1.2.1 or later,
538 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
542 Description: ASSERTION(me == md->md_me) failed in lnet_match_md()
546 Description: increase send queue size for ciblnd/openiblnd
550 Description: new userspace socklnd
551 Details : Old userspace tcpnal that resided in lnet/ulnds/socklnd replaced
552 with new one - usocklnd.
554 Severity : enhancement
556 Description: Console message flood
557 Details : Make cdls ratelimiting more tunable by adding several tunable in
558 procfs /proc/sys/lnet/console_{min,max}_delay_centisecs and
559 /proc/sys/lnet/console_backoff.
561 --------------------------------------------------------------------------------
563 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
565 * Support for networks:
566 socklnd - any kernel supported by Lustre,
567 qswlnd - Qsnet kernel modules 5.20 and later,
568 openiblnd - IbGold 1.8.2,
569 o2iblnd - OFED 1.1 and 1.2,
570 viblnd - Voltaire ibhost 3.4.5 and later,
571 ciblnd - Topspin 3.2.0,
572 iiblnd - Infiniserv 3.3 + PathBits patch,
573 gmlnd - GM 2.1.22 and later,
574 mxlnd - MX 1.2.1 or later,
575 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
579 Description: /proc/sys/lnet has non-sysctl entries
580 Details : Updating dump_kernel/daemon_file/debug_mb to use sysctl variables
584 Description: TOE Kernel panic by ksocklnd
585 Details : offloaded sockets provide their own implementation of sendpage,
586 can't call tcp_sendpage() directly
590 Description: kibnal_shutdown() doesn't finish; lconf --cleanup hangs
591 Details : races between lnd_shutdown and peer creation prevent
592 lnd_shutdown from finishing.
596 Description: open files rlimit 1024 reached while liblustre testing
597 Details : ulnds/socklnd must close open socket after unsuccessful
602 Description: build error
603 Details : fix typos in gmlnd, ptllnd and viblnd
605 ------------------------------------------------------------------------------
607 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
609 * Support for networks:
610 socklnd - kernels up to 2.6.16,
611 qswlnd - Qsnet kernel modules 5.20 and later,
612 openiblnd - IbGold 1.8.2,
613 o2iblnd - OFED 1.1 and 1.2
614 viblnd - Voltaire ibhost 3.4.5 and later,
615 ciblnd - Topspin 3.2.0,
616 iiblnd - Infiniserv 3.3 + PathBits patch,
617 gmlnd - GM 2.1.22 and later,
618 mxlnd - MX 1.2.1 or later,
619 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
621 2007-06-21 Cluster File Systems, Inc. <info@clusterfs.com>
623 * Support for networks:
624 socklnd - kernels up to 2.6.16,
625 qswlnd - Qsnet kernel modules 5.20 and later,
626 openiblnd - IbGold 1.8.2,
628 viblnd - Voltaire ibhost 3.4.5 and later,
629 ciblnd - Topspin 3.2.0,
630 iiblnd - Infiniserv 3.3 + PathBits patch,
631 gmlnd - GM 2.1.22 and later,
632 mxlnd - MX 1.2.1 or later,
633 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
637 Description: Initialize cpumask before use
641 Description: ASSERTION failures when upgrading to the patchless zero-copy
643 Details : This bug affects "rolling upgrades", causing an inconsistent
644 protocol version negotiation and subsequent assertion failure
645 during rolling upgrades after the first wave of upgrades.
649 Details : Change "dropped message" CERRORs to D_NETERROR so they are
650 logged instead of creating "console chatter" when a lustre
651 timeout races with normal RPC completion.
654 Details : lnet_clear_peer_table can wait forever if user forgets to
658 Details : libcfs_id2str should check pid against LNET_PID_ANY.
662 Description: added LNET self test
663 Details : landing b_self_test
668 Description: cfs_duration_{u,n}sec() wrongly calculate nanosecond part of
670 Details : do_div() macro is used incorrectly.
672 2007-04-23 Cluster File Systems, Inc. <info@clusterfs.com>
676 Description: make panic on lbug configurable
680 Description: Add OFED1.2 support to o2iblnd
681 Details : o2iblnd depends on OFED's modules, if out-tree OFED's modules
682 are installed (other than kernel's in-tree infiniband), there
683 could be some problem while insmod o2iblnd (mismatch CRC of
685 If extra Module.symvers is supported in kernel (i.e, 2.6.17),
686 this link provides solution:
687 https://bugs.openfabrics.org/show_bug.cgi?id=355
688 if extra Module.symvers is not supported in kernel, we will
689 have to run the script in bug 12316 to update
690 $LINUX/module.symvers before building o2iblnd.
691 More details about this are in bug 12316.
693 ------------------------------------------------------------------------------
695 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
696 * version 1.4.10 / 1.6.0
697 * Support for networks:
698 socklnd - kernels up to 2.6.16,
699 qswlnd - Qsnet kernel modules 5.20 and later,
700 openiblnd - IbGold 1.8.2,
702 viblnd - Voltaire ibhost 3.4.5 and later,
703 ciblnd - Topspin 3.2.0,
704 iiblnd - Infiniserv 3.3 + PathBits patch,
705 gmlnd - GM 2.1.22 and later,
706 mxlnd - MX 1.2.1 or later,
707 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
711 Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
712 possibly accessed in kptllnd_shutdown. Ptllnd should init
713 kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
717 Description: gmlnd ignored some transmit errors when finalizing lnet messages.
721 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
725 Description: the_lnet.ln_finalizing was not set when the current thread is
726 about to complete messages. It only affects multi-threaded
732 Description: Changed the default kqswlnd ntxmsg=512
737 Description: Assertion failure in kernel ptllnd caused by posting passive
738 bulk buffers before connection establishment complete.
743 Description: A race in kernel ptllnd between deleting a peer and posting
744 new communications for it could hang communications -
745 manifesting as "Unexpectedly long timeout" messages.
750 Description: Kernel ptllnd lock ordering issue could hang a node.
755 Description: node crash on socket teardown race
758 Frequency : 'lctl peer_list' issued on a mx net
760 Description: Enable lctl's peer_list for MXLND
763 Frequency : after Ptllnd timeouts and portals congestion
765 Description: Credit overflows
766 Details : This was a bug in ptllnd connection establishment. The fix
767 implements better peer stamps to disambiguate connection
768 establishment and ensure both peers enter the credit flow
769 state machine consistently.
774 Description: kptllnd didn't propagate some network errors up to LNET
775 Details : This bug was spotted while investigating 11394. The fix
776 ensures network errors on sends and bulk transfers are
777 propagated to LNET/lustre correctly.
779 Severity : enhancement
781 Description: Fixed console chatter in case of -ETIMEDOUT.
783 Severity : enhancement
785 Description: Added D_NETTRACE for recording network packet history
786 (initially only for ptllnd). Also a separate userspace
787 ptllnd facility to gather history which should really be
788 covered by D_NETTRACE too, if only CDEBUG recorded history in
794 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
795 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
796 callback can occur before a connection has actually been
797 established. This caused an assertion failure previously.
799 Severity : enhancement
801 Description: Multiple instances for o2iblnd
802 Details : Allow multiple instances of o2iblnd to enable networking over
803 multiple HCAs and routing between them.
807 Description: lnet deadlock in router_checker
808 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
809 into BH locks to eliminate potential deadlock caused by
810 ksocknal_data_ready() preempting code holding these locks.
814 Description: Millions of failed socklnd connection attempts cause a very slow FS
815 Details : added a new route flag ksnr_scheduled to distinguish from
816 ksnr_connecting, so that a peer connection request is only turned
817 down for race concerns when an active connection to the same peer
818 is under progress (instead of just being scheduled).
820 ------------------------------------------------------------------------------
822 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
824 * Support for networks:
825 socklnd - kernels up to 2.6.16
826 qswlnd - Qsnet kernel modules 5.20 and later
827 openiblnd - IbGold 1.8.2
829 viblnd - Voltaire ibhost 3.4.5 and later
830 ciblnd - Topspin 3.2.0
831 iiblnd - Infiniserv 3.3 + PathBits patch
832 gmlnd - GM 2.1.22 and later
833 mxlnd - MX 1.2.1 or later
834 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
837 Severity : major on XT3
839 Description: libcfs overwrites /proc/sys/portals
840 Details : libcfs created a symlink from /proc/sys/portals to
841 /proc/sys/lnet for backwards compatibility. This is no
842 longer required and makes the Cray portals /proc variables
847 Description: OFED FMR API change
848 Details : This changes parameter usage to reflect a change in
849 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
850 that FMR support is only used in experimental versions of the
851 o2iblnd - this change does not affect standard usage at all.
853 Severity : enhancement
855 Description: new ko2iblnd module parameter: ib_mtu
856 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
857 HCAs. You can avoid this problem by setting the MTU to 1024
858 using this module parameter.
860 Severity : enhancement
861 Bugzilla : 11118/11620
862 Description: ptllnd small request message buffer alignment fix
863 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
864 Round up small message size on sends in case this option
865 is not supported. 11620 was a defect in the initial
866 implementation which effectively asserted all peers had to be
867 running the correct protocol version which was fixed by always
868 NAK-ing such requests and handling any misalignments they
873 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
874 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
875 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
877 Severity : enhancement
879 Description: Patchless ZC(zero copy) socklnd
880 Details : New protocol for socklnd, socklnd can support zero copy without
881 kernel patch, it's compatible with old socklnd. Checksum is
882 moved from tunables to modparams.
886 Description: When ksocknal_del_peer() is called upon a peer whose
887 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
888 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
891 Frequency : when ptlrpc is under heavy use and runs out of request buffer
893 Description: In lnet_match_blocked_msg(), md can be used without holding a
897 Frequency : very rarely
899 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
900 If connd connects a route which has been closed by
901 ksocknal_shutdown(), ksocknal_create_routes() may create new
902 routes which hold references on the peer, causing shutdown
903 process to wait for peer to disappear forever.
905 Severity : enhancement
907 Description: Dump XT3 portals traces on kptllnd timeout
908 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
909 dump Cray portals debug traces to a file. The kptllnd module
910 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
911 is the basename of the dump file.
914 Frequency : infrequent
916 Description: kernel ptllnd fix bug in connection re-establishment
917 Details : Kernel ptllnd could produce protocol errors e.g. illegal
918 matchbits and/or violate the credit flow protocol when trying
919 to re-establish a connection with a peer after an error or
922 Severity : enhancement
924 Description: Allow /proc/sys/lnet/debug to be set symbolically
925 Details : Allow debug and subsystem debug values to be read/set by name
926 in addition to numerically, for ease of use.
929 Frequency : only in configurations with LNET routers
931 Description: routes automatically marked down and recovered
932 Details : In configurations with LNET routers if a router fails routers
933 now actively try to recover routes that are down, unless they
934 are marked down by an administrator.
936 ------------------------------------------------------------------------------
938 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
941 Frequency : very rarely, in configurations with LNET routers and TCP
943 Description: incorrect data written to files on OSTs
944 Details : In certain high-load conditions incorrect data may be written
945 to files on the OST when using TCP networks.
947 ------------------------------------------------------------------------------
949 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
951 - rework CDEBUG messages rate-limiting mechanism b=10375
952 - add per-socket tunables for socklnd if the kernel is patched b=10327
954 ------------------------------------------------------------------------------
956 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
958 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
959 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
960 - quiet LNET startup LNI message for liblustre b=10128
961 - Better console error messages if 'ip2nets' can't match an IP address
962 - Fixed overflow/use-before-set bugs in linux-time.h
963 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
964 - LNET teardown failed an assertion about the route table being empty
965 - Fixed a crash in LNetEQPoll(<invalid handle>)
966 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
967 - improve debug message for liblustre/Catamount nodes (b=10116)
969 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
970 * Configuration change for the XT3
971 The PTLLND is now used to run Lustre over Portals on the XT3.
972 The configure option(s) --with-cray-portals are no longer
973 used. Rather --with-portals=<path-to-portals-includes> is
974 used to enable building on the XT3. In addition to enable
975 XT3 specific features the option --enable-cray-xt3 must be
978 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
979 * Portals has been removed, replaced by LNET.
980 LNET is new networking infrastructure for Lustre, it includes a
981 reorganized network configuration mode (see the user
982 documentation for full details) as well as support for routing
983 between different network fabrics. Lustre Networking Devices
984 (LNDS) for the supported network fabrics have also been created
985 for this new infrastructure.
987 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
992 Frequency : rare (large Voltaire clusters only)
994 Description: the default number of reserved transmit descriptors was too low
995 for some large clusters
996 Details : As a workaround, the number was increased. A proper fix includes
999 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
1004 Frequency : occasional (large-scale events, cluster reboot, network failure)
1006 Description: too many error messages on console obscure actual problem and
1007 can slow down/panic server, or cause recovery to fail repeatedly
1008 Details : enable rate-limiting of console error messages, and some messages
1009 that were console errors now only go to the kernel log
1011 Severity : enhancement
1013 Description: add /proc/sys/portals/catastrophe entry which will report if
1014 that node has previously LBUGged
1016 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
1018 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
1020 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
1022 - handle error return code in kranal_check_fma_rx() (5915,6054)
1024 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
1026 - update vibnal (Voltaire IB NAL)
1027 - update gmnal (Myrinet NAL), gmnalid
1029 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
1031 * Landed portals:b_port_step as follows...
1033 - removed CFS_DECL_SPIN*
1034 just use 'spinlock_t' and initialise with spin_lock_init()
1036 - removed CFS_DECL_MUTEX*
1037 just use 'struct semaphore' and initialise with init_mutex()
1039 - removed CFS_DECL_RWSEM*
1040 just use 'struct rw_semaphore' and initialise with init_rwsem()
1042 - renamed cfs_sleep_chan -> cfs_waitq
1043 cfs_sleep_link -> cfs_waitlink
1045 - fixed race in linux version of arch-independent socknal
1046 (the ENOMEM/EAGAIN decision).
1048 - Didn't fix problems in Darwin version of arch-independent socknal
1049 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
1051 - removed libcfs types from non-socknal header files (only some types
1052 in the header files had been changed; the .c files hadn't been