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: the_lnet.ln_finalizing was not set when the current thread is
36 about to complete messages. It only affects multi-threaded
42 Description: Changed the default kqswlnd ntxmsg=512
47 Description: Assertion failure in kernel ptllnd caused by posting passive
48 bulk buffers before connection establishment complete.
53 Description: A race in kernel ptllnd between deleting a peer and posting
54 new communications for it could hang communications -
55 manifesting as "Unexpectedly long timeout" messages.
60 Description: Kernel ptllnd lock ordering issue could hang a node.
65 Description: node crash on socket teardown race
68 Frequency : 'lctl peer_list' issued on a mx net
70 Description: Enable lctl's peer_list for MXLND
73 Frequency : after Ptllnd timeouts and portals congestion
75 Description: Credit overflows
76 Details : This was a bug in ptllnd connection establishment. The fix
77 implements better peer stamps to disambiguate connection
78 establishment and ensure both peers enter the credit flow
79 state machine consistently.
84 Description: kptllnd didn't propagate some network errors up to LNET
85 Details : This bug was spotted while investigating 11394. The fix
86 ensures network errors on sends and bulk transfers are
87 propagated to LNET/lustre correctly.
89 Severity : enhancement
91 Description: Fixed console chatter in case of -ETIMEDOUT.
93 Severity : enhancement
95 Description: Added D_NETTRACE for recording network packet history
96 (initially only for ptllnd). Also a separate userspace
97 ptllnd facility to gather history which should really be
98 covered by D_NETTRACE too, if only CDEBUG recorded history in
104 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
105 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
106 callback can occur before a connection has actually been
107 established. This caused an assertion failure previously.
109 Severity : enhancement
111 Description: Multiple instances for o2iblnd
112 Details : Allow multiple instances of o2iblnd to enable networking over
113 multiple HCAs and routing between them.
117 Description: lnet deadlock in router_checker
118 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
119 into BH locks to eliminate potential deadlock caused by
120 ksocknal_data_ready() preempting code holding these locks.
124 Description: Millions of failed socklnd connection attempts cause a very slow FS
125 Details : added a new route flag ksnr_scheduled to distinguish from
126 ksnr_connecting, so that a peer connection request is only turned
127 down for race concerns when an active connection to the same peer
128 is under progress (instead of just being scheduled).
130 ------------------------------------------------------------------------------
132 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
134 * Support for networks:
135 socklnd - kernels up to 2.6.16
136 qswlnd - Qsnet kernel modules 5.20 and later
137 openiblnd - IbGold 1.8.2
139 viblnd - Voltaire ibhost 3.4.5 and later
140 ciblnd - Topspin 3.2.0
141 iiblnd - Infiniserv 3.3 + PathBits patch
142 gmlnd - GM 2.1.22 and later
143 mxlnd - MX 1.2.1 or later
144 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
147 Severity : major on XT3
149 Description: libcfs overwrites /proc/sys/portals
150 Details : libcfs created a symlink from /proc/sys/portals to
151 /proc/sys/lnet for backwards compatibility. This is no
152 longer required and makes the Cray portals /proc variables
157 Description: OFED FMR API change
158 Details : This changes parameter usage to reflect a change in
159 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
160 that FMR support is only used in experimental versions of the
161 o2iblnd - this change does not affect standard usage at all.
163 Severity : enhancement
165 Description: new ko2iblnd module parameter: ib_mtu
166 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
167 HCAs. You can avoid this problem by setting the MTU to 1024
168 using this module parameter.
170 Severity : enhancement
171 Bugzilla : 11118/11620
172 Description: ptllnd small request message buffer alignment fix
173 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
174 Round up small message size on sends in case this option
175 is not supported. 11620 was a defect in the initial
176 implementation which effectively asserted all peers had to be
177 running the correct protocol version which was fixed by always
178 NAK-ing such requests and handling any misalignments they
183 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
184 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
185 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
187 Severity : enhancement
189 Description: Patchless ZC(zero copy) socklnd
190 Details : New protocol for socklnd, socklnd can support zero copy without
191 kernel patch, it's compatible with old socklnd. Checksum is
192 moved from tunables to modparams.
196 Description: When ksocknal_del_peer() is called upon a peer whose
197 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
198 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
201 Frequency : when ptlrpc is under heavy use and runs out of request buffer
203 Description: In lnet_match_blocked_msg(), md can be used without holding a
207 Frequency : very rarely
209 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
210 If connd connects a route which has been closed by
211 ksocknal_shutdown(), ksocknal_create_routes() may create new
212 routes which hold references on the peer, causing shutdown
213 process to wait for peer to disappear forever.
215 Severity : enhancement
217 Description: Dump XT3 portals traces on kptllnd timeout
218 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
219 dump Cray portals debug traces to a file. The kptllnd module
220 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
221 is the basename of the dump file.
224 Frequency : infrequent
226 Description: kernel ptllnd fix bug in connection re-establishment
227 Details : Kernel ptllnd could produce protocol errors e.g. illegal
228 matchbits and/or violate the credit flow protocol when trying
229 to re-establish a connection with a peer after an error or
232 Severity : enhancement
234 Description: Allow /proc/sys/lnet/debug to be set symbolically
235 Details : Allow debug and subsystem debug values to be read/set by name
236 in addition to numerically, for ease of use.
239 Frequency : only in configurations with LNET routers
241 Description: routes automatically marked down and recovered
242 Details : In configurations with LNET routers if a router fails routers
243 now actively try to recover routes that are down, unless they
244 are marked down by an administrator.
246 ------------------------------------------------------------------------------
248 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
251 Frequency : very rarely, in configurations with LNET routers and TCP
253 Description: incorrect data written to files on OSTs
254 Details : In certain high-load conditions incorrect data may be written
255 to files on the OST when using TCP networks.
257 ------------------------------------------------------------------------------
259 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
261 - rework CDEBUG messages rate-limiting mechanism b=10375
262 - add per-socket tunables for socklnd if the kernel is patched b=10327
264 ------------------------------------------------------------------------------
266 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
268 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
269 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
270 - quiet LNET startup LNI message for liblustre b=10128
271 - Better console error messages if 'ip2nets' can't match an IP address
272 - Fixed overflow/use-before-set bugs in linux-time.h
273 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
274 - LNET teardown failed an assertion about the route table being empty
275 - Fixed a crash in LNetEQPoll(<invalid handle>)
276 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
277 - improve debug message for liblustre/Catamount nodes (b=10116)
279 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
280 * Configuration change for the XT3
281 The PTLLND is now used to run Lustre over Portals on the XT3.
282 The configure option(s) --with-cray-portals are no longer
283 used. Rather --with-portals=<path-to-portals-includes> is
284 used to enable building on the XT3. In addition to enable
285 XT3 specific features the option --enable-cray-xt3 must be
288 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
289 * Portals has been removed, replaced by LNET.
290 LNET is new networking infrastructure for Lustre, it includes a
291 reorganized network configuration mode (see the user
292 documentation for full details) as well as support for routing
293 between different network fabrics. Lustre Networking Devices
294 (LNDS) for the supported network fabrics have also been created
295 for this new infrastructure.
297 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
302 Frequency : rare (large Voltaire clusters only)
304 Description: the default number of reserved transmit descriptors was too low
305 for some large clusters
306 Details : As a workaround, the number was increased. A proper fix includes
309 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
314 Frequency : occasional (large-scale events, cluster reboot, network failure)
316 Description: too many error messages on console obscure actual problem and
317 can slow down/panic server, or cause recovery to fail repeatedly
318 Details : enable rate-limiting of console error messages, and some messages
319 that were console errors now only go to the kernel log
321 Severity : enhancement
323 Description: add /proc/sys/portals/catastrophe entry which will report if
324 that node has previously LBUGged
326 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
328 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
330 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
332 - handle error return code in kranal_check_fma_rx() (5915,6054)
334 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
336 - update vibnal (Voltaire IB NAL)
337 - update gmnal (Myrinet NAL), gmnalid
339 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
341 * Landed portals:b_port_step as follows...
343 - removed CFS_DECL_SPIN*
344 just use 'spinlock_t' and initialise with spin_lock_init()
346 - removed CFS_DECL_MUTEX*
347 just use 'struct semaphore' and initialise with init_mutex()
349 - removed CFS_DECL_RWSEM*
350 just use 'struct rw_semaphore' and initialise with init_rwsem()
352 - renamed cfs_sleep_chan -> cfs_waitq
353 cfs_sleep_link -> cfs_waitlink
355 - fixed race in linux version of arch-independent socknal
356 (the ENOMEM/EAGAIN decision).
358 - Didn't fix problems in Darwin version of arch-independent socknal
359 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
361 - removed libcfs types from non-socknal header files (only some types
362 in the header files had been changed; the .c files hadn't been