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
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
16 ------------------------------------------------------------------------------
18 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
19 * version 1.4.10 / 1.6.0
20 * Support for networks:
21 socklnd - kernels up to 2.6.16
22 qswlnd - Qsnet kernel modules 5.20 and later
23 openiblnd - IbGold 1.8.2
25 viblnd - Voltaire ibhost 3.4.5 and later
26 ciblnd - Topspin 3.2.0
27 iiblnd - Infiniserv 3.3 + PathBits patch
28 gmlnd - GM 2.1.22 and later
29 mxlnd - MX 1.2.1 or later
30 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
35 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
39 Description: the_lnet.ln_finalizing was not set when the current thread is
40 about to complete messages. It only affects multi-threaded
46 Description: Changed the default kqswlnd ntxmsg=512
51 Description: Assertion failure in kernel ptllnd caused by posting passive
52 bulk buffers before connection establishment complete.
57 Description: A race in kernel ptllnd between deleting a peer and posting
58 new communications for it could hang communications -
59 manifesting as "Unexpectedly long timeout" messages.
64 Description: Kernel ptllnd lock ordering issue could hang a node.
69 Description: node crash on socket teardown race
72 Frequency : 'lctl peer_list' issued on a mx net
74 Description: Enable lctl's peer_list for MXLND
77 Frequency : after Ptllnd timeouts and portals congestion
79 Description: Credit overflows
80 Details : This was a bug in ptllnd connection establishment. The fix
81 implements better peer stamps to disambiguate connection
82 establishment and ensure both peers enter the credit flow
83 state machine consistently.
88 Description: kptllnd didn't propagate some network errors up to LNET
89 Details : This bug was spotted while investigating 11394. The fix
90 ensures network errors on sends and bulk transfers are
91 propagated to LNET/lustre correctly.
93 Severity : enhancement
95 Description: Fixed console chatter in case of -ETIMEDOUT.
97 Severity : enhancement
99 Description: Added D_NETTRACE for recording network packet history
100 (initially only for ptllnd). Also a separate userspace
101 ptllnd facility to gather history which should really be
102 covered by D_NETTRACE too, if only CDEBUG recorded history in
108 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
109 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
110 callback can occur before a connection has actually been
111 established. This caused an assertion failure previously.
113 Severity : enhancement
115 Description: Multiple instances for o2iblnd
116 Details : Allow multiple instances of o2iblnd to enable networking over
117 multiple HCAs and routing between them.
121 Description: lnet deadlock in router_checker
122 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
123 into BH locks to eliminate potential deadlock caused by
124 ksocknal_data_ready() preempting code holding these locks.
128 Description: Millions of failed socklnd connection attempts cause a very slow FS
129 Details : added a new route flag ksnr_scheduled to distinguish from
130 ksnr_connecting, so that a peer connection request is only turned
131 down for race concerns when an active connection to the same peer
132 is under progress (instead of just being scheduled).
134 ------------------------------------------------------------------------------
136 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
138 * Support for networks:
139 socklnd - kernels up to 2.6.16
140 qswlnd - Qsnet kernel modules 5.20 and later
141 openiblnd - IbGold 1.8.2
143 viblnd - Voltaire ibhost 3.4.5 and later
144 ciblnd - Topspin 3.2.0
145 iiblnd - Infiniserv 3.3 + PathBits patch
146 gmlnd - GM 2.1.22 and later
147 mxlnd - MX 1.2.1 or later
148 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
151 Severity : major on XT3
153 Description: libcfs overwrites /proc/sys/portals
154 Details : libcfs created a symlink from /proc/sys/portals to
155 /proc/sys/lnet for backwards compatibility. This is no
156 longer required and makes the Cray portals /proc variables
161 Description: OFED FMR API change
162 Details : This changes parameter usage to reflect a change in
163 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
164 that FMR support is only used in experimental versions of the
165 o2iblnd - this change does not affect standard usage at all.
167 Severity : enhancement
169 Description: new ko2iblnd module parameter: ib_mtu
170 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
171 HCAs. You can avoid this problem by setting the MTU to 1024
172 using this module parameter.
174 Severity : enhancement
175 Bugzilla : 11118/11620
176 Description: ptllnd small request message buffer alignment fix
177 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
178 Round up small message size on sends in case this option
179 is not supported. 11620 was a defect in the initial
180 implementation which effectively asserted all peers had to be
181 running the correct protocol version which was fixed by always
182 NAK-ing such requests and handling any misalignments they
187 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
188 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
189 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
191 Severity : enhancement
193 Description: Patchless ZC(zero copy) socklnd
194 Details : New protocol for socklnd, socklnd can support zero copy without
195 kernel patch, it's compatible with old socklnd. Checksum is
196 moved from tunables to modparams.
200 Description: When ksocknal_del_peer() is called upon a peer whose
201 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
202 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
205 Frequency : when ptlrpc is under heavy use and runs out of request buffer
207 Description: In lnet_match_blocked_msg(), md can be used without holding a
211 Frequency : very rarely
213 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
214 If connd connects a route which has been closed by
215 ksocknal_shutdown(), ksocknal_create_routes() may create new
216 routes which hold references on the peer, causing shutdown
217 process to wait for peer to disappear forever.
219 Severity : enhancement
221 Description: Dump XT3 portals traces on kptllnd timeout
222 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
223 dump Cray portals debug traces to a file. The kptllnd module
224 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
225 is the basename of the dump file.
228 Frequency : infrequent
230 Description: kernel ptllnd fix bug in connection re-establishment
231 Details : Kernel ptllnd could produce protocol errors e.g. illegal
232 matchbits and/or violate the credit flow protocol when trying
233 to re-establish a connection with a peer after an error or
236 Severity : enhancement
238 Description: Allow /proc/sys/lnet/debug to be set symbolically
239 Details : Allow debug and subsystem debug values to be read/set by name
240 in addition to numerically, for ease of use.
243 Frequency : only in configurations with LNET routers
245 Description: routes automatically marked down and recovered
246 Details : In configurations with LNET routers if a router fails routers
247 now actively try to recover routes that are down, unless they
248 are marked down by an administrator.
250 ------------------------------------------------------------------------------
252 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
255 Frequency : very rarely, in configurations with LNET routers and TCP
257 Description: incorrect data written to files on OSTs
258 Details : In certain high-load conditions incorrect data may be written
259 to files on the OST when using TCP networks.
261 ------------------------------------------------------------------------------
263 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
265 - rework CDEBUG messages rate-limiting mechanism b=10375
266 - add per-socket tunables for socklnd if the kernel is patched b=10327
268 ------------------------------------------------------------------------------
270 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
272 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
273 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
274 - quiet LNET startup LNI message for liblustre b=10128
275 - Better console error messages if 'ip2nets' can't match an IP address
276 - Fixed overflow/use-before-set bugs in linux-time.h
277 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
278 - LNET teardown failed an assertion about the route table being empty
279 - Fixed a crash in LNetEQPoll(<invalid handle>)
280 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
281 - improve debug message for liblustre/Catamount nodes (b=10116)
283 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
284 * Configuration change for the XT3
285 The PTLLND is now used to run Lustre over Portals on the XT3.
286 The configure option(s) --with-cray-portals are no longer
287 used. Rather --with-portals=<path-to-portals-includes> is
288 used to enable building on the XT3. In addition to enable
289 XT3 specific features the option --enable-cray-xt3 must be
292 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
293 * Portals has been removed, replaced by LNET.
294 LNET is new networking infrastructure for Lustre, it includes a
295 reorganized network configuration mode (see the user
296 documentation for full details) as well as support for routing
297 between different network fabrics. Lustre Networking Devices
298 (LNDS) for the supported network fabrics have also been created
299 for this new infrastructure.
301 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
306 Frequency : rare (large Voltaire clusters only)
308 Description: the default number of reserved transmit descriptors was too low
309 for some large clusters
310 Details : As a workaround, the number was increased. A proper fix includes
313 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
318 Frequency : occasional (large-scale events, cluster reboot, network failure)
320 Description: too many error messages on console obscure actual problem and
321 can slow down/panic server, or cause recovery to fail repeatedly
322 Details : enable rate-limiting of console error messages, and some messages
323 that were console errors now only go to the kernel log
325 Severity : enhancement
327 Description: add /proc/sys/portals/catastrophe entry which will report if
328 that node has previously LBUGged
330 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
332 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
334 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
336 - handle error return code in kranal_check_fma_rx() (5915,6054)
338 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
340 - update vibnal (Voltaire IB NAL)
341 - update gmnal (Myrinet NAL), gmnalid
343 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
345 * Landed portals:b_port_step as follows...
347 - removed CFS_DECL_SPIN*
348 just use 'spinlock_t' and initialise with spin_lock_init()
350 - removed CFS_DECL_MUTEX*
351 just use 'struct semaphore' and initialise with init_mutex()
353 - removed CFS_DECL_RWSEM*
354 just use 'struct rw_semaphore' and initialise with init_rwsem()
356 - renamed cfs_sleep_chan -> cfs_waitq
357 cfs_sleep_link -> cfs_waitlink
359 - fixed race in linux version of arch-independent socknal
360 (the ENOMEM/EAGAIN decision).
362 - Didn't fix problems in Darwin version of arch-independent socknal
363 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
365 - removed libcfs types from non-socknal header files (only some types
366 in the header files had been changed; the .c files hadn't been