1 TBD 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
19 Description: kptllnd didn't propagate some network errors up to LNET
20 Details : This bug was spotted while investigating 11394. The fix
21 ensures network errors on sends and bulk transfers are
22 propagated to LNET/lustre correctly.
24 Severity : enhancement
26 Description: Fixed console chatter in case of -ETIMEDOUT.
28 Severity : enhancement
30 Description: Added D_NETTRACE for recording network packet history
31 (initially only for ptllnd). Also a separate userspace
32 ptllnd facility to gather history which should really be
33 covered by D_NETTRACE too, if only CDEBUG recorded history in
39 Description: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
40 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
41 callback can occur before a connection has actually been
42 established. This caused an assertion failure previously.
44 Severity : enhancement
46 Description: Multiple instances for o2iblnd
47 Details : Allow multiple instances of o2iblnd to enable networking over
48 multiple HCAs and routing between them.
52 Description: lnet deadlock in router_checker
53 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
54 into BH locks to eliminate potential deadlock caused by
55 ksocknal_data_ready() preempting code holding these locks.
59 Description: Millions of failed socklnd connection attempts cause a very slow FS
60 Details : added a new route flag ksnr_scheduled to distinguish from
61 ksnr_connecting, so that a peer connection request is only turned
62 down for race concerns when an active connection to the same peer
63 is under progress (instead of just being scheduled).
65 ------------------------------------------------------------------------------
67 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
69 * Support for networks:
70 socklnd - kernels up to 2.6.16
71 qswlnd - Qsnet kernel modules 5.20 and later
72 openiblnd - IbGold 1.8.2
74 viblnd - Voltaire ibhost 3.4.5 and later
75 ciblnd - Topspin 3.2.0
76 iiblnd - Infiniserv 3.3 + PathBits patch
77 gmlnd - GM 2.1.22 and later
78 mxlnd - MX 1.2.1 or later
79 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
82 Severity : major on XT3
84 Description: libcfs overwrites /proc/sys/portals
85 Details : libcfs created a symlink from /proc/sys/portals to
86 /proc/sys/lnet for backwards compatibility. This is no
87 longer required and makes the Cray portals /proc variables
92 Description: OFED FMR API change
93 Details : This changes parameter usage to reflect a change in
94 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
95 that FMR support is only used in experimental versions of the
96 o2iblnd - this change does not affect standard usage at all.
98 Severity : enhancement
100 Description: new ko2iblnd module parameter: ib_mtu
101 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
102 HCAs. You can avoid this problem by setting the MTU to 1024
103 using this module parameter.
105 Severity : enhancement
106 Bugzilla : 11118/11620
107 Description: ptllnd small request message buffer alignment fix
108 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
109 Round up small message size on sends in case this option
110 is not supported. 11620 was a defect in the initial
111 implementation which effectively asserted all peers had to be
112 running the correct protocol version which was fixed by always
113 NAK-ing such requests and handling any misalignments they
118 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
119 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
120 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
122 Severity : enhancement
124 Description: Patchless ZC(zero copy) socklnd
125 Details : New protocol for socklnd, socklnd can support zero copy without
126 kernel patch, it's compatible with old socklnd. Checksum is
127 moved from tunables to modparams.
131 Description: When ksocknal_del_peer() is called upon a peer whose
132 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
133 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
136 Frequency : when ptlrpc is under heavy use and runs out of request buffer
138 Description: In lnet_match_blocked_msg(), md can be used without holding a
142 Frequency : very rarely
144 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
145 If connd connects a route which has been closed by
146 ksocknal_shutdown(), ksocknal_create_routes() may create new
147 routes which hold references on the peer, causing shutdown
148 process to wait for peer to disappear forever.
150 Severity : enhancement
152 Description: Dump XT3 portals traces on kptllnd timeout
153 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
154 dump Cray portals debug traces to a file. The kptllnd module
155 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
156 is the basename of the dump file.
159 Frequency : infrequent
161 Description: kernel ptllnd fix bug in connection re-establishment
162 Details : Kernel ptllnd could produce protocol errors e.g. illegal
163 matchbits and/or violate the credit flow protocol when trying
164 to re-establish a connection with a peer after an error or
167 Severity : enhancement
169 Description: Allow /proc/sys/lnet/debug to be set symbolically
170 Details : Allow debug and subsystem debug values to be read/set by name
171 in addition to numerically, for ease of use.
174 Frequency : only in configurations with LNET routers
176 Description: routes automatically marked down and recovered
177 Details : In configurations with LNET routers if a router fails routers
178 now actively try to recover routes that are down, unless they
179 are marked down by an administrator.
181 ------------------------------------------------------------------------------
183 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
186 Frequency : very rarely, in configurations with LNET routers and TCP
188 Description: incorrect data written to files on OSTs
189 Details : In certain high-load conditions incorrect data may be written
190 to files on the OST when using TCP networks.
192 ------------------------------------------------------------------------------
194 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
196 - rework CDEBUG messages rate-limiting mechanism b=10375
197 - add per-socket tunables for socklnd if the kernel is patched b=10327
199 ------------------------------------------------------------------------------
201 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
203 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
204 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
205 - quiet LNET startup LNI message for liblustre b=10128
206 - Better console error messages if 'ip2nets' can't match an IP address
207 - Fixed overflow/use-before-set bugs in linux-time.h
208 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
209 - LNET teardown failed an assertion about the route table being empty
210 - Fixed a crash in LNetEQPoll(<invalid handle>)
211 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
212 - improve debug message for liblustre/Catamount nodes (b=10116)
214 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
215 * Configuration change for the XT3
216 The PTLLND is now used to run Lustre over Portals on the XT3.
217 The configure option(s) --with-cray-portals are no longer
218 used. Rather --with-portals=<path-to-portals-includes> is
219 used to enable building on the XT3. In addition to enable
220 XT3 specific features the option --enable-cray-xt3 must be
223 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
224 * Portals has been removed, replaced by LNET.
225 LNET is new networking infrastructure for Lustre, it includes a
226 reorganized network configuration mode (see the user
227 documentation for full details) as well as support for routing
228 between different network fabrics. Lustre Networking Devices
229 (LNDS) for the supported network fabrics have also been created
230 for this new infrastructure.
232 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
237 Frequency : rare (large Voltaire clusters only)
239 Description: the default number of reserved transmit descriptors was too low
240 for some large clusters
241 Details : As a workaround, the number was increased. A proper fix includes
244 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
249 Frequency : occasional (large-scale events, cluster reboot, network failure)
251 Description: too many error messages on console obscure actual problem and
252 can slow down/panic server, or cause recovery to fail repeatedly
253 Details : enable rate-limiting of console error messages, and some messages
254 that were console errors now only go to the kernel log
256 Severity : enhancement
258 Description: add /proc/sys/portals/catastrophe entry which will report if
259 that node has previously LBUGged
261 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
263 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
265 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
267 - handle error return code in kranal_check_fma_rx() (5915,6054)
269 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
271 - update vibnal (Voltaire IB NAL)
272 - update gmnal (Myrinet NAL), gmnalid
274 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
276 * Landed portals:b_port_step as follows...
278 - removed CFS_DECL_SPIN*
279 just use 'spinlock_t' and initialise with spin_lock_init()
281 - removed CFS_DECL_MUTEX*
282 just use 'struct semaphore' and initialise with init_mutex()
284 - removed CFS_DECL_RWSEM*
285 just use 'struct rw_semaphore' and initialise with init_rwsem()
287 - renamed cfs_sleep_chan -> cfs_waitq
288 cfs_sleep_link -> cfs_waitlink
290 - fixed race in linux version of arch-independent socknal
291 (the ENOMEM/EAGAIN decision).
293 - Didn't fix problems in Darwin version of arch-independent socknal
294 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
296 - removed libcfs types from non-socknal header files (only some types
297 in the header files had been changed; the .c files hadn't been