1 tbd Sun Microsystems, Inc.
4 2.6.16.60-0.27 (SLES 10),
5 2.6.18-92.1.10.el5 (RHEL 5),
6 2.6.22.14 vanilla (kernel.org).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.40.11-sun1
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
17 Description: Prevent import from entering FULL state when server in recovery
21 Description: service mount cannot take device name with ":"
22 Details : Only when device name contains ":/" will mount treat it as
28 Description: replace ptlrpcd with the statahead thread to interpret the async
29 statahead RPC callback
33 Frequency : on recovery
34 Description: I/O failures after umount during fail back
35 Details : if client reconnected to restarted server we need join to recovery
36 instead of find server handler is changed and process self eviction
37 with cancel all locks.
39 Severity : enhancement
41 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
43 Severity : enhancement
45 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
47 Severity : enhancement
49 Description: Upcall on Lustre log has been dumped
50 Details : Allow for a user mode script to be called once a Lustre log has
51 been dumped. It passes the filename of the dumped log to the
52 script, the location of the script can be specified via
53 /proc/sys/lnet/debug_log_upcall.
58 Description: avoid messages about idr_remove called for id which is not allocated.
59 Details : Move assigment s_dev for clustered nfs to end of initialization, for
60 avoid problem with error handling.
65 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
66 Details : When connection is reused this not moved from CONN_UNUSED_HASH into
67 CONN_USED_HASH and this prodice warning when put connection again
70 Severity : enhancement
72 Description: Export bytes_read/bytes_write count on OSC/OST.
76 Description: Early reply size mismatch, MGC loses connection
77 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
78 the connect flags are properly negotiated.
82 Description: Properly propagate oinfo flags from lov to osc for statfs
83 Details : restore missing copy oi_flags to lov requests.
85 Severity : enhancement
87 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
88 : l_getgroups(8), lst(8), routerstat(8)
89 Details : included man pages for llobdstat(8), llstat(8),
90 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
92 Severity : enhancement
94 Description: configure's --enable-quota should check the
95 : kernel .config for CONFIG_QUOTA
96 Details : configure is terminated if --enable-quota is passed but
97 : no quota support is in kernel
101 Description: Remove portals compatibility
102 Details : Remove portals compatibility, not interoperable with releases
107 Description: Resolve device initialization race
108 Details : Prevent proc handler from accessing devices added to the
109 obd_devs array but yet be intialized.
111 Severity : enhancement
113 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
115 Severity : enhancement
117 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
122 Description: Using wrong pointer in osc_brw_prep_request
123 Details : Access to array[-1] can produce panic if kernel compiled with
124 CONFIG_PAGE_ALLOC enabled
128 Description: Client runs out of low memory
129 Details : Consider only lowmem when counting initial number of llap pages
133 Description: *optional* service tags registration
134 Details : if the "service tags" package is installed on a Lustre node
135 When the filesystem is mounted, a local-node service tag will
136 be created. See http://inventory.sun.com/ for more information
137 about the Service Tags asset management system.
141 Description: Kernel BUG tries to release flock
142 Details : Lustre does not destroy flock lock before last reference goes
143 away. So always drop flock locks when client is evicted and
144 perform unlock regardless of successfulness of speaking to MDS.
148 Description: add recount protection for osc callbacks, so avoid panic on shutdown
152 Description: sanity test 65a fails if stripecount of -1 is set
153 Details : handle -1 striping on filesystem in ll_dirstripe_verify
158 Description: ASSERTION(CheckWriteback(page,cmd)) failed
159 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
162 Severity : enhancement
164 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
168 Description: do not process already freed flock
169 Details : flock can possibly be freed by another thread before it reaches
170 to ldlm_flock_completion_ast.
174 Description: LBUG during stress test
175 Details : Need properly lock accesses the flock deadlock detection list.
179 Description: oops in page fault handler
180 Details : kernel page fault handler can return two special 'pages' in error case, don't
181 try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
185 Description: timeout with invalidate import.
186 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which should be
187 handled by ptlrpcd. This produce long age waiting and -ETIMEOUT
188 ptlrpc_invalidate_import and as result LASSERT.
190 Severity : enhancement
192 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
196 Description: enable MGS and MDT services start separately
197 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
198 a co-located MGS without starting the MGS, which is a complement
199 to 'nosvc' mount option.
203 Frequency : after recovery
204 Description: precreate to many object's after del orphan.
205 Details : del orphan st in oscc last_id == next_id and this triger growing
206 count of precreated objects. Set flag LOW to skip increase count
207 of precreated objects.
211 Frequency : rare, on clear nid stats
212 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
213 Details : when clean nid stats sometimes try destroy live entry,
214 and this produce panic in free.
218 Description: Stack overflow during MDS log replay
219 ease stack pressure by using a thread dealing llog_process.
223 Description: wait until IO finished before start new when do lock cancel.
224 Details : VM protocol want old IO finished before start new, in this case
225 need wait until PG_writeback is cleared until check dirty flag and
226 call writepages in lock cancel callback.
228 Severity : enhancement
230 Description: using special macro for print time and cleanup in includes.
234 Description: mds_mfd_close() ASSERTION(rc == 0)
235 Details : In mds_mfd_close(), we need protect inode's writecount change
236 within its orphan write semaphore to prevent possible races.
240 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
244 Frequency : rare, on shutdown ost
245 Description: don't hit live lock with umount ost.
246 Details : shrink_dcache_parent can be in long loop with destroy dentries,
247 use shrink_dcache_sb instead.
251 Description: don't panic with use echo client
252 Details : echo client pass NULL as client nid pointer and this produce null
257 Description: fix build on ppc32
258 Details : compile code with -m64 flag produce wrong object file for ppc32.
262 Description: add message levels for liblustreapi
266 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
267 Details : Move the 'good_osts' check before the 'total_bavail' check. This
268 will result in an -EAGAIN and in the exit call path we call
269 alloc_rr() which will with increasing aggressiveness attempt to
270 aquire precreated objects on the minimum number of required OSCs.
274 Description: Use old size assignment to avoid deadlock
275 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
276 the scheduling while holding a spinlock. We do not need locking
277 for size in ll_update_inode() because size is only updated from
278 the MDS for directories or files without objects, so there is no
279 other place to do the update, and concurrent access to such inodes
280 are protected by the inode lock.
284 Description: resolve "_IOWR redefined" build error on SLES10
288 Description: dump the memory debugging after all modules are unloaded to
289 suppress false negative in conf_sanity test 39
291 Severity : enhancement
293 Description: build kernel-ib packages for OFED 1.3 in our release cycle
298 Description: fix SLES kernel versioning
299 Details : the kernel version for our SLES 10 kernel did not include a "-"
300 before the "smp" at the end. while this was not a problem in
301 general, it did mean that software trying to use the kernel
302 version to try to detect a vendor specific kernel would fail.
303 this was most evident by the OFED build scripts.
307 Description: Don't update lov_desc members until making sure they are valid
308 Details : When updating lov_desc members via proc fs, need fix their
309 validities before doing the real update.
313 Description: don't put request into delay list while invalidate in flight.
314 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
315 import in flight. this produce timeout for invalidate and sometimes
316 can cause stale data.
320 Frequency : on ppc only
321 Description: not convert ost objects for directory because it's not exist.
322 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
326 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
328 Severity : enhancement
330 Description: Update to support 2.6.22.14 vanilla kernel.
334 Frequency : rare, on recovery
335 Description: read procfs can produce deadlock in some situation
336 Details : Holding lprocfs lock which send rpc can produce block for destroy
337 obd objects and this also block reconnect with -EALREADY. This isn't
338 fix all lprocfs bugs - but make it rare.
340 Severity : enhancement
342 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
345 Frequency : frequent on X2 node
347 Description: mdc_set_open_replay_data LBUG
348 Details : Set replay data for requests that are eligible for replay.
352 Description: lustre_mgs: operation 101 on unconnected MGS
353 Details : When MGC is disconnected from MGS long enough, MGS will evict the
354 MGC, and late on MGC cannot successfully connect to MGS and a lot
355 of the error messages complaining that MGS is not connected.
359 Frequency : on network error
360 Description: panic with double free request if network error
361 Details : mdc_finish_enqueue is finish request if any network error ocuring,
362 but it's true only for synchronus enqueue, for async enqueue
363 (via ptlrpcd) this incorrect and ptlrpcd want finish request
366 Severity : enhancement
368 Description: client-side metadata stat-ahead during readdir(directory readahead)
369 Details : perform client-side metadata stat-ahead when the client detects
370 readdir and sequential stat of dir entries therein
373 Frequency : on start mds
375 Description: Implement get_info(last_id) in obdfilter.
378 Frequency : occasional
380 Description: Correctly check stale fid, not start epoch if ost not support SOM
381 Details : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
382 overwrited with new generated fid, but mds can anwer with one of these
383 two fids and both is not stale. setattr incorectly start epoch and
384 assume will be called done_writeting, but without SOM done_writing
388 Frequency : rare, depends on device drivers and load
390 Description: MDS or OSS nodes crash due to stack overflow
391 Details : Code changes in 1.8.0 increased the stack usage of some functions.
392 In some cases, in conjunction with device drivers that use a lot
393 of stack the MDS (or possibly OSS) service threads could overflow
394 the stack. One change which was identified to consume additional
395 stack has been reworked to avoid the extra stack usage.
398 Frequency : occasional
400 Description: Do not fail import if osc_interpret_create gets -EAGAIN
401 Details : If osc_interpret_create got -EAGAIN it immediately exits and
402 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
403 oscc_has_objects and see OSC has no objests and call
404 oscc_internal_create to resend create request.
406 Severity : enhancement
408 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
410 Severity : enhancement
412 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
415 Frequency : very rare
417 Description: Fix replay if there is an un-replied request and open
418 Details : In some cases, older replay request will revert the
419 mcd->mcd_last_xid on MDS which is used to record the client's
422 Severity : enhancement
424 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
426 Severity : enhancement
428 Description: Add rhel5 support to HEAD.
430 Serverity : enhancement
432 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
437 Description: Don't allow skipping OSTs if index has been specified.
438 Details : Don't allow skipping OSTs if index has been specified, make locking
439 in internal create lots better.
443 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
444 Details : access to bitfield in structure is always rounded to long
445 and this produce problem with not atomic change any bit.
449 Description: Lustre make rpms failed.
450 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
451 builds Lustre rpms from tarball.
454 Frequency : rare on shutdown ost
456 Description: If llog cancel was not send before clean_exports phase, this can
457 produce deadlock in llog code.
458 Details : If llog thread has last reference to obd and call class_import_put
459 this produce deadlock because llog_cleanup_commit_master wait when
460 last llog_commit_thread exited, but this never success because was
461 called from llog_commit_thread.
465 Description: allow userland application know is lost one of stripes.
466 Details : fill lvb_blocks with error code on ost and return it to
467 application if error flag found.
471 Description: NULL lov_tgts causing MDS oops
472 Details : more safe checks for NULL lov_tgts for avoid oops.
474 Severity : enhancement
476 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
480 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
481 Details : If an OST is down the MDS will hang indefinitely in
482 obd_statfs_rqset() waiting for the statfs data. While for
483 MDS QOS usage of statfs, it should not stuck in waiting.
485 Severity : enhancement
487 Description: remote_acl support
488 Details : Support ACL-based permission check for remote user.
489 Support setfacl/getfacl for remote user with the utils
490 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
491 the system "{s,g}etfacl" utils.
493 Severity : enhancement
495 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
497 Severity : enhancement
499 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
503 Description: make mgs_setparam() handle fsname containing dash
504 Details : fsname containing a dash does not work with lctl conf_param
506 Severity : enhancement
508 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
510 Severity : enhancement
512 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
516 Frequency : rare, at shutdown
517 Description: access already free / zero obd_namespace.
518 Details : if client_disconnect_export was called without force flag set,
519 and exist connect request in flight, this can produce access to
520 NULL pointer (or already free pointer) when connect_interpret
521 store ocd flags in obd_namespace.
525 Frequency : only at startup
526 Description: not alloc memory with spinlock held.
527 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
528 if any spinlock held.
530 Severity : enhancement
532 Description: make lustre randomly fail allocating memory
533 Details : Make lustre randomly failed allocating memory for testing purpose.
535 Severity : enhancement
537 Description: lost problems with lov objid file
538 Details : Fixes some scability and access to not inited memory problems
539 in work with lov objdid file.
544 Description: lfs find does not continue on file error
545 Details : Continue other files processing when a file/dir is absent.
549 Description: Inconsistent usage of lustre_pack_reply()
550 Details : Standardize the usage of lustre_pack_reply() such that it
551 always generate a CERROR on failure.
554 Frequency : occasional
556 Description: MDS hang or stay in waiting lock
557 Details : If client receive lock with CBPENDING flag ldlm need send lock
558 cancel as separate rpc, to avoid situation when cancel request
559 can't processed due all i/o threads stay in wait lock.
563 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
567 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
571 Description: Update to SLES9 kernel-2.6.5-7.287.3.
575 Description: Update to RHEL4 latest kernel.
577 Severity : enhancement
579 Description: Build SLES10 patchless client fails
580 Details : The configure was broken by run ./configure with
581 --with-linux-obj=.... argument for patchless client. When the
582 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
583 can't search header adequately. Use absolute path such as
584 -I($LINUX)/include instead.
588 Description: interrupt oig_wait produce painc on resend.
589 Details : brw_redo_request can be used for resend requests from ptlrpcd and
590 private set, and this produce situation when rq_ptlrpcd_data not
591 copyed to new allocated request and triggered LBUG on assert
592 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
593 ptlrpcd set if request is changed and can be safety changed to use
598 Description: LASSERT_{REQ,REP}SWAB macros are buggy
599 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
600 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
601 these macros and replace them with their difinitions instead.
605 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
606 Details : Update which_patch & target file for SLES10 latest kernel.
609 Description: Remove client patches from SLES 10 kernel.
610 Details : This causes SLES 10 clients to behave as patchless clients
611 even on a Lustre-patched (server) kernel.
613 Severity : enhancement
615 Description: self-adjustable client's lru lists
616 Details : use adaptive algorithm for managing client cached locks lru
617 lists according to current server load, other client's work
618 pattern, memory activities, etc. Both, server and client
619 side namespaces provide number of proc tunables for controlling
622 Severity : enhancement
624 Description: light-weight GSS support
625 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
626 and reduce performance overhead.
628 Severity : enhancement
630 Description: Linux keyring support
631 Details : Support using service of Linux keyring for Lustre GSS internal
632 context refresh/cache mechanism.
636 Description: Fix errors in lfs documentation
637 Details : Fixes man pages
641 Description: don't use GFP_* in generic Lustre code.
642 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
646 Description: obdclass is limited by single OBD_ALLOC(idarray)
647 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
651 Description: warnings with build patchless client with vanila 2.6.19 and up
652 Details : change old ctl_table style and replace ctl_table/ctl_table_header
653 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
657 Frequency : Only for SLES
658 Description: sanity_quota fail test_1
659 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
660 file, which makes TSTID[2] inunique.
664 Frequency : Only for SLES9
665 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
666 Details : Update target/ChangeLog/which_patch .
670 Frequency : Only for test_10 in sanity-quota.sh
671 Description: a bug in quota test code
672 Details : checking a test flag in a wrong place
676 Frequency : Only for test_13 in sanity-quota.sh
677 Description: a bug in quota test script
678 Details : shouldn't check overfull allocated quota
682 Description: jbd statistics
683 Details : Port older jbd statistics patch for sles10
685 Severity : enhancement
687 Description: Add printing inode info into message about error in writepage.
692 Description: reply_lock_interpret crash due to race with it and lock cancel.
693 Details : Do not replay locks that are being cancelled. Do not reference
694 locks by their address during replay, just by their handle.
698 Frequency : When flocks are used.
699 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
700 Details : Flock locks might destroy just granted lock if it could be merged
701 with another existing flock, this is done in completion handler,
702 so teach ldlm_cli_enquque_fini that this is a valid case for
708 Description: Oops in read and write path when failing to allocate lock.
709 Details : Check if lock allocation failed and return error back.
713 Description: lstripe command fails for valid OST index
714 Details : The stripe offset is compared to lov->desc.ld_tgt_count
715 instead of lov->desc.ld_active_tgt_count.
719 Description: sanity.sh failed test 103
720 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
725 Description: df doesn't work properly if diskfs blocksize != 4K
726 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
730 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
731 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
733 Severity : enhancement
735 Description: Adaptive timeouts
736 Details : RPC timeouts adapt to changing server load and network
737 conditions to reduce resend attempts and improve recovery time.
741 Description: llapi_file_create() does not allow some changes
742 Details : add llapi_file_open() that allows specifying the mode and
743 open flags, and also returns an open file handle.
747 Description: merge and cleanup kernel patches.
748 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
752 Description: Add journal checksum support.(Kernel part)
753 Details : The journal checksum feature adds two new flags i.e
754 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
755 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
756 indicates that the commit block contains the checksum for
757 the blocks described by the descriptor blocks. Now commit
758 record can be sent to disk without waiting for descriptor
759 blocks to be written to disk. This behavior is controlled
760 using JBD2_FEATURE_ASYNC_COMMIT flag.
764 Description: OSS needs mutliple precreate threads
765 Details : Add ability to start more than one create thread per OSS.
769 Description: Sanity.sh test_65k failed: setstripe should have succeeded
770 Details : execute lfs setstripe on client
774 Description: mds_obd_create error creating tmp object
775 Details : When the user sets quota on root, llog will be affected and can't
776 create files and write files.
779 Frequency : Always on ia64 patchless client, and possibly others.
781 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
782 Details : This allows the patchless client to be loaded on architectures
787 Description: RedHat Update kernel for RHEL5
788 Details : Add the kernel config file for RHEL5.
792 Description: RedHat Update kernel for RHEL5
793 Details : Modify the kernel config file more closer RHEL5.
797 Description: Build failure against Centos5 (RHEL5)
798 Details : Define PAGE_SIZE when it isn't present.
802 Description: test_7 on sanity-quota.sh can't be used on separate machines
803 Details : add facet to handle it
807 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
808 Details : with NFS, the anon dentry's parent was set to itself in
809 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
810 corresponding dentry other than use rec->ur_name.
812 Severity : enhancement
814 Description: lfs setstripe enhancement
815 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
819 Description: enable data checksumming by default
820 Details : enable checksum by default, allow --disable-checksum
821 configure option and "-o nochecksum" mount option. Checksums
822 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
826 Description: lustre support for RHEL5
827 Details : Add support for RHEL5.
831 Description: Client eviction due to failover config
832 Details : after a connection loss, the lustre client should attempt to
833 reconnect to the last active server first before trying the
834 other potential connections.
838 Description: when mds and osts use different quota unit(32bit and 64bit),
839 quota will be released repeatly.
840 Details : void sending multiple quota reqs to mds, which will keep the
841 status between the reqs.
845 Description: osts not allocated evenly to files
846 Details : change the condition to increase offset_idx
849 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
851 Description: Data corruption for OSTs that are formatted larger than 2TB
853 Details : When generating the bio request for lustre file writes the
854 sector number would overflow a temporary variable before being
855 used for the IO. The data reads correctly from Lustre (which
856 will overflow in a similar manner) but other file data or
857 filesystem metadata may be corrupted in some cases.
861 Description: Tune the kernel for good SCSI performance.
862 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
863 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
868 Description: rewrite ext2-derived code in obdclass/uuid.c
869 Details : rewrite inherited code (uuid parsing code from ext2 utils)
870 from scratch preserving functionality.
874 Description: "lfs find -obd UUID" prints directories
875 Details : "lfs find -obd UUID" will return all directory names instead
876 of just file names. It is incorrect because the directories
877 do not reside on the OSTs.
881 Description: use i_size_read and i_size_write in 2.6 port
882 Details : replace inode->i_size access with i_size_read/write()
885 Frequency : when removing large files
887 Description: scheduling issue during removal of large Lustre files
888 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
889 It causes scheduling issues when removing large files (17TB in the
893 Frequency : only with liblustre clients on XT3
895 Description: evictions taking too long
896 Details : allow llrd to evict clients directly on OSTs
899 Frequency : only on ppc
901 Description: /proc/fs/lustre/devices broken on ppc
902 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
903 We should make sure the type of 'index' is loff_t and then cast
904 explicitly as needed below. Do not assign an explicitly cast
908 Frequency : only for rhel5
910 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
911 Details : Modify the target file & which_kernel.
913 Severity : enhancement
915 Description: omit set fsid for export NFS
916 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
917 as FS_REQUIRES_DEV for avoid problems with generate fsid.
921 Frequency : Always, for kernels after 2.6.16
922 Description: Fix warning idr_remove called for id=.. which is not allocated.
923 Details : Last kernels save old s_dev before kill super and not allow
924 to restore from callback - restore it before call kill_anon_super.
928 Description: buffer overruns could theoretically occur
929 Details : llapi_semantic_traverse() modifies the "path" argument by
930 appending values to the end of the origin string, and a
931 overrun may occur. Adding buffer overrun check in liblustreapi.
935 Description: Fix error on 'ls .' at the top of the Lustre mount.
936 Details : Don't revalidate dentry if it is a root dentry.
940 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
941 Details : Modify vm-tunables-rhel4.patch.
945 Description: Kernel config for 2.6.18-vanilla.
946 Details : Modify targets/2.6-vanilla.target.in.
947 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
948 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
949 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
950 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
954 Description: improve handling recoverable errors
955 Details : if request processig with error which can be recoverable on server
956 request should be resend, otherwise page released from cache and
961 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
962 Details : Modify target file & which_patch.
963 A flaw was found in the IA32 system call emulation provided
964 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
965 value could be stored in the %RAX register, which could trigger an
966 out-of-bounds system call table access. An untrusted local user
967 could exploit this flaw to run code in the kernel
968 (ie a root privilege escalation). (CVE-2007-4573).
972 Description: O_DIRECT bypasses client statistics.
973 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
974 read_ahead_stats were updated. copied the stats section from
975 osc_send_oap_rpc() into async_internal().
979 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
983 Description: change order of libsysio includes
984 Details : '#include sysio.h' should always come before '#include xtio.h'
988 Description: handle "serious error: objid * already exists" more gracefully
989 Details : If LAST_ID value on disk is smaller than the objects existing in
990 the O/0/d* directories, it indicates disk corruption and causes an
991 LBUG(). If the object is 0-length, then we should use the existing
992 object. This will help to avoid a full fsck in most cases.
994 Severity : enhancement
996 Description: adapt the lustre_config script to support the upgrade case
997 Details : Add "-u" option for lustre_config script to support upgrading 1.4
998 server targets to 1.6 in parallel.
1002 Description: To avoid grant space > avaible space when the disk is almost
1003 full. Without this patch you might see the error "grant XXXX >
1004 available" or some LBUG about grant, when the disk is almost
1006 Details : In filter_check_grant, for non_grant cache write, we should
1007 check the left space by if (*left > ungranted + bytes), instead
1008 of (*left > ungranted), because only we are sure the left space
1009 is enough for another "bytes", then the ungrant space should be
1010 increase. In client, we should update cl_avail_grant only there
1011 is OBD_MD_FLGRANT in the reply.
1015 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1016 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1017 described in CVE-2007-4573. This problem could allow untrusted
1018 local users to gain root access.
1021 Frequency : when using O_DIRECT and quotas
1023 Description: Incorrect file ownership on O_DIRECT output files
1024 Details : block usage reported by 'lfs quota' does not take into account
1025 files that have been written with O_DIRECT.
1030 Description: touch file failed when fs is not full
1031 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1032 otherwise we can get ENOSP while fs is not full.
1036 Description: parallel lock callbacks
1037 Details : Instead of sending blocking and completion callbacks as separated
1038 requests, adding them to a set and sending in parallel.
1041 Frequency : only for Cray XT3
1042 Bugzilla : 12829/13455
1043 Description: Changing primary group doesn't change the group lustre assigns to
1045 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1046 overridden with the primary group provided by the group upcall,
1047 whereas the supplied fsgid can be trusted if it is in the list of
1048 supplementary groups returned by the group upcall.
1050 Severity : enhancement
1052 Description: Allow masking D_WARNING, D_ERROR messages from console
1053 Details : Console messages can now be disabled via lnet.printk.
1057 Description: User code with malformed file open parameter crashes client node
1058 Details : Before packing join_file req, all the related reference should be
1059 checked carefully in case some malformed flags cause fake
1060 join_file req on client.
1064 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1065 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1066 mode is equal to the granted mode. However, it should also check
1067 that LDLM_FL_CP_REQD is not set.
1071 Description: Heavy nfs access might result in deadlocks
1072 Details : After ELC code landed, it is now improper to enqueue any mds
1073 locks under och_sem, because enqueue might want to decide to
1074 cancel open locks for same inode we are holding och_sem for.
1078 Description: Client eviction while running blogbench
1079 Details : A lot of unlink operations with concurrent I/O can lead to a
1080 deadlock causing evictions. To address the problem, the number of
1081 oustanding OST_DESTROY requests is now throttled to
1082 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1083 callbacks are processed in priority.
1087 Description: enable ACLs on MDS by default
1088 Details : ACLs must be enabled on MDS by default.
1091 Frequency : PPC/PPC64 only
1093 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1094 Details : fix duplicated definitions between asm-ppc64/types.h and
1095 lustre_types.h on PPC.
1098 Frequency : PPC/PPC64 only
1100 Description: asm-ppc/segment.h does not exist
1101 Details : fix compile issue on PPC.
1105 Description: better handle error messages in extents code
1108 Frequency : RHEL4 only
1110 Description: mkfs is very slow on IA64/RHEL4
1111 Details : A performance regression has been discovered in the MPT Fusion
1112 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1113 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1114 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1116 Severity : enhancement
1118 Description: SNMP support enhancement
1119 Details : Adding total number of sampled request for an MDS node in snmp
1122 Severity : enhancement
1124 Description: Optimize ldlm waiting list processing for PR extent locks
1125 Details : When processing waiting list for read extent lock and meeting read
1126 lock that is same or wider to it that is not contended, skip
1127 processing rest of the list and immediatelly return current
1128 status of conflictness, since we are guaranteed there are no
1129 conflicting locks in the rest of the list.
1133 Description: Time out and refuse to reconnect
1134 Details : When the failover node is the primary node, it is possible
1135 to have two identical connections in imp_conn_list. We must
1136 compare not conn's pointers but NIDs, otherwise we can defeat
1137 connection throttling.
1141 Description: port llog fixes from b1_6 into HEAD
1142 Details : Port llog reference couting and some llog cleanups from b1_6
1143 (bug 10800) into HEAD, for protect from panic and access to already
1144 free llog structures.
1148 Description: Detect stride IO mode in read-ahead
1149 Details : When a client does stride read, read-ahead should detect that and
1150 read-ahead pages according to the detected stride pattern.
1154 Description: data checksumming impacts single node performance
1155 Details : add support for several checksum algorithm. Currently, only CRC32
1156 and Adler-32 are supported. The checksum type can be changed on
1157 the fly via /proc/fs/lustre/osc/*/checksum_type.
1161 Description: use adler32 for page checksums
1162 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1167 Description: build for x2 fails
1168 Details : fix compile issue on Cray systems.
1172 Description: Properly match for duplicate locks
1173 Details : Due to different lock order from skiplists code, we need to
1174 traverse entire list for now
1177 Frequency : only on PPC/SLES10
1179 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1180 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1181 be defined. Add a hack in mkfs_lustre.c to work around this compile
1186 Description: LASSERT on MDS when client holding flock lock dies
1187 Details : ldlm pool logic depends on number of granted locks equal to
1188 number of released locks which is not true for flock locks, so
1189 just exclude such locks from consideration.
1193 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1194 Details : Use fsfilt_sync() for both the whole filesystem sync and
1195 individual file sync to eliminate dangerous inode locking
1196 with I_LOCK that can lead to a deadlock.
1200 Description: performance in 1.6.3
1201 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1202 sure that 1MB requests are not fragmented by the block layer.
1204 Severity : enhancement
1206 Description: organize the server-side client stats on per-nid basis
1207 Details : Change the structure of stats under obdfilter and mds to
1215 The "uuid"s file would list the uuids of _active_ exports.
1216 And the clear entry is to clear all stats and stale nids.
1218 Severity : enhancement
1220 Description: eliminate client locks in face of contention
1221 Details : file contention detection and lockless i/o implementation
1222 for contended files.
1226 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1229 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1231 Description: MDS may lose file striping (and hence file data) in some cases
1232 Details : If there are additional extended attributes stored on the MDS,
1233 in particular ACLs, SELinux, or user attributes (if user_xattr
1234 is specified for the client mount options) then there is a risk
1235 of attribute loss. Additionally, the Lustre file striping
1236 needs to be larger than default (e.g. striped over all OSTs),
1237 and an additional attribute must be stored initially in the
1238 inode and then increase in size enough to be moved to the
1239 external attribute block (e.g. ACL growing in size) for file
1244 Description: skiplist implementation simplification
1245 Details : skiplists are used to group compatible locks on granted list
1246 that was implemented as tracking first and last lock of each lock group
1247 the patch changes that to using doubly linked lists
1251 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1252 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1253 because LASSERT on both the data supplied by a client, and the data
1254 on disk is dangerous and incorrect.
1256 Severity : enhancement
1258 Description: Slow truncate/writes to huge files at high offsets.
1259 Details : Directly associate cached pages to lock that protect those pages,
1260 this allows us to quickly find what pages to write and remove
1261 once lock callback is received.
1265 Description: more ldlm soft lockups
1266 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1267 starve other threads from the resource lock for a long time in
1268 case of long waiting queue, so change the debug level from
1269 D_OTHER to the less frequently used D_INFO.
1271 Severity : enhancement
1273 Description: add -gid, -group, -uid, -user options to lfs find
1277 Description: Hung threads in invalidate_inode_pages2_range
1278 Details : The direct IO path doesn't call check_rpcs to submit a new RPC once
1279 one is completed. As a result, some RPCs are stuck in the queue
1284 Description: filter threads hungs on waiting journal commit
1285 Details : Cleanup filter group llog code, then only filter group llog will
1286 be only created in the MDS/OST syncing process.
1290 Description: Procfs and llog threads access destoryed import sometimes.
1291 Details : Sync the import destoryed process with procfs and llog threads by
1292 the import refcount and semaphore.
1294 Severity : enhancement
1296 Description: openlock cache of b1_6 port to HEAD
1301 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1302 Details : If insertion of an extent fails, then discard the inode
1303 preallocation and free data blocks else it can lead to duplicate
1308 Description: don't always update ctime in ext3_xattr_set_handle()
1309 Details : Current xattr code updates the inode ctime in ext3_xattr_set_handle.
1310 In some cases the ctime should not be updated, for example for
1311 2.0->1.8 compatibility it is necessary to delete an xattr and it
1312 should not update the ctime.
1316 Bugzilla : 15713/16362
1317 Description: Assertion in iopen_connect_dentry in 1.6.3
1318 Details : looking up an inode via iopen with the wrong generation number can
1319 populate the dcache with a disconneced dentry while the inode
1320 number is in the process of being reallocated. This causes an
1321 assertion failure in iopen since the inode's dentry list contains
1322 both a connected and disconnected dentry.
1326 Description: assertion failure in ldlm_handle2lock()
1327 Details : fix a race between class_handle_unhash() and class_handle2object()
1328 introduced in lustre 1.6.5 by bug 13622.
1333 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1334 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1339 Description: oops in sd_iostats_seq_show()
1340 Details : unloading/reloading the scsi low level driver triggers a kernel
1341 bug when trying to access the sd iostat file.
1346 Description: Kernel panics during QLogic driver reload
1347 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1348 patch, causing memory corruption.
1353 Description: journal_dev option does not work in b1_6
1354 Details : pass mount option during pre-mount.
1356 Severity : enhancement
1358 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1359 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1360 extent information of a file. It can be used to map logical blocks
1361 in a file to physical blocks in the block device.
1365 Description: LDLM soft lockups - improvement
1366 Details : It is be possible to send the lock handle along with each read
1367 or write request because the client is already doing a lock match
1368 itself so there isn't any reason the OST should have to re-do that
1371 --------------------------------------------------------------------------------
1373 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
1375 * Support for kernels:
1376 2.6.5-7.283 (SLES 9),
1377 2.6.9-55.EL (RHEL 4),
1378 2.6.16.46-0.14 (SLES 10),
1379 2.6.18.8 vanilla (kernel.org)
1380 * Client support for unpatched kernels:
1381 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1382 2.6.16 - 2.6.22 vanilla (kernel.org)
1383 * Due to recently discovered recovery problems, we do not recommend
1384 using patchless RHEL 4 clients with this or any earlier release.
1385 * Recommended e2fsprogs version: 1.39.cfs8
1386 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1387 * Starting with this release, the ldiskfs backing filesystem required
1388 by Lustre is now in its own package, lustre-ldiskfs. This package
1389 should be installed. It is versioned separately from Lustre and
1390 may be released separately in future.
1394 Description: block reactivating mgc import until all deactivates complete
1395 Details : Fix race when failing back MDT/MGS to itself (testing)
1398 Frequency : at statup only
1400 Description: mds_lov_synchronize race leads to various problems
1401 Details : simultaneous MDT->OST connections at startup can cause the
1402 sync to abort, leaving the OSC in a bad state.
1404 Severity : enhancement
1406 Description: add optional extra BUILD_VERSION info
1407 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1408 to override the lustre version.
1411 Frequency : 2.6.18 servers only
1413 Description: ll_kern_mount() doesn't release the module reference
1414 Details : The ldiskfs module reference count never drops down to 0
1415 because ll_kern_mount() doesn't release the module reference.
1420 Description: server LBUG when using old ost_num_threads parameter
1421 Details : Accept the old ost_num_threads parameter but warn that it
1422 is deprecated, and fix an off-by-one error that caused an LBUG.
1427 Description: Transient SCSI error results in persistent IO issue
1428 Details : iobuf->dr_error is not reinitialized to 0 between two
1432 Frequency : sometimes when underlying device returns I/O errors
1434 Description: OSTs not going read-only during write failures
1435 Details : OSTs are not remounted read-only when the journal commit threads
1436 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1437 instead of the ext3 wrappers.
1441 Description: poor connect scaling with increasing client count
1442 Details : Don't run filter_grant_sanity_check for more than 100 exports
1443 to improve scaling for large numbers of clients.
1446 Frequency : SLES10 only
1448 Description: sanity-quota.sh quotacheck failed: rc = -22
1449 Details : Quotas cannot be enabled on SLES10.
1452 Frequency : liblustre clients only
1454 Description: getdirentries does not give error when run on compute nodes
1455 Details : getdirentries does not fail when the size specified as an argument
1456 is too small to contain at least one entry
1458 Severity : enhancement
1460 Description: Add LNET router traceability for debug purposes
1461 Details : If a checksum failure occurs with a router as part of the
1462 IO path, the NID of the last router that forwarded the bulk data
1463 is printed so it can be identified.
1468 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1469 Details : Due to a race condition, liblustre clients were occasionally
1470 evicted incorrectly.
1472 Severity : enhancement
1474 Description: lfs setstripe use optional parameters instead of postional
1477 Severity : enhancement
1479 Description: Nanosecond timestamp support for ldiskfs
1480 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1481 resolution timestamps. There is not yet support for this at
1482 the Lustre filesystem level.
1485 Frequency : during server recovery
1487 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1488 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1492 Frequency : nfs export on patchless client
1494 Description: connectathon hang when test nfs export over patchless client
1495 Details : Disconnected dentry cannot be found with lookup, so we do not need
1496 to unhash it or make it invalid
1499 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1500 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1501 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1506 Description: open req refcounting wrong on reconnect
1507 Details : If reconnect happened between getting open reply from server and
1508 call to mdc_set_replay_data in ll_file_open, we will schedule
1509 replay for unreferenced request that we are about to free.
1510 Subsequent close will crash in variety of ways.
1511 Check that request is still eligible for replay in
1512 mdc_set_replay_data().
1517 Description: disable writes to filesystem when reading health_check file
1518 Details : the default for reading the health_check proc file has changed
1519 to NOT do a journal transaction and write to disk, because this
1520 can cause reads of the /proc file to hang and block HA state
1521 checking on a healthy but otherwise heavily loaded system. It
1522 is possible to return to the previous behaviour during configure
1523 with --enable-health-write.
1525 Severity : enhancement
1527 Description: 64-bit inode version
1528 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1529 to the inode. This will be required for version-based recovery.
1534 Description: MDS fails to start if a duplicate client export is detected
1535 Details : in some rare cases it was possible for a client to connect to
1536 an MDS multiple times. Upon recovery the MDS would detect this
1537 and fail during startup. Handle this more gracefully.
1539 Severity : enhancement
1541 Description: Add -o localflock option to simulate old noflock
1543 Details : This will achieve local-only flock/fcntl locks
1549 Description: log_commit_thread vs filter_destroy race leads to crash
1550 Details : Take import reference before releasing llog record semaphore
1555 Description: Wrong request locking in request set processing
1556 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1557 imp_delayed_list, in this place should be used imp_lock.
1560 Frequency : when reconnection
1562 Description: Grant Leak when osc reconnect to OST
1563 Details : When osc reconnect ost, OST(filter) should check whether it
1564 should grant more space to client by comparing fed_grant and
1565 cl_avail_grant, and return the granted space to client instead
1566 of "new granted" space, because client will call osc_init_grant
1567 to update the client grant space info.
1570 Frequency : when client reconnect to OST
1572 Description: Grant Leak when osc do resend and replay bulk write
1573 Details : When osc reconnect to OST, OST(filter)should clear grant info of
1574 bulk write request, because the grant info while be sync between
1575 OSC and OST when reconnect, and we should ignore the grant info
1576 these of resend/replay write req.
1581 Description: Grant space more than avaiable left space sometimes.
1582 Details : When then OST is about to be full, if two bulk writing from
1583 different clients came to OST. Accord the avaliable space of the
1584 OST, the first req should be permitted, and the second one
1585 should be denied by ENOSPC. But if the seconde arrived before
1586 the first one is commited. The OST might wrongly permit second
1587 writing, which will cause grant space > avaiable space.
1590 Frequency : when client is evicted
1592 Description: Grant might be wrongly erased when osc is evicted by OST
1593 Details : when the import is evicted by server, it will fork another
1594 thread ptlrpc_invalidate_import_thread to invalidate the
1595 import, where the grant will be set to 0. While the original
1596 thread will update the grant it got when connecting. So if
1597 the former happened latter, the grant will be wrongly errased
1598 because of this race.
1603 Description: Checking Stale with correct fid
1604 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1605 to check whether it is stale, because sometimes, we want the
1606 enqueue happened anyway, and op_data.fid2 will not be initialized.
1608 Severity : enhancement
1610 Description: update patchless client
1611 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1614 Frequency : only with 2.4 kernel
1616 Description: random memory corruption
1617 Details : size of struct ll_inode_info is to big for union inode.u and this
1618 can be cause of random memory corruption.
1623 Description: Memory leak in recovery
1624 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1625 It should also check obd_fail before fsfilt_start, otherwise if
1626 fsfilt_start return -EROFS,(failover mds during mds recovery).
1627 then the req will return with repmsg->transno = 0 and rc = EROFS.
1628 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1629 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1630 be freed no matter whether fsfilt_commit success or not.
1633 Frequency : only with huge count clients
1635 Description: Prevents from taking the superblock lock in llap_from_page for
1637 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1638 for llap_from_page call in ll_removepage prevents from taking the
1639 superblock lock for a soon died page.
1644 Description: Not check open intent error before release open handle
1645 Details : in some rare cases, the open intent error is not checked before
1646 release open handle, which may cause
1647 ASSERTION(open_req->rq_transno != 0), because it tries to release
1648 the failed open handle.
1653 Description: Set cat log bitmap only after create log success.
1654 Details : in some rare cases, the cat log bitmap is set too early. and it
1655 should be set only after create log success.
1659 Description: Accessing a block bevice can re-enable I/O when Lustre is
1660 tearing down a device.
1661 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1666 Description: service threads may hog cpus when there are a lot of requests
1667 Details : Insert cond_resched to give other threads a chance to use some CPU
1672 Description: the cat log was not initialized in recovery
1673 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1674 unlink log on mds will not be initialized until mds post
1675 recovery. And also in mds post recovery, the unlink log will
1676 initialization will be done asynchronausly, so there will be race
1677 between add unlink log and unlink log initialization.
1681 Description: brw_stats were being printed incorrectly
1682 Details : brw_stats were being printed as log2 but all of them were not
1683 recorded as log2. Also remove some code duplication arising from
1684 filter_tally_{read,write}.
1688 Frequency : rare, only in recovery.
1689 Description: ASSERTION(req->rq_type != LI_POISON) failed
1690 Details : imp_lock should be held while iterating over imp_sending_list for
1691 prevent destroy request after get timeout in ptlrpc_queue_wait.
1695 Description: replay-single.sh test 52 fails
1696 Details : A lock's skiplist need to be cleanup when it being unlinked
1697 from its resource list.
1701 Description: Short directio read returns full requested size rather than
1703 Details : Direct I/O operations should return actual amount of bytes
1704 transferred rather than requested size.
1706 Severity : enhancement
1708 Description: metadata RPC reduction (e.g. for rm performance)
1709 Details : decrease the amount of synchronous RPC between clients and servers
1710 by canceling conflicing lock before the operation on the client
1711 and packing thier handles into the main operation RPC to server.
1713 Severity : enhancement
1715 Description: Async OSC create to avoid the blocking unnecessarily.
1716 Details : If a OST has no remain object, system will block on the creating
1717 when need to create a new object on this OST. Now, ways use
1718 pre-created objects when available, instead of blocking on an
1719 empty osc while others are not empty. If we must block, we block
1720 for the shortest possible period of time.
1722 Severity : enhancement
1724 Description: refine locking for avoid write wrong info into lov_objid file
1725 Details : fix possible races with add new target and write/update data in
1728 --------------------------------------------------------------------------------
1730 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1735 Frequency : on some architectures
1737 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1738 Details : Uninitialized flags sometimes cause configuration commands to
1742 Frequency : patchless clients only
1744 Description: missing __iget() symbol export
1745 Details : The __iget() symbol export is missing. To avoid the need for
1746 this on patchless clients the deathrow inode reaper is turned
1747 off, and we depend on the VM to clean up old inodes. This
1748 dependency was during via the fix for bug 12181.
1750 --------------------------------------------------------------------------------
1752 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1754 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1755 INTEROPERATE with older versions automatically. In many cases a
1756 special upgrade step is needed. Please read the
1757 user documentation before upgrading any part of a live system.
1758 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1759 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1760 * WARNING: Lustre configuration and startup changes are required with
1761 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1767 Severity : enhancement
1769 Description: MountConf
1770 Details : Lustre configuration is now managed via mkfs and mount
1771 commands instead of lmc and lconf. New obd types (MGS, MGC)
1772 are added for dynamic configuration management. See
1773 https://mail.clusterfs.com/wikis/lustre/MountConf for
1776 Severity : enhancement
1778 Description: dynamic OST addition
1779 Details : OSTs can now be added to a live filesystem
1781 Severity : enhancement
1783 Description: startup order invariance
1784 Details : MDTs and OSTs can be started in any order. Clients only
1785 require the MDT to complete startup.
1787 Severity : enhancement
1789 Description: parallel, asynchronous orphan cleanup
1790 Details : orphan cleanup is now performed in separate threads for each
1791 OST, allowing parallel non-blocking operation.
1793 Severity : enhancement
1795 Description: optimized stripe assignment
1796 Details : stripe assignments are now made based on ost space available,
1797 ost previous usage, and OSS previous usage, in order to try
1798 to optimize storage space and networking resources.
1800 Severity : enhancement
1802 Description: Permanently set tunables
1803 Details : All writable /proc/fs/lustre tunables can now be permanently
1804 set on a per-server basis, at mkfs time or on a live system.
1806 Severity : enhancement
1808 Description: Lustre message v2
1809 Details : Add lustre message format v2.
1811 Severity : enhancement
1813 Description: client OST exclusion list
1814 Details : Clients can be started with a list of OSTs that should be
1815 declared "inactive" for known non-responsive OSTs.
1819 Description: SPEC SFS validation failure on NFS v2 over lustre.
1820 Details : Changes the blocksize for regular files to be 2x RPC size,
1821 and not depend on stripe size.
1823 Severity : enhancement
1825 Description: Multiple MD RPCs in flight.
1826 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1827 To avoid overly-overloading MDS, introduce a limit on number of
1828 MDS RPCs in flight for a single client and add /proc controls
1829 to adjust this limit.
1831 Severity : enhancement
1833 Description: client read/write statistics
1834 Details : Add client read/write call usage stats for performance
1835 analysis of user processes.
1836 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1837 file access. extents_stats shows chunk size distribution.
1838 extents_stats_per_process show chunk size distribution per
1841 Severity : enhancement
1843 Description: mds statistics
1844 Details : Add detailed mds operations statistics in
1845 /proc/fs/lustre/mds/*/stats.
1849 Description: Failure of copying files with lustre special EAs.
1850 Details : Client side always return success for setxattr call for lustre
1851 special xattr (currently only "trusted.lov").
1856 Description: Refcount LNET uuids
1857 Details : The global LNET uuid list grew linearly with every startup;
1858 refcount repeated list entries instead of always adding to
1861 Severity : enhancement
1863 Description: Dynamic service threads
1864 Details : Within a small range, start extra service threads
1865 automatically when the request queue builds up.
1868 Frequency : mixed-endian client/server environments
1870 Description: mixed-endian crashes
1871 Details : The new msg_v2 system had some failures in mixed-endian
1874 Severity : enhancement
1876 Description: Easy OST removal
1877 Details : OSTs can be permanently deactivated with e.g. 'lctl
1878 conf_param lustre-OST0001.osc.active=0'
1880 Severity : enhancement
1882 Description: MGS proc entries
1883 Details : Added basic proc entries for the MGS showing what filesystems
1886 Severity : enhancement
1888 Description: provide MGS failover
1889 Details : Added config lock reacquisition after MGS server failover.
1891 Severity : enhancement
1893 Description: add Linux 2.4 support
1894 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1898 Description: a large application tries to do I/O to the same resource and dies
1899 in the middle of it.
1900 Details : Check the req->rq_arrival time after the call to
1901 ost_brw_lock_get(), but before we do anything about
1902 processing it & sending the BULK transfer request. This
1903 should help move old stale pending locks off the queue as
1904 quickly as obd_timeout.
1907 Frequency : when an incorrect nid is specified during startup
1909 Description: ptlrpc connect to non-existant node causes kernel crash
1910 Details : LNET can't be re-entered from an event callback, which
1911 happened when we expire a message after the export has been
1912 cleaned up. Instead, hand the zombie cleanup off to another
1915 Severity : enhancement
1917 Description: plain/inodebits lock performance improvement
1918 Details : Grouping plain/inodebits in granted list by their request modes
1919 and bits policy, thus improving the performance of search through
1923 Frequency : only if OST filesystem is corrupted
1925 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1926 Details : for a short time RPCs with bulk IO are in the replay list,
1927 but replay of bulk IOs is unimplemented. If the OST filesystem
1928 is corrupted due to disk cache incoherency and then replay is
1929 started it is possible to trip an assertion. Avoid putting
1930 committed RPCs into the replay list at all to avoid this issue.
1933 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1935 Description: System hang on startup
1936 Details : This bug allowed the liblustre (e.g. catamount) client to
1937 return to the app before handling all startup RPCs. This
1938 could leave the node unresponsive to lustre network traffic
1939 and manifested as a server ptllnd timeout.
1941 Severity : enhancement
1943 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1944 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1945 boolean to control whether to print peer debug info when a
1946 client's RPC times out.
1949 Frequency : only for kernels with patches from Lustre below 1.4.3
1951 Description: Remove old rdonly API
1952 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1955 Frequency : only for devices with external journals
1957 Description: Set external device read-only also
1958 Details : During a commanded failover stop, we set the disk device
1959 read-only while the server shuts down. We now also set any
1960 external journal device read-only at the same time.
1963 Frequency : when upgrading from 1.4 while trying to change parameters
1965 Description: The wrong (new) MDC name was used when setting parameters for
1966 upgraded MDT's. Also allows changing of OSC (and MDC)
1967 parameters if --writeconf is specified at tunefs upgrade time.
1970 Frequency : when setting specific ost indicies
1972 Description: QOS code breaks on skipped indicies
1973 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1974 created with --index need not be sequential.
1976 Severity : enhancement
1978 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1979 Details : The uninit_groups feature works in conjunction with the kernel
1980 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1981 up the pass1 processing of e2fsck. This is a read-only feature
1982 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1983 mount filesystems that have had this feature enabled.
1985 Severity : enhancement
1987 Description: Improve multi-block allocation algorithm to avoid fragmentation
1988 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1989 allocation locality and avoid filesystem fragmentation.
1991 ------------------------------------------------------------------------------
1993 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1995 * Support for kernels:
1996 2.4.21-47.0.1.EL (RHEL 3)
1997 2.6.5-7.283 (SLES 9)
1998 2.6.9-42.0.10.EL (RHEL 4)
1999 2.6.12.6 vanilla (kernel.org)
2000 2.6.16.27-0.9 (SLES 10)
2001 * Recommended e2fsprogs version: 1.39.cfs5
2003 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2007 Frequency : occasional, depends on client load and configuration
2008 Bugzilla : 12181, 12203
2009 Description: data loss for recently-modified files
2011 Details : In some cases it is possible that recently written or created
2012 files may not be written to disk in a timely manner (this should
2013 normally be within 30s unless client IO load is very high).
2014 The problem appears as zero-length files or files that are a
2015 multiple of 1MB in size after a client crash or client eviction
2016 that are missing data at the end of the file.
2018 This problem is more likely to be hit on clients where files are
2019 repeatedly created and unlinked in the same directory, clients
2020 have a large amount of RAM, have many CPUs, the filesystem has
2021 many OSTs, the clients are rebooted frequently, and/or the files
2022 are not accessed by other nodes after being written.
2024 The presence of the problem can be detected by looking at
2025 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2026 smaller than the second (nr_unused) then dirty files will not
2027 be flushed automatically to disk. "sync; sleep 10" should be
2028 run several times on the node before unmounting it to update
2029 Lustre (this is also safe to run on nodes without this problem).
2031 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2032 that can cause this same problem, so customers using that kernel
2033 also need to update the kernel in addition to Lustre. In order
2034 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2036 It is normal that files written just before a client crash (less
2037 than 30s) may not yet have been flushed to disk, even for local
2041 Frequency : frequent on thin XT3 nodes
2043 Description: UUID collision on thin XT3 Linux nodes
2044 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2045 randomly, since we relied on an insufficiently-seeded PRNG.
2050 Description: OSS hangs after "All ost request buffers busy"
2051 Details : A deadlock between quota and journal operations caused OSS
2052 hangs after printing "All ost request buffers busy."
2055 Frequency : always on liblustre builds
2057 Description: Cleanup compiler warnings on liblustre
2060 Frequency : always on liblustre builds on XT3
2062 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2066 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2067 > 32000 subdirectories
2068 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2069 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2070 subdirectory count crosses 32000. This will aid e2fsck to
2071 correctly handle more than 32000 subdirectories.
2074 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2076 Description: System hang on startup
2077 Details : This bug allowed the liblustre (e.g. catamount) client to
2078 return to the app before handling all startup RPCs. This
2079 could leave the node unresponsive to lustre network traffic
2080 and manifested as a server ptllnd timeout.
2082 Severity : enhancement
2084 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2085 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2086 boolean to control whether to print peer debug info when a
2087 client's RPC times out.
2092 Description: make O_SYNC working on 2.6 kernels
2093 Details : 2.6 kernels use different method for mark pages for write,
2094 so need add a code to lustre for O_SYNC work.
2099 Description: Failure to close file and release space on NFS
2100 Details : Put inode details into lock acquired in ll_intent_file_open.
2101 Use mdc_intent_lock in ll_intent_open to properly
2102 detect all kind of errors unhandled by mdc_enqueue.
2107 Description: proc file read during shutdown sometimes raced obd removal,
2109 Details : Add lock to prevent obd access after proc file removal.
2112 Frequency : Only for files larger than 4GB on 32-bit clients.
2114 Description: improperly doing page alignment of locks
2115 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2116 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2121 Description: RPCs being resent when they shouldn't be
2122 Details : Some RPCs that should not be resent are being resent. This
2123 can cause inconsistencies in the RPC state machine. Do not
2124 resend such requests.
2127 Frequency : rare, only with NFS export
2129 Description: Crash on NFS re-export node
2130 Details : under very unusual load conditions an assertion is hit in
2131 ll_intent_file_open()
2134 Frequency : only if OST filesystem is corrupted
2136 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2137 Details : for a short time RPCs with bulk IO are in the replay list,
2138 but replay of bulk IOs is unimplemented. If the OST filesystem
2139 is corrupted due to disk cache incoherency and then replay is
2140 started it is possible to trip an assertion. Avoid putting
2141 committed RPCs into the replay list at all to avoid this issue.
2146 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2147 Details : Large single O_DIRECT read and write calls can fail to allocate
2148 a sufficiently large buffer to process the request. In case of
2149 allocation failure the allocation is retried with a smaller
2150 buffer and broken into smaller requests.
2152 Severity : enhancement
2154 Description: Add -o localflock option to simulate old noflock behaviour.
2155 Details : This will achieve local-only flock/fcntl locks coherentness.
2160 Description: versioning check is incomplete
2161 Details : Checking the version difference of client vs. server, report
2162 error if the gap is too big.
2167 Description: add support PG_writeback bit
2168 Details : add support for PG_writeback bit for Lustre, for more carefull
2169 work with page cache in 2.6 kernel. This also fix some deadlocks
2170 and remove hack for work O_SYNC with 2.6 kernel.
2172 Severity : enhancement
2174 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2175 Details : The uninit_groups feature works in conjunction with the kernel
2176 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2177 up the pass1 processing of e2fsck. This is a read-only feature
2178 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2179 mount filesystems that have had this feature enabled.
2181 Severity : enhancement
2183 Description: Improve multi-block allocation algorithm to avoid fragmentation
2184 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2185 allocation locality and avoid filesystem fragmentation.
2187 ------------------------------------------------------------------------------
2189 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2191 * Support for kernels:
2192 2.6.9-42.0.3.EL (RHEL 4)
2193 2.6.5-7.276 (SLES 9)
2194 2.4.21-47.0.1.EL (RHEL 3)
2195 2.6.12.6 vanilla (kernel.org)
2196 2.6.16.21-0.8 (SLES10)
2197 * Recommended e2fsprogs version: 1.39.cfs2-0
2199 * The backwards-compatible /proc/sys/portals symlink has been removed
2200 in this release. Before upgrading, please ensure that you change
2201 any configuration scripts or /etc/sysctl.conf files that access
2202 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2203 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2204 in advance of the upgrade on any system running Lustre 1.4.6 or
2205 newer, since /proc/sys/lnet was added in that version.
2206 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2210 Frequency : only when quota is used
2212 Description: avoid scanning export list for quota master
2213 Details : Change the algorithms to avoid scanning export list in order
2214 to improve the efficiency.
2217 Frequency : MDS failover only, very rarely
2219 Description: "went back in time" messages on mds failover
2220 Details : The greatest transno may be lost when the current operation
2221 finishes with an error (transno==0) and the client's last_rcvd
2222 record is over-written. Save the greatest transno in the
2223 mds_last_transno for this case.
2226 Frequency : always for specific kernels and striping counts
2228 Description: client may get "Matching packet too big" without ACL support
2229 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2230 when trying to access files in certain configurations. The
2231 clients should in fact be denied when mounting because they do
2232 not understand ACLs.
2235 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2237 Description: many clients connecting with IO in progress causes connect timeouts
2238 Details : Avoid synchronous journal commits to avoid delays caused by many
2239 clients connecting/disconnecting when bulk IO is in progress.
2240 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2241 OST_IO_PORTAL to avoid delays behind potentially many pending
2245 Frequency : occasionally with multiple writers to a single file
2247 Description: shared writes to file may result in wrong size reported by stat()
2248 Details : Allow growing of kms when extent lock is cancelled
2251 Frequency : always with random mmap IO to multi-striped file
2253 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2254 Details : Only if the hole is at the end of OST object so that kms is too
2255 small. Fix is to increase kms accordingly in ll_nopage.
2258 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2260 Description: writes to a missing object would leak memory on the OST
2261 Details : If there is an inconsistency between the MDS and OST filesystems,
2262 such that the MDS references an object that doesn't exist, writes
2263 to that object will leak memory due to incorrect cleanup in the
2264 error handling path, eventually running out of memory on the OST.
2269 Description: Creating too long symlink causes lustre errors
2270 Details : Check symlink and name lengths before sending requests to MDS.
2273 Frequency : only if flock is enabled (not on by default)
2275 Description: posix locks not released on fd closure on 2.6.9+
2276 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2277 kernels, this caused such locks not to be released on fd close and
2278 then assertions on fs unmount about still used locks.
2281 Frequency : MDS failover only, very rarely
2283 Description: clients may get ASSERTION(granted_lock != NULL)
2284 Details : When request was taking a long time, and a client was resending
2285 a getattr by name lock request. The were multiple lock requests
2286 with the same client lock handle and
2287 mds_getattr_name->fixup_handle_for_resent_request found one of the
2288 lock handles but later failed with ASSERTION(granted_lock != NULL).
2293 Description: handle->h_buffer_credits > 0, assertion failure
2294 Details : h_buffer_credits is zero after truncate, causing assertion
2295 failure. This patch extends the transaction or creates a new
2299 Frequency : NFS re-export or patchless client
2300 Bugzilla : 11179, 10796
2301 Description: Crash on NFS re-export node (__d_move)
2302 Details : We do not want to hash the dentry if we don't have a lock.
2303 But if this dentry is later used in d_move, we'd hit uninitialised
2304 list head d_hash, so we just do this to init d_hash field but
2305 leave dentry unhashed.
2308 Frequency : NFS re-export or patchless client
2310 Description: NFS exports has problem with symbolic link
2311 Details : lustre client didn't properly install dentry when re-exported
2312 to NFS or running patchless client.
2315 Frequency : NFS re-export or patchless client
2317 Description: Various nfs/patchless fixes.
2318 Details : fixes reuse disconected alias for lookup process - this fixes
2319 warning "find_exported_dentry: npd != pd",
2320 fix permission error with open files at nfs.
2321 fix apply umask when do revalidate.
2324 Frequency : occasional
2326 Description: Crash on NFS re-export node
2327 Details : calling clear_page() on the wrong pointer triggered oops in
2328 generic_mapping_read().
2331 Frequency : rarely, using O_DIRECT IO
2333 Description: unaligned directio crashes client with LASSERT
2334 Details : check for unaligned buffers before trying any requests.
2337 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2339 Description: stale data returned from RAID cache
2340 Details : If only a small amount of IO is done to the RAID device before
2341 reading it again it is possible to get stale data from the RAID
2342 cache instead of reading it from disk.
2345 Frequency : always for sles10 kernel
2347 Description: sles10 support
2348 Details : ll_follow_link: compile fixes and using of nd_set_link
2349 under newer kernels.
2352 Frequency : depends on arch, kernel and compiler version, always on sles10
2355 Description: recursive or deep enough symlinks cause stack overflow
2356 Details : getting rid of large stack-allocated variable in
2360 Frequency : depends on hardware
2362 Description: lustre write performance loss in the SLES10 kernel
2363 Details : the performance loss is caused by using of write barriers in the
2364 ext3 code. The SLES10 kernel turns barrier support on by
2365 default. The fix is to undo that change for ldiskfs.
2367 ------------------------------------------------------------------------------
2369 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2371 * Support for kernels:
2372 2.6.9-42.0.3EL (RHEL 4)
2373 2.6.5-7.276 (SLES 9)
2374 2.4.21-47.0.1.EL (RHEL 3)
2375 2.6.12.6 vanilla (kernel.org)
2379 Frequency : quota enabled and large files being deleted
2381 Description: releasing more than 4GB of quota at once hangs OST
2382 Details : If a user deletes more than 4GB of files on a single OST it
2383 will cause the OST to spin in an infinite loop. Release
2384 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2389 Description: statfs data retrieved from /proc may be stale or zero
2390 Details : When reading per-device statfs data from /proc, in the
2391 {kbytes,files}_{total,free,avail} files, it may appear
2392 as zero or be out of date.
2395 Frequency : systems with MD RAID1 external journal devices
2397 Description: lconf's call to blkid is confused by RAID1 journal devices
2398 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2399 of returning all block devices that match the journal UUID.
2402 Frequency : always, for aggregate stripe size over 4GB
2404 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2405 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2406 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2409 Frequency : always if "lfs find" used on a local file/directory
2411 Description: "lfs find" segfaults if used on a local file/directory
2412 Details : The case where a directory component was not specified wasn't
2413 handled correctly. Handle this properly.
2416 Frequency : always on ppc64
2418 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2419 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2420 perform endianness conversion.
2423 Frequency : rarely (truncate to non-zero file size after write under load)
2424 Bugzilla : 10730, 10687
2425 Description: Files padded with zeros to next 4K multiple
2426 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2427 it is possible that files that are truncated to a non-zero size
2428 immediately after being written are filled with zero bytes beyond
2429 the truncated size. No file data is lost.
2431 Severity : enhancement
2433 Description: Allow recovery/failover for liblustre clients.
2434 Details : liblustre clients were unaware of failover configurations until
2437 Severity : enhancement
2439 Description: user file locks should fail when not mounting with flock option
2440 Details : Set up an error-returning stub in ll_file_operations.lock field
2441 to prevent incorrect behaviour when client is mounted without
2442 flock option. Also, set up properly f_op->flock field for
2446 Frequency : always on ia64
2448 Description: "lfs df" loops on printing out MDS statfs information
2449 Details : The obd_ioctl_data was not initialized and in some systems
2450 this caused a failure during the ioctl that did not return
2451 an error. Initialize the struct and return an error on failure.
2454 Frequency : SLES 9 only
2456 Description: Error of copying files with lustre special EAs as root
2457 Details : Client side always return success for setxattr call for lustre
2458 special xattr (currently only "trusted.lov").
2461 Frequency : rarely on clusters with both ia64+i386 clients
2463 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2464 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2465 clients are doing concurrent writes to the same file it is
2466 possible that the ia64 clients may LASSERT because the OST
2467 extent locks are not PAGE_SIZE aligned. Ensure that grown
2468 locks are always aligned on the request boundary.
2471 Frequency : specific use, occasional
2473 Description: Overwriting in use executable truncates on-disk binary image
2474 Details : If one node attempts to overwrite an executable in use by
2475 another node, we now correctly return ETXTBSY instead of
2476 truncating the file.
2478 Severity : enhancement
2480 Description: Async OSC create to avoid the blocking unnecessarily.
2481 Details : If a OST has no remain object, system will block on the creating
2482 when need to create a new object on this OST. Now, ways use
2483 pre-created objects when available, instead of blocking on an
2484 empty osc while others are not empty. If we must block, we block
2485 for the shortest possible period of time.
2490 Description: chmod on Lustre root is propagated to other clients
2491 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2492 invalid by the follow_mount time.
2495 Frequency : liblustre clients only
2497 Description: Race in 'instant cancel' lock handling could lead to such locks
2498 never to be granted in case of SMP MDS
2499 Details : Do not destroy not yet granted but cbpending locks in
2503 Frequency : replay/resend of open
2505 Description: non null lock assetion failure in mds_intent_policy
2506 Details : Trying to replay/resend lockless open requests resulted in
2507 mds_open() returning 0 with no lock. Now it sets a flag if
2508 a lock is going to be returned.
2510 Severity : enhancement
2512 Description: Checksum enhancements
2513 Details : New checksum enhancements allow for resending RPCs that failed
2516 Severity : enhancement
2518 Description: Tunables on number of dirty pages in cacche
2519 Details : Allow to set limit on number of dirty pages cached.
2524 Description: client crash on unmount - lock still has references
2525 Details : In some error handling cases it was possible to leak a lock
2526 reference on a client while accessing a file. This was not
2527 harmful to the client during operation, but would cause the
2528 client to crash when the filesystem is unmounted.
2531 Frequency : specific case, rare
2533 Description: ETXTBSY on mds though file not in use by client
2534 Details : ETXTBSY is no longer incorrectly returned when attempting to
2535 chmod or chown a directory that the user previously tried to
2536 execute or a currently-executing binary.
2539 Frequency : extremely rare except on liblustre-based clients
2541 Description: Lustre space not freed when files are deleted
2542 Details : Clean up open-unlinked files after client eviction. Previously
2543 the unlink was skipped and the files remained as orphans.
2548 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2549 Details : In some cases it was possible to send a blocking callback to
2550 a client doing a glimpse, even though that client didn't get
2551 a lock granted. When the glimpse lock is cancelled on the OST
2552 the freed lock is left on the waiting list and corrupted the list.
2555 Frequency : all core dumps
2557 Description: Broke core dumps to lustre
2558 Details : Negative dentry may be unhashed if parent does not have UPDATE
2559 lock, but some callers, e.g. do_coredump, expect dentry to be
2560 hashed after successful create, hash it in ll_create_it.
2562 ------------------------------------------------------------------------------
2564 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2566 * Support for kernels:
2567 2.6.9-42.0.2.EL (RHEL 4)
2568 2.6.5-7.276 (SLES 9)
2569 2.4.21-40.EL (RHEL 3)
2570 2.6.12.6 vanilla (kernel.org)
2574 Frequency : always on RHEL 3
2576 Description: Number of open files grows over time
2577 Details : The number of open files grows over time, whether or not
2578 Lustre is started. This was due to a filp leak introduced
2579 by one of our kernel patches.
2581 ------------------------------------------------------------------------------
2583 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2585 * Support for kernels:
2586 2.6.9-42.EL (RHEL 4)
2587 2.6.5-7.276 (SLES 9)
2588 2.4.21-40.EL (RHEL 3)
2589 2.6.12.6 vanilla (kernel.org)
2594 Bugzilla : 5719, 9635, 9792, 9684
2595 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2596 Details : If a server is under heavy load and cannot reply to new
2597 connection requests before the client resends the (re)connect,
2598 the connection handling code can behave badly if two service
2599 threads are concurrently handing separate (re)connections from
2600 the same client. Add better locking to the connection handling
2601 code, and ensure that only a single connection will be processed
2602 for a given client UUID, even if the lock is dropped.
2604 Severity : enhancement
2606 Description: add TCP zero-copy support to kernel
2607 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2608 sends if the hardware supports scatter-gather and checksumming.
2609 This allows socklnd to do client-write and server-read more
2610 efficiently and reduce CPU utilization from skbuf copying.
2613 Frequency : only if NFS exporting from client
2615 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2616 Details : If setting attributes on a file created under NFS that had
2617 never been opened it would be possible to oops the client
2618 if the file had no objects.
2621 Frequency : always for liblustre
2623 Description: liblustre client does MDS+OSTs setattr RPC for each write
2624 Details : When doing a write from a liblustre client, the client
2625 incorrectly issued an RPC to the MDS and each OST the file was
2626 striped over in order to update the timestamps. When writing
2627 with small chunks and many clients this could overwhelm the MDS
2628 with RPCs. In all cases it would slow down the write because
2629 these RPCs are unnecessary.
2631 Severity : enhancement
2633 Description: allow number of MDS service threads to be changed at module load
2634 Details : It is now possible to change the number of MDS service threads
2635 running. Adding "options mds mds_num_threads={N}" to the MDS's
2636 /etc/modprobe.conf will set the number of threads for the next
2637 time Lustre is restarted (assuming the "mds" module is also
2638 reloaded at that time). The default number of threads will
2639 stay the same, 32 for most systems.
2644 Description: OST crash if filesystem is unformatted or corrupt
2645 Details : If an OST is started on a device that has never been formatted
2646 or if the filesystem is corrupt and cannot even mount then the
2647 error handling cleanup routines would dereference a NULL pointer.
2652 Description: NULL pointer deref in llap_from_page.
2653 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2654 incorrect) mapping if the page was truncated/reclaimed while it was
2655 searched for. Check for this condition and skip such pages when
2656 doing readahead. Introduce extra check to llap_from_page() to
2657 verify page->mapping->host is non-NULL (so page is not anonymous).
2660 Frequency : Sometimes when using sys_sendfile
2662 Description: "page not covered by a lock" warnings from ll_readpage
2663 Details : sendfile called ll_readpage without right page locks present.
2664 Now we introduced ll_file_sendfile that does necessary locking
2665 around call to generic_file_sendfile() much like we do in
2669 Frequency : with certain MDS communication failures at client mount time
2671 Description: NULL pointer deref after failed client mount
2672 Details : a client connection request may delayed by the network layer
2673 and not be sent until after the PTLRPC layer has timed out the
2674 request. If the client fails the mount immediately it will try
2675 to clean up before the network times out the request. Add a
2676 reference from the request import to the obd device and delay
2677 the cleanup until the network drops the request.
2680 Frequency : occasionally during client (re)connect
2682 Description: assertion failure during client (re)connect
2683 Details : processing a client connection request may be delayed by the
2684 client or server longer than the client connect timeout. This
2685 causes the client to resend the connection request. If the
2686 original connection request is replied in this interval, the
2687 client may trip an assertion failure in ptlrpc_connect_interpret()
2688 which thought it would be the only running connect process.
2691 Frequency : only with obd_echo servers and clients that are rebooted
2693 Description: kernel BUG accessing uninitialized data structure
2694 Details : When running an obd_echo server it did not start the ping_evictor
2695 thread, and when a client was evicted an uninitialized data
2696 structure was accessed. Start the ping_evictor in the RPC
2697 service startup instead of the OBD startup.
2699 Severity : enhancement
2700 Bugzilla : 10193 (patchless)
2701 Description: Remove dependency on various unexported kernel interfaces.
2702 Details : No longer need reparent_to_init, exit_mm, exit_files,
2703 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2706 Frequency : rare (only users of deprecated and unsupported LDAP config)
2708 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2709 Details : LDAP apparently contains 'lustreName' attributes instead of
2710 'name'. A simple remapping of the name is sufficient.
2713 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2715 Description: waiting_locks_callback trips kernel BUG if client is evicted
2716 Details : Running with the dump_on_timeout debug flag turned on makes
2717 it possible that the waiting_locks_callback() can try to dump
2718 the Lustre kernel debug logs from an interrupt handler. Defer
2719 this log dumping to the expired_lock_main() thread.
2721 Severity : enhancement
2723 Description: Support NFS exporting on 2.6 kernels.
2724 Details : Implement non-rawops metadata methods for NFS server to use without
2725 changing NFS server code.
2728 Frequency : very rare (synthetic metadata workload only)
2730 Description: two racing renames might cause an MDS thread to deadlock
2731 Details : Running the "racer" program may cause one MDS thread to rename
2732 a file from being the source of a rename to being the target of
2733 a rename at exactly the same time that another thread is doing
2734 so, and the second thread has already enqueued these locks after
2735 doing a lookup of the target and is trying to relock them in
2736 order. Ensure that we don't try to re-lock the same resource.
2739 Frequency : only very large systems with liblustre clients
2741 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2742 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2743 all OSTs in parallel. This allows the eviction of stale liblustre
2744 clients to proceed much faster than if they were done in series,
2745 and also offers similar improvements for other set_info RPCs.
2750 Description: excessive CPU usage during initial read phase on client
2751 Details : During the initial read phase on a client, it would agressively
2752 retry readahead on the file, consuming too much CPU and impacting
2753 performance (since 1.4.5.8). Improve the readahead algorithm
2754 to avoid this, and also improve some other common cases (read
2755 of small files in particular, where "small" is files smaller than
2756 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2761 Description: MDS crash when receiving packet with unknown intent.
2762 Details : Do not LBUG in unknown intent case, just return -EFAULT
2764 Severity : enhancement
2765 Bugzilla : 9293, 9385
2766 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2767 Details : Do not serialize getattr (non-intent version) and statfs.
2770 Frequency : occasional, when OST network is overloaded/intermittent
2772 Description: client evicted by OST after bulk IO timeout
2773 Details : If a client sends a bulk IO request (read or write) the OST
2774 may evict the client if it is unresposive to its data GET/PUT
2775 request. This is incorrect if the network is overloaded (takes
2776 too long to transfer the RPC data) or dropped the OST GET/PUT
2777 request. There is no need to evict the client at all, since
2778 the pinger and/or lock callbacks will handle this, and the
2779 client can restart the bulk request.
2782 Frequency : Always when mmapping file with no objects
2784 Description: client crashes when mmapping file with no objects
2785 Details : Check that we actually have objects in a file before doing any
2786 operations on objects in ll_vm_open, ll_vm_close and
2792 Description: Request leak when working with deleted CWD
2793 Details : Introduce advanced request refcount tracking for requests
2794 referenced from lustre intent.
2796 Severity : Enhancement
2798 Description: Cache open file handles on client.
2799 Details : MDS now will return special lock along with openhandle, if
2800 requested and client is allowed to hold openhandle, even if unused,
2801 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2802 closing files for every read/write openration.
2804 Severity : Enhancement
2806 Description: Cache open negative dentries on client when possible.
2807 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2808 negative dentries on lock revocation.
2813 Description: Remounting a client read-only wasn't possible with a zconf mount
2814 Details : It wasn't possible to remount a client read-only with llmount.
2816 Severity : enhancement
2817 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2818 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2819 MPICH is included to add Lustre-specific ADIO interfaces.
2820 This is based closely on the UFS ADIO layer and only differs
2821 in file creation, in order to allow the OST striping to be set.
2822 This is user-contributed code and not supported by CFS.
2827 Description: extended inode attributes (immutable, append-only) work improperly
2828 when 2.4 and 2.6 kernels are used on client/server or vice versa
2829 Details : Introduce kernel-independent values for these flags.
2831 Severity : enhancement
2834 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2835 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2836 granularity. Now, Fractional MB granularity is be supported,
2837 this is very useful for low memory system.
2839 Severity : enhancement
2841 Description: Getattr by fid
2842 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2843 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2844 have locks and avoids getattr rpc requests for it on every stat.
2847 Frequency : Always, for filesystems larger than 2TB
2849 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2850 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2851 filesystems larger than 2TB to be mounted with mballoc enabled.
2854 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2856 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2857 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2858 filesystems larger than 2TB is mounted then the kernel will
2859 silently corrupt the start of the filesystem. CONFIG_LBD is
2860 enabled for all CFS-supported kernels, but the possibility of
2861 this happening with a modified kernel config exists.
2863 Severity : enhancement
2865 Description: add client O_DIRECT support for 2.6 kernels
2866 Details : It is now possible to do O_DIRECT reads and writes to files
2867 in the Lustre client mountpoint on 2.6 kernel clients.
2869 Severity : enhancement
2871 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2872 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2873 parallel, not waiting for response from every OST before sending
2874 a rpc to the next OST.
2879 Description: setattr vs write race when updating file timestamps
2880 Details : Client processes that update a file timestamp into the past
2881 right after writing to the file (e.g. tar) it is possible that
2882 the updated file modification time can be reset to the current
2883 time due to a race between processing the setattr and write RPC.
2885 Severity : enhancement
2887 Description: Bring 'lfs find' closer in line with regular Linux find.
2888 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2889 -print0 options and obtains all the needed info through the lustre
2892 Severity : enhancement
2894 Description: support up to 1024 configured devices on one node
2895 Details : change obd_dev array from statically allocated to dynamically
2896 allocated structs as they are first used to reduce memory usage
2901 Description: Flush dirty partially truncated pages during truncate
2902 Details : Immediatelly flush partially truncated pages in filter_setattr,
2903 this way we completely avoid having any pages in page cache on OST
2904 and can retire ugly workarounds during writes to flush such pages.
2909 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2910 Details : protect inode from truncation within vfs_unlink() context
2911 just take a reference before calling vfs_unlink() and release it
2912 when parent's i_sem is free.
2917 Description: last_id value checked outside lock on OST caused LASSERT failure
2918 Details : If there were multiple MDS->OST object precreate requests in
2919 flight, it was possible that the OST's last object id was checked
2920 outside a lock and incorrectly tripped an assertion. Move checks
2921 inside locks, and discard old precreate requests.
2924 Frequency : always, if extents are used on OSTs
2926 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2927 Details : OSTs using the extents format would not zero the high 16 bits of
2928 the index physical block number. This is not a problem for any
2929 OST filesystems smaller than 16TB, and no kernels support ext3
2930 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2931 new/modified files) and can be fixed for existing filesystems
2932 with e2fsprogs-1.39-cfs1.
2937 Description: import connection selection may be incorrect if timer wraps
2938 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2939 import connections to be ignored if the 32-bit jiffies counter
2940 wraps. Use a 64-bit jiffies counter.
2943 Frequency : very large clusters immediately after boot
2945 Description: LNET request buffers exhausted under heavy short-term load
2946 Details : If a large number of client requests are generated on a service
2947 that has previously never seen so many requests it is possible
2948 that the request buffer growth cannot keep up with the spike in
2949 demand. Instead of dropping incoming requests, they are held in
2950 the LND until the RPC service can accept more requests.
2953 Frequency : Sometimes during replay
2955 Description: Assertion failure in ll_local_open after replay.
2956 Details : If replay happened on an open request reply before we were able
2957 to set replay handler, reply will become not swabbed tripping the
2958 assertion in ll_local_open. Now we set the handler right after
2959 recognising of open request
2962 Frequency : very rare
2964 Description: kernel reports "badness in vsnprintf"
2965 Details : Reading from the "recovery_status" /proc file in small chunks
2966 may cause a negative length in lprocfs_obd_rd_recovery_status()
2967 call to vsnprintf() (which is otherwise harmless). Exit early
2968 if there is no more space in the output buffer.
2970 Severity : enhancement
2972 Description: clear OBD RPC statistics by writing to them
2973 Details : It is now possible to clear the OBD RPC statistics by writing
2974 to the "stats" file.
2979 Description: Client mtime is not the same on different clients after utimes
2980 Details : In some cases, the client was using the utimes() syscall on
2981 a file cached on another node. The clients now validate the
2982 ctime from the MDS + OSTs to determine which one is right.
2987 Description: Inability to activate failout mode
2988 Details : lconf script incorrectly assumed that in pythong string's numeric
2989 value is used in comparisons.
2992 Frequency : always with multiple stripes per file
2994 Description: Inefficient object allocation for mutli-stripe files
2995 Details : When selecting which OSTs to stripe files over, for files with
2996 a stripe count that divides evenly into the number of OSTs,
2997 the MDS is always picking the same starting OST for each file.
2998 Return the OST selection heuristic to the original design.
3003 Description: mount failures may take full timeout to return an error
3004 Details : Under some heavy load conditions it is possible that a
3005 failed mount can wait for the full obd_timeout interval,
3006 possibly several minutes, before reporting an error.
3007 Instead return an error as soon as the status is known.
3009 Frequency : quota enabled and large files being deleted
3011 Description: releasing more than 4GB of quota at once hangs OST
3012 Details : If a user deletes more than 4GB of files on a single OST it
3013 will cause the OST to spin in an infinite loop. Release
3014 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3019 Description: statfs data retrieved from /proc may be stale or zero
3020 Details : When reading per-device statfs data from /proc, in the
3021 {kbytes,files}_{total,free,avail} files, it may appear
3022 as zero or be out of date.
3025 Frequency : systems with MD RAID1 external journal devices
3027 Description: lconf's call to blkid is confused by RAID1 journal devices
3028 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3029 of returning all block devices that match the journal UUID.
3032 Frequency : always, for aggregate stripe size over 4GB
3034 Description: assertion fails when trying to use 4GB stripe size
3035 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
3036 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
3039 Frequency : always on ppc64
3041 Description: the first write on an ext3 filesystem with mballoc got stuck
3042 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3043 perform endianness conversion.
3045 ------------------------------------------------------------------------------
3047 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3049 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3050 INTEROPERATE with older versions automatically. Please read the
3051 user documentation before upgrading any part of a live system.
3052 * WARNING: Lustre networking configuration changes are required with
3053 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3056 * Support for kernels:
3057 2.6.9-22.0.2.EL (RHEL 4)
3058 2.6.5-7.244 (SLES 9)
3059 2.6.12.6 vanilla (kernel.org)
3062 Severity : enhancement
3063 Bugzilla : 7981/8208
3064 Description: Introduced Lustre Networking (LNET)
3065 Details : LNET is new networking infrastructure for Lustre, it includes
3066 a reorganized network configuration mode (see the user
3067 documentation for full details) as well as support for routing
3068 between different network fabrics. Lustre Networking Devices
3069 (LNDs) for the supported network fabrics have also been
3070 created for this new infrastructure.
3072 Severity : enhancement
3073 Description: Introduced Access control lists
3074 Details : clients can set ACLs on files and directories in order to have
3075 more fine-grained permissions than the standard Unix UGO+RWX.
3076 The MDS must be started with the "-o acl" mount option.
3078 Severity : enhancement
3079 Description: Introduced filesystem quotas
3080 Details : Administrators may now establish per-user quotas on the
3083 Severity : enhancement
3085 Description: Configuration change for the XT3
3086 The PTLLND is now used to run Lustre over Portals on the XT3
3087 The configure option(s) --with-cray-portals are no longer used.
3088 Rather --with-portals=<path-to-portals-includes> is used to
3089 enable building on the XT3. In addition to enable XT3 specific
3090 features the option --enable-cray-xt3 must be used.
3095 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3096 Details : A race between allocating a new llcd and re-getting the llcd_lock
3097 allowed another thread to grab newly-allocated llcd.
3099 Severity : enhancement
3101 Description: 2.6 OST async journal commit and locking fix to improve performance
3102 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3103 2.6 kernels are now async as they already were in 2.4 kernels so
3104 that they can commit concurrently with the network bulk transfer.
3105 For block-allocated files the filter allocation semaphore is held
3106 to avoid filesystem fragmentation during allocation. BKL lock
3107 removed for 2.6 xattr operations where it is no longer needed.
3112 Description: lconf incorrectly determined whether two IP networks could talk
3113 Details : In some more complicated routing and multiple-network
3114 configurations, lconf will avoid trying to make a network
3115 connection to a disjoint part of the IP space. It was doing the
3116 math incorrectly for one set of cases.
3121 Description: Fix for potential infinite loop processing records in an llog.
3122 Details : If an llog record is corrupted/zeroed, it is possible to loop
3123 forever in llog_process(). Validate the llog record length
3124 and skip the remainder of the block on error.
3127 Frequency : occasional (liblustre only)
3129 Description: liblustre could not open files whose last component is a symlink
3130 Details : sysio_path_walk() would incorrectly pass the open intent to
3131 intermediate path components.
3134 Frequency : rare (liblustre only with non-standard tuning)
3135 Bugzilla : 7201 (7350)
3136 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3137 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3138 after it has been released, only do this for referenced locks
3140 Severity : enhancement
3142 Description: specifying an (invalid) directory default stripe_size of -1
3143 would reset the directory default striping
3144 Details : stripe_size -1 was used internally to signal directory stripe
3145 removal, now use "all default" to signal dir stripe removal
3146 as a directory striping of "all default" is not useful
3149 Frequency : common for large clusters running liblustre clients
3151 Description: doing an ls when liblustre clients are running is slow
3152 Details : sending a glimpse AST to a liblustre client waits for every AST
3153 to time out, as liblustre clients will not respond. Since they
3154 cannot cache data we refresh the OST lock LVB from disk instead.
3156 Severity : enhancement
3158 Description: doing an ls at the same time as file IO can be slow
3159 Details : enqueue and other "small" requests can be blocked behind many
3160 large IO requests. Create a new OST IO portal for non-IO
3161 requests so they can be processed faster.
3164 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3166 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3167 created file with mode 000
3168 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3169 allow an ftruncate() as a non-root user to a file with mode 000.
3170 Lustre now respects this flag to disable mode checks when
3171 truncating a file owned by the user
3174 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3177 Description: Revoking locks from clients that went dead or catatonic might take
3179 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3180 cancellation of such locks instant on servers without waiting for
3181 any reply from clients. Clients drops these locks when cancel
3182 notification from server is received without replying.
3185 Frequency : liblustre-only, when liblustre client dies or becomes busy
3187 Description: Doing ls on Linux clients can take a long time with active
3189 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3190 granting such locks to it in the first place if possible. Locks
3191 are taken by proxy on the OST during the read or write and
3192 dropped immediately afterward. Add connect flags handling, do
3193 not grant locks to liblustre clients for glimpse ASTs.
3195 Severity : enhancement
3197 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3198 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3199 sequential reads and behaves badly with applications doing random
3200 reads. Improve it by reading ahead at least read region, and
3201 avoiding excessive large RPC for small reads.
3203 Severity : enhancement
3205 Description: Creating more than 1000 files for a single job may cause a load
3206 imbalance on the OSTs if there are also a large number of OSTs.
3207 Details : qos_prep_create() uses an OST index reseed value that is an
3208 even multiple of the number of available OSTs so that if the
3209 reseed happens in the middle of the object allocation it will
3210 still utilize the OSTs as uniformly as possible.
3215 Description: OST or MDS may oops in ping_evictor_main()
3216 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3217 but doesn't restart at beginning of obd_exports_timed list
3220 Severity : enhancement
3222 Description: improve by-nid export eviction on the MDS and OST
3223 Details : allow multiple exports with the same NID to be evicted at one
3224 time without re-searching the exports list.
3227 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3229 Description: MDS may oops in groups_free()
3230 Details : in rare race conditions a newly allocated group_info struct is
3231 freed again, and this can be NULL. The 2.4 compatibility code
3232 for groups_free() checked for a NULL pointer, but 2.6 did not.
3235 Frequency : common for liblustre clients doing little filesystem IO
3236 Bugzilla : 9352, 7313
3237 Description: server may evict liblustre clients accessing contended locks
3238 Details : if a client is granted a lock or receives a completion AST
3239 with a blocking AST already set it would not reply to the AST
3240 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3243 Frequency : lfs setstripe, only systems with more than 160 OSTs
3245 Description: unable to set striping with a starting offset beyond OST 160
3246 Details : llapi_create_file() incorrectly limited the starting stripe
3247 index to the maximum single-file stripe count.
3250 Frequency : LDAP users only
3252 Description: lconf did not handle in-kernel recovery with LDAP properly
3253 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3255 Severity : enhancement
3257 Description: bind OST threads to NUMA nodes to improve performance
3258 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3259 node and do their allocations there to localize memory access
3261 Severity : enhancement
3263 Description: llmount can determine client NID directly from Myrinet (GM)
3264 Details : the client NID code from gmnalnid was moved directly into
3265 llmount, removing the need to use this or specifying the
3266 client NID explicitly when mounting GM clients with zeroconf
3269 Frequency : if client is started with down MDS
3271 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3272 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3273 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3274 ptlrpc_import_delayed_req() because we didn't check if the
3275 request was interrupted, and we also didn't break out of the
3276 event loop if there was no timeout
3281 Description: data loss during non-page-aligned writes to a single file from
3282 both multiple nodes and multiple threads on one node at same time
3283 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3284 inconsistency led to false short-reads, that were cached and later
3285 used by ->prepare_write() to fill in partially written page,
3286 leading to data loss.
3289 Frequency : always, if lconf --abort_recovery used
3291 Description: lconf --abort_recovery fails with 'Operation not supported'
3292 Details : lconf was attempting to abort recovery on the MDT device and not
3295 Severity : enhancement
3297 Description: remove cleanup logs
3298 Details : replace lconf-generated cleanup logs with lustre internal
3299 cleanup routines. Eliminates the need for client-cleanup and
3302 Severity : enhancement
3304 Description: add support for EAs (user and system) on lustre filesystems
3305 Details : it is now possible to store extended attributes in the Lustre
3306 client filesystem, and with the user_xattr mount option it
3307 is possible to allow users to store EAs on their files also
3309 Severity : enhancement
3311 Description: Add possibility (config option) to show minimal available OST free
3313 Details : When compiled with --enable-mindf configure option, statfs(2)
3314 (and so, df) will return least minimal free space available from
3315 all OSTs as amount of free space on FS, instead of summary of
3316 free spaces of all OSTs.
3318 Severity : enhancement
3320 Description: do not expand extent locks acquired on OST-side
3321 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3322 by server: they are not cached anyway.
3325 Frequency : when mmap is used/binaries executed from Lustre
3327 Description: Unmmap pages before throwing them away from read cache.
3328 Details : llap_shrink cache now attempts to unmap pages before discarding
3329 them (if unmapping failed - do not discard). SLES9 kernel has
3330 extra checks that trigger if this unmapping is not done first.
3335 Description: lconf didn't resolve symlinks before checking to see whether a
3336 given mountpoint was already in use
3339 Frequency : when migrating failover services
3340 Bugzilla : 6395, 9514
3341 Description: When migrating a subset of services from a node (e.g. failback
3342 from a failover service node) the remaining services would
3343 time out and evict clients.
3344 Details : lconf --force (implied by --failover) sets the global obd_timeout
3345 to 5 seconds in order to quickly disconnect, but this caused
3346 other RPCs to time out too quickly. Do not change the global
3347 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3349 Severity : enhancement
3350 Frequency : if MDS is started with down OST
3351 Bugzilla : 9439,5706
3352 Description: Allow startup/shutdown of an MDS without depending on the
3353 availability of the OSTs.
3354 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3355 Add appropriate locking and lov-osc refcounts for safe
3356 cleaning. Add osc abort_inflight calls in case the
3357 synchronize never started.
3360 Frequency : occasional (Cray XT3 only)
3362 Description: root not authorized to access files in CRAY_PORTALS environment
3363 Details : The client process capabilities were not honoured on the MDS in
3364 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3365 been accessed by an authorized user then root was able to access
3366 the file on the local client also. The root user capabilities
3367 are now allowed on the MDS, as this environment has secure UID.
3370 Frequency : occasional
3372 Description: ldiskfs "too long searching" message happens too often
3373 Details : A debugging message (otherwise harmless) prints too often on
3374 the OST console. This has been reduced to only happen when
3375 there are fragmentation problems on the filesystem.
3380 Description: Division by zero in statfs when all OSCs are inactive
3381 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3382 lov_statfs divides by value returned by lov_get_stripecnt().
3386 Bugzilla : 9489, 3273
3387 Description: First write from each client to each OST was only 4kB in size,
3388 to initialize client writeback cache, which caused sub-optimal
3389 RPCs and poor layout on disk for the first writen file.
3390 Details : Clients now request an initial cache grant at (re)connect time
3391 and so that they can start streaming writes to the cache right
3392 away and always do full-sized RPCs if there is enough data.
3393 If the OST is rebooted the client also re-establishes its grant
3394 so that client cached writes will be honoured under the grant.
3399 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3400 Details : Now I/O RPCs go to different portal number and (presumably) fast
3401 lock requests (and glimses) and other RPCs get their own service
3402 threads pool that should be able to service those RPCs
3405 Severity : enhancement
3407 Description: Ability to exchange lustre version between client and servers and
3408 issue warnings at client side if client is too old. Also for
3409 liblustre clients there is ability to refuse connection of too old
3411 Details : New 'version' field is added to connect data structure that is
3412 filled with version info. That info is later checked by server and
3416 Frequency : rare, liblustre only.
3417 Bugzilla : 9296, 9581
3418 Description: Two simultaneous writes from liblustre at offset within same page
3419 might proceed at the same time overwriting eachother with stale
3421 Details : I/O lock withing llu_file_prwv was released too early, before data
3422 actually was hitting the wire. Extended lock-holding time until
3423 server acknowledges receiving data.
3426 Frequency : extremely rare. Never observed in practice.
3428 Description: avoid generating lustre_handle cookie of 0.
3429 Details : class_handle_hash() generates handle cookies by incrementing
3430 global counter, and can hit 0 occasionaly (this is unlikely, but
3431 not impossible, because initial value of cookie counter is
3432 selected randonly). Value of 0 is used as a sentinel meaning
3433 "unassigned handle" --- avoid it. Also coalesce two critical
3434 sections in this function into one.
3436 Severity : enhancement
3438 Description: allow liblustre clients to delegate truncate locking to OST
3439 Details : To avoid overhead of locking, liblustre client instructs OST to
3440 take extent lock in ost_punch() on client's behalf. New connection
3441 flag is added to handle backward compatibility.
3443 Severity : enhancement
3444 Bugzilla : 4928, 7341, 9758
3445 Description: allow number of OST service threads to be specified
3446 Details : a module parameter allows the number of OST service threads
3447 to be specified via "options ost ost_num_threads={N}" in the
3448 OSS's /etc/modules.conf or /etc/modprobe.conf.
3452 Bugzilla : 6146, 9635, 9895
3453 Description: servers crash with bad pointer in target_handle_connect()
3454 Details : In rare cases when a client is reconnecting it was possible that
3455 the connection request was the last reference for that export.
3456 We would temporarily drop the export reference and get a new
3457 one, but this may have been the last reference and the export
3458 was just destroyed. Get new reference before dropping old one.
3460 Severity : enhancement
3461 Frequency : if client is started with failover MDS
3463 Description: Allow multiple MDS hostnames in the mount command
3464 Details : Try to read the configuration from all specified MDS
3465 hostnames during a client mount in case the "primary"
3468 Severity : enhancement
3470 Description: Stop sending data to evicted clients as soon as possible.
3471 Details : Check if the client we are about to send or are sending data to
3472 was evicted already. (Check is done every second of waiting,
3473 for which l_wait_event interface was extended to allow checking
3474 of exit condition at specified intervals).
3477 Frequency : rare, normally only when NFS exporting is done from client
3479 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3481 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3484 Frequency : always, if srand() is called before liblustre initialization
3486 Description: Liblustre uses system PRNG disturbing its usage by user application
3487 Details : Introduce internal to lustre fast and high-quality PRNG for
3488 lustre usage and make liblustre and some other places in generic
3489 lustre code to use it.
3491 Severity : enhancement
3492 Bugzilla : 9477, 9557, 9870
3493 Description: Verify that the MDS configuration logs are updated when xml is
3494 Details : Check if the .xml configuration logs are newer than the config
3495 logs stored on the MDS and report an error if this is the case.
3496 Request --write-conf, or allow starting with --old_conf.
3498 Severity : enhancement
3500 Description: Handle symlinks in the path when checking if Lustre is mounted.
3501 Details : Resolve intermediate symlinks when checking if a client has
3502 mounted a filesystem to avoid duplicate client mounts.
3507 Description: lconf can hit an error exception but still return success.
3508 Details : The lconf command catches the Command error exception at the top
3509 level script context and will exit with the associated exit
3510 status, but doesn't ensure that this exit status is non-zero.
3515 Description: failure of ptlrpc thread startup can cause oops
3516 Details : Starting a ptlrpc service thread can fail if there are a large
3517 number of threads or the server memory is very fragmented.
3518 Handle this without oopsing.
3521 Frequency : always, only if liblustre and non-default acceptor port was used
3523 Description: liblustre cannot connect to servers with non-default acceptor port
3524 Details : tcpnal_set_default_params() was not called and was therefore
3525 ignoring the environment varaible TCPNAL_PORT, as well as other
3526 TCPNAL_ environment variables
3531 Description: two objects could be created on the same OST for a single file
3532 Details : If an OST is down, in some cases it was possible to create two
3533 objects on a single OST for a single file. No problems other
3534 than potential performance impact and spurious error messages.
3538 Bugzilla : 5681, 9562
3539 Description: Client may oops in ll_unhash_aliases
3540 Details : Client dcache may become inconsistent in race condition.
3541 In some cases "getcwd" can fail if the current directory is
3547 Description: Inode refcounting problems in NFS export code
3548 Details : link_raw functions used to call d_instantiate without obtaining
3549 extra inode reference first.
3553 Bugzilla : 9942, 9903
3554 Description: Referencing freed requests leading to crash, memleaks with NFS.
3555 Details : We used to require that call to ll_revalidate_it was always
3556 followed by ll_lookup_it. Also with revalidate_special() it is
3557 possible to call ll_revalidate_it() twice for the same dentry
3558 even if first occurence returned success. This fix changes semantic
3559 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3560 reference on a request referred from the intent.
3561 ll_intent_release() then releases such a request.
3564 Frequency : rare, normally benchmark loads only
3566 Description: unlinked inodes were kept in memory on the client
3567 Details : If a client is repeatedly creating and unlinking files it
3568 can accumulate a lot of stale inodes in the inode slab cache.
3569 If there is no other client load running this can cause the
3570 client node to run out of memory. Instead flush old inodes
3571 from client cache that have the same inode number as a new inode.
3574 Frequency : SLES9 2.6.5 kernel and long filenames only
3575 Bugzilla : 9969, 10379
3576 Description: utime reports stale NFS file handle
3577 Details : SLES9 uses out-of-dentry names in some cases, which confused
3578 the lustre dentry revalidation. Change it to always use the
3582 Frequency : rare, unless heavy write-truncate concurrency is continuous
3583 Bugzilla : 4180, 6984, 7171, 9963, 9331
3584 Description: OST becomes very slow and/or deadlocked during object unlink
3585 Details : filter_destroy() was holding onto the parent directory lock
3586 while truncating+unlinking objects. For very large objects this
3587 may block other threads for a long time and slow overall OST
3588 responsiveness. It may also be possible to get a lock ordering
3589 deadlock in this case, or run out of journal credits because of
3590 the combined truncate+unlink. Solution is to do object truncate
3591 first in one transaction without parent lock, and then do the
3592 final unlink in a new transaction with the parent lock. This
3593 reduces the lock hold time dramatically.
3596 Frequency : rare, 2.4 kernels only
3598 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3599 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3600 before unmounting their filesystem in order to avoid deadlock.
3601 The kernel_locked() function in 2.4 kernels only checks whether
3602 the kernel lock is held, not whether it is this process that is
3603 holding it as 2.6 kernels do.
3608 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3609 Details : The client ptlrpc code may be trying to reconnect to a down
3610 server before a previous connection attempt has timed out.
3611 Increase the reconnect interval to be longer than the connection
3612 timeout interval to avoid sending duplicate connections to
3616 Frequency : echo_client brw_test command
3618 Description: fix echo_client to work with OST preallocated code
3619 Details : OST preallocation code (5137) didn't take echo_client IO path
3620 into account: echo_client calls filter methods outside of any
3621 OST thread and, hence, there is no per-thread preallocated
3622 pages and buffers to use. Solution: hijack pga pages for IO. As
3623 a byproduct, this avoids unnecessary data copying.
3627 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3628 Description: Client can oops in mdc_commit_close() after open replay
3629 Details : It was possible for the MDS to return an open request with no
3630 transaction number in mds_finish_transno() if the client was
3631 evicted, but without actually returning an error. Clients
3632 would later try to replay that open and may trip an assertion
3633 Simplify the client close codepath, and always return an error
3634 from the MDS in case the open is not successful.
3637 Frequency : rare, 2.6 OSTs only
3639 Description: OST may deadlock under high load on fragmented files
3640 Details : If there was a heavy load and highly-fragmented OST filesystems
3641 it was possible to have all the OST threads deadlock waiting on
3642 allocation of biovecs, because the biovecs were not released
3643 until the entire RPC IO was completed. Instead, release biovecs
3644 as soon as they are complete to ensure forward IO progress.
3646 Severity : enhancement
3648 Description: Support for specifying external journal device at mount
3649 Details : If an OST or MDS device is formatted with an external journal
3650 device, this device major/minor is stored in the ext3 superblock
3651 and may not be valid for failover. Allow detecting and
3652 specifying the external journal at mount time.
3657 Description: Mounting an MDS with pending unlinked files may cause oops
3658 Details : target_finish_recovery() calls mds_postrecov() which returned
3659 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3660 considers this an error and immediately begins cleaning up the
3661 lov, just after starting the mds_lov process
3663 Severity : enhancement
3665 Description: Implement 'lfs df' to report actual free space on per-OST basis
3666 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3667 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3668 usage of objects; '-h' to report in human readable format.
3670 ------------------------------------------------------------------------------
3672 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3679 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3680 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3681 at startup during orphan unlinking. Instead, skip buddy update
3682 before setup, as it will be regenerated after recovery is complete.
3687 Description: If an OST is inactive, its locks might reference stale inodes.
3688 Details : lov_change_cbdata() must iterate over all namespaces, even if
3689 they are inactive to clear inode references from the lock.
3691 Severity : enhancement
3692 Frequency : occasional, if non-standard max_dirty_mb used
3694 Description: Client will block write RPCs if not enough grant
3695 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3696 then the client will block writes while waiting for another RPC
3697 to complete instead of consuming its dirty limit. With change
3698 we get improved performance when max_dirty_mb is small.
3700 Severity : enhancement
3701 Bugzilla : 3389, 6253
3702 Description: Add support for supplementary groups on the MDS.
3703 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3704 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3705 lookups for user supplementary groups into a cache.
3709 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3710 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3711 for liblustre O_EXCL is enforced by MDS.
3715 Description: readdir never returns NULL in liblustre.
3716 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3717 dentry in struct dirent.
3721 Frequency : liblustre only, depends on application IO pattern
3722 Description: liblustre clients evicted if not contacting servers
3723 Details : Don't put liblustre clients into the ping_evictor list, so
3724 they will not be evicted by the pinger ever.
3726 Severity : enhancement
3728 Description: Add ability to evict clients by NID from MDS.
3729 Details : By echoing "nid:$NID" string into
3730 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3731 $NID would be instantly evicted from this MDS and from all active
3732 OSTs connected to it.
3736 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3737 Details : lookup_it_finish() used to query file size from OSTs that was not
3742 Description: service threads change working directory to that of init
3743 Details : Starting lustre service threads may pin the working directory
3744 of the parent thread, making that filesystem busy. Threads
3745 now change to the working directory of init to avoid this.
3749 Frequency : during shutdown only
3750 Description: shutdown with a failed MDS or OST can cause unmount to hang
3751 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3756 Frequency : 2.6 only
3757 Description: chmod/chown may include an extra supplementary group
3758 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3759 supplementary group and if none is specified this is used.
3762 Bugzilla : 5479 (6816)
3764 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3765 Details : If lookup is in progress on a file that is unlinked we might try
3766 to revalidate the inode and fail in revalidate after lookup is
3767 complete and ll_file_open() enqueues the open again but
3768 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3771 Frequency : always, if lconf --abort_recovery used
3773 Description: lconf --abort_recovery fails with 'Operation not supported'
3774 Details : lconf was attempting to abort recovery on the MDT device and not
3777 ------------------------------------------------------------------------------
3779 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3784 Frequency : rare (only unsupported configurations with a node running as an
3786 Bugzilla : 6514, 5137
3787 Description: Mounting a Lustre file system on a node running as an OST could
3789 Details : OSTs now preallocates memory needed to write out data at
3790 startup, instead of when needed, to avoid having to
3791 allocate memory in possibly low memory situations.
3792 Specifically, if the file system is mounted on on OST,
3793 memory pressure could force it to try to write out data,
3794 which it needed to allocate memory to do. Due to the low
3795 memory, it would be unable to do so and the node would
3796 become unresponsive.
3798 Severity : enhancement
3800 Description: Addition of lconf --service command line option
3801 Details : lconf now accepts a '--service <arg>' option, which is
3802 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3804 Severity : enhancement
3806 Description: Failover mode is now the default for OSTs.
3807 Details : By default, OSTs will now run in failover mode. To return to
3808 the old behaviour, add '--failout' to the lmc line for OSTs.
3810 Severity : enhancement
3812 Description: Health checks are now provided for MDS and OSTs
3813 Details : Additional detailed health check information on MSD and OSTs
3814 is now provided through the procfs health_check value.
3817 Frequency : occasional, depends on IO load
3819 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3820 after numerous create/delete cycles
3821 Details : The ext3 inode allocation policy would not allocate new inodes
3822 very well on the OSTs because there are no new directories
3823 being created. Instead we look for groups with free space if
3824 the parent directories are nearly full.
3829 Description: Network or server problems during mount may cause partially
3830 mounted clients instead of returning an error.
3831 Details : The config llog parsing code may overwrite the error return
3832 code during mount error handling, returning success instead
3838 Description: MDS can fail to allocate large reply buffers
3839 Details : After long uptimes the MDS can fail to allocate large reply
3840 buffers (e.g. zconf client mount config records) due to memory
3841 fragmentation or consumption by the buffer cache. Preallocate
3842 some large reply buffers so that these replies can be sent even
3843 under memory pressure.
3847 Frequency : rare (liblustre)
3848 Description: fsx running with liblustre complained that using truncate() to
3849 extend the file doesn't work. This patch corrects that issue.
3850 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3851 ATTR_SIZE and lsm use in llu_setattr_raw.
3855 Frequency : rare, only 2.6 kernels
3856 Description: Unusual file access patterns on the MDS may result in inode
3857 data being lost in very rare circumstances.
3858 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3859 code in the 2.6 kernel caused the inode and/or EA data not to
3860 be read from disk, causing single-file corruption.
3864 Frequency : rare, only 2.6 filesystems using extents
3865 Description: Heavy concurrent write and delete load may cause data corruption.
3866 Details : It was possible under high-load situations to have an extent
3867 metadata block in the block device cache from a just-unlinked
3868 file overwrite a newly-allocated data block. We now unmap any
3869 metadata buffers that alias just-allocated data blocks.
3873 Frequency : filesystems with default stripe_count larger than 77
3874 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3875 Details : lconf specifies an inode size of 4096 bytes when the default
3876 stripe_count is larger than 77. This conflicts with the default
3877 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3878 this case to avoid pinning too much memory for large EAs.
3880 ------------------------------------------------------------------------------
3882 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3887 Frequency : rare (extremely heavy IO load with hundreds of clients)
3889 Description: Client is evicted, gets IO error writing to file
3890 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3891 caused clients to be evicted for AST timeouts. The fixes in
3892 bug 5192 mean we no longer need to have such short AST timeouts
3893 so ldlm_timeout has been increased.
3896 Frequency : occasional during --force or --failover shutdown under load
3897 Bugzilla : 5949, 4834
3898 Description: Server oops/LBUG if stopped with --force or --failover under load
3899 Details : a collection of import/export refcount and cleanup ordering
3900 issues fixed for safer force cleanup
3903 Frequency : only filesystems larger than 120 OSTs
3904 Bugzilla : 5990, 6223
3905 Description: lfs getstripe would oops on a very large filesystem
3906 Details : lov_getconfig used kfree on vmalloc'd memory
3909 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3910 Bugzilla : 6242, 6243
3911 Description: reading from files that had been truncated to a non-zero size
3912 but never opened returned no data
3913 Details : ll_file_read() reads zeros from no-object files to EOF
3918 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3919 Details : There's little harm in aborting MDS/OSS recovery and letting it
3920 try again, so I removed the LASSERT and return an error instead.
3922 Severity : enhancement
3924 Description: New debugging infrastructure for tracking down data corruption
3925 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3926 (b) cover more of the client-side code path, and (c) try to narrow
3927 down where problems occurred
3931 Bugzilla : 3819, 4364, 4397, 6313
3932 Description: Racing close and eviction MDS could cause assertion in mds_close
3933 Details : It was possible to get multiple mfd references during close and
3934 client eviction, leading to one thread referencing a freed mfd.
3936 Severity: : enhancement
3937 Bugzilla : 3262, 6359
3938 Description: Attempts to reconnect to servers are now more aggressive.
3939 Details : This builds on the enhanced upcall-less recovery that was added
3940 in 1.4.2. When trying to reconnect to servers, clients will
3941 now try each server in the failover group every 10 seconds. By
3942 default, clients would previously try one server every 25 seconds.
3947 Description: After recovery, certain operations trigger a failed
3948 assertion on a client.
3949 Details : Failing over an mds, using lconf -d --failover, while a
3950 client was doing a readdir() call would cause the client to
3951 LBUG after recovery completed and the readdir() was resent.
3953 Severity : enhancement
3955 Description: Default groups are now added by lconf
3956 Details : You can now run lconf --group <servicename> without having to
3957 manually add groups with lmc.
3960 Frequency : occasional
3962 Description: Nodes with an elan id of 0 trigger a failed assertion
3965 Frequency : always when accessing e.g. tty/console device nodes
3967 Description: tty and some other devices nodes cannot be used on lustre
3968 Details : file's private_data field is used by device data and lustre
3969 values in there got lost. New field was added to struct file to
3970 store fs-specific private data.
3973 Frequency : when exporting Lustre via NFS
3975 Description: NFSD failed occasionally when looking up a path component
3976 Details : NFSD is looking up ".." which was broken in ext3 directories
3977 that had grown large enough to become hashed.
3980 Frequency : Clusters with multiple interfaces not on the same subnet
3982 Description: Nodes will repeatedly try to reconnect to an interface which it
3983 cannot reach and report an error to the log.
3984 Details : Extra peer list entries will be created by lconf with some peers
3985 unreachable. lconf now validates the peer before adding it.
3988 Frequency : Only if a default stripe is set on the filesystem root.
3990 Description: Setting a default stripe on the filesystem root prevented the
3991 filesystem from being remounted.
3992 Details : The client was sending extra request flags in the root getattr
3993 request and did not allocate a reply buffer for the dir EA.
3996 Frequency : occasional, higher if lots of files are accessed by one client
3997 Bugzilla : 6159, 6097
3998 Description: Client trips assertion regarding lsm mismatch/magic
3999 Details : While revalidating inodes the VFS looks up inodes with ifind()
4000 and in rare cases can find an inode that is being freed.
4001 The ll_test_inode() code will free the lsm during ifind()
4002 when it finds an existing inode and then the VFS later attaches
4003 this free lsm to a new inode.
4007 Bugzilla : 6422, 7030
4008 Description: MDS deadlock between mkdir and client eviction
4009 Details : Creating a new file via mkdir or mknod (starting a transaction
4010 and getting the ns lock) can deadlock with client eviction
4011 (gets ns lock and trying to finish a synchronous transaction).
4014 Frequency : occasional
4015 Description: While starting a server, the fsfilt_ext3 module could not be
4017 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
4018 kernels. Previously, lconf would still use the ext3 name
4019 when trying to load modules. Now, it will correctly use
4020 ext3 on 2.4 and ldiskfs on 2.6.
4022 Severity : enhancement
4023 Description: The default stripe count has been changed to 1
4024 Details : The interpretation of the default stripe count (0, to lfs
4025 or lmc) has been changed to mean striping across a single
4026 OST, rather than all available. For general usage we have
4027 found a stripe count of 1 or 2 works best.
4029 Severity : enhancement
4030 Description: Add support for compiling against Cray portals.
4031 Details : Conditional compiling for some areas that are different
4035 Frequency : occasional
4036 Bugzilla : 6409, 6834
4037 Description: Creating files with an explicit stripe count may lead to
4038 a failed assertion on the MDS
4039 Details : If some OSTs are full or unavailable, creating files may
4040 trigger a failed assertion on the MDS. Now, Lustre will
4041 try to use other servers or return an error to the
4045 Frequency : occasional
4047 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4048 caused lock timeout during readahead (since 1.4.2).
4049 Details : Processes doing readahead might match a lock that hasn't been
4050 granted yet if there are overlapping and conflicting lock
4051 requests. The readahead process waits on ungranted lock
4052 (original lock is CBPENDING), while OST waits for that process
4053 to cancel CBPENDING read lock and eventually evicts client.
4055 Severity : enhancement
4057 Description: Initial enabling of flock support for clients
4058 Details : Implements fcntl advisory locking and file status functions.
4059 This feature is provided as an optional mount flag (default
4060 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4061 locking are implemented yet, and those that are are not guaranteed
4062 to be completely correct in production environments.
4063 mount -t lustre -o [flock|noflock] ...
4066 Frequency : occasional
4068 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4069 assertion in the ext3 JBD (journaling) layer.
4070 Details : The b_committed_data struct is protected by the big kernel lock
4071 in 2.4 kernels, serializing journal_commit_transaction() and
4072 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4073 access to this struct. In 2.6 kernels there is finer grained
4074 locking to improve SMP performance of the JBD layer.
4078 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4083 Description: merge and cleanup kernel patches.
4086 -----------------------------------------------------------------------------
4088 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4090 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4091 versions of Lustre. Please update all servers and clients to
4092 version 1.4.2 or later at the same time. You must also run
4093 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4094 to update the configuration logs.
4096 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4097 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4098 - swab llog records of type '0' so we get proper header size/idx (5861)
4099 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4100 - fix rename of one directory over another leaking an inode (5953)
4101 - avoid SetPageDirty on 2.6 (5981)
4102 - don't re-add just-being-destroyed locks to the waiting list (5653)
4103 - when creating new directories, inherit the parent's custom
4104 striping settings if present parent (3048)
4105 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4106 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4107 - don't hold client locks on temporary worklist from l_lru (5666)
4108 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4109 - automatically evict dead clients (5921)
4110 - Update file size properly in create+truncate+fstat case (6196)
4111 - Do not unhash mountpoint dentries, do not allow removal of
4113 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4114 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4115 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4116 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4117 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4118 - Correctly handle reads of files with no objects (6243)
4119 - lctl recover will also mark a device active if deactivate used (5933)
4121 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
4122 - allow --write-conf on an MDS with different nettype than client (5619)
4123 - don't write config llogs to MDS for mounts not from that MDS (5617)
4124 - lconf should create multiple TCP connections from a client (5201)
4125 - init scripts are now turned off by default; run chkconfig --on
4126 lustre and chkconfig --on lustrefs to use them
4127 - upcalls are no longer needed for clients to recover to failover
4129 - add --abort-recovery option to lconf to abort recovery on device
4131 - add support for an arbitrary number of OSTs (3026)
4132 - Quota support protocol changes.
4133 - forward compatibility changes to wire structs (6007)
4134 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
4135 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
4136 - improved "lustre status" script
4137 - initialize blocksize for non-regular files (6062)
4138 - added --disable-server and --disable-client configure options (5782)
4139 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
4140 - Vanilla 2.4.29 support
4141 - increase maximum number of obd devices to 520 (6242)
4142 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4143 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4144 SLES 9 2.6.5 kernels
4145 - limit stripes per file to 160 (the maximum EA size) (6093)
4147 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4150 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4151 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4152 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4153 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4154 - small ext3 extents cleanups and fixes (5733)
4155 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4156 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4157 - avoid needless client->OST connect, fix handle mismatch (5317)
4158 - fix DLM error path that led to out-of-sync client, long delays (5779)
4159 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4160 - fix several locking issues related to i_size (5492,5624,5654,5672)
4161 - don't move pending lock onto export if it is already evicted (5683)
4162 - fix kernel oops when creating .foo in unlinked directory (5548)
4163 - fix deadlock in obdfilter statistics vs. object create (5811)
4164 - use time_{before,after} to avoid timer jiffies wrap (5882)
4165 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4166 - Do not leak request if server was not able to process it (5154)
4167 - If mds_open unable to find parent dir, make that negative lookup(5154)
4168 - don't create new directories with extent-mapping (5909, 5936)
4170 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
4171 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
4172 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
4173 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
4174 - support for RHEL4 kernel on i686 (5773)
4175 - provide error messages when incompatible logs are encountered (5898)
4177 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
4178 * version 1.4.0.10 (1.4.1 release candidate 1)
4180 - don't keep a lock reference when lock is not granted (4238)
4181 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4182 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
4183 - drop import inflight refcount on signal_completed_replay error (5255)
4184 - unlock page after async write error during send (3677)
4185 - handle missing objects in filter_preprw_read properly (5265)
4186 - no transno return for symlink open, don't save no-trasno open (3440)
4187 - don't try to complete elan receive that already failed (4012)
4188 - free RPC server reply state on error (5406)
4189 - clean up thread from ptlrpc_start_thread() on error (5160)
4190 - readahead could read extra page into cache that wasn't ejected (5388)
4191 - prevent races in class_attach/setup/cleanup/detach (5260)
4192 - don't dereference de->d_inode after l_dput of de (5458)
4193 - use "int" for stripe value returned from lock_to_stripe (5544)
4194 - mballoc allocation and error-checking fixes in 2.6 (5504)
4195 - block device patches to fix I/O request sizes in 2.6 (5482)
4196 - look up hostnames for IB nals (5602)
4197 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
4198 - don't start multiple acceptors for the same port (5277)
4199 - fix incorrect LASSERT in mds_getattr_name (5635)
4200 - export a proc file for general "ping" checking (5628)
4201 - fix "lfs check" to not block when the MDS is down (5628)
4203 - service request history (4965)
4204 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4205 - create an "evict_client" /proc entry on OSTs, like the MDS has
4206 - fix mount usage message, return errors per mount(8) (5168)
4207 - change grep [] to grep "[]" in tests so they work in more UMLs
4208 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4209 - remove ancient LOV_MAGIC_V0 EA support (5047)
4210 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4211 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4212 - allow readahead allocations to fail when low on memory (5383)
4213 - mmap locking landed again, after considerable improvement (2828)
4214 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4216 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4219 - send OST transaction number in read/write reply to free req (4966)
4220 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4221 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4222 - the watchdog thread now runs as interruptible (5246)
4223 - flock/lockf fixes (but it's still disabled, pending 5135)
4224 - don't use EXT3 constants in llite code (5094)
4225 - memory shortage at startup could cause assertion (5176)
4227 - reorganization of lov code
4228 - single portals codebase
4230 - add extents/mballoc support (5025)
4231 - direct I/O reads in the obdfilter (4048)
4232 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4234 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4237 - send OST transaction number in read/write reply to free req (4966)
4238 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4239 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4240 - the watchdog thread now runs as interruptible (5246)
4241 - handle missing objects in filter_preprw_read properly (5265)
4242 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4243 - drop import inflight refcount on signal_completed_replay error (5255)
4244 - unlock page after async write error during send (3677)
4245 - return original error code on reconstructed replies (3761)
4246 - no transno return for symlink open, don't save no-trasno open (3440)
4248 - add pid to ldlm debugging output (4922)
4249 - bump the watchdog timeouts -- we can't handle 30sec yet
4250 - extra debugging for orphan dentry/inode bug (5259)
4252 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
4255 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4256 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4257 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
4258 - fix NULL dereference in /proc/sys/portals/routes (4827)
4259 - allow failed mdc_close() operations to be interrupted (4561)
4260 - stop precreate on OST before MDS would time out on it (4778)
4261 - don't send partial-page writes before EOF from client (4410)
4262 - discard client grant for sub-page writes on large-page clients (4520)
4263 - don't free dentries not owned by NFS code, check generation (4806)
4264 - fix lsm leak if mds_create_objects() fails (4801)
4265 - limit debug_daemon file size, always print CERROR messages (4789)
4266 - use transno after validating reply (3892)
4267 - process timed out requests if import state changes (3754)
4268 - update mtime on OST during writes, return in glimpse (4829)
4269 - add mkfsoptions to LDAP (4679)
4270 - use ->max_readahead method instead of zapping global ra (5039)
4271 - don't interrupt __l_wait_event() during strace
4273 - add software watchdogs to catch hung threads quickly (4941)
4274 - make lustrefs init script start after nfs is mounted
4275 - fix CWARN/ERROR duplication (4930)
4276 - return async write errors to application if possible (2248)
4277 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4278 - print NAL number in %x format (4645)
4279 - update barely-supported suse-2.4.21-171 series (4842)
4280 - support for sles 9 %post scripts
4281 - support for building 2.6 kernel-source packages
4282 - support for sles km_* packages
4284 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4287 - ignore -ENOENT errors in osc_destroy (3639)
4288 - notify osc create thread that OSC is being cleaned up (4600)
4289 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4290 - reconstruct ost_handle() like mds_handle() (4657)
4291 - create a new thread to do import eviction to avoid deadlock (3969)
4292 - let lconf resolve symlinked-to devices (4629)
4293 - don't unlink "objects" from directory with default EA (4554)
4294 - hold socknal file ref over connect in case target is down (4394)
4295 - allow more than 32000 subdirectories in a single directory (3244)
4296 - fix blocks count for O_DIRECT writes (3751)
4297 - OST returns ENOSPC from object create when no space left (4539)
4298 - don't send truncate RPC if file size isn't changing (4410)
4299 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4300 - bind to privileged port in socknal and tcpnal (3689)
4302 - rate limit CERROR/CWARN console message to avoid overload (4519)
4303 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4304 - basic mmap support (3918)
4305 - kernel patch series update from b1_4 (4711)
4307 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4310 - avoid crash during MDS cleanup with OST shut down (2775)
4311 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4312 - don't use bad inodes on the MDS (3744)
4313 - dynamic object preallocation to improve recovery speed (4236)
4314 - don't hold spinlock over lock dumping or change debug flags (4401)
4315 - don't zero obd_dev when it is force cleaned (3651)
4316 - print grants to console if they go negative (4431)
4317 - "lctl deactivate" will stop automatic recovery attempts (3406)
4318 - look for existing locks in ldlm_handle_enqueue() (3764)
4319 - don't resolve lock handle twice in recovery avoiding race (4401)
4320 - revalidate should check working dir is a directory (4134)
4322 - don't always mark "slow" obdfilter messages as errors (4418)
4324 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4327 - don't close LustreDB during write_conf until it is done (3860)
4328 - fix typo in lconf for_each_profile (3821)
4329 - allow dumping logs from multiple threads at one time (3820)
4330 - don't allow multiple threads in OSC recovery (3812)
4331 - fix debug_size parameters (3864)
4332 - fix mds_postrecov to initialize import for llog ctxt (3121)
4333 - replace config semaphore with spinlock (3306)
4334 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4335 - don't allow enqueue to complete on a destroyed export (3822)
4336 - down write_lock before checking llog header bitmap (3825)
4337 - recover from lock replay timeout (3764)
4338 - up llog sem before sending rpc (3652)
4339 - reduce ns lock hold times when setting kms (3267)
4340 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4341 - fix NULL deref and obd_dev leak on setup error (3312)
4342 - replace some LBUG about llog ops with error handling (3841)
4343 - don't match INVALID dentries from d_lookup and spin (3784)
4344 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4345 - fix invalid assertion in ptlrpc_set_wait (3880)
4347 - add libwrap support for the TCP acceptor (3996)
4348 - add /proc/sys/portals/routes for non-root route listing (3994)
4349 - allow setting MDS UUID in .xml (2580)
4350 - print the stack of a process that LBUGs (4228)
4352 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4355 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4356 - make sure to unset replay flag from failed open requests (3440)
4357 - if default stripe count is 0, use OST count for inode size (3636)
4358 - update parent mtime/ctime on client for create/unlink (2611)
4359 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4360 - free recovery state on server during a forced cleanup (3571)
4361 - unregister_reply for resent reqs (3063)
4362 - loop back devices mounting and status check on 2.6 (3563)
4363 - fix resource-creation race that can provoke i_size == 0 (3513)
4364 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4365 - more debugging for page-accounting assertion (3746)
4366 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4367 - avoid deadlock after precreation failure (3758)
4368 - fix race and lock order deadlock in orphan handling (3450, 3750)
4369 - add validity checks when grabbing inodes from l_ast_data (3599)
4371 - add /proc/.../recovery_status to obdfilter (3428)
4372 - lightweight CDEBUG infrastructure, debug daemon (3668)
4373 - change default OSC RPC parameters to be better on small clusters
4374 - turn off OST read cache for files smaller than 32MB
4375 - install man pages and include them in rpms (3100)
4376 - add new init script for (un)mounting lustre filesystems (2593)
4377 - run chkconfig in %post for init scripts (3701)
4378 - drop scimac NAL (unmaintained)
4380 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4383 - clean kiobufs before and after use (3485)
4384 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4385 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4386 - change init script to fail more clearly if not run as root (1528)
4387 - allow clients to reconnect during replay (1742)
4388 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4389 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4390 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4391 - fix cond_resched (3554)
4392 - only evict unfinished clients after recovery (3515)
4393 - allow bulk resend, prevent data loss (3570)
4394 - dynamic ptlrpc request buffer allocation (2102)
4395 - don't allow unlinking open directory if it isn't empty (2904)
4396 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4397 - remove extraneous obd dereference causing LASSERT failure (3334)
4398 - don't use get_cycles() when creating temp. files on the mds (3156)
4399 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4400 - handle EEXIST for set-stripe, set proper directory name (3336)
4402 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4403 - fix ksocknal_fmb_callback() error messages (2918)
4405 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4408 - don't copy lvb into (possibly NULL) reply on error (2983)
4409 - don't deref dentry after dput, don't free lvb on error (2922)
4410 - use the kms to determine writeback rpc length (2947)
4411 - increment oti_logcookies when osc is inactive (2948)
4412 - update client's i_blocks count via lvb messages (2543)
4413 - handle intent open/close of special files properly (1557)
4414 - mount MDS with errors=remount-ro, like obdfilter (2009)
4415 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4416 - don't use cancelling-locks' kms values (2947)
4417 - use highest lock extent for kms, not last one (2925)
4418 - don't dereference ERR_PTR() dentry in error handling path (3107)
4419 - fix thread race in portals_debug_dumplog() (3122)
4420 - create lprocfs device entries at setup instead of at attach (1519)
4421 - common AST error handler, don't evict client on completion race (3145)
4422 - zero nameidata in detach_mnt in 2.6 (3118)
4423 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4424 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4425 - initialize RPC timeout timer earlier for 2.6 (3219)
4426 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4427 - print nal/nid for unknown nid (3258)
4428 - additional checks for oscc recovery before doing precreate (3284)
4429 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4430 - don't crash in mdc_close for bad permissions on open (3285)
4431 - zero i_rdev for non-device files (3147)
4432 - clear page->private before handing to FS, better assertion (3119)
4433 - tune the read pipeline (3236)
4434 - fix incorrect decref of invalidated dentry (2350)
4435 - provide read-ahead stats and refine rpc in flight stats (3328)
4436 - don't hold journal transaction open across create RPC (3313)
4437 - update atime on MDS at close time (3265)
4438 - close LDAP connection when recovering to avoid server load (3315)
4439 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4440 - don't leak open file on MDS after open resend (3325)
4441 - serialize filter_precreate and filter_destroy_precreated (3329)
4442 - loop device shouldn't call sync_dev() for nul device (3092)
4443 - clear page cache after eviction (2766)
4444 - resynchronize MDS->OST in background (2824)
4445 - refuse to mount the same filesystem twice on same mountpoint (3394)
4446 - allow llmount to create routes for mounting behind routers (3320)
4447 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4448 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4449 - fix rare problem with rename on htree directories (3417)
4451 - allow default OST striping configuration per directory (1414)
4452 - fix compilation for qswnal for 2.6 kernels (3125)
4453 - increase maximum number of MDS request buffers for large systems
4454 - change liblustreapi to be useful for external progs like lfsck (3098)
4455 - increase local configuration timeout for slow disks (3353)
4456 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4458 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4461 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4462 - don't overwrite extent policy data in reply if lock was blocked (2901)
4463 - drop filter export grants atomically with removal from device (2663)
4464 - del obd_self_export from work_list in class_disconnect_exports (2908)
4465 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4466 - swab reply message in mdc_close, other PPC fixes (2464)
4467 - fix destroying of named logs (2325)
4468 - overwrite old logs when running lconf --write_conf (2264)
4469 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4470 - fix race in target_handle_connect (2898)
4471 - mds_reint_create() should take same inode create lock (2926)
4472 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4473 - don't close files for self_export to avoid uninitialized obd (2936)
4474 - allow MDS with the same name as client node (2939)
4475 - hold dentry reference for closed log files for unlink (2325)
4476 - reserve space for all logs during transactions (2059)
4477 - don't evict page beyond end of stripe extent (2925)
4478 - don't oops on a deleted current working directory (2399)
4479 - handle hard links to targets without a parent properly (2517)
4480 - don't dereference NULL lock when racing during eviction (2867)
4481 - don't grow lock extents when lots of conflicting locks (2919)
4483 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4486 - account for cache space usage on clients to avoid data loss (974)
4487 - lfsck support in lustre kernel code (2349)
4488 - reduce journal credits needed for BRW writes (2370)
4489 - orphan handling to avoid losing space on client/server crashes
4490 - ptlrpcd can be blocked, stopping ALL progress (2477)
4491 - use lock value blocks to assist in proper KMS, faster stat (1021)
4492 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4493 - recovery for initial connections (2355)
4494 - fixes for mds_cleanup_orphans (1934)
4495 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4496 - block all file creations until orphan recovery completes (1901)
4497 - client remove rq_connection from request struct (2423)
4498 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4499 - recovery timer race (2670)
4500 - mdc_close recovey bug (2532)
4501 - ptlrpc cleanup bug (2710)
4502 - mds timeout on local locks (2588)
4503 - namespace lock held during RPCs (2431)
4504 - handle interrupted sync write properly (2503)
4505 - don't try to handle a message that hasn't been replied to (2699)
4506 - client assert failure during cleanup after abort recovery (2701)
4507 - leak mdc device after failed mount (2712)
4508 - ptlrpc_check_set allows timedout requests to complete (2714)
4509 - wait for inflight reqs when ptlrpcd finishes (2710)
4510 - make sure unregistered services are removed from the srv_list
4511 - reset bulk XID's when resending them (caught by 1138 test)
4512 - unregister_bulk after timeout
4513 - fix lconf error (2694)
4514 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4515 - readahead locks pages, leaves pending causing memory pressure (2673)
4516 - increase OST request buffers to 4096 on large machines (2729)
4517 - fix up permission of existing directories in simple_mkdir (2661)
4518 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4519 - don't assign transno to errored transactions (2742)
4520 - don't delete objects on OST if given a bogus objid from MDS (2751)
4521 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4522 - if rq_no_resend, then timeout request after recovery (2432)
4523 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4524 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4525 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4526 - MDS recovery shouldn't delete objects that it has given out (2730)
4527 - if enqueue arrives after completion, don't clobber LVB (2819)
4528 - don't unlock pages twice when trigger_group_io returns error (2814)
4529 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4530 - don't write pages to disk if there was an error (1450)
4531 - don't ping imports that have recovery disabled (2676)
4532 - take buffered bytes into account when balancing socknal conn (2817)
4533 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4534 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4535 - remove little-endian swabbing of llog records (1987)
4536 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4537 - retry reposting request buffers if they fail (1191)
4538 - grow extent at grant time to avoid granting a revoked lock (2809)
4539 - lock revoke doesn't evict page if covered by a second lock (2765)
4540 - disable VM readahead to avoid reading outside lock extents (2805)
4542 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4543 - updated kernel patches for hp-2.4.20 kernel (2681)
4545 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4548 - fix truncated write corruption (2366)
4549 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4551 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4552 - ensure len(uuid) < 37 in lmc (1171)
4553 - fix ia64 OOPS in llog_test (2255)
4554 - zero end of page at obdfilter for partial page writes (2648)
4555 - don't leave stale dentries around after renames (bug 2428)
4556 - fix timeouts when evicting a client with a single lock held (2642)
4557 - set deadline for the initial HELLO message to drain (2634)
4558 - print out dotted-quad IP addresses in the socknal (2302)
4560 - additional debugging for MDS client eviction problem (2443)
4561 - fix mkfsoptions support for osts (2603, 2604)
4563 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4566 - add series for the vanilla 2.6.0 kernel
4567 - add series for the vanilla 2.4.24 kernel
4568 - add series for a cray x86/64 UL kernel drop
4569 - fix xattr patches for the vanilla 2.4.19 series
4571 - generate true UUIDs in lmc (1171)
4572 - have portals stack dumping break in UML (2466)
4573 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4574 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4575 - wake callback waiting threads on client eviction (2460)
4576 - Add --ptldebug and --subsystem to lmc (1719)
4577 - update assertion to allow safe interrupt allocation
4578 - set rq_no_resend for cancel requests (2432)
4579 - recalculate ptlrpcd timeout after resend (2494)
4580 - call vfs_rmdir when removing pending directories (2368)
4581 - fix renaming a file to itself (2429)
4582 - lmc creates a default one-stripe lov (2454)
4583 - expand procfs space to handle large clusters (2326)
4584 - increase UML stack to avoid overflow
4585 - update lconf's list of debug and subsystem masks
4586 - fix lfs find --obd (2510)
4587 - /proc tunable for disabling filter read caching (2591)
4588 - stop rpm packages from altering slapd.conf (2301)
4589 - disable nagle in the socknal under 0conf (2578)
4590 - choose mds inode size based on stripe count (2572)
4591 - fix kernel-source rpm problems (2516)
4593 - add --disable-doc to avoid pdf generation (2421)
4594 - update documentation, tests, type-os, comments
4595 - avoid format warnings on ia64
4596 - remove the TOE NAL
4597 - tiny code cleanups by removing unused fields
4599 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4602 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4603 - remove the most blatant lies from BUILDING file (2371)
4604 - change default debug level to reasonable production setting
4605 - reduce client side cache size to reduce cache flush time
4606 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4607 - make TCP zerocopy and pinger support enabled by default (2476)
4608 - sync writes completed after process exits caused crashes (2319)
4609 - maintain correct mount count on the MDS (2356)
4610 - backout 1557, because 2316 wasn't really fixed
4611 - better file I/O statistics gathering in /proc
4612 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4613 - another kernel patch to fix zero-copy TCP function export
4614 - don't take duplicate lock when processing re-sent getattr (2420)
4615 - lctl uses obd_self_export instead of creating new conn (2353)
4616 - MDS/OST recovery case which requires object creation asserted (2425)
4617 - move lfs from /usr/sbin to /usr/bin in packages
4618 - fix race between mds_client_add and mds_client_free (2417)
4619 - use kmalloc instead of slabs in portals (2430)
4620 - don't create duplicate records when a failover MDS is present (2442)
4621 - remove unnecessary mount age check (2332)
4622 - don't remove directory inodes from locks prematurely (2451)
4623 - don't break if MDS service name is the same as hostname (2103)
4624 - fix races in client write RPC generation when cache full (2482)
4626 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4629 - remove now-unused request->rq_obd (278)
4630 - if an allocation fails, print out how much memory we've used (1933)
4631 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4632 - add the "configurable stack size" patch to most series files (1256)
4633 - ability to write large log records, for 100+ OST configs (2306)
4634 - fix NULL deref when filter_prep fails (2314)
4635 - fix operator precedence error in filter_sync
4636 - dynamic allocation of socknal TX descriptors (2315)
4637 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4638 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4639 - ensure that CWARN messages in Portals always get to the syslog
4640 - __init/__exit are not for prototype decls (ldlm_init/exit)
4641 - x86-64 compile warning fixes
4642 - fix gateway LMC keyword conflict (2318)
4643 - fix MDS lock inversions in getattr/reint paths (1844)
4644 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4645 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4646 - DLM race condition prevented some lock evictions (2328)
4647 - ENOMEM detection and retry on socknal sends (2230)
4648 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4649 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4650 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4651 - better error messages when a client is rejected during recovery (1505)
4652 - avoid cancelling locks which were never granted, after failure (2330)
4653 - fix i_sem/journal inversion in mds_client_add (2333)
4654 - fix truncate/getattr lock cycle deadlock (2334)
4655 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4656 - fix two rare exit paths which could leak an l_lock() ref (2321)
4657 - fencepost error in MDS/OST orphan recovery (2226)
4658 - make log record alignment 8 bytes (1988)
4659 - lstripe now fails when requested offset > ost_count (2237)
4660 - ensure that all kernel series have a complete list.h (1607)
4661 - fix crashes in special-file operations (2316)
4662 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4664 - allow configurable automake binary, for testing new versions
4665 - small update to the lfs documentation
4667 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4669 * fix negative export reference count in fsfilt_sync (2312)
4671 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4672 * release candidate 0.9.1
4674 - orphans are moved into the PENDING directory for possible recovery
4675 - replayed opens now open by fid for orphan/rename safety (1042)
4676 - last close of an orphan inode generates a transno (683)
4677 - chdir() and mount() now pin the directory entry (1020)
4678 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4679 - discard very old requests without processing them (1502)
4680 - remove some common, well-understood CERRORs (1505)
4681 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4682 - clear "grant" flags in OST replies until OST grant code lands (1644)
4683 - fix read performance by not clobbering i_blksize on client (1598)
4684 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4685 - make LRU size a /proc tunable, clears locks when reduced (707)
4686 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4687 - server locks take references on exports now (1558)
4688 - build fixes for 2.4.20-rh trees (1663)
4689 - return an error from lov_create if all OSCs are inactive (1751)
4690 - fix import levels when a reconnect happens without a timeout (1597)
4691 - exit early from mds_open if we get a lookup error (1749)
4692 - partial page read at EOF wouldn't wait for disk before sending (1642)
4693 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4694 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4695 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4696 - write/truncate lock inversion (1639)
4697 - Don't auto-load obdclass, portals modules during cleanup (1495)
4698 - fix timestamps from jumping to "now" (1763)
4699 - extra journal assertions (1648)
4700 - add an extra multiunlink test (1771)
4701 - fix read_record/write_record API (1776)
4702 - fix leak of offset_extent, possible incorrect i_size later (1772)
4703 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4704 - Debugging for the kqswnal_get_idle_tx problems (1820)
4705 - Allow recovery to be attempted multiple times (1536)
4706 - Write out MDS last_rcvd file after it is first created (1600)
4707 - Fix tx_descriptor leak in failed transmit situations (1827)
4708 - ext3 journaling fixes for assertion failure after IO error (1871)
4709 - class_export_put() on freed export after completion AST error (1896)
4710 - Fix revalidate looping in VFS (1322)
4711 - Don't access a freed export during MDS_REINT timeout (1521)
4712 - Add open-unlink recovery support on the MDS (1673,1764)
4713 - Return an error if no MDS data was read from last_rcvd (1946)
4714 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4715 - Don't setup a disk device that doesn't match exported UUID (317)
4716 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4717 - avoid committing NULL handle in force close
4718 - local.sh is now a one-stripe LOV configuration
4719 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4720 - fix ext3 htree duplicate directory entry corruption (1516)
4721 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4722 - update to new LOV EA format (2097)
4723 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4725 2003-06-15 Phil Schwan <phil@clusterfs.com>
4728 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4729 - per-import/export recovery handling (958, 931, 959)
4730 - multiple last-rcvd slots, for serving multiple FSes (949)
4731 - connections are again shared between multiple imp/exports (963, 964)
4732 - "umount -f" would hang if any requests needed to be sent (393, 978)
4733 - avoid pinning large req buffer by copying for queued messages (989)
4734 - add "uuid" to "lctl device" command to help upcalls (991)
4735 - "open" RPCs with transnos would confuse recovery counters (1037)
4736 - do proper endian conversion of all wire messages (288, 340, 891)
4737 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4738 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4739 - fix ptlrpc_connection leak in target_handle_connect (1174)
4740 - fix import refcounting bug in OST and MDS cleanup (1134)
4741 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4742 - fix very unlikely obd_types race condition (501)
4743 - remove osc_open hack for echo_client (1187)
4744 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4745 - a failure in read_inode2 leads to deadlock (1139)
4746 - cancel ack-locks as soon as transaction is committed (1072)
4747 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4748 - make sure to commitrw after any preprw to avoid deadlock (1162)
4749 - failing to execute a file in a lustre FS would lock inode (1203)
4750 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4751 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4752 - obd_uuid2tgt would walk off the end of the list (1255)
4753 - on IA64 the osc would give portals incorrect bulk size (1258)
4754 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4755 - fix lock inversion caused by new llite matching code (1282)
4756 - limit the number of dirty pages on a client to 10MB (1286)
4757 - timed out locks were not being corrected cancelled (1289)
4758 - fix O_DIRECT above 4GB on IA-32 (1292)
4759 * major user-visible changes
4760 - fail out/fail over policy now controlled by the upcall (993)
4762 - add OBD_PING to check server availability and failure (954)
4763 - lustre messages are now sent in sending host order (288, 340, 891)
4764 - add eadatalen to MDS getattr reply (340)
4765 - OST read replies may contain second buffer, with per-page status (593)
4767 2003-03-11 Phil Schwan <phil@clusterfs.com>
4770 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4771 - failed open()s could cause deadlock; fixed (867, 869)
4772 - stop cancelling OST locks when files are closed (481)
4773 - overlapping XID spaces caused network corruption (851, 853)
4774 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4775 - setattr_raw added, to do single-RPC, server-side setattrs
4776 - lmc/lconf syntax change for OST UUIDs
4777 - fix crashy race condition between ptlrpc_free_req and osc_close
4778 - don't use request in mdc_enqueue if we hit a timeout (889)
4779 - don't set the inode i_size for regular files from the MDS (896)
4780 - handle out of order completion AST (842)
4781 - don't LBUG if a lock request times out after receiving AST (913)
4782 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4783 - if a bad lock AST arrives, send an error instead of dropping entirely
4784 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4785 - fix leak in bulk IO when only partially completed (899, 900, 926)
4786 - fix O_DIRECT for ia64 (55)
4787 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4788 - C-z after timeout could hang a process forever; fixed (977)
4790 - client-side I/O cache (678, 924, 929, 941, 970)
4792 - READPAGE and SETATTRs which don't take server-side locks get
4795 2003-02-11 Phil Schwan <phil@clusterfs.com>
4798 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4799 - Fix fsfilt_extN_readpage() to read a full page of directory
4800 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4801 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4802 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4803 - Most intent operations were removed, in favour of a new RPC mode
4804 that does a single RPC to the server and bypasses most of the VFS
4805 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4806 - Aggressively cancel local locks on DLM servers
4807 - mds_reint_unlink sends EA to the client if it's the last nlink.
4808 client uses that EA to unlink OST objects.
4809 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4810 - recursive symlinks were fixed (439)
4811 - fixed NULL deref in DEBUG_REQ
4812 - filter_update_lastobjid no longer calls sync, which annoyed extN
4813 - fixed multi-client small-writes to a single file problem (445)
4814 - fixed mtime updates during file writes (607)
4815 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4816 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4817 - fixed Linux OST size reporting problem (444, 656)
4818 - OST now updates object mtime with writes or setattr (607, 619)
4819 - client verifies file size before zeroing page past EOF (445)
4820 - OST now writes last allocated objid to disk with allocation (108)
4821 - LOV on echo now works (409)
4823 - mds_reint_unlink sends a new buffer, with the EA included. this
4824 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4825 set if a regular file was being unlinked, and it was the last link
4826 - use PtlGet from the target for bulk writes (315)
4827 - OST now updates object mtime with writes or setattr (607, 619)
4828 - LDLM now has a grant-time callback to revalidate locked items, if
4830 - Many MDS operations were reorganized to combat race conditions
4832 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4833 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4835 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4838 - Fully reactivate OST imports after reconnection (512, others)
4839 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4840 - More graceful error handling for truncating on dead OST (515)
4841 - Don't error out unless we're actually accessing dead stripes (474)
4842 - Fix garbage sizes when stripes are missing (410)
4843 - LRU counters were broken, causing constant lock purge (433, 432)
4844 - garbage on read from stripes with failed OSTs (441)
4845 - mark OSCs as active before reconnecting during recovery (438)
4846 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4847 - lfind did not preserve OST order in output (443)
4848 - symlinks cause hung clients, incorrect data (439)
4849 - stop dereferencing request after dropping refcount (457)
4850 - don't LASSERT(spin_is_locked) on non-SMP (455)
4851 - fixes for many rename() bugs
4852 - fstat didn't correctly synchronize attributes (399)
4853 - server must handle lock cancellation during blocking AST prep (487)
4854 - bulk descriptors were free()d too soon (511)
4855 - fix paths in lconf, which would load incorrect modules (451, 507)
4856 - fix confusing lconf 'host not found' error message (386)
4857 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4858 - fix race condition in mdc_blocking_ast() for inode access (526)
4859 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4860 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4861 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4862 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4863 - fix lstripe in several ways (kernel side) (527)
4864 - fix request leak in ldlm_cli_enqueue (262)
4865 - incorrect OSC was marked inactive after OST failure
4866 - call mds_fs_cleanup before unmounting filesystem (524)
4867 - fix races between taking ns_lock and ldlm_lock_change_resource
4868 - fix races updating LOV export open file list
4869 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4870 - fix recovery NULL deref in ldlm_cli_cancel_unused
4871 - fix some DLM races by using new hash table for lock handles (419)
4872 - permit the client to specify desired inodes, at replay
4873 - duplicate requests when we queue them for replay reintegration
4874 - fix last_rcvd offset calculation
4875 - sync after each recovered transaction, so we always make progress
4876 - never, not always, ERESTART requests without transnos
4877 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4879 - skip replay if the MDS says that the client is already connected
4880 - don't check for a recovery-enabled export to match lctl's UUID
4881 - don't INC_USE_COUNT for phantom exports
4882 - don't crash when cleaning up phantom exports (567)
4883 - don't double-finish or set replay data for errored mdc_open requests
4884 - abort requests when they time out, so we don't get old replies
4885 - send/receive replies for AST messages again
4886 - if the client says that it doesn't have the lock, cancel it on the
4888 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4889 mark it as destroyed, it will all work out when decref is called
4890 - fix module use counts (22, 581)
4892 - ASTs now expect a reply (server cancels lock on error reply)
4894 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4897 - fix many simultaneous client startup (392)
4898 - fix dentry->d_it clobbering
4899 - credentials weren't being shipped for readdir/getattr operations
4900 - remove invalid assertions triggered during some concurrent MD
4902 - proper Lustre versions added (336, 389)
4903 - fix memory leak for create error case (398)
4904 - fix LOV locking bug that would get cli/srv out of sync
4905 - fix echo client over LOV (409)
4906 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4907 - fix double-O_EXCL intent crash (424)
4908 - avoid sending multiple lock CANCELs (352)
4910 - MDS can do multi-client recovery (modulo bugs in new code)
4912 - many updates, edits, cleanups
4914 2002-11-18 Phil Schwan <phil@clusterfs.com>
4917 - fix null d_it dereference (346)
4918 - fix full OST/dbench hang (333)
4919 - fix permission problem with file removal (286)
4920 - fix removal of OSCs from LOV when they fail
4921 - fix NULL deref during bulk timeout (214)
4922 - fix problems related to multiple filesystems on one MDS (241)
4923 - fixed serious subtle metadata locking bugs
4924 - free locks on clients when inodes are removed due to memory
4926 - fix inode pointer in lock data (285)
4927 - partial support for multiple MDS on a single host (241)
4928 - data locks weren't cancelled at clear_inode time (290, 311)
4929 - intent locks could lead to unbounded lock growth (205)
4930 - added a maximum lock count, an LRU list, and a flusher
4931 - fix multiple rename (365)
4932 - properly abstracted the echo client
4933 - OSC locked 1 byte too many; fixed
4934 - rewrote brw callback code:
4935 - fixed recovery bugs related to LOVs (306)
4936 - fixed too-many-pages-in-one-write crash (191)
4937 - fixed (again) crash in sync_io_timeout (214)
4938 - probably fixed callback-related race (385)
4940 - Add capability to MDS protocol
4941 - LDLM cancellations and callbacks on different portals
4943 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4946 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4947 - defer open object destruction to close time (601981, 138)
4948 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4949 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4950 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4951 - don't call obd_setattr in ll_file_release for destroyed objects
4952 * protocol change to lustre_msg: move |version| and add |flags|
4953 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4954 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4955 * added replay of create, unlink, link and rename operations during
4956 MDS failover; recovery should be much more robust now
4957 * remove failed OSCs from LOVs (only lov_create uses this so far)
4958 * the lustre-HOWTO was brought (more) up to date (582544)
4960 2002-10-23 Phil Schwan <phil@clusterfs.com>
4963 - in-use dentries weren't being reused properly (617851)
4964 - prevent multiple LDLM setup (599178)
4965 - fix LOV size calculations for truncate (617853)
4966 - fix client handling of MDS intent errors (POSIX)
4967 - fix permission bug in lovstripe.c test (624321)
4968 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4969 - truncate past end of file could corrupt data
4970 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4971 - a race in recovery could return ETIMEDOUT to apps (623947)
4972 - building outside the source directory was fixed
4973 * the lustre-HOWTO was brought (more) up to date (582544)
4974 * major progress was made on recovery functionality
4976 2002-10-10 Phil Schwan <phil@clusterfs.com>
4979 - recovery deadlock fix
4980 - rm -rf causes LBUG fix (617817)
4981 - file open by multiple tasks fix (618962)
4982 - directory permissions bugs (602707 and 620007)
4983 - journal_stop fixed with locking (611313)
4984 - O_APPEND failures resolved (618273, perhaps 614459)
4985 - lconf PATH fix (619770)
4986 - IA64 build fix (621450)
4987 - RPC buffer sizes scale with amount of memory
4989 2002-10-01 Phil Schwan <phil@clusterfs.com>
4992 - locks would be cancelled without throwing away data pages,
4993 resulting in inconsistent data (605627)
4994 - inode attributes were not always being refreshed (605627, 612449)
4995 - lconf now continues to cleanup after lctl reports an error
4996 - MDS now enforces user permissions (602707)
4997 - lprocfs cleanup fixed, but not yet enabled (614157)
4998 - fixed infinite server hang, should a client not respond to an AST
4999 - avoid going into recovery if user calls readlink() with a buffer
5000 that's too small (613941)
5001 - AST RPCs no longer require replies (614867) -- this may be changed
5002 - don't crash server if client sends an IOV that's too big (611336)
5003 - fixed lock conversion deadlock (611892)
5004 - fixed the following of symlinks (614622)
5005 * recovery: the server can remove locks from a client that dies, other
5006 clients can make progress
5007 * more extN patch fixes
5008 * compile-time configurable ptlrpc buffer allocations
5010 - collaborative read cache document
5011 - Lustre Lite Performance CDR document-in-progress
5013 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5016 - fix typo in patch-2.4.18
5018 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5021 - clear ptlrpc request each time in handle_incoming_request()
5022 - unlink of files now destroys the object on the OST
5024 2002-09-19 Peter Braam <braam@clusterfs.com>
5026 * add hard link support
5027 * change obdfile creation method
5028 * kernel patch changed
5030 2002-09-19 Peter Braam <braam@clusterfs.com>
5033 - stack overflow bug in extN fixed
5035 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
5037 * documentation updates
5038 - add man pages for config tools
5039 - update tests/README to describe testing with new config tools
5040 - finish metadata API descriptions
5041 * bug fixes and cleanups
5042 - statfs workaround for 16TB limit
5043 - LOV stripe allocation improved, can stripe on subset of OSTs
5044 - LOV file size/IO offset was wrong for files > 4GB in size
5045 - object EA data was being dropped, caused files to be unreadable
5046 - memory overflow with non-LOV OST caused memory corruption
5047 - fixed regression tests to work with new config tools, obdfilter
5048 - fixed bug when directory size became larger than 1 block
5049 - fixed bug (for single client case) when PWD was deleted
5050 - invalidate local directory pages when doing intent-based ops
5051 - avoid LDLM oops when lock callback contained bad data
5053 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5055 * documentation updates
5056 * bug fixes and cleanups
5057 - configuration tools
5060 - 64-bit compile warnings
5061 - 64-bit internal statfs data
5063 * test_brw on persistent OST devices
5065 * lprocfs (disabled)
5067 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5069 * documentation updates
5070 * bug fixes and cleanups
5071 * configuration tools
5073 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5076 * many small fixes to 0_5_4
5077 * io/network handling
5078 * thinkos in MDS operations
5080 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5083 * crucial basic fixes to 0.5.3
5084 * IOR, Iozone work over Elan
5087 2002-08-07 Phil Schwan <phil@clusterfs.com>
5088 * version 0_5_3, our first alpha
5089 * we use the new Portals iovs
5090 * documentation updates
5091 * bug fixes and cleanups
5092 * small changes in the DLM wire protocol
5094 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5095 * version 0_5_1 with some initial stability,
5096 * locking on MD and file I/O.
5097 * documentation updates
5098 * several bug fixes since 0.5.0
5099 * small changes in wire protocol
5101 2002-07-18 Phil Schwan <phil@clusterfs.com>
5103 * delivered as Lustre Light Alpha
5104 * fixed a crash after handling invalid MDS requests
5105 * fixed directory pages for architectures with non-4k pages sizes
5107 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5108 * release version v0_4_4
5109 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5111 * Fixes a number of interruption problems with OST operations.
5112 * Update documentation for portals header changes
5113 * Move all wire protocol structs/defines to lustre_idl.h
5114 * Fixes symlink length bug.
5115 * Add tcpdump to repository.
5117 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5118 * release version v0_4_3
5119 * Fixes statfs for inodes on extN.
5120 * Fixes bug in runtests which would delete /etc/hosts.
5121 * Use 64-bit object IDs wherever possible (not into VFS though)
5122 Remove ost_get_info, which is unused by lustre, and out of date.
5124 2002-07-03 Peter Braam <braam@clusterfs.com>
5125 * release version v0_4_2 Fixes a lookup error (type not passed)
5126 * move forward to head of Portals
5127 * move forward to latest Lustre kernel
5129 2002-06-25 Peter Braam <braam@clusterfs.com>
5130 * release version v0_4_1. Hopefully stable on single node use.