1 tbd Cluster File Systems, Inc. <info@clusterfs.com>
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,
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
17 Description: new userspace socklnd
18 Details : Old userspace tcpnal that resided in lnet/ulnds/socklnd replaced
19 with new one - usocklnd.
21 --------------------------------------------------------------------------------
23 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
25 * Support for networks:
26 socklnd - any kernel supported by Lustre,
27 qswlnd - Qsnet kernel modules 5.20 and later,
28 openiblnd - IbGold 1.8.2,
29 o2iblnd - OFED 1.1 and 1.2,
30 viblnd - Voltaire ibhost 3.4.5 and later,
31 ciblnd - Topspin 3.2.0,
32 iiblnd - Infiniserv 3.3 + PathBits patch,
33 gmlnd - GM 2.1.22 and later,
34 mxlnd - MX 1.2.1 or later,
35 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
39 Description: /proc/sys/lnet has non-sysctl entries
40 Details : Updating dump_kernel/daemon_file/debug_mb to use sysctl variables
44 Description: TOE Kernel panic by ksocklnd
45 Details : offloaded sockets provide their own implementation of sendpage,
46 can't call tcp_sendpage() directly
50 Description: kibnal_shutdown() doesn't finish; lconf --cleanup hangs
51 Details : races between lnd_shutdown and peer creation prevent
52 lnd_shutdown from finishing.
56 Description: open files rlimit 1024 reached while liblustre testing
57 Details : ulnds/socklnd must close open socket after unsuccessful
62 Description: build error
63 Details : fix typos in gmlnd, ptllnd and viblnd
65 ------------------------------------------------------------------------------
67 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
69 * Support for networks:
70 socklnd - kernels up to 2.6.16,
71 qswlnd - Qsnet kernel modules 5.20 and later,
72 openiblnd - IbGold 1.8.2,
73 o2iblnd - OFED 1.1 and 1.2
74 viblnd - Voltaire ibhost 3.4.5 and later,
75 ciblnd - Topspin 3.2.0,
76 iiblnd - Infiniserv 3.3 + PathBits patch,
77 gmlnd - GM 2.1.22 and later,
78 mxlnd - MX 1.2.1 or later,
79 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
81 2007-06-21 Cluster File Systems, Inc. <info@clusterfs.com>
83 * Support for networks:
84 socklnd - kernels up to 2.6.16,
85 qswlnd - Qsnet kernel modules 5.20 and later,
86 openiblnd - IbGold 1.8.2,
88 viblnd - Voltaire ibhost 3.4.5 and later,
89 ciblnd - Topspin 3.2.0,
90 iiblnd - Infiniserv 3.3 + PathBits patch,
91 gmlnd - GM 2.1.22 and later,
92 mxlnd - MX 1.2.1 or later,
93 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
97 Description: Initialize cpumask before use
101 Description: ASSERTION failures when upgrading to the patchless zero-copy
103 Details : This bug affects "rolling upgrades", causing an inconsistent
104 protocol version negotiation and subsequent assertion failure
105 during rolling upgrades after the first wave of upgrades.
109 Details : Change "dropped message" CERRORs to D_NETERROR so they are
110 logged instead of creating "console chatter" when a lustre
111 timeout races with normal RPC completion.
114 Details : lnet_clear_peer_table can wait forever if user forgets to
118 Details : libcfs_id2str should check pid against LNET_PID_ANY.
122 Description: added LNET self test
123 Details : landing b_self_test
128 Description: cfs_duration_{u,n}sec() wrongly calculate nanosecond part of
130 Details : do_div() macro is used incorrectly.
132 2007-04-23 Cluster File Systems, Inc. <info@clusterfs.com>
136 Description: make panic on lbug configurable
140 Description: Add OFED1.2 support to o2iblnd
141 Details : o2iblnd depends on OFED's modules, if out-tree OFED's modules
142 are installed (other than kernel's in-tree infiniband), there
143 could be some problem while insmod o2iblnd (mismatch CRC of
145 If extra Module.symvers is supported in kernel (i.e, 2.6.17),
146 this link provides solution:
147 https://bugs.openfabrics.org/show_bug.cgi?id=355
148 if extra Module.symvers is not supported in kernel, we will
149 have to run the script in bug 12316 to update
150 $LINUX/module.symvers before building o2iblnd.
151 More details about this are in bug 12316.
153 ------------------------------------------------------------------------------
155 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
156 * version 1.4.10 / 1.6.0
157 * Support for networks:
158 socklnd - kernels up to 2.6.16,
159 qswlnd - Qsnet kernel modules 5.20 and later,
160 openiblnd - IbGold 1.8.2,
162 viblnd - Voltaire ibhost 3.4.5 and later,
163 ciblnd - Topspin 3.2.0,
164 iiblnd - Infiniserv 3.3 + PathBits patch,
165 gmlnd - GM 2.1.22 and later,
166 mxlnd - MX 1.2.1 or later,
167 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
171 Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
172 possibly accessed in kptllnd_shutdown. Ptllnd should init
173 kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
177 Description: gmlnd ignored some transmit errors when finalizing lnet messages.
181 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
185 Description: the_lnet.ln_finalizing was not set when the current thread is
186 about to complete messages. It only affects multi-threaded
192 Description: Changed the default kqswlnd ntxmsg=512
197 Description: Assertion failure in kernel ptllnd caused by posting passive
198 bulk buffers before connection establishment complete.
203 Description: A race in kernel ptllnd between deleting a peer and posting
204 new communications for it could hang communications -
205 manifesting as "Unexpectedly long timeout" messages.
210 Description: Kernel ptllnd lock ordering issue could hang a node.
215 Description: node crash on socket teardown race
218 Frequency : 'lctl peer_list' issued on a mx net
220 Description: Enable lctl's peer_list for MXLND
223 Frequency : after Ptllnd timeouts and portals congestion
225 Description: Credit overflows
226 Details : This was a bug in ptllnd connection establishment. The fix
227 implements better peer stamps to disambiguate connection
228 establishment and ensure both peers enter the credit flow
229 state machine consistently.
234 Description: kptllnd didn't propagate some network errors up to LNET
235 Details : This bug was spotted while investigating 11394. The fix
236 ensures network errors on sends and bulk transfers are
237 propagated to LNET/lustre correctly.
239 Severity : enhancement
241 Description: Fixed console chatter in case of -ETIMEDOUT.
243 Severity : enhancement
245 Description: Added D_NETTRACE for recording network packet history
246 (initially only for ptllnd). Also a separate userspace
247 ptllnd facility to gather history which should really be
248 covered by D_NETTRACE too, if only CDEBUG recorded history in
254 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
255 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
256 callback can occur before a connection has actually been
257 established. This caused an assertion failure previously.
259 Severity : enhancement
261 Description: Multiple instances for o2iblnd
262 Details : Allow multiple instances of o2iblnd to enable networking over
263 multiple HCAs and routing between them.
267 Description: lnet deadlock in router_checker
268 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
269 into BH locks to eliminate potential deadlock caused by
270 ksocknal_data_ready() preempting code holding these locks.
274 Description: Millions of failed socklnd connection attempts cause a very slow FS
275 Details : added a new route flag ksnr_scheduled to distinguish from
276 ksnr_connecting, so that a peer connection request is only turned
277 down for race concerns when an active connection to the same peer
278 is under progress (instead of just being scheduled).
280 ------------------------------------------------------------------------------
282 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
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
297 Severity : major on XT3
299 Description: libcfs overwrites /proc/sys/portals
300 Details : libcfs created a symlink from /proc/sys/portals to
301 /proc/sys/lnet for backwards compatibility. This is no
302 longer required and makes the Cray portals /proc variables
307 Description: OFED FMR API change
308 Details : This changes parameter usage to reflect a change in
309 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
310 that FMR support is only used in experimental versions of the
311 o2iblnd - this change does not affect standard usage at all.
313 Severity : enhancement
315 Description: new ko2iblnd module parameter: ib_mtu
316 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
317 HCAs. You can avoid this problem by setting the MTU to 1024
318 using this module parameter.
320 Severity : enhancement
321 Bugzilla : 11118/11620
322 Description: ptllnd small request message buffer alignment fix
323 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
324 Round up small message size on sends in case this option
325 is not supported. 11620 was a defect in the initial
326 implementation which effectively asserted all peers had to be
327 running the correct protocol version which was fixed by always
328 NAK-ing such requests and handling any misalignments they
333 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
334 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
335 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
337 Severity : enhancement
339 Description: Patchless ZC(zero copy) socklnd
340 Details : New protocol for socklnd, socklnd can support zero copy without
341 kernel patch, it's compatible with old socklnd. Checksum is
342 moved from tunables to modparams.
346 Description: When ksocknal_del_peer() is called upon a peer whose
347 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
348 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
351 Frequency : when ptlrpc is under heavy use and runs out of request buffer
353 Description: In lnet_match_blocked_msg(), md can be used without holding a
357 Frequency : very rarely
359 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
360 If connd connects a route which has been closed by
361 ksocknal_shutdown(), ksocknal_create_routes() may create new
362 routes which hold references on the peer, causing shutdown
363 process to wait for peer to disappear forever.
365 Severity : enhancement
367 Description: Dump XT3 portals traces on kptllnd timeout
368 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
369 dump Cray portals debug traces to a file. The kptllnd module
370 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
371 is the basename of the dump file.
374 Frequency : infrequent
376 Description: kernel ptllnd fix bug in connection re-establishment
377 Details : Kernel ptllnd could produce protocol errors e.g. illegal
378 matchbits and/or violate the credit flow protocol when trying
379 to re-establish a connection with a peer after an error or
382 Severity : enhancement
384 Description: Allow /proc/sys/lnet/debug to be set symbolically
385 Details : Allow debug and subsystem debug values to be read/set by name
386 in addition to numerically, for ease of use.
389 Frequency : only in configurations with LNET routers
391 Description: routes automatically marked down and recovered
392 Details : In configurations with LNET routers if a router fails routers
393 now actively try to recover routes that are down, unless they
394 are marked down by an administrator.
396 ------------------------------------------------------------------------------
398 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
401 Frequency : very rarely, in configurations with LNET routers and TCP
403 Description: incorrect data written to files on OSTs
404 Details : In certain high-load conditions incorrect data may be written
405 to files on the OST when using TCP networks.
407 ------------------------------------------------------------------------------
409 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
411 - rework CDEBUG messages rate-limiting mechanism b=10375
412 - add per-socket tunables for socklnd if the kernel is patched b=10327
414 ------------------------------------------------------------------------------
416 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
418 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
419 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
420 - quiet LNET startup LNI message for liblustre b=10128
421 - Better console error messages if 'ip2nets' can't match an IP address
422 - Fixed overflow/use-before-set bugs in linux-time.h
423 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
424 - LNET teardown failed an assertion about the route table being empty
425 - Fixed a crash in LNetEQPoll(<invalid handle>)
426 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
427 - improve debug message for liblustre/Catamount nodes (b=10116)
429 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
430 * Configuration change for the XT3
431 The PTLLND is now used to run Lustre over Portals on the XT3.
432 The configure option(s) --with-cray-portals are no longer
433 used. Rather --with-portals=<path-to-portals-includes> is
434 used to enable building on the XT3. In addition to enable
435 XT3 specific features the option --enable-cray-xt3 must be
438 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
439 * Portals has been removed, replaced by LNET.
440 LNET is new networking infrastructure for Lustre, it includes a
441 reorganized network configuration mode (see the user
442 documentation for full details) as well as support for routing
443 between different network fabrics. Lustre Networking Devices
444 (LNDS) for the supported network fabrics have also been created
445 for this new infrastructure.
447 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
452 Frequency : rare (large Voltaire clusters only)
454 Description: the default number of reserved transmit descriptors was too low
455 for some large clusters
456 Details : As a workaround, the number was increased. A proper fix includes
459 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
464 Frequency : occasional (large-scale events, cluster reboot, network failure)
466 Description: too many error messages on console obscure actual problem and
467 can slow down/panic server, or cause recovery to fail repeatedly
468 Details : enable rate-limiting of console error messages, and some messages
469 that were console errors now only go to the kernel log
471 Severity : enhancement
473 Description: add /proc/sys/portals/catastrophe entry which will report if
474 that node has previously LBUGged
476 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
478 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
480 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
482 - handle error return code in kranal_check_fma_rx() (5915,6054)
484 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
486 - update vibnal (Voltaire IB NAL)
487 - update gmnal (Myrinet NAL), gmnalid
489 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
491 * Landed portals:b_port_step as follows...
493 - removed CFS_DECL_SPIN*
494 just use 'spinlock_t' and initialise with spin_lock_init()
496 - removed CFS_DECL_MUTEX*
497 just use 'struct semaphore' and initialise with init_mutex()
499 - removed CFS_DECL_RWSEM*
500 just use 'struct rw_semaphore' and initialise with init_rwsem()
502 - renamed cfs_sleep_chan -> cfs_waitq
503 cfs_sleep_link -> cfs_waitlink
505 - fixed race in linux version of arch-independent socknal
506 (the ENOMEM/EAGAIN decision).
508 - Didn't fix problems in Darwin version of arch-independent socknal
509 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
511 - removed libcfs types from non-socknal header files (only some types
512 in the header files had been changed; the .c files hadn't been