1 2007-08-27 Cluster File Systems, Inc. <info@clusterfs.com>
3 * Support for networks:
4 socklnd - kernels up to 2.6.16
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
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
15 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
16 * version 1.4.11 / 1.6.1
17 * Support for networks:
18 socklnd - kernels up to 2.6.16
19 qswlnd - Qsnet kernel modules 5.20 and later
20 openiblnd - IbGold 1.8.2
21 o2iblnd - OFED 1.1 and 1.2
22 viblnd - Voltaire ibhost 3.4.5 and later
23 ciblnd - Topspin 3.2.0
24 iiblnd - Infiniserv 3.3 + PathBits patch
25 gmlnd - GM 2.1.22 and later
26 mxlnd - MX 1.2.1 or later
27 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
32 Description: ASSERTION failures when upgrading to the patchless zero-copy
34 Details : This bug affects "rolling upgrades", causing an inconsistent
35 protocol version negotiation and subsequent assertion failure
36 during rolling upgrades after the first wave of upgrades.
40 Details : Change "dropped message" CERRORs to D_NETERROR so they are
41 logged instead of creating "console chatter" when a lustre
42 timeout races with normal RPC completion.
45 Details : lnet_clear_peer_table can wait forever if user forgets to
49 Details : libcfs_id2str should check pid against LNET_PID_ANY.
53 Description: added LNET self test
54 Details : landing b_self_test
59 Description: cfs_duration_{u,n}sec() wrongly calculate nanosecond part of
61 Details : do_div() macro is used incorrectly.
63 2007-04-23 Cluster File Systems, Inc. <info@clusterfs.com>
67 Description: make panic on lbug configurable
71 Description: Add OFED1.2 support to o2iblnd
72 Details : o2iblnd depends on OFED's modules, if out-tree OFED's modules
73 are installed (other than kernel's in-tree infiniband), there
74 could be some problem while insmod o2iblnd (mismatch CRC of
76 If extra Module.symvers is supported in kernel (i.e, 2.6.17),
77 this link provides solution:
78 https://bugs.openfabrics.org/show_bug.cgi?id=355
79 if extra Module.symvers is not supported in kernel, we will
80 have to run the script in bug 12316 to update
81 $LINUX/module.symvers before building o2iblnd.
82 More details about this are in bug 12316.
84 ------------------------------------------------------------------------------
86 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
87 * version 1.4.10 / 1.6.0
88 * Support for networks:
89 socklnd - kernels up to 2.6.16
90 qswlnd - Qsnet kernel modules 5.20 and later
91 openiblnd - IbGold 1.8.2
93 viblnd - Voltaire ibhost 3.4.5 and later
94 ciblnd - Topspin 3.2.0
95 iiblnd - Infiniserv 3.3 + PathBits patch
96 gmlnd - GM 2.1.22 and later
97 mxlnd - MX 1.2.1 or later
98 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
103 Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
104 possibly accessed in kptllnd_shutdown. Ptllnd should init
105 kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
109 Description: gmlnd ignored some transmit errors when finalizing lnet messages.
113 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
117 Description: the_lnet.ln_finalizing was not set when the current thread is
118 about to complete messages. It only affects multi-threaded
124 Description: Changed the default kqswlnd ntxmsg=512
129 Description: Assertion failure in kernel ptllnd caused by posting passive
130 bulk buffers before connection establishment complete.
135 Description: A race in kernel ptllnd between deleting a peer and posting
136 new communications for it could hang communications -
137 manifesting as "Unexpectedly long timeout" messages.
142 Description: Kernel ptllnd lock ordering issue could hang a node.
147 Description: node crash on socket teardown race
150 Frequency : 'lctl peer_list' issued on a mx net
152 Description: Enable lctl's peer_list for MXLND
155 Frequency : after Ptllnd timeouts and portals congestion
157 Description: Credit overflows
158 Details : This was a bug in ptllnd connection establishment. The fix
159 implements better peer stamps to disambiguate connection
160 establishment and ensure both peers enter the credit flow
161 state machine consistently.
166 Description: kptllnd didn't propagate some network errors up to LNET
167 Details : This bug was spotted while investigating 11394. The fix
168 ensures network errors on sends and bulk transfers are
169 propagated to LNET/lustre correctly.
171 Severity : enhancement
173 Description: Fixed console chatter in case of -ETIMEDOUT.
175 Severity : enhancement
177 Description: Added D_NETTRACE for recording network packet history
178 (initially only for ptllnd). Also a separate userspace
179 ptllnd facility to gather history which should really be
180 covered by D_NETTRACE too, if only CDEBUG recorded history in
186 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
187 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
188 callback can occur before a connection has actually been
189 established. This caused an assertion failure previously.
191 Severity : enhancement
193 Description: Multiple instances for o2iblnd
194 Details : Allow multiple instances of o2iblnd to enable networking over
195 multiple HCAs and routing between them.
199 Description: lnet deadlock in router_checker
200 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
201 into BH locks to eliminate potential deadlock caused by
202 ksocknal_data_ready() preempting code holding these locks.
206 Description: Millions of failed socklnd connection attempts cause a very slow FS
207 Details : added a new route flag ksnr_scheduled to distinguish from
208 ksnr_connecting, so that a peer connection request is only turned
209 down for race concerns when an active connection to the same peer
210 is under progress (instead of just being scheduled).
212 ------------------------------------------------------------------------------
214 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
216 * Support for networks:
217 socklnd - kernels up to 2.6.16
218 qswlnd - Qsnet kernel modules 5.20 and later
219 openiblnd - IbGold 1.8.2
221 viblnd - Voltaire ibhost 3.4.5 and later
222 ciblnd - Topspin 3.2.0
223 iiblnd - Infiniserv 3.3 + PathBits patch
224 gmlnd - GM 2.1.22 and later
225 mxlnd - MX 1.2.1 or later
226 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
229 Severity : major on XT3
231 Description: libcfs overwrites /proc/sys/portals
232 Details : libcfs created a symlink from /proc/sys/portals to
233 /proc/sys/lnet for backwards compatibility. This is no
234 longer required and makes the Cray portals /proc variables
239 Description: OFED FMR API change
240 Details : This changes parameter usage to reflect a change in
241 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
242 that FMR support is only used in experimental versions of the
243 o2iblnd - this change does not affect standard usage at all.
245 Severity : enhancement
247 Description: new ko2iblnd module parameter: ib_mtu
248 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
249 HCAs. You can avoid this problem by setting the MTU to 1024
250 using this module parameter.
252 Severity : enhancement
253 Bugzilla : 11118/11620
254 Description: ptllnd small request message buffer alignment fix
255 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
256 Round up small message size on sends in case this option
257 is not supported. 11620 was a defect in the initial
258 implementation which effectively asserted all peers had to be
259 running the correct protocol version which was fixed by always
260 NAK-ing such requests and handling any misalignments they
265 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
266 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
267 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
269 Severity : enhancement
271 Description: Patchless ZC(zero copy) socklnd
272 Details : New protocol for socklnd, socklnd can support zero copy without
273 kernel patch, it's compatible with old socklnd. Checksum is
274 moved from tunables to modparams.
278 Description: When ksocknal_del_peer() is called upon a peer whose
279 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
280 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
283 Frequency : when ptlrpc is under heavy use and runs out of request buffer
285 Description: In lnet_match_blocked_msg(), md can be used without holding a
289 Frequency : very rarely
291 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
292 If connd connects a route which has been closed by
293 ksocknal_shutdown(), ksocknal_create_routes() may create new
294 routes which hold references on the peer, causing shutdown
295 process to wait for peer to disappear forever.
297 Severity : enhancement
299 Description: Dump XT3 portals traces on kptllnd timeout
300 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
301 dump Cray portals debug traces to a file. The kptllnd module
302 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
303 is the basename of the dump file.
306 Frequency : infrequent
308 Description: kernel ptllnd fix bug in connection re-establishment
309 Details : Kernel ptllnd could produce protocol errors e.g. illegal
310 matchbits and/or violate the credit flow protocol when trying
311 to re-establish a connection with a peer after an error or
314 Severity : enhancement
316 Description: Allow /proc/sys/lnet/debug to be set symbolically
317 Details : Allow debug and subsystem debug values to be read/set by name
318 in addition to numerically, for ease of use.
321 Frequency : only in configurations with LNET routers
323 Description: routes automatically marked down and recovered
324 Details : In configurations with LNET routers if a router fails routers
325 now actively try to recover routes that are down, unless they
326 are marked down by an administrator.
328 ------------------------------------------------------------------------------
330 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
333 Frequency : very rarely, in configurations with LNET routers and TCP
335 Description: incorrect data written to files on OSTs
336 Details : In certain high-load conditions incorrect data may be written
337 to files on the OST when using TCP networks.
339 ------------------------------------------------------------------------------
341 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
343 - rework CDEBUG messages rate-limiting mechanism b=10375
344 - add per-socket tunables for socklnd if the kernel is patched b=10327
346 ------------------------------------------------------------------------------
348 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
350 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
351 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
352 - quiet LNET startup LNI message for liblustre b=10128
353 - Better console error messages if 'ip2nets' can't match an IP address
354 - Fixed overflow/use-before-set bugs in linux-time.h
355 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
356 - LNET teardown failed an assertion about the route table being empty
357 - Fixed a crash in LNetEQPoll(<invalid handle>)
358 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
359 - improve debug message for liblustre/Catamount nodes (b=10116)
361 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
362 * Configuration change for the XT3
363 The PTLLND is now used to run Lustre over Portals on the XT3.
364 The configure option(s) --with-cray-portals are no longer
365 used. Rather --with-portals=<path-to-portals-includes> is
366 used to enable building on the XT3. In addition to enable
367 XT3 specific features the option --enable-cray-xt3 must be
370 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
371 * Portals has been removed, replaced by LNET.
372 LNET is new networking infrastructure for Lustre, it includes a
373 reorganized network configuration mode (see the user
374 documentation for full details) as well as support for routing
375 between different network fabrics. Lustre Networking Devices
376 (LNDS) for the supported network fabrics have also been created
377 for this new infrastructure.
379 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
384 Frequency : rare (large Voltaire clusters only)
386 Description: the default number of reserved transmit descriptors was too low
387 for some large clusters
388 Details : As a workaround, the number was increased. A proper fix includes
391 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
396 Frequency : occasional (large-scale events, cluster reboot, network failure)
398 Description: too many error messages on console obscure actual problem and
399 can slow down/panic server, or cause recovery to fail repeatedly
400 Details : enable rate-limiting of console error messages, and some messages
401 that were console errors now only go to the kernel log
403 Severity : enhancement
405 Description: add /proc/sys/portals/catastrophe entry which will report if
406 that node has previously LBUGged
408 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
410 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
412 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
414 - handle error return code in kranal_check_fma_rx() (5915,6054)
416 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
418 - update vibnal (Voltaire IB NAL)
419 - update gmnal (Myrinet NAL), gmnalid
421 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
423 * Landed portals:b_port_step as follows...
425 - removed CFS_DECL_SPIN*
426 just use 'spinlock_t' and initialise with spin_lock_init()
428 - removed CFS_DECL_MUTEX*
429 just use 'struct semaphore' and initialise with init_mutex()
431 - removed CFS_DECL_RWSEM*
432 just use 'struct rw_semaphore' and initialise with init_rwsem()
434 - renamed cfs_sleep_chan -> cfs_waitq
435 cfs_sleep_link -> cfs_waitlink
437 - fixed race in linux version of arch-independent socknal
438 (the ENOMEM/EAGAIN decision).
440 - Didn't fix problems in Darwin version of arch-independent socknal
441 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
443 - removed libcfs types from non-socknal header files (only some types
444 in the header files had been changed; the .c files hadn't been