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: A race in kernel ptllnd between deleting a peer and posting
37 new communications for it could hang communications -
38 manifesting as "Unexpectedly long timeout" messages.
43 Description: Kernel ptllnd lock ordering issue could hang a node.
48 Description: node crash on socket teardown race
51 Frequency : 'lctl peer_list' issued on a mx net
53 Description: Enable lctl's peer_list for MXLND
56 Frequency : after Ptllnd timeouts and portals congestion
58 Description: Credit overflows
59 Details : This was a bug in ptllnd connection establishment. The fix
60 implements better peer stamps to disambiguate connection
61 establishment and ensure both peers enter the credit flow
62 state machine consistently.
67 Description: kptllnd didn't propagate some network errors up to LNET
68 Details : This bug was spotted while investigating 11394. The fix
69 ensures network errors on sends and bulk transfers are
70 propagated to LNET/lustre correctly.
72 Severity : enhancement
74 Description: Fixed console chatter in case of -ETIMEDOUT.
76 Severity : enhancement
78 Description: Added D_NETTRACE for recording network packet history
79 (initially only for ptllnd). Also a separate userspace
80 ptllnd facility to gather history which should really be
81 covered by D_NETTRACE too, if only CDEBUG recorded history in
87 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
88 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
89 callback can occur before a connection has actually been
90 established. This caused an assertion failure previously.
92 Severity : enhancement
94 Description: Multiple instances for o2iblnd
95 Details : Allow multiple instances of o2iblnd to enable networking over
96 multiple HCAs and routing between them.
100 Description: lnet deadlock in router_checker
101 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
102 into BH locks to eliminate potential deadlock caused by
103 ksocknal_data_ready() preempting code holding these locks.
107 Description: Millions of failed socklnd connection attempts cause a very slow FS
108 Details : added a new route flag ksnr_scheduled to distinguish from
109 ksnr_connecting, so that a peer connection request is only turned
110 down for race concerns when an active connection to the same peer
111 is under progress (instead of just being scheduled).
113 ------------------------------------------------------------------------------
115 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
117 * Support for networks:
118 socklnd - kernels up to 2.6.16
119 qswlnd - Qsnet kernel modules 5.20 and later
120 openiblnd - IbGold 1.8.2
122 viblnd - Voltaire ibhost 3.4.5 and later
123 ciblnd - Topspin 3.2.0
124 iiblnd - Infiniserv 3.3 + PathBits patch
125 gmlnd - GM 2.1.22 and later
126 mxlnd - MX 1.2.1 or later
127 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
130 Severity : major on XT3
132 Description: libcfs overwrites /proc/sys/portals
133 Details : libcfs created a symlink from /proc/sys/portals to
134 /proc/sys/lnet for backwards compatibility. This is no
135 longer required and makes the Cray portals /proc variables
140 Description: OFED FMR API change
141 Details : This changes parameter usage to reflect a change in
142 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
143 that FMR support is only used in experimental versions of the
144 o2iblnd - this change does not affect standard usage at all.
146 Severity : enhancement
148 Description: new ko2iblnd module parameter: ib_mtu
149 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
150 HCAs. You can avoid this problem by setting the MTU to 1024
151 using this module parameter.
153 Severity : enhancement
154 Bugzilla : 11118/11620
155 Description: ptllnd small request message buffer alignment fix
156 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
157 Round up small message size on sends in case this option
158 is not supported. 11620 was a defect in the initial
159 implementation which effectively asserted all peers had to be
160 running the correct protocol version which was fixed by always
161 NAK-ing such requests and handling any misalignments they
166 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
167 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
168 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
170 Severity : enhancement
172 Description: Patchless ZC(zero copy) socklnd
173 Details : New protocol for socklnd, socklnd can support zero copy without
174 kernel patch, it's compatible with old socklnd. Checksum is
175 moved from tunables to modparams.
179 Description: When ksocknal_del_peer() is called upon a peer whose
180 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
181 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
184 Frequency : when ptlrpc is under heavy use and runs out of request buffer
186 Description: In lnet_match_blocked_msg(), md can be used without holding a
190 Frequency : very rarely
192 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
193 If connd connects a route which has been closed by
194 ksocknal_shutdown(), ksocknal_create_routes() may create new
195 routes which hold references on the peer, causing shutdown
196 process to wait for peer to disappear forever.
198 Severity : enhancement
200 Description: Dump XT3 portals traces on kptllnd timeout
201 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
202 dump Cray portals debug traces to a file. The kptllnd module
203 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
204 is the basename of the dump file.
207 Frequency : infrequent
209 Description: kernel ptllnd fix bug in connection re-establishment
210 Details : Kernel ptllnd could produce protocol errors e.g. illegal
211 matchbits and/or violate the credit flow protocol when trying
212 to re-establish a connection with a peer after an error or
215 Severity : enhancement
217 Description: Allow /proc/sys/lnet/debug to be set symbolically
218 Details : Allow debug and subsystem debug values to be read/set by name
219 in addition to numerically, for ease of use.
222 Frequency : only in configurations with LNET routers
224 Description: routes automatically marked down and recovered
225 Details : In configurations with LNET routers if a router fails routers
226 now actively try to recover routes that are down, unless they
227 are marked down by an administrator.
229 ------------------------------------------------------------------------------
231 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
234 Frequency : very rarely, in configurations with LNET routers and TCP
236 Description: incorrect data written to files on OSTs
237 Details : In certain high-load conditions incorrect data may be written
238 to files on the OST when using TCP networks.
240 ------------------------------------------------------------------------------
242 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
244 - rework CDEBUG messages rate-limiting mechanism b=10375
245 - add per-socket tunables for socklnd if the kernel is patched b=10327
247 ------------------------------------------------------------------------------
249 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
251 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
252 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
253 - quiet LNET startup LNI message for liblustre b=10128
254 - Better console error messages if 'ip2nets' can't match an IP address
255 - Fixed overflow/use-before-set bugs in linux-time.h
256 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
257 - LNET teardown failed an assertion about the route table being empty
258 - Fixed a crash in LNetEQPoll(<invalid handle>)
259 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
260 - improve debug message for liblustre/Catamount nodes (b=10116)
262 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
263 * Configuration change for the XT3
264 The PTLLND is now used to run Lustre over Portals on the XT3.
265 The configure option(s) --with-cray-portals are no longer
266 used. Rather --with-portals=<path-to-portals-includes> is
267 used to enable building on the XT3. In addition to enable
268 XT3 specific features the option --enable-cray-xt3 must be
271 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
272 * Portals has been removed, replaced by LNET.
273 LNET is new networking infrastructure for Lustre, it includes a
274 reorganized network configuration mode (see the user
275 documentation for full details) as well as support for routing
276 between different network fabrics. Lustre Networking Devices
277 (LNDS) for the supported network fabrics have also been created
278 for this new infrastructure.
280 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
285 Frequency : rare (large Voltaire clusters only)
287 Description: the default number of reserved transmit descriptors was too low
288 for some large clusters
289 Details : As a workaround, the number was increased. A proper fix includes
292 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
297 Frequency : occasional (large-scale events, cluster reboot, network failure)
299 Description: too many error messages on console obscure actual problem and
300 can slow down/panic server, or cause recovery to fail repeatedly
301 Details : enable rate-limiting of console error messages, and some messages
302 that were console errors now only go to the kernel log
304 Severity : enhancement
306 Description: add /proc/sys/portals/catastrophe entry which will report if
307 that node has previously LBUGged
309 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
311 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
313 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
315 - handle error return code in kranal_check_fma_rx() (5915,6054)
317 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
319 - update vibnal (Voltaire IB NAL)
320 - update gmnal (Myrinet NAL), gmnalid
322 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
324 * Landed portals:b_port_step as follows...
326 - removed CFS_DECL_SPIN*
327 just use 'spinlock_t' and initialise with spin_lock_init()
329 - removed CFS_DECL_MUTEX*
330 just use 'struct semaphore' and initialise with init_mutex()
332 - removed CFS_DECL_RWSEM*
333 just use 'struct rw_semaphore' and initialise with init_rwsem()
335 - renamed cfs_sleep_chan -> cfs_waitq
336 cfs_sleep_link -> cfs_waitlink
338 - fixed race in linux version of arch-independent socknal
339 (the ENOMEM/EAGAIN decision).
341 - Didn't fix problems in Darwin version of arch-independent socknal
342 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
344 - removed libcfs types from non-socknal header files (only some types
345 in the header files had been changed; the .c files hadn't been