1 2007-04-23 Cluster File Systems, Inc. <info@clusterfs.com>
2 * version 1.4.11 / 1.6.1
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
18 Description: make panic on lbug configurable
22 Description: Add OFED1.2 support to o2iblnd
23 Details : o2iblnd depends on OFED's modules, if out-tree OFED's modules
24 are installed (other than kernel's in-tree infiniband), there
25 could be some problem while insmod o2iblnd (mismatch CRC of
27 If extra Module.symvers is supported in kernel (i.e, 2.6.17),
28 this link provides solution:
29 https://bugs.openfabrics.org/show_bug.cgi?id=355
30 if extra Module.symvers is not supported in kernel, we will
31 have to run the script in bug 12316 to update
32 $LINUX/module.symvers before building o2iblnd.
33 More details about this are in bug 12316.
35 ------------------------------------------------------------------------------
37 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
38 * version 1.4.10 / 1.6.0
39 * Support for networks:
40 socklnd - kernels up to 2.6.16
41 qswlnd - Qsnet kernel modules 5.20 and later
42 openiblnd - IbGold 1.8.2
44 viblnd - Voltaire ibhost 3.4.5 and later
45 ciblnd - Topspin 3.2.0
46 iiblnd - Infiniserv 3.3 + PathBits patch
47 gmlnd - GM 2.1.22 and later
48 mxlnd - MX 1.2.1 or later
49 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
54 Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
55 possibly accessed in kptllnd_shutdown. Ptllnd should init
56 kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
60 Description: gmlnd ignored some transmit errors when finalizing lnet messages.
64 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
68 Description: the_lnet.ln_finalizing was not set when the current thread is
69 about to complete messages. It only affects multi-threaded
75 Description: Changed the default kqswlnd ntxmsg=512
80 Description: Assertion failure in kernel ptllnd caused by posting passive
81 bulk buffers before connection establishment complete.
86 Description: A race in kernel ptllnd between deleting a peer and posting
87 new communications for it could hang communications -
88 manifesting as "Unexpectedly long timeout" messages.
93 Description: Kernel ptllnd lock ordering issue could hang a node.
98 Description: node crash on socket teardown race
101 Frequency : 'lctl peer_list' issued on a mx net
103 Description: Enable lctl's peer_list for MXLND
106 Frequency : after Ptllnd timeouts and portals congestion
108 Description: Credit overflows
109 Details : This was a bug in ptllnd connection establishment. The fix
110 implements better peer stamps to disambiguate connection
111 establishment and ensure both peers enter the credit flow
112 state machine consistently.
117 Description: kptllnd didn't propagate some network errors up to LNET
118 Details : This bug was spotted while investigating 11394. The fix
119 ensures network errors on sends and bulk transfers are
120 propagated to LNET/lustre correctly.
122 Severity : enhancement
124 Description: Fixed console chatter in case of -ETIMEDOUT.
126 Severity : enhancement
128 Description: Added D_NETTRACE for recording network packet history
129 (initially only for ptllnd). Also a separate userspace
130 ptllnd facility to gather history which should really be
131 covered by D_NETTRACE too, if only CDEBUG recorded history in
137 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
138 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
139 callback can occur before a connection has actually been
140 established. This caused an assertion failure previously.
142 Severity : enhancement
144 Description: Multiple instances for o2iblnd
145 Details : Allow multiple instances of o2iblnd to enable networking over
146 multiple HCAs and routing between them.
150 Description: lnet deadlock in router_checker
151 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
152 into BH locks to eliminate potential deadlock caused by
153 ksocknal_data_ready() preempting code holding these locks.
157 Description: Millions of failed socklnd connection attempts cause a very slow FS
158 Details : added a new route flag ksnr_scheduled to distinguish from
159 ksnr_connecting, so that a peer connection request is only turned
160 down for race concerns when an active connection to the same peer
161 is under progress (instead of just being scheduled).
163 ------------------------------------------------------------------------------
165 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
167 * Support for networks:
168 socklnd - kernels up to 2.6.16
169 qswlnd - Qsnet kernel modules 5.20 and later
170 openiblnd - IbGold 1.8.2
172 viblnd - Voltaire ibhost 3.4.5 and later
173 ciblnd - Topspin 3.2.0
174 iiblnd - Infiniserv 3.3 + PathBits patch
175 gmlnd - GM 2.1.22 and later
176 mxlnd - MX 1.2.1 or later
177 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
180 Severity : major on XT3
182 Description: libcfs overwrites /proc/sys/portals
183 Details : libcfs created a symlink from /proc/sys/portals to
184 /proc/sys/lnet for backwards compatibility. This is no
185 longer required and makes the Cray portals /proc variables
190 Description: OFED FMR API change
191 Details : This changes parameter usage to reflect a change in
192 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
193 that FMR support is only used in experimental versions of the
194 o2iblnd - this change does not affect standard usage at all.
196 Severity : enhancement
198 Description: new ko2iblnd module parameter: ib_mtu
199 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
200 HCAs. You can avoid this problem by setting the MTU to 1024
201 using this module parameter.
203 Severity : enhancement
204 Bugzilla : 11118/11620
205 Description: ptllnd small request message buffer alignment fix
206 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
207 Round up small message size on sends in case this option
208 is not supported. 11620 was a defect in the initial
209 implementation which effectively asserted all peers had to be
210 running the correct protocol version which was fixed by always
211 NAK-ing such requests and handling any misalignments they
216 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
217 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
218 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
220 Severity : enhancement
222 Description: Patchless ZC(zero copy) socklnd
223 Details : New protocol for socklnd, socklnd can support zero copy without
224 kernel patch, it's compatible with old socklnd. Checksum is
225 moved from tunables to modparams.
229 Description: When ksocknal_del_peer() is called upon a peer whose
230 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
231 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
234 Frequency : when ptlrpc is under heavy use and runs out of request buffer
236 Description: In lnet_match_blocked_msg(), md can be used without holding a
240 Frequency : very rarely
242 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
243 If connd connects a route which has been closed by
244 ksocknal_shutdown(), ksocknal_create_routes() may create new
245 routes which hold references on the peer, causing shutdown
246 process to wait for peer to disappear forever.
248 Severity : enhancement
250 Description: Dump XT3 portals traces on kptllnd timeout
251 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
252 dump Cray portals debug traces to a file. The kptllnd module
253 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
254 is the basename of the dump file.
257 Frequency : infrequent
259 Description: kernel ptllnd fix bug in connection re-establishment
260 Details : Kernel ptllnd could produce protocol errors e.g. illegal
261 matchbits and/or violate the credit flow protocol when trying
262 to re-establish a connection with a peer after an error or
265 Severity : enhancement
267 Description: Allow /proc/sys/lnet/debug to be set symbolically
268 Details : Allow debug and subsystem debug values to be read/set by name
269 in addition to numerically, for ease of use.
272 Frequency : only in configurations with LNET routers
274 Description: routes automatically marked down and recovered
275 Details : In configurations with LNET routers if a router fails routers
276 now actively try to recover routes that are down, unless they
277 are marked down by an administrator.
279 ------------------------------------------------------------------------------
281 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
284 Frequency : very rarely, in configurations with LNET routers and TCP
286 Description: incorrect data written to files on OSTs
287 Details : In certain high-load conditions incorrect data may be written
288 to files on the OST when using TCP networks.
290 ------------------------------------------------------------------------------
292 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
294 - rework CDEBUG messages rate-limiting mechanism b=10375
295 - add per-socket tunables for socklnd if the kernel is patched b=10327
297 ------------------------------------------------------------------------------
299 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
301 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
302 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
303 - quiet LNET startup LNI message for liblustre b=10128
304 - Better console error messages if 'ip2nets' can't match an IP address
305 - Fixed overflow/use-before-set bugs in linux-time.h
306 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
307 - LNET teardown failed an assertion about the route table being empty
308 - Fixed a crash in LNetEQPoll(<invalid handle>)
309 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
310 - improve debug message for liblustre/Catamount nodes (b=10116)
312 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
313 * Configuration change for the XT3
314 The PTLLND is now used to run Lustre over Portals on the XT3.
315 The configure option(s) --with-cray-portals are no longer
316 used. Rather --with-portals=<path-to-portals-includes> is
317 used to enable building on the XT3. In addition to enable
318 XT3 specific features the option --enable-cray-xt3 must be
321 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
322 * Portals has been removed, replaced by LNET.
323 LNET is new networking infrastructure for Lustre, it includes a
324 reorganized network configuration mode (see the user
325 documentation for full details) as well as support for routing
326 between different network fabrics. Lustre Networking Devices
327 (LNDS) for the supported network fabrics have also been created
328 for this new infrastructure.
330 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
335 Frequency : rare (large Voltaire clusters only)
337 Description: the default number of reserved transmit descriptors was too low
338 for some large clusters
339 Details : As a workaround, the number was increased. A proper fix includes
342 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
347 Frequency : occasional (large-scale events, cluster reboot, network failure)
349 Description: too many error messages on console obscure actual problem and
350 can slow down/panic server, or cause recovery to fail repeatedly
351 Details : enable rate-limiting of console error messages, and some messages
352 that were console errors now only go to the kernel log
354 Severity : enhancement
356 Description: add /proc/sys/portals/catastrophe entry which will report if
357 that node has previously LBUGged
359 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
361 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
363 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
365 - handle error return code in kranal_check_fma_rx() (5915,6054)
367 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
369 - update vibnal (Voltaire IB NAL)
370 - update gmnal (Myrinet NAL), gmnalid
372 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
374 * Landed portals:b_port_step as follows...
376 - removed CFS_DECL_SPIN*
377 just use 'spinlock_t' and initialise with spin_lock_init()
379 - removed CFS_DECL_MUTEX*
380 just use 'struct semaphore' and initialise with init_mutex()
382 - removed CFS_DECL_RWSEM*
383 just use 'struct rw_semaphore' and initialise with init_rwsem()
385 - renamed cfs_sleep_chan -> cfs_waitq
386 cfs_sleep_link -> cfs_waitlink
388 - fixed race in linux version of arch-independent socknal
389 (the ENOMEM/EAGAIN decision).
391 - Didn't fix problems in Darwin version of arch-independent socknal
392 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
394 - removed libcfs types from non-socknal header files (only some types
395 in the header files had been changed; the .c files hadn't been