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
36 Description: Changed the default kqswlnd ntxmsg=512
41 Description: Assertion failure in kernel ptllnd caused by posting passive
42 bulk buffers before connection establishment complete.
47 Description: A race in kernel ptllnd between deleting a peer and posting
48 new communications for it could hang communications -
49 manifesting as "Unexpectedly long timeout" messages.
54 Description: Kernel ptllnd lock ordering issue could hang a node.
59 Description: node crash on socket teardown race
62 Frequency : 'lctl peer_list' issued on a mx net
64 Description: Enable lctl's peer_list for MXLND
67 Frequency : after Ptllnd timeouts and portals congestion
69 Description: Credit overflows
70 Details : This was a bug in ptllnd connection establishment. The fix
71 implements better peer stamps to disambiguate connection
72 establishment and ensure both peers enter the credit flow
73 state machine consistently.
78 Description: kptllnd didn't propagate some network errors up to LNET
79 Details : This bug was spotted while investigating 11394. The fix
80 ensures network errors on sends and bulk transfers are
81 propagated to LNET/lustre correctly.
83 Severity : enhancement
85 Description: Fixed console chatter in case of -ETIMEDOUT.
87 Severity : enhancement
89 Description: Added D_NETTRACE for recording network packet history
90 (initially only for ptllnd). Also a separate userspace
91 ptllnd facility to gather history which should really be
92 covered by D_NETTRACE too, if only CDEBUG recorded history in
98 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
99 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
100 callback can occur before a connection has actually been
101 established. This caused an assertion failure previously.
103 Severity : enhancement
105 Description: Multiple instances for o2iblnd
106 Details : Allow multiple instances of o2iblnd to enable networking over
107 multiple HCAs and routing between them.
111 Description: lnet deadlock in router_checker
112 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
113 into BH locks to eliminate potential deadlock caused by
114 ksocknal_data_ready() preempting code holding these locks.
118 Description: Millions of failed socklnd connection attempts cause a very slow FS
119 Details : added a new route flag ksnr_scheduled to distinguish from
120 ksnr_connecting, so that a peer connection request is only turned
121 down for race concerns when an active connection to the same peer
122 is under progress (instead of just being scheduled).
124 ------------------------------------------------------------------------------
126 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
128 * Support for networks:
129 socklnd - kernels up to 2.6.16
130 qswlnd - Qsnet kernel modules 5.20 and later
131 openiblnd - IbGold 1.8.2
133 viblnd - Voltaire ibhost 3.4.5 and later
134 ciblnd - Topspin 3.2.0
135 iiblnd - Infiniserv 3.3 + PathBits patch
136 gmlnd - GM 2.1.22 and later
137 mxlnd - MX 1.2.1 or later
138 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
141 Severity : major on XT3
143 Description: libcfs overwrites /proc/sys/portals
144 Details : libcfs created a symlink from /proc/sys/portals to
145 /proc/sys/lnet for backwards compatibility. This is no
146 longer required and makes the Cray portals /proc variables
151 Description: OFED FMR API change
152 Details : This changes parameter usage to reflect a change in
153 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
154 that FMR support is only used in experimental versions of the
155 o2iblnd - this change does not affect standard usage at all.
157 Severity : enhancement
159 Description: new ko2iblnd module parameter: ib_mtu
160 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
161 HCAs. You can avoid this problem by setting the MTU to 1024
162 using this module parameter.
164 Severity : enhancement
165 Bugzilla : 11118/11620
166 Description: ptllnd small request message buffer alignment fix
167 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
168 Round up small message size on sends in case this option
169 is not supported. 11620 was a defect in the initial
170 implementation which effectively asserted all peers had to be
171 running the correct protocol version which was fixed by always
172 NAK-ing such requests and handling any misalignments they
177 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
178 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
179 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
181 Severity : enhancement
183 Description: Patchless ZC(zero copy) socklnd
184 Details : New protocol for socklnd, socklnd can support zero copy without
185 kernel patch, it's compatible with old socklnd. Checksum is
186 moved from tunables to modparams.
190 Description: When ksocknal_del_peer() is called upon a peer whose
191 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
192 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
195 Frequency : when ptlrpc is under heavy use and runs out of request buffer
197 Description: In lnet_match_blocked_msg(), md can be used without holding a
201 Frequency : very rarely
203 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
204 If connd connects a route which has been closed by
205 ksocknal_shutdown(), ksocknal_create_routes() may create new
206 routes which hold references on the peer, causing shutdown
207 process to wait for peer to disappear forever.
209 Severity : enhancement
211 Description: Dump XT3 portals traces on kptllnd timeout
212 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
213 dump Cray portals debug traces to a file. The kptllnd module
214 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
215 is the basename of the dump file.
218 Frequency : infrequent
220 Description: kernel ptllnd fix bug in connection re-establishment
221 Details : Kernel ptllnd could produce protocol errors e.g. illegal
222 matchbits and/or violate the credit flow protocol when trying
223 to re-establish a connection with a peer after an error or
226 Severity : enhancement
228 Description: Allow /proc/sys/lnet/debug to be set symbolically
229 Details : Allow debug and subsystem debug values to be read/set by name
230 in addition to numerically, for ease of use.
233 Frequency : only in configurations with LNET routers
235 Description: routes automatically marked down and recovered
236 Details : In configurations with LNET routers if a router fails routers
237 now actively try to recover routes that are down, unless they
238 are marked down by an administrator.
240 ------------------------------------------------------------------------------
242 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
245 Frequency : very rarely, in configurations with LNET routers and TCP
247 Description: incorrect data written to files on OSTs
248 Details : In certain high-load conditions incorrect data may be written
249 to files on the OST when using TCP networks.
251 ------------------------------------------------------------------------------
253 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
255 - rework CDEBUG messages rate-limiting mechanism b=10375
256 - add per-socket tunables for socklnd if the kernel is patched b=10327
258 ------------------------------------------------------------------------------
260 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
262 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
263 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
264 - quiet LNET startup LNI message for liblustre b=10128
265 - Better console error messages if 'ip2nets' can't match an IP address
266 - Fixed overflow/use-before-set bugs in linux-time.h
267 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
268 - LNET teardown failed an assertion about the route table being empty
269 - Fixed a crash in LNetEQPoll(<invalid handle>)
270 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
271 - improve debug message for liblustre/Catamount nodes (b=10116)
273 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
274 * Configuration change for the XT3
275 The PTLLND is now used to run Lustre over Portals on the XT3.
276 The configure option(s) --with-cray-portals are no longer
277 used. Rather --with-portals=<path-to-portals-includes> is
278 used to enable building on the XT3. In addition to enable
279 XT3 specific features the option --enable-cray-xt3 must be
282 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
283 * Portals has been removed, replaced by LNET.
284 LNET is new networking infrastructure for Lustre, it includes a
285 reorganized network configuration mode (see the user
286 documentation for full details) as well as support for routing
287 between different network fabrics. Lustre Networking Devices
288 (LNDS) for the supported network fabrics have also been created
289 for this new infrastructure.
291 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
296 Frequency : rare (large Voltaire clusters only)
298 Description: the default number of reserved transmit descriptors was too low
299 for some large clusters
300 Details : As a workaround, the number was increased. A proper fix includes
303 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
308 Frequency : occasional (large-scale events, cluster reboot, network failure)
310 Description: too many error messages on console obscure actual problem and
311 can slow down/panic server, or cause recovery to fail repeatedly
312 Details : enable rate-limiting of console error messages, and some messages
313 that were console errors now only go to the kernel log
315 Severity : enhancement
317 Description: add /proc/sys/portals/catastrophe entry which will report if
318 that node has previously LBUGged
320 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
322 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
324 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
326 - handle error return code in kranal_check_fma_rx() (5915,6054)
328 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
330 - update vibnal (Voltaire IB NAL)
331 - update gmnal (Myrinet NAL), gmnalid
333 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
335 * Landed portals:b_port_step as follows...
337 - removed CFS_DECL_SPIN*
338 just use 'spinlock_t' and initialise with spin_lock_init()
340 - removed CFS_DECL_MUTEX*
341 just use 'struct semaphore' and initialise with init_mutex()
343 - removed CFS_DECL_RWSEM*
344 just use 'struct rw_semaphore' and initialise with init_rwsem()
346 - renamed cfs_sleep_chan -> cfs_waitq
347 cfs_sleep_link -> cfs_waitlink
349 - fixed race in linux version of arch-independent socknal
350 (the ENOMEM/EAGAIN decision).
352 - Didn't fix problems in Darwin version of arch-independent socknal
353 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
355 - removed libcfs types from non-socknal header files (only some types
356 in the header files had been changed; the .c files hadn't been