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: o2iblnd handle early RDMA_CM_EVENT_DISCONNECTED.
20 Details : If the fabric is lossy, an RDMA_CM_EVENT_DISCONNECTED
21 callback can occur before a connection has actually been
22 established. This caused an assertion failure previously.
24 Severity : enhancement
26 Description: Multiple instances for o2iblnd
27 Details : Allow multiple instances of o2iblnd to enable networking over
28 multiple HCAs and routing between them.
32 Description: lnet deadlock in router_checker
33 Details : turned ksnd_connd_lock, ksnd_reaper_lock, and ksock_net_t:ksnd_lock
34 into BH locks to eliminate potential deadlock caused by
35 ksocknal_data_ready() preempting code holding these locks.
39 Description: Millions of failed socklnd connection attempts cause a very slow FS
40 Details : added a new route flag ksnr_scheduled to distinguish from
41 ksnr_connecting, so that a peer connection request is only turned
42 down for race concerns when an active connection to the same peer
43 is under progress (instead of just being scheduled).
45 ------------------------------------------------------------------------------
47 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
49 * Support for networks:
50 socklnd - kernels up to 2.6.16
51 qswlnd - Qsnet kernel modules 5.20 and later
52 openiblnd - IbGold 1.8.2
54 viblnd - Voltaire ibhost 3.4.5 and later
55 ciblnd - Topspin 3.2.0
56 iiblnd - Infiniserv 3.3 + PathBits patch
57 gmlnd - GM 2.1.22 and later
58 mxlnd - MX 1.2.1 or later
59 ptllnd - Portals 3.3 / UNICOS/lc 1.5.x, 2.0.x
62 Severity : major on XT3
64 Description: libcfs overwrites /proc/sys/portals
65 Details : libcfs created a symlink from /proc/sys/portals to
66 /proc/sys/lnet for backwards compatibility. This is no
67 longer required and makes the Cray portals /proc variables
72 Description: OFED FMR API change
73 Details : This changes parameter usage to reflect a change in
74 ib_fmr_pool_map_phys() between OFED 1.0 and OFED 1.1. Note
75 that FMR support is only used in experimental versions of the
76 o2iblnd - this change does not affect standard usage at all.
78 Severity : enhancement
80 Description: new ko2iblnd module parameter: ib_mtu
81 Details : the default IB MTU of 2048 performs badly on 23108 Tavor
82 HCAs. You can avoid this problem by setting the MTU to 1024
83 using this module parameter.
85 Severity : enhancement
86 Bugzilla : 11118/11620
87 Description: ptllnd small request message buffer alignment fix
88 Details : Set the PTL_MD_LOCAL_ALIGN8 option on small message receives.
89 Round up small message size on sends in case this option
90 is not supported. 11620 was a defect in the initial
91 implementation which effectively asserted all peers had to be
92 running the correct protocol version which was fixed by always
93 NAK-ing such requests and handling any misalignments they
98 Description: When kib(nal|lnd)_del_peer() is called upon a peer whose
99 ibp_tx_queue is not empty, kib(nal|lnd)_destroy_peer()'s
100 'LASSERT(list_empty(&peer->ibp_tx_queue))' will fail.
102 Severity : enhancement
104 Description: Patchless ZC(zero copy) socklnd
105 Details : New protocol for socklnd, socklnd can support zero copy without
106 kernel patch, it's compatible with old socklnd. Checksum is
107 moved from tunables to modparams.
111 Description: When ksocknal_del_peer() is called upon a peer whose
112 ksnp_tx_queue is not empty, ksocknal_destroy_peer()'s
113 'LASSERT(list_empty(&peer->ksnp_tx_queue))' will fail.
116 Frequency : when ptlrpc is under heavy use and runs out of request buffer
118 Description: In lnet_match_blocked_msg(), md can be used without holding a
122 Frequency : very rarely
124 Description: If ksocknal_lib_setup_sock() fails, a ref on peer is lost.
125 If connd connects a route which has been closed by
126 ksocknal_shutdown(), ksocknal_create_routes() may create new
127 routes which hold references on the peer, causing shutdown
128 process to wait for peer to disappear forever.
130 Severity : enhancement
132 Description: Dump XT3 portals traces on kptllnd timeout
133 Details : Set the kptllnd module parameter "ptltrace_on_timeout=1" to
134 dump Cray portals debug traces to a file. The kptllnd module
135 parameter "ptltrace_basename", default "/tmp/lnet-ptltrace",
136 is the basename of the dump file.
139 Frequency : infrequent
141 Description: kernel ptllnd fix bug in connection re-establishment
142 Details : Kernel ptllnd could produce protocol errors e.g. illegal
143 matchbits and/or violate the credit flow protocol when trying
144 to re-establish a connection with a peer after an error or
147 Severity : enhancement
149 Description: Allow /proc/sys/lnet/debug to be set symbolically
150 Details : Allow debug and subsystem debug values to be read/set by name
151 in addition to numerically, for ease of use.
154 Frequency : only in configurations with LNET routers
156 Description: routes automatically marked down and recovered
157 Details : In configurations with LNET routers if a router fails routers
158 now actively try to recover routes that are down, unless they
159 are marked down by an administrator.
161 ------------------------------------------------------------------------------
163 2006-07-31 Cluster File Systems, Inc. <info@clusterfs.com>
165 - rework CDEBUG messages rate-limiting mechanism b=10375
166 - add per-socket tunables for socklnd if the kernel is patched b=10327
168 ------------------------------------------------------------------------------
170 2006-02-15 Cluster File Systems, Inc. <info@clusterfs.com>
172 - fix use of portals/lnet pid to avoid dropping RPCs b=10074
173 - iiblnd wasn't mapping all memory, resulting in comms errors b=9776
174 - quiet LNET startup LNI message for liblustre b=10128
175 - Better console error messages if 'ip2nets' can't match an IP address
176 - Fixed overflow/use-before-set bugs in linux-time.h
177 - Fixed ptllnd bug that wasn't initialising rx descriptors completely
178 - LNET teardown failed an assertion about the route table being empty
179 - Fixed a crash in LNetEQPoll(<invalid handle>)
180 - Future protocol compatibility work (b_rls146_lnetprotovrsn)
181 - improve debug message for liblustre/Catamount nodes (b=10116)
183 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
184 * Configuration change for the XT3
185 The PTLLND is now used to run Lustre over Portals on the XT3.
186 The configure option(s) --with-cray-portals are no longer
187 used. Rather --with-portals=<path-to-portals-includes> is
188 used to enable building on the XT3. In addition to enable
189 XT3 specific features the option --enable-cray-xt3 must be
192 2005-10-10 Cluster File Systems, Inc. <info@clusterfs.com>
193 * Portals has been removed, replaced by LNET.
194 LNET is new networking infrastructure for Lustre, it includes a
195 reorganized network configuration mode (see the user
196 documentation for full details) as well as support for routing
197 between different network fabrics. Lustre Networking Devices
198 (LNDS) for the supported network fabrics have also been created
199 for this new infrastructure.
201 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
206 Frequency : rare (large Voltaire clusters only)
208 Description: the default number of reserved transmit descriptors was too low
209 for some large clusters
210 Details : As a workaround, the number was increased. A proper fix includes
213 2005-06-02 Cluster File Systems, Inc. <info@clusterfs.com>
218 Frequency : occasional (large-scale events, cluster reboot, network failure)
220 Description: too many error messages on console obscure actual problem and
221 can slow down/panic server, or cause recovery to fail repeatedly
222 Details : enable rate-limiting of console error messages, and some messages
223 that were console errors now only go to the kernel log
225 Severity : enhancement
227 Description: add /proc/sys/portals/catastrophe entry which will report if
228 that node has previously LBUGged
230 2005-04-06 Cluster File Systems, Inc. <info@clusterfs.com>
232 - update gmnal to use PTL_MTU, fix module refcounting (b=5786)
234 2005-04-04 Cluster File Systems, Inc. <info@clusterfs.com>
236 - handle error return code in kranal_check_fma_rx() (5915,6054)
238 2005-02-04 Cluster File Systems, Inc. <info@clusterfs.com>
240 - update vibnal (Voltaire IB NAL)
241 - update gmnal (Myrinet NAL), gmnalid
243 2005-02-04 Eric Barton <eeb@bartonsoftware.com>
245 * Landed portals:b_port_step as follows...
247 - removed CFS_DECL_SPIN*
248 just use 'spinlock_t' and initialise with spin_lock_init()
250 - removed CFS_DECL_MUTEX*
251 just use 'struct semaphore' and initialise with init_mutex()
253 - removed CFS_DECL_RWSEM*
254 just use 'struct rw_semaphore' and initialise with init_rwsem()
256 - renamed cfs_sleep_chan -> cfs_waitq
257 cfs_sleep_link -> cfs_waitlink
259 - fixed race in linux version of arch-independent socknal
260 (the ENOMEM/EAGAIN decision).
262 - Didn't fix problems in Darwin version of arch-independent socknal
263 (resetting socket callbacks, eager ack hack, ENOMEM/EAGAIN decision)
265 - removed libcfs types from non-socknal header files (only some types
266 in the header files had been changed; the .c files hadn't been