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: LNET Selftest results in Soft lockup on OSS CPU
23 Details : only hits when 8 or more o2ib clients involved and a session is
24 torn down with 'lst end_session' without preceeding 'lst stop'.
28 Description: concurrent_sends in IB LNDs should not be changeable at run time
29 Details : concurrent_sends in IB LNDs should not be changeable at run time
33 Description: ptl_send_rpc hits LASSERT when ptl_send_buf fails
34 Details : only hits under out-of-memory situations
37 -------------------------------------------------------------------------------
40 04-26-2008 Sun Microsystems, Inc.
42 * Support for networks:
43 socklnd - any kernel supported by Lustre,
44 qswlnd - Qsnet kernel modules 5.20 and later,
45 openiblnd - IbGold 1.8.2,
46 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5
47 viblnd - Voltaire ibhost 3.4.5 and later,
48 ciblnd - Topspin 3.2.0,
49 iiblnd - Infiniserv 3.3 + PathBits patch,
50 gmlnd - GM 2.1.22 and later,
51 mxlnd - MX 1.2.1 or later,
52 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
56 Description: excessive debug information removed
57 Details : excessive debug information removed
61 Description: ksocknal_create_conn() hit ASSERTION during connection race
62 Details : ksocknal_create_conn() hit ASSERTION during connection race
66 Description: ksocknal_send_hello() hit ASSERTION while connecting race
67 Details : ksocknal_send_hello() hit ASSERTION while connecting race
71 Description: o2iblnd/ptllnd credit deadlock in a routed config.
72 Details : o2iblnd/ptllnd credit deadlock in a routed config.
76 Description: High load after starting lnet
77 Details : gmlnd should sleep in rx thread in interruptible way. Otherwise,
78 uptime utility reports high load that looks confusingly.
82 Description: ksocklnd fails to establish connection if accept_port is high
83 Details : PID remapping must not be done for active (outgoing) connections
85 --------------------------------------------------------------------------------
87 2008-01-11 Sun Microsystems, Inc.
89 * Support for networks:
90 socklnd - any kernel supported by Lustre,
91 qswlnd - Qsnet kernel modules 5.20 and later,
92 openiblnd - IbGold 1.8.2,
93 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5
94 viblnd - Voltaire ibhost 3.4.5 and later,
95 ciblnd - Topspin 3.2.0,
96 iiblnd - Infiniserv 3.3 + PathBits patch,
97 gmlnd - GM 2.1.22 and later,
98 mxlnd - MX 1.2.1 or later,
99 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
102 Description: liblustre network error
103 Details : liblustre clients should understand LNET_ACCEPT_PORT environment
104 variable even if they don't start lnet acceptor.
108 Description: Strange message from lnet (Ignoring prediction from the future)
109 Details : Incorrect calculation of peer's last_alive value in ksocklnd
111 --------------------------------------------------------------------------------
113 2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
115 * Support for networks:
116 socklnd - any kernel supported by Lustre,
117 qswlnd - Qsnet kernel modules 5.20 and later,
118 openiblnd - IbGold 1.8.2,
119 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5.
120 viblnd - Voltaire ibhost 3.4.5 and later,
121 ciblnd - Topspin 3.2.0,
122 iiblnd - Infiniserv 3.3 + PathBits patch,
123 gmlnd - GM 2.1.22 and later,
124 mxlnd - MX 1.2.1 or later,
125 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
129 Description: ASSERTION(me == md->md_me) failed in lnet_match_md()
133 Description: increase send queue size for ciblnd/openiblnd
137 Description: new userspace socklnd
138 Details : Old userspace tcpnal that resided in lnet/ulnds/socklnd replaced
139 with new one - usocklnd.
141 Severity : enhancement
143 Description: Console message flood
144 Details : Make cdls ratelimiting more tunable by adding several tunable in
145 procfs /proc/sys/lnet/console_{min,max}_delay_centisecs and
146 /proc/sys/lnet/console_backoff.
148 --------------------------------------------------------------------------------
150 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
152 * Support for networks:
153 socklnd - any kernel supported by Lustre,
154 qswlnd - Qsnet kernel modules 5.20 and later,
155 openiblnd - IbGold 1.8.2,
156 o2iblnd - OFED 1.1 and 1.2,
157 viblnd - Voltaire ibhost 3.4.5 and later,
158 ciblnd - Topspin 3.2.0,
159 iiblnd - Infiniserv 3.3 + PathBits patch,
160 gmlnd - GM 2.1.22 and later,
161 mxlnd - MX 1.2.1 or later,
162 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
166 Description: /proc/sys/lnet has non-sysctl entries
167 Details : Updating dump_kernel/daemon_file/debug_mb to use sysctl variables
171 Description: TOE Kernel panic by ksocklnd
172 Details : offloaded sockets provide their own implementation of sendpage,
173 can't call tcp_sendpage() directly
177 Description: kibnal_shutdown() doesn't finish; lconf --cleanup hangs
178 Details : races between lnd_shutdown and peer creation prevent
179 lnd_shutdown from finishing.
183 Description: open files rlimit 1024 reached while liblustre testing
184 Details : ulnds/socklnd must close open socket after unsuccessful
189 Description: build error
190 Details : fix typos in gmlnd, ptllnd and viblnd
192 ------------------------------------------------------------------------------
194 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
196 * Support for networks:
197 socklnd - kernels up to 2.6.16,
198 qswlnd - Qsnet kernel modules 5.20 and later,
199 openiblnd - IbGold 1.8.2,
200 o2iblnd - OFED 1.1 and 1.2
201 viblnd - Voltaire ibhost 3.4.5 and later,
202 ciblnd - Topspin 3.2.0,
203 iiblnd - Infiniserv 3.3 + PathBits patch,
204 gmlnd - GM 2.1.22 and later,
205 mxlnd - MX 1.2.1 or later,
206 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
208 2007-06-21 Cluster File Systems, Inc. <info@clusterfs.com>
210 * Support for networks:
211 socklnd - kernels up to 2.6.16,
212 qswlnd - Qsnet kernel modules 5.20 and later,
213 openiblnd - IbGold 1.8.2,
215 viblnd - Voltaire ibhost 3.4.5 and later,
216 ciblnd - Topspin 3.2.0,
217 iiblnd - Infiniserv 3.3 + PathBits patch,
218 gmlnd - GM 2.1.22 and later,
219 mxlnd - MX 1.2.1 or later,
220 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
224 Description: Initialize cpumask before use
228 Description: ASSERTION failures when upgrading to the patchless zero-copy
230 Details : This bug affects "rolling upgrades", causing an inconsistent
231 protocol version negotiation and subsequent assertion failure
232 during rolling upgrades after the first wave of upgrades.
236 Details : Change "dropped message" CERRORs to D_NETERROR so they are
237 logged instead of creating "console chatter" when a lustre
238 timeout races with normal RPC completion.
241 Details : lnet_clear_peer_table can wait forever if user forgets to
245 Details : libcfs_id2str should check pid against LNET_PID_ANY.
249 Description: added LNET self test
250 Details : landing b_self_test
255 Description: cfs_duration_{u,n}sec() wrongly calculate nanosecond part of
257 Details : do_div() macro is used incorrectly.
259 2007-04-23 Cluster File Systems, Inc. <info@clusterfs.com>
263 Description: make panic on lbug configurable
267 Description: Add OFED1.2 support to o2iblnd
268 Details : o2iblnd depends on OFED's modules, if out-tree OFED's modules
269 are installed (other than kernel's in-tree infiniband), there
270 could be some problem while insmod o2iblnd (mismatch CRC of
272 If extra Module.symvers is supported in kernel (i.e, 2.6.17),
273 this link provides solution:
274 https://bugs.openfabrics.org/show_bug.cgi?id=355
275 if extra Module.symvers is not supported in kernel, we will
276 have to run the script in bug 12316 to update
277 $LINUX/module.symvers before building o2iblnd.
278 More details about this are in bug 12316.
280 ------------------------------------------------------------------------------
282 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
283 * version 1.4.10 / 1.6.0
284 * Support for networks:
285 socklnd - kernels up to 2.6.16,
286 qswlnd - Qsnet kernel modules 5.20 and later,
287 openiblnd - IbGold 1.8.2,
289 viblnd - Voltaire ibhost 3.4.5 and later,
290 ciblnd - Topspin 3.2.0,
291 iiblnd - Infiniserv 3.3 + PathBits patch,
292 gmlnd - GM 2.1.22 and later,
293 mxlnd - MX 1.2.1 or later,
294 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
298 Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
299 possibly accessed in kptllnd_shutdown. Ptllnd should init
300 kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
304 Description: gmlnd ignored some transmit errors when finalizing lnet messages.
308 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
312 Description: the_lnet.ln_finalizing was not set when the current thread is
313 about to complete messages. It only affects multi-threaded
319 Description: Changed the default kqswlnd ntxmsg=512
324 Description: Assertion failure in kernel ptllnd caused by posting passive
325 bulk buffers before connection establishment complete.
330 Description: A race in kernel ptllnd between deleting a peer and posting
331 new communications for it could hang communications -
332 manifesting as "Unexpectedly long timeout" messages.
337 Description: Kernel ptllnd lock ordering issue could hang a node.
342 Description: node crash on socket teardown race
345 Frequency : 'lctl peer_list' issued on a mx net
347 Description: Enable lctl's peer_list for MXLND
350 Frequency : after Ptllnd timeouts and portals congestion
352 Description: Credit overflows
353 Details : This was a bug in ptllnd connection establishment. The fix
354 implements better peer stamps to disambiguate connection
355 establishment and ensure both peers enter the credit flow
356 state machine consistently.
361 Description: kptllnd didn't propagate some network errors up to LNET
362 Details : This bug was spotted while investigating 11394. The fix
363 ensures network errors on sends and bulk transfers are
364 propagated to LNET/lustre correctly.
366 Severity : enhancement
368 Description: Fixed console chatter in case of -ETIMEDOUT.
370 Severity : enhancement
372 Description: Added D_NETTRACE for recording network packet history
373 (initially only for ptllnd). Also a separate userspace
374 ptllnd facility to gather history which should really be
375 covered by D_NETTRACE too, if only CDEBUG recorded history in
381 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
382 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
383 callback can occur before a connection has actually been
384 established. This caused an assertion failure previously.
386 Severity : enhancement
388 Description: Multiple instances for o2iblnd
389 Details : Allow multiple instances of o2iblnd to enable networking over
390 multiple HCAs and routing between them.
394 Description: lnet deadlock in router_checker
395 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
396 into BH locks to eliminate potential deadlock caused by
397 ksocknal_data_ready() preempting code holding these locks.
401 Description: Millions of failed socklnd connection attempts cause a very slow FS
402 Details : added a new route flag ksnr_scheduled to distinguish from
403 ksnr_connecting, so that a peer connection request is only turned
404 down for race concerns when an active connection to the same peer
405 is under progress (instead of just being scheduled).
407 ------------------------------------------------------------------------------
409 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
411 * Support for networks:
412 socklnd - kernels up to 2.6.16
413 qswlnd - Qsnet kernel modules 5.20 and later
414 openiblnd - IbGold 1.8.2
416 viblnd - Voltaire ibhost 3.4.5 and later
417 ciblnd - Topspin 3.2.0
418 iiblnd - Infiniserv 3.3 + PathBits patch
419 gmlnd - GM 2.1.22 and later
420 mxlnd - MX 1.2.1 or later
421 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
424 Severity : major on XT3
426 Description: libcfs overwrites /proc/sys/portals
427 Details : libcfs created a symlink from /proc/sys/portals to
428 /proc/sys/lnet for backwards compatibility. This is no
429 longer required and makes the Cray portals /proc variables
434 Description: OFED FMR API change
435 Details : This changes parameter usage to reflect a change in
436 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
437 that FMR support is only used in experimental versions of the
438 o2iblnd - this change does not affect standard usage at all.
440 Severity : enhancement
442 Description: new ko2iblnd module parameter: ib_mtu
443 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
444 HCAs. You can avoid this problem by setting the MTU to 1024
445 using this module parameter.
447 Severity : enhancement
448 Bugzilla : 11118/11620
449 Description: ptllnd small request message buffer alignment fix
450 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
451 Round up small message size on sends in case this option
452 is not supported. 11620 was a defect in the initial
453 implementation which effectively asserted all peers had to be
454 running the correct protocol version which was fixed by always
455 NAK-ing such requests and handling any misalignments they
460 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
461 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
462 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
464 Severity : enhancement
466 Description: Patchless ZC(zero copy) socklnd
467 Details : New protocol for socklnd, socklnd can support zero copy without
468 kernel patch, it's compatible with old socklnd. Checksum is
469 moved from tunables to modparams.
473 Description: When ksocknal_del_peer() is called upon a peer whose
474 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
475 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
478 Frequency : when ptlrpc is under heavy use and runs out of request buffer
480 Description: In lnet_match_blocked_msg(), md can be used without holding a
484 Frequency : very rarely
486 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
487 If connd connects a route which has been closed by
488 ksocknal_shutdown(), ksocknal_create_routes() may create new
489 routes which hold references on the peer, causing shutdown
490 process to wait for peer to disappear forever.
492 Severity : enhancement
494 Description: Dump XT3 portals traces on kptllnd timeout
495 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
496 dump Cray portals debug traces to a file. The kptllnd module
497 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
498 is the basename of the dump file.
501 Frequency : infrequent
503 Description: kernel ptllnd fix bug in connection re-establishment
504 Details : Kernel ptllnd could produce protocol errors e.g. illegal
505 matchbits and/or violate the credit flow protocol when trying
506 to re-establish a connection with a peer after an error or
509 Severity : enhancement
511 Description: Allow /proc/sys/lnet/debug to be set symbolically
512 Details : Allow debug and subsystem debug values to be read/set by name
513 in addition to numerically, for ease of use.
516 Frequency : only in configurations with LNET routers
518 Description: routes automatically marked down and recovered
519 Details : In configurations with LNET routers if a router fails routers
520 now actively try to recover routes that are down, unless they
521 are marked down by an administrator.
523 ------------------------------------------------------------------------------
525 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
528 Frequency : very rarely, in configurations with LNET routers and TCP
530 Description: incorrect data written to files on OSTs
531 Details : In certain high-load conditions incorrect data may be written
532 to files on the OST when using TCP networks.
534 ------------------------------------------------------------------------------
536 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
538 - rework CDEBUG messages rate-limiting mechanism b=10375
539 - add per-socket tunables for socklnd if the kernel is patched b=10327
541 ------------------------------------------------------------------------------
543 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
545 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
546 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
547 - quiet LNET startup LNI message for liblustre b=10128
548 - Better console error messages if 'ip2nets' can't match an IP address
549 - Fixed overflow/use-before-set bugs in linux-time.h
550 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
551 - LNET teardown failed an assertion about the route table being empty
552 - Fixed a crash in LNetEQPoll(<invalid handle>)
553 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
554 - improve debug message for liblustre/Catamount nodes (b=10116)
556 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
557 * Configuration change for the XT3
558 The PTLLND is now used to run Lustre over Portals on the XT3.
559 The configure option(s) --with-cray-portals are no longer
560 used. Rather --with-portals=<path-to-portals-includes> is
561 used to enable building on the XT3. In addition to enable
562 XT3 specific features the option --enable-cray-xt3 must be
565 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
566 * Portals has been removed, replaced by LNET.
567 LNET is new networking infrastructure for Lustre, it includes a
568 reorganized network configuration mode (see the user
569 documentation for full details) as well as support for routing
570 between different network fabrics. Lustre Networking Devices
571 (LNDS) for the supported network fabrics have also been created
572 for this new infrastructure.
574 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
579 Frequency : rare (large Voltaire clusters only)
581 Description: the default number of reserved transmit descriptors was too low
582 for some large clusters
583 Details : As a workaround, the number was increased. A proper fix includes
586 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
591 Frequency : occasional (large-scale events, cluster reboot, network failure)
593 Description: too many error messages on console obscure actual problem and
594 can slow down/panic server, or cause recovery to fail repeatedly
595 Details : enable rate-limiting of console error messages, and some messages
596 that were console errors now only go to the kernel log
598 Severity : enhancement
600 Description: add /proc/sys/portals/catastrophe entry which will report if
601 that node has previously LBUGged
603 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
605 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
607 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
609 - handle error return code in kranal_check_fma_rx() (5915,6054)
611 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
613 - update vibnal (Voltaire IB NAL)
614 - update gmnal (Myrinet NAL), gmnalid
616 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
618 * Landed portals:b_port_step as follows...
620 - removed CFS_DECL_SPIN*
621 just use 'spinlock_t' and initialise with spin_lock_init()
623 - removed CFS_DECL_MUTEX*
624 just use 'struct semaphore' and initialise with init_mutex()
626 - removed CFS_DECL_RWSEM*
627 just use 'struct rw_semaphore' and initialise with init_rwsem()
629 - renamed cfs_sleep_chan -> cfs_waitq
630 cfs_sleep_link -> cfs_waitlink
632 - fixed race in linux version of arch-independent socknal
633 (the ENOMEM/EAGAIN decision).
635 - Didn't fix problems in Darwin version of arch-independent socknal
636 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
638 - removed libcfs types from non-socknal header files (only some types
639 in the header files had been changed; the .c files hadn't been