Whamcloud - gitweb
adilger [Tue, 3 Jan 2006 22:50:45 +0000 (22:50 +0000)]
Branch b_release_1_4_6
Ensure that statfs does not come out of local cached value.
b=9691
green [Tue, 3 Jan 2006 22:48:44 +0000 (22:48 +0000)]
b=9780
properly handle O_NOFOLLOW
green [Tue, 3 Jan 2006 22:46:30 +0000 (22:46 +0000)]
More sync-up with sysio head
alex [Tue, 3 Jan 2006 19:10:41 +0000 (19:10 +0000)]
b=9919
r=nikita (original fix)
hijacks pages from struct brw_page (allocated by echo_client_kbrw()) by filter_brw(). This
- fixes problem introduced by 5137 (filter_preprw_*() assumes that pages are preallocated),
- avoids duplicate allocation,
- avoids data copying.
Also fixes small defect in filter_iobuf_get().
adilger [Mon, 2 Jan 2006 08:04:54 +0000 (08:04 +0000)]
Branch b_release_1_4_6
Define $tdir and $tfile for sanityN.sh, bringing us one tiny step closer to
being able to use test-framework.sh for sanityN.sh.
adilger [Mon, 2 Jan 2006 07:48:32 +0000 (07:48 +0000)]
Branch b_release_1_4_6
Description: two objects could be created on the same OST for a single file
Details : If an OST is down, in some cases it was possible to create two
objects on a single OST for a single file. No problems other
than potential performance impact and spurious error messages.
b=9923
r=green
adilger [Mon, 2 Jan 2006 07:38:42 +0000 (07:38 +0000)]
Branch b_release_1_4_6
We should always be starting service threads when we call
ptlrpc_start_threads(). This caused a very tricky to notice "communications"
problem when an OST didn't start any service threads. This can only be a
code problem, and not from module parameter input.
r=hwartens
phil [Sat, 31 Dec 2005 16:33:00 +0000 (16:33 +0000)]
a local-tree-cleanup commit full of empty calories
green [Sat, 31 Dec 2005 13:44:12 +0000 (13:44 +0000)]
b=10018
Fix 64 bit compile problem
niu [Sat, 31 Dec 2005 05:03:09 +0000 (05:03 +0000)]
acquire block quota on owner of child also, in case of ea block counted in some kernel version.
b: 10017
yury [Fri, 30 Dec 2005 07:58:39 +0000 (07:58 +0000)]
b=9926
r=adilger
- do not confuse developer looking at logs by saying that namespace is freeing on last lock put.
- cleanups in mds_init_lov_desc()
- using stripes = min(tgt_count, LOV_MAX_STRIPE_COUNT) as a basis for calculating easize/cookiesize on MDS and client.
- removed redundant obd_set_info("mds_conn") in mds_notify()
green [Thu, 29 Dec 2005 22:00:01 +0000 (22:00 +0000)]
Change some CERROR()s that are not errors to CWARN()s in administrative
evictions path.
green [Thu, 29 Dec 2005 21:57:30 +0000 (21:57 +0000)]
Switch b_release_1_4_6 to use new libsysio.
green [Thu, 29 Dec 2005 21:50:49 +0000 (21:50 +0000)]
Branch: b_release_1_4_6
new libsysio from Sandia.
niu [Thu, 29 Dec 2005 06:32:12 +0000 (06:32 +0000)]
- typo fix
- message format fix for 'lfs quota -t'
adilger [Sat, 24 Dec 2005 20:07:28 +0000 (20:07 +0000)]
Branch b_release_1_4_6
Fix up acl check.
adilger [Fri, 23 Dec 2005 23:59:09 +0000 (23:59 +0000)]
Branch b_release_1_4_6
Use (effectively) ptlrpc_daemonize() when starting the ping evictor thread,
as this thread can do filesystem IO when evicting clients (update last_rcvd)
and as a result trip LASSERT in pop_ctxt().
We can't call ptlrpc_daemonize() directly from code in obdclass as that
results in module symbol circular dependencies. Looking into moving the
pinger code into ptlrpc entirely to avoid this issue.
b=9961
r=nathan
adilger [Fri, 23 Dec 2005 06:07:10 +0000 (06:07 +0000)]
Branch b_release_1_4_6
Fix error message argument order
b=9788
r=nathan (original fix)
niu [Fri, 23 Dec 2005 03:51:45 +0000 (03:51 +0000)]
make lustre_get_qids() support 2.6.12
b: 9812
brian [Thu, 22 Dec 2005 20:05:26 +0000 (20:05 +0000)]
b=9975
r=adilger
Add a --disable-quilt option to force ldiskfs patching using patch rather than
quilt even if it's found.
brian [Thu, 22 Dec 2005 19:30:46 +0000 (19:30 +0000)]
Revert this commit. The obdfilter module produces the following error on load:
/lib/modules/2.4.21-37.EL_lustre-b_release_1_4_6.200512222341smp/kernel/fs/lustre/obdfilter.o: unresolved symbol IS_KEY
yury [Thu, 22 Dec 2005 10:59:33 +0000 (10:59 +0000)]
- fixed debug message in ldlm_lock_put() to less confusing. Former looked
like this:
"final lock_put on destroyed lock, freeing namespace ..."
whereas it should like this:
"final lock_put on destroyed lock, freeing it. namespace ..."
- cleanups in mds_init_lov_desc()
- in mds_notify() remove obd_set_info("mds_conn") before calling mds_lov_start_synchronize()
as it will be called by mds_lov_start_synchronize() itself.
- in filter_set_info() using IS_KEY() macro
green [Wed, 21 Dec 2005 17:15:50 +0000 (17:15 +0000)]
Branch: b_release_1_4_6
r=adilger
Make llu_dir_do_readpage to match LCK_CR locks to conform to inodebits new
locking policy.
alex [Tue, 20 Dec 2005 21:51:34 +0000 (21:51 +0000)]
r=green
- remove dentry from hash before puting on orphan list (2.4 only)
adilger [Tue, 20 Dec 2005 09:27:56 +0000 (09:27 +0000)]
Branch b_release_1_4_6
Add /proc/fs/lustre/{mdc,osc}/*/connect_flags entries, so it is possible
to check what features are supported by a client.
adilger [Tue, 20 Dec 2005 07:13:38 +0000 (07:13 +0000)]
Branch b_release_1_4_6
Fix minor compile warnings.
yury [Mon, 19 Dec 2005 15:47:15 +0000 (15:47 +0000)]
- mostly debug stuff for #9926
- easize + lov_desc initialization is moved to mds_lov_update_desc()
niu [Mon, 19 Dec 2005 07:02:29 +0000 (07:02 +0000)]
read all old quota limits on quotacheck.
b: 9812
yury [Sat, 17 Dec 2005 17:06:57 +0000 (17:06 +0000)]
- lov_add_obd() should pass OBD_NOTIGY_ACTIVE instead of 1 as 1 means OBD_NOTIGY_INACTIVE, also not good to use magical digits.
This seems result of wrong backport from HEAD, as in HEAD obd_notify() has boolean flag (active/inactive) instead of event type.
This bug caused mds_notify() did nothing in the case of addind one more OST.
- fixed ENTRY/RETURN in lov_notify()
adilger [Sat, 17 Dec 2005 06:32:29 +0000 (06:32 +0000)]
Branch b_release_1_4_6
Print out filesystem label and UUID at server mount time for diagnosis.
b=9937
r=nathan
alex [Sat, 17 Dec 2005 00:39:42 +0000 (00:39 +0000)]
b=7225
- mballoc3 is landed
nic [Fri, 16 Dec 2005 21:14:09 +0000 (21:14 +0000)]
b=9933
r=adilger
call tcpnal_set_global_params() to parse things like TCPNAL_PORT
yury [Fri, 16 Dec 2005 15:07:40 +0000 (15:07 +0000)]
- fixed few EXIT/RETURN
- added more debug for cacthing diff >= 0 assert
nic [Fri, 16 Dec 2005 00:48:33 +0000 (00:48 +0000)]
r=brian
enable pthreads outside of liblustre so that lctl --threads works on a non
liblustre build
yury [Thu, 15 Dec 2005 15:01:45 +0000 (15:01 +0000)]
- in mds_lov_start_synchronize() if mds_lov_synchronize() thread did not start due to some reason, do not do
class_export_put(), that causes assert in case export has only one ref as export is still hashed. In fact, there
should be decref for previously made incref to @obd->obd_refcount and this fucntion seems not supposed to do
something else.
cliffw [Thu, 15 Dec 2005 01:00:42 +0000 (01:00 +0000)]
Changes for 1.4.6 so modprobe.conf is updated correctly and ptlrpc is
commented out. From Andreas
r=cliffw
nic [Wed, 14 Dec 2005 20:24:42 +0000 (20:24 +0000)]
r=adilger
seed the fsx operations from the time so that we vary the test sequence a bit
nic [Wed, 14 Dec 2005 16:19:33 +0000 (16:19 +0000)]
b=7238
we don't need to tell llmount about -o port=XXXX, lnet does that with module
options now
adilger [Wed, 14 Dec 2005 11:55:47 +0000 (11:55 +0000)]
Branch b_release_1_4_6
Description: failure of ptlrpc thread startup can cause oops
Details : Starting a ptlrpc service thread can fail if there are a large
number of threads or the server memory is very fragmented.
Handle this without oopsing.
b=9493
Tested extensively at Bull (512 OST service threads).
adilger [Wed, 14 Dec 2005 11:41:22 +0000 (11:41 +0000)]
Branch b_release_1_4_6
Fix error exit code of test 21 umount handler.
b=9931
adilger [Wed, 14 Dec 2005 09:35:44 +0000 (09:35 +0000)]
Branch b_release_1_4_6
Actually fix this problem
b=9692
yury [Wed, 14 Dec 2005 08:34:57 +0000 (08:34 +0000)]
- fixed possible skipped lquota_adjust() in filter_setattr_internal() due to fsfilt_setattr() turns off
all flags it set which also used to check if we need to adjust quota or not.
adilger [Wed, 14 Dec 2005 01:02:23 +0000 (01:02 +0000)]
Branch b_release_1_4_6
Allow test to run with existing dir.
b=9692
green [Tue, 13 Dec 2005 22:36:10 +0000 (22:36 +0000)]
r=adilger
Move lprocfs_register return value check under LPROCFS ifdef. IS_ERR threats
NULL as error too.
adilger [Tue, 13 Dec 2005 21:22:34 +0000 (21:22 +0000)]
Branch b_release_1_4_6
oops
adilger [Tue, 13 Dec 2005 21:17:15 +0000 (21:17 +0000)]
Branch b_release_1_4_6
Ignore genrated fle.
adilger [Tue, 13 Dec 2005 21:14:47 +0000 (21:14 +0000)]
Branch b_release_1_4_6
Verify that clients only enqueue LDLM_IBITS locks to MDS now. If server
does not handle OBD_CONNECT_IBITS then client DLM will convert this to
LDLM_PLAIN when sending RPC.
Handle IBITS locks in ldlm_lock_match() debug messages.
Verify lock type and do type conversion before performing lock replay.
b=9912
r=green
adilger [Tue, 13 Dec 2005 10:32:32 +0000 (10:32 +0000)]
Branch b_release_1_4_6
Fix merge breakage of join feature when IBITS locking was added.
b=9912
r=green
Move JOIN_FILE definitions away from potentially-conflicting VFS flags.
Minor cleanups in Join file code (line wrap, whitespace).
yury [Tue, 13 Dec 2005 09:05:02 +0000 (09:05 +0000)]
- small simplification in filter_setattr_internal()
yury [Tue, 13 Dec 2005 08:57:40 +0000 (08:57 +0000)]
- fixed type in comment
yury [Tue, 13 Dec 2005 08:35:18 +0000 (08:35 +0000)]
- former precreate pools stuff is returned back
- some additions to make quota work with formet creareate code
- some small fixes to filter_setattr() path
alex [Tue, 13 Dec 2005 00:14:57 +0000 (00:14 +0000)]
b=9562
r=adilger,nikita
- fixes in dcache interaction code. we need to lock dentries in 2.6.
also, we use __d_drop() instead of list_del() because the former
takes a care about actual dentry state (hashed/unhashed)
alex [Mon, 12 Dec 2005 23:28:33 +0000 (23:28 +0000)]
- disable mballoc "too long search" warning for a while
adilger [Mon, 12 Dec 2005 21:49:52 +0000 (21:49 +0000)]
Branch b_release_1_4_6
Found during CROW revert inspection/testing though not directly related to
that bug. ll_direct_io_24() calls ll_inode_fill_obdo() with VFS "rw" flag
(0=read, 1=write) and not OBD_BRW_READ=1, OBD_BRW_WRITE=2 flag.
b=9761
adilger [Mon, 12 Dec 2005 09:16:32 +0000 (09:16 +0000)]
Branch b_release_1_4_6
Severity : enhancement
Bugzilla : 9477, 9557, 9870
Description: Verify that the MDS configuration logs are updated when xml is
Details : Check if the .xml configuration logs are newer than the config
logs stored on the MDS and report an error if this is the case.
Request --write-conf, or allow starting with --old_conf.
r=adilger,nathan (lincent original patch, modified by adilger)
Severity : enhancement
Bugzilla : 6034
Description: Handle symlinks in the path when checking if Lustre is mounted.
Details : Resolve intermediate symlinks when checking if a client has
mounted a filesystem to avoid duplicate client mounts.
r=phil,adilger (Fergal original patch)
Severity : minor
Frequency : rare
Bugzilla : 9309
Description: lconf can hit an error exception but still return success.
Details : The lconf command catches the Command error exception at the top
level script context and will exit with the associated exit
status, but doesn't ensure that this exit status is non-zero.
r=phil,adilger (Fergal original patch)
Add "accept=all" when running from a developer tree to allow liblustre
testing, as insmod does not look at /etc/modules.conf for module options.
r=green
green [Sun, 11 Dec 2005 22:13:17 +0000 (22:13 +0000)]
b=9907
Fix __l_wait_event to correctly honor timeout value (broken by bug 9297 changes).
green [Sat, 10 Dec 2005 10:19:26 +0000 (10:19 +0000)]
b=9794
r=adilger
This patch uses a fast, high-quality PRNG internal to Lustre instead of the
glibc PRNG to avoid perturbing the glibc PRNG sequence for liblustre clients.
It also uses this PRNG for some uses in the lustre core code.
Introduces ll_rand() and ll_srand() with usage similar to rand(3) and srand(3)
brian [Fri, 9 Dec 2005 19:19:24 +0000 (19:19 +0000)]
b=9729
r=adilger
Merge the --enable-bgl changes from b1_4.
nic [Fri, 9 Dec 2005 18:36:13 +0000 (18:36 +0000)]
b=7283
r=adilger
allow tests to handle non-default ACCEPTOR_PORT
adilger [Fri, 9 Dec 2005 10:32:34 +0000 (10:32 +0000)]
Branch b_release_1_4_6
Minor compile warning fix.
adilger [Fri, 9 Dec 2005 10:26:31 +0000 (10:26 +0000)]
Branch b_release_1_4_6
Increase the default lustre DLM LRU size based on client CPU count.
For systems that have many CPUs there is expected to be a correspondingly
higher number of files open and in use for a given node.
b=9462
adilger [Fri, 9 Dec 2005 00:26:07 +0000 (00:26 +0000)]
Branch b_release_1_4_6
Increase RPC tunable maximums for large Bull clients.
Does not affect default values.
adilger [Fri, 9 Dec 2005 00:22:22 +0000 (00:22 +0000)]
Branch b_release_1_4_6
Add comment for OBD_CONENCT_JOIN.
Avoid self-referential CROW description.
Remove large comment for OBD_CONNECT_TRUNCLOCK, it "hides" below flags.
green [Thu, 8 Dec 2005 18:21:13 +0000 (18:21 +0000)]
r=adilger(original author)
fix mds_get_md to recognise empty lmm for files without objects.
This fixes error messages from chown on such files.
green [Thu, 8 Dec 2005 14:44:01 +0000 (14:44 +0000)]
b=9896
Fix recently broken test.
wangdi [Thu, 8 Dec 2005 09:20:10 +0000 (09:20 +0000)]
Branch: b_release_1_4_6
log: land b1_4_join to b_release_1_4_6
r=nikita, alex, adilger
green [Wed, 7 Dec 2005 20:35:01 +0000 (20:35 +0000)]
ug 9297 changelog update
green [Wed, 7 Dec 2005 14:38:59 +0000 (14:38 +0000)]
b=9297
r=adilger
Stop sending data to evicted clients as soon as possible.
green [Wed, 7 Dec 2005 13:40:22 +0000 (13:40 +0000)]
b=9826
Add new __craynv architecture to the list of 64bit architectures for XT3
green [Wed, 7 Dec 2005 13:39:20 +0000 (13:39 +0000)]
b=9827
Fix declarations of ldlm_cli_enqueue, ldlm_lock_match, ldlm_resource_get and
ldlm_lock_Create to match actual function definitions
(__u32 type -> ldlm_lock_t type)
brian [Wed, 7 Dec 2005 13:29:28 +0000 (13:29 +0000)]
Make CONFIG_IEEE1394_OHCI1394 a module.
Remove duplicate options. I should have checked for these before I added the new ones at the
end of the file in the first place.
Some minor rearrangement to agree with make oldconfig.
alex [Wed, 7 Dec 2005 10:05:36 +0000 (10:05 +0000)]
b=5681
r=adilger
- unhash dentry before __d_rehash() in iopen_connect_dentry(). otherwise
__d_rehash() will corrupt hash list as dentry can be already hashed
- locking fix suse's iopen
alex [Wed, 7 Dec 2005 09:38:06 +0000 (09:38 +0000)]
b=9446
- don't check health of devices being stopped
green [Tue, 6 Dec 2005 23:20:39 +0000 (23:20 +0000)]
b=9879
mds_connect_internal does not in fact need obd as its argument, because
export has it as ->exp_obd anyway.
green [Tue, 6 Dec 2005 23:18:37 +0000 (23:18 +0000)]
b=9879
r=adilger
Move all filter compat flags checks to filter_connect_internal.
Move call to filter_connect_internal in filter connect to be earlier that
!replayable check, so that it is always executed.
green [Tue, 6 Dec 2005 23:05:07 +0000 (23:05 +0000)]
b=9879
r=adilger
Test that we do not receive any plain lock requests from ibits-aware
clients.
Test that client and server have intersecting lock bits, and if not - server
drops IBITS compat flag. (if the flag is set and intersecting list of bits is
zero on client - just error message is issued).
We now also check that we do not return IBITS lock with sync callback to
liblustre clients.
cvs2svn [Tue, 6 Dec 2005 21:16:37 +0000 (21:16 +0000)]
This commit was manufactured by cvs2svn to create branch 'b_release_1_4_6'.
green [Tue, 6 Dec 2005 21:16:36 +0000 (21:16 +0000)]
reverting test commit
green [Tue, 6 Dec 2005 21:16:16 +0000 (21:16 +0000)]
test commit
brian [Tue, 6 Dec 2005 17:07:13 +0000 (17:07 +0000)]
Re-arrange options as make oldconfig will.
green [Tue, 6 Dec 2005 12:52:08 +0000 (12:52 +0000)]
b=9879
Introduce mds_reconnect to properly update export flags during reconnect.
green [Tue, 6 Dec 2005 10:35:59 +0000 (10:35 +0000)]
b=9879
Return error to client if it requests inode bits locks and we believe it
does not support such locks.
brian [Mon, 5 Dec 2005 18:14:40 +0000 (18:14 +0000)]
Remove duplicated and conflicting options.
brian [Mon, 5 Dec 2005 18:12:17 +0000 (18:12 +0000)]
CONFIG_IEEE1394_OHCI1394 needs to be a module too.
green [Mon, 5 Dec 2005 14:28:18 +0000 (14:28 +0000)]
Fix 2.6 build errors after b1_4 merge
green [Mon, 5 Dec 2005 12:37:10 +0000 (12:37 +0000)]
Update from b_hd_newconfig (b1_4)
green [Mon, 5 Dec 2005 11:51:45 +0000 (11:51 +0000)]
Updated to b1_4
green [Sun, 4 Dec 2005 22:54:59 +0000 (22:54 +0000)]
b=8001
r=adilger
Landing Inode bits compatibility stuff
brian [Sat, 3 Dec 2005 00:56:29 +0000 (00:56 +0000)]
b=9729
r=+adilger
Add an --enable-bgl flag for future BGL specific tunings and disable the -m64
compiler flag checking if set.
adilger [Fri, 2 Dec 2005 22:35:00 +0000 (22:35 +0000)]
Branch b_release_1_4_6
We always need to send OBD_CONNECT_ACL to connect to an MDS that is mounted
with "-o acl" (even just to get the config logs, currently).
b=9771
adilger [Fri, 2 Dec 2005 18:14:36 +0000 (18:14 +0000)]
Branch b_release_1_4_6
Fix 2.6.12 compat code.
ericm [Thu, 1 Dec 2005 21:40:30 +0000 (21:40 +0000)]
branch: b_release_1_4_6
- superblock need MS_POSIXACL flag
- check_acl() should return -EAGAIN in case of no ACL present or not
support ACL, to allow fall back triditional UGO checking.
ericm [Thu, 1 Dec 2005 21:37:28 +0000 (21:37 +0000)]
branch: b_release_1_4_6
from Andreas:
- grain control on which kernel use which ldiskfs series
ericm [Thu, 1 Dec 2005 21:34:43 +0000 (21:34 +0000)]
branch: b_release_1_4_6
from Andreas:
- fix sock_alloc() is not exported on vanilla-2.6.12
brian [Thu, 1 Dec 2005 13:42:54 +0000 (13:42 +0000)]
Rearrange kernel options to be the same as make oldconfig does.
adilger [Thu, 1 Dec 2005 08:35:53 +0000 (08:35 +0000)]
Branch b_release_1_4_6
Fix RPM build breakage
b=9771
adilger [Thu, 1 Dec 2005 08:32:33 +0000 (08:32 +0000)]
Branch b_release_1_4_6
Fix compile warning in ldiskfs/proc.c
ericm [Thu, 1 Dec 2005 02:23:27 +0000 (02:23 +0000)]
branch: b_release_1_4_6
add missing declaration of mds_quota_interface
ericm [Thu, 1 Dec 2005 02:09:24 +0000 (02:09 +0000)]
branch: b_release_1_4_6
fix compile error about ACL.
adilger [Wed, 30 Nov 2005 21:46:23 +0000 (21:46 +0000)]
Branch b_release_1_4_6
- fix build for 2.4 UML kernels that don't have ACL support