Whamcloud - gitweb
adilger [Mon, 30 May 2005 19:32:41 +0000 (19:32 +0000)]
Branch b1_4
Change lib64 search order to avoid stale .pyc files.
adilger [Sun, 29 May 2005 08:23:38 +0000 (08:23 +0000)]
Branch b1_4
Change build version to 1.4.2.4
green [Sun, 29 May 2005 07:56:25 +0000 (07:56 +0000)]
b=3790
Changelog update
green [Sun, 29 May 2005 07:51:24 +0000 (07:51 +0000)]
Branch: b1_4
b=3790
r=adilger
Introudce fsprivate member of struct file, to be used as filesystem's private data pointer
green [Sat, 28 May 2005 21:31:08 +0000 (21:31 +0000)]
Branch: b1_4
Instead of commenting out all LL_CDEBUG_PAGE usage, make LL_CDEBUG_PAGE
to work both in kernel and not.
jacob [Fri, 27 May 2005 22:45:48 +0000 (22:45 +0000)]
b=6412
check in the right file.
Severity : major
Frequency : occasional
Bugzilla : 6412
Description: Nodes with an elan id of 0 trigger a failed assertion
jacob [Fri, 27 May 2005 22:41:57 +0000 (22:41 +0000)]
b=6412
Severity : major
Frequency : occasional
Bugzilla : 6412
Description: Nodes with an elan id of 0 trigger a failed assertion
jacob [Fri, 27 May 2005 22:28:59 +0000 (22:28 +0000)]
b=6286
r=adilger
Severity : enhancement
Bugzilla : 6296
Description: Default groups are now added by lconf
Details : You can now run lconf --group <servicename> without having to
manually add groups with lmc.
jacob [Fri, 27 May 2005 21:34:59 +0000 (21:34 +0000)]
b=6377
r=adilger
Use obd_recovering instead of obd_recoverable_clients to check if the
obd is in recovery.
abort_recovery didn't set obd_recoverable_clients to 0 (which it now
does), so if recovery was aborted, the server would never ping-evict
clients (which can lead to extra exports, possibly requiring recovery
to be aborted).
jacob [Fri, 27 May 2005 21:22:33 +0000 (21:22 +0000)]
remove liblustre change entry. it's not enabled by default, and the
few customers whom it affects already know about it.
adilger [Fri, 27 May 2005 11:56:46 +0000 (11:56 +0000)]
Branch b1_4
Don't LASSERT OST in ptlrpc if a client sends a bad bulk request.
b=6345
adilger [Fri, 27 May 2005 11:53:38 +0000 (11:53 +0000)]
Branch b1_4 - merge from b_cray
Remove unused growth_count set_info.
adilger [Fri, 27 May 2005 11:40:06 +0000 (11:40 +0000)]
Branch b1_4
Make very verbose message an LCONSOLE_ERROR() for future use.
adilger [Fri, 27 May 2005 11:38:29 +0000 (11:38 +0000)]
Branch b1_4 - merge from b_cray
Use l_export_chain instead of l_pending chain on client for SYNC
ldlm_cancel_lru cancellations.
b=5666
adilger [Fri, 27 May 2005 11:35:45 +0000 (11:35 +0000)]
Branch b1_4
Initialize (currently unused) l_childof member of lock struct.
adilger [Fri, 27 May 2005 11:30:57 +0000 (11:30 +0000)]
Branch b1_4
Don't return an "error" (pid) from ptlrpc_import_recovery_state_machine()
from kernel_thread().
r=alex (original fix)
adilger [Fri, 27 May 2005 11:17:41 +0000 (11:17 +0000)]
Branch b1_4
Always include liblustre in DIST_SUBDIRS
adilger [Fri, 27 May 2005 11:10:03 +0000 (11:10 +0000)]
Branch b1_4
Fix .netmatch method to work with older Pythons.
Only reduce OST inode counts for larger OSTs.
b=5044
adilger [Thu, 26 May 2005 23:33:14 +0000 (23:33 +0000)]
Branch b1_4
Don't complain about llog padding record swabbing.
green [Thu, 26 May 2005 21:07:30 +0000 (21:07 +0000)]
Branch: b1_4
Liblustre revival.
Liblustre can be compiled and works now.
lwang [Thu, 26 May 2005 09:23:00 +0000 (09:23 +0000)]
5541: extra peer list
adilger [Wed, 25 May 2005 19:38:15 +0000 (19:38 +0000)]
Branch b1_4
Quiet spurious lov_iocontrol() error messages for PPC64 TCGETS, et. al. ioctls.
Print file offsets in both decimal and hex for debugging.
b=5947
jacob [Wed, 25 May 2005 15:57:45 +0000 (15:57 +0000)]
b=6371
r=adilger, r=phil
Severity : major
Frequency : rare
Bugzilla : 6371
Description: After recovery, certain operations trigger a failed
assertion on a client.
Details : Failing over an mds, using lconf -d --failover, while a
client was doing a readdir() call would cause the client to
LBUG after recovery completed and the readdir() was resent.
adilger [Wed, 25 May 2005 15:33:38 +0000 (15:33 +0000)]
Branch b1_4_bug3389
Move upcall stuff to lustre/lvfs/upcall_cache.c instead of mds_groups.c.
Rename all of the functions upcall_cache_* to match HEAD.
Pass upcall_cache * as parameter instead of mds_obd *.
Improve some error messages.
b=3389
b=6367 (default striping on root dir fix included)
r=phil
eeb [Wed, 25 May 2005 15:05:59 +0000 (15:05 +0000)]
* changed __clear_page_ll_data(page) to make page_cache_release(page)
the last thing it does.
adilger [Tue, 24 May 2005 22:24:48 +0000 (22:24 +0000)]
Branch b1_4
Comments and other purely cosmetic fixes.
adilger [Tue, 24 May 2005 22:04:05 +0000 (22:04 +0000)]
Branch b1_4
Comment what med_open_lock is locking.
adilger [Tue, 24 May 2005 16:03:45 +0000 (16:03 +0000)]
Branch b1_4_bug3389
Add support for supplementary groups via upcall on the MDS.
b=3389, b=6253
adilger [Sat, 21 May 2005 08:20:35 +0000 (08:20 +0000)]
Branch b1_4
Use 64-bit interface for truncate, so that it is possible to ./truncate
files larger than 2GB in UML.
adilger [Sat, 21 May 2005 08:16:46 +0000 (08:16 +0000)]
Branch b1_4
Print out numeric device if no name can be found for 2.6.
r=nathan
jacob [Fri, 20 May 2005 23:26:29 +0000 (23:26 +0000)]
fix formatting
jacob [Fri, 20 May 2005 23:25:36 +0000 (23:25 +0000)]
b=6359
r=adilger, phil
Severity: : enhancement
Bugzilla : 3262, 6359
Description: Attempts to reconnect to servers are now more aggressive.
Details : This builds on the enhanced upcall-less recovery that was added
in 1.4.2. When trying to reconnect to servers, clients will
now try each server in the failover group every 10 seconds. By
default, clients would previously try one server every 25 seconds.
adilger [Fri, 20 May 2005 23:07:42 +0000 (23:07 +0000)]
Branch b1_4
Do it right.
adilger [Fri, 20 May 2005 23:06:35 +0000 (23:06 +0000)]
Branch b1_4
Increase fsx size for acceptance-small.sh to something reasonable.
phil [Fri, 20 May 2005 22:39:47 +0000 (22:39 +0000)]
add the now-required -DLUSTRE_UTILS=1 to the Lustre build as well
jacob [Fri, 20 May 2005 20:47:48 +0000 (20:47 +0000)]
print more useful things
adilger [Fri, 20 May 2005 20:16:27 +0000 (20:16 +0000)]
Branch b1_4
Print a useful error on failure.
adilger [Fri, 20 May 2005 17:38:57 +0000 (17:38 +0000)]
Branch b1_4
Remove some extraneous logic in mdc_rename_pack(). We already check this
same thing in ll_i2uctxt() as called by ll{,u}_prepare_mdc_op_data() from
ll{,u}_rename_raw(). Verified with POSIX this has no effect.
adilger [Fri, 20 May 2005 17:17:53 +0000 (17:17 +0000)]
Branch b1_4
Remove unused and duplicating-function file simple.c. All of this has
been moved into lvfs*.c
jacob [Thu, 19 May 2005 20:56:34 +0000 (20:56 +0000)]
b=4855
r=phil (original patch)
handle crc32_le being a module or disabled.
jacob [Thu, 19 May 2005 17:35:56 +0000 (17:35 +0000)]
b=6349
r=adilger
print the nid instead of the export uuid (if available) when
ping-evicting clients
adilger [Wed, 18 May 2005 17:11:14 +0000 (17:11 +0000)]
Branch b1_4
It was possible to get multiple mfd references during close and client
eviction, leading to either mds_close() or mds_mfd_close() referencing
a freed mfd.
Now we remove the mfd from the mfd_list and the handle hash under lock so
that once it starts on the road to destruction it is not possible to get a
new reference to it. Also add comments on usage of mfd-related functions!
b=3819, b=4364, b=4397, b=6313
r=green
jacob [Wed, 18 May 2005 00:51:54 +0000 (00:51 +0000)]
r=phil
stop lying just to be popular.
phil [Tue, 17 May 2005 16:25:58 +0000 (16:25 +0000)]
lsy reported that LL_CDEBUG_PAGE in osc_checksum_bulk wouldn't compile,
because the outer scope defines a variable "page_count", and LL_CDEBUG_PAGE
wants to call the VM function page_count().
I still can't figure out why it works fine for me, but I've stopped trying.
Changing the name of the variable is easier.
lwang [Tue, 17 May 2005 08:43:37 +0000 (08:43 +0000)]
fix up for readline (the old one can not handle long input)
phil [Tue, 17 May 2005 04:04:09 +0000 (04:04 +0000)]
b=5902
r=adilger
Many improvements to the code that checksums I/O, among them:
- add a switch to enable or disable them at runtime
- check the pages while under llite's control, and the buffers while under
the osc/ost's
- if the server and client see different checksums, try to figure out where
it went wrong
- use the kernel crc32 routine
phil [Tue, 17 May 2005 03:41:19 +0000 (03:41 +0000)]
b=6200
A bug in MDS/OSS recovery could cause the OSS to fail a (ctxt == NULL)
assertion.
There's little harm in just aborting MDS/OSS recovery and letting
it try again next time, so I removed the LASSERT and return an error instead.
We can revisit it when we put the knife in the llog code for real.
adilger [Mon, 16 May 2005 22:57:48 +0000 (22:57 +0000)]
Branch b1_4
Disable fsync-under-recovery test, which is preventing remaining tests
from running.
b=5494
r=nathan
adilger [Mon, 16 May 2005 18:22:42 +0000 (18:22 +0000)]
Branch b1_4
Decrease the ldlm_timeout if obd_timeout is dropped. It doesn't really make
sense to have AST timeouts longer than the RPC timeouts, but is mostly only
noticable in our regression tests (recovery-small.sh test 10 in particular).
b=6172
adilger [Mon, 16 May 2005 18:01:46 +0000 (18:01 +0000)]
Branch b1_4
Use a timeout instead of killing writemany - the OST failover may not have
happened yet.
r=nathan
adilger [Mon, 16 May 2005 17:54:59 +0000 (17:54 +0000)]
Branch b1_4
Make sure that test 52 writemany will exit at some point.
Put writemany usage message at top of code so it is easily found.
r=nathan
larry [Sun, 15 May 2005 23:21:51 +0000 (23:21 +0000)]
*** empty log message ***
adilger [Sat, 14 May 2005 10:32:37 +0000 (10:32 +0000)]
Branch b1_4
Add regression test for reading from a file with no objects (reads zeros,
stops at EOF).
b=6242, b=6243
jacob [Fri, 13 May 2005 18:54:53 +0000 (18:54 +0000)]
update to new errata
nathan [Fri, 13 May 2005 18:11:51 +0000 (18:11 +0000)]
Branch b1_4
b=6301
r=adilger
Prevent damage due to racing export disconnects
jacob [Fri, 13 May 2005 18:04:05 +0000 (18:04 +0000)]
land b_release_1_4_2:
- qsnet patches for sles 2.6 and rhel 2.4
adilger [Thu, 12 May 2005 23:13:24 +0000 (23:13 +0000)]
Branch b1_4
Move all pagecache vs. directio handling into filter_clear_page_cache()
and add comments about what is going on here and why.
Use the fs-specific ->invalidatepage called from truncate_complete_page()
instead of calling block_invalidatepage() directly.
Wait on page for the 2.6 code, and use 2.6 functions and have compat
macros for 2.4 instead.
b=4892
r=green
adilger [Thu, 12 May 2005 21:28:49 +0000 (21:28 +0000)]
Branch b1_4
Update ChangeLog to new format.
adilger [Thu, 12 May 2005 20:42:39 +0000 (20:42 +0000)]
Branch b1_4
Add obd_name to message.
b=6301
adilger [Wed, 11 May 2005 22:04:03 +0000 (22:04 +0000)]
Branch b1_4
Ensure the PING_INTERVAL is at least 1s, otherwise obd_timeout < 4 can
cause severe heartburn.
green [Wed, 11 May 2005 21:16:30 +0000 (21:16 +0000)]
Branch: b1_4
b=6243
r=adilger
Properly return 0 for reads that are totally after EOF.
adilger [Wed, 11 May 2005 19:23:49 +0000 (19:23 +0000)]
Branch b1_4
Revert previous commit, it was confusingly correct.
b=6289
adilger [Wed, 11 May 2005 18:28:46 +0000 (18:28 +0000)]
Branch b1_4
Pass OBD_BRW_WRITE to filter_alloc_iobuf() from preprw_write(). This
parameter is actually unused at this time, so no real effect.
b=6289
r=fergal (original change)
adilger [Wed, 11 May 2005 18:03:07 +0000 (18:03 +0000)]
Branch b1_4
Update build version to 1.4.2.2bgl
adilger [Wed, 11 May 2005 17:51:37 +0000 (17:51 +0000)]
Branch b1_4
Fix typo in bug number.
nathan [Wed, 11 May 2005 00:07:36 +0000 (00:07 +0000)]
Branch b1_4
b=5493
add asserts to help track this down
nathan [Tue, 10 May 2005 21:28:44 +0000 (21:28 +0000)]
Branch b1_4
b=5949
adilger [Tue, 10 May 2005 20:32:23 +0000 (20:32 +0000)]
Branch b1_4
Increase the LDLM AST timeout. With the lock reordering in bug 5492 et.al.
we now hold ns_lock over vmtruncate.
b=6172
r=phil
nathan [Tue, 10 May 2005 15:25:39 +0000 (15:25 +0000)]
Branch b1_4
b=6223
r=adilger
Fix lfs find to handle arbitrary number of targets
adilger [Mon, 9 May 2005 06:05:24 +0000 (06:05 +0000)]
Branch b1_4
Fix compile warning.
green [Sun, 8 May 2005 20:49:11 +0000 (20:49 +0000)]
2.4.29 does not need smp_num_siblings export patch as this is already exported
green [Sun, 8 May 2005 20:33:40 +0000 (20:33 +0000)]
Move CONFIG_KALLSYMS question from ISDN menu to kernel hacking menu
in 2.4.29 patch
green [Sun, 8 May 2005 20:22:48 +0000 (20:22 +0000)]
kallsyms patch for 2.4.29
Separated uml & non-uml 2.4.29 series
green [Fri, 6 May 2005 20:22:39 +0000 (20:22 +0000)]
b=5907
r=adilger
Testcase for the bug.
adilger [Fri, 6 May 2005 14:55:49 +0000 (14:55 +0000)]
fix subsystem type
adilger [Thu, 5 May 2005 22:28:36 +0000 (22:28 +0000)]
Branch: b1_4
Add bug reference for workaround.
b=6155
nathan [Thu, 5 May 2005 21:52:47 +0000 (21:52 +0000)]
Branch b1_4
b=5949
r=adilger
Various fixes to make --failover under load safer
green [Thu, 5 May 2005 20:54:22 +0000 (20:54 +0000)]
b=6243
Export clear_user_skas as well (uml-only again)
phil [Thu, 5 May 2005 20:50:41 +0000 (20:50 +0000)]
- I committed the ext3-patch-fuzz patch, but forgot to add it to the series
- add the recent clear_user export patch to my uml series
adilger [Thu, 5 May 2005 20:16:49 +0000 (20:16 +0000)]
Branch: b1_4
Remove unused function. class_find_client_obd_notype() does the same and
is more generic.
b=6242
adilger [Thu, 5 May 2005 19:41:27 +0000 (19:41 +0000)]
Branch: b1_4
Reorder ll_async_page slightly so the ints are together (slight savings on
64-bit arches).
adilger [Thu, 5 May 2005 18:51:17 +0000 (18:51 +0000)]
Branch: b1_4
Limit the number of stripes in a file to the maximum that will fit into
a 4kB EA. This is needed until we can store larger EAs in ext3 (bug 4424).
b=6093
adilger [Thu, 5 May 2005 18:36:35 +0000 (18:36 +0000)]
Branch: b1_4
Update build version to 1.4.1.14.
adilger [Thu, 5 May 2005 18:25:02 +0000 (18:25 +0000)]
Branch: b1_4
Make it easier to see dev_read_only set on device.
adilger [Thu, 5 May 2005 18:12:01 +0000 (18:12 +0000)]
Branch: b1_4
Placeholder from b_cray so it doesn't get reused.
phil [Thu, 5 May 2005 18:07:11 +0000 (18:07 +0000)]
b=5902
Remove the tcp-zero-copy patch from the 2.4.21-suse-171 series. We are
virtually certain that the corruption is caused in the TCP stack, and we are
completely unable to reproduce it with zero-copy disabled. Because this series
is of almost zero interest, just remove the patch and move on.
adilger [Thu, 5 May 2005 18:01:27 +0000 (18:01 +0000)]
Branch: b1_4
Remove hopelessly obsolete test program.
jacob [Thu, 5 May 2005 17:47:32 +0000 (17:47 +0000)]
turn quadrics drivers on
adilger [Thu, 5 May 2005 17:44:17 +0000 (17:44 +0000)]
Branch: b1_4
Using 'lctl recover' should mark a device active if it was previously
deactivated.
Add some debugging to ptlrpc_pinger_main() to help analyse bug 5933.
b=5933
r=nathan
jacob [Thu, 5 May 2005 17:41:08 +0000 (17:41 +0000)]
b=6233
originally from oleg, r=jacob
Build quota_context as part of ptlrpc, so it can be shared by the mds
and obdfilter without being linked into both, which sometimes causes
build failures on lester.
adilger [Thu, 5 May 2005 16:44:59 +0000 (16:44 +0000)]
Branch: b1_4
Bump the maximum number of devices as a short-term solution. The long-term
solution to allocate these structs is under implementation but not ready
for a 1.4.2 release.
b=6242
green [Thu, 5 May 2005 05:55:08 +0000 (05:55 +0000)]
b=6243
r=adilger
Correctly return zero-filled regions when doing reads of files with no objects.
Export clear_user_tt symbol in UML as it is now needed by lustre.
jacob [Wed, 4 May 2005 23:04:04 +0000 (23:04 +0000)]
add qsnet patch for rhel-2.4.
adilger [Wed, 4 May 2005 21:28:01 +0000 (21:28 +0000)]
Branch: b1_4
Commit the routed.sh script to CVS for reference, remove old configs.
jacob [Wed, 4 May 2005 19:49:57 +0000 (19:49 +0000)]
update to latest errata; patches apply and no config changes needed
adilger [Wed, 4 May 2005 08:30:17 +0000 (08:30 +0000)]
Branch: b1_4
Don't clobber last_rcvd upon mds_client_add() failure. This was caused by
bad error handling in mds_connect() not setting med_mcd = NULL after freeing
it, and then zeroing the "client data" at med_off = 0. Includes test.
This commit adds some safety checks to {mds,filter}_finish_transno()
and reconciles the differences between the mds and obdfilter last_rcvd
functions.
b=6086
r=nathan
jacob [Tue, 3 May 2005 22:16:07 +0000 (22:16 +0000)]
b=3031
- this is what i should have done originally
nathan [Tue, 3 May 2005 22:13:12 +0000 (22:13 +0000)]
Branch b1_4
b=6223
Fix this related one too.
nathan [Tue, 3 May 2005 21:54:49 +0000 (21:54 +0000)]
b=6223
This should fix the crash (need to use vfree, not kfree). But it won't
allow lfs getstripe to work for 420 OSTs -- OBD_MAX_IOCTL_BUFFER of 8k allows a
max of around 180 OBDs.
jacob [Tue, 3 May 2005 21:46:41 +0000 (21:46 +0000)]
b=3031
- remove .flags files on make clean