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).
16 Frequency : Only on FC kernels 2.6.22+
18 Description: oops in statahead
19 Details : Do not drop reference count for the dentry from VFS when lookup,
20 VFS will do that by itself.
22 Severity : enhancement
24 Description: Generic /proc file permissions
25 Details : Set /Proc file permissions in a more generic way to enable non-
26 root users operate on some /proc files.
30 Description: Hitting mdc_commit_close() ASSERTION
31 Details : Properly handle request reference release in
32 ll_release_openhandle().
36 Frequency : only patchless client
37 Description: add workaround for race between add/remove dentry from hash
39 Severity : enhancement
41 Description: Allow OST glimpses to return PW locks
45 Description: LBUG when llog conf file is full
46 Details : When llog bitmap is full, ENOSPC should be returned for plain
51 Description: Prevent import from entering FULL state when server in recovery
55 Description: service mount cannot take device name with ":"
56 Details : Only when device name contains ":/" will mount treat it as
62 Description: replace ptlrpcd with the statahead thread to interpret the async
63 statahead RPC callback
67 Frequency : on recovery
68 Description: I/O failures after umount during fail back
69 Details : if client reconnected to restarted server we need join to recovery
70 instead of find server handler is changed and process self eviction
71 with cancel all locks.
73 Severity : enhancement
75 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
77 Severity : enhancement
79 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
81 Severity : enhancement
83 Description: Upcall on Lustre log has been dumped
84 Details : Allow for a user mode script to be called once a Lustre log has
85 been dumped. It passes the filename of the dumped log to the
86 script, the location of the script can be specified via
87 /proc/sys/lnet/debug_log_upcall.
92 Description: avoid idr_remove called for id which is not allocated.
93 Details : Move assigment s_dev for clustered nfs to end of initialization,
94 to avoid problem with error handling.
99 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
100 Details : When connection is reused this not moved from CONN_UNUSED_HASH
101 into CONN_USED_HASH and this prodice warning when put connection
102 again in unused hash.
104 Severity : enhancement
106 Description: File striping can now be set to use an arbitrary pool of OSTs.
108 Severity : enhancement
110 Description: Export bytes_read/bytes_write count on OSC/OST.
114 Description: Early reply size mismatch, MGC loses connection
115 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
116 the connect flags are properly negotiated.
120 Description: Properly propagate oinfo flags from lov to osc for statfs
121 Details : restore missing copy oi_flags to lov requests.
123 Severity : enhancement
125 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
126 : l_getgroups(8), lst(8), routerstat(8)
127 Details : included man pages for llobdstat(8), llstat(8),
128 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
130 Severity : enhancement
132 Description: configure's --enable-quota should check the
133 : kernel .config for CONFIG_QUOTA
134 Details : configure is terminated if --enable-quota is passed but
135 : no quota support is in kernel
139 Description: Remove portals compatibility
140 Details : Remove portals compatibility, not interoperable with releases
145 Description: Resolve device initialization race
146 Details : Prevent proc handler from accessing devices added to the
147 obd_devs array but yet be intialized.
149 Severity : enhancement
151 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
153 Severity : enhancement
155 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
160 Description: Using wrong pointer in osc_brw_prep_request
161 Details : Access to array[-1] can produce panic if kernel compiled with
162 CONFIG_PAGE_ALLOC enabled
166 Description: Client runs out of low memory
167 Details : Consider only lowmem when counting initial number of llap pages
171 Description: *optional* service tags registration
172 Details : if the "service tags" package is installed on a Lustre node
173 When the filesystem is mounted, a local-node service tag will
174 be created. See http://inventory.sun.com/ for more information
175 about the Service Tags asset management system.
179 Description: Kernel BUG tries to release flock
180 Details : Lustre does not destroy flock lock before last reference goes
181 away. So always drop flock locks when client is evicted and
182 perform unlock regardless of successfulness of speaking to MDS.
186 Description: add recount protection for osc callbacks, so avoid panic on shutdown
190 Description: sanity test 65a fails if stripecount of -1 is set
191 Details : handle -1 striping on filesystem in ll_dirstripe_verify
196 Description: ASSERTION(CheckWriteback(page,cmd)) failed
197 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
200 Severity : enhancement
202 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
206 Description: do not process already freed flock
207 Details : flock can possibly be freed by another thread before it reaches
208 to ldlm_flock_completion_ast.
212 Description: LBUG during stress test
213 Details : Need properly lock accesses the flock deadlock detection list.
217 Description: oops in page fault handler
218 Details : kernel page fault handler can return two special 'pages' in error case, don't
219 try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
223 Description: timeout with invalidate import.
224 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which should be
225 handled by ptlrpcd. This produce long age waiting and -ETIMEOUT
226 ptlrpc_invalidate_import and as result LASSERT.
228 Severity : enhancement
230 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
234 Description: enable MGS and MDT services start separately
235 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
236 a co-located MGS without starting the MGS, which is a complement
237 to 'nosvc' mount option.
241 Frequency : after recovery
242 Description: precreate to many object's after del orphan.
243 Details : del orphan st in oscc last_id == next_id and this triger growing
244 count of precreated objects. Set flag LOW to skip increase count
245 of precreated objects.
249 Frequency : rare, on clear nid stats
250 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
251 Details : when clean nid stats sometimes try destroy live entry,
252 and this produce panic in free.
256 Description: Stack overflow during MDS log replay
257 ease stack pressure by using a thread dealing llog_process.
261 Description: wait until IO finished before start new when do lock cancel.
262 Details : VM protocol want old IO finished before start new, in this case
263 need wait until PG_writeback is cleared until check dirty flag and
264 call writepages in lock cancel callback.
266 Severity : enhancement
268 Description: using special macro for print time and cleanup in includes.
272 Description: mds_mfd_close() ASSERTION(rc == 0)
273 Details : In mds_mfd_close(), we need protect inode's writecount change
274 within its orphan write semaphore to prevent possible races.
278 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
282 Frequency : rare, on shutdown ost
283 Description: don't hit live lock with umount ost.
284 Details : shrink_dcache_parent can be in long loop with destroy dentries,
285 use shrink_dcache_sb instead.
289 Description: don't panic with use echo client
290 Details : echo client pass NULL as client nid pointer and this produce null
295 Description: fix build on ppc32
296 Details : compile code with -m64 flag produce wrong object file for ppc32.
300 Description: add message levels for liblustreapi
304 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
305 Details : Move the 'good_osts' check before the 'total_bavail' check. This
306 will result in an -EAGAIN and in the exit call path we call
307 alloc_rr() which will with increasing aggressiveness attempt to
308 aquire precreated objects on the minimum number of required OSCs.
312 Description: Use old size assignment to avoid deadlock
313 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
314 the scheduling while holding a spinlock. We do not need locking
315 for size in ll_update_inode() because size is only updated from
316 the MDS for directories or files without objects, so there is no
317 other place to do the update, and concurrent access to such inodes
318 are protected by the inode lock.
322 Description: resolve "_IOWR redefined" build error on SLES10
326 Description: dump the memory debugging after all modules are unloaded to
327 suppress false negative in conf_sanity test 39
329 Severity : enhancement
331 Description: build kernel-ib packages for OFED 1.3 in our release cycle
336 Description: fix SLES kernel versioning
337 Details : the kernel version for our SLES 10 kernel did not include a "-"
338 before the "smp" at the end. while this was not a problem in
339 general, it did mean that software trying to use the kernel
340 version to try to detect a vendor specific kernel would fail.
341 this was most evident by the OFED build scripts.
345 Description: Don't update lov_desc members until making sure they are valid
346 Details : When updating lov_desc members via proc fs, need fix their
347 validities before doing the real update.
351 Description: don't put request into delay list while invalidate in flight.
352 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
353 import in flight. this produce timeout for invalidate and sometimes
354 can cause stale data.
358 Frequency : on ppc only
359 Description: not convert ost objects for directory because it's not exist.
360 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
364 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
366 Severity : enhancement
368 Description: Update to support 2.6.22.14 vanilla kernel.
372 Frequency : rare, on recovery
373 Description: read procfs can produce deadlock in some situation
374 Details : Holding lprocfs lock which send rpc can produce block for destroy
375 obd objects and this also block reconnect with -EALREADY. This isn't
376 fix all lprocfs bugs - but make it rare.
378 Severity : enhancement
380 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
383 Frequency : frequent on X2 node
385 Description: mdc_set_open_replay_data LBUG
386 Details : Set replay data for requests that are eligible for replay.
390 Description: lustre_mgs: operation 101 on unconnected MGS
391 Details : When MGC is disconnected from MGS long enough, MGS will evict the
392 MGC, and late on MGC cannot successfully connect to MGS and a lot
393 of the error messages complaining that MGS is not connected.
397 Frequency : on network error
398 Description: panic with double free request if network error
399 Details : mdc_finish_enqueue is finish request if any network error ocuring,
400 but it's true only for synchronus enqueue, for async enqueue
401 (via ptlrpcd) this incorrect and ptlrpcd want finish request
404 Severity : enhancement
406 Description: client-side metadata stat-ahead during readdir(directory readahead)
407 Details : perform client-side metadata stat-ahead when the client detects
408 readdir and sequential stat of dir entries therein
411 Frequency : on start mds
413 Description: Implement get_info(last_id) in obdfilter.
416 Frequency : occasional
418 Description: Correctly check stale fid, not start epoch if ost not support SOM
419 Details : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
420 overwrited with new generated fid, but mds can anwer with one of these
421 two fids and both is not stale. setattr incorectly start epoch and
422 assume will be called done_writeting, but without SOM done_writing
426 Frequency : rare, depends on device drivers and load
428 Description: MDS or OSS nodes crash due to stack overflow
429 Details : Code changes in 1.8.0 increased the stack usage of some functions.
430 In some cases, in conjunction with device drivers that use a lot
431 of stack the MDS (or possibly OSS) service threads could overflow
432 the stack. One change which was identified to consume additional
433 stack has been reworked to avoid the extra stack usage.
436 Frequency : occasional
438 Description: Do not fail import if osc_interpret_create gets -EAGAIN
439 Details : If osc_interpret_create got -EAGAIN it immediately exits and
440 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
441 oscc_has_objects and see OSC has no objests and call
442 oscc_internal_create to resend create request.
444 Severity : enhancement
446 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
448 Severity : enhancement
450 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
453 Frequency : very rare
455 Description: Fix replay if there is an un-replied request and open
456 Details : In some cases, older replay request will revert the
457 mcd->mcd_last_xid on MDS which is used to record the client's
460 Severity : enhancement
462 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
464 Severity : enhancement
466 Description: Add rhel5 support to HEAD.
468 Serverity : enhancement
470 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
475 Description: Don't allow skipping OSTs if index has been specified.
476 Details : Don't allow skipping OSTs if index has been specified, make locking
477 in internal create lots better.
481 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
482 Details : access to bitfield in structure is always rounded to long
483 and this produce problem with not atomic change any bit.
487 Description: Lustre make rpms failed.
488 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
489 builds Lustre rpms from tarball.
492 Frequency : rare on shutdown ost
494 Description: If llog cancel was not send before clean_exports phase, this can
495 produce deadlock in llog code.
496 Details : If llog thread has last reference to obd and call class_import_put
497 this produce deadlock because llog_cleanup_commit_master wait when
498 last llog_commit_thread exited, but this never success because was
499 called from llog_commit_thread.
503 Description: allow userland application know is lost one of stripes.
504 Details : fill lvb_blocks with error code on ost and return it to
505 application if error flag found.
509 Description: NULL lov_tgts causing MDS oops
510 Details : more safe checks for NULL lov_tgts for avoid oops.
512 Severity : enhancement
514 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
518 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
519 Details : If an OST is down the MDS will hang indefinitely in
520 obd_statfs_rqset() waiting for the statfs data. While for
521 MDS QOS usage of statfs, it should not stuck in waiting.
523 Severity : enhancement
525 Description: remote_acl support
526 Details : Support ACL-based permission check for remote user.
527 Support setfacl/getfacl for remote user with the utils
528 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
529 the system "{s,g}etfacl" utils.
531 Severity : enhancement
533 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
535 Severity : enhancement
537 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
541 Description: make mgs_setparam() handle fsname containing dash
542 Details : fsname containing a dash does not work with lctl conf_param
544 Severity : enhancement
546 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
548 Severity : enhancement
550 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
554 Frequency : rare, at shutdown
555 Description: access already free / zero obd_namespace.
556 Details : if client_disconnect_export was called without force flag set,
557 and exist connect request in flight, this can produce access to
558 NULL pointer (or already free pointer) when connect_interpret
559 store ocd flags in obd_namespace.
563 Frequency : only at startup
564 Description: not alloc memory with spinlock held.
565 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
566 if any spinlock held.
568 Severity : enhancement
570 Description: make lustre randomly fail allocating memory
571 Details : Make lustre randomly failed allocating memory for testing purpose.
573 Severity : enhancement
575 Description: lost problems with lov objid file
576 Details : Fixes some scability and access to not inited memory problems
577 in work with lov objdid file.
582 Description: lfs find does not continue on file error
583 Details : Continue other files processing when a file/dir is absent.
587 Description: Inconsistent usage of lustre_pack_reply()
588 Details : Standardize the usage of lustre_pack_reply() such that it
589 always generate a CERROR on failure.
592 Frequency : occasional
594 Description: MDS hang or stay in waiting lock
595 Details : If client receive lock with CBPENDING flag ldlm need send lock
596 cancel as separate rpc, to avoid situation when cancel request
597 can't processed due all i/o threads stay in wait lock.
601 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
605 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
609 Description: Update to SLES9 kernel-2.6.5-7.287.3.
613 Description: Update to RHEL4 latest kernel.
615 Severity : enhancement
617 Description: Build SLES10 patchless client fails
618 Details : The configure was broken by run ./configure with
619 --with-linux-obj=.... argument for patchless client. When the
620 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
621 can't search header adequately. Use absolute path such as
622 -I($LINUX)/include instead.
626 Description: interrupt oig_wait produce painc on resend.
627 Details : brw_redo_request can be used for resend requests from ptlrpcd and
628 private set, and this produce situation when rq_ptlrpcd_data not
629 copyed to new allocated request and triggered LBUG on assert
630 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
631 ptlrpcd set if request is changed and can be safety changed to use
636 Description: LASSERT_{REQ,REP}SWAB macros are buggy
637 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
638 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
639 these macros and replace them with their difinitions instead.
643 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
644 Details : Update which_patch & target file for SLES10 latest kernel.
647 Description: Remove client patches from SLES 10 kernel.
648 Details : This causes SLES 10 clients to behave as patchless clients
649 even on a Lustre-patched (server) kernel.
651 Severity : enhancement
653 Description: self-adjustable client's lru lists
654 Details : use adaptive algorithm for managing client cached locks lru
655 lists according to current server load, other client's work
656 pattern, memory activities, etc. Both, server and client
657 side namespaces provide number of proc tunables for controlling
660 Severity : enhancement
662 Description: light-weight GSS support
663 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
664 and reduce performance overhead.
666 Severity : enhancement
668 Description: Linux keyring support
669 Details : Support using service of Linux keyring for Lustre GSS internal
670 context refresh/cache mechanism.
674 Description: Fix errors in lfs documentation
675 Details : Fixes man pages
679 Description: don't use GFP_* in generic Lustre code.
680 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
684 Description: obdclass is limited by single OBD_ALLOC(idarray)
685 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
689 Description: warnings with build patchless client with vanila 2.6.19 and up
690 Details : change old ctl_table style and replace ctl_table/ctl_table_header
691 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
695 Frequency : Only for SLES
696 Description: sanity_quota fail test_1
697 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
698 file, which makes TSTID[2] inunique.
702 Frequency : Only for SLES9
703 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
704 Details : Update target/ChangeLog/which_patch .
708 Frequency : Only for test_10 in sanity-quota.sh
709 Description: a bug in quota test code
710 Details : checking a test flag in a wrong place
714 Frequency : Only for test_13 in sanity-quota.sh
715 Description: a bug in quota test script
716 Details : shouldn't check overfull allocated quota
720 Description: jbd statistics
721 Details : Port older jbd statistics patch for sles10
723 Severity : enhancement
725 Description: Add printing inode info into message about error in writepage.
730 Description: reply_lock_interpret crash due to race with it and lock cancel.
731 Details : Do not replay locks that are being cancelled. Do not reference
732 locks by their address during replay, just by their handle.
736 Frequency : When flocks are used.
737 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
738 Details : Flock locks might destroy just granted lock if it could be merged
739 with another existing flock, this is done in completion handler,
740 so teach ldlm_cli_enquque_fini that this is a valid case for
746 Description: Oops in read and write path when failing to allocate lock.
747 Details : Check if lock allocation failed and return error back.
751 Description: lstripe command fails for valid OST index
752 Details : The stripe offset is compared to lov->desc.ld_tgt_count
753 instead of lov->desc.ld_active_tgt_count.
757 Description: sanity.sh failed test 103
758 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
763 Description: df doesn't work properly if diskfs blocksize != 4K
764 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
768 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
769 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
771 Severity : enhancement
773 Description: Adaptive timeouts
774 Details : RPC timeouts adapt to changing server load and network
775 conditions to reduce resend attempts and improve recovery time.
779 Description: llapi_file_create() does not allow some changes
780 Details : add llapi_file_open() that allows specifying the mode and
781 open flags, and also returns an open file handle.
785 Description: merge and cleanup kernel patches.
786 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
790 Description: Add journal checksum support.(Kernel part)
791 Details : The journal checksum feature adds two new flags i.e
792 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
793 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
794 indicates that the commit block contains the checksum for
795 the blocks described by the descriptor blocks. Now commit
796 record can be sent to disk without waiting for descriptor
797 blocks to be written to disk. This behavior is controlled
798 using JBD2_FEATURE_ASYNC_COMMIT flag.
802 Description: OSS needs mutliple precreate threads
803 Details : Add ability to start more than one create thread per OSS.
807 Description: Sanity.sh test_65k failed: setstripe should have succeeded
808 Details : execute lfs setstripe on client
812 Description: mds_obd_create error creating tmp object
813 Details : When the user sets quota on root, llog will be affected and can't
814 create files and write files.
817 Frequency : Always on ia64 patchless client, and possibly others.
819 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
820 Details : This allows the patchless client to be loaded on architectures
825 Description: RedHat Update kernel for RHEL5
826 Details : Add the kernel config file for RHEL5.
830 Description: RedHat Update kernel for RHEL5
831 Details : Modify the kernel config file more closer RHEL5.
835 Description: Build failure against Centos5 (RHEL5)
836 Details : Define PAGE_SIZE when it isn't present.
840 Description: test_7 on sanity-quota.sh can't be used on separate machines
841 Details : add facet to handle it
845 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
846 Details : with NFS, the anon dentry's parent was set to itself in
847 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
848 corresponding dentry other than use rec->ur_name.
850 Severity : enhancement
852 Description: lfs setstripe enhancement
853 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
857 Description: enable data checksumming by default
858 Details : enable checksum by default, allow --disable-checksum
859 configure option and "-o nochecksum" mount option. Checksums
860 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
864 Description: lustre support for RHEL5
865 Details : Add support for RHEL5.
869 Description: Client eviction due to failover config
870 Details : after a connection loss, the lustre client should attempt to
871 reconnect to the last active server first before trying the
872 other potential connections.
876 Description: when mds and osts use different quota unit(32bit and 64bit),
877 quota will be released repeatly.
878 Details : void sending multiple quota reqs to mds, which will keep the
879 status between the reqs.
883 Description: osts not allocated evenly to files
884 Details : change the condition to increase offset_idx
887 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
889 Description: Data corruption for OSTs that are formatted larger than 2TB
891 Details : When generating the bio request for lustre file writes the
892 sector number would overflow a temporary variable before being
893 used for the IO. The data reads correctly from Lustre (which
894 will overflow in a similar manner) but other file data or
895 filesystem metadata may be corrupted in some cases.
899 Description: Tune the kernel for good SCSI performance.
900 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
901 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
906 Description: rewrite ext2-derived code in obdclass/uuid.c
907 Details : rewrite inherited code (uuid parsing code from ext2 utils)
908 from scratch preserving functionality.
912 Description: "lfs find -obd UUID" prints directories
913 Details : "lfs find -obd UUID" will return all directory names instead
914 of just file names. It is incorrect because the directories
915 do not reside on the OSTs.
919 Description: use i_size_read and i_size_write in 2.6 port
920 Details : replace inode->i_size access with i_size_read/write()
923 Frequency : when removing large files
925 Description: scheduling issue during removal of large Lustre files
926 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
927 It causes scheduling issues when removing large files (17TB in the
931 Frequency : only with liblustre clients on XT3
933 Description: evictions taking too long
934 Details : allow llrd to evict clients directly on OSTs
937 Frequency : only on ppc
939 Description: /proc/fs/lustre/devices broken on ppc
940 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
941 We should make sure the type of 'index' is loff_t and then cast
942 explicitly as needed below. Do not assign an explicitly cast
946 Frequency : only for rhel5
948 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
949 Details : Modify the target file & which_kernel.
951 Severity : enhancement
953 Description: omit set fsid for export NFS
954 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
955 as FS_REQUIRES_DEV for avoid problems with generate fsid.
959 Frequency : Always, for kernels after 2.6.16
960 Description: Fix warning idr_remove called for id=.. which is not allocated.
961 Details : Last kernels save old s_dev before kill super and not allow
962 to restore from callback - restore it before call kill_anon_super.
966 Description: buffer overruns could theoretically occur
967 Details : llapi_semantic_traverse() modifies the "path" argument by
968 appending values to the end of the origin string, and a
969 overrun may occur. Adding buffer overrun check in liblustreapi.
973 Description: Fix error on 'ls .' at the top of the Lustre mount.
974 Details : Don't revalidate dentry if it is a root dentry.
978 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
979 Details : Modify vm-tunables-rhel4.patch.
983 Description: Kernel config for 2.6.18-vanilla.
984 Details : Modify targets/2.6-vanilla.target.in.
985 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
986 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
987 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
988 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
992 Description: improve handling recoverable errors
993 Details : if request processig with error which can be recoverable on server
994 request should be resend, otherwise page released from cache and
999 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1000 Details : Modify target file & which_patch.
1001 A flaw was found in the IA32 system call emulation provided
1002 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1003 value could be stored in the %RAX register, which could trigger an
1004 out-of-bounds system call table access. An untrusted local user
1005 could exploit this flaw to run code in the kernel
1006 (ie a root privilege escalation). (CVE-2007-4573).
1010 Description: O_DIRECT bypasses client statistics.
1011 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1012 read_ahead_stats were updated. copied the stats section from
1013 osc_send_oap_rpc() into async_internal().
1017 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1021 Description: change order of libsysio includes
1022 Details : '#include sysio.h' should always come before '#include xtio.h'
1026 Description: handle "serious error: objid * already exists" more gracefully
1027 Details : If LAST_ID value on disk is smaller than the objects existing in
1028 the O/0/d* directories, it indicates disk corruption and causes an
1029 LBUG(). If the object is 0-length, then we should use the existing
1030 object. This will help to avoid a full fsck in most cases.
1032 Severity : enhancement
1034 Description: adapt the lustre_config script to support the upgrade case
1035 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1036 server targets to 1.6 in parallel.
1040 Description: To avoid grant space > avaible space when the disk is almost
1041 full. Without this patch you might see the error "grant XXXX >
1042 available" or some LBUG about grant, when the disk is almost
1044 Details : In filter_check_grant, for non_grant cache write, we should
1045 check the left space by if (*left > ungranted + bytes), instead
1046 of (*left > ungranted), because only we are sure the left space
1047 is enough for another "bytes", then the ungrant space should be
1048 increase. In client, we should update cl_avail_grant only there
1049 is OBD_MD_FLGRANT in the reply.
1053 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1054 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1055 described in CVE-2007-4573. This problem could allow untrusted
1056 local users to gain root access.
1059 Frequency : when using O_DIRECT and quotas
1061 Description: Incorrect file ownership on O_DIRECT output files
1062 Details : block usage reported by 'lfs quota' does not take into account
1063 files that have been written with O_DIRECT.
1068 Description: touch file failed when fs is not full
1069 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1070 otherwise we can get ENOSP while fs is not full.
1074 Description: parallel lock callbacks
1075 Details : Instead of sending blocking and completion callbacks as separated
1076 requests, adding them to a set and sending in parallel.
1079 Frequency : only for Cray XT3
1080 Bugzilla : 12829/13455
1081 Description: Changing primary group doesn't change the group lustre assigns to
1083 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1084 overridden with the primary group provided by the group upcall,
1085 whereas the supplied fsgid can be trusted if it is in the list of
1086 supplementary groups returned by the group upcall.
1088 Severity : enhancement
1090 Description: Allow masking D_WARNING, D_ERROR messages from console
1091 Details : Console messages can now be disabled via lnet.printk.
1095 Description: User code with malformed file open parameter crashes client node
1096 Details : Before packing join_file req, all the related reference should be
1097 checked carefully in case some malformed flags cause fake
1098 join_file req on client.
1102 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1103 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1104 mode is equal to the granted mode. However, it should also check
1105 that LDLM_FL_CP_REQD is not set.
1109 Description: Heavy nfs access might result in deadlocks
1110 Details : After ELC code landed, it is now improper to enqueue any mds
1111 locks under och_sem, because enqueue might want to decide to
1112 cancel open locks for same inode we are holding och_sem for.
1116 Description: Client eviction while running blogbench
1117 Details : A lot of unlink operations with concurrent I/O can lead to a
1118 deadlock causing evictions. To address the problem, the number of
1119 oustanding OST_DESTROY requests is now throttled to
1120 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1121 callbacks are processed in priority.
1125 Description: enable ACLs on MDS by default
1126 Details : ACLs must be enabled on MDS by default.
1129 Frequency : PPC/PPC64 only
1131 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1132 Details : fix duplicated definitions between asm-ppc64/types.h and
1133 lustre_types.h on PPC.
1136 Frequency : PPC/PPC64 only
1138 Description: asm-ppc/segment.h does not exist
1139 Details : fix compile issue on PPC.
1143 Description: better handle error messages in extents code
1146 Frequency : RHEL4 only
1148 Description: mkfs is very slow on IA64/RHEL4
1149 Details : A performance regression has been discovered in the MPT Fusion
1150 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1151 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1152 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1154 Severity : enhancement
1156 Description: SNMP support enhancement
1157 Details : Adding total number of sampled request for an MDS node in snmp
1160 Severity : enhancement
1162 Description: Optimize ldlm waiting list processing for PR extent locks
1163 Details : When processing waiting list for read extent lock and meeting read
1164 lock that is same or wider to it that is not contended, skip
1165 processing rest of the list and immediatelly return current
1166 status of conflictness, since we are guaranteed there are no
1167 conflicting locks in the rest of the list.
1171 Description: Time out and refuse to reconnect
1172 Details : When the failover node is the primary node, it is possible
1173 to have two identical connections in imp_conn_list. We must
1174 compare not conn's pointers but NIDs, otherwise we can defeat
1175 connection throttling.
1179 Description: port llog fixes from b1_6 into HEAD
1180 Details : Port llog reference couting and some llog cleanups from b1_6
1181 (bug 10800) into HEAD, for protect from panic and access to already
1182 free llog structures.
1186 Description: Detect stride IO mode in read-ahead
1187 Details : When a client does stride read, read-ahead should detect that and
1188 read-ahead pages according to the detected stride pattern.
1192 Description: data checksumming impacts single node performance
1193 Details : add support for several checksum algorithm. Currently, only CRC32
1194 and Adler-32 are supported. The checksum type can be changed on
1195 the fly via /proc/fs/lustre/osc/*/checksum_type.
1199 Description: use adler32 for page checksums
1200 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1205 Description: build for x2 fails
1206 Details : fix compile issue on Cray systems.
1210 Description: Properly match for duplicate locks
1211 Details : Due to different lock order from skiplists code, we need to
1212 traverse entire list for now
1215 Frequency : only on PPC/SLES10
1217 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1218 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1219 be defined. Add a hack in mkfs_lustre.c to work around this compile
1224 Description: LASSERT on MDS when client holding flock lock dies
1225 Details : ldlm pool logic depends on number of granted locks equal to
1226 number of released locks which is not true for flock locks, so
1227 just exclude such locks from consideration.
1231 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1232 Details : Use fsfilt_sync() for both the whole filesystem sync and
1233 individual file sync to eliminate dangerous inode locking
1234 with I_LOCK that can lead to a deadlock.
1238 Description: performance in 1.6.3
1239 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1240 sure that 1MB requests are not fragmented by the block layer.
1242 Severity : enhancement
1244 Description: organize the server-side client stats on per-nid basis
1245 Details : Change the structure of stats under obdfilter and mds to
1253 The "uuid"s file would list the uuids of _active_ exports.
1254 And the clear entry is to clear all stats and stale nids.
1256 Severity : enhancement
1258 Description: eliminate client locks in face of contention
1259 Details : file contention detection and lockless i/o implementation
1260 for contended files.
1264 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1267 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1269 Description: MDS may lose file striping (and hence file data) in some cases
1270 Details : If there are additional extended attributes stored on the MDS,
1271 in particular ACLs, SELinux, or user attributes (if user_xattr
1272 is specified for the client mount options) then there is a risk
1273 of attribute loss. Additionally, the Lustre file striping
1274 needs to be larger than default (e.g. striped over all OSTs),
1275 and an additional attribute must be stored initially in the
1276 inode and then increase in size enough to be moved to the
1277 external attribute block (e.g. ACL growing in size) for file
1282 Description: skiplist implementation simplification
1283 Details : skiplists are used to group compatible locks on granted list
1284 that was implemented as tracking first and last lock of each
1285 lock group the patch changes that to using doubly linked lists
1289 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1290 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1291 because LASSERT on both the data supplied by a client, and the
1292 data on disk is dangerous and incorrect.
1294 Severity : enhancement
1296 Description: Slow truncate/writes to huge files at high offsets.
1297 Details : Directly associate cached pages to lock that protect those pages,
1298 this allows us to quickly find what pages to write and remove
1299 once lock callback is received.
1303 Description: more ldlm soft lockups
1304 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1305 starve other threads from the resource lock for a long time in
1306 case of long waiting queue, so change the debug level from
1307 D_OTHER to the less frequently used D_INFO.
1309 Severity : enhancement
1311 Description: add -gid, -group, -uid, -user options to lfs find
1315 Description: Hung threads in invalidate_inode_pages2_range
1316 Details : The direct IO path doesn't call check_rpcs to submit a new RPC once
1317 one is completed. As a result, some RPCs are stuck in the queue
1322 Description: filter threads hungs on waiting journal commit
1323 Details : Cleanup filter group llog code, then only filter group llog will
1324 be only created in the MDS/OST syncing process.
1328 Description: Procfs and llog threads access destoryed import sometimes.
1329 Details : Sync the import destoryed process with procfs and llog threads by
1330 the import refcount and semaphore.
1332 Severity : enhancement
1334 Description: openlock cache of b1_6 port to HEAD
1339 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1340 Details : If insertion of an extent fails, then discard the inode
1341 preallocation and free data blocks else it can lead to duplicate
1346 Description: don't always update ctime in ext3_xattr_set_handle()
1347 Details : Current xattr code updates the inode ctime in ext3_xattr_set_handle.
1348 In some cases the ctime should not be updated, for example for
1349 2.0->1.8 compatibility it is necessary to delete an xattr and it
1350 should not update the ctime.
1354 Bugzilla : 15713/16362
1355 Description: Assertion in iopen_connect_dentry in 1.6.3
1356 Details : looking up an inode via iopen with the wrong generation number can
1357 populate the dcache with a disconneced dentry while the inode
1358 number is in the process of being reallocated. This causes an
1359 assertion failure in iopen since the inode's dentry list contains
1360 both a connected and disconnected dentry.
1364 Description: assertion failure in ldlm_handle2lock()
1365 Details : fix a race between class_handle_unhash() and class_handle2object()
1366 introduced in lustre 1.6.5 by bug 13622.
1371 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1372 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1377 Description: oops in sd_iostats_seq_show()
1378 Details : unloading/reloading the scsi low level driver triggers a kernel
1379 bug when trying to access the sd iostat file.
1384 Description: Kernel panics during QLogic driver reload
1385 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1386 patch, causing memory corruption.
1391 Description: journal_dev option does not work in b1_6
1392 Details : pass mount option during pre-mount.
1394 Severity : enhancement
1396 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1397 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1398 extent information of a file. It can be used to map logical blocks
1399 in a file to physical blocks in the block device.
1403 Description: LDLM soft lockups - improvement
1404 Details : It is be possible to send the lock handle along with each read
1405 or write request because the client is already doing a lock match
1406 itself so there isn't any reason the OST should have to re-do that
1410 Frequency : only on Cray X2
1412 Description: X2 build failures
1413 Details : fix build failures on Cray X2.
1417 Description: xid & resent requests
1418 Details : Initialize RPC XID from clock at startup (randomly if clock is
1421 Severity : enhancement
1423 Description: Add lustre_start utility to start or stop multiple Lustre servers
1428 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
1429 Details : In case of memory pressure, list_del() can be called twice on
1430 req->rq_history_list, causing a kernel oops.
1432 --------------------------------------------------------------------------------
1434 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
1436 * Support for kernels:
1437 2.6.5-7.283 (SLES 9),
1438 2.6.9-55.EL (RHEL 4),
1439 2.6.16.46-0.14 (SLES 10),
1440 2.6.18.8 vanilla (kernel.org)
1441 * Client support for unpatched kernels:
1442 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1443 2.6.16 - 2.6.22 vanilla (kernel.org)
1444 * Due to recently discovered recovery problems, we do not recommend
1445 using patchless RHEL 4 clients with this or any earlier release.
1446 * Recommended e2fsprogs version: 1.39.cfs8
1447 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1448 * Starting with this release, the ldiskfs backing filesystem required
1449 by Lustre is now in its own package, lustre-ldiskfs. This package
1450 should be installed. It is versioned separately from Lustre and
1451 may be released separately in future.
1455 Description: block reactivating mgc import until all deactivates complete
1456 Details : Fix race when failing back MDT/MGS to itself (testing)
1459 Frequency : at statup only
1461 Description: mds_lov_synchronize race leads to various problems
1462 Details : simultaneous MDT->OST connections at startup can cause the
1463 sync to abort, leaving the OSC in a bad state.
1465 Severity : enhancement
1467 Description: add optional extra BUILD_VERSION info
1468 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1469 to override the lustre version.
1472 Frequency : 2.6.18 servers only
1474 Description: ll_kern_mount() doesn't release the module reference
1475 Details : The ldiskfs module reference count never drops down to 0
1476 because ll_kern_mount() doesn't release the module reference.
1481 Description: server LBUG when using old ost_num_threads parameter
1482 Details : Accept the old ost_num_threads parameter but warn that it
1483 is deprecated, and fix an off-by-one error that caused an LBUG.
1488 Description: Transient SCSI error results in persistent IO issue
1489 Details : iobuf->dr_error is not reinitialized to 0 between two
1493 Frequency : sometimes when underlying device returns I/O errors
1495 Description: OSTs not going read-only during write failures
1496 Details : OSTs are not remounted read-only when the journal commit threads
1497 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1498 instead of the ext3 wrappers.
1502 Description: poor connect scaling with increasing client count
1503 Details : Don't run filter_grant_sanity_check for more than 100 exports
1504 to improve scaling for large numbers of clients.
1507 Frequency : SLES10 only
1509 Description: sanity-quota.sh quotacheck failed: rc = -22
1510 Details : Quotas cannot be enabled on SLES10.
1513 Frequency : liblustre clients only
1515 Description: getdirentries does not give error when run on compute nodes
1516 Details : getdirentries does not fail when the size specified as an argument
1517 is too small to contain at least one entry
1519 Severity : enhancement
1521 Description: Add LNET router traceability for debug purposes
1522 Details : If a checksum failure occurs with a router as part of the
1523 IO path, the NID of the last router that forwarded the bulk data
1524 is printed so it can be identified.
1529 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1530 Details : Due to a race condition, liblustre clients were occasionally
1531 evicted incorrectly.
1533 Severity : enhancement
1535 Description: lfs setstripe use optional parameters instead of postional
1538 Severity : enhancement
1540 Description: Nanosecond timestamp support for ldiskfs
1541 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1542 resolution timestamps. There is not yet support for this at
1543 the Lustre filesystem level.
1546 Frequency : during server recovery
1548 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1549 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1553 Frequency : nfs export on patchless client
1555 Description: connectathon hang when test nfs export over patchless client
1556 Details : Disconnected dentry cannot be found with lookup, so we do not need
1557 to unhash it or make it invalid
1560 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1561 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1562 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1567 Description: open req refcounting wrong on reconnect
1568 Details : If reconnect happened between getting open reply from server and
1569 call to mdc_set_replay_data in ll_file_open, we will schedule
1570 replay for unreferenced request that we are about to free.
1571 Subsequent close will crash in variety of ways.
1572 Check that request is still eligible for replay in
1573 mdc_set_replay_data().
1578 Description: disable writes to filesystem when reading health_check file
1579 Details : the default for reading the health_check proc file has changed
1580 to NOT do a journal transaction and write to disk, because this
1581 can cause reads of the /proc file to hang and block HA state
1582 checking on a healthy but otherwise heavily loaded system. It
1583 is possible to return to the previous behaviour during configure
1584 with --enable-health-write.
1586 Severity : enhancement
1588 Description: 64-bit inode version
1589 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1590 to the inode. This will be required for version-based recovery.
1595 Description: MDS fails to start if a duplicate client export is detected
1596 Details : in some rare cases it was possible for a client to connect to
1597 an MDS multiple times. Upon recovery the MDS would detect this
1598 and fail during startup. Handle this more gracefully.
1600 Severity : enhancement
1602 Description: Add -o localflock option to simulate old noflock
1604 Details : This will achieve local-only flock/fcntl locks
1610 Description: log_commit_thread vs filter_destroy race leads to crash
1611 Details : Take import reference before releasing llog record semaphore
1616 Description: Wrong request locking in request set processing
1617 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1618 imp_delayed_list, in this place should be used imp_lock.
1621 Frequency : when reconnection
1623 Description: Grant Leak when osc reconnect to OST
1624 Details : When osc reconnect ost, OST(filter) should check whether it
1625 should grant more space to client by comparing fed_grant and
1626 cl_avail_grant, and return the granted space to client instead
1627 of "new granted" space, because client will call osc_init_grant
1628 to update the client grant space info.
1631 Frequency : when client reconnect to OST
1633 Description: Grant Leak when osc do resend and replay bulk write
1634 Details : When osc reconnect to OST, OST(filter)should clear grant info of
1635 bulk write request, because the grant info while be sync between
1636 OSC and OST when reconnect, and we should ignore the grant info
1637 these of resend/replay write req.
1642 Description: Grant space more than avaiable left space sometimes.
1643 Details : When then OST is about to be full, if two bulk writing from
1644 different clients came to OST. Accord the avaliable space of the
1645 OST, the first req should be permitted, and the second one
1646 should be denied by ENOSPC. But if the seconde arrived before
1647 the first one is commited. The OST might wrongly permit second
1648 writing, which will cause grant space > avaiable space.
1651 Frequency : when client is evicted
1653 Description: Grant might be wrongly erased when osc is evicted by OST
1654 Details : when the import is evicted by server, it will fork another
1655 thread ptlrpc_invalidate_import_thread to invalidate the
1656 import, where the grant will be set to 0. While the original
1657 thread will update the grant it got when connecting. So if
1658 the former happened latter, the grant will be wrongly errased
1659 because of this race.
1664 Description: Checking Stale with correct fid
1665 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1666 to check whether it is stale, because sometimes, we want the
1667 enqueue happened anyway, and op_data.fid2 will not be initialized.
1669 Severity : enhancement
1671 Description: update patchless client
1672 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1675 Frequency : only with 2.4 kernel
1677 Description: random memory corruption
1678 Details : size of struct ll_inode_info is to big for union inode.u and this
1679 can be cause of random memory corruption.
1684 Description: Memory leak in recovery
1685 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1686 It should also check obd_fail before fsfilt_start, otherwise if
1687 fsfilt_start return -EROFS,(failover mds during mds recovery).
1688 then the req will return with repmsg->transno = 0 and rc = EROFS.
1689 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1690 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1691 be freed no matter whether fsfilt_commit success or not.
1694 Frequency : only with huge count clients
1696 Description: Prevents from taking the superblock lock in llap_from_page for
1698 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1699 for llap_from_page call in ll_removepage prevents from taking the
1700 superblock lock for a soon died page.
1705 Description: Not check open intent error before release open handle
1706 Details : in some rare cases, the open intent error is not checked before
1707 release open handle, which may cause
1708 ASSERTION(open_req->rq_transno != 0), because it tries to release
1709 the failed open handle.
1714 Description: Set cat log bitmap only after create log success.
1715 Details : in some rare cases, the cat log bitmap is set too early. and it
1716 should be set only after create log success.
1720 Description: Accessing a block bevice can re-enable I/O when Lustre is
1721 tearing down a device.
1722 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1727 Description: service threads may hog cpus when there are a lot of requests
1728 Details : Insert cond_resched to give other threads a chance to use some CPU
1733 Description: the cat log was not initialized in recovery
1734 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1735 unlink log on mds will not be initialized until mds post
1736 recovery. And also in mds post recovery, the unlink log will
1737 initialization will be done asynchronausly, so there will be race
1738 between add unlink log and unlink log initialization.
1742 Description: brw_stats were being printed incorrectly
1743 Details : brw_stats were being printed as log2 but all of them were not
1744 recorded as log2. Also remove some code duplication arising from
1745 filter_tally_{read,write}.
1749 Frequency : rare, only in recovery.
1750 Description: ASSERTION(req->rq_type != LI_POISON) failed
1751 Details : imp_lock should be held while iterating over imp_sending_list for
1752 prevent destroy request after get timeout in ptlrpc_queue_wait.
1756 Description: replay-single.sh test 52 fails
1757 Details : A lock's skiplist need to be cleanup when it being unlinked
1758 from its resource list.
1762 Description: Short directio read returns full requested size rather than
1764 Details : Direct I/O operations should return actual amount of bytes
1765 transferred rather than requested size.
1767 Severity : enhancement
1769 Description: metadata RPC reduction (e.g. for rm performance)
1770 Details : decrease the amount of synchronous RPC between clients and servers
1771 by canceling conflicing lock before the operation on the client
1772 and packing thier handles into the main operation RPC to server.
1774 Severity : enhancement
1776 Description: Async OSC create to avoid the blocking unnecessarily.
1777 Details : If a OST has no remain object, system will block on the creating
1778 when need to create a new object on this OST. Now, ways use
1779 pre-created objects when available, instead of blocking on an
1780 empty osc while others are not empty. If we must block, we block
1781 for the shortest possible period of time.
1785 Description: improve handling recoverable errors
1786 Details : if request processig with error which can be recoverable on server
1787 request should be resend, otherwise page released from cache and
1790 Severity : enhancement
1792 Description: refine locking for avoid write wrong info into lov_objid file
1793 Details : fix possible races with add new target and write/update data in
1796 --------------------------------------------------------------------------------
1798 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1803 Frequency : on some architectures
1805 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1806 Details : Uninitialized flags sometimes cause configuration commands to
1810 Frequency : patchless clients only
1812 Description: missing __iget() symbol export
1813 Details : The __iget() symbol export is missing. To avoid the need for
1814 this on patchless clients the deathrow inode reaper is turned
1815 off, and we depend on the VM to clean up old inodes. This
1816 dependency was during via the fix for bug 12181.
1818 --------------------------------------------------------------------------------
1820 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1822 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1823 INTEROPERATE with older versions automatically. In many cases a
1824 special upgrade step is needed. Please read the
1825 user documentation before upgrading any part of a live system.
1826 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1827 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1828 * WARNING: Lustre configuration and startup changes are required with
1829 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1833 Severity : enhancement
1835 Description: MountConf
1836 Details : Lustre configuration is now managed via mkfs and mount
1837 commands instead of lmc and lconf. New obd types (MGS, MGC)
1838 are added for dynamic configuration management. See
1839 https://mail.clusterfs.com/wikis/lustre/MountConf for
1842 Severity : enhancement
1844 Description: dynamic OST addition
1845 Details : OSTs can now be added to a live filesystem
1847 Severity : enhancement
1849 Description: startup order invariance
1850 Details : MDTs and OSTs can be started in any order. Clients only
1851 require the MDT to complete startup.
1853 Severity : enhancement
1855 Description: parallel, asynchronous orphan cleanup
1856 Details : orphan cleanup is now performed in separate threads for each
1857 OST, allowing parallel non-blocking operation.
1859 Severity : enhancement
1861 Description: optimized stripe assignment
1862 Details : stripe assignments are now made based on ost space available,
1863 ost previous usage, and OSS previous usage, in order to try
1864 to optimize storage space and networking resources.
1866 Severity : enhancement
1868 Description: Permanently set tunables
1869 Details : All writable /proc/fs/lustre tunables can now be permanently
1870 set on a per-server basis, at mkfs time or on a live system.
1872 Severity : enhancement
1874 Description: Lustre message v2
1875 Details : Add lustre message format v2.
1877 Severity : enhancement
1879 Description: client OST exclusion list
1880 Details : Clients can be started with a list of OSTs that should be
1881 declared "inactive" for known non-responsive OSTs.
1885 Description: SPEC SFS validation failure on NFS v2 over lustre.
1886 Details : Changes the blocksize for regular files to be 2x RPC size,
1887 and not depend on stripe size.
1889 Severity : enhancement
1891 Description: Multiple MD RPCs in flight.
1892 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1893 To avoid overly-overloading MDS, introduce a limit on number of
1894 MDS RPCs in flight for a single client and add /proc controls
1895 to adjust this limit.
1897 Severity : enhancement
1899 Description: client read/write statistics
1900 Details : Add client read/write call usage stats for performance
1901 analysis of user processes.
1902 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1903 file access. extents_stats shows chunk size distribution.
1904 extents_stats_per_process show chunk size distribution per
1907 Severity : enhancement
1909 Description: mds statistics
1910 Details : Add detailed mds operations statistics in
1911 /proc/fs/lustre/mds/*/stats.
1915 Description: Failure of copying files with lustre special EAs.
1916 Details : Client side always return success for setxattr call for lustre
1917 special xattr (currently only "trusted.lov").
1922 Description: Refcount LNET uuids
1923 Details : The global LNET uuid list grew linearly with every startup;
1924 refcount repeated list entries instead of always adding to
1927 Severity : enhancement
1929 Description: Dynamic service threads
1930 Details : Within a small range, start extra service threads
1931 automatically when the request queue builds up.
1934 Frequency : mixed-endian client/server environments
1936 Description: mixed-endian crashes
1937 Details : The new msg_v2 system had some failures in mixed-endian
1940 Severity : enhancement
1942 Description: Easy OST removal
1943 Details : OSTs can be permanently deactivated with e.g. 'lctl
1944 conf_param lustre-OST0001.osc.active=0'
1946 Severity : enhancement
1948 Description: MGS proc entries
1949 Details : Added basic proc entries for the MGS showing what filesystems
1952 Severity : enhancement
1954 Description: provide MGS failover
1955 Details : Added config lock reacquisition after MGS server failover.
1957 Severity : enhancement
1959 Description: add Linux 2.4 support
1960 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1964 Description: a large application tries to do I/O to the same resource and dies
1965 in the middle of it.
1966 Details : Check the req->rq_arrival time after the call to
1967 ost_brw_lock_get(), but before we do anything about
1968 processing it & sending the BULK transfer request. This
1969 should help move old stale pending locks off the queue as
1970 quickly as obd_timeout.
1973 Frequency : when an incorrect nid is specified during startup
1975 Description: ptlrpc connect to non-existant node causes kernel crash
1976 Details : LNET can't be re-entered from an event callback, which
1977 happened when we expire a message after the export has been
1978 cleaned up. Instead, hand the zombie cleanup off to another
1981 Severity : enhancement
1983 Description: plain/inodebits lock performance improvement
1984 Details : Grouping plain/inodebits in granted list by their request modes
1985 and bits policy, thus improving the performance of search through
1989 Frequency : only if OST filesystem is corrupted
1991 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1992 Details : for a short time RPCs with bulk IO are in the replay list,
1993 but replay of bulk IOs is unimplemented. If the OST filesystem
1994 is corrupted due to disk cache incoherency and then replay is
1995 started it is possible to trip an assertion. Avoid putting
1996 committed RPCs into the replay list at all to avoid this issue.
1999 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2001 Description: System hang on startup
2002 Details : This bug allowed the liblustre (e.g. catamount) client to
2003 return to the app before handling all startup RPCs. This
2004 could leave the node unresponsive to lustre network traffic
2005 and manifested as a server ptllnd timeout.
2007 Severity : enhancement
2009 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2010 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
2011 boolean to control whether to print peer debug info when a
2012 client's RPC times out.
2015 Frequency : only for kernels with patches from Lustre below 1.4.3
2017 Description: Remove old rdonly API
2018 Details : Remove old rdonly API which unused from at least lustre 1.4.3
2021 Frequency : only for devices with external journals
2023 Description: Set external device read-only also
2024 Details : During a commanded failover stop, we set the disk device
2025 read-only while the server shuts down. We now also set any
2026 external journal device read-only at the same time.
2029 Frequency : when upgrading from 1.4 while trying to change parameters
2031 Description: The wrong (new) MDC name was used when setting parameters for
2032 upgraded MDT's. Also allows changing of OSC (and MDC)
2033 parameters if --writeconf is specified at tunefs upgrade time.
2036 Frequency : when setting specific ost indicies
2038 Description: QOS code breaks on skipped indicies
2039 Details : Add checks for missing OST indicies in the QOS code, so OSTs
2040 created with --index need not be sequential.
2042 Severity : enhancement
2044 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2045 Details : The uninit_groups feature works in conjunction with the kernel
2046 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2047 up the pass1 processing of e2fsck. This is a read-only feature
2048 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2049 mount filesystems that have had this feature enabled.
2051 Severity : enhancement
2053 Description: Improve multi-block allocation algorithm to avoid fragmentation
2054 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2055 allocation locality and avoid filesystem fragmentation.
2057 ------------------------------------------------------------------------------
2059 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
2061 * Support for kernels:
2062 2.4.21-47.0.1.EL (RHEL 3)
2063 2.6.5-7.283 (SLES 9)
2064 2.6.9-42.0.10.EL (RHEL 4)
2065 2.6.12.6 vanilla (kernel.org)
2066 2.6.16.27-0.9 (SLES 10)
2067 * Recommended e2fsprogs version: 1.39.cfs5
2069 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2073 Frequency : occasional, depends on client load and configuration
2074 Bugzilla : 12181, 12203
2075 Description: data loss for recently-modified files
2077 Details : In some cases it is possible that recently written or created
2078 files may not be written to disk in a timely manner (this should
2079 normally be within 30s unless client IO load is very high).
2080 The problem appears as zero-length files or files that are a
2081 multiple of 1MB in size after a client crash or client eviction
2082 that are missing data at the end of the file.
2084 This problem is more likely to be hit on clients where files are
2085 repeatedly created and unlinked in the same directory, clients
2086 have a large amount of RAM, have many CPUs, the filesystem has
2087 many OSTs, the clients are rebooted frequently, and/or the files
2088 are not accessed by other nodes after being written.
2090 The presence of the problem can be detected by looking at
2091 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2092 smaller than the second (nr_unused) then dirty files will not
2093 be flushed automatically to disk. "sync; sleep 10" should be
2094 run several times on the node before unmounting it to update
2095 Lustre (this is also safe to run on nodes without this problem).
2097 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2098 that can cause this same problem, so customers using that kernel
2099 also need to update the kernel in addition to Lustre. In order
2100 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2102 It is normal that files written just before a client crash (less
2103 than 30s) may not yet have been flushed to disk, even for local
2107 Frequency : frequent on thin XT3 nodes
2109 Description: UUID collision on thin XT3 Linux nodes
2110 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2111 randomly, since we relied on an insufficiently-seeded PRNG.
2116 Description: OSS hangs after "All ost request buffers busy"
2117 Details : A deadlock between quota and journal operations caused OSS
2118 hangs after printing "All ost request buffers busy."
2121 Frequency : always on liblustre builds
2123 Description: Cleanup compiler warnings on liblustre
2126 Frequency : always on liblustre builds on XT3
2128 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2132 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2133 > 32000 subdirectories
2134 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2135 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2136 subdirectory count crosses 32000. This will aid e2fsck to
2137 correctly handle more than 32000 subdirectories.
2140 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2142 Description: System hang on startup
2143 Details : This bug allowed the liblustre (e.g. catamount) client to
2144 return to the app before handling all startup RPCs. This
2145 could leave the node unresponsive to lustre network traffic
2146 and manifested as a server ptllnd timeout.
2148 Severity : enhancement
2150 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2151 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2152 boolean to control whether to print peer debug info when a
2153 client's RPC times out.
2158 Description: make O_SYNC working on 2.6 kernels
2159 Details : 2.6 kernels use different method for mark pages for write,
2160 so need add a code to lustre for O_SYNC work.
2165 Description: Failure to close file and release space on NFS
2166 Details : Put inode details into lock acquired in ll_intent_file_open.
2167 Use mdc_intent_lock in ll_intent_open to properly
2168 detect all kind of errors unhandled by mdc_enqueue.
2173 Description: proc file read during shutdown sometimes raced obd removal,
2175 Details : Add lock to prevent obd access after proc file removal.
2178 Frequency : Only for files larger than 4GB on 32-bit clients.
2180 Description: improperly doing page alignment of locks
2181 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2182 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2187 Description: RPCs being resent when they shouldn't be
2188 Details : Some RPCs that should not be resent are being resent. This
2189 can cause inconsistencies in the RPC state machine. Do not
2190 resend such requests.
2193 Frequency : rare, only with NFS export
2195 Description: Crash on NFS re-export node
2196 Details : under very unusual load conditions an assertion is hit in
2197 ll_intent_file_open()
2200 Frequency : only if OST filesystem is corrupted
2202 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2203 Details : for a short time RPCs with bulk IO are in the replay list,
2204 but replay of bulk IOs is unimplemented. If the OST filesystem
2205 is corrupted due to disk cache incoherency and then replay is
2206 started it is possible to trip an assertion. Avoid putting
2207 committed RPCs into the replay list at all to avoid this issue.
2212 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2213 Details : Large single O_DIRECT read and write calls can fail to allocate
2214 a sufficiently large buffer to process the request. In case of
2215 allocation failure the allocation is retried with a smaller
2216 buffer and broken into smaller requests.
2218 Severity : enhancement
2220 Description: Add -o localflock option to simulate old noflock behaviour.
2221 Details : This will achieve local-only flock/fcntl locks coherentness.
2226 Description: versioning check is incomplete
2227 Details : Checking the version difference of client vs. server, report
2228 error if the gap is too big.
2233 Description: add support PG_writeback bit
2234 Details : add support for PG_writeback bit for Lustre, for more carefull
2235 work with page cache in 2.6 kernel. This also fix some deadlocks
2236 and remove hack for work O_SYNC with 2.6 kernel.
2238 Severity : enhancement
2240 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2241 Details : The uninit_groups feature works in conjunction with the kernel
2242 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2243 up the pass1 processing of e2fsck. This is a read-only feature
2244 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2245 mount filesystems that have had this feature enabled.
2247 Severity : enhancement
2249 Description: Improve multi-block allocation algorithm to avoid fragmentation
2250 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2251 allocation locality and avoid filesystem fragmentation.
2253 ------------------------------------------------------------------------------
2255 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2257 * Support for kernels:
2258 2.6.9-42.0.3.EL (RHEL 4)
2259 2.6.5-7.276 (SLES 9)
2260 2.4.21-47.0.1.EL (RHEL 3)
2261 2.6.12.6 vanilla (kernel.org)
2262 2.6.16.21-0.8 (SLES10)
2263 * Recommended e2fsprogs version: 1.39.cfs2-0
2265 * The backwards-compatible /proc/sys/portals symlink has been removed
2266 in this release. Before upgrading, please ensure that you change
2267 any configuration scripts or /etc/sysctl.conf files that access
2268 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2269 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2270 in advance of the upgrade on any system running Lustre 1.4.6 or
2271 newer, since /proc/sys/lnet was added in that version.
2272 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2276 Frequency : only when quota is used
2278 Description: avoid scanning export list for quota master
2279 Details : Change the algorithms to avoid scanning export list in order
2280 to improve the efficiency.
2283 Frequency : MDS failover only, very rarely
2285 Description: "went back in time" messages on mds failover
2286 Details : The greatest transno may be lost when the current operation
2287 finishes with an error (transno==0) and the client's last_rcvd
2288 record is over-written. Save the greatest transno in the
2289 mds_last_transno for this case.
2292 Frequency : always for specific kernels and striping counts
2294 Description: client may get "Matching packet too big" without ACL support
2295 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2296 when trying to access files in certain configurations. The
2297 clients should in fact be denied when mounting because they do
2298 not understand ACLs.
2301 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2303 Description: many clients connecting with IO in progress causes connect timeouts
2304 Details : Avoid synchronous journal commits to avoid delays caused by many
2305 clients connecting/disconnecting when bulk IO is in progress.
2306 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2307 OST_IO_PORTAL to avoid delays behind potentially many pending
2311 Frequency : occasionally with multiple writers to a single file
2313 Description: shared writes to file may result in wrong size reported by stat()
2314 Details : Allow growing of kms when extent lock is cancelled
2317 Frequency : always with random mmap IO to multi-striped file
2319 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2320 Details : Only if the hole is at the end of OST object so that kms is too
2321 small. Fix is to increase kms accordingly in ll_nopage.
2324 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2326 Description: writes to a missing object would leak memory on the OST
2327 Details : If there is an inconsistency between the MDS and OST filesystems,
2328 such that the MDS references an object that doesn't exist, writes
2329 to that object will leak memory due to incorrect cleanup in the
2330 error handling path, eventually running out of memory on the OST.
2335 Description: Creating too long symlink causes lustre errors
2336 Details : Check symlink and name lengths before sending requests to MDS.
2339 Frequency : only if flock is enabled (not on by default)
2341 Description: posix locks not released on fd closure on 2.6.9+
2342 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2343 kernels, this caused such locks not to be released on fd close and
2344 then assertions on fs unmount about still used locks.
2347 Frequency : MDS failover only, very rarely
2349 Description: clients may get ASSERTION(granted_lock != NULL)
2350 Details : When request was taking a long time, and a client was resending
2351 a getattr by name lock request. The were multiple lock requests
2352 with the same client lock handle and
2353 mds_getattr_name->fixup_handle_for_resent_request found one of the
2354 lock handles but later failed with ASSERTION(granted_lock != NULL).
2359 Description: handle->h_buffer_credits > 0, assertion failure
2360 Details : h_buffer_credits is zero after truncate, causing assertion
2361 failure. This patch extends the transaction or creates a new
2365 Frequency : NFS re-export or patchless client
2366 Bugzilla : 11179, 10796
2367 Description: Crash on NFS re-export node (__d_move)
2368 Details : We do not want to hash the dentry if we don't have a lock.
2369 But if this dentry is later used in d_move, we'd hit uninitialised
2370 list head d_hash, so we just do this to init d_hash field but
2371 leave dentry unhashed.
2374 Frequency : NFS re-export or patchless client
2376 Description: NFS exports has problem with symbolic link
2377 Details : lustre client didn't properly install dentry when re-exported
2378 to NFS or running patchless client.
2381 Frequency : NFS re-export or patchless client
2383 Description: Various nfs/patchless fixes.
2384 Details : fixes reuse disconected alias for lookup process - this fixes
2385 warning "find_exported_dentry: npd != pd",
2386 fix permission error with open files at nfs.
2387 fix apply umask when do revalidate.
2390 Frequency : occasional
2392 Description: Crash on NFS re-export node
2393 Details : calling clear_page() on the wrong pointer triggered oops in
2394 generic_mapping_read().
2397 Frequency : rarely, using O_DIRECT IO
2399 Description: unaligned directio crashes client with LASSERT
2400 Details : check for unaligned buffers before trying any requests.
2403 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2405 Description: stale data returned from RAID cache
2406 Details : If only a small amount of IO is done to the RAID device before
2407 reading it again it is possible to get stale data from the RAID
2408 cache instead of reading it from disk.
2411 Frequency : always for sles10 kernel
2413 Description: sles10 support
2414 Details : ll_follow_link: compile fixes and using of nd_set_link
2415 under newer kernels.
2418 Frequency : depends on arch, kernel and compiler version, always on sles10
2421 Description: recursive or deep enough symlinks cause stack overflow
2422 Details : getting rid of large stack-allocated variable in
2426 Frequency : depends on hardware
2428 Description: lustre write performance loss in the SLES10 kernel
2429 Details : the performance loss is caused by using of write barriers in the
2430 ext3 code. The SLES10 kernel turns barrier support on by
2431 default. The fix is to undo that change for ldiskfs.
2433 ------------------------------------------------------------------------------
2435 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2437 * Support for kernels:
2438 2.6.9-42.0.3EL (RHEL 4)
2439 2.6.5-7.276 (SLES 9)
2440 2.4.21-47.0.1.EL (RHEL 3)
2441 2.6.12.6 vanilla (kernel.org)
2445 Frequency : quota enabled and large files being deleted
2447 Description: releasing more than 4GB of quota at once hangs OST
2448 Details : If a user deletes more than 4GB of files on a single OST it
2449 will cause the OST to spin in an infinite loop. Release
2450 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2455 Description: statfs data retrieved from /proc may be stale or zero
2456 Details : When reading per-device statfs data from /proc, in the
2457 {kbytes,files}_{total,free,avail} files, it may appear
2458 as zero or be out of date.
2461 Frequency : systems with MD RAID1 external journal devices
2463 Description: lconf's call to blkid is confused by RAID1 journal devices
2464 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2465 of returning all block devices that match the journal UUID.
2468 Frequency : always, for aggregate stripe size over 4GB
2470 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2471 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2472 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2475 Frequency : always if "lfs find" used on a local file/directory
2477 Description: "lfs find" segfaults if used on a local file/directory
2478 Details : The case where a directory component was not specified wasn't
2479 handled correctly. Handle this properly.
2482 Frequency : always on ppc64
2484 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2485 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2486 perform endianness conversion.
2489 Frequency : rarely (truncate to non-zero file size after write under load)
2490 Bugzilla : 10730, 10687
2491 Description: Files padded with zeros to next 4K multiple
2492 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2493 it is possible that files that are truncated to a non-zero size
2494 immediately after being written are filled with zero bytes beyond
2495 the truncated size. No file data is lost.
2497 Severity : enhancement
2499 Description: Allow recovery/failover for liblustre clients.
2500 Details : liblustre clients were unaware of failover configurations until
2503 Severity : enhancement
2505 Description: user file locks should fail when not mounting with flock option
2506 Details : Set up an error-returning stub in ll_file_operations.lock field
2507 to prevent incorrect behaviour when client is mounted without
2508 flock option. Also, set up properly f_op->flock field for
2512 Frequency : always on ia64
2514 Description: "lfs df" loops on printing out MDS statfs information
2515 Details : The obd_ioctl_data was not initialized and in some systems
2516 this caused a failure during the ioctl that did not return
2517 an error. Initialize the struct and return an error on failure.
2520 Frequency : SLES 9 only
2522 Description: Error of copying files with lustre special EAs as root
2523 Details : Client side always return success for setxattr call for lustre
2524 special xattr (currently only "trusted.lov").
2527 Frequency : rarely on clusters with both ia64+i386 clients
2529 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2530 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2531 clients are doing concurrent writes to the same file it is
2532 possible that the ia64 clients may LASSERT because the OST
2533 extent locks are not PAGE_SIZE aligned. Ensure that grown
2534 locks are always aligned on the request boundary.
2537 Frequency : specific use, occasional
2539 Description: Overwriting in use executable truncates on-disk binary image
2540 Details : If one node attempts to overwrite an executable in use by
2541 another node, we now correctly return ETXTBSY instead of
2542 truncating the file.
2544 Severity : enhancement
2546 Description: Async OSC create to avoid the blocking unnecessarily.
2547 Details : If a OST has no remain object, system will block on the creating
2548 when need to create a new object on this OST. Now, ways use
2549 pre-created objects when available, instead of blocking on an
2550 empty osc while others are not empty. If we must block, we block
2551 for the shortest possible period of time.
2556 Description: chmod on Lustre root is propagated to other clients
2557 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2558 invalid by the follow_mount time.
2561 Frequency : liblustre clients only
2563 Description: Race in 'instant cancel' lock handling could lead to such locks
2564 never to be granted in case of SMP MDS
2565 Details : Do not destroy not yet granted but cbpending locks in
2569 Frequency : replay/resend of open
2571 Description: non null lock assetion failure in mds_intent_policy
2572 Details : Trying to replay/resend lockless open requests resulted in
2573 mds_open() returning 0 with no lock. Now it sets a flag if
2574 a lock is going to be returned.
2576 Severity : enhancement
2578 Description: Checksum enhancements
2579 Details : New checksum enhancements allow for resending RPCs that failed
2582 Severity : enhancement
2584 Description: Tunables on number of dirty pages in cacche
2585 Details : Allow to set limit on number of dirty pages cached.
2590 Description: client crash on unmount - lock still has references
2591 Details : In some error handling cases it was possible to leak a lock
2592 reference on a client while accessing a file. This was not
2593 harmful to the client during operation, but would cause the
2594 client to crash when the filesystem is unmounted.
2597 Frequency : specific case, rare
2599 Description: ETXTBSY on mds though file not in use by client
2600 Details : ETXTBSY is no longer incorrectly returned when attempting to
2601 chmod or chown a directory that the user previously tried to
2602 execute or a currently-executing binary.
2605 Frequency : extremely rare except on liblustre-based clients
2607 Description: Lustre space not freed when files are deleted
2608 Details : Clean up open-unlinked files after client eviction. Previously
2609 the unlink was skipped and the files remained as orphans.
2614 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2615 Details : In some cases it was possible to send a blocking callback to
2616 a client doing a glimpse, even though that client didn't get
2617 a lock granted. When the glimpse lock is cancelled on the OST
2618 the freed lock is left on the waiting list and corrupted the list.
2621 Frequency : all core dumps
2623 Description: Broke core dumps to lustre
2624 Details : Negative dentry may be unhashed if parent does not have UPDATE
2625 lock, but some callers, e.g. do_coredump, expect dentry to be
2626 hashed after successful create, hash it in ll_create_it.
2628 ------------------------------------------------------------------------------
2630 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2632 * Support for kernels:
2633 2.6.9-42.0.2.EL (RHEL 4)
2634 2.6.5-7.276 (SLES 9)
2635 2.4.21-40.EL (RHEL 3)
2636 2.6.12.6 vanilla (kernel.org)
2640 Frequency : always on RHEL 3
2642 Description: Number of open files grows over time
2643 Details : The number of open files grows over time, whether or not
2644 Lustre is started. This was due to a filp leak introduced
2645 by one of our kernel patches.
2647 ------------------------------------------------------------------------------
2649 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2651 * Support for kernels:
2652 2.6.9-42.EL (RHEL 4)
2653 2.6.5-7.276 (SLES 9)
2654 2.4.21-40.EL (RHEL 3)
2655 2.6.12.6 vanilla (kernel.org)
2660 Bugzilla : 5719, 9635, 9792, 9684
2661 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2662 Details : If a server is under heavy load and cannot reply to new
2663 connection requests before the client resends the (re)connect,
2664 the connection handling code can behave badly if two service
2665 threads are concurrently handing separate (re)connections from
2666 the same client. Add better locking to the connection handling
2667 code, and ensure that only a single connection will be processed
2668 for a given client UUID, even if the lock is dropped.
2670 Severity : enhancement
2672 Description: add TCP zero-copy support to kernel
2673 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2674 sends if the hardware supports scatter-gather and checksumming.
2675 This allows socklnd to do client-write and server-read more
2676 efficiently and reduce CPU utilization from skbuf copying.
2679 Frequency : only if NFS exporting from client
2681 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2682 Details : If setting attributes on a file created under NFS that had
2683 never been opened it would be possible to oops the client
2684 if the file had no objects.
2687 Frequency : always for liblustre
2689 Description: liblustre client does MDS+OSTs setattr RPC for each write
2690 Details : When doing a write from a liblustre client, the client
2691 incorrectly issued an RPC to the MDS and each OST the file was
2692 striped over in order to update the timestamps. When writing
2693 with small chunks and many clients this could overwhelm the MDS
2694 with RPCs. In all cases it would slow down the write because
2695 these RPCs are unnecessary.
2697 Severity : enhancement
2699 Description: allow number of MDS service threads to be changed at module load
2700 Details : It is now possible to change the number of MDS service threads
2701 running. Adding "options mds mds_num_threads={N}" to the MDS's
2702 /etc/modprobe.conf will set the number of threads for the next
2703 time Lustre is restarted (assuming the "mds" module is also
2704 reloaded at that time). The default number of threads will
2705 stay the same, 32 for most systems.
2710 Description: OST crash if filesystem is unformatted or corrupt
2711 Details : If an OST is started on a device that has never been formatted
2712 or if the filesystem is corrupt and cannot even mount then the
2713 error handling cleanup routines would dereference a NULL pointer.
2718 Description: NULL pointer deref in llap_from_page.
2719 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2720 incorrect) mapping if the page was truncated/reclaimed while it was
2721 searched for. Check for this condition and skip such pages when
2722 doing readahead. Introduce extra check to llap_from_page() to
2723 verify page->mapping->host is non-NULL (so page is not anonymous).
2726 Frequency : Sometimes when using sys_sendfile
2728 Description: "page not covered by a lock" warnings from ll_readpage
2729 Details : sendfile called ll_readpage without right page locks present.
2730 Now we introduced ll_file_sendfile that does necessary locking
2731 around call to generic_file_sendfile() much like we do in
2735 Frequency : with certain MDS communication failures at client mount time
2737 Description: NULL pointer deref after failed client mount
2738 Details : a client connection request may delayed by the network layer
2739 and not be sent until after the PTLRPC layer has timed out the
2740 request. If the client fails the mount immediately it will try
2741 to clean up before the network times out the request. Add a
2742 reference from the request import to the obd device and delay
2743 the cleanup until the network drops the request.
2746 Frequency : occasionally during client (re)connect
2748 Description: assertion failure during client (re)connect
2749 Details : processing a client connection request may be delayed by the
2750 client or server longer than the client connect timeout. This
2751 causes the client to resend the connection request. If the
2752 original connection request is replied in this interval, the
2753 client may trip an assertion failure in ptlrpc_connect_interpret()
2754 which thought it would be the only running connect process.
2757 Frequency : only with obd_echo servers and clients that are rebooted
2759 Description: kernel BUG accessing uninitialized data structure
2760 Details : When running an obd_echo server it did not start the ping_evictor
2761 thread, and when a client was evicted an uninitialized data
2762 structure was accessed. Start the ping_evictor in the RPC
2763 service startup instead of the OBD startup.
2765 Severity : enhancement
2766 Bugzilla : 10193 (patchless)
2767 Description: Remove dependency on various unexported kernel interfaces.
2768 Details : No longer need reparent_to_init, exit_mm, exit_files,
2769 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2772 Frequency : rare (only users of deprecated and unsupported LDAP config)
2774 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2775 Details : LDAP apparently contains 'lustreName' attributes instead of
2776 'name'. A simple remapping of the name is sufficient.
2779 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2781 Description: waiting_locks_callback trips kernel BUG if client is evicted
2782 Details : Running with the dump_on_timeout debug flag turned on makes
2783 it possible that the waiting_locks_callback() can try to dump
2784 the Lustre kernel debug logs from an interrupt handler. Defer
2785 this log dumping to the expired_lock_main() thread.
2787 Severity : enhancement
2789 Description: Support NFS exporting on 2.6 kernels.
2790 Details : Implement non-rawops metadata methods for NFS server to use without
2791 changing NFS server code.
2794 Frequency : very rare (synthetic metadata workload only)
2796 Description: two racing renames might cause an MDS thread to deadlock
2797 Details : Running the "racer" program may cause one MDS thread to rename
2798 a file from being the source of a rename to being the target of
2799 a rename at exactly the same time that another thread is doing
2800 so, and the second thread has already enqueued these locks after
2801 doing a lookup of the target and is trying to relock them in
2802 order. Ensure that we don't try to re-lock the same resource.
2805 Frequency : only very large systems with liblustre clients
2807 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2808 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2809 all OSTs in parallel. This allows the eviction of stale liblustre
2810 clients to proceed much faster than if they were done in series,
2811 and also offers similar improvements for other set_info RPCs.
2816 Description: excessive CPU usage during initial read phase on client
2817 Details : During the initial read phase on a client, it would agressively
2818 retry readahead on the file, consuming too much CPU and impacting
2819 performance (since 1.4.5.8). Improve the readahead algorithm
2820 to avoid this, and also improve some other common cases (read
2821 of small files in particular, where "small" is files smaller than
2822 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2827 Description: MDS crash when receiving packet with unknown intent.
2828 Details : Do not LBUG in unknown intent case, just return -EFAULT
2830 Severity : enhancement
2831 Bugzilla : 9293, 9385
2832 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2833 Details : Do not serialize getattr (non-intent version) and statfs.
2836 Frequency : occasional, when OST network is overloaded/intermittent
2838 Description: client evicted by OST after bulk IO timeout
2839 Details : If a client sends a bulk IO request (read or write) the OST
2840 may evict the client if it is unresposive to its data GET/PUT
2841 request. This is incorrect if the network is overloaded (takes
2842 too long to transfer the RPC data) or dropped the OST GET/PUT
2843 request. There is no need to evict the client at all, since
2844 the pinger and/or lock callbacks will handle this, and the
2845 client can restart the bulk request.
2848 Frequency : Always when mmapping file with no objects
2850 Description: client crashes when mmapping file with no objects
2851 Details : Check that we actually have objects in a file before doing any
2852 operations on objects in ll_vm_open, ll_vm_close and
2858 Description: Request leak when working with deleted CWD
2859 Details : Introduce advanced request refcount tracking for requests
2860 referenced from lustre intent.
2862 Severity : Enhancement
2864 Description: Cache open file handles on client.
2865 Details : MDS now will return special lock along with openhandle, if
2866 requested and client is allowed to hold openhandle, even if unused,
2867 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2868 closing files for every read/write openration.
2870 Severity : Enhancement
2872 Description: Cache open negative dentries on client when possible.
2873 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2874 negative dentries on lock revocation.
2879 Description: Remounting a client read-only wasn't possible with a zconf mount
2880 Details : It wasn't possible to remount a client read-only with llmount.
2882 Severity : enhancement
2883 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2884 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2885 MPICH is included to add Lustre-specific ADIO interfaces.
2886 This is based closely on the UFS ADIO layer and only differs
2887 in file creation, in order to allow the OST striping to be set.
2888 This is user-contributed code and not supported by CFS.
2893 Description: extended inode attributes (immutable, append-only) work improperly
2894 when 2.4 and 2.6 kernels are used on client/server or vice versa
2895 Details : Introduce kernel-independent values for these flags.
2897 Severity : enhancement
2900 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2901 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2902 granularity. Now, Fractional MB granularity is be supported,
2903 this is very useful for low memory system.
2905 Severity : enhancement
2907 Description: Getattr by fid
2908 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2909 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2910 have locks and avoids getattr rpc requests for it on every stat.
2913 Frequency : Always, for filesystems larger than 2TB
2915 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2916 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2917 filesystems larger than 2TB to be mounted with mballoc enabled.
2920 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2922 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2923 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2924 filesystems larger than 2TB is mounted then the kernel will
2925 silently corrupt the start of the filesystem. CONFIG_LBD is
2926 enabled for all CFS-supported kernels, but the possibility of
2927 this happening with a modified kernel config exists.
2929 Severity : enhancement
2931 Description: add client O_DIRECT support for 2.6 kernels
2932 Details : It is now possible to do O_DIRECT reads and writes to files
2933 in the Lustre client mountpoint on 2.6 kernel clients.
2935 Severity : enhancement
2937 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2938 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2939 parallel, not waiting for response from every OST before sending
2940 a rpc to the next OST.
2945 Description: setattr vs write race when updating file timestamps
2946 Details : Client processes that update a file timestamp into the past
2947 right after writing to the file (e.g. tar) it is possible that
2948 the updated file modification time can be reset to the current
2949 time due to a race between processing the setattr and write RPC.
2951 Severity : enhancement
2953 Description: Bring 'lfs find' closer in line with regular Linux find.
2954 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2955 -print0 options and obtains all the needed info through the lustre
2958 Severity : enhancement
2960 Description: support up to 1024 configured devices on one node
2961 Details : change obd_dev array from statically allocated to dynamically
2962 allocated structs as they are first used to reduce memory usage
2967 Description: Flush dirty partially truncated pages during truncate
2968 Details : Immediatelly flush partially truncated pages in filter_setattr,
2969 this way we completely avoid having any pages in page cache on OST
2970 and can retire ugly workarounds during writes to flush such pages.
2975 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2976 Details : protect inode from truncation within vfs_unlink() context
2977 just take a reference before calling vfs_unlink() and release it
2978 when parent's i_sem is free.
2983 Description: last_id value checked outside lock on OST caused LASSERT failure
2984 Details : If there were multiple MDS->OST object precreate requests in
2985 flight, it was possible that the OST's last object id was checked
2986 outside a lock and incorrectly tripped an assertion. Move checks
2987 inside locks, and discard old precreate requests.
2990 Frequency : always, if extents are used on OSTs
2992 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2993 Details : OSTs using the extents format would not zero the high 16 bits of
2994 the index physical block number. This is not a problem for any
2995 OST filesystems smaller than 16TB, and no kernels support ext3
2996 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2997 new/modified files) and can be fixed for existing filesystems
2998 with e2fsprogs-1.39-cfs1.
3003 Description: import connection selection may be incorrect if timer wraps
3004 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
3005 import connections to be ignored if the 32-bit jiffies counter
3006 wraps. Use a 64-bit jiffies counter.
3009 Frequency : very large clusters immediately after boot
3011 Description: LNET request buffers exhausted under heavy short-term load
3012 Details : If a large number of client requests are generated on a service
3013 that has previously never seen so many requests it is possible
3014 that the request buffer growth cannot keep up with the spike in
3015 demand. Instead of dropping incoming requests, they are held in
3016 the LND until the RPC service can accept more requests.
3019 Frequency : Sometimes during replay
3021 Description: Assertion failure in ll_local_open after replay.
3022 Details : If replay happened on an open request reply before we were able
3023 to set replay handler, reply will become not swabbed tripping the
3024 assertion in ll_local_open. Now we set the handler right after
3025 recognising of open request
3028 Frequency : very rare
3030 Description: kernel reports "badness in vsnprintf"
3031 Details : Reading from the "recovery_status" /proc file in small chunks
3032 may cause a negative length in lprocfs_obd_rd_recovery_status()
3033 call to vsnprintf() (which is otherwise harmless). Exit early
3034 if there is no more space in the output buffer.
3036 Severity : enhancement
3038 Description: clear OBD RPC statistics by writing to them
3039 Details : It is now possible to clear the OBD RPC statistics by writing
3040 to the "stats" file.
3045 Description: Client mtime is not the same on different clients after utimes
3046 Details : In some cases, the client was using the utimes() syscall on
3047 a file cached on another node. The clients now validate the
3048 ctime from the MDS + OSTs to determine which one is right.
3053 Description: Inability to activate failout mode
3054 Details : lconf script incorrectly assumed that in pythong string's numeric
3055 value is used in comparisons.
3058 Frequency : always with multiple stripes per file
3060 Description: Inefficient object allocation for mutli-stripe files
3061 Details : When selecting which OSTs to stripe files over, for files with
3062 a stripe count that divides evenly into the number of OSTs,
3063 the MDS is always picking the same starting OST for each file.
3064 Return the OST selection heuristic to the original design.
3069 Description: mount failures may take full timeout to return an error
3070 Details : Under some heavy load conditions it is possible that a
3071 failed mount can wait for the full obd_timeout interval,
3072 possibly several minutes, before reporting an error.
3073 Instead return an error as soon as the status is known.
3075 Frequency : quota enabled and large files being deleted
3077 Description: releasing more than 4GB of quota at once hangs OST
3078 Details : If a user deletes more than 4GB of files on a single OST it
3079 will cause the OST to spin in an infinite loop. Release
3080 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3085 Description: statfs data retrieved from /proc may be stale or zero
3086 Details : When reading per-device statfs data from /proc, in the
3087 {kbytes,files}_{total,free,avail} files, it may appear
3088 as zero or be out of date.
3091 Frequency : systems with MD RAID1 external journal devices
3093 Description: lconf's call to blkid is confused by RAID1 journal devices
3094 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3095 of returning all block devices that match the journal UUID.
3098 Frequency : always, for aggregate stripe size over 4GB
3100 Description: assertion fails when trying to use 4GB stripe size
3101 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
3102 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
3105 Frequency : always on ppc64
3107 Description: the first write on an ext3 filesystem with mballoc got stuck
3108 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3109 perform endianness conversion.
3111 ------------------------------------------------------------------------------
3113 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3115 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3116 INTEROPERATE with older versions automatically. Please read the
3117 user documentation before upgrading any part of a live system.
3118 * WARNING: Lustre networking configuration changes are required with
3119 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3122 * Support for kernels:
3123 2.6.9-22.0.2.EL (RHEL 4)
3124 2.6.5-7.244 (SLES 9)
3125 2.6.12.6 vanilla (kernel.org)
3128 Severity : enhancement
3129 Bugzilla : 7981/8208
3130 Description: Introduced Lustre Networking (LNET)
3131 Details : LNET is new networking infrastructure for Lustre, it includes
3132 a reorganized network configuration mode (see the user
3133 documentation for full details) as well as support for routing
3134 between different network fabrics. Lustre Networking Devices
3135 (LNDs) for the supported network fabrics have also been
3136 created for this new infrastructure.
3138 Severity : enhancement
3139 Description: Introduced Access control lists
3140 Details : clients can set ACLs on files and directories in order to have
3141 more fine-grained permissions than the standard Unix UGO+RWX.
3142 The MDS must be started with the "-o acl" mount option.
3144 Severity : enhancement
3145 Description: Introduced filesystem quotas
3146 Details : Administrators may now establish per-user quotas on the
3149 Severity : enhancement
3151 Description: Configuration change for the XT3
3152 The PTLLND is now used to run Lustre over Portals on the XT3
3153 The configure option(s) --with-cray-portals are no longer used.
3154 Rather --with-portals=<path-to-portals-includes> is used to
3155 enable building on the XT3. In addition to enable XT3 specific
3156 features the option --enable-cray-xt3 must be used.
3161 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3162 Details : A race between allocating a new llcd and re-getting the llcd_lock
3163 allowed another thread to grab newly-allocated llcd.
3165 Severity : enhancement
3167 Description: 2.6 OST async journal commit and locking fix to improve performance
3168 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3169 2.6 kernels are now async as they already were in 2.4 kernels so
3170 that they can commit concurrently with the network bulk transfer.
3171 For block-allocated files the filter allocation semaphore is held
3172 to avoid filesystem fragmentation during allocation. BKL lock
3173 removed for 2.6 xattr operations where it is no longer needed.
3178 Description: lconf incorrectly determined whether two IP networks could talk
3179 Details : In some more complicated routing and multiple-network
3180 configurations, lconf will avoid trying to make a network
3181 connection to a disjoint part of the IP space. It was doing the
3182 math incorrectly for one set of cases.
3187 Description: Fix for potential infinite loop processing records in an llog.
3188 Details : If an llog record is corrupted/zeroed, it is possible to loop
3189 forever in llog_process(). Validate the llog record length
3190 and skip the remainder of the block on error.
3193 Frequency : occasional (liblustre only)
3195 Description: liblustre could not open files whose last component is a symlink
3196 Details : sysio_path_walk() would incorrectly pass the open intent to
3197 intermediate path components.
3200 Frequency : rare (liblustre only with non-standard tuning)
3201 Bugzilla : 7201 (7350)
3202 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3203 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3204 after it has been released, only do this for referenced locks
3206 Severity : enhancement
3208 Description: specifying an (invalid) directory default stripe_size of -1
3209 would reset the directory default striping
3210 Details : stripe_size -1 was used internally to signal directory stripe
3211 removal, now use "all default" to signal dir stripe removal
3212 as a directory striping of "all default" is not useful
3215 Frequency : common for large clusters running liblustre clients
3217 Description: doing an ls when liblustre clients are running is slow
3218 Details : sending a glimpse AST to a liblustre client waits for every AST
3219 to time out, as liblustre clients will not respond. Since they
3220 cannot cache data we refresh the OST lock LVB from disk instead.
3222 Severity : enhancement
3224 Description: doing an ls at the same time as file IO can be slow
3225 Details : enqueue and other "small" requests can be blocked behind many
3226 large IO requests. Create a new OST IO portal for non-IO
3227 requests so they can be processed faster.
3230 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3232 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3233 created file with mode 000
3234 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3235 allow an ftruncate() as a non-root user to a file with mode 000.
3236 Lustre now respects this flag to disable mode checks when
3237 truncating a file owned by the user
3240 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3243 Description: Revoking locks from clients that went dead or catatonic might take
3245 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3246 cancellation of such locks instant on servers without waiting for
3247 any reply from clients. Clients drops these locks when cancel
3248 notification from server is received without replying.
3251 Frequency : liblustre-only, when liblustre client dies or becomes busy
3253 Description: Doing ls on Linux clients can take a long time with active
3255 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3256 granting such locks to it in the first place if possible. Locks
3257 are taken by proxy on the OST during the read or write and
3258 dropped immediately afterward. Add connect flags handling, do
3259 not grant locks to liblustre clients for glimpse ASTs.
3261 Severity : enhancement
3263 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3264 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3265 sequential reads and behaves badly with applications doing random
3266 reads. Improve it by reading ahead at least read region, and
3267 avoiding excessive large RPC for small reads.
3269 Severity : enhancement
3271 Description: Creating more than 1000 files for a single job may cause a load
3272 imbalance on the OSTs if there are also a large number of OSTs.
3273 Details : qos_prep_create() uses an OST index reseed value that is an
3274 even multiple of the number of available OSTs so that if the
3275 reseed happens in the middle of the object allocation it will
3276 still utilize the OSTs as uniformly as possible.
3281 Description: OST or MDS may oops in ping_evictor_main()
3282 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3283 but doesn't restart at beginning of obd_exports_timed list
3286 Severity : enhancement
3288 Description: improve by-nid export eviction on the MDS and OST
3289 Details : allow multiple exports with the same NID to be evicted at one
3290 time without re-searching the exports list.
3293 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3295 Description: MDS may oops in groups_free()
3296 Details : in rare race conditions a newly allocated group_info struct is
3297 freed again, and this can be NULL. The 2.4 compatibility code
3298 for groups_free() checked for a NULL pointer, but 2.6 did not.
3301 Frequency : common for liblustre clients doing little filesystem IO
3302 Bugzilla : 9352, 7313
3303 Description: server may evict liblustre clients accessing contended locks
3304 Details : if a client is granted a lock or receives a completion AST
3305 with a blocking AST already set it would not reply to the AST
3306 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3309 Frequency : lfs setstripe, only systems with more than 160 OSTs
3311 Description: unable to set striping with a starting offset beyond OST 160
3312 Details : llapi_create_file() incorrectly limited the starting stripe
3313 index to the maximum single-file stripe count.
3316 Frequency : LDAP users only
3318 Description: lconf did not handle in-kernel recovery with LDAP properly
3319 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3321 Severity : enhancement
3323 Description: bind OST threads to NUMA nodes to improve performance
3324 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3325 node and do their allocations there to localize memory access
3327 Severity : enhancement
3329 Description: llmount can determine client NID directly from Myrinet (GM)
3330 Details : the client NID code from gmnalnid was moved directly into
3331 llmount, removing the need to use this or specifying the
3332 client NID explicitly when mounting GM clients with zeroconf
3335 Frequency : if client is started with down MDS
3337 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3338 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3339 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3340 ptlrpc_import_delayed_req() because we didn't check if the
3341 request was interrupted, and we also didn't break out of the
3342 event loop if there was no timeout
3347 Description: data loss during non-page-aligned writes to a single file from
3348 both multiple nodes and multiple threads on one node at same time
3349 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3350 inconsistency led to false short-reads, that were cached and later
3351 used by ->prepare_write() to fill in partially written page,
3352 leading to data loss.
3355 Frequency : always, if lconf --abort_recovery used
3357 Description: lconf --abort_recovery fails with 'Operation not supported'
3358 Details : lconf was attempting to abort recovery on the MDT device and not
3361 Severity : enhancement
3363 Description: remove cleanup logs
3364 Details : replace lconf-generated cleanup logs with lustre internal
3365 cleanup routines. Eliminates the need for client-cleanup and
3368 Severity : enhancement
3370 Description: add support for EAs (user and system) on lustre filesystems
3371 Details : it is now possible to store extended attributes in the Lustre
3372 client filesystem, and with the user_xattr mount option it
3373 is possible to allow users to store EAs on their files also
3375 Severity : enhancement
3377 Description: Add possibility (config option) to show minimal available OST free
3379 Details : When compiled with --enable-mindf configure option, statfs(2)
3380 (and so, df) will return least minimal free space available from
3381 all OSTs as amount of free space on FS, instead of summary of
3382 free spaces of all OSTs.
3384 Severity : enhancement
3386 Description: do not expand extent locks acquired on OST-side
3387 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3388 by server: they are not cached anyway.
3391 Frequency : when mmap is used/binaries executed from Lustre
3393 Description: Unmmap pages before throwing them away from read cache.
3394 Details : llap_shrink cache now attempts to unmap pages before discarding
3395 them (if unmapping failed - do not discard). SLES9 kernel has
3396 extra checks that trigger if this unmapping is not done first.
3401 Description: lconf didn't resolve symlinks before checking to see whether a
3402 given mountpoint was already in use
3405 Frequency : when migrating failover services
3406 Bugzilla : 6395, 9514
3407 Description: When migrating a subset of services from a node (e.g. failback
3408 from a failover service node) the remaining services would
3409 time out and evict clients.
3410 Details : lconf --force (implied by --failover) sets the global obd_timeout
3411 to 5 seconds in order to quickly disconnect, but this caused
3412 other RPCs to time out too quickly. Do not change the global
3413 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3415 Severity : enhancement
3416 Frequency : if MDS is started with down OST
3417 Bugzilla : 9439,5706
3418 Description: Allow startup/shutdown of an MDS without depending on the
3419 availability of the OSTs.
3420 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3421 Add appropriate locking and lov-osc refcounts for safe
3422 cleaning. Add osc abort_inflight calls in case the
3423 synchronize never started.
3426 Frequency : occasional (Cray XT3 only)
3428 Description: root not authorized to access files in CRAY_PORTALS environment
3429 Details : The client process capabilities were not honoured on the MDS in
3430 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3431 been accessed by an authorized user then root was able to access
3432 the file on the local client also. The root user capabilities
3433 are now allowed on the MDS, as this environment has secure UID.
3436 Frequency : occasional
3438 Description: ldiskfs "too long searching" message happens too often
3439 Details : A debugging message (otherwise harmless) prints too often on
3440 the OST console. This has been reduced to only happen when
3441 there are fragmentation problems on the filesystem.
3446 Description: Division by zero in statfs when all OSCs are inactive
3447 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3448 lov_statfs divides by value returned by lov_get_stripecnt().
3452 Bugzilla : 9489, 3273
3453 Description: First write from each client to each OST was only 4kB in size,
3454 to initialize client writeback cache, which caused sub-optimal
3455 RPCs and poor layout on disk for the first writen file.
3456 Details : Clients now request an initial cache grant at (re)connect time
3457 and so that they can start streaming writes to the cache right
3458 away and always do full-sized RPCs if there is enough data.
3459 If the OST is rebooted the client also re-establishes its grant
3460 so that client cached writes will be honoured under the grant.
3465 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3466 Details : Now I/O RPCs go to different portal number and (presumably) fast
3467 lock requests (and glimses) and other RPCs get their own service
3468 threads pool that should be able to service those RPCs
3471 Severity : enhancement
3473 Description: Ability to exchange lustre version between client and servers and
3474 issue warnings at client side if client is too old. Also for
3475 liblustre clients there is ability to refuse connection of too old
3477 Details : New 'version' field is added to connect data structure that is
3478 filled with version info. That info is later checked by server and
3482 Frequency : rare, liblustre only.
3483 Bugzilla : 9296, 9581
3484 Description: Two simultaneous writes from liblustre at offset within same page
3485 might proceed at the same time overwriting eachother with stale
3487 Details : I/O lock withing llu_file_prwv was released too early, before data
3488 actually was hitting the wire. Extended lock-holding time until
3489 server acknowledges receiving data.
3492 Frequency : extremely rare. Never observed in practice.
3494 Description: avoid generating lustre_handle cookie of 0.
3495 Details : class_handle_hash() generates handle cookies by incrementing
3496 global counter, and can hit 0 occasionaly (this is unlikely, but
3497 not impossible, because initial value of cookie counter is
3498 selected randonly). Value of 0 is used as a sentinel meaning
3499 "unassigned handle" --- avoid it. Also coalesce two critical
3500 sections in this function into one.
3502 Severity : enhancement
3504 Description: allow liblustre clients to delegate truncate locking to OST
3505 Details : To avoid overhead of locking, liblustre client instructs OST to
3506 take extent lock in ost_punch() on client's behalf. New connection
3507 flag is added to handle backward compatibility.
3509 Severity : enhancement
3510 Bugzilla : 4928, 7341, 9758
3511 Description: allow number of OST service threads to be specified
3512 Details : a module parameter allows the number of OST service threads
3513 to be specified via "options ost ost_num_threads={N}" in the
3514 OSS's /etc/modules.conf or /etc/modprobe.conf.
3518 Bugzilla : 6146, 9635, 9895
3519 Description: servers crash with bad pointer in target_handle_connect()
3520 Details : In rare cases when a client is reconnecting it was possible that
3521 the connection request was the last reference for that export.
3522 We would temporarily drop the export reference and get a new
3523 one, but this may have been the last reference and the export
3524 was just destroyed. Get new reference before dropping old one.
3526 Severity : enhancement
3527 Frequency : if client is started with failover MDS
3529 Description: Allow multiple MDS hostnames in the mount command
3530 Details : Try to read the configuration from all specified MDS
3531 hostnames during a client mount in case the "primary"
3534 Severity : enhancement
3536 Description: Stop sending data to evicted clients as soon as possible.
3537 Details : Check if the client we are about to send or are sending data to
3538 was evicted already. (Check is done every second of waiting,
3539 for which l_wait_event interface was extended to allow checking
3540 of exit condition at specified intervals).
3543 Frequency : rare, normally only when NFS exporting is done from client
3545 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3547 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3550 Frequency : always, if srand() is called before liblustre initialization
3552 Description: Liblustre uses system PRNG disturbing its usage by user application
3553 Details : Introduce internal to lustre fast and high-quality PRNG for
3554 lustre usage and make liblustre and some other places in generic
3555 lustre code to use it.
3557 Severity : enhancement
3558 Bugzilla : 9477, 9557, 9870
3559 Description: Verify that the MDS configuration logs are updated when xml is
3560 Details : Check if the .xml configuration logs are newer than the config
3561 logs stored on the MDS and report an error if this is the case.
3562 Request --write-conf, or allow starting with --old_conf.
3564 Severity : enhancement
3566 Description: Handle symlinks in the path when checking if Lustre is mounted.
3567 Details : Resolve intermediate symlinks when checking if a client has
3568 mounted a filesystem to avoid duplicate client mounts.
3573 Description: lconf can hit an error exception but still return success.
3574 Details : The lconf command catches the Command error exception at the top
3575 level script context and will exit with the associated exit
3576 status, but doesn't ensure that this exit status is non-zero.
3581 Description: failure of ptlrpc thread startup can cause oops
3582 Details : Starting a ptlrpc service thread can fail if there are a large
3583 number of threads or the server memory is very fragmented.
3584 Handle this without oopsing.
3587 Frequency : always, only if liblustre and non-default acceptor port was used
3589 Description: liblustre cannot connect to servers with non-default acceptor port
3590 Details : tcpnal_set_default_params() was not called and was therefore
3591 ignoring the environment varaible TCPNAL_PORT, as well as other
3592 TCPNAL_ environment variables
3597 Description: two objects could be created on the same OST for a single file
3598 Details : If an OST is down, in some cases it was possible to create two
3599 objects on a single OST for a single file. No problems other
3600 than potential performance impact and spurious error messages.
3604 Bugzilla : 5681, 9562
3605 Description: Client may oops in ll_unhash_aliases
3606 Details : Client dcache may become inconsistent in race condition.
3607 In some cases "getcwd" can fail if the current directory is
3613 Description: Inode refcounting problems in NFS export code
3614 Details : link_raw functions used to call d_instantiate without obtaining
3615 extra inode reference first.
3619 Bugzilla : 9942, 9903
3620 Description: Referencing freed requests leading to crash, memleaks with NFS.
3621 Details : We used to require that call to ll_revalidate_it was always
3622 followed by ll_lookup_it. Also with revalidate_special() it is
3623 possible to call ll_revalidate_it() twice for the same dentry
3624 even if first occurence returned success. This fix changes semantic
3625 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3626 reference on a request referred from the intent.
3627 ll_intent_release() then releases such a request.
3630 Frequency : rare, normally benchmark loads only
3632 Description: unlinked inodes were kept in memory on the client
3633 Details : If a client is repeatedly creating and unlinking files it
3634 can accumulate a lot of stale inodes in the inode slab cache.
3635 If there is no other client load running this can cause the
3636 client node to run out of memory. Instead flush old inodes
3637 from client cache that have the same inode number as a new inode.
3640 Frequency : SLES9 2.6.5 kernel and long filenames only
3641 Bugzilla : 9969, 10379
3642 Description: utime reports stale NFS file handle
3643 Details : SLES9 uses out-of-dentry names in some cases, which confused
3644 the lustre dentry revalidation. Change it to always use the
3648 Frequency : rare, unless heavy write-truncate concurrency is continuous
3649 Bugzilla : 4180, 6984, 7171, 9963, 9331
3650 Description: OST becomes very slow and/or deadlocked during object unlink
3651 Details : filter_destroy() was holding onto the parent directory lock
3652 while truncating+unlinking objects. For very large objects this
3653 may block other threads for a long time and slow overall OST
3654 responsiveness. It may also be possible to get a lock ordering
3655 deadlock in this case, or run out of journal credits because of
3656 the combined truncate+unlink. Solution is to do object truncate
3657 first in one transaction without parent lock, and then do the
3658 final unlink in a new transaction with the parent lock. This
3659 reduces the lock hold time dramatically.
3662 Frequency : rare, 2.4 kernels only
3664 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3665 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3666 before unmounting their filesystem in order to avoid deadlock.
3667 The kernel_locked() function in 2.4 kernels only checks whether
3668 the kernel lock is held, not whether it is this process that is
3669 holding it as 2.6 kernels do.
3674 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3675 Details : The client ptlrpc code may be trying to reconnect to a down
3676 server before a previous connection attempt has timed out.
3677 Increase the reconnect interval to be longer than the connection
3678 timeout interval to avoid sending duplicate connections to
3682 Frequency : echo_client brw_test command
3684 Description: fix echo_client to work with OST preallocated code
3685 Details : OST preallocation code (5137) didn't take echo_client IO path
3686 into account: echo_client calls filter methods outside of any
3687 OST thread and, hence, there is no per-thread preallocated
3688 pages and buffers to use. Solution: hijack pga pages for IO. As
3689 a byproduct, this avoids unnecessary data copying.
3693 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3694 Description: Client can oops in mdc_commit_close() after open replay
3695 Details : It was possible for the MDS to return an open request with no
3696 transaction number in mds_finish_transno() if the client was
3697 evicted, but without actually returning an error. Clients
3698 would later try to replay that open and may trip an assertion
3699 Simplify the client close codepath, and always return an error
3700 from the MDS in case the open is not successful.
3703 Frequency : rare, 2.6 OSTs only
3705 Description: OST may deadlock under high load on fragmented files
3706 Details : If there was a heavy load and highly-fragmented OST filesystems
3707 it was possible to have all the OST threads deadlock waiting on
3708 allocation of biovecs, because the biovecs were not released
3709 until the entire RPC IO was completed. Instead, release biovecs
3710 as soon as they are complete to ensure forward IO progress.
3712 Severity : enhancement
3714 Description: Support for specifying external journal device at mount
3715 Details : If an OST or MDS device is formatted with an external journal
3716 device, this device major/minor is stored in the ext3 superblock
3717 and may not be valid for failover. Allow detecting and
3718 specifying the external journal at mount time.
3723 Description: Mounting an MDS with pending unlinked files may cause oops
3724 Details : target_finish_recovery() calls mds_postrecov() which returned
3725 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3726 considers this an error and immediately begins cleaning up the
3727 lov, just after starting the mds_lov process
3729 Severity : enhancement
3731 Description: Implement 'lfs df' to report actual free space on per-OST basis
3732 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3733 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3734 usage of objects; '-h' to report in human readable format.
3736 ------------------------------------------------------------------------------
3738 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3745 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3746 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3747 at startup during orphan unlinking. Instead, skip buddy update
3748 before setup, as it will be regenerated after recovery is complete.
3753 Description: If an OST is inactive, its locks might reference stale inodes.
3754 Details : lov_change_cbdata() must iterate over all namespaces, even if
3755 they are inactive to clear inode references from the lock.
3757 Severity : enhancement
3758 Frequency : occasional, if non-standard max_dirty_mb used
3760 Description: Client will block write RPCs if not enough grant
3761 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3762 then the client will block writes while waiting for another RPC
3763 to complete instead of consuming its dirty limit. With change
3764 we get improved performance when max_dirty_mb is small.
3766 Severity : enhancement
3767 Bugzilla : 3389, 6253
3768 Description: Add support for supplementary groups on the MDS.
3769 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3770 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3771 lookups for user supplementary groups into a cache.
3775 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3776 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3777 for liblustre O_EXCL is enforced by MDS.
3781 Description: readdir never returns NULL in liblustre.
3782 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3783 dentry in struct dirent.
3787 Frequency : liblustre only, depends on application IO pattern
3788 Description: liblustre clients evicted if not contacting servers
3789 Details : Don't put liblustre clients into the ping_evictor list, so
3790 they will not be evicted by the pinger ever.
3792 Severity : enhancement
3794 Description: Add ability to evict clients by NID from MDS.
3795 Details : By echoing "nid:$NID" string into
3796 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3797 $NID would be instantly evicted from this MDS and from all active
3798 OSTs connected to it.
3802 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3803 Details : lookup_it_finish() used to query file size from OSTs that was not
3808 Description: service threads change working directory to that of init
3809 Details : Starting lustre service threads may pin the working directory
3810 of the parent thread, making that filesystem busy. Threads
3811 now change to the working directory of init to avoid this.
3815 Frequency : during shutdown only
3816 Description: shutdown with a failed MDS or OST can cause unmount to hang
3817 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3822 Frequency : 2.6 only
3823 Description: chmod/chown may include an extra supplementary group
3824 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3825 supplementary group and if none is specified this is used.
3828 Bugzilla : 5479 (6816)
3830 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3831 Details : If lookup is in progress on a file that is unlinked we might try
3832 to revalidate the inode and fail in revalidate after lookup is
3833 complete and ll_file_open() enqueues the open again but
3834 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3837 Frequency : always, if lconf --abort_recovery used
3839 Description: lconf --abort_recovery fails with 'Operation not supported'
3840 Details : lconf was attempting to abort recovery on the MDT device and not
3843 ------------------------------------------------------------------------------
3845 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3850 Frequency : rare (only unsupported configurations with a node running as an
3852 Bugzilla : 6514, 5137
3853 Description: Mounting a Lustre file system on a node running as an OST could
3855 Details : OSTs now preallocates memory needed to write out data at
3856 startup, instead of when needed, to avoid having to
3857 allocate memory in possibly low memory situations.
3858 Specifically, if the file system is mounted on on OST,
3859 memory pressure could force it to try to write out data,
3860 which it needed to allocate memory to do. Due to the low
3861 memory, it would be unable to do so and the node would
3862 become unresponsive.
3864 Severity : enhancement
3866 Description: Addition of lconf --service command line option
3867 Details : lconf now accepts a '--service <arg>' option, which is
3868 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3870 Severity : enhancement
3872 Description: Failover mode is now the default for OSTs.
3873 Details : By default, OSTs will now run in failover mode. To return to
3874 the old behaviour, add '--failout' to the lmc line for OSTs.
3876 Severity : enhancement
3878 Description: Health checks are now provided for MDS and OSTs
3879 Details : Additional detailed health check information on MSD and OSTs
3880 is now provided through the procfs health_check value.
3883 Frequency : occasional, depends on IO load
3885 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3886 after numerous create/delete cycles
3887 Details : The ext3 inode allocation policy would not allocate new inodes
3888 very well on the OSTs because there are no new directories
3889 being created. Instead we look for groups with free space if
3890 the parent directories are nearly full.
3895 Description: Network or server problems during mount may cause partially
3896 mounted clients instead of returning an error.
3897 Details : The config llog parsing code may overwrite the error return
3898 code during mount error handling, returning success instead
3904 Description: MDS can fail to allocate large reply buffers
3905 Details : After long uptimes the MDS can fail to allocate large reply
3906 buffers (e.g. zconf client mount config records) due to memory
3907 fragmentation or consumption by the buffer cache. Preallocate
3908 some large reply buffers so that these replies can be sent even
3909 under memory pressure.
3913 Frequency : rare (liblustre)
3914 Description: fsx running with liblustre complained that using truncate() to
3915 extend the file doesn't work. This patch corrects that issue.
3916 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3917 ATTR_SIZE and lsm use in llu_setattr_raw.
3921 Frequency : rare, only 2.6 kernels
3922 Description: Unusual file access patterns on the MDS may result in inode
3923 data being lost in very rare circumstances.
3924 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3925 code in the 2.6 kernel caused the inode and/or EA data not to
3926 be read from disk, causing single-file corruption.
3930 Frequency : rare, only 2.6 filesystems using extents
3931 Description: Heavy concurrent write and delete load may cause data corruption.
3932 Details : It was possible under high-load situations to have an extent
3933 metadata block in the block device cache from a just-unlinked
3934 file overwrite a newly-allocated data block. We now unmap any
3935 metadata buffers that alias just-allocated data blocks.
3939 Frequency : filesystems with default stripe_count larger than 77
3940 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3941 Details : lconf specifies an inode size of 4096 bytes when the default
3942 stripe_count is larger than 77. This conflicts with the default
3943 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3944 this case to avoid pinning too much memory for large EAs.
3946 ------------------------------------------------------------------------------
3948 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3953 Frequency : rare (extremely heavy IO load with hundreds of clients)
3955 Description: Client is evicted, gets IO error writing to file
3956 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3957 caused clients to be evicted for AST timeouts. The fixes in
3958 bug 5192 mean we no longer need to have such short AST timeouts
3959 so ldlm_timeout has been increased.
3962 Frequency : occasional during --force or --failover shutdown under load
3963 Bugzilla : 5949, 4834
3964 Description: Server oops/LBUG if stopped with --force or --failover under load
3965 Details : a collection of import/export refcount and cleanup ordering
3966 issues fixed for safer force cleanup
3969 Frequency : only filesystems larger than 120 OSTs
3970 Bugzilla : 5990, 6223
3971 Description: lfs getstripe would oops on a very large filesystem
3972 Details : lov_getconfig used kfree on vmalloc'd memory
3975 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3976 Bugzilla : 6242, 6243
3977 Description: reading from files that had been truncated to a non-zero size
3978 but never opened returned no data
3979 Details : ll_file_read() reads zeros from no-object files to EOF
3984 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3985 Details : There's little harm in aborting MDS/OSS recovery and letting it
3986 try again, so I removed the LASSERT and return an error instead.
3988 Severity : enhancement
3990 Description: New debugging infrastructure for tracking down data corruption
3991 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3992 (b) cover more of the client-side code path, and (c) try to narrow
3993 down where problems occurred
3997 Bugzilla : 3819, 4364, 4397, 6313
3998 Description: Racing close and eviction MDS could cause assertion in mds_close
3999 Details : It was possible to get multiple mfd references during close and
4000 client eviction, leading to one thread referencing a freed mfd.
4002 Severity: : enhancement
4003 Bugzilla : 3262, 6359
4004 Description: Attempts to reconnect to servers are now more aggressive.
4005 Details : This builds on the enhanced upcall-less recovery that was added
4006 in 1.4.2. When trying to reconnect to servers, clients will
4007 now try each server in the failover group every 10 seconds. By
4008 default, clients would previously try one server every 25 seconds.
4013 Description: After recovery, certain operations trigger a failed
4014 assertion on a client.
4015 Details : Failing over an mds, using lconf -d --failover, while a
4016 client was doing a readdir() call would cause the client to
4017 LBUG after recovery completed and the readdir() was resent.
4019 Severity : enhancement
4021 Description: Default groups are now added by lconf
4022 Details : You can now run lconf --group <servicename> without having to
4023 manually add groups with lmc.
4026 Frequency : occasional
4028 Description: Nodes with an elan id of 0 trigger a failed assertion
4031 Frequency : always when accessing e.g. tty/console device nodes
4033 Description: tty and some other devices nodes cannot be used on lustre
4034 Details : file's private_data field is used by device data and lustre
4035 values in there got lost. New field was added to struct file to
4036 store fs-specific private data.
4039 Frequency : when exporting Lustre via NFS
4041 Description: NFSD failed occasionally when looking up a path component
4042 Details : NFSD is looking up ".." which was broken in ext3 directories
4043 that had grown large enough to become hashed.
4046 Frequency : Clusters with multiple interfaces not on the same subnet
4048 Description: Nodes will repeatedly try to reconnect to an interface which it
4049 cannot reach and report an error to the log.
4050 Details : Extra peer list entries will be created by lconf with some peers
4051 unreachable. lconf now validates the peer before adding it.
4054 Frequency : Only if a default stripe is set on the filesystem root.
4056 Description: Setting a default stripe on the filesystem root prevented the
4057 filesystem from being remounted.
4058 Details : The client was sending extra request flags in the root getattr
4059 request and did not allocate a reply buffer for the dir EA.
4062 Frequency : occasional, higher if lots of files are accessed by one client
4063 Bugzilla : 6159, 6097
4064 Description: Client trips assertion regarding lsm mismatch/magic
4065 Details : While revalidating inodes the VFS looks up inodes with ifind()
4066 and in rare cases can find an inode that is being freed.
4067 The ll_test_inode() code will free the lsm during ifind()
4068 when it finds an existing inode and then the VFS later attaches
4069 this free lsm to a new inode.
4073 Bugzilla : 6422, 7030
4074 Description: MDS deadlock between mkdir and client eviction
4075 Details : Creating a new file via mkdir or mknod (starting a transaction
4076 and getting the ns lock) can deadlock with client eviction
4077 (gets ns lock and trying to finish a synchronous transaction).
4080 Frequency : occasional
4081 Description: While starting a server, the fsfilt_ext3 module could not be
4083 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
4084 kernels. Previously, lconf would still use the ext3 name
4085 when trying to load modules. Now, it will correctly use
4086 ext3 on 2.4 and ldiskfs on 2.6.
4088 Severity : enhancement
4089 Description: The default stripe count has been changed to 1
4090 Details : The interpretation of the default stripe count (0, to lfs
4091 or lmc) has been changed to mean striping across a single
4092 OST, rather than all available. For general usage we have
4093 found a stripe count of 1 or 2 works best.
4095 Severity : enhancement
4096 Description: Add support for compiling against Cray portals.
4097 Details : Conditional compiling for some areas that are different
4101 Frequency : occasional
4102 Bugzilla : 6409, 6834
4103 Description: Creating files with an explicit stripe count may lead to
4104 a failed assertion on the MDS
4105 Details : If some OSTs are full or unavailable, creating files may
4106 trigger a failed assertion on the MDS. Now, Lustre will
4107 try to use other servers or return an error to the
4111 Frequency : occasional
4113 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4114 caused lock timeout during readahead (since 1.4.2).
4115 Details : Processes doing readahead might match a lock that hasn't been
4116 granted yet if there are overlapping and conflicting lock
4117 requests. The readahead process waits on ungranted lock
4118 (original lock is CBPENDING), while OST waits for that process
4119 to cancel CBPENDING read lock and eventually evicts client.
4121 Severity : enhancement
4123 Description: Initial enabling of flock support for clients
4124 Details : Implements fcntl advisory locking and file status functions.
4125 This feature is provided as an optional mount flag (default
4126 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4127 locking are implemented yet, and those that are are not guaranteed
4128 to be completely correct in production environments.
4129 mount -t lustre -o [flock|noflock] ...
4132 Frequency : occasional
4134 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4135 assertion in the ext3 JBD (journaling) layer.
4136 Details : The b_committed_data struct is protected by the big kernel lock
4137 in 2.4 kernels, serializing journal_commit_transaction() and
4138 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4139 access to this struct. In 2.6 kernels there is finer grained
4140 locking to improve SMP performance of the JBD layer.
4144 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4149 Description: merge and cleanup kernel patches.
4152 -----------------------------------------------------------------------------
4154 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4156 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4157 versions of Lustre. Please update all servers and clients to
4158 version 1.4.2 or later at the same time. You must also run
4159 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4160 to update the configuration logs.
4162 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4163 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4164 - swab llog records of type '0' so we get proper header size/idx (5861)
4165 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4166 - fix rename of one directory over another leaking an inode (5953)
4167 - avoid SetPageDirty on 2.6 (5981)
4168 - don't re-add just-being-destroyed locks to the waiting list (5653)
4169 - when creating new directories, inherit the parent's custom
4170 striping settings if present parent (3048)
4171 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4172 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4173 - don't hold client locks on temporary worklist from l_lru (5666)
4174 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4175 - automatically evict dead clients (5921)
4176 - Update file size properly in create+truncate+fstat case (6196)
4177 - Do not unhash mountpoint dentries, do not allow removal of
4179 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4180 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4181 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4182 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4183 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4184 - Correctly handle reads of files with no objects (6243)
4185 - lctl recover will also mark a device active if deactivate used (5933)
4187 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
4188 - allow --write-conf on an MDS with different nettype than client (5619)
4189 - don't write config llogs to MDS for mounts not from that MDS (5617)
4190 - lconf should create multiple TCP connections from a client (5201)
4191 - init scripts are now turned off by default; run chkconfig --on
4192 lustre and chkconfig --on lustrefs to use them
4193 - upcalls are no longer needed for clients to recover to failover
4195 - add --abort-recovery option to lconf to abort recovery on device
4197 - add support for an arbitrary number of OSTs (3026)
4198 - Quota support protocol changes.
4199 - forward compatibility changes to wire structs (6007)
4200 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
4201 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
4202 - improved "lustre status" script
4203 - initialize blocksize for non-regular files (6062)
4204 - added --disable-server and --disable-client configure options (5782)
4205 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
4206 - Vanilla 2.4.29 support
4207 - increase maximum number of obd devices to 520 (6242)
4208 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4209 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4210 SLES 9 2.6.5 kernels
4211 - limit stripes per file to 160 (the maximum EA size) (6093)
4213 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4216 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4217 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4218 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4219 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4220 - small ext3 extents cleanups and fixes (5733)
4221 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4222 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4223 - avoid needless client->OST connect, fix handle mismatch (5317)
4224 - fix DLM error path that led to out-of-sync client, long delays (5779)
4225 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4226 - fix several locking issues related to i_size (5492,5624,5654,5672)
4227 - don't move pending lock onto export if it is already evicted (5683)
4228 - fix kernel oops when creating .foo in unlinked directory (5548)
4229 - fix deadlock in obdfilter statistics vs. object create (5811)
4230 - use time_{before,after} to avoid timer jiffies wrap (5882)
4231 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4232 - Do not leak request if server was not able to process it (5154)
4233 - If mds_open unable to find parent dir, make that negative lookup(5154)
4234 - don't create new directories with extent-mapping (5909, 5936)
4236 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
4237 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
4238 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
4239 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
4240 - support for RHEL4 kernel on i686 (5773)
4241 - provide error messages when incompatible logs are encountered (5898)
4243 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
4244 * version 1.4.0.10 (1.4.1 release candidate 1)
4246 - don't keep a lock reference when lock is not granted (4238)
4247 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4248 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
4249 - drop import inflight refcount on signal_completed_replay error (5255)
4250 - unlock page after async write error during send (3677)
4251 - handle missing objects in filter_preprw_read properly (5265)
4252 - no transno return for symlink open, don't save no-trasno open (3440)
4253 - don't try to complete elan receive that already failed (4012)
4254 - free RPC server reply state on error (5406)
4255 - clean up thread from ptlrpc_start_thread() on error (5160)
4256 - readahead could read extra page into cache that wasn't ejected (5388)
4257 - prevent races in class_attach/setup/cleanup/detach (5260)
4258 - don't dereference de->d_inode after l_dput of de (5458)
4259 - use "int" for stripe value returned from lock_to_stripe (5544)
4260 - mballoc allocation and error-checking fixes in 2.6 (5504)
4261 - block device patches to fix I/O request sizes in 2.6 (5482)
4262 - look up hostnames for IB nals (5602)
4263 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
4264 - don't start multiple acceptors for the same port (5277)
4265 - fix incorrect LASSERT in mds_getattr_name (5635)
4266 - export a proc file for general "ping" checking (5628)
4267 - fix "lfs check" to not block when the MDS is down (5628)
4269 - service request history (4965)
4270 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4271 - create an "evict_client" /proc entry on OSTs, like the MDS has
4272 - fix mount usage message, return errors per mount(8) (5168)
4273 - change grep [] to grep "[]" in tests so they work in more UMLs
4274 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4275 - remove ancient LOV_MAGIC_V0 EA support (5047)
4276 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4277 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4278 - allow readahead allocations to fail when low on memory (5383)
4279 - mmap locking landed again, after considerable improvement (2828)
4280 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4282 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4285 - send OST transaction number in read/write reply to free req (4966)
4286 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4287 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4288 - the watchdog thread now runs as interruptible (5246)
4289 - flock/lockf fixes (but it's still disabled, pending 5135)
4290 - don't use EXT3 constants in llite code (5094)
4291 - memory shortage at startup could cause assertion (5176)
4293 - reorganization of lov code
4294 - single portals codebase
4296 - add extents/mballoc support (5025)
4297 - direct I/O reads in the obdfilter (4048)
4298 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4300 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4303 - send OST transaction number in read/write reply to free req (4966)
4304 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4305 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4306 - the watchdog thread now runs as interruptible (5246)
4307 - handle missing objects in filter_preprw_read properly (5265)
4308 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4309 - drop import inflight refcount on signal_completed_replay error (5255)
4310 - unlock page after async write error during send (3677)
4311 - return original error code on reconstructed replies (3761)
4312 - no transno return for symlink open, don't save no-trasno open (3440)
4314 - add pid to ldlm debugging output (4922)
4315 - bump the watchdog timeouts -- we can't handle 30sec yet
4316 - extra debugging for orphan dentry/inode bug (5259)
4318 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
4321 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4322 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4323 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
4324 - fix NULL dereference in /proc/sys/portals/routes (4827)
4325 - allow failed mdc_close() operations to be interrupted (4561)
4326 - stop precreate on OST before MDS would time out on it (4778)
4327 - don't send partial-page writes before EOF from client (4410)
4328 - discard client grant for sub-page writes on large-page clients (4520)
4329 - don't free dentries not owned by NFS code, check generation (4806)
4330 - fix lsm leak if mds_create_objects() fails (4801)
4331 - limit debug_daemon file size, always print CERROR messages (4789)
4332 - use transno after validating reply (3892)
4333 - process timed out requests if import state changes (3754)
4334 - update mtime on OST during writes, return in glimpse (4829)
4335 - add mkfsoptions to LDAP (4679)
4336 - use ->max_readahead method instead of zapping global ra (5039)
4337 - don't interrupt __l_wait_event() during strace
4339 - add software watchdogs to catch hung threads quickly (4941)
4340 - make lustrefs init script start after nfs is mounted
4341 - fix CWARN/ERROR duplication (4930)
4342 - return async write errors to application if possible (2248)
4343 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4344 - print NAL number in %x format (4645)
4345 - update barely-supported suse-2.4.21-171 series (4842)
4346 - support for sles 9 %post scripts
4347 - support for building 2.6 kernel-source packages
4348 - support for sles km_* packages
4350 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4353 - ignore -ENOENT errors in osc_destroy (3639)
4354 - notify osc create thread that OSC is being cleaned up (4600)
4355 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4356 - reconstruct ost_handle() like mds_handle() (4657)
4357 - create a new thread to do import eviction to avoid deadlock (3969)
4358 - let lconf resolve symlinked-to devices (4629)
4359 - don't unlink "objects" from directory with default EA (4554)
4360 - hold socknal file ref over connect in case target is down (4394)
4361 - allow more than 32000 subdirectories in a single directory (3244)
4362 - fix blocks count for O_DIRECT writes (3751)
4363 - OST returns ENOSPC from object create when no space left (4539)
4364 - don't send truncate RPC if file size isn't changing (4410)
4365 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4366 - bind to privileged port in socknal and tcpnal (3689)
4368 - rate limit CERROR/CWARN console message to avoid overload (4519)
4369 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4370 - basic mmap support (3918)
4371 - kernel patch series update from b1_4 (4711)
4373 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4376 - avoid crash during MDS cleanup with OST shut down (2775)
4377 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4378 - don't use bad inodes on the MDS (3744)
4379 - dynamic object preallocation to improve recovery speed (4236)
4380 - don't hold spinlock over lock dumping or change debug flags (4401)
4381 - don't zero obd_dev when it is force cleaned (3651)
4382 - print grants to console if they go negative (4431)
4383 - "lctl deactivate" will stop automatic recovery attempts (3406)
4384 - look for existing locks in ldlm_handle_enqueue() (3764)
4385 - don't resolve lock handle twice in recovery avoiding race (4401)
4386 - revalidate should check working dir is a directory (4134)
4388 - don't always mark "slow" obdfilter messages as errors (4418)
4390 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4393 - don't close LustreDB during write_conf until it is done (3860)
4394 - fix typo in lconf for_each_profile (3821)
4395 - allow dumping logs from multiple threads at one time (3820)
4396 - don't allow multiple threads in OSC recovery (3812)
4397 - fix debug_size parameters (3864)
4398 - fix mds_postrecov to initialize import for llog ctxt (3121)
4399 - replace config semaphore with spinlock (3306)
4400 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4401 - don't allow enqueue to complete on a destroyed export (3822)
4402 - down write_lock before checking llog header bitmap (3825)
4403 - recover from lock replay timeout (3764)
4404 - up llog sem before sending rpc (3652)
4405 - reduce ns lock hold times when setting kms (3267)
4406 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4407 - fix NULL deref and obd_dev leak on setup error (3312)
4408 - replace some LBUG about llog ops with error handling (3841)
4409 - don't match INVALID dentries from d_lookup and spin (3784)
4410 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4411 - fix invalid assertion in ptlrpc_set_wait (3880)
4413 - add libwrap support for the TCP acceptor (3996)
4414 - add /proc/sys/portals/routes for non-root route listing (3994)
4415 - allow setting MDS UUID in .xml (2580)
4416 - print the stack of a process that LBUGs (4228)
4418 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4421 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4422 - make sure to unset replay flag from failed open requests (3440)
4423 - if default stripe count is 0, use OST count for inode size (3636)
4424 - update parent mtime/ctime on client for create/unlink (2611)
4425 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4426 - free recovery state on server during a forced cleanup (3571)
4427 - unregister_reply for resent reqs (3063)
4428 - loop back devices mounting and status check on 2.6 (3563)
4429 - fix resource-creation race that can provoke i_size == 0 (3513)
4430 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4431 - more debugging for page-accounting assertion (3746)
4432 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4433 - avoid deadlock after precreation failure (3758)
4434 - fix race and lock order deadlock in orphan handling (3450, 3750)
4435 - add validity checks when grabbing inodes from l_ast_data (3599)
4437 - add /proc/.../recovery_status to obdfilter (3428)
4438 - lightweight CDEBUG infrastructure, debug daemon (3668)
4439 - change default OSC RPC parameters to be better on small clusters
4440 - turn off OST read cache for files smaller than 32MB
4441 - install man pages and include them in rpms (3100)
4442 - add new init script for (un)mounting lustre filesystems (2593)
4443 - run chkconfig in %post for init scripts (3701)
4444 - drop scimac NAL (unmaintained)
4446 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4449 - clean kiobufs before and after use (3485)
4450 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4451 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4452 - change init script to fail more clearly if not run as root (1528)
4453 - allow clients to reconnect during replay (1742)
4454 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4455 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4456 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4457 - fix cond_resched (3554)
4458 - only evict unfinished clients after recovery (3515)
4459 - allow bulk resend, prevent data loss (3570)
4460 - dynamic ptlrpc request buffer allocation (2102)
4461 - don't allow unlinking open directory if it isn't empty (2904)
4462 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4463 - remove extraneous obd dereference causing LASSERT failure (3334)
4464 - don't use get_cycles() when creating temp. files on the mds (3156)
4465 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4466 - handle EEXIST for set-stripe, set proper directory name (3336)
4468 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4469 - fix ksocknal_fmb_callback() error messages (2918)
4471 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4474 - don't copy lvb into (possibly NULL) reply on error (2983)
4475 - don't deref dentry after dput, don't free lvb on error (2922)
4476 - use the kms to determine writeback rpc length (2947)
4477 - increment oti_logcookies when osc is inactive (2948)
4478 - update client's i_blocks count via lvb messages (2543)
4479 - handle intent open/close of special files properly (1557)
4480 - mount MDS with errors=remount-ro, like obdfilter (2009)
4481 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4482 - don't use cancelling-locks' kms values (2947)
4483 - use highest lock extent for kms, not last one (2925)
4484 - don't dereference ERR_PTR() dentry in error handling path (3107)
4485 - fix thread race in portals_debug_dumplog() (3122)
4486 - create lprocfs device entries at setup instead of at attach (1519)
4487 - common AST error handler, don't evict client on completion race (3145)
4488 - zero nameidata in detach_mnt in 2.6 (3118)
4489 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4490 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4491 - initialize RPC timeout timer earlier for 2.6 (3219)
4492 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4493 - print nal/nid for unknown nid (3258)
4494 - additional checks for oscc recovery before doing precreate (3284)
4495 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4496 - don't crash in mdc_close for bad permissions on open (3285)
4497 - zero i_rdev for non-device files (3147)
4498 - clear page->private before handing to FS, better assertion (3119)
4499 - tune the read pipeline (3236)
4500 - fix incorrect decref of invalidated dentry (2350)
4501 - provide read-ahead stats and refine rpc in flight stats (3328)
4502 - don't hold journal transaction open across create RPC (3313)
4503 - update atime on MDS at close time (3265)
4504 - close LDAP connection when recovering to avoid server load (3315)
4505 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4506 - don't leak open file on MDS after open resend (3325)
4507 - serialize filter_precreate and filter_destroy_precreated (3329)
4508 - loop device shouldn't call sync_dev() for nul device (3092)
4509 - clear page cache after eviction (2766)
4510 - resynchronize MDS->OST in background (2824)
4511 - refuse to mount the same filesystem twice on same mountpoint (3394)
4512 - allow llmount to create routes for mounting behind routers (3320)
4513 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4514 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4515 - fix rare problem with rename on htree directories (3417)
4517 - allow default OST striping configuration per directory (1414)
4518 - fix compilation for qswnal for 2.6 kernels (3125)
4519 - increase maximum number of MDS request buffers for large systems
4520 - change liblustreapi to be useful for external progs like lfsck (3098)
4521 - increase local configuration timeout for slow disks (3353)
4522 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4524 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4527 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4528 - don't overwrite extent policy data in reply if lock was blocked (2901)
4529 - drop filter export grants atomically with removal from device (2663)
4530 - del obd_self_export from work_list in class_disconnect_exports (2908)
4531 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4532 - swab reply message in mdc_close, other PPC fixes (2464)
4533 - fix destroying of named logs (2325)
4534 - overwrite old logs when running lconf --write_conf (2264)
4535 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4536 - fix race in target_handle_connect (2898)
4537 - mds_reint_create() should take same inode create lock (2926)
4538 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4539 - don't close files for self_export to avoid uninitialized obd (2936)
4540 - allow MDS with the same name as client node (2939)
4541 - hold dentry reference for closed log files for unlink (2325)
4542 - reserve space for all logs during transactions (2059)
4543 - don't evict page beyond end of stripe extent (2925)
4544 - don't oops on a deleted current working directory (2399)
4545 - handle hard links to targets without a parent properly (2517)
4546 - don't dereference NULL lock when racing during eviction (2867)
4547 - don't grow lock extents when lots of conflicting locks (2919)
4549 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4552 - account for cache space usage on clients to avoid data loss (974)
4553 - lfsck support in lustre kernel code (2349)
4554 - reduce journal credits needed for BRW writes (2370)
4555 - orphan handling to avoid losing space on client/server crashes
4556 - ptlrpcd can be blocked, stopping ALL progress (2477)
4557 - use lock value blocks to assist in proper KMS, faster stat (1021)
4558 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4559 - recovery for initial connections (2355)
4560 - fixes for mds_cleanup_orphans (1934)
4561 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4562 - block all file creations until orphan recovery completes (1901)
4563 - client remove rq_connection from request struct (2423)
4564 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4565 - recovery timer race (2670)
4566 - mdc_close recovey bug (2532)
4567 - ptlrpc cleanup bug (2710)
4568 - mds timeout on local locks (2588)
4569 - namespace lock held during RPCs (2431)
4570 - handle interrupted sync write properly (2503)
4571 - don't try to handle a message that hasn't been replied to (2699)
4572 - client assert failure during cleanup after abort recovery (2701)
4573 - leak mdc device after failed mount (2712)
4574 - ptlrpc_check_set allows timedout requests to complete (2714)
4575 - wait for inflight reqs when ptlrpcd finishes (2710)
4576 - make sure unregistered services are removed from the srv_list
4577 - reset bulk XID's when resending them (caught by 1138 test)
4578 - unregister_bulk after timeout
4579 - fix lconf error (2694)
4580 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4581 - readahead locks pages, leaves pending causing memory pressure (2673)
4582 - increase OST request buffers to 4096 on large machines (2729)
4583 - fix up permission of existing directories in simple_mkdir (2661)
4584 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4585 - don't assign transno to errored transactions (2742)
4586 - don't delete objects on OST if given a bogus objid from MDS (2751)
4587 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4588 - if rq_no_resend, then timeout request after recovery (2432)
4589 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4590 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4591 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4592 - MDS recovery shouldn't delete objects that it has given out (2730)
4593 - if enqueue arrives after completion, don't clobber LVB (2819)
4594 - don't unlock pages twice when trigger_group_io returns error (2814)
4595 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4596 - don't write pages to disk if there was an error (1450)
4597 - don't ping imports that have recovery disabled (2676)
4598 - take buffered bytes into account when balancing socknal conn (2817)
4599 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4600 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4601 - remove little-endian swabbing of llog records (1987)
4602 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4603 - retry reposting request buffers if they fail (1191)
4604 - grow extent at grant time to avoid granting a revoked lock (2809)
4605 - lock revoke doesn't evict page if covered by a second lock (2765)
4606 - disable VM readahead to avoid reading outside lock extents (2805)
4608 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4609 - updated kernel patches for hp-2.4.20 kernel (2681)
4611 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4614 - fix truncated write corruption (2366)
4615 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4617 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4618 - ensure len(uuid) < 37 in lmc (1171)
4619 - fix ia64 OOPS in llog_test (2255)
4620 - zero end of page at obdfilter for partial page writes (2648)
4621 - don't leave stale dentries around after renames (bug 2428)
4622 - fix timeouts when evicting a client with a single lock held (2642)
4623 - set deadline for the initial HELLO message to drain (2634)
4624 - print out dotted-quad IP addresses in the socknal (2302)
4626 - additional debugging for MDS client eviction problem (2443)
4627 - fix mkfsoptions support for osts (2603, 2604)
4629 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4632 - add series for the vanilla 2.6.0 kernel
4633 - add series for the vanilla 2.4.24 kernel
4634 - add series for a cray x86/64 UL kernel drop
4635 - fix xattr patches for the vanilla 2.4.19 series
4637 - generate true UUIDs in lmc (1171)
4638 - have portals stack dumping break in UML (2466)
4639 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4640 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4641 - wake callback waiting threads on client eviction (2460)
4642 - Add --ptldebug and --subsystem to lmc (1719)
4643 - update assertion to allow safe interrupt allocation
4644 - set rq_no_resend for cancel requests (2432)
4645 - recalculate ptlrpcd timeout after resend (2494)
4646 - call vfs_rmdir when removing pending directories (2368)
4647 - fix renaming a file to itself (2429)
4648 - lmc creates a default one-stripe lov (2454)
4649 - expand procfs space to handle large clusters (2326)
4650 - increase UML stack to avoid overflow
4651 - update lconf's list of debug and subsystem masks
4652 - fix lfs find --obd (2510)
4653 - /proc tunable for disabling filter read caching (2591)
4654 - stop rpm packages from altering slapd.conf (2301)
4655 - disable nagle in the socknal under 0conf (2578)
4656 - choose mds inode size based on stripe count (2572)
4657 - fix kernel-source rpm problems (2516)
4659 - add --disable-doc to avoid pdf generation (2421)
4660 - update documentation, tests, type-os, comments
4661 - avoid format warnings on ia64
4662 - remove the TOE NAL
4663 - tiny code cleanups by removing unused fields
4665 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4668 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4669 - remove the most blatant lies from BUILDING file (2371)
4670 - change default debug level to reasonable production setting
4671 - reduce client side cache size to reduce cache flush time
4672 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4673 - make TCP zerocopy and pinger support enabled by default (2476)
4674 - sync writes completed after process exits caused crashes (2319)
4675 - maintain correct mount count on the MDS (2356)
4676 - backout 1557, because 2316 wasn't really fixed
4677 - better file I/O statistics gathering in /proc
4678 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4679 - another kernel patch to fix zero-copy TCP function export
4680 - don't take duplicate lock when processing re-sent getattr (2420)
4681 - lctl uses obd_self_export instead of creating new conn (2353)
4682 - MDS/OST recovery case which requires object creation asserted (2425)
4683 - move lfs from /usr/sbin to /usr/bin in packages
4684 - fix race between mds_client_add and mds_client_free (2417)
4685 - use kmalloc instead of slabs in portals (2430)
4686 - don't create duplicate records when a failover MDS is present (2442)
4687 - remove unnecessary mount age check (2332)
4688 - don't remove directory inodes from locks prematurely (2451)
4689 - don't break if MDS service name is the same as hostname (2103)
4690 - fix races in client write RPC generation when cache full (2482)
4692 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4695 - remove now-unused request->rq_obd (278)
4696 - if an allocation fails, print out how much memory we've used (1933)
4697 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4698 - add the "configurable stack size" patch to most series files (1256)
4699 - ability to write large log records, for 100+ OST configs (2306)
4700 - fix NULL deref when filter_prep fails (2314)
4701 - fix operator precedence error in filter_sync
4702 - dynamic allocation of socknal TX descriptors (2315)
4703 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4704 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4705 - ensure that CWARN messages in Portals always get to the syslog
4706 - __init/__exit are not for prototype decls (ldlm_init/exit)
4707 - x86-64 compile warning fixes
4708 - fix gateway LMC keyword conflict (2318)
4709 - fix MDS lock inversions in getattr/reint paths (1844)
4710 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4711 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4712 - DLM race condition prevented some lock evictions (2328)
4713 - ENOMEM detection and retry on socknal sends (2230)
4714 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4715 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4716 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4717 - better error messages when a client is rejected during recovery (1505)
4718 - avoid cancelling locks which were never granted, after failure (2330)
4719 - fix i_sem/journal inversion in mds_client_add (2333)
4720 - fix truncate/getattr lock cycle deadlock (2334)
4721 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4722 - fix two rare exit paths which could leak an l_lock() ref (2321)
4723 - fencepost error in MDS/OST orphan recovery (2226)
4724 - make log record alignment 8 bytes (1988)
4725 - lstripe now fails when requested offset > ost_count (2237)
4726 - ensure that all kernel series have a complete list.h (1607)
4727 - fix crashes in special-file operations (2316)
4728 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4730 - allow configurable automake binary, for testing new versions
4731 - small update to the lfs documentation
4733 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4735 * fix negative export reference count in fsfilt_sync (2312)
4737 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4738 * release candidate 0.9.1
4740 - orphans are moved into the PENDING directory for possible recovery
4741 - replayed opens now open by fid for orphan/rename safety (1042)
4742 - last close of an orphan inode generates a transno (683)
4743 - chdir() and mount() now pin the directory entry (1020)
4744 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4745 - discard very old requests without processing them (1502)
4746 - remove some common, well-understood CERRORs (1505)
4747 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4748 - clear "grant" flags in OST replies until OST grant code lands (1644)
4749 - fix read performance by not clobbering i_blksize on client (1598)
4750 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4751 - make LRU size a /proc tunable, clears locks when reduced (707)
4752 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4753 - server locks take references on exports now (1558)
4754 - build fixes for 2.4.20-rh trees (1663)
4755 - return an error from lov_create if all OSCs are inactive (1751)
4756 - fix import levels when a reconnect happens without a timeout (1597)
4757 - exit early from mds_open if we get a lookup error (1749)
4758 - partial page read at EOF wouldn't wait for disk before sending (1642)
4759 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4760 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4761 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4762 - write/truncate lock inversion (1639)
4763 - Don't auto-load obdclass, portals modules during cleanup (1495)
4764 - fix timestamps from jumping to "now" (1763)
4765 - extra journal assertions (1648)
4766 - add an extra multiunlink test (1771)
4767 - fix read_record/write_record API (1776)
4768 - fix leak of offset_extent, possible incorrect i_size later (1772)
4769 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4770 - Debugging for the kqswnal_get_idle_tx problems (1820)
4771 - Allow recovery to be attempted multiple times (1536)
4772 - Write out MDS last_rcvd file after it is first created (1600)
4773 - Fix tx_descriptor leak in failed transmit situations (1827)
4774 - ext3 journaling fixes for assertion failure after IO error (1871)
4775 - class_export_put() on freed export after completion AST error (1896)
4776 - Fix revalidate looping in VFS (1322)
4777 - Don't access a freed export during MDS_REINT timeout (1521)
4778 - Add open-unlink recovery support on the MDS (1673,1764)
4779 - Return an error if no MDS data was read from last_rcvd (1946)
4780 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4781 - Don't setup a disk device that doesn't match exported UUID (317)
4782 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4783 - avoid committing NULL handle in force close
4784 - local.sh is now a one-stripe LOV configuration
4785 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4786 - fix ext3 htree duplicate directory entry corruption (1516)
4787 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4788 - update to new LOV EA format (2097)
4789 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4791 2003-06-15 Phil Schwan <phil@clusterfs.com>
4794 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4795 - per-import/export recovery handling (958, 931, 959)
4796 - multiple last-rcvd slots, for serving multiple FSes (949)
4797 - connections are again shared between multiple imp/exports (963, 964)
4798 - "umount -f" would hang if any requests needed to be sent (393, 978)
4799 - avoid pinning large req buffer by copying for queued messages (989)
4800 - add "uuid" to "lctl device" command to help upcalls (991)
4801 - "open" RPCs with transnos would confuse recovery counters (1037)
4802 - do proper endian conversion of all wire messages (288, 340, 891)
4803 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4804 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4805 - fix ptlrpc_connection leak in target_handle_connect (1174)
4806 - fix import refcounting bug in OST and MDS cleanup (1134)
4807 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4808 - fix very unlikely obd_types race condition (501)
4809 - remove osc_open hack for echo_client (1187)
4810 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4811 - a failure in read_inode2 leads to deadlock (1139)
4812 - cancel ack-locks as soon as transaction is committed (1072)
4813 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4814 - make sure to commitrw after any preprw to avoid deadlock (1162)
4815 - failing to execute a file in a lustre FS would lock inode (1203)
4816 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4817 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4818 - obd_uuid2tgt would walk off the end of the list (1255)
4819 - on IA64 the osc would give portals incorrect bulk size (1258)
4820 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4821 - fix lock inversion caused by new llite matching code (1282)
4822 - limit the number of dirty pages on a client to 10MB (1286)
4823 - timed out locks were not being corrected cancelled (1289)
4824 - fix O_DIRECT above 4GB on IA-32 (1292)
4825 * major user-visible changes
4826 - fail out/fail over policy now controlled by the upcall (993)
4828 - add OBD_PING to check server availability and failure (954)
4829 - lustre messages are now sent in sending host order (288, 340, 891)
4830 - add eadatalen to MDS getattr reply (340)
4831 - OST read replies may contain second buffer, with per-page status (593)
4833 2003-03-11 Phil Schwan <phil@clusterfs.com>
4836 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4837 - failed open()s could cause deadlock; fixed (867, 869)
4838 - stop cancelling OST locks when files are closed (481)
4839 - overlapping XID spaces caused network corruption (851, 853)
4840 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4841 - setattr_raw added, to do single-RPC, server-side setattrs
4842 - lmc/lconf syntax change for OST UUIDs
4843 - fix crashy race condition between ptlrpc_free_req and osc_close
4844 - don't use request in mdc_enqueue if we hit a timeout (889)
4845 - don't set the inode i_size for regular files from the MDS (896)
4846 - handle out of order completion AST (842)
4847 - don't LBUG if a lock request times out after receiving AST (913)
4848 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4849 - if a bad lock AST arrives, send an error instead of dropping entirely
4850 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4851 - fix leak in bulk IO when only partially completed (899, 900, 926)
4852 - fix O_DIRECT for ia64 (55)
4853 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4854 - C-z after timeout could hang a process forever; fixed (977)
4856 - client-side I/O cache (678, 924, 929, 941, 970)
4858 - READPAGE and SETATTRs which don't take server-side locks get
4861 2003-02-11 Phil Schwan <phil@clusterfs.com>
4864 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4865 - Fix fsfilt_extN_readpage() to read a full page of directory
4866 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4867 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4868 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4869 - Most intent operations were removed, in favour of a new RPC mode
4870 that does a single RPC to the server and bypasses most of the VFS
4871 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4872 - Aggressively cancel local locks on DLM servers
4873 - mds_reint_unlink sends EA to the client if it's the last nlink.
4874 client uses that EA to unlink OST objects.
4875 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4876 - recursive symlinks were fixed (439)
4877 - fixed NULL deref in DEBUG_REQ
4878 - filter_update_lastobjid no longer calls sync, which annoyed extN
4879 - fixed multi-client small-writes to a single file problem (445)
4880 - fixed mtime updates during file writes (607)
4881 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4882 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4883 - fixed Linux OST size reporting problem (444, 656)
4884 - OST now updates object mtime with writes or setattr (607, 619)
4885 - client verifies file size before zeroing page past EOF (445)
4886 - OST now writes last allocated objid to disk with allocation (108)
4887 - LOV on echo now works (409)
4889 - mds_reint_unlink sends a new buffer, with the EA included. this
4890 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4891 set if a regular file was being unlinked, and it was the last link
4892 - use PtlGet from the target for bulk writes (315)
4893 - OST now updates object mtime with writes or setattr (607, 619)
4894 - LDLM now has a grant-time callback to revalidate locked items, if
4896 - Many MDS operations were reorganized to combat race conditions
4898 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4899 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4901 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4904 - Fully reactivate OST imports after reconnection (512, others)
4905 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4906 - More graceful error handling for truncating on dead OST (515)
4907 - Don't error out unless we're actually accessing dead stripes (474)
4908 - Fix garbage sizes when stripes are missing (410)
4909 - LRU counters were broken, causing constant lock purge (433, 432)
4910 - garbage on read from stripes with failed OSTs (441)
4911 - mark OSCs as active before reconnecting during recovery (438)
4912 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4913 - lfind did not preserve OST order in output (443)
4914 - symlinks cause hung clients, incorrect data (439)
4915 - stop dereferencing request after dropping refcount (457)
4916 - don't LASSERT(spin_is_locked) on non-SMP (455)
4917 - fixes for many rename() bugs
4918 - fstat didn't correctly synchronize attributes (399)
4919 - server must handle lock cancellation during blocking AST prep (487)
4920 - bulk descriptors were free()d too soon (511)
4921 - fix paths in lconf, which would load incorrect modules (451, 507)
4922 - fix confusing lconf 'host not found' error message (386)
4923 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4924 - fix race condition in mdc_blocking_ast() for inode access (526)
4925 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4926 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4927 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4928 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4929 - fix lstripe in several ways (kernel side) (527)
4930 - fix request leak in ldlm_cli_enqueue (262)
4931 - incorrect OSC was marked inactive after OST failure
4932 - call mds_fs_cleanup before unmounting filesystem (524)
4933 - fix races between taking ns_lock and ldlm_lock_change_resource
4934 - fix races updating LOV export open file list
4935 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4936 - fix recovery NULL deref in ldlm_cli_cancel_unused
4937 - fix some DLM races by using new hash table for lock handles (419)
4938 - permit the client to specify desired inodes, at replay
4939 - duplicate requests when we queue them for replay reintegration
4940 - fix last_rcvd offset calculation
4941 - sync after each recovered transaction, so we always make progress
4942 - never, not always, ERESTART requests without transnos
4943 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4945 - skip replay if the MDS says that the client is already connected
4946 - don't check for a recovery-enabled export to match lctl's UUID
4947 - don't INC_USE_COUNT for phantom exports
4948 - don't crash when cleaning up phantom exports (567)
4949 - don't double-finish or set replay data for errored mdc_open requests
4950 - abort requests when they time out, so we don't get old replies
4951 - send/receive replies for AST messages again
4952 - if the client says that it doesn't have the lock, cancel it on the
4954 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4955 mark it as destroyed, it will all work out when decref is called
4956 - fix module use counts (22, 581)
4958 - ASTs now expect a reply (server cancels lock on error reply)
4960 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4963 - fix many simultaneous client startup (392)
4964 - fix dentry->d_it clobbering
4965 - credentials weren't being shipped for readdir/getattr operations
4966 - remove invalid assertions triggered during some concurrent MD
4968 - proper Lustre versions added (336, 389)
4969 - fix memory leak for create error case (398)
4970 - fix LOV locking bug that would get cli/srv out of sync
4971 - fix echo client over LOV (409)
4972 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4973 - fix double-O_EXCL intent crash (424)
4974 - avoid sending multiple lock CANCELs (352)
4976 - MDS can do multi-client recovery (modulo bugs in new code)
4978 - many updates, edits, cleanups
4980 2002-11-18 Phil Schwan <phil@clusterfs.com>
4983 - fix null d_it dereference (346)
4984 - fix full OST/dbench hang (333)
4985 - fix permission problem with file removal (286)
4986 - fix removal of OSCs from LOV when they fail
4987 - fix NULL deref during bulk timeout (214)
4988 - fix problems related to multiple filesystems on one MDS (241)
4989 - fixed serious subtle metadata locking bugs
4990 - free locks on clients when inodes are removed due to memory
4992 - fix inode pointer in lock data (285)
4993 - partial support for multiple MDS on a single host (241)
4994 - data locks weren't cancelled at clear_inode time (290, 311)
4995 - intent locks could lead to unbounded lock growth (205)
4996 - added a maximum lock count, an LRU list, and a flusher
4997 - fix multiple rename (365)
4998 - properly abstracted the echo client
4999 - OSC locked 1 byte too many; fixed
5000 - rewrote brw callback code:
5001 - fixed recovery bugs related to LOVs (306)
5002 - fixed too-many-pages-in-one-write crash (191)
5003 - fixed (again) crash in sync_io_timeout (214)
5004 - probably fixed callback-related race (385)
5006 - Add capability to MDS protocol
5007 - LDLM cancellations and callbacks on different portals
5009 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
5012 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
5013 - defer open object destruction to close time (601981, 138)
5014 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
5015 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
5016 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
5017 - don't call obd_setattr in ll_file_release for destroyed objects
5018 * protocol change to lustre_msg: move |version| and add |flags|
5019 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
5020 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
5021 * added replay of create, unlink, link and rename operations during
5022 MDS failover; recovery should be much more robust now
5023 * remove failed OSCs from LOVs (only lov_create uses this so far)
5024 * the lustre-HOWTO was brought (more) up to date (582544)
5026 2002-10-23 Phil Schwan <phil@clusterfs.com>
5029 - in-use dentries weren't being reused properly (617851)
5030 - prevent multiple LDLM setup (599178)
5031 - fix LOV size calculations for truncate (617853)
5032 - fix client handling of MDS intent errors (POSIX)
5033 - fix permission bug in lovstripe.c test (624321)
5034 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
5035 - truncate past end of file could corrupt data
5036 - proper cleanup after timeouts, crashes, etc (592524, 550815)
5037 - a race in recovery could return ETIMEDOUT to apps (623947)
5038 - building outside the source directory was fixed
5039 * the lustre-HOWTO was brought (more) up to date (582544)
5040 * major progress was made on recovery functionality
5042 2002-10-10 Phil Schwan <phil@clusterfs.com>
5045 - recovery deadlock fix
5046 - rm -rf causes LBUG fix (617817)
5047 - file open by multiple tasks fix (618962)
5048 - directory permissions bugs (602707 and 620007)
5049 - journal_stop fixed with locking (611313)
5050 - O_APPEND failures resolved (618273, perhaps 614459)
5051 - lconf PATH fix (619770)
5052 - IA64 build fix (621450)
5053 - RPC buffer sizes scale with amount of memory
5055 2002-10-01 Phil Schwan <phil@clusterfs.com>
5058 - locks would be cancelled without throwing away data pages,
5059 resulting in inconsistent data (605627)
5060 - inode attributes were not always being refreshed (605627, 612449)
5061 - lconf now continues to cleanup after lctl reports an error
5062 - MDS now enforces user permissions (602707)
5063 - lprocfs cleanup fixed, but not yet enabled (614157)
5064 - fixed infinite server hang, should a client not respond to an AST
5065 - avoid going into recovery if user calls readlink() with a buffer
5066 that's too small (613941)
5067 - AST RPCs no longer require replies (614867) -- this may be changed
5068 - don't crash server if client sends an IOV that's too big (611336)
5069 - fixed lock conversion deadlock (611892)
5070 - fixed the following of symlinks (614622)
5071 * recovery: the server can remove locks from a client that dies, other
5072 clients can make progress
5073 * more extN patch fixes
5074 * compile-time configurable ptlrpc buffer allocations
5076 - collaborative read cache document
5077 - Lustre Lite Performance CDR document-in-progress
5079 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5082 - fix typo in patch-2.4.18
5084 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5087 - clear ptlrpc request each time in handle_incoming_request()
5088 - unlink of files now destroys the object on the OST
5090 2002-09-19 Peter Braam <braam@clusterfs.com>
5092 * add hard link support
5093 * change obdfile creation method
5094 * kernel patch changed
5096 2002-09-19 Peter Braam <braam@clusterfs.com>
5099 - stack overflow bug in extN fixed
5101 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
5103 * documentation updates
5104 - add man pages for config tools
5105 - update tests/README to describe testing with new config tools
5106 - finish metadata API descriptions
5107 * bug fixes and cleanups
5108 - statfs workaround for 16TB limit
5109 - LOV stripe allocation improved, can stripe on subset of OSTs
5110 - LOV file size/IO offset was wrong for files > 4GB in size
5111 - object EA data was being dropped, caused files to be unreadable
5112 - memory overflow with non-LOV OST caused memory corruption
5113 - fixed regression tests to work with new config tools, obdfilter
5114 - fixed bug when directory size became larger than 1 block
5115 - fixed bug (for single client case) when PWD was deleted
5116 - invalidate local directory pages when doing intent-based ops
5117 - avoid LDLM oops when lock callback contained bad data
5119 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5121 * documentation updates
5122 * bug fixes and cleanups
5123 - configuration tools
5126 - 64-bit compile warnings
5127 - 64-bit internal statfs data
5129 * test_brw on persistent OST devices
5131 * lprocfs (disabled)
5133 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5135 * documentation updates
5136 * bug fixes and cleanups
5137 * configuration tools
5139 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5142 * many small fixes to 0_5_4
5143 * io/network handling
5144 * thinkos in MDS operations
5146 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5149 * crucial basic fixes to 0.5.3
5150 * IOR, Iozone work over Elan
5153 2002-08-07 Phil Schwan <phil@clusterfs.com>
5154 * version 0_5_3, our first alpha
5155 * we use the new Portals iovs
5156 * documentation updates
5157 * bug fixes and cleanups
5158 * small changes in the DLM wire protocol
5160 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5161 * version 0_5_1 with some initial stability,
5162 * locking on MD and file I/O.
5163 * documentation updates
5164 * several bug fixes since 0.5.0
5165 * small changes in wire protocol
5167 2002-07-18 Phil Schwan <phil@clusterfs.com>
5169 * delivered as Lustre Light Alpha
5170 * fixed a crash after handling invalid MDS requests
5171 * fixed directory pages for architectures with non-4k pages sizes
5173 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5174 * release version v0_4_4
5175 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5177 * Fixes a number of interruption problems with OST operations.
5178 * Update documentation for portals header changes
5179 * Move all wire protocol structs/defines to lustre_idl.h
5180 * Fixes symlink length bug.
5181 * Add tcpdump to repository.
5183 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5184 * release version v0_4_3
5185 * Fixes statfs for inodes on extN.
5186 * Fixes bug in runtests which would delete /etc/hosts.
5187 * Use 64-bit object IDs wherever possible (not into VFS though)
5188 Remove ost_get_info, which is unused by lustre, and out of date.
5190 2002-07-03 Peter Braam <braam@clusterfs.com>
5191 * release version v0_4_2 Fixes a lookup error (type not passed)
5192 * move forward to head of Portals
5193 * move forward to latest Lustre kernel
5195 2002-06-25 Peter Braam <braam@clusterfs.com>
5196 * release version v0_4_1. Hopefully stable on single node use.