Whamcloud - gitweb
eeb [Tue, 21 Jun 2005 11:48:33 +0000 (11:48 +0000)]
* debugged single acceptor
* fixed PtlGetId() to skip the loopback NI
eeb [Mon, 20 Jun 2005 20:23:42 +0000 (20:23 +0000)]
* first cut single acceptor for SOCKNAL, RANAL, OPENIBNAL, with
backwards compatibility.
eeb [Fri, 17 Jun 2005 09:29:28 +0000 (09:29 +0000)]
* Updated vibnal from b1_4 to use FMR
eeb [Sat, 11 Jun 2005 13:15:26 +0000 (13:15 +0000)]
* vibnal: added module parameters for some low-level IB tunables (bug #6443)
* vibnal: changed some default tunables (bug #6474)
eeb [Wed, 8 Jun 2005 19:04:22 +0000 (19:04 +0000)]
* noticed difference between userspace and kernel x86_64 __u64 etc typdefs
and adjusted LPU64 etc in <libcfs/linux/kp30.h> appropriately
* libcfs_pause() -> inline cfs_pause() in <libcfs/linux/linux-prim.h> and
made more use of it
* cleaned up NAL loading/instantiated console messages
* fixed reconnect_interval in connection NALs, where I'd left out * HZ
* made ???_npeers atomic again; using ???_global lock round the decrement in
???_destroy_peer() could deadlock.
* fixed uninitialised list in vib:kibnal_peer_connect_failed()
* changed to "nicer" format for /proc/sys/portals/routes
* more x86_64 testing
eeb [Tue, 7 Jun 2005 23:18:56 +0000 (23:18 +0000)]
* Got rid of portals/router; getting it to build in portals/portals using
symlinks didn't work on 2.6 and router.c has changed enough for the diffs
not to make sense any more (and life is too short).
* reworked how reconnect timeouts work in all connection based NALs, to close
out a bug where reconnection could be delayed until jiffies wraps.
* tested vibnal with a single echo client + server @ LLNL on 2.6, x86_64
* make configure-on-demand rather than configure-on-load the default; modprobe
in 2.6 does locking and deadlocks if the module it is loading tries to load
more modules in its init sequence.
* removed some unused vars etc caught by better warnings when building for
2.6/x86_64
eeb [Mon, 6 Jun 2005 21:22:25 +0000 (21:22 +0000)]
* brought openibnal into the newconfig fold
* moved socket address query function out of NALs and into libcfs
eeb [Fri, 3 Jun 2005 11:20:52 +0000 (11:20 +0000)]
* merged 6361 vibnal fixes (IBNAL_OUS_DST_RD = 1, HCA names)
* vibnal to use HCA name as the interface name
- add hca and IPoIB interface name prefixes to tunables
- use 1:1 HCA number and IPoIB interface number to determine IP address
of given HCA
eeb [Fri, 3 Jun 2005 09:03:44 +0000 (09:03 +0000)]
* merged in 5858 fix (vibnal arp retries)
eeb [Thu, 2 Jun 2005 17:57:59 +0000 (17:57 +0000)]
* vibnal takes a single interface name which defaults to 'ipoib0'
eeb [Thu, 2 Jun 2005 07:09:25 +0000 (07:09 +0000)]
* moved #defines in qswnal.h around to separate fixed constants
from ones that can be set via modparams
* placeholder vibnal for newconfig; still need to get the IPoIB
IP address somehow.
eeb [Wed, 1 Jun 2005 21:23:21 +0000 (21:23 +0000)]
* Added interface and socket queries to libcfs (libcfs/linux-tcpip.c). Some
of this came from socknal_lib-linux.c but was generalised a little so all
the NALs that use TCP/IP to do connection establishment can use it.
CAVEAT EMPTOR! Just the linux versions are done; the darwin versions are
not implemented yet.
Changed socknal and ranal to use this.
* Brought ranal up-to-date with new config + mod params for all tunables.
Ranal gets its local address from the IP of a (single) specified
interface, or the first "suitable" one found if no interface is specified.
eeb [Thu, 19 May 2005 17:53:51 +0000 (17:53 +0000)]
* fixed use of %n at end of string in nidstrings.c
eeb [Thu, 19 May 2005 14:05:33 +0000 (14:05 +0000)]
* changed router lookups to always return the local destination
socknal + qswnal changes to match
simplified "will this be forwarded" predicate in ptl_send()
* fixed bug in multi-instance socknal to pass the correct NI to the tx
teardown function to allow for the case where the tx hasn't even had a conn
assigned yet.
* fixed routerstat stats formatting
* removed osc::check_elan_limit() and replaced with max_unfragmented_pages()
which actually works. NB it depend on the assumption that file and page
offsets are aligned!
* fixed lconf to stay away from portals, except to 'modprobe' for it first in
'load_module()', and to rmmod all the NALs + portals when ptlrpc gets
unloaded.
eeb [Mon, 16 May 2005 07:58:32 +0000 (07:58 +0000)]
* removed any trace of route table file
* changed 'routes=' syntax to drop 'route' keyword: syntax is now just a
list of directives separated by newlines or ';', with comments (#), where
each directive is <networks> <gateways> [<gateways> ...]
eeb [Sun, 15 May 2005 18:13:41 +0000 (18:13 +0000)]
* changed route table into a 2-level list; first by target net, then
by gateway.
eeb [Sun, 15 May 2005 15:51:09 +0000 (15:51 +0000)]
* ensure route tables print in same order as constructed
eeb [Sun, 15 May 2005 12:27:07 +0000 (12:27 +0000)]
* changed 'routes=' syntax...
(a) to support pdsh-style range expansion e.g....
[a,b,c,d-e,f,g,h-i]
(b) to allow target network expansion so that all the remote networks
accessible through the same set of gateways can be included in the
same route command.
eeb [Wed, 11 May 2005 20:12:29 +0000 (20:12 +0000)]
* multiple TCP networks pass initial tests
eeb [Wed, 11 May 2005 16:14:34 +0000 (16:14 +0000)]
* lonal part of portals
eeb [Wed, 11 May 2005 12:17:47 +0000 (12:17 +0000)]
* moved lonal into portals module
eeb [Wed, 11 May 2005 12:15:44 +0000 (12:15 +0000)]
* lonal moved into portals module
eeb [Wed, 11 May 2005 06:33:13 +0000 (06:33 +0000)]
* first cut multiple TCP networks
eeb [Wed, 11 May 2005 06:31:57 +0000 (06:31 +0000)]
* First cut multiple TCP networks
eeb [Sat, 7 May 2005 18:10:20 +0000 (18:10 +0000)]
* allow [a,b,c,...] in gateway NID ranges
* fixed routerstat
eeb [Sat, 7 May 2005 03:57:57 +0000 (03:57 +0000)]
* changed router to do "obvious" routing automatically if enabled
* tested routing between elan and tcp networks
eeb [Fri, 6 May 2005 16:32:15 +0000 (16:32 +0000)]
* module loading working with request_module() :))
* tcp interface enumeration/query working
eeb [Fri, 6 May 2005 14:25:14 +0000 (14:25 +0000)]
* placeholder (digging out IP interface info)
eeb [Thu, 5 May 2005 17:57:46 +0000 (17:57 +0000)]
* using request_module() to load the NAL (thanks Andreas)
eeb [Thu, 5 May 2005 17:12:51 +0000 (17:12 +0000)]
* portals auto-loads required NALs (taken from 'networks=')
eeb [Thu, 5 May 2005 06:43:36 +0000 (06:43 +0000)]
* first cut "networks" and "routes" portals module params
eeb [Thu, 5 May 2005 06:06:00 +0000 (06:06 +0000)]
* placeholder
eeb [Wed, 4 May 2005 09:49:28 +0000 (09:49 +0000)]
* placeholder
eeb [Fri, 29 Apr 2005 19:34:49 +0000 (19:34 +0000)]
* libcfs/nidstrings.c does nid parsing/printing. It's included
in libptlctl.a for userspace utils too.
* addr@net - style nids working with > 1 local network
* router part of portals; ptl_send() uses it to chose the NI, but NALs
still do am-i-there-yet forwarding. NAL interface changed to match.
* reworked nal commands into PtlCtl(ni, cmd, arg) and rewired portals ioctl
handling. PtlFailNid, now just another PtlCtl. NAL interface includes
ctl handler.
* commented out all lconf portals config, except setting the NID (pending
socknal working out its own interfaces)
* removed all portals config from zeroconf and llmount
* "lctl network" on its own lists all local NIDs (replaces shownid).
Most (all?) portals config commands still work from lctl.
* simplified NAL setup/teardown (no separate nal_cmd and router init)
eeb [Fri, 29 Apr 2005 15:19:45 +0000 (15:19 +0000)]
* placeholder commit
eeb [Tue, 26 Apr 2005 18:28:47 +0000 (18:28 +0000)]
* kernel socknal acceptor; commented out lconf acceptor start/stop
* removed socknal's use of sock->file; replaced with atomic refcount in conn
* moved more socknal tunables into module params/sysfs/sysctl
* consistent use/naming of socknal refcounting fns
eeb [Tue, 26 Apr 2005 02:52:47 +0000 (02:52 +0000)]
* temporarily disabled socknal explicit peer table setup until lconf
fixed to stop doing it
* added socknal support for peer IP == local_address(NID)
* added CFS_MODULE_PARM macros
* changed socknal and qswnal tunables to allow setting via module
parameters
eeb [Thu, 21 Apr 2005 17:14:20 +0000 (17:14 +0000)]
* new-config; simple 2 node echo test working with single API interface
eeb [Thu, 21 Apr 2005 00:35:17 +0000 (00:35 +0000)]
* new network config snapshot
half-way through lustre uses single NI
cvs2svn [Tue, 19 Apr 2005 23:55:58 +0000 (23:55 +0000)]
This commit was manufactured by cvs2svn to create branch 'b_hd_newconfig'.
jacob [Tue, 19 Apr 2005 23:55:57 +0000 (23:55 +0000)]
add support for sles 8:
- new target for sles 8 kernel
- post(un) scripts for kernel rpms
- fixup quota build on sles 8 and sles 9
- update patches for a more recent sles 8 kernel
jacob [Tue, 19 Apr 2005 23:25:14 +0000 (23:25 +0000)]
b=b=6065
r=eeb
r=adilger
remove noops when stderr isn't present; just fail instead.
ericm [Tue, 19 Apr 2005 22:14:36 +0000 (22:14 +0000)]
from Brian: adjust default lsd location to /usr/sbin instead of /sbin/
yury [Tue, 19 Apr 2005 12:27:07 +0000 (12:27 +0000)]
- added comment in mds_open() and GNS mount points.
- cleanups in sanity-gns.sh
- added one more GNS test - mounting directory with absent mount object
yury [Tue, 19 Apr 2005 12:08:05 +0000 (12:08 +0000)]
- cosmetic changes in mdapi DLD
- more fixes in GNS stuff:
- fixes with refcount onto @dchild in ll_gns_mount_object() for error control paths
- fixing using accessing @dchild in the case of IS_ERR(dchild)
- disabled checking for GNS file in mds_open(), thus GNS mount points are also
opening by now. This is needed, because in taking lock we can't reliable to detect
GNS objects on MDS. Checking just suid bit and skip open such a dirs is not correct
as one may create structure like dir/dir/dir and mark all of them by suid.
- cleanups in sanity-gns.sh. Added few new tests for odd conditions like mount object
is symlink, dir, etc. Added test for changing mount object name to "." and "..".
Grouping tests into 2 groups:
(1) general GNS tests - mounting/umounting on timeout with different conditions
with possible deadlocks, races, etc.
(2) checking for odd conditions like recursive suid bit marked mount points, etc
yury [Mon, 18 Apr 2005 16:37:32 +0000 (16:37 +0000)]
- more fixes and improvements in GNS stuff:
- small optimization in ll_gns_wait_for_mount(), changing return code semantics to
usual 0 - ok, < 0 error.
- handled possible nested mount with returning -ERETSARTSYS and causing endless loop
of mounting.
- check if found mount object is regular file to prevent from evil cases like it is
suid bit marked directory, etc.
- fix with completting all waiters for dentry mounting in progress.
- added check for "." and ".." names in setting gns_object_name.
- fixed default gns_object_name value.
yury [Mon, 18 Apr 2005 12:40:50 +0000 (12:40 +0000)]
- fixed returning @retval in ll_lookup_it() in the case of -ERESTARTSYS
yury [Mon, 18 Apr 2005 12:38:33 +0000 (12:38 +0000)]
- fixed missed comma
yury [Mon, 18 Apr 2005 12:13:53 +0000 (12:13 +0000)]
- many fixes in GNS code after Nikita's code review. They are the following:
- removed ll_d_lookup() as it is not used anymore.
- removed struts from real_lookup() with calling revalidate() after lookup if dentry
is marked GNS_MOUNTING. By now all this dentry marking is not needed anymore.
- ll_gns_mount_object() is now called from ll_revalidate_it() and ll_lookup_it()
- fixed deadlock when one thread T1 waits for mount completion and another one T2
created by upcall script waits for dentry->d_inode->i_sem. By now -ERESTARTSYS
is returned for T2 and others and whren they restart syscall, dentry is mounted
already or some error occured and GNS state is READY for mount.
- fixes in comments and header of llite_gns.c
- fixed possible incorrect spin_unlock() in ll_gns_mount_object()
- using lookup_one_len() for looking for mount object instead of d_lookup() which
looks only in dcache.
- fixed handling of negative mount object dentry in ll_gns_mount_object()
- fixed readding mount object content. By now kernel_read() is used instead of
readpage approach.
- waking up all possible waiters on finishing GNS mount. Nevertheless in current
stuff nobody seem waits for mount (-RESTARTSYS is used instead), completion
exists and it is semanticaly correct to wake up all its waiters. In fact this
is needed for possible future using. If it happen that nobody will wait on it
and it is counted not needed, it will be removed.
ericm [Fri, 15 Apr 2005 22:22:57 +0000 (22:22 +0000)]
from Alex: temporarily fix for mds hanging on mountain: spinlock might
be called in interrupt context.
tappro [Wed, 13 Apr 2005 17:14:15 +0000 (17:14 +0000)]
smfs support in basic tests
MDS_BACKFSTYPE and OST_BACKFSTYPE are defined for this.
FSTYPE is ldiskfs for 2.6 kernel by default
yury [Wed, 13 Apr 2005 15:48:15 +0000 (15:48 +0000)]
- fix for sanityN.sh test_16 from Oleg (patch from #4982)
yury [Wed, 13 Apr 2005 15:25:46 +0000 (15:25 +0000)]
- default filesystem on 2.6.x is ldiskfs, not ext3
- small fix about check_type() function
tappro [Wed, 13 Apr 2005 13:02:39 +0000 (13:02 +0000)]
b=5881
-replace CERROR with CDEBUG()
-check strings before using.
tappro [Wed, 13 Apr 2005 12:23:46 +0000 (12:23 +0000)]
b=5881
remove untested stuff. Additional testing is needed at first.
yury [Wed, 13 Apr 2005 11:27:14 +0000 (11:27 +0000)]
- added patches from #5492 and 5654
tappro [Wed, 13 Apr 2005 10:08:34 +0000 (10:08 +0000)]
b=6143
r=wangdi
changes for smfs support:
1. realdev in prepare() is set to device name, not to 'smfs'. This fixes bug 6143.
2. backdev is not used, instead of it real device name used for --dev parameter in smfs configuration.
yury [Wed, 13 Apr 2005 09:49:50 +0000 (09:49 +0000)]
- compile fixes for --enable-liblustre
tappro [Wed, 13 Apr 2005 07:01:32 +0000 (07:01 +0000)]
b=5881
update kml and lru stuff in accordance with new DLD for SMFS
remove unused file journal.c
add changes to Makefile
using new options retrival method in smfs_lib
tappro [Wed, 13 Apr 2005 06:58:43 +0000 (06:58 +0000)]
b=5881
r=alex
changes related to iopen support, fc3 compatibility and new smfs plugin API
tappro [Wed, 13 Apr 2005 06:53:34 +0000 (06:53 +0000)]
b=5881
replace several 'if' that check non-allowed situations with LASSERT()
tappro [Wed, 13 Apr 2005 06:51:33 +0000 (06:51 +0000)]
b=5881
adding sets of smfs fsfilt methods
adds some new HOOKs
tappro [Wed, 13 Apr 2005 06:49:38 +0000 (06:49 +0000)]
b=5881
move smfs methods to smfs dir.
tappro [Wed, 13 Apr 2005 06:42:52 +0000 (06:42 +0000)]
b=5881
r=alex
d=24-33
support for iopen
FC3 compatibility
tappro [Wed, 13 Apr 2005 06:40:04 +0000 (06:40 +0000)]
b=5881
r=alex
d=24-33
support for iopen and cmd in smfs methods.
tappro [Wed, 13 Apr 2005 06:35:51 +0000 (06:35 +0000)]
b=5881
r=alex
d=26-33
support for iopen/cmd/wantedi for smfs
adilger [Tue, 12 Apr 2005 23:54:43 +0000 (23:54 +0000)]
Branch: b1_4_5921
Handle "./lustrecvs -r foo" instead of choking.
eeb [Tue, 12 Apr 2005 13:39:38 +0000 (13:39 +0000)]
* removed unused lib_nal_t abstraction
alex [Tue, 12 Apr 2005 13:05:21 +0000 (13:05 +0000)]
b=6136
r=<awaiting from eeb>
- use correspondent list anchor
eeb [Tue, 12 Apr 2005 10:59:46 +0000 (10:59 +0000)]
* fix for 4580 vibnal NULL deref in arp callback
yury [Sat, 9 Apr 2005 11:57:34 +0000 (11:57 +0000)]
- added sanity-gns.sh test_5a
wangdi [Fri, 8 Apr 2005 11:56:49 +0000 (11:56 +0000)]
Branch HEAD
add 2.6.10 fc3 kernel config
yury [Fri, 8 Apr 2005 08:24:21 +0000 (08:24 +0000)]
- fixed lossing error code in llog processing during zconf mount. This caused
accessing not initialized MDCs in the case of errors.
wangdi [Fri, 8 Apr 2005 07:46:17 +0000 (07:46 +0000)]
Branch: HEAD
1)minor fix and cleanup dev_rd_only kernel patch on HEAD
2)add sec support on kernel config
3)some cleanup
alex [Thu, 7 Apr 2005 16:40:42 +0000 (16:40 +0000)]
- disable CDEBUG() ratelimit for a while
wangdi [Thu, 7 Apr 2005 15:41:12 +0000 (15:41 +0000)]
Branch: HEAD
add signal bug fix to uml patch
yury [Thu, 7 Apr 2005 10:54:01 +0000 (10:54 +0000)]
- cleanups in sanity-gns.sh
yury [Thu, 7 Apr 2005 10:49:37 +0000 (10:49 +0000)]
- removed hardcoded checking for ".mntinfo" on MDS to see if we are trying to open GNS mount object. By now only suid bit on directory is checked. This should not break anything, as it is not used for any purpose.
- cleanups, moving not needed to be widely known llite functions to llite_internal.h
- added test_4a to sanity-gns.sh. It checks how GNS mounting works with 20 concurrent threads aiming to mount the same object.
wangdi [Thu, 7 Apr 2005 10:19:37 +0000 (10:19 +0000)]
Branch: HEAD
minor fix in lkcd patch
wangdi [Thu, 7 Apr 2005 09:13:45 +0000 (09:13 +0000)]
Branch: HEAD
some patches fix for lustre in FC3
adilger [Wed, 6 Apr 2005 23:43:55 +0000 (23:43 +0000)]
Branch: b1_4
- update gmnal to use PTL_MTU, fix module refcounting, for HP
b=5786
r=mdoyle (original patch)
wangdi [Wed, 6 Apr 2005 16:44:03 +0000 (16:44 +0000)]
Branch HEAD
update dev_read_only patch from b1_4, thanks for nathan's reminding
wangdi [Wed, 6 Apr 2005 16:25:49 +0000 (16:25 +0000)]
Branch: HEAD
minor fix in Makefile for uml patch
wangdi [Wed, 6 Apr 2005 16:18:43 +0000 (16:18 +0000)]
Branch: HEAD
add uml and vmware kernel config for FC3
wangdi [Wed, 6 Apr 2005 16:10:05 +0000 (16:10 +0000)]
Branch HEAD
add lustre FC3 series
wangdi [Wed, 6 Apr 2005 16:05:44 +0000 (16:05 +0000)]
Branch: HEAD
add FC3 kernel patches
adilger [Wed, 6 Apr 2005 11:57:42 +0000 (11:57 +0000)]
Branch: b_cray
Allow utils to be built with cray portals.
b=5918
r=bogl (original change)
eeb [Wed, 6 Apr 2005 11:40:59 +0000 (11:40 +0000)]
* Updated gmnalnid.c with patch from HP
eeb [Wed, 6 Apr 2005 10:49:52 +0000 (10:49 +0000)]
* changed vibnal compilation checks to allow gcc < 3.0.0 and
allow HIGHMEM4G on x86 (bug 5852)
eeb [Tue, 5 Apr 2005 20:33:42 +0000 (20:33 +0000)]
* bringing llmount.c up to date for RapidArray (XD1) and IB Nals
jacob [Tue, 5 Apr 2005 20:33:04 +0000 (20:33 +0000)]
b=5667
always include our local ioctl.h if _IOW isn't defined, which is
probably only catamount.
eeb [Tue, 5 Apr 2005 19:01:15 +0000 (19:01 +0000)]
* Updated lconf symbolic debug masks to be consistent with portals HEAD
eeb [Tue, 5 Apr 2005 17:54:32 +0000 (17:54 +0000)]
* 5852 vibnal highmem fix to always use page_to_phys() when calculating
physical memory addresses, but conditionally supress sign extension if
page_to_phys() returns a 32 bit address.
* removed confusing (and obsolete) use of vmalloc_to_page() in vibnal's
kibnal_kvaddr_to_page().
* removed some unused variables in vibnal
yury [Tue, 5 Apr 2005 15:25:17 +0000 (15:25 +0000)]
- cleanups in sanity-fid.sh
- added more tests into sanity-gns.sh, splitting general test onto few smaller ones.
eeb [Tue, 5 Apr 2005 10:44:21 +0000 (10:44 +0000)]
* fixed libcfs watchdog CWARN; units are mS, not uS
adilger [Tue, 5 Apr 2005 02:56:30 +0000 (02:56 +0000)]
Branch: b_cray
b=5915,6054
eeb [Tue, 5 Apr 2005 02:00:02 +0000 (02:00 +0000)]
* ranal connection setup changes from Igor (bug 5915)
jacob [Mon, 4 Apr 2005 23:36:27 +0000 (23:36 +0000)]
b1_4_bgl is dead
jacob [Mon, 4 Apr 2005 23:33:18 +0000 (23:33 +0000)]
land b1_4_bgl on HEAD (20050404_1913)
adilger [Mon, 4 Apr 2005 22:28:25 +0000 (22:28 +0000)]
Branch: b_cray
Grab new cray snapshot tags from same place as b_cray.
b=6053