Whamcloud - gitweb
fs/lustre-release.git
21 years ago* Merged HEAD
eeb [Thu, 4 Dec 2003 11:06:55 +0000 (11:06 +0000)]
*   Merged HEAD

*   "clobbered" (as Phil would say) kernel_patches/patches

21 years agoInstrumentation for reproducing and verifying 1897 (open-count leaked if close
shaver [Wed, 3 Dec 2003 21:24:57 +0000 (21:24 +0000)]
Instrumentation for reproducing and verifying 1897 (open-count leaked if close
is interrupted on the client). r=robert.

21 years ago* Ensured CWARN messages always get to the system log.
eeb [Tue, 2 Dec 2003 23:07:59 +0000 (23:07 +0000)]
*   Ensured CWARN messages always get to the system log.

    NB this only affects socknal EOF messages currently, since the NALs'
    subsystem debug is disabled by default.  Hopefully the appearance
    of these messages (which should be relatively infrequent and provide
    good post-mortem evidence) will not spread too much FUD...

21 years ago* fixed a missing initialiser that makes autoconnect crash in a CDEBUG
eeb [Tue, 2 Dec 2003 22:59:53 +0000 (22:59 +0000)]
*    fixed a missing initialiser that makes autoconnect crash in a CDEBUG
     if it fails to create a connection

*    moved tx_resid initialisation to launch_packet

21 years ago- error var was declared after statements. gcc-2.96 dislikes this
alex [Tue, 2 Dec 2003 20:59:24 +0000 (20:59 +0000)]
- error var was declared after statements. gcc-2.96 dislikes this
  b2294

21 years ago- fixes against gfp_memalloc patches
alex [Tue, 2 Dec 2003 20:36:29 +0000 (20:36 +0000)]
- fixes against gfp_memalloc patches
- gfp_memalloc is added into vanilla-2.4.20 series

21 years ago- use -fPIC to build liblustre to stop the .os from containing relocations
zab [Tue, 2 Dec 2003 19:48:24 +0000 (19:48 +0000)]
- use -fPIC to build liblustre to stop the .os from containing relocations
  that x86_64 can't put in .sos
- don't include ldlm *and* ptlrpc libs in liblustre.{a,so}

21 years agob=2310
jacob [Tue, 2 Dec 2003 19:28:34 +0000 (19:28 +0000)]
b=2310
r=phil

fix a missed case in the GFP_MEMALLOC patch which can cause ksocknald
to block on kswapd (which is bad)

21 years ago* Dynamic allocation of socknal tx descs.
eeb [Tue, 2 Dec 2003 15:27:59 +0000 (15:27 +0000)]
*   Dynamic allocation of socknal tx descs.

*   Statically allocated socknal forwarded message buffers now allocated
    individually rather than 1 large chunk to avoid store fragmentation.

*   Some function renaming in socknal_cb.c to avoid name conflicts.

*   added memhog gfp flags

*   lmc --gw_cluster_id => lmc --gateway_cluster_id to avoid conflict with
    lmc --gw

21 years agoUpdated the document for some additional options.
radhika [Tue, 2 Dec 2003 00:20:06 +0000 (00:20 +0000)]
Updated the document for some additional options.

21 years agoremove the show_mem() debugging; it's not exported from all kernels,
phil [Mon, 1 Dec 2003 22:58:47 +0000 (22:58 +0000)]
remove the show_mem() debugging; it's not exported from all kernels,
and I've already learned what I needed to learn

21 years ago* Minor socknal cleanup
eeb [Mon, 1 Dec 2003 16:43:30 +0000 (16:43 +0000)]
*   Minor socknal cleanup

*   Added lctl::memhog to allocate kernel memory (freed on request or close)

21 years ago- configurable stacksize against all the series except chaos-2.4.18
alex [Mon, 1 Dec 2003 15:39:02 +0000 (15:39 +0000)]
- configurable stacksize against all the series except chaos-2.4.18
  only i386 is affected

21 years agob=1933
phil [Sun, 30 Nov 2003 22:56:01 +0000 (22:56 +0000)]
b=1933
If an allocation fails, print information about how much memory we
have allocated, and call show_mem()

21 years agouse PORTAL_SLAB_FREE now that we also use _ALLOC
phil [Sun, 30 Nov 2003 22:01:53 +0000 (22:01 +0000)]
use PORTAL_SLAB_FREE now that we also use _ALLOC

21 years ago* Added lwt stop now prints the wallclock time of the first trace record
eeb [Sun, 30 Nov 2003 18:23:50 +0000 (18:23 +0000)]
*    Added lwt stop now prints the wallclock time of the first trace record

*    Turned off lwt support by default

21 years agob=278
phil [Sat, 29 Nov 2003 09:15:51 +0000 (09:15 +0000)]
b=278
More merging from b_eq, including the removal of ptlrpc_request->rq_obd

21 years agob=1933
phil [Sat, 29 Nov 2003 04:28:12 +0000 (04:28 +0000)]
b=1933
The lib_*_alloc functions in portals were calling kmem_cache_alloc(GFP_NOFS)
instead of PORTAL_SLAB_ALLOC, which uses GFP_MEMALLOC. These calls are _more_
likely to return NULL, because they won't use the emergency pool and can't
descend into the file system to free memory.

Unsurprisingly, these lib functions eventually failed to allocate at NCSA, so
this is not just a theoretical risk.

21 years agomerge part of b_eq to b_devel:
phil [Fri, 28 Nov 2003 21:53:50 +0000 (21:53 +0000)]
merge part of b_eq to b_devel:

removes ptlrpc_pack_msg(), adds ptlrpc_pack_request() and ptlrpc_pack_reply()

21 years agomore innocent b_eq to b_devel merging:
phil [Fri, 28 Nov 2003 21:51:40 +0000 (21:51 +0000)]
more innocent b_eq to b_devel merging:
 - updates to liblustre-specific files
 - updates to #ifdef-ed code
 - Makefiles et al

21 years agob=1993
ccooper [Fri, 28 Nov 2003 21:40:22 +0000 (21:40 +0000)]
b=1993

- wrapper for sanity that will submit test output to buffalo. Based on an original script by sice for UML that I updated to work with real hardware as well.

Requires: recent checkout of ltest module, b_tdevel branch. If you plan to use this wrapper with UML, you must be able to DSH into UML from your host machine.

Here is an example command line I use to run the wrapper on mdev:

LTESTDIR=/home/ccooper/b_tdevel NETWORKTYPE=elan  sh sanity-buffalo.sh --sender=ccooper@clusterfs.com --extra-params="START=' ' CLEAN=' ' SANITYLOG=/tmp/sanity.log" --target=mdev10 --config=lov

...and that yielded the following buffalo-sandbox report:

https://buffalo-sandbox.clusterfs.com/display_report.pl?report_id=121

Reports are sent to the buffalo-sandbox by default to avoid cluttering buffalo proper with bogus results, which inevitably happens when you're first trying to get reporting setup on a new cluster. You can toggle the RECIPIENTS variable in the source once the wrapper is reporting properly.

21 years agorename lop_precleanup to lop_sync and do related changes
tianying [Fri, 28 Nov 2003 17:52:39 +0000 (17:52 +0000)]
rename lop_precleanup to lop_sync and do related changes

21 years ago 1. add lop_precleanup - llog_obd_repl_precleanup for filter
tianying [Fri, 28 Nov 2003 15:25:15 +0000 (15:25 +0000)]
   1. add lop_precleanup - llog_obd_repl_precleanup for filter
   2. delay llog_precleanup after fsfilt_sync in filter_disconnect

21 years ago* Fixed ksocknal_process_transmit() rc in error case
eeb [Fri, 28 Nov 2003 14:46:46 +0000 (14:46 +0000)]
*    Fixed ksocknal_process_transmit() rc in error case

21 years agob=2254
phil [Fri, 28 Nov 2003 06:54:21 +0000 (06:54 +0000)]
b=2254
At truncate time, ext3 is zeroing indirect blocks and writing a
transaction to the journal.  At some point, perhaps when that
transaction commits, it gives the dirty buffer for that indirect block
to the buffer cache to write to disk.  Meanwhile, having marked that
block as unused, it reallocates it to us as a normal data block into
which IOR puts some data.

The obdfilter writes that block with brw_kiovec, which writes
immediately to the disk with no regard for what data might be in the
buffer cache.  Shortly thereafter, the buffer cache writes the block
of zeroes over top our valuable data.

The correct fix is to modify our special ext3 block allocation code to
look in the buffer cache for us, and discard any pending writes to the
newly-allocated blocks, much like the direct I/O code does.  As a
workaround, for kernels which do not yet have this change, I added
some code to the obdfilter to do this after the call to
ext3_map_inode_page returns.

This introduces kernel version 32, but doesn't force an upgrade.  I
updated the kernel patches for 2.4.18 and 2.4.20, but not 2.6.  I also:
 - tested the ext3_map_inode_page change on vanilla-2.4.20
 - tested the workaround change on chaos-2.4.18
 - compile-tested a version 32 chaos-2.4.18 kernel

21 years agoSlightly less obvious, but still very innocent, parts of b_eq:
phil [Thu, 27 Nov 2003 06:31:34 +0000 (06:31 +0000)]
Slightly less obvious, but still very innocent, parts of b_eq:

 - code or includes in the main build, but contained in #if __KERNEL__
 - userspace-only portals code
 - very few and minor other changes, such as renaming a function

21 years agob=2303: Unregister replies when skipping ping-requests from disconnected
shaver [Thu, 27 Nov 2003 01:05:52 +0000 (01:05 +0000)]
b=2303: Unregister replies when skipping ping-requests from disconnected
        imports, to avoid tripping assertions when we free the requests.
r=robert

21 years ago* Fixed socknal forwarding race in the new version of
eeb [Wed, 26 Nov 2003 21:42:28 +0000 (21:42 +0000)]
*   Fixed socknal forwarding race in the new version of
    ksocknal_process_receive().

21 years agobacking out eeb's socknal checkin from yesterday:
phil [Wed, 26 Nov 2003 18:52:12 +0000 (18:52 +0000)]
backing out eeb's socknal checkin from yesterday:

11/25/2003 12:22 eeb lustre/portals/knals/socknal/socknal.c 1.1.2.13
11/25/2003 12:22 eeb lustre/portals/knals/socknal/socknal_cb.c 1.1.2.16

LLNL found a race condition in overnight testing, causes LBUG.

21 years agob=2296
rread [Wed, 26 Nov 2003 01:26:46 +0000 (01:26 +0000)]
b=2296
r=shaver

- Build ldlm into ptlrpc.
- Remove all the ptlrpc_ldlm_* hooks

This  almost certainly breaks building when $(SRDIR) != $(OBJDIR).

21 years ago- unregister the elan nal
rread [Tue, 25 Nov 2003 20:29:39 +0000 (20:29 +0000)]
- unregister the elan nal

21 years ago* Always hold kss_lock while setting/inspecting ksnc_{rx,tx}_ready.
eeb [Tue, 25 Nov 2003 17:22:29 +0000 (17:22 +0000)]
*    Always hold kss_lock while setting/inspecting ksnc_{rx,tx}_ready.

*    Removed kss_lock ops from ksocknal_process_{transmit,receive}().
     ksocknal_scheduler() does this locking now.

*    Included the --single_socket lconf flag.

21 years agoliblustre: as andreas pointed out, open() need to convert O_XXX flags
ericm [Tue, 25 Nov 2003 14:43:38 +0000 (14:43 +0000)]
liblustre: as andreas pointed out, open() need to convert O_XXX flags
to FMODE_XXX which is internally used by lustre. remove the original
hacking code.

21 years ago1. move llog_origin_handle_cancel to llog_server.c
tianying [Tue, 25 Nov 2003 12:51:54 +0000 (12:51 +0000)]
1. move llog_origin_handle_cancel to llog_server.c
2. destroy useless plain logs during both llog_cleanup and llog_setup
3. change some CERRORs to CWARNs while just print out status information

21 years agob=1941
jacob [Tue, 25 Nov 2003 07:12:15 +0000 (07:12 +0000)]
b=1941

Improvements from jerrifer's patch, with some changes.

 - use lmc --batch for performance goodness
 - remove 'read_conf' as it is confusing and redundant
 - misc cleanups, improvements

21 years agocleanup clients correctly in insanity.sh
rread [Mon, 24 Nov 2003 23:57:02 +0000 (23:57 +0000)]
cleanup clients correctly in insanity.sh
set the mds upcall, so the mds can recovery from ost failures

21 years ago- port onto virgin 2.4.22
alex [Mon, 24 Nov 2003 22:13:36 +0000 (22:13 +0000)]
- port onto virgin 2.4.22

21 years agoMove device type specific llog ioctl codes from generic obdclass to mds/obdfilter...
niu [Mon, 24 Nov 2003 08:59:21 +0000 (08:59 +0000)]
Move device type specific llog ioctl codes from generic obdclass to mds/obdfilter/mdc.

21 years agob=1883
phil [Mon, 24 Nov 2003 04:45:54 +0000 (04:45 +0000)]
b=1883
- Adds llite_close.c; Zach's originally, somewhat heavily hacked
- Adds an MDS_DONE_WRITING rpc and handler
- Adds client-side code for storing the MDS FID and epoch in the
  obdo's inline field in BRW_WRITE requests

21 years agob=2192
phil [Mon, 24 Nov 2003 00:24:53 +0000 (00:24 +0000)]
b=2192
ll_inode_getattr (-> ll_lsm_getattr) now takes an obdo, which it fills with the
attributes.  The caller is responsible, if it wishes, for updating the fields
in the inode, and the one previous caller now contains that extra code.  We'll
see whether additional abstraction is required during the KMS work.

This is a prerequisite for the first 30% of the KMS that we decided to do as a
fix for this bug.

21 years agoliblustre: forgot to commit last night: workaround O_WRONLY (b2282)
ericm [Sat, 22 Nov 2003 07:09:16 +0000 (07:09 +0000)]
liblustre: forgot to commit last night: workaround O_WRONLY (b2282)

21 years agoadd special device inode lustre handler
wangdi [Fri, 21 Nov 2003 18:20:14 +0000 (18:20 +0000)]
add special device inode lustre handler

21 years ago- patch to fix 64-bit pointer arithmetic bug in the ext3 extended
alex [Fri, 21 Nov 2003 15:06:24 +0000 (15:06 +0000)]
- patch to fix 64-bit pointer arithmetic bug in the ext3 extended
  attributes code.

21 years ago- rename the patch
alex [Fri, 21 Nov 2003 14:42:56 +0000 (14:42 +0000)]
- rename the patch

21 years ago- patches against 2.4.19-pre1 and series
alex [Fri, 21 Nov 2003 14:36:01 +0000 (14:36 +0000)]
- patches against 2.4.19-pre1 and series

21 years agomove client_df to test-framework.
rread [Fri, 21 Nov 2003 09:10:57 +0000 (09:10 +0000)]
move client_df to test-framework.

21 years ago- integrate support for FAILURE_MODE into test-framework
rread [Fri, 21 Nov 2003 08:42:40 +0000 (08:42 +0000)]
- integrate support for FAILURE_MODE into test-framework
- add configs for mdev

21 years ago- cleanup old active files
rread [Fri, 21 Nov 2003 03:27:14 +0000 (03:27 +0000)]
- cleanup old active files
- cleanup insanity test_0
- add checks to make sure the nodes are up during setup

21 years agoadd usage() to test_framework, and use getopts cause getopt is broken
rread [Fri, 21 Nov 2003 01:04:56 +0000 (01:04 +0000)]
add usage() to test_framework, and use getopts cause getopt is broken
insanity fixes for hardware

21 years agoSmall test and upcall fixes; we fail over during a write now.
shaver [Thu, 20 Nov 2003 21:41:14 +0000 (21:41 +0000)]
Small test and upcall fixes; we fail over during a write now.

21 years agoFix bulk RPC checksum code and page poisoning for debugging bug 2254.
adilger [Thu, 20 Nov 2003 12:23:22 +0000 (12:23 +0000)]
Fix bulk RPC checksum code and page poisoning for debugging bug 2254.
b=1703,b=2254

21 years ago- Integrate insanity.sh with test-framework
rread [Thu, 20 Nov 2003 07:43:00 +0000 (07:43 +0000)]
- Integrate insanity.sh with test-framework
- Centralize the config data to files in tests/cfg/
- Different config file can be used by setting CONFIG=file or
  using command line option:   --config <file>
- FAILURE_MODE option for insanity.sh:
  FAILURE_MODE=SOFT - shutdown services with lconf --force
  FAILURE_MODE=HARD - power down the service nodes using $POWER_DOWN
     and $POWER_UP to reboot.
  This feature should be integrated with the other tests.

- replay-single.sh, replay-dual.sh, and conf-sanity.sh all pass on hardware
- replay-ost-dual.sh and insanity.sh don't pass yet

21 years agobug: 1944
wangchao [Thu, 20 Nov 2003 06:06:55 +0000 (06:06 +0000)]
bug: 1944
reviewer: Andreas
Add "lfs osts" and "lfs check <mds|osts|servers>" to lfs.

21 years agoadd lbuild scripts, please read lbuild --help
wangdi [Thu, 20 Nov 2003 03:25:17 +0000 (03:25 +0000)]
add lbuild scripts, please read lbuild --help

21 years ago- during disconnect, downgrade import level early for non-replayable
rread [Thu, 20 Nov 2003 00:50:56 +0000 (00:50 +0000)]
- during disconnect, downgrade import level early for non-replayable
obds. This prevents ptlrpc from automatically reconnecting on a
network failure.

21 years agoFix local-large-inode.sh script so it is actually using large inodes (my bad),
adilger [Wed, 19 Nov 2003 21:40:51 +0000 (21:40 +0000)]
Fix local-large-inode.sh script so it is actually using large inodes (my bad),
although in my defence acc-sm.sh would have broken until people updated their
kernel, which has mostly happened now.

Make the other configs create an MDS large enough to hold inodes for bonnie++
with all-external EA data.

21 years agoReturn GFP_NOFS fix to socknal that was lost in autoconn code shuffle.
adilger [Wed, 19 Nov 2003 21:36:22 +0000 (21:36 +0000)]
Return GFP_NOFS fix to socknal that was lost in autoconn code shuffle.
b=1933

21 years agoMDS-side OSCs always ping their OSTs for prompt failure detection.
shaver [Wed, 19 Nov 2003 18:53:50 +0000 (18:53 +0000)]
MDS-side OSCs always ping their OSTs for prompt failure detection.

21 years agoliblustre: release it lock a bit later, abvoid reference to null lock,
ericm [Wed, 19 Nov 2003 11:38:36 +0000 (11:38 +0000)]
liblustre: release it lock a bit later, abvoid reference to null lock,
should be the reason of LASSERT in simul test.

21 years agoadd conf-sanity.sh
wangdi [Wed, 19 Nov 2003 11:10:09 +0000 (11:10 +0000)]
add conf-sanity.sh

21 years agoliblustre: minor fix for replay_single test, even 60 to too much for vmware.
ericm [Wed, 19 Nov 2003 09:57:20 +0000 (09:57 +0000)]
liblustre: minor fix for replay_single test, even 60 to too much for vmware.

21 years ago- Remove last references to OBD_CLASS_UUID. Now the obd->obd_uuid is
rread [Tue, 18 Nov 2003 20:36:11 +0000 (20:36 +0000)]
- Remove last references to OBD_CLASS_UUID.  Now the obd->obd_uuid is
used as the client uuid for self and lctl exports.

21 years ago- fix the osc histogram spinlock inits that my uml build missed.
zab [Tue, 18 Nov 2003 18:43:55 +0000 (18:43 +0000)]
- fix the osc histogram spinlock inits that my uml build missed.

take a shot.

21 years agob=2247: detect misbehaving clients that reuse UUIDs _and_ import handles (such
shaver [Tue, 18 Nov 2003 18:30:56 +0000 (18:30 +0000)]
b=2247: detect misbehaving clients that reuse UUIDs _and_ import handles (such
        as libtest before the srand fix) and error, rather than LBUGing.

21 years agoinsanity.sh - this is test17 turned into a regular test, if you could
rread [Tue, 18 Nov 2003 09:52:39 +0000 (09:52 +0000)]
insanity.sh - this is test17 turned into a regular test, if you could
call this test regular. Still a work in progress, but does at least
setup/cleanup.

21 years agoliblustre: fix test_lock_cancel for zeroconf.
ericm [Tue, 18 Nov 2003 08:58:34 +0000 (08:58 +0000)]
liblustre: fix test_lock_cancel for zeroconf.

21 years ago- now devices can use memory allocated via uaf
alex [Mon, 17 Nov 2003 23:28:28 +0000 (23:28 +0000)]
- now devices can use memory allocated via uaf

21 years agoAllow current b_devel to work with older (read LLNL) kernels until they are
adilger [Mon, 17 Nov 2003 19:26:28 +0000 (19:26 +0000)]
Allow current b_devel to work with older (read LLNL) kernels until they are
updated.  Bumped LUSTRE_KERNEL_VERSION to 31 so we know which kernels have
the required patches, but don't make it a hard requirement.
b=1933

21 years ago- lustre patch of b1933
alex [Mon, 17 Nov 2003 17:32:56 +0000 (17:32 +0000)]
- lustre patch of b1933

NOTE: don't forget to apply gfp_memalloc patche first

21 years ago- kernel part of b1933
alex [Mon, 17 Nov 2003 17:32:24 +0000 (17:32 +0000)]
- kernel part of b1933

21 years agob: 2194
niu [Mon, 17 Nov 2003 06:42:02 +0000 (06:42 +0000)]
b: 2194
r: Peter

Add lctl interface for llog.

21 years agoSigh, patch I had submitted (which I grabbed) was not what was added to tree
adilger [Sun, 16 Nov 2003 23:18:42 +0000 (23:18 +0000)]
Sigh, patch I had submitted (which I grabbed) was not what was added to tree
(minor changes from SCT on the way).

21 years agoSlight context diff for journal callback patch from vanilla, add to suse series.
adilger [Sun, 16 Nov 2003 23:01:14 +0000 (23:01 +0000)]
Slight context diff for journal callback patch from vanilla, add to suse series.

21 years agoAdd journal callback patch for SuSE 2.4.19 ia64 kernel (in vanilla 2.4.20).
adilger [Sun, 16 Nov 2003 22:55:54 +0000 (22:55 +0000)]
Add journal callback patch for SuSE 2.4.19 ia64 kernel (in vanilla 2.4.20).

21 years agoNew patch files for SuSE 2.4.19 ia64 kernel as used by SDSC. Mostly cosmetic
adilger [Sun, 16 Nov 2003 22:08:54 +0000 (22:08 +0000)]
New patch files for SuSE 2.4.19 ia64 kernel as used by SDSC.  Mostly cosmetic
except vfs_intent still used the old-style path_init()/path_walk(), and it
already had EA patches applied.

21 years agoNew series for SuSE-2.4.19 ia64 kernel for SDSC machines.
adilger [Sun, 16 Nov 2003 21:59:19 +0000 (21:59 +0000)]
New series for SuSE-2.4.19 ia64 kernel for SDSC machines.

21 years agoRemove gratuitous newline hunk, which causes a conflict on SuSE-2.4.19.
adilger [Sun, 16 Nov 2003 21:43:24 +0000 (21:43 +0000)]
Remove gratuitous newline hunk, which causes a conflict on SuSE-2.4.19.

21 years ago* Plugged a spurious timeout hole
eeb [Sun, 16 Nov 2003 17:45:55 +0000 (17:45 +0000)]
*   Plugged a spurious timeout hole

21 years ago- lightly-tested, undocumented debuggin tool to catch use-after-free bugs
alex [Sun, 16 Nov 2003 16:40:30 +0000 (16:40 +0000)]
- lightly-tested, undocumented debuggin tool to catch use-after-free bugs

NOTE: documentation will be available soon

21 years agob=2250
rread [Sat, 15 Nov 2003 01:17:52 +0000 (01:17 +0000)]
b=2250

create lctl commands set_lustre_upcall and set_timeout, so they
can be saved as part of the 0conf log.

21 years agoon hardware, the script runs too fast for these multiops to die, cause
rread [Sat, 15 Nov 2003 01:05:26 +0000 (01:05 +0000)]
on hardware, the script runs too fast for these multiops to die, cause
the /mnt/lustre to be busy.  So, have test 32 cause recovery on the client and wait
a bit before finishing.

21 years ago- Don't wake up the pinger, if it's not enabled
rread [Fri, 14 Nov 2003 23:55:42 +0000 (23:55 +0000)]
- Don't wake up the pinger, if it's not enabled

21 years agoMake sure lwt_event_t packs into a page evenly on ia64.
adilger [Fri, 14 Nov 2003 23:09:31 +0000 (23:09 +0000)]
Make sure lwt_event_t packs into a page evenly on ia64.

21 years agoSome more wire constant checks.
adilger [Fri, 14 Nov 2003 23:03:25 +0000 (23:03 +0000)]
Some more wire constant checks.
b=1321

21 years agoMore test cleanups.
rread [Fri, 14 Nov 2003 22:32:24 +0000 (22:32 +0000)]
More test cleanups.

21 years agoWake up the pinger after adding an import, to reset the timer.
rread [Fri, 14 Nov 2003 22:04:23 +0000 (22:04 +0000)]
Wake up the pinger after adding an import, to reset the timer.

21 years agoRe-add lgd_len in case it is needed at some point (conflict resolution error).
adilger [Fri, 14 Nov 2003 21:49:34 +0000 (21:49 +0000)]
Re-add lgd_len in case it is needed at some point (conflict resolution error).
b=1821

21 years ago- remove reference to lgd_len, which has disappeared
rread [Fri, 14 Nov 2003 20:52:11 +0000 (20:52 +0000)]
- remove reference to lgd_len, which has disappeared

21 years agoInteroperability for different PAGE_SIZE/wordsize machines. Tested on ia64
adilger [Fri, 14 Nov 2003 20:12:30 +0000 (20:12 +0000)]
Interoperability for different PAGE_SIZE/wordsize machines.  Tested on ia64
and i386 separately, and with ia64 client + i386 MDS/OST.

Mostly aligning structs to have 64-bit fields aligned on 64-bit boundaries.
Remove some VFS constants and replace them with Lustre constants on the wire.
Since the MDS doesn't really open files itself, we don't need to convert from
the wire constants back to local flags at all.

Frobbing of niobufs on the targets to split them into PAGE_SIZE chunks (this
may be a problem on large PAGE_SIZE servers with small PAGE_SIZE clients,
not sure yet).

I have tested and this appears to be compatible with old filesystems.
At worst we should only need another --write_conf on the MDS.

b=686, b=1821, b=1343, b=2042

21 years agoUse PTL_MD_MAX_PAGES instead of PTL_MD_MAX_IOV where needed, so we send the
adilger [Fri, 14 Nov 2003 19:27:01 +0000 (19:27 +0000)]
Use PTL_MD_MAX_PAGES instead of PTL_MD_MAX_IOV where needed, so we send the
right number of pages over the wire for ia64/i386 interoperability.
b=1821

21 years agoliblustre: update lrun script.
ericm [Fri, 14 Nov 2003 15:05:48 +0000 (15:05 +0000)]
liblustre: update lrun script.

21 years agoliblustre:
ericm [Fri, 14 Nov 2003 11:13:26 +0000 (11:13 +0000)]
liblustre:
  add more test items into replay_single.

21 years agoliblustre: add more test in replay_single, and some other adjustment.
ericm [Fri, 14 Nov 2003 09:37:33 +0000 (09:37 +0000)]
liblustre: add more test in replay_single, and some other adjustment.

21 years agoI touched one thing in lustre_mds.h, and I was dismayed to see the
phil [Fri, 14 Nov 2003 09:24:31 +0000 (09:24 +0000)]
I touched one thing in lustre_mds.h, and I was dismayed to see the
entire source tree rebuild!

It took just a few minutes to remove a number of #includes which
violated all manner of abstraction boundary.  ericm assures me that I
have not broken anything major in liblustre.

Also, I moved a good chunk of inline functions into llite_internal.h
(which is where I secretly believe that most of lustre_lite.h will end up),
and renamed ll_ino2fid (which no longer takes an inode) to mdc_pack_fid.

21 years ago- fix del_route
rread [Fri, 14 Nov 2003 01:12:31 +0000 (01:12 +0000)]
- fix del_route

21 years ago- update this patch to apply against vanilla-2.4.22
alex [Thu, 13 Nov 2003 20:49:31 +0000 (20:49 +0000)]
- update this patch to apply against vanilla-2.4.22

21 years agoSilence most of the warnings in the x86_64 build, and kill some Real Bugs in
zab [Thu, 13 Nov 2003 20:21:48 +0000 (20:21 +0000)]
Silence most of the warnings in the x86_64 build, and kill some Real Bugs in
the process.  (sanity checked in an x86 uml build, as well)

- include linux/init.h from some modules that use __init and company
- x86_64 has 64 bit longs but uses 'long long' for u64 in the kernel.  we
  add its own specific section to LP*64 definitions.  Sadly, there are still
  a huge number of warnings from this gcc when we try to LP*64 print uint64_t
  obdo arguments.
- use min_t in some defines to avoid the duplicate const prefacing that comes
  with nesting min()
- LPX64 printing -1UL doesn't work so well
- i_nlink can be any of unsigned short, int, or long.  when printing cast
  to unsigned long.
- kern_addr_valid takes a 'unsigned long' argument and shifts it.. sigh.
- don't #define EXPORT_SYMTAB in the source, that's the job of the kernel
  makefiles
- x86_64/suse has THREAD_SIZE defined in userspace
- s/long flags/unsigned long flags/.  Again.
- va_list ap = NULL; is not legal, as far as I know.  it doesn't seem
  to actually do anything in this code path either.
- x86_64 passes CONFIG_X86 but had thread->rsp as the stack pointer.  this
  code is going away eventually anyway.
- print a pointer with %p, not %x
- some portals callback pointer definitions used size_t while some nal's
  functions used ptl_size_t in the function defintions.  I switched the
  functions over to avoid pushing changes on all the nals.

21 years ago- ide_end_request() was used wrong way: fixed
alex [Thu, 13 Nov 2003 15:30:09 +0000 (15:30 +0000)]
- ide_end_request() was used wrong way: fixed

NOTE: tested on the all related series

21 years ago- kmem_cache_validate patch has been removed from all the series
alex [Thu, 13 Nov 2003 15:06:15 +0000 (15:06 +0000)]
- kmem_cache_validate patch has been removed from all the series

NOTE: all the supported series still build

21 years agoliblustre: zconf adjustment for recovery_small and replay_single.
ericm [Thu, 13 Nov 2003 13:01:48 +0000 (13:01 +0000)]
liblustre: zconf adjustment for recovery_small and replay_single.