1 tbd Sun Microsystems, Inc.
4 2.6.16.60-0.23 (SLES 10),
5 2.6.18-92.1.6.el5 (RHEL 5),
6 2.6.22.14 vanilla (kernel.org).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.40.11-sun1
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
15 Severity : enhancement
17 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
19 Severity : enhancement
21 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
26 Description: Using wrong pointer in osc_brw_prep_request
27 Details : Access to array[-1] can produce panic if kernel compiled with
28 CONFIG_PAGE_ALLOC enabled
32 Description: Client runs out of low memory
33 Details : Consider only lowmem when counting initial number of llap pages
37 Description: *optional* service tags registration
38 Details : if the "service tags" package is installed on a Lustre node
39 When the filesystem is mounted, a local-node service tag will
40 be created. See http://inventory.sun.com/ for more information
41 about the Service Tags asset management system.
45 Description: Kernel BUG tries to release flock
46 Details : Lustre does not destroy flock lock before last reference goes
47 away. So always drop flock locks when client is evicted and
48 perform unlock regardless of successfulness of speaking to MDS.
52 Description: add recount protection for osc callbacks, so avoid panic on shutdown
56 Description: sanity test 65a fails if stripecount of -1 is set
57 Details : handle -1 striping on filesystem in ll_dirstripe_verify
62 Description: ASSERTION(CheckWriteback(page,cmd)) failed
63 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
66 Severity : enhancement
68 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
72 Description: do not process already freed flock
73 Details : flock can possibly be freed by another thread before it reaches
74 to ldlm_flock_completion_ast.
78 Description: LBUG during stress test
79 Details : Need properly lock accesses the flock deadlock detection list.
83 Description: oops in page fault handler
84 Details : kernel page fault handler can return two special 'pages' in error case, don't
85 try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
89 Description: timeout with invalidate import.
90 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which should be
91 handled by ptlrpcd. This produce long age waiting and -ETIMEOUT
92 ptlrpc_invalidate_import and as result LASSERT.
94 Severity : enhancement
96 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
100 Description: enable MGS and MDT services start separately
101 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
102 a co-located MGS without starting the MGS, which is a complement
103 to 'nosvc' mount option.
107 Frequency : after recovery
108 Description: precreate to many object's after del orphan.
109 Details : del orphan st in oscc last_id == next_id and this triger growing
110 count of precreated objects. Set flag LOW to skip increase count
111 of precreated objects.
115 Frequency : rare, on clear nid stats
116 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
117 Details : when clean nid stats sometimes try destroy live entry,
118 and this produce panic in free.
122 Description: Stack overflow during MDS log replay
123 ease stack pressure by using a thread dealing llog_process.
127 Description: wait until IO finished before start new when do lock cancel.
128 Details : VM protocol want old IO finished before start new, in this case
129 need wait until PG_writeback is cleared until check dirty flag and
130 call writepages in lock cancel callback.
132 Severity : enhancement
134 Description: using special macro for print time and cleanup in includes.
138 Description: mds_mfd_close() ASSERTION(rc == 0)
139 Details : In mds_mfd_close(), we need protect inode's writecount change
140 within its orphan write semaphore to prevent possible races.
144 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
148 Frequency : rare, on shutdown ost
149 Description: don't hit live lock with umount ost.
150 Details : shrink_dcache_parent can be in long loop with destroy dentries,
151 use shrink_dcache_sb instead.
155 Description: don't panic with use echo client
156 Details : echo client pass NULL as client nid pointer and this produce null
161 Description: fix build on ppc32
162 Details : compile code with -m64 flag produce wrong object file for ppc32.
166 Description: add message levels for liblustreapi
170 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
171 Details : Move the 'good_osts' check before the 'total_bavail' check. This
172 will result in an -EAGAIN and in the exit call path we call
173 alloc_rr() which will with increasing aggressiveness attempt to
174 aquire precreated objects on the minimum number of required OSCs.
178 Description: Use old size assignment to avoid deadlock
179 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
180 the scheduling while holding a spinlock. We do not need locking
181 for size in ll_update_inode() because size is only updated from
182 the MDS for directories or files without objects, so there is no
183 other place to do the update, and concurrent access to such inodes
184 are protected by the inode lock.
188 Description: resolve "_IOWR redefined" build error on SLES10
192 Description: dump the memory debugging after all modules are unloaded to
193 suppress false negative in conf_sanity test 39
195 Severity : enhancement
197 Description: build kernel-ib packages for OFED 1.3 in our release cycle
202 Description: fix SLES kernel versioning
203 Details : the kernel version for our SLES 10 kernel did not include a "-"
204 before the "smp" at the end. while this was not a problem in
205 general, it did mean that software trying to use the kernel
206 version to try to detect a vendor specific kernel would fail.
207 this was most evident by the OFED build scripts.
211 Description: Don't update lov_desc members until making sure they are valid
212 Details : When updating lov_desc members via proc fs, need fix their
213 validities before doing the real update.
217 Description: don't put request into delay list while invalidate in flight.
218 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
219 import in flight. this produce timeout for invalidate and sometimes
220 can cause stale data.
224 Frequency : on ppc only
225 Description: not convert ost objects for directory because it's not exist.
226 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
230 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
232 Severity : enhancement
234 Description: Update to support 2.6.22.14 vanilla kernel.
238 Frequency : rare, on recovery
239 Description: read procfs can produce deadlock in some situation
240 Details : Holding lprocfs lock which send rpc can produce block for destroy
241 obd objects and this also block reconnect with -EALREADY. This isn't
242 fix all lprocfs bugs - but make it rare.
244 Severity : enhancement
246 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
249 Frequency : frequent on X2 node
251 Description: mdc_set_open_replay_data LBUG
252 Details : Set replay data for requests that are eligible for replay.
256 Description: lustre_mgs: operation 101 on unconnected MGS
257 Details : When MGC is disconnected from MGS long enough, MGS will evict the
258 MGC, and late on MGC cannot successfully connect to MGS and a lot
259 of the error messages complaining that MGS is not connected.
263 Frequency : on network error
264 Description: panic with double free request if network error
265 Details : mdc_finish_enqueue is finish request if any network error ocuring,
266 but it's true only for synchronus enqueue, for async enqueue
267 (via ptlrpcd) this incorrect and ptlrpcd want finish request
270 Severity : enhancement
272 Description: client-side metadata stat-ahead during readdir(directory readahead)
273 Details : perform client-side metadata stat-ahead when the client detects
274 readdir and sequential stat of dir entries therein
277 Frequency : on start mds
279 Description: Implement get_info(last_id) in obdfilter.
282 Frequency : occasional
284 Description: Correctly check stale fid, not start epoch if ost not support SOM
285 Details : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
286 overwrited with new generated fid, but mds can anwer with one of these
287 two fids and both is not stale. setattr incorectly start epoch and
288 assume will be called done_writeting, but without SOM done_writing
292 Frequency : rare, depends on device drivers and load
294 Description: MDS or OSS nodes crash due to stack overflow
295 Details : Code changes in 1.8.0 increased the stack usage of some functions.
296 In some cases, in conjunction with device drivers that use a lot
297 of stack the MDS (or possibly OSS) service threads could overflow
298 the stack. One change which was identified to consume additional
299 stack has been reworked to avoid the extra stack usage.
302 Frequency : occasional
304 Description: Do not fail import if osc_interpret_create gets -EAGAIN
305 Details : If osc_interpret_create got -EAGAIN it immediately exits and
306 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
307 oscc_has_objects and see OSC has no objests and call
308 oscc_internal_create to resend create request.
310 Severity : enhancement
312 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
314 Severity : enhancement
316 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
319 Frequency : very rare
321 Description: Fix replay if there is an un-replied request and open
322 Details : In some cases, older replay request will revert the
323 mcd->mcd_last_xid on MDS which is used to record the client's
326 Severity : enhancement
328 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
330 Severity : enhancement
332 Description: Add rhel5 support to HEAD.
334 Serverity : enhancement
336 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
341 Description: Don't allow skipping OSTs if index has been specified.
342 Details : Don't allow skipping OSTs if index has been specified, make locking
343 in internal create lots better.
347 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
348 Details : access to bitfield in structure is always rounded to long
349 and this produce problem with not atomic change any bit.
353 Description: Lustre make rpms failed.
354 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
355 builds Lustre rpms from tarball.
358 Frequency : rare on shutdown ost
360 Description: If llog cancel was not send before clean_exports phase, this can
361 produce deadlock in llog code.
362 Details : If llog thread has last reference to obd and call class_import_put
363 this produce deadlock because llog_cleanup_commit_master wait when
364 last llog_commit_thread exited, but this never success because was
365 called from llog_commit_thread.
369 Description: allow userland application know is lost one of stripes.
370 Details : fill lvb_blocks with error code on ost and return it to
371 application if error flag found.
375 Description: NULL lov_tgts causing MDS oops
376 Details : more safe checks for NULL lov_tgts for avoid oops.
378 Severity : enhancement
380 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
384 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
385 Details : If an OST is down the MDS will hang indefinitely in
386 obd_statfs_rqset() waiting for the statfs data. While for
387 MDS QOS usage of statfs, it should not stuck in waiting.
389 Severity : enhancement
391 Description: remote_acl support
392 Details : Support ACL-based permission check for remote user.
393 Support setfacl/getfacl for remote user with the utils
394 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
395 the system "{s,g}etfacl" utils.
397 Severity : enhancement
399 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
401 Severity : enhancement
403 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
407 Description: make mgs_setparam() handle fsname containing dash
408 Details : fsname containing a dash does not work with lctl conf_param
410 Severity : enhancement
412 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
414 Severity : enhancement
416 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
420 Frequency : rare, at shutdown
421 Description: access already free / zero obd_namespace.
422 Details : if client_disconnect_export was called without force flag set,
423 and exist connect request in flight, this can produce access to
424 NULL pointer (or already free pointer) when connect_interpret
425 store ocd flags in obd_namespace.
429 Frequency : only at startup
430 Description: not alloc memory with spinlock held.
431 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
432 if any spinlock held.
434 Severity : enhancement
436 Description: make lustre randomly fail allocating memory
437 Details : Make lustre randomly failed allocating memory for testing purpose.
439 Severity : enhancement
441 Description: lost problems with lov objid file
442 Details : Fixes some scability and access to not inited memory problems
443 in work with lov objdid file.
448 Description: lfs find does not continue on file error
449 Details : Continue other files processing when a file/dir is absent.
453 Description: Inconsistent usage of lustre_pack_reply()
454 Details : Standardize the usage of lustre_pack_reply() such that it
455 always generate a CERROR on failure.
458 Frequency : occasional
460 Description: MDS hang or stay in waiting lock
461 Details : If client receive lock with CBPENDING flag ldlm need send lock
462 cancel as separate rpc, to avoid situation when cancel request
463 can't processed due all i/o threads stay in wait lock.
467 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
471 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
475 Description: Update to SLES9 kernel-2.6.5-7.287.3.
479 Description: Update to RHEL4 latest kernel.
481 Severity : enhancement
483 Description: Build SLES10 patchless client fails
484 Details : The configure was broken by run ./configure with
485 --with-linux-obj=.... argument for patchless client. When the
486 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
487 can't search header adequately. Use absolute path such as
488 -I($LINUX)/include instead.
492 Description: interrupt oig_wait produce painc on resend.
493 Details : brw_redo_request can be used for resend requests from ptlrpcd and
494 private set, and this produce situation when rq_ptlrpcd_data not
495 copyed to new allocated request and triggered LBUG on assert
496 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
497 ptlrpcd set if request is changed and can be safety changed to use
502 Description: LASSERT_{REQ,REP}SWAB macros are buggy
503 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
504 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
505 these macros and replace them with their difinitions instead.
509 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
510 Details : Update which_patch & target file for SLES10 latest kernel.
513 Description: Remove client patches from SLES 10 kernel.
514 Details : This causes SLES 10 clients to behave as patchless clients
515 even on a Lustre-patched (server) kernel.
517 Severity : enhancement
519 Description: self-adjustable client's lru lists
520 Details : use adaptive algorithm for managing client cached locks lru
521 lists according to current server load, other client's work
522 pattern, memory activities, etc. Both, server and client
523 side namespaces provide number of proc tunables for controlling
526 Severity : enhancement
528 Description: light-weight GSS support
529 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
530 and reduce performance overhead.
532 Severity : enhancement
534 Description: Linux keyring support
535 Details : Support using service of Linux keyring for Lustre GSS internal
536 context refresh/cache mechanism.
540 Description: Fix errors in lfs documentation
541 Details : Fixes man pages
545 Description: don't use GFP_* in generic Lustre code.
546 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
550 Description: obdclass is limited by single OBD_ALLOC(idarray)
551 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
555 Description: warnings with build patchless client with vanila 2.6.19 and up
556 Details : change old ctl_table style and replace ctl_table/ctl_table_header
557 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
561 Frequency : Only for SLES
562 Description: sanity_quota fail test_1
563 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
564 file, which makes TSTID[2] inunique.
568 Frequency : Only for SLES9
569 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
570 Details : Update target/ChangeLog/which_patch .
574 Frequency : Only for test_10 in sanity-quota.sh
575 Description: a bug in quota test code
576 Details : checking a test flag in a wrong place
580 Frequency : Only for test_13 in sanity-quota.sh
581 Description: a bug in quota test script
582 Details : shouldn't check overfull allocated quota
586 Description: jbd statistics
587 Details : Port older jbd statistics patch for sles10
589 Severity : enhancement
591 Description: Add printing inode info into message about error in writepage.
596 Description: reply_lock_interpret crash due to race with it and lock cancel.
597 Details : Do not replay locks that are being cancelled. Do not reference
598 locks by their address during replay, just by their handle.
602 Frequency : When flocks are used.
603 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
604 Details : Flock locks might destroy just granted lock if it could be merged
605 with another existing flock, this is done in completion handler,
606 so teach ldlm_cli_enquque_fini that this is a valid case for
612 Description: Oops in read and write path when failing to allocate lock.
613 Details : Check if lock allocation failed and return error back.
617 Description: lstripe command fails for valid OST index
618 Details : The stripe offset is compared to lov->desc.ld_tgt_count
619 instead of lov->desc.ld_active_tgt_count.
623 Description: sanity.sh failed test 103
624 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
629 Description: df doesn't work properly if diskfs blocksize != 4K
630 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
634 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
635 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
637 Severity : enhancement
639 Description: Adaptive timeouts
640 Details : RPC timeouts adapt to changing server load and network
641 conditions to reduce resend attempts and improve recovery time.
645 Description: llapi_file_create() does not allow some changes
646 Details : add llapi_file_open() that allows specifying the mode and
647 open flags, and also returns an open file handle.
651 Description: merge and cleanup kernel patches.
652 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
656 Description: Add journal checksum support.(Kernel part)
657 Details : The journal checksum feature adds two new flags i.e
658 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
659 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
660 indicates that the commit block contains the checksum for
661 the blocks described by the descriptor blocks. Now commit
662 record can be sent to disk without waiting for descriptor
663 blocks to be written to disk. This behavior is controlled
664 using JBD2_FEATURE_ASYNC_COMMIT flag.
668 Description: OSS needs mutliple precreate threads
669 Details : Add ability to start more than one create thread per OSS.
673 Description: Sanity.sh test_65k failed: setstripe should have succeeded
674 Details : execute lfs setstripe on client
678 Description: mds_obd_create error creating tmp object
679 Details : When the user sets quota on root, llog will be affected and can't
680 create files and write files.
683 Frequency : Always on ia64 patchless client, and possibly others.
685 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
686 Details : This allows the patchless client to be loaded on architectures
691 Description: RedHat Update kernel for RHEL5
692 Details : Add the kernel config file for RHEL5.
696 Description: RedHat Update kernel for RHEL5
697 Details : Modify the kernel config file more closer RHEL5.
701 Description: Build failure against Centos5 (RHEL5)
702 Details : Define PAGE_SIZE when it isn't present.
706 Description: test_7 on sanity-quota.sh can't be used on separate machines
707 Details : add facet to handle it
711 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
712 Details : with NFS, the anon dentry's parent was set to itself in
713 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
714 corresponding dentry other than use rec->ur_name.
716 Severity : enhancement
718 Description: lfs setstripe enhancement
719 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
723 Description: enable data checksumming by default
724 Details : enable checksum by default, allow --disable-checksum
725 configure option and "-o nochecksum" mount option. Checksums
726 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
730 Description: lustre support for RHEL5
731 Details : Add support for RHEL5.
735 Description: Client eviction due to failover config
736 Details : after a connection loss, the lustre client should attempt to
737 reconnect to the last active server first before trying the
738 other potential connections.
742 Description: when mds and osts use different quota unit(32bit and 64bit),
743 quota will be released repeatly.
744 Details : void sending multiple quota reqs to mds, which will keep the
745 status between the reqs.
749 Description: osts not allocated evenly to files
750 Details : change the condition to increase offset_idx
753 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
755 Description: Data corruption for OSTs that are formatted larger than 2TB
757 Details : When generating the bio request for lustre file writes the
758 sector number would overflow a temporary variable before being
759 used for the IO. The data reads correctly from Lustre (which
760 will overflow in a similar manner) but other file data or
761 filesystem metadata may be corrupted in some cases.
765 Description: Tune the kernel for good SCSI performance.
766 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
767 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
772 Description: rewrite ext2-derived code in obdclass/uuid.c
773 Details : rewrite inherited code (uuid parsing code from ext2 utils)
774 from scratch preserving functionality.
778 Description: "lfs find -obd UUID" prints directories
779 Details : "lfs find -obd UUID" will return all directory names instead
780 of just file names. It is incorrect because the directories
781 do not reside on the OSTs.
785 Description: use i_size_read and i_size_write in 2.6 port
786 Details : replace inode->i_size access with i_size_read/write()
789 Frequency : when removing large files
791 Description: scheduling issue during removal of large Lustre files
792 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
793 It causes scheduling issues when removing large files (17TB in the
797 Frequency : only with liblustre clients on XT3
799 Description: evictions taking too long
800 Details : allow llrd to evict clients directly on OSTs
803 Frequency : only on ppc
805 Description: /proc/fs/lustre/devices broken on ppc
806 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
807 We should make sure the type of 'index' is loff_t and then cast
808 explicitly as needed below. Do not assign an explicitly cast
812 Frequency : only for rhel5
814 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
815 Details : Modify the target file & which_kernel.
817 Severity : enhancement
819 Description: omit set fsid for export NFS
820 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
821 as FS_REQUIRES_DEV for avoid problems with generate fsid.
825 Frequency : Always, for kernels after 2.6.16
826 Description: Fix warning idr_remove called for id=.. which is not allocated.
827 Details : Last kernels save old s_dev before kill super and not allow
828 to restore from callback - restore it before call kill_anon_super.
832 Description: buffer overruns could theoretically occur
833 Details : llapi_semantic_traverse() modifies the "path" argument by
834 appending values to the end of the origin string, and a
835 overrun may occur. Adding buffer overrun check in liblustreapi.
839 Description: Fix error on 'ls .' at the top of the Lustre mount.
840 Details : Don't revalidate dentry if it is a root dentry.
844 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
845 Details : Modify vm-tunables-rhel4.patch.
849 Description: Kernel config for 2.6.18-vanilla.
850 Details : Modify targets/2.6-vanilla.target.in.
851 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
852 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
853 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
854 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
858 Description: improve handling recoverable errors
859 Details : if request processig with error which can be recoverable on server
860 request should be resend, otherwise page released from cache and
865 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
866 Details : Modify target file & which_patch.
867 A flaw was found in the IA32 system call emulation provided
868 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
869 value could be stored in the %RAX register, which could trigger an
870 out-of-bounds system call table access. An untrusted local user
871 could exploit this flaw to run code in the kernel
872 (ie a root privilege escalation). (CVE-2007-4573).
876 Description: O_DIRECT bypasses client statistics.
877 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
878 read_ahead_stats were updated. copied the stats section from
879 osc_send_oap_rpc() into async_internal().
883 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
887 Description: change order of libsysio includes
888 Details : '#include sysio.h' should always come before '#include xtio.h'
892 Description: handle "serious error: objid * already exists" more gracefully
893 Details : If LAST_ID value on disk is smaller than the objects existing in
894 the O/0/d* directories, it indicates disk corruption and causes an
895 LBUG(). If the object is 0-length, then we should use the existing
896 object. This will help to avoid a full fsck in most cases.
898 Severity : enhancement
900 Description: adapt the lustre_config script to support the upgrade case
901 Details : Add "-u" option for lustre_config script to support upgrading 1.4
902 server targets to 1.6 in parallel.
906 Description: To avoid grant space > avaible space when the disk is almost
907 full. Without this patch you might see the error "grant XXXX >
908 available" or some LBUG about grant, when the disk is almost
910 Details : In filter_check_grant, for non_grant cache write, we should
911 check the left space by if (*left > ungranted + bytes), instead
912 of (*left > ungranted), because only we are sure the left space
913 is enough for another "bytes", then the ungrant space should be
914 increase. In client, we should update cl_avail_grant only there
915 is OBD_MD_FLGRANT in the reply.
919 Description: Update RHEL 4 kernel to fix local root privilege escalation.
920 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
921 described in CVE-2007-4573. This problem could allow untrusted
922 local users to gain root access.
925 Frequency : when using O_DIRECT and quotas
927 Description: Incorrect file ownership on O_DIRECT output files
928 Details : block usage reported by 'lfs quota' does not take into account
929 files that have been written with O_DIRECT.
934 Description: touch file failed when fs is not full
935 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
936 otherwise we can get ENOSP while fs is not full.
940 Description: parallel lock callbacks
941 Details : Instead of sending blocking and completion callbacks as separated
942 requests, adding them to a set and sending in parallel.
945 Frequency : only for Cray XT3
946 Bugzilla : 12829/13455
947 Description: Changing primary group doesn't change the group lustre assigns to
949 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
950 overridden with the primary group provided by the group upcall,
951 whereas the supplied fsgid can be trusted if it is in the list of
952 supplementary groups returned by the group upcall.
954 Severity : enhancement
956 Description: Allow masking D_WARNING, D_ERROR messages from console
957 Details : Console messages can now be disabled via lnet.printk.
961 Description: User code with malformed file open parameter crashes client node
962 Details : Before packing join_file req, all the related reference should be
963 checked carefully in case some malformed flags cause fake
964 join_file req on client.
968 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
969 Details : ldlm_completion_ast() assumes that a lock is granted when the req
970 mode is equal to the granted mode. However, it should also check
971 that LDLM_FL_CP_REQD is not set.
975 Description: Heavy nfs access might result in deadlocks
976 Details : After ELC code landed, it is now improper to enqueue any mds
977 locks under och_sem, because enqueue might want to decide to
978 cancel open locks for same inode we are holding och_sem for.
982 Description: Client eviction while running blogbench
983 Details : A lot of unlink operations with concurrent I/O can lead to a
984 deadlock causing evictions. To address the problem, the number of
985 oustanding OST_DESTROY requests is now throttled to
986 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
987 callbacks are processed in priority.
991 Description: enable ACLs on MDS by default
992 Details : ACLs must be enabled on MDS by default.
995 Frequency : PPC/PPC64 only
997 Description: conflicts between asm-ppc64/types.h and lustre_types.h
998 Details : fix duplicated definitions between asm-ppc64/types.h and
999 lustre_types.h on PPC.
1002 Frequency : PPC/PPC64 only
1004 Description: asm-ppc/segment.h does not exist
1005 Details : fix compile issue on PPC.
1009 Description: better handle error messages in extents code
1012 Frequency : RHEL4 only
1014 Description: mkfs is very slow on IA64/RHEL4
1015 Details : A performance regression has been discovered in the MPT Fusion
1016 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1017 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1018 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1020 Severity : enhancement
1022 Description: SNMP support enhancement
1023 Details : Adding total number of sampled request for an MDS node in snmp
1026 Severity : enhancement
1028 Description: Optimize ldlm waiting list processing for PR extent locks
1029 Details : When processing waiting list for read extent lock and meeting read
1030 lock that is same or wider to it that is not contended, skip
1031 processing rest of the list and immediatelly return current
1032 status of conflictness, since we are guaranteed there are no
1033 conflicting locks in the rest of the list.
1037 Description: Time out and refuse to reconnect
1038 Details : When the failover node is the primary node, it is possible
1039 to have two identical connections in imp_conn_list. We must
1040 compare not conn's pointers but NIDs, otherwise we can defeat
1041 connection throttling.
1045 Description: port llog fixes from b1_6 into HEAD
1046 Details : Port llog reference couting and some llog cleanups from b1_6
1047 (bug 10800) into HEAD, for protect from panic and access to already
1048 free llog structures.
1052 Description: Detect stride IO mode in read-ahead
1053 Details : When a client does stride read, read-ahead should detect that and
1054 read-ahead pages according to the detected stride pattern.
1058 Description: data checksumming impacts single node performance
1059 Details : add support for several checksum algorithm. Currently, only CRC32
1060 and Adler-32 are supported. The checksum type can be changed on
1061 the fly via /proc/fs/lustre/osc/*/checksum_type.
1065 Description: use adler32 for page checksums
1066 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1071 Description: build for x2 fails
1072 Details : fix compile issue on Cray systems.
1076 Description: Properly match for duplicate locks
1077 Details : Due to different lock order from skiplists code, we need to
1078 traverse entire list for now
1081 Frequency : only on PPC/SLES10
1083 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1084 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1085 be defined. Add a hack in mkfs_lustre.c to work around this compile
1090 Description: LASSERT on MDS when client holding flock lock dies
1091 Details : ldlm pool logic depends on number of granted locks equal to
1092 number of released locks which is not true for flock locks, so
1093 just exclude such locks from consideration.
1097 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1098 Details : Use fsfilt_sync() for both the whole filesystem sync and
1099 individual file sync to eliminate dangerous inode locking
1100 with I_LOCK that can lead to a deadlock.
1104 Description: performance in 1.6.3
1105 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1106 sure that 1MB requests are not fragmented by the block layer.
1108 Severity : enhancement
1110 Description: organize the server-side client stats on per-nid basis
1111 Details : Change the structure of stats under obdfilter and mds to
1119 The "uuid"s file would list the uuids of _active_ exports.
1120 And the clear entry is to clear all stats and stale nids.
1122 Severity : enhancement
1124 Description: eliminate client locks in face of contention
1125 Details : file contention detection and lockless i/o implementation
1126 for contended files.
1130 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1133 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1135 Description: MDS may lose file striping (and hence file data) in some cases
1136 Details : If there are additional extended attributes stored on the MDS,
1137 in particular ACLs, SELinux, or user attributes (if user_xattr
1138 is specified for the client mount options) then there is a risk
1139 of attribute loss. Additionally, the Lustre file striping
1140 needs to be larger than default (e.g. striped over all OSTs),
1141 and an additional attribute must be stored initially in the
1142 inode and then increase in size enough to be moved to the
1143 external attribute block (e.g. ACL growing in size) for file
1148 Description: skiplist implementation simplification
1149 Details : skiplists are used to group compatible locks on granted list
1150 that was implemented as tracking first and last lock of each lock group
1151 the patch changes that to using doubly linked lists
1155 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1156 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1157 because LASSERT on both the data supplied by a client, and the data
1158 on disk is dangerous and incorrect.
1160 Severity : enhancement
1162 Description: Slow truncate/writes to huge files at high offsets.
1163 Details : Directly associate cached pages to lock that protect those pages,
1164 this allows us to quickly find what pages to write and remove
1165 once lock callback is received.
1169 Description: more ldlm soft lockups
1170 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1171 starve other threads from the resource lock for a long time in
1172 case of long waiting queue, so change the debug level from
1173 D_OTHER to the less frequently used D_INFO.
1175 Severity : enhancement
1177 Description: add -gid, -group, -uid, -user options to lfs find
1181 Description: Hung threads in invalidate_inode_pages2_range
1182 Details : The direct IO path doesn't call check_rpcs to submit a new RPC once
1183 one is completed. As a result, some RPCs are stuck in the queue
1188 Description: filter threads hungs on waiting journal commit
1189 Details : Cleanup filter group llog code, then only filter group llog will
1190 be only created in the MDS/OST syncing process.
1194 Description: Procfs and llog threads access destoryed import sometimes.
1195 Details : Sync the import destoryed process with procfs and llog threads by
1196 the import refcount and semaphore.
1198 Severity : enhancement
1200 Description: openlock cache of b1_6 port to HEAD
1205 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1206 Details : If insertion of an extent fails, then discard the inode
1207 preallocation and free data blocks else it can lead to duplicate
1212 Description: don't always update ctime in ext3_xattr_set_handle()
1213 Details : Current xattr code updates the inode ctime in ext3_xattr_set_handle.
1214 In some cases the ctime should not be updated, for example for
1215 2.0->1.8 compatibility it is necessary to delete an xattr and it
1216 should not update the ctime.
1220 Bugzilla : 15713/16362
1221 Description: Assertion in iopen_connect_dentry in 1.6.3
1222 Details : looking up an inode via iopen with the wrong generation number can
1223 populate the dcache with a disconneced dentry while the inode
1224 number is in the process of being reallocated. This causes an
1225 assertion failure in iopen since the inode's dentry list contains
1226 both a connected and disconnected dentry.
1228 --------------------------------------------------------------------------------
1230 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
1232 * Support for kernels:
1233 2.6.5-7.283 (SLES 9),
1234 2.6.9-55.EL (RHEL 4),
1235 2.6.16.46-0.14 (SLES 10),
1236 2.6.18.8 vanilla (kernel.org)
1237 * Client support for unpatched kernels:
1238 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1239 2.6.16 - 2.6.22 vanilla (kernel.org)
1240 * Due to recently discovered recovery problems, we do not recommend
1241 using patchless RHEL 4 clients with this or any earlier release.
1242 * Recommended e2fsprogs version: 1.39.cfs8
1243 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1244 * Starting with this release, the ldiskfs backing filesystem required
1245 by Lustre is now in its own package, lustre-ldiskfs. This package
1246 should be installed. It is versioned separately from Lustre and
1247 may be released separately in future.
1251 Description: block reactivating mgc import until all deactivates complete
1252 Details : Fix race when failing back MDT/MGS to itself (testing)
1255 Frequency : at statup only
1257 Description: mds_lov_synchronize race leads to various problems
1258 Details : simultaneous MDT->OST connections at startup can cause the
1259 sync to abort, leaving the OSC in a bad state.
1261 Severity : enhancement
1263 Description: add optional extra BUILD_VERSION info
1264 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1265 to override the lustre version.
1268 Frequency : 2.6.18 servers only
1270 Description: ll_kern_mount() doesn't release the module reference
1271 Details : The ldiskfs module reference count never drops down to 0
1272 because ll_kern_mount() doesn't release the module reference.
1277 Description: server LBUG when using old ost_num_threads parameter
1278 Details : Accept the old ost_num_threads parameter but warn that it
1279 is deprecated, and fix an off-by-one error that caused an LBUG.
1284 Description: Transient SCSI error results in persistent IO issue
1285 Details : iobuf->dr_error is not reinitialized to 0 between two
1289 Frequency : sometimes when underlying device returns I/O errors
1291 Description: OSTs not going read-only during write failures
1292 Details : OSTs are not remounted read-only when the journal commit threads
1293 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1294 instead of the ext3 wrappers.
1298 Description: poor connect scaling with increasing client count
1299 Details : Don't run filter_grant_sanity_check for more than 100 exports
1300 to improve scaling for large numbers of clients.
1303 Frequency : SLES10 only
1305 Description: sanity-quota.sh quotacheck failed: rc = -22
1306 Details : Quotas cannot be enabled on SLES10.
1309 Frequency : liblustre clients only
1311 Description: getdirentries does not give error when run on compute nodes
1312 Details : getdirentries does not fail when the size specified as an argument
1313 is too small to contain at least one entry
1315 Severity : enhancement
1317 Description: Add LNET router traceability for debug purposes
1318 Details : If a checksum failure occurs with a router as part of the
1319 IO path, the NID of the last router that forwarded the bulk data
1320 is printed so it can be identified.
1325 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1326 Details : Due to a race condition, liblustre clients were occasionally
1327 evicted incorrectly.
1329 Severity : enhancement
1331 Description: lfs setstripe use optional parameters instead of postional
1334 Severity : enhancement
1336 Description: Nanosecond timestamp support for ldiskfs
1337 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1338 resolution timestamps. There is not yet support for this at
1339 the Lustre filesystem level.
1342 Frequency : during server recovery
1344 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1345 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1349 Frequency : nfs export on patchless client
1351 Description: connectathon hang when test nfs export over patchless client
1352 Details : Disconnected dentry cannot be found with lookup, so we do not need
1353 to unhash it or make it invalid
1356 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1357 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1358 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1363 Description: open req refcounting wrong on reconnect
1364 Details : If reconnect happened between getting open reply from server and
1365 call to mdc_set_replay_data in ll_file_open, we will schedule
1366 replay for unreferenced request that we are about to free.
1367 Subsequent close will crash in variety of ways.
1368 Check that request is still eligible for replay in
1369 mdc_set_replay_data().
1374 Description: disable writes to filesystem when reading health_check file
1375 Details : the default for reading the health_check proc file has changed
1376 to NOT do a journal transaction and write to disk, because this
1377 can cause reads of the /proc file to hang and block HA state
1378 checking on a healthy but otherwise heavily loaded system. It
1379 is possible to return to the previous behaviour during configure
1380 with --enable-health-write.
1382 Severity : enhancement
1384 Description: 64-bit inode version
1385 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1386 to the inode. This will be required for version-based recovery.
1391 Description: MDS fails to start if a duplicate client export is detected
1392 Details : in some rare cases it was possible for a client to connect to
1393 an MDS multiple times. Upon recovery the MDS would detect this
1394 and fail during startup. Handle this more gracefully.
1396 Severity : enhancement
1398 Description: Add -o localflock option to simulate old noflock
1400 Details : This will achieve local-only flock/fcntl locks
1406 Description: log_commit_thread vs filter_destroy race leads to crash
1407 Details : Take import reference before releasing llog record semaphore
1412 Description: Wrong request locking in request set processing
1413 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1414 imp_delayed_list, in this place should be used imp_lock.
1417 Frequency : when reconnection
1419 Description: Grant Leak when osc reconnect to OST
1420 Details : When osc reconnect ost, OST(filter) should check whether it
1421 should grant more space to client by comparing fed_grant and
1422 cl_avail_grant, and return the granted space to client instead
1423 of "new granted" space, because client will call osc_init_grant
1424 to update the client grant space info.
1427 Frequency : when client reconnect to OST
1429 Description: Grant Leak when osc do resend and replay bulk write
1430 Details : When osc reconnect to OST, OST(filter)should clear grant info of
1431 bulk write request, because the grant info while be sync between
1432 OSC and OST when reconnect, and we should ignore the grant info
1433 these of resend/replay write req.
1438 Description: Grant space more than avaiable left space sometimes.
1439 Details : When then OST is about to be full, if two bulk writing from
1440 different clients came to OST. Accord the avaliable space of the
1441 OST, the first req should be permitted, and the second one
1442 should be denied by ENOSPC. But if the seconde arrived before
1443 the first one is commited. The OST might wrongly permit second
1444 writing, which will cause grant space > avaiable space.
1447 Frequency : when client is evicted
1449 Description: Grant might be wrongly erased when osc is evicted by OST
1450 Details : when the import is evicted by server, it will fork another
1451 thread ptlrpc_invalidate_import_thread to invalidate the
1452 import, where the grant will be set to 0. While the original
1453 thread will update the grant it got when connecting. So if
1454 the former happened latter, the grant will be wrongly errased
1455 because of this race.
1460 Description: Checking Stale with correct fid
1461 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1462 to check whether it is stale, because sometimes, we want the
1463 enqueue happened anyway, and op_data.fid2 will not be initialized.
1465 Severity : enhancement
1467 Description: update patchless client
1468 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1471 Frequency : only with 2.4 kernel
1473 Description: random memory corruption
1474 Details : size of struct ll_inode_info is to big for union inode.u and this
1475 can be cause of random memory corruption.
1480 Description: Memory leak in recovery
1481 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1482 It should also check obd_fail before fsfilt_start, otherwise if
1483 fsfilt_start return -EROFS,(failover mds during mds recovery).
1484 then the req will return with repmsg->transno = 0 and rc = EROFS.
1485 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1486 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1487 be freed no matter whether fsfilt_commit success or not.
1490 Frequency : only with huge count clients
1492 Description: Prevents from taking the superblock lock in llap_from_page for
1494 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1495 for llap_from_page call in ll_removepage prevents from taking the
1496 superblock lock for a soon died page.
1501 Description: Not check open intent error before release open handle
1502 Details : in some rare cases, the open intent error is not checked before
1503 release open handle, which may cause
1504 ASSERTION(open_req->rq_transno != 0), because it tries to release
1505 the failed open handle.
1510 Description: Set cat log bitmap only after create log success.
1511 Details : in some rare cases, the cat log bitmap is set too early. and it
1512 should be set only after create log success.
1516 Description: Accessing a block bevice can re-enable I/O when Lustre is
1517 tearing down a device.
1518 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1523 Description: service threads may hog cpus when there are a lot of requests
1524 Details : Insert cond_resched to give other threads a chance to use some CPU
1529 Description: the cat log was not initialized in recovery
1530 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1531 unlink log on mds will not be initialized until mds post
1532 recovery. And also in mds post recovery, the unlink log will
1533 initialization will be done asynchronausly, so there will be race
1534 between add unlink log and unlink log initialization.
1538 Description: brw_stats were being printed incorrectly
1539 Details : brw_stats were being printed as log2 but all of them were not
1540 recorded as log2. Also remove some code duplication arising from
1541 filter_tally_{read,write}.
1545 Frequency : rare, only in recovery.
1546 Description: ASSERTION(req->rq_type != LI_POISON) failed
1547 Details : imp_lock should be held while iterating over imp_sending_list for
1548 prevent destroy request after get timeout in ptlrpc_queue_wait.
1552 Description: replay-single.sh test 52 fails
1553 Details : A lock's skiplist need to be cleanup when it being unlinked
1554 from its resource list.
1558 Description: Short directio read returns full requested size rather than
1560 Details : Direct I/O operations should return actual amount of bytes
1561 transferred rather than requested size.
1563 Severity : enhancement
1565 Description: metadata RPC reduction (e.g. for rm performance)
1566 Details : decrease the amount of synchronous RPC between clients and servers
1567 by canceling conflicing lock before the operation on the client
1568 and packing thier handles into the main operation RPC to server.
1570 Severity : enhancement
1572 Description: Async OSC create to avoid the blocking unnecessarily.
1573 Details : If a OST has no remain object, system will block on the creating
1574 when need to create a new object on this OST. Now, ways use
1575 pre-created objects when available, instead of blocking on an
1576 empty osc while others are not empty. If we must block, we block
1577 for the shortest possible period of time.
1579 Severity : enhancement
1581 Description: refine locking for avoid write wrong info into lov_objid file
1582 Details : fix possible races with add new target and write/update data in
1585 --------------------------------------------------------------------------------
1587 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1592 Frequency : on some architectures
1594 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1595 Details : Uninitialized flags sometimes cause configuration commands to
1599 Frequency : patchless clients only
1601 Description: missing __iget() symbol export
1602 Details : The __iget() symbol export is missing. To avoid the need for
1603 this on patchless clients the deathrow inode reaper is turned
1604 off, and we depend on the VM to clean up old inodes. This
1605 dependency was during via the fix for bug 12181.
1607 --------------------------------------------------------------------------------
1609 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1611 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1612 INTEROPERATE with older versions automatically. In many cases a
1613 special upgrade step is needed. Please read the
1614 user documentation before upgrading any part of a live system.
1615 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1616 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1617 * WARNING: Lustre configuration and startup changes are required with
1618 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1624 Severity : enhancement
1626 Description: MountConf
1627 Details : Lustre configuration is now managed via mkfs and mount
1628 commands instead of lmc and lconf. New obd types (MGS, MGC)
1629 are added for dynamic configuration management. See
1630 https://mail.clusterfs.com/wikis/lustre/MountConf for
1633 Severity : enhancement
1635 Description: dynamic OST addition
1636 Details : OSTs can now be added to a live filesystem
1638 Severity : enhancement
1640 Description: startup order invariance
1641 Details : MDTs and OSTs can be started in any order. Clients only
1642 require the MDT to complete startup.
1644 Severity : enhancement
1646 Description: parallel, asynchronous orphan cleanup
1647 Details : orphan cleanup is now performed in separate threads for each
1648 OST, allowing parallel non-blocking operation.
1650 Severity : enhancement
1652 Description: optimized stripe assignment
1653 Details : stripe assignments are now made based on ost space available,
1654 ost previous usage, and OSS previous usage, in order to try
1655 to optimize storage space and networking resources.
1657 Severity : enhancement
1659 Description: Permanently set tunables
1660 Details : All writable /proc/fs/lustre tunables can now be permanently
1661 set on a per-server basis, at mkfs time or on a live system.
1663 Severity : enhancement
1665 Description: Lustre message v2
1666 Details : Add lustre message format v2.
1668 Severity : enhancement
1670 Description: client OST exclusion list
1671 Details : Clients can be started with a list of OSTs that should be
1672 declared "inactive" for known non-responsive OSTs.
1676 Description: SPEC SFS validation failure on NFS v2 over lustre.
1677 Details : Changes the blocksize for regular files to be 2x RPC size,
1678 and not depend on stripe size.
1680 Severity : enhancement
1682 Description: Multiple MD RPCs in flight.
1683 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1684 To avoid overly-overloading MDS, introduce a limit on number of
1685 MDS RPCs in flight for a single client and add /proc controls
1686 to adjust this limit.
1688 Severity : enhancement
1690 Description: client read/write statistics
1691 Details : Add client read/write call usage stats for performance
1692 analysis of user processes.
1693 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1694 file access. extents_stats shows chunk size distribution.
1695 extents_stats_per_process show chunk size distribution per
1698 Severity : enhancement
1700 Description: mds statistics
1701 Details : Add detailed mds operations statistics in
1702 /proc/fs/lustre/mds/*/stats.
1706 Description: Failure of copying files with lustre special EAs.
1707 Details : Client side always return success for setxattr call for lustre
1708 special xattr (currently only "trusted.lov").
1713 Description: Refcount LNET uuids
1714 Details : The global LNET uuid list grew linearly with every startup;
1715 refcount repeated list entries instead of always adding to
1718 Severity : enhancement
1720 Description: Dynamic service threads
1721 Details : Within a small range, start extra service threads
1722 automatically when the request queue builds up.
1725 Frequency : mixed-endian client/server environments
1727 Description: mixed-endian crashes
1728 Details : The new msg_v2 system had some failures in mixed-endian
1731 Severity : enhancement
1733 Description: Easy OST removal
1734 Details : OSTs can be permanently deactivated with e.g. 'lctl
1735 conf_param lustre-OST0001.osc.active=0'
1737 Severity : enhancement
1739 Description: MGS proc entries
1740 Details : Added basic proc entries for the MGS showing what filesystems
1743 Severity : enhancement
1745 Description: provide MGS failover
1746 Details : Added config lock reacquisition after MGS server failover.
1748 Severity : enhancement
1750 Description: add Linux 2.4 support
1751 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1755 Description: a large application tries to do I/O to the same resource and dies
1756 in the middle of it.
1757 Details : Check the req->rq_arrival time after the call to
1758 ost_brw_lock_get(), but before we do anything about
1759 processing it & sending the BULK transfer request. This
1760 should help move old stale pending locks off the queue as
1761 quickly as obd_timeout.
1764 Frequency : when an incorrect nid is specified during startup
1766 Description: ptlrpc connect to non-existant node causes kernel crash
1767 Details : LNET can't be re-entered from an event callback, which
1768 happened when we expire a message after the export has been
1769 cleaned up. Instead, hand the zombie cleanup off to another
1772 Severity : enhancement
1774 Description: plain/inodebits lock performance improvement
1775 Details : Grouping plain/inodebits in granted list by their request modes
1776 and bits policy, thus improving the performance of search through
1780 Frequency : only if OST filesystem is corrupted
1782 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1783 Details : for a short time RPCs with bulk IO are in the replay list,
1784 but replay of bulk IOs is unimplemented. If the OST filesystem
1785 is corrupted due to disk cache incoherency and then replay is
1786 started it is possible to trip an assertion. Avoid putting
1787 committed RPCs into the replay list at all to avoid this issue.
1790 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1792 Description: System hang on startup
1793 Details : This bug allowed the liblustre (e.g. catamount) client to
1794 return to the app before handling all startup RPCs. This
1795 could leave the node unresponsive to lustre network traffic
1796 and manifested as a server ptllnd timeout.
1798 Severity : enhancement
1800 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1801 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1802 boolean to control whether to print peer debug info when a
1803 client's RPC times out.
1806 Frequency : only for kernels with patches from Lustre below 1.4.3
1808 Description: Remove old rdonly API
1809 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1812 Frequency : only for devices with external journals
1814 Description: Set external device read-only also
1815 Details : During a commanded failover stop, we set the disk device
1816 read-only while the server shuts down. We now also set any
1817 external journal device read-only at the same time.
1820 Frequency : when upgrading from 1.4 while trying to change parameters
1822 Description: The wrong (new) MDC name was used when setting parameters for
1823 upgraded MDT's. Also allows changing of OSC (and MDC)
1824 parameters if --writeconf is specified at tunefs upgrade time.
1827 Frequency : when setting specific ost indicies
1829 Description: QOS code breaks on skipped indicies
1830 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1831 created with --index need not be sequential.
1833 Severity : enhancement
1835 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1836 Details : The uninit_groups feature works in conjunction with the kernel
1837 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1838 up the pass1 processing of e2fsck. This is a read-only feature
1839 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1840 mount filesystems that have had this feature enabled.
1842 Severity : enhancement
1844 Description: Improve multi-block allocation algorithm to avoid fragmentation
1845 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1846 allocation locality and avoid filesystem fragmentation.
1848 ------------------------------------------------------------------------------
1850 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1852 * Support for kernels:
1853 2.4.21-47.0.1.EL (RHEL 3)
1854 2.6.5-7.283 (SLES 9)
1855 2.6.9-42.0.10.EL (RHEL 4)
1856 2.6.12.6 vanilla (kernel.org)
1857 2.6.16.27-0.9 (SLES 10)
1858 * Recommended e2fsprogs version: 1.39.cfs5
1860 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1864 Frequency : occasional, depends on client load and configuration
1865 Bugzilla : 12181, 12203
1866 Description: data loss for recently-modified files
1868 Details : In some cases it is possible that recently written or created
1869 files may not be written to disk in a timely manner (this should
1870 normally be within 30s unless client IO load is very high).
1871 The problem appears as zero-length files or files that are a
1872 multiple of 1MB in size after a client crash or client eviction
1873 that are missing data at the end of the file.
1875 This problem is more likely to be hit on clients where files are
1876 repeatedly created and unlinked in the same directory, clients
1877 have a large amount of RAM, have many CPUs, the filesystem has
1878 many OSTs, the clients are rebooted frequently, and/or the files
1879 are not accessed by other nodes after being written.
1881 The presence of the problem can be detected by looking at
1882 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1883 smaller than the second (nr_unused) then dirty files will not
1884 be flushed automatically to disk. "sync; sleep 10" should be
1885 run several times on the node before unmounting it to update
1886 Lustre (this is also safe to run on nodes without this problem).
1888 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1889 that can cause this same problem, so customers using that kernel
1890 also need to update the kernel in addition to Lustre. In order
1891 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1893 It is normal that files written just before a client crash (less
1894 than 30s) may not yet have been flushed to disk, even for local
1898 Frequency : frequent on thin XT3 nodes
1900 Description: UUID collision on thin XT3 Linux nodes
1901 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1902 randomly, since we relied on an insufficiently-seeded PRNG.
1907 Description: OSS hangs after "All ost request buffers busy"
1908 Details : A deadlock between quota and journal operations caused OSS
1909 hangs after printing "All ost request buffers busy."
1912 Frequency : always on liblustre builds
1914 Description: Cleanup compiler warnings on liblustre
1917 Frequency : always on liblustre builds on XT3
1919 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1923 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1924 > 32000 subdirectories
1925 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1926 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1927 subdirectory count crosses 32000. This will aid e2fsck to
1928 correctly handle more than 32000 subdirectories.
1931 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1933 Description: System hang on startup
1934 Details : This bug allowed the liblustre (e.g. catamount) client to
1935 return to the app before handling all startup RPCs. This
1936 could leave the node unresponsive to lustre network traffic
1937 and manifested as a server ptllnd timeout.
1939 Severity : enhancement
1941 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1942 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1943 boolean to control whether to print peer debug info when a
1944 client's RPC times out.
1949 Description: make O_SYNC working on 2.6 kernels
1950 Details : 2.6 kernels use different method for mark pages for write,
1951 so need add a code to lustre for O_SYNC work.
1956 Description: Failure to close file and release space on NFS
1957 Details : Put inode details into lock acquired in ll_intent_file_open.
1958 Use mdc_intent_lock in ll_intent_open to properly
1959 detect all kind of errors unhandled by mdc_enqueue.
1964 Description: proc file read during shutdown sometimes raced obd removal,
1966 Details : Add lock to prevent obd access after proc file removal.
1969 Frequency : Only for files larger than 4GB on 32-bit clients.
1971 Description: improperly doing page alignment of locks
1972 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1973 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1978 Description: RPCs being resent when they shouldn't be
1979 Details : Some RPCs that should not be resent are being resent. This
1980 can cause inconsistencies in the RPC state machine. Do not
1981 resend such requests.
1984 Frequency : rare, only with NFS export
1986 Description: Crash on NFS re-export node
1987 Details : under very unusual load conditions an assertion is hit in
1988 ll_intent_file_open()
1991 Frequency : only if OST filesystem is corrupted
1993 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1994 Details : for a short time RPCs with bulk IO are in the replay list,
1995 but replay of bulk IOs is unimplemented. If the OST filesystem
1996 is corrupted due to disk cache incoherency and then replay is
1997 started it is possible to trip an assertion. Avoid putting
1998 committed RPCs into the replay list at all to avoid this issue.
2003 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2004 Details : Large single O_DIRECT read and write calls can fail to allocate
2005 a sufficiently large buffer to process the request. In case of
2006 allocation failure the allocation is retried with a smaller
2007 buffer and broken into smaller requests.
2009 Severity : enhancement
2011 Description: Add -o localflock option to simulate old noflock behaviour.
2012 Details : This will achieve local-only flock/fcntl locks coherentness.
2017 Description: versioning check is incomplete
2018 Details : Checking the version difference of client vs. server, report
2019 error if the gap is too big.
2024 Description: add support PG_writeback bit
2025 Details : add support for PG_writeback bit for Lustre, for more carefull
2026 work with page cache in 2.6 kernel. This also fix some deadlocks
2027 and remove hack for work O_SYNC with 2.6 kernel.
2029 Severity : enhancement
2031 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2032 Details : The uninit_groups feature works in conjunction with the kernel
2033 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2034 up the pass1 processing of e2fsck. This is a read-only feature
2035 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2036 mount filesystems that have had this feature enabled.
2038 Severity : enhancement
2040 Description: Improve multi-block allocation algorithm to avoid fragmentation
2041 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2042 allocation locality and avoid filesystem fragmentation.
2044 ------------------------------------------------------------------------------
2046 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2048 * Support for kernels:
2049 2.6.9-42.0.3.EL (RHEL 4)
2050 2.6.5-7.276 (SLES 9)
2051 2.4.21-47.0.1.EL (RHEL 3)
2052 2.6.12.6 vanilla (kernel.org)
2053 2.6.16.21-0.8 (SLES10)
2054 * Recommended e2fsprogs version: 1.39.cfs2-0
2056 * The backwards-compatible /proc/sys/portals symlink has been removed
2057 in this release. Before upgrading, please ensure that you change
2058 any configuration scripts or /etc/sysctl.conf files that access
2059 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2060 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2061 in advance of the upgrade on any system running Lustre 1.4.6 or
2062 newer, since /proc/sys/lnet was added in that version.
2063 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2067 Frequency : only when quota is used
2069 Description: avoid scanning export list for quota master
2070 Details : Change the algorithms to avoid scanning export list in order
2071 to improve the efficiency.
2074 Frequency : MDS failover only, very rarely
2076 Description: "went back in time" messages on mds failover
2077 Details : The greatest transno may be lost when the current operation
2078 finishes with an error (transno==0) and the client's last_rcvd
2079 record is over-written. Save the greatest transno in the
2080 mds_last_transno for this case.
2083 Frequency : always for specific kernels and striping counts
2085 Description: client may get "Matching packet too big" without ACL support
2086 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2087 when trying to access files in certain configurations. The
2088 clients should in fact be denied when mounting because they do
2089 not understand ACLs.
2092 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2094 Description: many clients connecting with IO in progress causes connect timeouts
2095 Details : Avoid synchronous journal commits to avoid delays caused by many
2096 clients connecting/disconnecting when bulk IO is in progress.
2097 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2098 OST_IO_PORTAL to avoid delays behind potentially many pending
2102 Frequency : occasionally with multiple writers to a single file
2104 Description: shared writes to file may result in wrong size reported by stat()
2105 Details : Allow growing of kms when extent lock is cancelled
2108 Frequency : always with random mmap IO to multi-striped file
2110 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2111 Details : Only if the hole is at the end of OST object so that kms is too
2112 small. Fix is to increase kms accordingly in ll_nopage.
2115 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2117 Description: writes to a missing object would leak memory on the OST
2118 Details : If there is an inconsistency between the MDS and OST filesystems,
2119 such that the MDS references an object that doesn't exist, writes
2120 to that object will leak memory due to incorrect cleanup in the
2121 error handling path, eventually running out of memory on the OST.
2126 Description: Creating too long symlink causes lustre errors
2127 Details : Check symlink and name lengths before sending requests to MDS.
2130 Frequency : only if flock is enabled (not on by default)
2132 Description: posix locks not released on fd closure on 2.6.9+
2133 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2134 kernels, this caused such locks not to be released on fd close and
2135 then assertions on fs unmount about still used locks.
2138 Frequency : MDS failover only, very rarely
2140 Description: clients may get ASSERTION(granted_lock != NULL)
2141 Details : When request was taking a long time, and a client was resending
2142 a getattr by name lock request. The were multiple lock requests
2143 with the same client lock handle and
2144 mds_getattr_name->fixup_handle_for_resent_request found one of the
2145 lock handles but later failed with ASSERTION(granted_lock != NULL).
2150 Description: handle->h_buffer_credits > 0, assertion failure
2151 Details : h_buffer_credits is zero after truncate, causing assertion
2152 failure. This patch extends the transaction or creates a new
2156 Frequency : NFS re-export or patchless client
2157 Bugzilla : 11179, 10796
2158 Description: Crash on NFS re-export node (__d_move)
2159 Details : We do not want to hash the dentry if we don't have a lock.
2160 But if this dentry is later used in d_move, we'd hit uninitialised
2161 list head d_hash, so we just do this to init d_hash field but
2162 leave dentry unhashed.
2165 Frequency : NFS re-export or patchless client
2167 Description: NFS exports has problem with symbolic link
2168 Details : lustre client didn't properly install dentry when re-exported
2169 to NFS or running patchless client.
2172 Frequency : NFS re-export or patchless client
2174 Description: Various nfs/patchless fixes.
2175 Details : fixes reuse disconected alias for lookup process - this fixes
2176 warning "find_exported_dentry: npd != pd",
2177 fix permission error with open files at nfs.
2178 fix apply umask when do revalidate.
2181 Frequency : occasional
2183 Description: Crash on NFS re-export node
2184 Details : calling clear_page() on the wrong pointer triggered oops in
2185 generic_mapping_read().
2188 Frequency : rarely, using O_DIRECT IO
2190 Description: unaligned directio crashes client with LASSERT
2191 Details : check for unaligned buffers before trying any requests.
2194 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2196 Description: stale data returned from RAID cache
2197 Details : If only a small amount of IO is done to the RAID device before
2198 reading it again it is possible to get stale data from the RAID
2199 cache instead of reading it from disk.
2202 Frequency : always for sles10 kernel
2204 Description: sles10 support
2205 Details : ll_follow_link: compile fixes and using of nd_set_link
2206 under newer kernels.
2209 Frequency : depends on arch, kernel and compiler version, always on sles10
2212 Description: recursive or deep enough symlinks cause stack overflow
2213 Details : getting rid of large stack-allocated variable in
2217 Frequency : depends on hardware
2219 Description: lustre write performance loss in the SLES10 kernel
2220 Details : the performance loss is caused by using of write barriers in the
2221 ext3 code. The SLES10 kernel turns barrier support on by
2222 default. The fix is to undo that change for ldiskfs.
2224 ------------------------------------------------------------------------------
2226 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2228 * Support for kernels:
2229 2.6.9-42.0.3EL (RHEL 4)
2230 2.6.5-7.276 (SLES 9)
2231 2.4.21-47.0.1.EL (RHEL 3)
2232 2.6.12.6 vanilla (kernel.org)
2236 Frequency : quota enabled and large files being deleted
2238 Description: releasing more than 4GB of quota at once hangs OST
2239 Details : If a user deletes more than 4GB of files on a single OST it
2240 will cause the OST to spin in an infinite loop. Release
2241 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2246 Description: statfs data retrieved from /proc may be stale or zero
2247 Details : When reading per-device statfs data from /proc, in the
2248 {kbytes,files}_{total,free,avail} files, it may appear
2249 as zero or be out of date.
2252 Frequency : systems with MD RAID1 external journal devices
2254 Description: lconf's call to blkid is confused by RAID1 journal devices
2255 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2256 of returning all block devices that match the journal UUID.
2259 Frequency : always, for aggregate stripe size over 4GB
2261 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2262 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2263 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2266 Frequency : always if "lfs find" used on a local file/directory
2268 Description: "lfs find" segfaults if used on a local file/directory
2269 Details : The case where a directory component was not specified wasn't
2270 handled correctly. Handle this properly.
2273 Frequency : always on ppc64
2275 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2276 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2277 perform endianness conversion.
2280 Frequency : rarely (truncate to non-zero file size after write under load)
2281 Bugzilla : 10730, 10687
2282 Description: Files padded with zeros to next 4K multiple
2283 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2284 it is possible that files that are truncated to a non-zero size
2285 immediately after being written are filled with zero bytes beyond
2286 the truncated size. No file data is lost.
2288 Severity : enhancement
2290 Description: Allow recovery/failover for liblustre clients.
2291 Details : liblustre clients were unaware of failover configurations until
2294 Severity : enhancement
2296 Description: user file locks should fail when not mounting with flock option
2297 Details : Set up an error-returning stub in ll_file_operations.lock field
2298 to prevent incorrect behaviour when client is mounted without
2299 flock option. Also, set up properly f_op->flock field for
2303 Frequency : always on ia64
2305 Description: "lfs df" loops on printing out MDS statfs information
2306 Details : The obd_ioctl_data was not initialized and in some systems
2307 this caused a failure during the ioctl that did not return
2308 an error. Initialize the struct and return an error on failure.
2311 Frequency : SLES 9 only
2313 Description: Error of copying files with lustre special EAs as root
2314 Details : Client side always return success for setxattr call for lustre
2315 special xattr (currently only "trusted.lov").
2318 Frequency : rarely on clusters with both ia64+i386 clients
2320 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2321 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2322 clients are doing concurrent writes to the same file it is
2323 possible that the ia64 clients may LASSERT because the OST
2324 extent locks are not PAGE_SIZE aligned. Ensure that grown
2325 locks are always aligned on the request boundary.
2328 Frequency : specific use, occasional
2330 Description: Overwriting in use executable truncates on-disk binary image
2331 Details : If one node attempts to overwrite an executable in use by
2332 another node, we now correctly return ETXTBSY instead of
2333 truncating the file.
2335 Severity : enhancement
2337 Description: Async OSC create to avoid the blocking unnecessarily.
2338 Details : If a OST has no remain object, system will block on the creating
2339 when need to create a new object on this OST. Now, ways use
2340 pre-created objects when available, instead of blocking on an
2341 empty osc while others are not empty. If we must block, we block
2342 for the shortest possible period of time.
2347 Description: chmod on Lustre root is propagated to other clients
2348 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2349 invalid by the follow_mount time.
2352 Frequency : liblustre clients only
2354 Description: Race in 'instant cancel' lock handling could lead to such locks
2355 never to be granted in case of SMP MDS
2356 Details : Do not destroy not yet granted but cbpending locks in
2360 Frequency : replay/resend of open
2362 Description: non null lock assetion failure in mds_intent_policy
2363 Details : Trying to replay/resend lockless open requests resulted in
2364 mds_open() returning 0 with no lock. Now it sets a flag if
2365 a lock is going to be returned.
2367 Severity : enhancement
2369 Description: Checksum enhancements
2370 Details : New checksum enhancements allow for resending RPCs that failed
2373 Severity : enhancement
2375 Description: Tunables on number of dirty pages in cacche
2376 Details : Allow to set limit on number of dirty pages cached.
2381 Description: client crash on unmount - lock still has references
2382 Details : In some error handling cases it was possible to leak a lock
2383 reference on a client while accessing a file. This was not
2384 harmful to the client during operation, but would cause the
2385 client to crash when the filesystem is unmounted.
2388 Frequency : specific case, rare
2390 Description: ETXTBSY on mds though file not in use by client
2391 Details : ETXTBSY is no longer incorrectly returned when attempting to
2392 chmod or chown a directory that the user previously tried to
2393 execute or a currently-executing binary.
2396 Frequency : extremely rare except on liblustre-based clients
2398 Description: Lustre space not freed when files are deleted
2399 Details : Clean up open-unlinked files after client eviction. Previously
2400 the unlink was skipped and the files remained as orphans.
2405 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2406 Details : In some cases it was possible to send a blocking callback to
2407 a client doing a glimpse, even though that client didn't get
2408 a lock granted. When the glimpse lock is cancelled on the OST
2409 the freed lock is left on the waiting list and corrupted the list.
2412 Frequency : all core dumps
2414 Description: Broke core dumps to lustre
2415 Details : Negative dentry may be unhashed if parent does not have UPDATE
2416 lock, but some callers, e.g. do_coredump, expect dentry to be
2417 hashed after successful create, hash it in ll_create_it.
2419 ------------------------------------------------------------------------------
2421 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2423 * Support for kernels:
2424 2.6.9-42.0.2.EL (RHEL 4)
2425 2.6.5-7.276 (SLES 9)
2426 2.4.21-40.EL (RHEL 3)
2427 2.6.12.6 vanilla (kernel.org)
2431 Frequency : always on RHEL 3
2433 Description: Number of open files grows over time
2434 Details : The number of open files grows over time, whether or not
2435 Lustre is started. This was due to a filp leak introduced
2436 by one of our kernel patches.
2438 ------------------------------------------------------------------------------
2440 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2442 * Support for kernels:
2443 2.6.9-42.EL (RHEL 4)
2444 2.6.5-7.276 (SLES 9)
2445 2.4.21-40.EL (RHEL 3)
2446 2.6.12.6 vanilla (kernel.org)
2451 Bugzilla : 5719, 9635, 9792, 9684
2452 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2453 Details : If a server is under heavy load and cannot reply to new
2454 connection requests before the client resends the (re)connect,
2455 the connection handling code can behave badly if two service
2456 threads are concurrently handing separate (re)connections from
2457 the same client. Add better locking to the connection handling
2458 code, and ensure that only a single connection will be processed
2459 for a given client UUID, even if the lock is dropped.
2461 Severity : enhancement
2463 Description: add TCP zero-copy support to kernel
2464 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2465 sends if the hardware supports scatter-gather and checksumming.
2466 This allows socklnd to do client-write and server-read more
2467 efficiently and reduce CPU utilization from skbuf copying.
2470 Frequency : only if NFS exporting from client
2472 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2473 Details : If setting attributes on a file created under NFS that had
2474 never been opened it would be possible to oops the client
2475 if the file had no objects.
2478 Frequency : always for liblustre
2480 Description: liblustre client does MDS+OSTs setattr RPC for each write
2481 Details : When doing a write from a liblustre client, the client
2482 incorrectly issued an RPC to the MDS and each OST the file was
2483 striped over in order to update the timestamps. When writing
2484 with small chunks and many clients this could overwhelm the MDS
2485 with RPCs. In all cases it would slow down the write because
2486 these RPCs are unnecessary.
2488 Severity : enhancement
2490 Description: allow number of MDS service threads to be changed at module load
2491 Details : It is now possible to change the number of MDS service threads
2492 running. Adding "options mds mds_num_threads={N}" to the MDS's
2493 /etc/modprobe.conf will set the number of threads for the next
2494 time Lustre is restarted (assuming the "mds" module is also
2495 reloaded at that time). The default number of threads will
2496 stay the same, 32 for most systems.
2501 Description: OST crash if filesystem is unformatted or corrupt
2502 Details : If an OST is started on a device that has never been formatted
2503 or if the filesystem is corrupt and cannot even mount then the
2504 error handling cleanup routines would dereference a NULL pointer.
2509 Description: NULL pointer deref in llap_from_page.
2510 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2511 incorrect) mapping if the page was truncated/reclaimed while it was
2512 searched for. Check for this condition and skip such pages when
2513 doing readahead. Introduce extra check to llap_from_page() to
2514 verify page->mapping->host is non-NULL (so page is not anonymous).
2517 Frequency : Sometimes when using sys_sendfile
2519 Description: "page not covered by a lock" warnings from ll_readpage
2520 Details : sendfile called ll_readpage without right page locks present.
2521 Now we introduced ll_file_sendfile that does necessary locking
2522 around call to generic_file_sendfile() much like we do in
2526 Frequency : with certain MDS communication failures at client mount time
2528 Description: NULL pointer deref after failed client mount
2529 Details : a client connection request may delayed by the network layer
2530 and not be sent until after the PTLRPC layer has timed out the
2531 request. If the client fails the mount immediately it will try
2532 to clean up before the network times out the request. Add a
2533 reference from the request import to the obd device and delay
2534 the cleanup until the network drops the request.
2537 Frequency : occasionally during client (re)connect
2539 Description: assertion failure during client (re)connect
2540 Details : processing a client connection request may be delayed by the
2541 client or server longer than the client connect timeout. This
2542 causes the client to resend the connection request. If the
2543 original connection request is replied in this interval, the
2544 client may trip an assertion failure in ptlrpc_connect_interpret()
2545 which thought it would be the only running connect process.
2548 Frequency : only with obd_echo servers and clients that are rebooted
2550 Description: kernel BUG accessing uninitialized data structure
2551 Details : When running an obd_echo server it did not start the ping_evictor
2552 thread, and when a client was evicted an uninitialized data
2553 structure was accessed. Start the ping_evictor in the RPC
2554 service startup instead of the OBD startup.
2556 Severity : enhancement
2557 Bugzilla : 10193 (patchless)
2558 Description: Remove dependency on various unexported kernel interfaces.
2559 Details : No longer need reparent_to_init, exit_mm, exit_files,
2560 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2563 Frequency : rare (only users of deprecated and unsupported LDAP config)
2565 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2566 Details : LDAP apparently contains 'lustreName' attributes instead of
2567 'name'. A simple remapping of the name is sufficient.
2570 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2572 Description: waiting_locks_callback trips kernel BUG if client is evicted
2573 Details : Running with the dump_on_timeout debug flag turned on makes
2574 it possible that the waiting_locks_callback() can try to dump
2575 the Lustre kernel debug logs from an interrupt handler. Defer
2576 this log dumping to the expired_lock_main() thread.
2578 Severity : enhancement
2580 Description: Support NFS exporting on 2.6 kernels.
2581 Details : Implement non-rawops metadata methods for NFS server to use without
2582 changing NFS server code.
2585 Frequency : very rare (synthetic metadata workload only)
2587 Description: two racing renames might cause an MDS thread to deadlock
2588 Details : Running the "racer" program may cause one MDS thread to rename
2589 a file from being the source of a rename to being the target of
2590 a rename at exactly the same time that another thread is doing
2591 so, and the second thread has already enqueued these locks after
2592 doing a lookup of the target and is trying to relock them in
2593 order. Ensure that we don't try to re-lock the same resource.
2596 Frequency : only very large systems with liblustre clients
2598 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2599 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2600 all OSTs in parallel. This allows the eviction of stale liblustre
2601 clients to proceed much faster than if they were done in series,
2602 and also offers similar improvements for other set_info RPCs.
2607 Description: excessive CPU usage during initial read phase on client
2608 Details : During the initial read phase on a client, it would agressively
2609 retry readahead on the file, consuming too much CPU and impacting
2610 performance (since 1.4.5.8). Improve the readahead algorithm
2611 to avoid this, and also improve some other common cases (read
2612 of small files in particular, where "small" is files smaller than
2613 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2618 Description: MDS crash when receiving packet with unknown intent.
2619 Details : Do not LBUG in unknown intent case, just return -EFAULT
2621 Severity : enhancement
2622 Bugzilla : 9293, 9385
2623 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2624 Details : Do not serialize getattr (non-intent version) and statfs.
2627 Frequency : occasional, when OST network is overloaded/intermittent
2629 Description: client evicted by OST after bulk IO timeout
2630 Details : If a client sends a bulk IO request (read or write) the OST
2631 may evict the client if it is unresposive to its data GET/PUT
2632 request. This is incorrect if the network is overloaded (takes
2633 too long to transfer the RPC data) or dropped the OST GET/PUT
2634 request. There is no need to evict the client at all, since
2635 the pinger and/or lock callbacks will handle this, and the
2636 client can restart the bulk request.
2639 Frequency : Always when mmapping file with no objects
2641 Description: client crashes when mmapping file with no objects
2642 Details : Check that we actually have objects in a file before doing any
2643 operations on objects in ll_vm_open, ll_vm_close and
2649 Description: Request leak when working with deleted CWD
2650 Details : Introduce advanced request refcount tracking for requests
2651 referenced from lustre intent.
2653 Severity : Enhancement
2655 Description: Cache open file handles on client.
2656 Details : MDS now will return special lock along with openhandle, if
2657 requested and client is allowed to hold openhandle, even if unused,
2658 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2659 closing files for every read/write openration.
2661 Severity : Enhancement
2663 Description: Cache open negative dentries on client when possible.
2664 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2665 negative dentries on lock revocation.
2670 Description: Remounting a client read-only wasn't possible with a zconf mount
2671 Details : It wasn't possible to remount a client read-only with llmount.
2673 Severity : enhancement
2674 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2675 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2676 MPICH is included to add Lustre-specific ADIO interfaces.
2677 This is based closely on the UFS ADIO layer and only differs
2678 in file creation, in order to allow the OST striping to be set.
2679 This is user-contributed code and not supported by CFS.
2684 Description: extended inode attributes (immutable, append-only) work improperly
2685 when 2.4 and 2.6 kernels are used on client/server or vice versa
2686 Details : Introduce kernel-independent values for these flags.
2688 Severity : enhancement
2691 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2692 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2693 granularity. Now, Fractional MB granularity is be supported,
2694 this is very useful for low memory system.
2696 Severity : enhancement
2698 Description: Getattr by fid
2699 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2700 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2701 have locks and avoids getattr rpc requests for it on every stat.
2704 Frequency : Always, for filesystems larger than 2TB
2706 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2707 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2708 filesystems larger than 2TB to be mounted with mballoc enabled.
2711 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2713 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2714 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2715 filesystems larger than 2TB is mounted then the kernel will
2716 silently corrupt the start of the filesystem. CONFIG_LBD is
2717 enabled for all CFS-supported kernels, but the possibility of
2718 this happening with a modified kernel config exists.
2720 Severity : enhancement
2722 Description: add client O_DIRECT support for 2.6 kernels
2723 Details : It is now possible to do O_DIRECT reads and writes to files
2724 in the Lustre client mountpoint on 2.6 kernel clients.
2726 Severity : enhancement
2728 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2729 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2730 parallel, not waiting for response from every OST before sending
2731 a rpc to the next OST.
2736 Description: setattr vs write race when updating file timestamps
2737 Details : Client processes that update a file timestamp into the past
2738 right after writing to the file (e.g. tar) it is possible that
2739 the updated file modification time can be reset to the current
2740 time due to a race between processing the setattr and write RPC.
2742 Severity : enhancement
2744 Description: Bring 'lfs find' closer in line with regular Linux find.
2745 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2746 -print0 options and obtains all the needed info through the lustre
2749 Severity : enhancement
2751 Description: support up to 1024 configured devices on one node
2752 Details : change obd_dev array from statically allocated to dynamically
2753 allocated structs as they are first used to reduce memory usage
2758 Description: Flush dirty partially truncated pages during truncate
2759 Details : Immediatelly flush partially truncated pages in filter_setattr,
2760 this way we completely avoid having any pages in page cache on OST
2761 and can retire ugly workarounds during writes to flush such pages.
2766 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2767 Details : protect inode from truncation within vfs_unlink() context
2768 just take a reference before calling vfs_unlink() and release it
2769 when parent's i_sem is free.
2774 Description: last_id value checked outside lock on OST caused LASSERT failure
2775 Details : If there were multiple MDS->OST object precreate requests in
2776 flight, it was possible that the OST's last object id was checked
2777 outside a lock and incorrectly tripped an assertion. Move checks
2778 inside locks, and discard old precreate requests.
2781 Frequency : always, if extents are used on OSTs
2783 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2784 Details : OSTs using the extents format would not zero the high 16 bits of
2785 the index physical block number. This is not a problem for any
2786 OST filesystems smaller than 16TB, and no kernels support ext3
2787 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2788 new/modified files) and can be fixed for existing filesystems
2789 with e2fsprogs-1.39-cfs1.
2794 Description: import connection selection may be incorrect if timer wraps
2795 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2796 import connections to be ignored if the 32-bit jiffies counter
2797 wraps. Use a 64-bit jiffies counter.
2800 Frequency : very large clusters immediately after boot
2802 Description: LNET request buffers exhausted under heavy short-term load
2803 Details : If a large number of client requests are generated on a service
2804 that has previously never seen so many requests it is possible
2805 that the request buffer growth cannot keep up with the spike in
2806 demand. Instead of dropping incoming requests, they are held in
2807 the LND until the RPC service can accept more requests.
2810 Frequency : Sometimes during replay
2812 Description: Assertion failure in ll_local_open after replay.
2813 Details : If replay happened on an open request reply before we were able
2814 to set replay handler, reply will become not swabbed tripping the
2815 assertion in ll_local_open. Now we set the handler right after
2816 recognising of open request
2819 Frequency : very rare
2821 Description: kernel reports "badness in vsnprintf"
2822 Details : Reading from the "recovery_status" /proc file in small chunks
2823 may cause a negative length in lprocfs_obd_rd_recovery_status()
2824 call to vsnprintf() (which is otherwise harmless). Exit early
2825 if there is no more space in the output buffer.
2827 Severity : enhancement
2829 Description: clear OBD RPC statistics by writing to them
2830 Details : It is now possible to clear the OBD RPC statistics by writing
2831 to the "stats" file.
2836 Description: Client mtime is not the same on different clients after utimes
2837 Details : In some cases, the client was using the utimes() syscall on
2838 a file cached on another node. The clients now validate the
2839 ctime from the MDS + OSTs to determine which one is right.
2844 Description: Inability to activate failout mode
2845 Details : lconf script incorrectly assumed that in pythong string's numeric
2846 value is used in comparisons.
2849 Frequency : always with multiple stripes per file
2851 Description: Inefficient object allocation for mutli-stripe files
2852 Details : When selecting which OSTs to stripe files over, for files with
2853 a stripe count that divides evenly into the number of OSTs,
2854 the MDS is always picking the same starting OST for each file.
2855 Return the OST selection heuristic to the original design.
2860 Description: mount failures may take full timeout to return an error
2861 Details : Under some heavy load conditions it is possible that a
2862 failed mount can wait for the full obd_timeout interval,
2863 possibly several minutes, before reporting an error.
2864 Instead return an error as soon as the status is known.
2866 Frequency : quota enabled and large files being deleted
2868 Description: releasing more than 4GB of quota at once hangs OST
2869 Details : If a user deletes more than 4GB of files on a single OST it
2870 will cause the OST to spin in an infinite loop. Release
2871 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2876 Description: statfs data retrieved from /proc may be stale or zero
2877 Details : When reading per-device statfs data from /proc, in the
2878 {kbytes,files}_{total,free,avail} files, it may appear
2879 as zero or be out of date.
2882 Frequency : systems with MD RAID1 external journal devices
2884 Description: lconf's call to blkid is confused by RAID1 journal devices
2885 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2886 of returning all block devices that match the journal UUID.
2889 Frequency : always, for aggregate stripe size over 4GB
2891 Description: assertion fails when trying to use 4GB stripe size
2892 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2893 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2896 Frequency : always on ppc64
2898 Description: the first write on an ext3 filesystem with mballoc got stuck
2899 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2900 perform endianness conversion.
2902 ------------------------------------------------------------------------------
2904 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2906 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2907 INTEROPERATE with older versions automatically. Please read the
2908 user documentation before upgrading any part of a live system.
2909 * WARNING: Lustre networking configuration changes are required with
2910 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2913 * Support for kernels:
2914 2.6.9-22.0.2.EL (RHEL 4)
2915 2.6.5-7.244 (SLES 9)
2916 2.6.12.6 vanilla (kernel.org)
2919 Severity : enhancement
2920 Bugzilla : 7981/8208
2921 Description: Introduced Lustre Networking (LNET)
2922 Details : LNET is new networking infrastructure for Lustre, it includes
2923 a reorganized network configuration mode (see the user
2924 documentation for full details) as well as support for routing
2925 between different network fabrics. Lustre Networking Devices
2926 (LNDs) for the supported network fabrics have also been
2927 created for this new infrastructure.
2929 Severity : enhancement
2930 Description: Introduced Access control lists
2931 Details : clients can set ACLs on files and directories in order to have
2932 more fine-grained permissions than the standard Unix UGO+RWX.
2933 The MDS must be started with the "-o acl" mount option.
2935 Severity : enhancement
2936 Description: Introduced filesystem quotas
2937 Details : Administrators may now establish per-user quotas on the
2940 Severity : enhancement
2942 Description: Configuration change for the XT3
2943 The PTLLND is now used to run Lustre over Portals on the XT3
2944 The configure option(s) --with-cray-portals are no longer used.
2945 Rather --with-portals=<path-to-portals-includes> is used to
2946 enable building on the XT3. In addition to enable XT3 specific
2947 features the option --enable-cray-xt3 must be used.
2952 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2953 Details : A race between allocating a new llcd and re-getting the llcd_lock
2954 allowed another thread to grab newly-allocated llcd.
2956 Severity : enhancement
2958 Description: 2.6 OST async journal commit and locking fix to improve performance
2959 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2960 2.6 kernels are now async as they already were in 2.4 kernels so
2961 that they can commit concurrently with the network bulk transfer.
2962 For block-allocated files the filter allocation semaphore is held
2963 to avoid filesystem fragmentation during allocation. BKL lock
2964 removed for 2.6 xattr operations where it is no longer needed.
2969 Description: lconf incorrectly determined whether two IP networks could talk
2970 Details : In some more complicated routing and multiple-network
2971 configurations, lconf will avoid trying to make a network
2972 connection to a disjoint part of the IP space. It was doing the
2973 math incorrectly for one set of cases.
2978 Description: Fix for potential infinite loop processing records in an llog.
2979 Details : If an llog record is corrupted/zeroed, it is possible to loop
2980 forever in llog_process(). Validate the llog record length
2981 and skip the remainder of the block on error.
2984 Frequency : occasional (liblustre only)
2986 Description: liblustre could not open files whose last component is a symlink
2987 Details : sysio_path_walk() would incorrectly pass the open intent to
2988 intermediate path components.
2991 Frequency : rare (liblustre only with non-standard tuning)
2992 Bugzilla : 7201 (7350)
2993 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2994 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2995 after it has been released, only do this for referenced locks
2997 Severity : enhancement
2999 Description: specifying an (invalid) directory default stripe_size of -1
3000 would reset the directory default striping
3001 Details : stripe_size -1 was used internally to signal directory stripe
3002 removal, now use "all default" to signal dir stripe removal
3003 as a directory striping of "all default" is not useful
3006 Frequency : common for large clusters running liblustre clients
3008 Description: doing an ls when liblustre clients are running is slow
3009 Details : sending a glimpse AST to a liblustre client waits for every AST
3010 to time out, as liblustre clients will not respond. Since they
3011 cannot cache data we refresh the OST lock LVB from disk instead.
3013 Severity : enhancement
3015 Description: doing an ls at the same time as file IO can be slow
3016 Details : enqueue and other "small" requests can be blocked behind many
3017 large IO requests. Create a new OST IO portal for non-IO
3018 requests so they can be processed faster.
3021 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3023 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3024 created file with mode 000
3025 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3026 allow an ftruncate() as a non-root user to a file with mode 000.
3027 Lustre now respects this flag to disable mode checks when
3028 truncating a file owned by the user
3031 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3034 Description: Revoking locks from clients that went dead or catatonic might take
3036 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3037 cancellation of such locks instant on servers without waiting for
3038 any reply from clients. Clients drops these locks when cancel
3039 notification from server is received without replying.
3042 Frequency : liblustre-only, when liblustre client dies or becomes busy
3044 Description: Doing ls on Linux clients can take a long time with active
3046 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3047 granting such locks to it in the first place if possible. Locks
3048 are taken by proxy on the OST during the read or write and
3049 dropped immediately afterward. Add connect flags handling, do
3050 not grant locks to liblustre clients for glimpse ASTs.
3052 Severity : enhancement
3054 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3055 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3056 sequential reads and behaves badly with applications doing random
3057 reads. Improve it by reading ahead at least read region, and
3058 avoiding excessive large RPC for small reads.
3060 Severity : enhancement
3062 Description: Creating more than 1000 files for a single job may cause a load
3063 imbalance on the OSTs if there are also a large number of OSTs.
3064 Details : qos_prep_create() uses an OST index reseed value that is an
3065 even multiple of the number of available OSTs so that if the
3066 reseed happens in the middle of the object allocation it will
3067 still utilize the OSTs as uniformly as possible.
3072 Description: OST or MDS may oops in ping_evictor_main()
3073 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3074 but doesn't restart at beginning of obd_exports_timed list
3077 Severity : enhancement
3079 Description: improve by-nid export eviction on the MDS and OST
3080 Details : allow multiple exports with the same NID to be evicted at one
3081 time without re-searching the exports list.
3084 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3086 Description: MDS may oops in groups_free()
3087 Details : in rare race conditions a newly allocated group_info struct is
3088 freed again, and this can be NULL. The 2.4 compatibility code
3089 for groups_free() checked for a NULL pointer, but 2.6 did not.
3092 Frequency : common for liblustre clients doing little filesystem IO
3093 Bugzilla : 9352, 7313
3094 Description: server may evict liblustre clients accessing contended locks
3095 Details : if a client is granted a lock or receives a completion AST
3096 with a blocking AST already set it would not reply to the AST
3097 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3100 Frequency : lfs setstripe, only systems with more than 160 OSTs
3102 Description: unable to set striping with a starting offset beyond OST 160
3103 Details : llapi_create_file() incorrectly limited the starting stripe
3104 index to the maximum single-file stripe count.
3107 Frequency : LDAP users only
3109 Description: lconf did not handle in-kernel recovery with LDAP properly
3110 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3112 Severity : enhancement
3114 Description: bind OST threads to NUMA nodes to improve performance
3115 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3116 node and do their allocations there to localize memory access
3118 Severity : enhancement
3120 Description: llmount can determine client NID directly from Myrinet (GM)
3121 Details : the client NID code from gmnalnid was moved directly into
3122 llmount, removing the need to use this or specifying the
3123 client NID explicitly when mounting GM clients with zeroconf
3126 Frequency : if client is started with down MDS
3128 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3129 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3130 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3131 ptlrpc_import_delayed_req() because we didn't check if the
3132 request was interrupted, and we also didn't break out of the
3133 event loop if there was no timeout
3138 Description: data loss during non-page-aligned writes to a single file from
3139 both multiple nodes and multiple threads on one node at same time
3140 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3141 inconsistency led to false short-reads, that were cached and later
3142 used by ->prepare_write() to fill in partially written page,
3143 leading to data loss.
3146 Frequency : always, if lconf --abort_recovery used
3148 Description: lconf --abort_recovery fails with 'Operation not supported'
3149 Details : lconf was attempting to abort recovery on the MDT device and not
3152 Severity : enhancement
3154 Description: remove cleanup logs
3155 Details : replace lconf-generated cleanup logs with lustre internal
3156 cleanup routines. Eliminates the need for client-cleanup and
3159 Severity : enhancement
3161 Description: add support for EAs (user and system) on lustre filesystems
3162 Details : it is now possible to store extended attributes in the Lustre
3163 client filesystem, and with the user_xattr mount option it
3164 is possible to allow users to store EAs on their files also
3166 Severity : enhancement
3168 Description: Add possibility (config option) to show minimal available OST free
3170 Details : When compiled with --enable-mindf configure option, statfs(2)
3171 (and so, df) will return least minimal free space available from
3172 all OSTs as amount of free space on FS, instead of summary of
3173 free spaces of all OSTs.
3175 Severity : enhancement
3177 Description: do not expand extent locks acquired on OST-side
3178 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3179 by server: they are not cached anyway.
3182 Frequency : when mmap is used/binaries executed from Lustre
3184 Description: Unmmap pages before throwing them away from read cache.
3185 Details : llap_shrink cache now attempts to unmap pages before discarding
3186 them (if unmapping failed - do not discard). SLES9 kernel has
3187 extra checks that trigger if this unmapping is not done first.
3192 Description: lconf didn't resolve symlinks before checking to see whether a
3193 given mountpoint was already in use
3196 Frequency : when migrating failover services
3197 Bugzilla : 6395, 9514
3198 Description: When migrating a subset of services from a node (e.g. failback
3199 from a failover service node) the remaining services would
3200 time out and evict clients.
3201 Details : lconf --force (implied by --failover) sets the global obd_timeout
3202 to 5 seconds in order to quickly disconnect, but this caused
3203 other RPCs to time out too quickly. Do not change the global
3204 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3206 Severity : enhancement
3207 Frequency : if MDS is started with down OST
3208 Bugzilla : 9439,5706
3209 Description: Allow startup/shutdown of an MDS without depending on the
3210 availability of the OSTs.
3211 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3212 Add appropriate locking and lov-osc refcounts for safe
3213 cleaning. Add osc abort_inflight calls in case the
3214 synchronize never started.
3217 Frequency : occasional (Cray XT3 only)
3219 Description: root not authorized to access files in CRAY_PORTALS environment
3220 Details : The client process capabilities were not honoured on the MDS in
3221 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3222 been accessed by an authorized user then root was able to access
3223 the file on the local client also. The root user capabilities
3224 are now allowed on the MDS, as this environment has secure UID.
3227 Frequency : occasional
3229 Description: ldiskfs "too long searching" message happens too often
3230 Details : A debugging message (otherwise harmless) prints too often on
3231 the OST console. This has been reduced to only happen when
3232 there are fragmentation problems on the filesystem.
3237 Description: Division by zero in statfs when all OSCs are inactive
3238 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3239 lov_statfs divides by value returned by lov_get_stripecnt().
3243 Bugzilla : 9489, 3273
3244 Description: First write from each client to each OST was only 4kB in size,
3245 to initialize client writeback cache, which caused sub-optimal
3246 RPCs and poor layout on disk for the first writen file.
3247 Details : Clients now request an initial cache grant at (re)connect time
3248 and so that they can start streaming writes to the cache right
3249 away and always do full-sized RPCs if there is enough data.
3250 If the OST is rebooted the client also re-establishes its grant
3251 so that client cached writes will be honoured under the grant.
3256 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3257 Details : Now I/O RPCs go to different portal number and (presumably) fast
3258 lock requests (and glimses) and other RPCs get their own service
3259 threads pool that should be able to service those RPCs
3262 Severity : enhancement
3264 Description: Ability to exchange lustre version between client and servers and
3265 issue warnings at client side if client is too old. Also for
3266 liblustre clients there is ability to refuse connection of too old
3268 Details : New 'version' field is added to connect data structure that is
3269 filled with version info. That info is later checked by server and
3273 Frequency : rare, liblustre only.
3274 Bugzilla : 9296, 9581
3275 Description: Two simultaneous writes from liblustre at offset within same page
3276 might proceed at the same time overwriting eachother with stale
3278 Details : I/O lock withing llu_file_prwv was released too early, before data
3279 actually was hitting the wire. Extended lock-holding time until
3280 server acknowledges receiving data.
3283 Frequency : extremely rare. Never observed in practice.
3285 Description: avoid generating lustre_handle cookie of 0.
3286 Details : class_handle_hash() generates handle cookies by incrementing
3287 global counter, and can hit 0 occasionaly (this is unlikely, but
3288 not impossible, because initial value of cookie counter is
3289 selected randonly). Value of 0 is used as a sentinel meaning
3290 "unassigned handle" --- avoid it. Also coalesce two critical
3291 sections in this function into one.
3293 Severity : enhancement
3295 Description: allow liblustre clients to delegate truncate locking to OST
3296 Details : To avoid overhead of locking, liblustre client instructs OST to
3297 take extent lock in ost_punch() on client's behalf. New connection
3298 flag is added to handle backward compatibility.
3300 Severity : enhancement
3301 Bugzilla : 4928, 7341, 9758
3302 Description: allow number of OST service threads to be specified
3303 Details : a module parameter allows the number of OST service threads
3304 to be specified via "options ost ost_num_threads={N}" in the
3305 OSS's /etc/modules.conf or /etc/modprobe.conf.
3309 Bugzilla : 6146, 9635, 9895
3310 Description: servers crash with bad pointer in target_handle_connect()
3311 Details : In rare cases when a client is reconnecting it was possible that
3312 the connection request was the last reference for that export.
3313 We would temporarily drop the export reference and get a new
3314 one, but this may have been the last reference and the export
3315 was just destroyed. Get new reference before dropping old one.
3317 Severity : enhancement
3318 Frequency : if client is started with failover MDS
3320 Description: Allow multiple MDS hostnames in the mount command
3321 Details : Try to read the configuration from all specified MDS
3322 hostnames during a client mount in case the "primary"
3325 Severity : enhancement
3327 Description: Stop sending data to evicted clients as soon as possible.
3328 Details : Check if the client we are about to send or are sending data to
3329 was evicted already. (Check is done every second of waiting,
3330 for which l_wait_event interface was extended to allow checking
3331 of exit condition at specified intervals).
3334 Frequency : rare, normally only when NFS exporting is done from client
3336 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3338 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3341 Frequency : always, if srand() is called before liblustre initialization
3343 Description: Liblustre uses system PRNG disturbing its usage by user application
3344 Details : Introduce internal to lustre fast and high-quality PRNG for
3345 lustre usage and make liblustre and some other places in generic
3346 lustre code to use it.
3348 Severity : enhancement
3349 Bugzilla : 9477, 9557, 9870
3350 Description: Verify that the MDS configuration logs are updated when xml is
3351 Details : Check if the .xml configuration logs are newer than the config
3352 logs stored on the MDS and report an error if this is the case.
3353 Request --write-conf, or allow starting with --old_conf.
3355 Severity : enhancement
3357 Description: Handle symlinks in the path when checking if Lustre is mounted.
3358 Details : Resolve intermediate symlinks when checking if a client has
3359 mounted a filesystem to avoid duplicate client mounts.
3364 Description: lconf can hit an error exception but still return success.
3365 Details : The lconf command catches the Command error exception at the top
3366 level script context and will exit with the associated exit
3367 status, but doesn't ensure that this exit status is non-zero.
3372 Description: failure of ptlrpc thread startup can cause oops
3373 Details : Starting a ptlrpc service thread can fail if there are a large
3374 number of threads or the server memory is very fragmented.
3375 Handle this without oopsing.
3378 Frequency : always, only if liblustre and non-default acceptor port was used
3380 Description: liblustre cannot connect to servers with non-default acceptor port
3381 Details : tcpnal_set_default_params() was not called and was therefore
3382 ignoring the environment varaible TCPNAL_PORT, as well as other
3383 TCPNAL_ environment variables
3388 Description: two objects could be created on the same OST for a single file
3389 Details : If an OST is down, in some cases it was possible to create two
3390 objects on a single OST for a single file. No problems other
3391 than potential performance impact and spurious error messages.
3395 Bugzilla : 5681, 9562
3396 Description: Client may oops in ll_unhash_aliases
3397 Details : Client dcache may become inconsistent in race condition.
3398 In some cases "getcwd" can fail if the current directory is
3404 Description: Inode refcounting problems in NFS export code
3405 Details : link_raw functions used to call d_instantiate without obtaining
3406 extra inode reference first.
3410 Bugzilla : 9942, 9903
3411 Description: Referencing freed requests leading to crash, memleaks with NFS.
3412 Details : We used to require that call to ll_revalidate_it was always
3413 followed by ll_lookup_it. Also with revalidate_special() it is
3414 possible to call ll_revalidate_it() twice for the same dentry
3415 even if first occurence returned success. This fix changes semantic
3416 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3417 reference on a request referred from the intent.
3418 ll_intent_release() then releases such a request.
3421 Frequency : rare, normally benchmark loads only
3423 Description: unlinked inodes were kept in memory on the client
3424 Details : If a client is repeatedly creating and unlinking files it
3425 can accumulate a lot of stale inodes in the inode slab cache.
3426 If there is no other client load running this can cause the
3427 client node to run out of memory. Instead flush old inodes
3428 from client cache that have the same inode number as a new inode.
3431 Frequency : SLES9 2.6.5 kernel and long filenames only
3432 Bugzilla : 9969, 10379
3433 Description: utime reports stale NFS file handle
3434 Details : SLES9 uses out-of-dentry names in some cases, which confused
3435 the lustre dentry revalidation. Change it to always use the
3439 Frequency : rare, unless heavy write-truncate concurrency is continuous
3440 Bugzilla : 4180, 6984, 7171, 9963, 9331
3441 Description: OST becomes very slow and/or deadlocked during object unlink
3442 Details : filter_destroy() was holding onto the parent directory lock
3443 while truncating+unlinking objects. For very large objects this
3444 may block other threads for a long time and slow overall OST
3445 responsiveness. It may also be possible to get a lock ordering
3446 deadlock in this case, or run out of journal credits because of
3447 the combined truncate+unlink. Solution is to do object truncate
3448 first in one transaction without parent lock, and then do the
3449 final unlink in a new transaction with the parent lock. This
3450 reduces the lock hold time dramatically.
3453 Frequency : rare, 2.4 kernels only
3455 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3456 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3457 before unmounting their filesystem in order to avoid deadlock.
3458 The kernel_locked() function in 2.4 kernels only checks whether
3459 the kernel lock is held, not whether it is this process that is
3460 holding it as 2.6 kernels do.
3465 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3466 Details : The client ptlrpc code may be trying to reconnect to a down
3467 server before a previous connection attempt has timed out.
3468 Increase the reconnect interval to be longer than the connection
3469 timeout interval to avoid sending duplicate connections to
3473 Frequency : echo_client brw_test command
3475 Description: fix echo_client to work with OST preallocated code
3476 Details : OST preallocation code (5137) didn't take echo_client IO path
3477 into account: echo_client calls filter methods outside of any
3478 OST thread and, hence, there is no per-thread preallocated
3479 pages and buffers to use. Solution: hijack pga pages for IO. As
3480 a byproduct, this avoids unnecessary data copying.
3484 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3485 Description: Client can oops in mdc_commit_close() after open replay
3486 Details : It was possible for the MDS to return an open request with no
3487 transaction number in mds_finish_transno() if the client was
3488 evicted, but without actually returning an error. Clients
3489 would later try to replay that open and may trip an assertion
3490 Simplify the client close codepath, and always return an error
3491 from the MDS in case the open is not successful.
3494 Frequency : rare, 2.6 OSTs only
3496 Description: OST may deadlock under high load on fragmented files
3497 Details : If there was a heavy load and highly-fragmented OST filesystems
3498 it was possible to have all the OST threads deadlock waiting on
3499 allocation of biovecs, because the biovecs were not released
3500 until the entire RPC IO was completed. Instead, release biovecs
3501 as soon as they are complete to ensure forward IO progress.
3503 Severity : enhancement
3505 Description: Support for specifying external journal device at mount
3506 Details : If an OST or MDS device is formatted with an external journal
3507 device, this device major/minor is stored in the ext3 superblock
3508 and may not be valid for failover. Allow detecting and
3509 specifying the external journal at mount time.
3514 Description: Mounting an MDS with pending unlinked files may cause oops
3515 Details : target_finish_recovery() calls mds_postrecov() which returned
3516 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3517 considers this an error and immediately begins cleaning up the
3518 lov, just after starting the mds_lov process
3520 Severity : enhancement
3522 Description: Implement 'lfs df' to report actual free space on per-OST basis
3523 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3524 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3525 usage of objects; '-h' to report in human readable format.
3527 ------------------------------------------------------------------------------
3529 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3536 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3537 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3538 at startup during orphan unlinking. Instead, skip buddy update
3539 before setup, as it will be regenerated after recovery is complete.
3544 Description: If an OST is inactive, its locks might reference stale inodes.
3545 Details : lov_change_cbdata() must iterate over all namespaces, even if
3546 they are inactive to clear inode references from the lock.
3548 Severity : enhancement
3549 Frequency : occasional, if non-standard max_dirty_mb used
3551 Description: Client will block write RPCs if not enough grant
3552 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3553 then the client will block writes while waiting for another RPC
3554 to complete instead of consuming its dirty limit. With change
3555 we get improved performance when max_dirty_mb is small.
3557 Severity : enhancement
3558 Bugzilla : 3389, 6253
3559 Description: Add support for supplementary groups on the MDS.
3560 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3561 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3562 lookups for user supplementary groups into a cache.
3566 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3567 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3568 for liblustre O_EXCL is enforced by MDS.
3572 Description: readdir never returns NULL in liblustre.
3573 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3574 dentry in struct dirent.
3578 Frequency : liblustre only, depends on application IO pattern
3579 Description: liblustre clients evicted if not contacting servers
3580 Details : Don't put liblustre clients into the ping_evictor list, so
3581 they will not be evicted by the pinger ever.
3583 Severity : enhancement
3585 Description: Add ability to evict clients by NID from MDS.
3586 Details : By echoing "nid:$NID" string into
3587 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3588 $NID would be instantly evicted from this MDS and from all active
3589 OSTs connected to it.
3593 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3594 Details : lookup_it_finish() used to query file size from OSTs that was not
3599 Description: service threads change working directory to that of init
3600 Details : Starting lustre service threads may pin the working directory
3601 of the parent thread, making that filesystem busy. Threads
3602 now change to the working directory of init to avoid this.
3606 Frequency : during shutdown only
3607 Description: shutdown with a failed MDS or OST can cause unmount to hang
3608 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3613 Frequency : 2.6 only
3614 Description: chmod/chown may include an extra supplementary group
3615 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3616 supplementary group and if none is specified this is used.
3619 Bugzilla : 5479 (6816)
3621 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3622 Details : If lookup is in progress on a file that is unlinked we might try
3623 to revalidate the inode and fail in revalidate after lookup is
3624 complete and ll_file_open() enqueues the open again but
3625 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3628 Frequency : always, if lconf --abort_recovery used
3630 Description: lconf --abort_recovery fails with 'Operation not supported'
3631 Details : lconf was attempting to abort recovery on the MDT device and not
3634 ------------------------------------------------------------------------------
3636 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3641 Frequency : rare (only unsupported configurations with a node running as an
3643 Bugzilla : 6514, 5137
3644 Description: Mounting a Lustre file system on a node running as an OST could
3646 Details : OSTs now preallocates memory needed to write out data at
3647 startup, instead of when needed, to avoid having to
3648 allocate memory in possibly low memory situations.
3649 Specifically, if the file system is mounted on on OST,
3650 memory pressure could force it to try to write out data,
3651 which it needed to allocate memory to do. Due to the low
3652 memory, it would be unable to do so and the node would
3653 become unresponsive.
3655 Severity : enhancement
3657 Description: Addition of lconf --service command line option
3658 Details : lconf now accepts a '--service <arg>' option, which is
3659 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3661 Severity : enhancement
3663 Description: Failover mode is now the default for OSTs.
3664 Details : By default, OSTs will now run in failover mode. To return to
3665 the old behaviour, add '--failout' to the lmc line for OSTs.
3667 Severity : enhancement
3669 Description: Health checks are now provided for MDS and OSTs
3670 Details : Additional detailed health check information on MSD and OSTs
3671 is now provided through the procfs health_check value.
3674 Frequency : occasional, depends on IO load
3676 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3677 after numerous create/delete cycles
3678 Details : The ext3 inode allocation policy would not allocate new inodes
3679 very well on the OSTs because there are no new directories
3680 being created. Instead we look for groups with free space if
3681 the parent directories are nearly full.
3686 Description: Network or server problems during mount may cause partially
3687 mounted clients instead of returning an error.
3688 Details : The config llog parsing code may overwrite the error return
3689 code during mount error handling, returning success instead
3695 Description: MDS can fail to allocate large reply buffers
3696 Details : After long uptimes the MDS can fail to allocate large reply
3697 buffers (e.g. zconf client mount config records) due to memory
3698 fragmentation or consumption by the buffer cache. Preallocate
3699 some large reply buffers so that these replies can be sent even
3700 under memory pressure.
3704 Frequency : rare (liblustre)
3705 Description: fsx running with liblustre complained that using truncate() to
3706 extend the file doesn't work. This patch corrects that issue.
3707 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3708 ATTR_SIZE and lsm use in llu_setattr_raw.
3712 Frequency : rare, only 2.6 kernels
3713 Description: Unusual file access patterns on the MDS may result in inode
3714 data being lost in very rare circumstances.
3715 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3716 code in the 2.6 kernel caused the inode and/or EA data not to
3717 be read from disk, causing single-file corruption.
3721 Frequency : rare, only 2.6 filesystems using extents
3722 Description: Heavy concurrent write and delete load may cause data corruption.
3723 Details : It was possible under high-load situations to have an extent
3724 metadata block in the block device cache from a just-unlinked
3725 file overwrite a newly-allocated data block. We now unmap any
3726 metadata buffers that alias just-allocated data blocks.
3730 Frequency : filesystems with default stripe_count larger than 77
3731 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3732 Details : lconf specifies an inode size of 4096 bytes when the default
3733 stripe_count is larger than 77. This conflicts with the default
3734 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3735 this case to avoid pinning too much memory for large EAs.
3737 ------------------------------------------------------------------------------
3739 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3744 Frequency : rare (extremely heavy IO load with hundreds of clients)
3746 Description: Client is evicted, gets IO error writing to file
3747 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3748 caused clients to be evicted for AST timeouts. The fixes in
3749 bug 5192 mean we no longer need to have such short AST timeouts
3750 so ldlm_timeout has been increased.
3753 Frequency : occasional during --force or --failover shutdown under load
3754 Bugzilla : 5949, 4834
3755 Description: Server oops/LBUG if stopped with --force or --failover under load
3756 Details : a collection of import/export refcount and cleanup ordering
3757 issues fixed for safer force cleanup
3760 Frequency : only filesystems larger than 120 OSTs
3761 Bugzilla : 5990, 6223
3762 Description: lfs getstripe would oops on a very large filesystem
3763 Details : lov_getconfig used kfree on vmalloc'd memory
3766 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3767 Bugzilla : 6242, 6243
3768 Description: reading from files that had been truncated to a non-zero size
3769 but never opened returned no data
3770 Details : ll_file_read() reads zeros from no-object files to EOF
3775 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3776 Details : There's little harm in aborting MDS/OSS recovery and letting it
3777 try again, so I removed the LASSERT and return an error instead.
3779 Severity : enhancement
3781 Description: New debugging infrastructure for tracking down data corruption
3782 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3783 (b) cover more of the client-side code path, and (c) try to narrow
3784 down where problems occurred
3788 Bugzilla : 3819, 4364, 4397, 6313
3789 Description: Racing close and eviction MDS could cause assertion in mds_close
3790 Details : It was possible to get multiple mfd references during close and
3791 client eviction, leading to one thread referencing a freed mfd.
3793 Severity: : enhancement
3794 Bugzilla : 3262, 6359
3795 Description: Attempts to reconnect to servers are now more aggressive.
3796 Details : This builds on the enhanced upcall-less recovery that was added
3797 in 1.4.2. When trying to reconnect to servers, clients will
3798 now try each server in the failover group every 10 seconds. By
3799 default, clients would previously try one server every 25 seconds.
3804 Description: After recovery, certain operations trigger a failed
3805 assertion on a client.
3806 Details : Failing over an mds, using lconf -d --failover, while a
3807 client was doing a readdir() call would cause the client to
3808 LBUG after recovery completed and the readdir() was resent.
3810 Severity : enhancement
3812 Description: Default groups are now added by lconf
3813 Details : You can now run lconf --group <servicename> without having to
3814 manually add groups with lmc.
3817 Frequency : occasional
3819 Description: Nodes with an elan id of 0 trigger a failed assertion
3822 Frequency : always when accessing e.g. tty/console device nodes
3824 Description: tty and some other devices nodes cannot be used on lustre
3825 Details : file's private_data field is used by device data and lustre
3826 values in there got lost. New field was added to struct file to
3827 store fs-specific private data.
3830 Frequency : when exporting Lustre via NFS
3832 Description: NFSD failed occasionally when looking up a path component
3833 Details : NFSD is looking up ".." which was broken in ext3 directories
3834 that had grown large enough to become hashed.
3837 Frequency : Clusters with multiple interfaces not on the same subnet
3839 Description: Nodes will repeatedly try to reconnect to an interface which it
3840 cannot reach and report an error to the log.
3841 Details : Extra peer list entries will be created by lconf with some peers
3842 unreachable. lconf now validates the peer before adding it.
3845 Frequency : Only if a default stripe is set on the filesystem root.
3847 Description: Setting a default stripe on the filesystem root prevented the
3848 filesystem from being remounted.
3849 Details : The client was sending extra request flags in the root getattr
3850 request and did not allocate a reply buffer for the dir EA.
3853 Frequency : occasional, higher if lots of files are accessed by one client
3854 Bugzilla : 6159, 6097
3855 Description: Client trips assertion regarding lsm mismatch/magic
3856 Details : While revalidating inodes the VFS looks up inodes with ifind()
3857 and in rare cases can find an inode that is being freed.
3858 The ll_test_inode() code will free the lsm during ifind()
3859 when it finds an existing inode and then the VFS later attaches
3860 this free lsm to a new inode.
3864 Bugzilla : 6422, 7030
3865 Description: MDS deadlock between mkdir and client eviction
3866 Details : Creating a new file via mkdir or mknod (starting a transaction
3867 and getting the ns lock) can deadlock with client eviction
3868 (gets ns lock and trying to finish a synchronous transaction).
3871 Frequency : occasional
3872 Description: While starting a server, the fsfilt_ext3 module could not be
3874 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3875 kernels. Previously, lconf would still use the ext3 name
3876 when trying to load modules. Now, it will correctly use
3877 ext3 on 2.4 and ldiskfs on 2.6.
3879 Severity : enhancement
3880 Description: The default stripe count has been changed to 1
3881 Details : The interpretation of the default stripe count (0, to lfs
3882 or lmc) has been changed to mean striping across a single
3883 OST, rather than all available. For general usage we have
3884 found a stripe count of 1 or 2 works best.
3886 Severity : enhancement
3887 Description: Add support for compiling against Cray portals.
3888 Details : Conditional compiling for some areas that are different
3892 Frequency : occasional
3893 Bugzilla : 6409, 6834
3894 Description: Creating files with an explicit stripe count may lead to
3895 a failed assertion on the MDS
3896 Details : If some OSTs are full or unavailable, creating files may
3897 trigger a failed assertion on the MDS. Now, Lustre will
3898 try to use other servers or return an error to the
3902 Frequency : occasional
3904 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3905 caused lock timeout during readahead (since 1.4.2).
3906 Details : Processes doing readahead might match a lock that hasn't been
3907 granted yet if there are overlapping and conflicting lock
3908 requests. The readahead process waits on ungranted lock
3909 (original lock is CBPENDING), while OST waits for that process
3910 to cancel CBPENDING read lock and eventually evicts client.
3912 Severity : enhancement
3914 Description: Initial enabling of flock support for clients
3915 Details : Implements fcntl advisory locking and file status functions.
3916 This feature is provided as an optional mount flag (default
3917 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3918 locking are implemented yet, and those that are are not guaranteed
3919 to be completely correct in production environments.
3920 mount -t lustre -o [flock|noflock] ...
3923 Frequency : occasional
3925 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3926 assertion in the ext3 JBD (journaling) layer.
3927 Details : The b_committed_data struct is protected by the big kernel lock
3928 in 2.4 kernels, serializing journal_commit_transaction() and
3929 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3930 access to this struct. In 2.6 kernels there is finer grained
3931 locking to improve SMP performance of the JBD layer.
3935 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3940 Description: merge and cleanup kernel patches.
3943 -----------------------------------------------------------------------------
3945 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3947 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3948 versions of Lustre. Please update all servers and clients to
3949 version 1.4.2 or later at the same time. You must also run
3950 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3951 to update the configuration logs.
3953 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3954 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3955 - swab llog records of type '0' so we get proper header size/idx (5861)
3956 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3957 - fix rename of one directory over another leaking an inode (5953)
3958 - avoid SetPageDirty on 2.6 (5981)
3959 - don't re-add just-being-destroyed locks to the waiting list (5653)
3960 - when creating new directories, inherit the parent's custom
3961 striping settings if present parent (3048)
3962 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3963 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3964 - don't hold client locks on temporary worklist from l_lru (5666)
3965 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3966 - automatically evict dead clients (5921)
3967 - Update file size properly in create+truncate+fstat case (6196)
3968 - Do not unhash mountpoint dentries, do not allow removal of
3970 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3971 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3972 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3973 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3974 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3975 - Correctly handle reads of files with no objects (6243)
3976 - lctl recover will also mark a device active if deactivate used (5933)
3978 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3979 - allow --write-conf on an MDS with different nettype than client (5619)
3980 - don't write config llogs to MDS for mounts not from that MDS (5617)
3981 - lconf should create multiple TCP connections from a client (5201)
3982 - init scripts are now turned off by default; run chkconfig --on
3983 lustre and chkconfig --on lustrefs to use them
3984 - upcalls are no longer needed for clients to recover to failover
3986 - add --abort-recovery option to lconf to abort recovery on device
3988 - add support for an arbitrary number of OSTs (3026)
3989 - Quota support protocol changes.
3990 - forward compatibility changes to wire structs (6007)
3991 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3992 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3993 - improved "lustre status" script
3994 - initialize blocksize for non-regular files (6062)
3995 - added --disable-server and --disable-client configure options (5782)
3996 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3997 - Vanilla 2.4.29 support
3998 - increase maximum number of obd devices to 520 (6242)
3999 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4000 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4001 SLES 9 2.6.5 kernels
4002 - limit stripes per file to 160 (the maximum EA size) (6093)
4004 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4007 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4008 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4009 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4010 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4011 - small ext3 extents cleanups and fixes (5733)
4012 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4013 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4014 - avoid needless client->OST connect, fix handle mismatch (5317)
4015 - fix DLM error path that led to out-of-sync client, long delays (5779)
4016 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4017 - fix several locking issues related to i_size (5492,5624,5654,5672)
4018 - don't move pending lock onto export if it is already evicted (5683)
4019 - fix kernel oops when creating .foo in unlinked directory (5548)
4020 - fix deadlock in obdfilter statistics vs. object create (5811)
4021 - use time_{before,after} to avoid timer jiffies wrap (5882)
4022 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4023 - Do not leak request if server was not able to process it (5154)
4024 - If mds_open unable to find parent dir, make that negative lookup(5154)
4025 - don't create new directories with extent-mapping (5909, 5936)
4027 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
4028 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
4029 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
4030 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
4031 - support for RHEL4 kernel on i686 (5773)
4032 - provide error messages when incompatible logs are encountered (5898)
4034 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
4035 * version 1.4.0.10 (1.4.1 release candidate 1)
4037 - don't keep a lock reference when lock is not granted (4238)
4038 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4039 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
4040 - drop import inflight refcount on signal_completed_replay error (5255)
4041 - unlock page after async write error during send (3677)
4042 - handle missing objects in filter_preprw_read properly (5265)
4043 - no transno return for symlink open, don't save no-trasno open (3440)
4044 - don't try to complete elan receive that already failed (4012)
4045 - free RPC server reply state on error (5406)
4046 - clean up thread from ptlrpc_start_thread() on error (5160)
4047 - readahead could read extra page into cache that wasn't ejected (5388)
4048 - prevent races in class_attach/setup/cleanup/detach (5260)
4049 - don't dereference de->d_inode after l_dput of de (5458)
4050 - use "int" for stripe value returned from lock_to_stripe (5544)
4051 - mballoc allocation and error-checking fixes in 2.6 (5504)
4052 - block device patches to fix I/O request sizes in 2.6 (5482)
4053 - look up hostnames for IB nals (5602)
4054 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
4055 - don't start multiple acceptors for the same port (5277)
4056 - fix incorrect LASSERT in mds_getattr_name (5635)
4057 - export a proc file for general "ping" checking (5628)
4058 - fix "lfs check" to not block when the MDS is down (5628)
4060 - service request history (4965)
4061 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4062 - create an "evict_client" /proc entry on OSTs, like the MDS has
4063 - fix mount usage message, return errors per mount(8) (5168)
4064 - change grep [] to grep "[]" in tests so they work in more UMLs
4065 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4066 - remove ancient LOV_MAGIC_V0 EA support (5047)
4067 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4068 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4069 - allow readahead allocations to fail when low on memory (5383)
4070 - mmap locking landed again, after considerable improvement (2828)
4071 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4073 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4076 - send OST transaction number in read/write reply to free req (4966)
4077 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4078 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4079 - the watchdog thread now runs as interruptible (5246)
4080 - flock/lockf fixes (but it's still disabled, pending 5135)
4081 - don't use EXT3 constants in llite code (5094)
4082 - memory shortage at startup could cause assertion (5176)
4084 - reorganization of lov code
4085 - single portals codebase
4087 - add extents/mballoc support (5025)
4088 - direct I/O reads in the obdfilter (4048)
4089 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4091 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4094 - send OST transaction number in read/write reply to free req (4966)
4095 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4096 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4097 - the watchdog thread now runs as interruptible (5246)
4098 - handle missing objects in filter_preprw_read properly (5265)
4099 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4100 - drop import inflight refcount on signal_completed_replay error (5255)
4101 - unlock page after async write error during send (3677)
4102 - return original error code on reconstructed replies (3761)
4103 - no transno return for symlink open, don't save no-trasno open (3440)
4105 - add pid to ldlm debugging output (4922)
4106 - bump the watchdog timeouts -- we can't handle 30sec yet
4107 - extra debugging for orphan dentry/inode bug (5259)
4109 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
4112 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4113 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4114 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
4115 - fix NULL dereference in /proc/sys/portals/routes (4827)
4116 - allow failed mdc_close() operations to be interrupted (4561)
4117 - stop precreate on OST before MDS would time out on it (4778)
4118 - don't send partial-page writes before EOF from client (4410)
4119 - discard client grant for sub-page writes on large-page clients (4520)
4120 - don't free dentries not owned by NFS code, check generation (4806)
4121 - fix lsm leak if mds_create_objects() fails (4801)
4122 - limit debug_daemon file size, always print CERROR messages (4789)
4123 - use transno after validating reply (3892)
4124 - process timed out requests if import state changes (3754)
4125 - update mtime on OST during writes, return in glimpse (4829)
4126 - add mkfsoptions to LDAP (4679)
4127 - use ->max_readahead method instead of zapping global ra (5039)
4128 - don't interrupt __l_wait_event() during strace
4130 - add software watchdogs to catch hung threads quickly (4941)
4131 - make lustrefs init script start after nfs is mounted
4132 - fix CWARN/ERROR duplication (4930)
4133 - return async write errors to application if possible (2248)
4134 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4135 - print NAL number in %x format (4645)
4136 - update barely-supported suse-2.4.21-171 series (4842)
4137 - support for sles 9 %post scripts
4138 - support for building 2.6 kernel-source packages
4139 - support for sles km_* packages
4141 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4144 - ignore -ENOENT errors in osc_destroy (3639)
4145 - notify osc create thread that OSC is being cleaned up (4600)
4146 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4147 - reconstruct ost_handle() like mds_handle() (4657)
4148 - create a new thread to do import eviction to avoid deadlock (3969)
4149 - let lconf resolve symlinked-to devices (4629)
4150 - don't unlink "objects" from directory with default EA (4554)
4151 - hold socknal file ref over connect in case target is down (4394)
4152 - allow more than 32000 subdirectories in a single directory (3244)
4153 - fix blocks count for O_DIRECT writes (3751)
4154 - OST returns ENOSPC from object create when no space left (4539)
4155 - don't send truncate RPC if file size isn't changing (4410)
4156 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4157 - bind to privileged port in socknal and tcpnal (3689)
4159 - rate limit CERROR/CWARN console message to avoid overload (4519)
4160 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4161 - basic mmap support (3918)
4162 - kernel patch series update from b1_4 (4711)
4164 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4167 - avoid crash during MDS cleanup with OST shut down (2775)
4168 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4169 - don't use bad inodes on the MDS (3744)
4170 - dynamic object preallocation to improve recovery speed (4236)
4171 - don't hold spinlock over lock dumping or change debug flags (4401)
4172 - don't zero obd_dev when it is force cleaned (3651)
4173 - print grants to console if they go negative (4431)
4174 - "lctl deactivate" will stop automatic recovery attempts (3406)
4175 - look for existing locks in ldlm_handle_enqueue() (3764)
4176 - don't resolve lock handle twice in recovery avoiding race (4401)
4177 - revalidate should check working dir is a directory (4134)
4179 - don't always mark "slow" obdfilter messages as errors (4418)
4181 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4184 - don't close LustreDB during write_conf until it is done (3860)
4185 - fix typo in lconf for_each_profile (3821)
4186 - allow dumping logs from multiple threads at one time (3820)
4187 - don't allow multiple threads in OSC recovery (3812)
4188 - fix debug_size parameters (3864)
4189 - fix mds_postrecov to initialize import for llog ctxt (3121)
4190 - replace config semaphore with spinlock (3306)
4191 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4192 - don't allow enqueue to complete on a destroyed export (3822)
4193 - down write_lock before checking llog header bitmap (3825)
4194 - recover from lock replay timeout (3764)
4195 - up llog sem before sending rpc (3652)
4196 - reduce ns lock hold times when setting kms (3267)
4197 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4198 - fix NULL deref and obd_dev leak on setup error (3312)
4199 - replace some LBUG about llog ops with error handling (3841)
4200 - don't match INVALID dentries from d_lookup and spin (3784)
4201 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4202 - fix invalid assertion in ptlrpc_set_wait (3880)
4204 - add libwrap support for the TCP acceptor (3996)
4205 - add /proc/sys/portals/routes for non-root route listing (3994)
4206 - allow setting MDS UUID in .xml (2580)
4207 - print the stack of a process that LBUGs (4228)
4209 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4212 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4213 - make sure to unset replay flag from failed open requests (3440)
4214 - if default stripe count is 0, use OST count for inode size (3636)
4215 - update parent mtime/ctime on client for create/unlink (2611)
4216 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4217 - free recovery state on server during a forced cleanup (3571)
4218 - unregister_reply for resent reqs (3063)
4219 - loop back devices mounting and status check on 2.6 (3563)
4220 - fix resource-creation race that can provoke i_size == 0 (3513)
4221 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4222 - more debugging for page-accounting assertion (3746)
4223 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4224 - avoid deadlock after precreation failure (3758)
4225 - fix race and lock order deadlock in orphan handling (3450, 3750)
4226 - add validity checks when grabbing inodes from l_ast_data (3599)
4228 - add /proc/.../recovery_status to obdfilter (3428)
4229 - lightweight CDEBUG infrastructure, debug daemon (3668)
4230 - change default OSC RPC parameters to be better on small clusters
4231 - turn off OST read cache for files smaller than 32MB
4232 - install man pages and include them in rpms (3100)
4233 - add new init script for (un)mounting lustre filesystems (2593)
4234 - run chkconfig in %post for init scripts (3701)
4235 - drop scimac NAL (unmaintained)
4237 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4240 - clean kiobufs before and after use (3485)
4241 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4242 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4243 - change init script to fail more clearly if not run as root (1528)
4244 - allow clients to reconnect during replay (1742)
4245 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4246 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4247 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4248 - fix cond_resched (3554)
4249 - only evict unfinished clients after recovery (3515)
4250 - allow bulk resend, prevent data loss (3570)
4251 - dynamic ptlrpc request buffer allocation (2102)
4252 - don't allow unlinking open directory if it isn't empty (2904)
4253 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4254 - remove extraneous obd dereference causing LASSERT failure (3334)
4255 - don't use get_cycles() when creating temp. files on the mds (3156)
4256 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4257 - handle EEXIST for set-stripe, set proper directory name (3336)
4259 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4260 - fix ksocknal_fmb_callback() error messages (2918)
4262 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4265 - don't copy lvb into (possibly NULL) reply on error (2983)
4266 - don't deref dentry after dput, don't free lvb on error (2922)
4267 - use the kms to determine writeback rpc length (2947)
4268 - increment oti_logcookies when osc is inactive (2948)
4269 - update client's i_blocks count via lvb messages (2543)
4270 - handle intent open/close of special files properly (1557)
4271 - mount MDS with errors=remount-ro, like obdfilter (2009)
4272 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4273 - don't use cancelling-locks' kms values (2947)
4274 - use highest lock extent for kms, not last one (2925)
4275 - don't dereference ERR_PTR() dentry in error handling path (3107)
4276 - fix thread race in portals_debug_dumplog() (3122)
4277 - create lprocfs device entries at setup instead of at attach (1519)
4278 - common AST error handler, don't evict client on completion race (3145)
4279 - zero nameidata in detach_mnt in 2.6 (3118)
4280 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4281 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4282 - initialize RPC timeout timer earlier for 2.6 (3219)
4283 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4284 - print nal/nid for unknown nid (3258)
4285 - additional checks for oscc recovery before doing precreate (3284)
4286 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4287 - don't crash in mdc_close for bad permissions on open (3285)
4288 - zero i_rdev for non-device files (3147)
4289 - clear page->private before handing to FS, better assertion (3119)
4290 - tune the read pipeline (3236)
4291 - fix incorrect decref of invalidated dentry (2350)
4292 - provide read-ahead stats and refine rpc in flight stats (3328)
4293 - don't hold journal transaction open across create RPC (3313)
4294 - update atime on MDS at close time (3265)
4295 - close LDAP connection when recovering to avoid server load (3315)
4296 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4297 - don't leak open file on MDS after open resend (3325)
4298 - serialize filter_precreate and filter_destroy_precreated (3329)
4299 - loop device shouldn't call sync_dev() for nul device (3092)
4300 - clear page cache after eviction (2766)
4301 - resynchronize MDS->OST in background (2824)
4302 - refuse to mount the same filesystem twice on same mountpoint (3394)
4303 - allow llmount to create routes for mounting behind routers (3320)
4304 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4305 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4306 - fix rare problem with rename on htree directories (3417)
4308 - allow default OST striping configuration per directory (1414)
4309 - fix compilation for qswnal for 2.6 kernels (3125)
4310 - increase maximum number of MDS request buffers for large systems
4311 - change liblustreapi to be useful for external progs like lfsck (3098)
4312 - increase local configuration timeout for slow disks (3353)
4313 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4315 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4318 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4319 - don't overwrite extent policy data in reply if lock was blocked (2901)
4320 - drop filter export grants atomically with removal from device (2663)
4321 - del obd_self_export from work_list in class_disconnect_exports (2908)
4322 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4323 - swab reply message in mdc_close, other PPC fixes (2464)
4324 - fix destroying of named logs (2325)
4325 - overwrite old logs when running lconf --write_conf (2264)
4326 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4327 - fix race in target_handle_connect (2898)
4328 - mds_reint_create() should take same inode create lock (2926)
4329 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4330 - don't close files for self_export to avoid uninitialized obd (2936)
4331 - allow MDS with the same name as client node (2939)
4332 - hold dentry reference for closed log files for unlink (2325)
4333 - reserve space for all logs during transactions (2059)
4334 - don't evict page beyond end of stripe extent (2925)
4335 - don't oops on a deleted current working directory (2399)
4336 - handle hard links to targets without a parent properly (2517)
4337 - don't dereference NULL lock when racing during eviction (2867)
4338 - don't grow lock extents when lots of conflicting locks (2919)
4340 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4343 - account for cache space usage on clients to avoid data loss (974)
4344 - lfsck support in lustre kernel code (2349)
4345 - reduce journal credits needed for BRW writes (2370)
4346 - orphan handling to avoid losing space on client/server crashes
4347 - ptlrpcd can be blocked, stopping ALL progress (2477)
4348 - use lock value blocks to assist in proper KMS, faster stat (1021)
4349 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4350 - recovery for initial connections (2355)
4351 - fixes for mds_cleanup_orphans (1934)
4352 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4353 - block all file creations until orphan recovery completes (1901)
4354 - client remove rq_connection from request struct (2423)
4355 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4356 - recovery timer race (2670)
4357 - mdc_close recovey bug (2532)
4358 - ptlrpc cleanup bug (2710)
4359 - mds timeout on local locks (2588)
4360 - namespace lock held during RPCs (2431)
4361 - handle interrupted sync write properly (2503)
4362 - don't try to handle a message that hasn't been replied to (2699)
4363 - client assert failure during cleanup after abort recovery (2701)
4364 - leak mdc device after failed mount (2712)
4365 - ptlrpc_check_set allows timedout requests to complete (2714)
4366 - wait for inflight reqs when ptlrpcd finishes (2710)
4367 - make sure unregistered services are removed from the srv_list
4368 - reset bulk XID's when resending them (caught by 1138 test)
4369 - unregister_bulk after timeout
4370 - fix lconf error (2694)
4371 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4372 - readahead locks pages, leaves pending causing memory pressure (2673)
4373 - increase OST request buffers to 4096 on large machines (2729)
4374 - fix up permission of existing directories in simple_mkdir (2661)
4375 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4376 - don't assign transno to errored transactions (2742)
4377 - don't delete objects on OST if given a bogus objid from MDS (2751)
4378 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4379 - if rq_no_resend, then timeout request after recovery (2432)
4380 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4381 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4382 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4383 - MDS recovery shouldn't delete objects that it has given out (2730)
4384 - if enqueue arrives after completion, don't clobber LVB (2819)
4385 - don't unlock pages twice when trigger_group_io returns error (2814)
4386 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4387 - don't write pages to disk if there was an error (1450)
4388 - don't ping imports that have recovery disabled (2676)
4389 - take buffered bytes into account when balancing socknal conn (2817)
4390 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4391 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4392 - remove little-endian swabbing of llog records (1987)
4393 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4394 - retry reposting request buffers if they fail (1191)
4395 - grow extent at grant time to avoid granting a revoked lock (2809)
4396 - lock revoke doesn't evict page if covered by a second lock (2765)
4397 - disable VM readahead to avoid reading outside lock extents (2805)
4399 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4400 - updated kernel patches for hp-2.4.20 kernel (2681)
4402 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4405 - fix truncated write corruption (2366)
4406 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4408 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4409 - ensure len(uuid) < 37 in lmc (1171)
4410 - fix ia64 OOPS in llog_test (2255)
4411 - zero end of page at obdfilter for partial page writes (2648)
4412 - don't leave stale dentries around after renames (bug 2428)
4413 - fix timeouts when evicting a client with a single lock held (2642)
4414 - set deadline for the initial HELLO message to drain (2634)
4415 - print out dotted-quad IP addresses in the socknal (2302)
4417 - additional debugging for MDS client eviction problem (2443)
4418 - fix mkfsoptions support for osts (2603, 2604)
4420 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4423 - add series for the vanilla 2.6.0 kernel
4424 - add series for the vanilla 2.4.24 kernel
4425 - add series for a cray x86/64 UL kernel drop
4426 - fix xattr patches for the vanilla 2.4.19 series
4428 - generate true UUIDs in lmc (1171)
4429 - have portals stack dumping break in UML (2466)
4430 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4431 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4432 - wake callback waiting threads on client eviction (2460)
4433 - Add --ptldebug and --subsystem to lmc (1719)
4434 - update assertion to allow safe interrupt allocation
4435 - set rq_no_resend for cancel requests (2432)
4436 - recalculate ptlrpcd timeout after resend (2494)
4437 - call vfs_rmdir when removing pending directories (2368)
4438 - fix renaming a file to itself (2429)
4439 - lmc creates a default one-stripe lov (2454)
4440 - expand procfs space to handle large clusters (2326)
4441 - increase UML stack to avoid overflow
4442 - update lconf's list of debug and subsystem masks
4443 - fix lfs find --obd (2510)
4444 - /proc tunable for disabling filter read caching (2591)
4445 - stop rpm packages from altering slapd.conf (2301)
4446 - disable nagle in the socknal under 0conf (2578)
4447 - choose mds inode size based on stripe count (2572)
4448 - fix kernel-source rpm problems (2516)
4450 - add --disable-doc to avoid pdf generation (2421)
4451 - update documentation, tests, type-os, comments
4452 - avoid format warnings on ia64
4453 - remove the TOE NAL
4454 - tiny code cleanups by removing unused fields
4456 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4459 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4460 - remove the most blatant lies from BUILDING file (2371)
4461 - change default debug level to reasonable production setting
4462 - reduce client side cache size to reduce cache flush time
4463 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4464 - make TCP zerocopy and pinger support enabled by default (2476)
4465 - sync writes completed after process exits caused crashes (2319)
4466 - maintain correct mount count on the MDS (2356)
4467 - backout 1557, because 2316 wasn't really fixed
4468 - better file I/O statistics gathering in /proc
4469 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4470 - another kernel patch to fix zero-copy TCP function export
4471 - don't take duplicate lock when processing re-sent getattr (2420)
4472 - lctl uses obd_self_export instead of creating new conn (2353)
4473 - MDS/OST recovery case which requires object creation asserted (2425)
4474 - move lfs from /usr/sbin to /usr/bin in packages
4475 - fix race between mds_client_add and mds_client_free (2417)
4476 - use kmalloc instead of slabs in portals (2430)
4477 - don't create duplicate records when a failover MDS is present (2442)
4478 - remove unnecessary mount age check (2332)
4479 - don't remove directory inodes from locks prematurely (2451)
4480 - don't break if MDS service name is the same as hostname (2103)
4481 - fix races in client write RPC generation when cache full (2482)
4483 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4486 - remove now-unused request->rq_obd (278)
4487 - if an allocation fails, print out how much memory we've used (1933)
4488 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4489 - add the "configurable stack size" patch to most series files (1256)
4490 - ability to write large log records, for 100+ OST configs (2306)
4491 - fix NULL deref when filter_prep fails (2314)
4492 - fix operator precedence error in filter_sync
4493 - dynamic allocation of socknal TX descriptors (2315)
4494 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4495 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4496 - ensure that CWARN messages in Portals always get to the syslog
4497 - __init/__exit are not for prototype decls (ldlm_init/exit)
4498 - x86-64 compile warning fixes
4499 - fix gateway LMC keyword conflict (2318)
4500 - fix MDS lock inversions in getattr/reint paths (1844)
4501 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4502 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4503 - DLM race condition prevented some lock evictions (2328)
4504 - ENOMEM detection and retry on socknal sends (2230)
4505 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4506 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4507 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4508 - better error messages when a client is rejected during recovery (1505)
4509 - avoid cancelling locks which were never granted, after failure (2330)
4510 - fix i_sem/journal inversion in mds_client_add (2333)
4511 - fix truncate/getattr lock cycle deadlock (2334)
4512 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4513 - fix two rare exit paths which could leak an l_lock() ref (2321)
4514 - fencepost error in MDS/OST orphan recovery (2226)
4515 - make log record alignment 8 bytes (1988)
4516 - lstripe now fails when requested offset > ost_count (2237)
4517 - ensure that all kernel series have a complete list.h (1607)
4518 - fix crashes in special-file operations (2316)
4519 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4521 - allow configurable automake binary, for testing new versions
4522 - small update to the lfs documentation
4524 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4526 * fix negative export reference count in fsfilt_sync (2312)
4528 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4529 * release candidate 0.9.1
4531 - orphans are moved into the PENDING directory for possible recovery
4532 - replayed opens now open by fid for orphan/rename safety (1042)
4533 - last close of an orphan inode generates a transno (683)
4534 - chdir() and mount() now pin the directory entry (1020)
4535 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4536 - discard very old requests without processing them (1502)
4537 - remove some common, well-understood CERRORs (1505)
4538 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4539 - clear "grant" flags in OST replies until OST grant code lands (1644)
4540 - fix read performance by not clobbering i_blksize on client (1598)
4541 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4542 - make LRU size a /proc tunable, clears locks when reduced (707)
4543 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4544 - server locks take references on exports now (1558)
4545 - build fixes for 2.4.20-rh trees (1663)
4546 - return an error from lov_create if all OSCs are inactive (1751)
4547 - fix import levels when a reconnect happens without a timeout (1597)
4548 - exit early from mds_open if we get a lookup error (1749)
4549 - partial page read at EOF wouldn't wait for disk before sending (1642)
4550 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4551 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4552 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4553 - write/truncate lock inversion (1639)
4554 - Don't auto-load obdclass, portals modules during cleanup (1495)
4555 - fix timestamps from jumping to "now" (1763)
4556 - extra journal assertions (1648)
4557 - add an extra multiunlink test (1771)
4558 - fix read_record/write_record API (1776)
4559 - fix leak of offset_extent, possible incorrect i_size later (1772)
4560 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4561 - Debugging for the kqswnal_get_idle_tx problems (1820)
4562 - Allow recovery to be attempted multiple times (1536)
4563 - Write out MDS last_rcvd file after it is first created (1600)
4564 - Fix tx_descriptor leak in failed transmit situations (1827)
4565 - ext3 journaling fixes for assertion failure after IO error (1871)
4566 - class_export_put() on freed export after completion AST error (1896)
4567 - Fix revalidate looping in VFS (1322)
4568 - Don't access a freed export during MDS_REINT timeout (1521)
4569 - Add open-unlink recovery support on the MDS (1673,1764)
4570 - Return an error if no MDS data was read from last_rcvd (1946)
4571 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4572 - Don't setup a disk device that doesn't match exported UUID (317)
4573 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4574 - avoid committing NULL handle in force close
4575 - local.sh is now a one-stripe LOV configuration
4576 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4577 - fix ext3 htree duplicate directory entry corruption (1516)
4578 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4579 - update to new LOV EA format (2097)
4580 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4582 2003-06-15 Phil Schwan <phil@clusterfs.com>
4585 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4586 - per-import/export recovery handling (958, 931, 959)
4587 - multiple last-rcvd slots, for serving multiple FSes (949)
4588 - connections are again shared between multiple imp/exports (963, 964)
4589 - "umount -f" would hang if any requests needed to be sent (393, 978)
4590 - avoid pinning large req buffer by copying for queued messages (989)
4591 - add "uuid" to "lctl device" command to help upcalls (991)
4592 - "open" RPCs with transnos would confuse recovery counters (1037)
4593 - do proper endian conversion of all wire messages (288, 340, 891)
4594 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4595 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4596 - fix ptlrpc_connection leak in target_handle_connect (1174)
4597 - fix import refcounting bug in OST and MDS cleanup (1134)
4598 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4599 - fix very unlikely obd_types race condition (501)
4600 - remove osc_open hack for echo_client (1187)
4601 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4602 - a failure in read_inode2 leads to deadlock (1139)
4603 - cancel ack-locks as soon as transaction is committed (1072)
4604 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4605 - make sure to commitrw after any preprw to avoid deadlock (1162)
4606 - failing to execute a file in a lustre FS would lock inode (1203)
4607 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4608 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4609 - obd_uuid2tgt would walk off the end of the list (1255)
4610 - on IA64 the osc would give portals incorrect bulk size (1258)
4611 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4612 - fix lock inversion caused by new llite matching code (1282)
4613 - limit the number of dirty pages on a client to 10MB (1286)
4614 - timed out locks were not being corrected cancelled (1289)
4615 - fix O_DIRECT above 4GB on IA-32 (1292)
4616 * major user-visible changes
4617 - fail out/fail over policy now controlled by the upcall (993)
4619 - add OBD_PING to check server availability and failure (954)
4620 - lustre messages are now sent in sending host order (288, 340, 891)
4621 - add eadatalen to MDS getattr reply (340)
4622 - OST read replies may contain second buffer, with per-page status (593)
4624 2003-03-11 Phil Schwan <phil@clusterfs.com>
4627 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4628 - failed open()s could cause deadlock; fixed (867, 869)
4629 - stop cancelling OST locks when files are closed (481)
4630 - overlapping XID spaces caused network corruption (851, 853)
4631 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4632 - setattr_raw added, to do single-RPC, server-side setattrs
4633 - lmc/lconf syntax change for OST UUIDs
4634 - fix crashy race condition between ptlrpc_free_req and osc_close
4635 - don't use request in mdc_enqueue if we hit a timeout (889)
4636 - don't set the inode i_size for regular files from the MDS (896)
4637 - handle out of order completion AST (842)
4638 - don't LBUG if a lock request times out after receiving AST (913)
4639 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4640 - if a bad lock AST arrives, send an error instead of dropping entirely
4641 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4642 - fix leak in bulk IO when only partially completed (899, 900, 926)
4643 - fix O_DIRECT for ia64 (55)
4644 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4645 - C-z after timeout could hang a process forever; fixed (977)
4647 - client-side I/O cache (678, 924, 929, 941, 970)
4649 - READPAGE and SETATTRs which don't take server-side locks get
4652 2003-02-11 Phil Schwan <phil@clusterfs.com>
4655 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4656 - Fix fsfilt_extN_readpage() to read a full page of directory
4657 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4658 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4659 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4660 - Most intent operations were removed, in favour of a new RPC mode
4661 that does a single RPC to the server and bypasses most of the VFS
4662 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4663 - Aggressively cancel local locks on DLM servers
4664 - mds_reint_unlink sends EA to the client if it's the last nlink.
4665 client uses that EA to unlink OST objects.
4666 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4667 - recursive symlinks were fixed (439)
4668 - fixed NULL deref in DEBUG_REQ
4669 - filter_update_lastobjid no longer calls sync, which annoyed extN
4670 - fixed multi-client small-writes to a single file problem (445)
4671 - fixed mtime updates during file writes (607)
4672 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4673 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4674 - fixed Linux OST size reporting problem (444, 656)
4675 - OST now updates object mtime with writes or setattr (607, 619)
4676 - client verifies file size before zeroing page past EOF (445)
4677 - OST now writes last allocated objid to disk with allocation (108)
4678 - LOV on echo now works (409)
4680 - mds_reint_unlink sends a new buffer, with the EA included. this
4681 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4682 set if a regular file was being unlinked, and it was the last link
4683 - use PtlGet from the target for bulk writes (315)
4684 - OST now updates object mtime with writes or setattr (607, 619)
4685 - LDLM now has a grant-time callback to revalidate locked items, if
4687 - Many MDS operations were reorganized to combat race conditions
4689 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4690 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4692 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4695 - Fully reactivate OST imports after reconnection (512, others)
4696 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4697 - More graceful error handling for truncating on dead OST (515)
4698 - Don't error out unless we're actually accessing dead stripes (474)
4699 - Fix garbage sizes when stripes are missing (410)
4700 - LRU counters were broken, causing constant lock purge (433, 432)
4701 - garbage on read from stripes with failed OSTs (441)
4702 - mark OSCs as active before reconnecting during recovery (438)
4703 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4704 - lfind did not preserve OST order in output (443)
4705 - symlinks cause hung clients, incorrect data (439)
4706 - stop dereferencing request after dropping refcount (457)
4707 - don't LASSERT(spin_is_locked) on non-SMP (455)
4708 - fixes for many rename() bugs
4709 - fstat didn't correctly synchronize attributes (399)
4710 - server must handle lock cancellation during blocking AST prep (487)
4711 - bulk descriptors were free()d too soon (511)
4712 - fix paths in lconf, which would load incorrect modules (451, 507)
4713 - fix confusing lconf 'host not found' error message (386)
4714 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4715 - fix race condition in mdc_blocking_ast() for inode access (526)
4716 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4717 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4718 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4719 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4720 - fix lstripe in several ways (kernel side) (527)
4721 - fix request leak in ldlm_cli_enqueue (262)
4722 - incorrect OSC was marked inactive after OST failure
4723 - call mds_fs_cleanup before unmounting filesystem (524)
4724 - fix races between taking ns_lock and ldlm_lock_change_resource
4725 - fix races updating LOV export open file list
4726 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4727 - fix recovery NULL deref in ldlm_cli_cancel_unused
4728 - fix some DLM races by using new hash table for lock handles (419)
4729 - permit the client to specify desired inodes, at replay
4730 - duplicate requests when we queue them for replay reintegration
4731 - fix last_rcvd offset calculation
4732 - sync after each recovered transaction, so we always make progress
4733 - never, not always, ERESTART requests without transnos
4734 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4736 - skip replay if the MDS says that the client is already connected
4737 - don't check for a recovery-enabled export to match lctl's UUID
4738 - don't INC_USE_COUNT for phantom exports
4739 - don't crash when cleaning up phantom exports (567)
4740 - don't double-finish or set replay data for errored mdc_open requests
4741 - abort requests when they time out, so we don't get old replies
4742 - send/receive replies for AST messages again
4743 - if the client says that it doesn't have the lock, cancel it on the
4745 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4746 mark it as destroyed, it will all work out when decref is called
4747 - fix module use counts (22, 581)
4749 - ASTs now expect a reply (server cancels lock on error reply)
4751 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4754 - fix many simultaneous client startup (392)
4755 - fix dentry->d_it clobbering
4756 - credentials weren't being shipped for readdir/getattr operations
4757 - remove invalid assertions triggered during some concurrent MD
4759 - proper Lustre versions added (336, 389)
4760 - fix memory leak for create error case (398)
4761 - fix LOV locking bug that would get cli/srv out of sync
4762 - fix echo client over LOV (409)
4763 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4764 - fix double-O_EXCL intent crash (424)
4765 - avoid sending multiple lock CANCELs (352)
4767 - MDS can do multi-client recovery (modulo bugs in new code)
4769 - many updates, edits, cleanups
4771 2002-11-18 Phil Schwan <phil@clusterfs.com>
4774 - fix null d_it dereference (346)
4775 - fix full OST/dbench hang (333)
4776 - fix permission problem with file removal (286)
4777 - fix removal of OSCs from LOV when they fail
4778 - fix NULL deref during bulk timeout (214)
4779 - fix problems related to multiple filesystems on one MDS (241)
4780 - fixed serious subtle metadata locking bugs
4781 - free locks on clients when inodes are removed due to memory
4783 - fix inode pointer in lock data (285)
4784 - partial support for multiple MDS on a single host (241)
4785 - data locks weren't cancelled at clear_inode time (290, 311)
4786 - intent locks could lead to unbounded lock growth (205)
4787 - added a maximum lock count, an LRU list, and a flusher
4788 - fix multiple rename (365)
4789 - properly abstracted the echo client
4790 - OSC locked 1 byte too many; fixed
4791 - rewrote brw callback code:
4792 - fixed recovery bugs related to LOVs (306)
4793 - fixed too-many-pages-in-one-write crash (191)
4794 - fixed (again) crash in sync_io_timeout (214)
4795 - probably fixed callback-related race (385)
4797 - Add capability to MDS protocol
4798 - LDLM cancellations and callbacks on different portals
4800 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4803 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4804 - defer open object destruction to close time (601981, 138)
4805 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4806 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4807 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4808 - don't call obd_setattr in ll_file_release for destroyed objects
4809 * protocol change to lustre_msg: move |version| and add |flags|
4810 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4811 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4812 * added replay of create, unlink, link and rename operations during
4813 MDS failover; recovery should be much more robust now
4814 * remove failed OSCs from LOVs (only lov_create uses this so far)
4815 * the lustre-HOWTO was brought (more) up to date (582544)
4817 2002-10-23 Phil Schwan <phil@clusterfs.com>
4820 - in-use dentries weren't being reused properly (617851)
4821 - prevent multiple LDLM setup (599178)
4822 - fix LOV size calculations for truncate (617853)
4823 - fix client handling of MDS intent errors (POSIX)
4824 - fix permission bug in lovstripe.c test (624321)
4825 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4826 - truncate past end of file could corrupt data
4827 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4828 - a race in recovery could return ETIMEDOUT to apps (623947)
4829 - building outside the source directory was fixed
4830 * the lustre-HOWTO was brought (more) up to date (582544)
4831 * major progress was made on recovery functionality
4833 2002-10-10 Phil Schwan <phil@clusterfs.com>
4836 - recovery deadlock fix
4837 - rm -rf causes LBUG fix (617817)
4838 - file open by multiple tasks fix (618962)
4839 - directory permissions bugs (602707 and 620007)
4840 - journal_stop fixed with locking (611313)
4841 - O_APPEND failures resolved (618273, perhaps 614459)
4842 - lconf PATH fix (619770)
4843 - IA64 build fix (621450)
4844 - RPC buffer sizes scale with amount of memory
4846 2002-10-01 Phil Schwan <phil@clusterfs.com>
4849 - locks would be cancelled without throwing away data pages,
4850 resulting in inconsistent data (605627)
4851 - inode attributes were not always being refreshed (605627, 612449)
4852 - lconf now continues to cleanup after lctl reports an error
4853 - MDS now enforces user permissions (602707)
4854 - lprocfs cleanup fixed, but not yet enabled (614157)
4855 - fixed infinite server hang, should a client not respond to an AST
4856 - avoid going into recovery if user calls readlink() with a buffer
4857 that's too small (613941)
4858 - AST RPCs no longer require replies (614867) -- this may be changed
4859 - don't crash server if client sends an IOV that's too big (611336)
4860 - fixed lock conversion deadlock (611892)
4861 - fixed the following of symlinks (614622)
4862 * recovery: the server can remove locks from a client that dies, other
4863 clients can make progress
4864 * more extN patch fixes
4865 * compile-time configurable ptlrpc buffer allocations
4867 - collaborative read cache document
4868 - Lustre Lite Performance CDR document-in-progress
4870 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4873 - fix typo in patch-2.4.18
4875 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4878 - clear ptlrpc request each time in handle_incoming_request()
4879 - unlink of files now destroys the object on the OST
4881 2002-09-19 Peter Braam <braam@clusterfs.com>
4883 * add hard link support
4884 * change obdfile creation method
4885 * kernel patch changed
4887 2002-09-19 Peter Braam <braam@clusterfs.com>
4890 - stack overflow bug in extN fixed
4892 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4894 * documentation updates
4895 - add man pages for config tools
4896 - update tests/README to describe testing with new config tools
4897 - finish metadata API descriptions
4898 * bug fixes and cleanups
4899 - statfs workaround for 16TB limit
4900 - LOV stripe allocation improved, can stripe on subset of OSTs
4901 - LOV file size/IO offset was wrong for files > 4GB in size
4902 - object EA data was being dropped, caused files to be unreadable
4903 - memory overflow with non-LOV OST caused memory corruption
4904 - fixed regression tests to work with new config tools, obdfilter
4905 - fixed bug when directory size became larger than 1 block
4906 - fixed bug (for single client case) when PWD was deleted
4907 - invalidate local directory pages when doing intent-based ops
4908 - avoid LDLM oops when lock callback contained bad data
4910 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4912 * documentation updates
4913 * bug fixes and cleanups
4914 - configuration tools
4917 - 64-bit compile warnings
4918 - 64-bit internal statfs data
4920 * test_brw on persistent OST devices
4922 * lprocfs (disabled)
4924 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4926 * documentation updates
4927 * bug fixes and cleanups
4928 * configuration tools
4930 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4933 * many small fixes to 0_5_4
4934 * io/network handling
4935 * thinkos in MDS operations
4937 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4940 * crucial basic fixes to 0.5.3
4941 * IOR, Iozone work over Elan
4944 2002-08-07 Phil Schwan <phil@clusterfs.com>
4945 * version 0_5_3, our first alpha
4946 * we use the new Portals iovs
4947 * documentation updates
4948 * bug fixes and cleanups
4949 * small changes in the DLM wire protocol
4951 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4952 * version 0_5_1 with some initial stability,
4953 * locking on MD and file I/O.
4954 * documentation updates
4955 * several bug fixes since 0.5.0
4956 * small changes in wire protocol
4958 2002-07-18 Phil Schwan <phil@clusterfs.com>
4960 * delivered as Lustre Light Alpha
4961 * fixed a crash after handling invalid MDS requests
4962 * fixed directory pages for architectures with non-4k pages sizes
4964 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4965 * release version v0_4_4
4966 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4968 * Fixes a number of interruption problems with OST operations.
4969 * Update documentation for portals header changes
4970 * Move all wire protocol structs/defines to lustre_idl.h
4971 * Fixes symlink length bug.
4972 * Add tcpdump to repository.
4974 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4975 * release version v0_4_3
4976 * Fixes statfs for inodes on extN.
4977 * Fixes bug in runtests which would delete /etc/hosts.
4978 * Use 64-bit object IDs wherever possible (not into VFS though)
4979 Remove ost_get_info, which is unused by lustre, and out of date.
4981 2002-07-03 Peter Braam <braam@clusterfs.com>
4982 * release version v0_4_2 Fixes a lookup error (type not passed)
4983 * move forward to head of Portals
4984 * move forward to latest Lustre kernel
4986 2002-06-25 Peter Braam <braam@clusterfs.com>
4987 * release version v0_4_1. Hopefully stable on single node use.