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 and 1.2.0, 1.2.5
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
16 Description: ksocklnd fails to establish connection if accept_port is high
17 Details : PID remapping must not be done for active (outgoing) connections
19 --------------------------------------------------------------------------------
21 2008-01-11 Sun Microsystems, Inc.
23 * Support for networks:
24 socklnd - any kernel supported by Lustre,
25 qswlnd - Qsnet kernel modules 5.20 and later,
26 openiblnd - IbGold 1.8.2,
27 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5
28 viblnd - Voltaire ibhost 3.4.5 and later,
29 ciblnd - Topspin 3.2.0,
30 iiblnd - Infiniserv 3.3 + PathBits patch,
31 gmlnd - GM 2.1.22 and later,
32 mxlnd - MX 1.2.1 or later,
33 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
36 Description: liblustre network error
37 Details : liblustre clients should understand LNET_ACCEPT_PORT environment
38 variable even if they don't start lnet acceptor.
42 Description: Strange message from lnet (Ignoring prediction from the future)
43 Details : Incorrect calculation of peer's last_alive value in ksocklnd
45 --------------------------------------------------------------------------------
47 2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
49 * Support for networks:
50 socklnd - any kernel supported by Lustre,
51 qswlnd - Qsnet kernel modules 5.20 and later,
52 openiblnd - IbGold 1.8.2,
53 o2iblnd - OFED 1.1 and 1.2.0, 1.2.5.
54 viblnd - Voltaire ibhost 3.4.5 and later,
55 ciblnd - Topspin 3.2.0,
56 iiblnd - Infiniserv 3.3 + PathBits patch,
57 gmlnd - GM 2.1.22 and later,
58 mxlnd - MX 1.2.1 or later,
59 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
63 Description: ASSERTION(me == md->md_me) failed in lnet_match_md()
67 Description: increase send queue size for ciblnd/openiblnd
71 Description: new userspace socklnd
72 Details : Old userspace tcpnal that resided in lnet/ulnds/socklnd replaced
73 with new one - usocklnd.
75 Severity : enhancement
77 Description: Console message flood
78 Details : Make cdls ratelimiting more tunable by adding several tunable in
79 procfs /proc/sys/lnet/console_{min,max}_delay_centisecs and
80 /proc/sys/lnet/console_backoff.
82 --------------------------------------------------------------------------------
84 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
86 * Support for networks:
87 socklnd - any kernel supported by Lustre,
88 qswlnd - Qsnet kernel modules 5.20 and later,
89 openiblnd - IbGold 1.8.2,
90 o2iblnd - OFED 1.1 and 1.2,
91 viblnd - Voltaire ibhost 3.4.5 and later,
92 ciblnd - Topspin 3.2.0,
93 iiblnd - Infiniserv 3.3 + PathBits patch,
94 gmlnd - GM 2.1.22 and later,
95 mxlnd - MX 1.2.1 or later,
96 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
100 Description: /proc/sys/lnet has non-sysctl entries
101 Details : Updating dump_kernel/daemon_file/debug_mb to use sysctl variables
105 Description: TOE Kernel panic by ksocklnd
106 Details : offloaded sockets provide their own implementation of sendpage,
107 can't call tcp_sendpage() directly
111 Description: kibnal_shutdown() doesn't finish; lconf --cleanup hangs
112 Details : races between lnd_shutdown and peer creation prevent
113 lnd_shutdown from finishing.
117 Description: open files rlimit 1024 reached while liblustre testing
118 Details : ulnds/socklnd must close open socket after unsuccessful
123 Description: build error
124 Details : fix typos in gmlnd, ptllnd and viblnd
126 ------------------------------------------------------------------------------
128 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
130 * Support for networks:
131 socklnd - kernels up to 2.6.16,
132 qswlnd - Qsnet kernel modules 5.20 and later,
133 openiblnd - IbGold 1.8.2,
134 o2iblnd - OFED 1.1 and 1.2
135 viblnd - Voltaire ibhost 3.4.5 and later,
136 ciblnd - Topspin 3.2.0,
137 iiblnd - Infiniserv 3.3 + PathBits patch,
138 gmlnd - GM 2.1.22 and later,
139 mxlnd - MX 1.2.1 or later,
140 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
142 2007-06-21 Cluster File Systems, Inc. <info@clusterfs.com>
144 * Support for networks:
145 socklnd - kernels up to 2.6.16,
146 qswlnd - Qsnet kernel modules 5.20 and later,
147 openiblnd - IbGold 1.8.2,
149 viblnd - Voltaire ibhost 3.4.5 and later,
150 ciblnd - Topspin 3.2.0,
151 iiblnd - Infiniserv 3.3 + PathBits patch,
152 gmlnd - GM 2.1.22 and later,
153 mxlnd - MX 1.2.1 or later,
154 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
158 Description: Initialize cpumask before use
162 Description: ASSERTION failures when upgrading to the patchless zero-copy
164 Details : This bug affects "rolling upgrades", causing an inconsistent
165 protocol version negotiation and subsequent assertion failure
166 during rolling upgrades after the first wave of upgrades.
170 Details : Change "dropped message" CERRORs to D_NETERROR so they are
171 logged instead of creating "console chatter" when a lustre
172 timeout races with normal RPC completion.
175 Details : lnet_clear_peer_table can wait forever if user forgets to
179 Details : libcfs_id2str should check pid against LNET_PID_ANY.
183 Description: added LNET self test
184 Details : landing b_self_test
189 Description: cfs_duration_{u,n}sec() wrongly calculate nanosecond part of
191 Details : do_div() macro is used incorrectly.
193 2007-04-23 Cluster File Systems, Inc. <info@clusterfs.com>
197 Description: make panic on lbug configurable
201 Description: Add OFED1.2 support to o2iblnd
202 Details : o2iblnd depends on OFED's modules, if out-tree OFED's modules
203 are installed (other than kernel's in-tree infiniband), there
204 could be some problem while insmod o2iblnd (mismatch CRC of
206 If extra Module.symvers is supported in kernel (i.e, 2.6.17),
207 this link provides solution:
208 https://bugs.openfabrics.org/show_bug.cgi?id=355
209 if extra Module.symvers is not supported in kernel, we will
210 have to run the script in bug 12316 to update
211 $LINUX/module.symvers before building o2iblnd.
212 More details about this are in bug 12316.
214 ------------------------------------------------------------------------------
216 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
217 * version 1.4.10 / 1.6.0
218 * Support for networks:
219 socklnd - kernels up to 2.6.16,
220 qswlnd - Qsnet kernel modules 5.20 and later,
221 openiblnd - IbGold 1.8.2,
223 viblnd - Voltaire ibhost 3.4.5 and later,
224 ciblnd - Topspin 3.2.0,
225 iiblnd - Infiniserv 3.3 + PathBits patch,
226 gmlnd - GM 2.1.22 and later,
227 mxlnd - MX 1.2.1 or later,
228 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
232 Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
233 possibly accessed in kptllnd_shutdown. Ptllnd should init
234 kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
238 Description: gmlnd ignored some transmit errors when finalizing lnet messages.
242 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
246 Description: the_lnet.ln_finalizing was not set when the current thread is
247 about to complete messages. It only affects multi-threaded
253 Description: Changed the default kqswlnd ntxmsg=512
258 Description: Assertion failure in kernel ptllnd caused by posting passive
259 bulk buffers before connection establishment complete.
264 Description: A race in kernel ptllnd between deleting a peer and posting
265 new communications for it could hang communications -
266 manifesting as "Unexpectedly long timeout" messages.
271 Description: Kernel ptllnd lock ordering issue could hang a node.
276 Description: node crash on socket teardown race
279 Frequency : 'lctl peer_list' issued on a mx net
281 Description: Enable lctl's peer_list for MXLND
284 Frequency : after Ptllnd timeouts and portals congestion
286 Description: Credit overflows
287 Details : This was a bug in ptllnd connection establishment. The fix
288 implements better peer stamps to disambiguate connection
289 establishment and ensure both peers enter the credit flow
290 state machine consistently.
295 Description: kptllnd didn't propagate some network errors up to LNET
296 Details : This bug was spotted while investigating 11394. The fix
297 ensures network errors on sends and bulk transfers are
298 propagated to LNET/lustre correctly.
300 Severity : enhancement
302 Description: Fixed console chatter in case of -ETIMEDOUT.
304 Severity : enhancement
306 Description: Added D_NETTRACE for recording network packet history
307 (initially only for ptllnd). Also a separate userspace
308 ptllnd facility to gather history which should really be
309 covered by D_NETTRACE too, if only CDEBUG recorded history in
315 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
316 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
317 callback can occur before a connection has actually been
318 established. This caused an assertion failure previously.
320 Severity : enhancement
322 Description: Multiple instances for o2iblnd
323 Details : Allow multiple instances of o2iblnd to enable networking over
324 multiple HCAs and routing between them.
328 Description: lnet deadlock in router_checker
329 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
330 into BH locks to eliminate potential deadlock caused by
331 ksocknal_data_ready() preempting code holding these locks.
335 Description: Millions of failed socklnd connection attempts cause a very slow FS
336 Details : added a new route flag ksnr_scheduled to distinguish from
337 ksnr_connecting, so that a peer connection request is only turned
338 down for race concerns when an active connection to the same peer
339 is under progress (instead of just being scheduled).
341 ------------------------------------------------------------------------------
343 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
345 * Support for networks:
346 socklnd - kernels up to 2.6.16
347 qswlnd - Qsnet kernel modules 5.20 and later
348 openiblnd - IbGold 1.8.2
350 viblnd - Voltaire ibhost 3.4.5 and later
351 ciblnd - Topspin 3.2.0
352 iiblnd - Infiniserv 3.3 + PathBits patch
353 gmlnd - GM 2.1.22 and later
354 mxlnd - MX 1.2.1 or later
355 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
358 Severity : major on XT3
360 Description: libcfs overwrites /proc/sys/portals
361 Details : libcfs created a symlink from /proc/sys/portals to
362 /proc/sys/lnet for backwards compatibility. This is no
363 longer required and makes the Cray portals /proc variables
368 Description: OFED FMR API change
369 Details : This changes parameter usage to reflect a change in
370 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
371 that FMR support is only used in experimental versions of the
372 o2iblnd - this change does not affect standard usage at all.
374 Severity : enhancement
376 Description: new ko2iblnd module parameter: ib_mtu
377 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
378 HCAs. You can avoid this problem by setting the MTU to 1024
379 using this module parameter.
381 Severity : enhancement
382 Bugzilla : 11118/11620
383 Description: ptllnd small request message buffer alignment fix
384 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
385 Round up small message size on sends in case this option
386 is not supported. 11620 was a defect in the initial
387 implementation which effectively asserted all peers had to be
388 running the correct protocol version which was fixed by always
389 NAK-ing such requests and handling any misalignments they
394 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
395 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
396 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
398 Severity : enhancement
400 Description: Patchless ZC(zero copy) socklnd
401 Details : New protocol for socklnd, socklnd can support zero copy without
402 kernel patch, it's compatible with old socklnd. Checksum is
403 moved from tunables to modparams.
407 Description: When ksocknal_del_peer() is called upon a peer whose
408 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
409 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
412 Frequency : when ptlrpc is under heavy use and runs out of request buffer
414 Description: In lnet_match_blocked_msg(), md can be used without holding a
418 Frequency : very rarely
420 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
421 If connd connects a route which has been closed by
422 ksocknal_shutdown(), ksocknal_create_routes() may create new
423 routes which hold references on the peer, causing shutdown
424 process to wait for peer to disappear forever.
426 Severity : enhancement
428 Description: Dump XT3 portals traces on kptllnd timeout
429 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
430 dump Cray portals debug traces to a file. The kptllnd module
431 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
432 is the basename of the dump file.
435 Frequency : infrequent
437 Description: kernel ptllnd fix bug in connection re-establishment
438 Details : Kernel ptllnd could produce protocol errors e.g. illegal
439 matchbits and/or violate the credit flow protocol when trying
440 to re-establish a connection with a peer after an error or
443 Severity : enhancement
445 Description: Allow /proc/sys/lnet/debug to be set symbolically
446 Details : Allow debug and subsystem debug values to be read/set by name
447 in addition to numerically, for ease of use.
450 Frequency : only in configurations with LNET routers
452 Description: routes automatically marked down and recovered
453 Details : In configurations with LNET routers if a router fails routers
454 now actively try to recover routes that are down, unless they
455 are marked down by an administrator.
457 ------------------------------------------------------------------------------
459 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
462 Frequency : very rarely, in configurations with LNET routers and TCP
464 Description: incorrect data written to files on OSTs
465 Details : In certain high-load conditions incorrect data may be written
466 to files on the OST when using TCP networks.
468 ------------------------------------------------------------------------------
470 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
472 - rework CDEBUG messages rate-limiting mechanism b=10375
473 - add per-socket tunables for socklnd if the kernel is patched b=10327
475 ------------------------------------------------------------------------------
477 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
479 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
480 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
481 - quiet LNET startup LNI message for liblustre b=10128
482 - Better console error messages if 'ip2nets' can't match an IP address
483 - Fixed overflow/use-before-set bugs in linux-time.h
484 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
485 - LNET teardown failed an assertion about the route table being empty
486 - Fixed a crash in LNetEQPoll(<invalid handle>)
487 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
488 - improve debug message for liblustre/Catamount nodes (b=10116)
490 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
491 * Configuration change for the XT3
492 The PTLLND is now used to run Lustre over Portals on the XT3.
493 The configure option(s) --with-cray-portals are no longer
494 used. Rather --with-portals=<path-to-portals-includes> is
495 used to enable building on the XT3. In addition to enable
496 XT3 specific features the option --enable-cray-xt3 must be
499 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
500 * Portals has been removed, replaced by LNET.
501 LNET is new networking infrastructure for Lustre, it includes a
502 reorganized network configuration mode (see the user
503 documentation for full details) as well as support for routing
504 between different network fabrics. Lustre Networking Devices
505 (LNDS) for the supported network fabrics have also been created
506 for this new infrastructure.
508 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
513 Frequency : rare (large Voltaire clusters only)
515 Description: the default number of reserved transmit descriptors was too low
516 for some large clusters
517 Details : As a workaround, the number was increased. A proper fix includes
520 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
525 Frequency : occasional (large-scale events, cluster reboot, network failure)
527 Description: too many error messages on console obscure actual problem and
528 can slow down/panic server, or cause recovery to fail repeatedly
529 Details : enable rate-limiting of console error messages, and some messages
530 that were console errors now only go to the kernel log
532 Severity : enhancement
534 Description: add /proc/sys/portals/catastrophe entry which will report if
535 that node has previously LBUGged
537 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
539 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
541 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
543 - handle error return code in kranal_check_fma_rx() (5915,6054)
545 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
547 - update vibnal (Voltaire IB NAL)
548 - update gmnal (Myrinet NAL), gmnalid
550 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
552 * Landed portals:b_port_step as follows...
554 - removed CFS_DECL_SPIN*
555 just use 'spinlock_t' and initialise with spin_lock_init()
557 - removed CFS_DECL_MUTEX*
558 just use 'struct semaphore' and initialise with init_mutex()
560 - removed CFS_DECL_RWSEM*
561 just use 'struct rw_semaphore' and initialise with init_rwsem()
563 - renamed cfs_sleep_chan -> cfs_waitq
564 cfs_sleep_link -> cfs_waitlink
566 - fixed race in linux version of arch-independent socknal
567 (the ENOMEM/EAGAIN decision).
569 - Didn't fix problems in Darwin version of arch-independent socknal
570 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
572 - removed libcfs types from non-socknal header files (only some types
573 in the header files had been changed; the .c files hadn't been