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).
15 Severity : enhancement
17 Description: Generic /proc file permissions
18 Details : Set /Proc file permissions in a more generic way to enable non-
19 root users operate on some /proc files.
23 Description: Hitting mdc_commit_close() ASSERTION
24 Details : Properly handle request reference release in
25 ll_release_openhandle().
29 Frequency : only patchless client
30 Description: add workaround for race between add/remove dentry from hash
32 Severity : enhancement
34 Description: Allow OST glimpses to return PW locks
38 Description: LBUG when llog conf file is full
39 Details : When llog bitmap is full, ENOSPC should be returned for plain
44 Description: Prevent import from entering FULL state when server in recovery
48 Description: service mount cannot take device name with ":"
49 Details : Only when device name contains ":/" will mount treat it as
55 Description: replace ptlrpcd with the statahead thread to interpret the async
56 statahead RPC callback
60 Frequency : on recovery
61 Description: I/O failures after umount during fail back
62 Details : if client reconnected to restarted server we need join to recovery
63 instead of find server handler is changed and process self eviction
64 with cancel all locks.
66 Severity : enhancement
68 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
70 Severity : enhancement
72 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
74 Severity : enhancement
76 Description: Upcall on Lustre log has been dumped
77 Details : Allow for a user mode script to be called once a Lustre log has
78 been dumped. It passes the filename of the dumped log to the
79 script, the location of the script can be specified via
80 /proc/sys/lnet/debug_log_upcall.
85 Description: avoid idr_remove called for id which is not allocated.
86 Details : Move assigment s_dev for clustered nfs to end of initialization,
87 to avoid problem with error handling.
92 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
93 Details : When connection is reused this not moved from CONN_UNUSED_HASH
94 into CONN_USED_HASH and this prodice warning when put connection
97 Severity : enhancement
99 Description: File striping can now be set to use an arbitrary pool of OSTs.
101 Severity : enhancement
103 Description: Export bytes_read/bytes_write count on OSC/OST.
107 Description: Early reply size mismatch, MGC loses connection
108 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
109 the connect flags are properly negotiated.
113 Description: Properly propagate oinfo flags from lov to osc for statfs
114 Details : restore missing copy oi_flags to lov requests.
116 Severity : enhancement
118 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
119 : l_getgroups(8), lst(8), routerstat(8)
120 Details : included man pages for llobdstat(8), llstat(8),
121 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
123 Severity : enhancement
125 Description: configure's --enable-quota should check the
126 : kernel .config for CONFIG_QUOTA
127 Details : configure is terminated if --enable-quota is passed but
128 : no quota support is in kernel
132 Description: Remove portals compatibility
133 Details : Remove portals compatibility, not interoperable with releases
138 Description: Resolve device initialization race
139 Details : Prevent proc handler from accessing devices added to the
140 obd_devs array but yet be intialized.
142 Severity : enhancement
144 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
146 Severity : enhancement
148 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
153 Description: Using wrong pointer in osc_brw_prep_request
154 Details : Access to array[-1] can produce panic if kernel compiled with
155 CONFIG_PAGE_ALLOC enabled
159 Description: Client runs out of low memory
160 Details : Consider only lowmem when counting initial number of llap pages
164 Description: *optional* service tags registration
165 Details : if the "service tags" package is installed on a Lustre node
166 When the filesystem is mounted, a local-node service tag will
167 be created. See http://inventory.sun.com/ for more information
168 about the Service Tags asset management system.
172 Description: Kernel BUG tries to release flock
173 Details : Lustre does not destroy flock lock before last reference goes
174 away. So always drop flock locks when client is evicted and
175 perform unlock regardless of successfulness of speaking to MDS.
179 Description: add recount protection for osc callbacks, so avoid panic on shutdown
183 Description: sanity test 65a fails if stripecount of -1 is set
184 Details : handle -1 striping on filesystem in ll_dirstripe_verify
189 Description: ASSERTION(CheckWriteback(page,cmd)) failed
190 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
193 Severity : enhancement
195 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
199 Description: do not process already freed flock
200 Details : flock can possibly be freed by another thread before it reaches
201 to ldlm_flock_completion_ast.
205 Description: LBUG during stress test
206 Details : Need properly lock accesses the flock deadlock detection list.
210 Description: oops in page fault handler
211 Details : kernel page fault handler can return two special 'pages' in error case, don't
212 try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
216 Description: timeout with invalidate import.
217 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which should be
218 handled by ptlrpcd. This produce long age waiting and -ETIMEOUT
219 ptlrpc_invalidate_import and as result LASSERT.
221 Severity : enhancement
223 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
227 Description: enable MGS and MDT services start separately
228 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
229 a co-located MGS without starting the MGS, which is a complement
230 to 'nosvc' mount option.
234 Frequency : after recovery
235 Description: precreate to many object's after del orphan.
236 Details : del orphan st in oscc last_id == next_id and this triger growing
237 count of precreated objects. Set flag LOW to skip increase count
238 of precreated objects.
242 Frequency : rare, on clear nid stats
243 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
244 Details : when clean nid stats sometimes try destroy live entry,
245 and this produce panic in free.
249 Description: Stack overflow during MDS log replay
250 ease stack pressure by using a thread dealing llog_process.
254 Description: wait until IO finished before start new when do lock cancel.
255 Details : VM protocol want old IO finished before start new, in this case
256 need wait until PG_writeback is cleared until check dirty flag and
257 call writepages in lock cancel callback.
259 Severity : enhancement
261 Description: using special macro for print time and cleanup in includes.
265 Description: mds_mfd_close() ASSERTION(rc == 0)
266 Details : In mds_mfd_close(), we need protect inode's writecount change
267 within its orphan write semaphore to prevent possible races.
271 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
275 Frequency : rare, on shutdown ost
276 Description: don't hit live lock with umount ost.
277 Details : shrink_dcache_parent can be in long loop with destroy dentries,
278 use shrink_dcache_sb instead.
282 Description: don't panic with use echo client
283 Details : echo client pass NULL as client nid pointer and this produce null
288 Description: fix build on ppc32
289 Details : compile code with -m64 flag produce wrong object file for ppc32.
293 Description: add message levels for liblustreapi
297 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
298 Details : Move the 'good_osts' check before the 'total_bavail' check. This
299 will result in an -EAGAIN and in the exit call path we call
300 alloc_rr() which will with increasing aggressiveness attempt to
301 aquire precreated objects on the minimum number of required OSCs.
305 Description: Use old size assignment to avoid deadlock
306 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
307 the scheduling while holding a spinlock. We do not need locking
308 for size in ll_update_inode() because size is only updated from
309 the MDS for directories or files without objects, so there is no
310 other place to do the update, and concurrent access to such inodes
311 are protected by the inode lock.
315 Description: resolve "_IOWR redefined" build error on SLES10
319 Description: dump the memory debugging after all modules are unloaded to
320 suppress false negative in conf_sanity test 39
322 Severity : enhancement
324 Description: build kernel-ib packages for OFED 1.3 in our release cycle
329 Description: fix SLES kernel versioning
330 Details : the kernel version for our SLES 10 kernel did not include a "-"
331 before the "smp" at the end. while this was not a problem in
332 general, it did mean that software trying to use the kernel
333 version to try to detect a vendor specific kernel would fail.
334 this was most evident by the OFED build scripts.
338 Description: Don't update lov_desc members until making sure they are valid
339 Details : When updating lov_desc members via proc fs, need fix their
340 validities before doing the real update.
344 Description: don't put request into delay list while invalidate in flight.
345 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
346 import in flight. this produce timeout for invalidate and sometimes
347 can cause stale data.
351 Frequency : on ppc only
352 Description: not convert ost objects for directory because it's not exist.
353 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
357 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
359 Severity : enhancement
361 Description: Update to support 2.6.22.14 vanilla kernel.
365 Frequency : rare, on recovery
366 Description: read procfs can produce deadlock in some situation
367 Details : Holding lprocfs lock which send rpc can produce block for destroy
368 obd objects and this also block reconnect with -EALREADY. This isn't
369 fix all lprocfs bugs - but make it rare.
371 Severity : enhancement
373 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
376 Frequency : frequent on X2 node
378 Description: mdc_set_open_replay_data LBUG
379 Details : Set replay data for requests that are eligible for replay.
383 Description: lustre_mgs: operation 101 on unconnected MGS
384 Details : When MGC is disconnected from MGS long enough, MGS will evict the
385 MGC, and late on MGC cannot successfully connect to MGS and a lot
386 of the error messages complaining that MGS is not connected.
390 Frequency : on network error
391 Description: panic with double free request if network error
392 Details : mdc_finish_enqueue is finish request if any network error ocuring,
393 but it's true only for synchronus enqueue, for async enqueue
394 (via ptlrpcd) this incorrect and ptlrpcd want finish request
397 Severity : enhancement
399 Description: client-side metadata stat-ahead during readdir(directory readahead)
400 Details : perform client-side metadata stat-ahead when the client detects
401 readdir and sequential stat of dir entries therein
404 Frequency : on start mds
406 Description: Implement get_info(last_id) in obdfilter.
409 Frequency : occasional
411 Description: Correctly check stale fid, not start epoch if ost not support SOM
412 Details : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
413 overwrited with new generated fid, but mds can anwer with one of these
414 two fids and both is not stale. setattr incorectly start epoch and
415 assume will be called done_writeting, but without SOM done_writing
419 Frequency : rare, depends on device drivers and load
421 Description: MDS or OSS nodes crash due to stack overflow
422 Details : Code changes in 1.8.0 increased the stack usage of some functions.
423 In some cases, in conjunction with device drivers that use a lot
424 of stack the MDS (or possibly OSS) service threads could overflow
425 the stack. One change which was identified to consume additional
426 stack has been reworked to avoid the extra stack usage.
429 Frequency : occasional
431 Description: Do not fail import if osc_interpret_create gets -EAGAIN
432 Details : If osc_interpret_create got -EAGAIN it immediately exits and
433 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
434 oscc_has_objects and see OSC has no objests and call
435 oscc_internal_create to resend create request.
437 Severity : enhancement
439 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
441 Severity : enhancement
443 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
446 Frequency : very rare
448 Description: Fix replay if there is an un-replied request and open
449 Details : In some cases, older replay request will revert the
450 mcd->mcd_last_xid on MDS which is used to record the client's
453 Severity : enhancement
455 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
457 Severity : enhancement
459 Description: Add rhel5 support to HEAD.
461 Serverity : enhancement
463 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
468 Description: Don't allow skipping OSTs if index has been specified.
469 Details : Don't allow skipping OSTs if index has been specified, make locking
470 in internal create lots better.
474 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
475 Details : access to bitfield in structure is always rounded to long
476 and this produce problem with not atomic change any bit.
480 Description: Lustre make rpms failed.
481 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
482 builds Lustre rpms from tarball.
485 Frequency : rare on shutdown ost
487 Description: If llog cancel was not send before clean_exports phase, this can
488 produce deadlock in llog code.
489 Details : If llog thread has last reference to obd and call class_import_put
490 this produce deadlock because llog_cleanup_commit_master wait when
491 last llog_commit_thread exited, but this never success because was
492 called from llog_commit_thread.
496 Description: allow userland application know is lost one of stripes.
497 Details : fill lvb_blocks with error code on ost and return it to
498 application if error flag found.
502 Description: NULL lov_tgts causing MDS oops
503 Details : more safe checks for NULL lov_tgts for avoid oops.
505 Severity : enhancement
507 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
511 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
512 Details : If an OST is down the MDS will hang indefinitely in
513 obd_statfs_rqset() waiting for the statfs data. While for
514 MDS QOS usage of statfs, it should not stuck in waiting.
516 Severity : enhancement
518 Description: remote_acl support
519 Details : Support ACL-based permission check for remote user.
520 Support setfacl/getfacl for remote user with the utils
521 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
522 the system "{s,g}etfacl" utils.
524 Severity : enhancement
526 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
528 Severity : enhancement
530 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
534 Description: make mgs_setparam() handle fsname containing dash
535 Details : fsname containing a dash does not work with lctl conf_param
537 Severity : enhancement
539 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
541 Severity : enhancement
543 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
547 Frequency : rare, at shutdown
548 Description: access already free / zero obd_namespace.
549 Details : if client_disconnect_export was called without force flag set,
550 and exist connect request in flight, this can produce access to
551 NULL pointer (or already free pointer) when connect_interpret
552 store ocd flags in obd_namespace.
556 Frequency : only at startup
557 Description: not alloc memory with spinlock held.
558 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
559 if any spinlock held.
561 Severity : enhancement
563 Description: make lustre randomly fail allocating memory
564 Details : Make lustre randomly failed allocating memory for testing purpose.
566 Severity : enhancement
568 Description: lost problems with lov objid file
569 Details : Fixes some scability and access to not inited memory problems
570 in work with lov objdid file.
575 Description: lfs find does not continue on file error
576 Details : Continue other files processing when a file/dir is absent.
580 Description: Inconsistent usage of lustre_pack_reply()
581 Details : Standardize the usage of lustre_pack_reply() such that it
582 always generate a CERROR on failure.
585 Frequency : occasional
587 Description: MDS hang or stay in waiting lock
588 Details : If client receive lock with CBPENDING flag ldlm need send lock
589 cancel as separate rpc, to avoid situation when cancel request
590 can't processed due all i/o threads stay in wait lock.
594 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
598 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
602 Description: Update to SLES9 kernel-2.6.5-7.287.3.
606 Description: Update to RHEL4 latest kernel.
608 Severity : enhancement
610 Description: Build SLES10 patchless client fails
611 Details : The configure was broken by run ./configure with
612 --with-linux-obj=.... argument for patchless client. When the
613 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
614 can't search header adequately. Use absolute path such as
615 -I($LINUX)/include instead.
619 Description: interrupt oig_wait produce painc on resend.
620 Details : brw_redo_request can be used for resend requests from ptlrpcd and
621 private set, and this produce situation when rq_ptlrpcd_data not
622 copyed to new allocated request and triggered LBUG on assert
623 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
624 ptlrpcd set if request is changed and can be safety changed to use
629 Description: LASSERT_{REQ,REP}SWAB macros are buggy
630 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
631 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
632 these macros and replace them with their difinitions instead.
636 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
637 Details : Update which_patch & target file for SLES10 latest kernel.
640 Description: Remove client patches from SLES 10 kernel.
641 Details : This causes SLES 10 clients to behave as patchless clients
642 even on a Lustre-patched (server) kernel.
644 Severity : enhancement
646 Description: self-adjustable client's lru lists
647 Details : use adaptive algorithm for managing client cached locks lru
648 lists according to current server load, other client's work
649 pattern, memory activities, etc. Both, server and client
650 side namespaces provide number of proc tunables for controlling
653 Severity : enhancement
655 Description: light-weight GSS support
656 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
657 and reduce performance overhead.
659 Severity : enhancement
661 Description: Linux keyring support
662 Details : Support using service of Linux keyring for Lustre GSS internal
663 context refresh/cache mechanism.
667 Description: Fix errors in lfs documentation
668 Details : Fixes man pages
672 Description: don't use GFP_* in generic Lustre code.
673 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
677 Description: obdclass is limited by single OBD_ALLOC(idarray)
678 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
682 Description: warnings with build patchless client with vanila 2.6.19 and up
683 Details : change old ctl_table style and replace ctl_table/ctl_table_header
684 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
688 Frequency : Only for SLES
689 Description: sanity_quota fail test_1
690 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
691 file, which makes TSTID[2] inunique.
695 Frequency : Only for SLES9
696 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
697 Details : Update target/ChangeLog/which_patch .
701 Frequency : Only for test_10 in sanity-quota.sh
702 Description: a bug in quota test code
703 Details : checking a test flag in a wrong place
707 Frequency : Only for test_13 in sanity-quota.sh
708 Description: a bug in quota test script
709 Details : shouldn't check overfull allocated quota
713 Description: jbd statistics
714 Details : Port older jbd statistics patch for sles10
716 Severity : enhancement
718 Description: Add printing inode info into message about error in writepage.
723 Description: reply_lock_interpret crash due to race with it and lock cancel.
724 Details : Do not replay locks that are being cancelled. Do not reference
725 locks by their address during replay, just by their handle.
729 Frequency : When flocks are used.
730 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
731 Details : Flock locks might destroy just granted lock if it could be merged
732 with another existing flock, this is done in completion handler,
733 so teach ldlm_cli_enquque_fini that this is a valid case for
739 Description: Oops in read and write path when failing to allocate lock.
740 Details : Check if lock allocation failed and return error back.
744 Description: lstripe command fails for valid OST index
745 Details : The stripe offset is compared to lov->desc.ld_tgt_count
746 instead of lov->desc.ld_active_tgt_count.
750 Description: sanity.sh failed test 103
751 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
756 Description: df doesn't work properly if diskfs blocksize != 4K
757 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
761 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
762 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
764 Severity : enhancement
766 Description: Adaptive timeouts
767 Details : RPC timeouts adapt to changing server load and network
768 conditions to reduce resend attempts and improve recovery time.
772 Description: llapi_file_create() does not allow some changes
773 Details : add llapi_file_open() that allows specifying the mode and
774 open flags, and also returns an open file handle.
778 Description: merge and cleanup kernel patches.
779 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
783 Description: Add journal checksum support.(Kernel part)
784 Details : The journal checksum feature adds two new flags i.e
785 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
786 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
787 indicates that the commit block contains the checksum for
788 the blocks described by the descriptor blocks. Now commit
789 record can be sent to disk without waiting for descriptor
790 blocks to be written to disk. This behavior is controlled
791 using JBD2_FEATURE_ASYNC_COMMIT flag.
795 Description: OSS needs mutliple precreate threads
796 Details : Add ability to start more than one create thread per OSS.
800 Description: Sanity.sh test_65k failed: setstripe should have succeeded
801 Details : execute lfs setstripe on client
805 Description: mds_obd_create error creating tmp object
806 Details : When the user sets quota on root, llog will be affected and can't
807 create files and write files.
810 Frequency : Always on ia64 patchless client, and possibly others.
812 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
813 Details : This allows the patchless client to be loaded on architectures
818 Description: RedHat Update kernel for RHEL5
819 Details : Add the kernel config file for RHEL5.
823 Description: RedHat Update kernel for RHEL5
824 Details : Modify the kernel config file more closer RHEL5.
828 Description: Build failure against Centos5 (RHEL5)
829 Details : Define PAGE_SIZE when it isn't present.
833 Description: test_7 on sanity-quota.sh can't be used on separate machines
834 Details : add facet to handle it
838 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
839 Details : with NFS, the anon dentry's parent was set to itself in
840 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
841 corresponding dentry other than use rec->ur_name.
843 Severity : enhancement
845 Description: lfs setstripe enhancement
846 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
850 Description: enable data checksumming by default
851 Details : enable checksum by default, allow --disable-checksum
852 configure option and "-o nochecksum" mount option. Checksums
853 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
857 Description: lustre support for RHEL5
858 Details : Add support for RHEL5.
862 Description: Client eviction due to failover config
863 Details : after a connection loss, the lustre client should attempt to
864 reconnect to the last active server first before trying the
865 other potential connections.
869 Description: when mds and osts use different quota unit(32bit and 64bit),
870 quota will be released repeatly.
871 Details : void sending multiple quota reqs to mds, which will keep the
872 status between the reqs.
876 Description: osts not allocated evenly to files
877 Details : change the condition to increase offset_idx
880 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
882 Description: Data corruption for OSTs that are formatted larger than 2TB
884 Details : When generating the bio request for lustre file writes the
885 sector number would overflow a temporary variable before being
886 used for the IO. The data reads correctly from Lustre (which
887 will overflow in a similar manner) but other file data or
888 filesystem metadata may be corrupted in some cases.
892 Description: Tune the kernel for good SCSI performance.
893 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
894 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
899 Description: rewrite ext2-derived code in obdclass/uuid.c
900 Details : rewrite inherited code (uuid parsing code from ext2 utils)
901 from scratch preserving functionality.
905 Description: "lfs find -obd UUID" prints directories
906 Details : "lfs find -obd UUID" will return all directory names instead
907 of just file names. It is incorrect because the directories
908 do not reside on the OSTs.
912 Description: use i_size_read and i_size_write in 2.6 port
913 Details : replace inode->i_size access with i_size_read/write()
916 Frequency : when removing large files
918 Description: scheduling issue during removal of large Lustre files
919 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
920 It causes scheduling issues when removing large files (17TB in the
924 Frequency : only with liblustre clients on XT3
926 Description: evictions taking too long
927 Details : allow llrd to evict clients directly on OSTs
930 Frequency : only on ppc
932 Description: /proc/fs/lustre/devices broken on ppc
933 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
934 We should make sure the type of 'index' is loff_t and then cast
935 explicitly as needed below. Do not assign an explicitly cast
939 Frequency : only for rhel5
941 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
942 Details : Modify the target file & which_kernel.
944 Severity : enhancement
946 Description: omit set fsid for export NFS
947 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
948 as FS_REQUIRES_DEV for avoid problems with generate fsid.
952 Frequency : Always, for kernels after 2.6.16
953 Description: Fix warning idr_remove called for id=.. which is not allocated.
954 Details : Last kernels save old s_dev before kill super and not allow
955 to restore from callback - restore it before call kill_anon_super.
959 Description: buffer overruns could theoretically occur
960 Details : llapi_semantic_traverse() modifies the "path" argument by
961 appending values to the end of the origin string, and a
962 overrun may occur. Adding buffer overrun check in liblustreapi.
966 Description: Fix error on 'ls .' at the top of the Lustre mount.
967 Details : Don't revalidate dentry if it is a root dentry.
971 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
972 Details : Modify vm-tunables-rhel4.patch.
976 Description: Kernel config for 2.6.18-vanilla.
977 Details : Modify targets/2.6-vanilla.target.in.
978 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
979 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
980 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
981 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
985 Description: improve handling recoverable errors
986 Details : if request processig with error which can be recoverable on server
987 request should be resend, otherwise page released from cache and
992 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
993 Details : Modify target file & which_patch.
994 A flaw was found in the IA32 system call emulation provided
995 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
996 value could be stored in the %RAX register, which could trigger an
997 out-of-bounds system call table access. An untrusted local user
998 could exploit this flaw to run code in the kernel
999 (ie a root privilege escalation). (CVE-2007-4573).
1003 Description: O_DIRECT bypasses client statistics.
1004 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1005 read_ahead_stats were updated. copied the stats section from
1006 osc_send_oap_rpc() into async_internal().
1010 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1014 Description: change order of libsysio includes
1015 Details : '#include sysio.h' should always come before '#include xtio.h'
1019 Description: handle "serious error: objid * already exists" more gracefully
1020 Details : If LAST_ID value on disk is smaller than the objects existing in
1021 the O/0/d* directories, it indicates disk corruption and causes an
1022 LBUG(). If the object is 0-length, then we should use the existing
1023 object. This will help to avoid a full fsck in most cases.
1025 Severity : enhancement
1027 Description: adapt the lustre_config script to support the upgrade case
1028 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1029 server targets to 1.6 in parallel.
1033 Description: To avoid grant space > avaible space when the disk is almost
1034 full. Without this patch you might see the error "grant XXXX >
1035 available" or some LBUG about grant, when the disk is almost
1037 Details : In filter_check_grant, for non_grant cache write, we should
1038 check the left space by if (*left > ungranted + bytes), instead
1039 of (*left > ungranted), because only we are sure the left space
1040 is enough for another "bytes", then the ungrant space should be
1041 increase. In client, we should update cl_avail_grant only there
1042 is OBD_MD_FLGRANT in the reply.
1046 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1047 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1048 described in CVE-2007-4573. This problem could allow untrusted
1049 local users to gain root access.
1052 Frequency : when using O_DIRECT and quotas
1054 Description: Incorrect file ownership on O_DIRECT output files
1055 Details : block usage reported by 'lfs quota' does not take into account
1056 files that have been written with O_DIRECT.
1061 Description: touch file failed when fs is not full
1062 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1063 otherwise we can get ENOSP while fs is not full.
1067 Description: parallel lock callbacks
1068 Details : Instead of sending blocking and completion callbacks as separated
1069 requests, adding them to a set and sending in parallel.
1072 Frequency : only for Cray XT3
1073 Bugzilla : 12829/13455
1074 Description: Changing primary group doesn't change the group lustre assigns to
1076 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1077 overridden with the primary group provided by the group upcall,
1078 whereas the supplied fsgid can be trusted if it is in the list of
1079 supplementary groups returned by the group upcall.
1081 Severity : enhancement
1083 Description: Allow masking D_WARNING, D_ERROR messages from console
1084 Details : Console messages can now be disabled via lnet.printk.
1088 Description: User code with malformed file open parameter crashes client node
1089 Details : Before packing join_file req, all the related reference should be
1090 checked carefully in case some malformed flags cause fake
1091 join_file req on client.
1095 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1096 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1097 mode is equal to the granted mode. However, it should also check
1098 that LDLM_FL_CP_REQD is not set.
1102 Description: Heavy nfs access might result in deadlocks
1103 Details : After ELC code landed, it is now improper to enqueue any mds
1104 locks under och_sem, because enqueue might want to decide to
1105 cancel open locks for same inode we are holding och_sem for.
1109 Description: Client eviction while running blogbench
1110 Details : A lot of unlink operations with concurrent I/O can lead to a
1111 deadlock causing evictions. To address the problem, the number of
1112 oustanding OST_DESTROY requests is now throttled to
1113 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1114 callbacks are processed in priority.
1118 Description: enable ACLs on MDS by default
1119 Details : ACLs must be enabled on MDS by default.
1122 Frequency : PPC/PPC64 only
1124 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1125 Details : fix duplicated definitions between asm-ppc64/types.h and
1126 lustre_types.h on PPC.
1129 Frequency : PPC/PPC64 only
1131 Description: asm-ppc/segment.h does not exist
1132 Details : fix compile issue on PPC.
1136 Description: better handle error messages in extents code
1139 Frequency : RHEL4 only
1141 Description: mkfs is very slow on IA64/RHEL4
1142 Details : A performance regression has been discovered in the MPT Fusion
1143 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1144 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1145 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1147 Severity : enhancement
1149 Description: SNMP support enhancement
1150 Details : Adding total number of sampled request for an MDS node in snmp
1153 Severity : enhancement
1155 Description: Optimize ldlm waiting list processing for PR extent locks
1156 Details : When processing waiting list for read extent lock and meeting read
1157 lock that is same or wider to it that is not contended, skip
1158 processing rest of the list and immediatelly return current
1159 status of conflictness, since we are guaranteed there are no
1160 conflicting locks in the rest of the list.
1164 Description: Time out and refuse to reconnect
1165 Details : When the failover node is the primary node, it is possible
1166 to have two identical connections in imp_conn_list. We must
1167 compare not conn's pointers but NIDs, otherwise we can defeat
1168 connection throttling.
1172 Description: port llog fixes from b1_6 into HEAD
1173 Details : Port llog reference couting and some llog cleanups from b1_6
1174 (bug 10800) into HEAD, for protect from panic and access to already
1175 free llog structures.
1179 Description: Detect stride IO mode in read-ahead
1180 Details : When a client does stride read, read-ahead should detect that and
1181 read-ahead pages according to the detected stride pattern.
1185 Description: data checksumming impacts single node performance
1186 Details : add support for several checksum algorithm. Currently, only CRC32
1187 and Adler-32 are supported. The checksum type can be changed on
1188 the fly via /proc/fs/lustre/osc/*/checksum_type.
1192 Description: use adler32 for page checksums
1193 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1198 Description: build for x2 fails
1199 Details : fix compile issue on Cray systems.
1203 Description: Properly match for duplicate locks
1204 Details : Due to different lock order from skiplists code, we need to
1205 traverse entire list for now
1208 Frequency : only on PPC/SLES10
1210 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1211 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1212 be defined. Add a hack in mkfs_lustre.c to work around this compile
1217 Description: LASSERT on MDS when client holding flock lock dies
1218 Details : ldlm pool logic depends on number of granted locks equal to
1219 number of released locks which is not true for flock locks, so
1220 just exclude such locks from consideration.
1224 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1225 Details : Use fsfilt_sync() for both the whole filesystem sync and
1226 individual file sync to eliminate dangerous inode locking
1227 with I_LOCK that can lead to a deadlock.
1231 Description: performance in 1.6.3
1232 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1233 sure that 1MB requests are not fragmented by the block layer.
1235 Severity : enhancement
1237 Description: organize the server-side client stats on per-nid basis
1238 Details : Change the structure of stats under obdfilter and mds to
1246 The "uuid"s file would list the uuids of _active_ exports.
1247 And the clear entry is to clear all stats and stale nids.
1249 Severity : enhancement
1251 Description: eliminate client locks in face of contention
1252 Details : file contention detection and lockless i/o implementation
1253 for contended files.
1257 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1260 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1262 Description: MDS may lose file striping (and hence file data) in some cases
1263 Details : If there are additional extended attributes stored on the MDS,
1264 in particular ACLs, SELinux, or user attributes (if user_xattr
1265 is specified for the client mount options) then there is a risk
1266 of attribute loss. Additionally, the Lustre file striping
1267 needs to be larger than default (e.g. striped over all OSTs),
1268 and an additional attribute must be stored initially in the
1269 inode and then increase in size enough to be moved to the
1270 external attribute block (e.g. ACL growing in size) for file
1275 Description: skiplist implementation simplification
1276 Details : skiplists are used to group compatible locks on granted list
1277 that was implemented as tracking first and last lock of each
1278 lock group the patch changes that to using doubly linked lists
1282 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1283 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1284 because LASSERT on both the data supplied by a client, and the
1285 data on disk is dangerous and incorrect.
1287 Severity : enhancement
1289 Description: Slow truncate/writes to huge files at high offsets.
1290 Details : Directly associate cached pages to lock that protect those pages,
1291 this allows us to quickly find what pages to write and remove
1292 once lock callback is received.
1296 Description: more ldlm soft lockups
1297 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1298 starve other threads from the resource lock for a long time in
1299 case of long waiting queue, so change the debug level from
1300 D_OTHER to the less frequently used D_INFO.
1302 Severity : enhancement
1304 Description: add -gid, -group, -uid, -user options to lfs find
1308 Description: Hung threads in invalidate_inode_pages2_range
1309 Details : The direct IO path doesn't call check_rpcs to submit a new RPC once
1310 one is completed. As a result, some RPCs are stuck in the queue
1315 Description: filter threads hungs on waiting journal commit
1316 Details : Cleanup filter group llog code, then only filter group llog will
1317 be only created in the MDS/OST syncing process.
1321 Description: Procfs and llog threads access destoryed import sometimes.
1322 Details : Sync the import destoryed process with procfs and llog threads by
1323 the import refcount and semaphore.
1325 Severity : enhancement
1327 Description: openlock cache of b1_6 port to HEAD
1332 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1333 Details : If insertion of an extent fails, then discard the inode
1334 preallocation and free data blocks else it can lead to duplicate
1339 Description: don't always update ctime in ext3_xattr_set_handle()
1340 Details : Current xattr code updates the inode ctime in ext3_xattr_set_handle.
1341 In some cases the ctime should not be updated, for example for
1342 2.0->1.8 compatibility it is necessary to delete an xattr and it
1343 should not update the ctime.
1347 Bugzilla : 15713/16362
1348 Description: Assertion in iopen_connect_dentry in 1.6.3
1349 Details : looking up an inode via iopen with the wrong generation number can
1350 populate the dcache with a disconneced dentry while the inode
1351 number is in the process of being reallocated. This causes an
1352 assertion failure in iopen since the inode's dentry list contains
1353 both a connected and disconnected dentry.
1357 Description: assertion failure in ldlm_handle2lock()
1358 Details : fix a race between class_handle_unhash() and class_handle2object()
1359 introduced in lustre 1.6.5 by bug 13622.
1364 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1365 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1370 Description: oops in sd_iostats_seq_show()
1371 Details : unloading/reloading the scsi low level driver triggers a kernel
1372 bug when trying to access the sd iostat file.
1377 Description: Kernel panics during QLogic driver reload
1378 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1379 patch, causing memory corruption.
1384 Description: journal_dev option does not work in b1_6
1385 Details : pass mount option during pre-mount.
1387 Severity : enhancement
1389 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1390 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1391 extent information of a file. It can be used to map logical blocks
1392 in a file to physical blocks in the block device.
1396 Description: LDLM soft lockups - improvement
1397 Details : It is be possible to send the lock handle along with each read
1398 or write request because the client is already doing a lock match
1399 itself so there isn't any reason the OST should have to re-do that
1403 Frequency : only on Cray X2
1405 Description: X2 build failures
1406 Details : fix build failures on Cray X2.
1410 Description: xid & resent requests
1411 Details : Initialize RPC XID from clock at startup (randomly if clock is
1414 Severity : enhancement
1416 Description: Add lustre_start utility to start or stop multiple Lustre servers
1421 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
1422 Details : In case of memory pressure, list_del() can be called twice on
1423 req->rq_history_list, causing a kernel oops.
1425 --------------------------------------------------------------------------------
1427 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
1429 * Support for kernels:
1430 2.6.5-7.283 (SLES 9),
1431 2.6.9-55.EL (RHEL 4),
1432 2.6.16.46-0.14 (SLES 10),
1433 2.6.18.8 vanilla (kernel.org)
1434 * Client support for unpatched kernels:
1435 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1436 2.6.16 - 2.6.22 vanilla (kernel.org)
1437 * Due to recently discovered recovery problems, we do not recommend
1438 using patchless RHEL 4 clients with this or any earlier release.
1439 * Recommended e2fsprogs version: 1.39.cfs8
1440 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1441 * Starting with this release, the ldiskfs backing filesystem required
1442 by Lustre is now in its own package, lustre-ldiskfs. This package
1443 should be installed. It is versioned separately from Lustre and
1444 may be released separately in future.
1448 Description: block reactivating mgc import until all deactivates complete
1449 Details : Fix race when failing back MDT/MGS to itself (testing)
1452 Frequency : at statup only
1454 Description: mds_lov_synchronize race leads to various problems
1455 Details : simultaneous MDT->OST connections at startup can cause the
1456 sync to abort, leaving the OSC in a bad state.
1458 Severity : enhancement
1460 Description: add optional extra BUILD_VERSION info
1461 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1462 to override the lustre version.
1465 Frequency : 2.6.18 servers only
1467 Description: ll_kern_mount() doesn't release the module reference
1468 Details : The ldiskfs module reference count never drops down to 0
1469 because ll_kern_mount() doesn't release the module reference.
1474 Description: server LBUG when using old ost_num_threads parameter
1475 Details : Accept the old ost_num_threads parameter but warn that it
1476 is deprecated, and fix an off-by-one error that caused an LBUG.
1481 Description: Transient SCSI error results in persistent IO issue
1482 Details : iobuf->dr_error is not reinitialized to 0 between two
1486 Frequency : sometimes when underlying device returns I/O errors
1488 Description: OSTs not going read-only during write failures
1489 Details : OSTs are not remounted read-only when the journal commit threads
1490 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1491 instead of the ext3 wrappers.
1495 Description: poor connect scaling with increasing client count
1496 Details : Don't run filter_grant_sanity_check for more than 100 exports
1497 to improve scaling for large numbers of clients.
1500 Frequency : SLES10 only
1502 Description: sanity-quota.sh quotacheck failed: rc = -22
1503 Details : Quotas cannot be enabled on SLES10.
1506 Frequency : liblustre clients only
1508 Description: getdirentries does not give error when run on compute nodes
1509 Details : getdirentries does not fail when the size specified as an argument
1510 is too small to contain at least one entry
1512 Severity : enhancement
1514 Description: Add LNET router traceability for debug purposes
1515 Details : If a checksum failure occurs with a router as part of the
1516 IO path, the NID of the last router that forwarded the bulk data
1517 is printed so it can be identified.
1522 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1523 Details : Due to a race condition, liblustre clients were occasionally
1524 evicted incorrectly.
1526 Severity : enhancement
1528 Description: lfs setstripe use optional parameters instead of postional
1531 Severity : enhancement
1533 Description: Nanosecond timestamp support for ldiskfs
1534 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1535 resolution timestamps. There is not yet support for this at
1536 the Lustre filesystem level.
1539 Frequency : during server recovery
1541 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1542 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1546 Frequency : nfs export on patchless client
1548 Description: connectathon hang when test nfs export over patchless client
1549 Details : Disconnected dentry cannot be found with lookup, so we do not need
1550 to unhash it or make it invalid
1553 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1554 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1555 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1560 Description: open req refcounting wrong on reconnect
1561 Details : If reconnect happened between getting open reply from server and
1562 call to mdc_set_replay_data in ll_file_open, we will schedule
1563 replay for unreferenced request that we are about to free.
1564 Subsequent close will crash in variety of ways.
1565 Check that request is still eligible for replay in
1566 mdc_set_replay_data().
1571 Description: disable writes to filesystem when reading health_check file
1572 Details : the default for reading the health_check proc file has changed
1573 to NOT do a journal transaction and write to disk, because this
1574 can cause reads of the /proc file to hang and block HA state
1575 checking on a healthy but otherwise heavily loaded system. It
1576 is possible to return to the previous behaviour during configure
1577 with --enable-health-write.
1579 Severity : enhancement
1581 Description: 64-bit inode version
1582 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1583 to the inode. This will be required for version-based recovery.
1588 Description: MDS fails to start if a duplicate client export is detected
1589 Details : in some rare cases it was possible for a client to connect to
1590 an MDS multiple times. Upon recovery the MDS would detect this
1591 and fail during startup. Handle this more gracefully.
1593 Severity : enhancement
1595 Description: Add -o localflock option to simulate old noflock
1597 Details : This will achieve local-only flock/fcntl locks
1603 Description: log_commit_thread vs filter_destroy race leads to crash
1604 Details : Take import reference before releasing llog record semaphore
1609 Description: Wrong request locking in request set processing
1610 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1611 imp_delayed_list, in this place should be used imp_lock.
1614 Frequency : when reconnection
1616 Description: Grant Leak when osc reconnect to OST
1617 Details : When osc reconnect ost, OST(filter) should check whether it
1618 should grant more space to client by comparing fed_grant and
1619 cl_avail_grant, and return the granted space to client instead
1620 of "new granted" space, because client will call osc_init_grant
1621 to update the client grant space info.
1624 Frequency : when client reconnect to OST
1626 Description: Grant Leak when osc do resend and replay bulk write
1627 Details : When osc reconnect to OST, OST(filter)should clear grant info of
1628 bulk write request, because the grant info while be sync between
1629 OSC and OST when reconnect, and we should ignore the grant info
1630 these of resend/replay write req.
1635 Description: Grant space more than avaiable left space sometimes.
1636 Details : When then OST is about to be full, if two bulk writing from
1637 different clients came to OST. Accord the avaliable space of the
1638 OST, the first req should be permitted, and the second one
1639 should be denied by ENOSPC. But if the seconde arrived before
1640 the first one is commited. The OST might wrongly permit second
1641 writing, which will cause grant space > avaiable space.
1644 Frequency : when client is evicted
1646 Description: Grant might be wrongly erased when osc is evicted by OST
1647 Details : when the import is evicted by server, it will fork another
1648 thread ptlrpc_invalidate_import_thread to invalidate the
1649 import, where the grant will be set to 0. While the original
1650 thread will update the grant it got when connecting. So if
1651 the former happened latter, the grant will be wrongly errased
1652 because of this race.
1657 Description: Checking Stale with correct fid
1658 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1659 to check whether it is stale, because sometimes, we want the
1660 enqueue happened anyway, and op_data.fid2 will not be initialized.
1662 Severity : enhancement
1664 Description: update patchless client
1665 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1668 Frequency : only with 2.4 kernel
1670 Description: random memory corruption
1671 Details : size of struct ll_inode_info is to big for union inode.u and this
1672 can be cause of random memory corruption.
1677 Description: Memory leak in recovery
1678 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1679 It should also check obd_fail before fsfilt_start, otherwise if
1680 fsfilt_start return -EROFS,(failover mds during mds recovery).
1681 then the req will return with repmsg->transno = 0 and rc = EROFS.
1682 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1683 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1684 be freed no matter whether fsfilt_commit success or not.
1687 Frequency : only with huge count clients
1689 Description: Prevents from taking the superblock lock in llap_from_page for
1691 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1692 for llap_from_page call in ll_removepage prevents from taking the
1693 superblock lock for a soon died page.
1698 Description: Not check open intent error before release open handle
1699 Details : in some rare cases, the open intent error is not checked before
1700 release open handle, which may cause
1701 ASSERTION(open_req->rq_transno != 0), because it tries to release
1702 the failed open handle.
1707 Description: Set cat log bitmap only after create log success.
1708 Details : in some rare cases, the cat log bitmap is set too early. and it
1709 should be set only after create log success.
1713 Description: Accessing a block bevice can re-enable I/O when Lustre is
1714 tearing down a device.
1715 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1720 Description: service threads may hog cpus when there are a lot of requests
1721 Details : Insert cond_resched to give other threads a chance to use some CPU
1726 Description: the cat log was not initialized in recovery
1727 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1728 unlink log on mds will not be initialized until mds post
1729 recovery. And also in mds post recovery, the unlink log will
1730 initialization will be done asynchronausly, so there will be race
1731 between add unlink log and unlink log initialization.
1735 Description: brw_stats were being printed incorrectly
1736 Details : brw_stats were being printed as log2 but all of them were not
1737 recorded as log2. Also remove some code duplication arising from
1738 filter_tally_{read,write}.
1742 Frequency : rare, only in recovery.
1743 Description: ASSERTION(req->rq_type != LI_POISON) failed
1744 Details : imp_lock should be held while iterating over imp_sending_list for
1745 prevent destroy request after get timeout in ptlrpc_queue_wait.
1749 Description: replay-single.sh test 52 fails
1750 Details : A lock's skiplist need to be cleanup when it being unlinked
1751 from its resource list.
1755 Description: Short directio read returns full requested size rather than
1757 Details : Direct I/O operations should return actual amount of bytes
1758 transferred rather than requested size.
1760 Severity : enhancement
1762 Description: metadata RPC reduction (e.g. for rm performance)
1763 Details : decrease the amount of synchronous RPC between clients and servers
1764 by canceling conflicing lock before the operation on the client
1765 and packing thier handles into the main operation RPC to server.
1767 Severity : enhancement
1769 Description: Async OSC create to avoid the blocking unnecessarily.
1770 Details : If a OST has no remain object, system will block on the creating
1771 when need to create a new object on this OST. Now, ways use
1772 pre-created objects when available, instead of blocking on an
1773 empty osc while others are not empty. If we must block, we block
1774 for the shortest possible period of time.
1778 Description: improve handling recoverable errors
1779 Details : if request processig with error which can be recoverable on server
1780 request should be resend, otherwise page released from cache and
1783 Severity : enhancement
1785 Description: refine locking for avoid write wrong info into lov_objid file
1786 Details : fix possible races with add new target and write/update data in
1789 --------------------------------------------------------------------------------
1791 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1796 Frequency : on some architectures
1798 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1799 Details : Uninitialized flags sometimes cause configuration commands to
1803 Frequency : patchless clients only
1805 Description: missing __iget() symbol export
1806 Details : The __iget() symbol export is missing. To avoid the need for
1807 this on patchless clients the deathrow inode reaper is turned
1808 off, and we depend on the VM to clean up old inodes. This
1809 dependency was during via the fix for bug 12181.
1811 --------------------------------------------------------------------------------
1813 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1815 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1816 INTEROPERATE with older versions automatically. In many cases a
1817 special upgrade step is needed. Please read the
1818 user documentation before upgrading any part of a live system.
1819 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1820 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1821 * WARNING: Lustre configuration and startup changes are required with
1822 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1826 Severity : enhancement
1828 Description: MountConf
1829 Details : Lustre configuration is now managed via mkfs and mount
1830 commands instead of lmc and lconf. New obd types (MGS, MGC)
1831 are added for dynamic configuration management. See
1832 https://mail.clusterfs.com/wikis/lustre/MountConf for
1835 Severity : enhancement
1837 Description: dynamic OST addition
1838 Details : OSTs can now be added to a live filesystem
1840 Severity : enhancement
1842 Description: startup order invariance
1843 Details : MDTs and OSTs can be started in any order. Clients only
1844 require the MDT to complete startup.
1846 Severity : enhancement
1848 Description: parallel, asynchronous orphan cleanup
1849 Details : orphan cleanup is now performed in separate threads for each
1850 OST, allowing parallel non-blocking operation.
1852 Severity : enhancement
1854 Description: optimized stripe assignment
1855 Details : stripe assignments are now made based on ost space available,
1856 ost previous usage, and OSS previous usage, in order to try
1857 to optimize storage space and networking resources.
1859 Severity : enhancement
1861 Description: Permanently set tunables
1862 Details : All writable /proc/fs/lustre tunables can now be permanently
1863 set on a per-server basis, at mkfs time or on a live system.
1865 Severity : enhancement
1867 Description: Lustre message v2
1868 Details : Add lustre message format v2.
1870 Severity : enhancement
1872 Description: client OST exclusion list
1873 Details : Clients can be started with a list of OSTs that should be
1874 declared "inactive" for known non-responsive OSTs.
1878 Description: SPEC SFS validation failure on NFS v2 over lustre.
1879 Details : Changes the blocksize for regular files to be 2x RPC size,
1880 and not depend on stripe size.
1882 Severity : enhancement
1884 Description: Multiple MD RPCs in flight.
1885 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1886 To avoid overly-overloading MDS, introduce a limit on number of
1887 MDS RPCs in flight for a single client and add /proc controls
1888 to adjust this limit.
1890 Severity : enhancement
1892 Description: client read/write statistics
1893 Details : Add client read/write call usage stats for performance
1894 analysis of user processes.
1895 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1896 file access. extents_stats shows chunk size distribution.
1897 extents_stats_per_process show chunk size distribution per
1900 Severity : enhancement
1902 Description: mds statistics
1903 Details : Add detailed mds operations statistics in
1904 /proc/fs/lustre/mds/*/stats.
1908 Description: Failure of copying files with lustre special EAs.
1909 Details : Client side always return success for setxattr call for lustre
1910 special xattr (currently only "trusted.lov").
1915 Description: Refcount LNET uuids
1916 Details : The global LNET uuid list grew linearly with every startup;
1917 refcount repeated list entries instead of always adding to
1920 Severity : enhancement
1922 Description: Dynamic service threads
1923 Details : Within a small range, start extra service threads
1924 automatically when the request queue builds up.
1927 Frequency : mixed-endian client/server environments
1929 Description: mixed-endian crashes
1930 Details : The new msg_v2 system had some failures in mixed-endian
1933 Severity : enhancement
1935 Description: Easy OST removal
1936 Details : OSTs can be permanently deactivated with e.g. 'lctl
1937 conf_param lustre-OST0001.osc.active=0'
1939 Severity : enhancement
1941 Description: MGS proc entries
1942 Details : Added basic proc entries for the MGS showing what filesystems
1945 Severity : enhancement
1947 Description: provide MGS failover
1948 Details : Added config lock reacquisition after MGS server failover.
1950 Severity : enhancement
1952 Description: add Linux 2.4 support
1953 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1957 Description: a large application tries to do I/O to the same resource and dies
1958 in the middle of it.
1959 Details : Check the req->rq_arrival time after the call to
1960 ost_brw_lock_get(), but before we do anything about
1961 processing it & sending the BULK transfer request. This
1962 should help move old stale pending locks off the queue as
1963 quickly as obd_timeout.
1966 Frequency : when an incorrect nid is specified during startup
1968 Description: ptlrpc connect to non-existant node causes kernel crash
1969 Details : LNET can't be re-entered from an event callback, which
1970 happened when we expire a message after the export has been
1971 cleaned up. Instead, hand the zombie cleanup off to another
1974 Severity : enhancement
1976 Description: plain/inodebits lock performance improvement
1977 Details : Grouping plain/inodebits in granted list by their request modes
1978 and bits policy, thus improving the performance of search through
1982 Frequency : only if OST filesystem is corrupted
1984 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1985 Details : for a short time RPCs with bulk IO are in the replay list,
1986 but replay of bulk IOs is unimplemented. If the OST filesystem
1987 is corrupted due to disk cache incoherency and then replay is
1988 started it is possible to trip an assertion. Avoid putting
1989 committed RPCs into the replay list at all to avoid this issue.
1992 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1994 Description: System hang on startup
1995 Details : This bug allowed the liblustre (e.g. catamount) client to
1996 return to the app before handling all startup RPCs. This
1997 could leave the node unresponsive to lustre network traffic
1998 and manifested as a server ptllnd timeout.
2000 Severity : enhancement
2002 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2003 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
2004 boolean to control whether to print peer debug info when a
2005 client's RPC times out.
2008 Frequency : only for kernels with patches from Lustre below 1.4.3
2010 Description: Remove old rdonly API
2011 Details : Remove old rdonly API which unused from at least lustre 1.4.3
2014 Frequency : only for devices with external journals
2016 Description: Set external device read-only also
2017 Details : During a commanded failover stop, we set the disk device
2018 read-only while the server shuts down. We now also set any
2019 external journal device read-only at the same time.
2022 Frequency : when upgrading from 1.4 while trying to change parameters
2024 Description: The wrong (new) MDC name was used when setting parameters for
2025 upgraded MDT's. Also allows changing of OSC (and MDC)
2026 parameters if --writeconf is specified at tunefs upgrade time.
2029 Frequency : when setting specific ost indicies
2031 Description: QOS code breaks on skipped indicies
2032 Details : Add checks for missing OST indicies in the QOS code, so OSTs
2033 created with --index need not be sequential.
2035 Severity : enhancement
2037 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2038 Details : The uninit_groups feature works in conjunction with the kernel
2039 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2040 up the pass1 processing of e2fsck. This is a read-only feature
2041 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2042 mount filesystems that have had this feature enabled.
2044 Severity : enhancement
2046 Description: Improve multi-block allocation algorithm to avoid fragmentation
2047 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2048 allocation locality and avoid filesystem fragmentation.
2050 ------------------------------------------------------------------------------
2052 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
2054 * Support for kernels:
2055 2.4.21-47.0.1.EL (RHEL 3)
2056 2.6.5-7.283 (SLES 9)
2057 2.6.9-42.0.10.EL (RHEL 4)
2058 2.6.12.6 vanilla (kernel.org)
2059 2.6.16.27-0.9 (SLES 10)
2060 * Recommended e2fsprogs version: 1.39.cfs5
2062 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2066 Frequency : occasional, depends on client load and configuration
2067 Bugzilla : 12181, 12203
2068 Description: data loss for recently-modified files
2070 Details : In some cases it is possible that recently written or created
2071 files may not be written to disk in a timely manner (this should
2072 normally be within 30s unless client IO load is very high).
2073 The problem appears as zero-length files or files that are a
2074 multiple of 1MB in size after a client crash or client eviction
2075 that are missing data at the end of the file.
2077 This problem is more likely to be hit on clients where files are
2078 repeatedly created and unlinked in the same directory, clients
2079 have a large amount of RAM, have many CPUs, the filesystem has
2080 many OSTs, the clients are rebooted frequently, and/or the files
2081 are not accessed by other nodes after being written.
2083 The presence of the problem can be detected by looking at
2084 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2085 smaller than the second (nr_unused) then dirty files will not
2086 be flushed automatically to disk. "sync; sleep 10" should be
2087 run several times on the node before unmounting it to update
2088 Lustre (this is also safe to run on nodes without this problem).
2090 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2091 that can cause this same problem, so customers using that kernel
2092 also need to update the kernel in addition to Lustre. In order
2093 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2095 It is normal that files written just before a client crash (less
2096 than 30s) may not yet have been flushed to disk, even for local
2100 Frequency : frequent on thin XT3 nodes
2102 Description: UUID collision on thin XT3 Linux nodes
2103 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2104 randomly, since we relied on an insufficiently-seeded PRNG.
2109 Description: OSS hangs after "All ost request buffers busy"
2110 Details : A deadlock between quota and journal operations caused OSS
2111 hangs after printing "All ost request buffers busy."
2114 Frequency : always on liblustre builds
2116 Description: Cleanup compiler warnings on liblustre
2119 Frequency : always on liblustre builds on XT3
2121 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2125 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2126 > 32000 subdirectories
2127 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2128 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2129 subdirectory count crosses 32000. This will aid e2fsck to
2130 correctly handle more than 32000 subdirectories.
2133 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2135 Description: System hang on startup
2136 Details : This bug allowed the liblustre (e.g. catamount) client to
2137 return to the app before handling all startup RPCs. This
2138 could leave the node unresponsive to lustre network traffic
2139 and manifested as a server ptllnd timeout.
2141 Severity : enhancement
2143 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2144 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2145 boolean to control whether to print peer debug info when a
2146 client's RPC times out.
2151 Description: make O_SYNC working on 2.6 kernels
2152 Details : 2.6 kernels use different method for mark pages for write,
2153 so need add a code to lustre for O_SYNC work.
2158 Description: Failure to close file and release space on NFS
2159 Details : Put inode details into lock acquired in ll_intent_file_open.
2160 Use mdc_intent_lock in ll_intent_open to properly
2161 detect all kind of errors unhandled by mdc_enqueue.
2166 Description: proc file read during shutdown sometimes raced obd removal,
2168 Details : Add lock to prevent obd access after proc file removal.
2171 Frequency : Only for files larger than 4GB on 32-bit clients.
2173 Description: improperly doing page alignment of locks
2174 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2175 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2180 Description: RPCs being resent when they shouldn't be
2181 Details : Some RPCs that should not be resent are being resent. This
2182 can cause inconsistencies in the RPC state machine. Do not
2183 resend such requests.
2186 Frequency : rare, only with NFS export
2188 Description: Crash on NFS re-export node
2189 Details : under very unusual load conditions an assertion is hit in
2190 ll_intent_file_open()
2193 Frequency : only if OST filesystem is corrupted
2195 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2196 Details : for a short time RPCs with bulk IO are in the replay list,
2197 but replay of bulk IOs is unimplemented. If the OST filesystem
2198 is corrupted due to disk cache incoherency and then replay is
2199 started it is possible to trip an assertion. Avoid putting
2200 committed RPCs into the replay list at all to avoid this issue.
2205 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2206 Details : Large single O_DIRECT read and write calls can fail to allocate
2207 a sufficiently large buffer to process the request. In case of
2208 allocation failure the allocation is retried with a smaller
2209 buffer and broken into smaller requests.
2211 Severity : enhancement
2213 Description: Add -o localflock option to simulate old noflock behaviour.
2214 Details : This will achieve local-only flock/fcntl locks coherentness.
2219 Description: versioning check is incomplete
2220 Details : Checking the version difference of client vs. server, report
2221 error if the gap is too big.
2226 Description: add support PG_writeback bit
2227 Details : add support for PG_writeback bit for Lustre, for more carefull
2228 work with page cache in 2.6 kernel. This also fix some deadlocks
2229 and remove hack for work O_SYNC with 2.6 kernel.
2231 Severity : enhancement
2233 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2234 Details : The uninit_groups feature works in conjunction with the kernel
2235 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2236 up the pass1 processing of e2fsck. This is a read-only feature
2237 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2238 mount filesystems that have had this feature enabled.
2240 Severity : enhancement
2242 Description: Improve multi-block allocation algorithm to avoid fragmentation
2243 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2244 allocation locality and avoid filesystem fragmentation.
2246 ------------------------------------------------------------------------------
2248 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2250 * Support for kernels:
2251 2.6.9-42.0.3.EL (RHEL 4)
2252 2.6.5-7.276 (SLES 9)
2253 2.4.21-47.0.1.EL (RHEL 3)
2254 2.6.12.6 vanilla (kernel.org)
2255 2.6.16.21-0.8 (SLES10)
2256 * Recommended e2fsprogs version: 1.39.cfs2-0
2258 * The backwards-compatible /proc/sys/portals symlink has been removed
2259 in this release. Before upgrading, please ensure that you change
2260 any configuration scripts or /etc/sysctl.conf files that access
2261 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2262 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2263 in advance of the upgrade on any system running Lustre 1.4.6 or
2264 newer, since /proc/sys/lnet was added in that version.
2265 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2269 Frequency : only when quota is used
2271 Description: avoid scanning export list for quota master
2272 Details : Change the algorithms to avoid scanning export list in order
2273 to improve the efficiency.
2276 Frequency : MDS failover only, very rarely
2278 Description: "went back in time" messages on mds failover
2279 Details : The greatest transno may be lost when the current operation
2280 finishes with an error (transno==0) and the client's last_rcvd
2281 record is over-written. Save the greatest transno in the
2282 mds_last_transno for this case.
2285 Frequency : always for specific kernels and striping counts
2287 Description: client may get "Matching packet too big" without ACL support
2288 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2289 when trying to access files in certain configurations. The
2290 clients should in fact be denied when mounting because they do
2291 not understand ACLs.
2294 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2296 Description: many clients connecting with IO in progress causes connect timeouts
2297 Details : Avoid synchronous journal commits to avoid delays caused by many
2298 clients connecting/disconnecting when bulk IO is in progress.
2299 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2300 OST_IO_PORTAL to avoid delays behind potentially many pending
2304 Frequency : occasionally with multiple writers to a single file
2306 Description: shared writes to file may result in wrong size reported by stat()
2307 Details : Allow growing of kms when extent lock is cancelled
2310 Frequency : always with random mmap IO to multi-striped file
2312 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2313 Details : Only if the hole is at the end of OST object so that kms is too
2314 small. Fix is to increase kms accordingly in ll_nopage.
2317 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2319 Description: writes to a missing object would leak memory on the OST
2320 Details : If there is an inconsistency between the MDS and OST filesystems,
2321 such that the MDS references an object that doesn't exist, writes
2322 to that object will leak memory due to incorrect cleanup in the
2323 error handling path, eventually running out of memory on the OST.
2328 Description: Creating too long symlink causes lustre errors
2329 Details : Check symlink and name lengths before sending requests to MDS.
2332 Frequency : only if flock is enabled (not on by default)
2334 Description: posix locks not released on fd closure on 2.6.9+
2335 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2336 kernels, this caused such locks not to be released on fd close and
2337 then assertions on fs unmount about still used locks.
2340 Frequency : MDS failover only, very rarely
2342 Description: clients may get ASSERTION(granted_lock != NULL)
2343 Details : When request was taking a long time, and a client was resending
2344 a getattr by name lock request. The were multiple lock requests
2345 with the same client lock handle and
2346 mds_getattr_name->fixup_handle_for_resent_request found one of the
2347 lock handles but later failed with ASSERTION(granted_lock != NULL).
2352 Description: handle->h_buffer_credits > 0, assertion failure
2353 Details : h_buffer_credits is zero after truncate, causing assertion
2354 failure. This patch extends the transaction or creates a new
2358 Frequency : NFS re-export or patchless client
2359 Bugzilla : 11179, 10796
2360 Description: Crash on NFS re-export node (__d_move)
2361 Details : We do not want to hash the dentry if we don't have a lock.
2362 But if this dentry is later used in d_move, we'd hit uninitialised
2363 list head d_hash, so we just do this to init d_hash field but
2364 leave dentry unhashed.
2367 Frequency : NFS re-export or patchless client
2369 Description: NFS exports has problem with symbolic link
2370 Details : lustre client didn't properly install dentry when re-exported
2371 to NFS or running patchless client.
2374 Frequency : NFS re-export or patchless client
2376 Description: Various nfs/patchless fixes.
2377 Details : fixes reuse disconected alias for lookup process - this fixes
2378 warning "find_exported_dentry: npd != pd",
2379 fix permission error with open files at nfs.
2380 fix apply umask when do revalidate.
2383 Frequency : occasional
2385 Description: Crash on NFS re-export node
2386 Details : calling clear_page() on the wrong pointer triggered oops in
2387 generic_mapping_read().
2390 Frequency : rarely, using O_DIRECT IO
2392 Description: unaligned directio crashes client with LASSERT
2393 Details : check for unaligned buffers before trying any requests.
2396 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2398 Description: stale data returned from RAID cache
2399 Details : If only a small amount of IO is done to the RAID device before
2400 reading it again it is possible to get stale data from the RAID
2401 cache instead of reading it from disk.
2404 Frequency : always for sles10 kernel
2406 Description: sles10 support
2407 Details : ll_follow_link: compile fixes and using of nd_set_link
2408 under newer kernels.
2411 Frequency : depends on arch, kernel and compiler version, always on sles10
2414 Description: recursive or deep enough symlinks cause stack overflow
2415 Details : getting rid of large stack-allocated variable in
2419 Frequency : depends on hardware
2421 Description: lustre write performance loss in the SLES10 kernel
2422 Details : the performance loss is caused by using of write barriers in the
2423 ext3 code. The SLES10 kernel turns barrier support on by
2424 default. The fix is to undo that change for ldiskfs.
2426 ------------------------------------------------------------------------------
2428 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2430 * Support for kernels:
2431 2.6.9-42.0.3EL (RHEL 4)
2432 2.6.5-7.276 (SLES 9)
2433 2.4.21-47.0.1.EL (RHEL 3)
2434 2.6.12.6 vanilla (kernel.org)
2438 Frequency : quota enabled and large files being deleted
2440 Description: releasing more than 4GB of quota at once hangs OST
2441 Details : If a user deletes more than 4GB of files on a single OST it
2442 will cause the OST to spin in an infinite loop. Release
2443 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2448 Description: statfs data retrieved from /proc may be stale or zero
2449 Details : When reading per-device statfs data from /proc, in the
2450 {kbytes,files}_{total,free,avail} files, it may appear
2451 as zero or be out of date.
2454 Frequency : systems with MD RAID1 external journal devices
2456 Description: lconf's call to blkid is confused by RAID1 journal devices
2457 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2458 of returning all block devices that match the journal UUID.
2461 Frequency : always, for aggregate stripe size over 4GB
2463 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2464 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2465 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2468 Frequency : always if "lfs find" used on a local file/directory
2470 Description: "lfs find" segfaults if used on a local file/directory
2471 Details : The case where a directory component was not specified wasn't
2472 handled correctly. Handle this properly.
2475 Frequency : always on ppc64
2477 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2478 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2479 perform endianness conversion.
2482 Frequency : rarely (truncate to non-zero file size after write under load)
2483 Bugzilla : 10730, 10687
2484 Description: Files padded with zeros to next 4K multiple
2485 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2486 it is possible that files that are truncated to a non-zero size
2487 immediately after being written are filled with zero bytes beyond
2488 the truncated size. No file data is lost.
2490 Severity : enhancement
2492 Description: Allow recovery/failover for liblustre clients.
2493 Details : liblustre clients were unaware of failover configurations until
2496 Severity : enhancement
2498 Description: user file locks should fail when not mounting with flock option
2499 Details : Set up an error-returning stub in ll_file_operations.lock field
2500 to prevent incorrect behaviour when client is mounted without
2501 flock option. Also, set up properly f_op->flock field for
2505 Frequency : always on ia64
2507 Description: "lfs df" loops on printing out MDS statfs information
2508 Details : The obd_ioctl_data was not initialized and in some systems
2509 this caused a failure during the ioctl that did not return
2510 an error. Initialize the struct and return an error on failure.
2513 Frequency : SLES 9 only
2515 Description: Error of copying files with lustre special EAs as root
2516 Details : Client side always return success for setxattr call for lustre
2517 special xattr (currently only "trusted.lov").
2520 Frequency : rarely on clusters with both ia64+i386 clients
2522 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2523 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2524 clients are doing concurrent writes to the same file it is
2525 possible that the ia64 clients may LASSERT because the OST
2526 extent locks are not PAGE_SIZE aligned. Ensure that grown
2527 locks are always aligned on the request boundary.
2530 Frequency : specific use, occasional
2532 Description: Overwriting in use executable truncates on-disk binary image
2533 Details : If one node attempts to overwrite an executable in use by
2534 another node, we now correctly return ETXTBSY instead of
2535 truncating the file.
2537 Severity : enhancement
2539 Description: Async OSC create to avoid the blocking unnecessarily.
2540 Details : If a OST has no remain object, system will block on the creating
2541 when need to create a new object on this OST. Now, ways use
2542 pre-created objects when available, instead of blocking on an
2543 empty osc while others are not empty. If we must block, we block
2544 for the shortest possible period of time.
2549 Description: chmod on Lustre root is propagated to other clients
2550 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2551 invalid by the follow_mount time.
2554 Frequency : liblustre clients only
2556 Description: Race in 'instant cancel' lock handling could lead to such locks
2557 never to be granted in case of SMP MDS
2558 Details : Do not destroy not yet granted but cbpending locks in
2562 Frequency : replay/resend of open
2564 Description: non null lock assetion failure in mds_intent_policy
2565 Details : Trying to replay/resend lockless open requests resulted in
2566 mds_open() returning 0 with no lock. Now it sets a flag if
2567 a lock is going to be returned.
2569 Severity : enhancement
2571 Description: Checksum enhancements
2572 Details : New checksum enhancements allow for resending RPCs that failed
2575 Severity : enhancement
2577 Description: Tunables on number of dirty pages in cacche
2578 Details : Allow to set limit on number of dirty pages cached.
2583 Description: client crash on unmount - lock still has references
2584 Details : In some error handling cases it was possible to leak a lock
2585 reference on a client while accessing a file. This was not
2586 harmful to the client during operation, but would cause the
2587 client to crash when the filesystem is unmounted.
2590 Frequency : specific case, rare
2592 Description: ETXTBSY on mds though file not in use by client
2593 Details : ETXTBSY is no longer incorrectly returned when attempting to
2594 chmod or chown a directory that the user previously tried to
2595 execute or a currently-executing binary.
2598 Frequency : extremely rare except on liblustre-based clients
2600 Description: Lustre space not freed when files are deleted
2601 Details : Clean up open-unlinked files after client eviction. Previously
2602 the unlink was skipped and the files remained as orphans.
2607 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2608 Details : In some cases it was possible to send a blocking callback to
2609 a client doing a glimpse, even though that client didn't get
2610 a lock granted. When the glimpse lock is cancelled on the OST
2611 the freed lock is left on the waiting list and corrupted the list.
2614 Frequency : all core dumps
2616 Description: Broke core dumps to lustre
2617 Details : Negative dentry may be unhashed if parent does not have UPDATE
2618 lock, but some callers, e.g. do_coredump, expect dentry to be
2619 hashed after successful create, hash it in ll_create_it.
2621 ------------------------------------------------------------------------------
2623 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2625 * Support for kernels:
2626 2.6.9-42.0.2.EL (RHEL 4)
2627 2.6.5-7.276 (SLES 9)
2628 2.4.21-40.EL (RHEL 3)
2629 2.6.12.6 vanilla (kernel.org)
2633 Frequency : always on RHEL 3
2635 Description: Number of open files grows over time
2636 Details : The number of open files grows over time, whether or not
2637 Lustre is started. This was due to a filp leak introduced
2638 by one of our kernel patches.
2640 ------------------------------------------------------------------------------
2642 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2644 * Support for kernels:
2645 2.6.9-42.EL (RHEL 4)
2646 2.6.5-7.276 (SLES 9)
2647 2.4.21-40.EL (RHEL 3)
2648 2.6.12.6 vanilla (kernel.org)
2653 Bugzilla : 5719, 9635, 9792, 9684
2654 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2655 Details : If a server is under heavy load and cannot reply to new
2656 connection requests before the client resends the (re)connect,
2657 the connection handling code can behave badly if two service
2658 threads are concurrently handing separate (re)connections from
2659 the same client. Add better locking to the connection handling
2660 code, and ensure that only a single connection will be processed
2661 for a given client UUID, even if the lock is dropped.
2663 Severity : enhancement
2665 Description: add TCP zero-copy support to kernel
2666 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2667 sends if the hardware supports scatter-gather and checksumming.
2668 This allows socklnd to do client-write and server-read more
2669 efficiently and reduce CPU utilization from skbuf copying.
2672 Frequency : only if NFS exporting from client
2674 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2675 Details : If setting attributes on a file created under NFS that had
2676 never been opened it would be possible to oops the client
2677 if the file had no objects.
2680 Frequency : always for liblustre
2682 Description: liblustre client does MDS+OSTs setattr RPC for each write
2683 Details : When doing a write from a liblustre client, the client
2684 incorrectly issued an RPC to the MDS and each OST the file was
2685 striped over in order to update the timestamps. When writing
2686 with small chunks and many clients this could overwhelm the MDS
2687 with RPCs. In all cases it would slow down the write because
2688 these RPCs are unnecessary.
2690 Severity : enhancement
2692 Description: allow number of MDS service threads to be changed at module load
2693 Details : It is now possible to change the number of MDS service threads
2694 running. Adding "options mds mds_num_threads={N}" to the MDS's
2695 /etc/modprobe.conf will set the number of threads for the next
2696 time Lustre is restarted (assuming the "mds" module is also
2697 reloaded at that time). The default number of threads will
2698 stay the same, 32 for most systems.
2703 Description: OST crash if filesystem is unformatted or corrupt
2704 Details : If an OST is started on a device that has never been formatted
2705 or if the filesystem is corrupt and cannot even mount then the
2706 error handling cleanup routines would dereference a NULL pointer.
2711 Description: NULL pointer deref in llap_from_page.
2712 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2713 incorrect) mapping if the page was truncated/reclaimed while it was
2714 searched for. Check for this condition and skip such pages when
2715 doing readahead. Introduce extra check to llap_from_page() to
2716 verify page->mapping->host is non-NULL (so page is not anonymous).
2719 Frequency : Sometimes when using sys_sendfile
2721 Description: "page not covered by a lock" warnings from ll_readpage
2722 Details : sendfile called ll_readpage without right page locks present.
2723 Now we introduced ll_file_sendfile that does necessary locking
2724 around call to generic_file_sendfile() much like we do in
2728 Frequency : with certain MDS communication failures at client mount time
2730 Description: NULL pointer deref after failed client mount
2731 Details : a client connection request may delayed by the network layer
2732 and not be sent until after the PTLRPC layer has timed out the
2733 request. If the client fails the mount immediately it will try
2734 to clean up before the network times out the request. Add a
2735 reference from the request import to the obd device and delay
2736 the cleanup until the network drops the request.
2739 Frequency : occasionally during client (re)connect
2741 Description: assertion failure during client (re)connect
2742 Details : processing a client connection request may be delayed by the
2743 client or server longer than the client connect timeout. This
2744 causes the client to resend the connection request. If the
2745 original connection request is replied in this interval, the
2746 client may trip an assertion failure in ptlrpc_connect_interpret()
2747 which thought it would be the only running connect process.
2750 Frequency : only with obd_echo servers and clients that are rebooted
2752 Description: kernel BUG accessing uninitialized data structure
2753 Details : When running an obd_echo server it did not start the ping_evictor
2754 thread, and when a client was evicted an uninitialized data
2755 structure was accessed. Start the ping_evictor in the RPC
2756 service startup instead of the OBD startup.
2758 Severity : enhancement
2759 Bugzilla : 10193 (patchless)
2760 Description: Remove dependency on various unexported kernel interfaces.
2761 Details : No longer need reparent_to_init, exit_mm, exit_files,
2762 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2765 Frequency : rare (only users of deprecated and unsupported LDAP config)
2767 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2768 Details : LDAP apparently contains 'lustreName' attributes instead of
2769 'name'. A simple remapping of the name is sufficient.
2772 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2774 Description: waiting_locks_callback trips kernel BUG if client is evicted
2775 Details : Running with the dump_on_timeout debug flag turned on makes
2776 it possible that the waiting_locks_callback() can try to dump
2777 the Lustre kernel debug logs from an interrupt handler. Defer
2778 this log dumping to the expired_lock_main() thread.
2780 Severity : enhancement
2782 Description: Support NFS exporting on 2.6 kernels.
2783 Details : Implement non-rawops metadata methods for NFS server to use without
2784 changing NFS server code.
2787 Frequency : very rare (synthetic metadata workload only)
2789 Description: two racing renames might cause an MDS thread to deadlock
2790 Details : Running the "racer" program may cause one MDS thread to rename
2791 a file from being the source of a rename to being the target of
2792 a rename at exactly the same time that another thread is doing
2793 so, and the second thread has already enqueued these locks after
2794 doing a lookup of the target and is trying to relock them in
2795 order. Ensure that we don't try to re-lock the same resource.
2798 Frequency : only very large systems with liblustre clients
2800 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2801 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2802 all OSTs in parallel. This allows the eviction of stale liblustre
2803 clients to proceed much faster than if they were done in series,
2804 and also offers similar improvements for other set_info RPCs.
2809 Description: excessive CPU usage during initial read phase on client
2810 Details : During the initial read phase on a client, it would agressively
2811 retry readahead on the file, consuming too much CPU and impacting
2812 performance (since 1.4.5.8). Improve the readahead algorithm
2813 to avoid this, and also improve some other common cases (read
2814 of small files in particular, where "small" is files smaller than
2815 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2820 Description: MDS crash when receiving packet with unknown intent.
2821 Details : Do not LBUG in unknown intent case, just return -EFAULT
2823 Severity : enhancement
2824 Bugzilla : 9293, 9385
2825 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2826 Details : Do not serialize getattr (non-intent version) and statfs.
2829 Frequency : occasional, when OST network is overloaded/intermittent
2831 Description: client evicted by OST after bulk IO timeout
2832 Details : If a client sends a bulk IO request (read or write) the OST
2833 may evict the client if it is unresposive to its data GET/PUT
2834 request. This is incorrect if the network is overloaded (takes
2835 too long to transfer the RPC data) or dropped the OST GET/PUT
2836 request. There is no need to evict the client at all, since
2837 the pinger and/or lock callbacks will handle this, and the
2838 client can restart the bulk request.
2841 Frequency : Always when mmapping file with no objects
2843 Description: client crashes when mmapping file with no objects
2844 Details : Check that we actually have objects in a file before doing any
2845 operations on objects in ll_vm_open, ll_vm_close and
2851 Description: Request leak when working with deleted CWD
2852 Details : Introduce advanced request refcount tracking for requests
2853 referenced from lustre intent.
2855 Severity : Enhancement
2857 Description: Cache open file handles on client.
2858 Details : MDS now will return special lock along with openhandle, if
2859 requested and client is allowed to hold openhandle, even if unused,
2860 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2861 closing files for every read/write openration.
2863 Severity : Enhancement
2865 Description: Cache open negative dentries on client when possible.
2866 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2867 negative dentries on lock revocation.
2872 Description: Remounting a client read-only wasn't possible with a zconf mount
2873 Details : It wasn't possible to remount a client read-only with llmount.
2875 Severity : enhancement
2876 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2877 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2878 MPICH is included to add Lustre-specific ADIO interfaces.
2879 This is based closely on the UFS ADIO layer and only differs
2880 in file creation, in order to allow the OST striping to be set.
2881 This is user-contributed code and not supported by CFS.
2886 Description: extended inode attributes (immutable, append-only) work improperly
2887 when 2.4 and 2.6 kernels are used on client/server or vice versa
2888 Details : Introduce kernel-independent values for these flags.
2890 Severity : enhancement
2893 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2894 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2895 granularity. Now, Fractional MB granularity is be supported,
2896 this is very useful for low memory system.
2898 Severity : enhancement
2900 Description: Getattr by fid
2901 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2902 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2903 have locks and avoids getattr rpc requests for it on every stat.
2906 Frequency : Always, for filesystems larger than 2TB
2908 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2909 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2910 filesystems larger than 2TB to be mounted with mballoc enabled.
2913 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2915 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2916 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2917 filesystems larger than 2TB is mounted then the kernel will
2918 silently corrupt the start of the filesystem. CONFIG_LBD is
2919 enabled for all CFS-supported kernels, but the possibility of
2920 this happening with a modified kernel config exists.
2922 Severity : enhancement
2924 Description: add client O_DIRECT support for 2.6 kernels
2925 Details : It is now possible to do O_DIRECT reads and writes to files
2926 in the Lustre client mountpoint on 2.6 kernel clients.
2928 Severity : enhancement
2930 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2931 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2932 parallel, not waiting for response from every OST before sending
2933 a rpc to the next OST.
2938 Description: setattr vs write race when updating file timestamps
2939 Details : Client processes that update a file timestamp into the past
2940 right after writing to the file (e.g. tar) it is possible that
2941 the updated file modification time can be reset to the current
2942 time due to a race between processing the setattr and write RPC.
2944 Severity : enhancement
2946 Description: Bring 'lfs find' closer in line with regular Linux find.
2947 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2948 -print0 options and obtains all the needed info through the lustre
2951 Severity : enhancement
2953 Description: support up to 1024 configured devices on one node
2954 Details : change obd_dev array from statically allocated to dynamically
2955 allocated structs as they are first used to reduce memory usage
2960 Description: Flush dirty partially truncated pages during truncate
2961 Details : Immediatelly flush partially truncated pages in filter_setattr,
2962 this way we completely avoid having any pages in page cache on OST
2963 and can retire ugly workarounds during writes to flush such pages.
2968 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2969 Details : protect inode from truncation within vfs_unlink() context
2970 just take a reference before calling vfs_unlink() and release it
2971 when parent's i_sem is free.
2976 Description: last_id value checked outside lock on OST caused LASSERT failure
2977 Details : If there were multiple MDS->OST object precreate requests in
2978 flight, it was possible that the OST's last object id was checked
2979 outside a lock and incorrectly tripped an assertion. Move checks
2980 inside locks, and discard old precreate requests.
2983 Frequency : always, if extents are used on OSTs
2985 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2986 Details : OSTs using the extents format would not zero the high 16 bits of
2987 the index physical block number. This is not a problem for any
2988 OST filesystems smaller than 16TB, and no kernels support ext3
2989 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2990 new/modified files) and can be fixed for existing filesystems
2991 with e2fsprogs-1.39-cfs1.
2996 Description: import connection selection may be incorrect if timer wraps
2997 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2998 import connections to be ignored if the 32-bit jiffies counter
2999 wraps. Use a 64-bit jiffies counter.
3002 Frequency : very large clusters immediately after boot
3004 Description: LNET request buffers exhausted under heavy short-term load
3005 Details : If a large number of client requests are generated on a service
3006 that has previously never seen so many requests it is possible
3007 that the request buffer growth cannot keep up with the spike in
3008 demand. Instead of dropping incoming requests, they are held in
3009 the LND until the RPC service can accept more requests.
3012 Frequency : Sometimes during replay
3014 Description: Assertion failure in ll_local_open after replay.
3015 Details : If replay happened on an open request reply before we were able
3016 to set replay handler, reply will become not swabbed tripping the
3017 assertion in ll_local_open. Now we set the handler right after
3018 recognising of open request
3021 Frequency : very rare
3023 Description: kernel reports "badness in vsnprintf"
3024 Details : Reading from the "recovery_status" /proc file in small chunks
3025 may cause a negative length in lprocfs_obd_rd_recovery_status()
3026 call to vsnprintf() (which is otherwise harmless). Exit early
3027 if there is no more space in the output buffer.
3029 Severity : enhancement
3031 Description: clear OBD RPC statistics by writing to them
3032 Details : It is now possible to clear the OBD RPC statistics by writing
3033 to the "stats" file.
3038 Description: Client mtime is not the same on different clients after utimes
3039 Details : In some cases, the client was using the utimes() syscall on
3040 a file cached on another node. The clients now validate the
3041 ctime from the MDS + OSTs to determine which one is right.
3046 Description: Inability to activate failout mode
3047 Details : lconf script incorrectly assumed that in pythong string's numeric
3048 value is used in comparisons.
3051 Frequency : always with multiple stripes per file
3053 Description: Inefficient object allocation for mutli-stripe files
3054 Details : When selecting which OSTs to stripe files over, for files with
3055 a stripe count that divides evenly into the number of OSTs,
3056 the MDS is always picking the same starting OST for each file.
3057 Return the OST selection heuristic to the original design.
3062 Description: mount failures may take full timeout to return an error
3063 Details : Under some heavy load conditions it is possible that a
3064 failed mount can wait for the full obd_timeout interval,
3065 possibly several minutes, before reporting an error.
3066 Instead return an error as soon as the status is known.
3068 Frequency : quota enabled and large files being deleted
3070 Description: releasing more than 4GB of quota at once hangs OST
3071 Details : If a user deletes more than 4GB of files on a single OST it
3072 will cause the OST to spin in an infinite loop. Release
3073 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3078 Description: statfs data retrieved from /proc may be stale or zero
3079 Details : When reading per-device statfs data from /proc, in the
3080 {kbytes,files}_{total,free,avail} files, it may appear
3081 as zero or be out of date.
3084 Frequency : systems with MD RAID1 external journal devices
3086 Description: lconf's call to blkid is confused by RAID1 journal devices
3087 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3088 of returning all block devices that match the journal UUID.
3091 Frequency : always, for aggregate stripe size over 4GB
3093 Description: assertion fails when trying to use 4GB stripe size
3094 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
3095 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
3098 Frequency : always on ppc64
3100 Description: the first write on an ext3 filesystem with mballoc got stuck
3101 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3102 perform endianness conversion.
3104 ------------------------------------------------------------------------------
3106 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3108 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3109 INTEROPERATE with older versions automatically. Please read the
3110 user documentation before upgrading any part of a live system.
3111 * WARNING: Lustre networking configuration changes are required with
3112 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3115 * Support for kernels:
3116 2.6.9-22.0.2.EL (RHEL 4)
3117 2.6.5-7.244 (SLES 9)
3118 2.6.12.6 vanilla (kernel.org)
3121 Severity : enhancement
3122 Bugzilla : 7981/8208
3123 Description: Introduced Lustre Networking (LNET)
3124 Details : LNET is new networking infrastructure for Lustre, it includes
3125 a reorganized network configuration mode (see the user
3126 documentation for full details) as well as support for routing
3127 between different network fabrics. Lustre Networking Devices
3128 (LNDs) for the supported network fabrics have also been
3129 created for this new infrastructure.
3131 Severity : enhancement
3132 Description: Introduced Access control lists
3133 Details : clients can set ACLs on files and directories in order to have
3134 more fine-grained permissions than the standard Unix UGO+RWX.
3135 The MDS must be started with the "-o acl" mount option.
3137 Severity : enhancement
3138 Description: Introduced filesystem quotas
3139 Details : Administrators may now establish per-user quotas on the
3142 Severity : enhancement
3144 Description: Configuration change for the XT3
3145 The PTLLND is now used to run Lustre over Portals on the XT3
3146 The configure option(s) --with-cray-portals are no longer used.
3147 Rather --with-portals=<path-to-portals-includes> is used to
3148 enable building on the XT3. In addition to enable XT3 specific
3149 features the option --enable-cray-xt3 must be used.
3154 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3155 Details : A race between allocating a new llcd and re-getting the llcd_lock
3156 allowed another thread to grab newly-allocated llcd.
3158 Severity : enhancement
3160 Description: 2.6 OST async journal commit and locking fix to improve performance
3161 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3162 2.6 kernels are now async as they already were in 2.4 kernels so
3163 that they can commit concurrently with the network bulk transfer.
3164 For block-allocated files the filter allocation semaphore is held
3165 to avoid filesystem fragmentation during allocation. BKL lock
3166 removed for 2.6 xattr operations where it is no longer needed.
3171 Description: lconf incorrectly determined whether two IP networks could talk
3172 Details : In some more complicated routing and multiple-network
3173 configurations, lconf will avoid trying to make a network
3174 connection to a disjoint part of the IP space. It was doing the
3175 math incorrectly for one set of cases.
3180 Description: Fix for potential infinite loop processing records in an llog.
3181 Details : If an llog record is corrupted/zeroed, it is possible to loop
3182 forever in llog_process(). Validate the llog record length
3183 and skip the remainder of the block on error.
3186 Frequency : occasional (liblustre only)
3188 Description: liblustre could not open files whose last component is a symlink
3189 Details : sysio_path_walk() would incorrectly pass the open intent to
3190 intermediate path components.
3193 Frequency : rare (liblustre only with non-standard tuning)
3194 Bugzilla : 7201 (7350)
3195 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3196 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3197 after it has been released, only do this for referenced locks
3199 Severity : enhancement
3201 Description: specifying an (invalid) directory default stripe_size of -1
3202 would reset the directory default striping
3203 Details : stripe_size -1 was used internally to signal directory stripe
3204 removal, now use "all default" to signal dir stripe removal
3205 as a directory striping of "all default" is not useful
3208 Frequency : common for large clusters running liblustre clients
3210 Description: doing an ls when liblustre clients are running is slow
3211 Details : sending a glimpse AST to a liblustre client waits for every AST
3212 to time out, as liblustre clients will not respond. Since they
3213 cannot cache data we refresh the OST lock LVB from disk instead.
3215 Severity : enhancement
3217 Description: doing an ls at the same time as file IO can be slow
3218 Details : enqueue and other "small" requests can be blocked behind many
3219 large IO requests. Create a new OST IO portal for non-IO
3220 requests so they can be processed faster.
3223 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3225 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3226 created file with mode 000
3227 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3228 allow an ftruncate() as a non-root user to a file with mode 000.
3229 Lustre now respects this flag to disable mode checks when
3230 truncating a file owned by the user
3233 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3236 Description: Revoking locks from clients that went dead or catatonic might take
3238 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3239 cancellation of such locks instant on servers without waiting for
3240 any reply from clients. Clients drops these locks when cancel
3241 notification from server is received without replying.
3244 Frequency : liblustre-only, when liblustre client dies or becomes busy
3246 Description: Doing ls on Linux clients can take a long time with active
3248 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3249 granting such locks to it in the first place if possible. Locks
3250 are taken by proxy on the OST during the read or write and
3251 dropped immediately afterward. Add connect flags handling, do
3252 not grant locks to liblustre clients for glimpse ASTs.
3254 Severity : enhancement
3256 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3257 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3258 sequential reads and behaves badly with applications doing random
3259 reads. Improve it by reading ahead at least read region, and
3260 avoiding excessive large RPC for small reads.
3262 Severity : enhancement
3264 Description: Creating more than 1000 files for a single job may cause a load
3265 imbalance on the OSTs if there are also a large number of OSTs.
3266 Details : qos_prep_create() uses an OST index reseed value that is an
3267 even multiple of the number of available OSTs so that if the
3268 reseed happens in the middle of the object allocation it will
3269 still utilize the OSTs as uniformly as possible.
3274 Description: OST or MDS may oops in ping_evictor_main()
3275 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3276 but doesn't restart at beginning of obd_exports_timed list
3279 Severity : enhancement
3281 Description: improve by-nid export eviction on the MDS and OST
3282 Details : allow multiple exports with the same NID to be evicted at one
3283 time without re-searching the exports list.
3286 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3288 Description: MDS may oops in groups_free()
3289 Details : in rare race conditions a newly allocated group_info struct is
3290 freed again, and this can be NULL. The 2.4 compatibility code
3291 for groups_free() checked for a NULL pointer, but 2.6 did not.
3294 Frequency : common for liblustre clients doing little filesystem IO
3295 Bugzilla : 9352, 7313
3296 Description: server may evict liblustre clients accessing contended locks
3297 Details : if a client is granted a lock or receives a completion AST
3298 with a blocking AST already set it would not reply to the AST
3299 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3302 Frequency : lfs setstripe, only systems with more than 160 OSTs
3304 Description: unable to set striping with a starting offset beyond OST 160
3305 Details : llapi_create_file() incorrectly limited the starting stripe
3306 index to the maximum single-file stripe count.
3309 Frequency : LDAP users only
3311 Description: lconf did not handle in-kernel recovery with LDAP properly
3312 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3314 Severity : enhancement
3316 Description: bind OST threads to NUMA nodes to improve performance
3317 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3318 node and do their allocations there to localize memory access
3320 Severity : enhancement
3322 Description: llmount can determine client NID directly from Myrinet (GM)
3323 Details : the client NID code from gmnalnid was moved directly into
3324 llmount, removing the need to use this or specifying the
3325 client NID explicitly when mounting GM clients with zeroconf
3328 Frequency : if client is started with down MDS
3330 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3331 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3332 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3333 ptlrpc_import_delayed_req() because we didn't check if the
3334 request was interrupted, and we also didn't break out of the
3335 event loop if there was no timeout
3340 Description: data loss during non-page-aligned writes to a single file from
3341 both multiple nodes and multiple threads on one node at same time
3342 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3343 inconsistency led to false short-reads, that were cached and later
3344 used by ->prepare_write() to fill in partially written page,
3345 leading to data loss.
3348 Frequency : always, if lconf --abort_recovery used
3350 Description: lconf --abort_recovery fails with 'Operation not supported'
3351 Details : lconf was attempting to abort recovery on the MDT device and not
3354 Severity : enhancement
3356 Description: remove cleanup logs
3357 Details : replace lconf-generated cleanup logs with lustre internal
3358 cleanup routines. Eliminates the need for client-cleanup and
3361 Severity : enhancement
3363 Description: add support for EAs (user and system) on lustre filesystems
3364 Details : it is now possible to store extended attributes in the Lustre
3365 client filesystem, and with the user_xattr mount option it
3366 is possible to allow users to store EAs on their files also
3368 Severity : enhancement
3370 Description: Add possibility (config option) to show minimal available OST free
3372 Details : When compiled with --enable-mindf configure option, statfs(2)
3373 (and so, df) will return least minimal free space available from
3374 all OSTs as amount of free space on FS, instead of summary of
3375 free spaces of all OSTs.
3377 Severity : enhancement
3379 Description: do not expand extent locks acquired on OST-side
3380 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3381 by server: they are not cached anyway.
3384 Frequency : when mmap is used/binaries executed from Lustre
3386 Description: Unmmap pages before throwing them away from read cache.
3387 Details : llap_shrink cache now attempts to unmap pages before discarding
3388 them (if unmapping failed - do not discard). SLES9 kernel has
3389 extra checks that trigger if this unmapping is not done first.
3394 Description: lconf didn't resolve symlinks before checking to see whether a
3395 given mountpoint was already in use
3398 Frequency : when migrating failover services
3399 Bugzilla : 6395, 9514
3400 Description: When migrating a subset of services from a node (e.g. failback
3401 from a failover service node) the remaining services would
3402 time out and evict clients.
3403 Details : lconf --force (implied by --failover) sets the global obd_timeout
3404 to 5 seconds in order to quickly disconnect, but this caused
3405 other RPCs to time out too quickly. Do not change the global
3406 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3408 Severity : enhancement
3409 Frequency : if MDS is started with down OST
3410 Bugzilla : 9439,5706
3411 Description: Allow startup/shutdown of an MDS without depending on the
3412 availability of the OSTs.
3413 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3414 Add appropriate locking and lov-osc refcounts for safe
3415 cleaning. Add osc abort_inflight calls in case the
3416 synchronize never started.
3419 Frequency : occasional (Cray XT3 only)
3421 Description: root not authorized to access files in CRAY_PORTALS environment
3422 Details : The client process capabilities were not honoured on the MDS in
3423 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3424 been accessed by an authorized user then root was able to access
3425 the file on the local client also. The root user capabilities
3426 are now allowed on the MDS, as this environment has secure UID.
3429 Frequency : occasional
3431 Description: ldiskfs "too long searching" message happens too often
3432 Details : A debugging message (otherwise harmless) prints too often on
3433 the OST console. This has been reduced to only happen when
3434 there are fragmentation problems on the filesystem.
3439 Description: Division by zero in statfs when all OSCs are inactive
3440 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3441 lov_statfs divides by value returned by lov_get_stripecnt().
3445 Bugzilla : 9489, 3273
3446 Description: First write from each client to each OST was only 4kB in size,
3447 to initialize client writeback cache, which caused sub-optimal
3448 RPCs and poor layout on disk for the first writen file.
3449 Details : Clients now request an initial cache grant at (re)connect time
3450 and so that they can start streaming writes to the cache right
3451 away and always do full-sized RPCs if there is enough data.
3452 If the OST is rebooted the client also re-establishes its grant
3453 so that client cached writes will be honoured under the grant.
3458 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3459 Details : Now I/O RPCs go to different portal number and (presumably) fast
3460 lock requests (and glimses) and other RPCs get their own service
3461 threads pool that should be able to service those RPCs
3464 Severity : enhancement
3466 Description: Ability to exchange lustre version between client and servers and
3467 issue warnings at client side if client is too old. Also for
3468 liblustre clients there is ability to refuse connection of too old
3470 Details : New 'version' field is added to connect data structure that is
3471 filled with version info. That info is later checked by server and
3475 Frequency : rare, liblustre only.
3476 Bugzilla : 9296, 9581
3477 Description: Two simultaneous writes from liblustre at offset within same page
3478 might proceed at the same time overwriting eachother with stale
3480 Details : I/O lock withing llu_file_prwv was released too early, before data
3481 actually was hitting the wire. Extended lock-holding time until
3482 server acknowledges receiving data.
3485 Frequency : extremely rare. Never observed in practice.
3487 Description: avoid generating lustre_handle cookie of 0.
3488 Details : class_handle_hash() generates handle cookies by incrementing
3489 global counter, and can hit 0 occasionaly (this is unlikely, but
3490 not impossible, because initial value of cookie counter is
3491 selected randonly). Value of 0 is used as a sentinel meaning
3492 "unassigned handle" --- avoid it. Also coalesce two critical
3493 sections in this function into one.
3495 Severity : enhancement
3497 Description: allow liblustre clients to delegate truncate locking to OST
3498 Details : To avoid overhead of locking, liblustre client instructs OST to
3499 take extent lock in ost_punch() on client's behalf. New connection
3500 flag is added to handle backward compatibility.
3502 Severity : enhancement
3503 Bugzilla : 4928, 7341, 9758
3504 Description: allow number of OST service threads to be specified
3505 Details : a module parameter allows the number of OST service threads
3506 to be specified via "options ost ost_num_threads={N}" in the
3507 OSS's /etc/modules.conf or /etc/modprobe.conf.
3511 Bugzilla : 6146, 9635, 9895
3512 Description: servers crash with bad pointer in target_handle_connect()
3513 Details : In rare cases when a client is reconnecting it was possible that
3514 the connection request was the last reference for that export.
3515 We would temporarily drop the export reference and get a new
3516 one, but this may have been the last reference and the export
3517 was just destroyed. Get new reference before dropping old one.
3519 Severity : enhancement
3520 Frequency : if client is started with failover MDS
3522 Description: Allow multiple MDS hostnames in the mount command
3523 Details : Try to read the configuration from all specified MDS
3524 hostnames during a client mount in case the "primary"
3527 Severity : enhancement
3529 Description: Stop sending data to evicted clients as soon as possible.
3530 Details : Check if the client we are about to send or are sending data to
3531 was evicted already. (Check is done every second of waiting,
3532 for which l_wait_event interface was extended to allow checking
3533 of exit condition at specified intervals).
3536 Frequency : rare, normally only when NFS exporting is done from client
3538 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3540 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3543 Frequency : always, if srand() is called before liblustre initialization
3545 Description: Liblustre uses system PRNG disturbing its usage by user application
3546 Details : Introduce internal to lustre fast and high-quality PRNG for
3547 lustre usage and make liblustre and some other places in generic
3548 lustre code to use it.
3550 Severity : enhancement
3551 Bugzilla : 9477, 9557, 9870
3552 Description: Verify that the MDS configuration logs are updated when xml is
3553 Details : Check if the .xml configuration logs are newer than the config
3554 logs stored on the MDS and report an error if this is the case.
3555 Request --write-conf, or allow starting with --old_conf.
3557 Severity : enhancement
3559 Description: Handle symlinks in the path when checking if Lustre is mounted.
3560 Details : Resolve intermediate symlinks when checking if a client has
3561 mounted a filesystem to avoid duplicate client mounts.
3566 Description: lconf can hit an error exception but still return success.
3567 Details : The lconf command catches the Command error exception at the top
3568 level script context and will exit with the associated exit
3569 status, but doesn't ensure that this exit status is non-zero.
3574 Description: failure of ptlrpc thread startup can cause oops
3575 Details : Starting a ptlrpc service thread can fail if there are a large
3576 number of threads or the server memory is very fragmented.
3577 Handle this without oopsing.
3580 Frequency : always, only if liblustre and non-default acceptor port was used
3582 Description: liblustre cannot connect to servers with non-default acceptor port
3583 Details : tcpnal_set_default_params() was not called and was therefore
3584 ignoring the environment varaible TCPNAL_PORT, as well as other
3585 TCPNAL_ environment variables
3590 Description: two objects could be created on the same OST for a single file
3591 Details : If an OST is down, in some cases it was possible to create two
3592 objects on a single OST for a single file. No problems other
3593 than potential performance impact and spurious error messages.
3597 Bugzilla : 5681, 9562
3598 Description: Client may oops in ll_unhash_aliases
3599 Details : Client dcache may become inconsistent in race condition.
3600 In some cases "getcwd" can fail if the current directory is
3606 Description: Inode refcounting problems in NFS export code
3607 Details : link_raw functions used to call d_instantiate without obtaining
3608 extra inode reference first.
3612 Bugzilla : 9942, 9903
3613 Description: Referencing freed requests leading to crash, memleaks with NFS.
3614 Details : We used to require that call to ll_revalidate_it was always
3615 followed by ll_lookup_it. Also with revalidate_special() it is
3616 possible to call ll_revalidate_it() twice for the same dentry
3617 even if first occurence returned success. This fix changes semantic
3618 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3619 reference on a request referred from the intent.
3620 ll_intent_release() then releases such a request.
3623 Frequency : rare, normally benchmark loads only
3625 Description: unlinked inodes were kept in memory on the client
3626 Details : If a client is repeatedly creating and unlinking files it
3627 can accumulate a lot of stale inodes in the inode slab cache.
3628 If there is no other client load running this can cause the
3629 client node to run out of memory. Instead flush old inodes
3630 from client cache that have the same inode number as a new inode.
3633 Frequency : SLES9 2.6.5 kernel and long filenames only
3634 Bugzilla : 9969, 10379
3635 Description: utime reports stale NFS file handle
3636 Details : SLES9 uses out-of-dentry names in some cases, which confused
3637 the lustre dentry revalidation. Change it to always use the
3641 Frequency : rare, unless heavy write-truncate concurrency is continuous
3642 Bugzilla : 4180, 6984, 7171, 9963, 9331
3643 Description: OST becomes very slow and/or deadlocked during object unlink
3644 Details : filter_destroy() was holding onto the parent directory lock
3645 while truncating+unlinking objects. For very large objects this
3646 may block other threads for a long time and slow overall OST
3647 responsiveness. It may also be possible to get a lock ordering
3648 deadlock in this case, or run out of journal credits because of
3649 the combined truncate+unlink. Solution is to do object truncate
3650 first in one transaction without parent lock, and then do the
3651 final unlink in a new transaction with the parent lock. This
3652 reduces the lock hold time dramatically.
3655 Frequency : rare, 2.4 kernels only
3657 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3658 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3659 before unmounting their filesystem in order to avoid deadlock.
3660 The kernel_locked() function in 2.4 kernels only checks whether
3661 the kernel lock is held, not whether it is this process that is
3662 holding it as 2.6 kernels do.
3667 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3668 Details : The client ptlrpc code may be trying to reconnect to a down
3669 server before a previous connection attempt has timed out.
3670 Increase the reconnect interval to be longer than the connection
3671 timeout interval to avoid sending duplicate connections to
3675 Frequency : echo_client brw_test command
3677 Description: fix echo_client to work with OST preallocated code
3678 Details : OST preallocation code (5137) didn't take echo_client IO path
3679 into account: echo_client calls filter methods outside of any
3680 OST thread and, hence, there is no per-thread preallocated
3681 pages and buffers to use. Solution: hijack pga pages for IO. As
3682 a byproduct, this avoids unnecessary data copying.
3686 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3687 Description: Client can oops in mdc_commit_close() after open replay
3688 Details : It was possible for the MDS to return an open request with no
3689 transaction number in mds_finish_transno() if the client was
3690 evicted, but without actually returning an error. Clients
3691 would later try to replay that open and may trip an assertion
3692 Simplify the client close codepath, and always return an error
3693 from the MDS in case the open is not successful.
3696 Frequency : rare, 2.6 OSTs only
3698 Description: OST may deadlock under high load on fragmented files
3699 Details : If there was a heavy load and highly-fragmented OST filesystems
3700 it was possible to have all the OST threads deadlock waiting on
3701 allocation of biovecs, because the biovecs were not released
3702 until the entire RPC IO was completed. Instead, release biovecs
3703 as soon as they are complete to ensure forward IO progress.
3705 Severity : enhancement
3707 Description: Support for specifying external journal device at mount
3708 Details : If an OST or MDS device is formatted with an external journal
3709 device, this device major/minor is stored in the ext3 superblock
3710 and may not be valid for failover. Allow detecting and
3711 specifying the external journal at mount time.
3716 Description: Mounting an MDS with pending unlinked files may cause oops
3717 Details : target_finish_recovery() calls mds_postrecov() which returned
3718 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3719 considers this an error and immediately begins cleaning up the
3720 lov, just after starting the mds_lov process
3722 Severity : enhancement
3724 Description: Implement 'lfs df' to report actual free space on per-OST basis
3725 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3726 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3727 usage of objects; '-h' to report in human readable format.
3729 ------------------------------------------------------------------------------
3731 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3738 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3739 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3740 at startup during orphan unlinking. Instead, skip buddy update
3741 before setup, as it will be regenerated after recovery is complete.
3746 Description: If an OST is inactive, its locks might reference stale inodes.
3747 Details : lov_change_cbdata() must iterate over all namespaces, even if
3748 they are inactive to clear inode references from the lock.
3750 Severity : enhancement
3751 Frequency : occasional, if non-standard max_dirty_mb used
3753 Description: Client will block write RPCs if not enough grant
3754 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3755 then the client will block writes while waiting for another RPC
3756 to complete instead of consuming its dirty limit. With change
3757 we get improved performance when max_dirty_mb is small.
3759 Severity : enhancement
3760 Bugzilla : 3389, 6253
3761 Description: Add support for supplementary groups on the MDS.
3762 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3763 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3764 lookups for user supplementary groups into a cache.
3768 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3769 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3770 for liblustre O_EXCL is enforced by MDS.
3774 Description: readdir never returns NULL in liblustre.
3775 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3776 dentry in struct dirent.
3780 Frequency : liblustre only, depends on application IO pattern
3781 Description: liblustre clients evicted if not contacting servers
3782 Details : Don't put liblustre clients into the ping_evictor list, so
3783 they will not be evicted by the pinger ever.
3785 Severity : enhancement
3787 Description: Add ability to evict clients by NID from MDS.
3788 Details : By echoing "nid:$NID" string into
3789 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3790 $NID would be instantly evicted from this MDS and from all active
3791 OSTs connected to it.
3795 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3796 Details : lookup_it_finish() used to query file size from OSTs that was not
3801 Description: service threads change working directory to that of init
3802 Details : Starting lustre service threads may pin the working directory
3803 of the parent thread, making that filesystem busy. Threads
3804 now change to the working directory of init to avoid this.
3808 Frequency : during shutdown only
3809 Description: shutdown with a failed MDS or OST can cause unmount to hang
3810 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3815 Frequency : 2.6 only
3816 Description: chmod/chown may include an extra supplementary group
3817 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3818 supplementary group and if none is specified this is used.
3821 Bugzilla : 5479 (6816)
3823 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3824 Details : If lookup is in progress on a file that is unlinked we might try
3825 to revalidate the inode and fail in revalidate after lookup is
3826 complete and ll_file_open() enqueues the open again but
3827 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3830 Frequency : always, if lconf --abort_recovery used
3832 Description: lconf --abort_recovery fails with 'Operation not supported'
3833 Details : lconf was attempting to abort recovery on the MDT device and not
3836 ------------------------------------------------------------------------------
3838 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3843 Frequency : rare (only unsupported configurations with a node running as an
3845 Bugzilla : 6514, 5137
3846 Description: Mounting a Lustre file system on a node running as an OST could
3848 Details : OSTs now preallocates memory needed to write out data at
3849 startup, instead of when needed, to avoid having to
3850 allocate memory in possibly low memory situations.
3851 Specifically, if the file system is mounted on on OST,
3852 memory pressure could force it to try to write out data,
3853 which it needed to allocate memory to do. Due to the low
3854 memory, it would be unable to do so and the node would
3855 become unresponsive.
3857 Severity : enhancement
3859 Description: Addition of lconf --service command line option
3860 Details : lconf now accepts a '--service <arg>' option, which is
3861 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3863 Severity : enhancement
3865 Description: Failover mode is now the default for OSTs.
3866 Details : By default, OSTs will now run in failover mode. To return to
3867 the old behaviour, add '--failout' to the lmc line for OSTs.
3869 Severity : enhancement
3871 Description: Health checks are now provided for MDS and OSTs
3872 Details : Additional detailed health check information on MSD and OSTs
3873 is now provided through the procfs health_check value.
3876 Frequency : occasional, depends on IO load
3878 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3879 after numerous create/delete cycles
3880 Details : The ext3 inode allocation policy would not allocate new inodes
3881 very well on the OSTs because there are no new directories
3882 being created. Instead we look for groups with free space if
3883 the parent directories are nearly full.
3888 Description: Network or server problems during mount may cause partially
3889 mounted clients instead of returning an error.
3890 Details : The config llog parsing code may overwrite the error return
3891 code during mount error handling, returning success instead
3897 Description: MDS can fail to allocate large reply buffers
3898 Details : After long uptimes the MDS can fail to allocate large reply
3899 buffers (e.g. zconf client mount config records) due to memory
3900 fragmentation or consumption by the buffer cache. Preallocate
3901 some large reply buffers so that these replies can be sent even
3902 under memory pressure.
3906 Frequency : rare (liblustre)
3907 Description: fsx running with liblustre complained that using truncate() to
3908 extend the file doesn't work. This patch corrects that issue.
3909 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3910 ATTR_SIZE and lsm use in llu_setattr_raw.
3914 Frequency : rare, only 2.6 kernels
3915 Description: Unusual file access patterns on the MDS may result in inode
3916 data being lost in very rare circumstances.
3917 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3918 code in the 2.6 kernel caused the inode and/or EA data not to
3919 be read from disk, causing single-file corruption.
3923 Frequency : rare, only 2.6 filesystems using extents
3924 Description: Heavy concurrent write and delete load may cause data corruption.
3925 Details : It was possible under high-load situations to have an extent
3926 metadata block in the block device cache from a just-unlinked
3927 file overwrite a newly-allocated data block. We now unmap any
3928 metadata buffers that alias just-allocated data blocks.
3932 Frequency : filesystems with default stripe_count larger than 77
3933 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3934 Details : lconf specifies an inode size of 4096 bytes when the default
3935 stripe_count is larger than 77. This conflicts with the default
3936 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3937 this case to avoid pinning too much memory for large EAs.
3939 ------------------------------------------------------------------------------
3941 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3946 Frequency : rare (extremely heavy IO load with hundreds of clients)
3948 Description: Client is evicted, gets IO error writing to file
3949 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3950 caused clients to be evicted for AST timeouts. The fixes in
3951 bug 5192 mean we no longer need to have such short AST timeouts
3952 so ldlm_timeout has been increased.
3955 Frequency : occasional during --force or --failover shutdown under load
3956 Bugzilla : 5949, 4834
3957 Description: Server oops/LBUG if stopped with --force or --failover under load
3958 Details : a collection of import/export refcount and cleanup ordering
3959 issues fixed for safer force cleanup
3962 Frequency : only filesystems larger than 120 OSTs
3963 Bugzilla : 5990, 6223
3964 Description: lfs getstripe would oops on a very large filesystem
3965 Details : lov_getconfig used kfree on vmalloc'd memory
3968 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3969 Bugzilla : 6242, 6243
3970 Description: reading from files that had been truncated to a non-zero size
3971 but never opened returned no data
3972 Details : ll_file_read() reads zeros from no-object files to EOF
3977 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3978 Details : There's little harm in aborting MDS/OSS recovery and letting it
3979 try again, so I removed the LASSERT and return an error instead.
3981 Severity : enhancement
3983 Description: New debugging infrastructure for tracking down data corruption
3984 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3985 (b) cover more of the client-side code path, and (c) try to narrow
3986 down where problems occurred
3990 Bugzilla : 3819, 4364, 4397, 6313
3991 Description: Racing close and eviction MDS could cause assertion in mds_close
3992 Details : It was possible to get multiple mfd references during close and
3993 client eviction, leading to one thread referencing a freed mfd.
3995 Severity: : enhancement
3996 Bugzilla : 3262, 6359
3997 Description: Attempts to reconnect to servers are now more aggressive.
3998 Details : This builds on the enhanced upcall-less recovery that was added
3999 in 1.4.2. When trying to reconnect to servers, clients will
4000 now try each server in the failover group every 10 seconds. By
4001 default, clients would previously try one server every 25 seconds.
4006 Description: After recovery, certain operations trigger a failed
4007 assertion on a client.
4008 Details : Failing over an mds, using lconf -d --failover, while a
4009 client was doing a readdir() call would cause the client to
4010 LBUG after recovery completed and the readdir() was resent.
4012 Severity : enhancement
4014 Description: Default groups are now added by lconf
4015 Details : You can now run lconf --group <servicename> without having to
4016 manually add groups with lmc.
4019 Frequency : occasional
4021 Description: Nodes with an elan id of 0 trigger a failed assertion
4024 Frequency : always when accessing e.g. tty/console device nodes
4026 Description: tty and some other devices nodes cannot be used on lustre
4027 Details : file's private_data field is used by device data and lustre
4028 values in there got lost. New field was added to struct file to
4029 store fs-specific private data.
4032 Frequency : when exporting Lustre via NFS
4034 Description: NFSD failed occasionally when looking up a path component
4035 Details : NFSD is looking up ".." which was broken in ext3 directories
4036 that had grown large enough to become hashed.
4039 Frequency : Clusters with multiple interfaces not on the same subnet
4041 Description: Nodes will repeatedly try to reconnect to an interface which it
4042 cannot reach and report an error to the log.
4043 Details : Extra peer list entries will be created by lconf with some peers
4044 unreachable. lconf now validates the peer before adding it.
4047 Frequency : Only if a default stripe is set on the filesystem root.
4049 Description: Setting a default stripe on the filesystem root prevented the
4050 filesystem from being remounted.
4051 Details : The client was sending extra request flags in the root getattr
4052 request and did not allocate a reply buffer for the dir EA.
4055 Frequency : occasional, higher if lots of files are accessed by one client
4056 Bugzilla : 6159, 6097
4057 Description: Client trips assertion regarding lsm mismatch/magic
4058 Details : While revalidating inodes the VFS looks up inodes with ifind()
4059 and in rare cases can find an inode that is being freed.
4060 The ll_test_inode() code will free the lsm during ifind()
4061 when it finds an existing inode and then the VFS later attaches
4062 this free lsm to a new inode.
4066 Bugzilla : 6422, 7030
4067 Description: MDS deadlock between mkdir and client eviction
4068 Details : Creating a new file via mkdir or mknod (starting a transaction
4069 and getting the ns lock) can deadlock with client eviction
4070 (gets ns lock and trying to finish a synchronous transaction).
4073 Frequency : occasional
4074 Description: While starting a server, the fsfilt_ext3 module could not be
4076 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
4077 kernels. Previously, lconf would still use the ext3 name
4078 when trying to load modules. Now, it will correctly use
4079 ext3 on 2.4 and ldiskfs on 2.6.
4081 Severity : enhancement
4082 Description: The default stripe count has been changed to 1
4083 Details : The interpretation of the default stripe count (0, to lfs
4084 or lmc) has been changed to mean striping across a single
4085 OST, rather than all available. For general usage we have
4086 found a stripe count of 1 or 2 works best.
4088 Severity : enhancement
4089 Description: Add support for compiling against Cray portals.
4090 Details : Conditional compiling for some areas that are different
4094 Frequency : occasional
4095 Bugzilla : 6409, 6834
4096 Description: Creating files with an explicit stripe count may lead to
4097 a failed assertion on the MDS
4098 Details : If some OSTs are full or unavailable, creating files may
4099 trigger a failed assertion on the MDS. Now, Lustre will
4100 try to use other servers or return an error to the
4104 Frequency : occasional
4106 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4107 caused lock timeout during readahead (since 1.4.2).
4108 Details : Processes doing readahead might match a lock that hasn't been
4109 granted yet if there are overlapping and conflicting lock
4110 requests. The readahead process waits on ungranted lock
4111 (original lock is CBPENDING), while OST waits for that process
4112 to cancel CBPENDING read lock and eventually evicts client.
4114 Severity : enhancement
4116 Description: Initial enabling of flock support for clients
4117 Details : Implements fcntl advisory locking and file status functions.
4118 This feature is provided as an optional mount flag (default
4119 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4120 locking are implemented yet, and those that are are not guaranteed
4121 to be completely correct in production environments.
4122 mount -t lustre -o [flock|noflock] ...
4125 Frequency : occasional
4127 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4128 assertion in the ext3 JBD (journaling) layer.
4129 Details : The b_committed_data struct is protected by the big kernel lock
4130 in 2.4 kernels, serializing journal_commit_transaction() and
4131 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4132 access to this struct. In 2.6 kernels there is finer grained
4133 locking to improve SMP performance of the JBD layer.
4137 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4142 Description: merge and cleanup kernel patches.
4145 -----------------------------------------------------------------------------
4147 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4149 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4150 versions of Lustre. Please update all servers and clients to
4151 version 1.4.2 or later at the same time. You must also run
4152 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4153 to update the configuration logs.
4155 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4156 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4157 - swab llog records of type '0' so we get proper header size/idx (5861)
4158 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4159 - fix rename of one directory over another leaking an inode (5953)
4160 - avoid SetPageDirty on 2.6 (5981)
4161 - don't re-add just-being-destroyed locks to the waiting list (5653)
4162 - when creating new directories, inherit the parent's custom
4163 striping settings if present parent (3048)
4164 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4165 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4166 - don't hold client locks on temporary worklist from l_lru (5666)
4167 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4168 - automatically evict dead clients (5921)
4169 - Update file size properly in create+truncate+fstat case (6196)
4170 - Do not unhash mountpoint dentries, do not allow removal of
4172 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4173 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4174 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4175 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4176 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4177 - Correctly handle reads of files with no objects (6243)
4178 - lctl recover will also mark a device active if deactivate used (5933)
4180 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
4181 - allow --write-conf on an MDS with different nettype than client (5619)
4182 - don't write config llogs to MDS for mounts not from that MDS (5617)
4183 - lconf should create multiple TCP connections from a client (5201)
4184 - init scripts are now turned off by default; run chkconfig --on
4185 lustre and chkconfig --on lustrefs to use them
4186 - upcalls are no longer needed for clients to recover to failover
4188 - add --abort-recovery option to lconf to abort recovery on device
4190 - add support for an arbitrary number of OSTs (3026)
4191 - Quota support protocol changes.
4192 - forward compatibility changes to wire structs (6007)
4193 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
4194 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
4195 - improved "lustre status" script
4196 - initialize blocksize for non-regular files (6062)
4197 - added --disable-server and --disable-client configure options (5782)
4198 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
4199 - Vanilla 2.4.29 support
4200 - increase maximum number of obd devices to 520 (6242)
4201 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4202 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4203 SLES 9 2.6.5 kernels
4204 - limit stripes per file to 160 (the maximum EA size) (6093)
4206 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4209 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4210 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4211 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4212 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4213 - small ext3 extents cleanups and fixes (5733)
4214 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4215 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4216 - avoid needless client->OST connect, fix handle mismatch (5317)
4217 - fix DLM error path that led to out-of-sync client, long delays (5779)
4218 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4219 - fix several locking issues related to i_size (5492,5624,5654,5672)
4220 - don't move pending lock onto export if it is already evicted (5683)
4221 - fix kernel oops when creating .foo in unlinked directory (5548)
4222 - fix deadlock in obdfilter statistics vs. object create (5811)
4223 - use time_{before,after} to avoid timer jiffies wrap (5882)
4224 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4225 - Do not leak request if server was not able to process it (5154)
4226 - If mds_open unable to find parent dir, make that negative lookup(5154)
4227 - don't create new directories with extent-mapping (5909, 5936)
4229 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
4230 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
4231 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
4232 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
4233 - support for RHEL4 kernel on i686 (5773)
4234 - provide error messages when incompatible logs are encountered (5898)
4236 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
4237 * version 1.4.0.10 (1.4.1 release candidate 1)
4239 - don't keep a lock reference when lock is not granted (4238)
4240 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4241 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
4242 - drop import inflight refcount on signal_completed_replay error (5255)
4243 - unlock page after async write error during send (3677)
4244 - handle missing objects in filter_preprw_read properly (5265)
4245 - no transno return for symlink open, don't save no-trasno open (3440)
4246 - don't try to complete elan receive that already failed (4012)
4247 - free RPC server reply state on error (5406)
4248 - clean up thread from ptlrpc_start_thread() on error (5160)
4249 - readahead could read extra page into cache that wasn't ejected (5388)
4250 - prevent races in class_attach/setup/cleanup/detach (5260)
4251 - don't dereference de->d_inode after l_dput of de (5458)
4252 - use "int" for stripe value returned from lock_to_stripe (5544)
4253 - mballoc allocation and error-checking fixes in 2.6 (5504)
4254 - block device patches to fix I/O request sizes in 2.6 (5482)
4255 - look up hostnames for IB nals (5602)
4256 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
4257 - don't start multiple acceptors for the same port (5277)
4258 - fix incorrect LASSERT in mds_getattr_name (5635)
4259 - export a proc file for general "ping" checking (5628)
4260 - fix "lfs check" to not block when the MDS is down (5628)
4262 - service request history (4965)
4263 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4264 - create an "evict_client" /proc entry on OSTs, like the MDS has
4265 - fix mount usage message, return errors per mount(8) (5168)
4266 - change grep [] to grep "[]" in tests so they work in more UMLs
4267 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4268 - remove ancient LOV_MAGIC_V0 EA support (5047)
4269 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4270 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4271 - allow readahead allocations to fail when low on memory (5383)
4272 - mmap locking landed again, after considerable improvement (2828)
4273 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4275 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4278 - send OST transaction number in read/write reply to free req (4966)
4279 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4280 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4281 - the watchdog thread now runs as interruptible (5246)
4282 - flock/lockf fixes (but it's still disabled, pending 5135)
4283 - don't use EXT3 constants in llite code (5094)
4284 - memory shortage at startup could cause assertion (5176)
4286 - reorganization of lov code
4287 - single portals codebase
4289 - add extents/mballoc support (5025)
4290 - direct I/O reads in the obdfilter (4048)
4291 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4293 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4296 - send OST transaction number in read/write reply to free req (4966)
4297 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4298 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4299 - the watchdog thread now runs as interruptible (5246)
4300 - handle missing objects in filter_preprw_read properly (5265)
4301 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4302 - drop import inflight refcount on signal_completed_replay error (5255)
4303 - unlock page after async write error during send (3677)
4304 - return original error code on reconstructed replies (3761)
4305 - no transno return for symlink open, don't save no-trasno open (3440)
4307 - add pid to ldlm debugging output (4922)
4308 - bump the watchdog timeouts -- we can't handle 30sec yet
4309 - extra debugging for orphan dentry/inode bug (5259)
4311 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
4314 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4315 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4316 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
4317 - fix NULL dereference in /proc/sys/portals/routes (4827)
4318 - allow failed mdc_close() operations to be interrupted (4561)
4319 - stop precreate on OST before MDS would time out on it (4778)
4320 - don't send partial-page writes before EOF from client (4410)
4321 - discard client grant for sub-page writes on large-page clients (4520)
4322 - don't free dentries not owned by NFS code, check generation (4806)
4323 - fix lsm leak if mds_create_objects() fails (4801)
4324 - limit debug_daemon file size, always print CERROR messages (4789)
4325 - use transno after validating reply (3892)
4326 - process timed out requests if import state changes (3754)
4327 - update mtime on OST during writes, return in glimpse (4829)
4328 - add mkfsoptions to LDAP (4679)
4329 - use ->max_readahead method instead of zapping global ra (5039)
4330 - don't interrupt __l_wait_event() during strace
4332 - add software watchdogs to catch hung threads quickly (4941)
4333 - make lustrefs init script start after nfs is mounted
4334 - fix CWARN/ERROR duplication (4930)
4335 - return async write errors to application if possible (2248)
4336 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4337 - print NAL number in %x format (4645)
4338 - update barely-supported suse-2.4.21-171 series (4842)
4339 - support for sles 9 %post scripts
4340 - support for building 2.6 kernel-source packages
4341 - support for sles km_* packages
4343 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4346 - ignore -ENOENT errors in osc_destroy (3639)
4347 - notify osc create thread that OSC is being cleaned up (4600)
4348 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4349 - reconstruct ost_handle() like mds_handle() (4657)
4350 - create a new thread to do import eviction to avoid deadlock (3969)
4351 - let lconf resolve symlinked-to devices (4629)
4352 - don't unlink "objects" from directory with default EA (4554)
4353 - hold socknal file ref over connect in case target is down (4394)
4354 - allow more than 32000 subdirectories in a single directory (3244)
4355 - fix blocks count for O_DIRECT writes (3751)
4356 - OST returns ENOSPC from object create when no space left (4539)
4357 - don't send truncate RPC if file size isn't changing (4410)
4358 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4359 - bind to privileged port in socknal and tcpnal (3689)
4361 - rate limit CERROR/CWARN console message to avoid overload (4519)
4362 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4363 - basic mmap support (3918)
4364 - kernel patch series update from b1_4 (4711)
4366 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4369 - avoid crash during MDS cleanup with OST shut down (2775)
4370 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4371 - don't use bad inodes on the MDS (3744)
4372 - dynamic object preallocation to improve recovery speed (4236)
4373 - don't hold spinlock over lock dumping or change debug flags (4401)
4374 - don't zero obd_dev when it is force cleaned (3651)
4375 - print grants to console if they go negative (4431)
4376 - "lctl deactivate" will stop automatic recovery attempts (3406)
4377 - look for existing locks in ldlm_handle_enqueue() (3764)
4378 - don't resolve lock handle twice in recovery avoiding race (4401)
4379 - revalidate should check working dir is a directory (4134)
4381 - don't always mark "slow" obdfilter messages as errors (4418)
4383 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4386 - don't close LustreDB during write_conf until it is done (3860)
4387 - fix typo in lconf for_each_profile (3821)
4388 - allow dumping logs from multiple threads at one time (3820)
4389 - don't allow multiple threads in OSC recovery (3812)
4390 - fix debug_size parameters (3864)
4391 - fix mds_postrecov to initialize import for llog ctxt (3121)
4392 - replace config semaphore with spinlock (3306)
4393 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4394 - don't allow enqueue to complete on a destroyed export (3822)
4395 - down write_lock before checking llog header bitmap (3825)
4396 - recover from lock replay timeout (3764)
4397 - up llog sem before sending rpc (3652)
4398 - reduce ns lock hold times when setting kms (3267)
4399 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4400 - fix NULL deref and obd_dev leak on setup error (3312)
4401 - replace some LBUG about llog ops with error handling (3841)
4402 - don't match INVALID dentries from d_lookup and spin (3784)
4403 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4404 - fix invalid assertion in ptlrpc_set_wait (3880)
4406 - add libwrap support for the TCP acceptor (3996)
4407 - add /proc/sys/portals/routes for non-root route listing (3994)
4408 - allow setting MDS UUID in .xml (2580)
4409 - print the stack of a process that LBUGs (4228)
4411 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4414 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4415 - make sure to unset replay flag from failed open requests (3440)
4416 - if default stripe count is 0, use OST count for inode size (3636)
4417 - update parent mtime/ctime on client for create/unlink (2611)
4418 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4419 - free recovery state on server during a forced cleanup (3571)
4420 - unregister_reply for resent reqs (3063)
4421 - loop back devices mounting and status check on 2.6 (3563)
4422 - fix resource-creation race that can provoke i_size == 0 (3513)
4423 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4424 - more debugging for page-accounting assertion (3746)
4425 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4426 - avoid deadlock after precreation failure (3758)
4427 - fix race and lock order deadlock in orphan handling (3450, 3750)
4428 - add validity checks when grabbing inodes from l_ast_data (3599)
4430 - add /proc/.../recovery_status to obdfilter (3428)
4431 - lightweight CDEBUG infrastructure, debug daemon (3668)
4432 - change default OSC RPC parameters to be better on small clusters
4433 - turn off OST read cache for files smaller than 32MB
4434 - install man pages and include them in rpms (3100)
4435 - add new init script for (un)mounting lustre filesystems (2593)
4436 - run chkconfig in %post for init scripts (3701)
4437 - drop scimac NAL (unmaintained)
4439 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4442 - clean kiobufs before and after use (3485)
4443 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4444 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4445 - change init script to fail more clearly if not run as root (1528)
4446 - allow clients to reconnect during replay (1742)
4447 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4448 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4449 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4450 - fix cond_resched (3554)
4451 - only evict unfinished clients after recovery (3515)
4452 - allow bulk resend, prevent data loss (3570)
4453 - dynamic ptlrpc request buffer allocation (2102)
4454 - don't allow unlinking open directory if it isn't empty (2904)
4455 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4456 - remove extraneous obd dereference causing LASSERT failure (3334)
4457 - don't use get_cycles() when creating temp. files on the mds (3156)
4458 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4459 - handle EEXIST for set-stripe, set proper directory name (3336)
4461 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4462 - fix ksocknal_fmb_callback() error messages (2918)
4464 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4467 - don't copy lvb into (possibly NULL) reply on error (2983)
4468 - don't deref dentry after dput, don't free lvb on error (2922)
4469 - use the kms to determine writeback rpc length (2947)
4470 - increment oti_logcookies when osc is inactive (2948)
4471 - update client's i_blocks count via lvb messages (2543)
4472 - handle intent open/close of special files properly (1557)
4473 - mount MDS with errors=remount-ro, like obdfilter (2009)
4474 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4475 - don't use cancelling-locks' kms values (2947)
4476 - use highest lock extent for kms, not last one (2925)
4477 - don't dereference ERR_PTR() dentry in error handling path (3107)
4478 - fix thread race in portals_debug_dumplog() (3122)
4479 - create lprocfs device entries at setup instead of at attach (1519)
4480 - common AST error handler, don't evict client on completion race (3145)
4481 - zero nameidata in detach_mnt in 2.6 (3118)
4482 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4483 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4484 - initialize RPC timeout timer earlier for 2.6 (3219)
4485 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4486 - print nal/nid for unknown nid (3258)
4487 - additional checks for oscc recovery before doing precreate (3284)
4488 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4489 - don't crash in mdc_close for bad permissions on open (3285)
4490 - zero i_rdev for non-device files (3147)
4491 - clear page->private before handing to FS, better assertion (3119)
4492 - tune the read pipeline (3236)
4493 - fix incorrect decref of invalidated dentry (2350)
4494 - provide read-ahead stats and refine rpc in flight stats (3328)
4495 - don't hold journal transaction open across create RPC (3313)
4496 - update atime on MDS at close time (3265)
4497 - close LDAP connection when recovering to avoid server load (3315)
4498 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4499 - don't leak open file on MDS after open resend (3325)
4500 - serialize filter_precreate and filter_destroy_precreated (3329)
4501 - loop device shouldn't call sync_dev() for nul device (3092)
4502 - clear page cache after eviction (2766)
4503 - resynchronize MDS->OST in background (2824)
4504 - refuse to mount the same filesystem twice on same mountpoint (3394)
4505 - allow llmount to create routes for mounting behind routers (3320)
4506 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4507 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4508 - fix rare problem with rename on htree directories (3417)
4510 - allow default OST striping configuration per directory (1414)
4511 - fix compilation for qswnal for 2.6 kernels (3125)
4512 - increase maximum number of MDS request buffers for large systems
4513 - change liblustreapi to be useful for external progs like lfsck (3098)
4514 - increase local configuration timeout for slow disks (3353)
4515 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4517 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4520 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4521 - don't overwrite extent policy data in reply if lock was blocked (2901)
4522 - drop filter export grants atomically with removal from device (2663)
4523 - del obd_self_export from work_list in class_disconnect_exports (2908)
4524 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4525 - swab reply message in mdc_close, other PPC fixes (2464)
4526 - fix destroying of named logs (2325)
4527 - overwrite old logs when running lconf --write_conf (2264)
4528 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4529 - fix race in target_handle_connect (2898)
4530 - mds_reint_create() should take same inode create lock (2926)
4531 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4532 - don't close files for self_export to avoid uninitialized obd (2936)
4533 - allow MDS with the same name as client node (2939)
4534 - hold dentry reference for closed log files for unlink (2325)
4535 - reserve space for all logs during transactions (2059)
4536 - don't evict page beyond end of stripe extent (2925)
4537 - don't oops on a deleted current working directory (2399)
4538 - handle hard links to targets without a parent properly (2517)
4539 - don't dereference NULL lock when racing during eviction (2867)
4540 - don't grow lock extents when lots of conflicting locks (2919)
4542 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4545 - account for cache space usage on clients to avoid data loss (974)
4546 - lfsck support in lustre kernel code (2349)
4547 - reduce journal credits needed for BRW writes (2370)
4548 - orphan handling to avoid losing space on client/server crashes
4549 - ptlrpcd can be blocked, stopping ALL progress (2477)
4550 - use lock value blocks to assist in proper KMS, faster stat (1021)
4551 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4552 - recovery for initial connections (2355)
4553 - fixes for mds_cleanup_orphans (1934)
4554 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4555 - block all file creations until orphan recovery completes (1901)
4556 - client remove rq_connection from request struct (2423)
4557 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4558 - recovery timer race (2670)
4559 - mdc_close recovey bug (2532)
4560 - ptlrpc cleanup bug (2710)
4561 - mds timeout on local locks (2588)
4562 - namespace lock held during RPCs (2431)
4563 - handle interrupted sync write properly (2503)
4564 - don't try to handle a message that hasn't been replied to (2699)
4565 - client assert failure during cleanup after abort recovery (2701)
4566 - leak mdc device after failed mount (2712)
4567 - ptlrpc_check_set allows timedout requests to complete (2714)
4568 - wait for inflight reqs when ptlrpcd finishes (2710)
4569 - make sure unregistered services are removed from the srv_list
4570 - reset bulk XID's when resending them (caught by 1138 test)
4571 - unregister_bulk after timeout
4572 - fix lconf error (2694)
4573 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4574 - readahead locks pages, leaves pending causing memory pressure (2673)
4575 - increase OST request buffers to 4096 on large machines (2729)
4576 - fix up permission of existing directories in simple_mkdir (2661)
4577 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4578 - don't assign transno to errored transactions (2742)
4579 - don't delete objects on OST if given a bogus objid from MDS (2751)
4580 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4581 - if rq_no_resend, then timeout request after recovery (2432)
4582 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4583 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4584 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4585 - MDS recovery shouldn't delete objects that it has given out (2730)
4586 - if enqueue arrives after completion, don't clobber LVB (2819)
4587 - don't unlock pages twice when trigger_group_io returns error (2814)
4588 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4589 - don't write pages to disk if there was an error (1450)
4590 - don't ping imports that have recovery disabled (2676)
4591 - take buffered bytes into account when balancing socknal conn (2817)
4592 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4593 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4594 - remove little-endian swabbing of llog records (1987)
4595 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4596 - retry reposting request buffers if they fail (1191)
4597 - grow extent at grant time to avoid granting a revoked lock (2809)
4598 - lock revoke doesn't evict page if covered by a second lock (2765)
4599 - disable VM readahead to avoid reading outside lock extents (2805)
4601 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4602 - updated kernel patches for hp-2.4.20 kernel (2681)
4604 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4607 - fix truncated write corruption (2366)
4608 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4610 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4611 - ensure len(uuid) < 37 in lmc (1171)
4612 - fix ia64 OOPS in llog_test (2255)
4613 - zero end of page at obdfilter for partial page writes (2648)
4614 - don't leave stale dentries around after renames (bug 2428)
4615 - fix timeouts when evicting a client with a single lock held (2642)
4616 - set deadline for the initial HELLO message to drain (2634)
4617 - print out dotted-quad IP addresses in the socknal (2302)
4619 - additional debugging for MDS client eviction problem (2443)
4620 - fix mkfsoptions support for osts (2603, 2604)
4622 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4625 - add series for the vanilla 2.6.0 kernel
4626 - add series for the vanilla 2.4.24 kernel
4627 - add series for a cray x86/64 UL kernel drop
4628 - fix xattr patches for the vanilla 2.4.19 series
4630 - generate true UUIDs in lmc (1171)
4631 - have portals stack dumping break in UML (2466)
4632 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4633 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4634 - wake callback waiting threads on client eviction (2460)
4635 - Add --ptldebug and --subsystem to lmc (1719)
4636 - update assertion to allow safe interrupt allocation
4637 - set rq_no_resend for cancel requests (2432)
4638 - recalculate ptlrpcd timeout after resend (2494)
4639 - call vfs_rmdir when removing pending directories (2368)
4640 - fix renaming a file to itself (2429)
4641 - lmc creates a default one-stripe lov (2454)
4642 - expand procfs space to handle large clusters (2326)
4643 - increase UML stack to avoid overflow
4644 - update lconf's list of debug and subsystem masks
4645 - fix lfs find --obd (2510)
4646 - /proc tunable for disabling filter read caching (2591)
4647 - stop rpm packages from altering slapd.conf (2301)
4648 - disable nagle in the socknal under 0conf (2578)
4649 - choose mds inode size based on stripe count (2572)
4650 - fix kernel-source rpm problems (2516)
4652 - add --disable-doc to avoid pdf generation (2421)
4653 - update documentation, tests, type-os, comments
4654 - avoid format warnings on ia64
4655 - remove the TOE NAL
4656 - tiny code cleanups by removing unused fields
4658 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4661 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4662 - remove the most blatant lies from BUILDING file (2371)
4663 - change default debug level to reasonable production setting
4664 - reduce client side cache size to reduce cache flush time
4665 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4666 - make TCP zerocopy and pinger support enabled by default (2476)
4667 - sync writes completed after process exits caused crashes (2319)
4668 - maintain correct mount count on the MDS (2356)
4669 - backout 1557, because 2316 wasn't really fixed
4670 - better file I/O statistics gathering in /proc
4671 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4672 - another kernel patch to fix zero-copy TCP function export
4673 - don't take duplicate lock when processing re-sent getattr (2420)
4674 - lctl uses obd_self_export instead of creating new conn (2353)
4675 - MDS/OST recovery case which requires object creation asserted (2425)
4676 - move lfs from /usr/sbin to /usr/bin in packages
4677 - fix race between mds_client_add and mds_client_free (2417)
4678 - use kmalloc instead of slabs in portals (2430)
4679 - don't create duplicate records when a failover MDS is present (2442)
4680 - remove unnecessary mount age check (2332)
4681 - don't remove directory inodes from locks prematurely (2451)
4682 - don't break if MDS service name is the same as hostname (2103)
4683 - fix races in client write RPC generation when cache full (2482)
4685 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4688 - remove now-unused request->rq_obd (278)
4689 - if an allocation fails, print out how much memory we've used (1933)
4690 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4691 - add the "configurable stack size" patch to most series files (1256)
4692 - ability to write large log records, for 100+ OST configs (2306)
4693 - fix NULL deref when filter_prep fails (2314)
4694 - fix operator precedence error in filter_sync
4695 - dynamic allocation of socknal TX descriptors (2315)
4696 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4697 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4698 - ensure that CWARN messages in Portals always get to the syslog
4699 - __init/__exit are not for prototype decls (ldlm_init/exit)
4700 - x86-64 compile warning fixes
4701 - fix gateway LMC keyword conflict (2318)
4702 - fix MDS lock inversions in getattr/reint paths (1844)
4703 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4704 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4705 - DLM race condition prevented some lock evictions (2328)
4706 - ENOMEM detection and retry on socknal sends (2230)
4707 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4708 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4709 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4710 - better error messages when a client is rejected during recovery (1505)
4711 - avoid cancelling locks which were never granted, after failure (2330)
4712 - fix i_sem/journal inversion in mds_client_add (2333)
4713 - fix truncate/getattr lock cycle deadlock (2334)
4714 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4715 - fix two rare exit paths which could leak an l_lock() ref (2321)
4716 - fencepost error in MDS/OST orphan recovery (2226)
4717 - make log record alignment 8 bytes (1988)
4718 - lstripe now fails when requested offset > ost_count (2237)
4719 - ensure that all kernel series have a complete list.h (1607)
4720 - fix crashes in special-file operations (2316)
4721 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4723 - allow configurable automake binary, for testing new versions
4724 - small update to the lfs documentation
4726 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4728 * fix negative export reference count in fsfilt_sync (2312)
4730 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4731 * release candidate 0.9.1
4733 - orphans are moved into the PENDING directory for possible recovery
4734 - replayed opens now open by fid for orphan/rename safety (1042)
4735 - last close of an orphan inode generates a transno (683)
4736 - chdir() and mount() now pin the directory entry (1020)
4737 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4738 - discard very old requests without processing them (1502)
4739 - remove some common, well-understood CERRORs (1505)
4740 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4741 - clear "grant" flags in OST replies until OST grant code lands (1644)
4742 - fix read performance by not clobbering i_blksize on client (1598)
4743 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4744 - make LRU size a /proc tunable, clears locks when reduced (707)
4745 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4746 - server locks take references on exports now (1558)
4747 - build fixes for 2.4.20-rh trees (1663)
4748 - return an error from lov_create if all OSCs are inactive (1751)
4749 - fix import levels when a reconnect happens without a timeout (1597)
4750 - exit early from mds_open if we get a lookup error (1749)
4751 - partial page read at EOF wouldn't wait for disk before sending (1642)
4752 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4753 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4754 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4755 - write/truncate lock inversion (1639)
4756 - Don't auto-load obdclass, portals modules during cleanup (1495)
4757 - fix timestamps from jumping to "now" (1763)
4758 - extra journal assertions (1648)
4759 - add an extra multiunlink test (1771)
4760 - fix read_record/write_record API (1776)
4761 - fix leak of offset_extent, possible incorrect i_size later (1772)
4762 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4763 - Debugging for the kqswnal_get_idle_tx problems (1820)
4764 - Allow recovery to be attempted multiple times (1536)
4765 - Write out MDS last_rcvd file after it is first created (1600)
4766 - Fix tx_descriptor leak in failed transmit situations (1827)
4767 - ext3 journaling fixes for assertion failure after IO error (1871)
4768 - class_export_put() on freed export after completion AST error (1896)
4769 - Fix revalidate looping in VFS (1322)
4770 - Don't access a freed export during MDS_REINT timeout (1521)
4771 - Add open-unlink recovery support on the MDS (1673,1764)
4772 - Return an error if no MDS data was read from last_rcvd (1946)
4773 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4774 - Don't setup a disk device that doesn't match exported UUID (317)
4775 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4776 - avoid committing NULL handle in force close
4777 - local.sh is now a one-stripe LOV configuration
4778 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4779 - fix ext3 htree duplicate directory entry corruption (1516)
4780 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4781 - update to new LOV EA format (2097)
4782 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4784 2003-06-15 Phil Schwan <phil@clusterfs.com>
4787 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4788 - per-import/export recovery handling (958, 931, 959)
4789 - multiple last-rcvd slots, for serving multiple FSes (949)
4790 - connections are again shared between multiple imp/exports (963, 964)
4791 - "umount -f" would hang if any requests needed to be sent (393, 978)
4792 - avoid pinning large req buffer by copying for queued messages (989)
4793 - add "uuid" to "lctl device" command to help upcalls (991)
4794 - "open" RPCs with transnos would confuse recovery counters (1037)
4795 - do proper endian conversion of all wire messages (288, 340, 891)
4796 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4797 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4798 - fix ptlrpc_connection leak in target_handle_connect (1174)
4799 - fix import refcounting bug in OST and MDS cleanup (1134)
4800 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4801 - fix very unlikely obd_types race condition (501)
4802 - remove osc_open hack for echo_client (1187)
4803 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4804 - a failure in read_inode2 leads to deadlock (1139)
4805 - cancel ack-locks as soon as transaction is committed (1072)
4806 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4807 - make sure to commitrw after any preprw to avoid deadlock (1162)
4808 - failing to execute a file in a lustre FS would lock inode (1203)
4809 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4810 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4811 - obd_uuid2tgt would walk off the end of the list (1255)
4812 - on IA64 the osc would give portals incorrect bulk size (1258)
4813 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4814 - fix lock inversion caused by new llite matching code (1282)
4815 - limit the number of dirty pages on a client to 10MB (1286)
4816 - timed out locks were not being corrected cancelled (1289)
4817 - fix O_DIRECT above 4GB on IA-32 (1292)
4818 * major user-visible changes
4819 - fail out/fail over policy now controlled by the upcall (993)
4821 - add OBD_PING to check server availability and failure (954)
4822 - lustre messages are now sent in sending host order (288, 340, 891)
4823 - add eadatalen to MDS getattr reply (340)
4824 - OST read replies may contain second buffer, with per-page status (593)
4826 2003-03-11 Phil Schwan <phil@clusterfs.com>
4829 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4830 - failed open()s could cause deadlock; fixed (867, 869)
4831 - stop cancelling OST locks when files are closed (481)
4832 - overlapping XID spaces caused network corruption (851, 853)
4833 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4834 - setattr_raw added, to do single-RPC, server-side setattrs
4835 - lmc/lconf syntax change for OST UUIDs
4836 - fix crashy race condition between ptlrpc_free_req and osc_close
4837 - don't use request in mdc_enqueue if we hit a timeout (889)
4838 - don't set the inode i_size for regular files from the MDS (896)
4839 - handle out of order completion AST (842)
4840 - don't LBUG if a lock request times out after receiving AST (913)
4841 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4842 - if a bad lock AST arrives, send an error instead of dropping entirely
4843 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4844 - fix leak in bulk IO when only partially completed (899, 900, 926)
4845 - fix O_DIRECT for ia64 (55)
4846 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4847 - C-z after timeout could hang a process forever; fixed (977)
4849 - client-side I/O cache (678, 924, 929, 941, 970)
4851 - READPAGE and SETATTRs which don't take server-side locks get
4854 2003-02-11 Phil Schwan <phil@clusterfs.com>
4857 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4858 - Fix fsfilt_extN_readpage() to read a full page of directory
4859 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4860 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4861 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4862 - Most intent operations were removed, in favour of a new RPC mode
4863 that does a single RPC to the server and bypasses most of the VFS
4864 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4865 - Aggressively cancel local locks on DLM servers
4866 - mds_reint_unlink sends EA to the client if it's the last nlink.
4867 client uses that EA to unlink OST objects.
4868 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4869 - recursive symlinks were fixed (439)
4870 - fixed NULL deref in DEBUG_REQ
4871 - filter_update_lastobjid no longer calls sync, which annoyed extN
4872 - fixed multi-client small-writes to a single file problem (445)
4873 - fixed mtime updates during file writes (607)
4874 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4875 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4876 - fixed Linux OST size reporting problem (444, 656)
4877 - OST now updates object mtime with writes or setattr (607, 619)
4878 - client verifies file size before zeroing page past EOF (445)
4879 - OST now writes last allocated objid to disk with allocation (108)
4880 - LOV on echo now works (409)
4882 - mds_reint_unlink sends a new buffer, with the EA included. this
4883 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4884 set if a regular file was being unlinked, and it was the last link
4885 - use PtlGet from the target for bulk writes (315)
4886 - OST now updates object mtime with writes or setattr (607, 619)
4887 - LDLM now has a grant-time callback to revalidate locked items, if
4889 - Many MDS operations were reorganized to combat race conditions
4891 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4892 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4894 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4897 - Fully reactivate OST imports after reconnection (512, others)
4898 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4899 - More graceful error handling for truncating on dead OST (515)
4900 - Don't error out unless we're actually accessing dead stripes (474)
4901 - Fix garbage sizes when stripes are missing (410)
4902 - LRU counters were broken, causing constant lock purge (433, 432)
4903 - garbage on read from stripes with failed OSTs (441)
4904 - mark OSCs as active before reconnecting during recovery (438)
4905 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4906 - lfind did not preserve OST order in output (443)
4907 - symlinks cause hung clients, incorrect data (439)
4908 - stop dereferencing request after dropping refcount (457)
4909 - don't LASSERT(spin_is_locked) on non-SMP (455)
4910 - fixes for many rename() bugs
4911 - fstat didn't correctly synchronize attributes (399)
4912 - server must handle lock cancellation during blocking AST prep (487)
4913 - bulk descriptors were free()d too soon (511)
4914 - fix paths in lconf, which would load incorrect modules (451, 507)
4915 - fix confusing lconf 'host not found' error message (386)
4916 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4917 - fix race condition in mdc_blocking_ast() for inode access (526)
4918 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4919 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4920 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4921 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4922 - fix lstripe in several ways (kernel side) (527)
4923 - fix request leak in ldlm_cli_enqueue (262)
4924 - incorrect OSC was marked inactive after OST failure
4925 - call mds_fs_cleanup before unmounting filesystem (524)
4926 - fix races between taking ns_lock and ldlm_lock_change_resource
4927 - fix races updating LOV export open file list
4928 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4929 - fix recovery NULL deref in ldlm_cli_cancel_unused
4930 - fix some DLM races by using new hash table for lock handles (419)
4931 - permit the client to specify desired inodes, at replay
4932 - duplicate requests when we queue them for replay reintegration
4933 - fix last_rcvd offset calculation
4934 - sync after each recovered transaction, so we always make progress
4935 - never, not always, ERESTART requests without transnos
4936 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4938 - skip replay if the MDS says that the client is already connected
4939 - don't check for a recovery-enabled export to match lctl's UUID
4940 - don't INC_USE_COUNT for phantom exports
4941 - don't crash when cleaning up phantom exports (567)
4942 - don't double-finish or set replay data for errored mdc_open requests
4943 - abort requests when they time out, so we don't get old replies
4944 - send/receive replies for AST messages again
4945 - if the client says that it doesn't have the lock, cancel it on the
4947 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4948 mark it as destroyed, it will all work out when decref is called
4949 - fix module use counts (22, 581)
4951 - ASTs now expect a reply (server cancels lock on error reply)
4953 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4956 - fix many simultaneous client startup (392)
4957 - fix dentry->d_it clobbering
4958 - credentials weren't being shipped for readdir/getattr operations
4959 - remove invalid assertions triggered during some concurrent MD
4961 - proper Lustre versions added (336, 389)
4962 - fix memory leak for create error case (398)
4963 - fix LOV locking bug that would get cli/srv out of sync
4964 - fix echo client over LOV (409)
4965 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4966 - fix double-O_EXCL intent crash (424)
4967 - avoid sending multiple lock CANCELs (352)
4969 - MDS can do multi-client recovery (modulo bugs in new code)
4971 - many updates, edits, cleanups
4973 2002-11-18 Phil Schwan <phil@clusterfs.com>
4976 - fix null d_it dereference (346)
4977 - fix full OST/dbench hang (333)
4978 - fix permission problem with file removal (286)
4979 - fix removal of OSCs from LOV when they fail
4980 - fix NULL deref during bulk timeout (214)
4981 - fix problems related to multiple filesystems on one MDS (241)
4982 - fixed serious subtle metadata locking bugs
4983 - free locks on clients when inodes are removed due to memory
4985 - fix inode pointer in lock data (285)
4986 - partial support for multiple MDS on a single host (241)
4987 - data locks weren't cancelled at clear_inode time (290, 311)
4988 - intent locks could lead to unbounded lock growth (205)
4989 - added a maximum lock count, an LRU list, and a flusher
4990 - fix multiple rename (365)
4991 - properly abstracted the echo client
4992 - OSC locked 1 byte too many; fixed
4993 - rewrote brw callback code:
4994 - fixed recovery bugs related to LOVs (306)
4995 - fixed too-many-pages-in-one-write crash (191)
4996 - fixed (again) crash in sync_io_timeout (214)
4997 - probably fixed callback-related race (385)
4999 - Add capability to MDS protocol
5000 - LDLM cancellations and callbacks on different portals
5002 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
5005 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
5006 - defer open object destruction to close time (601981, 138)
5007 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
5008 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
5009 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
5010 - don't call obd_setattr in ll_file_release for destroyed objects
5011 * protocol change to lustre_msg: move |version| and add |flags|
5012 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
5013 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
5014 * added replay of create, unlink, link and rename operations during
5015 MDS failover; recovery should be much more robust now
5016 * remove failed OSCs from LOVs (only lov_create uses this so far)
5017 * the lustre-HOWTO was brought (more) up to date (582544)
5019 2002-10-23 Phil Schwan <phil@clusterfs.com>
5022 - in-use dentries weren't being reused properly (617851)
5023 - prevent multiple LDLM setup (599178)
5024 - fix LOV size calculations for truncate (617853)
5025 - fix client handling of MDS intent errors (POSIX)
5026 - fix permission bug in lovstripe.c test (624321)
5027 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
5028 - truncate past end of file could corrupt data
5029 - proper cleanup after timeouts, crashes, etc (592524, 550815)
5030 - a race in recovery could return ETIMEDOUT to apps (623947)
5031 - building outside the source directory was fixed
5032 * the lustre-HOWTO was brought (more) up to date (582544)
5033 * major progress was made on recovery functionality
5035 2002-10-10 Phil Schwan <phil@clusterfs.com>
5038 - recovery deadlock fix
5039 - rm -rf causes LBUG fix (617817)
5040 - file open by multiple tasks fix (618962)
5041 - directory permissions bugs (602707 and 620007)
5042 - journal_stop fixed with locking (611313)
5043 - O_APPEND failures resolved (618273, perhaps 614459)
5044 - lconf PATH fix (619770)
5045 - IA64 build fix (621450)
5046 - RPC buffer sizes scale with amount of memory
5048 2002-10-01 Phil Schwan <phil@clusterfs.com>
5051 - locks would be cancelled without throwing away data pages,
5052 resulting in inconsistent data (605627)
5053 - inode attributes were not always being refreshed (605627, 612449)
5054 - lconf now continues to cleanup after lctl reports an error
5055 - MDS now enforces user permissions (602707)
5056 - lprocfs cleanup fixed, but not yet enabled (614157)
5057 - fixed infinite server hang, should a client not respond to an AST
5058 - avoid going into recovery if user calls readlink() with a buffer
5059 that's too small (613941)
5060 - AST RPCs no longer require replies (614867) -- this may be changed
5061 - don't crash server if client sends an IOV that's too big (611336)
5062 - fixed lock conversion deadlock (611892)
5063 - fixed the following of symlinks (614622)
5064 * recovery: the server can remove locks from a client that dies, other
5065 clients can make progress
5066 * more extN patch fixes
5067 * compile-time configurable ptlrpc buffer allocations
5069 - collaborative read cache document
5070 - Lustre Lite Performance CDR document-in-progress
5072 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5075 - fix typo in patch-2.4.18
5077 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5080 - clear ptlrpc request each time in handle_incoming_request()
5081 - unlink of files now destroys the object on the OST
5083 2002-09-19 Peter Braam <braam@clusterfs.com>
5085 * add hard link support
5086 * change obdfile creation method
5087 * kernel patch changed
5089 2002-09-19 Peter Braam <braam@clusterfs.com>
5092 - stack overflow bug in extN fixed
5094 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
5096 * documentation updates
5097 - add man pages for config tools
5098 - update tests/README to describe testing with new config tools
5099 - finish metadata API descriptions
5100 * bug fixes and cleanups
5101 - statfs workaround for 16TB limit
5102 - LOV stripe allocation improved, can stripe on subset of OSTs
5103 - LOV file size/IO offset was wrong for files > 4GB in size
5104 - object EA data was being dropped, caused files to be unreadable
5105 - memory overflow with non-LOV OST caused memory corruption
5106 - fixed regression tests to work with new config tools, obdfilter
5107 - fixed bug when directory size became larger than 1 block
5108 - fixed bug (for single client case) when PWD was deleted
5109 - invalidate local directory pages when doing intent-based ops
5110 - avoid LDLM oops when lock callback contained bad data
5112 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5114 * documentation updates
5115 * bug fixes and cleanups
5116 - configuration tools
5119 - 64-bit compile warnings
5120 - 64-bit internal statfs data
5122 * test_brw on persistent OST devices
5124 * lprocfs (disabled)
5126 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5128 * documentation updates
5129 * bug fixes and cleanups
5130 * configuration tools
5132 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5135 * many small fixes to 0_5_4
5136 * io/network handling
5137 * thinkos in MDS operations
5139 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5142 * crucial basic fixes to 0.5.3
5143 * IOR, Iozone work over Elan
5146 2002-08-07 Phil Schwan <phil@clusterfs.com>
5147 * version 0_5_3, our first alpha
5148 * we use the new Portals iovs
5149 * documentation updates
5150 * bug fixes and cleanups
5151 * small changes in the DLM wire protocol
5153 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5154 * version 0_5_1 with some initial stability,
5155 * locking on MD and file I/O.
5156 * documentation updates
5157 * several bug fixes since 0.5.0
5158 * small changes in wire protocol
5160 2002-07-18 Phil Schwan <phil@clusterfs.com>
5162 * delivered as Lustre Light Alpha
5163 * fixed a crash after handling invalid MDS requests
5164 * fixed directory pages for architectures with non-4k pages sizes
5166 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5167 * release version v0_4_4
5168 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5170 * Fixes a number of interruption problems with OST operations.
5171 * Update documentation for portals header changes
5172 * Move all wire protocol structs/defines to lustre_idl.h
5173 * Fixes symlink length bug.
5174 * Add tcpdump to repository.
5176 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5177 * release version v0_4_3
5178 * Fixes statfs for inodes on extN.
5179 * Fixes bug in runtests which would delete /etc/hosts.
5180 * Use 64-bit object IDs wherever possible (not into VFS though)
5181 Remove ost_get_info, which is unused by lustre, and out of date.
5183 2002-07-03 Peter Braam <braam@clusterfs.com>
5184 * release version v0_4_2 Fixes a lookup error (type not passed)
5185 * move forward to head of Portals
5186 * move forward to latest Lustre kernel
5188 2002-06-25 Peter Braam <braam@clusterfs.com>
5189 * release version v0_4_1. Hopefully stable on single node use.