1 2007-04-23 Cluster File Systems, Inc. <info@clusterfs.com>
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 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
17 * version 1.4.10 / 1.6.0
18 * Support for networks:
19 socklnd - kernels up to 2.6.16
20 qswlnd - Qsnet kernel modules 5.20 and later
21 openiblnd - IbGold 1.8.2
23 viblnd - Voltaire ibhost 3.4.5 and later
24 ciblnd - Topspin 3.2.0
25 iiblnd - Infiniserv 3.3 + PathBits patch
26 gmlnd - GM 2.1.22 and later
27 mxlnd - MX 1.2.1 or later
28 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
32 Frequency : 'lctl peer_list' issued on a mx net
34 Description: Enable lctl's peer_list for MXLND
37 Frequency : after Ptllnd timeouts and portals congestion
39 Description: Credit overflows
40 Details : This was a bug in ptllnd connection establishment. The fix
41 implements better peer stamps to disambiguate connection
42 establishment and ensure both peers enter the credit flow
43 state machine consistently.
48 Description: kptllnd didn't propagate some network errors up to LNET
49 Details : This bug was spotted while investigating 11394. The fix
50 ensures network errors on sends and bulk transfers are
51 propagated to LNET/lustre correctly.
53 Severity : enhancement
55 Description: Fixed console chatter in case of -ETIMEDOUT.
57 Severity : enhancement
59 Description: Added D_NETTRACE for recording network packet history
60 (initially only for ptllnd). Also a separate userspace
61 ptllnd facility to gather history which should really be
62 covered by D_NETTRACE too, if only CDEBUG recorded history in
68 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
69 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
70 callback can occur before a connection has actually been
71 established. This caused an assertion failure previously.
73 Severity : enhancement
75 Description: Multiple instances for o2iblnd
76 Details : Allow multiple instances of o2iblnd to enable networking over
77 multiple HCAs and routing between them.
81 Description: lnet deadlock in router_checker
82 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
83 into BH locks to eliminate potential deadlock caused by
84 ksocknal_data_ready() preempting code holding these locks.
88 Description: Millions of failed socklnd connection attempts cause a very slow FS
89 Details : added a new route flag ksnr_scheduled to distinguish from
90 ksnr_connecting, so that a peer connection request is only turned
91 down for race concerns when an active connection to the same peer
92 is under progress (instead of just being scheduled).
94 ------------------------------------------------------------------------------
96 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
98 * Support for networks:
99 socklnd - kernels up to 2.6.16
100 qswlnd - Qsnet kernel modules 5.20 and later
101 openiblnd - IbGold 1.8.2
103 viblnd - Voltaire ibhost 3.4.5 and later
104 ciblnd - Topspin 3.2.0
105 iiblnd - Infiniserv 3.3 + PathBits patch
106 gmlnd - GM 2.1.22 and later
107 mxlnd - MX 1.2.1 or later
108 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
111 Severity : major on XT3
113 Description: libcfs overwrites /proc/sys/portals
114 Details : libcfs created a symlink from /proc/sys/portals to
115 /proc/sys/lnet for backwards compatibility. This is no
116 longer required and makes the Cray portals /proc variables
121 Description: OFED FMR API change
122 Details : This changes parameter usage to reflect a change in
123 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
124 that FMR support is only used in experimental versions of the
125 o2iblnd - this change does not affect standard usage at all.
127 Severity : enhancement
129 Description: new ko2iblnd module parameter: ib_mtu
130 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
131 HCAs. You can avoid this problem by setting the MTU to 1024
132 using this module parameter.
134 Severity : enhancement
135 Bugzilla : 11118/11620
136 Description: ptllnd small request message buffer alignment fix
137 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
138 Round up small message size on sends in case this option
139 is not supported. 11620 was a defect in the initial
140 implementation which effectively asserted all peers had to be
141 running the correct protocol version which was fixed by always
142 NAK-ing such requests and handling any misalignments they
147 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
148 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
149 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
151 Severity : enhancement
153 Description: Patchless ZC(zero copy) socklnd
154 Details : New protocol for socklnd, socklnd can support zero copy without
155 kernel patch, it's compatible with old socklnd. Checksum is
156 moved from tunables to modparams.
160 Description: When ksocknal_del_peer() is called upon a peer whose
161 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
162 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
165 Frequency : when ptlrpc is under heavy use and runs out of request buffer
167 Description: In lnet_match_blocked_msg(), md can be used without holding a
171 Frequency : very rarely
173 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
174 If connd connects a route which has been closed by
175 ksocknal_shutdown(), ksocknal_create_routes() may create new
176 routes which hold references on the peer, causing shutdown
177 process to wait for peer to disappear forever.
179 Severity : enhancement
181 Description: Dump XT3 portals traces on kptllnd timeout
182 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
183 dump Cray portals debug traces to a file. The kptllnd module
184 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
185 is the basename of the dump file.
188 Frequency : infrequent
190 Description: kernel ptllnd fix bug in connection re-establishment
191 Details : Kernel ptllnd could produce protocol errors e.g. illegal
192 matchbits and/or violate the credit flow protocol when trying
193 to re-establish a connection with a peer after an error or
196 Severity : enhancement
198 Description: Allow /proc/sys/lnet/debug to be set symbolically
199 Details : Allow debug and subsystem debug values to be read/set by name
200 in addition to numerically, for ease of use.
203 Frequency : only in configurations with LNET routers
205 Description: routes automatically marked down and recovered
206 Details : In configurations with LNET routers if a router fails routers
207 now actively try to recover routes that are down, unless they
208 are marked down by an administrator.
210 ------------------------------------------------------------------------------
212 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
215 Frequency : very rarely, in configurations with LNET routers and TCP
217 Description: incorrect data written to files on OSTs
218 Details : In certain high-load conditions incorrect data may be written
219 to files on the OST when using TCP networks.
221 ------------------------------------------------------------------------------
223 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
225 - rework CDEBUG messages rate-limiting mechanism b=10375
226 - add per-socket tunables for socklnd if the kernel is patched b=10327
228 ------------------------------------------------------------------------------
230 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
232 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
233 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
234 - quiet LNET startup LNI message for liblustre b=10128
235 - Better console error messages if 'ip2nets' can't match an IP address
236 - Fixed overflow/use-before-set bugs in linux-time.h
237 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
238 - LNET teardown failed an assertion about the route table being empty
239 - Fixed a crash in LNetEQPoll(<invalid handle>)
240 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
241 - improve debug message for liblustre/Catamount nodes (b=10116)
243 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
244 * Configuration change for the XT3
245 The PTLLND is now used to run Lustre over Portals on the XT3.
246 The configure option(s) --with-cray-portals are no longer
247 used. Rather --with-portals=<path-to-portals-includes> is
248 used to enable building on the XT3. In addition to enable
249 XT3 specific features the option --enable-cray-xt3 must be
252 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
253 * Portals has been removed, replaced by LNET.
254 LNET is new networking infrastructure for Lustre, it includes a
255 reorganized network configuration mode (see the user
256 documentation for full details) as well as support for routing
257 between different network fabrics. Lustre Networking Devices
258 (LNDS) for the supported network fabrics have also been created
259 for this new infrastructure.
261 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
266 Frequency : rare (large Voltaire clusters only)
268 Description: the default number of reserved transmit descriptors was too low
269 for some large clusters
270 Details : As a workaround, the number was increased. A proper fix includes
273 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
278 Frequency : occasional (large-scale events, cluster reboot, network failure)
280 Description: too many error messages on console obscure actual problem and
281 can slow down/panic server, or cause recovery to fail repeatedly
282 Details : enable rate-limiting of console error messages, and some messages
283 that were console errors now only go to the kernel log
285 Severity : enhancement
287 Description: add /proc/sys/portals/catastrophe entry which will report if
288 that node has previously LBUGged
290 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
292 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
294 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
296 - handle error return code in kranal_check_fma_rx() (5915,6054)
298 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
300 - update vibnal (Voltaire IB NAL)
301 - update gmnal (Myrinet NAL), gmnalid
303 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
305 * Landed portals:b_port_step as follows...
307 - removed CFS_DECL_SPIN*
308 just use 'spinlock_t' and initialise with spin_lock_init()
310 - removed CFS_DECL_MUTEX*
311 just use 'struct semaphore' and initialise with init_mutex()
313 - removed CFS_DECL_RWSEM*
314 just use 'struct rw_semaphore' and initialise with init_rwsem()
316 - renamed cfs_sleep_chan -> cfs_waitq
317 cfs_sleep_link -> cfs_waitlink
319 - fixed race in linux version of arch-independent socknal
320 (the ENOMEM/EAGAIN decision).
322 - Didn't fix problems in Darwin version of arch-independent socknal
323 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
325 - removed libcfs types from non-socknal header files (only some types
326 in the header files had been changed; the .c files hadn't been