Whamcloud - gitweb
adilger [Fri, 14 Jun 2002 07:49:18 +0000 (07:49 +0000)]
Fix a crash in the POSIX test.
adilger [Fri, 14 Jun 2002 07:30:16 +0000 (07:30 +0000)]
Another part of the statfs story, disabled.
adilger [Fri, 14 Jun 2002 07:26:20 +0000 (07:26 +0000)]
Probable fix for storing 64-bit objids on the MDS for ext3. Probably needs
to be revised to fit the new LOV format (if there is enough space for it).
adilger [Fri, 14 Jun 2002 03:24:50 +0000 (03:24 +0000)]
Create the filesystem ROOT directory with 0755 permissions (the same as a
normal filesystem).
braam [Thu, 13 Jun 2002 23:08:15 +0000 (23:08 +0000)]
add intent related update to NFS server.
braam [Thu, 13 Jun 2002 21:32:32 +0000 (21:32 +0000)]
- Update md intent locking for chaos and lum patches to latest
version. Nothing terribly new apart from a first, failed, shot at
handling renames.
- Slightly improved locking documentation
pschwan [Thu, 13 Jun 2002 21:05:11 +0000 (21:05 +0000)]
- Added the obdo inline code to the OSC, for the non-lov-using case
- use extN MDS
- fixups to the EA code that I broke earlier
pschwan [Thu, 13 Jun 2002 19:56:41 +0000 (19:56 +0000)]
- Fixes obdecho
- Fixes server-side write leaks
- removes ext2 and ext3 MDS module insertion from common.sh
pschwan [Thu, 13 Jun 2002 19:25:10 +0000 (19:25 +0000)]
WARNING: This commit breaks everything. It will be back in shape within 12
hours, we hope, but update at your own (extreme) risk.
- Adds more unfinished LOV functionality
- Changes MDS EAs to contain LOV data
- Adds new brw callbacks
- Fixes to work with the Portals tip
adilger [Wed, 12 Jun 2002 20:12:05 +0000 (20:12 +0000)]
Do kunmap() on pages kmapp()ed in echo_preprw().
adilger [Wed, 12 Jun 2002 20:06:50 +0000 (20:06 +0000)]
Don't override MDS return code with transaction close code on error.
adilger [Wed, 12 Jun 2002 20:05:47 +0000 (20:05 +0000)]
Don't open files (i.e. directories) on the MDS with O_DIRECT.
adilger [Wed, 12 Jun 2002 20:04:21 +0000 (20:04 +0000)]
Make sure we clean up all the generated extN files on "make distclean"
adilger [Wed, 12 Jun 2002 19:21:58 +0000 (19:21 +0000)]
Commented-out mdc_statfs method. Not sure of how to set up all the portals
stuff, but it should be close.
adilger [Wed, 12 Jun 2002 19:20:53 +0000 (19:20 +0000)]
Minor cleanups.
Add commented-out call to mdc_statfs() for ll_statfs() - it isn't complete.
adilger [Wed, 12 Jun 2002 19:20:00 +0000 (19:20 +0000)]
Part of the O_DIRECT fix. The page indices iobuf->maplist[i]->index
do not appear to contain valid offsets into the file. Use the "blocknr"
value per generic_direct_IO instead.
adilger [Wed, 12 Jun 2002 19:11:30 +0000 (19:11 +0000)]
Whitespace cleanup
adilger [Wed, 12 Jun 2002 19:10:20 +0000 (19:10 +0000)]
Don't dereference niobuf_local page if it is not set in debug code.
adilger [Wed, 12 Jun 2002 19:07:45 +0000 (19:07 +0000)]
Add fsx binary to list of ignored files
adilger [Wed, 12 Jun 2002 08:27:38 +0000 (08:27 +0000)]
Remove typo file.
adilger [Wed, 12 Jun 2002 07:25:46 +0000 (07:25 +0000)]
common.sh: tell us when we are mounting
runiozone: allow the size parameter to be given on the command-line
adilger [Wed, 12 Jun 2002 00:17:40 +0000 (00:17 +0000)]
Check the f_mode field instead of the f_flags field when using FMODE_ masks.
adilger [Tue, 11 Jun 2002 19:40:12 +0000 (19:40 +0000)]
Merge changes made in portals/linux/utils/parser.c into lustre/utils/parser.c
adilger [Tue, 11 Jun 2002 17:24:10 +0000 (17:24 +0000)]
Clean up the right portals network type if it is set (default tcp).
adilger [Tue, 11 Jun 2002 17:23:29 +0000 (17:23 +0000)]
Remove compiler warning for fsx on linux.
adilger [Mon, 10 Jun 2002 23:38:04 +0000 (23:38 +0000)]
Add connid and xid to timeout messages.
adilger [Mon, 10 Jun 2002 22:39:21 +0000 (22:39 +0000)]
Same fix for lustre_get_page_write().
adilger [Mon, 10 Jun 2002 22:37:52 +0000 (22:37 +0000)]
Don't kmap() a this page twice. We kunmap() in lustre_put_page() so we
do the kmap() in all cases.
adilger [Mon, 10 Jun 2002 22:35:12 +0000 (22:35 +0000)]
Add prototypes for the debug code in lib/debug.c.
adilger [Mon, 10 Jun 2002 22:34:38 +0000 (22:34 +0000)]
Dump ->addr as a pointer instead of a long long.
adilger [Mon, 10 Jun 2002 22:31:34 +0000 (22:31 +0000)]
Use OBD_ALLOC for BRW vector test code.
Turn device name error into a debug message to avoid confusion.
adilger [Sat, 8 Jun 2002 01:11:26 +0000 (01:11 +0000)]
Add fsx test.
adilger [Sat, 8 Jun 2002 01:10:25 +0000 (01:10 +0000)]
Move VM stuff into lib/page.c.
adilger [Sat, 8 Jun 2002 01:05:58 +0000 (01:05 +0000)]
Exit out if there are setup errors.
adilger [Sat, 8 Jun 2002 01:03:09 +0000 (01:03 +0000)]
Save filesystem context before writing to last_rcvd file.
adilger [Sat, 8 Jun 2002 00:58:00 +0000 (00:58 +0000)]
Major fixups for multi-page I/Os in filterobd.
A few stubs for statfs on the filesystem.
adilger [Sat, 8 Jun 2002 00:44:02 +0000 (00:44 +0000)]
Fix some minor llite bugs. Should fix fsx problems.
adilger [Sat, 8 Jun 2002 00:39:35 +0000 (00:39 +0000)]
Make obdctl return the error code from the most recently executed command
in interactive mode. This allows scripts using HERE documents to abort
properly during setup.
adilger [Fri, 7 Jun 2002 17:22:33 +0000 (17:22 +0000)]
Transcription error.
braam [Fri, 7 Jun 2002 17:17:49 +0000 (17:17 +0000)]
Updated patch to remove extraneous changes.
adilger [Fri, 7 Jun 2002 00:15:44 +0000 (00:15 +0000)]
Don't stop checking for operations to commit if we hit an operation
with reqno == 0.
Peter - the question remains why we aren't already skipping these
open operations because of the PTL_RPC_FL_REPLAY flag? Is it because
they are closes or something?
adilger [Thu, 6 Jun 2002 21:40:27 +0000 (21:40 +0000)]
Subtract the OBD_FREE'd memory from the total before printing it.
braam [Thu, 6 Jun 2002 21:37:58 +0000 (21:37 +0000)]
- add more handling of intents (for chmod style functions)
- update to the head
adilger [Thu, 6 Jun 2002 21:31:39 +0000 (21:31 +0000)]
Rename "bulk" to "desc" to avoid confusion with the local "bulk" variable.
braam [Thu, 6 Jun 2002 19:26:17 +0000 (19:26 +0000)]
Change paths to use Lustre pathname instead of obd
braam [Thu, 6 Jun 2002 19:23:16 +0000 (19:23 +0000)]
Bring Makefile.am forward to move extN before mds.
adilger [Thu, 6 Jun 2002 08:16:58 +0000 (08:16 +0000)]
We weren't freeing local_nb in ost_brw_read(). This becomes obvious if you
run test_brw with larger vector sizes and large numbers of iterations.
adilger [Thu, 6 Jun 2002 08:13:38 +0000 (08:13 +0000)]
Remove ___wait_on_page entirely. If we ever need it again, it will be
different anyways.
adilger [Wed, 5 Jun 2002 23:38:04 +0000 (23:38 +0000)]
Get rid of local definition of ___wait_on_page. It existed at least as far
back as 2.4.0-prerelease and all 2.4.0-test kernels. It also breaks with
current RH RMAP kernels as well.
braam [Wed, 5 Jun 2002 22:48:02 +0000 (22:48 +0000)]
- fix Marcus' bug. Will move t_last_stable forward.
adilger [Wed, 5 Jun 2002 20:27:41 +0000 (20:27 +0000)]
Change the cleanup scripts to use the debug_kernel instead of get_debug.
adilger [Mon, 3 Jun 2002 22:36:53 +0000 (22:36 +0000)]
Move ioctl values out of the way of standard ext2/ext3 ioctl numbers.
adilger [Mon, 3 Jun 2002 22:18:57 +0000 (22:18 +0000)]
Add llite file flag to allow skipping the locking for file read/write ops.
The ioctl LL_IOC_SETFLAGS and LL_IOC_CLRFLAGS can be used to set and clear
the LL_FILE_IGNORE_LOCK flag on a file, and LL_IOC_GETFLAGS can be used to
retrieve the current flags value. These values are defined in lustre_lite.h.
Also add a small optimization - we do not update the file atime on reads if
it would not change the actual atime value (i.e. do not update this field
more often than once per second). We also avoid updating the atime on R/O
mounted filesystems or if the "noatime" mount option is used.
adilger [Mon, 3 Jun 2002 19:57:11 +0000 (19:57 +0000)]
Replace usage of "page" with "bulk" where "page" was actually a "struct
ptlrpc_bulk_page" to avoid (mental) confusion with the kernel "struct page".
adilger [Mon, 3 Jun 2002 19:03:26 +0000 (19:03 +0000)]
Change object "refcounts" to be dgets instead of igets, where possible,
because igets are not only more expensive (need to lock the inode, etc),
but you need to do the dentry lookup to find the inode in the first place.
Also cleaned up access to atomic counters to use "atomic_read(count)"
instead of count.value. Purely cosmetic, since we don't really need
the atomicness of the counters for debug output.
adilger [Fri, 31 May 2002 21:14:49 +0000 (21:14 +0000)]
Fix block leak on truncate - we weren't setting the correct mode on the
file and that was causing the truncate to be skipped.
adilger [Fri, 31 May 2002 11:05:03 +0000 (11:05 +0000)]
Fix the RH-chaos kernel oops. This was caused by trying to enable slab
poisoning (in order to catch errors where we dereference freed objects).
Sadly, the slab.c code checks for "invalid" flags and BUGs if any are set
(only doing so at runtime) rather than just not defining them and catching
the error at compile time or just ignoring it entirely.
adilger [Fri, 31 May 2002 09:57:03 +0000 (09:57 +0000)]
Fix minor typo in chaos1 patch.
adilger [Thu, 30 May 2002 22:22:04 +0000 (22:22 +0000)]
Remove debugging that slipped into checkin.
adilger [Thu, 30 May 2002 22:18:41 +0000 (22:18 +0000)]
Check RPC reply status for both MDC and OSC. This allows us to return
errors "properly" from the MDS and OST to the filesystem/kernel/application.
This finally appears to handle out-of-space errors on the MDS and OST
reasonably well.
Still to do at some point: OS agnostic error value encoding/decoding.
adilger [Thu, 30 May 2002 22:13:59 +0000 (22:13 +0000)]
Create a plain-text version of lustre-HOWTO.
adilger [Thu, 30 May 2002 21:48:55 +0000 (21:48 +0000)]
Argh, why can't C do what I mean and not what I say?
behlendo [Thu, 30 May 2002 21:24:41 +0000 (21:24 +0000)]
A 2.4.18-chaos1 redhat/llnl patch that cleanly applies.
adilger [Thu, 30 May 2002 20:44:39 +0000 (20:44 +0000)]
Clean up the object on the OST if there is a failure adding it to the MDS.
adilger [Thu, 30 May 2002 20:43:16 +0000 (20:43 +0000)]
Clean up in the right order on the failure path.
adilger [Thu, 30 May 2002 19:31:24 +0000 (19:31 +0000)]
Don't oops if we have a NULL lock.
adilger [Thu, 30 May 2002 19:30:01 +0000 (19:30 +0000)]
Clean up error llite open/release handling code. Still not 100% sure about
rpc request usage/cleanup, but I asked Phil to look at that.
adilger [Thu, 30 May 2002 19:00:09 +0000 (19:00 +0000)]
Set the dir_index flag on extN filesystems if we are creating them.
adilger [Thu, 30 May 2002 18:13:39 +0000 (18:13 +0000)]
Remove no-longer-needed inode operations (they previously had extN EA VFS
functions in them).
adilger [Wed, 29 May 2002 18:57:58 +0000 (18:57 +0000)]
Update the journal callback code in patch-2.4.9-chaos14. While it is not
necessary to update to this version of the journal callback if the older
version is applied, this does the callbacks outside the journal semaphore
and is the version likely to make it into ext3 proper.
adilger [Wed, 29 May 2002 18:38:08 +0000 (18:38 +0000)]
Remove all of the extended attribute changes, as they are all contained in
the extN code directly. Unfortunately, there is no easy way to know if the
generic kernel includes the EA VFS changes, so extN will not have those for
now.
adilger [Wed, 29 May 2002 18:30:03 +0000 (18:30 +0000)]
Remove VFS extended attribute interface so that extN will build even if the
kernel patch does not have the EA stuff in it (we no longer need this
interface as obdfilter calls the exported EA functions directly).
adilger [Wed, 29 May 2002 18:23:18 +0000 (18:23 +0000)]
Make mds_ext3.c and mds_extN.c build when no journal callback patch applied.
adilger [Wed, 29 May 2002 18:17:15 +0000 (18:17 +0000)]
Print when we are done the mount and start copying.
adilger [Wed, 29 May 2002 10:31:09 +0000 (10:31 +0000)]
Make the MDS filesystem interface code be a separate module. This allows
us to load MDS even if the extN module is not building properly.
adilger [Wed, 29 May 2002 10:27:57 +0000 (10:27 +0000)]
Add separate file for MDS filesystem interaction routines.
adilger [Wed, 29 May 2002 10:25:42 +0000 (10:25 +0000)]
Make xattr.c depend explicitly on patch-stamp, so that the patch part is
done before we try and build the results. Otherwise we have problems with
"make -j2"
adilger [Wed, 29 May 2002 10:22:13 +0000 (10:22 +0000)]
Add exports for extended attribute functions. This should be part of the
next official EA patch already (it was in ext2 but missing from ext3 EAs).
adilger [Wed, 29 May 2002 10:10:50 +0000 (10:10 +0000)]
Add super_operations to struct mds_obd so that we can free the allocated
struct after the superblock has been freed.
adilger [Wed, 29 May 2002 10:04:37 +0000 (10:04 +0000)]
Add mds filesystem helper modules to setup and cleanup.
adilger [Tue, 28 May 2002 23:28:08 +0000 (23:28 +0000)]
Add interface to extN filesystem which uses extended attributes to store
OST object IDs.
adilger [Tue, 28 May 2002 23:26:45 +0000 (23:26 +0000)]
Do proper cleanup in ll_read_super() if there are errors mounting the
filesystem.
braam [Mon, 27 May 2002 17:11:37 +0000 (17:11 +0000)]
changes to accomodate new directory names
pschwan [Mon, 27 May 2002 17:09:21 +0000 (17:09 +0000)]
Last week's LOV hackery
pschwan [Mon, 27 May 2002 16:48:11 +0000 (16:48 +0000)]
This commit was generated by cvs2svn to compensate for changes in r1070,
which included commits to RCS files with non-trunk default branches.
pschwan [Mon, 27 May 2002 16:48:11 +0000 (16:48 +0000)]
Initial lustretop import
pschwan [Sun, 26 May 2002 21:39:24 +0000 (21:39 +0000)]
- LBUG() on failed niobuf allocation
- Fix uninitialized variable mistake
braam [Sat, 25 May 2002 19:08:49 +0000 (19:08 +0000)]
Hooks for intent-based locks
pschwan [Sat, 25 May 2002 15:31:48 +0000 (15:31 +0000)]
- Quiet some journaling CERRORs
- Fix alignment of rpc buffers
adilger [Fri, 24 May 2002 20:06:54 +0000 (20:06 +0000)]
Fix patch problem.
pschwan [Fri, 24 May 2002 19:40:42 +0000 (19:40 +0000)]
- Quiet some journal verbosity
- Fixed (I hope) some ia64 unaligned access warnings
braam [Fri, 24 May 2002 19:20:55 +0000 (19:20 +0000)]
- another attempt to get the blocksize right
braam [Fri, 24 May 2002 19:08:26 +0000 (19:08 +0000)]
Try to get working chunk size for IA64 where PAGE_CACHE_SIZE is 4096
while PAGE_SIZE is 8192 (why ?).
pschwan [Fri, 24 May 2002 18:56:31 +0000 (18:56 +0000)]
- Fixed some 64bit warnings
- Fixed ldlm EXPORT table
- fixed small obdctl bug
adilger [Fri, 24 May 2002 18:46:33 +0000 (18:46 +0000)]
Fix refcount problem. It wasn't dentries that were causing the problem,
but an unbalanced push_ctxt() increasing the mnt->mnt_count and not
allowing us to free the superblock. Argh.
pschwan [Fri, 24 May 2002 17:10:31 +0000 (17:10 +0000)]
I am a superstar.
pschwan [Fri, 24 May 2002 17:09:48 +0000 (17:09 +0000)]
It did indeed fix it; here's the ones that I forgot.
pschwan [Fri, 24 May 2002 16:58:06 +0000 (16:58 +0000)]
Do lots of explicit EXPORT_SYMBOLs to see if this cures ia64 problems
adilger [Thu, 23 May 2002 23:21:05 +0000 (23:21 +0000)]
Don't BUG() if we just run out of space.
Some minor changes made when using page index/offset debugging (which is
not included here).
adilger [Thu, 23 May 2002 23:18:01 +0000 (23:18 +0000)]
Add the xattr files to the list of dependencies.
Add a comment that the export patch can fail (this happens if the kernel
source has already been patched to export ext3_bread()).