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
18 Description: make panic on lbug configurable
20 ------------------------------------------------------------------------------
22 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
23 * version 1.4.10 / 1.6.0
24 * Support for networks:
25 socklnd - kernels up to 2.6.16
26 qswlnd - Qsnet kernel modules 5.20 and later
27 openiblnd - IbGold 1.8.2
29 viblnd - Voltaire ibhost 3.4.5 and later
30 ciblnd - Topspin 3.2.0
31 iiblnd - Infiniserv 3.3 + PathBits patch
32 gmlnd - GM 2.1.22 and later
33 mxlnd - MX 1.2.1 or later
34 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
39 Description: Ptllnd didn't init kptllnd_data.kptl_idle_txs before it could be
40 possibly accessed in kptllnd_shutdown. Ptllnd should init
41 kptllnd_data.kptl_ptlid2str_lock before calling kptllnd_ptlid2str.
45 Description: gmlnd ignored some transmit errors when finalizing lnet messages.
49 Description: ptllnd logs a piece of incorrect debug info in kptllnd_peer_handle_hello.
53 Description: the_lnet.ln_finalizing was not set when the current thread is
54 about to complete messages. It only affects multi-threaded
60 Description: Changed the default kqswlnd ntxmsg=512
65 Description: Assertion failure in kernel ptllnd caused by posting passive
66 bulk buffers before connection establishment complete.
71 Description: A race in kernel ptllnd between deleting a peer and posting
72 new communications for it could hang communications -
73 manifesting as "Unexpectedly long timeout" messages.
78 Description: Kernel ptllnd lock ordering issue could hang a node.
83 Description: node crash on socket teardown race
86 Frequency : 'lctl peer_list' issued on a mx net
88 Description: Enable lctl's peer_list for MXLND
91 Frequency : after Ptllnd timeouts and portals congestion
93 Description: Credit overflows
94 Details : This was a bug in ptllnd connection establishment. The fix
95 implements better peer stamps to disambiguate connection
96 establishment and ensure both peers enter the credit flow
97 state machine consistently.
102 Description: kptllnd didn't propagate some network errors up to LNET
103 Details : This bug was spotted while investigating 11394. The fix
104 ensures network errors on sends and bulk transfers are
105 propagated to LNET/lustre correctly.
107 Severity : enhancement
109 Description: Fixed console chatter in case of -ETIMEDOUT.
111 Severity : enhancement
113 Description: Added D_NETTRACE for recording network packet history
114 (initially only for ptllnd). Also a separate userspace
115 ptllnd facility to gather history which should really be
116 covered by D_NETTRACE too, if only CDEBUG recorded history in
122 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
123 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
124 callback can occur before a connection has actually been
125 established. This caused an assertion failure previously.
127 Severity : enhancement
129 Description: Multiple instances for o2iblnd
130 Details : Allow multiple instances of o2iblnd to enable networking over
131 multiple HCAs and routing between them.
135 Description: lnet deadlock in router_checker
136 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
137 into BH locks to eliminate potential deadlock caused by
138 ksocknal_data_ready() preempting code holding these locks.
142 Description: Millions of failed socklnd connection attempts cause a very slow FS
143 Details : added a new route flag ksnr_scheduled to distinguish from
144 ksnr_connecting, so that a peer connection request is only turned
145 down for race concerns when an active connection to the same peer
146 is under progress (instead of just being scheduled).
148 ------------------------------------------------------------------------------
150 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
152 * Support for networks:
153 socklnd - kernels up to 2.6.16
154 qswlnd - Qsnet kernel modules 5.20 and later
155 openiblnd - IbGold 1.8.2
157 viblnd - Voltaire ibhost 3.4.5 and later
158 ciblnd - Topspin 3.2.0
159 iiblnd - Infiniserv 3.3 + PathBits patch
160 gmlnd - GM 2.1.22 and later
161 mxlnd - MX 1.2.1 or later
162 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
165 Severity : major on XT3
167 Description: libcfs overwrites /proc/sys/portals
168 Details : libcfs created a symlink from /proc/sys/portals to
169 /proc/sys/lnet for backwards compatibility. This is no
170 longer required and makes the Cray portals /proc variables
175 Description: OFED FMR API change
176 Details : This changes parameter usage to reflect a change in
177 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
178 that FMR support is only used in experimental versions of the
179 o2iblnd - this change does not affect standard usage at all.
181 Severity : enhancement
183 Description: new ko2iblnd module parameter: ib_mtu
184 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
185 HCAs. You can avoid this problem by setting the MTU to 1024
186 using this module parameter.
188 Severity : enhancement
189 Bugzilla : 11118/11620
190 Description: ptllnd small request message buffer alignment fix
191 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
192 Round up small message size on sends in case this option
193 is not supported. 11620 was a defect in the initial
194 implementation which effectively asserted all peers had to be
195 running the correct protocol version which was fixed by always
196 NAK-ing such requests and handling any misalignments they
201 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
202 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
203 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
205 Severity : enhancement
207 Description: Patchless ZC(zero copy) socklnd
208 Details : New protocol for socklnd, socklnd can support zero copy without
209 kernel patch, it's compatible with old socklnd. Checksum is
210 moved from tunables to modparams.
214 Description: When ksocknal_del_peer() is called upon a peer whose
215 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
216 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
219 Frequency : when ptlrpc is under heavy use and runs out of request buffer
221 Description: In lnet_match_blocked_msg(), md can be used without holding a
225 Frequency : very rarely
227 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
228 If connd connects a route which has been closed by
229 ksocknal_shutdown(), ksocknal_create_routes() may create new
230 routes which hold references on the peer, causing shutdown
231 process to wait for peer to disappear forever.
233 Severity : enhancement
235 Description: Dump XT3 portals traces on kptllnd timeout
236 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
237 dump Cray portals debug traces to a file. The kptllnd module
238 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
239 is the basename of the dump file.
242 Frequency : infrequent
244 Description: kernel ptllnd fix bug in connection re-establishment
245 Details : Kernel ptllnd could produce protocol errors e.g. illegal
246 matchbits and/or violate the credit flow protocol when trying
247 to re-establish a connection with a peer after an error or
250 Severity : enhancement
252 Description: Allow /proc/sys/lnet/debug to be set symbolically
253 Details : Allow debug and subsystem debug values to be read/set by name
254 in addition to numerically, for ease of use.
257 Frequency : only in configurations with LNET routers
259 Description: routes automatically marked down and recovered
260 Details : In configurations with LNET routers if a router fails routers
261 now actively try to recover routes that are down, unless they
262 are marked down by an administrator.
264 ------------------------------------------------------------------------------
266 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
269 Frequency : very rarely, in configurations with LNET routers and TCP
271 Description: incorrect data written to files on OSTs
272 Details : In certain high-load conditions incorrect data may be written
273 to files on the OST when using TCP networks.
275 ------------------------------------------------------------------------------
277 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
279 - rework CDEBUG messages rate-limiting mechanism b=10375
280 - add per-socket tunables for socklnd if the kernel is patched b=10327
282 ------------------------------------------------------------------------------
284 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
286 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
287 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
288 - quiet LNET startup LNI message for liblustre b=10128
289 - Better console error messages if 'ip2nets' can't match an IP address
290 - Fixed overflow/use-before-set bugs in linux-time.h
291 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
292 - LNET teardown failed an assertion about the route table being empty
293 - Fixed a crash in LNetEQPoll(<invalid handle>)
294 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
295 - improve debug message for liblustre/Catamount nodes (b=10116)
297 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
298 * Configuration change for the XT3
299 The PTLLND is now used to run Lustre over Portals on the XT3.
300 The configure option(s) --with-cray-portals are no longer
301 used. Rather --with-portals=<path-to-portals-includes> is
302 used to enable building on the XT3. In addition to enable
303 XT3 specific features the option --enable-cray-xt3 must be
306 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
307 * Portals has been removed, replaced by LNET.
308 LNET is new networking infrastructure for Lustre, it includes a
309 reorganized network configuration mode (see the user
310 documentation for full details) as well as support for routing
311 between different network fabrics. Lustre Networking Devices
312 (LNDS) for the supported network fabrics have also been created
313 for this new infrastructure.
315 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
320 Frequency : rare (large Voltaire clusters only)
322 Description: the default number of reserved transmit descriptors was too low
323 for some large clusters
324 Details : As a workaround, the number was increased. A proper fix includes
327 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
332 Frequency : occasional (large-scale events, cluster reboot, network failure)
334 Description: too many error messages on console obscure actual problem and
335 can slow down/panic server, or cause recovery to fail repeatedly
336 Details : enable rate-limiting of console error messages, and some messages
337 that were console errors now only go to the kernel log
339 Severity : enhancement
341 Description: add /proc/sys/portals/catastrophe entry which will report if
342 that node has previously LBUGged
344 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
346 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
348 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
350 - handle error return code in kranal_check_fma_rx() (5915,6054)
352 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
354 - update vibnal (Voltaire IB NAL)
355 - update gmnal (Myrinet NAL), gmnalid
357 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
359 * Landed portals:b_port_step as follows...
361 - removed CFS_DECL_SPIN*
362 just use 'spinlock_t' and initialise with spin_lock_init()
364 - removed CFS_DECL_MUTEX*
365 just use 'struct semaphore' and initialise with init_mutex()
367 - removed CFS_DECL_RWSEM*
368 just use 'struct rw_semaphore' and initialise with init_rwsem()
370 - renamed cfs_sleep_chan -> cfs_waitq
371 cfs_sleep_link -> cfs_waitlink
373 - fixed race in linux version of arch-independent socknal
374 (the ENOMEM/EAGAIN decision).
376 - Didn't fix problems in Darwin version of arch-independent socknal
377 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
379 - removed libcfs types from non-socknal header files (only some types
380 in the header files had been changed; the .c files hadn't been