1 tbd Sun Microsystems, 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 1.1, 1.2.0, 1.2.5, and 1.3
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 1.2.1 or later,
13 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
22 Description: uptllnd credit overflow fix
23 Details : kptl_msg_t::ptlm_credits could be overflown by uptllnd since
28 Description: socklnd prtocol version 3
29 Details : With current protocol V2, connections on router can be
30 blocked and can't receive any incoming messages when there is no
31 more router buffer, so ZC-ACK can't be handled (LNet message
32 can't be finalized) and will cause deadlock on router.
33 Protocol V3 has a dedicated connection for emergency messages
34 like ZC-ACK to router, messages on this dedicated connection
35 don't need any credit so will never be blocked. Also, V3 can send
36 keepalive ping in specified period for router healthy checking.
40 Description: workaround for OOM from o2iblnd
41 Details : OFED needs allocate big chunk of memory for QP while creating
42 connection for o2iblnd, OOM can happen if no such a contiguous
44 QP size is decided by concurrent_sends and max_fragments of
45 o2iblnd, now we permit user to specify smaller value for
46 concurrent_sends of o2iblnd(i.e: concurrent_sends=7), which
47 will decrease memory block size required by creating QP.
51 Description: Support Zerocopy receive of Chelsio device
52 Details : Chelsio driver can support zerocopy for iov[1] if it's
53 contiguous and large enough.
57 Description: fix credit flow deadlock in uptllnd
61 Description: finalize network operation in reasonable time
62 Details : conf-sanity test_32a couldn't stop ost and mds because it
63 tried to access non-existent peer and tcp connect took
64 quite long before timing out.
68 Description: Remove portals compatibility
69 Details : Remove portals compatibility, not interoperable with releases
74 Description: Continuous recovery on 33 of 413 nodes after lustre oss failure
75 Details : Lost reference on conn prevents peer from being destroyed, which
76 could prevent new peer creation if peer count has reached upper
81 Description: LNET Selftest results in Soft lockup on OSS CPU
82 Details : only hits when 8 or more o2ib clients involved and a session is
83 torn down with 'lst end_session' without preceeding 'lst stop'.
87 Description: concurrent_sends in IB LNDs should not be changeable at run time
88 Details : concurrent_sends in IB LNDs should not be changeable at run time
92 Description: ptl_send_rpc hits LASSERT when ptl_send_buf fails
93 Details : only hits under out-of-memory situations
96 -------------------------------------------------------------------------------
99 04-26-2008 Sun Microsystems, Inc.
101 * Support for networks:
102 socklnd - any kernel supported by Lustre,
103 qswlnd - Qsnet kernel modules 5.20 and later,
104 openiblnd - IbGold 1.8.2,
105 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5
106 viblnd - Voltaire ibhost 3.4.5 and later,
107 ciblnd - Topspin 3.2.0,
108 iiblnd - Infiniserv 3.3 + PathBits patch,
109 gmlnd - GM 2.1.22 and later,
110 mxlnd - MX 1.2.1 or later,
111 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
115 Description: excessive debug information removed
116 Details : excessive debug information removed
120 Description: ksocknal_create_conn() hit ASSERTION during connection race
121 Details : ksocknal_create_conn() hit ASSERTION during connection race
125 Description: ksocknal_send_hello() hit ASSERTION while connecting race
126 Details : ksocknal_send_hello() hit ASSERTION while connecting race
130 Description: o2iblnd/ptllnd credit deadlock in a routed config.
131 Details : o2iblnd/ptllnd credit deadlock in a routed config.
135 Description: High load after starting lnet
136 Details : gmlnd should sleep in rx thread in interruptible way. Otherwise,
137 uptime utility reports high load that looks confusingly.
141 Description: ksocklnd fails to establish connection if accept_port is high
142 Details : PID remapping must not be done for active (outgoing) connections
144 --------------------------------------------------------------------------------
146 2008-01-11 Sun Microsystems, Inc.
148 * Support for networks:
149 socklnd - any kernel supported by Lustre,
150 qswlnd - Qsnet kernel modules 5.20 and later,
151 openiblnd - IbGold 1.8.2,
152 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5
153 viblnd - Voltaire ibhost 3.4.5 and later,
154 ciblnd - Topspin 3.2.0,
155 iiblnd - Infiniserv 3.3 + PathBits patch,
156 gmlnd - GM 2.1.22 and later,
157 mxlnd - MX 1.2.1 or later,
158 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
161 Description: liblustre network error
162 Details : liblustre clients should understand LNET_ACCEPT_PORT environment
163 variable even if they don't start lnet acceptor.
167 Description: Strange message from lnet (Ignoring prediction from the future)
168 Details : Incorrect calculation of peer's last_alive value in ksocklnd
170 --------------------------------------------------------------------------------
172 2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
174 * Support for networks:
175 socklnd - any kernel supported by Lustre,
176 qswlnd - Qsnet kernel modules 5.20 and later,
177 openiblnd - IbGold 1.8.2,
178 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5.
179 viblnd - Voltaire ibhost 3.4.5 and later,
180 ciblnd - Topspin 3.2.0,
181 iiblnd - Infiniserv 3.3 + PathBits patch,
182 gmlnd - GM 2.1.22 and later,
183 mxlnd - MX 1.2.1 or later,
184 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
188 Description: ASSERTION(me == md->md_me) failed in lnet_match_md()
192 Description: increase send queue size for ciblnd/openiblnd
196 Description: new userspace socklnd
197 Details : Old userspace tcpnal that resided in lnet/ulnds/socklnd replaced
198 with new one - usocklnd.
200 Severity : enhancement
202 Description: Console message flood
203 Details : Make cdls ratelimiting more tunable by adding several tunable in
204 procfs /proc/sys/lnet/console_{min,max}_delay_centisecs and
205 /proc/sys/lnet/console_backoff.
207 --------------------------------------------------------------------------------
209 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
211 * Support for networks:
212 socklnd - any kernel supported by Lustre,
213 qswlnd - Qsnet kernel modules 5.20 and later,
214 openiblnd - IbGold 1.8.2,
215 o2iblnd - OFED 1.1 and 1.2,
216 viblnd - Voltaire ibhost 3.4.5 and later,
217 ciblnd - Topspin 3.2.0,
218 iiblnd - Infiniserv 3.3 + PathBits patch,
219 gmlnd - GM 2.1.22 and later,
220 mxlnd - MX 1.2.1 or later,
221 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
225 Description: /proc/sys/lnet has non-sysctl entries
226 Details : Updating dump_kernel/daemon_file/debug_mb to use sysctl variables
230 Description: TOE Kernel panic by ksocklnd
231 Details : offloaded sockets provide their own implementation of sendpage,
232 can't call tcp_sendpage() directly
236 Description: kibnal_shutdown() doesn't finish; lconf --cleanup hangs
237 Details : races between lnd_shutdown and peer creation prevent
238 lnd_shutdown from finishing.
242 Description: open files rlimit 1024 reached while liblustre testing
243 Details : ulnds/socklnd must close open socket after unsuccessful
248 Description: build error
249 Details : fix typos in gmlnd, ptllnd and viblnd
251 ------------------------------------------------------------------------------
253 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
255 * Support for networks:
256 socklnd - kernels up to 2.6.16,
257 qswlnd - Qsnet kernel modules 5.20 and later,
258 openiblnd - IbGold 1.8.2,
259 o2iblnd - OFED 1.1 and 1.2
260 viblnd - Voltaire ibhost 3.4.5 and later,
261 ciblnd - Topspin 3.2.0,
262 iiblnd - Infiniserv 3.3 + PathBits patch,
263 gmlnd - GM 2.1.22 and later,
264 mxlnd - MX 1.2.1 or later,
265 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
267 2007-06-21 Cluster File Systems, Inc. <info@clusterfs.com>
269 * Support for networks:
270 socklnd - kernels up to 2.6.16,
271 qswlnd - Qsnet kernel modules 5.20 and later,
272 openiblnd - IbGold 1.8.2,
274 viblnd - Voltaire ibhost 3.4.5 and later,
275 ciblnd - Topspin 3.2.0,
276 iiblnd - Infiniserv 3.3 + PathBits patch,
277 gmlnd - GM 2.1.22 and later,
278 mxlnd - MX 1.2.1 or later,
279 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
283 Description: Initialize cpumask before use
287 Description: ASSERTION failures when upgrading to the patchless zero-copy
289 Details : This bug affects "rolling upgrades", causing an inconsistent
290 protocol version negotiation and subsequent assertion failure
291 during rolling upgrades after the first wave of upgrades.
295 Details : Change "dropped message" CERRORs to D_NETERROR so they are
296 logged instead of creating "console chatter" when a lustre
297 timeout races with normal RPC completion.
300 Details : lnet_clear_peer_table can wait forever if user forgets to
304 Details : libcfs_id2str should check pid against LNET_PID_ANY.
308 Description: added LNET self test
309 Details : landing b_self_test
314 Description: cfs_duration_{u,n}sec() wrongly calculate nanosecond part of
316 Details : do_div() macro is used incorrectly.
318 2007-04-23 Cluster File Systems, Inc. <info@clusterfs.com>
322 Description: make panic on lbug configurable
326 Description: Add OFED1.2 support to o2iblnd
327 Details : o2iblnd depends on OFED's modules, if out-tree OFED's modules
328 are installed (other than kernel's in-tree infiniband), there
329 could be some problem while insmod o2iblnd (mismatch CRC of
331 If extra Module.symvers is supported in kernel (i.e, 2.6.17),
332 this link provides solution:
333 https://bugs.openfabrics.org/show_bug.cgi?id=355
334 if extra Module.symvers is not supported in kernel, we will
335 have to run the script in bug 12316 to update
336 $LINUX/module.symvers before building o2iblnd.
337 More details about this are in bug 12316.
339 ------------------------------------------------------------------------------
341 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
342 * version 1.4.10 / 1.6.0
343 * Support for networks:
344 socklnd - kernels up to 2.6.16,
345 qswlnd - Qsnet kernel modules 5.20 and later,
346 openiblnd - IbGold 1.8.2,
348 viblnd - Voltaire ibhost 3.4.5 and later,
349 ciblnd - Topspin 3.2.0,
350 iiblnd - Infiniserv 3.3 + PathBits patch,
351 gmlnd - GM 2.1.22 and later,
352 mxlnd - MX 1.2.1 or later,
353 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
357 Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
358 possibly accessed in kptllnd_shutdown. Ptllnd should init
359 kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
363 Description: gmlnd ignored some transmit errors when finalizing lnet messages.
367 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
371 Description: the_lnet.ln_finalizing was not set when the current thread is
372 about to complete messages. It only affects multi-threaded
378 Description: Changed the default kqswlnd ntxmsg=512
383 Description: Assertion failure in kernel ptllnd caused by posting passive
384 bulk buffers before connection establishment complete.
389 Description: A race in kernel ptllnd between deleting a peer and posting
390 new communications for it could hang communications -
391 manifesting as "Unexpectedly long timeout" messages.
396 Description: Kernel ptllnd lock ordering issue could hang a node.
401 Description: node crash on socket teardown race
404 Frequency : 'lctl peer_list' issued on a mx net
406 Description: Enable lctl's peer_list for MXLND
409 Frequency : after Ptllnd timeouts and portals congestion
411 Description: Credit overflows
412 Details : This was a bug in ptllnd connection establishment. The fix
413 implements better peer stamps to disambiguate connection
414 establishment and ensure both peers enter the credit flow
415 state machine consistently.
420 Description: kptllnd didn't propagate some network errors up to LNET
421 Details : This bug was spotted while investigating 11394. The fix
422 ensures network errors on sends and bulk transfers are
423 propagated to LNET/lustre correctly.
425 Severity : enhancement
427 Description: Fixed console chatter in case of -ETIMEDOUT.
429 Severity : enhancement
431 Description: Added D_NETTRACE for recording network packet history
432 (initially only for ptllnd). Also a separate userspace
433 ptllnd facility to gather history which should really be
434 covered by D_NETTRACE too, if only CDEBUG recorded history in
440 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
441 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
442 callback can occur before a connection has actually been
443 established. This caused an assertion failure previously.
445 Severity : enhancement
447 Description: Multiple instances for o2iblnd
448 Details : Allow multiple instances of o2iblnd to enable networking over
449 multiple HCAs and routing between them.
453 Description: lnet deadlock in router_checker
454 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
455 into BH locks to eliminate potential deadlock caused by
456 ksocknal_data_ready() preempting code holding these locks.
460 Description: Millions of failed socklnd connection attempts cause a very slow FS
461 Details : added a new route flag ksnr_scheduled to distinguish from
462 ksnr_connecting, so that a peer connection request is only turned
463 down for race concerns when an active connection to the same peer
464 is under progress (instead of just being scheduled).
466 ------------------------------------------------------------------------------
468 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
470 * Support for networks:
471 socklnd - kernels up to 2.6.16
472 qswlnd - Qsnet kernel modules 5.20 and later
473 openiblnd - IbGold 1.8.2
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
483 Severity : major on XT3
485 Description: libcfs overwrites /proc/sys/portals
486 Details : libcfs created a symlink from /proc/sys/portals to
487 /proc/sys/lnet for backwards compatibility. This is no
488 longer required and makes the Cray portals /proc variables
493 Description: OFED FMR API change
494 Details : This changes parameter usage to reflect a change in
495 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
496 that FMR support is only used in experimental versions of the
497 o2iblnd - this change does not affect standard usage at all.
499 Severity : enhancement
501 Description: new ko2iblnd module parameter: ib_mtu
502 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
503 HCAs. You can avoid this problem by setting the MTU to 1024
504 using this module parameter.
506 Severity : enhancement
507 Bugzilla : 11118/11620
508 Description: ptllnd small request message buffer alignment fix
509 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
510 Round up small message size on sends in case this option
511 is not supported. 11620 was a defect in the initial
512 implementation which effectively asserted all peers had to be
513 running the correct protocol version which was fixed by always
514 NAK-ing such requests and handling any misalignments they
519 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
520 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
521 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
523 Severity : enhancement
525 Description: Patchless ZC(zero copy) socklnd
526 Details : New protocol for socklnd, socklnd can support zero copy without
527 kernel patch, it's compatible with old socklnd. Checksum is
528 moved from tunables to modparams.
532 Description: When ksocknal_del_peer() is called upon a peer whose
533 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
534 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
537 Frequency : when ptlrpc is under heavy use and runs out of request buffer
539 Description: In lnet_match_blocked_msg(), md can be used without holding a
543 Frequency : very rarely
545 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
546 If connd connects a route which has been closed by
547 ksocknal_shutdown(), ksocknal_create_routes() may create new
548 routes which hold references on the peer, causing shutdown
549 process to wait for peer to disappear forever.
551 Severity : enhancement
553 Description: Dump XT3 portals traces on kptllnd timeout
554 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
555 dump Cray portals debug traces to a file. The kptllnd module
556 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
557 is the basename of the dump file.
560 Frequency : infrequent
562 Description: kernel ptllnd fix bug in connection re-establishment
563 Details : Kernel ptllnd could produce protocol errors e.g. illegal
564 matchbits and/or violate the credit flow protocol when trying
565 to re-establish a connection with a peer after an error or
568 Severity : enhancement
570 Description: Allow /proc/sys/lnet/debug to be set symbolically
571 Details : Allow debug and subsystem debug values to be read/set by name
572 in addition to numerically, for ease of use.
575 Frequency : only in configurations with LNET routers
577 Description: routes automatically marked down and recovered
578 Details : In configurations with LNET routers if a router fails routers
579 now actively try to recover routes that are down, unless they
580 are marked down by an administrator.
582 ------------------------------------------------------------------------------
584 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
587 Frequency : very rarely, in configurations with LNET routers and TCP
589 Description: incorrect data written to files on OSTs
590 Details : In certain high-load conditions incorrect data may be written
591 to files on the OST when using TCP networks.
593 ------------------------------------------------------------------------------
595 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
597 - rework CDEBUG messages rate-limiting mechanism b=10375
598 - add per-socket tunables for socklnd if the kernel is patched b=10327
600 ------------------------------------------------------------------------------
602 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
604 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
605 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
606 - quiet LNET startup LNI message for liblustre b=10128
607 - Better console error messages if 'ip2nets' can't match an IP address
608 - Fixed overflow/use-before-set bugs in linux-time.h
609 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
610 - LNET teardown failed an assertion about the route table being empty
611 - Fixed a crash in LNetEQPoll(<invalid handle>)
612 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
613 - improve debug message for liblustre/Catamount nodes (b=10116)
615 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
616 * Configuration change for the XT3
617 The PTLLND is now used to run Lustre over Portals on the XT3.
618 The configure option(s) --with-cray-portals are no longer
619 used. Rather --with-portals=<path-to-portals-includes> is
620 used to enable building on the XT3. In addition to enable
621 XT3 specific features the option --enable-cray-xt3 must be
624 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
625 * Portals has been removed, replaced by LNET.
626 LNET is new networking infrastructure for Lustre, it includes a
627 reorganized network configuration mode (see the user
628 documentation for full details) as well as support for routing
629 between different network fabrics. Lustre Networking Devices
630 (LNDS) for the supported network fabrics have also been created
631 for this new infrastructure.
633 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
638 Frequency : rare (large Voltaire clusters only)
640 Description: the default number of reserved transmit descriptors was too low
641 for some large clusters
642 Details : As a workaround, the number was increased. A proper fix includes
645 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
650 Frequency : occasional (large-scale events, cluster reboot, network failure)
652 Description: too many error messages on console obscure actual problem and
653 can slow down/panic server, or cause recovery to fail repeatedly
654 Details : enable rate-limiting of console error messages, and some messages
655 that were console errors now only go to the kernel log
657 Severity : enhancement
659 Description: add /proc/sys/portals/catastrophe entry which will report if
660 that node has previously LBUGged
662 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
664 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
666 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
668 - handle error return code in kranal_check_fma_rx() (5915,6054)
670 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
672 - update vibnal (Voltaire IB NAL)
673 - update gmnal (Myrinet NAL), gmnalid
675 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
677 * Landed portals:b_port_step as follows...
679 - removed CFS_DECL_SPIN*
680 just use 'spinlock_t' and initialise with spin_lock_init()
682 - removed CFS_DECL_MUTEX*
683 just use 'struct semaphore' and initialise with init_mutex()
685 - removed CFS_DECL_RWSEM*
686 just use 'struct rw_semaphore' and initialise with init_rwsem()
688 - renamed cfs_sleep_chan -> cfs_waitq
689 cfs_sleep_link -> cfs_waitlink
691 - fixed race in linux version of arch-independent socknal
692 (the ENOMEM/EAGAIN decision).
694 - Didn't fix problems in Darwin version of arch-independent socknal
695 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
697 - removed libcfs types from non-socknal header files (only some types
698 in the header files had been changed; the .c files hadn't been