Whamcloud - gitweb
adilger [Wed, 26 Oct 2005 18:35:46 +0000 (18:35 +0000)]
Branch b1_4
Don't compare .xml mtimes, they will always be different.
Split "long UUID" and "consistent .xml" into separate subtests.
nathan [Wed, 26 Oct 2005 16:46:51 +0000 (16:46 +0000)]
Branch b1_4
(lin)print lov_setup info
adilger [Wed, 26 Oct 2005 09:29:33 +0000 (09:29 +0000)]
Branch b1_4
Description: When migrating a subset of services from a node (e.g. failback
from a failover service node) the remaining services would
time out and evict clients.
Details : lconf --force (implied by --failover) sets the global obd_timeout
to 5 seconds in order to quickly disconnect, but this caused
other RPCs to time out too quickly. Do not change the global
obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
b=6395, b=9514
adilger [Wed, 26 Oct 2005 00:30:10 +0000 (00:30 +0000)]
Branch b1_4
Use "if (likely(!ext3_mb_agressive))" since it is likely that this is not set
(it defaults to off). Also, check this first since it is the likely case.
nic [Tue, 25 Oct 2005 22:54:10 +0000 (22:54 +0000)]
p=alex
- fix for write performance slowdown on 2.6 kernels. Return early when not in
'aggressive' mode and default aggressive mode to off
adilger [Tue, 25 Oct 2005 19:57:43 +0000 (19:57 +0000)]
Branch b1_4
Don't get ll_inode_size_lock() in ll_update_inode() as this can be called
with inode_lock (spinlock) held and deadlock. This was protecting the
setting of lli_smd to prevent ll_inode_size_unlock() from inconsistently
calling lov_stripe_unlock() when it was never locked because lli_smd changed
since ll_inode_size_lock() was called.
We now avoid this race by only ever calling ll_inode_size_lock() with
lli_smd already set, or with "lock_lsm = 0" so we don't care if it changes
between lock and unlock. This makes sense in any case, because if there is
no lli_smd we shouldn't be doing glimpse/enqueue on the OSTs anyways.
b=9547
r=nikita
nathan [Tue, 25 Oct 2005 19:37:04 +0000 (19:37 +0000)]
Branch b1_4
b=9477
r=adilger
- robustify mtime check
- forgot to add back in part of --service= checks
nathan [Tue, 25 Oct 2005 18:22:05 +0000 (18:22 +0000)]
Branch b1_4
b=8080
r=adilger
Change magic for new lmd
adilger [Tue, 25 Oct 2005 06:39:23 +0000 (06:39 +0000)]
Branch b1_4
Fix bug 9482 regression for fix to 2.6 llap_shrink_cache() page cleanup.
b=6450, b=9482
r=green
ericm [Mon, 24 Oct 2005 17:09:22 +0000 (17:09 +0000)]
branch: b1_4
lconf recognize option --user_xattr (fake) to make ltest happy.
brian [Mon, 24 Oct 2005 16:27:30 +0000 (16:27 +0000)]
Make release 1.4.5.92 on current b1_4 head.
eeb [Sun, 23 Oct 2005 14:08:15 +0000 (14:08 +0000)]
* socklnd:
fixed connection race that can occur with multiple routers
changed 'typed_conns' module parameter to RO
* nidstrings:
allow 0xnnnn parsing of numerical NIDs
libcfs_num_addr2str -> libcfs_decnum_addr2str (LO, QSW, PTL)
libcfs_hexnum_addr2str (GM)
* gmlnd:
change from decimal to hex representation of GM addresses (they're the
lowest 4 bytes of the NIC's MAC address).
* router:
compare routers first on # uncompleted bytes, then credits
for better load balance.
* llmount:
fixed error message (it's a NID, not a host)
fixed bug in checking mdx & profile string lengths
adilger [Sat, 22 Oct 2005 07:36:18 +0000 (07:36 +0000)]
Branch b1_4
Save "options" from the wrath of strtok() so we can save them into /etc/mtab.
This is needed for buffalo testing of new features, among other things.
pjkirner [Fri, 21 Oct 2005 13:48:57 +0000 (13:48 +0000)]
* Make echo_test a proper liblustre test, now that it works even on Catamount!
nikita [Fri, 21 Oct 2005 10:34:56 +0000 (10:34 +0000)]
fix unnecessary line-wrap
nikita [Fri, 21 Oct 2005 10:13:09 +0000 (10:13 +0000)]
ENTRY has to be matched by RETURN()
niu [Fri, 21 Oct 2005 08:33:59 +0000 (08:33 +0000)]
Disable compile userspace quota stuff for b1_4. This is short-term solution
for the <linux/quota.h> problem before b1_4_quota lands.
-b 9542
-r adilger
nathan [Thu, 20 Oct 2005 19:17:37 +0000 (19:17 +0000)]
Branch b1_4
allow auto-loading of lustre/LNET modules
nathan [Thu, 20 Oct 2005 18:23:35 +0000 (18:23 +0000)]
Branch b1_4
b=none
fix 'make dist' err with llog_reader
pjkirner [Thu, 20 Oct 2005 02:13:54 +0000 (02:13 +0000)]
* Added command line support for liblustre echo_test
* Added "hack" to allow echo_test to build on Catamount
green [Wed, 19 Oct 2005 18:58:42 +0000 (18:58 +0000)]
b=9482
r=adilger
Check that there is no page_mapped() before trying to define it for 2.4 kernels.
nathan [Wed, 19 Oct 2005 00:41:25 +0000 (00:41 +0000)]
Branch b1_4
b=9477
These lines should not have been removed in the 9477 patch. They prevent
--select from limiting services in linux 2.4
nikita [Tue, 18 Oct 2005 19:27:26 +0000 (19:27 +0000)]
Forgotten chunks of 7133 patch.
nathan [Tue, 18 Oct 2005 19:14:35 +0000 (19:14 +0000)]
Branch b1_4
b=9477
exempt bad/incomplete nids from network ping test
nathan [Tue, 18 Oct 2005 18:26:37 +0000 (18:26 +0000)]
b=8080
- modules aren't as automatic as they should be for linux 2.4
- for some reason python in linux 2.4 didn't understand this lmc construct
nathan [Tue, 18 Oct 2005 16:38:26 +0000 (16:38 +0000)]
b=8080
Landing LNET (b1_4_newconfig)
green [Mon, 17 Oct 2005 21:26:53 +0000 (21:26 +0000)]
Branch: b1_4
b=9482
r=adilger
Try to unmap pages before discarding them in llap_shrink_cache.
nikita [Sat, 15 Oct 2005 17:26:04 +0000 (17:26 +0000)]
An optimization proposed by Andreas: do not grow (as part of ldlm lock policy)
extent locks acquired on server (e.g., OST-side locks introduced by previous
7311 fixes): server-side locks are not cached and would only conflict with
other threads.
b=7311
r=adilger
green [Fri, 14 Oct 2005 20:07:42 +0000 (20:07 +0000)]
b=7293
r=adilger
Add possibility (config option) to show minimal available OST free space
alex [Fri, 14 Oct 2005 11:36:42 +0000 (11:36 +0000)]
b=7314
r=adilger,alex (original patch from Brian Behlendorf)
- adds ldiskfs tunnables for mballoc
adilger [Thu, 13 Oct 2005 23:04:41 +0000 (23:04 +0000)]
Branch b1_4
Update build version to 1.4.5.8, last release before LNET/b1_4_neconfig landing.
adilger [Thu, 13 Oct 2005 23:02:31 +0000 (23:02 +0000)]
Branch b1_4
Missing locking for direct IO.
Disable locking check for liblustre, it has many callsites that do not lock
and lov_stripe_{un,}lock() are noops there anyways.
b=5047
ericm [Thu, 13 Oct 2005 22:39:36 +0000 (22:39 +0000)]
branch: b1_4
xattr: minor fix, add comment.
adilger [Thu, 13 Oct 2005 22:37:23 +0000 (22:37 +0000)]
Branch b1_4
- add user_xattr to lov.sh, uml.sh client mount options
- don't run xattr sanity test 102 when not root or mounted with user_xattr
- add nouser_xattr mount option to allow user_xattr to be unset
- add check for linux/xattr_acl.h + compat (doesn't exist on BG/L)
- fix test 56, 65a with default striping on parent directory
- add ChangeLog entry for gmnalnid support to llmount.c
- add ChangeLog entry for xattr support
b=7979, b=8592, b=9504, b=9505
cliffw [Thu, 13 Oct 2005 22:12:10 +0000 (22:12 +0000)]
b=9508
r=adilger@clusterfs.com
Reverted 2.4 kernels
cliffw [Thu, 13 Oct 2005 21:53:49 +0000 (21:53 +0000)]
b=9508
r=adilger@clusterfs.com
When both gcc33 and gcc32 are available, we should use gcc33
nikita [Thu, 13 Oct 2005 14:39:20 +0000 (14:39 +0000)]
move misplaced comment.
ericm [Thu, 13 Oct 2005 05:23:42 +0000 (05:23 +0000)]
branch: b1_4
land b1_4_xattr: support manipulating user extended attributes.
nikita [Wed, 12 Oct 2005 20:46:16 +0000 (20:46 +0000)]
cleanup llap_from_page():
- add explicit LLAP_FROM_REMOVEPAGE
- make llap_from_page() static
- fix inverted condition in llap_from_page().
b=5047
r=adilger
nathan [Wed, 12 Oct 2005 20:45:30 +0000 (20:45 +0000)]
Branch b1_4
b=9445
nikita [Wed, 12 Oct 2005 19:02:39 +0000 (19:02 +0000)]
remove unneeded conditional compilation wrappers.
nic [Wed, 12 Oct 2005 18:11:47 +0000 (18:11 +0000)]
b=7047
p=adilger
r=nic
fix typo that was preventing lconf --abort_recovery from working
nic [Wed, 12 Oct 2005 17:58:10 +0000 (17:58 +0000)]
b=7047
p=adilger
r=nic
fix typo that was preventing lconf --abort_recovery from working
nikita [Wed, 12 Oct 2005 10:59:38 +0000 (10:59 +0000)]
typo fix.
nikita [Wed, 12 Oct 2005 10:55:00 +0000 (10:55 +0000)]
Add locking to provide consistency between kms and lsm.
b=5047
r=nikita
r=adilger
nic [Mon, 10 Oct 2005 23:59:13 +0000 (23:59 +0000)]
make sure we actually build the drivers...
nic [Mon, 10 Oct 2005 22:16:12 +0000 (22:16 +0000)]
add qsnet patch for 2.6-rhel4
nic [Mon, 10 Oct 2005 20:22:44 +0000 (20:22 +0000)]
update to latest update from Suse.
nikita [Mon, 10 Oct 2005 06:52:34 +0000 (06:52 +0000)]
check returned value
nikita [Mon, 10 Oct 2005 06:36:45 +0000 (06:36 +0000)]
liblustre/tests/sanity.c: add test 51 to test for regression in
ldlm_cli_enqueue() introduced by 7311 fix.
nikita [Sun, 9 Oct 2005 20:32:40 +0000 (20:32 +0000)]
Fix wrong assertion added to ldlm_cli_enqueue() by patch from bug 7311. Also
fix few outdated references in comments.
b=7311
r=adilger
nathan [Mon, 3 Oct 2005 19:31:27 +0000 (19:31 +0000)]
b=3289
r=adilger
report recovery time remaining in /proc/.../recovery
adilger [Sat, 1 Oct 2005 06:09:53 +0000 (06:09 +0000)]
Branch b1_4
Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
ptlrpc_import_delayed_req() because we didn't check if the
request was interrupted, and we also didn't break out of the
event loop if there was no timeout.
__l_wait_event() changes match those recently made in HEAD.
b=7184
r=devesh
nathan [Fri, 30 Sep 2005 22:40:52 +0000 (22:40 +0000)]
b=9445
r=adilger
remove mds and client cleanup logs
adilger [Fri, 30 Sep 2005 18:19:13 +0000 (18:19 +0000)]
Branch b1_4
Update build version to 1.4.5.7.
nikita [Fri, 30 Sep 2005 10:54:45 +0000 (10:54 +0000)]
reorder changelog entries: new entries go to the end of current release list
nikita [Fri, 30 Sep 2005 10:51:05 +0000 (10:51 +0000)]
remove assertion that does not buy us much, while potentially hampering interoperability
adilger [Fri, 30 Sep 2005 10:37:05 +0000 (10:37 +0000)]
Branch b1_4
Shouldn't have been committed.
b=7342
adilger [Fri, 30 Sep 2005 10:26:10 +0000 (10:26 +0000)]
Branch b1_4
Description: bind OST threads to NUMA nodes to improve performance
Details : all OST threads are uniformly bound to CPUs on a single NUMA
node and do their allocations there to localize memory access
b=7342
adilger [Fri, 30 Sep 2005 10:10:20 +0000 (10:10 +0000)]
Branch b1_4
Use actual page size instead of hard-coded 4096 bytes.
nathan [Fri, 30 Sep 2005 00:34:22 +0000 (00:34 +0000)]
Branch b1_4
fix for nonexistent modules.conf
adilger [Thu, 29 Sep 2005 23:43:28 +0000 (23:43 +0000)]
Branch b1_4
Description: lconf did not handle in-kernel recovery with LDAP properly
Details : lconf/LustreDB get_refs() is searching the wrong namespace
b=6163
adilger [Thu, 29 Sep 2005 23:22:27 +0000 (23:22 +0000)]
Branch b1_4
Remove left-over function (replaced by ldlm_glimpse_ast() in latest patch).
b=7311
adilger [Thu, 29 Sep 2005 21:29:30 +0000 (21:29 +0000)]
Branch b1_4
Description: unable to set striping with a starting offset beyond OST 160
Details : llapi_create_file() incorrectly limited the starting stripe
index to the maximum single-file stripe count.
b=9440
r=behlendo
nathan [Thu, 29 Sep 2005 15:34:38 +0000 (15:34 +0000)]
b=9428
r=adilger
at initial connect, try all failover servers quickly
nikita [Thu, 29 Sep 2005 13:20:56 +0000 (13:20 +0000)]
oops... revert chunk committed by mistake (some local debugging stuff).
nikita [Thu, 29 Sep 2005 12:58:24 +0000 (12:58 +0000)]
Latest OST-side locking with connection flags.
b=7311
r=adiler
adilger [Thu, 29 Sep 2005 06:37:35 +0000 (06:37 +0000)]
Branch b1_4
Description: server may evict liblustre clients accessing contended locks
Details : if a client is granted a lock or receives a lock completion AST
with a blocking AST pending it would not reply to the AST for
LDLM_FL_CANCEL_ON_BLOCK locks causing the server to time out on
the AST (it only cancels when sending an explicit blocking AST).
If enough such locks were processed it would cause clients to
be evicted. It now replies to such ASTs and cancels when done.
b=9352, b=7313
r=nikita, bogl
adilger [Wed, 28 Sep 2005 18:26:07 +0000 (18:26 +0000)]
Branch b1_4
Remove last vestiges of "mgmt" and "mgmt_cli", which was never used. The
lustre/mgmt directory has been an empty shell for a long time.
The new mountconfig code uses "mgc" and "mgs".
adilger [Tue, 27 Sep 2005 23:46:59 +0000 (23:46 +0000)]
Branch b1_4
Description: MDS may oops in groups_free()
Details : in rare race conditions a newly allocated group_info struct is
freed again, and this can be NULL. The 2.4 compatibility code
for groups_free() checked for a NULL pointer, but 2.6 did not.
b=7273
adilger [Tue, 27 Sep 2005 23:42:47 +0000 (23:42 +0000)]
Branch b1_4
Lustre fixes for compiling against a 2.6.12 kernel from Bull.
b=6864
adilger [Tue, 27 Sep 2005 22:16:07 +0000 (22:16 +0000)]
Branch b1_4
Lustre fixes for compiling against a 2.6.12 kernel from Bull.
b=6864
adilger [Tue, 27 Sep 2005 20:14:54 +0000 (20:14 +0000)]
Branch b1_4
Fix stripe test program to properly handle filesystems with default stripe
count = -1 (which should result in a full-OST striping).
b=9359
adilger [Tue, 27 Sep 2005 19:41:39 +0000 (19:41 +0000)]
Branch b1_4
Add liblustre_wait_event() calls before entering all liblustre API functions
to ensure that pending ASTs from LDLM_FL_CANCEL_ON_BLOCK locks are handled
before we do any local lock matching. Also add liblustre_wait_event() calls
just before exiting Lustre code to handle any remaining items before returning
to the uninterruptible client code.
b=9352, b=7313
r=green
adilger [Tue, 27 Sep 2005 00:11:34 +0000 (00:11 +0000)]
Branch b1_4
Don't try to walk directory default EA with "lfs find --obd ...".
Tested by HP.
b=9382
adilger [Mon, 26 Sep 2005 23:09:42 +0000 (23:09 +0000)]
Branch b1_4
Put LDLM_FL definitions in numerical order to avoid potential duplication
of values.
b=7313
eeb [Mon, 26 Sep 2005 08:30:15 +0000 (08:30 +0000)]
* GM zeroconf mount fixes
eeb [Mon, 26 Sep 2005 08:10:06 +0000 (08:10 +0000)]
* Backed out previous commit; it included the wrong files
eeb [Mon, 26 Sep 2005 08:01:22 +0000 (08:01 +0000)]
* Added vibnal arp patch from 8206
cliffw [Fri, 23 Sep 2005 18:18:44 +0000 (18:18 +0000)]
GM nal for Sandia
adilger [Fri, 23 Sep 2005 17:20:26 +0000 (17:20 +0000)]
Branch b1_4
Update build version for 1.4.5.5 tag
adilger [Fri, 23 Sep 2005 09:34:07 +0000 (09:34 +0000)]
Branch b1_4
Create liblustre test files with O_LARGEFILE so they can grown > 2GB.
Clean up the t23 test file.
b=9339
adilger [Fri, 23 Sep 2005 09:02:49 +0000 (09:02 +0000)]
Branch b1_4
Description: improve by-nid export eviction on the MDS and OST
Details : allow multiple exports with the same NID to be evicted at one
time without re-searching the exports list.
b=7304
r=green, tested at Sandia
adilger [Wed, 21 Sep 2005 17:45:05 +0000 (17:45 +0000)]
Branch b1_4
Update usage and docs for lfs setstripe -d.
adilger [Wed, 21 Sep 2005 07:55:37 +0000 (07:55 +0000)]
Branch b1_4
b=8322
r=nathan
Description: OST or MDS may oops in ping_evictor_main()
Details : ping_evictor_main() drops obd_dev_lock if deleting a stale
export but doesn't restart at beginning of obd_exports_timed
list afterward.
The list_for_each_safe() macro is only safe for the removal of the current
entry and not safe if some other entry (in particular the next one)
is removed. As class_fail_export() will immediately result in the export
being removed from the obd_exports_timed list (via class_unlink_export())
we are OK to restart processing at the start of the list each time.
The extra pet_lock around pet_exp references in code are not strictly
necessary, but rather precautionary and for consistency when accessing
pet_exp.
adilger [Wed, 21 Sep 2005 07:16:08 +0000 (07:16 +0000)]
Branch b1_4
Description: Creating more than 1000 files for a single job may cause a load
imbalance on the OSTs if there are also a large number of OSTs.
Details : qos_prep_create() uses an OST index reseed value that is an
even multiple of the number of available OSTs so that if the
reseed happens in the middle of the object allocation it will
still utilize the OSTs as uniformly as possible.
b=8330
r=behlendorf (tested on BG/L)
adilger [Wed, 21 Sep 2005 07:11:53 +0000 (07:11 +0000)]
Branch b1_4
Put 3min timelimit on random-read sanity test for slow systems (ala UML).
b=6252
adilger [Wed, 21 Sep 2005 06:22:46 +0000 (06:22 +0000)]
Branch b1_4
Improve error message if fsfilt_ext3_write_record fails to start a transaction.
b=8317 (debugging of)
r=phil
adilger [Tue, 20 Sep 2005 23:55:30 +0000 (23:55 +0000)]
Branch b1_4
Add definition for pgoff_t, which isn't defined for 2.4 kernels.
b=6252
nikita [Tue, 20 Sep 2005 13:24:54 +0000 (13:24 +0000)]
Land changes to the read-ahead algorithm improving its behavior for random
reads:
- always try to read-ahead at least file region that will be read by read(2)
call.
- try to detect random reads, and avoid excessive read-ahead in that case.
b=6252
r=adilger
nikita [Mon, 19 Sep 2005 14:09:16 +0000 (14:09 +0000)]
A script to run kernel builds as a benchmark.
adilger [Thu, 15 Sep 2005 19:08:53 +0000 (19:08 +0000)]
Branch b1_4
Fix liblustre sanity unaligned write test. Was previously not really testing
the 1TB write offset because of 32-bit lseek offset truncation.
b=7279
nikita [Thu, 15 Sep 2005 08:19:02 +0000 (08:19 +0000)]
use local variable
adilger [Wed, 14 Sep 2005 21:45:49 +0000 (21:45 +0000)]
Branch b1_4
Add support for F_{GET,SET}LK{,W}64 to fcntl because the t23() use of
_FILE_OFFSET_BITS=64 caused these macros to be changed in the header.
b=7279
adilger [Wed, 14 Sep 2005 21:18:37 +0000 (21:18 +0000)]
Branch b1_4
Use 64-bit variable for libsysio lseek64() internal return value.
Update liblustre sanity.c to use 64-bit IO functions where needed.
b=7279
adilger [Wed, 14 Sep 2005 20:46:52 +0000 (20:46 +0000)]
Branch b1_4
Move ChangeLog comment to end of 1.4.6 release notes.
Fix CFS_PAGE_MASK for extent start in ost_get_extent_lock().
Add LDLM_FL_CBPENDING for all granted liblustre extent locks.
Add OBD_CONNECT_SRVLOCK for OST only.
Add misc comments from final patch.
b=7311
r=nikita (original patch)
nikita [Wed, 14 Sep 2005 10:42:52 +0000 (10:42 +0000)]
version of 7311 fix, tested by Cray, with few minor modifications from newer
not-yet tested version:
- comments;
- uninitialized .l_extent in lustre_build_lock_params();
- extent lock boundaries are better to be page aligned in ost_brw_lock_get().
b=7311
r=adilger
adilger [Wed, 14 Sep 2005 08:56:29 +0000 (08:56 +0000)]
Branch b1_4
Update liblustre 2GB lseek test to also do SEEK_SET 0 afterward (per bug).
I'm unable to reproduce a problem with this test on 2.6.9.
b=7279
green [Tue, 13 Sep 2005 16:53:20 +0000 (16:53 +0000)]
Branch: b1_4
b=7313
r=adilger
Allow locks flagged in a certain way (used by liblustre) to be cancelled without
waiting for reply from client.
This is a prototype version of code. We land it so that it will get into next
code drop to Cray.
adilger [Tue, 13 Sep 2005 16:13:35 +0000 (16:13 +0000)]
Branch b1_4
Remove lower limit on ldlm_timeout value for liblustre. Tested at Sandia.
b=7201