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: Hitting mdc_commit_close() ASSERTION
18 Details : Properly handle request reference release in
19 ll_release_openhandle().
23 Frequency : only patchless client
24 Description: add workaround for race between add/remove dentry from hash
26 Severity : enhancement
28 Description: Allow OST glimpses to return PW locks
32 Description: LBUG when llog conf file is full
33 Details : When llog bitmap is full, ENOSPC should be returned for plain
38 Description: Prevent import from entering FULL state when server in recovery
42 Description: service mount cannot take device name with ":"
43 Details : Only when device name contains ":/" will mount treat it as
49 Description: replace ptlrpcd with the statahead thread to interpret the async
50 statahead RPC callback
54 Frequency : on recovery
55 Description: I/O failures after umount during fail back
56 Details : if client reconnected to restarted server we need join to recovery
57 instead of find server handler is changed and process self eviction
58 with cancel all locks.
60 Severity : enhancement
62 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
64 Severity : enhancement
66 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
68 Severity : enhancement
70 Description: Upcall on Lustre log has been dumped
71 Details : Allow for a user mode script to be called once a Lustre log has
72 been dumped. It passes the filename of the dumped log to the
73 script, the location of the script can be specified via
74 /proc/sys/lnet/debug_log_upcall.
79 Description: avoid idr_remove called for id which is not allocated.
80 Details : Move assigment s_dev for clustered nfs to end of initialization,
81 to avoid problem with error handling.
86 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
87 Details : When connection is reused this not moved from CONN_UNUSED_HASH
88 into CONN_USED_HASH and this prodice warning when put connection
91 Severity : enhancement
93 Description: Export bytes_read/bytes_write count on OSC/OST.
97 Description: Early reply size mismatch, MGC loses connection
98 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
99 the connect flags are properly negotiated.
103 Description: Properly propagate oinfo flags from lov to osc for statfs
104 Details : restore missing copy oi_flags to lov requests.
106 Severity : enhancement
108 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
109 : l_getgroups(8), lst(8), routerstat(8)
110 Details : included man pages for llobdstat(8), llstat(8),
111 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
113 Severity : enhancement
115 Description: configure's --enable-quota should check the
116 : kernel .config for CONFIG_QUOTA
117 Details : configure is terminated if --enable-quota is passed but
118 : no quota support is in kernel
122 Description: Remove portals compatibility
123 Details : Remove portals compatibility, not interoperable with releases
128 Description: Resolve device initialization race
129 Details : Prevent proc handler from accessing devices added to the
130 obd_devs array but yet be intialized.
132 Severity : enhancement
134 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
136 Severity : enhancement
138 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
143 Description: Using wrong pointer in osc_brw_prep_request
144 Details : Access to array[-1] can produce panic if kernel compiled with
145 CONFIG_PAGE_ALLOC enabled
149 Description: Client runs out of low memory
150 Details : Consider only lowmem when counting initial number of llap pages
154 Description: *optional* service tags registration
155 Details : if the "service tags" package is installed on a Lustre node
156 When the filesystem is mounted, a local-node service tag will
157 be created. See http://inventory.sun.com/ for more information
158 about the Service Tags asset management system.
162 Description: Kernel BUG tries to release flock
163 Details : Lustre does not destroy flock lock before last reference goes
164 away. So always drop flock locks when client is evicted and
165 perform unlock regardless of successfulness of speaking to MDS.
169 Description: add recount protection for osc callbacks, so avoid panic on shutdown
173 Description: sanity test 65a fails if stripecount of -1 is set
174 Details : handle -1 striping on filesystem in ll_dirstripe_verify
179 Description: ASSERTION(CheckWriteback(page,cmd)) failed
180 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
183 Severity : enhancement
185 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
189 Description: do not process already freed flock
190 Details : flock can possibly be freed by another thread before it reaches
191 to ldlm_flock_completion_ast.
195 Description: LBUG during stress test
196 Details : Need properly lock accesses the flock deadlock detection list.
200 Description: oops in page fault handler
201 Details : kernel page fault handler can return two special 'pages' in error case, don't
202 try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
206 Description: timeout with invalidate import.
207 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which should be
208 handled by ptlrpcd. This produce long age waiting and -ETIMEOUT
209 ptlrpc_invalidate_import and as result LASSERT.
211 Severity : enhancement
213 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
217 Description: enable MGS and MDT services start separately
218 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
219 a co-located MGS without starting the MGS, which is a complement
220 to 'nosvc' mount option.
224 Frequency : after recovery
225 Description: precreate to many object's after del orphan.
226 Details : del orphan st in oscc last_id == next_id and this triger growing
227 count of precreated objects. Set flag LOW to skip increase count
228 of precreated objects.
232 Frequency : rare, on clear nid stats
233 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
234 Details : when clean nid stats sometimes try destroy live entry,
235 and this produce panic in free.
239 Description: Stack overflow during MDS log replay
240 ease stack pressure by using a thread dealing llog_process.
244 Description: wait until IO finished before start new when do lock cancel.
245 Details : VM protocol want old IO finished before start new, in this case
246 need wait until PG_writeback is cleared until check dirty flag and
247 call writepages in lock cancel callback.
249 Severity : enhancement
251 Description: using special macro for print time and cleanup in includes.
255 Description: mds_mfd_close() ASSERTION(rc == 0)
256 Details : In mds_mfd_close(), we need protect inode's writecount change
257 within its orphan write semaphore to prevent possible races.
261 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
265 Frequency : rare, on shutdown ost
266 Description: don't hit live lock with umount ost.
267 Details : shrink_dcache_parent can be in long loop with destroy dentries,
268 use shrink_dcache_sb instead.
272 Description: don't panic with use echo client
273 Details : echo client pass NULL as client nid pointer and this produce null
278 Description: fix build on ppc32
279 Details : compile code with -m64 flag produce wrong object file for ppc32.
283 Description: add message levels for liblustreapi
287 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
288 Details : Move the 'good_osts' check before the 'total_bavail' check. This
289 will result in an -EAGAIN and in the exit call path we call
290 alloc_rr() which will with increasing aggressiveness attempt to
291 aquire precreated objects on the minimum number of required OSCs.
295 Description: Use old size assignment to avoid deadlock
296 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
297 the scheduling while holding a spinlock. We do not need locking
298 for size in ll_update_inode() because size is only updated from
299 the MDS for directories or files without objects, so there is no
300 other place to do the update, and concurrent access to such inodes
301 are protected by the inode lock.
305 Description: resolve "_IOWR redefined" build error on SLES10
309 Description: dump the memory debugging after all modules are unloaded to
310 suppress false negative in conf_sanity test 39
312 Severity : enhancement
314 Description: build kernel-ib packages for OFED 1.3 in our release cycle
319 Description: fix SLES kernel versioning
320 Details : the kernel version for our SLES 10 kernel did not include a "-"
321 before the "smp" at the end. while this was not a problem in
322 general, it did mean that software trying to use the kernel
323 version to try to detect a vendor specific kernel would fail.
324 this was most evident by the OFED build scripts.
328 Description: Don't update lov_desc members until making sure they are valid
329 Details : When updating lov_desc members via proc fs, need fix their
330 validities before doing the real update.
334 Description: don't put request into delay list while invalidate in flight.
335 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
336 import in flight. this produce timeout for invalidate and sometimes
337 can cause stale data.
341 Frequency : on ppc only
342 Description: not convert ost objects for directory because it's not exist.
343 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
347 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
349 Severity : enhancement
351 Description: Update to support 2.6.22.14 vanilla kernel.
355 Frequency : rare, on recovery
356 Description: read procfs can produce deadlock in some situation
357 Details : Holding lprocfs lock which send rpc can produce block for destroy
358 obd objects and this also block reconnect with -EALREADY. This isn't
359 fix all lprocfs bugs - but make it rare.
361 Severity : enhancement
363 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
366 Frequency : frequent on X2 node
368 Description: mdc_set_open_replay_data LBUG
369 Details : Set replay data for requests that are eligible for replay.
373 Description: lustre_mgs: operation 101 on unconnected MGS
374 Details : When MGC is disconnected from MGS long enough, MGS will evict the
375 MGC, and late on MGC cannot successfully connect to MGS and a lot
376 of the error messages complaining that MGS is not connected.
380 Frequency : on network error
381 Description: panic with double free request if network error
382 Details : mdc_finish_enqueue is finish request if any network error ocuring,
383 but it's true only for synchronus enqueue, for async enqueue
384 (via ptlrpcd) this incorrect and ptlrpcd want finish request
387 Severity : enhancement
389 Description: client-side metadata stat-ahead during readdir(directory readahead)
390 Details : perform client-side metadata stat-ahead when the client detects
391 readdir and sequential stat of dir entries therein
394 Frequency : on start mds
396 Description: Implement get_info(last_id) in obdfilter.
399 Frequency : occasional
401 Description: Correctly check stale fid, not start epoch if ost not support SOM
402 Details : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
403 overwrited with new generated fid, but mds can anwer with one of these
404 two fids and both is not stale. setattr incorectly start epoch and
405 assume will be called done_writeting, but without SOM done_writing
409 Frequency : rare, depends on device drivers and load
411 Description: MDS or OSS nodes crash due to stack overflow
412 Details : Code changes in 1.8.0 increased the stack usage of some functions.
413 In some cases, in conjunction with device drivers that use a lot
414 of stack the MDS (or possibly OSS) service threads could overflow
415 the stack. One change which was identified to consume additional
416 stack has been reworked to avoid the extra stack usage.
419 Frequency : occasional
421 Description: Do not fail import if osc_interpret_create gets -EAGAIN
422 Details : If osc_interpret_create got -EAGAIN it immediately exits and
423 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
424 oscc_has_objects and see OSC has no objests and call
425 oscc_internal_create to resend create request.
427 Severity : enhancement
429 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
431 Severity : enhancement
433 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
436 Frequency : very rare
438 Description: Fix replay if there is an un-replied request and open
439 Details : In some cases, older replay request will revert the
440 mcd->mcd_last_xid on MDS which is used to record the client's
443 Severity : enhancement
445 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
447 Severity : enhancement
449 Description: Add rhel5 support to HEAD.
451 Serverity : enhancement
453 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
458 Description: Don't allow skipping OSTs if index has been specified.
459 Details : Don't allow skipping OSTs if index has been specified, make locking
460 in internal create lots better.
464 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
465 Details : access to bitfield in structure is always rounded to long
466 and this produce problem with not atomic change any bit.
470 Description: Lustre make rpms failed.
471 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
472 builds Lustre rpms from tarball.
475 Frequency : rare on shutdown ost
477 Description: If llog cancel was not send before clean_exports phase, this can
478 produce deadlock in llog code.
479 Details : If llog thread has last reference to obd and call class_import_put
480 this produce deadlock because llog_cleanup_commit_master wait when
481 last llog_commit_thread exited, but this never success because was
482 called from llog_commit_thread.
486 Description: allow userland application know is lost one of stripes.
487 Details : fill lvb_blocks with error code on ost and return it to
488 application if error flag found.
492 Description: NULL lov_tgts causing MDS oops
493 Details : more safe checks for NULL lov_tgts for avoid oops.
495 Severity : enhancement
497 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
501 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
502 Details : If an OST is down the MDS will hang indefinitely in
503 obd_statfs_rqset() waiting for the statfs data. While for
504 MDS QOS usage of statfs, it should not stuck in waiting.
506 Severity : enhancement
508 Description: remote_acl support
509 Details : Support ACL-based permission check for remote user.
510 Support setfacl/getfacl for remote user with the utils
511 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
512 the system "{s,g}etfacl" utils.
514 Severity : enhancement
516 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
518 Severity : enhancement
520 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
524 Description: make mgs_setparam() handle fsname containing dash
525 Details : fsname containing a dash does not work with lctl conf_param
527 Severity : enhancement
529 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
531 Severity : enhancement
533 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
537 Frequency : rare, at shutdown
538 Description: access already free / zero obd_namespace.
539 Details : if client_disconnect_export was called without force flag set,
540 and exist connect request in flight, this can produce access to
541 NULL pointer (or already free pointer) when connect_interpret
542 store ocd flags in obd_namespace.
546 Frequency : only at startup
547 Description: not alloc memory with spinlock held.
548 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
549 if any spinlock held.
551 Severity : enhancement
553 Description: make lustre randomly fail allocating memory
554 Details : Make lustre randomly failed allocating memory for testing purpose.
556 Severity : enhancement
558 Description: lost problems with lov objid file
559 Details : Fixes some scability and access to not inited memory problems
560 in work with lov objdid file.
565 Description: lfs find does not continue on file error
566 Details : Continue other files processing when a file/dir is absent.
570 Description: Inconsistent usage of lustre_pack_reply()
571 Details : Standardize the usage of lustre_pack_reply() such that it
572 always generate a CERROR on failure.
575 Frequency : occasional
577 Description: MDS hang or stay in waiting lock
578 Details : If client receive lock with CBPENDING flag ldlm need send lock
579 cancel as separate rpc, to avoid situation when cancel request
580 can't processed due all i/o threads stay in wait lock.
584 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
588 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
592 Description: Update to SLES9 kernel-2.6.5-7.287.3.
596 Description: Update to RHEL4 latest kernel.
598 Severity : enhancement
600 Description: Build SLES10 patchless client fails
601 Details : The configure was broken by run ./configure with
602 --with-linux-obj=.... argument for patchless client. When the
603 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
604 can't search header adequately. Use absolute path such as
605 -I($LINUX)/include instead.
609 Description: interrupt oig_wait produce painc on resend.
610 Details : brw_redo_request can be used for resend requests from ptlrpcd and
611 private set, and this produce situation when rq_ptlrpcd_data not
612 copyed to new allocated request and triggered LBUG on assert
613 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
614 ptlrpcd set if request is changed and can be safety changed to use
619 Description: LASSERT_{REQ,REP}SWAB macros are buggy
620 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
621 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
622 these macros and replace them with their difinitions instead.
626 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
627 Details : Update which_patch & target file for SLES10 latest kernel.
630 Description: Remove client patches from SLES 10 kernel.
631 Details : This causes SLES 10 clients to behave as patchless clients
632 even on a Lustre-patched (server) kernel.
634 Severity : enhancement
636 Description: self-adjustable client's lru lists
637 Details : use adaptive algorithm for managing client cached locks lru
638 lists according to current server load, other client's work
639 pattern, memory activities, etc. Both, server and client
640 side namespaces provide number of proc tunables for controlling
643 Severity : enhancement
645 Description: light-weight GSS support
646 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
647 and reduce performance overhead.
649 Severity : enhancement
651 Description: Linux keyring support
652 Details : Support using service of Linux keyring for Lustre GSS internal
653 context refresh/cache mechanism.
657 Description: Fix errors in lfs documentation
658 Details : Fixes man pages
662 Description: don't use GFP_* in generic Lustre code.
663 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
667 Description: obdclass is limited by single OBD_ALLOC(idarray)
668 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
672 Description: warnings with build patchless client with vanila 2.6.19 and up
673 Details : change old ctl_table style and replace ctl_table/ctl_table_header
674 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
678 Frequency : Only for SLES
679 Description: sanity_quota fail test_1
680 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
681 file, which makes TSTID[2] inunique.
685 Frequency : Only for SLES9
686 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
687 Details : Update target/ChangeLog/which_patch .
691 Frequency : Only for test_10 in sanity-quota.sh
692 Description: a bug in quota test code
693 Details : checking a test flag in a wrong place
697 Frequency : Only for test_13 in sanity-quota.sh
698 Description: a bug in quota test script
699 Details : shouldn't check overfull allocated quota
703 Description: jbd statistics
704 Details : Port older jbd statistics patch for sles10
706 Severity : enhancement
708 Description: Add printing inode info into message about error in writepage.
713 Description: reply_lock_interpret crash due to race with it and lock cancel.
714 Details : Do not replay locks that are being cancelled. Do not reference
715 locks by their address during replay, just by their handle.
719 Frequency : When flocks are used.
720 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
721 Details : Flock locks might destroy just granted lock if it could be merged
722 with another existing flock, this is done in completion handler,
723 so teach ldlm_cli_enquque_fini that this is a valid case for
729 Description: Oops in read and write path when failing to allocate lock.
730 Details : Check if lock allocation failed and return error back.
734 Description: lstripe command fails for valid OST index
735 Details : The stripe offset is compared to lov->desc.ld_tgt_count
736 instead of lov->desc.ld_active_tgt_count.
740 Description: sanity.sh failed test 103
741 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
746 Description: df doesn't work properly if diskfs blocksize != 4K
747 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
751 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
752 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
754 Severity : enhancement
756 Description: Adaptive timeouts
757 Details : RPC timeouts adapt to changing server load and network
758 conditions to reduce resend attempts and improve recovery time.
762 Description: llapi_file_create() does not allow some changes
763 Details : add llapi_file_open() that allows specifying the mode and
764 open flags, and also returns an open file handle.
768 Description: merge and cleanup kernel patches.
769 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
773 Description: Add journal checksum support.(Kernel part)
774 Details : The journal checksum feature adds two new flags i.e
775 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
776 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
777 indicates that the commit block contains the checksum for
778 the blocks described by the descriptor blocks. Now commit
779 record can be sent to disk without waiting for descriptor
780 blocks to be written to disk. This behavior is controlled
781 using JBD2_FEATURE_ASYNC_COMMIT flag.
785 Description: OSS needs mutliple precreate threads
786 Details : Add ability to start more than one create thread per OSS.
790 Description: Sanity.sh test_65k failed: setstripe should have succeeded
791 Details : execute lfs setstripe on client
795 Description: mds_obd_create error creating tmp object
796 Details : When the user sets quota on root, llog will be affected and can't
797 create files and write files.
800 Frequency : Always on ia64 patchless client, and possibly others.
802 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
803 Details : This allows the patchless client to be loaded on architectures
808 Description: RedHat Update kernel for RHEL5
809 Details : Add the kernel config file for RHEL5.
813 Description: RedHat Update kernel for RHEL5
814 Details : Modify the kernel config file more closer RHEL5.
818 Description: Build failure against Centos5 (RHEL5)
819 Details : Define PAGE_SIZE when it isn't present.
823 Description: test_7 on sanity-quota.sh can't be used on separate machines
824 Details : add facet to handle it
828 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
829 Details : with NFS, the anon dentry's parent was set to itself in
830 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
831 corresponding dentry other than use rec->ur_name.
833 Severity : enhancement
835 Description: lfs setstripe enhancement
836 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
840 Description: enable data checksumming by default
841 Details : enable checksum by default, allow --disable-checksum
842 configure option and "-o nochecksum" mount option. Checksums
843 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
847 Description: lustre support for RHEL5
848 Details : Add support for RHEL5.
852 Description: Client eviction due to failover config
853 Details : after a connection loss, the lustre client should attempt to
854 reconnect to the last active server first before trying the
855 other potential connections.
859 Description: when mds and osts use different quota unit(32bit and 64bit),
860 quota will be released repeatly.
861 Details : void sending multiple quota reqs to mds, which will keep the
862 status between the reqs.
866 Description: osts not allocated evenly to files
867 Details : change the condition to increase offset_idx
870 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
872 Description: Data corruption for OSTs that are formatted larger than 2TB
874 Details : When generating the bio request for lustre file writes the
875 sector number would overflow a temporary variable before being
876 used for the IO. The data reads correctly from Lustre (which
877 will overflow in a similar manner) but other file data or
878 filesystem metadata may be corrupted in some cases.
882 Description: Tune the kernel for good SCSI performance.
883 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
884 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
889 Description: rewrite ext2-derived code in obdclass/uuid.c
890 Details : rewrite inherited code (uuid parsing code from ext2 utils)
891 from scratch preserving functionality.
895 Description: "lfs find -obd UUID" prints directories
896 Details : "lfs find -obd UUID" will return all directory names instead
897 of just file names. It is incorrect because the directories
898 do not reside on the OSTs.
902 Description: use i_size_read and i_size_write in 2.6 port
903 Details : replace inode->i_size access with i_size_read/write()
906 Frequency : when removing large files
908 Description: scheduling issue during removal of large Lustre files
909 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
910 It causes scheduling issues when removing large files (17TB in the
914 Frequency : only with liblustre clients on XT3
916 Description: evictions taking too long
917 Details : allow llrd to evict clients directly on OSTs
920 Frequency : only on ppc
922 Description: /proc/fs/lustre/devices broken on ppc
923 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
924 We should make sure the type of 'index' is loff_t and then cast
925 explicitly as needed below. Do not assign an explicitly cast
929 Frequency : only for rhel5
931 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
932 Details : Modify the target file & which_kernel.
934 Severity : enhancement
936 Description: omit set fsid for export NFS
937 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
938 as FS_REQUIRES_DEV for avoid problems with generate fsid.
942 Frequency : Always, for kernels after 2.6.16
943 Description: Fix warning idr_remove called for id=.. which is not allocated.
944 Details : Last kernels save old s_dev before kill super and not allow
945 to restore from callback - restore it before call kill_anon_super.
949 Description: buffer overruns could theoretically occur
950 Details : llapi_semantic_traverse() modifies the "path" argument by
951 appending values to the end of the origin string, and a
952 overrun may occur. Adding buffer overrun check in liblustreapi.
956 Description: Fix error on 'ls .' at the top of the Lustre mount.
957 Details : Don't revalidate dentry if it is a root dentry.
961 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
962 Details : Modify vm-tunables-rhel4.patch.
966 Description: Kernel config for 2.6.18-vanilla.
967 Details : Modify targets/2.6-vanilla.target.in.
968 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
969 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
970 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
971 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
975 Description: improve handling recoverable errors
976 Details : if request processig with error which can be recoverable on server
977 request should be resend, otherwise page released from cache and
982 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
983 Details : Modify target file & which_patch.
984 A flaw was found in the IA32 system call emulation provided
985 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
986 value could be stored in the %RAX register, which could trigger an
987 out-of-bounds system call table access. An untrusted local user
988 could exploit this flaw to run code in the kernel
989 (ie a root privilege escalation). (CVE-2007-4573).
993 Description: O_DIRECT bypasses client statistics.
994 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
995 read_ahead_stats were updated. copied the stats section from
996 osc_send_oap_rpc() into async_internal().
1000 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1004 Description: change order of libsysio includes
1005 Details : '#include sysio.h' should always come before '#include xtio.h'
1009 Description: handle "serious error: objid * already exists" more gracefully
1010 Details : If LAST_ID value on disk is smaller than the objects existing in
1011 the O/0/d* directories, it indicates disk corruption and causes an
1012 LBUG(). If the object is 0-length, then we should use the existing
1013 object. This will help to avoid a full fsck in most cases.
1015 Severity : enhancement
1017 Description: adapt the lustre_config script to support the upgrade case
1018 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1019 server targets to 1.6 in parallel.
1023 Description: To avoid grant space > avaible space when the disk is almost
1024 full. Without this patch you might see the error "grant XXXX >
1025 available" or some LBUG about grant, when the disk is almost
1027 Details : In filter_check_grant, for non_grant cache write, we should
1028 check the left space by if (*left > ungranted + bytes), instead
1029 of (*left > ungranted), because only we are sure the left space
1030 is enough for another "bytes", then the ungrant space should be
1031 increase. In client, we should update cl_avail_grant only there
1032 is OBD_MD_FLGRANT in the reply.
1036 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1037 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1038 described in CVE-2007-4573. This problem could allow untrusted
1039 local users to gain root access.
1042 Frequency : when using O_DIRECT and quotas
1044 Description: Incorrect file ownership on O_DIRECT output files
1045 Details : block usage reported by 'lfs quota' does not take into account
1046 files that have been written with O_DIRECT.
1051 Description: touch file failed when fs is not full
1052 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1053 otherwise we can get ENOSP while fs is not full.
1057 Description: parallel lock callbacks
1058 Details : Instead of sending blocking and completion callbacks as separated
1059 requests, adding them to a set and sending in parallel.
1062 Frequency : only for Cray XT3
1063 Bugzilla : 12829/13455
1064 Description: Changing primary group doesn't change the group lustre assigns to
1066 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1067 overridden with the primary group provided by the group upcall,
1068 whereas the supplied fsgid can be trusted if it is in the list of
1069 supplementary groups returned by the group upcall.
1071 Severity : enhancement
1073 Description: Allow masking D_WARNING, D_ERROR messages from console
1074 Details : Console messages can now be disabled via lnet.printk.
1078 Description: User code with malformed file open parameter crashes client node
1079 Details : Before packing join_file req, all the related reference should be
1080 checked carefully in case some malformed flags cause fake
1081 join_file req on client.
1085 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1086 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1087 mode is equal to the granted mode. However, it should also check
1088 that LDLM_FL_CP_REQD is not set.
1092 Description: Heavy nfs access might result in deadlocks
1093 Details : After ELC code landed, it is now improper to enqueue any mds
1094 locks under och_sem, because enqueue might want to decide to
1095 cancel open locks for same inode we are holding och_sem for.
1099 Description: Client eviction while running blogbench
1100 Details : A lot of unlink operations with concurrent I/O can lead to a
1101 deadlock causing evictions. To address the problem, the number of
1102 oustanding OST_DESTROY requests is now throttled to
1103 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1104 callbacks are processed in priority.
1108 Description: enable ACLs on MDS by default
1109 Details : ACLs must be enabled on MDS by default.
1112 Frequency : PPC/PPC64 only
1114 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1115 Details : fix duplicated definitions between asm-ppc64/types.h and
1116 lustre_types.h on PPC.
1119 Frequency : PPC/PPC64 only
1121 Description: asm-ppc/segment.h does not exist
1122 Details : fix compile issue on PPC.
1126 Description: better handle error messages in extents code
1129 Frequency : RHEL4 only
1131 Description: mkfs is very slow on IA64/RHEL4
1132 Details : A performance regression has been discovered in the MPT Fusion
1133 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1134 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1135 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1137 Severity : enhancement
1139 Description: SNMP support enhancement
1140 Details : Adding total number of sampled request for an MDS node in snmp
1143 Severity : enhancement
1145 Description: Optimize ldlm waiting list processing for PR extent locks
1146 Details : When processing waiting list for read extent lock and meeting read
1147 lock that is same or wider to it that is not contended, skip
1148 processing rest of the list and immediatelly return current
1149 status of conflictness, since we are guaranteed there are no
1150 conflicting locks in the rest of the list.
1154 Description: Time out and refuse to reconnect
1155 Details : When the failover node is the primary node, it is possible
1156 to have two identical connections in imp_conn_list. We must
1157 compare not conn's pointers but NIDs, otherwise we can defeat
1158 connection throttling.
1162 Description: port llog fixes from b1_6 into HEAD
1163 Details : Port llog reference couting and some llog cleanups from b1_6
1164 (bug 10800) into HEAD, for protect from panic and access to already
1165 free llog structures.
1169 Description: Detect stride IO mode in read-ahead
1170 Details : When a client does stride read, read-ahead should detect that and
1171 read-ahead pages according to the detected stride pattern.
1175 Description: data checksumming impacts single node performance
1176 Details : add support for several checksum algorithm. Currently, only CRC32
1177 and Adler-32 are supported. The checksum type can be changed on
1178 the fly via /proc/fs/lustre/osc/*/checksum_type.
1182 Description: use adler32 for page checksums
1183 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1188 Description: build for x2 fails
1189 Details : fix compile issue on Cray systems.
1193 Description: Properly match for duplicate locks
1194 Details : Due to different lock order from skiplists code, we need to
1195 traverse entire list for now
1198 Frequency : only on PPC/SLES10
1200 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1201 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1202 be defined. Add a hack in mkfs_lustre.c to work around this compile
1207 Description: LASSERT on MDS when client holding flock lock dies
1208 Details : ldlm pool logic depends on number of granted locks equal to
1209 number of released locks which is not true for flock locks, so
1210 just exclude such locks from consideration.
1214 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1215 Details : Use fsfilt_sync() for both the whole filesystem sync and
1216 individual file sync to eliminate dangerous inode locking
1217 with I_LOCK that can lead to a deadlock.
1221 Description: performance in 1.6.3
1222 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1223 sure that 1MB requests are not fragmented by the block layer.
1225 Severity : enhancement
1227 Description: organize the server-side client stats on per-nid basis
1228 Details : Change the structure of stats under obdfilter and mds to
1236 The "uuid"s file would list the uuids of _active_ exports.
1237 And the clear entry is to clear all stats and stale nids.
1239 Severity : enhancement
1241 Description: eliminate client locks in face of contention
1242 Details : file contention detection and lockless i/o implementation
1243 for contended files.
1247 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1250 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1252 Description: MDS may lose file striping (and hence file data) in some cases
1253 Details : If there are additional extended attributes stored on the MDS,
1254 in particular ACLs, SELinux, or user attributes (if user_xattr
1255 is specified for the client mount options) then there is a risk
1256 of attribute loss. Additionally, the Lustre file striping
1257 needs to be larger than default (e.g. striped over all OSTs),
1258 and an additional attribute must be stored initially in the
1259 inode and then increase in size enough to be moved to the
1260 external attribute block (e.g. ACL growing in size) for file
1265 Description: skiplist implementation simplification
1266 Details : skiplists are used to group compatible locks on granted list
1267 that was implemented as tracking first and last lock of each
1268 lock group the patch changes that to using doubly linked lists
1272 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1273 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1274 because LASSERT on both the data supplied by a client, and the
1275 data on disk is dangerous and incorrect.
1277 Severity : enhancement
1279 Description: Slow truncate/writes to huge files at high offsets.
1280 Details : Directly associate cached pages to lock that protect those pages,
1281 this allows us to quickly find what pages to write and remove
1282 once lock callback is received.
1286 Description: more ldlm soft lockups
1287 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1288 starve other threads from the resource lock for a long time in
1289 case of long waiting queue, so change the debug level from
1290 D_OTHER to the less frequently used D_INFO.
1292 Severity : enhancement
1294 Description: add -gid, -group, -uid, -user options to lfs find
1298 Description: Hung threads in invalidate_inode_pages2_range
1299 Details : The direct IO path doesn't call check_rpcs to submit a new RPC once
1300 one is completed. As a result, some RPCs are stuck in the queue
1305 Description: filter threads hungs on waiting journal commit
1306 Details : Cleanup filter group llog code, then only filter group llog will
1307 be only created in the MDS/OST syncing process.
1311 Description: Procfs and llog threads access destoryed import sometimes.
1312 Details : Sync the import destoryed process with procfs and llog threads by
1313 the import refcount and semaphore.
1315 Severity : enhancement
1317 Description: openlock cache of b1_6 port to HEAD
1322 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1323 Details : If insertion of an extent fails, then discard the inode
1324 preallocation and free data blocks else it can lead to duplicate
1329 Description: don't always update ctime in ext3_xattr_set_handle()
1330 Details : Current xattr code updates the inode ctime in ext3_xattr_set_handle.
1331 In some cases the ctime should not be updated, for example for
1332 2.0->1.8 compatibility it is necessary to delete an xattr and it
1333 should not update the ctime.
1337 Bugzilla : 15713/16362
1338 Description: Assertion in iopen_connect_dentry in 1.6.3
1339 Details : looking up an inode via iopen with the wrong generation number can
1340 populate the dcache with a disconneced dentry while the inode
1341 number is in the process of being reallocated. This causes an
1342 assertion failure in iopen since the inode's dentry list contains
1343 both a connected and disconnected dentry.
1347 Description: assertion failure in ldlm_handle2lock()
1348 Details : fix a race between class_handle_unhash() and class_handle2object()
1349 introduced in lustre 1.6.5 by bug 13622.
1354 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1355 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1360 Description: oops in sd_iostats_seq_show()
1361 Details : unloading/reloading the scsi low level driver triggers a kernel
1362 bug when trying to access the sd iostat file.
1367 Description: Kernel panics during QLogic driver reload
1368 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1369 patch, causing memory corruption.
1374 Description: journal_dev option does not work in b1_6
1375 Details : pass mount option during pre-mount.
1377 Severity : enhancement
1379 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1380 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1381 extent information of a file. It can be used to map logical blocks
1382 in a file to physical blocks in the block device.
1386 Description: LDLM soft lockups - improvement
1387 Details : It is be possible to send the lock handle along with each read
1388 or write request because the client is already doing a lock match
1389 itself so there isn't any reason the OST should have to re-do that
1395 Description: X2 build failures
1396 Details : fix build failures on Cray X2.
1398 --------------------------------------------------------------------------------
1400 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
1402 * Support for kernels:
1403 2.6.5-7.283 (SLES 9),
1404 2.6.9-55.EL (RHEL 4),
1405 2.6.16.46-0.14 (SLES 10),
1406 2.6.18.8 vanilla (kernel.org)
1407 * Client support for unpatched kernels:
1408 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1409 2.6.16 - 2.6.22 vanilla (kernel.org)
1410 * Due to recently discovered recovery problems, we do not recommend
1411 using patchless RHEL 4 clients with this or any earlier release.
1412 * Recommended e2fsprogs version: 1.39.cfs8
1413 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1414 * Starting with this release, the ldiskfs backing filesystem required
1415 by Lustre is now in its own package, lustre-ldiskfs. This package
1416 should be installed. It is versioned separately from Lustre and
1417 may be released separately in future.
1421 Description: block reactivating mgc import until all deactivates complete
1422 Details : Fix race when failing back MDT/MGS to itself (testing)
1425 Frequency : at statup only
1427 Description: mds_lov_synchronize race leads to various problems
1428 Details : simultaneous MDT->OST connections at startup can cause the
1429 sync to abort, leaving the OSC in a bad state.
1431 Severity : enhancement
1433 Description: add optional extra BUILD_VERSION info
1434 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1435 to override the lustre version.
1438 Frequency : 2.6.18 servers only
1440 Description: ll_kern_mount() doesn't release the module reference
1441 Details : The ldiskfs module reference count never drops down to 0
1442 because ll_kern_mount() doesn't release the module reference.
1447 Description: server LBUG when using old ost_num_threads parameter
1448 Details : Accept the old ost_num_threads parameter but warn that it
1449 is deprecated, and fix an off-by-one error that caused an LBUG.
1454 Description: Transient SCSI error results in persistent IO issue
1455 Details : iobuf->dr_error is not reinitialized to 0 between two
1459 Frequency : sometimes when underlying device returns I/O errors
1461 Description: OSTs not going read-only during write failures
1462 Details : OSTs are not remounted read-only when the journal commit threads
1463 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1464 instead of the ext3 wrappers.
1468 Description: poor connect scaling with increasing client count
1469 Details : Don't run filter_grant_sanity_check for more than 100 exports
1470 to improve scaling for large numbers of clients.
1473 Frequency : SLES10 only
1475 Description: sanity-quota.sh quotacheck failed: rc = -22
1476 Details : Quotas cannot be enabled on SLES10.
1479 Frequency : liblustre clients only
1481 Description: getdirentries does not give error when run on compute nodes
1482 Details : getdirentries does not fail when the size specified as an argument
1483 is too small to contain at least one entry
1485 Severity : enhancement
1487 Description: Add LNET router traceability for debug purposes
1488 Details : If a checksum failure occurs with a router as part of the
1489 IO path, the NID of the last router that forwarded the bulk data
1490 is printed so it can be identified.
1495 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1496 Details : Due to a race condition, liblustre clients were occasionally
1497 evicted incorrectly.
1499 Severity : enhancement
1501 Description: lfs setstripe use optional parameters instead of postional
1504 Severity : enhancement
1506 Description: Nanosecond timestamp support for ldiskfs
1507 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1508 resolution timestamps. There is not yet support for this at
1509 the Lustre filesystem level.
1512 Frequency : during server recovery
1514 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1515 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1519 Frequency : nfs export on patchless client
1521 Description: connectathon hang when test nfs export over patchless client
1522 Details : Disconnected dentry cannot be found with lookup, so we do not need
1523 to unhash it or make it invalid
1526 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1527 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1528 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1533 Description: open req refcounting wrong on reconnect
1534 Details : If reconnect happened between getting open reply from server and
1535 call to mdc_set_replay_data in ll_file_open, we will schedule
1536 replay for unreferenced request that we are about to free.
1537 Subsequent close will crash in variety of ways.
1538 Check that request is still eligible for replay in
1539 mdc_set_replay_data().
1544 Description: disable writes to filesystem when reading health_check file
1545 Details : the default for reading the health_check proc file has changed
1546 to NOT do a journal transaction and write to disk, because this
1547 can cause reads of the /proc file to hang and block HA state
1548 checking on a healthy but otherwise heavily loaded system. It
1549 is possible to return to the previous behaviour during configure
1550 with --enable-health-write.
1552 Severity : enhancement
1554 Description: 64-bit inode version
1555 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1556 to the inode. This will be required for version-based recovery.
1561 Description: MDS fails to start if a duplicate client export is detected
1562 Details : in some rare cases it was possible for a client to connect to
1563 an MDS multiple times. Upon recovery the MDS would detect this
1564 and fail during startup. Handle this more gracefully.
1566 Severity : enhancement
1568 Description: Add -o localflock option to simulate old noflock
1570 Details : This will achieve local-only flock/fcntl locks
1576 Description: log_commit_thread vs filter_destroy race leads to crash
1577 Details : Take import reference before releasing llog record semaphore
1582 Description: Wrong request locking in request set processing
1583 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1584 imp_delayed_list, in this place should be used imp_lock.
1587 Frequency : when reconnection
1589 Description: Grant Leak when osc reconnect to OST
1590 Details : When osc reconnect ost, OST(filter) should check whether it
1591 should grant more space to client by comparing fed_grant and
1592 cl_avail_grant, and return the granted space to client instead
1593 of "new granted" space, because client will call osc_init_grant
1594 to update the client grant space info.
1597 Frequency : when client reconnect to OST
1599 Description: Grant Leak when osc do resend and replay bulk write
1600 Details : When osc reconnect to OST, OST(filter)should clear grant info of
1601 bulk write request, because the grant info while be sync between
1602 OSC and OST when reconnect, and we should ignore the grant info
1603 these of resend/replay write req.
1608 Description: Grant space more than avaiable left space sometimes.
1609 Details : When then OST is about to be full, if two bulk writing from
1610 different clients came to OST. Accord the avaliable space of the
1611 OST, the first req should be permitted, and the second one
1612 should be denied by ENOSPC. But if the seconde arrived before
1613 the first one is commited. The OST might wrongly permit second
1614 writing, which will cause grant space > avaiable space.
1617 Frequency : when client is evicted
1619 Description: Grant might be wrongly erased when osc is evicted by OST
1620 Details : when the import is evicted by server, it will fork another
1621 thread ptlrpc_invalidate_import_thread to invalidate the
1622 import, where the grant will be set to 0. While the original
1623 thread will update the grant it got when connecting. So if
1624 the former happened latter, the grant will be wrongly errased
1625 because of this race.
1630 Description: Checking Stale with correct fid
1631 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1632 to check whether it is stale, because sometimes, we want the
1633 enqueue happened anyway, and op_data.fid2 will not be initialized.
1635 Severity : enhancement
1637 Description: update patchless client
1638 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1641 Frequency : only with 2.4 kernel
1643 Description: random memory corruption
1644 Details : size of struct ll_inode_info is to big for union inode.u and this
1645 can be cause of random memory corruption.
1650 Description: Memory leak in recovery
1651 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1652 It should also check obd_fail before fsfilt_start, otherwise if
1653 fsfilt_start return -EROFS,(failover mds during mds recovery).
1654 then the req will return with repmsg->transno = 0 and rc = EROFS.
1655 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1656 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1657 be freed no matter whether fsfilt_commit success or not.
1660 Frequency : only with huge count clients
1662 Description: Prevents from taking the superblock lock in llap_from_page for
1664 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1665 for llap_from_page call in ll_removepage prevents from taking the
1666 superblock lock for a soon died page.
1671 Description: Not check open intent error before release open handle
1672 Details : in some rare cases, the open intent error is not checked before
1673 release open handle, which may cause
1674 ASSERTION(open_req->rq_transno != 0), because it tries to release
1675 the failed open handle.
1680 Description: Set cat log bitmap only after create log success.
1681 Details : in some rare cases, the cat log bitmap is set too early. and it
1682 should be set only after create log success.
1686 Description: Accessing a block bevice can re-enable I/O when Lustre is
1687 tearing down a device.
1688 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1693 Description: service threads may hog cpus when there are a lot of requests
1694 Details : Insert cond_resched to give other threads a chance to use some CPU
1699 Description: the cat log was not initialized in recovery
1700 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1701 unlink log on mds will not be initialized until mds post
1702 recovery. And also in mds post recovery, the unlink log will
1703 initialization will be done asynchronausly, so there will be race
1704 between add unlink log and unlink log initialization.
1708 Description: brw_stats were being printed incorrectly
1709 Details : brw_stats were being printed as log2 but all of them were not
1710 recorded as log2. Also remove some code duplication arising from
1711 filter_tally_{read,write}.
1715 Frequency : rare, only in recovery.
1716 Description: ASSERTION(req->rq_type != LI_POISON) failed
1717 Details : imp_lock should be held while iterating over imp_sending_list for
1718 prevent destroy request after get timeout in ptlrpc_queue_wait.
1722 Description: replay-single.sh test 52 fails
1723 Details : A lock's skiplist need to be cleanup when it being unlinked
1724 from its resource list.
1728 Description: Short directio read returns full requested size rather than
1730 Details : Direct I/O operations should return actual amount of bytes
1731 transferred rather than requested size.
1733 Severity : enhancement
1735 Description: metadata RPC reduction (e.g. for rm performance)
1736 Details : decrease the amount of synchronous RPC between clients and servers
1737 by canceling conflicing lock before the operation on the client
1738 and packing thier handles into the main operation RPC to server.
1740 Severity : enhancement
1742 Description: Async OSC create to avoid the blocking unnecessarily.
1743 Details : If a OST has no remain object, system will block on the creating
1744 when need to create a new object on this OST. Now, ways use
1745 pre-created objects when available, instead of blocking on an
1746 empty osc while others are not empty. If we must block, we block
1747 for the shortest possible period of time.
1751 Description: improve handling recoverable errors
1752 Details : if request processig with error which can be recoverable on server
1753 request should be resend, otherwise page released from cache and
1756 Severity : enhancement
1758 Description: refine locking for avoid write wrong info into lov_objid file
1759 Details : fix possible races with add new target and write/update data in
1762 --------------------------------------------------------------------------------
1764 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1769 Frequency : on some architectures
1771 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1772 Details : Uninitialized flags sometimes cause configuration commands to
1776 Frequency : patchless clients only
1778 Description: missing __iget() symbol export
1779 Details : The __iget() symbol export is missing. To avoid the need for
1780 this on patchless clients the deathrow inode reaper is turned
1781 off, and we depend on the VM to clean up old inodes. This
1782 dependency was during via the fix for bug 12181.
1784 --------------------------------------------------------------------------------
1786 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1788 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1789 INTEROPERATE with older versions automatically. In many cases a
1790 special upgrade step is needed. Please read the
1791 user documentation before upgrading any part of a live system.
1792 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1793 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1794 * WARNING: Lustre configuration and startup changes are required with
1795 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1801 Severity : enhancement
1803 Description: MountConf
1804 Details : Lustre configuration is now managed via mkfs and mount
1805 commands instead of lmc and lconf. New obd types (MGS, MGC)
1806 are added for dynamic configuration management. See
1807 https://mail.clusterfs.com/wikis/lustre/MountConf for
1810 Severity : enhancement
1812 Description: dynamic OST addition
1813 Details : OSTs can now be added to a live filesystem
1815 Severity : enhancement
1817 Description: startup order invariance
1818 Details : MDTs and OSTs can be started in any order. Clients only
1819 require the MDT to complete startup.
1821 Severity : enhancement
1823 Description: parallel, asynchronous orphan cleanup
1824 Details : orphan cleanup is now performed in separate threads for each
1825 OST, allowing parallel non-blocking operation.
1827 Severity : enhancement
1829 Description: optimized stripe assignment
1830 Details : stripe assignments are now made based on ost space available,
1831 ost previous usage, and OSS previous usage, in order to try
1832 to optimize storage space and networking resources.
1834 Severity : enhancement
1836 Description: Permanently set tunables
1837 Details : All writable /proc/fs/lustre tunables can now be permanently
1838 set on a per-server basis, at mkfs time or on a live system.
1840 Severity : enhancement
1842 Description: Lustre message v2
1843 Details : Add lustre message format v2.
1845 Severity : enhancement
1847 Description: client OST exclusion list
1848 Details : Clients can be started with a list of OSTs that should be
1849 declared "inactive" for known non-responsive OSTs.
1853 Description: SPEC SFS validation failure on NFS v2 over lustre.
1854 Details : Changes the blocksize for regular files to be 2x RPC size,
1855 and not depend on stripe size.
1857 Severity : enhancement
1859 Description: Multiple MD RPCs in flight.
1860 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1861 To avoid overly-overloading MDS, introduce a limit on number of
1862 MDS RPCs in flight for a single client and add /proc controls
1863 to adjust this limit.
1865 Severity : enhancement
1867 Description: client read/write statistics
1868 Details : Add client read/write call usage stats for performance
1869 analysis of user processes.
1870 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1871 file access. extents_stats shows chunk size distribution.
1872 extents_stats_per_process show chunk size distribution per
1875 Severity : enhancement
1877 Description: mds statistics
1878 Details : Add detailed mds operations statistics in
1879 /proc/fs/lustre/mds/*/stats.
1883 Description: Failure of copying files with lustre special EAs.
1884 Details : Client side always return success for setxattr call for lustre
1885 special xattr (currently only "trusted.lov").
1890 Description: Refcount LNET uuids
1891 Details : The global LNET uuid list grew linearly with every startup;
1892 refcount repeated list entries instead of always adding to
1895 Severity : enhancement
1897 Description: Dynamic service threads
1898 Details : Within a small range, start extra service threads
1899 automatically when the request queue builds up.
1902 Frequency : mixed-endian client/server environments
1904 Description: mixed-endian crashes
1905 Details : The new msg_v2 system had some failures in mixed-endian
1908 Severity : enhancement
1910 Description: Easy OST removal
1911 Details : OSTs can be permanently deactivated with e.g. 'lctl
1912 conf_param lustre-OST0001.osc.active=0'
1914 Severity : enhancement
1916 Description: MGS proc entries
1917 Details : Added basic proc entries for the MGS showing what filesystems
1920 Severity : enhancement
1922 Description: provide MGS failover
1923 Details : Added config lock reacquisition after MGS server failover.
1925 Severity : enhancement
1927 Description: add Linux 2.4 support
1928 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1932 Description: a large application tries to do I/O to the same resource and dies
1933 in the middle of it.
1934 Details : Check the req->rq_arrival time after the call to
1935 ost_brw_lock_get(), but before we do anything about
1936 processing it & sending the BULK transfer request. This
1937 should help move old stale pending locks off the queue as
1938 quickly as obd_timeout.
1941 Frequency : when an incorrect nid is specified during startup
1943 Description: ptlrpc connect to non-existant node causes kernel crash
1944 Details : LNET can't be re-entered from an event callback, which
1945 happened when we expire a message after the export has been
1946 cleaned up. Instead, hand the zombie cleanup off to another
1949 Severity : enhancement
1951 Description: plain/inodebits lock performance improvement
1952 Details : Grouping plain/inodebits in granted list by their request modes
1953 and bits policy, thus improving the performance of search through
1957 Frequency : only if OST filesystem is corrupted
1959 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1960 Details : for a short time RPCs with bulk IO are in the replay list,
1961 but replay of bulk IOs is unimplemented. If the OST filesystem
1962 is corrupted due to disk cache incoherency and then replay is
1963 started it is possible to trip an assertion. Avoid putting
1964 committed RPCs into the replay list at all to avoid this issue.
1967 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1969 Description: System hang on startup
1970 Details : This bug allowed the liblustre (e.g. catamount) client to
1971 return to the app before handling all startup RPCs. This
1972 could leave the node unresponsive to lustre network traffic
1973 and manifested as a server ptllnd timeout.
1975 Severity : enhancement
1977 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1978 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1979 boolean to control whether to print peer debug info when a
1980 client's RPC times out.
1983 Frequency : only for kernels with patches from Lustre below 1.4.3
1985 Description: Remove old rdonly API
1986 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1989 Frequency : only for devices with external journals
1991 Description: Set external device read-only also
1992 Details : During a commanded failover stop, we set the disk device
1993 read-only while the server shuts down. We now also set any
1994 external journal device read-only at the same time.
1997 Frequency : when upgrading from 1.4 while trying to change parameters
1999 Description: The wrong (new) MDC name was used when setting parameters for
2000 upgraded MDT's. Also allows changing of OSC (and MDC)
2001 parameters if --writeconf is specified at tunefs upgrade time.
2004 Frequency : when setting specific ost indicies
2006 Description: QOS code breaks on skipped indicies
2007 Details : Add checks for missing OST indicies in the QOS code, so OSTs
2008 created with --index need not be sequential.
2010 Severity : enhancement
2012 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2013 Details : The uninit_groups feature works in conjunction with the kernel
2014 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2015 up the pass1 processing of e2fsck. This is a read-only feature
2016 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2017 mount filesystems that have had this feature enabled.
2019 Severity : enhancement
2021 Description: Improve multi-block allocation algorithm to avoid fragmentation
2022 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2023 allocation locality and avoid filesystem fragmentation.
2025 ------------------------------------------------------------------------------
2027 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
2029 * Support for kernels:
2030 2.4.21-47.0.1.EL (RHEL 3)
2031 2.6.5-7.283 (SLES 9)
2032 2.6.9-42.0.10.EL (RHEL 4)
2033 2.6.12.6 vanilla (kernel.org)
2034 2.6.16.27-0.9 (SLES 10)
2035 * Recommended e2fsprogs version: 1.39.cfs5
2037 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2041 Frequency : occasional, depends on client load and configuration
2042 Bugzilla : 12181, 12203
2043 Description: data loss for recently-modified files
2045 Details : In some cases it is possible that recently written or created
2046 files may not be written to disk in a timely manner (this should
2047 normally be within 30s unless client IO load is very high).
2048 The problem appears as zero-length files or files that are a
2049 multiple of 1MB in size after a client crash or client eviction
2050 that are missing data at the end of the file.
2052 This problem is more likely to be hit on clients where files are
2053 repeatedly created and unlinked in the same directory, clients
2054 have a large amount of RAM, have many CPUs, the filesystem has
2055 many OSTs, the clients are rebooted frequently, and/or the files
2056 are not accessed by other nodes after being written.
2058 The presence of the problem can be detected by looking at
2059 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2060 smaller than the second (nr_unused) then dirty files will not
2061 be flushed automatically to disk. "sync; sleep 10" should be
2062 run several times on the node before unmounting it to update
2063 Lustre (this is also safe to run on nodes without this problem).
2065 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2066 that can cause this same problem, so customers using that kernel
2067 also need to update the kernel in addition to Lustre. In order
2068 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2070 It is normal that files written just before a client crash (less
2071 than 30s) may not yet have been flushed to disk, even for local
2075 Frequency : frequent on thin XT3 nodes
2077 Description: UUID collision on thin XT3 Linux nodes
2078 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2079 randomly, since we relied on an insufficiently-seeded PRNG.
2084 Description: OSS hangs after "All ost request buffers busy"
2085 Details : A deadlock between quota and journal operations caused OSS
2086 hangs after printing "All ost request buffers busy."
2089 Frequency : always on liblustre builds
2091 Description: Cleanup compiler warnings on liblustre
2094 Frequency : always on liblustre builds on XT3
2096 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2100 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2101 > 32000 subdirectories
2102 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2103 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2104 subdirectory count crosses 32000. This will aid e2fsck to
2105 correctly handle more than 32000 subdirectories.
2108 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2110 Description: System hang on startup
2111 Details : This bug allowed the liblustre (e.g. catamount) client to
2112 return to the app before handling all startup RPCs. This
2113 could leave the node unresponsive to lustre network traffic
2114 and manifested as a server ptllnd timeout.
2116 Severity : enhancement
2118 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2119 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2120 boolean to control whether to print peer debug info when a
2121 client's RPC times out.
2126 Description: make O_SYNC working on 2.6 kernels
2127 Details : 2.6 kernels use different method for mark pages for write,
2128 so need add a code to lustre for O_SYNC work.
2133 Description: Failure to close file and release space on NFS
2134 Details : Put inode details into lock acquired in ll_intent_file_open.
2135 Use mdc_intent_lock in ll_intent_open to properly
2136 detect all kind of errors unhandled by mdc_enqueue.
2141 Description: proc file read during shutdown sometimes raced obd removal,
2143 Details : Add lock to prevent obd access after proc file removal.
2146 Frequency : Only for files larger than 4GB on 32-bit clients.
2148 Description: improperly doing page alignment of locks
2149 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2150 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2155 Description: RPCs being resent when they shouldn't be
2156 Details : Some RPCs that should not be resent are being resent. This
2157 can cause inconsistencies in the RPC state machine. Do not
2158 resend such requests.
2161 Frequency : rare, only with NFS export
2163 Description: Crash on NFS re-export node
2164 Details : under very unusual load conditions an assertion is hit in
2165 ll_intent_file_open()
2168 Frequency : only if OST filesystem is corrupted
2170 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2171 Details : for a short time RPCs with bulk IO are in the replay list,
2172 but replay of bulk IOs is unimplemented. If the OST filesystem
2173 is corrupted due to disk cache incoherency and then replay is
2174 started it is possible to trip an assertion. Avoid putting
2175 committed RPCs into the replay list at all to avoid this issue.
2180 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2181 Details : Large single O_DIRECT read and write calls can fail to allocate
2182 a sufficiently large buffer to process the request. In case of
2183 allocation failure the allocation is retried with a smaller
2184 buffer and broken into smaller requests.
2186 Severity : enhancement
2188 Description: Add -o localflock option to simulate old noflock behaviour.
2189 Details : This will achieve local-only flock/fcntl locks coherentness.
2194 Description: versioning check is incomplete
2195 Details : Checking the version difference of client vs. server, report
2196 error if the gap is too big.
2201 Description: add support PG_writeback bit
2202 Details : add support for PG_writeback bit for Lustre, for more carefull
2203 work with page cache in 2.6 kernel. This also fix some deadlocks
2204 and remove hack for work O_SYNC with 2.6 kernel.
2206 Severity : enhancement
2208 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2209 Details : The uninit_groups feature works in conjunction with the kernel
2210 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2211 up the pass1 processing of e2fsck. This is a read-only feature
2212 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2213 mount filesystems that have had this feature enabled.
2215 Severity : enhancement
2217 Description: Improve multi-block allocation algorithm to avoid fragmentation
2218 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2219 allocation locality and avoid filesystem fragmentation.
2221 ------------------------------------------------------------------------------
2223 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2225 * Support for kernels:
2226 2.6.9-42.0.3.EL (RHEL 4)
2227 2.6.5-7.276 (SLES 9)
2228 2.4.21-47.0.1.EL (RHEL 3)
2229 2.6.12.6 vanilla (kernel.org)
2230 2.6.16.21-0.8 (SLES10)
2231 * Recommended e2fsprogs version: 1.39.cfs2-0
2233 * The backwards-compatible /proc/sys/portals symlink has been removed
2234 in this release. Before upgrading, please ensure that you change
2235 any configuration scripts or /etc/sysctl.conf files that access
2236 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2237 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2238 in advance of the upgrade on any system running Lustre 1.4.6 or
2239 newer, since /proc/sys/lnet was added in that version.
2240 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2244 Frequency : only when quota is used
2246 Description: avoid scanning export list for quota master
2247 Details : Change the algorithms to avoid scanning export list in order
2248 to improve the efficiency.
2251 Frequency : MDS failover only, very rarely
2253 Description: "went back in time" messages on mds failover
2254 Details : The greatest transno may be lost when the current operation
2255 finishes with an error (transno==0) and the client's last_rcvd
2256 record is over-written. Save the greatest transno in the
2257 mds_last_transno for this case.
2260 Frequency : always for specific kernels and striping counts
2262 Description: client may get "Matching packet too big" without ACL support
2263 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2264 when trying to access files in certain configurations. The
2265 clients should in fact be denied when mounting because they do
2266 not understand ACLs.
2269 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2271 Description: many clients connecting with IO in progress causes connect timeouts
2272 Details : Avoid synchronous journal commits to avoid delays caused by many
2273 clients connecting/disconnecting when bulk IO is in progress.
2274 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2275 OST_IO_PORTAL to avoid delays behind potentially many pending
2279 Frequency : occasionally with multiple writers to a single file
2281 Description: shared writes to file may result in wrong size reported by stat()
2282 Details : Allow growing of kms when extent lock is cancelled
2285 Frequency : always with random mmap IO to multi-striped file
2287 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2288 Details : Only if the hole is at the end of OST object so that kms is too
2289 small. Fix is to increase kms accordingly in ll_nopage.
2292 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2294 Description: writes to a missing object would leak memory on the OST
2295 Details : If there is an inconsistency between the MDS and OST filesystems,
2296 such that the MDS references an object that doesn't exist, writes
2297 to that object will leak memory due to incorrect cleanup in the
2298 error handling path, eventually running out of memory on the OST.
2303 Description: Creating too long symlink causes lustre errors
2304 Details : Check symlink and name lengths before sending requests to MDS.
2307 Frequency : only if flock is enabled (not on by default)
2309 Description: posix locks not released on fd closure on 2.6.9+
2310 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2311 kernels, this caused such locks not to be released on fd close and
2312 then assertions on fs unmount about still used locks.
2315 Frequency : MDS failover only, very rarely
2317 Description: clients may get ASSERTION(granted_lock != NULL)
2318 Details : When request was taking a long time, and a client was resending
2319 a getattr by name lock request. The were multiple lock requests
2320 with the same client lock handle and
2321 mds_getattr_name->fixup_handle_for_resent_request found one of the
2322 lock handles but later failed with ASSERTION(granted_lock != NULL).
2327 Description: handle->h_buffer_credits > 0, assertion failure
2328 Details : h_buffer_credits is zero after truncate, causing assertion
2329 failure. This patch extends the transaction or creates a new
2333 Frequency : NFS re-export or patchless client
2334 Bugzilla : 11179, 10796
2335 Description: Crash on NFS re-export node (__d_move)
2336 Details : We do not want to hash the dentry if we don't have a lock.
2337 But if this dentry is later used in d_move, we'd hit uninitialised
2338 list head d_hash, so we just do this to init d_hash field but
2339 leave dentry unhashed.
2342 Frequency : NFS re-export or patchless client
2344 Description: NFS exports has problem with symbolic link
2345 Details : lustre client didn't properly install dentry when re-exported
2346 to NFS or running patchless client.
2349 Frequency : NFS re-export or patchless client
2351 Description: Various nfs/patchless fixes.
2352 Details : fixes reuse disconected alias for lookup process - this fixes
2353 warning "find_exported_dentry: npd != pd",
2354 fix permission error with open files at nfs.
2355 fix apply umask when do revalidate.
2358 Frequency : occasional
2360 Description: Crash on NFS re-export node
2361 Details : calling clear_page() on the wrong pointer triggered oops in
2362 generic_mapping_read().
2365 Frequency : rarely, using O_DIRECT IO
2367 Description: unaligned directio crashes client with LASSERT
2368 Details : check for unaligned buffers before trying any requests.
2371 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2373 Description: stale data returned from RAID cache
2374 Details : If only a small amount of IO is done to the RAID device before
2375 reading it again it is possible to get stale data from the RAID
2376 cache instead of reading it from disk.
2379 Frequency : always for sles10 kernel
2381 Description: sles10 support
2382 Details : ll_follow_link: compile fixes and using of nd_set_link
2383 under newer kernels.
2386 Frequency : depends on arch, kernel and compiler version, always on sles10
2389 Description: recursive or deep enough symlinks cause stack overflow
2390 Details : getting rid of large stack-allocated variable in
2394 Frequency : depends on hardware
2396 Description: lustre write performance loss in the SLES10 kernel
2397 Details : the performance loss is caused by using of write barriers in the
2398 ext3 code. The SLES10 kernel turns barrier support on by
2399 default. The fix is to undo that change for ldiskfs.
2401 ------------------------------------------------------------------------------
2403 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2405 * Support for kernels:
2406 2.6.9-42.0.3EL (RHEL 4)
2407 2.6.5-7.276 (SLES 9)
2408 2.4.21-47.0.1.EL (RHEL 3)
2409 2.6.12.6 vanilla (kernel.org)
2413 Frequency : quota enabled and large files being deleted
2415 Description: releasing more than 4GB of quota at once hangs OST
2416 Details : If a user deletes more than 4GB of files on a single OST it
2417 will cause the OST to spin in an infinite loop. Release
2418 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2423 Description: statfs data retrieved from /proc may be stale or zero
2424 Details : When reading per-device statfs data from /proc, in the
2425 {kbytes,files}_{total,free,avail} files, it may appear
2426 as zero or be out of date.
2429 Frequency : systems with MD RAID1 external journal devices
2431 Description: lconf's call to blkid is confused by RAID1 journal devices
2432 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2433 of returning all block devices that match the journal UUID.
2436 Frequency : always, for aggregate stripe size over 4GB
2438 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2439 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2440 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2443 Frequency : always if "lfs find" used on a local file/directory
2445 Description: "lfs find" segfaults if used on a local file/directory
2446 Details : The case where a directory component was not specified wasn't
2447 handled correctly. Handle this properly.
2450 Frequency : always on ppc64
2452 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2453 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2454 perform endianness conversion.
2457 Frequency : rarely (truncate to non-zero file size after write under load)
2458 Bugzilla : 10730, 10687
2459 Description: Files padded with zeros to next 4K multiple
2460 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2461 it is possible that files that are truncated to a non-zero size
2462 immediately after being written are filled with zero bytes beyond
2463 the truncated size. No file data is lost.
2465 Severity : enhancement
2467 Description: Allow recovery/failover for liblustre clients.
2468 Details : liblustre clients were unaware of failover configurations until
2471 Severity : enhancement
2473 Description: user file locks should fail when not mounting with flock option
2474 Details : Set up an error-returning stub in ll_file_operations.lock field
2475 to prevent incorrect behaviour when client is mounted without
2476 flock option. Also, set up properly f_op->flock field for
2480 Frequency : always on ia64
2482 Description: "lfs df" loops on printing out MDS statfs information
2483 Details : The obd_ioctl_data was not initialized and in some systems
2484 this caused a failure during the ioctl that did not return
2485 an error. Initialize the struct and return an error on failure.
2488 Frequency : SLES 9 only
2490 Description: Error of copying files with lustre special EAs as root
2491 Details : Client side always return success for setxattr call for lustre
2492 special xattr (currently only "trusted.lov").
2495 Frequency : rarely on clusters with both ia64+i386 clients
2497 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2498 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2499 clients are doing concurrent writes to the same file it is
2500 possible that the ia64 clients may LASSERT because the OST
2501 extent locks are not PAGE_SIZE aligned. Ensure that grown
2502 locks are always aligned on the request boundary.
2505 Frequency : specific use, occasional
2507 Description: Overwriting in use executable truncates on-disk binary image
2508 Details : If one node attempts to overwrite an executable in use by
2509 another node, we now correctly return ETXTBSY instead of
2510 truncating the file.
2512 Severity : enhancement
2514 Description: Async OSC create to avoid the blocking unnecessarily.
2515 Details : If a OST has no remain object, system will block on the creating
2516 when need to create a new object on this OST. Now, ways use
2517 pre-created objects when available, instead of blocking on an
2518 empty osc while others are not empty. If we must block, we block
2519 for the shortest possible period of time.
2524 Description: chmod on Lustre root is propagated to other clients
2525 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2526 invalid by the follow_mount time.
2529 Frequency : liblustre clients only
2531 Description: Race in 'instant cancel' lock handling could lead to such locks
2532 never to be granted in case of SMP MDS
2533 Details : Do not destroy not yet granted but cbpending locks in
2537 Frequency : replay/resend of open
2539 Description: non null lock assetion failure in mds_intent_policy
2540 Details : Trying to replay/resend lockless open requests resulted in
2541 mds_open() returning 0 with no lock. Now it sets a flag if
2542 a lock is going to be returned.
2544 Severity : enhancement
2546 Description: Checksum enhancements
2547 Details : New checksum enhancements allow for resending RPCs that failed
2550 Severity : enhancement
2552 Description: Tunables on number of dirty pages in cacche
2553 Details : Allow to set limit on number of dirty pages cached.
2558 Description: client crash on unmount - lock still has references
2559 Details : In some error handling cases it was possible to leak a lock
2560 reference on a client while accessing a file. This was not
2561 harmful to the client during operation, but would cause the
2562 client to crash when the filesystem is unmounted.
2565 Frequency : specific case, rare
2567 Description: ETXTBSY on mds though file not in use by client
2568 Details : ETXTBSY is no longer incorrectly returned when attempting to
2569 chmod or chown a directory that the user previously tried to
2570 execute or a currently-executing binary.
2573 Frequency : extremely rare except on liblustre-based clients
2575 Description: Lustre space not freed when files are deleted
2576 Details : Clean up open-unlinked files after client eviction. Previously
2577 the unlink was skipped and the files remained as orphans.
2582 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2583 Details : In some cases it was possible to send a blocking callback to
2584 a client doing a glimpse, even though that client didn't get
2585 a lock granted. When the glimpse lock is cancelled on the OST
2586 the freed lock is left on the waiting list and corrupted the list.
2589 Frequency : all core dumps
2591 Description: Broke core dumps to lustre
2592 Details : Negative dentry may be unhashed if parent does not have UPDATE
2593 lock, but some callers, e.g. do_coredump, expect dentry to be
2594 hashed after successful create, hash it in ll_create_it.
2596 ------------------------------------------------------------------------------
2598 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2600 * Support for kernels:
2601 2.6.9-42.0.2.EL (RHEL 4)
2602 2.6.5-7.276 (SLES 9)
2603 2.4.21-40.EL (RHEL 3)
2604 2.6.12.6 vanilla (kernel.org)
2608 Frequency : always on RHEL 3
2610 Description: Number of open files grows over time
2611 Details : The number of open files grows over time, whether or not
2612 Lustre is started. This was due to a filp leak introduced
2613 by one of our kernel patches.
2615 ------------------------------------------------------------------------------
2617 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2619 * Support for kernels:
2620 2.6.9-42.EL (RHEL 4)
2621 2.6.5-7.276 (SLES 9)
2622 2.4.21-40.EL (RHEL 3)
2623 2.6.12.6 vanilla (kernel.org)
2628 Bugzilla : 5719, 9635, 9792, 9684
2629 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2630 Details : If a server is under heavy load and cannot reply to new
2631 connection requests before the client resends the (re)connect,
2632 the connection handling code can behave badly if two service
2633 threads are concurrently handing separate (re)connections from
2634 the same client. Add better locking to the connection handling
2635 code, and ensure that only a single connection will be processed
2636 for a given client UUID, even if the lock is dropped.
2638 Severity : enhancement
2640 Description: add TCP zero-copy support to kernel
2641 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2642 sends if the hardware supports scatter-gather and checksumming.
2643 This allows socklnd to do client-write and server-read more
2644 efficiently and reduce CPU utilization from skbuf copying.
2647 Frequency : only if NFS exporting from client
2649 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2650 Details : If setting attributes on a file created under NFS that had
2651 never been opened it would be possible to oops the client
2652 if the file had no objects.
2655 Frequency : always for liblustre
2657 Description: liblustre client does MDS+OSTs setattr RPC for each write
2658 Details : When doing a write from a liblustre client, the client
2659 incorrectly issued an RPC to the MDS and each OST the file was
2660 striped over in order to update the timestamps. When writing
2661 with small chunks and many clients this could overwhelm the MDS
2662 with RPCs. In all cases it would slow down the write because
2663 these RPCs are unnecessary.
2665 Severity : enhancement
2667 Description: allow number of MDS service threads to be changed at module load
2668 Details : It is now possible to change the number of MDS service threads
2669 running. Adding "options mds mds_num_threads={N}" to the MDS's
2670 /etc/modprobe.conf will set the number of threads for the next
2671 time Lustre is restarted (assuming the "mds" module is also
2672 reloaded at that time). The default number of threads will
2673 stay the same, 32 for most systems.
2678 Description: OST crash if filesystem is unformatted or corrupt
2679 Details : If an OST is started on a device that has never been formatted
2680 or if the filesystem is corrupt and cannot even mount then the
2681 error handling cleanup routines would dereference a NULL pointer.
2686 Description: NULL pointer deref in llap_from_page.
2687 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2688 incorrect) mapping if the page was truncated/reclaimed while it was
2689 searched for. Check for this condition and skip such pages when
2690 doing readahead. Introduce extra check to llap_from_page() to
2691 verify page->mapping->host is non-NULL (so page is not anonymous).
2694 Frequency : Sometimes when using sys_sendfile
2696 Description: "page not covered by a lock" warnings from ll_readpage
2697 Details : sendfile called ll_readpage without right page locks present.
2698 Now we introduced ll_file_sendfile that does necessary locking
2699 around call to generic_file_sendfile() much like we do in
2703 Frequency : with certain MDS communication failures at client mount time
2705 Description: NULL pointer deref after failed client mount
2706 Details : a client connection request may delayed by the network layer
2707 and not be sent until after the PTLRPC layer has timed out the
2708 request. If the client fails the mount immediately it will try
2709 to clean up before the network times out the request. Add a
2710 reference from the request import to the obd device and delay
2711 the cleanup until the network drops the request.
2714 Frequency : occasionally during client (re)connect
2716 Description: assertion failure during client (re)connect
2717 Details : processing a client connection request may be delayed by the
2718 client or server longer than the client connect timeout. This
2719 causes the client to resend the connection request. If the
2720 original connection request is replied in this interval, the
2721 client may trip an assertion failure in ptlrpc_connect_interpret()
2722 which thought it would be the only running connect process.
2725 Frequency : only with obd_echo servers and clients that are rebooted
2727 Description: kernel BUG accessing uninitialized data structure
2728 Details : When running an obd_echo server it did not start the ping_evictor
2729 thread, and when a client was evicted an uninitialized data
2730 structure was accessed. Start the ping_evictor in the RPC
2731 service startup instead of the OBD startup.
2733 Severity : enhancement
2734 Bugzilla : 10193 (patchless)
2735 Description: Remove dependency on various unexported kernel interfaces.
2736 Details : No longer need reparent_to_init, exit_mm, exit_files,
2737 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2740 Frequency : rare (only users of deprecated and unsupported LDAP config)
2742 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2743 Details : LDAP apparently contains 'lustreName' attributes instead of
2744 'name'. A simple remapping of the name is sufficient.
2747 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2749 Description: waiting_locks_callback trips kernel BUG if client is evicted
2750 Details : Running with the dump_on_timeout debug flag turned on makes
2751 it possible that the waiting_locks_callback() can try to dump
2752 the Lustre kernel debug logs from an interrupt handler. Defer
2753 this log dumping to the expired_lock_main() thread.
2755 Severity : enhancement
2757 Description: Support NFS exporting on 2.6 kernels.
2758 Details : Implement non-rawops metadata methods for NFS server to use without
2759 changing NFS server code.
2762 Frequency : very rare (synthetic metadata workload only)
2764 Description: two racing renames might cause an MDS thread to deadlock
2765 Details : Running the "racer" program may cause one MDS thread to rename
2766 a file from being the source of a rename to being the target of
2767 a rename at exactly the same time that another thread is doing
2768 so, and the second thread has already enqueued these locks after
2769 doing a lookup of the target and is trying to relock them in
2770 order. Ensure that we don't try to re-lock the same resource.
2773 Frequency : only very large systems with liblustre clients
2775 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2776 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2777 all OSTs in parallel. This allows the eviction of stale liblustre
2778 clients to proceed much faster than if they were done in series,
2779 and also offers similar improvements for other set_info RPCs.
2784 Description: excessive CPU usage during initial read phase on client
2785 Details : During the initial read phase on a client, it would agressively
2786 retry readahead on the file, consuming too much CPU and impacting
2787 performance (since 1.4.5.8). Improve the readahead algorithm
2788 to avoid this, and also improve some other common cases (read
2789 of small files in particular, where "small" is files smaller than
2790 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2795 Description: MDS crash when receiving packet with unknown intent.
2796 Details : Do not LBUG in unknown intent case, just return -EFAULT
2798 Severity : enhancement
2799 Bugzilla : 9293, 9385
2800 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2801 Details : Do not serialize getattr (non-intent version) and statfs.
2804 Frequency : occasional, when OST network is overloaded/intermittent
2806 Description: client evicted by OST after bulk IO timeout
2807 Details : If a client sends a bulk IO request (read or write) the OST
2808 may evict the client if it is unresposive to its data GET/PUT
2809 request. This is incorrect if the network is overloaded (takes
2810 too long to transfer the RPC data) or dropped the OST GET/PUT
2811 request. There is no need to evict the client at all, since
2812 the pinger and/or lock callbacks will handle this, and the
2813 client can restart the bulk request.
2816 Frequency : Always when mmapping file with no objects
2818 Description: client crashes when mmapping file with no objects
2819 Details : Check that we actually have objects in a file before doing any
2820 operations on objects in ll_vm_open, ll_vm_close and
2826 Description: Request leak when working with deleted CWD
2827 Details : Introduce advanced request refcount tracking for requests
2828 referenced from lustre intent.
2830 Severity : Enhancement
2832 Description: Cache open file handles on client.
2833 Details : MDS now will return special lock along with openhandle, if
2834 requested and client is allowed to hold openhandle, even if unused,
2835 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2836 closing files for every read/write openration.
2838 Severity : Enhancement
2840 Description: Cache open negative dentries on client when possible.
2841 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2842 negative dentries on lock revocation.
2847 Description: Remounting a client read-only wasn't possible with a zconf mount
2848 Details : It wasn't possible to remount a client read-only with llmount.
2850 Severity : enhancement
2851 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2852 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2853 MPICH is included to add Lustre-specific ADIO interfaces.
2854 This is based closely on the UFS ADIO layer and only differs
2855 in file creation, in order to allow the OST striping to be set.
2856 This is user-contributed code and not supported by CFS.
2861 Description: extended inode attributes (immutable, append-only) work improperly
2862 when 2.4 and 2.6 kernels are used on client/server or vice versa
2863 Details : Introduce kernel-independent values for these flags.
2865 Severity : enhancement
2868 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2869 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2870 granularity. Now, Fractional MB granularity is be supported,
2871 this is very useful for low memory system.
2873 Severity : enhancement
2875 Description: Getattr by fid
2876 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2877 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2878 have locks and avoids getattr rpc requests for it on every stat.
2881 Frequency : Always, for filesystems larger than 2TB
2883 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2884 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2885 filesystems larger than 2TB to be mounted with mballoc enabled.
2888 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2890 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2891 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2892 filesystems larger than 2TB is mounted then the kernel will
2893 silently corrupt the start of the filesystem. CONFIG_LBD is
2894 enabled for all CFS-supported kernels, but the possibility of
2895 this happening with a modified kernel config exists.
2897 Severity : enhancement
2899 Description: add client O_DIRECT support for 2.6 kernels
2900 Details : It is now possible to do O_DIRECT reads and writes to files
2901 in the Lustre client mountpoint on 2.6 kernel clients.
2903 Severity : enhancement
2905 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2906 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2907 parallel, not waiting for response from every OST before sending
2908 a rpc to the next OST.
2913 Description: setattr vs write race when updating file timestamps
2914 Details : Client processes that update a file timestamp into the past
2915 right after writing to the file (e.g. tar) it is possible that
2916 the updated file modification time can be reset to the current
2917 time due to a race between processing the setattr and write RPC.
2919 Severity : enhancement
2921 Description: Bring 'lfs find' closer in line with regular Linux find.
2922 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2923 -print0 options and obtains all the needed info through the lustre
2926 Severity : enhancement
2928 Description: support up to 1024 configured devices on one node
2929 Details : change obd_dev array from statically allocated to dynamically
2930 allocated structs as they are first used to reduce memory usage
2935 Description: Flush dirty partially truncated pages during truncate
2936 Details : Immediatelly flush partially truncated pages in filter_setattr,
2937 this way we completely avoid having any pages in page cache on OST
2938 and can retire ugly workarounds during writes to flush such pages.
2943 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2944 Details : protect inode from truncation within vfs_unlink() context
2945 just take a reference before calling vfs_unlink() and release it
2946 when parent's i_sem is free.
2951 Description: last_id value checked outside lock on OST caused LASSERT failure
2952 Details : If there were multiple MDS->OST object precreate requests in
2953 flight, it was possible that the OST's last object id was checked
2954 outside a lock and incorrectly tripped an assertion. Move checks
2955 inside locks, and discard old precreate requests.
2958 Frequency : always, if extents are used on OSTs
2960 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2961 Details : OSTs using the extents format would not zero the high 16 bits of
2962 the index physical block number. This is not a problem for any
2963 OST filesystems smaller than 16TB, and no kernels support ext3
2964 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2965 new/modified files) and can be fixed for existing filesystems
2966 with e2fsprogs-1.39-cfs1.
2971 Description: import connection selection may be incorrect if timer wraps
2972 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2973 import connections to be ignored if the 32-bit jiffies counter
2974 wraps. Use a 64-bit jiffies counter.
2977 Frequency : very large clusters immediately after boot
2979 Description: LNET request buffers exhausted under heavy short-term load
2980 Details : If a large number of client requests are generated on a service
2981 that has previously never seen so many requests it is possible
2982 that the request buffer growth cannot keep up with the spike in
2983 demand. Instead of dropping incoming requests, they are held in
2984 the LND until the RPC service can accept more requests.
2987 Frequency : Sometimes during replay
2989 Description: Assertion failure in ll_local_open after replay.
2990 Details : If replay happened on an open request reply before we were able
2991 to set replay handler, reply will become not swabbed tripping the
2992 assertion in ll_local_open. Now we set the handler right after
2993 recognising of open request
2996 Frequency : very rare
2998 Description: kernel reports "badness in vsnprintf"
2999 Details : Reading from the "recovery_status" /proc file in small chunks
3000 may cause a negative length in lprocfs_obd_rd_recovery_status()
3001 call to vsnprintf() (which is otherwise harmless). Exit early
3002 if there is no more space in the output buffer.
3004 Severity : enhancement
3006 Description: clear OBD RPC statistics by writing to them
3007 Details : It is now possible to clear the OBD RPC statistics by writing
3008 to the "stats" file.
3013 Description: Client mtime is not the same on different clients after utimes
3014 Details : In some cases, the client was using the utimes() syscall on
3015 a file cached on another node. The clients now validate the
3016 ctime from the MDS + OSTs to determine which one is right.
3021 Description: Inability to activate failout mode
3022 Details : lconf script incorrectly assumed that in pythong string's numeric
3023 value is used in comparisons.
3026 Frequency : always with multiple stripes per file
3028 Description: Inefficient object allocation for mutli-stripe files
3029 Details : When selecting which OSTs to stripe files over, for files with
3030 a stripe count that divides evenly into the number of OSTs,
3031 the MDS is always picking the same starting OST for each file.
3032 Return the OST selection heuristic to the original design.
3037 Description: mount failures may take full timeout to return an error
3038 Details : Under some heavy load conditions it is possible that a
3039 failed mount can wait for the full obd_timeout interval,
3040 possibly several minutes, before reporting an error.
3041 Instead return an error as soon as the status is known.
3043 Frequency : quota enabled and large files being deleted
3045 Description: releasing more than 4GB of quota at once hangs OST
3046 Details : If a user deletes more than 4GB of files on a single OST it
3047 will cause the OST to spin in an infinite loop. Release
3048 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3053 Description: statfs data retrieved from /proc may be stale or zero
3054 Details : When reading per-device statfs data from /proc, in the
3055 {kbytes,files}_{total,free,avail} files, it may appear
3056 as zero or be out of date.
3059 Frequency : systems with MD RAID1 external journal devices
3061 Description: lconf's call to blkid is confused by RAID1 journal devices
3062 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3063 of returning all block devices that match the journal UUID.
3066 Frequency : always, for aggregate stripe size over 4GB
3068 Description: assertion fails when trying to use 4GB stripe size
3069 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
3070 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
3073 Frequency : always on ppc64
3075 Description: the first write on an ext3 filesystem with mballoc got stuck
3076 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3077 perform endianness conversion.
3079 ------------------------------------------------------------------------------
3081 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3083 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3084 INTEROPERATE with older versions automatically. Please read the
3085 user documentation before upgrading any part of a live system.
3086 * WARNING: Lustre networking configuration changes are required with
3087 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3090 * Support for kernels:
3091 2.6.9-22.0.2.EL (RHEL 4)
3092 2.6.5-7.244 (SLES 9)
3093 2.6.12.6 vanilla (kernel.org)
3096 Severity : enhancement
3097 Bugzilla : 7981/8208
3098 Description: Introduced Lustre Networking (LNET)
3099 Details : LNET is new networking infrastructure for Lustre, it includes
3100 a reorganized network configuration mode (see the user
3101 documentation for full details) as well as support for routing
3102 between different network fabrics. Lustre Networking Devices
3103 (LNDs) for the supported network fabrics have also been
3104 created for this new infrastructure.
3106 Severity : enhancement
3107 Description: Introduced Access control lists
3108 Details : clients can set ACLs on files and directories in order to have
3109 more fine-grained permissions than the standard Unix UGO+RWX.
3110 The MDS must be started with the "-o acl" mount option.
3112 Severity : enhancement
3113 Description: Introduced filesystem quotas
3114 Details : Administrators may now establish per-user quotas on the
3117 Severity : enhancement
3119 Description: Configuration change for the XT3
3120 The PTLLND is now used to run Lustre over Portals on the XT3
3121 The configure option(s) --with-cray-portals are no longer used.
3122 Rather --with-portals=<path-to-portals-includes> is used to
3123 enable building on the XT3. In addition to enable XT3 specific
3124 features the option --enable-cray-xt3 must be used.
3129 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3130 Details : A race between allocating a new llcd and re-getting the llcd_lock
3131 allowed another thread to grab newly-allocated llcd.
3133 Severity : enhancement
3135 Description: 2.6 OST async journal commit and locking fix to improve performance
3136 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3137 2.6 kernels are now async as they already were in 2.4 kernels so
3138 that they can commit concurrently with the network bulk transfer.
3139 For block-allocated files the filter allocation semaphore is held
3140 to avoid filesystem fragmentation during allocation. BKL lock
3141 removed for 2.6 xattr operations where it is no longer needed.
3146 Description: lconf incorrectly determined whether two IP networks could talk
3147 Details : In some more complicated routing and multiple-network
3148 configurations, lconf will avoid trying to make a network
3149 connection to a disjoint part of the IP space. It was doing the
3150 math incorrectly for one set of cases.
3155 Description: Fix for potential infinite loop processing records in an llog.
3156 Details : If an llog record is corrupted/zeroed, it is possible to loop
3157 forever in llog_process(). Validate the llog record length
3158 and skip the remainder of the block on error.
3161 Frequency : occasional (liblustre only)
3163 Description: liblustre could not open files whose last component is a symlink
3164 Details : sysio_path_walk() would incorrectly pass the open intent to
3165 intermediate path components.
3168 Frequency : rare (liblustre only with non-standard tuning)
3169 Bugzilla : 7201 (7350)
3170 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3171 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3172 after it has been released, only do this for referenced locks
3174 Severity : enhancement
3176 Description: specifying an (invalid) directory default stripe_size of -1
3177 would reset the directory default striping
3178 Details : stripe_size -1 was used internally to signal directory stripe
3179 removal, now use "all default" to signal dir stripe removal
3180 as a directory striping of "all default" is not useful
3183 Frequency : common for large clusters running liblustre clients
3185 Description: doing an ls when liblustre clients are running is slow
3186 Details : sending a glimpse AST to a liblustre client waits for every AST
3187 to time out, as liblustre clients will not respond. Since they
3188 cannot cache data we refresh the OST lock LVB from disk instead.
3190 Severity : enhancement
3192 Description: doing an ls at the same time as file IO can be slow
3193 Details : enqueue and other "small" requests can be blocked behind many
3194 large IO requests. Create a new OST IO portal for non-IO
3195 requests so they can be processed faster.
3198 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3200 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3201 created file with mode 000
3202 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3203 allow an ftruncate() as a non-root user to a file with mode 000.
3204 Lustre now respects this flag to disable mode checks when
3205 truncating a file owned by the user
3208 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3211 Description: Revoking locks from clients that went dead or catatonic might take
3213 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3214 cancellation of such locks instant on servers without waiting for
3215 any reply from clients. Clients drops these locks when cancel
3216 notification from server is received without replying.
3219 Frequency : liblustre-only, when liblustre client dies or becomes busy
3221 Description: Doing ls on Linux clients can take a long time with active
3223 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3224 granting such locks to it in the first place if possible. Locks
3225 are taken by proxy on the OST during the read or write and
3226 dropped immediately afterward. Add connect flags handling, do
3227 not grant locks to liblustre clients for glimpse ASTs.
3229 Severity : enhancement
3231 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3232 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3233 sequential reads and behaves badly with applications doing random
3234 reads. Improve it by reading ahead at least read region, and
3235 avoiding excessive large RPC for small reads.
3237 Severity : enhancement
3239 Description: Creating more than 1000 files for a single job may cause a load
3240 imbalance on the OSTs if there are also a large number of OSTs.
3241 Details : qos_prep_create() uses an OST index reseed value that is an
3242 even multiple of the number of available OSTs so that if the
3243 reseed happens in the middle of the object allocation it will
3244 still utilize the OSTs as uniformly as possible.
3249 Description: OST or MDS may oops in ping_evictor_main()
3250 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3251 but doesn't restart at beginning of obd_exports_timed list
3254 Severity : enhancement
3256 Description: improve by-nid export eviction on the MDS and OST
3257 Details : allow multiple exports with the same NID to be evicted at one
3258 time without re-searching the exports list.
3261 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3263 Description: MDS may oops in groups_free()
3264 Details : in rare race conditions a newly allocated group_info struct is
3265 freed again, and this can be NULL. The 2.4 compatibility code
3266 for groups_free() checked for a NULL pointer, but 2.6 did not.
3269 Frequency : common for liblustre clients doing little filesystem IO
3270 Bugzilla : 9352, 7313
3271 Description: server may evict liblustre clients accessing contended locks
3272 Details : if a client is granted a lock or receives a completion AST
3273 with a blocking AST already set it would not reply to the AST
3274 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3277 Frequency : lfs setstripe, only systems with more than 160 OSTs
3279 Description: unable to set striping with a starting offset beyond OST 160
3280 Details : llapi_create_file() incorrectly limited the starting stripe
3281 index to the maximum single-file stripe count.
3284 Frequency : LDAP users only
3286 Description: lconf did not handle in-kernel recovery with LDAP properly
3287 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3289 Severity : enhancement
3291 Description: bind OST threads to NUMA nodes to improve performance
3292 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3293 node and do their allocations there to localize memory access
3295 Severity : enhancement
3297 Description: llmount can determine client NID directly from Myrinet (GM)
3298 Details : the client NID code from gmnalnid was moved directly into
3299 llmount, removing the need to use this or specifying the
3300 client NID explicitly when mounting GM clients with zeroconf
3303 Frequency : if client is started with down MDS
3305 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3306 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3307 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3308 ptlrpc_import_delayed_req() because we didn't check if the
3309 request was interrupted, and we also didn't break out of the
3310 event loop if there was no timeout
3315 Description: data loss during non-page-aligned writes to a single file from
3316 both multiple nodes and multiple threads on one node at same time
3317 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3318 inconsistency led to false short-reads, that were cached and later
3319 used by ->prepare_write() to fill in partially written page,
3320 leading to data loss.
3323 Frequency : always, if lconf --abort_recovery used
3325 Description: lconf --abort_recovery fails with 'Operation not supported'
3326 Details : lconf was attempting to abort recovery on the MDT device and not
3329 Severity : enhancement
3331 Description: remove cleanup logs
3332 Details : replace lconf-generated cleanup logs with lustre internal
3333 cleanup routines. Eliminates the need for client-cleanup and
3336 Severity : enhancement
3338 Description: add support for EAs (user and system) on lustre filesystems
3339 Details : it is now possible to store extended attributes in the Lustre
3340 client filesystem, and with the user_xattr mount option it
3341 is possible to allow users to store EAs on their files also
3343 Severity : enhancement
3345 Description: Add possibility (config option) to show minimal available OST free
3347 Details : When compiled with --enable-mindf configure option, statfs(2)
3348 (and so, df) will return least minimal free space available from
3349 all OSTs as amount of free space on FS, instead of summary of
3350 free spaces of all OSTs.
3352 Severity : enhancement
3354 Description: do not expand extent locks acquired on OST-side
3355 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3356 by server: they are not cached anyway.
3359 Frequency : when mmap is used/binaries executed from Lustre
3361 Description: Unmmap pages before throwing them away from read cache.
3362 Details : llap_shrink cache now attempts to unmap pages before discarding
3363 them (if unmapping failed - do not discard). SLES9 kernel has
3364 extra checks that trigger if this unmapping is not done first.
3369 Description: lconf didn't resolve symlinks before checking to see whether a
3370 given mountpoint was already in use
3373 Frequency : when migrating failover services
3374 Bugzilla : 6395, 9514
3375 Description: When migrating a subset of services from a node (e.g. failback
3376 from a failover service node) the remaining services would
3377 time out and evict clients.
3378 Details : lconf --force (implied by --failover) sets the global obd_timeout
3379 to 5 seconds in order to quickly disconnect, but this caused
3380 other RPCs to time out too quickly. Do not change the global
3381 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3383 Severity : enhancement
3384 Frequency : if MDS is started with down OST
3385 Bugzilla : 9439,5706
3386 Description: Allow startup/shutdown of an MDS without depending on the
3387 availability of the OSTs.
3388 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3389 Add appropriate locking and lov-osc refcounts for safe
3390 cleaning. Add osc abort_inflight calls in case the
3391 synchronize never started.
3394 Frequency : occasional (Cray XT3 only)
3396 Description: root not authorized to access files in CRAY_PORTALS environment
3397 Details : The client process capabilities were not honoured on the MDS in
3398 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3399 been accessed by an authorized user then root was able to access
3400 the file on the local client also. The root user capabilities
3401 are now allowed on the MDS, as this environment has secure UID.
3404 Frequency : occasional
3406 Description: ldiskfs "too long searching" message happens too often
3407 Details : A debugging message (otherwise harmless) prints too often on
3408 the OST console. This has been reduced to only happen when
3409 there are fragmentation problems on the filesystem.
3414 Description: Division by zero in statfs when all OSCs are inactive
3415 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3416 lov_statfs divides by value returned by lov_get_stripecnt().
3420 Bugzilla : 9489, 3273
3421 Description: First write from each client to each OST was only 4kB in size,
3422 to initialize client writeback cache, which caused sub-optimal
3423 RPCs and poor layout on disk for the first writen file.
3424 Details : Clients now request an initial cache grant at (re)connect time
3425 and so that they can start streaming writes to the cache right
3426 away and always do full-sized RPCs if there is enough data.
3427 If the OST is rebooted the client also re-establishes its grant
3428 so that client cached writes will be honoured under the grant.
3433 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3434 Details : Now I/O RPCs go to different portal number and (presumably) fast
3435 lock requests (and glimses) and other RPCs get their own service
3436 threads pool that should be able to service those RPCs
3439 Severity : enhancement
3441 Description: Ability to exchange lustre version between client and servers and
3442 issue warnings at client side if client is too old. Also for
3443 liblustre clients there is ability to refuse connection of too old
3445 Details : New 'version' field is added to connect data structure that is
3446 filled with version info. That info is later checked by server and
3450 Frequency : rare, liblustre only.
3451 Bugzilla : 9296, 9581
3452 Description: Two simultaneous writes from liblustre at offset within same page
3453 might proceed at the same time overwriting eachother with stale
3455 Details : I/O lock withing llu_file_prwv was released too early, before data
3456 actually was hitting the wire. Extended lock-holding time until
3457 server acknowledges receiving data.
3460 Frequency : extremely rare. Never observed in practice.
3462 Description: avoid generating lustre_handle cookie of 0.
3463 Details : class_handle_hash() generates handle cookies by incrementing
3464 global counter, and can hit 0 occasionaly (this is unlikely, but
3465 not impossible, because initial value of cookie counter is
3466 selected randonly). Value of 0 is used as a sentinel meaning
3467 "unassigned handle" --- avoid it. Also coalesce two critical
3468 sections in this function into one.
3470 Severity : enhancement
3472 Description: allow liblustre clients to delegate truncate locking to OST
3473 Details : To avoid overhead of locking, liblustre client instructs OST to
3474 take extent lock in ost_punch() on client's behalf. New connection
3475 flag is added to handle backward compatibility.
3477 Severity : enhancement
3478 Bugzilla : 4928, 7341, 9758
3479 Description: allow number of OST service threads to be specified
3480 Details : a module parameter allows the number of OST service threads
3481 to be specified via "options ost ost_num_threads={N}" in the
3482 OSS's /etc/modules.conf or /etc/modprobe.conf.
3486 Bugzilla : 6146, 9635, 9895
3487 Description: servers crash with bad pointer in target_handle_connect()
3488 Details : In rare cases when a client is reconnecting it was possible that
3489 the connection request was the last reference for that export.
3490 We would temporarily drop the export reference and get a new
3491 one, but this may have been the last reference and the export
3492 was just destroyed. Get new reference before dropping old one.
3494 Severity : enhancement
3495 Frequency : if client is started with failover MDS
3497 Description: Allow multiple MDS hostnames in the mount command
3498 Details : Try to read the configuration from all specified MDS
3499 hostnames during a client mount in case the "primary"
3502 Severity : enhancement
3504 Description: Stop sending data to evicted clients as soon as possible.
3505 Details : Check if the client we are about to send or are sending data to
3506 was evicted already. (Check is done every second of waiting,
3507 for which l_wait_event interface was extended to allow checking
3508 of exit condition at specified intervals).
3511 Frequency : rare, normally only when NFS exporting is done from client
3513 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3515 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3518 Frequency : always, if srand() is called before liblustre initialization
3520 Description: Liblustre uses system PRNG disturbing its usage by user application
3521 Details : Introduce internal to lustre fast and high-quality PRNG for
3522 lustre usage and make liblustre and some other places in generic
3523 lustre code to use it.
3525 Severity : enhancement
3526 Bugzilla : 9477, 9557, 9870
3527 Description: Verify that the MDS configuration logs are updated when xml is
3528 Details : Check if the .xml configuration logs are newer than the config
3529 logs stored on the MDS and report an error if this is the case.
3530 Request --write-conf, or allow starting with --old_conf.
3532 Severity : enhancement
3534 Description: Handle symlinks in the path when checking if Lustre is mounted.
3535 Details : Resolve intermediate symlinks when checking if a client has
3536 mounted a filesystem to avoid duplicate client mounts.
3541 Description: lconf can hit an error exception but still return success.
3542 Details : The lconf command catches the Command error exception at the top
3543 level script context and will exit with the associated exit
3544 status, but doesn't ensure that this exit status is non-zero.
3549 Description: failure of ptlrpc thread startup can cause oops
3550 Details : Starting a ptlrpc service thread can fail if there are a large
3551 number of threads or the server memory is very fragmented.
3552 Handle this without oopsing.
3555 Frequency : always, only if liblustre and non-default acceptor port was used
3557 Description: liblustre cannot connect to servers with non-default acceptor port
3558 Details : tcpnal_set_default_params() was not called and was therefore
3559 ignoring the environment varaible TCPNAL_PORT, as well as other
3560 TCPNAL_ environment variables
3565 Description: two objects could be created on the same OST for a single file
3566 Details : If an OST is down, in some cases it was possible to create two
3567 objects on a single OST for a single file. No problems other
3568 than potential performance impact and spurious error messages.
3572 Bugzilla : 5681, 9562
3573 Description: Client may oops in ll_unhash_aliases
3574 Details : Client dcache may become inconsistent in race condition.
3575 In some cases "getcwd" can fail if the current directory is
3581 Description: Inode refcounting problems in NFS export code
3582 Details : link_raw functions used to call d_instantiate without obtaining
3583 extra inode reference first.
3587 Bugzilla : 9942, 9903
3588 Description: Referencing freed requests leading to crash, memleaks with NFS.
3589 Details : We used to require that call to ll_revalidate_it was always
3590 followed by ll_lookup_it. Also with revalidate_special() it is
3591 possible to call ll_revalidate_it() twice for the same dentry
3592 even if first occurence returned success. This fix changes semantic
3593 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3594 reference on a request referred from the intent.
3595 ll_intent_release() then releases such a request.
3598 Frequency : rare, normally benchmark loads only
3600 Description: unlinked inodes were kept in memory on the client
3601 Details : If a client is repeatedly creating and unlinking files it
3602 can accumulate a lot of stale inodes in the inode slab cache.
3603 If there is no other client load running this can cause the
3604 client node to run out of memory. Instead flush old inodes
3605 from client cache that have the same inode number as a new inode.
3608 Frequency : SLES9 2.6.5 kernel and long filenames only
3609 Bugzilla : 9969, 10379
3610 Description: utime reports stale NFS file handle
3611 Details : SLES9 uses out-of-dentry names in some cases, which confused
3612 the lustre dentry revalidation. Change it to always use the
3616 Frequency : rare, unless heavy write-truncate concurrency is continuous
3617 Bugzilla : 4180, 6984, 7171, 9963, 9331
3618 Description: OST becomes very slow and/or deadlocked during object unlink
3619 Details : filter_destroy() was holding onto the parent directory lock
3620 while truncating+unlinking objects. For very large objects this
3621 may block other threads for a long time and slow overall OST
3622 responsiveness. It may also be possible to get a lock ordering
3623 deadlock in this case, or run out of journal credits because of
3624 the combined truncate+unlink. Solution is to do object truncate
3625 first in one transaction without parent lock, and then do the
3626 final unlink in a new transaction with the parent lock. This
3627 reduces the lock hold time dramatically.
3630 Frequency : rare, 2.4 kernels only
3632 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3633 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3634 before unmounting their filesystem in order to avoid deadlock.
3635 The kernel_locked() function in 2.4 kernels only checks whether
3636 the kernel lock is held, not whether it is this process that is
3637 holding it as 2.6 kernels do.
3642 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3643 Details : The client ptlrpc code may be trying to reconnect to a down
3644 server before a previous connection attempt has timed out.
3645 Increase the reconnect interval to be longer than the connection
3646 timeout interval to avoid sending duplicate connections to
3650 Frequency : echo_client brw_test command
3652 Description: fix echo_client to work with OST preallocated code
3653 Details : OST preallocation code (5137) didn't take echo_client IO path
3654 into account: echo_client calls filter methods outside of any
3655 OST thread and, hence, there is no per-thread preallocated
3656 pages and buffers to use. Solution: hijack pga pages for IO. As
3657 a byproduct, this avoids unnecessary data copying.
3661 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3662 Description: Client can oops in mdc_commit_close() after open replay
3663 Details : It was possible for the MDS to return an open request with no
3664 transaction number in mds_finish_transno() if the client was
3665 evicted, but without actually returning an error. Clients
3666 would later try to replay that open and may trip an assertion
3667 Simplify the client close codepath, and always return an error
3668 from the MDS in case the open is not successful.
3671 Frequency : rare, 2.6 OSTs only
3673 Description: OST may deadlock under high load on fragmented files
3674 Details : If there was a heavy load and highly-fragmented OST filesystems
3675 it was possible to have all the OST threads deadlock waiting on
3676 allocation of biovecs, because the biovecs were not released
3677 until the entire RPC IO was completed. Instead, release biovecs
3678 as soon as they are complete to ensure forward IO progress.
3680 Severity : enhancement
3682 Description: Support for specifying external journal device at mount
3683 Details : If an OST or MDS device is formatted with an external journal
3684 device, this device major/minor is stored in the ext3 superblock
3685 and may not be valid for failover. Allow detecting and
3686 specifying the external journal at mount time.
3691 Description: Mounting an MDS with pending unlinked files may cause oops
3692 Details : target_finish_recovery() calls mds_postrecov() which returned
3693 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3694 considers this an error and immediately begins cleaning up the
3695 lov, just after starting the mds_lov process
3697 Severity : enhancement
3699 Description: Implement 'lfs df' to report actual free space on per-OST basis
3700 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3701 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3702 usage of objects; '-h' to report in human readable format.
3704 ------------------------------------------------------------------------------
3706 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3713 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3714 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3715 at startup during orphan unlinking. Instead, skip buddy update
3716 before setup, as it will be regenerated after recovery is complete.
3721 Description: If an OST is inactive, its locks might reference stale inodes.
3722 Details : lov_change_cbdata() must iterate over all namespaces, even if
3723 they are inactive to clear inode references from the lock.
3725 Severity : enhancement
3726 Frequency : occasional, if non-standard max_dirty_mb used
3728 Description: Client will block write RPCs if not enough grant
3729 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3730 then the client will block writes while waiting for another RPC
3731 to complete instead of consuming its dirty limit. With change
3732 we get improved performance when max_dirty_mb is small.
3734 Severity : enhancement
3735 Bugzilla : 3389, 6253
3736 Description: Add support for supplementary groups on the MDS.
3737 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3738 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3739 lookups for user supplementary groups into a cache.
3743 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3744 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3745 for liblustre O_EXCL is enforced by MDS.
3749 Description: readdir never returns NULL in liblustre.
3750 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3751 dentry in struct dirent.
3755 Frequency : liblustre only, depends on application IO pattern
3756 Description: liblustre clients evicted if not contacting servers
3757 Details : Don't put liblustre clients into the ping_evictor list, so
3758 they will not be evicted by the pinger ever.
3760 Severity : enhancement
3762 Description: Add ability to evict clients by NID from MDS.
3763 Details : By echoing "nid:$NID" string into
3764 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3765 $NID would be instantly evicted from this MDS and from all active
3766 OSTs connected to it.
3770 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3771 Details : lookup_it_finish() used to query file size from OSTs that was not
3776 Description: service threads change working directory to that of init
3777 Details : Starting lustre service threads may pin the working directory
3778 of the parent thread, making that filesystem busy. Threads
3779 now change to the working directory of init to avoid this.
3783 Frequency : during shutdown only
3784 Description: shutdown with a failed MDS or OST can cause unmount to hang
3785 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3790 Frequency : 2.6 only
3791 Description: chmod/chown may include an extra supplementary group
3792 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3793 supplementary group and if none is specified this is used.
3796 Bugzilla : 5479 (6816)
3798 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3799 Details : If lookup is in progress on a file that is unlinked we might try
3800 to revalidate the inode and fail in revalidate after lookup is
3801 complete and ll_file_open() enqueues the open again but
3802 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3805 Frequency : always, if lconf --abort_recovery used
3807 Description: lconf --abort_recovery fails with 'Operation not supported'
3808 Details : lconf was attempting to abort recovery on the MDT device and not
3811 ------------------------------------------------------------------------------
3813 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3818 Frequency : rare (only unsupported configurations with a node running as an
3820 Bugzilla : 6514, 5137
3821 Description: Mounting a Lustre file system on a node running as an OST could
3823 Details : OSTs now preallocates memory needed to write out data at
3824 startup, instead of when needed, to avoid having to
3825 allocate memory in possibly low memory situations.
3826 Specifically, if the file system is mounted on on OST,
3827 memory pressure could force it to try to write out data,
3828 which it needed to allocate memory to do. Due to the low
3829 memory, it would be unable to do so and the node would
3830 become unresponsive.
3832 Severity : enhancement
3834 Description: Addition of lconf --service command line option
3835 Details : lconf now accepts a '--service <arg>' option, which is
3836 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3838 Severity : enhancement
3840 Description: Failover mode is now the default for OSTs.
3841 Details : By default, OSTs will now run in failover mode. To return to
3842 the old behaviour, add '--failout' to the lmc line for OSTs.
3844 Severity : enhancement
3846 Description: Health checks are now provided for MDS and OSTs
3847 Details : Additional detailed health check information on MSD and OSTs
3848 is now provided through the procfs health_check value.
3851 Frequency : occasional, depends on IO load
3853 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3854 after numerous create/delete cycles
3855 Details : The ext3 inode allocation policy would not allocate new inodes
3856 very well on the OSTs because there are no new directories
3857 being created. Instead we look for groups with free space if
3858 the parent directories are nearly full.
3863 Description: Network or server problems during mount may cause partially
3864 mounted clients instead of returning an error.
3865 Details : The config llog parsing code may overwrite the error return
3866 code during mount error handling, returning success instead
3872 Description: MDS can fail to allocate large reply buffers
3873 Details : After long uptimes the MDS can fail to allocate large reply
3874 buffers (e.g. zconf client mount config records) due to memory
3875 fragmentation or consumption by the buffer cache. Preallocate
3876 some large reply buffers so that these replies can be sent even
3877 under memory pressure.
3881 Frequency : rare (liblustre)
3882 Description: fsx running with liblustre complained that using truncate() to
3883 extend the file doesn't work. This patch corrects that issue.
3884 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3885 ATTR_SIZE and lsm use in llu_setattr_raw.
3889 Frequency : rare, only 2.6 kernels
3890 Description: Unusual file access patterns on the MDS may result in inode
3891 data being lost in very rare circumstances.
3892 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3893 code in the 2.6 kernel caused the inode and/or EA data not to
3894 be read from disk, causing single-file corruption.
3898 Frequency : rare, only 2.6 filesystems using extents
3899 Description: Heavy concurrent write and delete load may cause data corruption.
3900 Details : It was possible under high-load situations to have an extent
3901 metadata block in the block device cache from a just-unlinked
3902 file overwrite a newly-allocated data block. We now unmap any
3903 metadata buffers that alias just-allocated data blocks.
3907 Frequency : filesystems with default stripe_count larger than 77
3908 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3909 Details : lconf specifies an inode size of 4096 bytes when the default
3910 stripe_count is larger than 77. This conflicts with the default
3911 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3912 this case to avoid pinning too much memory for large EAs.
3914 ------------------------------------------------------------------------------
3916 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3921 Frequency : rare (extremely heavy IO load with hundreds of clients)
3923 Description: Client is evicted, gets IO error writing to file
3924 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3925 caused clients to be evicted for AST timeouts. The fixes in
3926 bug 5192 mean we no longer need to have such short AST timeouts
3927 so ldlm_timeout has been increased.
3930 Frequency : occasional during --force or --failover shutdown under load
3931 Bugzilla : 5949, 4834
3932 Description: Server oops/LBUG if stopped with --force or --failover under load
3933 Details : a collection of import/export refcount and cleanup ordering
3934 issues fixed for safer force cleanup
3937 Frequency : only filesystems larger than 120 OSTs
3938 Bugzilla : 5990, 6223
3939 Description: lfs getstripe would oops on a very large filesystem
3940 Details : lov_getconfig used kfree on vmalloc'd memory
3943 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3944 Bugzilla : 6242, 6243
3945 Description: reading from files that had been truncated to a non-zero size
3946 but never opened returned no data
3947 Details : ll_file_read() reads zeros from no-object files to EOF
3952 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3953 Details : There's little harm in aborting MDS/OSS recovery and letting it
3954 try again, so I removed the LASSERT and return an error instead.
3956 Severity : enhancement
3958 Description: New debugging infrastructure for tracking down data corruption
3959 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3960 (b) cover more of the client-side code path, and (c) try to narrow
3961 down where problems occurred
3965 Bugzilla : 3819, 4364, 4397, 6313
3966 Description: Racing close and eviction MDS could cause assertion in mds_close
3967 Details : It was possible to get multiple mfd references during close and
3968 client eviction, leading to one thread referencing a freed mfd.
3970 Severity: : enhancement
3971 Bugzilla : 3262, 6359
3972 Description: Attempts to reconnect to servers are now more aggressive.
3973 Details : This builds on the enhanced upcall-less recovery that was added
3974 in 1.4.2. When trying to reconnect to servers, clients will
3975 now try each server in the failover group every 10 seconds. By
3976 default, clients would previously try one server every 25 seconds.
3981 Description: After recovery, certain operations trigger a failed
3982 assertion on a client.
3983 Details : Failing over an mds, using lconf -d --failover, while a
3984 client was doing a readdir() call would cause the client to
3985 LBUG after recovery completed and the readdir() was resent.
3987 Severity : enhancement
3989 Description: Default groups are now added by lconf
3990 Details : You can now run lconf --group <servicename> without having to
3991 manually add groups with lmc.
3994 Frequency : occasional
3996 Description: Nodes with an elan id of 0 trigger a failed assertion
3999 Frequency : always when accessing e.g. tty/console device nodes
4001 Description: tty and some other devices nodes cannot be used on lustre
4002 Details : file's private_data field is used by device data and lustre
4003 values in there got lost. New field was added to struct file to
4004 store fs-specific private data.
4007 Frequency : when exporting Lustre via NFS
4009 Description: NFSD failed occasionally when looking up a path component
4010 Details : NFSD is looking up ".." which was broken in ext3 directories
4011 that had grown large enough to become hashed.
4014 Frequency : Clusters with multiple interfaces not on the same subnet
4016 Description: Nodes will repeatedly try to reconnect to an interface which it
4017 cannot reach and report an error to the log.
4018 Details : Extra peer list entries will be created by lconf with some peers
4019 unreachable. lconf now validates the peer before adding it.
4022 Frequency : Only if a default stripe is set on the filesystem root.
4024 Description: Setting a default stripe on the filesystem root prevented the
4025 filesystem from being remounted.
4026 Details : The client was sending extra request flags in the root getattr
4027 request and did not allocate a reply buffer for the dir EA.
4030 Frequency : occasional, higher if lots of files are accessed by one client
4031 Bugzilla : 6159, 6097
4032 Description: Client trips assertion regarding lsm mismatch/magic
4033 Details : While revalidating inodes the VFS looks up inodes with ifind()
4034 and in rare cases can find an inode that is being freed.
4035 The ll_test_inode() code will free the lsm during ifind()
4036 when it finds an existing inode and then the VFS later attaches
4037 this free lsm to a new inode.
4041 Bugzilla : 6422, 7030
4042 Description: MDS deadlock between mkdir and client eviction
4043 Details : Creating a new file via mkdir or mknod (starting a transaction
4044 and getting the ns lock) can deadlock with client eviction
4045 (gets ns lock and trying to finish a synchronous transaction).
4048 Frequency : occasional
4049 Description: While starting a server, the fsfilt_ext3 module could not be
4051 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
4052 kernels. Previously, lconf would still use the ext3 name
4053 when trying to load modules. Now, it will correctly use
4054 ext3 on 2.4 and ldiskfs on 2.6.
4056 Severity : enhancement
4057 Description: The default stripe count has been changed to 1
4058 Details : The interpretation of the default stripe count (0, to lfs
4059 or lmc) has been changed to mean striping across a single
4060 OST, rather than all available. For general usage we have
4061 found a stripe count of 1 or 2 works best.
4063 Severity : enhancement
4064 Description: Add support for compiling against Cray portals.
4065 Details : Conditional compiling for some areas that are different
4069 Frequency : occasional
4070 Bugzilla : 6409, 6834
4071 Description: Creating files with an explicit stripe count may lead to
4072 a failed assertion on the MDS
4073 Details : If some OSTs are full or unavailable, creating files may
4074 trigger a failed assertion on the MDS. Now, Lustre will
4075 try to use other servers or return an error to the
4079 Frequency : occasional
4081 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4082 caused lock timeout during readahead (since 1.4.2).
4083 Details : Processes doing readahead might match a lock that hasn't been
4084 granted yet if there are overlapping and conflicting lock
4085 requests. The readahead process waits on ungranted lock
4086 (original lock is CBPENDING), while OST waits for that process
4087 to cancel CBPENDING read lock and eventually evicts client.
4089 Severity : enhancement
4091 Description: Initial enabling of flock support for clients
4092 Details : Implements fcntl advisory locking and file status functions.
4093 This feature is provided as an optional mount flag (default
4094 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4095 locking are implemented yet, and those that are are not guaranteed
4096 to be completely correct in production environments.
4097 mount -t lustre -o [flock|noflock] ...
4100 Frequency : occasional
4102 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4103 assertion in the ext3 JBD (journaling) layer.
4104 Details : The b_committed_data struct is protected by the big kernel lock
4105 in 2.4 kernels, serializing journal_commit_transaction() and
4106 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4107 access to this struct. In 2.6 kernels there is finer grained
4108 locking to improve SMP performance of the JBD layer.
4112 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4117 Description: merge and cleanup kernel patches.
4120 -----------------------------------------------------------------------------
4122 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4124 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4125 versions of Lustre. Please update all servers and clients to
4126 version 1.4.2 or later at the same time. You must also run
4127 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4128 to update the configuration logs.
4130 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4131 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4132 - swab llog records of type '0' so we get proper header size/idx (5861)
4133 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4134 - fix rename of one directory over another leaking an inode (5953)
4135 - avoid SetPageDirty on 2.6 (5981)
4136 - don't re-add just-being-destroyed locks to the waiting list (5653)
4137 - when creating new directories, inherit the parent's custom
4138 striping settings if present parent (3048)
4139 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4140 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4141 - don't hold client locks on temporary worklist from l_lru (5666)
4142 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4143 - automatically evict dead clients (5921)
4144 - Update file size properly in create+truncate+fstat case (6196)
4145 - Do not unhash mountpoint dentries, do not allow removal of
4147 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4148 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4149 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4150 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4151 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4152 - Correctly handle reads of files with no objects (6243)
4153 - lctl recover will also mark a device active if deactivate used (5933)
4155 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
4156 - allow --write-conf on an MDS with different nettype than client (5619)
4157 - don't write config llogs to MDS for mounts not from that MDS (5617)
4158 - lconf should create multiple TCP connections from a client (5201)
4159 - init scripts are now turned off by default; run chkconfig --on
4160 lustre and chkconfig --on lustrefs to use them
4161 - upcalls are no longer needed for clients to recover to failover
4163 - add --abort-recovery option to lconf to abort recovery on device
4165 - add support for an arbitrary number of OSTs (3026)
4166 - Quota support protocol changes.
4167 - forward compatibility changes to wire structs (6007)
4168 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
4169 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
4170 - improved "lustre status" script
4171 - initialize blocksize for non-regular files (6062)
4172 - added --disable-server and --disable-client configure options (5782)
4173 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
4174 - Vanilla 2.4.29 support
4175 - increase maximum number of obd devices to 520 (6242)
4176 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4177 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4178 SLES 9 2.6.5 kernels
4179 - limit stripes per file to 160 (the maximum EA size) (6093)
4181 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4184 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4185 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4186 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4187 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4188 - small ext3 extents cleanups and fixes (5733)
4189 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4190 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4191 - avoid needless client->OST connect, fix handle mismatch (5317)
4192 - fix DLM error path that led to out-of-sync client, long delays (5779)
4193 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4194 - fix several locking issues related to i_size (5492,5624,5654,5672)
4195 - don't move pending lock onto export if it is already evicted (5683)
4196 - fix kernel oops when creating .foo in unlinked directory (5548)
4197 - fix deadlock in obdfilter statistics vs. object create (5811)
4198 - use time_{before,after} to avoid timer jiffies wrap (5882)
4199 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4200 - Do not leak request if server was not able to process it (5154)
4201 - If mds_open unable to find parent dir, make that negative lookup(5154)
4202 - don't create new directories with extent-mapping (5909, 5936)
4204 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
4205 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
4206 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
4207 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
4208 - support for RHEL4 kernel on i686 (5773)
4209 - provide error messages when incompatible logs are encountered (5898)
4211 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
4212 * version 1.4.0.10 (1.4.1 release candidate 1)
4214 - don't keep a lock reference when lock is not granted (4238)
4215 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4216 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
4217 - drop import inflight refcount on signal_completed_replay error (5255)
4218 - unlock page after async write error during send (3677)
4219 - handle missing objects in filter_preprw_read properly (5265)
4220 - no transno return for symlink open, don't save no-trasno open (3440)
4221 - don't try to complete elan receive that already failed (4012)
4222 - free RPC server reply state on error (5406)
4223 - clean up thread from ptlrpc_start_thread() on error (5160)
4224 - readahead could read extra page into cache that wasn't ejected (5388)
4225 - prevent races in class_attach/setup/cleanup/detach (5260)
4226 - don't dereference de->d_inode after l_dput of de (5458)
4227 - use "int" for stripe value returned from lock_to_stripe (5544)
4228 - mballoc allocation and error-checking fixes in 2.6 (5504)
4229 - block device patches to fix I/O request sizes in 2.6 (5482)
4230 - look up hostnames for IB nals (5602)
4231 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
4232 - don't start multiple acceptors for the same port (5277)
4233 - fix incorrect LASSERT in mds_getattr_name (5635)
4234 - export a proc file for general "ping" checking (5628)
4235 - fix "lfs check" to not block when the MDS is down (5628)
4237 - service request history (4965)
4238 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4239 - create an "evict_client" /proc entry on OSTs, like the MDS has
4240 - fix mount usage message, return errors per mount(8) (5168)
4241 - change grep [] to grep "[]" in tests so they work in more UMLs
4242 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4243 - remove ancient LOV_MAGIC_V0 EA support (5047)
4244 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4245 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4246 - allow readahead allocations to fail when low on memory (5383)
4247 - mmap locking landed again, after considerable improvement (2828)
4248 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4250 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4253 - send OST transaction number in read/write reply to free req (4966)
4254 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4255 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4256 - the watchdog thread now runs as interruptible (5246)
4257 - flock/lockf fixes (but it's still disabled, pending 5135)
4258 - don't use EXT3 constants in llite code (5094)
4259 - memory shortage at startup could cause assertion (5176)
4261 - reorganization of lov code
4262 - single portals codebase
4264 - add extents/mballoc support (5025)
4265 - direct I/O reads in the obdfilter (4048)
4266 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4268 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4271 - send OST transaction number in read/write reply to free req (4966)
4272 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4273 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4274 - the watchdog thread now runs as interruptible (5246)
4275 - handle missing objects in filter_preprw_read properly (5265)
4276 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4277 - drop import inflight refcount on signal_completed_replay error (5255)
4278 - unlock page after async write error during send (3677)
4279 - return original error code on reconstructed replies (3761)
4280 - no transno return for symlink open, don't save no-trasno open (3440)
4282 - add pid to ldlm debugging output (4922)
4283 - bump the watchdog timeouts -- we can't handle 30sec yet
4284 - extra debugging for orphan dentry/inode bug (5259)
4286 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
4289 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4290 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4291 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
4292 - fix NULL dereference in /proc/sys/portals/routes (4827)
4293 - allow failed mdc_close() operations to be interrupted (4561)
4294 - stop precreate on OST before MDS would time out on it (4778)
4295 - don't send partial-page writes before EOF from client (4410)
4296 - discard client grant for sub-page writes on large-page clients (4520)
4297 - don't free dentries not owned by NFS code, check generation (4806)
4298 - fix lsm leak if mds_create_objects() fails (4801)
4299 - limit debug_daemon file size, always print CERROR messages (4789)
4300 - use transno after validating reply (3892)
4301 - process timed out requests if import state changes (3754)
4302 - update mtime on OST during writes, return in glimpse (4829)
4303 - add mkfsoptions to LDAP (4679)
4304 - use ->max_readahead method instead of zapping global ra (5039)
4305 - don't interrupt __l_wait_event() during strace
4307 - add software watchdogs to catch hung threads quickly (4941)
4308 - make lustrefs init script start after nfs is mounted
4309 - fix CWARN/ERROR duplication (4930)
4310 - return async write errors to application if possible (2248)
4311 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4312 - print NAL number in %x format (4645)
4313 - update barely-supported suse-2.4.21-171 series (4842)
4314 - support for sles 9 %post scripts
4315 - support for building 2.6 kernel-source packages
4316 - support for sles km_* packages
4318 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4321 - ignore -ENOENT errors in osc_destroy (3639)
4322 - notify osc create thread that OSC is being cleaned up (4600)
4323 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4324 - reconstruct ost_handle() like mds_handle() (4657)
4325 - create a new thread to do import eviction to avoid deadlock (3969)
4326 - let lconf resolve symlinked-to devices (4629)
4327 - don't unlink "objects" from directory with default EA (4554)
4328 - hold socknal file ref over connect in case target is down (4394)
4329 - allow more than 32000 subdirectories in a single directory (3244)
4330 - fix blocks count for O_DIRECT writes (3751)
4331 - OST returns ENOSPC from object create when no space left (4539)
4332 - don't send truncate RPC if file size isn't changing (4410)
4333 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4334 - bind to privileged port in socknal and tcpnal (3689)
4336 - rate limit CERROR/CWARN console message to avoid overload (4519)
4337 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4338 - basic mmap support (3918)
4339 - kernel patch series update from b1_4 (4711)
4341 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4344 - avoid crash during MDS cleanup with OST shut down (2775)
4345 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4346 - don't use bad inodes on the MDS (3744)
4347 - dynamic object preallocation to improve recovery speed (4236)
4348 - don't hold spinlock over lock dumping or change debug flags (4401)
4349 - don't zero obd_dev when it is force cleaned (3651)
4350 - print grants to console if they go negative (4431)
4351 - "lctl deactivate" will stop automatic recovery attempts (3406)
4352 - look for existing locks in ldlm_handle_enqueue() (3764)
4353 - don't resolve lock handle twice in recovery avoiding race (4401)
4354 - revalidate should check working dir is a directory (4134)
4356 - don't always mark "slow" obdfilter messages as errors (4418)
4358 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4361 - don't close LustreDB during write_conf until it is done (3860)
4362 - fix typo in lconf for_each_profile (3821)
4363 - allow dumping logs from multiple threads at one time (3820)
4364 - don't allow multiple threads in OSC recovery (3812)
4365 - fix debug_size parameters (3864)
4366 - fix mds_postrecov to initialize import for llog ctxt (3121)
4367 - replace config semaphore with spinlock (3306)
4368 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4369 - don't allow enqueue to complete on a destroyed export (3822)
4370 - down write_lock before checking llog header bitmap (3825)
4371 - recover from lock replay timeout (3764)
4372 - up llog sem before sending rpc (3652)
4373 - reduce ns lock hold times when setting kms (3267)
4374 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4375 - fix NULL deref and obd_dev leak on setup error (3312)
4376 - replace some LBUG about llog ops with error handling (3841)
4377 - don't match INVALID dentries from d_lookup and spin (3784)
4378 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4379 - fix invalid assertion in ptlrpc_set_wait (3880)
4381 - add libwrap support for the TCP acceptor (3996)
4382 - add /proc/sys/portals/routes for non-root route listing (3994)
4383 - allow setting MDS UUID in .xml (2580)
4384 - print the stack of a process that LBUGs (4228)
4386 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4389 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4390 - make sure to unset replay flag from failed open requests (3440)
4391 - if default stripe count is 0, use OST count for inode size (3636)
4392 - update parent mtime/ctime on client for create/unlink (2611)
4393 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4394 - free recovery state on server during a forced cleanup (3571)
4395 - unregister_reply for resent reqs (3063)
4396 - loop back devices mounting and status check on 2.6 (3563)
4397 - fix resource-creation race that can provoke i_size == 0 (3513)
4398 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4399 - more debugging for page-accounting assertion (3746)
4400 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4401 - avoid deadlock after precreation failure (3758)
4402 - fix race and lock order deadlock in orphan handling (3450, 3750)
4403 - add validity checks when grabbing inodes from l_ast_data (3599)
4405 - add /proc/.../recovery_status to obdfilter (3428)
4406 - lightweight CDEBUG infrastructure, debug daemon (3668)
4407 - change default OSC RPC parameters to be better on small clusters
4408 - turn off OST read cache for files smaller than 32MB
4409 - install man pages and include them in rpms (3100)
4410 - add new init script for (un)mounting lustre filesystems (2593)
4411 - run chkconfig in %post for init scripts (3701)
4412 - drop scimac NAL (unmaintained)
4414 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4417 - clean kiobufs before and after use (3485)
4418 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4419 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4420 - change init script to fail more clearly if not run as root (1528)
4421 - allow clients to reconnect during replay (1742)
4422 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4423 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4424 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4425 - fix cond_resched (3554)
4426 - only evict unfinished clients after recovery (3515)
4427 - allow bulk resend, prevent data loss (3570)
4428 - dynamic ptlrpc request buffer allocation (2102)
4429 - don't allow unlinking open directory if it isn't empty (2904)
4430 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4431 - remove extraneous obd dereference causing LASSERT failure (3334)
4432 - don't use get_cycles() when creating temp. files on the mds (3156)
4433 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4434 - handle EEXIST for set-stripe, set proper directory name (3336)
4436 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4437 - fix ksocknal_fmb_callback() error messages (2918)
4439 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4442 - don't copy lvb into (possibly NULL) reply on error (2983)
4443 - don't deref dentry after dput, don't free lvb on error (2922)
4444 - use the kms to determine writeback rpc length (2947)
4445 - increment oti_logcookies when osc is inactive (2948)
4446 - update client's i_blocks count via lvb messages (2543)
4447 - handle intent open/close of special files properly (1557)
4448 - mount MDS with errors=remount-ro, like obdfilter (2009)
4449 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4450 - don't use cancelling-locks' kms values (2947)
4451 - use highest lock extent for kms, not last one (2925)
4452 - don't dereference ERR_PTR() dentry in error handling path (3107)
4453 - fix thread race in portals_debug_dumplog() (3122)
4454 - create lprocfs device entries at setup instead of at attach (1519)
4455 - common AST error handler, don't evict client on completion race (3145)
4456 - zero nameidata in detach_mnt in 2.6 (3118)
4457 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4458 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4459 - initialize RPC timeout timer earlier for 2.6 (3219)
4460 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4461 - print nal/nid for unknown nid (3258)
4462 - additional checks for oscc recovery before doing precreate (3284)
4463 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4464 - don't crash in mdc_close for bad permissions on open (3285)
4465 - zero i_rdev for non-device files (3147)
4466 - clear page->private before handing to FS, better assertion (3119)
4467 - tune the read pipeline (3236)
4468 - fix incorrect decref of invalidated dentry (2350)
4469 - provide read-ahead stats and refine rpc in flight stats (3328)
4470 - don't hold journal transaction open across create RPC (3313)
4471 - update atime on MDS at close time (3265)
4472 - close LDAP connection when recovering to avoid server load (3315)
4473 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4474 - don't leak open file on MDS after open resend (3325)
4475 - serialize filter_precreate and filter_destroy_precreated (3329)
4476 - loop device shouldn't call sync_dev() for nul device (3092)
4477 - clear page cache after eviction (2766)
4478 - resynchronize MDS->OST in background (2824)
4479 - refuse to mount the same filesystem twice on same mountpoint (3394)
4480 - allow llmount to create routes for mounting behind routers (3320)
4481 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4482 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4483 - fix rare problem with rename on htree directories (3417)
4485 - allow default OST striping configuration per directory (1414)
4486 - fix compilation for qswnal for 2.6 kernels (3125)
4487 - increase maximum number of MDS request buffers for large systems
4488 - change liblustreapi to be useful for external progs like lfsck (3098)
4489 - increase local configuration timeout for slow disks (3353)
4490 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4492 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4495 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4496 - don't overwrite extent policy data in reply if lock was blocked (2901)
4497 - drop filter export grants atomically with removal from device (2663)
4498 - del obd_self_export from work_list in class_disconnect_exports (2908)
4499 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4500 - swab reply message in mdc_close, other PPC fixes (2464)
4501 - fix destroying of named logs (2325)
4502 - overwrite old logs when running lconf --write_conf (2264)
4503 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4504 - fix race in target_handle_connect (2898)
4505 - mds_reint_create() should take same inode create lock (2926)
4506 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4507 - don't close files for self_export to avoid uninitialized obd (2936)
4508 - allow MDS with the same name as client node (2939)
4509 - hold dentry reference for closed log files for unlink (2325)
4510 - reserve space for all logs during transactions (2059)
4511 - don't evict page beyond end of stripe extent (2925)
4512 - don't oops on a deleted current working directory (2399)
4513 - handle hard links to targets without a parent properly (2517)
4514 - don't dereference NULL lock when racing during eviction (2867)
4515 - don't grow lock extents when lots of conflicting locks (2919)
4517 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4520 - account for cache space usage on clients to avoid data loss (974)
4521 - lfsck support in lustre kernel code (2349)
4522 - reduce journal credits needed for BRW writes (2370)
4523 - orphan handling to avoid losing space on client/server crashes
4524 - ptlrpcd can be blocked, stopping ALL progress (2477)
4525 - use lock value blocks to assist in proper KMS, faster stat (1021)
4526 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4527 - recovery for initial connections (2355)
4528 - fixes for mds_cleanup_orphans (1934)
4529 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4530 - block all file creations until orphan recovery completes (1901)
4531 - client remove rq_connection from request struct (2423)
4532 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4533 - recovery timer race (2670)
4534 - mdc_close recovey bug (2532)
4535 - ptlrpc cleanup bug (2710)
4536 - mds timeout on local locks (2588)
4537 - namespace lock held during RPCs (2431)
4538 - handle interrupted sync write properly (2503)
4539 - don't try to handle a message that hasn't been replied to (2699)
4540 - client assert failure during cleanup after abort recovery (2701)
4541 - leak mdc device after failed mount (2712)
4542 - ptlrpc_check_set allows timedout requests to complete (2714)
4543 - wait for inflight reqs when ptlrpcd finishes (2710)
4544 - make sure unregistered services are removed from the srv_list
4545 - reset bulk XID's when resending them (caught by 1138 test)
4546 - unregister_bulk after timeout
4547 - fix lconf error (2694)
4548 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4549 - readahead locks pages, leaves pending causing memory pressure (2673)
4550 - increase OST request buffers to 4096 on large machines (2729)
4551 - fix up permission of existing directories in simple_mkdir (2661)
4552 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4553 - don't assign transno to errored transactions (2742)
4554 - don't delete objects on OST if given a bogus objid from MDS (2751)
4555 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4556 - if rq_no_resend, then timeout request after recovery (2432)
4557 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4558 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4559 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4560 - MDS recovery shouldn't delete objects that it has given out (2730)
4561 - if enqueue arrives after completion, don't clobber LVB (2819)
4562 - don't unlock pages twice when trigger_group_io returns error (2814)
4563 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4564 - don't write pages to disk if there was an error (1450)
4565 - don't ping imports that have recovery disabled (2676)
4566 - take buffered bytes into account when balancing socknal conn (2817)
4567 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4568 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4569 - remove little-endian swabbing of llog records (1987)
4570 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4571 - retry reposting request buffers if they fail (1191)
4572 - grow extent at grant time to avoid granting a revoked lock (2809)
4573 - lock revoke doesn't evict page if covered by a second lock (2765)
4574 - disable VM readahead to avoid reading outside lock extents (2805)
4576 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4577 - updated kernel patches for hp-2.4.20 kernel (2681)
4579 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4582 - fix truncated write corruption (2366)
4583 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4585 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4586 - ensure len(uuid) < 37 in lmc (1171)
4587 - fix ia64 OOPS in llog_test (2255)
4588 - zero end of page at obdfilter for partial page writes (2648)
4589 - don't leave stale dentries around after renames (bug 2428)
4590 - fix timeouts when evicting a client with a single lock held (2642)
4591 - set deadline for the initial HELLO message to drain (2634)
4592 - print out dotted-quad IP addresses in the socknal (2302)
4594 - additional debugging for MDS client eviction problem (2443)
4595 - fix mkfsoptions support for osts (2603, 2604)
4597 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4600 - add series for the vanilla 2.6.0 kernel
4601 - add series for the vanilla 2.4.24 kernel
4602 - add series for a cray x86/64 UL kernel drop
4603 - fix xattr patches for the vanilla 2.4.19 series
4605 - generate true UUIDs in lmc (1171)
4606 - have portals stack dumping break in UML (2466)
4607 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4608 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4609 - wake callback waiting threads on client eviction (2460)
4610 - Add --ptldebug and --subsystem to lmc (1719)
4611 - update assertion to allow safe interrupt allocation
4612 - set rq_no_resend for cancel requests (2432)
4613 - recalculate ptlrpcd timeout after resend (2494)
4614 - call vfs_rmdir when removing pending directories (2368)
4615 - fix renaming a file to itself (2429)
4616 - lmc creates a default one-stripe lov (2454)
4617 - expand procfs space to handle large clusters (2326)
4618 - increase UML stack to avoid overflow
4619 - update lconf's list of debug and subsystem masks
4620 - fix lfs find --obd (2510)
4621 - /proc tunable for disabling filter read caching (2591)
4622 - stop rpm packages from altering slapd.conf (2301)
4623 - disable nagle in the socknal under 0conf (2578)
4624 - choose mds inode size based on stripe count (2572)
4625 - fix kernel-source rpm problems (2516)
4627 - add --disable-doc to avoid pdf generation (2421)
4628 - update documentation, tests, type-os, comments
4629 - avoid format warnings on ia64
4630 - remove the TOE NAL
4631 - tiny code cleanups by removing unused fields
4633 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4636 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4637 - remove the most blatant lies from BUILDING file (2371)
4638 - change default debug level to reasonable production setting
4639 - reduce client side cache size to reduce cache flush time
4640 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4641 - make TCP zerocopy and pinger support enabled by default (2476)
4642 - sync writes completed after process exits caused crashes (2319)
4643 - maintain correct mount count on the MDS (2356)
4644 - backout 1557, because 2316 wasn't really fixed
4645 - better file I/O statistics gathering in /proc
4646 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4647 - another kernel patch to fix zero-copy TCP function export
4648 - don't take duplicate lock when processing re-sent getattr (2420)
4649 - lctl uses obd_self_export instead of creating new conn (2353)
4650 - MDS/OST recovery case which requires object creation asserted (2425)
4651 - move lfs from /usr/sbin to /usr/bin in packages
4652 - fix race between mds_client_add and mds_client_free (2417)
4653 - use kmalloc instead of slabs in portals (2430)
4654 - don't create duplicate records when a failover MDS is present (2442)
4655 - remove unnecessary mount age check (2332)
4656 - don't remove directory inodes from locks prematurely (2451)
4657 - don't break if MDS service name is the same as hostname (2103)
4658 - fix races in client write RPC generation when cache full (2482)
4660 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4663 - remove now-unused request->rq_obd (278)
4664 - if an allocation fails, print out how much memory we've used (1933)
4665 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4666 - add the "configurable stack size" patch to most series files (1256)
4667 - ability to write large log records, for 100+ OST configs (2306)
4668 - fix NULL deref when filter_prep fails (2314)
4669 - fix operator precedence error in filter_sync
4670 - dynamic allocation of socknal TX descriptors (2315)
4671 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4672 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4673 - ensure that CWARN messages in Portals always get to the syslog
4674 - __init/__exit are not for prototype decls (ldlm_init/exit)
4675 - x86-64 compile warning fixes
4676 - fix gateway LMC keyword conflict (2318)
4677 - fix MDS lock inversions in getattr/reint paths (1844)
4678 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4679 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4680 - DLM race condition prevented some lock evictions (2328)
4681 - ENOMEM detection and retry on socknal sends (2230)
4682 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4683 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4684 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4685 - better error messages when a client is rejected during recovery (1505)
4686 - avoid cancelling locks which were never granted, after failure (2330)
4687 - fix i_sem/journal inversion in mds_client_add (2333)
4688 - fix truncate/getattr lock cycle deadlock (2334)
4689 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4690 - fix two rare exit paths which could leak an l_lock() ref (2321)
4691 - fencepost error in MDS/OST orphan recovery (2226)
4692 - make log record alignment 8 bytes (1988)
4693 - lstripe now fails when requested offset > ost_count (2237)
4694 - ensure that all kernel series have a complete list.h (1607)
4695 - fix crashes in special-file operations (2316)
4696 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4698 - allow configurable automake binary, for testing new versions
4699 - small update to the lfs documentation
4701 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4703 * fix negative export reference count in fsfilt_sync (2312)
4705 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4706 * release candidate 0.9.1
4708 - orphans are moved into the PENDING directory for possible recovery
4709 - replayed opens now open by fid for orphan/rename safety (1042)
4710 - last close of an orphan inode generates a transno (683)
4711 - chdir() and mount() now pin the directory entry (1020)
4712 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4713 - discard very old requests without processing them (1502)
4714 - remove some common, well-understood CERRORs (1505)
4715 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4716 - clear "grant" flags in OST replies until OST grant code lands (1644)
4717 - fix read performance by not clobbering i_blksize on client (1598)
4718 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4719 - make LRU size a /proc tunable, clears locks when reduced (707)
4720 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4721 - server locks take references on exports now (1558)
4722 - build fixes for 2.4.20-rh trees (1663)
4723 - return an error from lov_create if all OSCs are inactive (1751)
4724 - fix import levels when a reconnect happens without a timeout (1597)
4725 - exit early from mds_open if we get a lookup error (1749)
4726 - partial page read at EOF wouldn't wait for disk before sending (1642)
4727 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4728 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4729 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4730 - write/truncate lock inversion (1639)
4731 - Don't auto-load obdclass, portals modules during cleanup (1495)
4732 - fix timestamps from jumping to "now" (1763)
4733 - extra journal assertions (1648)
4734 - add an extra multiunlink test (1771)
4735 - fix read_record/write_record API (1776)
4736 - fix leak of offset_extent, possible incorrect i_size later (1772)
4737 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4738 - Debugging for the kqswnal_get_idle_tx problems (1820)
4739 - Allow recovery to be attempted multiple times (1536)
4740 - Write out MDS last_rcvd file after it is first created (1600)
4741 - Fix tx_descriptor leak in failed transmit situations (1827)
4742 - ext3 journaling fixes for assertion failure after IO error (1871)
4743 - class_export_put() on freed export after completion AST error (1896)
4744 - Fix revalidate looping in VFS (1322)
4745 - Don't access a freed export during MDS_REINT timeout (1521)
4746 - Add open-unlink recovery support on the MDS (1673,1764)
4747 - Return an error if no MDS data was read from last_rcvd (1946)
4748 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4749 - Don't setup a disk device that doesn't match exported UUID (317)
4750 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4751 - avoid committing NULL handle in force close
4752 - local.sh is now a one-stripe LOV configuration
4753 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4754 - fix ext3 htree duplicate directory entry corruption (1516)
4755 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4756 - update to new LOV EA format (2097)
4757 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4759 2003-06-15 Phil Schwan <phil@clusterfs.com>
4762 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4763 - per-import/export recovery handling (958, 931, 959)
4764 - multiple last-rcvd slots, for serving multiple FSes (949)
4765 - connections are again shared between multiple imp/exports (963, 964)
4766 - "umount -f" would hang if any requests needed to be sent (393, 978)
4767 - avoid pinning large req buffer by copying for queued messages (989)
4768 - add "uuid" to "lctl device" command to help upcalls (991)
4769 - "open" RPCs with transnos would confuse recovery counters (1037)
4770 - do proper endian conversion of all wire messages (288, 340, 891)
4771 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4772 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4773 - fix ptlrpc_connection leak in target_handle_connect (1174)
4774 - fix import refcounting bug in OST and MDS cleanup (1134)
4775 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4776 - fix very unlikely obd_types race condition (501)
4777 - remove osc_open hack for echo_client (1187)
4778 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4779 - a failure in read_inode2 leads to deadlock (1139)
4780 - cancel ack-locks as soon as transaction is committed (1072)
4781 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4782 - make sure to commitrw after any preprw to avoid deadlock (1162)
4783 - failing to execute a file in a lustre FS would lock inode (1203)
4784 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4785 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4786 - obd_uuid2tgt would walk off the end of the list (1255)
4787 - on IA64 the osc would give portals incorrect bulk size (1258)
4788 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4789 - fix lock inversion caused by new llite matching code (1282)
4790 - limit the number of dirty pages on a client to 10MB (1286)
4791 - timed out locks were not being corrected cancelled (1289)
4792 - fix O_DIRECT above 4GB on IA-32 (1292)
4793 * major user-visible changes
4794 - fail out/fail over policy now controlled by the upcall (993)
4796 - add OBD_PING to check server availability and failure (954)
4797 - lustre messages are now sent in sending host order (288, 340, 891)
4798 - add eadatalen to MDS getattr reply (340)
4799 - OST read replies may contain second buffer, with per-page status (593)
4801 2003-03-11 Phil Schwan <phil@clusterfs.com>
4804 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4805 - failed open()s could cause deadlock; fixed (867, 869)
4806 - stop cancelling OST locks when files are closed (481)
4807 - overlapping XID spaces caused network corruption (851, 853)
4808 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4809 - setattr_raw added, to do single-RPC, server-side setattrs
4810 - lmc/lconf syntax change for OST UUIDs
4811 - fix crashy race condition between ptlrpc_free_req and osc_close
4812 - don't use request in mdc_enqueue if we hit a timeout (889)
4813 - don't set the inode i_size for regular files from the MDS (896)
4814 - handle out of order completion AST (842)
4815 - don't LBUG if a lock request times out after receiving AST (913)
4816 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4817 - if a bad lock AST arrives, send an error instead of dropping entirely
4818 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4819 - fix leak in bulk IO when only partially completed (899, 900, 926)
4820 - fix O_DIRECT for ia64 (55)
4821 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4822 - C-z after timeout could hang a process forever; fixed (977)
4824 - client-side I/O cache (678, 924, 929, 941, 970)
4826 - READPAGE and SETATTRs which don't take server-side locks get
4829 2003-02-11 Phil Schwan <phil@clusterfs.com>
4832 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4833 - Fix fsfilt_extN_readpage() to read a full page of directory
4834 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4835 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4836 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4837 - Most intent operations were removed, in favour of a new RPC mode
4838 that does a single RPC to the server and bypasses most of the VFS
4839 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4840 - Aggressively cancel local locks on DLM servers
4841 - mds_reint_unlink sends EA to the client if it's the last nlink.
4842 client uses that EA to unlink OST objects.
4843 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4844 - recursive symlinks were fixed (439)
4845 - fixed NULL deref in DEBUG_REQ
4846 - filter_update_lastobjid no longer calls sync, which annoyed extN
4847 - fixed multi-client small-writes to a single file problem (445)
4848 - fixed mtime updates during file writes (607)
4849 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4850 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4851 - fixed Linux OST size reporting problem (444, 656)
4852 - OST now updates object mtime with writes or setattr (607, 619)
4853 - client verifies file size before zeroing page past EOF (445)
4854 - OST now writes last allocated objid to disk with allocation (108)
4855 - LOV on echo now works (409)
4857 - mds_reint_unlink sends a new buffer, with the EA included. this
4858 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4859 set if a regular file was being unlinked, and it was the last link
4860 - use PtlGet from the target for bulk writes (315)
4861 - OST now updates object mtime with writes or setattr (607, 619)
4862 - LDLM now has a grant-time callback to revalidate locked items, if
4864 - Many MDS operations were reorganized to combat race conditions
4866 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4867 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4869 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4872 - Fully reactivate OST imports after reconnection (512, others)
4873 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4874 - More graceful error handling for truncating on dead OST (515)
4875 - Don't error out unless we're actually accessing dead stripes (474)
4876 - Fix garbage sizes when stripes are missing (410)
4877 - LRU counters were broken, causing constant lock purge (433, 432)
4878 - garbage on read from stripes with failed OSTs (441)
4879 - mark OSCs as active before reconnecting during recovery (438)
4880 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4881 - lfind did not preserve OST order in output (443)
4882 - symlinks cause hung clients, incorrect data (439)
4883 - stop dereferencing request after dropping refcount (457)
4884 - don't LASSERT(spin_is_locked) on non-SMP (455)
4885 - fixes for many rename() bugs
4886 - fstat didn't correctly synchronize attributes (399)
4887 - server must handle lock cancellation during blocking AST prep (487)
4888 - bulk descriptors were free()d too soon (511)
4889 - fix paths in lconf, which would load incorrect modules (451, 507)
4890 - fix confusing lconf 'host not found' error message (386)
4891 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4892 - fix race condition in mdc_blocking_ast() for inode access (526)
4893 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4894 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4895 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4896 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4897 - fix lstripe in several ways (kernel side) (527)
4898 - fix request leak in ldlm_cli_enqueue (262)
4899 - incorrect OSC was marked inactive after OST failure
4900 - call mds_fs_cleanup before unmounting filesystem (524)
4901 - fix races between taking ns_lock and ldlm_lock_change_resource
4902 - fix races updating LOV export open file list
4903 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4904 - fix recovery NULL deref in ldlm_cli_cancel_unused
4905 - fix some DLM races by using new hash table for lock handles (419)
4906 - permit the client to specify desired inodes, at replay
4907 - duplicate requests when we queue them for replay reintegration
4908 - fix last_rcvd offset calculation
4909 - sync after each recovered transaction, so we always make progress
4910 - never, not always, ERESTART requests without transnos
4911 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4913 - skip replay if the MDS says that the client is already connected
4914 - don't check for a recovery-enabled export to match lctl's UUID
4915 - don't INC_USE_COUNT for phantom exports
4916 - don't crash when cleaning up phantom exports (567)
4917 - don't double-finish or set replay data for errored mdc_open requests
4918 - abort requests when they time out, so we don't get old replies
4919 - send/receive replies for AST messages again
4920 - if the client says that it doesn't have the lock, cancel it on the
4922 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4923 mark it as destroyed, it will all work out when decref is called
4924 - fix module use counts (22, 581)
4926 - ASTs now expect a reply (server cancels lock on error reply)
4928 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4931 - fix many simultaneous client startup (392)
4932 - fix dentry->d_it clobbering
4933 - credentials weren't being shipped for readdir/getattr operations
4934 - remove invalid assertions triggered during some concurrent MD
4936 - proper Lustre versions added (336, 389)
4937 - fix memory leak for create error case (398)
4938 - fix LOV locking bug that would get cli/srv out of sync
4939 - fix echo client over LOV (409)
4940 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4941 - fix double-O_EXCL intent crash (424)
4942 - avoid sending multiple lock CANCELs (352)
4944 - MDS can do multi-client recovery (modulo bugs in new code)
4946 - many updates, edits, cleanups
4948 2002-11-18 Phil Schwan <phil@clusterfs.com>
4951 - fix null d_it dereference (346)
4952 - fix full OST/dbench hang (333)
4953 - fix permission problem with file removal (286)
4954 - fix removal of OSCs from LOV when they fail
4955 - fix NULL deref during bulk timeout (214)
4956 - fix problems related to multiple filesystems on one MDS (241)
4957 - fixed serious subtle metadata locking bugs
4958 - free locks on clients when inodes are removed due to memory
4960 - fix inode pointer in lock data (285)
4961 - partial support for multiple MDS on a single host (241)
4962 - data locks weren't cancelled at clear_inode time (290, 311)
4963 - intent locks could lead to unbounded lock growth (205)
4964 - added a maximum lock count, an LRU list, and a flusher
4965 - fix multiple rename (365)
4966 - properly abstracted the echo client
4967 - OSC locked 1 byte too many; fixed
4968 - rewrote brw callback code:
4969 - fixed recovery bugs related to LOVs (306)
4970 - fixed too-many-pages-in-one-write crash (191)
4971 - fixed (again) crash in sync_io_timeout (214)
4972 - probably fixed callback-related race (385)
4974 - Add capability to MDS protocol
4975 - LDLM cancellations and callbacks on different portals
4977 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4980 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4981 - defer open object destruction to close time (601981, 138)
4982 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4983 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4984 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4985 - don't call obd_setattr in ll_file_release for destroyed objects
4986 * protocol change to lustre_msg: move |version| and add |flags|
4987 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4988 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4989 * added replay of create, unlink, link and rename operations during
4990 MDS failover; recovery should be much more robust now
4991 * remove failed OSCs from LOVs (only lov_create uses this so far)
4992 * the lustre-HOWTO was brought (more) up to date (582544)
4994 2002-10-23 Phil Schwan <phil@clusterfs.com>
4997 - in-use dentries weren't being reused properly (617851)
4998 - prevent multiple LDLM setup (599178)
4999 - fix LOV size calculations for truncate (617853)
5000 - fix client handling of MDS intent errors (POSIX)
5001 - fix permission bug in lovstripe.c test (624321)
5002 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
5003 - truncate past end of file could corrupt data
5004 - proper cleanup after timeouts, crashes, etc (592524, 550815)
5005 - a race in recovery could return ETIMEDOUT to apps (623947)
5006 - building outside the source directory was fixed
5007 * the lustre-HOWTO was brought (more) up to date (582544)
5008 * major progress was made on recovery functionality
5010 2002-10-10 Phil Schwan <phil@clusterfs.com>
5013 - recovery deadlock fix
5014 - rm -rf causes LBUG fix (617817)
5015 - file open by multiple tasks fix (618962)
5016 - directory permissions bugs (602707 and 620007)
5017 - journal_stop fixed with locking (611313)
5018 - O_APPEND failures resolved (618273, perhaps 614459)
5019 - lconf PATH fix (619770)
5020 - IA64 build fix (621450)
5021 - RPC buffer sizes scale with amount of memory
5023 2002-10-01 Phil Schwan <phil@clusterfs.com>
5026 - locks would be cancelled without throwing away data pages,
5027 resulting in inconsistent data (605627)
5028 - inode attributes were not always being refreshed (605627, 612449)
5029 - lconf now continues to cleanup after lctl reports an error
5030 - MDS now enforces user permissions (602707)
5031 - lprocfs cleanup fixed, but not yet enabled (614157)
5032 - fixed infinite server hang, should a client not respond to an AST
5033 - avoid going into recovery if user calls readlink() with a buffer
5034 that's too small (613941)
5035 - AST RPCs no longer require replies (614867) -- this may be changed
5036 - don't crash server if client sends an IOV that's too big (611336)
5037 - fixed lock conversion deadlock (611892)
5038 - fixed the following of symlinks (614622)
5039 * recovery: the server can remove locks from a client that dies, other
5040 clients can make progress
5041 * more extN patch fixes
5042 * compile-time configurable ptlrpc buffer allocations
5044 - collaborative read cache document
5045 - Lustre Lite Performance CDR document-in-progress
5047 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5050 - fix typo in patch-2.4.18
5052 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5055 - clear ptlrpc request each time in handle_incoming_request()
5056 - unlink of files now destroys the object on the OST
5058 2002-09-19 Peter Braam <braam@clusterfs.com>
5060 * add hard link support
5061 * change obdfile creation method
5062 * kernel patch changed
5064 2002-09-19 Peter Braam <braam@clusterfs.com>
5067 - stack overflow bug in extN fixed
5069 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
5071 * documentation updates
5072 - add man pages for config tools
5073 - update tests/README to describe testing with new config tools
5074 - finish metadata API descriptions
5075 * bug fixes and cleanups
5076 - statfs workaround for 16TB limit
5077 - LOV stripe allocation improved, can stripe on subset of OSTs
5078 - LOV file size/IO offset was wrong for files > 4GB in size
5079 - object EA data was being dropped, caused files to be unreadable
5080 - memory overflow with non-LOV OST caused memory corruption
5081 - fixed regression tests to work with new config tools, obdfilter
5082 - fixed bug when directory size became larger than 1 block
5083 - fixed bug (for single client case) when PWD was deleted
5084 - invalidate local directory pages when doing intent-based ops
5085 - avoid LDLM oops when lock callback contained bad data
5087 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5089 * documentation updates
5090 * bug fixes and cleanups
5091 - configuration tools
5094 - 64-bit compile warnings
5095 - 64-bit internal statfs data
5097 * test_brw on persistent OST devices
5099 * lprocfs (disabled)
5101 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5103 * documentation updates
5104 * bug fixes and cleanups
5105 * configuration tools
5107 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5110 * many small fixes to 0_5_4
5111 * io/network handling
5112 * thinkos in MDS operations
5114 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5117 * crucial basic fixes to 0.5.3
5118 * IOR, Iozone work over Elan
5121 2002-08-07 Phil Schwan <phil@clusterfs.com>
5122 * version 0_5_3, our first alpha
5123 * we use the new Portals iovs
5124 * documentation updates
5125 * bug fixes and cleanups
5126 * small changes in the DLM wire protocol
5128 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5129 * version 0_5_1 with some initial stability,
5130 * locking on MD and file I/O.
5131 * documentation updates
5132 * several bug fixes since 0.5.0
5133 * small changes in wire protocol
5135 2002-07-18 Phil Schwan <phil@clusterfs.com>
5137 * delivered as Lustre Light Alpha
5138 * fixed a crash after handling invalid MDS requests
5139 * fixed directory pages for architectures with non-4k pages sizes
5141 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5142 * release version v0_4_4
5143 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5145 * Fixes a number of interruption problems with OST operations.
5146 * Update documentation for portals header changes
5147 * Move all wire protocol structs/defines to lustre_idl.h
5148 * Fixes symlink length bug.
5149 * Add tcpdump to repository.
5151 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5152 * release version v0_4_3
5153 * Fixes statfs for inodes on extN.
5154 * Fixes bug in runtests which would delete /etc/hosts.
5155 * Use 64-bit object IDs wherever possible (not into VFS though)
5156 Remove ost_get_info, which is unused by lustre, and out of date.
5158 2002-07-03 Peter Braam <braam@clusterfs.com>
5159 * release version v0_4_2 Fixes a lookup error (type not passed)
5160 * move forward to head of Portals
5161 * move forward to latest Lustre kernel
5163 2002-06-25 Peter Braam <braam@clusterfs.com>
5164 * release version v0_4_1. Hopefully stable on single node use.