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 Frequency : on recovery
18 Description: I/O failures after umount during fail back
19 Details : if client reconnected to restarted server we need join to recovery
20 instead of find server handler is changed and process self eviction
21 with cancel all locks.
23 Severity : enhancement
25 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
27 Severity : enhancement
29 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
31 Severity : enhancement
33 Description: Upcall on Lustre log has been dumped
34 Details : Allow for a user mode script to be called once a Lustre log has
35 been dumped. It passes the filename of the dumped log to the
36 script, the location of the script can be specified via
37 /proc/sys/lnet/debug_log_upcall.
42 Description: avoid messages about idr_remove called for id which is not allocated.
43 Details : Move assigment s_dev for clustered nfs to end of initialization, for
44 avoid problem with error handling.
49 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
50 Details : When connection is reused this not moved from CONN_UNUSED_HASH into
51 CONN_USED_HASH and this prodice warning when put connection again
54 Severity : enhancement
56 Description: Export bytes_read/bytes_write count on OSC/OST.
60 Description: Early reply size mismatch, MGC loses connection
61 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
62 the connect flags are properly negotiated.
66 Description: Properly propagate oinfo flags from lov to osc for statfs
67 Details : restore missing copy oi_flags to lov requests.
69 Severity : enhancement
71 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
72 : l_getgroups(8), lst(8), routerstat(8)
73 Details : included man pages for llobdstat(8), llstat(8),
74 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
76 Severity : enhancement
78 Description: configure's --enable-quota should check the
79 : kernel .config for CONFIG_QUOTA
80 Details : configure is terminated if --enable-quota is passed but
81 : no quota support is in kernel
85 Description: Remove portals compatibility
86 Details : Remove portals compatibility, not interoperable with releases
91 Description: Resolve device initialization race
92 Details : Prevent proc handler from accessing devices added to the
93 obd_devs array but yet be intialized.
95 Severity : enhancement
97 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
99 Severity : enhancement
101 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
106 Description: Using wrong pointer in osc_brw_prep_request
107 Details : Access to array[-1] can produce panic if kernel compiled with
108 CONFIG_PAGE_ALLOC enabled
112 Description: Client runs out of low memory
113 Details : Consider only lowmem when counting initial number of llap pages
117 Description: *optional* service tags registration
118 Details : if the "service tags" package is installed on a Lustre node
119 When the filesystem is mounted, a local-node service tag will
120 be created. See http://inventory.sun.com/ for more information
121 about the Service Tags asset management system.
125 Description: Kernel BUG tries to release flock
126 Details : Lustre does not destroy flock lock before last reference goes
127 away. So always drop flock locks when client is evicted and
128 perform unlock regardless of successfulness of speaking to MDS.
132 Description: add recount protection for osc callbacks, so avoid panic on shutdown
136 Description: sanity test 65a fails if stripecount of -1 is set
137 Details : handle -1 striping on filesystem in ll_dirstripe_verify
142 Description: ASSERTION(CheckWriteback(page,cmd)) failed
143 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
146 Severity : enhancement
148 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
152 Description: do not process already freed flock
153 Details : flock can possibly be freed by another thread before it reaches
154 to ldlm_flock_completion_ast.
158 Description: LBUG during stress test
159 Details : Need properly lock accesses the flock deadlock detection list.
163 Description: oops in page fault handler
164 Details : kernel page fault handler can return two special 'pages' in error case, don't
165 try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
169 Description: timeout with invalidate import.
170 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which should be
171 handled by ptlrpcd. This produce long age waiting and -ETIMEOUT
172 ptlrpc_invalidate_import and as result LASSERT.
174 Severity : enhancement
176 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
180 Description: enable MGS and MDT services start separately
181 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
182 a co-located MGS without starting the MGS, which is a complement
183 to 'nosvc' mount option.
187 Frequency : after recovery
188 Description: precreate to many object's after del orphan.
189 Details : del orphan st in oscc last_id == next_id and this triger growing
190 count of precreated objects. Set flag LOW to skip increase count
191 of precreated objects.
195 Frequency : rare, on clear nid stats
196 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
197 Details : when clean nid stats sometimes try destroy live entry,
198 and this produce panic in free.
202 Description: Stack overflow during MDS log replay
203 ease stack pressure by using a thread dealing llog_process.
207 Description: wait until IO finished before start new when do lock cancel.
208 Details : VM protocol want old IO finished before start new, in this case
209 need wait until PG_writeback is cleared until check dirty flag and
210 call writepages in lock cancel callback.
212 Severity : enhancement
214 Description: using special macro for print time and cleanup in includes.
218 Description: mds_mfd_close() ASSERTION(rc == 0)
219 Details : In mds_mfd_close(), we need protect inode's writecount change
220 within its orphan write semaphore to prevent possible races.
224 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
228 Frequency : rare, on shutdown ost
229 Description: don't hit live lock with umount ost.
230 Details : shrink_dcache_parent can be in long loop with destroy dentries,
231 use shrink_dcache_sb instead.
235 Description: don't panic with use echo client
236 Details : echo client pass NULL as client nid pointer and this produce null
241 Description: fix build on ppc32
242 Details : compile code with -m64 flag produce wrong object file for ppc32.
246 Description: add message levels for liblustreapi
250 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
251 Details : Move the 'good_osts' check before the 'total_bavail' check. This
252 will result in an -EAGAIN and in the exit call path we call
253 alloc_rr() which will with increasing aggressiveness attempt to
254 aquire precreated objects on the minimum number of required OSCs.
258 Description: Use old size assignment to avoid deadlock
259 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
260 the scheduling while holding a spinlock. We do not need locking
261 for size in ll_update_inode() because size is only updated from
262 the MDS for directories or files without objects, so there is no
263 other place to do the update, and concurrent access to such inodes
264 are protected by the inode lock.
268 Description: resolve "_IOWR redefined" build error on SLES10
272 Description: dump the memory debugging after all modules are unloaded to
273 suppress false negative in conf_sanity test 39
275 Severity : enhancement
277 Description: build kernel-ib packages for OFED 1.3 in our release cycle
282 Description: fix SLES kernel versioning
283 Details : the kernel version for our SLES 10 kernel did not include a "-"
284 before the "smp" at the end. while this was not a problem in
285 general, it did mean that software trying to use the kernel
286 version to try to detect a vendor specific kernel would fail.
287 this was most evident by the OFED build scripts.
291 Description: Don't update lov_desc members until making sure they are valid
292 Details : When updating lov_desc members via proc fs, need fix their
293 validities before doing the real update.
297 Description: don't put request into delay list while invalidate in flight.
298 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
299 import in flight. this produce timeout for invalidate and sometimes
300 can cause stale data.
304 Frequency : on ppc only
305 Description: not convert ost objects for directory because it's not exist.
306 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
310 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
312 Severity : enhancement
314 Description: Update to support 2.6.22.14 vanilla kernel.
318 Frequency : rare, on recovery
319 Description: read procfs can produce deadlock in some situation
320 Details : Holding lprocfs lock which send rpc can produce block for destroy
321 obd objects and this also block reconnect with -EALREADY. This isn't
322 fix all lprocfs bugs - but make it rare.
324 Severity : enhancement
326 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
329 Frequency : frequent on X2 node
331 Description: mdc_set_open_replay_data LBUG
332 Details : Set replay data for requests that are eligible for replay.
336 Description: lustre_mgs: operation 101 on unconnected MGS
337 Details : When MGC is disconnected from MGS long enough, MGS will evict the
338 MGC, and late on MGC cannot successfully connect to MGS and a lot
339 of the error messages complaining that MGS is not connected.
343 Frequency : on network error
344 Description: panic with double free request if network error
345 Details : mdc_finish_enqueue is finish request if any network error ocuring,
346 but it's true only for synchronus enqueue, for async enqueue
347 (via ptlrpcd) this incorrect and ptlrpcd want finish request
350 Severity : enhancement
352 Description: client-side metadata stat-ahead during readdir(directory readahead)
353 Details : perform client-side metadata stat-ahead when the client detects
354 readdir and sequential stat of dir entries therein
357 Frequency : on start mds
359 Description: Implement get_info(last_id) in obdfilter.
362 Frequency : occasional
364 Description: Correctly check stale fid, not start epoch if ost not support SOM
365 Details : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
366 overwrited with new generated fid, but mds can anwer with one of these
367 two fids and both is not stale. setattr incorectly start epoch and
368 assume will be called done_writeting, but without SOM done_writing
372 Frequency : rare, depends on device drivers and load
374 Description: MDS or OSS nodes crash due to stack overflow
375 Details : Code changes in 1.8.0 increased the stack usage of some functions.
376 In some cases, in conjunction with device drivers that use a lot
377 of stack the MDS (or possibly OSS) service threads could overflow
378 the stack. One change which was identified to consume additional
379 stack has been reworked to avoid the extra stack usage.
382 Frequency : occasional
384 Description: Do not fail import if osc_interpret_create gets -EAGAIN
385 Details : If osc_interpret_create got -EAGAIN it immediately exits and
386 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
387 oscc_has_objects and see OSC has no objests and call
388 oscc_internal_create to resend create request.
390 Severity : enhancement
392 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
394 Severity : enhancement
396 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
399 Frequency : very rare
401 Description: Fix replay if there is an un-replied request and open
402 Details : In some cases, older replay request will revert the
403 mcd->mcd_last_xid on MDS which is used to record the client's
406 Severity : enhancement
408 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
410 Severity : enhancement
412 Description: Add rhel5 support to HEAD.
414 Serverity : enhancement
416 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
421 Description: Don't allow skipping OSTs if index has been specified.
422 Details : Don't allow skipping OSTs if index has been specified, make locking
423 in internal create lots better.
427 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
428 Details : access to bitfield in structure is always rounded to long
429 and this produce problem with not atomic change any bit.
433 Description: Lustre make rpms failed.
434 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
435 builds Lustre rpms from tarball.
438 Frequency : rare on shutdown ost
440 Description: If llog cancel was not send before clean_exports phase, this can
441 produce deadlock in llog code.
442 Details : If llog thread has last reference to obd and call class_import_put
443 this produce deadlock because llog_cleanup_commit_master wait when
444 last llog_commit_thread exited, but this never success because was
445 called from llog_commit_thread.
449 Description: allow userland application know is lost one of stripes.
450 Details : fill lvb_blocks with error code on ost and return it to
451 application if error flag found.
455 Description: NULL lov_tgts causing MDS oops
456 Details : more safe checks for NULL lov_tgts for avoid oops.
458 Severity : enhancement
460 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
464 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
465 Details : If an OST is down the MDS will hang indefinitely in
466 obd_statfs_rqset() waiting for the statfs data. While for
467 MDS QOS usage of statfs, it should not stuck in waiting.
469 Severity : enhancement
471 Description: remote_acl support
472 Details : Support ACL-based permission check for remote user.
473 Support setfacl/getfacl for remote user with the utils
474 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
475 the system "{s,g}etfacl" utils.
477 Severity : enhancement
479 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
481 Severity : enhancement
483 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
487 Description: make mgs_setparam() handle fsname containing dash
488 Details : fsname containing a dash does not work with lctl conf_param
490 Severity : enhancement
492 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
494 Severity : enhancement
496 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
500 Frequency : rare, at shutdown
501 Description: access already free / zero obd_namespace.
502 Details : if client_disconnect_export was called without force flag set,
503 and exist connect request in flight, this can produce access to
504 NULL pointer (or already free pointer) when connect_interpret
505 store ocd flags in obd_namespace.
509 Frequency : only at startup
510 Description: not alloc memory with spinlock held.
511 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
512 if any spinlock held.
514 Severity : enhancement
516 Description: make lustre randomly fail allocating memory
517 Details : Make lustre randomly failed allocating memory for testing purpose.
519 Severity : enhancement
521 Description: lost problems with lov objid file
522 Details : Fixes some scability and access to not inited memory problems
523 in work with lov objdid file.
528 Description: lfs find does not continue on file error
529 Details : Continue other files processing when a file/dir is absent.
533 Description: Inconsistent usage of lustre_pack_reply()
534 Details : Standardize the usage of lustre_pack_reply() such that it
535 always generate a CERROR on failure.
538 Frequency : occasional
540 Description: MDS hang or stay in waiting lock
541 Details : If client receive lock with CBPENDING flag ldlm need send lock
542 cancel as separate rpc, to avoid situation when cancel request
543 can't processed due all i/o threads stay in wait lock.
547 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
551 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
555 Description: Update to SLES9 kernel-2.6.5-7.287.3.
559 Description: Update to RHEL4 latest kernel.
561 Severity : enhancement
563 Description: Build SLES10 patchless client fails
564 Details : The configure was broken by run ./configure with
565 --with-linux-obj=.... argument for patchless client. When the
566 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
567 can't search header adequately. Use absolute path such as
568 -I($LINUX)/include instead.
572 Description: interrupt oig_wait produce painc on resend.
573 Details : brw_redo_request can be used for resend requests from ptlrpcd and
574 private set, and this produce situation when rq_ptlrpcd_data not
575 copyed to new allocated request and triggered LBUG on assert
576 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
577 ptlrpcd set if request is changed and can be safety changed to use
582 Description: LASSERT_{REQ,REP}SWAB macros are buggy
583 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
584 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
585 these macros and replace them with their difinitions instead.
589 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
590 Details : Update which_patch & target file for SLES10 latest kernel.
593 Description: Remove client patches from SLES 10 kernel.
594 Details : This causes SLES 10 clients to behave as patchless clients
595 even on a Lustre-patched (server) kernel.
597 Severity : enhancement
599 Description: self-adjustable client's lru lists
600 Details : use adaptive algorithm for managing client cached locks lru
601 lists according to current server load, other client's work
602 pattern, memory activities, etc. Both, server and client
603 side namespaces provide number of proc tunables for controlling
606 Severity : enhancement
608 Description: light-weight GSS support
609 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
610 and reduce performance overhead.
612 Severity : enhancement
614 Description: Linux keyring support
615 Details : Support using service of Linux keyring for Lustre GSS internal
616 context refresh/cache mechanism.
620 Description: Fix errors in lfs documentation
621 Details : Fixes man pages
625 Description: don't use GFP_* in generic Lustre code.
626 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
630 Description: obdclass is limited by single OBD_ALLOC(idarray)
631 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
635 Description: warnings with build patchless client with vanila 2.6.19 and up
636 Details : change old ctl_table style and replace ctl_table/ctl_table_header
637 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
641 Frequency : Only for SLES
642 Description: sanity_quota fail test_1
643 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
644 file, which makes TSTID[2] inunique.
648 Frequency : Only for SLES9
649 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
650 Details : Update target/ChangeLog/which_patch .
654 Frequency : Only for test_10 in sanity-quota.sh
655 Description: a bug in quota test code
656 Details : checking a test flag in a wrong place
660 Frequency : Only for test_13 in sanity-quota.sh
661 Description: a bug in quota test script
662 Details : shouldn't check overfull allocated quota
666 Description: jbd statistics
667 Details : Port older jbd statistics patch for sles10
669 Severity : enhancement
671 Description: Add printing inode info into message about error in writepage.
676 Description: reply_lock_interpret crash due to race with it and lock cancel.
677 Details : Do not replay locks that are being cancelled. Do not reference
678 locks by their address during replay, just by their handle.
682 Frequency : When flocks are used.
683 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
684 Details : Flock locks might destroy just granted lock if it could be merged
685 with another existing flock, this is done in completion handler,
686 so teach ldlm_cli_enquque_fini that this is a valid case for
692 Description: Oops in read and write path when failing to allocate lock.
693 Details : Check if lock allocation failed and return error back.
697 Description: lstripe command fails for valid OST index
698 Details : The stripe offset is compared to lov->desc.ld_tgt_count
699 instead of lov->desc.ld_active_tgt_count.
703 Description: sanity.sh failed test 103
704 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
709 Description: df doesn't work properly if diskfs blocksize != 4K
710 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
714 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
715 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
717 Severity : enhancement
719 Description: Adaptive timeouts
720 Details : RPC timeouts adapt to changing server load and network
721 conditions to reduce resend attempts and improve recovery time.
725 Description: llapi_file_create() does not allow some changes
726 Details : add llapi_file_open() that allows specifying the mode and
727 open flags, and also returns an open file handle.
731 Description: merge and cleanup kernel patches.
732 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
736 Description: Add journal checksum support.(Kernel part)
737 Details : The journal checksum feature adds two new flags i.e
738 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
739 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
740 indicates that the commit block contains the checksum for
741 the blocks described by the descriptor blocks. Now commit
742 record can be sent to disk without waiting for descriptor
743 blocks to be written to disk. This behavior is controlled
744 using JBD2_FEATURE_ASYNC_COMMIT flag.
748 Description: OSS needs mutliple precreate threads
749 Details : Add ability to start more than one create thread per OSS.
753 Description: Sanity.sh test_65k failed: setstripe should have succeeded
754 Details : execute lfs setstripe on client
758 Description: mds_obd_create error creating tmp object
759 Details : When the user sets quota on root, llog will be affected and can't
760 create files and write files.
763 Frequency : Always on ia64 patchless client, and possibly others.
765 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
766 Details : This allows the patchless client to be loaded on architectures
771 Description: RedHat Update kernel for RHEL5
772 Details : Add the kernel config file for RHEL5.
776 Description: RedHat Update kernel for RHEL5
777 Details : Modify the kernel config file more closer RHEL5.
781 Description: Build failure against Centos5 (RHEL5)
782 Details : Define PAGE_SIZE when it isn't present.
786 Description: test_7 on sanity-quota.sh can't be used on separate machines
787 Details : add facet to handle it
791 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
792 Details : with NFS, the anon dentry's parent was set to itself in
793 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
794 corresponding dentry other than use rec->ur_name.
796 Severity : enhancement
798 Description: lfs setstripe enhancement
799 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
803 Description: enable data checksumming by default
804 Details : enable checksum by default, allow --disable-checksum
805 configure option and "-o nochecksum" mount option. Checksums
806 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
810 Description: lustre support for RHEL5
811 Details : Add support for RHEL5.
815 Description: Client eviction due to failover config
816 Details : after a connection loss, the lustre client should attempt to
817 reconnect to the last active server first before trying the
818 other potential connections.
822 Description: when mds and osts use different quota unit(32bit and 64bit),
823 quota will be released repeatly.
824 Details : void sending multiple quota reqs to mds, which will keep the
825 status between the reqs.
829 Description: osts not allocated evenly to files
830 Details : change the condition to increase offset_idx
833 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
835 Description: Data corruption for OSTs that are formatted larger than 2TB
837 Details : When generating the bio request for lustre file writes the
838 sector number would overflow a temporary variable before being
839 used for the IO. The data reads correctly from Lustre (which
840 will overflow in a similar manner) but other file data or
841 filesystem metadata may be corrupted in some cases.
845 Description: Tune the kernel for good SCSI performance.
846 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
847 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
852 Description: rewrite ext2-derived code in obdclass/uuid.c
853 Details : rewrite inherited code (uuid parsing code from ext2 utils)
854 from scratch preserving functionality.
858 Description: "lfs find -obd UUID" prints directories
859 Details : "lfs find -obd UUID" will return all directory names instead
860 of just file names. It is incorrect because the directories
861 do not reside on the OSTs.
865 Description: use i_size_read and i_size_write in 2.6 port
866 Details : replace inode->i_size access with i_size_read/write()
869 Frequency : when removing large files
871 Description: scheduling issue during removal of large Lustre files
872 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
873 It causes scheduling issues when removing large files (17TB in the
877 Frequency : only with liblustre clients on XT3
879 Description: evictions taking too long
880 Details : allow llrd to evict clients directly on OSTs
883 Frequency : only on ppc
885 Description: /proc/fs/lustre/devices broken on ppc
886 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
887 We should make sure the type of 'index' is loff_t and then cast
888 explicitly as needed below. Do not assign an explicitly cast
892 Frequency : only for rhel5
894 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
895 Details : Modify the target file & which_kernel.
897 Severity : enhancement
899 Description: omit set fsid for export NFS
900 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
901 as FS_REQUIRES_DEV for avoid problems with generate fsid.
905 Frequency : Always, for kernels after 2.6.16
906 Description: Fix warning idr_remove called for id=.. which is not allocated.
907 Details : Last kernels save old s_dev before kill super and not allow
908 to restore from callback - restore it before call kill_anon_super.
912 Description: buffer overruns could theoretically occur
913 Details : llapi_semantic_traverse() modifies the "path" argument by
914 appending values to the end of the origin string, and a
915 overrun may occur. Adding buffer overrun check in liblustreapi.
919 Description: Fix error on 'ls .' at the top of the Lustre mount.
920 Details : Don't revalidate dentry if it is a root dentry.
924 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
925 Details : Modify vm-tunables-rhel4.patch.
929 Description: Kernel config for 2.6.18-vanilla.
930 Details : Modify targets/2.6-vanilla.target.in.
931 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
932 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
933 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
934 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
938 Description: improve handling recoverable errors
939 Details : if request processig with error which can be recoverable on server
940 request should be resend, otherwise page released from cache and
945 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
946 Details : Modify target file & which_patch.
947 A flaw was found in the IA32 system call emulation provided
948 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
949 value could be stored in the %RAX register, which could trigger an
950 out-of-bounds system call table access. An untrusted local user
951 could exploit this flaw to run code in the kernel
952 (ie a root privilege escalation). (CVE-2007-4573).
956 Description: O_DIRECT bypasses client statistics.
957 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
958 read_ahead_stats were updated. copied the stats section from
959 osc_send_oap_rpc() into async_internal().
963 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
967 Description: change order of libsysio includes
968 Details : '#include sysio.h' should always come before '#include xtio.h'
972 Description: handle "serious error: objid * already exists" more gracefully
973 Details : If LAST_ID value on disk is smaller than the objects existing in
974 the O/0/d* directories, it indicates disk corruption and causes an
975 LBUG(). If the object is 0-length, then we should use the existing
976 object. This will help to avoid a full fsck in most cases.
978 Severity : enhancement
980 Description: adapt the lustre_config script to support the upgrade case
981 Details : Add "-u" option for lustre_config script to support upgrading 1.4
982 server targets to 1.6 in parallel.
986 Description: To avoid grant space > avaible space when the disk is almost
987 full. Without this patch you might see the error "grant XXXX >
988 available" or some LBUG about grant, when the disk is almost
990 Details : In filter_check_grant, for non_grant cache write, we should
991 check the left space by if (*left > ungranted + bytes), instead
992 of (*left > ungranted), because only we are sure the left space
993 is enough for another "bytes", then the ungrant space should be
994 increase. In client, we should update cl_avail_grant only there
995 is OBD_MD_FLGRANT in the reply.
999 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1000 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1001 described in CVE-2007-4573. This problem could allow untrusted
1002 local users to gain root access.
1005 Frequency : when using O_DIRECT and quotas
1007 Description: Incorrect file ownership on O_DIRECT output files
1008 Details : block usage reported by 'lfs quota' does not take into account
1009 files that have been written with O_DIRECT.
1014 Description: touch file failed when fs is not full
1015 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1016 otherwise we can get ENOSP while fs is not full.
1020 Description: parallel lock callbacks
1021 Details : Instead of sending blocking and completion callbacks as separated
1022 requests, adding them to a set and sending in parallel.
1025 Frequency : only for Cray XT3
1026 Bugzilla : 12829/13455
1027 Description: Changing primary group doesn't change the group lustre assigns to
1029 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1030 overridden with the primary group provided by the group upcall,
1031 whereas the supplied fsgid can be trusted if it is in the list of
1032 supplementary groups returned by the group upcall.
1034 Severity : enhancement
1036 Description: Allow masking D_WARNING, D_ERROR messages from console
1037 Details : Console messages can now be disabled via lnet.printk.
1041 Description: User code with malformed file open parameter crashes client node
1042 Details : Before packing join_file req, all the related reference should be
1043 checked carefully in case some malformed flags cause fake
1044 join_file req on client.
1048 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1049 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1050 mode is equal to the granted mode. However, it should also check
1051 that LDLM_FL_CP_REQD is not set.
1055 Description: Heavy nfs access might result in deadlocks
1056 Details : After ELC code landed, it is now improper to enqueue any mds
1057 locks under och_sem, because enqueue might want to decide to
1058 cancel open locks for same inode we are holding och_sem for.
1062 Description: Client eviction while running blogbench
1063 Details : A lot of unlink operations with concurrent I/O can lead to a
1064 deadlock causing evictions. To address the problem, the number of
1065 oustanding OST_DESTROY requests is now throttled to
1066 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1067 callbacks are processed in priority.
1071 Description: enable ACLs on MDS by default
1072 Details : ACLs must be enabled on MDS by default.
1075 Frequency : PPC/PPC64 only
1077 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1078 Details : fix duplicated definitions between asm-ppc64/types.h and
1079 lustre_types.h on PPC.
1082 Frequency : PPC/PPC64 only
1084 Description: asm-ppc/segment.h does not exist
1085 Details : fix compile issue on PPC.
1089 Description: better handle error messages in extents code
1092 Frequency : RHEL4 only
1094 Description: mkfs is very slow on IA64/RHEL4
1095 Details : A performance regression has been discovered in the MPT Fusion
1096 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1097 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1098 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1100 Severity : enhancement
1102 Description: SNMP support enhancement
1103 Details : Adding total number of sampled request for an MDS node in snmp
1106 Severity : enhancement
1108 Description: Optimize ldlm waiting list processing for PR extent locks
1109 Details : When processing waiting list for read extent lock and meeting read
1110 lock that is same or wider to it that is not contended, skip
1111 processing rest of the list and immediatelly return current
1112 status of conflictness, since we are guaranteed there are no
1113 conflicting locks in the rest of the list.
1117 Description: Time out and refuse to reconnect
1118 Details : When the failover node is the primary node, it is possible
1119 to have two identical connections in imp_conn_list. We must
1120 compare not conn's pointers but NIDs, otherwise we can defeat
1121 connection throttling.
1125 Description: port llog fixes from b1_6 into HEAD
1126 Details : Port llog reference couting and some llog cleanups from b1_6
1127 (bug 10800) into HEAD, for protect from panic and access to already
1128 free llog structures.
1132 Description: Detect stride IO mode in read-ahead
1133 Details : When a client does stride read, read-ahead should detect that and
1134 read-ahead pages according to the detected stride pattern.
1138 Description: data checksumming impacts single node performance
1139 Details : add support for several checksum algorithm. Currently, only CRC32
1140 and Adler-32 are supported. The checksum type can be changed on
1141 the fly via /proc/fs/lustre/osc/*/checksum_type.
1145 Description: use adler32 for page checksums
1146 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1151 Description: build for x2 fails
1152 Details : fix compile issue on Cray systems.
1156 Description: Properly match for duplicate locks
1157 Details : Due to different lock order from skiplists code, we need to
1158 traverse entire list for now
1161 Frequency : only on PPC/SLES10
1163 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1164 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1165 be defined. Add a hack in mkfs_lustre.c to work around this compile
1170 Description: LASSERT on MDS when client holding flock lock dies
1171 Details : ldlm pool logic depends on number of granted locks equal to
1172 number of released locks which is not true for flock locks, so
1173 just exclude such locks from consideration.
1177 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1178 Details : Use fsfilt_sync() for both the whole filesystem sync and
1179 individual file sync to eliminate dangerous inode locking
1180 with I_LOCK that can lead to a deadlock.
1184 Description: performance in 1.6.3
1185 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1186 sure that 1MB requests are not fragmented by the block layer.
1188 Severity : enhancement
1190 Description: organize the server-side client stats on per-nid basis
1191 Details : Change the structure of stats under obdfilter and mds to
1199 The "uuid"s file would list the uuids of _active_ exports.
1200 And the clear entry is to clear all stats and stale nids.
1202 Severity : enhancement
1204 Description: eliminate client locks in face of contention
1205 Details : file contention detection and lockless i/o implementation
1206 for contended files.
1210 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1213 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1215 Description: MDS may lose file striping (and hence file data) in some cases
1216 Details : If there are additional extended attributes stored on the MDS,
1217 in particular ACLs, SELinux, or user attributes (if user_xattr
1218 is specified for the client mount options) then there is a risk
1219 of attribute loss. Additionally, the Lustre file striping
1220 needs to be larger than default (e.g. striped over all OSTs),
1221 and an additional attribute must be stored initially in the
1222 inode and then increase in size enough to be moved to the
1223 external attribute block (e.g. ACL growing in size) for file
1228 Description: skiplist implementation simplification
1229 Details : skiplists are used to group compatible locks on granted list
1230 that was implemented as tracking first and last lock of each lock group
1231 the patch changes that to using doubly linked lists
1235 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1236 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1237 because LASSERT on both the data supplied by a client, and the data
1238 on disk is dangerous and incorrect.
1240 Severity : enhancement
1242 Description: Slow truncate/writes to huge files at high offsets.
1243 Details : Directly associate cached pages to lock that protect those pages,
1244 this allows us to quickly find what pages to write and remove
1245 once lock callback is received.
1249 Description: more ldlm soft lockups
1250 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1251 starve other threads from the resource lock for a long time in
1252 case of long waiting queue, so change the debug level from
1253 D_OTHER to the less frequently used D_INFO.
1255 Severity : enhancement
1257 Description: add -gid, -group, -uid, -user options to lfs find
1261 Description: Hung threads in invalidate_inode_pages2_range
1262 Details : The direct IO path doesn't call check_rpcs to submit a new RPC once
1263 one is completed. As a result, some RPCs are stuck in the queue
1268 Description: filter threads hungs on waiting journal commit
1269 Details : Cleanup filter group llog code, then only filter group llog will
1270 be only created in the MDS/OST syncing process.
1274 Description: Procfs and llog threads access destoryed import sometimes.
1275 Details : Sync the import destoryed process with procfs and llog threads by
1276 the import refcount and semaphore.
1278 Severity : enhancement
1280 Description: openlock cache of b1_6 port to HEAD
1285 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1286 Details : If insertion of an extent fails, then discard the inode
1287 preallocation and free data blocks else it can lead to duplicate
1292 Description: don't always update ctime in ext3_xattr_set_handle()
1293 Details : Current xattr code updates the inode ctime in ext3_xattr_set_handle.
1294 In some cases the ctime should not be updated, for example for
1295 2.0->1.8 compatibility it is necessary to delete an xattr and it
1296 should not update the ctime.
1300 Bugzilla : 15713/16362
1301 Description: Assertion in iopen_connect_dentry in 1.6.3
1302 Details : looking up an inode via iopen with the wrong generation number can
1303 populate the dcache with a disconneced dentry while the inode
1304 number is in the process of being reallocated. This causes an
1305 assertion failure in iopen since the inode's dentry list contains
1306 both a connected and disconnected dentry.
1310 Description: assertion failure in ldlm_handle2lock()
1311 Details : fix a race between class_handle_unhash() and class_handle2object()
1312 introduced in lustre 1.6.5 by bug 13622.
1317 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1318 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1323 Description: oops in sd_iostats_seq_show()
1324 Details : unloading/reloading the scsi low level driver triggers a kernel
1325 bug when trying to access the sd iostat file.
1330 Description: Kernel panics during QLogic driver reload
1331 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1332 patch, causing memory corruption.
1337 Description: journal_dev option does not work in b1_6
1338 Details : pass mount option during pre-mount.
1340 Severity : enhancement
1342 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1343 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1344 extent information of a file. It can be used to map logical blocks
1345 in a file to physical blocks in the block device.
1349 Description: LDLM soft lockups - improvement
1350 Details : It is be possible to send the lock handle along with each read
1351 or write request because the client is already doing a lock match
1352 itself so there isn't any reason the OST should have to re-do that
1355 --------------------------------------------------------------------------------
1357 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
1359 * Support for kernels:
1360 2.6.5-7.283 (SLES 9),
1361 2.6.9-55.EL (RHEL 4),
1362 2.6.16.46-0.14 (SLES 10),
1363 2.6.18.8 vanilla (kernel.org)
1364 * Client support for unpatched kernels:
1365 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1366 2.6.16 - 2.6.22 vanilla (kernel.org)
1367 * Due to recently discovered recovery problems, we do not recommend
1368 using patchless RHEL 4 clients with this or any earlier release.
1369 * Recommended e2fsprogs version: 1.39.cfs8
1370 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1371 * Starting with this release, the ldiskfs backing filesystem required
1372 by Lustre is now in its own package, lustre-ldiskfs. This package
1373 should be installed. It is versioned separately from Lustre and
1374 may be released separately in future.
1378 Description: block reactivating mgc import until all deactivates complete
1379 Details : Fix race when failing back MDT/MGS to itself (testing)
1382 Frequency : at statup only
1384 Description: mds_lov_synchronize race leads to various problems
1385 Details : simultaneous MDT->OST connections at startup can cause the
1386 sync to abort, leaving the OSC in a bad state.
1388 Severity : enhancement
1390 Description: add optional extra BUILD_VERSION info
1391 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1392 to override the lustre version.
1395 Frequency : 2.6.18 servers only
1397 Description: ll_kern_mount() doesn't release the module reference
1398 Details : The ldiskfs module reference count never drops down to 0
1399 because ll_kern_mount() doesn't release the module reference.
1404 Description: server LBUG when using old ost_num_threads parameter
1405 Details : Accept the old ost_num_threads parameter but warn that it
1406 is deprecated, and fix an off-by-one error that caused an LBUG.
1411 Description: Transient SCSI error results in persistent IO issue
1412 Details : iobuf->dr_error is not reinitialized to 0 between two
1416 Frequency : sometimes when underlying device returns I/O errors
1418 Description: OSTs not going read-only during write failures
1419 Details : OSTs are not remounted read-only when the journal commit threads
1420 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1421 instead of the ext3 wrappers.
1425 Description: poor connect scaling with increasing client count
1426 Details : Don't run filter_grant_sanity_check for more than 100 exports
1427 to improve scaling for large numbers of clients.
1430 Frequency : SLES10 only
1432 Description: sanity-quota.sh quotacheck failed: rc = -22
1433 Details : Quotas cannot be enabled on SLES10.
1436 Frequency : liblustre clients only
1438 Description: getdirentries does not give error when run on compute nodes
1439 Details : getdirentries does not fail when the size specified as an argument
1440 is too small to contain at least one entry
1442 Severity : enhancement
1444 Description: Add LNET router traceability for debug purposes
1445 Details : If a checksum failure occurs with a router as part of the
1446 IO path, the NID of the last router that forwarded the bulk data
1447 is printed so it can be identified.
1452 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1453 Details : Due to a race condition, liblustre clients were occasionally
1454 evicted incorrectly.
1456 Severity : enhancement
1458 Description: lfs setstripe use optional parameters instead of postional
1461 Severity : enhancement
1463 Description: Nanosecond timestamp support for ldiskfs
1464 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1465 resolution timestamps. There is not yet support for this at
1466 the Lustre filesystem level.
1469 Frequency : during server recovery
1471 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1472 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1476 Frequency : nfs export on patchless client
1478 Description: connectathon hang when test nfs export over patchless client
1479 Details : Disconnected dentry cannot be found with lookup, so we do not need
1480 to unhash it or make it invalid
1483 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1484 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1485 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1490 Description: open req refcounting wrong on reconnect
1491 Details : If reconnect happened between getting open reply from server and
1492 call to mdc_set_replay_data in ll_file_open, we will schedule
1493 replay for unreferenced request that we are about to free.
1494 Subsequent close will crash in variety of ways.
1495 Check that request is still eligible for replay in
1496 mdc_set_replay_data().
1501 Description: disable writes to filesystem when reading health_check file
1502 Details : the default for reading the health_check proc file has changed
1503 to NOT do a journal transaction and write to disk, because this
1504 can cause reads of the /proc file to hang and block HA state
1505 checking on a healthy but otherwise heavily loaded system. It
1506 is possible to return to the previous behaviour during configure
1507 with --enable-health-write.
1509 Severity : enhancement
1511 Description: 64-bit inode version
1512 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1513 to the inode. This will be required for version-based recovery.
1518 Description: MDS fails to start if a duplicate client export is detected
1519 Details : in some rare cases it was possible for a client to connect to
1520 an MDS multiple times. Upon recovery the MDS would detect this
1521 and fail during startup. Handle this more gracefully.
1523 Severity : enhancement
1525 Description: Add -o localflock option to simulate old noflock
1527 Details : This will achieve local-only flock/fcntl locks
1533 Description: log_commit_thread vs filter_destroy race leads to crash
1534 Details : Take import reference before releasing llog record semaphore
1539 Description: Wrong request locking in request set processing
1540 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1541 imp_delayed_list, in this place should be used imp_lock.
1544 Frequency : when reconnection
1546 Description: Grant Leak when osc reconnect to OST
1547 Details : When osc reconnect ost, OST(filter) should check whether it
1548 should grant more space to client by comparing fed_grant and
1549 cl_avail_grant, and return the granted space to client instead
1550 of "new granted" space, because client will call osc_init_grant
1551 to update the client grant space info.
1554 Frequency : when client reconnect to OST
1556 Description: Grant Leak when osc do resend and replay bulk write
1557 Details : When osc reconnect to OST, OST(filter)should clear grant info of
1558 bulk write request, because the grant info while be sync between
1559 OSC and OST when reconnect, and we should ignore the grant info
1560 these of resend/replay write req.
1565 Description: Grant space more than avaiable left space sometimes.
1566 Details : When then OST is about to be full, if two bulk writing from
1567 different clients came to OST. Accord the avaliable space of the
1568 OST, the first req should be permitted, and the second one
1569 should be denied by ENOSPC. But if the seconde arrived before
1570 the first one is commited. The OST might wrongly permit second
1571 writing, which will cause grant space > avaiable space.
1574 Frequency : when client is evicted
1576 Description: Grant might be wrongly erased when osc is evicted by OST
1577 Details : when the import is evicted by server, it will fork another
1578 thread ptlrpc_invalidate_import_thread to invalidate the
1579 import, where the grant will be set to 0. While the original
1580 thread will update the grant it got when connecting. So if
1581 the former happened latter, the grant will be wrongly errased
1582 because of this race.
1587 Description: Checking Stale with correct fid
1588 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1589 to check whether it is stale, because sometimes, we want the
1590 enqueue happened anyway, and op_data.fid2 will not be initialized.
1592 Severity : enhancement
1594 Description: update patchless client
1595 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1598 Frequency : only with 2.4 kernel
1600 Description: random memory corruption
1601 Details : size of struct ll_inode_info is to big for union inode.u and this
1602 can be cause of random memory corruption.
1607 Description: Memory leak in recovery
1608 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1609 It should also check obd_fail before fsfilt_start, otherwise if
1610 fsfilt_start return -EROFS,(failover mds during mds recovery).
1611 then the req will return with repmsg->transno = 0 and rc = EROFS.
1612 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1613 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1614 be freed no matter whether fsfilt_commit success or not.
1617 Frequency : only with huge count clients
1619 Description: Prevents from taking the superblock lock in llap_from_page for
1621 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1622 for llap_from_page call in ll_removepage prevents from taking the
1623 superblock lock for a soon died page.
1628 Description: Not check open intent error before release open handle
1629 Details : in some rare cases, the open intent error is not checked before
1630 release open handle, which may cause
1631 ASSERTION(open_req->rq_transno != 0), because it tries to release
1632 the failed open handle.
1637 Description: Set cat log bitmap only after create log success.
1638 Details : in some rare cases, the cat log bitmap is set too early. and it
1639 should be set only after create log success.
1643 Description: Accessing a block bevice can re-enable I/O when Lustre is
1644 tearing down a device.
1645 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1650 Description: service threads may hog cpus when there are a lot of requests
1651 Details : Insert cond_resched to give other threads a chance to use some CPU
1656 Description: the cat log was not initialized in recovery
1657 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1658 unlink log on mds will not be initialized until mds post
1659 recovery. And also in mds post recovery, the unlink log will
1660 initialization will be done asynchronausly, so there will be race
1661 between add unlink log and unlink log initialization.
1665 Description: brw_stats were being printed incorrectly
1666 Details : brw_stats were being printed as log2 but all of them were not
1667 recorded as log2. Also remove some code duplication arising from
1668 filter_tally_{read,write}.
1672 Frequency : rare, only in recovery.
1673 Description: ASSERTION(req->rq_type != LI_POISON) failed
1674 Details : imp_lock should be held while iterating over imp_sending_list for
1675 prevent destroy request after get timeout in ptlrpc_queue_wait.
1679 Description: replay-single.sh test 52 fails
1680 Details : A lock's skiplist need to be cleanup when it being unlinked
1681 from its resource list.
1685 Description: Short directio read returns full requested size rather than
1687 Details : Direct I/O operations should return actual amount of bytes
1688 transferred rather than requested size.
1690 Severity : enhancement
1692 Description: metadata RPC reduction (e.g. for rm performance)
1693 Details : decrease the amount of synchronous RPC between clients and servers
1694 by canceling conflicing lock before the operation on the client
1695 and packing thier handles into the main operation RPC to server.
1697 Severity : enhancement
1699 Description: Async OSC create to avoid the blocking unnecessarily.
1700 Details : If a OST has no remain object, system will block on the creating
1701 when need to create a new object on this OST. Now, ways use
1702 pre-created objects when available, instead of blocking on an
1703 empty osc while others are not empty. If we must block, we block
1704 for the shortest possible period of time.
1706 Severity : enhancement
1708 Description: refine locking for avoid write wrong info into lov_objid file
1709 Details : fix possible races with add new target and write/update data in
1712 --------------------------------------------------------------------------------
1714 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1719 Frequency : on some architectures
1721 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1722 Details : Uninitialized flags sometimes cause configuration commands to
1726 Frequency : patchless clients only
1728 Description: missing __iget() symbol export
1729 Details : The __iget() symbol export is missing. To avoid the need for
1730 this on patchless clients the deathrow inode reaper is turned
1731 off, and we depend on the VM to clean up old inodes. This
1732 dependency was during via the fix for bug 12181.
1734 --------------------------------------------------------------------------------
1736 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1738 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1739 INTEROPERATE with older versions automatically. In many cases a
1740 special upgrade step is needed. Please read the
1741 user documentation before upgrading any part of a live system.
1742 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1743 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1744 * WARNING: Lustre configuration and startup changes are required with
1745 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1751 Severity : enhancement
1753 Description: MountConf
1754 Details : Lustre configuration is now managed via mkfs and mount
1755 commands instead of lmc and lconf. New obd types (MGS, MGC)
1756 are added for dynamic configuration management. See
1757 https://mail.clusterfs.com/wikis/lustre/MountConf for
1760 Severity : enhancement
1762 Description: dynamic OST addition
1763 Details : OSTs can now be added to a live filesystem
1765 Severity : enhancement
1767 Description: startup order invariance
1768 Details : MDTs and OSTs can be started in any order. Clients only
1769 require the MDT to complete startup.
1771 Severity : enhancement
1773 Description: parallel, asynchronous orphan cleanup
1774 Details : orphan cleanup is now performed in separate threads for each
1775 OST, allowing parallel non-blocking operation.
1777 Severity : enhancement
1779 Description: optimized stripe assignment
1780 Details : stripe assignments are now made based on ost space available,
1781 ost previous usage, and OSS previous usage, in order to try
1782 to optimize storage space and networking resources.
1784 Severity : enhancement
1786 Description: Permanently set tunables
1787 Details : All writable /proc/fs/lustre tunables can now be permanently
1788 set on a per-server basis, at mkfs time or on a live system.
1790 Severity : enhancement
1792 Description: Lustre message v2
1793 Details : Add lustre message format v2.
1795 Severity : enhancement
1797 Description: client OST exclusion list
1798 Details : Clients can be started with a list of OSTs that should be
1799 declared "inactive" for known non-responsive OSTs.
1803 Description: SPEC SFS validation failure on NFS v2 over lustre.
1804 Details : Changes the blocksize for regular files to be 2x RPC size,
1805 and not depend on stripe size.
1807 Severity : enhancement
1809 Description: Multiple MD RPCs in flight.
1810 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1811 To avoid overly-overloading MDS, introduce a limit on number of
1812 MDS RPCs in flight for a single client and add /proc controls
1813 to adjust this limit.
1815 Severity : enhancement
1817 Description: client read/write statistics
1818 Details : Add client read/write call usage stats for performance
1819 analysis of user processes.
1820 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1821 file access. extents_stats shows chunk size distribution.
1822 extents_stats_per_process show chunk size distribution per
1825 Severity : enhancement
1827 Description: mds statistics
1828 Details : Add detailed mds operations statistics in
1829 /proc/fs/lustre/mds/*/stats.
1833 Description: Failure of copying files with lustre special EAs.
1834 Details : Client side always return success for setxattr call for lustre
1835 special xattr (currently only "trusted.lov").
1840 Description: Refcount LNET uuids
1841 Details : The global LNET uuid list grew linearly with every startup;
1842 refcount repeated list entries instead of always adding to
1845 Severity : enhancement
1847 Description: Dynamic service threads
1848 Details : Within a small range, start extra service threads
1849 automatically when the request queue builds up.
1852 Frequency : mixed-endian client/server environments
1854 Description: mixed-endian crashes
1855 Details : The new msg_v2 system had some failures in mixed-endian
1858 Severity : enhancement
1860 Description: Easy OST removal
1861 Details : OSTs can be permanently deactivated with e.g. 'lctl
1862 conf_param lustre-OST0001.osc.active=0'
1864 Severity : enhancement
1866 Description: MGS proc entries
1867 Details : Added basic proc entries for the MGS showing what filesystems
1870 Severity : enhancement
1872 Description: provide MGS failover
1873 Details : Added config lock reacquisition after MGS server failover.
1875 Severity : enhancement
1877 Description: add Linux 2.4 support
1878 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1882 Description: a large application tries to do I/O to the same resource and dies
1883 in the middle of it.
1884 Details : Check the req->rq_arrival time after the call to
1885 ost_brw_lock_get(), but before we do anything about
1886 processing it & sending the BULK transfer request. This
1887 should help move old stale pending locks off the queue as
1888 quickly as obd_timeout.
1891 Frequency : when an incorrect nid is specified during startup
1893 Description: ptlrpc connect to non-existant node causes kernel crash
1894 Details : LNET can't be re-entered from an event callback, which
1895 happened when we expire a message after the export has been
1896 cleaned up. Instead, hand the zombie cleanup off to another
1899 Severity : enhancement
1901 Description: plain/inodebits lock performance improvement
1902 Details : Grouping plain/inodebits in granted list by their request modes
1903 and bits policy, thus improving the performance of search through
1907 Frequency : only if OST filesystem is corrupted
1909 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1910 Details : for a short time RPCs with bulk IO are in the replay list,
1911 but replay of bulk IOs is unimplemented. If the OST filesystem
1912 is corrupted due to disk cache incoherency and then replay is
1913 started it is possible to trip an assertion. Avoid putting
1914 committed RPCs into the replay list at all to avoid this issue.
1917 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1919 Description: System hang on startup
1920 Details : This bug allowed the liblustre (e.g. catamount) client to
1921 return to the app before handling all startup RPCs. This
1922 could leave the node unresponsive to lustre network traffic
1923 and manifested as a server ptllnd timeout.
1925 Severity : enhancement
1927 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1928 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1929 boolean to control whether to print peer debug info when a
1930 client's RPC times out.
1933 Frequency : only for kernels with patches from Lustre below 1.4.3
1935 Description: Remove old rdonly API
1936 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1939 Frequency : only for devices with external journals
1941 Description: Set external device read-only also
1942 Details : During a commanded failover stop, we set the disk device
1943 read-only while the server shuts down. We now also set any
1944 external journal device read-only at the same time.
1947 Frequency : when upgrading from 1.4 while trying to change parameters
1949 Description: The wrong (new) MDC name was used when setting parameters for
1950 upgraded MDT's. Also allows changing of OSC (and MDC)
1951 parameters if --writeconf is specified at tunefs upgrade time.
1954 Frequency : when setting specific ost indicies
1956 Description: QOS code breaks on skipped indicies
1957 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1958 created with --index need not be sequential.
1960 Severity : enhancement
1962 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1963 Details : The uninit_groups feature works in conjunction with the kernel
1964 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1965 up the pass1 processing of e2fsck. This is a read-only feature
1966 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1967 mount filesystems that have had this feature enabled.
1969 Severity : enhancement
1971 Description: Improve multi-block allocation algorithm to avoid fragmentation
1972 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1973 allocation locality and avoid filesystem fragmentation.
1975 ------------------------------------------------------------------------------
1977 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1979 * Support for kernels:
1980 2.4.21-47.0.1.EL (RHEL 3)
1981 2.6.5-7.283 (SLES 9)
1982 2.6.9-42.0.10.EL (RHEL 4)
1983 2.6.12.6 vanilla (kernel.org)
1984 2.6.16.27-0.9 (SLES 10)
1985 * Recommended e2fsprogs version: 1.39.cfs5
1987 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1991 Frequency : occasional, depends on client load and configuration
1992 Bugzilla : 12181, 12203
1993 Description: data loss for recently-modified files
1995 Details : In some cases it is possible that recently written or created
1996 files may not be written to disk in a timely manner (this should
1997 normally be within 30s unless client IO load is very high).
1998 The problem appears as zero-length files or files that are a
1999 multiple of 1MB in size after a client crash or client eviction
2000 that are missing data at the end of the file.
2002 This problem is more likely to be hit on clients where files are
2003 repeatedly created and unlinked in the same directory, clients
2004 have a large amount of RAM, have many CPUs, the filesystem has
2005 many OSTs, the clients are rebooted frequently, and/or the files
2006 are not accessed by other nodes after being written.
2008 The presence of the problem can be detected by looking at
2009 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2010 smaller than the second (nr_unused) then dirty files will not
2011 be flushed automatically to disk. "sync; sleep 10" should be
2012 run several times on the node before unmounting it to update
2013 Lustre (this is also safe to run on nodes without this problem).
2015 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2016 that can cause this same problem, so customers using that kernel
2017 also need to update the kernel in addition to Lustre. In order
2018 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2020 It is normal that files written just before a client crash (less
2021 than 30s) may not yet have been flushed to disk, even for local
2025 Frequency : frequent on thin XT3 nodes
2027 Description: UUID collision on thin XT3 Linux nodes
2028 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2029 randomly, since we relied on an insufficiently-seeded PRNG.
2034 Description: OSS hangs after "All ost request buffers busy"
2035 Details : A deadlock between quota and journal operations caused OSS
2036 hangs after printing "All ost request buffers busy."
2039 Frequency : always on liblustre builds
2041 Description: Cleanup compiler warnings on liblustre
2044 Frequency : always on liblustre builds on XT3
2046 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2050 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2051 > 32000 subdirectories
2052 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2053 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2054 subdirectory count crosses 32000. This will aid e2fsck to
2055 correctly handle more than 32000 subdirectories.
2058 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2060 Description: System hang on startup
2061 Details : This bug allowed the liblustre (e.g. catamount) client to
2062 return to the app before handling all startup RPCs. This
2063 could leave the node unresponsive to lustre network traffic
2064 and manifested as a server ptllnd timeout.
2066 Severity : enhancement
2068 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2069 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2070 boolean to control whether to print peer debug info when a
2071 client's RPC times out.
2076 Description: make O_SYNC working on 2.6 kernels
2077 Details : 2.6 kernels use different method for mark pages for write,
2078 so need add a code to lustre for O_SYNC work.
2083 Description: Failure to close file and release space on NFS
2084 Details : Put inode details into lock acquired in ll_intent_file_open.
2085 Use mdc_intent_lock in ll_intent_open to properly
2086 detect all kind of errors unhandled by mdc_enqueue.
2091 Description: proc file read during shutdown sometimes raced obd removal,
2093 Details : Add lock to prevent obd access after proc file removal.
2096 Frequency : Only for files larger than 4GB on 32-bit clients.
2098 Description: improperly doing page alignment of locks
2099 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2100 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2105 Description: RPCs being resent when they shouldn't be
2106 Details : Some RPCs that should not be resent are being resent. This
2107 can cause inconsistencies in the RPC state machine. Do not
2108 resend such requests.
2111 Frequency : rare, only with NFS export
2113 Description: Crash on NFS re-export node
2114 Details : under very unusual load conditions an assertion is hit in
2115 ll_intent_file_open()
2118 Frequency : only if OST filesystem is corrupted
2120 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2121 Details : for a short time RPCs with bulk IO are in the replay list,
2122 but replay of bulk IOs is unimplemented. If the OST filesystem
2123 is corrupted due to disk cache incoherency and then replay is
2124 started it is possible to trip an assertion. Avoid putting
2125 committed RPCs into the replay list at all to avoid this issue.
2130 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2131 Details : Large single O_DIRECT read and write calls can fail to allocate
2132 a sufficiently large buffer to process the request. In case of
2133 allocation failure the allocation is retried with a smaller
2134 buffer and broken into smaller requests.
2136 Severity : enhancement
2138 Description: Add -o localflock option to simulate old noflock behaviour.
2139 Details : This will achieve local-only flock/fcntl locks coherentness.
2144 Description: versioning check is incomplete
2145 Details : Checking the version difference of client vs. server, report
2146 error if the gap is too big.
2151 Description: add support PG_writeback bit
2152 Details : add support for PG_writeback bit for Lustre, for more carefull
2153 work with page cache in 2.6 kernel. This also fix some deadlocks
2154 and remove hack for work O_SYNC with 2.6 kernel.
2156 Severity : enhancement
2158 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2159 Details : The uninit_groups feature works in conjunction with the kernel
2160 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2161 up the pass1 processing of e2fsck. This is a read-only feature
2162 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2163 mount filesystems that have had this feature enabled.
2165 Severity : enhancement
2167 Description: Improve multi-block allocation algorithm to avoid fragmentation
2168 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2169 allocation locality and avoid filesystem fragmentation.
2171 ------------------------------------------------------------------------------
2173 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2175 * Support for kernels:
2176 2.6.9-42.0.3.EL (RHEL 4)
2177 2.6.5-7.276 (SLES 9)
2178 2.4.21-47.0.1.EL (RHEL 3)
2179 2.6.12.6 vanilla (kernel.org)
2180 2.6.16.21-0.8 (SLES10)
2181 * Recommended e2fsprogs version: 1.39.cfs2-0
2183 * The backwards-compatible /proc/sys/portals symlink has been removed
2184 in this release. Before upgrading, please ensure that you change
2185 any configuration scripts or /etc/sysctl.conf files that access
2186 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2187 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2188 in advance of the upgrade on any system running Lustre 1.4.6 or
2189 newer, since /proc/sys/lnet was added in that version.
2190 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2194 Frequency : only when quota is used
2196 Description: avoid scanning export list for quota master
2197 Details : Change the algorithms to avoid scanning export list in order
2198 to improve the efficiency.
2201 Frequency : MDS failover only, very rarely
2203 Description: "went back in time" messages on mds failover
2204 Details : The greatest transno may be lost when the current operation
2205 finishes with an error (transno==0) and the client's last_rcvd
2206 record is over-written. Save the greatest transno in the
2207 mds_last_transno for this case.
2210 Frequency : always for specific kernels and striping counts
2212 Description: client may get "Matching packet too big" without ACL support
2213 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2214 when trying to access files in certain configurations. The
2215 clients should in fact be denied when mounting because they do
2216 not understand ACLs.
2219 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2221 Description: many clients connecting with IO in progress causes connect timeouts
2222 Details : Avoid synchronous journal commits to avoid delays caused by many
2223 clients connecting/disconnecting when bulk IO is in progress.
2224 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2225 OST_IO_PORTAL to avoid delays behind potentially many pending
2229 Frequency : occasionally with multiple writers to a single file
2231 Description: shared writes to file may result in wrong size reported by stat()
2232 Details : Allow growing of kms when extent lock is cancelled
2235 Frequency : always with random mmap IO to multi-striped file
2237 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2238 Details : Only if the hole is at the end of OST object so that kms is too
2239 small. Fix is to increase kms accordingly in ll_nopage.
2242 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2244 Description: writes to a missing object would leak memory on the OST
2245 Details : If there is an inconsistency between the MDS and OST filesystems,
2246 such that the MDS references an object that doesn't exist, writes
2247 to that object will leak memory due to incorrect cleanup in the
2248 error handling path, eventually running out of memory on the OST.
2253 Description: Creating too long symlink causes lustre errors
2254 Details : Check symlink and name lengths before sending requests to MDS.
2257 Frequency : only if flock is enabled (not on by default)
2259 Description: posix locks not released on fd closure on 2.6.9+
2260 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2261 kernels, this caused such locks not to be released on fd close and
2262 then assertions on fs unmount about still used locks.
2265 Frequency : MDS failover only, very rarely
2267 Description: clients may get ASSERTION(granted_lock != NULL)
2268 Details : When request was taking a long time, and a client was resending
2269 a getattr by name lock request. The were multiple lock requests
2270 with the same client lock handle and
2271 mds_getattr_name->fixup_handle_for_resent_request found one of the
2272 lock handles but later failed with ASSERTION(granted_lock != NULL).
2277 Description: handle->h_buffer_credits > 0, assertion failure
2278 Details : h_buffer_credits is zero after truncate, causing assertion
2279 failure. This patch extends the transaction or creates a new
2283 Frequency : NFS re-export or patchless client
2284 Bugzilla : 11179, 10796
2285 Description: Crash on NFS re-export node (__d_move)
2286 Details : We do not want to hash the dentry if we don't have a lock.
2287 But if this dentry is later used in d_move, we'd hit uninitialised
2288 list head d_hash, so we just do this to init d_hash field but
2289 leave dentry unhashed.
2292 Frequency : NFS re-export or patchless client
2294 Description: NFS exports has problem with symbolic link
2295 Details : lustre client didn't properly install dentry when re-exported
2296 to NFS or running patchless client.
2299 Frequency : NFS re-export or patchless client
2301 Description: Various nfs/patchless fixes.
2302 Details : fixes reuse disconected alias for lookup process - this fixes
2303 warning "find_exported_dentry: npd != pd",
2304 fix permission error with open files at nfs.
2305 fix apply umask when do revalidate.
2308 Frequency : occasional
2310 Description: Crash on NFS re-export node
2311 Details : calling clear_page() on the wrong pointer triggered oops in
2312 generic_mapping_read().
2315 Frequency : rarely, using O_DIRECT IO
2317 Description: unaligned directio crashes client with LASSERT
2318 Details : check for unaligned buffers before trying any requests.
2321 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2323 Description: stale data returned from RAID cache
2324 Details : If only a small amount of IO is done to the RAID device before
2325 reading it again it is possible to get stale data from the RAID
2326 cache instead of reading it from disk.
2329 Frequency : always for sles10 kernel
2331 Description: sles10 support
2332 Details : ll_follow_link: compile fixes and using of nd_set_link
2333 under newer kernels.
2336 Frequency : depends on arch, kernel and compiler version, always on sles10
2339 Description: recursive or deep enough symlinks cause stack overflow
2340 Details : getting rid of large stack-allocated variable in
2344 Frequency : depends on hardware
2346 Description: lustre write performance loss in the SLES10 kernel
2347 Details : the performance loss is caused by using of write barriers in the
2348 ext3 code. The SLES10 kernel turns barrier support on by
2349 default. The fix is to undo that change for ldiskfs.
2351 ------------------------------------------------------------------------------
2353 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2355 * Support for kernels:
2356 2.6.9-42.0.3EL (RHEL 4)
2357 2.6.5-7.276 (SLES 9)
2358 2.4.21-47.0.1.EL (RHEL 3)
2359 2.6.12.6 vanilla (kernel.org)
2363 Frequency : quota enabled and large files being deleted
2365 Description: releasing more than 4GB of quota at once hangs OST
2366 Details : If a user deletes more than 4GB of files on a single OST it
2367 will cause the OST to spin in an infinite loop. Release
2368 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2373 Description: statfs data retrieved from /proc may be stale or zero
2374 Details : When reading per-device statfs data from /proc, in the
2375 {kbytes,files}_{total,free,avail} files, it may appear
2376 as zero or be out of date.
2379 Frequency : systems with MD RAID1 external journal devices
2381 Description: lconf's call to blkid is confused by RAID1 journal devices
2382 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2383 of returning all block devices that match the journal UUID.
2386 Frequency : always, for aggregate stripe size over 4GB
2388 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2389 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2390 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2393 Frequency : always if "lfs find" used on a local file/directory
2395 Description: "lfs find" segfaults if used on a local file/directory
2396 Details : The case where a directory component was not specified wasn't
2397 handled correctly. Handle this properly.
2400 Frequency : always on ppc64
2402 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2403 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2404 perform endianness conversion.
2407 Frequency : rarely (truncate to non-zero file size after write under load)
2408 Bugzilla : 10730, 10687
2409 Description: Files padded with zeros to next 4K multiple
2410 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2411 it is possible that files that are truncated to a non-zero size
2412 immediately after being written are filled with zero bytes beyond
2413 the truncated size. No file data is lost.
2415 Severity : enhancement
2417 Description: Allow recovery/failover for liblustre clients.
2418 Details : liblustre clients were unaware of failover configurations until
2421 Severity : enhancement
2423 Description: user file locks should fail when not mounting with flock option
2424 Details : Set up an error-returning stub in ll_file_operations.lock field
2425 to prevent incorrect behaviour when client is mounted without
2426 flock option. Also, set up properly f_op->flock field for
2430 Frequency : always on ia64
2432 Description: "lfs df" loops on printing out MDS statfs information
2433 Details : The obd_ioctl_data was not initialized and in some systems
2434 this caused a failure during the ioctl that did not return
2435 an error. Initialize the struct and return an error on failure.
2438 Frequency : SLES 9 only
2440 Description: Error of copying files with lustre special EAs as root
2441 Details : Client side always return success for setxattr call for lustre
2442 special xattr (currently only "trusted.lov").
2445 Frequency : rarely on clusters with both ia64+i386 clients
2447 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2448 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2449 clients are doing concurrent writes to the same file it is
2450 possible that the ia64 clients may LASSERT because the OST
2451 extent locks are not PAGE_SIZE aligned. Ensure that grown
2452 locks are always aligned on the request boundary.
2455 Frequency : specific use, occasional
2457 Description: Overwriting in use executable truncates on-disk binary image
2458 Details : If one node attempts to overwrite an executable in use by
2459 another node, we now correctly return ETXTBSY instead of
2460 truncating the file.
2462 Severity : enhancement
2464 Description: Async OSC create to avoid the blocking unnecessarily.
2465 Details : If a OST has no remain object, system will block on the creating
2466 when need to create a new object on this OST. Now, ways use
2467 pre-created objects when available, instead of blocking on an
2468 empty osc while others are not empty. If we must block, we block
2469 for the shortest possible period of time.
2474 Description: chmod on Lustre root is propagated to other clients
2475 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2476 invalid by the follow_mount time.
2479 Frequency : liblustre clients only
2481 Description: Race in 'instant cancel' lock handling could lead to such locks
2482 never to be granted in case of SMP MDS
2483 Details : Do not destroy not yet granted but cbpending locks in
2487 Frequency : replay/resend of open
2489 Description: non null lock assetion failure in mds_intent_policy
2490 Details : Trying to replay/resend lockless open requests resulted in
2491 mds_open() returning 0 with no lock. Now it sets a flag if
2492 a lock is going to be returned.
2494 Severity : enhancement
2496 Description: Checksum enhancements
2497 Details : New checksum enhancements allow for resending RPCs that failed
2500 Severity : enhancement
2502 Description: Tunables on number of dirty pages in cacche
2503 Details : Allow to set limit on number of dirty pages cached.
2508 Description: client crash on unmount - lock still has references
2509 Details : In some error handling cases it was possible to leak a lock
2510 reference on a client while accessing a file. This was not
2511 harmful to the client during operation, but would cause the
2512 client to crash when the filesystem is unmounted.
2515 Frequency : specific case, rare
2517 Description: ETXTBSY on mds though file not in use by client
2518 Details : ETXTBSY is no longer incorrectly returned when attempting to
2519 chmod or chown a directory that the user previously tried to
2520 execute or a currently-executing binary.
2523 Frequency : extremely rare except on liblustre-based clients
2525 Description: Lustre space not freed when files are deleted
2526 Details : Clean up open-unlinked files after client eviction. Previously
2527 the unlink was skipped and the files remained as orphans.
2532 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2533 Details : In some cases it was possible to send a blocking callback to
2534 a client doing a glimpse, even though that client didn't get
2535 a lock granted. When the glimpse lock is cancelled on the OST
2536 the freed lock is left on the waiting list and corrupted the list.
2539 Frequency : all core dumps
2541 Description: Broke core dumps to lustre
2542 Details : Negative dentry may be unhashed if parent does not have UPDATE
2543 lock, but some callers, e.g. do_coredump, expect dentry to be
2544 hashed after successful create, hash it in ll_create_it.
2546 ------------------------------------------------------------------------------
2548 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2550 * Support for kernels:
2551 2.6.9-42.0.2.EL (RHEL 4)
2552 2.6.5-7.276 (SLES 9)
2553 2.4.21-40.EL (RHEL 3)
2554 2.6.12.6 vanilla (kernel.org)
2558 Frequency : always on RHEL 3
2560 Description: Number of open files grows over time
2561 Details : The number of open files grows over time, whether or not
2562 Lustre is started. This was due to a filp leak introduced
2563 by one of our kernel patches.
2565 ------------------------------------------------------------------------------
2567 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2569 * Support for kernels:
2570 2.6.9-42.EL (RHEL 4)
2571 2.6.5-7.276 (SLES 9)
2572 2.4.21-40.EL (RHEL 3)
2573 2.6.12.6 vanilla (kernel.org)
2578 Bugzilla : 5719, 9635, 9792, 9684
2579 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2580 Details : If a server is under heavy load and cannot reply to new
2581 connection requests before the client resends the (re)connect,
2582 the connection handling code can behave badly if two service
2583 threads are concurrently handing separate (re)connections from
2584 the same client. Add better locking to the connection handling
2585 code, and ensure that only a single connection will be processed
2586 for a given client UUID, even if the lock is dropped.
2588 Severity : enhancement
2590 Description: add TCP zero-copy support to kernel
2591 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2592 sends if the hardware supports scatter-gather and checksumming.
2593 This allows socklnd to do client-write and server-read more
2594 efficiently and reduce CPU utilization from skbuf copying.
2597 Frequency : only if NFS exporting from client
2599 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2600 Details : If setting attributes on a file created under NFS that had
2601 never been opened it would be possible to oops the client
2602 if the file had no objects.
2605 Frequency : always for liblustre
2607 Description: liblustre client does MDS+OSTs setattr RPC for each write
2608 Details : When doing a write from a liblustre client, the client
2609 incorrectly issued an RPC to the MDS and each OST the file was
2610 striped over in order to update the timestamps. When writing
2611 with small chunks and many clients this could overwhelm the MDS
2612 with RPCs. In all cases it would slow down the write because
2613 these RPCs are unnecessary.
2615 Severity : enhancement
2617 Description: allow number of MDS service threads to be changed at module load
2618 Details : It is now possible to change the number of MDS service threads
2619 running. Adding "options mds mds_num_threads={N}" to the MDS's
2620 /etc/modprobe.conf will set the number of threads for the next
2621 time Lustre is restarted (assuming the "mds" module is also
2622 reloaded at that time). The default number of threads will
2623 stay the same, 32 for most systems.
2628 Description: OST crash if filesystem is unformatted or corrupt
2629 Details : If an OST is started on a device that has never been formatted
2630 or if the filesystem is corrupt and cannot even mount then the
2631 error handling cleanup routines would dereference a NULL pointer.
2636 Description: NULL pointer deref in llap_from_page.
2637 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2638 incorrect) mapping if the page was truncated/reclaimed while it was
2639 searched for. Check for this condition and skip such pages when
2640 doing readahead. Introduce extra check to llap_from_page() to
2641 verify page->mapping->host is non-NULL (so page is not anonymous).
2644 Frequency : Sometimes when using sys_sendfile
2646 Description: "page not covered by a lock" warnings from ll_readpage
2647 Details : sendfile called ll_readpage without right page locks present.
2648 Now we introduced ll_file_sendfile that does necessary locking
2649 around call to generic_file_sendfile() much like we do in
2653 Frequency : with certain MDS communication failures at client mount time
2655 Description: NULL pointer deref after failed client mount
2656 Details : a client connection request may delayed by the network layer
2657 and not be sent until after the PTLRPC layer has timed out the
2658 request. If the client fails the mount immediately it will try
2659 to clean up before the network times out the request. Add a
2660 reference from the request import to the obd device and delay
2661 the cleanup until the network drops the request.
2664 Frequency : occasionally during client (re)connect
2666 Description: assertion failure during client (re)connect
2667 Details : processing a client connection request may be delayed by the
2668 client or server longer than the client connect timeout. This
2669 causes the client to resend the connection request. If the
2670 original connection request is replied in this interval, the
2671 client may trip an assertion failure in ptlrpc_connect_interpret()
2672 which thought it would be the only running connect process.
2675 Frequency : only with obd_echo servers and clients that are rebooted
2677 Description: kernel BUG accessing uninitialized data structure
2678 Details : When running an obd_echo server it did not start the ping_evictor
2679 thread, and when a client was evicted an uninitialized data
2680 structure was accessed. Start the ping_evictor in the RPC
2681 service startup instead of the OBD startup.
2683 Severity : enhancement
2684 Bugzilla : 10193 (patchless)
2685 Description: Remove dependency on various unexported kernel interfaces.
2686 Details : No longer need reparent_to_init, exit_mm, exit_files,
2687 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2690 Frequency : rare (only users of deprecated and unsupported LDAP config)
2692 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2693 Details : LDAP apparently contains 'lustreName' attributes instead of
2694 'name'. A simple remapping of the name is sufficient.
2697 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2699 Description: waiting_locks_callback trips kernel BUG if client is evicted
2700 Details : Running with the dump_on_timeout debug flag turned on makes
2701 it possible that the waiting_locks_callback() can try to dump
2702 the Lustre kernel debug logs from an interrupt handler. Defer
2703 this log dumping to the expired_lock_main() thread.
2705 Severity : enhancement
2707 Description: Support NFS exporting on 2.6 kernels.
2708 Details : Implement non-rawops metadata methods for NFS server to use without
2709 changing NFS server code.
2712 Frequency : very rare (synthetic metadata workload only)
2714 Description: two racing renames might cause an MDS thread to deadlock
2715 Details : Running the "racer" program may cause one MDS thread to rename
2716 a file from being the source of a rename to being the target of
2717 a rename at exactly the same time that another thread is doing
2718 so, and the second thread has already enqueued these locks after
2719 doing a lookup of the target and is trying to relock them in
2720 order. Ensure that we don't try to re-lock the same resource.
2723 Frequency : only very large systems with liblustre clients
2725 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2726 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2727 all OSTs in parallel. This allows the eviction of stale liblustre
2728 clients to proceed much faster than if they were done in series,
2729 and also offers similar improvements for other set_info RPCs.
2734 Description: excessive CPU usage during initial read phase on client
2735 Details : During the initial read phase on a client, it would agressively
2736 retry readahead on the file, consuming too much CPU and impacting
2737 performance (since 1.4.5.8). Improve the readahead algorithm
2738 to avoid this, and also improve some other common cases (read
2739 of small files in particular, where "small" is files smaller than
2740 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2745 Description: MDS crash when receiving packet with unknown intent.
2746 Details : Do not LBUG in unknown intent case, just return -EFAULT
2748 Severity : enhancement
2749 Bugzilla : 9293, 9385
2750 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2751 Details : Do not serialize getattr (non-intent version) and statfs.
2754 Frequency : occasional, when OST network is overloaded/intermittent
2756 Description: client evicted by OST after bulk IO timeout
2757 Details : If a client sends a bulk IO request (read or write) the OST
2758 may evict the client if it is unresposive to its data GET/PUT
2759 request. This is incorrect if the network is overloaded (takes
2760 too long to transfer the RPC data) or dropped the OST GET/PUT
2761 request. There is no need to evict the client at all, since
2762 the pinger and/or lock callbacks will handle this, and the
2763 client can restart the bulk request.
2766 Frequency : Always when mmapping file with no objects
2768 Description: client crashes when mmapping file with no objects
2769 Details : Check that we actually have objects in a file before doing any
2770 operations on objects in ll_vm_open, ll_vm_close and
2776 Description: Request leak when working with deleted CWD
2777 Details : Introduce advanced request refcount tracking for requests
2778 referenced from lustre intent.
2780 Severity : Enhancement
2782 Description: Cache open file handles on client.
2783 Details : MDS now will return special lock along with openhandle, if
2784 requested and client is allowed to hold openhandle, even if unused,
2785 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2786 closing files for every read/write openration.
2788 Severity : Enhancement
2790 Description: Cache open negative dentries on client when possible.
2791 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2792 negative dentries on lock revocation.
2797 Description: Remounting a client read-only wasn't possible with a zconf mount
2798 Details : It wasn't possible to remount a client read-only with llmount.
2800 Severity : enhancement
2801 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2802 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2803 MPICH is included to add Lustre-specific ADIO interfaces.
2804 This is based closely on the UFS ADIO layer and only differs
2805 in file creation, in order to allow the OST striping to be set.
2806 This is user-contributed code and not supported by CFS.
2811 Description: extended inode attributes (immutable, append-only) work improperly
2812 when 2.4 and 2.6 kernels are used on client/server or vice versa
2813 Details : Introduce kernel-independent values for these flags.
2815 Severity : enhancement
2818 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2819 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2820 granularity. Now, Fractional MB granularity is be supported,
2821 this is very useful for low memory system.
2823 Severity : enhancement
2825 Description: Getattr by fid
2826 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2827 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2828 have locks and avoids getattr rpc requests for it on every stat.
2831 Frequency : Always, for filesystems larger than 2TB
2833 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2834 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2835 filesystems larger than 2TB to be mounted with mballoc enabled.
2838 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2840 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2841 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2842 filesystems larger than 2TB is mounted then the kernel will
2843 silently corrupt the start of the filesystem. CONFIG_LBD is
2844 enabled for all CFS-supported kernels, but the possibility of
2845 this happening with a modified kernel config exists.
2847 Severity : enhancement
2849 Description: add client O_DIRECT support for 2.6 kernels
2850 Details : It is now possible to do O_DIRECT reads and writes to files
2851 in the Lustre client mountpoint on 2.6 kernel clients.
2853 Severity : enhancement
2855 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2856 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2857 parallel, not waiting for response from every OST before sending
2858 a rpc to the next OST.
2863 Description: setattr vs write race when updating file timestamps
2864 Details : Client processes that update a file timestamp into the past
2865 right after writing to the file (e.g. tar) it is possible that
2866 the updated file modification time can be reset to the current
2867 time due to a race between processing the setattr and write RPC.
2869 Severity : enhancement
2871 Description: Bring 'lfs find' closer in line with regular Linux find.
2872 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2873 -print0 options and obtains all the needed info through the lustre
2876 Severity : enhancement
2878 Description: support up to 1024 configured devices on one node
2879 Details : change obd_dev array from statically allocated to dynamically
2880 allocated structs as they are first used to reduce memory usage
2885 Description: Flush dirty partially truncated pages during truncate
2886 Details : Immediatelly flush partially truncated pages in filter_setattr,
2887 this way we completely avoid having any pages in page cache on OST
2888 and can retire ugly workarounds during writes to flush such pages.
2893 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2894 Details : protect inode from truncation within vfs_unlink() context
2895 just take a reference before calling vfs_unlink() and release it
2896 when parent's i_sem is free.
2901 Description: last_id value checked outside lock on OST caused LASSERT failure
2902 Details : If there were multiple MDS->OST object precreate requests in
2903 flight, it was possible that the OST's last object id was checked
2904 outside a lock and incorrectly tripped an assertion. Move checks
2905 inside locks, and discard old precreate requests.
2908 Frequency : always, if extents are used on OSTs
2910 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2911 Details : OSTs using the extents format would not zero the high 16 bits of
2912 the index physical block number. This is not a problem for any
2913 OST filesystems smaller than 16TB, and no kernels support ext3
2914 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2915 new/modified files) and can be fixed for existing filesystems
2916 with e2fsprogs-1.39-cfs1.
2921 Description: import connection selection may be incorrect if timer wraps
2922 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2923 import connections to be ignored if the 32-bit jiffies counter
2924 wraps. Use a 64-bit jiffies counter.
2927 Frequency : very large clusters immediately after boot
2929 Description: LNET request buffers exhausted under heavy short-term load
2930 Details : If a large number of client requests are generated on a service
2931 that has previously never seen so many requests it is possible
2932 that the request buffer growth cannot keep up with the spike in
2933 demand. Instead of dropping incoming requests, they are held in
2934 the LND until the RPC service can accept more requests.
2937 Frequency : Sometimes during replay
2939 Description: Assertion failure in ll_local_open after replay.
2940 Details : If replay happened on an open request reply before we were able
2941 to set replay handler, reply will become not swabbed tripping the
2942 assertion in ll_local_open. Now we set the handler right after
2943 recognising of open request
2946 Frequency : very rare
2948 Description: kernel reports "badness in vsnprintf"
2949 Details : Reading from the "recovery_status" /proc file in small chunks
2950 may cause a negative length in lprocfs_obd_rd_recovery_status()
2951 call to vsnprintf() (which is otherwise harmless). Exit early
2952 if there is no more space in the output buffer.
2954 Severity : enhancement
2956 Description: clear OBD RPC statistics by writing to them
2957 Details : It is now possible to clear the OBD RPC statistics by writing
2958 to the "stats" file.
2963 Description: Client mtime is not the same on different clients after utimes
2964 Details : In some cases, the client was using the utimes() syscall on
2965 a file cached on another node. The clients now validate the
2966 ctime from the MDS + OSTs to determine which one is right.
2971 Description: Inability to activate failout mode
2972 Details : lconf script incorrectly assumed that in pythong string's numeric
2973 value is used in comparisons.
2976 Frequency : always with multiple stripes per file
2978 Description: Inefficient object allocation for mutli-stripe files
2979 Details : When selecting which OSTs to stripe files over, for files with
2980 a stripe count that divides evenly into the number of OSTs,
2981 the MDS is always picking the same starting OST for each file.
2982 Return the OST selection heuristic to the original design.
2987 Description: mount failures may take full timeout to return an error
2988 Details : Under some heavy load conditions it is possible that a
2989 failed mount can wait for the full obd_timeout interval,
2990 possibly several minutes, before reporting an error.
2991 Instead return an error as soon as the status is known.
2993 Frequency : quota enabled and large files being deleted
2995 Description: releasing more than 4GB of quota at once hangs OST
2996 Details : If a user deletes more than 4GB of files on a single OST it
2997 will cause the OST to spin in an infinite loop. Release
2998 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3003 Description: statfs data retrieved from /proc may be stale or zero
3004 Details : When reading per-device statfs data from /proc, in the
3005 {kbytes,files}_{total,free,avail} files, it may appear
3006 as zero or be out of date.
3009 Frequency : systems with MD RAID1 external journal devices
3011 Description: lconf's call to blkid is confused by RAID1 journal devices
3012 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3013 of returning all block devices that match the journal UUID.
3016 Frequency : always, for aggregate stripe size over 4GB
3018 Description: assertion fails when trying to use 4GB stripe size
3019 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
3020 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
3023 Frequency : always on ppc64
3025 Description: the first write on an ext3 filesystem with mballoc got stuck
3026 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3027 perform endianness conversion.
3029 ------------------------------------------------------------------------------
3031 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3033 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3034 INTEROPERATE with older versions automatically. Please read the
3035 user documentation before upgrading any part of a live system.
3036 * WARNING: Lustre networking configuration changes are required with
3037 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3040 * Support for kernels:
3041 2.6.9-22.0.2.EL (RHEL 4)
3042 2.6.5-7.244 (SLES 9)
3043 2.6.12.6 vanilla (kernel.org)
3046 Severity : enhancement
3047 Bugzilla : 7981/8208
3048 Description: Introduced Lustre Networking (LNET)
3049 Details : LNET is new networking infrastructure for Lustre, it includes
3050 a reorganized network configuration mode (see the user
3051 documentation for full details) as well as support for routing
3052 between different network fabrics. Lustre Networking Devices
3053 (LNDs) for the supported network fabrics have also been
3054 created for this new infrastructure.
3056 Severity : enhancement
3057 Description: Introduced Access control lists
3058 Details : clients can set ACLs on files and directories in order to have
3059 more fine-grained permissions than the standard Unix UGO+RWX.
3060 The MDS must be started with the "-o acl" mount option.
3062 Severity : enhancement
3063 Description: Introduced filesystem quotas
3064 Details : Administrators may now establish per-user quotas on the
3067 Severity : enhancement
3069 Description: Configuration change for the XT3
3070 The PTLLND is now used to run Lustre over Portals on the XT3
3071 The configure option(s) --with-cray-portals are no longer used.
3072 Rather --with-portals=<path-to-portals-includes> is used to
3073 enable building on the XT3. In addition to enable XT3 specific
3074 features the option --enable-cray-xt3 must be used.
3079 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3080 Details : A race between allocating a new llcd and re-getting the llcd_lock
3081 allowed another thread to grab newly-allocated llcd.
3083 Severity : enhancement
3085 Description: 2.6 OST async journal commit and locking fix to improve performance
3086 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3087 2.6 kernels are now async as they already were in 2.4 kernels so
3088 that they can commit concurrently with the network bulk transfer.
3089 For block-allocated files the filter allocation semaphore is held
3090 to avoid filesystem fragmentation during allocation. BKL lock
3091 removed for 2.6 xattr operations where it is no longer needed.
3096 Description: lconf incorrectly determined whether two IP networks could talk
3097 Details : In some more complicated routing and multiple-network
3098 configurations, lconf will avoid trying to make a network
3099 connection to a disjoint part of the IP space. It was doing the
3100 math incorrectly for one set of cases.
3105 Description: Fix for potential infinite loop processing records in an llog.
3106 Details : If an llog record is corrupted/zeroed, it is possible to loop
3107 forever in llog_process(). Validate the llog record length
3108 and skip the remainder of the block on error.
3111 Frequency : occasional (liblustre only)
3113 Description: liblustre could not open files whose last component is a symlink
3114 Details : sysio_path_walk() would incorrectly pass the open intent to
3115 intermediate path components.
3118 Frequency : rare (liblustre only with non-standard tuning)
3119 Bugzilla : 7201 (7350)
3120 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3121 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3122 after it has been released, only do this for referenced locks
3124 Severity : enhancement
3126 Description: specifying an (invalid) directory default stripe_size of -1
3127 would reset the directory default striping
3128 Details : stripe_size -1 was used internally to signal directory stripe
3129 removal, now use "all default" to signal dir stripe removal
3130 as a directory striping of "all default" is not useful
3133 Frequency : common for large clusters running liblustre clients
3135 Description: doing an ls when liblustre clients are running is slow
3136 Details : sending a glimpse AST to a liblustre client waits for every AST
3137 to time out, as liblustre clients will not respond. Since they
3138 cannot cache data we refresh the OST lock LVB from disk instead.
3140 Severity : enhancement
3142 Description: doing an ls at the same time as file IO can be slow
3143 Details : enqueue and other "small" requests can be blocked behind many
3144 large IO requests. Create a new OST IO portal for non-IO
3145 requests so they can be processed faster.
3148 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3150 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3151 created file with mode 000
3152 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3153 allow an ftruncate() as a non-root user to a file with mode 000.
3154 Lustre now respects this flag to disable mode checks when
3155 truncating a file owned by the user
3158 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3161 Description: Revoking locks from clients that went dead or catatonic might take
3163 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3164 cancellation of such locks instant on servers without waiting for
3165 any reply from clients. Clients drops these locks when cancel
3166 notification from server is received without replying.
3169 Frequency : liblustre-only, when liblustre client dies or becomes busy
3171 Description: Doing ls on Linux clients can take a long time with active
3173 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3174 granting such locks to it in the first place if possible. Locks
3175 are taken by proxy on the OST during the read or write and
3176 dropped immediately afterward. Add connect flags handling, do
3177 not grant locks to liblustre clients for glimpse ASTs.
3179 Severity : enhancement
3181 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3182 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3183 sequential reads and behaves badly with applications doing random
3184 reads. Improve it by reading ahead at least read region, and
3185 avoiding excessive large RPC for small reads.
3187 Severity : enhancement
3189 Description: Creating more than 1000 files for a single job may cause a load
3190 imbalance on the OSTs if there are also a large number of OSTs.
3191 Details : qos_prep_create() uses an OST index reseed value that is an
3192 even multiple of the number of available OSTs so that if the
3193 reseed happens in the middle of the object allocation it will
3194 still utilize the OSTs as uniformly as possible.
3199 Description: OST or MDS may oops in ping_evictor_main()
3200 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3201 but doesn't restart at beginning of obd_exports_timed list
3204 Severity : enhancement
3206 Description: improve by-nid export eviction on the MDS and OST
3207 Details : allow multiple exports with the same NID to be evicted at one
3208 time without re-searching the exports list.
3211 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3213 Description: MDS may oops in groups_free()
3214 Details : in rare race conditions a newly allocated group_info struct is
3215 freed again, and this can be NULL. The 2.4 compatibility code
3216 for groups_free() checked for a NULL pointer, but 2.6 did not.
3219 Frequency : common for liblustre clients doing little filesystem IO
3220 Bugzilla : 9352, 7313
3221 Description: server may evict liblustre clients accessing contended locks
3222 Details : if a client is granted a lock or receives a completion AST
3223 with a blocking AST already set it would not reply to the AST
3224 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3227 Frequency : lfs setstripe, only systems with more than 160 OSTs
3229 Description: unable to set striping with a starting offset beyond OST 160
3230 Details : llapi_create_file() incorrectly limited the starting stripe
3231 index to the maximum single-file stripe count.
3234 Frequency : LDAP users only
3236 Description: lconf did not handle in-kernel recovery with LDAP properly
3237 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3239 Severity : enhancement
3241 Description: bind OST threads to NUMA nodes to improve performance
3242 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3243 node and do their allocations there to localize memory access
3245 Severity : enhancement
3247 Description: llmount can determine client NID directly from Myrinet (GM)
3248 Details : the client NID code from gmnalnid was moved directly into
3249 llmount, removing the need to use this or specifying the
3250 client NID explicitly when mounting GM clients with zeroconf
3253 Frequency : if client is started with down MDS
3255 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3256 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3257 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3258 ptlrpc_import_delayed_req() because we didn't check if the
3259 request was interrupted, and we also didn't break out of the
3260 event loop if there was no timeout
3265 Description: data loss during non-page-aligned writes to a single file from
3266 both multiple nodes and multiple threads on one node at same time
3267 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3268 inconsistency led to false short-reads, that were cached and later
3269 used by ->prepare_write() to fill in partially written page,
3270 leading to data loss.
3273 Frequency : always, if lconf --abort_recovery used
3275 Description: lconf --abort_recovery fails with 'Operation not supported'
3276 Details : lconf was attempting to abort recovery on the MDT device and not
3279 Severity : enhancement
3281 Description: remove cleanup logs
3282 Details : replace lconf-generated cleanup logs with lustre internal
3283 cleanup routines. Eliminates the need for client-cleanup and
3286 Severity : enhancement
3288 Description: add support for EAs (user and system) on lustre filesystems
3289 Details : it is now possible to store extended attributes in the Lustre
3290 client filesystem, and with the user_xattr mount option it
3291 is possible to allow users to store EAs on their files also
3293 Severity : enhancement
3295 Description: Add possibility (config option) to show minimal available OST free
3297 Details : When compiled with --enable-mindf configure option, statfs(2)
3298 (and so, df) will return least minimal free space available from
3299 all OSTs as amount of free space on FS, instead of summary of
3300 free spaces of all OSTs.
3302 Severity : enhancement
3304 Description: do not expand extent locks acquired on OST-side
3305 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3306 by server: they are not cached anyway.
3309 Frequency : when mmap is used/binaries executed from Lustre
3311 Description: Unmmap pages before throwing them away from read cache.
3312 Details : llap_shrink cache now attempts to unmap pages before discarding
3313 them (if unmapping failed - do not discard). SLES9 kernel has
3314 extra checks that trigger if this unmapping is not done first.
3319 Description: lconf didn't resolve symlinks before checking to see whether a
3320 given mountpoint was already in use
3323 Frequency : when migrating failover services
3324 Bugzilla : 6395, 9514
3325 Description: When migrating a subset of services from a node (e.g. failback
3326 from a failover service node) the remaining services would
3327 time out and evict clients.
3328 Details : lconf --force (implied by --failover) sets the global obd_timeout
3329 to 5 seconds in order to quickly disconnect, but this caused
3330 other RPCs to time out too quickly. Do not change the global
3331 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3333 Severity : enhancement
3334 Frequency : if MDS is started with down OST
3335 Bugzilla : 9439,5706
3336 Description: Allow startup/shutdown of an MDS without depending on the
3337 availability of the OSTs.
3338 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3339 Add appropriate locking and lov-osc refcounts for safe
3340 cleaning. Add osc abort_inflight calls in case the
3341 synchronize never started.
3344 Frequency : occasional (Cray XT3 only)
3346 Description: root not authorized to access files in CRAY_PORTALS environment
3347 Details : The client process capabilities were not honoured on the MDS in
3348 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3349 been accessed by an authorized user then root was able to access
3350 the file on the local client also. The root user capabilities
3351 are now allowed on the MDS, as this environment has secure UID.
3354 Frequency : occasional
3356 Description: ldiskfs "too long searching" message happens too often
3357 Details : A debugging message (otherwise harmless) prints too often on
3358 the OST console. This has been reduced to only happen when
3359 there are fragmentation problems on the filesystem.
3364 Description: Division by zero in statfs when all OSCs are inactive
3365 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3366 lov_statfs divides by value returned by lov_get_stripecnt().
3370 Bugzilla : 9489, 3273
3371 Description: First write from each client to each OST was only 4kB in size,
3372 to initialize client writeback cache, which caused sub-optimal
3373 RPCs and poor layout on disk for the first writen file.
3374 Details : Clients now request an initial cache grant at (re)connect time
3375 and so that they can start streaming writes to the cache right
3376 away and always do full-sized RPCs if there is enough data.
3377 If the OST is rebooted the client also re-establishes its grant
3378 so that client cached writes will be honoured under the grant.
3383 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3384 Details : Now I/O RPCs go to different portal number and (presumably) fast
3385 lock requests (and glimses) and other RPCs get their own service
3386 threads pool that should be able to service those RPCs
3389 Severity : enhancement
3391 Description: Ability to exchange lustre version between client and servers and
3392 issue warnings at client side if client is too old. Also for
3393 liblustre clients there is ability to refuse connection of too old
3395 Details : New 'version' field is added to connect data structure that is
3396 filled with version info. That info is later checked by server and
3400 Frequency : rare, liblustre only.
3401 Bugzilla : 9296, 9581
3402 Description: Two simultaneous writes from liblustre at offset within same page
3403 might proceed at the same time overwriting eachother with stale
3405 Details : I/O lock withing llu_file_prwv was released too early, before data
3406 actually was hitting the wire. Extended lock-holding time until
3407 server acknowledges receiving data.
3410 Frequency : extremely rare. Never observed in practice.
3412 Description: avoid generating lustre_handle cookie of 0.
3413 Details : class_handle_hash() generates handle cookies by incrementing
3414 global counter, and can hit 0 occasionaly (this is unlikely, but
3415 not impossible, because initial value of cookie counter is
3416 selected randonly). Value of 0 is used as a sentinel meaning
3417 "unassigned handle" --- avoid it. Also coalesce two critical
3418 sections in this function into one.
3420 Severity : enhancement
3422 Description: allow liblustre clients to delegate truncate locking to OST
3423 Details : To avoid overhead of locking, liblustre client instructs OST to
3424 take extent lock in ost_punch() on client's behalf. New connection
3425 flag is added to handle backward compatibility.
3427 Severity : enhancement
3428 Bugzilla : 4928, 7341, 9758
3429 Description: allow number of OST service threads to be specified
3430 Details : a module parameter allows the number of OST service threads
3431 to be specified via "options ost ost_num_threads={N}" in the
3432 OSS's /etc/modules.conf or /etc/modprobe.conf.
3436 Bugzilla : 6146, 9635, 9895
3437 Description: servers crash with bad pointer in target_handle_connect()
3438 Details : In rare cases when a client is reconnecting it was possible that
3439 the connection request was the last reference for that export.
3440 We would temporarily drop the export reference and get a new
3441 one, but this may have been the last reference and the export
3442 was just destroyed. Get new reference before dropping old one.
3444 Severity : enhancement
3445 Frequency : if client is started with failover MDS
3447 Description: Allow multiple MDS hostnames in the mount command
3448 Details : Try to read the configuration from all specified MDS
3449 hostnames during a client mount in case the "primary"
3452 Severity : enhancement
3454 Description: Stop sending data to evicted clients as soon as possible.
3455 Details : Check if the client we are about to send or are sending data to
3456 was evicted already. (Check is done every second of waiting,
3457 for which l_wait_event interface was extended to allow checking
3458 of exit condition at specified intervals).
3461 Frequency : rare, normally only when NFS exporting is done from client
3463 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3465 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3468 Frequency : always, if srand() is called before liblustre initialization
3470 Description: Liblustre uses system PRNG disturbing its usage by user application
3471 Details : Introduce internal to lustre fast and high-quality PRNG for
3472 lustre usage and make liblustre and some other places in generic
3473 lustre code to use it.
3475 Severity : enhancement
3476 Bugzilla : 9477, 9557, 9870
3477 Description: Verify that the MDS configuration logs are updated when xml is
3478 Details : Check if the .xml configuration logs are newer than the config
3479 logs stored on the MDS and report an error if this is the case.
3480 Request --write-conf, or allow starting with --old_conf.
3482 Severity : enhancement
3484 Description: Handle symlinks in the path when checking if Lustre is mounted.
3485 Details : Resolve intermediate symlinks when checking if a client has
3486 mounted a filesystem to avoid duplicate client mounts.
3491 Description: lconf can hit an error exception but still return success.
3492 Details : The lconf command catches the Command error exception at the top
3493 level script context and will exit with the associated exit
3494 status, but doesn't ensure that this exit status is non-zero.
3499 Description: failure of ptlrpc thread startup can cause oops
3500 Details : Starting a ptlrpc service thread can fail if there are a large
3501 number of threads or the server memory is very fragmented.
3502 Handle this without oopsing.
3505 Frequency : always, only if liblustre and non-default acceptor port was used
3507 Description: liblustre cannot connect to servers with non-default acceptor port
3508 Details : tcpnal_set_default_params() was not called and was therefore
3509 ignoring the environment varaible TCPNAL_PORT, as well as other
3510 TCPNAL_ environment variables
3515 Description: two objects could be created on the same OST for a single file
3516 Details : If an OST is down, in some cases it was possible to create two
3517 objects on a single OST for a single file. No problems other
3518 than potential performance impact and spurious error messages.
3522 Bugzilla : 5681, 9562
3523 Description: Client may oops in ll_unhash_aliases
3524 Details : Client dcache may become inconsistent in race condition.
3525 In some cases "getcwd" can fail if the current directory is
3531 Description: Inode refcounting problems in NFS export code
3532 Details : link_raw functions used to call d_instantiate without obtaining
3533 extra inode reference first.
3537 Bugzilla : 9942, 9903
3538 Description: Referencing freed requests leading to crash, memleaks with NFS.
3539 Details : We used to require that call to ll_revalidate_it was always
3540 followed by ll_lookup_it. Also with revalidate_special() it is
3541 possible to call ll_revalidate_it() twice for the same dentry
3542 even if first occurence returned success. This fix changes semantic
3543 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3544 reference on a request referred from the intent.
3545 ll_intent_release() then releases such a request.
3548 Frequency : rare, normally benchmark loads only
3550 Description: unlinked inodes were kept in memory on the client
3551 Details : If a client is repeatedly creating and unlinking files it
3552 can accumulate a lot of stale inodes in the inode slab cache.
3553 If there is no other client load running this can cause the
3554 client node to run out of memory. Instead flush old inodes
3555 from client cache that have the same inode number as a new inode.
3558 Frequency : SLES9 2.6.5 kernel and long filenames only
3559 Bugzilla : 9969, 10379
3560 Description: utime reports stale NFS file handle
3561 Details : SLES9 uses out-of-dentry names in some cases, which confused
3562 the lustre dentry revalidation. Change it to always use the
3566 Frequency : rare, unless heavy write-truncate concurrency is continuous
3567 Bugzilla : 4180, 6984, 7171, 9963, 9331
3568 Description: OST becomes very slow and/or deadlocked during object unlink
3569 Details : filter_destroy() was holding onto the parent directory lock
3570 while truncating+unlinking objects. For very large objects this
3571 may block other threads for a long time and slow overall OST
3572 responsiveness. It may also be possible to get a lock ordering
3573 deadlock in this case, or run out of journal credits because of
3574 the combined truncate+unlink. Solution is to do object truncate
3575 first in one transaction without parent lock, and then do the
3576 final unlink in a new transaction with the parent lock. This
3577 reduces the lock hold time dramatically.
3580 Frequency : rare, 2.4 kernels only
3582 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3583 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3584 before unmounting their filesystem in order to avoid deadlock.
3585 The kernel_locked() function in 2.4 kernels only checks whether
3586 the kernel lock is held, not whether it is this process that is
3587 holding it as 2.6 kernels do.
3592 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3593 Details : The client ptlrpc code may be trying to reconnect to a down
3594 server before a previous connection attempt has timed out.
3595 Increase the reconnect interval to be longer than the connection
3596 timeout interval to avoid sending duplicate connections to
3600 Frequency : echo_client brw_test command
3602 Description: fix echo_client to work with OST preallocated code
3603 Details : OST preallocation code (5137) didn't take echo_client IO path
3604 into account: echo_client calls filter methods outside of any
3605 OST thread and, hence, there is no per-thread preallocated
3606 pages and buffers to use. Solution: hijack pga pages for IO. As
3607 a byproduct, this avoids unnecessary data copying.
3611 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3612 Description: Client can oops in mdc_commit_close() after open replay
3613 Details : It was possible for the MDS to return an open request with no
3614 transaction number in mds_finish_transno() if the client was
3615 evicted, but without actually returning an error. Clients
3616 would later try to replay that open and may trip an assertion
3617 Simplify the client close codepath, and always return an error
3618 from the MDS in case the open is not successful.
3621 Frequency : rare, 2.6 OSTs only
3623 Description: OST may deadlock under high load on fragmented files
3624 Details : If there was a heavy load and highly-fragmented OST filesystems
3625 it was possible to have all the OST threads deadlock waiting on
3626 allocation of biovecs, because the biovecs were not released
3627 until the entire RPC IO was completed. Instead, release biovecs
3628 as soon as they are complete to ensure forward IO progress.
3630 Severity : enhancement
3632 Description: Support for specifying external journal device at mount
3633 Details : If an OST or MDS device is formatted with an external journal
3634 device, this device major/minor is stored in the ext3 superblock
3635 and may not be valid for failover. Allow detecting and
3636 specifying the external journal at mount time.
3641 Description: Mounting an MDS with pending unlinked files may cause oops
3642 Details : target_finish_recovery() calls mds_postrecov() which returned
3643 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3644 considers this an error and immediately begins cleaning up the
3645 lov, just after starting the mds_lov process
3647 Severity : enhancement
3649 Description: Implement 'lfs df' to report actual free space on per-OST basis
3650 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3651 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3652 usage of objects; '-h' to report in human readable format.
3654 ------------------------------------------------------------------------------
3656 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3663 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3664 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3665 at startup during orphan unlinking. Instead, skip buddy update
3666 before setup, as it will be regenerated after recovery is complete.
3671 Description: If an OST is inactive, its locks might reference stale inodes.
3672 Details : lov_change_cbdata() must iterate over all namespaces, even if
3673 they are inactive to clear inode references from the lock.
3675 Severity : enhancement
3676 Frequency : occasional, if non-standard max_dirty_mb used
3678 Description: Client will block write RPCs if not enough grant
3679 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3680 then the client will block writes while waiting for another RPC
3681 to complete instead of consuming its dirty limit. With change
3682 we get improved performance when max_dirty_mb is small.
3684 Severity : enhancement
3685 Bugzilla : 3389, 6253
3686 Description: Add support for supplementary groups on the MDS.
3687 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3688 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3689 lookups for user supplementary groups into a cache.
3693 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3694 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3695 for liblustre O_EXCL is enforced by MDS.
3699 Description: readdir never returns NULL in liblustre.
3700 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3701 dentry in struct dirent.
3705 Frequency : liblustre only, depends on application IO pattern
3706 Description: liblustre clients evicted if not contacting servers
3707 Details : Don't put liblustre clients into the ping_evictor list, so
3708 they will not be evicted by the pinger ever.
3710 Severity : enhancement
3712 Description: Add ability to evict clients by NID from MDS.
3713 Details : By echoing "nid:$NID" string into
3714 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3715 $NID would be instantly evicted from this MDS and from all active
3716 OSTs connected to it.
3720 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3721 Details : lookup_it_finish() used to query file size from OSTs that was not
3726 Description: service threads change working directory to that of init
3727 Details : Starting lustre service threads may pin the working directory
3728 of the parent thread, making that filesystem busy. Threads
3729 now change to the working directory of init to avoid this.
3733 Frequency : during shutdown only
3734 Description: shutdown with a failed MDS or OST can cause unmount to hang
3735 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3740 Frequency : 2.6 only
3741 Description: chmod/chown may include an extra supplementary group
3742 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3743 supplementary group and if none is specified this is used.
3746 Bugzilla : 5479 (6816)
3748 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3749 Details : If lookup is in progress on a file that is unlinked we might try
3750 to revalidate the inode and fail in revalidate after lookup is
3751 complete and ll_file_open() enqueues the open again but
3752 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3755 Frequency : always, if lconf --abort_recovery used
3757 Description: lconf --abort_recovery fails with 'Operation not supported'
3758 Details : lconf was attempting to abort recovery on the MDT device and not
3761 ------------------------------------------------------------------------------
3763 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3768 Frequency : rare (only unsupported configurations with a node running as an
3770 Bugzilla : 6514, 5137
3771 Description: Mounting a Lustre file system on a node running as an OST could
3773 Details : OSTs now preallocates memory needed to write out data at
3774 startup, instead of when needed, to avoid having to
3775 allocate memory in possibly low memory situations.
3776 Specifically, if the file system is mounted on on OST,
3777 memory pressure could force it to try to write out data,
3778 which it needed to allocate memory to do. Due to the low
3779 memory, it would be unable to do so and the node would
3780 become unresponsive.
3782 Severity : enhancement
3784 Description: Addition of lconf --service command line option
3785 Details : lconf now accepts a '--service <arg>' option, which is
3786 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3788 Severity : enhancement
3790 Description: Failover mode is now the default for OSTs.
3791 Details : By default, OSTs will now run in failover mode. To return to
3792 the old behaviour, add '--failout' to the lmc line for OSTs.
3794 Severity : enhancement
3796 Description: Health checks are now provided for MDS and OSTs
3797 Details : Additional detailed health check information on MSD and OSTs
3798 is now provided through the procfs health_check value.
3801 Frequency : occasional, depends on IO load
3803 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3804 after numerous create/delete cycles
3805 Details : The ext3 inode allocation policy would not allocate new inodes
3806 very well on the OSTs because there are no new directories
3807 being created. Instead we look for groups with free space if
3808 the parent directories are nearly full.
3813 Description: Network or server problems during mount may cause partially
3814 mounted clients instead of returning an error.
3815 Details : The config llog parsing code may overwrite the error return
3816 code during mount error handling, returning success instead
3822 Description: MDS can fail to allocate large reply buffers
3823 Details : After long uptimes the MDS can fail to allocate large reply
3824 buffers (e.g. zconf client mount config records) due to memory
3825 fragmentation or consumption by the buffer cache. Preallocate
3826 some large reply buffers so that these replies can be sent even
3827 under memory pressure.
3831 Frequency : rare (liblustre)
3832 Description: fsx running with liblustre complained that using truncate() to
3833 extend the file doesn't work. This patch corrects that issue.
3834 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3835 ATTR_SIZE and lsm use in llu_setattr_raw.
3839 Frequency : rare, only 2.6 kernels
3840 Description: Unusual file access patterns on the MDS may result in inode
3841 data being lost in very rare circumstances.
3842 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3843 code in the 2.6 kernel caused the inode and/or EA data not to
3844 be read from disk, causing single-file corruption.
3848 Frequency : rare, only 2.6 filesystems using extents
3849 Description: Heavy concurrent write and delete load may cause data corruption.
3850 Details : It was possible under high-load situations to have an extent
3851 metadata block in the block device cache from a just-unlinked
3852 file overwrite a newly-allocated data block. We now unmap any
3853 metadata buffers that alias just-allocated data blocks.
3857 Frequency : filesystems with default stripe_count larger than 77
3858 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3859 Details : lconf specifies an inode size of 4096 bytes when the default
3860 stripe_count is larger than 77. This conflicts with the default
3861 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3862 this case to avoid pinning too much memory for large EAs.
3864 ------------------------------------------------------------------------------
3866 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3871 Frequency : rare (extremely heavy IO load with hundreds of clients)
3873 Description: Client is evicted, gets IO error writing to file
3874 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3875 caused clients to be evicted for AST timeouts. The fixes in
3876 bug 5192 mean we no longer need to have such short AST timeouts
3877 so ldlm_timeout has been increased.
3880 Frequency : occasional during --force or --failover shutdown under load
3881 Bugzilla : 5949, 4834
3882 Description: Server oops/LBUG if stopped with --force or --failover under load
3883 Details : a collection of import/export refcount and cleanup ordering
3884 issues fixed for safer force cleanup
3887 Frequency : only filesystems larger than 120 OSTs
3888 Bugzilla : 5990, 6223
3889 Description: lfs getstripe would oops on a very large filesystem
3890 Details : lov_getconfig used kfree on vmalloc'd memory
3893 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3894 Bugzilla : 6242, 6243
3895 Description: reading from files that had been truncated to a non-zero size
3896 but never opened returned no data
3897 Details : ll_file_read() reads zeros from no-object files to EOF
3902 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3903 Details : There's little harm in aborting MDS/OSS recovery and letting it
3904 try again, so I removed the LASSERT and return an error instead.
3906 Severity : enhancement
3908 Description: New debugging infrastructure for tracking down data corruption
3909 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3910 (b) cover more of the client-side code path, and (c) try to narrow
3911 down where problems occurred
3915 Bugzilla : 3819, 4364, 4397, 6313
3916 Description: Racing close and eviction MDS could cause assertion in mds_close
3917 Details : It was possible to get multiple mfd references during close and
3918 client eviction, leading to one thread referencing a freed mfd.
3920 Severity: : enhancement
3921 Bugzilla : 3262, 6359
3922 Description: Attempts to reconnect to servers are now more aggressive.
3923 Details : This builds on the enhanced upcall-less recovery that was added
3924 in 1.4.2. When trying to reconnect to servers, clients will
3925 now try each server in the failover group every 10 seconds. By
3926 default, clients would previously try one server every 25 seconds.
3931 Description: After recovery, certain operations trigger a failed
3932 assertion on a client.
3933 Details : Failing over an mds, using lconf -d --failover, while a
3934 client was doing a readdir() call would cause the client to
3935 LBUG after recovery completed and the readdir() was resent.
3937 Severity : enhancement
3939 Description: Default groups are now added by lconf
3940 Details : You can now run lconf --group <servicename> without having to
3941 manually add groups with lmc.
3944 Frequency : occasional
3946 Description: Nodes with an elan id of 0 trigger a failed assertion
3949 Frequency : always when accessing e.g. tty/console device nodes
3951 Description: tty and some other devices nodes cannot be used on lustre
3952 Details : file's private_data field is used by device data and lustre
3953 values in there got lost. New field was added to struct file to
3954 store fs-specific private data.
3957 Frequency : when exporting Lustre via NFS
3959 Description: NFSD failed occasionally when looking up a path component
3960 Details : NFSD is looking up ".." which was broken in ext3 directories
3961 that had grown large enough to become hashed.
3964 Frequency : Clusters with multiple interfaces not on the same subnet
3966 Description: Nodes will repeatedly try to reconnect to an interface which it
3967 cannot reach and report an error to the log.
3968 Details : Extra peer list entries will be created by lconf with some peers
3969 unreachable. lconf now validates the peer before adding it.
3972 Frequency : Only if a default stripe is set on the filesystem root.
3974 Description: Setting a default stripe on the filesystem root prevented the
3975 filesystem from being remounted.
3976 Details : The client was sending extra request flags in the root getattr
3977 request and did not allocate a reply buffer for the dir EA.
3980 Frequency : occasional, higher if lots of files are accessed by one client
3981 Bugzilla : 6159, 6097
3982 Description: Client trips assertion regarding lsm mismatch/magic
3983 Details : While revalidating inodes the VFS looks up inodes with ifind()
3984 and in rare cases can find an inode that is being freed.
3985 The ll_test_inode() code will free the lsm during ifind()
3986 when it finds an existing inode and then the VFS later attaches
3987 this free lsm to a new inode.
3991 Bugzilla : 6422, 7030
3992 Description: MDS deadlock between mkdir and client eviction
3993 Details : Creating a new file via mkdir or mknod (starting a transaction
3994 and getting the ns lock) can deadlock with client eviction
3995 (gets ns lock and trying to finish a synchronous transaction).
3998 Frequency : occasional
3999 Description: While starting a server, the fsfilt_ext3 module could not be
4001 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
4002 kernels. Previously, lconf would still use the ext3 name
4003 when trying to load modules. Now, it will correctly use
4004 ext3 on 2.4 and ldiskfs on 2.6.
4006 Severity : enhancement
4007 Description: The default stripe count has been changed to 1
4008 Details : The interpretation of the default stripe count (0, to lfs
4009 or lmc) has been changed to mean striping across a single
4010 OST, rather than all available. For general usage we have
4011 found a stripe count of 1 or 2 works best.
4013 Severity : enhancement
4014 Description: Add support for compiling against Cray portals.
4015 Details : Conditional compiling for some areas that are different
4019 Frequency : occasional
4020 Bugzilla : 6409, 6834
4021 Description: Creating files with an explicit stripe count may lead to
4022 a failed assertion on the MDS
4023 Details : If some OSTs are full or unavailable, creating files may
4024 trigger a failed assertion on the MDS. Now, Lustre will
4025 try to use other servers or return an error to the
4029 Frequency : occasional
4031 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4032 caused lock timeout during readahead (since 1.4.2).
4033 Details : Processes doing readahead might match a lock that hasn't been
4034 granted yet if there are overlapping and conflicting lock
4035 requests. The readahead process waits on ungranted lock
4036 (original lock is CBPENDING), while OST waits for that process
4037 to cancel CBPENDING read lock and eventually evicts client.
4039 Severity : enhancement
4041 Description: Initial enabling of flock support for clients
4042 Details : Implements fcntl advisory locking and file status functions.
4043 This feature is provided as an optional mount flag (default
4044 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4045 locking are implemented yet, and those that are are not guaranteed
4046 to be completely correct in production environments.
4047 mount -t lustre -o [flock|noflock] ...
4050 Frequency : occasional
4052 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4053 assertion in the ext3 JBD (journaling) layer.
4054 Details : The b_committed_data struct is protected by the big kernel lock
4055 in 2.4 kernels, serializing journal_commit_transaction() and
4056 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4057 access to this struct. In 2.6 kernels there is finer grained
4058 locking to improve SMP performance of the JBD layer.
4062 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4067 Description: merge and cleanup kernel patches.
4070 -----------------------------------------------------------------------------
4072 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4074 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4075 versions of Lustre. Please update all servers and clients to
4076 version 1.4.2 or later at the same time. You must also run
4077 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4078 to update the configuration logs.
4080 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4081 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4082 - swab llog records of type '0' so we get proper header size/idx (5861)
4083 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4084 - fix rename of one directory over another leaking an inode (5953)
4085 - avoid SetPageDirty on 2.6 (5981)
4086 - don't re-add just-being-destroyed locks to the waiting list (5653)
4087 - when creating new directories, inherit the parent's custom
4088 striping settings if present parent (3048)
4089 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4090 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4091 - don't hold client locks on temporary worklist from l_lru (5666)
4092 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4093 - automatically evict dead clients (5921)
4094 - Update file size properly in create+truncate+fstat case (6196)
4095 - Do not unhash mountpoint dentries, do not allow removal of
4097 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4098 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4099 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4100 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4101 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4102 - Correctly handle reads of files with no objects (6243)
4103 - lctl recover will also mark a device active if deactivate used (5933)
4105 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
4106 - allow --write-conf on an MDS with different nettype than client (5619)
4107 - don't write config llogs to MDS for mounts not from that MDS (5617)
4108 - lconf should create multiple TCP connections from a client (5201)
4109 - init scripts are now turned off by default; run chkconfig --on
4110 lustre and chkconfig --on lustrefs to use them
4111 - upcalls are no longer needed for clients to recover to failover
4113 - add --abort-recovery option to lconf to abort recovery on device
4115 - add support for an arbitrary number of OSTs (3026)
4116 - Quota support protocol changes.
4117 - forward compatibility changes to wire structs (6007)
4118 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
4119 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
4120 - improved "lustre status" script
4121 - initialize blocksize for non-regular files (6062)
4122 - added --disable-server and --disable-client configure options (5782)
4123 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
4124 - Vanilla 2.4.29 support
4125 - increase maximum number of obd devices to 520 (6242)
4126 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4127 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4128 SLES 9 2.6.5 kernels
4129 - limit stripes per file to 160 (the maximum EA size) (6093)
4131 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4134 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4135 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4136 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4137 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4138 - small ext3 extents cleanups and fixes (5733)
4139 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4140 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4141 - avoid needless client->OST connect, fix handle mismatch (5317)
4142 - fix DLM error path that led to out-of-sync client, long delays (5779)
4143 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4144 - fix several locking issues related to i_size (5492,5624,5654,5672)
4145 - don't move pending lock onto export if it is already evicted (5683)
4146 - fix kernel oops when creating .foo in unlinked directory (5548)
4147 - fix deadlock in obdfilter statistics vs. object create (5811)
4148 - use time_{before,after} to avoid timer jiffies wrap (5882)
4149 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4150 - Do not leak request if server was not able to process it (5154)
4151 - If mds_open unable to find parent dir, make that negative lookup(5154)
4152 - don't create new directories with extent-mapping (5909, 5936)
4154 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
4155 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
4156 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
4157 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
4158 - support for RHEL4 kernel on i686 (5773)
4159 - provide error messages when incompatible logs are encountered (5898)
4161 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
4162 * version 1.4.0.10 (1.4.1 release candidate 1)
4164 - don't keep a lock reference when lock is not granted (4238)
4165 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4166 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
4167 - drop import inflight refcount on signal_completed_replay error (5255)
4168 - unlock page after async write error during send (3677)
4169 - handle missing objects in filter_preprw_read properly (5265)
4170 - no transno return for symlink open, don't save no-trasno open (3440)
4171 - don't try to complete elan receive that already failed (4012)
4172 - free RPC server reply state on error (5406)
4173 - clean up thread from ptlrpc_start_thread() on error (5160)
4174 - readahead could read extra page into cache that wasn't ejected (5388)
4175 - prevent races in class_attach/setup/cleanup/detach (5260)
4176 - don't dereference de->d_inode after l_dput of de (5458)
4177 - use "int" for stripe value returned from lock_to_stripe (5544)
4178 - mballoc allocation and error-checking fixes in 2.6 (5504)
4179 - block device patches to fix I/O request sizes in 2.6 (5482)
4180 - look up hostnames for IB nals (5602)
4181 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
4182 - don't start multiple acceptors for the same port (5277)
4183 - fix incorrect LASSERT in mds_getattr_name (5635)
4184 - export a proc file for general "ping" checking (5628)
4185 - fix "lfs check" to not block when the MDS is down (5628)
4187 - service request history (4965)
4188 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4189 - create an "evict_client" /proc entry on OSTs, like the MDS has
4190 - fix mount usage message, return errors per mount(8) (5168)
4191 - change grep [] to grep "[]" in tests so they work in more UMLs
4192 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4193 - remove ancient LOV_MAGIC_V0 EA support (5047)
4194 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4195 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4196 - allow readahead allocations to fail when low on memory (5383)
4197 - mmap locking landed again, after considerable improvement (2828)
4198 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4200 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4203 - send OST transaction number in read/write reply to free req (4966)
4204 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4205 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4206 - the watchdog thread now runs as interruptible (5246)
4207 - flock/lockf fixes (but it's still disabled, pending 5135)
4208 - don't use EXT3 constants in llite code (5094)
4209 - memory shortage at startup could cause assertion (5176)
4211 - reorganization of lov code
4212 - single portals codebase
4214 - add extents/mballoc support (5025)
4215 - direct I/O reads in the obdfilter (4048)
4216 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4218 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4221 - send OST transaction number in read/write reply to free req (4966)
4222 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4223 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4224 - the watchdog thread now runs as interruptible (5246)
4225 - handle missing objects in filter_preprw_read properly (5265)
4226 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4227 - drop import inflight refcount on signal_completed_replay error (5255)
4228 - unlock page after async write error during send (3677)
4229 - return original error code on reconstructed replies (3761)
4230 - no transno return for symlink open, don't save no-trasno open (3440)
4232 - add pid to ldlm debugging output (4922)
4233 - bump the watchdog timeouts -- we can't handle 30sec yet
4234 - extra debugging for orphan dentry/inode bug (5259)
4236 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
4239 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4240 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4241 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
4242 - fix NULL dereference in /proc/sys/portals/routes (4827)
4243 - allow failed mdc_close() operations to be interrupted (4561)
4244 - stop precreate on OST before MDS would time out on it (4778)
4245 - don't send partial-page writes before EOF from client (4410)
4246 - discard client grant for sub-page writes on large-page clients (4520)
4247 - don't free dentries not owned by NFS code, check generation (4806)
4248 - fix lsm leak if mds_create_objects() fails (4801)
4249 - limit debug_daemon file size, always print CERROR messages (4789)
4250 - use transno after validating reply (3892)
4251 - process timed out requests if import state changes (3754)
4252 - update mtime on OST during writes, return in glimpse (4829)
4253 - add mkfsoptions to LDAP (4679)
4254 - use ->max_readahead method instead of zapping global ra (5039)
4255 - don't interrupt __l_wait_event() during strace
4257 - add software watchdogs to catch hung threads quickly (4941)
4258 - make lustrefs init script start after nfs is mounted
4259 - fix CWARN/ERROR duplication (4930)
4260 - return async write errors to application if possible (2248)
4261 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4262 - print NAL number in %x format (4645)
4263 - update barely-supported suse-2.4.21-171 series (4842)
4264 - support for sles 9 %post scripts
4265 - support for building 2.6 kernel-source packages
4266 - support for sles km_* packages
4268 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4271 - ignore -ENOENT errors in osc_destroy (3639)
4272 - notify osc create thread that OSC is being cleaned up (4600)
4273 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4274 - reconstruct ost_handle() like mds_handle() (4657)
4275 - create a new thread to do import eviction to avoid deadlock (3969)
4276 - let lconf resolve symlinked-to devices (4629)
4277 - don't unlink "objects" from directory with default EA (4554)
4278 - hold socknal file ref over connect in case target is down (4394)
4279 - allow more than 32000 subdirectories in a single directory (3244)
4280 - fix blocks count for O_DIRECT writes (3751)
4281 - OST returns ENOSPC from object create when no space left (4539)
4282 - don't send truncate RPC if file size isn't changing (4410)
4283 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4284 - bind to privileged port in socknal and tcpnal (3689)
4286 - rate limit CERROR/CWARN console message to avoid overload (4519)
4287 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4288 - basic mmap support (3918)
4289 - kernel patch series update from b1_4 (4711)
4291 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4294 - avoid crash during MDS cleanup with OST shut down (2775)
4295 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4296 - don't use bad inodes on the MDS (3744)
4297 - dynamic object preallocation to improve recovery speed (4236)
4298 - don't hold spinlock over lock dumping or change debug flags (4401)
4299 - don't zero obd_dev when it is force cleaned (3651)
4300 - print grants to console if they go negative (4431)
4301 - "lctl deactivate" will stop automatic recovery attempts (3406)
4302 - look for existing locks in ldlm_handle_enqueue() (3764)
4303 - don't resolve lock handle twice in recovery avoiding race (4401)
4304 - revalidate should check working dir is a directory (4134)
4306 - don't always mark "slow" obdfilter messages as errors (4418)
4308 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4311 - don't close LustreDB during write_conf until it is done (3860)
4312 - fix typo in lconf for_each_profile (3821)
4313 - allow dumping logs from multiple threads at one time (3820)
4314 - don't allow multiple threads in OSC recovery (3812)
4315 - fix debug_size parameters (3864)
4316 - fix mds_postrecov to initialize import for llog ctxt (3121)
4317 - replace config semaphore with spinlock (3306)
4318 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4319 - don't allow enqueue to complete on a destroyed export (3822)
4320 - down write_lock before checking llog header bitmap (3825)
4321 - recover from lock replay timeout (3764)
4322 - up llog sem before sending rpc (3652)
4323 - reduce ns lock hold times when setting kms (3267)
4324 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4325 - fix NULL deref and obd_dev leak on setup error (3312)
4326 - replace some LBUG about llog ops with error handling (3841)
4327 - don't match INVALID dentries from d_lookup and spin (3784)
4328 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4329 - fix invalid assertion in ptlrpc_set_wait (3880)
4331 - add libwrap support for the TCP acceptor (3996)
4332 - add /proc/sys/portals/routes for non-root route listing (3994)
4333 - allow setting MDS UUID in .xml (2580)
4334 - print the stack of a process that LBUGs (4228)
4336 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4339 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4340 - make sure to unset replay flag from failed open requests (3440)
4341 - if default stripe count is 0, use OST count for inode size (3636)
4342 - update parent mtime/ctime on client for create/unlink (2611)
4343 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4344 - free recovery state on server during a forced cleanup (3571)
4345 - unregister_reply for resent reqs (3063)
4346 - loop back devices mounting and status check on 2.6 (3563)
4347 - fix resource-creation race that can provoke i_size == 0 (3513)
4348 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4349 - more debugging for page-accounting assertion (3746)
4350 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4351 - avoid deadlock after precreation failure (3758)
4352 - fix race and lock order deadlock in orphan handling (3450, 3750)
4353 - add validity checks when grabbing inodes from l_ast_data (3599)
4355 - add /proc/.../recovery_status to obdfilter (3428)
4356 - lightweight CDEBUG infrastructure, debug daemon (3668)
4357 - change default OSC RPC parameters to be better on small clusters
4358 - turn off OST read cache for files smaller than 32MB
4359 - install man pages and include them in rpms (3100)
4360 - add new init script for (un)mounting lustre filesystems (2593)
4361 - run chkconfig in %post for init scripts (3701)
4362 - drop scimac NAL (unmaintained)
4364 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4367 - clean kiobufs before and after use (3485)
4368 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4369 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4370 - change init script to fail more clearly if not run as root (1528)
4371 - allow clients to reconnect during replay (1742)
4372 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4373 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4374 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4375 - fix cond_resched (3554)
4376 - only evict unfinished clients after recovery (3515)
4377 - allow bulk resend, prevent data loss (3570)
4378 - dynamic ptlrpc request buffer allocation (2102)
4379 - don't allow unlinking open directory if it isn't empty (2904)
4380 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4381 - remove extraneous obd dereference causing LASSERT failure (3334)
4382 - don't use get_cycles() when creating temp. files on the mds (3156)
4383 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4384 - handle EEXIST for set-stripe, set proper directory name (3336)
4386 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4387 - fix ksocknal_fmb_callback() error messages (2918)
4389 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4392 - don't copy lvb into (possibly NULL) reply on error (2983)
4393 - don't deref dentry after dput, don't free lvb on error (2922)
4394 - use the kms to determine writeback rpc length (2947)
4395 - increment oti_logcookies when osc is inactive (2948)
4396 - update client's i_blocks count via lvb messages (2543)
4397 - handle intent open/close of special files properly (1557)
4398 - mount MDS with errors=remount-ro, like obdfilter (2009)
4399 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4400 - don't use cancelling-locks' kms values (2947)
4401 - use highest lock extent for kms, not last one (2925)
4402 - don't dereference ERR_PTR() dentry in error handling path (3107)
4403 - fix thread race in portals_debug_dumplog() (3122)
4404 - create lprocfs device entries at setup instead of at attach (1519)
4405 - common AST error handler, don't evict client on completion race (3145)
4406 - zero nameidata in detach_mnt in 2.6 (3118)
4407 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4408 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4409 - initialize RPC timeout timer earlier for 2.6 (3219)
4410 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4411 - print nal/nid for unknown nid (3258)
4412 - additional checks for oscc recovery before doing precreate (3284)
4413 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4414 - don't crash in mdc_close for bad permissions on open (3285)
4415 - zero i_rdev for non-device files (3147)
4416 - clear page->private before handing to FS, better assertion (3119)
4417 - tune the read pipeline (3236)
4418 - fix incorrect decref of invalidated dentry (2350)
4419 - provide read-ahead stats and refine rpc in flight stats (3328)
4420 - don't hold journal transaction open across create RPC (3313)
4421 - update atime on MDS at close time (3265)
4422 - close LDAP connection when recovering to avoid server load (3315)
4423 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4424 - don't leak open file on MDS after open resend (3325)
4425 - serialize filter_precreate and filter_destroy_precreated (3329)
4426 - loop device shouldn't call sync_dev() for nul device (3092)
4427 - clear page cache after eviction (2766)
4428 - resynchronize MDS->OST in background (2824)
4429 - refuse to mount the same filesystem twice on same mountpoint (3394)
4430 - allow llmount to create routes for mounting behind routers (3320)
4431 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4432 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4433 - fix rare problem with rename on htree directories (3417)
4435 - allow default OST striping configuration per directory (1414)
4436 - fix compilation for qswnal for 2.6 kernels (3125)
4437 - increase maximum number of MDS request buffers for large systems
4438 - change liblustreapi to be useful for external progs like lfsck (3098)
4439 - increase local configuration timeout for slow disks (3353)
4440 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4442 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4445 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4446 - don't overwrite extent policy data in reply if lock was blocked (2901)
4447 - drop filter export grants atomically with removal from device (2663)
4448 - del obd_self_export from work_list in class_disconnect_exports (2908)
4449 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4450 - swab reply message in mdc_close, other PPC fixes (2464)
4451 - fix destroying of named logs (2325)
4452 - overwrite old logs when running lconf --write_conf (2264)
4453 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4454 - fix race in target_handle_connect (2898)
4455 - mds_reint_create() should take same inode create lock (2926)
4456 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4457 - don't close files for self_export to avoid uninitialized obd (2936)
4458 - allow MDS with the same name as client node (2939)
4459 - hold dentry reference for closed log files for unlink (2325)
4460 - reserve space for all logs during transactions (2059)
4461 - don't evict page beyond end of stripe extent (2925)
4462 - don't oops on a deleted current working directory (2399)
4463 - handle hard links to targets without a parent properly (2517)
4464 - don't dereference NULL lock when racing during eviction (2867)
4465 - don't grow lock extents when lots of conflicting locks (2919)
4467 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4470 - account for cache space usage on clients to avoid data loss (974)
4471 - lfsck support in lustre kernel code (2349)
4472 - reduce journal credits needed for BRW writes (2370)
4473 - orphan handling to avoid losing space on client/server crashes
4474 - ptlrpcd can be blocked, stopping ALL progress (2477)
4475 - use lock value blocks to assist in proper KMS, faster stat (1021)
4476 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4477 - recovery for initial connections (2355)
4478 - fixes for mds_cleanup_orphans (1934)
4479 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4480 - block all file creations until orphan recovery completes (1901)
4481 - client remove rq_connection from request struct (2423)
4482 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4483 - recovery timer race (2670)
4484 - mdc_close recovey bug (2532)
4485 - ptlrpc cleanup bug (2710)
4486 - mds timeout on local locks (2588)
4487 - namespace lock held during RPCs (2431)
4488 - handle interrupted sync write properly (2503)
4489 - don't try to handle a message that hasn't been replied to (2699)
4490 - client assert failure during cleanup after abort recovery (2701)
4491 - leak mdc device after failed mount (2712)
4492 - ptlrpc_check_set allows timedout requests to complete (2714)
4493 - wait for inflight reqs when ptlrpcd finishes (2710)
4494 - make sure unregistered services are removed from the srv_list
4495 - reset bulk XID's when resending them (caught by 1138 test)
4496 - unregister_bulk after timeout
4497 - fix lconf error (2694)
4498 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4499 - readahead locks pages, leaves pending causing memory pressure (2673)
4500 - increase OST request buffers to 4096 on large machines (2729)
4501 - fix up permission of existing directories in simple_mkdir (2661)
4502 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4503 - don't assign transno to errored transactions (2742)
4504 - don't delete objects on OST if given a bogus objid from MDS (2751)
4505 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4506 - if rq_no_resend, then timeout request after recovery (2432)
4507 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4508 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4509 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4510 - MDS recovery shouldn't delete objects that it has given out (2730)
4511 - if enqueue arrives after completion, don't clobber LVB (2819)
4512 - don't unlock pages twice when trigger_group_io returns error (2814)
4513 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4514 - don't write pages to disk if there was an error (1450)
4515 - don't ping imports that have recovery disabled (2676)
4516 - take buffered bytes into account when balancing socknal conn (2817)
4517 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4518 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4519 - remove little-endian swabbing of llog records (1987)
4520 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4521 - retry reposting request buffers if they fail (1191)
4522 - grow extent at grant time to avoid granting a revoked lock (2809)
4523 - lock revoke doesn't evict page if covered by a second lock (2765)
4524 - disable VM readahead to avoid reading outside lock extents (2805)
4526 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4527 - updated kernel patches for hp-2.4.20 kernel (2681)
4529 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4532 - fix truncated write corruption (2366)
4533 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4535 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4536 - ensure len(uuid) < 37 in lmc (1171)
4537 - fix ia64 OOPS in llog_test (2255)
4538 - zero end of page at obdfilter for partial page writes (2648)
4539 - don't leave stale dentries around after renames (bug 2428)
4540 - fix timeouts when evicting a client with a single lock held (2642)
4541 - set deadline for the initial HELLO message to drain (2634)
4542 - print out dotted-quad IP addresses in the socknal (2302)
4544 - additional debugging for MDS client eviction problem (2443)
4545 - fix mkfsoptions support for osts (2603, 2604)
4547 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4550 - add series for the vanilla 2.6.0 kernel
4551 - add series for the vanilla 2.4.24 kernel
4552 - add series for a cray x86/64 UL kernel drop
4553 - fix xattr patches for the vanilla 2.4.19 series
4555 - generate true UUIDs in lmc (1171)
4556 - have portals stack dumping break in UML (2466)
4557 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4558 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4559 - wake callback waiting threads on client eviction (2460)
4560 - Add --ptldebug and --subsystem to lmc (1719)
4561 - update assertion to allow safe interrupt allocation
4562 - set rq_no_resend for cancel requests (2432)
4563 - recalculate ptlrpcd timeout after resend (2494)
4564 - call vfs_rmdir when removing pending directories (2368)
4565 - fix renaming a file to itself (2429)
4566 - lmc creates a default one-stripe lov (2454)
4567 - expand procfs space to handle large clusters (2326)
4568 - increase UML stack to avoid overflow
4569 - update lconf's list of debug and subsystem masks
4570 - fix lfs find --obd (2510)
4571 - /proc tunable for disabling filter read caching (2591)
4572 - stop rpm packages from altering slapd.conf (2301)
4573 - disable nagle in the socknal under 0conf (2578)
4574 - choose mds inode size based on stripe count (2572)
4575 - fix kernel-source rpm problems (2516)
4577 - add --disable-doc to avoid pdf generation (2421)
4578 - update documentation, tests, type-os, comments
4579 - avoid format warnings on ia64
4580 - remove the TOE NAL
4581 - tiny code cleanups by removing unused fields
4583 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4586 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4587 - remove the most blatant lies from BUILDING file (2371)
4588 - change default debug level to reasonable production setting
4589 - reduce client side cache size to reduce cache flush time
4590 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4591 - make TCP zerocopy and pinger support enabled by default (2476)
4592 - sync writes completed after process exits caused crashes (2319)
4593 - maintain correct mount count on the MDS (2356)
4594 - backout 1557, because 2316 wasn't really fixed
4595 - better file I/O statistics gathering in /proc
4596 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4597 - another kernel patch to fix zero-copy TCP function export
4598 - don't take duplicate lock when processing re-sent getattr (2420)
4599 - lctl uses obd_self_export instead of creating new conn (2353)
4600 - MDS/OST recovery case which requires object creation asserted (2425)
4601 - move lfs from /usr/sbin to /usr/bin in packages
4602 - fix race between mds_client_add and mds_client_free (2417)
4603 - use kmalloc instead of slabs in portals (2430)
4604 - don't create duplicate records when a failover MDS is present (2442)
4605 - remove unnecessary mount age check (2332)
4606 - don't remove directory inodes from locks prematurely (2451)
4607 - don't break if MDS service name is the same as hostname (2103)
4608 - fix races in client write RPC generation when cache full (2482)
4610 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4613 - remove now-unused request->rq_obd (278)
4614 - if an allocation fails, print out how much memory we've used (1933)
4615 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4616 - add the "configurable stack size" patch to most series files (1256)
4617 - ability to write large log records, for 100+ OST configs (2306)
4618 - fix NULL deref when filter_prep fails (2314)
4619 - fix operator precedence error in filter_sync
4620 - dynamic allocation of socknal TX descriptors (2315)
4621 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4622 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4623 - ensure that CWARN messages in Portals always get to the syslog
4624 - __init/__exit are not for prototype decls (ldlm_init/exit)
4625 - x86-64 compile warning fixes
4626 - fix gateway LMC keyword conflict (2318)
4627 - fix MDS lock inversions in getattr/reint paths (1844)
4628 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4629 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4630 - DLM race condition prevented some lock evictions (2328)
4631 - ENOMEM detection and retry on socknal sends (2230)
4632 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4633 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4634 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4635 - better error messages when a client is rejected during recovery (1505)
4636 - avoid cancelling locks which were never granted, after failure (2330)
4637 - fix i_sem/journal inversion in mds_client_add (2333)
4638 - fix truncate/getattr lock cycle deadlock (2334)
4639 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4640 - fix two rare exit paths which could leak an l_lock() ref (2321)
4641 - fencepost error in MDS/OST orphan recovery (2226)
4642 - make log record alignment 8 bytes (1988)
4643 - lstripe now fails when requested offset > ost_count (2237)
4644 - ensure that all kernel series have a complete list.h (1607)
4645 - fix crashes in special-file operations (2316)
4646 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4648 - allow configurable automake binary, for testing new versions
4649 - small update to the lfs documentation
4651 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4653 * fix negative export reference count in fsfilt_sync (2312)
4655 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4656 * release candidate 0.9.1
4658 - orphans are moved into the PENDING directory for possible recovery
4659 - replayed opens now open by fid for orphan/rename safety (1042)
4660 - last close of an orphan inode generates a transno (683)
4661 - chdir() and mount() now pin the directory entry (1020)
4662 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4663 - discard very old requests without processing them (1502)
4664 - remove some common, well-understood CERRORs (1505)
4665 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4666 - clear "grant" flags in OST replies until OST grant code lands (1644)
4667 - fix read performance by not clobbering i_blksize on client (1598)
4668 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4669 - make LRU size a /proc tunable, clears locks when reduced (707)
4670 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4671 - server locks take references on exports now (1558)
4672 - build fixes for 2.4.20-rh trees (1663)
4673 - return an error from lov_create if all OSCs are inactive (1751)
4674 - fix import levels when a reconnect happens without a timeout (1597)
4675 - exit early from mds_open if we get a lookup error (1749)
4676 - partial page read at EOF wouldn't wait for disk before sending (1642)
4677 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4678 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4679 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4680 - write/truncate lock inversion (1639)
4681 - Don't auto-load obdclass, portals modules during cleanup (1495)
4682 - fix timestamps from jumping to "now" (1763)
4683 - extra journal assertions (1648)
4684 - add an extra multiunlink test (1771)
4685 - fix read_record/write_record API (1776)
4686 - fix leak of offset_extent, possible incorrect i_size later (1772)
4687 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4688 - Debugging for the kqswnal_get_idle_tx problems (1820)
4689 - Allow recovery to be attempted multiple times (1536)
4690 - Write out MDS last_rcvd file after it is first created (1600)
4691 - Fix tx_descriptor leak in failed transmit situations (1827)
4692 - ext3 journaling fixes for assertion failure after IO error (1871)
4693 - class_export_put() on freed export after completion AST error (1896)
4694 - Fix revalidate looping in VFS (1322)
4695 - Don't access a freed export during MDS_REINT timeout (1521)
4696 - Add open-unlink recovery support on the MDS (1673,1764)
4697 - Return an error if no MDS data was read from last_rcvd (1946)
4698 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4699 - Don't setup a disk device that doesn't match exported UUID (317)
4700 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4701 - avoid committing NULL handle in force close
4702 - local.sh is now a one-stripe LOV configuration
4703 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4704 - fix ext3 htree duplicate directory entry corruption (1516)
4705 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4706 - update to new LOV EA format (2097)
4707 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4709 2003-06-15 Phil Schwan <phil@clusterfs.com>
4712 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4713 - per-import/export recovery handling (958, 931, 959)
4714 - multiple last-rcvd slots, for serving multiple FSes (949)
4715 - connections are again shared between multiple imp/exports (963, 964)
4716 - "umount -f" would hang if any requests needed to be sent (393, 978)
4717 - avoid pinning large req buffer by copying for queued messages (989)
4718 - add "uuid" to "lctl device" command to help upcalls (991)
4719 - "open" RPCs with transnos would confuse recovery counters (1037)
4720 - do proper endian conversion of all wire messages (288, 340, 891)
4721 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4722 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4723 - fix ptlrpc_connection leak in target_handle_connect (1174)
4724 - fix import refcounting bug in OST and MDS cleanup (1134)
4725 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4726 - fix very unlikely obd_types race condition (501)
4727 - remove osc_open hack for echo_client (1187)
4728 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4729 - a failure in read_inode2 leads to deadlock (1139)
4730 - cancel ack-locks as soon as transaction is committed (1072)
4731 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4732 - make sure to commitrw after any preprw to avoid deadlock (1162)
4733 - failing to execute a file in a lustre FS would lock inode (1203)
4734 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4735 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4736 - obd_uuid2tgt would walk off the end of the list (1255)
4737 - on IA64 the osc would give portals incorrect bulk size (1258)
4738 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4739 - fix lock inversion caused by new llite matching code (1282)
4740 - limit the number of dirty pages on a client to 10MB (1286)
4741 - timed out locks were not being corrected cancelled (1289)
4742 - fix O_DIRECT above 4GB on IA-32 (1292)
4743 * major user-visible changes
4744 - fail out/fail over policy now controlled by the upcall (993)
4746 - add OBD_PING to check server availability and failure (954)
4747 - lustre messages are now sent in sending host order (288, 340, 891)
4748 - add eadatalen to MDS getattr reply (340)
4749 - OST read replies may contain second buffer, with per-page status (593)
4751 2003-03-11 Phil Schwan <phil@clusterfs.com>
4754 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4755 - failed open()s could cause deadlock; fixed (867, 869)
4756 - stop cancelling OST locks when files are closed (481)
4757 - overlapping XID spaces caused network corruption (851, 853)
4758 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4759 - setattr_raw added, to do single-RPC, server-side setattrs
4760 - lmc/lconf syntax change for OST UUIDs
4761 - fix crashy race condition between ptlrpc_free_req and osc_close
4762 - don't use request in mdc_enqueue if we hit a timeout (889)
4763 - don't set the inode i_size for regular files from the MDS (896)
4764 - handle out of order completion AST (842)
4765 - don't LBUG if a lock request times out after receiving AST (913)
4766 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4767 - if a bad lock AST arrives, send an error instead of dropping entirely
4768 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4769 - fix leak in bulk IO when only partially completed (899, 900, 926)
4770 - fix O_DIRECT for ia64 (55)
4771 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4772 - C-z after timeout could hang a process forever; fixed (977)
4774 - client-side I/O cache (678, 924, 929, 941, 970)
4776 - READPAGE and SETATTRs which don't take server-side locks get
4779 2003-02-11 Phil Schwan <phil@clusterfs.com>
4782 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4783 - Fix fsfilt_extN_readpage() to read a full page of directory
4784 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4785 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4786 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4787 - Most intent operations were removed, in favour of a new RPC mode
4788 that does a single RPC to the server and bypasses most of the VFS
4789 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4790 - Aggressively cancel local locks on DLM servers
4791 - mds_reint_unlink sends EA to the client if it's the last nlink.
4792 client uses that EA to unlink OST objects.
4793 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4794 - recursive symlinks were fixed (439)
4795 - fixed NULL deref in DEBUG_REQ
4796 - filter_update_lastobjid no longer calls sync, which annoyed extN
4797 - fixed multi-client small-writes to a single file problem (445)
4798 - fixed mtime updates during file writes (607)
4799 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4800 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4801 - fixed Linux OST size reporting problem (444, 656)
4802 - OST now updates object mtime with writes or setattr (607, 619)
4803 - client verifies file size before zeroing page past EOF (445)
4804 - OST now writes last allocated objid to disk with allocation (108)
4805 - LOV on echo now works (409)
4807 - mds_reint_unlink sends a new buffer, with the EA included. this
4808 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4809 set if a regular file was being unlinked, and it was the last link
4810 - use PtlGet from the target for bulk writes (315)
4811 - OST now updates object mtime with writes or setattr (607, 619)
4812 - LDLM now has a grant-time callback to revalidate locked items, if
4814 - Many MDS operations were reorganized to combat race conditions
4816 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4817 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4819 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4822 - Fully reactivate OST imports after reconnection (512, others)
4823 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4824 - More graceful error handling for truncating on dead OST (515)
4825 - Don't error out unless we're actually accessing dead stripes (474)
4826 - Fix garbage sizes when stripes are missing (410)
4827 - LRU counters were broken, causing constant lock purge (433, 432)
4828 - garbage on read from stripes with failed OSTs (441)
4829 - mark OSCs as active before reconnecting during recovery (438)
4830 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4831 - lfind did not preserve OST order in output (443)
4832 - symlinks cause hung clients, incorrect data (439)
4833 - stop dereferencing request after dropping refcount (457)
4834 - don't LASSERT(spin_is_locked) on non-SMP (455)
4835 - fixes for many rename() bugs
4836 - fstat didn't correctly synchronize attributes (399)
4837 - server must handle lock cancellation during blocking AST prep (487)
4838 - bulk descriptors were free()d too soon (511)
4839 - fix paths in lconf, which would load incorrect modules (451, 507)
4840 - fix confusing lconf 'host not found' error message (386)
4841 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4842 - fix race condition in mdc_blocking_ast() for inode access (526)
4843 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4844 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4845 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4846 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4847 - fix lstripe in several ways (kernel side) (527)
4848 - fix request leak in ldlm_cli_enqueue (262)
4849 - incorrect OSC was marked inactive after OST failure
4850 - call mds_fs_cleanup before unmounting filesystem (524)
4851 - fix races between taking ns_lock and ldlm_lock_change_resource
4852 - fix races updating LOV export open file list
4853 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4854 - fix recovery NULL deref in ldlm_cli_cancel_unused
4855 - fix some DLM races by using new hash table for lock handles (419)
4856 - permit the client to specify desired inodes, at replay
4857 - duplicate requests when we queue them for replay reintegration
4858 - fix last_rcvd offset calculation
4859 - sync after each recovered transaction, so we always make progress
4860 - never, not always, ERESTART requests without transnos
4861 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4863 - skip replay if the MDS says that the client is already connected
4864 - don't check for a recovery-enabled export to match lctl's UUID
4865 - don't INC_USE_COUNT for phantom exports
4866 - don't crash when cleaning up phantom exports (567)
4867 - don't double-finish or set replay data for errored mdc_open requests
4868 - abort requests when they time out, so we don't get old replies
4869 - send/receive replies for AST messages again
4870 - if the client says that it doesn't have the lock, cancel it on the
4872 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4873 mark it as destroyed, it will all work out when decref is called
4874 - fix module use counts (22, 581)
4876 - ASTs now expect a reply (server cancels lock on error reply)
4878 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4881 - fix many simultaneous client startup (392)
4882 - fix dentry->d_it clobbering
4883 - credentials weren't being shipped for readdir/getattr operations
4884 - remove invalid assertions triggered during some concurrent MD
4886 - proper Lustre versions added (336, 389)
4887 - fix memory leak for create error case (398)
4888 - fix LOV locking bug that would get cli/srv out of sync
4889 - fix echo client over LOV (409)
4890 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4891 - fix double-O_EXCL intent crash (424)
4892 - avoid sending multiple lock CANCELs (352)
4894 - MDS can do multi-client recovery (modulo bugs in new code)
4896 - many updates, edits, cleanups
4898 2002-11-18 Phil Schwan <phil@clusterfs.com>
4901 - fix null d_it dereference (346)
4902 - fix full OST/dbench hang (333)
4903 - fix permission problem with file removal (286)
4904 - fix removal of OSCs from LOV when they fail
4905 - fix NULL deref during bulk timeout (214)
4906 - fix problems related to multiple filesystems on one MDS (241)
4907 - fixed serious subtle metadata locking bugs
4908 - free locks on clients when inodes are removed due to memory
4910 - fix inode pointer in lock data (285)
4911 - partial support for multiple MDS on a single host (241)
4912 - data locks weren't cancelled at clear_inode time (290, 311)
4913 - intent locks could lead to unbounded lock growth (205)
4914 - added a maximum lock count, an LRU list, and a flusher
4915 - fix multiple rename (365)
4916 - properly abstracted the echo client
4917 - OSC locked 1 byte too many; fixed
4918 - rewrote brw callback code:
4919 - fixed recovery bugs related to LOVs (306)
4920 - fixed too-many-pages-in-one-write crash (191)
4921 - fixed (again) crash in sync_io_timeout (214)
4922 - probably fixed callback-related race (385)
4924 - Add capability to MDS protocol
4925 - LDLM cancellations and callbacks on different portals
4927 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4930 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4931 - defer open object destruction to close time (601981, 138)
4932 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4933 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4934 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4935 - don't call obd_setattr in ll_file_release for destroyed objects
4936 * protocol change to lustre_msg: move |version| and add |flags|
4937 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4938 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4939 * added replay of create, unlink, link and rename operations during
4940 MDS failover; recovery should be much more robust now
4941 * remove failed OSCs from LOVs (only lov_create uses this so far)
4942 * the lustre-HOWTO was brought (more) up to date (582544)
4944 2002-10-23 Phil Schwan <phil@clusterfs.com>
4947 - in-use dentries weren't being reused properly (617851)
4948 - prevent multiple LDLM setup (599178)
4949 - fix LOV size calculations for truncate (617853)
4950 - fix client handling of MDS intent errors (POSIX)
4951 - fix permission bug in lovstripe.c test (624321)
4952 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4953 - truncate past end of file could corrupt data
4954 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4955 - a race in recovery could return ETIMEDOUT to apps (623947)
4956 - building outside the source directory was fixed
4957 * the lustre-HOWTO was brought (more) up to date (582544)
4958 * major progress was made on recovery functionality
4960 2002-10-10 Phil Schwan <phil@clusterfs.com>
4963 - recovery deadlock fix
4964 - rm -rf causes LBUG fix (617817)
4965 - file open by multiple tasks fix (618962)
4966 - directory permissions bugs (602707 and 620007)
4967 - journal_stop fixed with locking (611313)
4968 - O_APPEND failures resolved (618273, perhaps 614459)
4969 - lconf PATH fix (619770)
4970 - IA64 build fix (621450)
4971 - RPC buffer sizes scale with amount of memory
4973 2002-10-01 Phil Schwan <phil@clusterfs.com>
4976 - locks would be cancelled without throwing away data pages,
4977 resulting in inconsistent data (605627)
4978 - inode attributes were not always being refreshed (605627, 612449)
4979 - lconf now continues to cleanup after lctl reports an error
4980 - MDS now enforces user permissions (602707)
4981 - lprocfs cleanup fixed, but not yet enabled (614157)
4982 - fixed infinite server hang, should a client not respond to an AST
4983 - avoid going into recovery if user calls readlink() with a buffer
4984 that's too small (613941)
4985 - AST RPCs no longer require replies (614867) -- this may be changed
4986 - don't crash server if client sends an IOV that's too big (611336)
4987 - fixed lock conversion deadlock (611892)
4988 - fixed the following of symlinks (614622)
4989 * recovery: the server can remove locks from a client that dies, other
4990 clients can make progress
4991 * more extN patch fixes
4992 * compile-time configurable ptlrpc buffer allocations
4994 - collaborative read cache document
4995 - Lustre Lite Performance CDR document-in-progress
4997 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5000 - fix typo in patch-2.4.18
5002 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5005 - clear ptlrpc request each time in handle_incoming_request()
5006 - unlink of files now destroys the object on the OST
5008 2002-09-19 Peter Braam <braam@clusterfs.com>
5010 * add hard link support
5011 * change obdfile creation method
5012 * kernel patch changed
5014 2002-09-19 Peter Braam <braam@clusterfs.com>
5017 - stack overflow bug in extN fixed
5019 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
5021 * documentation updates
5022 - add man pages for config tools
5023 - update tests/README to describe testing with new config tools
5024 - finish metadata API descriptions
5025 * bug fixes and cleanups
5026 - statfs workaround for 16TB limit
5027 - LOV stripe allocation improved, can stripe on subset of OSTs
5028 - LOV file size/IO offset was wrong for files > 4GB in size
5029 - object EA data was being dropped, caused files to be unreadable
5030 - memory overflow with non-LOV OST caused memory corruption
5031 - fixed regression tests to work with new config tools, obdfilter
5032 - fixed bug when directory size became larger than 1 block
5033 - fixed bug (for single client case) when PWD was deleted
5034 - invalidate local directory pages when doing intent-based ops
5035 - avoid LDLM oops when lock callback contained bad data
5037 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5039 * documentation updates
5040 * bug fixes and cleanups
5041 - configuration tools
5044 - 64-bit compile warnings
5045 - 64-bit internal statfs data
5047 * test_brw on persistent OST devices
5049 * lprocfs (disabled)
5051 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5053 * documentation updates
5054 * bug fixes and cleanups
5055 * configuration tools
5057 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5060 * many small fixes to 0_5_4
5061 * io/network handling
5062 * thinkos in MDS operations
5064 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5067 * crucial basic fixes to 0.5.3
5068 * IOR, Iozone work over Elan
5071 2002-08-07 Phil Schwan <phil@clusterfs.com>
5072 * version 0_5_3, our first alpha
5073 * we use the new Portals iovs
5074 * documentation updates
5075 * bug fixes and cleanups
5076 * small changes in the DLM wire protocol
5078 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5079 * version 0_5_1 with some initial stability,
5080 * locking on MD and file I/O.
5081 * documentation updates
5082 * several bug fixes since 0.5.0
5083 * small changes in wire protocol
5085 2002-07-18 Phil Schwan <phil@clusterfs.com>
5087 * delivered as Lustre Light Alpha
5088 * fixed a crash after handling invalid MDS requests
5089 * fixed directory pages for architectures with non-4k pages sizes
5091 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5092 * release version v0_4_4
5093 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5095 * Fixes a number of interruption problems with OST operations.
5096 * Update documentation for portals header changes
5097 * Move all wire protocol structs/defines to lustre_idl.h
5098 * Fixes symlink length bug.
5099 * Add tcpdump to repository.
5101 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5102 * release version v0_4_3
5103 * Fixes statfs for inodes on extN.
5104 * Fixes bug in runtests which would delete /etc/hosts.
5105 * Use 64-bit object IDs wherever possible (not into VFS though)
5106 Remove ost_get_info, which is unused by lustre, and out of date.
5108 2002-07-03 Peter Braam <braam@clusterfs.com>
5109 * release version v0_4_2 Fixes a lookup error (type not passed)
5110 * move forward to head of Portals
5111 * move forward to latest Lustre kernel
5113 2002-06-25 Peter Braam <braam@clusterfs.com>
5114 * release version v0_4_1. Hopefully stable on single node use.