1 tbd Sun Microsystems, Inc.
4 2.6.16.60-0.27 (SLES 10),
5 2.6.18-92.1.10.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).
17 Description: LBUG when llog conf file is full
18 Details : When llog bitmap is full, ENOSPC should be returned for plain
23 Description: Prevent import from entering FULL state when server in recovery
27 Description: service mount cannot take device name with ":"
28 Details : Only when device name contains ":/" will mount treat it as
34 Description: replace ptlrpcd with the statahead thread to interpret the async
35 statahead RPC callback
39 Frequency : on recovery
40 Description: I/O failures after umount during fail back
41 Details : if client reconnected to restarted server we need join to recovery
42 instead of find server handler is changed and process self eviction
43 with cancel all locks.
45 Severity : enhancement
47 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
49 Severity : enhancement
51 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
53 Severity : enhancement
55 Description: Upcall on Lustre log has been dumped
56 Details : Allow for a user mode script to be called once a Lustre log has
57 been dumped. It passes the filename of the dumped log to the
58 script, the location of the script can be specified via
59 /proc/sys/lnet/debug_log_upcall.
64 Description: avoid idr_remove called for id which is not allocated.
65 Details : Move assigment s_dev for clustered nfs to end of initialization,
66 to avoid problem with error handling.
71 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
72 Details : When connection is reused this not moved from CONN_UNUSED_HASH
73 into CONN_USED_HASH and this prodice warning when put connection
76 Severity : enhancement
78 Description: Export bytes_read/bytes_write count on OSC/OST.
82 Description: Early reply size mismatch, MGC loses connection
83 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
84 the connect flags are properly negotiated.
88 Description: Properly propagate oinfo flags from lov to osc for statfs
89 Details : restore missing copy oi_flags to lov requests.
91 Severity : enhancement
93 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
94 : l_getgroups(8), lst(8), routerstat(8)
95 Details : included man pages for llobdstat(8), llstat(8),
96 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
98 Severity : enhancement
100 Description: configure's --enable-quota should check the
101 : kernel .config for CONFIG_QUOTA
102 Details : configure is terminated if --enable-quota is passed but
103 : no quota support is in kernel
107 Description: Remove portals compatibility
108 Details : Remove portals compatibility, not interoperable with releases
113 Description: Resolve device initialization race
114 Details : Prevent proc handler from accessing devices added to the
115 obd_devs array but yet be intialized.
117 Severity : enhancement
119 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
121 Severity : enhancement
123 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
128 Description: Using wrong pointer in osc_brw_prep_request
129 Details : Access to array[-1] can produce panic if kernel compiled with
130 CONFIG_PAGE_ALLOC enabled
134 Description: Client runs out of low memory
135 Details : Consider only lowmem when counting initial number of llap pages
139 Description: *optional* service tags registration
140 Details : if the "service tags" package is installed on a Lustre node
141 When the filesystem is mounted, a local-node service tag will
142 be created. See http://inventory.sun.com/ for more information
143 about the Service Tags asset management system.
147 Description: Kernel BUG tries to release flock
148 Details : Lustre does not destroy flock lock before last reference goes
149 away. So always drop flock locks when client is evicted and
150 perform unlock regardless of successfulness of speaking to MDS.
154 Description: add recount protection for osc callbacks, so avoid panic on shutdown
158 Description: sanity test 65a fails if stripecount of -1 is set
159 Details : handle -1 striping on filesystem in ll_dirstripe_verify
164 Description: ASSERTION(CheckWriteback(page,cmd)) failed
165 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
168 Severity : enhancement
170 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
174 Description: do not process already freed flock
175 Details : flock can possibly be freed by another thread before it reaches
176 to ldlm_flock_completion_ast.
180 Description: LBUG during stress test
181 Details : Need properly lock accesses the flock deadlock detection list.
185 Description: oops in page fault handler
186 Details : kernel page fault handler can return two special 'pages' in error case, don't
187 try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
191 Description: timeout with invalidate import.
192 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which should be
193 handled by ptlrpcd. This produce long age waiting and -ETIMEOUT
194 ptlrpc_invalidate_import and as result LASSERT.
196 Severity : enhancement
198 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
202 Description: enable MGS and MDT services start separately
203 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
204 a co-located MGS without starting the MGS, which is a complement
205 to 'nosvc' mount option.
209 Frequency : after recovery
210 Description: precreate to many object's after del orphan.
211 Details : del orphan st in oscc last_id == next_id and this triger growing
212 count of precreated objects. Set flag LOW to skip increase count
213 of precreated objects.
217 Frequency : rare, on clear nid stats
218 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
219 Details : when clean nid stats sometimes try destroy live entry,
220 and this produce panic in free.
224 Description: Stack overflow during MDS log replay
225 ease stack pressure by using a thread dealing llog_process.
229 Description: wait until IO finished before start new when do lock cancel.
230 Details : VM protocol want old IO finished before start new, in this case
231 need wait until PG_writeback is cleared until check dirty flag and
232 call writepages in lock cancel callback.
234 Severity : enhancement
236 Description: using special macro for print time and cleanup in includes.
240 Description: mds_mfd_close() ASSERTION(rc == 0)
241 Details : In mds_mfd_close(), we need protect inode's writecount change
242 within its orphan write semaphore to prevent possible races.
246 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
250 Frequency : rare, on shutdown ost
251 Description: don't hit live lock with umount ost.
252 Details : shrink_dcache_parent can be in long loop with destroy dentries,
253 use shrink_dcache_sb instead.
257 Description: don't panic with use echo client
258 Details : echo client pass NULL as client nid pointer and this produce null
263 Description: fix build on ppc32
264 Details : compile code with -m64 flag produce wrong object file for ppc32.
268 Description: add message levels for liblustreapi
272 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
273 Details : Move the 'good_osts' check before the 'total_bavail' check. This
274 will result in an -EAGAIN and in the exit call path we call
275 alloc_rr() which will with increasing aggressiveness attempt to
276 aquire precreated objects on the minimum number of required OSCs.
280 Description: Use old size assignment to avoid deadlock
281 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
282 the scheduling while holding a spinlock. We do not need locking
283 for size in ll_update_inode() because size is only updated from
284 the MDS for directories or files without objects, so there is no
285 other place to do the update, and concurrent access to such inodes
286 are protected by the inode lock.
290 Description: resolve "_IOWR redefined" build error on SLES10
294 Description: dump the memory debugging after all modules are unloaded to
295 suppress false negative in conf_sanity test 39
297 Severity : enhancement
299 Description: build kernel-ib packages for OFED 1.3 in our release cycle
304 Description: fix SLES kernel versioning
305 Details : the kernel version for our SLES 10 kernel did not include a "-"
306 before the "smp" at the end. while this was not a problem in
307 general, it did mean that software trying to use the kernel
308 version to try to detect a vendor specific kernel would fail.
309 this was most evident by the OFED build scripts.
313 Description: Don't update lov_desc members until making sure they are valid
314 Details : When updating lov_desc members via proc fs, need fix their
315 validities before doing the real update.
319 Description: don't put request into delay list while invalidate in flight.
320 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
321 import in flight. this produce timeout for invalidate and sometimes
322 can cause stale data.
326 Frequency : on ppc only
327 Description: not convert ost objects for directory because it's not exist.
328 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
332 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
334 Severity : enhancement
336 Description: Update to support 2.6.22.14 vanilla kernel.
340 Frequency : rare, on recovery
341 Description: read procfs can produce deadlock in some situation
342 Details : Holding lprocfs lock which send rpc can produce block for destroy
343 obd objects and this also block reconnect with -EALREADY. This isn't
344 fix all lprocfs bugs - but make it rare.
346 Severity : enhancement
348 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
351 Frequency : frequent on X2 node
353 Description: mdc_set_open_replay_data LBUG
354 Details : Set replay data for requests that are eligible for replay.
358 Description: lustre_mgs: operation 101 on unconnected MGS
359 Details : When MGC is disconnected from MGS long enough, MGS will evict the
360 MGC, and late on MGC cannot successfully connect to MGS and a lot
361 of the error messages complaining that MGS is not connected.
365 Frequency : on network error
366 Description: panic with double free request if network error
367 Details : mdc_finish_enqueue is finish request if any network error ocuring,
368 but it's true only for synchronus enqueue, for async enqueue
369 (via ptlrpcd) this incorrect and ptlrpcd want finish request
372 Severity : enhancement
374 Description: client-side metadata stat-ahead during readdir(directory readahead)
375 Details : perform client-side metadata stat-ahead when the client detects
376 readdir and sequential stat of dir entries therein
379 Frequency : on start mds
381 Description: Implement get_info(last_id) in obdfilter.
384 Frequency : occasional
386 Description: Correctly check stale fid, not start epoch if ost not support SOM
387 Details : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
388 overwrited with new generated fid, but mds can anwer with one of these
389 two fids and both is not stale. setattr incorectly start epoch and
390 assume will be called done_writeting, but without SOM done_writing
394 Frequency : rare, depends on device drivers and load
396 Description: MDS or OSS nodes crash due to stack overflow
397 Details : Code changes in 1.8.0 increased the stack usage of some functions.
398 In some cases, in conjunction with device drivers that use a lot
399 of stack the MDS (or possibly OSS) service threads could overflow
400 the stack. One change which was identified to consume additional
401 stack has been reworked to avoid the extra stack usage.
404 Frequency : occasional
406 Description: Do not fail import if osc_interpret_create gets -EAGAIN
407 Details : If osc_interpret_create got -EAGAIN it immediately exits and
408 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
409 oscc_has_objects and see OSC has no objests and call
410 oscc_internal_create to resend create request.
412 Severity : enhancement
414 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
416 Severity : enhancement
418 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
421 Frequency : very rare
423 Description: Fix replay if there is an un-replied request and open
424 Details : In some cases, older replay request will revert the
425 mcd->mcd_last_xid on MDS which is used to record the client's
428 Severity : enhancement
430 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
432 Severity : enhancement
434 Description: Add rhel5 support to HEAD.
436 Serverity : enhancement
438 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
443 Description: Don't allow skipping OSTs if index has been specified.
444 Details : Don't allow skipping OSTs if index has been specified, make locking
445 in internal create lots better.
449 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
450 Details : access to bitfield in structure is always rounded to long
451 and this produce problem with not atomic change any bit.
455 Description: Lustre make rpms failed.
456 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
457 builds Lustre rpms from tarball.
460 Frequency : rare on shutdown ost
462 Description: If llog cancel was not send before clean_exports phase, this can
463 produce deadlock in llog code.
464 Details : If llog thread has last reference to obd and call class_import_put
465 this produce deadlock because llog_cleanup_commit_master wait when
466 last llog_commit_thread exited, but this never success because was
467 called from llog_commit_thread.
471 Description: allow userland application know is lost one of stripes.
472 Details : fill lvb_blocks with error code on ost and return it to
473 application if error flag found.
477 Description: NULL lov_tgts causing MDS oops
478 Details : more safe checks for NULL lov_tgts for avoid oops.
480 Severity : enhancement
482 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
486 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
487 Details : If an OST is down the MDS will hang indefinitely in
488 obd_statfs_rqset() waiting for the statfs data. While for
489 MDS QOS usage of statfs, it should not stuck in waiting.
491 Severity : enhancement
493 Description: remote_acl support
494 Details : Support ACL-based permission check for remote user.
495 Support setfacl/getfacl for remote user with the utils
496 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
497 the system "{s,g}etfacl" utils.
499 Severity : enhancement
501 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
503 Severity : enhancement
505 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
509 Description: make mgs_setparam() handle fsname containing dash
510 Details : fsname containing a dash does not work with lctl conf_param
512 Severity : enhancement
514 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
516 Severity : enhancement
518 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
522 Frequency : rare, at shutdown
523 Description: access already free / zero obd_namespace.
524 Details : if client_disconnect_export was called without force flag set,
525 and exist connect request in flight, this can produce access to
526 NULL pointer (or already free pointer) when connect_interpret
527 store ocd flags in obd_namespace.
531 Frequency : only at startup
532 Description: not alloc memory with spinlock held.
533 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
534 if any spinlock held.
536 Severity : enhancement
538 Description: make lustre randomly fail allocating memory
539 Details : Make lustre randomly failed allocating memory for testing purpose.
541 Severity : enhancement
543 Description: lost problems with lov objid file
544 Details : Fixes some scability and access to not inited memory problems
545 in work with lov objdid file.
550 Description: lfs find does not continue on file error
551 Details : Continue other files processing when a file/dir is absent.
555 Description: Inconsistent usage of lustre_pack_reply()
556 Details : Standardize the usage of lustre_pack_reply() such that it
557 always generate a CERROR on failure.
560 Frequency : occasional
562 Description: MDS hang or stay in waiting lock
563 Details : If client receive lock with CBPENDING flag ldlm need send lock
564 cancel as separate rpc, to avoid situation when cancel request
565 can't processed due all i/o threads stay in wait lock.
569 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
573 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
577 Description: Update to SLES9 kernel-2.6.5-7.287.3.
581 Description: Update to RHEL4 latest kernel.
583 Severity : enhancement
585 Description: Build SLES10 patchless client fails
586 Details : The configure was broken by run ./configure with
587 --with-linux-obj=.... argument for patchless client. When the
588 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
589 can't search header adequately. Use absolute path such as
590 -I($LINUX)/include instead.
594 Description: interrupt oig_wait produce painc on resend.
595 Details : brw_redo_request can be used for resend requests from ptlrpcd and
596 private set, and this produce situation when rq_ptlrpcd_data not
597 copyed to new allocated request and triggered LBUG on assert
598 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
599 ptlrpcd set if request is changed and can be safety changed to use
604 Description: LASSERT_{REQ,REP}SWAB macros are buggy
605 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
606 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
607 these macros and replace them with their difinitions instead.
611 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
612 Details : Update which_patch & target file for SLES10 latest kernel.
615 Description: Remove client patches from SLES 10 kernel.
616 Details : This causes SLES 10 clients to behave as patchless clients
617 even on a Lustre-patched (server) kernel.
619 Severity : enhancement
621 Description: self-adjustable client's lru lists
622 Details : use adaptive algorithm for managing client cached locks lru
623 lists according to current server load, other client's work
624 pattern, memory activities, etc. Both, server and client
625 side namespaces provide number of proc tunables for controlling
628 Severity : enhancement
630 Description: light-weight GSS support
631 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
632 and reduce performance overhead.
634 Severity : enhancement
636 Description: Linux keyring support
637 Details : Support using service of Linux keyring for Lustre GSS internal
638 context refresh/cache mechanism.
642 Description: Fix errors in lfs documentation
643 Details : Fixes man pages
647 Description: don't use GFP_* in generic Lustre code.
648 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
652 Description: obdclass is limited by single OBD_ALLOC(idarray)
653 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
657 Description: warnings with build patchless client with vanila 2.6.19 and up
658 Details : change old ctl_table style and replace ctl_table/ctl_table_header
659 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
663 Frequency : Only for SLES
664 Description: sanity_quota fail test_1
665 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
666 file, which makes TSTID[2] inunique.
670 Frequency : Only for SLES9
671 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
672 Details : Update target/ChangeLog/which_patch .
676 Frequency : Only for test_10 in sanity-quota.sh
677 Description: a bug in quota test code
678 Details : checking a test flag in a wrong place
682 Frequency : Only for test_13 in sanity-quota.sh
683 Description: a bug in quota test script
684 Details : shouldn't check overfull allocated quota
688 Description: jbd statistics
689 Details : Port older jbd statistics patch for sles10
691 Severity : enhancement
693 Description: Add printing inode info into message about error in writepage.
698 Description: reply_lock_interpret crash due to race with it and lock cancel.
699 Details : Do not replay locks that are being cancelled. Do not reference
700 locks by their address during replay, just by their handle.
704 Frequency : When flocks are used.
705 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
706 Details : Flock locks might destroy just granted lock if it could be merged
707 with another existing flock, this is done in completion handler,
708 so teach ldlm_cli_enquque_fini that this is a valid case for
714 Description: Oops in read and write path when failing to allocate lock.
715 Details : Check if lock allocation failed and return error back.
719 Description: lstripe command fails for valid OST index
720 Details : The stripe offset is compared to lov->desc.ld_tgt_count
721 instead of lov->desc.ld_active_tgt_count.
725 Description: sanity.sh failed test 103
726 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
731 Description: df doesn't work properly if diskfs blocksize != 4K
732 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
736 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
737 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
739 Severity : enhancement
741 Description: Adaptive timeouts
742 Details : RPC timeouts adapt to changing server load and network
743 conditions to reduce resend attempts and improve recovery time.
747 Description: llapi_file_create() does not allow some changes
748 Details : add llapi_file_open() that allows specifying the mode and
749 open flags, and also returns an open file handle.
753 Description: merge and cleanup kernel patches.
754 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
758 Description: Add journal checksum support.(Kernel part)
759 Details : The journal checksum feature adds two new flags i.e
760 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
761 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
762 indicates that the commit block contains the checksum for
763 the blocks described by the descriptor blocks. Now commit
764 record can be sent to disk without waiting for descriptor
765 blocks to be written to disk. This behavior is controlled
766 using JBD2_FEATURE_ASYNC_COMMIT flag.
770 Description: OSS needs mutliple precreate threads
771 Details : Add ability to start more than one create thread per OSS.
775 Description: Sanity.sh test_65k failed: setstripe should have succeeded
776 Details : execute lfs setstripe on client
780 Description: mds_obd_create error creating tmp object
781 Details : When the user sets quota on root, llog will be affected and can't
782 create files and write files.
785 Frequency : Always on ia64 patchless client, and possibly others.
787 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
788 Details : This allows the patchless client to be loaded on architectures
793 Description: RedHat Update kernel for RHEL5
794 Details : Add the kernel config file for RHEL5.
798 Description: RedHat Update kernel for RHEL5
799 Details : Modify the kernel config file more closer RHEL5.
803 Description: Build failure against Centos5 (RHEL5)
804 Details : Define PAGE_SIZE when it isn't present.
808 Description: test_7 on sanity-quota.sh can't be used on separate machines
809 Details : add facet to handle it
813 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
814 Details : with NFS, the anon dentry's parent was set to itself in
815 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
816 corresponding dentry other than use rec->ur_name.
818 Severity : enhancement
820 Description: lfs setstripe enhancement
821 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
825 Description: enable data checksumming by default
826 Details : enable checksum by default, allow --disable-checksum
827 configure option and "-o nochecksum" mount option. Checksums
828 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
832 Description: lustre support for RHEL5
833 Details : Add support for RHEL5.
837 Description: Client eviction due to failover config
838 Details : after a connection loss, the lustre client should attempt to
839 reconnect to the last active server first before trying the
840 other potential connections.
844 Description: when mds and osts use different quota unit(32bit and 64bit),
845 quota will be released repeatly.
846 Details : void sending multiple quota reqs to mds, which will keep the
847 status between the reqs.
851 Description: osts not allocated evenly to files
852 Details : change the condition to increase offset_idx
855 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
857 Description: Data corruption for OSTs that are formatted larger than 2TB
859 Details : When generating the bio request for lustre file writes the
860 sector number would overflow a temporary variable before being
861 used for the IO. The data reads correctly from Lustre (which
862 will overflow in a similar manner) but other file data or
863 filesystem metadata may be corrupted in some cases.
867 Description: Tune the kernel for good SCSI performance.
868 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
869 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
874 Description: rewrite ext2-derived code in obdclass/uuid.c
875 Details : rewrite inherited code (uuid parsing code from ext2 utils)
876 from scratch preserving functionality.
880 Description: "lfs find -obd UUID" prints directories
881 Details : "lfs find -obd UUID" will return all directory names instead
882 of just file names. It is incorrect because the directories
883 do not reside on the OSTs.
887 Description: use i_size_read and i_size_write in 2.6 port
888 Details : replace inode->i_size access with i_size_read/write()
891 Frequency : when removing large files
893 Description: scheduling issue during removal of large Lustre files
894 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
895 It causes scheduling issues when removing large files (17TB in the
899 Frequency : only with liblustre clients on XT3
901 Description: evictions taking too long
902 Details : allow llrd to evict clients directly on OSTs
905 Frequency : only on ppc
907 Description: /proc/fs/lustre/devices broken on ppc
908 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
909 We should make sure the type of 'index' is loff_t and then cast
910 explicitly as needed below. Do not assign an explicitly cast
914 Frequency : only for rhel5
916 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
917 Details : Modify the target file & which_kernel.
919 Severity : enhancement
921 Description: omit set fsid for export NFS
922 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
923 as FS_REQUIRES_DEV for avoid problems with generate fsid.
927 Frequency : Always, for kernels after 2.6.16
928 Description: Fix warning idr_remove called for id=.. which is not allocated.
929 Details : Last kernels save old s_dev before kill super and not allow
930 to restore from callback - restore it before call kill_anon_super.
934 Description: buffer overruns could theoretically occur
935 Details : llapi_semantic_traverse() modifies the "path" argument by
936 appending values to the end of the origin string, and a
937 overrun may occur. Adding buffer overrun check in liblustreapi.
941 Description: Fix error on 'ls .' at the top of the Lustre mount.
942 Details : Don't revalidate dentry if it is a root dentry.
946 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
947 Details : Modify vm-tunables-rhel4.patch.
951 Description: Kernel config for 2.6.18-vanilla.
952 Details : Modify targets/2.6-vanilla.target.in.
953 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
954 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
955 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
956 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
960 Description: improve handling recoverable errors
961 Details : if request processig with error which can be recoverable on server
962 request should be resend, otherwise page released from cache and
967 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
968 Details : Modify target file & which_patch.
969 A flaw was found in the IA32 system call emulation provided
970 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
971 value could be stored in the %RAX register, which could trigger an
972 out-of-bounds system call table access. An untrusted local user
973 could exploit this flaw to run code in the kernel
974 (ie a root privilege escalation). (CVE-2007-4573).
978 Description: O_DIRECT bypasses client statistics.
979 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
980 read_ahead_stats were updated. copied the stats section from
981 osc_send_oap_rpc() into async_internal().
985 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
989 Description: change order of libsysio includes
990 Details : '#include sysio.h' should always come before '#include xtio.h'
994 Description: handle "serious error: objid * already exists" more gracefully
995 Details : If LAST_ID value on disk is smaller than the objects existing in
996 the O/0/d* directories, it indicates disk corruption and causes an
997 LBUG(). If the object is 0-length, then we should use the existing
998 object. This will help to avoid a full fsck in most cases.
1000 Severity : enhancement
1002 Description: adapt the lustre_config script to support the upgrade case
1003 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1004 server targets to 1.6 in parallel.
1008 Description: To avoid grant space > avaible space when the disk is almost
1009 full. Without this patch you might see the error "grant XXXX >
1010 available" or some LBUG about grant, when the disk is almost
1012 Details : In filter_check_grant, for non_grant cache write, we should
1013 check the left space by if (*left > ungranted + bytes), instead
1014 of (*left > ungranted), because only we are sure the left space
1015 is enough for another "bytes", then the ungrant space should be
1016 increase. In client, we should update cl_avail_grant only there
1017 is OBD_MD_FLGRANT in the reply.
1021 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1022 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1023 described in CVE-2007-4573. This problem could allow untrusted
1024 local users to gain root access.
1027 Frequency : when using O_DIRECT and quotas
1029 Description: Incorrect file ownership on O_DIRECT output files
1030 Details : block usage reported by 'lfs quota' does not take into account
1031 files that have been written with O_DIRECT.
1036 Description: touch file failed when fs is not full
1037 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1038 otherwise we can get ENOSP while fs is not full.
1042 Description: parallel lock callbacks
1043 Details : Instead of sending blocking and completion callbacks as separated
1044 requests, adding them to a set and sending in parallel.
1047 Frequency : only for Cray XT3
1048 Bugzilla : 12829/13455
1049 Description: Changing primary group doesn't change the group lustre assigns to
1051 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1052 overridden with the primary group provided by the group upcall,
1053 whereas the supplied fsgid can be trusted if it is in the list of
1054 supplementary groups returned by the group upcall.
1056 Severity : enhancement
1058 Description: Allow masking D_WARNING, D_ERROR messages from console
1059 Details : Console messages can now be disabled via lnet.printk.
1063 Description: User code with malformed file open parameter crashes client node
1064 Details : Before packing join_file req, all the related reference should be
1065 checked carefully in case some malformed flags cause fake
1066 join_file req on client.
1070 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1071 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1072 mode is equal to the granted mode. However, it should also check
1073 that LDLM_FL_CP_REQD is not set.
1077 Description: Heavy nfs access might result in deadlocks
1078 Details : After ELC code landed, it is now improper to enqueue any mds
1079 locks under och_sem, because enqueue might want to decide to
1080 cancel open locks for same inode we are holding och_sem for.
1084 Description: Client eviction while running blogbench
1085 Details : A lot of unlink operations with concurrent I/O can lead to a
1086 deadlock causing evictions. To address the problem, the number of
1087 oustanding OST_DESTROY requests is now throttled to
1088 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1089 callbacks are processed in priority.
1093 Description: enable ACLs on MDS by default
1094 Details : ACLs must be enabled on MDS by default.
1097 Frequency : PPC/PPC64 only
1099 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1100 Details : fix duplicated definitions between asm-ppc64/types.h and
1101 lustre_types.h on PPC.
1104 Frequency : PPC/PPC64 only
1106 Description: asm-ppc/segment.h does not exist
1107 Details : fix compile issue on PPC.
1111 Description: better handle error messages in extents code
1114 Frequency : RHEL4 only
1116 Description: mkfs is very slow on IA64/RHEL4
1117 Details : A performance regression has been discovered in the MPT Fusion
1118 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1119 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1120 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1122 Severity : enhancement
1124 Description: SNMP support enhancement
1125 Details : Adding total number of sampled request for an MDS node in snmp
1128 Severity : enhancement
1130 Description: Optimize ldlm waiting list processing for PR extent locks
1131 Details : When processing waiting list for read extent lock and meeting read
1132 lock that is same or wider to it that is not contended, skip
1133 processing rest of the list and immediatelly return current
1134 status of conflictness, since we are guaranteed there are no
1135 conflicting locks in the rest of the list.
1139 Description: Time out and refuse to reconnect
1140 Details : When the failover node is the primary node, it is possible
1141 to have two identical connections in imp_conn_list. We must
1142 compare not conn's pointers but NIDs, otherwise we can defeat
1143 connection throttling.
1147 Description: port llog fixes from b1_6 into HEAD
1148 Details : Port llog reference couting and some llog cleanups from b1_6
1149 (bug 10800) into HEAD, for protect from panic and access to already
1150 free llog structures.
1154 Description: Detect stride IO mode in read-ahead
1155 Details : When a client does stride read, read-ahead should detect that and
1156 read-ahead pages according to the detected stride pattern.
1160 Description: data checksumming impacts single node performance
1161 Details : add support for several checksum algorithm. Currently, only CRC32
1162 and Adler-32 are supported. The checksum type can be changed on
1163 the fly via /proc/fs/lustre/osc/*/checksum_type.
1167 Description: use adler32 for page checksums
1168 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1173 Description: build for x2 fails
1174 Details : fix compile issue on Cray systems.
1178 Description: Properly match for duplicate locks
1179 Details : Due to different lock order from skiplists code, we need to
1180 traverse entire list for now
1183 Frequency : only on PPC/SLES10
1185 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1186 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1187 be defined. Add a hack in mkfs_lustre.c to work around this compile
1192 Description: LASSERT on MDS when client holding flock lock dies
1193 Details : ldlm pool logic depends on number of granted locks equal to
1194 number of released locks which is not true for flock locks, so
1195 just exclude such locks from consideration.
1199 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1200 Details : Use fsfilt_sync() for both the whole filesystem sync and
1201 individual file sync to eliminate dangerous inode locking
1202 with I_LOCK that can lead to a deadlock.
1206 Description: performance in 1.6.3
1207 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1208 sure that 1MB requests are not fragmented by the block layer.
1210 Severity : enhancement
1212 Description: organize the server-side client stats on per-nid basis
1213 Details : Change the structure of stats under obdfilter and mds to
1221 The "uuid"s file would list the uuids of _active_ exports.
1222 And the clear entry is to clear all stats and stale nids.
1224 Severity : enhancement
1226 Description: eliminate client locks in face of contention
1227 Details : file contention detection and lockless i/o implementation
1228 for contended files.
1232 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1235 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1237 Description: MDS may lose file striping (and hence file data) in some cases
1238 Details : If there are additional extended attributes stored on the MDS,
1239 in particular ACLs, SELinux, or user attributes (if user_xattr
1240 is specified for the client mount options) then there is a risk
1241 of attribute loss. Additionally, the Lustre file striping
1242 needs to be larger than default (e.g. striped over all OSTs),
1243 and an additional attribute must be stored initially in the
1244 inode and then increase in size enough to be moved to the
1245 external attribute block (e.g. ACL growing in size) for file
1250 Description: skiplist implementation simplification
1251 Details : skiplists are used to group compatible locks on granted list
1252 that was implemented as tracking first and last lock of each
1253 lock group the patch changes that to using doubly linked lists
1257 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1258 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1259 because LASSERT on both the data supplied by a client, and the
1260 data on disk is dangerous and incorrect.
1262 Severity : enhancement
1264 Description: Slow truncate/writes to huge files at high offsets.
1265 Details : Directly associate cached pages to lock that protect those pages,
1266 this allows us to quickly find what pages to write and remove
1267 once lock callback is received.
1271 Description: more ldlm soft lockups
1272 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1273 starve other threads from the resource lock for a long time in
1274 case of long waiting queue, so change the debug level from
1275 D_OTHER to the less frequently used D_INFO.
1277 Severity : enhancement
1279 Description: add -gid, -group, -uid, -user options to lfs find
1283 Description: Hung threads in invalidate_inode_pages2_range
1284 Details : The direct IO path doesn't call check_rpcs to submit a new RPC once
1285 one is completed. As a result, some RPCs are stuck in the queue
1290 Description: filter threads hungs on waiting journal commit
1291 Details : Cleanup filter group llog code, then only filter group llog will
1292 be only created in the MDS/OST syncing process.
1296 Description: Procfs and llog threads access destoryed import sometimes.
1297 Details : Sync the import destoryed process with procfs and llog threads by
1298 the import refcount and semaphore.
1300 Severity : enhancement
1302 Description: openlock cache of b1_6 port to HEAD
1307 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1308 Details : If insertion of an extent fails, then discard the inode
1309 preallocation and free data blocks else it can lead to duplicate
1314 Description: don't always update ctime in ext3_xattr_set_handle()
1315 Details : Current xattr code updates the inode ctime in ext3_xattr_set_handle.
1316 In some cases the ctime should not be updated, for example for
1317 2.0->1.8 compatibility it is necessary to delete an xattr and it
1318 should not update the ctime.
1322 Bugzilla : 15713/16362
1323 Description: Assertion in iopen_connect_dentry in 1.6.3
1324 Details : looking up an inode via iopen with the wrong generation number can
1325 populate the dcache with a disconneced dentry while the inode
1326 number is in the process of being reallocated. This causes an
1327 assertion failure in iopen since the inode's dentry list contains
1328 both a connected and disconnected dentry.
1332 Description: assertion failure in ldlm_handle2lock()
1333 Details : fix a race between class_handle_unhash() and class_handle2object()
1334 introduced in lustre 1.6.5 by bug 13622.
1339 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1340 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1345 Description: oops in sd_iostats_seq_show()
1346 Details : unloading/reloading the scsi low level driver triggers a kernel
1347 bug when trying to access the sd iostat file.
1352 Description: Kernel panics during QLogic driver reload
1353 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1354 patch, causing memory corruption.
1359 Description: journal_dev option does not work in b1_6
1360 Details : pass mount option during pre-mount.
1362 Severity : enhancement
1364 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1365 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1366 extent information of a file. It can be used to map logical blocks
1367 in a file to physical blocks in the block device.
1371 Description: LDLM soft lockups - improvement
1372 Details : It is be possible to send the lock handle along with each read
1373 or write request because the client is already doing a lock match
1374 itself so there isn't any reason the OST should have to re-do that
1377 --------------------------------------------------------------------------------
1379 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
1381 * Support for kernels:
1382 2.6.5-7.283 (SLES 9),
1383 2.6.9-55.EL (RHEL 4),
1384 2.6.16.46-0.14 (SLES 10),
1385 2.6.18.8 vanilla (kernel.org)
1386 * Client support for unpatched kernels:
1387 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1388 2.6.16 - 2.6.22 vanilla (kernel.org)
1389 * Due to recently discovered recovery problems, we do not recommend
1390 using patchless RHEL 4 clients with this or any earlier release.
1391 * Recommended e2fsprogs version: 1.39.cfs8
1392 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1393 * Starting with this release, the ldiskfs backing filesystem required
1394 by Lustre is now in its own package, lustre-ldiskfs. This package
1395 should be installed. It is versioned separately from Lustre and
1396 may be released separately in future.
1400 Description: block reactivating mgc import until all deactivates complete
1401 Details : Fix race when failing back MDT/MGS to itself (testing)
1404 Frequency : at statup only
1406 Description: mds_lov_synchronize race leads to various problems
1407 Details : simultaneous MDT->OST connections at startup can cause the
1408 sync to abort, leaving the OSC in a bad state.
1410 Severity : enhancement
1412 Description: add optional extra BUILD_VERSION info
1413 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1414 to override the lustre version.
1417 Frequency : 2.6.18 servers only
1419 Description: ll_kern_mount() doesn't release the module reference
1420 Details : The ldiskfs module reference count never drops down to 0
1421 because ll_kern_mount() doesn't release the module reference.
1426 Description: server LBUG when using old ost_num_threads parameter
1427 Details : Accept the old ost_num_threads parameter but warn that it
1428 is deprecated, and fix an off-by-one error that caused an LBUG.
1433 Description: Transient SCSI error results in persistent IO issue
1434 Details : iobuf->dr_error is not reinitialized to 0 between two
1438 Frequency : sometimes when underlying device returns I/O errors
1440 Description: OSTs not going read-only during write failures
1441 Details : OSTs are not remounted read-only when the journal commit threads
1442 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1443 instead of the ext3 wrappers.
1447 Description: poor connect scaling with increasing client count
1448 Details : Don't run filter_grant_sanity_check for more than 100 exports
1449 to improve scaling for large numbers of clients.
1452 Frequency : SLES10 only
1454 Description: sanity-quota.sh quotacheck failed: rc = -22
1455 Details : Quotas cannot be enabled on SLES10.
1458 Frequency : liblustre clients only
1460 Description: getdirentries does not give error when run on compute nodes
1461 Details : getdirentries does not fail when the size specified as an argument
1462 is too small to contain at least one entry
1464 Severity : enhancement
1466 Description: Add LNET router traceability for debug purposes
1467 Details : If a checksum failure occurs with a router as part of the
1468 IO path, the NID of the last router that forwarded the bulk data
1469 is printed so it can be identified.
1474 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1475 Details : Due to a race condition, liblustre clients were occasionally
1476 evicted incorrectly.
1478 Severity : enhancement
1480 Description: lfs setstripe use optional parameters instead of postional
1483 Severity : enhancement
1485 Description: Nanosecond timestamp support for ldiskfs
1486 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1487 resolution timestamps. There is not yet support for this at
1488 the Lustre filesystem level.
1491 Frequency : during server recovery
1493 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1494 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1498 Frequency : nfs export on patchless client
1500 Description: connectathon hang when test nfs export over patchless client
1501 Details : Disconnected dentry cannot be found with lookup, so we do not need
1502 to unhash it or make it invalid
1505 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1506 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1507 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1512 Description: open req refcounting wrong on reconnect
1513 Details : If reconnect happened between getting open reply from server and
1514 call to mdc_set_replay_data in ll_file_open, we will schedule
1515 replay for unreferenced request that we are about to free.
1516 Subsequent close will crash in variety of ways.
1517 Check that request is still eligible for replay in
1518 mdc_set_replay_data().
1523 Description: disable writes to filesystem when reading health_check file
1524 Details : the default for reading the health_check proc file has changed
1525 to NOT do a journal transaction and write to disk, because this
1526 can cause reads of the /proc file to hang and block HA state
1527 checking on a healthy but otherwise heavily loaded system. It
1528 is possible to return to the previous behaviour during configure
1529 with --enable-health-write.
1531 Severity : enhancement
1533 Description: 64-bit inode version
1534 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1535 to the inode. This will be required for version-based recovery.
1540 Description: MDS fails to start if a duplicate client export is detected
1541 Details : in some rare cases it was possible for a client to connect to
1542 an MDS multiple times. Upon recovery the MDS would detect this
1543 and fail during startup. Handle this more gracefully.
1545 Severity : enhancement
1547 Description: Add -o localflock option to simulate old noflock
1549 Details : This will achieve local-only flock/fcntl locks
1555 Description: log_commit_thread vs filter_destroy race leads to crash
1556 Details : Take import reference before releasing llog record semaphore
1561 Description: Wrong request locking in request set processing
1562 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1563 imp_delayed_list, in this place should be used imp_lock.
1566 Frequency : when reconnection
1568 Description: Grant Leak when osc reconnect to OST
1569 Details : When osc reconnect ost, OST(filter) should check whether it
1570 should grant more space to client by comparing fed_grant and
1571 cl_avail_grant, and return the granted space to client instead
1572 of "new granted" space, because client will call osc_init_grant
1573 to update the client grant space info.
1576 Frequency : when client reconnect to OST
1578 Description: Grant Leak when osc do resend and replay bulk write
1579 Details : When osc reconnect to OST, OST(filter)should clear grant info of
1580 bulk write request, because the grant info while be sync between
1581 OSC and OST when reconnect, and we should ignore the grant info
1582 these of resend/replay write req.
1587 Description: Grant space more than avaiable left space sometimes.
1588 Details : When then OST is about to be full, if two bulk writing from
1589 different clients came to OST. Accord the avaliable space of the
1590 OST, the first req should be permitted, and the second one
1591 should be denied by ENOSPC. But if the seconde arrived before
1592 the first one is commited. The OST might wrongly permit second
1593 writing, which will cause grant space > avaiable space.
1596 Frequency : when client is evicted
1598 Description: Grant might be wrongly erased when osc is evicted by OST
1599 Details : when the import is evicted by server, it will fork another
1600 thread ptlrpc_invalidate_import_thread to invalidate the
1601 import, where the grant will be set to 0. While the original
1602 thread will update the grant it got when connecting. So if
1603 the former happened latter, the grant will be wrongly errased
1604 because of this race.
1609 Description: Checking Stale with correct fid
1610 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1611 to check whether it is stale, because sometimes, we want the
1612 enqueue happened anyway, and op_data.fid2 will not be initialized.
1614 Severity : enhancement
1616 Description: update patchless client
1617 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1620 Frequency : only with 2.4 kernel
1622 Description: random memory corruption
1623 Details : size of struct ll_inode_info is to big for union inode.u and this
1624 can be cause of random memory corruption.
1629 Description: Memory leak in recovery
1630 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1631 It should also check obd_fail before fsfilt_start, otherwise if
1632 fsfilt_start return -EROFS,(failover mds during mds recovery).
1633 then the req will return with repmsg->transno = 0 and rc = EROFS.
1634 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1635 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1636 be freed no matter whether fsfilt_commit success or not.
1639 Frequency : only with huge count clients
1641 Description: Prevents from taking the superblock lock in llap_from_page for
1643 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1644 for llap_from_page call in ll_removepage prevents from taking the
1645 superblock lock for a soon died page.
1650 Description: Not check open intent error before release open handle
1651 Details : in some rare cases, the open intent error is not checked before
1652 release open handle, which may cause
1653 ASSERTION(open_req->rq_transno != 0), because it tries to release
1654 the failed open handle.
1659 Description: Set cat log bitmap only after create log success.
1660 Details : in some rare cases, the cat log bitmap is set too early. and it
1661 should be set only after create log success.
1665 Description: Accessing a block bevice can re-enable I/O when Lustre is
1666 tearing down a device.
1667 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1672 Description: service threads may hog cpus when there are a lot of requests
1673 Details : Insert cond_resched to give other threads a chance to use some CPU
1678 Description: the cat log was not initialized in recovery
1679 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1680 unlink log on mds will not be initialized until mds post
1681 recovery. And also in mds post recovery, the unlink log will
1682 initialization will be done asynchronausly, so there will be race
1683 between add unlink log and unlink log initialization.
1687 Description: brw_stats were being printed incorrectly
1688 Details : brw_stats were being printed as log2 but all of them were not
1689 recorded as log2. Also remove some code duplication arising from
1690 filter_tally_{read,write}.
1694 Frequency : rare, only in recovery.
1695 Description: ASSERTION(req->rq_type != LI_POISON) failed
1696 Details : imp_lock should be held while iterating over imp_sending_list for
1697 prevent destroy request after get timeout in ptlrpc_queue_wait.
1701 Description: replay-single.sh test 52 fails
1702 Details : A lock's skiplist need to be cleanup when it being unlinked
1703 from its resource list.
1707 Description: Short directio read returns full requested size rather than
1709 Details : Direct I/O operations should return actual amount of bytes
1710 transferred rather than requested size.
1712 Severity : enhancement
1714 Description: metadata RPC reduction (e.g. for rm performance)
1715 Details : decrease the amount of synchronous RPC between clients and servers
1716 by canceling conflicing lock before the operation on the client
1717 and packing thier handles into the main operation RPC to server.
1719 Severity : enhancement
1721 Description: Async OSC create to avoid the blocking unnecessarily.
1722 Details : If a OST has no remain object, system will block on the creating
1723 when need to create a new object on this OST. Now, ways use
1724 pre-created objects when available, instead of blocking on an
1725 empty osc while others are not empty. If we must block, we block
1726 for the shortest possible period of time.
1730 Description: improve handling recoverable errors
1731 Details : if request processig with error which can be recoverable on server
1732 request should be resend, otherwise page released from cache and
1735 Severity : enhancement
1737 Description: refine locking for avoid write wrong info into lov_objid file
1738 Details : fix possible races with add new target and write/update data in
1741 --------------------------------------------------------------------------------
1743 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1748 Frequency : on some architectures
1750 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1751 Details : Uninitialized flags sometimes cause configuration commands to
1755 Frequency : patchless clients only
1757 Description: missing __iget() symbol export
1758 Details : The __iget() symbol export is missing. To avoid the need for
1759 this on patchless clients the deathrow inode reaper is turned
1760 off, and we depend on the VM to clean up old inodes. This
1761 dependency was during via the fix for bug 12181.
1763 --------------------------------------------------------------------------------
1765 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1767 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1768 INTEROPERATE with older versions automatically. In many cases a
1769 special upgrade step is needed. Please read the
1770 user documentation before upgrading any part of a live system.
1771 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1772 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1773 * WARNING: Lustre configuration and startup changes are required with
1774 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1780 Severity : enhancement
1782 Description: MountConf
1783 Details : Lustre configuration is now managed via mkfs and mount
1784 commands instead of lmc and lconf. New obd types (MGS, MGC)
1785 are added for dynamic configuration management. See
1786 https://mail.clusterfs.com/wikis/lustre/MountConf for
1789 Severity : enhancement
1791 Description: dynamic OST addition
1792 Details : OSTs can now be added to a live filesystem
1794 Severity : enhancement
1796 Description: startup order invariance
1797 Details : MDTs and OSTs can be started in any order. Clients only
1798 require the MDT to complete startup.
1800 Severity : enhancement
1802 Description: parallel, asynchronous orphan cleanup
1803 Details : orphan cleanup is now performed in separate threads for each
1804 OST, allowing parallel non-blocking operation.
1806 Severity : enhancement
1808 Description: optimized stripe assignment
1809 Details : stripe assignments are now made based on ost space available,
1810 ost previous usage, and OSS previous usage, in order to try
1811 to optimize storage space and networking resources.
1813 Severity : enhancement
1815 Description: Permanently set tunables
1816 Details : All writable /proc/fs/lustre tunables can now be permanently
1817 set on a per-server basis, at mkfs time or on a live system.
1819 Severity : enhancement
1821 Description: Lustre message v2
1822 Details : Add lustre message format v2.
1824 Severity : enhancement
1826 Description: client OST exclusion list
1827 Details : Clients can be started with a list of OSTs that should be
1828 declared "inactive" for known non-responsive OSTs.
1832 Description: SPEC SFS validation failure on NFS v2 over lustre.
1833 Details : Changes the blocksize for regular files to be 2x RPC size,
1834 and not depend on stripe size.
1836 Severity : enhancement
1838 Description: Multiple MD RPCs in flight.
1839 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1840 To avoid overly-overloading MDS, introduce a limit on number of
1841 MDS RPCs in flight for a single client and add /proc controls
1842 to adjust this limit.
1844 Severity : enhancement
1846 Description: client read/write statistics
1847 Details : Add client read/write call usage stats for performance
1848 analysis of user processes.
1849 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1850 file access. extents_stats shows chunk size distribution.
1851 extents_stats_per_process show chunk size distribution per
1854 Severity : enhancement
1856 Description: mds statistics
1857 Details : Add detailed mds operations statistics in
1858 /proc/fs/lustre/mds/*/stats.
1862 Description: Failure of copying files with lustre special EAs.
1863 Details : Client side always return success for setxattr call for lustre
1864 special xattr (currently only "trusted.lov").
1869 Description: Refcount LNET uuids
1870 Details : The global LNET uuid list grew linearly with every startup;
1871 refcount repeated list entries instead of always adding to
1874 Severity : enhancement
1876 Description: Dynamic service threads
1877 Details : Within a small range, start extra service threads
1878 automatically when the request queue builds up.
1881 Frequency : mixed-endian client/server environments
1883 Description: mixed-endian crashes
1884 Details : The new msg_v2 system had some failures in mixed-endian
1887 Severity : enhancement
1889 Description: Easy OST removal
1890 Details : OSTs can be permanently deactivated with e.g. 'lctl
1891 conf_param lustre-OST0001.osc.active=0'
1893 Severity : enhancement
1895 Description: MGS proc entries
1896 Details : Added basic proc entries for the MGS showing what filesystems
1899 Severity : enhancement
1901 Description: provide MGS failover
1902 Details : Added config lock reacquisition after MGS server failover.
1904 Severity : enhancement
1906 Description: add Linux 2.4 support
1907 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1911 Description: a large application tries to do I/O to the same resource and dies
1912 in the middle of it.
1913 Details : Check the req->rq_arrival time after the call to
1914 ost_brw_lock_get(), but before we do anything about
1915 processing it & sending the BULK transfer request. This
1916 should help move old stale pending locks off the queue as
1917 quickly as obd_timeout.
1920 Frequency : when an incorrect nid is specified during startup
1922 Description: ptlrpc connect to non-existant node causes kernel crash
1923 Details : LNET can't be re-entered from an event callback, which
1924 happened when we expire a message after the export has been
1925 cleaned up. Instead, hand the zombie cleanup off to another
1928 Severity : enhancement
1930 Description: plain/inodebits lock performance improvement
1931 Details : Grouping plain/inodebits in granted list by their request modes
1932 and bits policy, thus improving the performance of search through
1936 Frequency : only if OST filesystem is corrupted
1938 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1939 Details : for a short time RPCs with bulk IO are in the replay list,
1940 but replay of bulk IOs is unimplemented. If the OST filesystem
1941 is corrupted due to disk cache incoherency and then replay is
1942 started it is possible to trip an assertion. Avoid putting
1943 committed RPCs into the replay list at all to avoid this issue.
1946 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1948 Description: System hang on startup
1949 Details : This bug allowed the liblustre (e.g. catamount) client to
1950 return to the app before handling all startup RPCs. This
1951 could leave the node unresponsive to lustre network traffic
1952 and manifested as a server ptllnd timeout.
1954 Severity : enhancement
1956 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1957 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1958 boolean to control whether to print peer debug info when a
1959 client's RPC times out.
1962 Frequency : only for kernels with patches from Lustre below 1.4.3
1964 Description: Remove old rdonly API
1965 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1968 Frequency : only for devices with external journals
1970 Description: Set external device read-only also
1971 Details : During a commanded failover stop, we set the disk device
1972 read-only while the server shuts down. We now also set any
1973 external journal device read-only at the same time.
1976 Frequency : when upgrading from 1.4 while trying to change parameters
1978 Description: The wrong (new) MDC name was used when setting parameters for
1979 upgraded MDT's. Also allows changing of OSC (and MDC)
1980 parameters if --writeconf is specified at tunefs upgrade time.
1983 Frequency : when setting specific ost indicies
1985 Description: QOS code breaks on skipped indicies
1986 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1987 created with --index need not be sequential.
1989 Severity : enhancement
1991 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1992 Details : The uninit_groups feature works in conjunction with the kernel
1993 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1994 up the pass1 processing of e2fsck. This is a read-only feature
1995 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1996 mount filesystems that have had this feature enabled.
1998 Severity : enhancement
2000 Description: Improve multi-block allocation algorithm to avoid fragmentation
2001 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2002 allocation locality and avoid filesystem fragmentation.
2004 ------------------------------------------------------------------------------
2006 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
2008 * Support for kernels:
2009 2.4.21-47.0.1.EL (RHEL 3)
2010 2.6.5-7.283 (SLES 9)
2011 2.6.9-42.0.10.EL (RHEL 4)
2012 2.6.12.6 vanilla (kernel.org)
2013 2.6.16.27-0.9 (SLES 10)
2014 * Recommended e2fsprogs version: 1.39.cfs5
2016 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2020 Frequency : occasional, depends on client load and configuration
2021 Bugzilla : 12181, 12203
2022 Description: data loss for recently-modified files
2024 Details : In some cases it is possible that recently written or created
2025 files may not be written to disk in a timely manner (this should
2026 normally be within 30s unless client IO load is very high).
2027 The problem appears as zero-length files or files that are a
2028 multiple of 1MB in size after a client crash or client eviction
2029 that are missing data at the end of the file.
2031 This problem is more likely to be hit on clients where files are
2032 repeatedly created and unlinked in the same directory, clients
2033 have a large amount of RAM, have many CPUs, the filesystem has
2034 many OSTs, the clients are rebooted frequently, and/or the files
2035 are not accessed by other nodes after being written.
2037 The presence of the problem can be detected by looking at
2038 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2039 smaller than the second (nr_unused) then dirty files will not
2040 be flushed automatically to disk. "sync; sleep 10" should be
2041 run several times on the node before unmounting it to update
2042 Lustre (this is also safe to run on nodes without this problem).
2044 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2045 that can cause this same problem, so customers using that kernel
2046 also need to update the kernel in addition to Lustre. In order
2047 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2049 It is normal that files written just before a client crash (less
2050 than 30s) may not yet have been flushed to disk, even for local
2054 Frequency : frequent on thin XT3 nodes
2056 Description: UUID collision on thin XT3 Linux nodes
2057 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2058 randomly, since we relied on an insufficiently-seeded PRNG.
2063 Description: OSS hangs after "All ost request buffers busy"
2064 Details : A deadlock between quota and journal operations caused OSS
2065 hangs after printing "All ost request buffers busy."
2068 Frequency : always on liblustre builds
2070 Description: Cleanup compiler warnings on liblustre
2073 Frequency : always on liblustre builds on XT3
2075 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2079 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2080 > 32000 subdirectories
2081 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2082 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2083 subdirectory count crosses 32000. This will aid e2fsck to
2084 correctly handle more than 32000 subdirectories.
2087 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2089 Description: System hang on startup
2090 Details : This bug allowed the liblustre (e.g. catamount) client to
2091 return to the app before handling all startup RPCs. This
2092 could leave the node unresponsive to lustre network traffic
2093 and manifested as a server ptllnd timeout.
2095 Severity : enhancement
2097 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2098 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2099 boolean to control whether to print peer debug info when a
2100 client's RPC times out.
2105 Description: make O_SYNC working on 2.6 kernels
2106 Details : 2.6 kernels use different method for mark pages for write,
2107 so need add a code to lustre for O_SYNC work.
2112 Description: Failure to close file and release space on NFS
2113 Details : Put inode details into lock acquired in ll_intent_file_open.
2114 Use mdc_intent_lock in ll_intent_open to properly
2115 detect all kind of errors unhandled by mdc_enqueue.
2120 Description: proc file read during shutdown sometimes raced obd removal,
2122 Details : Add lock to prevent obd access after proc file removal.
2125 Frequency : Only for files larger than 4GB on 32-bit clients.
2127 Description: improperly doing page alignment of locks
2128 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2129 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2134 Description: RPCs being resent when they shouldn't be
2135 Details : Some RPCs that should not be resent are being resent. This
2136 can cause inconsistencies in the RPC state machine. Do not
2137 resend such requests.
2140 Frequency : rare, only with NFS export
2142 Description: Crash on NFS re-export node
2143 Details : under very unusual load conditions an assertion is hit in
2144 ll_intent_file_open()
2147 Frequency : only if OST filesystem is corrupted
2149 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2150 Details : for a short time RPCs with bulk IO are in the replay list,
2151 but replay of bulk IOs is unimplemented. If the OST filesystem
2152 is corrupted due to disk cache incoherency and then replay is
2153 started it is possible to trip an assertion. Avoid putting
2154 committed RPCs into the replay list at all to avoid this issue.
2159 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2160 Details : Large single O_DIRECT read and write calls can fail to allocate
2161 a sufficiently large buffer to process the request. In case of
2162 allocation failure the allocation is retried with a smaller
2163 buffer and broken into smaller requests.
2165 Severity : enhancement
2167 Description: Add -o localflock option to simulate old noflock behaviour.
2168 Details : This will achieve local-only flock/fcntl locks coherentness.
2173 Description: versioning check is incomplete
2174 Details : Checking the version difference of client vs. server, report
2175 error if the gap is too big.
2180 Description: add support PG_writeback bit
2181 Details : add support for PG_writeback bit for Lustre, for more carefull
2182 work with page cache in 2.6 kernel. This also fix some deadlocks
2183 and remove hack for work O_SYNC with 2.6 kernel.
2185 Severity : enhancement
2187 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2188 Details : The uninit_groups feature works in conjunction with the kernel
2189 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2190 up the pass1 processing of e2fsck. This is a read-only feature
2191 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2192 mount filesystems that have had this feature enabled.
2194 Severity : enhancement
2196 Description: Improve multi-block allocation algorithm to avoid fragmentation
2197 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2198 allocation locality and avoid filesystem fragmentation.
2200 ------------------------------------------------------------------------------
2202 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2204 * Support for kernels:
2205 2.6.9-42.0.3.EL (RHEL 4)
2206 2.6.5-7.276 (SLES 9)
2207 2.4.21-47.0.1.EL (RHEL 3)
2208 2.6.12.6 vanilla (kernel.org)
2209 2.6.16.21-0.8 (SLES10)
2210 * Recommended e2fsprogs version: 1.39.cfs2-0
2212 * The backwards-compatible /proc/sys/portals symlink has been removed
2213 in this release. Before upgrading, please ensure that you change
2214 any configuration scripts or /etc/sysctl.conf files that access
2215 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2216 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2217 in advance of the upgrade on any system running Lustre 1.4.6 or
2218 newer, since /proc/sys/lnet was added in that version.
2219 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2223 Frequency : only when quota is used
2225 Description: avoid scanning export list for quota master
2226 Details : Change the algorithms to avoid scanning export list in order
2227 to improve the efficiency.
2230 Frequency : MDS failover only, very rarely
2232 Description: "went back in time" messages on mds failover
2233 Details : The greatest transno may be lost when the current operation
2234 finishes with an error (transno==0) and the client's last_rcvd
2235 record is over-written. Save the greatest transno in the
2236 mds_last_transno for this case.
2239 Frequency : always for specific kernels and striping counts
2241 Description: client may get "Matching packet too big" without ACL support
2242 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2243 when trying to access files in certain configurations. The
2244 clients should in fact be denied when mounting because they do
2245 not understand ACLs.
2248 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2250 Description: many clients connecting with IO in progress causes connect timeouts
2251 Details : Avoid synchronous journal commits to avoid delays caused by many
2252 clients connecting/disconnecting when bulk IO is in progress.
2253 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2254 OST_IO_PORTAL to avoid delays behind potentially many pending
2258 Frequency : occasionally with multiple writers to a single file
2260 Description: shared writes to file may result in wrong size reported by stat()
2261 Details : Allow growing of kms when extent lock is cancelled
2264 Frequency : always with random mmap IO to multi-striped file
2266 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2267 Details : Only if the hole is at the end of OST object so that kms is too
2268 small. Fix is to increase kms accordingly in ll_nopage.
2271 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2273 Description: writes to a missing object would leak memory on the OST
2274 Details : If there is an inconsistency between the MDS and OST filesystems,
2275 such that the MDS references an object that doesn't exist, writes
2276 to that object will leak memory due to incorrect cleanup in the
2277 error handling path, eventually running out of memory on the OST.
2282 Description: Creating too long symlink causes lustre errors
2283 Details : Check symlink and name lengths before sending requests to MDS.
2286 Frequency : only if flock is enabled (not on by default)
2288 Description: posix locks not released on fd closure on 2.6.9+
2289 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2290 kernels, this caused such locks not to be released on fd close and
2291 then assertions on fs unmount about still used locks.
2294 Frequency : MDS failover only, very rarely
2296 Description: clients may get ASSERTION(granted_lock != NULL)
2297 Details : When request was taking a long time, and a client was resending
2298 a getattr by name lock request. The were multiple lock requests
2299 with the same client lock handle and
2300 mds_getattr_name->fixup_handle_for_resent_request found one of the
2301 lock handles but later failed with ASSERTION(granted_lock != NULL).
2306 Description: handle->h_buffer_credits > 0, assertion failure
2307 Details : h_buffer_credits is zero after truncate, causing assertion
2308 failure. This patch extends the transaction or creates a new
2312 Frequency : NFS re-export or patchless client
2313 Bugzilla : 11179, 10796
2314 Description: Crash on NFS re-export node (__d_move)
2315 Details : We do not want to hash the dentry if we don't have a lock.
2316 But if this dentry is later used in d_move, we'd hit uninitialised
2317 list head d_hash, so we just do this to init d_hash field but
2318 leave dentry unhashed.
2321 Frequency : NFS re-export or patchless client
2323 Description: NFS exports has problem with symbolic link
2324 Details : lustre client didn't properly install dentry when re-exported
2325 to NFS or running patchless client.
2328 Frequency : NFS re-export or patchless client
2330 Description: Various nfs/patchless fixes.
2331 Details : fixes reuse disconected alias for lookup process - this fixes
2332 warning "find_exported_dentry: npd != pd",
2333 fix permission error with open files at nfs.
2334 fix apply umask when do revalidate.
2337 Frequency : occasional
2339 Description: Crash on NFS re-export node
2340 Details : calling clear_page() on the wrong pointer triggered oops in
2341 generic_mapping_read().
2344 Frequency : rarely, using O_DIRECT IO
2346 Description: unaligned directio crashes client with LASSERT
2347 Details : check for unaligned buffers before trying any requests.
2350 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2352 Description: stale data returned from RAID cache
2353 Details : If only a small amount of IO is done to the RAID device before
2354 reading it again it is possible to get stale data from the RAID
2355 cache instead of reading it from disk.
2358 Frequency : always for sles10 kernel
2360 Description: sles10 support
2361 Details : ll_follow_link: compile fixes and using of nd_set_link
2362 under newer kernels.
2365 Frequency : depends on arch, kernel and compiler version, always on sles10
2368 Description: recursive or deep enough symlinks cause stack overflow
2369 Details : getting rid of large stack-allocated variable in
2373 Frequency : depends on hardware
2375 Description: lustre write performance loss in the SLES10 kernel
2376 Details : the performance loss is caused by using of write barriers in the
2377 ext3 code. The SLES10 kernel turns barrier support on by
2378 default. The fix is to undo that change for ldiskfs.
2380 ------------------------------------------------------------------------------
2382 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2384 * Support for kernels:
2385 2.6.9-42.0.3EL (RHEL 4)
2386 2.6.5-7.276 (SLES 9)
2387 2.4.21-47.0.1.EL (RHEL 3)
2388 2.6.12.6 vanilla (kernel.org)
2392 Frequency : quota enabled and large files being deleted
2394 Description: releasing more than 4GB of quota at once hangs OST
2395 Details : If a user deletes more than 4GB of files on a single OST it
2396 will cause the OST to spin in an infinite loop. Release
2397 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2402 Description: statfs data retrieved from /proc may be stale or zero
2403 Details : When reading per-device statfs data from /proc, in the
2404 {kbytes,files}_{total,free,avail} files, it may appear
2405 as zero or be out of date.
2408 Frequency : systems with MD RAID1 external journal devices
2410 Description: lconf's call to blkid is confused by RAID1 journal devices
2411 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2412 of returning all block devices that match the journal UUID.
2415 Frequency : always, for aggregate stripe size over 4GB
2417 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2418 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2419 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2422 Frequency : always if "lfs find" used on a local file/directory
2424 Description: "lfs find" segfaults if used on a local file/directory
2425 Details : The case where a directory component was not specified wasn't
2426 handled correctly. Handle this properly.
2429 Frequency : always on ppc64
2431 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2432 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2433 perform endianness conversion.
2436 Frequency : rarely (truncate to non-zero file size after write under load)
2437 Bugzilla : 10730, 10687
2438 Description: Files padded with zeros to next 4K multiple
2439 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2440 it is possible that files that are truncated to a non-zero size
2441 immediately after being written are filled with zero bytes beyond
2442 the truncated size. No file data is lost.
2444 Severity : enhancement
2446 Description: Allow recovery/failover for liblustre clients.
2447 Details : liblustre clients were unaware of failover configurations until
2450 Severity : enhancement
2452 Description: user file locks should fail when not mounting with flock option
2453 Details : Set up an error-returning stub in ll_file_operations.lock field
2454 to prevent incorrect behaviour when client is mounted without
2455 flock option. Also, set up properly f_op->flock field for
2459 Frequency : always on ia64
2461 Description: "lfs df" loops on printing out MDS statfs information
2462 Details : The obd_ioctl_data was not initialized and in some systems
2463 this caused a failure during the ioctl that did not return
2464 an error. Initialize the struct and return an error on failure.
2467 Frequency : SLES 9 only
2469 Description: Error of copying files with lustre special EAs as root
2470 Details : Client side always return success for setxattr call for lustre
2471 special xattr (currently only "trusted.lov").
2474 Frequency : rarely on clusters with both ia64+i386 clients
2476 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2477 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2478 clients are doing concurrent writes to the same file it is
2479 possible that the ia64 clients may LASSERT because the OST
2480 extent locks are not PAGE_SIZE aligned. Ensure that grown
2481 locks are always aligned on the request boundary.
2484 Frequency : specific use, occasional
2486 Description: Overwriting in use executable truncates on-disk binary image
2487 Details : If one node attempts to overwrite an executable in use by
2488 another node, we now correctly return ETXTBSY instead of
2489 truncating the file.
2491 Severity : enhancement
2493 Description: Async OSC create to avoid the blocking unnecessarily.
2494 Details : If a OST has no remain object, system will block on the creating
2495 when need to create a new object on this OST. Now, ways use
2496 pre-created objects when available, instead of blocking on an
2497 empty osc while others are not empty. If we must block, we block
2498 for the shortest possible period of time.
2503 Description: chmod on Lustre root is propagated to other clients
2504 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2505 invalid by the follow_mount time.
2508 Frequency : liblustre clients only
2510 Description: Race in 'instant cancel' lock handling could lead to such locks
2511 never to be granted in case of SMP MDS
2512 Details : Do not destroy not yet granted but cbpending locks in
2516 Frequency : replay/resend of open
2518 Description: non null lock assetion failure in mds_intent_policy
2519 Details : Trying to replay/resend lockless open requests resulted in
2520 mds_open() returning 0 with no lock. Now it sets a flag if
2521 a lock is going to be returned.
2523 Severity : enhancement
2525 Description: Checksum enhancements
2526 Details : New checksum enhancements allow for resending RPCs that failed
2529 Severity : enhancement
2531 Description: Tunables on number of dirty pages in cacche
2532 Details : Allow to set limit on number of dirty pages cached.
2537 Description: client crash on unmount - lock still has references
2538 Details : In some error handling cases it was possible to leak a lock
2539 reference on a client while accessing a file. This was not
2540 harmful to the client during operation, but would cause the
2541 client to crash when the filesystem is unmounted.
2544 Frequency : specific case, rare
2546 Description: ETXTBSY on mds though file not in use by client
2547 Details : ETXTBSY is no longer incorrectly returned when attempting to
2548 chmod or chown a directory that the user previously tried to
2549 execute or a currently-executing binary.
2552 Frequency : extremely rare except on liblustre-based clients
2554 Description: Lustre space not freed when files are deleted
2555 Details : Clean up open-unlinked files after client eviction. Previously
2556 the unlink was skipped and the files remained as orphans.
2561 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2562 Details : In some cases it was possible to send a blocking callback to
2563 a client doing a glimpse, even though that client didn't get
2564 a lock granted. When the glimpse lock is cancelled on the OST
2565 the freed lock is left on the waiting list and corrupted the list.
2568 Frequency : all core dumps
2570 Description: Broke core dumps to lustre
2571 Details : Negative dentry may be unhashed if parent does not have UPDATE
2572 lock, but some callers, e.g. do_coredump, expect dentry to be
2573 hashed after successful create, hash it in ll_create_it.
2575 ------------------------------------------------------------------------------
2577 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2579 * Support for kernels:
2580 2.6.9-42.0.2.EL (RHEL 4)
2581 2.6.5-7.276 (SLES 9)
2582 2.4.21-40.EL (RHEL 3)
2583 2.6.12.6 vanilla (kernel.org)
2587 Frequency : always on RHEL 3
2589 Description: Number of open files grows over time
2590 Details : The number of open files grows over time, whether or not
2591 Lustre is started. This was due to a filp leak introduced
2592 by one of our kernel patches.
2594 ------------------------------------------------------------------------------
2596 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2598 * Support for kernels:
2599 2.6.9-42.EL (RHEL 4)
2600 2.6.5-7.276 (SLES 9)
2601 2.4.21-40.EL (RHEL 3)
2602 2.6.12.6 vanilla (kernel.org)
2607 Bugzilla : 5719, 9635, 9792, 9684
2608 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2609 Details : If a server is under heavy load and cannot reply to new
2610 connection requests before the client resends the (re)connect,
2611 the connection handling code can behave badly if two service
2612 threads are concurrently handing separate (re)connections from
2613 the same client. Add better locking to the connection handling
2614 code, and ensure that only a single connection will be processed
2615 for a given client UUID, even if the lock is dropped.
2617 Severity : enhancement
2619 Description: add TCP zero-copy support to kernel
2620 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2621 sends if the hardware supports scatter-gather and checksumming.
2622 This allows socklnd to do client-write and server-read more
2623 efficiently and reduce CPU utilization from skbuf copying.
2626 Frequency : only if NFS exporting from client
2628 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2629 Details : If setting attributes on a file created under NFS that had
2630 never been opened it would be possible to oops the client
2631 if the file had no objects.
2634 Frequency : always for liblustre
2636 Description: liblustre client does MDS+OSTs setattr RPC for each write
2637 Details : When doing a write from a liblustre client, the client
2638 incorrectly issued an RPC to the MDS and each OST the file was
2639 striped over in order to update the timestamps. When writing
2640 with small chunks and many clients this could overwhelm the MDS
2641 with RPCs. In all cases it would slow down the write because
2642 these RPCs are unnecessary.
2644 Severity : enhancement
2646 Description: allow number of MDS service threads to be changed at module load
2647 Details : It is now possible to change the number of MDS service threads
2648 running. Adding "options mds mds_num_threads={N}" to the MDS's
2649 /etc/modprobe.conf will set the number of threads for the next
2650 time Lustre is restarted (assuming the "mds" module is also
2651 reloaded at that time). The default number of threads will
2652 stay the same, 32 for most systems.
2657 Description: OST crash if filesystem is unformatted or corrupt
2658 Details : If an OST is started on a device that has never been formatted
2659 or if the filesystem is corrupt and cannot even mount then the
2660 error handling cleanup routines would dereference a NULL pointer.
2665 Description: NULL pointer deref in llap_from_page.
2666 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2667 incorrect) mapping if the page was truncated/reclaimed while it was
2668 searched for. Check for this condition and skip such pages when
2669 doing readahead. Introduce extra check to llap_from_page() to
2670 verify page->mapping->host is non-NULL (so page is not anonymous).
2673 Frequency : Sometimes when using sys_sendfile
2675 Description: "page not covered by a lock" warnings from ll_readpage
2676 Details : sendfile called ll_readpage without right page locks present.
2677 Now we introduced ll_file_sendfile that does necessary locking
2678 around call to generic_file_sendfile() much like we do in
2682 Frequency : with certain MDS communication failures at client mount time
2684 Description: NULL pointer deref after failed client mount
2685 Details : a client connection request may delayed by the network layer
2686 and not be sent until after the PTLRPC layer has timed out the
2687 request. If the client fails the mount immediately it will try
2688 to clean up before the network times out the request. Add a
2689 reference from the request import to the obd device and delay
2690 the cleanup until the network drops the request.
2693 Frequency : occasionally during client (re)connect
2695 Description: assertion failure during client (re)connect
2696 Details : processing a client connection request may be delayed by the
2697 client or server longer than the client connect timeout. This
2698 causes the client to resend the connection request. If the
2699 original connection request is replied in this interval, the
2700 client may trip an assertion failure in ptlrpc_connect_interpret()
2701 which thought it would be the only running connect process.
2704 Frequency : only with obd_echo servers and clients that are rebooted
2706 Description: kernel BUG accessing uninitialized data structure
2707 Details : When running an obd_echo server it did not start the ping_evictor
2708 thread, and when a client was evicted an uninitialized data
2709 structure was accessed. Start the ping_evictor in the RPC
2710 service startup instead of the OBD startup.
2712 Severity : enhancement
2713 Bugzilla : 10193 (patchless)
2714 Description: Remove dependency on various unexported kernel interfaces.
2715 Details : No longer need reparent_to_init, exit_mm, exit_files,
2716 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2719 Frequency : rare (only users of deprecated and unsupported LDAP config)
2721 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2722 Details : LDAP apparently contains 'lustreName' attributes instead of
2723 'name'. A simple remapping of the name is sufficient.
2726 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2728 Description: waiting_locks_callback trips kernel BUG if client is evicted
2729 Details : Running with the dump_on_timeout debug flag turned on makes
2730 it possible that the waiting_locks_callback() can try to dump
2731 the Lustre kernel debug logs from an interrupt handler. Defer
2732 this log dumping to the expired_lock_main() thread.
2734 Severity : enhancement
2736 Description: Support NFS exporting on 2.6 kernels.
2737 Details : Implement non-rawops metadata methods for NFS server to use without
2738 changing NFS server code.
2741 Frequency : very rare (synthetic metadata workload only)
2743 Description: two racing renames might cause an MDS thread to deadlock
2744 Details : Running the "racer" program may cause one MDS thread to rename
2745 a file from being the source of a rename to being the target of
2746 a rename at exactly the same time that another thread is doing
2747 so, and the second thread has already enqueued these locks after
2748 doing a lookup of the target and is trying to relock them in
2749 order. Ensure that we don't try to re-lock the same resource.
2752 Frequency : only very large systems with liblustre clients
2754 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2755 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2756 all OSTs in parallel. This allows the eviction of stale liblustre
2757 clients to proceed much faster than if they were done in series,
2758 and also offers similar improvements for other set_info RPCs.
2763 Description: excessive CPU usage during initial read phase on client
2764 Details : During the initial read phase on a client, it would agressively
2765 retry readahead on the file, consuming too much CPU and impacting
2766 performance (since 1.4.5.8). Improve the readahead algorithm
2767 to avoid this, and also improve some other common cases (read
2768 of small files in particular, where "small" is files smaller than
2769 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2774 Description: MDS crash when receiving packet with unknown intent.
2775 Details : Do not LBUG in unknown intent case, just return -EFAULT
2777 Severity : enhancement
2778 Bugzilla : 9293, 9385
2779 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2780 Details : Do not serialize getattr (non-intent version) and statfs.
2783 Frequency : occasional, when OST network is overloaded/intermittent
2785 Description: client evicted by OST after bulk IO timeout
2786 Details : If a client sends a bulk IO request (read or write) the OST
2787 may evict the client if it is unresposive to its data GET/PUT
2788 request. This is incorrect if the network is overloaded (takes
2789 too long to transfer the RPC data) or dropped the OST GET/PUT
2790 request. There is no need to evict the client at all, since
2791 the pinger and/or lock callbacks will handle this, and the
2792 client can restart the bulk request.
2795 Frequency : Always when mmapping file with no objects
2797 Description: client crashes when mmapping file with no objects
2798 Details : Check that we actually have objects in a file before doing any
2799 operations on objects in ll_vm_open, ll_vm_close and
2805 Description: Request leak when working with deleted CWD
2806 Details : Introduce advanced request refcount tracking for requests
2807 referenced from lustre intent.
2809 Severity : Enhancement
2811 Description: Cache open file handles on client.
2812 Details : MDS now will return special lock along with openhandle, if
2813 requested and client is allowed to hold openhandle, even if unused,
2814 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2815 closing files for every read/write openration.
2817 Severity : Enhancement
2819 Description: Cache open negative dentries on client when possible.
2820 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2821 negative dentries on lock revocation.
2826 Description: Remounting a client read-only wasn't possible with a zconf mount
2827 Details : It wasn't possible to remount a client read-only with llmount.
2829 Severity : enhancement
2830 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2831 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2832 MPICH is included to add Lustre-specific ADIO interfaces.
2833 This is based closely on the UFS ADIO layer and only differs
2834 in file creation, in order to allow the OST striping to be set.
2835 This is user-contributed code and not supported by CFS.
2840 Description: extended inode attributes (immutable, append-only) work improperly
2841 when 2.4 and 2.6 kernels are used on client/server or vice versa
2842 Details : Introduce kernel-independent values for these flags.
2844 Severity : enhancement
2847 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2848 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2849 granularity. Now, Fractional MB granularity is be supported,
2850 this is very useful for low memory system.
2852 Severity : enhancement
2854 Description: Getattr by fid
2855 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2856 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2857 have locks and avoids getattr rpc requests for it on every stat.
2860 Frequency : Always, for filesystems larger than 2TB
2862 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2863 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2864 filesystems larger than 2TB to be mounted with mballoc enabled.
2867 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2869 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2870 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2871 filesystems larger than 2TB is mounted then the kernel will
2872 silently corrupt the start of the filesystem. CONFIG_LBD is
2873 enabled for all CFS-supported kernels, but the possibility of
2874 this happening with a modified kernel config exists.
2876 Severity : enhancement
2878 Description: add client O_DIRECT support for 2.6 kernels
2879 Details : It is now possible to do O_DIRECT reads and writes to files
2880 in the Lustre client mountpoint on 2.6 kernel clients.
2882 Severity : enhancement
2884 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2885 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2886 parallel, not waiting for response from every OST before sending
2887 a rpc to the next OST.
2892 Description: setattr vs write race when updating file timestamps
2893 Details : Client processes that update a file timestamp into the past
2894 right after writing to the file (e.g. tar) it is possible that
2895 the updated file modification time can be reset to the current
2896 time due to a race between processing the setattr and write RPC.
2898 Severity : enhancement
2900 Description: Bring 'lfs find' closer in line with regular Linux find.
2901 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2902 -print0 options and obtains all the needed info through the lustre
2905 Severity : enhancement
2907 Description: support up to 1024 configured devices on one node
2908 Details : change obd_dev array from statically allocated to dynamically
2909 allocated structs as they are first used to reduce memory usage
2914 Description: Flush dirty partially truncated pages during truncate
2915 Details : Immediatelly flush partially truncated pages in filter_setattr,
2916 this way we completely avoid having any pages in page cache on OST
2917 and can retire ugly workarounds during writes to flush such pages.
2922 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2923 Details : protect inode from truncation within vfs_unlink() context
2924 just take a reference before calling vfs_unlink() and release it
2925 when parent's i_sem is free.
2930 Description: last_id value checked outside lock on OST caused LASSERT failure
2931 Details : If there were multiple MDS->OST object precreate requests in
2932 flight, it was possible that the OST's last object id was checked
2933 outside a lock and incorrectly tripped an assertion. Move checks
2934 inside locks, and discard old precreate requests.
2937 Frequency : always, if extents are used on OSTs
2939 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2940 Details : OSTs using the extents format would not zero the high 16 bits of
2941 the index physical block number. This is not a problem for any
2942 OST filesystems smaller than 16TB, and no kernels support ext3
2943 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2944 new/modified files) and can be fixed for existing filesystems
2945 with e2fsprogs-1.39-cfs1.
2950 Description: import connection selection may be incorrect if timer wraps
2951 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2952 import connections to be ignored if the 32-bit jiffies counter
2953 wraps. Use a 64-bit jiffies counter.
2956 Frequency : very large clusters immediately after boot
2958 Description: LNET request buffers exhausted under heavy short-term load
2959 Details : If a large number of client requests are generated on a service
2960 that has previously never seen so many requests it is possible
2961 that the request buffer growth cannot keep up with the spike in
2962 demand. Instead of dropping incoming requests, they are held in
2963 the LND until the RPC service can accept more requests.
2966 Frequency : Sometimes during replay
2968 Description: Assertion failure in ll_local_open after replay.
2969 Details : If replay happened on an open request reply before we were able
2970 to set replay handler, reply will become not swabbed tripping the
2971 assertion in ll_local_open. Now we set the handler right after
2972 recognising of open request
2975 Frequency : very rare
2977 Description: kernel reports "badness in vsnprintf"
2978 Details : Reading from the "recovery_status" /proc file in small chunks
2979 may cause a negative length in lprocfs_obd_rd_recovery_status()
2980 call to vsnprintf() (which is otherwise harmless). Exit early
2981 if there is no more space in the output buffer.
2983 Severity : enhancement
2985 Description: clear OBD RPC statistics by writing to them
2986 Details : It is now possible to clear the OBD RPC statistics by writing
2987 to the "stats" file.
2992 Description: Client mtime is not the same on different clients after utimes
2993 Details : In some cases, the client was using the utimes() syscall on
2994 a file cached on another node. The clients now validate the
2995 ctime from the MDS + OSTs to determine which one is right.
3000 Description: Inability to activate failout mode
3001 Details : lconf script incorrectly assumed that in pythong string's numeric
3002 value is used in comparisons.
3005 Frequency : always with multiple stripes per file
3007 Description: Inefficient object allocation for mutli-stripe files
3008 Details : When selecting which OSTs to stripe files over, for files with
3009 a stripe count that divides evenly into the number of OSTs,
3010 the MDS is always picking the same starting OST for each file.
3011 Return the OST selection heuristic to the original design.
3016 Description: mount failures may take full timeout to return an error
3017 Details : Under some heavy load conditions it is possible that a
3018 failed mount can wait for the full obd_timeout interval,
3019 possibly several minutes, before reporting an error.
3020 Instead return an error as soon as the status is known.
3022 Frequency : quota enabled and large files being deleted
3024 Description: releasing more than 4GB of quota at once hangs OST
3025 Details : If a user deletes more than 4GB of files on a single OST it
3026 will cause the OST to spin in an infinite loop. Release
3027 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3032 Description: statfs data retrieved from /proc may be stale or zero
3033 Details : When reading per-device statfs data from /proc, in the
3034 {kbytes,files}_{total,free,avail} files, it may appear
3035 as zero or be out of date.
3038 Frequency : systems with MD RAID1 external journal devices
3040 Description: lconf's call to blkid is confused by RAID1 journal devices
3041 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3042 of returning all block devices that match the journal UUID.
3045 Frequency : always, for aggregate stripe size over 4GB
3047 Description: assertion fails when trying to use 4GB stripe size
3048 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
3049 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
3052 Frequency : always on ppc64
3054 Description: the first write on an ext3 filesystem with mballoc got stuck
3055 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3056 perform endianness conversion.
3058 ------------------------------------------------------------------------------
3060 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3062 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3063 INTEROPERATE with older versions automatically. Please read the
3064 user documentation before upgrading any part of a live system.
3065 * WARNING: Lustre networking configuration changes are required with
3066 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3069 * Support for kernels:
3070 2.6.9-22.0.2.EL (RHEL 4)
3071 2.6.5-7.244 (SLES 9)
3072 2.6.12.6 vanilla (kernel.org)
3075 Severity : enhancement
3076 Bugzilla : 7981/8208
3077 Description: Introduced Lustre Networking (LNET)
3078 Details : LNET is new networking infrastructure for Lustre, it includes
3079 a reorganized network configuration mode (see the user
3080 documentation for full details) as well as support for routing
3081 between different network fabrics. Lustre Networking Devices
3082 (LNDs) for the supported network fabrics have also been
3083 created for this new infrastructure.
3085 Severity : enhancement
3086 Description: Introduced Access control lists
3087 Details : clients can set ACLs on files and directories in order to have
3088 more fine-grained permissions than the standard Unix UGO+RWX.
3089 The MDS must be started with the "-o acl" mount option.
3091 Severity : enhancement
3092 Description: Introduced filesystem quotas
3093 Details : Administrators may now establish per-user quotas on the
3096 Severity : enhancement
3098 Description: Configuration change for the XT3
3099 The PTLLND is now used to run Lustre over Portals on the XT3
3100 The configure option(s) --with-cray-portals are no longer used.
3101 Rather --with-portals=<path-to-portals-includes> is used to
3102 enable building on the XT3. In addition to enable XT3 specific
3103 features the option --enable-cray-xt3 must be used.
3108 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3109 Details : A race between allocating a new llcd and re-getting the llcd_lock
3110 allowed another thread to grab newly-allocated llcd.
3112 Severity : enhancement
3114 Description: 2.6 OST async journal commit and locking fix to improve performance
3115 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3116 2.6 kernels are now async as they already were in 2.4 kernels so
3117 that they can commit concurrently with the network bulk transfer.
3118 For block-allocated files the filter allocation semaphore is held
3119 to avoid filesystem fragmentation during allocation. BKL lock
3120 removed for 2.6 xattr operations where it is no longer needed.
3125 Description: lconf incorrectly determined whether two IP networks could talk
3126 Details : In some more complicated routing and multiple-network
3127 configurations, lconf will avoid trying to make a network
3128 connection to a disjoint part of the IP space. It was doing the
3129 math incorrectly for one set of cases.
3134 Description: Fix for potential infinite loop processing records in an llog.
3135 Details : If an llog record is corrupted/zeroed, it is possible to loop
3136 forever in llog_process(). Validate the llog record length
3137 and skip the remainder of the block on error.
3140 Frequency : occasional (liblustre only)
3142 Description: liblustre could not open files whose last component is a symlink
3143 Details : sysio_path_walk() would incorrectly pass the open intent to
3144 intermediate path components.
3147 Frequency : rare (liblustre only with non-standard tuning)
3148 Bugzilla : 7201 (7350)
3149 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3150 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3151 after it has been released, only do this for referenced locks
3153 Severity : enhancement
3155 Description: specifying an (invalid) directory default stripe_size of -1
3156 would reset the directory default striping
3157 Details : stripe_size -1 was used internally to signal directory stripe
3158 removal, now use "all default" to signal dir stripe removal
3159 as a directory striping of "all default" is not useful
3162 Frequency : common for large clusters running liblustre clients
3164 Description: doing an ls when liblustre clients are running is slow
3165 Details : sending a glimpse AST to a liblustre client waits for every AST
3166 to time out, as liblustre clients will not respond. Since they
3167 cannot cache data we refresh the OST lock LVB from disk instead.
3169 Severity : enhancement
3171 Description: doing an ls at the same time as file IO can be slow
3172 Details : enqueue and other "small" requests can be blocked behind many
3173 large IO requests. Create a new OST IO portal for non-IO
3174 requests so they can be processed faster.
3177 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3179 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3180 created file with mode 000
3181 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3182 allow an ftruncate() as a non-root user to a file with mode 000.
3183 Lustre now respects this flag to disable mode checks when
3184 truncating a file owned by the user
3187 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3190 Description: Revoking locks from clients that went dead or catatonic might take
3192 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3193 cancellation of such locks instant on servers without waiting for
3194 any reply from clients. Clients drops these locks when cancel
3195 notification from server is received without replying.
3198 Frequency : liblustre-only, when liblustre client dies or becomes busy
3200 Description: Doing ls on Linux clients can take a long time with active
3202 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3203 granting such locks to it in the first place if possible. Locks
3204 are taken by proxy on the OST during the read or write and
3205 dropped immediately afterward. Add connect flags handling, do
3206 not grant locks to liblustre clients for glimpse ASTs.
3208 Severity : enhancement
3210 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3211 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3212 sequential reads and behaves badly with applications doing random
3213 reads. Improve it by reading ahead at least read region, and
3214 avoiding excessive large RPC for small reads.
3216 Severity : enhancement
3218 Description: Creating more than 1000 files for a single job may cause a load
3219 imbalance on the OSTs if there are also a large number of OSTs.
3220 Details : qos_prep_create() uses an OST index reseed value that is an
3221 even multiple of the number of available OSTs so that if the
3222 reseed happens in the middle of the object allocation it will
3223 still utilize the OSTs as uniformly as possible.
3228 Description: OST or MDS may oops in ping_evictor_main()
3229 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3230 but doesn't restart at beginning of obd_exports_timed list
3233 Severity : enhancement
3235 Description: improve by-nid export eviction on the MDS and OST
3236 Details : allow multiple exports with the same NID to be evicted at one
3237 time without re-searching the exports list.
3240 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3242 Description: MDS may oops in groups_free()
3243 Details : in rare race conditions a newly allocated group_info struct is
3244 freed again, and this can be NULL. The 2.4 compatibility code
3245 for groups_free() checked for a NULL pointer, but 2.6 did not.
3248 Frequency : common for liblustre clients doing little filesystem IO
3249 Bugzilla : 9352, 7313
3250 Description: server may evict liblustre clients accessing contended locks
3251 Details : if a client is granted a lock or receives a completion AST
3252 with a blocking AST already set it would not reply to the AST
3253 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3256 Frequency : lfs setstripe, only systems with more than 160 OSTs
3258 Description: unable to set striping with a starting offset beyond OST 160
3259 Details : llapi_create_file() incorrectly limited the starting stripe
3260 index to the maximum single-file stripe count.
3263 Frequency : LDAP users only
3265 Description: lconf did not handle in-kernel recovery with LDAP properly
3266 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3268 Severity : enhancement
3270 Description: bind OST threads to NUMA nodes to improve performance
3271 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3272 node and do their allocations there to localize memory access
3274 Severity : enhancement
3276 Description: llmount can determine client NID directly from Myrinet (GM)
3277 Details : the client NID code from gmnalnid was moved directly into
3278 llmount, removing the need to use this or specifying the
3279 client NID explicitly when mounting GM clients with zeroconf
3282 Frequency : if client is started with down MDS
3284 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3285 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3286 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3287 ptlrpc_import_delayed_req() because we didn't check if the
3288 request was interrupted, and we also didn't break out of the
3289 event loop if there was no timeout
3294 Description: data loss during non-page-aligned writes to a single file from
3295 both multiple nodes and multiple threads on one node at same time
3296 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3297 inconsistency led to false short-reads, that were cached and later
3298 used by ->prepare_write() to fill in partially written page,
3299 leading to data loss.
3302 Frequency : always, if lconf --abort_recovery used
3304 Description: lconf --abort_recovery fails with 'Operation not supported'
3305 Details : lconf was attempting to abort recovery on the MDT device and not
3308 Severity : enhancement
3310 Description: remove cleanup logs
3311 Details : replace lconf-generated cleanup logs with lustre internal
3312 cleanup routines. Eliminates the need for client-cleanup and
3315 Severity : enhancement
3317 Description: add support for EAs (user and system) on lustre filesystems
3318 Details : it is now possible to store extended attributes in the Lustre
3319 client filesystem, and with the user_xattr mount option it
3320 is possible to allow users to store EAs on their files also
3322 Severity : enhancement
3324 Description: Add possibility (config option) to show minimal available OST free
3326 Details : When compiled with --enable-mindf configure option, statfs(2)
3327 (and so, df) will return least minimal free space available from
3328 all OSTs as amount of free space on FS, instead of summary of
3329 free spaces of all OSTs.
3331 Severity : enhancement
3333 Description: do not expand extent locks acquired on OST-side
3334 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3335 by server: they are not cached anyway.
3338 Frequency : when mmap is used/binaries executed from Lustre
3340 Description: Unmmap pages before throwing them away from read cache.
3341 Details : llap_shrink cache now attempts to unmap pages before discarding
3342 them (if unmapping failed - do not discard). SLES9 kernel has
3343 extra checks that trigger if this unmapping is not done first.
3348 Description: lconf didn't resolve symlinks before checking to see whether a
3349 given mountpoint was already in use
3352 Frequency : when migrating failover services
3353 Bugzilla : 6395, 9514
3354 Description: When migrating a subset of services from a node (e.g. failback
3355 from a failover service node) the remaining services would
3356 time out and evict clients.
3357 Details : lconf --force (implied by --failover) sets the global obd_timeout
3358 to 5 seconds in order to quickly disconnect, but this caused
3359 other RPCs to time out too quickly. Do not change the global
3360 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3362 Severity : enhancement
3363 Frequency : if MDS is started with down OST
3364 Bugzilla : 9439,5706
3365 Description: Allow startup/shutdown of an MDS without depending on the
3366 availability of the OSTs.
3367 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3368 Add appropriate locking and lov-osc refcounts for safe
3369 cleaning. Add osc abort_inflight calls in case the
3370 synchronize never started.
3373 Frequency : occasional (Cray XT3 only)
3375 Description: root not authorized to access files in CRAY_PORTALS environment
3376 Details : The client process capabilities were not honoured on the MDS in
3377 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3378 been accessed by an authorized user then root was able to access
3379 the file on the local client also. The root user capabilities
3380 are now allowed on the MDS, as this environment has secure UID.
3383 Frequency : occasional
3385 Description: ldiskfs "too long searching" message happens too often
3386 Details : A debugging message (otherwise harmless) prints too often on
3387 the OST console. This has been reduced to only happen when
3388 there are fragmentation problems on the filesystem.
3393 Description: Division by zero in statfs when all OSCs are inactive
3394 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3395 lov_statfs divides by value returned by lov_get_stripecnt().
3399 Bugzilla : 9489, 3273
3400 Description: First write from each client to each OST was only 4kB in size,
3401 to initialize client writeback cache, which caused sub-optimal
3402 RPCs and poor layout on disk for the first writen file.
3403 Details : Clients now request an initial cache grant at (re)connect time
3404 and so that they can start streaming writes to the cache right
3405 away and always do full-sized RPCs if there is enough data.
3406 If the OST is rebooted the client also re-establishes its grant
3407 so that client cached writes will be honoured under the grant.
3412 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3413 Details : Now I/O RPCs go to different portal number and (presumably) fast
3414 lock requests (and glimses) and other RPCs get their own service
3415 threads pool that should be able to service those RPCs
3418 Severity : enhancement
3420 Description: Ability to exchange lustre version between client and servers and
3421 issue warnings at client side if client is too old. Also for
3422 liblustre clients there is ability to refuse connection of too old
3424 Details : New 'version' field is added to connect data structure that is
3425 filled with version info. That info is later checked by server and
3429 Frequency : rare, liblustre only.
3430 Bugzilla : 9296, 9581
3431 Description: Two simultaneous writes from liblustre at offset within same page
3432 might proceed at the same time overwriting eachother with stale
3434 Details : I/O lock withing llu_file_prwv was released too early, before data
3435 actually was hitting the wire. Extended lock-holding time until
3436 server acknowledges receiving data.
3439 Frequency : extremely rare. Never observed in practice.
3441 Description: avoid generating lustre_handle cookie of 0.
3442 Details : class_handle_hash() generates handle cookies by incrementing
3443 global counter, and can hit 0 occasionaly (this is unlikely, but
3444 not impossible, because initial value of cookie counter is
3445 selected randonly). Value of 0 is used as a sentinel meaning
3446 "unassigned handle" --- avoid it. Also coalesce two critical
3447 sections in this function into one.
3449 Severity : enhancement
3451 Description: allow liblustre clients to delegate truncate locking to OST
3452 Details : To avoid overhead of locking, liblustre client instructs OST to
3453 take extent lock in ost_punch() on client's behalf. New connection
3454 flag is added to handle backward compatibility.
3456 Severity : enhancement
3457 Bugzilla : 4928, 7341, 9758
3458 Description: allow number of OST service threads to be specified
3459 Details : a module parameter allows the number of OST service threads
3460 to be specified via "options ost ost_num_threads={N}" in the
3461 OSS's /etc/modules.conf or /etc/modprobe.conf.
3465 Bugzilla : 6146, 9635, 9895
3466 Description: servers crash with bad pointer in target_handle_connect()
3467 Details : In rare cases when a client is reconnecting it was possible that
3468 the connection request was the last reference for that export.
3469 We would temporarily drop the export reference and get a new
3470 one, but this may have been the last reference and the export
3471 was just destroyed. Get new reference before dropping old one.
3473 Severity : enhancement
3474 Frequency : if client is started with failover MDS
3476 Description: Allow multiple MDS hostnames in the mount command
3477 Details : Try to read the configuration from all specified MDS
3478 hostnames during a client mount in case the "primary"
3481 Severity : enhancement
3483 Description: Stop sending data to evicted clients as soon as possible.
3484 Details : Check if the client we are about to send or are sending data to
3485 was evicted already. (Check is done every second of waiting,
3486 for which l_wait_event interface was extended to allow checking
3487 of exit condition at specified intervals).
3490 Frequency : rare, normally only when NFS exporting is done from client
3492 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3494 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3497 Frequency : always, if srand() is called before liblustre initialization
3499 Description: Liblustre uses system PRNG disturbing its usage by user application
3500 Details : Introduce internal to lustre fast and high-quality PRNG for
3501 lustre usage and make liblustre and some other places in generic
3502 lustre code to use it.
3504 Severity : enhancement
3505 Bugzilla : 9477, 9557, 9870
3506 Description: Verify that the MDS configuration logs are updated when xml is
3507 Details : Check if the .xml configuration logs are newer than the config
3508 logs stored on the MDS and report an error if this is the case.
3509 Request --write-conf, or allow starting with --old_conf.
3511 Severity : enhancement
3513 Description: Handle symlinks in the path when checking if Lustre is mounted.
3514 Details : Resolve intermediate symlinks when checking if a client has
3515 mounted a filesystem to avoid duplicate client mounts.
3520 Description: lconf can hit an error exception but still return success.
3521 Details : The lconf command catches the Command error exception at the top
3522 level script context and will exit with the associated exit
3523 status, but doesn't ensure that this exit status is non-zero.
3528 Description: failure of ptlrpc thread startup can cause oops
3529 Details : Starting a ptlrpc service thread can fail if there are a large
3530 number of threads or the server memory is very fragmented.
3531 Handle this without oopsing.
3534 Frequency : always, only if liblustre and non-default acceptor port was used
3536 Description: liblustre cannot connect to servers with non-default acceptor port
3537 Details : tcpnal_set_default_params() was not called and was therefore
3538 ignoring the environment varaible TCPNAL_PORT, as well as other
3539 TCPNAL_ environment variables
3544 Description: two objects could be created on the same OST for a single file
3545 Details : If an OST is down, in some cases it was possible to create two
3546 objects on a single OST for a single file. No problems other
3547 than potential performance impact and spurious error messages.
3551 Bugzilla : 5681, 9562
3552 Description: Client may oops in ll_unhash_aliases
3553 Details : Client dcache may become inconsistent in race condition.
3554 In some cases "getcwd" can fail if the current directory is
3560 Description: Inode refcounting problems in NFS export code
3561 Details : link_raw functions used to call d_instantiate without obtaining
3562 extra inode reference first.
3566 Bugzilla : 9942, 9903
3567 Description: Referencing freed requests leading to crash, memleaks with NFS.
3568 Details : We used to require that call to ll_revalidate_it was always
3569 followed by ll_lookup_it. Also with revalidate_special() it is
3570 possible to call ll_revalidate_it() twice for the same dentry
3571 even if first occurence returned success. This fix changes semantic
3572 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3573 reference on a request referred from the intent.
3574 ll_intent_release() then releases such a request.
3577 Frequency : rare, normally benchmark loads only
3579 Description: unlinked inodes were kept in memory on the client
3580 Details : If a client is repeatedly creating and unlinking files it
3581 can accumulate a lot of stale inodes in the inode slab cache.
3582 If there is no other client load running this can cause the
3583 client node to run out of memory. Instead flush old inodes
3584 from client cache that have the same inode number as a new inode.
3587 Frequency : SLES9 2.6.5 kernel and long filenames only
3588 Bugzilla : 9969, 10379
3589 Description: utime reports stale NFS file handle
3590 Details : SLES9 uses out-of-dentry names in some cases, which confused
3591 the lustre dentry revalidation. Change it to always use the
3595 Frequency : rare, unless heavy write-truncate concurrency is continuous
3596 Bugzilla : 4180, 6984, 7171, 9963, 9331
3597 Description: OST becomes very slow and/or deadlocked during object unlink
3598 Details : filter_destroy() was holding onto the parent directory lock
3599 while truncating+unlinking objects. For very large objects this
3600 may block other threads for a long time and slow overall OST
3601 responsiveness. It may also be possible to get a lock ordering
3602 deadlock in this case, or run out of journal credits because of
3603 the combined truncate+unlink. Solution is to do object truncate
3604 first in one transaction without parent lock, and then do the
3605 final unlink in a new transaction with the parent lock. This
3606 reduces the lock hold time dramatically.
3609 Frequency : rare, 2.4 kernels only
3611 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3612 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3613 before unmounting their filesystem in order to avoid deadlock.
3614 The kernel_locked() function in 2.4 kernels only checks whether
3615 the kernel lock is held, not whether it is this process that is
3616 holding it as 2.6 kernels do.
3621 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3622 Details : The client ptlrpc code may be trying to reconnect to a down
3623 server before a previous connection attempt has timed out.
3624 Increase the reconnect interval to be longer than the connection
3625 timeout interval to avoid sending duplicate connections to
3629 Frequency : echo_client brw_test command
3631 Description: fix echo_client to work with OST preallocated code
3632 Details : OST preallocation code (5137) didn't take echo_client IO path
3633 into account: echo_client calls filter methods outside of any
3634 OST thread and, hence, there is no per-thread preallocated
3635 pages and buffers to use. Solution: hijack pga pages for IO. As
3636 a byproduct, this avoids unnecessary data copying.
3640 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3641 Description: Client can oops in mdc_commit_close() after open replay
3642 Details : It was possible for the MDS to return an open request with no
3643 transaction number in mds_finish_transno() if the client was
3644 evicted, but without actually returning an error. Clients
3645 would later try to replay that open and may trip an assertion
3646 Simplify the client close codepath, and always return an error
3647 from the MDS in case the open is not successful.
3650 Frequency : rare, 2.6 OSTs only
3652 Description: OST may deadlock under high load on fragmented files
3653 Details : If there was a heavy load and highly-fragmented OST filesystems
3654 it was possible to have all the OST threads deadlock waiting on
3655 allocation of biovecs, because the biovecs were not released
3656 until the entire RPC IO was completed. Instead, release biovecs
3657 as soon as they are complete to ensure forward IO progress.
3659 Severity : enhancement
3661 Description: Support for specifying external journal device at mount
3662 Details : If an OST or MDS device is formatted with an external journal
3663 device, this device major/minor is stored in the ext3 superblock
3664 and may not be valid for failover. Allow detecting and
3665 specifying the external journal at mount time.
3670 Description: Mounting an MDS with pending unlinked files may cause oops
3671 Details : target_finish_recovery() calls mds_postrecov() which returned
3672 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3673 considers this an error and immediately begins cleaning up the
3674 lov, just after starting the mds_lov process
3676 Severity : enhancement
3678 Description: Implement 'lfs df' to report actual free space on per-OST basis
3679 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3680 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3681 usage of objects; '-h' to report in human readable format.
3683 ------------------------------------------------------------------------------
3685 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3692 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3693 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3694 at startup during orphan unlinking. Instead, skip buddy update
3695 before setup, as it will be regenerated after recovery is complete.
3700 Description: If an OST is inactive, its locks might reference stale inodes.
3701 Details : lov_change_cbdata() must iterate over all namespaces, even if
3702 they are inactive to clear inode references from the lock.
3704 Severity : enhancement
3705 Frequency : occasional, if non-standard max_dirty_mb used
3707 Description: Client will block write RPCs if not enough grant
3708 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3709 then the client will block writes while waiting for another RPC
3710 to complete instead of consuming its dirty limit. With change
3711 we get improved performance when max_dirty_mb is small.
3713 Severity : enhancement
3714 Bugzilla : 3389, 6253
3715 Description: Add support for supplementary groups on the MDS.
3716 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3717 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3718 lookups for user supplementary groups into a cache.
3722 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3723 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3724 for liblustre O_EXCL is enforced by MDS.
3728 Description: readdir never returns NULL in liblustre.
3729 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3730 dentry in struct dirent.
3734 Frequency : liblustre only, depends on application IO pattern
3735 Description: liblustre clients evicted if not contacting servers
3736 Details : Don't put liblustre clients into the ping_evictor list, so
3737 they will not be evicted by the pinger ever.
3739 Severity : enhancement
3741 Description: Add ability to evict clients by NID from MDS.
3742 Details : By echoing "nid:$NID" string into
3743 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3744 $NID would be instantly evicted from this MDS and from all active
3745 OSTs connected to it.
3749 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3750 Details : lookup_it_finish() used to query file size from OSTs that was not
3755 Description: service threads change working directory to that of init
3756 Details : Starting lustre service threads may pin the working directory
3757 of the parent thread, making that filesystem busy. Threads
3758 now change to the working directory of init to avoid this.
3762 Frequency : during shutdown only
3763 Description: shutdown with a failed MDS or OST can cause unmount to hang
3764 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3769 Frequency : 2.6 only
3770 Description: chmod/chown may include an extra supplementary group
3771 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3772 supplementary group and if none is specified this is used.
3775 Bugzilla : 5479 (6816)
3777 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3778 Details : If lookup is in progress on a file that is unlinked we might try
3779 to revalidate the inode and fail in revalidate after lookup is
3780 complete and ll_file_open() enqueues the open again but
3781 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3784 Frequency : always, if lconf --abort_recovery used
3786 Description: lconf --abort_recovery fails with 'Operation not supported'
3787 Details : lconf was attempting to abort recovery on the MDT device and not
3790 ------------------------------------------------------------------------------
3792 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3797 Frequency : rare (only unsupported configurations with a node running as an
3799 Bugzilla : 6514, 5137
3800 Description: Mounting a Lustre file system on a node running as an OST could
3802 Details : OSTs now preallocates memory needed to write out data at
3803 startup, instead of when needed, to avoid having to
3804 allocate memory in possibly low memory situations.
3805 Specifically, if the file system is mounted on on OST,
3806 memory pressure could force it to try to write out data,
3807 which it needed to allocate memory to do. Due to the low
3808 memory, it would be unable to do so and the node would
3809 become unresponsive.
3811 Severity : enhancement
3813 Description: Addition of lconf --service command line option
3814 Details : lconf now accepts a '--service <arg>' option, which is
3815 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3817 Severity : enhancement
3819 Description: Failover mode is now the default for OSTs.
3820 Details : By default, OSTs will now run in failover mode. To return to
3821 the old behaviour, add '--failout' to the lmc line for OSTs.
3823 Severity : enhancement
3825 Description: Health checks are now provided for MDS and OSTs
3826 Details : Additional detailed health check information on MSD and OSTs
3827 is now provided through the procfs health_check value.
3830 Frequency : occasional, depends on IO load
3832 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3833 after numerous create/delete cycles
3834 Details : The ext3 inode allocation policy would not allocate new inodes
3835 very well on the OSTs because there are no new directories
3836 being created. Instead we look for groups with free space if
3837 the parent directories are nearly full.
3842 Description: Network or server problems during mount may cause partially
3843 mounted clients instead of returning an error.
3844 Details : The config llog parsing code may overwrite the error return
3845 code during mount error handling, returning success instead
3851 Description: MDS can fail to allocate large reply buffers
3852 Details : After long uptimes the MDS can fail to allocate large reply
3853 buffers (e.g. zconf client mount config records) due to memory
3854 fragmentation or consumption by the buffer cache. Preallocate
3855 some large reply buffers so that these replies can be sent even
3856 under memory pressure.
3860 Frequency : rare (liblustre)
3861 Description: fsx running with liblustre complained that using truncate() to
3862 extend the file doesn't work. This patch corrects that issue.
3863 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3864 ATTR_SIZE and lsm use in llu_setattr_raw.
3868 Frequency : rare, only 2.6 kernels
3869 Description: Unusual file access patterns on the MDS may result in inode
3870 data being lost in very rare circumstances.
3871 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3872 code in the 2.6 kernel caused the inode and/or EA data not to
3873 be read from disk, causing single-file corruption.
3877 Frequency : rare, only 2.6 filesystems using extents
3878 Description: Heavy concurrent write and delete load may cause data corruption.
3879 Details : It was possible under high-load situations to have an extent
3880 metadata block in the block device cache from a just-unlinked
3881 file overwrite a newly-allocated data block. We now unmap any
3882 metadata buffers that alias just-allocated data blocks.
3886 Frequency : filesystems with default stripe_count larger than 77
3887 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3888 Details : lconf specifies an inode size of 4096 bytes when the default
3889 stripe_count is larger than 77. This conflicts with the default
3890 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3891 this case to avoid pinning too much memory for large EAs.
3893 ------------------------------------------------------------------------------
3895 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3900 Frequency : rare (extremely heavy IO load with hundreds of clients)
3902 Description: Client is evicted, gets IO error writing to file
3903 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3904 caused clients to be evicted for AST timeouts. The fixes in
3905 bug 5192 mean we no longer need to have such short AST timeouts
3906 so ldlm_timeout has been increased.
3909 Frequency : occasional during --force or --failover shutdown under load
3910 Bugzilla : 5949, 4834
3911 Description: Server oops/LBUG if stopped with --force or --failover under load
3912 Details : a collection of import/export refcount and cleanup ordering
3913 issues fixed for safer force cleanup
3916 Frequency : only filesystems larger than 120 OSTs
3917 Bugzilla : 5990, 6223
3918 Description: lfs getstripe would oops on a very large filesystem
3919 Details : lov_getconfig used kfree on vmalloc'd memory
3922 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3923 Bugzilla : 6242, 6243
3924 Description: reading from files that had been truncated to a non-zero size
3925 but never opened returned no data
3926 Details : ll_file_read() reads zeros from no-object files to EOF
3931 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3932 Details : There's little harm in aborting MDS/OSS recovery and letting it
3933 try again, so I removed the LASSERT and return an error instead.
3935 Severity : enhancement
3937 Description: New debugging infrastructure for tracking down data corruption
3938 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3939 (b) cover more of the client-side code path, and (c) try to narrow
3940 down where problems occurred
3944 Bugzilla : 3819, 4364, 4397, 6313
3945 Description: Racing close and eviction MDS could cause assertion in mds_close
3946 Details : It was possible to get multiple mfd references during close and
3947 client eviction, leading to one thread referencing a freed mfd.
3949 Severity: : enhancement
3950 Bugzilla : 3262, 6359
3951 Description: Attempts to reconnect to servers are now more aggressive.
3952 Details : This builds on the enhanced upcall-less recovery that was added
3953 in 1.4.2. When trying to reconnect to servers, clients will
3954 now try each server in the failover group every 10 seconds. By
3955 default, clients would previously try one server every 25 seconds.
3960 Description: After recovery, certain operations trigger a failed
3961 assertion on a client.
3962 Details : Failing over an mds, using lconf -d --failover, while a
3963 client was doing a readdir() call would cause the client to
3964 LBUG after recovery completed and the readdir() was resent.
3966 Severity : enhancement
3968 Description: Default groups are now added by lconf
3969 Details : You can now run lconf --group <servicename> without having to
3970 manually add groups with lmc.
3973 Frequency : occasional
3975 Description: Nodes with an elan id of 0 trigger a failed assertion
3978 Frequency : always when accessing e.g. tty/console device nodes
3980 Description: tty and some other devices nodes cannot be used on lustre
3981 Details : file's private_data field is used by device data and lustre
3982 values in there got lost. New field was added to struct file to
3983 store fs-specific private data.
3986 Frequency : when exporting Lustre via NFS
3988 Description: NFSD failed occasionally when looking up a path component
3989 Details : NFSD is looking up ".." which was broken in ext3 directories
3990 that had grown large enough to become hashed.
3993 Frequency : Clusters with multiple interfaces not on the same subnet
3995 Description: Nodes will repeatedly try to reconnect to an interface which it
3996 cannot reach and report an error to the log.
3997 Details : Extra peer list entries will be created by lconf with some peers
3998 unreachable. lconf now validates the peer before adding it.
4001 Frequency : Only if a default stripe is set on the filesystem root.
4003 Description: Setting a default stripe on the filesystem root prevented the
4004 filesystem from being remounted.
4005 Details : The client was sending extra request flags in the root getattr
4006 request and did not allocate a reply buffer for the dir EA.
4009 Frequency : occasional, higher if lots of files are accessed by one client
4010 Bugzilla : 6159, 6097
4011 Description: Client trips assertion regarding lsm mismatch/magic
4012 Details : While revalidating inodes the VFS looks up inodes with ifind()
4013 and in rare cases can find an inode that is being freed.
4014 The ll_test_inode() code will free the lsm during ifind()
4015 when it finds an existing inode and then the VFS later attaches
4016 this free lsm to a new inode.
4020 Bugzilla : 6422, 7030
4021 Description: MDS deadlock between mkdir and client eviction
4022 Details : Creating a new file via mkdir or mknod (starting a transaction
4023 and getting the ns lock) can deadlock with client eviction
4024 (gets ns lock and trying to finish a synchronous transaction).
4027 Frequency : occasional
4028 Description: While starting a server, the fsfilt_ext3 module could not be
4030 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
4031 kernels. Previously, lconf would still use the ext3 name
4032 when trying to load modules. Now, it will correctly use
4033 ext3 on 2.4 and ldiskfs on 2.6.
4035 Severity : enhancement
4036 Description: The default stripe count has been changed to 1
4037 Details : The interpretation of the default stripe count (0, to lfs
4038 or lmc) has been changed to mean striping across a single
4039 OST, rather than all available. For general usage we have
4040 found a stripe count of 1 or 2 works best.
4042 Severity : enhancement
4043 Description: Add support for compiling against Cray portals.
4044 Details : Conditional compiling for some areas that are different
4048 Frequency : occasional
4049 Bugzilla : 6409, 6834
4050 Description: Creating files with an explicit stripe count may lead to
4051 a failed assertion on the MDS
4052 Details : If some OSTs are full or unavailable, creating files may
4053 trigger a failed assertion on the MDS. Now, Lustre will
4054 try to use other servers or return an error to the
4058 Frequency : occasional
4060 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4061 caused lock timeout during readahead (since 1.4.2).
4062 Details : Processes doing readahead might match a lock that hasn't been
4063 granted yet if there are overlapping and conflicting lock
4064 requests. The readahead process waits on ungranted lock
4065 (original lock is CBPENDING), while OST waits for that process
4066 to cancel CBPENDING read lock and eventually evicts client.
4068 Severity : enhancement
4070 Description: Initial enabling of flock support for clients
4071 Details : Implements fcntl advisory locking and file status functions.
4072 This feature is provided as an optional mount flag (default
4073 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4074 locking are implemented yet, and those that are are not guaranteed
4075 to be completely correct in production environments.
4076 mount -t lustre -o [flock|noflock] ...
4079 Frequency : occasional
4081 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4082 assertion in the ext3 JBD (journaling) layer.
4083 Details : The b_committed_data struct is protected by the big kernel lock
4084 in 2.4 kernels, serializing journal_commit_transaction() and
4085 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4086 access to this struct. In 2.6 kernels there is finer grained
4087 locking to improve SMP performance of the JBD layer.
4091 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4096 Description: merge and cleanup kernel patches.
4099 -----------------------------------------------------------------------------
4101 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4103 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4104 versions of Lustre. Please update all servers and clients to
4105 version 1.4.2 or later at the same time. You must also run
4106 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4107 to update the configuration logs.
4109 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4110 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4111 - swab llog records of type '0' so we get proper header size/idx (5861)
4112 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4113 - fix rename of one directory over another leaking an inode (5953)
4114 - avoid SetPageDirty on 2.6 (5981)
4115 - don't re-add just-being-destroyed locks to the waiting list (5653)
4116 - when creating new directories, inherit the parent's custom
4117 striping settings if present parent (3048)
4118 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4119 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4120 - don't hold client locks on temporary worklist from l_lru (5666)
4121 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4122 - automatically evict dead clients (5921)
4123 - Update file size properly in create+truncate+fstat case (6196)
4124 - Do not unhash mountpoint dentries, do not allow removal of
4126 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4127 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4128 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4129 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4130 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4131 - Correctly handle reads of files with no objects (6243)
4132 - lctl recover will also mark a device active if deactivate used (5933)
4134 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
4135 - allow --write-conf on an MDS with different nettype than client (5619)
4136 - don't write config llogs to MDS for mounts not from that MDS (5617)
4137 - lconf should create multiple TCP connections from a client (5201)
4138 - init scripts are now turned off by default; run chkconfig --on
4139 lustre and chkconfig --on lustrefs to use them
4140 - upcalls are no longer needed for clients to recover to failover
4142 - add --abort-recovery option to lconf to abort recovery on device
4144 - add support for an arbitrary number of OSTs (3026)
4145 - Quota support protocol changes.
4146 - forward compatibility changes to wire structs (6007)
4147 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
4148 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
4149 - improved "lustre status" script
4150 - initialize blocksize for non-regular files (6062)
4151 - added --disable-server and --disable-client configure options (5782)
4152 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
4153 - Vanilla 2.4.29 support
4154 - increase maximum number of obd devices to 520 (6242)
4155 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4156 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4157 SLES 9 2.6.5 kernels
4158 - limit stripes per file to 160 (the maximum EA size) (6093)
4160 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4163 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4164 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4165 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4166 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4167 - small ext3 extents cleanups and fixes (5733)
4168 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4169 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4170 - avoid needless client->OST connect, fix handle mismatch (5317)
4171 - fix DLM error path that led to out-of-sync client, long delays (5779)
4172 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4173 - fix several locking issues related to i_size (5492,5624,5654,5672)
4174 - don't move pending lock onto export if it is already evicted (5683)
4175 - fix kernel oops when creating .foo in unlinked directory (5548)
4176 - fix deadlock in obdfilter statistics vs. object create (5811)
4177 - use time_{before,after} to avoid timer jiffies wrap (5882)
4178 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4179 - Do not leak request if server was not able to process it (5154)
4180 - If mds_open unable to find parent dir, make that negative lookup(5154)
4181 - don't create new directories with extent-mapping (5909, 5936)
4183 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
4184 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
4185 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
4186 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
4187 - support for RHEL4 kernel on i686 (5773)
4188 - provide error messages when incompatible logs are encountered (5898)
4190 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
4191 * version 1.4.0.10 (1.4.1 release candidate 1)
4193 - don't keep a lock reference when lock is not granted (4238)
4194 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4195 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
4196 - drop import inflight refcount on signal_completed_replay error (5255)
4197 - unlock page after async write error during send (3677)
4198 - handle missing objects in filter_preprw_read properly (5265)
4199 - no transno return for symlink open, don't save no-trasno open (3440)
4200 - don't try to complete elan receive that already failed (4012)
4201 - free RPC server reply state on error (5406)
4202 - clean up thread from ptlrpc_start_thread() on error (5160)
4203 - readahead could read extra page into cache that wasn't ejected (5388)
4204 - prevent races in class_attach/setup/cleanup/detach (5260)
4205 - don't dereference de->d_inode after l_dput of de (5458)
4206 - use "int" for stripe value returned from lock_to_stripe (5544)
4207 - mballoc allocation and error-checking fixes in 2.6 (5504)
4208 - block device patches to fix I/O request sizes in 2.6 (5482)
4209 - look up hostnames for IB nals (5602)
4210 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
4211 - don't start multiple acceptors for the same port (5277)
4212 - fix incorrect LASSERT in mds_getattr_name (5635)
4213 - export a proc file for general "ping" checking (5628)
4214 - fix "lfs check" to not block when the MDS is down (5628)
4216 - service request history (4965)
4217 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4218 - create an "evict_client" /proc entry on OSTs, like the MDS has
4219 - fix mount usage message, return errors per mount(8) (5168)
4220 - change grep [] to grep "[]" in tests so they work in more UMLs
4221 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4222 - remove ancient LOV_MAGIC_V0 EA support (5047)
4223 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4224 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4225 - allow readahead allocations to fail when low on memory (5383)
4226 - mmap locking landed again, after considerable improvement (2828)
4227 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4229 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4232 - send OST transaction number in read/write reply to free req (4966)
4233 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4234 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4235 - the watchdog thread now runs as interruptible (5246)
4236 - flock/lockf fixes (but it's still disabled, pending 5135)
4237 - don't use EXT3 constants in llite code (5094)
4238 - memory shortage at startup could cause assertion (5176)
4240 - reorganization of lov code
4241 - single portals codebase
4243 - add extents/mballoc support (5025)
4244 - direct I/O reads in the obdfilter (4048)
4245 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4247 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4250 - send OST transaction number in read/write reply to free req (4966)
4251 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4252 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4253 - the watchdog thread now runs as interruptible (5246)
4254 - handle missing objects in filter_preprw_read properly (5265)
4255 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4256 - drop import inflight refcount on signal_completed_replay error (5255)
4257 - unlock page after async write error during send (3677)
4258 - return original error code on reconstructed replies (3761)
4259 - no transno return for symlink open, don't save no-trasno open (3440)
4261 - add pid to ldlm debugging output (4922)
4262 - bump the watchdog timeouts -- we can't handle 30sec yet
4263 - extra debugging for orphan dentry/inode bug (5259)
4265 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
4268 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4269 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4270 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
4271 - fix NULL dereference in /proc/sys/portals/routes (4827)
4272 - allow failed mdc_close() operations to be interrupted (4561)
4273 - stop precreate on OST before MDS would time out on it (4778)
4274 - don't send partial-page writes before EOF from client (4410)
4275 - discard client grant for sub-page writes on large-page clients (4520)
4276 - don't free dentries not owned by NFS code, check generation (4806)
4277 - fix lsm leak if mds_create_objects() fails (4801)
4278 - limit debug_daemon file size, always print CERROR messages (4789)
4279 - use transno after validating reply (3892)
4280 - process timed out requests if import state changes (3754)
4281 - update mtime on OST during writes, return in glimpse (4829)
4282 - add mkfsoptions to LDAP (4679)
4283 - use ->max_readahead method instead of zapping global ra (5039)
4284 - don't interrupt __l_wait_event() during strace
4286 - add software watchdogs to catch hung threads quickly (4941)
4287 - make lustrefs init script start after nfs is mounted
4288 - fix CWARN/ERROR duplication (4930)
4289 - return async write errors to application if possible (2248)
4290 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4291 - print NAL number in %x format (4645)
4292 - update barely-supported suse-2.4.21-171 series (4842)
4293 - support for sles 9 %post scripts
4294 - support for building 2.6 kernel-source packages
4295 - support for sles km_* packages
4297 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4300 - ignore -ENOENT errors in osc_destroy (3639)
4301 - notify osc create thread that OSC is being cleaned up (4600)
4302 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4303 - reconstruct ost_handle() like mds_handle() (4657)
4304 - create a new thread to do import eviction to avoid deadlock (3969)
4305 - let lconf resolve symlinked-to devices (4629)
4306 - don't unlink "objects" from directory with default EA (4554)
4307 - hold socknal file ref over connect in case target is down (4394)
4308 - allow more than 32000 subdirectories in a single directory (3244)
4309 - fix blocks count for O_DIRECT writes (3751)
4310 - OST returns ENOSPC from object create when no space left (4539)
4311 - don't send truncate RPC if file size isn't changing (4410)
4312 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4313 - bind to privileged port in socknal and tcpnal (3689)
4315 - rate limit CERROR/CWARN console message to avoid overload (4519)
4316 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4317 - basic mmap support (3918)
4318 - kernel patch series update from b1_4 (4711)
4320 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4323 - avoid crash during MDS cleanup with OST shut down (2775)
4324 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4325 - don't use bad inodes on the MDS (3744)
4326 - dynamic object preallocation to improve recovery speed (4236)
4327 - don't hold spinlock over lock dumping or change debug flags (4401)
4328 - don't zero obd_dev when it is force cleaned (3651)
4329 - print grants to console if they go negative (4431)
4330 - "lctl deactivate" will stop automatic recovery attempts (3406)
4331 - look for existing locks in ldlm_handle_enqueue() (3764)
4332 - don't resolve lock handle twice in recovery avoiding race (4401)
4333 - revalidate should check working dir is a directory (4134)
4335 - don't always mark "slow" obdfilter messages as errors (4418)
4337 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4340 - don't close LustreDB during write_conf until it is done (3860)
4341 - fix typo in lconf for_each_profile (3821)
4342 - allow dumping logs from multiple threads at one time (3820)
4343 - don't allow multiple threads in OSC recovery (3812)
4344 - fix debug_size parameters (3864)
4345 - fix mds_postrecov to initialize import for llog ctxt (3121)
4346 - replace config semaphore with spinlock (3306)
4347 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4348 - don't allow enqueue to complete on a destroyed export (3822)
4349 - down write_lock before checking llog header bitmap (3825)
4350 - recover from lock replay timeout (3764)
4351 - up llog sem before sending rpc (3652)
4352 - reduce ns lock hold times when setting kms (3267)
4353 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4354 - fix NULL deref and obd_dev leak on setup error (3312)
4355 - replace some LBUG about llog ops with error handling (3841)
4356 - don't match INVALID dentries from d_lookup and spin (3784)
4357 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4358 - fix invalid assertion in ptlrpc_set_wait (3880)
4360 - add libwrap support for the TCP acceptor (3996)
4361 - add /proc/sys/portals/routes for non-root route listing (3994)
4362 - allow setting MDS UUID in .xml (2580)
4363 - print the stack of a process that LBUGs (4228)
4365 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4368 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4369 - make sure to unset replay flag from failed open requests (3440)
4370 - if default stripe count is 0, use OST count for inode size (3636)
4371 - update parent mtime/ctime on client for create/unlink (2611)
4372 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4373 - free recovery state on server during a forced cleanup (3571)
4374 - unregister_reply for resent reqs (3063)
4375 - loop back devices mounting and status check on 2.6 (3563)
4376 - fix resource-creation race that can provoke i_size == 0 (3513)
4377 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4378 - more debugging for page-accounting assertion (3746)
4379 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4380 - avoid deadlock after precreation failure (3758)
4381 - fix race and lock order deadlock in orphan handling (3450, 3750)
4382 - add validity checks when grabbing inodes from l_ast_data (3599)
4384 - add /proc/.../recovery_status to obdfilter (3428)
4385 - lightweight CDEBUG infrastructure, debug daemon (3668)
4386 - change default OSC RPC parameters to be better on small clusters
4387 - turn off OST read cache for files smaller than 32MB
4388 - install man pages and include them in rpms (3100)
4389 - add new init script for (un)mounting lustre filesystems (2593)
4390 - run chkconfig in %post for init scripts (3701)
4391 - drop scimac NAL (unmaintained)
4393 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4396 - clean kiobufs before and after use (3485)
4397 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4398 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4399 - change init script to fail more clearly if not run as root (1528)
4400 - allow clients to reconnect during replay (1742)
4401 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4402 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4403 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4404 - fix cond_resched (3554)
4405 - only evict unfinished clients after recovery (3515)
4406 - allow bulk resend, prevent data loss (3570)
4407 - dynamic ptlrpc request buffer allocation (2102)
4408 - don't allow unlinking open directory if it isn't empty (2904)
4409 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4410 - remove extraneous obd dereference causing LASSERT failure (3334)
4411 - don't use get_cycles() when creating temp. files on the mds (3156)
4412 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4413 - handle EEXIST for set-stripe, set proper directory name (3336)
4415 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4416 - fix ksocknal_fmb_callback() error messages (2918)
4418 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4421 - don't copy lvb into (possibly NULL) reply on error (2983)
4422 - don't deref dentry after dput, don't free lvb on error (2922)
4423 - use the kms to determine writeback rpc length (2947)
4424 - increment oti_logcookies when osc is inactive (2948)
4425 - update client's i_blocks count via lvb messages (2543)
4426 - handle intent open/close of special files properly (1557)
4427 - mount MDS with errors=remount-ro, like obdfilter (2009)
4428 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4429 - don't use cancelling-locks' kms values (2947)
4430 - use highest lock extent for kms, not last one (2925)
4431 - don't dereference ERR_PTR() dentry in error handling path (3107)
4432 - fix thread race in portals_debug_dumplog() (3122)
4433 - create lprocfs device entries at setup instead of at attach (1519)
4434 - common AST error handler, don't evict client on completion race (3145)
4435 - zero nameidata in detach_mnt in 2.6 (3118)
4436 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4437 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4438 - initialize RPC timeout timer earlier for 2.6 (3219)
4439 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4440 - print nal/nid for unknown nid (3258)
4441 - additional checks for oscc recovery before doing precreate (3284)
4442 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4443 - don't crash in mdc_close for bad permissions on open (3285)
4444 - zero i_rdev for non-device files (3147)
4445 - clear page->private before handing to FS, better assertion (3119)
4446 - tune the read pipeline (3236)
4447 - fix incorrect decref of invalidated dentry (2350)
4448 - provide read-ahead stats and refine rpc in flight stats (3328)
4449 - don't hold journal transaction open across create RPC (3313)
4450 - update atime on MDS at close time (3265)
4451 - close LDAP connection when recovering to avoid server load (3315)
4452 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4453 - don't leak open file on MDS after open resend (3325)
4454 - serialize filter_precreate and filter_destroy_precreated (3329)
4455 - loop device shouldn't call sync_dev() for nul device (3092)
4456 - clear page cache after eviction (2766)
4457 - resynchronize MDS->OST in background (2824)
4458 - refuse to mount the same filesystem twice on same mountpoint (3394)
4459 - allow llmount to create routes for mounting behind routers (3320)
4460 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4461 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4462 - fix rare problem with rename on htree directories (3417)
4464 - allow default OST striping configuration per directory (1414)
4465 - fix compilation for qswnal for 2.6 kernels (3125)
4466 - increase maximum number of MDS request buffers for large systems
4467 - change liblustreapi to be useful for external progs like lfsck (3098)
4468 - increase local configuration timeout for slow disks (3353)
4469 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4471 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4474 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4475 - don't overwrite extent policy data in reply if lock was blocked (2901)
4476 - drop filter export grants atomically with removal from device (2663)
4477 - del obd_self_export from work_list in class_disconnect_exports (2908)
4478 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4479 - swab reply message in mdc_close, other PPC fixes (2464)
4480 - fix destroying of named logs (2325)
4481 - overwrite old logs when running lconf --write_conf (2264)
4482 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4483 - fix race in target_handle_connect (2898)
4484 - mds_reint_create() should take same inode create lock (2926)
4485 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4486 - don't close files for self_export to avoid uninitialized obd (2936)
4487 - allow MDS with the same name as client node (2939)
4488 - hold dentry reference for closed log files for unlink (2325)
4489 - reserve space for all logs during transactions (2059)
4490 - don't evict page beyond end of stripe extent (2925)
4491 - don't oops on a deleted current working directory (2399)
4492 - handle hard links to targets without a parent properly (2517)
4493 - don't dereference NULL lock when racing during eviction (2867)
4494 - don't grow lock extents when lots of conflicting locks (2919)
4496 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4499 - account for cache space usage on clients to avoid data loss (974)
4500 - lfsck support in lustre kernel code (2349)
4501 - reduce journal credits needed for BRW writes (2370)
4502 - orphan handling to avoid losing space on client/server crashes
4503 - ptlrpcd can be blocked, stopping ALL progress (2477)
4504 - use lock value blocks to assist in proper KMS, faster stat (1021)
4505 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4506 - recovery for initial connections (2355)
4507 - fixes for mds_cleanup_orphans (1934)
4508 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4509 - block all file creations until orphan recovery completes (1901)
4510 - client remove rq_connection from request struct (2423)
4511 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4512 - recovery timer race (2670)
4513 - mdc_close recovey bug (2532)
4514 - ptlrpc cleanup bug (2710)
4515 - mds timeout on local locks (2588)
4516 - namespace lock held during RPCs (2431)
4517 - handle interrupted sync write properly (2503)
4518 - don't try to handle a message that hasn't been replied to (2699)
4519 - client assert failure during cleanup after abort recovery (2701)
4520 - leak mdc device after failed mount (2712)
4521 - ptlrpc_check_set allows timedout requests to complete (2714)
4522 - wait for inflight reqs when ptlrpcd finishes (2710)
4523 - make sure unregistered services are removed from the srv_list
4524 - reset bulk XID's when resending them (caught by 1138 test)
4525 - unregister_bulk after timeout
4526 - fix lconf error (2694)
4527 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4528 - readahead locks pages, leaves pending causing memory pressure (2673)
4529 - increase OST request buffers to 4096 on large machines (2729)
4530 - fix up permission of existing directories in simple_mkdir (2661)
4531 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4532 - don't assign transno to errored transactions (2742)
4533 - don't delete objects on OST if given a bogus objid from MDS (2751)
4534 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4535 - if rq_no_resend, then timeout request after recovery (2432)
4536 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4537 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4538 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4539 - MDS recovery shouldn't delete objects that it has given out (2730)
4540 - if enqueue arrives after completion, don't clobber LVB (2819)
4541 - don't unlock pages twice when trigger_group_io returns error (2814)
4542 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4543 - don't write pages to disk if there was an error (1450)
4544 - don't ping imports that have recovery disabled (2676)
4545 - take buffered bytes into account when balancing socknal conn (2817)
4546 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4547 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4548 - remove little-endian swabbing of llog records (1987)
4549 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4550 - retry reposting request buffers if they fail (1191)
4551 - grow extent at grant time to avoid granting a revoked lock (2809)
4552 - lock revoke doesn't evict page if covered by a second lock (2765)
4553 - disable VM readahead to avoid reading outside lock extents (2805)
4555 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4556 - updated kernel patches for hp-2.4.20 kernel (2681)
4558 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4561 - fix truncated write corruption (2366)
4562 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4564 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4565 - ensure len(uuid) < 37 in lmc (1171)
4566 - fix ia64 OOPS in llog_test (2255)
4567 - zero end of page at obdfilter for partial page writes (2648)
4568 - don't leave stale dentries around after renames (bug 2428)
4569 - fix timeouts when evicting a client with a single lock held (2642)
4570 - set deadline for the initial HELLO message to drain (2634)
4571 - print out dotted-quad IP addresses in the socknal (2302)
4573 - additional debugging for MDS client eviction problem (2443)
4574 - fix mkfsoptions support for osts (2603, 2604)
4576 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4579 - add series for the vanilla 2.6.0 kernel
4580 - add series for the vanilla 2.4.24 kernel
4581 - add series for a cray x86/64 UL kernel drop
4582 - fix xattr patches for the vanilla 2.4.19 series
4584 - generate true UUIDs in lmc (1171)
4585 - have portals stack dumping break in UML (2466)
4586 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4587 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4588 - wake callback waiting threads on client eviction (2460)
4589 - Add --ptldebug and --subsystem to lmc (1719)
4590 - update assertion to allow safe interrupt allocation
4591 - set rq_no_resend for cancel requests (2432)
4592 - recalculate ptlrpcd timeout after resend (2494)
4593 - call vfs_rmdir when removing pending directories (2368)
4594 - fix renaming a file to itself (2429)
4595 - lmc creates a default one-stripe lov (2454)
4596 - expand procfs space to handle large clusters (2326)
4597 - increase UML stack to avoid overflow
4598 - update lconf's list of debug and subsystem masks
4599 - fix lfs find --obd (2510)
4600 - /proc tunable for disabling filter read caching (2591)
4601 - stop rpm packages from altering slapd.conf (2301)
4602 - disable nagle in the socknal under 0conf (2578)
4603 - choose mds inode size based on stripe count (2572)
4604 - fix kernel-source rpm problems (2516)
4606 - add --disable-doc to avoid pdf generation (2421)
4607 - update documentation, tests, type-os, comments
4608 - avoid format warnings on ia64
4609 - remove the TOE NAL
4610 - tiny code cleanups by removing unused fields
4612 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4615 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4616 - remove the most blatant lies from BUILDING file (2371)
4617 - change default debug level to reasonable production setting
4618 - reduce client side cache size to reduce cache flush time
4619 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4620 - make TCP zerocopy and pinger support enabled by default (2476)
4621 - sync writes completed after process exits caused crashes (2319)
4622 - maintain correct mount count on the MDS (2356)
4623 - backout 1557, because 2316 wasn't really fixed
4624 - better file I/O statistics gathering in /proc
4625 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4626 - another kernel patch to fix zero-copy TCP function export
4627 - don't take duplicate lock when processing re-sent getattr (2420)
4628 - lctl uses obd_self_export instead of creating new conn (2353)
4629 - MDS/OST recovery case which requires object creation asserted (2425)
4630 - move lfs from /usr/sbin to /usr/bin in packages
4631 - fix race between mds_client_add and mds_client_free (2417)
4632 - use kmalloc instead of slabs in portals (2430)
4633 - don't create duplicate records when a failover MDS is present (2442)
4634 - remove unnecessary mount age check (2332)
4635 - don't remove directory inodes from locks prematurely (2451)
4636 - don't break if MDS service name is the same as hostname (2103)
4637 - fix races in client write RPC generation when cache full (2482)
4639 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4642 - remove now-unused request->rq_obd (278)
4643 - if an allocation fails, print out how much memory we've used (1933)
4644 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4645 - add the "configurable stack size" patch to most series files (1256)
4646 - ability to write large log records, for 100+ OST configs (2306)
4647 - fix NULL deref when filter_prep fails (2314)
4648 - fix operator precedence error in filter_sync
4649 - dynamic allocation of socknal TX descriptors (2315)
4650 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4651 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4652 - ensure that CWARN messages in Portals always get to the syslog
4653 - __init/__exit are not for prototype decls (ldlm_init/exit)
4654 - x86-64 compile warning fixes
4655 - fix gateway LMC keyword conflict (2318)
4656 - fix MDS lock inversions in getattr/reint paths (1844)
4657 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4658 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4659 - DLM race condition prevented some lock evictions (2328)
4660 - ENOMEM detection and retry on socknal sends (2230)
4661 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4662 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4663 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4664 - better error messages when a client is rejected during recovery (1505)
4665 - avoid cancelling locks which were never granted, after failure (2330)
4666 - fix i_sem/journal inversion in mds_client_add (2333)
4667 - fix truncate/getattr lock cycle deadlock (2334)
4668 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4669 - fix two rare exit paths which could leak an l_lock() ref (2321)
4670 - fencepost error in MDS/OST orphan recovery (2226)
4671 - make log record alignment 8 bytes (1988)
4672 - lstripe now fails when requested offset > ost_count (2237)
4673 - ensure that all kernel series have a complete list.h (1607)
4674 - fix crashes in special-file operations (2316)
4675 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4677 - allow configurable automake binary, for testing new versions
4678 - small update to the lfs documentation
4680 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4682 * fix negative export reference count in fsfilt_sync (2312)
4684 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4685 * release candidate 0.9.1
4687 - orphans are moved into the PENDING directory for possible recovery
4688 - replayed opens now open by fid for orphan/rename safety (1042)
4689 - last close of an orphan inode generates a transno (683)
4690 - chdir() and mount() now pin the directory entry (1020)
4691 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4692 - discard very old requests without processing them (1502)
4693 - remove some common, well-understood CERRORs (1505)
4694 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4695 - clear "grant" flags in OST replies until OST grant code lands (1644)
4696 - fix read performance by not clobbering i_blksize on client (1598)
4697 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4698 - make LRU size a /proc tunable, clears locks when reduced (707)
4699 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4700 - server locks take references on exports now (1558)
4701 - build fixes for 2.4.20-rh trees (1663)
4702 - return an error from lov_create if all OSCs are inactive (1751)
4703 - fix import levels when a reconnect happens without a timeout (1597)
4704 - exit early from mds_open if we get a lookup error (1749)
4705 - partial page read at EOF wouldn't wait for disk before sending (1642)
4706 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4707 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4708 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4709 - write/truncate lock inversion (1639)
4710 - Don't auto-load obdclass, portals modules during cleanup (1495)
4711 - fix timestamps from jumping to "now" (1763)
4712 - extra journal assertions (1648)
4713 - add an extra multiunlink test (1771)
4714 - fix read_record/write_record API (1776)
4715 - fix leak of offset_extent, possible incorrect i_size later (1772)
4716 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4717 - Debugging for the kqswnal_get_idle_tx problems (1820)
4718 - Allow recovery to be attempted multiple times (1536)
4719 - Write out MDS last_rcvd file after it is first created (1600)
4720 - Fix tx_descriptor leak in failed transmit situations (1827)
4721 - ext3 journaling fixes for assertion failure after IO error (1871)
4722 - class_export_put() on freed export after completion AST error (1896)
4723 - Fix revalidate looping in VFS (1322)
4724 - Don't access a freed export during MDS_REINT timeout (1521)
4725 - Add open-unlink recovery support on the MDS (1673,1764)
4726 - Return an error if no MDS data was read from last_rcvd (1946)
4727 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4728 - Don't setup a disk device that doesn't match exported UUID (317)
4729 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4730 - avoid committing NULL handle in force close
4731 - local.sh is now a one-stripe LOV configuration
4732 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4733 - fix ext3 htree duplicate directory entry corruption (1516)
4734 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4735 - update to new LOV EA format (2097)
4736 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4738 2003-06-15 Phil Schwan <phil@clusterfs.com>
4741 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4742 - per-import/export recovery handling (958, 931, 959)
4743 - multiple last-rcvd slots, for serving multiple FSes (949)
4744 - connections are again shared between multiple imp/exports (963, 964)
4745 - "umount -f" would hang if any requests needed to be sent (393, 978)
4746 - avoid pinning large req buffer by copying for queued messages (989)
4747 - add "uuid" to "lctl device" command to help upcalls (991)
4748 - "open" RPCs with transnos would confuse recovery counters (1037)
4749 - do proper endian conversion of all wire messages (288, 340, 891)
4750 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4751 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4752 - fix ptlrpc_connection leak in target_handle_connect (1174)
4753 - fix import refcounting bug in OST and MDS cleanup (1134)
4754 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4755 - fix very unlikely obd_types race condition (501)
4756 - remove osc_open hack for echo_client (1187)
4757 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4758 - a failure in read_inode2 leads to deadlock (1139)
4759 - cancel ack-locks as soon as transaction is committed (1072)
4760 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4761 - make sure to commitrw after any preprw to avoid deadlock (1162)
4762 - failing to execute a file in a lustre FS would lock inode (1203)
4763 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4764 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4765 - obd_uuid2tgt would walk off the end of the list (1255)
4766 - on IA64 the osc would give portals incorrect bulk size (1258)
4767 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4768 - fix lock inversion caused by new llite matching code (1282)
4769 - limit the number of dirty pages on a client to 10MB (1286)
4770 - timed out locks were not being corrected cancelled (1289)
4771 - fix O_DIRECT above 4GB on IA-32 (1292)
4772 * major user-visible changes
4773 - fail out/fail over policy now controlled by the upcall (993)
4775 - add OBD_PING to check server availability and failure (954)
4776 - lustre messages are now sent in sending host order (288, 340, 891)
4777 - add eadatalen to MDS getattr reply (340)
4778 - OST read replies may contain second buffer, with per-page status (593)
4780 2003-03-11 Phil Schwan <phil@clusterfs.com>
4783 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4784 - failed open()s could cause deadlock; fixed (867, 869)
4785 - stop cancelling OST locks when files are closed (481)
4786 - overlapping XID spaces caused network corruption (851, 853)
4787 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4788 - setattr_raw added, to do single-RPC, server-side setattrs
4789 - lmc/lconf syntax change for OST UUIDs
4790 - fix crashy race condition between ptlrpc_free_req and osc_close
4791 - don't use request in mdc_enqueue if we hit a timeout (889)
4792 - don't set the inode i_size for regular files from the MDS (896)
4793 - handle out of order completion AST (842)
4794 - don't LBUG if a lock request times out after receiving AST (913)
4795 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4796 - if a bad lock AST arrives, send an error instead of dropping entirely
4797 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4798 - fix leak in bulk IO when only partially completed (899, 900, 926)
4799 - fix O_DIRECT for ia64 (55)
4800 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4801 - C-z after timeout could hang a process forever; fixed (977)
4803 - client-side I/O cache (678, 924, 929, 941, 970)
4805 - READPAGE and SETATTRs which don't take server-side locks get
4808 2003-02-11 Phil Schwan <phil@clusterfs.com>
4811 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4812 - Fix fsfilt_extN_readpage() to read a full page of directory
4813 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4814 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4815 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4816 - Most intent operations were removed, in favour of a new RPC mode
4817 that does a single RPC to the server and bypasses most of the VFS
4818 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4819 - Aggressively cancel local locks on DLM servers
4820 - mds_reint_unlink sends EA to the client if it's the last nlink.
4821 client uses that EA to unlink OST objects.
4822 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4823 - recursive symlinks were fixed (439)
4824 - fixed NULL deref in DEBUG_REQ
4825 - filter_update_lastobjid no longer calls sync, which annoyed extN
4826 - fixed multi-client small-writes to a single file problem (445)
4827 - fixed mtime updates during file writes (607)
4828 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4829 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4830 - fixed Linux OST size reporting problem (444, 656)
4831 - OST now updates object mtime with writes or setattr (607, 619)
4832 - client verifies file size before zeroing page past EOF (445)
4833 - OST now writes last allocated objid to disk with allocation (108)
4834 - LOV on echo now works (409)
4836 - mds_reint_unlink sends a new buffer, with the EA included. this
4837 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4838 set if a regular file was being unlinked, and it was the last link
4839 - use PtlGet from the target for bulk writes (315)
4840 - OST now updates object mtime with writes or setattr (607, 619)
4841 - LDLM now has a grant-time callback to revalidate locked items, if
4843 - Many MDS operations were reorganized to combat race conditions
4845 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4846 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4848 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4851 - Fully reactivate OST imports after reconnection (512, others)
4852 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4853 - More graceful error handling for truncating on dead OST (515)
4854 - Don't error out unless we're actually accessing dead stripes (474)
4855 - Fix garbage sizes when stripes are missing (410)
4856 - LRU counters were broken, causing constant lock purge (433, 432)
4857 - garbage on read from stripes with failed OSTs (441)
4858 - mark OSCs as active before reconnecting during recovery (438)
4859 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4860 - lfind did not preserve OST order in output (443)
4861 - symlinks cause hung clients, incorrect data (439)
4862 - stop dereferencing request after dropping refcount (457)
4863 - don't LASSERT(spin_is_locked) on non-SMP (455)
4864 - fixes for many rename() bugs
4865 - fstat didn't correctly synchronize attributes (399)
4866 - server must handle lock cancellation during blocking AST prep (487)
4867 - bulk descriptors were free()d too soon (511)
4868 - fix paths in lconf, which would load incorrect modules (451, 507)
4869 - fix confusing lconf 'host not found' error message (386)
4870 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4871 - fix race condition in mdc_blocking_ast() for inode access (526)
4872 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4873 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4874 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4875 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4876 - fix lstripe in several ways (kernel side) (527)
4877 - fix request leak in ldlm_cli_enqueue (262)
4878 - incorrect OSC was marked inactive after OST failure
4879 - call mds_fs_cleanup before unmounting filesystem (524)
4880 - fix races between taking ns_lock and ldlm_lock_change_resource
4881 - fix races updating LOV export open file list
4882 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4883 - fix recovery NULL deref in ldlm_cli_cancel_unused
4884 - fix some DLM races by using new hash table for lock handles (419)
4885 - permit the client to specify desired inodes, at replay
4886 - duplicate requests when we queue them for replay reintegration
4887 - fix last_rcvd offset calculation
4888 - sync after each recovered transaction, so we always make progress
4889 - never, not always, ERESTART requests without transnos
4890 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4892 - skip replay if the MDS says that the client is already connected
4893 - don't check for a recovery-enabled export to match lctl's UUID
4894 - don't INC_USE_COUNT for phantom exports
4895 - don't crash when cleaning up phantom exports (567)
4896 - don't double-finish or set replay data for errored mdc_open requests
4897 - abort requests when they time out, so we don't get old replies
4898 - send/receive replies for AST messages again
4899 - if the client says that it doesn't have the lock, cancel it on the
4901 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4902 mark it as destroyed, it will all work out when decref is called
4903 - fix module use counts (22, 581)
4905 - ASTs now expect a reply (server cancels lock on error reply)
4907 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4910 - fix many simultaneous client startup (392)
4911 - fix dentry->d_it clobbering
4912 - credentials weren't being shipped for readdir/getattr operations
4913 - remove invalid assertions triggered during some concurrent MD
4915 - proper Lustre versions added (336, 389)
4916 - fix memory leak for create error case (398)
4917 - fix LOV locking bug that would get cli/srv out of sync
4918 - fix echo client over LOV (409)
4919 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4920 - fix double-O_EXCL intent crash (424)
4921 - avoid sending multiple lock CANCELs (352)
4923 - MDS can do multi-client recovery (modulo bugs in new code)
4925 - many updates, edits, cleanups
4927 2002-11-18 Phil Schwan <phil@clusterfs.com>
4930 - fix null d_it dereference (346)
4931 - fix full OST/dbench hang (333)
4932 - fix permission problem with file removal (286)
4933 - fix removal of OSCs from LOV when they fail
4934 - fix NULL deref during bulk timeout (214)
4935 - fix problems related to multiple filesystems on one MDS (241)
4936 - fixed serious subtle metadata locking bugs
4937 - free locks on clients when inodes are removed due to memory
4939 - fix inode pointer in lock data (285)
4940 - partial support for multiple MDS on a single host (241)
4941 - data locks weren't cancelled at clear_inode time (290, 311)
4942 - intent locks could lead to unbounded lock growth (205)
4943 - added a maximum lock count, an LRU list, and a flusher
4944 - fix multiple rename (365)
4945 - properly abstracted the echo client
4946 - OSC locked 1 byte too many; fixed
4947 - rewrote brw callback code:
4948 - fixed recovery bugs related to LOVs (306)
4949 - fixed too-many-pages-in-one-write crash (191)
4950 - fixed (again) crash in sync_io_timeout (214)
4951 - probably fixed callback-related race (385)
4953 - Add capability to MDS protocol
4954 - LDLM cancellations and callbacks on different portals
4956 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4959 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4960 - defer open object destruction to close time (601981, 138)
4961 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4962 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4963 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4964 - don't call obd_setattr in ll_file_release for destroyed objects
4965 * protocol change to lustre_msg: move |version| and add |flags|
4966 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4967 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4968 * added replay of create, unlink, link and rename operations during
4969 MDS failover; recovery should be much more robust now
4970 * remove failed OSCs from LOVs (only lov_create uses this so far)
4971 * the lustre-HOWTO was brought (more) up to date (582544)
4973 2002-10-23 Phil Schwan <phil@clusterfs.com>
4976 - in-use dentries weren't being reused properly (617851)
4977 - prevent multiple LDLM setup (599178)
4978 - fix LOV size calculations for truncate (617853)
4979 - fix client handling of MDS intent errors (POSIX)
4980 - fix permission bug in lovstripe.c test (624321)
4981 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4982 - truncate past end of file could corrupt data
4983 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4984 - a race in recovery could return ETIMEDOUT to apps (623947)
4985 - building outside the source directory was fixed
4986 * the lustre-HOWTO was brought (more) up to date (582544)
4987 * major progress was made on recovery functionality
4989 2002-10-10 Phil Schwan <phil@clusterfs.com>
4992 - recovery deadlock fix
4993 - rm -rf causes LBUG fix (617817)
4994 - file open by multiple tasks fix (618962)
4995 - directory permissions bugs (602707 and 620007)
4996 - journal_stop fixed with locking (611313)
4997 - O_APPEND failures resolved (618273, perhaps 614459)
4998 - lconf PATH fix (619770)
4999 - IA64 build fix (621450)
5000 - RPC buffer sizes scale with amount of memory
5002 2002-10-01 Phil Schwan <phil@clusterfs.com>
5005 - locks would be cancelled without throwing away data pages,
5006 resulting in inconsistent data (605627)
5007 - inode attributes were not always being refreshed (605627, 612449)
5008 - lconf now continues to cleanup after lctl reports an error
5009 - MDS now enforces user permissions (602707)
5010 - lprocfs cleanup fixed, but not yet enabled (614157)
5011 - fixed infinite server hang, should a client not respond to an AST
5012 - avoid going into recovery if user calls readlink() with a buffer
5013 that's too small (613941)
5014 - AST RPCs no longer require replies (614867) -- this may be changed
5015 - don't crash server if client sends an IOV that's too big (611336)
5016 - fixed lock conversion deadlock (611892)
5017 - fixed the following of symlinks (614622)
5018 * recovery: the server can remove locks from a client that dies, other
5019 clients can make progress
5020 * more extN patch fixes
5021 * compile-time configurable ptlrpc buffer allocations
5023 - collaborative read cache document
5024 - Lustre Lite Performance CDR document-in-progress
5026 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5029 - fix typo in patch-2.4.18
5031 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5034 - clear ptlrpc request each time in handle_incoming_request()
5035 - unlink of files now destroys the object on the OST
5037 2002-09-19 Peter Braam <braam@clusterfs.com>
5039 * add hard link support
5040 * change obdfile creation method
5041 * kernel patch changed
5043 2002-09-19 Peter Braam <braam@clusterfs.com>
5046 - stack overflow bug in extN fixed
5048 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
5050 * documentation updates
5051 - add man pages for config tools
5052 - update tests/README to describe testing with new config tools
5053 - finish metadata API descriptions
5054 * bug fixes and cleanups
5055 - statfs workaround for 16TB limit
5056 - LOV stripe allocation improved, can stripe on subset of OSTs
5057 - LOV file size/IO offset was wrong for files > 4GB in size
5058 - object EA data was being dropped, caused files to be unreadable
5059 - memory overflow with non-LOV OST caused memory corruption
5060 - fixed regression tests to work with new config tools, obdfilter
5061 - fixed bug when directory size became larger than 1 block
5062 - fixed bug (for single client case) when PWD was deleted
5063 - invalidate local directory pages when doing intent-based ops
5064 - avoid LDLM oops when lock callback contained bad data
5066 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5068 * documentation updates
5069 * bug fixes and cleanups
5070 - configuration tools
5073 - 64-bit compile warnings
5074 - 64-bit internal statfs data
5076 * test_brw on persistent OST devices
5078 * lprocfs (disabled)
5080 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5082 * documentation updates
5083 * bug fixes and cleanups
5084 * configuration tools
5086 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5089 * many small fixes to 0_5_4
5090 * io/network handling
5091 * thinkos in MDS operations
5093 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5096 * crucial basic fixes to 0.5.3
5097 * IOR, Iozone work over Elan
5100 2002-08-07 Phil Schwan <phil@clusterfs.com>
5101 * version 0_5_3, our first alpha
5102 * we use the new Portals iovs
5103 * documentation updates
5104 * bug fixes and cleanups
5105 * small changes in the DLM wire protocol
5107 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5108 * version 0_5_1 with some initial stability,
5109 * locking on MD and file I/O.
5110 * documentation updates
5111 * several bug fixes since 0.5.0
5112 * small changes in wire protocol
5114 2002-07-18 Phil Schwan <phil@clusterfs.com>
5116 * delivered as Lustre Light Alpha
5117 * fixed a crash after handling invalid MDS requests
5118 * fixed directory pages for architectures with non-4k pages sizes
5120 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5121 * release version v0_4_4
5122 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5124 * Fixes a number of interruption problems with OST operations.
5125 * Update documentation for portals header changes
5126 * Move all wire protocol structs/defines to lustre_idl.h
5127 * Fixes symlink length bug.
5128 * Add tcpdump to repository.
5130 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5131 * release version v0_4_3
5132 * Fixes statfs for inodes on extN.
5133 * Fixes bug in runtests which would delete /etc/hosts.
5134 * Use 64-bit object IDs wherever possible (not into VFS though)
5135 Remove ost_get_info, which is unused by lustre, and out of date.
5137 2002-07-03 Peter Braam <braam@clusterfs.com>
5138 * release version v0_4_2 Fixes a lookup error (type not passed)
5139 * move forward to head of Portals
5140 * move forward to latest Lustre kernel
5142 2002-06-25 Peter Braam <braam@clusterfs.com>
5143 * release version v0_4_1. Hopefully stable on single node use.