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: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
19 Severity : enhancement
21 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
23 Severity : enhancement
25 Description: Upcall on Lustre log has been dumped
26 Details : Allow for a user mode script to be called once a Lustre log has
27 been dumped. It passes the filename of the dumped log to the
28 script, the location of the script can be specified via
29 /proc/sys/lnet/debug_log_upcall.
34 Description: avoid messages about idr_remove called for id which is not allocated.
35 Details : Move assigment s_dev for clustered nfs to end of initialization, for
36 avoid problem with error handling.
41 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
42 Details : When connection is reused this not moved from CONN_UNUSED_HASH into
43 CONN_USED_HASH and this prodice warning when put connection again
46 Severity : enhancement
48 Description: Export bytes_read/bytes_write count on OSC/OST.
52 Description: Early reply size mismatch, MGC loses connection
53 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
54 the connect flags are properly negotiated.
58 Description: Properly propagate oinfo flags from lov to osc for statfs
59 Details : restore missing copy oi_flags to lov requests.
61 Severity : enhancement
63 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
64 : l_getgroups(8), lst(8), routerstat(8)
65 Details : included man pages for llobdstat(8), llstat(8),
66 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
68 Severity : enhancement
70 Description: configure's --enable-quota should check the
71 : kernel .config for CONFIG_QUOTA
72 Details : configure is terminated if --enable-quota is passed but
73 : no quota support is in kernel
77 Description: Remove portals compatibility
78 Details : Remove portals compatibility, not interoperable with releases
83 Description: Resolve device initialization race
84 Details : Prevent proc handler from accessing devices added to the
85 obd_devs array but yet be intialized.
87 Severity : enhancement
89 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
91 Severity : enhancement
93 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
98 Description: Using wrong pointer in osc_brw_prep_request
99 Details : Access to array[-1] can produce panic if kernel compiled with
100 CONFIG_PAGE_ALLOC enabled
104 Description: Client runs out of low memory
105 Details : Consider only lowmem when counting initial number of llap pages
109 Description: *optional* service tags registration
110 Details : if the "service tags" package is installed on a Lustre node
111 When the filesystem is mounted, a local-node service tag will
112 be created. See http://inventory.sun.com/ for more information
113 about the Service Tags asset management system.
117 Description: Kernel BUG tries to release flock
118 Details : Lustre does not destroy flock lock before last reference goes
119 away. So always drop flock locks when client is evicted and
120 perform unlock regardless of successfulness of speaking to MDS.
124 Description: add recount protection for osc callbacks, so avoid panic on shutdown
128 Description: sanity test 65a fails if stripecount of -1 is set
129 Details : handle -1 striping on filesystem in ll_dirstripe_verify
134 Description: ASSERTION(CheckWriteback(page,cmd)) failed
135 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
138 Severity : enhancement
140 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
144 Description: do not process already freed flock
145 Details : flock can possibly be freed by another thread before it reaches
146 to ldlm_flock_completion_ast.
150 Description: LBUG during stress test
151 Details : Need properly lock accesses the flock deadlock detection list.
155 Description: oops in page fault handler
156 Details : kernel page fault handler can return two special 'pages' in error case, don't
157 try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
161 Description: timeout with invalidate import.
162 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which should be
163 handled by ptlrpcd. This produce long age waiting and -ETIMEOUT
164 ptlrpc_invalidate_import and as result LASSERT.
166 Severity : enhancement
168 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
172 Description: enable MGS and MDT services start separately
173 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
174 a co-located MGS without starting the MGS, which is a complement
175 to 'nosvc' mount option.
179 Frequency : after recovery
180 Description: precreate to many object's after del orphan.
181 Details : del orphan st in oscc last_id == next_id and this triger growing
182 count of precreated objects. Set flag LOW to skip increase count
183 of precreated objects.
187 Frequency : rare, on clear nid stats
188 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
189 Details : when clean nid stats sometimes try destroy live entry,
190 and this produce panic in free.
194 Description: Stack overflow during MDS log replay
195 ease stack pressure by using a thread dealing llog_process.
199 Description: wait until IO finished before start new when do lock cancel.
200 Details : VM protocol want old IO finished before start new, in this case
201 need wait until PG_writeback is cleared until check dirty flag and
202 call writepages in lock cancel callback.
204 Severity : enhancement
206 Description: using special macro for print time and cleanup in includes.
210 Description: mds_mfd_close() ASSERTION(rc == 0)
211 Details : In mds_mfd_close(), we need protect inode's writecount change
212 within its orphan write semaphore to prevent possible races.
216 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
220 Frequency : rare, on shutdown ost
221 Description: don't hit live lock with umount ost.
222 Details : shrink_dcache_parent can be in long loop with destroy dentries,
223 use shrink_dcache_sb instead.
227 Description: don't panic with use echo client
228 Details : echo client pass NULL as client nid pointer and this produce null
233 Description: fix build on ppc32
234 Details : compile code with -m64 flag produce wrong object file for ppc32.
238 Description: add message levels for liblustreapi
242 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
243 Details : Move the 'good_osts' check before the 'total_bavail' check. This
244 will result in an -EAGAIN and in the exit call path we call
245 alloc_rr() which will with increasing aggressiveness attempt to
246 aquire precreated objects on the minimum number of required OSCs.
250 Description: Use old size assignment to avoid deadlock
251 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
252 the scheduling while holding a spinlock. We do not need locking
253 for size in ll_update_inode() because size is only updated from
254 the MDS for directories or files without objects, so there is no
255 other place to do the update, and concurrent access to such inodes
256 are protected by the inode lock.
260 Description: resolve "_IOWR redefined" build error on SLES10
264 Description: dump the memory debugging after all modules are unloaded to
265 suppress false negative in conf_sanity test 39
267 Severity : enhancement
269 Description: build kernel-ib packages for OFED 1.3 in our release cycle
274 Description: fix SLES kernel versioning
275 Details : the kernel version for our SLES 10 kernel did not include a "-"
276 before the "smp" at the end. while this was not a problem in
277 general, it did mean that software trying to use the kernel
278 version to try to detect a vendor specific kernel would fail.
279 this was most evident by the OFED build scripts.
283 Description: Don't update lov_desc members until making sure they are valid
284 Details : When updating lov_desc members via proc fs, need fix their
285 validities before doing the real update.
289 Description: don't put request into delay list while invalidate in flight.
290 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
291 import in flight. this produce timeout for invalidate and sometimes
292 can cause stale data.
296 Frequency : on ppc only
297 Description: not convert ost objects for directory because it's not exist.
298 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
302 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
304 Severity : enhancement
306 Description: Update to support 2.6.22.14 vanilla kernel.
310 Frequency : rare, on recovery
311 Description: read procfs can produce deadlock in some situation
312 Details : Holding lprocfs lock which send rpc can produce block for destroy
313 obd objects and this also block reconnect with -EALREADY. This isn't
314 fix all lprocfs bugs - but make it rare.
316 Severity : enhancement
318 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
321 Frequency : frequent on X2 node
323 Description: mdc_set_open_replay_data LBUG
324 Details : Set replay data for requests that are eligible for replay.
328 Description: lustre_mgs: operation 101 on unconnected MGS
329 Details : When MGC is disconnected from MGS long enough, MGS will evict the
330 MGC, and late on MGC cannot successfully connect to MGS and a lot
331 of the error messages complaining that MGS is not connected.
335 Frequency : on network error
336 Description: panic with double free request if network error
337 Details : mdc_finish_enqueue is finish request if any network error ocuring,
338 but it's true only for synchronus enqueue, for async enqueue
339 (via ptlrpcd) this incorrect and ptlrpcd want finish request
342 Severity : enhancement
344 Description: client-side metadata stat-ahead during readdir(directory readahead)
345 Details : perform client-side metadata stat-ahead when the client detects
346 readdir and sequential stat of dir entries therein
349 Frequency : on start mds
351 Description: Implement get_info(last_id) in obdfilter.
354 Frequency : occasional
356 Description: Correctly check stale fid, not start epoch if ost not support SOM
357 Details : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
358 overwrited with new generated fid, but mds can anwer with one of these
359 two fids and both is not stale. setattr incorectly start epoch and
360 assume will be called done_writeting, but without SOM done_writing
364 Frequency : rare, depends on device drivers and load
366 Description: MDS or OSS nodes crash due to stack overflow
367 Details : Code changes in 1.8.0 increased the stack usage of some functions.
368 In some cases, in conjunction with device drivers that use a lot
369 of stack the MDS (or possibly OSS) service threads could overflow
370 the stack. One change which was identified to consume additional
371 stack has been reworked to avoid the extra stack usage.
374 Frequency : occasional
376 Description: Do not fail import if osc_interpret_create gets -EAGAIN
377 Details : If osc_interpret_create got -EAGAIN it immediately exits and
378 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
379 oscc_has_objects and see OSC has no objests and call
380 oscc_internal_create to resend create request.
382 Severity : enhancement
384 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
386 Severity : enhancement
388 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
391 Frequency : very rare
393 Description: Fix replay if there is an un-replied request and open
394 Details : In some cases, older replay request will revert the
395 mcd->mcd_last_xid on MDS which is used to record the client's
398 Severity : enhancement
400 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
402 Severity : enhancement
404 Description: Add rhel5 support to HEAD.
406 Serverity : enhancement
408 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
413 Description: Don't allow skipping OSTs if index has been specified.
414 Details : Don't allow skipping OSTs if index has been specified, make locking
415 in internal create lots better.
419 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
420 Details : access to bitfield in structure is always rounded to long
421 and this produce problem with not atomic change any bit.
425 Description: Lustre make rpms failed.
426 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
427 builds Lustre rpms from tarball.
430 Frequency : rare on shutdown ost
432 Description: If llog cancel was not send before clean_exports phase, this can
433 produce deadlock in llog code.
434 Details : If llog thread has last reference to obd and call class_import_put
435 this produce deadlock because llog_cleanup_commit_master wait when
436 last llog_commit_thread exited, but this never success because was
437 called from llog_commit_thread.
441 Description: allow userland application know is lost one of stripes.
442 Details : fill lvb_blocks with error code on ost and return it to
443 application if error flag found.
447 Description: NULL lov_tgts causing MDS oops
448 Details : more safe checks for NULL lov_tgts for avoid oops.
450 Severity : enhancement
452 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
456 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
457 Details : If an OST is down the MDS will hang indefinitely in
458 obd_statfs_rqset() waiting for the statfs data. While for
459 MDS QOS usage of statfs, it should not stuck in waiting.
461 Severity : enhancement
463 Description: remote_acl support
464 Details : Support ACL-based permission check for remote user.
465 Support setfacl/getfacl for remote user with the utils
466 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
467 the system "{s,g}etfacl" utils.
469 Severity : enhancement
471 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
473 Severity : enhancement
475 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
479 Description: make mgs_setparam() handle fsname containing dash
480 Details : fsname containing a dash does not work with lctl conf_param
482 Severity : enhancement
484 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
486 Severity : enhancement
488 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
492 Frequency : rare, at shutdown
493 Description: access already free / zero obd_namespace.
494 Details : if client_disconnect_export was called without force flag set,
495 and exist connect request in flight, this can produce access to
496 NULL pointer (or already free pointer) when connect_interpret
497 store ocd flags in obd_namespace.
501 Frequency : only at startup
502 Description: not alloc memory with spinlock held.
503 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
504 if any spinlock held.
506 Severity : enhancement
508 Description: make lustre randomly fail allocating memory
509 Details : Make lustre randomly failed allocating memory for testing purpose.
511 Severity : enhancement
513 Description: lost problems with lov objid file
514 Details : Fixes some scability and access to not inited memory problems
515 in work with lov objdid file.
520 Description: lfs find does not continue on file error
521 Details : Continue other files processing when a file/dir is absent.
525 Description: Inconsistent usage of lustre_pack_reply()
526 Details : Standardize the usage of lustre_pack_reply() such that it
527 always generate a CERROR on failure.
530 Frequency : occasional
532 Description: MDS hang or stay in waiting lock
533 Details : If client receive lock with CBPENDING flag ldlm need send lock
534 cancel as separate rpc, to avoid situation when cancel request
535 can't processed due all i/o threads stay in wait lock.
539 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
543 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
547 Description: Update to SLES9 kernel-2.6.5-7.287.3.
551 Description: Update to RHEL4 latest kernel.
553 Severity : enhancement
555 Description: Build SLES10 patchless client fails
556 Details : The configure was broken by run ./configure with
557 --with-linux-obj=.... argument for patchless client. When the
558 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
559 can't search header adequately. Use absolute path such as
560 -I($LINUX)/include instead.
564 Description: interrupt oig_wait produce painc on resend.
565 Details : brw_redo_request can be used for resend requests from ptlrpcd and
566 private set, and this produce situation when rq_ptlrpcd_data not
567 copyed to new allocated request and triggered LBUG on assert
568 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
569 ptlrpcd set if request is changed and can be safety changed to use
574 Description: LASSERT_{REQ,REP}SWAB macros are buggy
575 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
576 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
577 these macros and replace them with their difinitions instead.
581 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
582 Details : Update which_patch & target file for SLES10 latest kernel.
585 Description: Remove client patches from SLES 10 kernel.
586 Details : This causes SLES 10 clients to behave as patchless clients
587 even on a Lustre-patched (server) kernel.
589 Severity : enhancement
591 Description: self-adjustable client's lru lists
592 Details : use adaptive algorithm for managing client cached locks lru
593 lists according to current server load, other client's work
594 pattern, memory activities, etc. Both, server and client
595 side namespaces provide number of proc tunables for controlling
598 Severity : enhancement
600 Description: light-weight GSS support
601 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
602 and reduce performance overhead.
604 Severity : enhancement
606 Description: Linux keyring support
607 Details : Support using service of Linux keyring for Lustre GSS internal
608 context refresh/cache mechanism.
612 Description: Fix errors in lfs documentation
613 Details : Fixes man pages
617 Description: don't use GFP_* in generic Lustre code.
618 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
622 Description: obdclass is limited by single OBD_ALLOC(idarray)
623 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
627 Description: warnings with build patchless client with vanila 2.6.19 and up
628 Details : change old ctl_table style and replace ctl_table/ctl_table_header
629 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
633 Frequency : Only for SLES
634 Description: sanity_quota fail test_1
635 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
636 file, which makes TSTID[2] inunique.
640 Frequency : Only for SLES9
641 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
642 Details : Update target/ChangeLog/which_patch .
646 Frequency : Only for test_10 in sanity-quota.sh
647 Description: a bug in quota test code
648 Details : checking a test flag in a wrong place
652 Frequency : Only for test_13 in sanity-quota.sh
653 Description: a bug in quota test script
654 Details : shouldn't check overfull allocated quota
658 Description: jbd statistics
659 Details : Port older jbd statistics patch for sles10
661 Severity : enhancement
663 Description: Add printing inode info into message about error in writepage.
668 Description: reply_lock_interpret crash due to race with it and lock cancel.
669 Details : Do not replay locks that are being cancelled. Do not reference
670 locks by their address during replay, just by their handle.
674 Frequency : When flocks are used.
675 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
676 Details : Flock locks might destroy just granted lock if it could be merged
677 with another existing flock, this is done in completion handler,
678 so teach ldlm_cli_enquque_fini that this is a valid case for
684 Description: Oops in read and write path when failing to allocate lock.
685 Details : Check if lock allocation failed and return error back.
689 Description: lstripe command fails for valid OST index
690 Details : The stripe offset is compared to lov->desc.ld_tgt_count
691 instead of lov->desc.ld_active_tgt_count.
695 Description: sanity.sh failed test 103
696 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
701 Description: df doesn't work properly if diskfs blocksize != 4K
702 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
706 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
707 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
709 Severity : enhancement
711 Description: Adaptive timeouts
712 Details : RPC timeouts adapt to changing server load and network
713 conditions to reduce resend attempts and improve recovery time.
717 Description: llapi_file_create() does not allow some changes
718 Details : add llapi_file_open() that allows specifying the mode and
719 open flags, and also returns an open file handle.
723 Description: merge and cleanup kernel patches.
724 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
728 Description: Add journal checksum support.(Kernel part)
729 Details : The journal checksum feature adds two new flags i.e
730 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
731 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
732 indicates that the commit block contains the checksum for
733 the blocks described by the descriptor blocks. Now commit
734 record can be sent to disk without waiting for descriptor
735 blocks to be written to disk. This behavior is controlled
736 using JBD2_FEATURE_ASYNC_COMMIT flag.
740 Description: OSS needs mutliple precreate threads
741 Details : Add ability to start more than one create thread per OSS.
745 Description: Sanity.sh test_65k failed: setstripe should have succeeded
746 Details : execute lfs setstripe on client
750 Description: mds_obd_create error creating tmp object
751 Details : When the user sets quota on root, llog will be affected and can't
752 create files and write files.
755 Frequency : Always on ia64 patchless client, and possibly others.
757 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
758 Details : This allows the patchless client to be loaded on architectures
763 Description: RedHat Update kernel for RHEL5
764 Details : Add the kernel config file for RHEL5.
768 Description: RedHat Update kernel for RHEL5
769 Details : Modify the kernel config file more closer RHEL5.
773 Description: Build failure against Centos5 (RHEL5)
774 Details : Define PAGE_SIZE when it isn't present.
778 Description: test_7 on sanity-quota.sh can't be used on separate machines
779 Details : add facet to handle it
783 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
784 Details : with NFS, the anon dentry's parent was set to itself in
785 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
786 corresponding dentry other than use rec->ur_name.
788 Severity : enhancement
790 Description: lfs setstripe enhancement
791 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
795 Description: enable data checksumming by default
796 Details : enable checksum by default, allow --disable-checksum
797 configure option and "-o nochecksum" mount option. Checksums
798 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
802 Description: lustre support for RHEL5
803 Details : Add support for RHEL5.
807 Description: Client eviction due to failover config
808 Details : after a connection loss, the lustre client should attempt to
809 reconnect to the last active server first before trying the
810 other potential connections.
814 Description: when mds and osts use different quota unit(32bit and 64bit),
815 quota will be released repeatly.
816 Details : void sending multiple quota reqs to mds, which will keep the
817 status between the reqs.
821 Description: osts not allocated evenly to files
822 Details : change the condition to increase offset_idx
825 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
827 Description: Data corruption for OSTs that are formatted larger than 2TB
829 Details : When generating the bio request for lustre file writes the
830 sector number would overflow a temporary variable before being
831 used for the IO. The data reads correctly from Lustre (which
832 will overflow in a similar manner) but other file data or
833 filesystem metadata may be corrupted in some cases.
837 Description: Tune the kernel for good SCSI performance.
838 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
839 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
844 Description: rewrite ext2-derived code in obdclass/uuid.c
845 Details : rewrite inherited code (uuid parsing code from ext2 utils)
846 from scratch preserving functionality.
850 Description: "lfs find -obd UUID" prints directories
851 Details : "lfs find -obd UUID" will return all directory names instead
852 of just file names. It is incorrect because the directories
853 do not reside on the OSTs.
857 Description: use i_size_read and i_size_write in 2.6 port
858 Details : replace inode->i_size access with i_size_read/write()
861 Frequency : when removing large files
863 Description: scheduling issue during removal of large Lustre files
864 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
865 It causes scheduling issues when removing large files (17TB in the
869 Frequency : only with liblustre clients on XT3
871 Description: evictions taking too long
872 Details : allow llrd to evict clients directly on OSTs
875 Frequency : only on ppc
877 Description: /proc/fs/lustre/devices broken on ppc
878 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
879 We should make sure the type of 'index' is loff_t and then cast
880 explicitly as needed below. Do not assign an explicitly cast
884 Frequency : only for rhel5
886 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
887 Details : Modify the target file & which_kernel.
889 Severity : enhancement
891 Description: omit set fsid for export NFS
892 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
893 as FS_REQUIRES_DEV for avoid problems with generate fsid.
897 Frequency : Always, for kernels after 2.6.16
898 Description: Fix warning idr_remove called for id=.. which is not allocated.
899 Details : Last kernels save old s_dev before kill super and not allow
900 to restore from callback - restore it before call kill_anon_super.
904 Description: buffer overruns could theoretically occur
905 Details : llapi_semantic_traverse() modifies the "path" argument by
906 appending values to the end of the origin string, and a
907 overrun may occur. Adding buffer overrun check in liblustreapi.
911 Description: Fix error on 'ls .' at the top of the Lustre mount.
912 Details : Don't revalidate dentry if it is a root dentry.
916 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
917 Details : Modify vm-tunables-rhel4.patch.
921 Description: Kernel config for 2.6.18-vanilla.
922 Details : Modify targets/2.6-vanilla.target.in.
923 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
924 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
925 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
926 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
930 Description: improve handling recoverable errors
931 Details : if request processig with error which can be recoverable on server
932 request should be resend, otherwise page released from cache and
937 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
938 Details : Modify target file & which_patch.
939 A flaw was found in the IA32 system call emulation provided
940 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
941 value could be stored in the %RAX register, which could trigger an
942 out-of-bounds system call table access. An untrusted local user
943 could exploit this flaw to run code in the kernel
944 (ie a root privilege escalation). (CVE-2007-4573).
948 Description: O_DIRECT bypasses client statistics.
949 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
950 read_ahead_stats were updated. copied the stats section from
951 osc_send_oap_rpc() into async_internal().
955 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
959 Description: change order of libsysio includes
960 Details : '#include sysio.h' should always come before '#include xtio.h'
964 Description: handle "serious error: objid * already exists" more gracefully
965 Details : If LAST_ID value on disk is smaller than the objects existing in
966 the O/0/d* directories, it indicates disk corruption and causes an
967 LBUG(). If the object is 0-length, then we should use the existing
968 object. This will help to avoid a full fsck in most cases.
970 Severity : enhancement
972 Description: adapt the lustre_config script to support the upgrade case
973 Details : Add "-u" option for lustre_config script to support upgrading 1.4
974 server targets to 1.6 in parallel.
978 Description: To avoid grant space > avaible space when the disk is almost
979 full. Without this patch you might see the error "grant XXXX >
980 available" or some LBUG about grant, when the disk is almost
982 Details : In filter_check_grant, for non_grant cache write, we should
983 check the left space by if (*left > ungranted + bytes), instead
984 of (*left > ungranted), because only we are sure the left space
985 is enough for another "bytes", then the ungrant space should be
986 increase. In client, we should update cl_avail_grant only there
987 is OBD_MD_FLGRANT in the reply.
991 Description: Update RHEL 4 kernel to fix local root privilege escalation.
992 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
993 described in CVE-2007-4573. This problem could allow untrusted
994 local users to gain root access.
997 Frequency : when using O_DIRECT and quotas
999 Description: Incorrect file ownership on O_DIRECT output files
1000 Details : block usage reported by 'lfs quota' does not take into account
1001 files that have been written with O_DIRECT.
1006 Description: touch file failed when fs is not full
1007 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1008 otherwise we can get ENOSP while fs is not full.
1012 Description: parallel lock callbacks
1013 Details : Instead of sending blocking and completion callbacks as separated
1014 requests, adding them to a set and sending in parallel.
1017 Frequency : only for Cray XT3
1018 Bugzilla : 12829/13455
1019 Description: Changing primary group doesn't change the group lustre assigns to
1021 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1022 overridden with the primary group provided by the group upcall,
1023 whereas the supplied fsgid can be trusted if it is in the list of
1024 supplementary groups returned by the group upcall.
1026 Severity : enhancement
1028 Description: Allow masking D_WARNING, D_ERROR messages from console
1029 Details : Console messages can now be disabled via lnet.printk.
1033 Description: User code with malformed file open parameter crashes client node
1034 Details : Before packing join_file req, all the related reference should be
1035 checked carefully in case some malformed flags cause fake
1036 join_file req on client.
1040 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1041 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1042 mode is equal to the granted mode. However, it should also check
1043 that LDLM_FL_CP_REQD is not set.
1047 Description: Heavy nfs access might result in deadlocks
1048 Details : After ELC code landed, it is now improper to enqueue any mds
1049 locks under och_sem, because enqueue might want to decide to
1050 cancel open locks for same inode we are holding och_sem for.
1054 Description: Client eviction while running blogbench
1055 Details : A lot of unlink operations with concurrent I/O can lead to a
1056 deadlock causing evictions. To address the problem, the number of
1057 oustanding OST_DESTROY requests is now throttled to
1058 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1059 callbacks are processed in priority.
1063 Description: enable ACLs on MDS by default
1064 Details : ACLs must be enabled on MDS by default.
1067 Frequency : PPC/PPC64 only
1069 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1070 Details : fix duplicated definitions between asm-ppc64/types.h and
1071 lustre_types.h on PPC.
1074 Frequency : PPC/PPC64 only
1076 Description: asm-ppc/segment.h does not exist
1077 Details : fix compile issue on PPC.
1081 Description: better handle error messages in extents code
1084 Frequency : RHEL4 only
1086 Description: mkfs is very slow on IA64/RHEL4
1087 Details : A performance regression has been discovered in the MPT Fusion
1088 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1089 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1090 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1092 Severity : enhancement
1094 Description: SNMP support enhancement
1095 Details : Adding total number of sampled request for an MDS node in snmp
1098 Severity : enhancement
1100 Description: Optimize ldlm waiting list processing for PR extent locks
1101 Details : When processing waiting list for read extent lock and meeting read
1102 lock that is same or wider to it that is not contended, skip
1103 processing rest of the list and immediatelly return current
1104 status of conflictness, since we are guaranteed there are no
1105 conflicting locks in the rest of the list.
1109 Description: Time out and refuse to reconnect
1110 Details : When the failover node is the primary node, it is possible
1111 to have two identical connections in imp_conn_list. We must
1112 compare not conn's pointers but NIDs, otherwise we can defeat
1113 connection throttling.
1117 Description: port llog fixes from b1_6 into HEAD
1118 Details : Port llog reference couting and some llog cleanups from b1_6
1119 (bug 10800) into HEAD, for protect from panic and access to already
1120 free llog structures.
1124 Description: Detect stride IO mode in read-ahead
1125 Details : When a client does stride read, read-ahead should detect that and
1126 read-ahead pages according to the detected stride pattern.
1130 Description: data checksumming impacts single node performance
1131 Details : add support for several checksum algorithm. Currently, only CRC32
1132 and Adler-32 are supported. The checksum type can be changed on
1133 the fly via /proc/fs/lustre/osc/*/checksum_type.
1137 Description: use adler32 for page checksums
1138 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1143 Description: build for x2 fails
1144 Details : fix compile issue on Cray systems.
1148 Description: Properly match for duplicate locks
1149 Details : Due to different lock order from skiplists code, we need to
1150 traverse entire list for now
1153 Frequency : only on PPC/SLES10
1155 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1156 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1157 be defined. Add a hack in mkfs_lustre.c to work around this compile
1162 Description: LASSERT on MDS when client holding flock lock dies
1163 Details : ldlm pool logic depends on number of granted locks equal to
1164 number of released locks which is not true for flock locks, so
1165 just exclude such locks from consideration.
1169 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1170 Details : Use fsfilt_sync() for both the whole filesystem sync and
1171 individual file sync to eliminate dangerous inode locking
1172 with I_LOCK that can lead to a deadlock.
1176 Description: performance in 1.6.3
1177 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1178 sure that 1MB requests are not fragmented by the block layer.
1180 Severity : enhancement
1182 Description: organize the server-side client stats on per-nid basis
1183 Details : Change the structure of stats under obdfilter and mds to
1191 The "uuid"s file would list the uuids of _active_ exports.
1192 And the clear entry is to clear all stats and stale nids.
1194 Severity : enhancement
1196 Description: eliminate client locks in face of contention
1197 Details : file contention detection and lockless i/o implementation
1198 for contended files.
1202 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1205 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1207 Description: MDS may lose file striping (and hence file data) in some cases
1208 Details : If there are additional extended attributes stored on the MDS,
1209 in particular ACLs, SELinux, or user attributes (if user_xattr
1210 is specified for the client mount options) then there is a risk
1211 of attribute loss. Additionally, the Lustre file striping
1212 needs to be larger than default (e.g. striped over all OSTs),
1213 and an additional attribute must be stored initially in the
1214 inode and then increase in size enough to be moved to the
1215 external attribute block (e.g. ACL growing in size) for file
1220 Description: skiplist implementation simplification
1221 Details : skiplists are used to group compatible locks on granted list
1222 that was implemented as tracking first and last lock of each lock group
1223 the patch changes that to using doubly linked lists
1227 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1228 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1229 because LASSERT on both the data supplied by a client, and the data
1230 on disk is dangerous and incorrect.
1232 Severity : enhancement
1234 Description: Slow truncate/writes to huge files at high offsets.
1235 Details : Directly associate cached pages to lock that protect those pages,
1236 this allows us to quickly find what pages to write and remove
1237 once lock callback is received.
1241 Description: more ldlm soft lockups
1242 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1243 starve other threads from the resource lock for a long time in
1244 case of long waiting queue, so change the debug level from
1245 D_OTHER to the less frequently used D_INFO.
1247 Severity : enhancement
1249 Description: add -gid, -group, -uid, -user options to lfs find
1253 Description: Hung threads in invalidate_inode_pages2_range
1254 Details : The direct IO path doesn't call check_rpcs to submit a new RPC once
1255 one is completed. As a result, some RPCs are stuck in the queue
1260 Description: filter threads hungs on waiting journal commit
1261 Details : Cleanup filter group llog code, then only filter group llog will
1262 be only created in the MDS/OST syncing process.
1266 Description: Procfs and llog threads access destoryed import sometimes.
1267 Details : Sync the import destoryed process with procfs and llog threads by
1268 the import refcount and semaphore.
1270 Severity : enhancement
1272 Description: openlock cache of b1_6 port to HEAD
1277 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1278 Details : If insertion of an extent fails, then discard the inode
1279 preallocation and free data blocks else it can lead to duplicate
1284 Description: don't always update ctime in ext3_xattr_set_handle()
1285 Details : Current xattr code updates the inode ctime in ext3_xattr_set_handle.
1286 In some cases the ctime should not be updated, for example for
1287 2.0->1.8 compatibility it is necessary to delete an xattr and it
1288 should not update the ctime.
1292 Bugzilla : 15713/16362
1293 Description: Assertion in iopen_connect_dentry in 1.6.3
1294 Details : looking up an inode via iopen with the wrong generation number can
1295 populate the dcache with a disconneced dentry while the inode
1296 number is in the process of being reallocated. This causes an
1297 assertion failure in iopen since the inode's dentry list contains
1298 both a connected and disconnected dentry.
1302 Description: assertion failure in ldlm_handle2lock()
1303 Details : fix a race between class_handle_unhash() and class_handle2object()
1304 introduced in lustre 1.6.5 by bug 13622.
1309 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1310 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1315 Description: oops in sd_iostats_seq_show()
1316 Details : unloading/reloading the scsi low level driver triggers a kernel
1317 bug when trying to access the sd iostat file.
1322 Description: Kernel panics during QLogic driver reload
1323 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1324 patch, causing memory corruption.
1329 Description: journal_dev option does not work in b1_6
1330 Details : pass mount option during pre-mount.
1332 Severity : enhancement
1334 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1335 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1336 extent information of a file. It can be used to map logical blocks
1337 in a file to physical blocks in the block device.
1341 Description: LDLM soft lockups - improvement
1342 Details : It is be possible to send the lock handle along with each read
1343 or write request because the client is already doing a lock match
1344 itself so there isn't any reason the OST should have to re-do that
1347 --------------------------------------------------------------------------------
1349 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
1351 * Support for kernels:
1352 2.6.5-7.283 (SLES 9),
1353 2.6.9-55.EL (RHEL 4),
1354 2.6.16.46-0.14 (SLES 10),
1355 2.6.18.8 vanilla (kernel.org)
1356 * Client support for unpatched kernels:
1357 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1358 2.6.16 - 2.6.22 vanilla (kernel.org)
1359 * Due to recently discovered recovery problems, we do not recommend
1360 using patchless RHEL 4 clients with this or any earlier release.
1361 * Recommended e2fsprogs version: 1.39.cfs8
1362 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1363 * Starting with this release, the ldiskfs backing filesystem required
1364 by Lustre is now in its own package, lustre-ldiskfs. This package
1365 should be installed. It is versioned separately from Lustre and
1366 may be released separately in future.
1370 Description: block reactivating mgc import until all deactivates complete
1371 Details : Fix race when failing back MDT/MGS to itself (testing)
1374 Frequency : at statup only
1376 Description: mds_lov_synchronize race leads to various problems
1377 Details : simultaneous MDT->OST connections at startup can cause the
1378 sync to abort, leaving the OSC in a bad state.
1380 Severity : enhancement
1382 Description: add optional extra BUILD_VERSION info
1383 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1384 to override the lustre version.
1387 Frequency : 2.6.18 servers only
1389 Description: ll_kern_mount() doesn't release the module reference
1390 Details : The ldiskfs module reference count never drops down to 0
1391 because ll_kern_mount() doesn't release the module reference.
1396 Description: server LBUG when using old ost_num_threads parameter
1397 Details : Accept the old ost_num_threads parameter but warn that it
1398 is deprecated, and fix an off-by-one error that caused an LBUG.
1403 Description: Transient SCSI error results in persistent IO issue
1404 Details : iobuf->dr_error is not reinitialized to 0 between two
1408 Frequency : sometimes when underlying device returns I/O errors
1410 Description: OSTs not going read-only during write failures
1411 Details : OSTs are not remounted read-only when the journal commit threads
1412 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1413 instead of the ext3 wrappers.
1417 Description: poor connect scaling with increasing client count
1418 Details : Don't run filter_grant_sanity_check for more than 100 exports
1419 to improve scaling for large numbers of clients.
1422 Frequency : SLES10 only
1424 Description: sanity-quota.sh quotacheck failed: rc = -22
1425 Details : Quotas cannot be enabled on SLES10.
1428 Frequency : liblustre clients only
1430 Description: getdirentries does not give error when run on compute nodes
1431 Details : getdirentries does not fail when the size specified as an argument
1432 is too small to contain at least one entry
1434 Severity : enhancement
1436 Description: Add LNET router traceability for debug purposes
1437 Details : If a checksum failure occurs with a router as part of the
1438 IO path, the NID of the last router that forwarded the bulk data
1439 is printed so it can be identified.
1444 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1445 Details : Due to a race condition, liblustre clients were occasionally
1446 evicted incorrectly.
1448 Severity : enhancement
1450 Description: lfs setstripe use optional parameters instead of postional
1453 Severity : enhancement
1455 Description: Nanosecond timestamp support for ldiskfs
1456 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1457 resolution timestamps. There is not yet support for this at
1458 the Lustre filesystem level.
1461 Frequency : during server recovery
1463 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1464 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1468 Frequency : nfs export on patchless client
1470 Description: connectathon hang when test nfs export over patchless client
1471 Details : Disconnected dentry cannot be found with lookup, so we do not need
1472 to unhash it or make it invalid
1475 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1476 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1477 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1482 Description: open req refcounting wrong on reconnect
1483 Details : If reconnect happened between getting open reply from server and
1484 call to mdc_set_replay_data in ll_file_open, we will schedule
1485 replay for unreferenced request that we are about to free.
1486 Subsequent close will crash in variety of ways.
1487 Check that request is still eligible for replay in
1488 mdc_set_replay_data().
1493 Description: disable writes to filesystem when reading health_check file
1494 Details : the default for reading the health_check proc file has changed
1495 to NOT do a journal transaction and write to disk, because this
1496 can cause reads of the /proc file to hang and block HA state
1497 checking on a healthy but otherwise heavily loaded system. It
1498 is possible to return to the previous behaviour during configure
1499 with --enable-health-write.
1501 Severity : enhancement
1503 Description: 64-bit inode version
1504 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1505 to the inode. This will be required for version-based recovery.
1510 Description: MDS fails to start if a duplicate client export is detected
1511 Details : in some rare cases it was possible for a client to connect to
1512 an MDS multiple times. Upon recovery the MDS would detect this
1513 and fail during startup. Handle this more gracefully.
1515 Severity : enhancement
1517 Description: Add -o localflock option to simulate old noflock
1519 Details : This will achieve local-only flock/fcntl locks
1525 Description: log_commit_thread vs filter_destroy race leads to crash
1526 Details : Take import reference before releasing llog record semaphore
1531 Description: Wrong request locking in request set processing
1532 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1533 imp_delayed_list, in this place should be used imp_lock.
1536 Frequency : when reconnection
1538 Description: Grant Leak when osc reconnect to OST
1539 Details : When osc reconnect ost, OST(filter) should check whether it
1540 should grant more space to client by comparing fed_grant and
1541 cl_avail_grant, and return the granted space to client instead
1542 of "new granted" space, because client will call osc_init_grant
1543 to update the client grant space info.
1546 Frequency : when client reconnect to OST
1548 Description: Grant Leak when osc do resend and replay bulk write
1549 Details : When osc reconnect to OST, OST(filter)should clear grant info of
1550 bulk write request, because the grant info while be sync between
1551 OSC and OST when reconnect, and we should ignore the grant info
1552 these of resend/replay write req.
1557 Description: Grant space more than avaiable left space sometimes.
1558 Details : When then OST is about to be full, if two bulk writing from
1559 different clients came to OST. Accord the avaliable space of the
1560 OST, the first req should be permitted, and the second one
1561 should be denied by ENOSPC. But if the seconde arrived before
1562 the first one is commited. The OST might wrongly permit second
1563 writing, which will cause grant space > avaiable space.
1566 Frequency : when client is evicted
1568 Description: Grant might be wrongly erased when osc is evicted by OST
1569 Details : when the import is evicted by server, it will fork another
1570 thread ptlrpc_invalidate_import_thread to invalidate the
1571 import, where the grant will be set to 0. While the original
1572 thread will update the grant it got when connecting. So if
1573 the former happened latter, the grant will be wrongly errased
1574 because of this race.
1579 Description: Checking Stale with correct fid
1580 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1581 to check whether it is stale, because sometimes, we want the
1582 enqueue happened anyway, and op_data.fid2 will not be initialized.
1584 Severity : enhancement
1586 Description: update patchless client
1587 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1590 Frequency : only with 2.4 kernel
1592 Description: random memory corruption
1593 Details : size of struct ll_inode_info is to big for union inode.u and this
1594 can be cause of random memory corruption.
1599 Description: Memory leak in recovery
1600 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1601 It should also check obd_fail before fsfilt_start, otherwise if
1602 fsfilt_start return -EROFS,(failover mds during mds recovery).
1603 then the req will return with repmsg->transno = 0 and rc = EROFS.
1604 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1605 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1606 be freed no matter whether fsfilt_commit success or not.
1609 Frequency : only with huge count clients
1611 Description: Prevents from taking the superblock lock in llap_from_page for
1613 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1614 for llap_from_page call in ll_removepage prevents from taking the
1615 superblock lock for a soon died page.
1620 Description: Not check open intent error before release open handle
1621 Details : in some rare cases, the open intent error is not checked before
1622 release open handle, which may cause
1623 ASSERTION(open_req->rq_transno != 0), because it tries to release
1624 the failed open handle.
1629 Description: Set cat log bitmap only after create log success.
1630 Details : in some rare cases, the cat log bitmap is set too early. and it
1631 should be set only after create log success.
1635 Description: Accessing a block bevice can re-enable I/O when Lustre is
1636 tearing down a device.
1637 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1642 Description: service threads may hog cpus when there are a lot of requests
1643 Details : Insert cond_resched to give other threads a chance to use some CPU
1648 Description: the cat log was not initialized in recovery
1649 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1650 unlink log on mds will not be initialized until mds post
1651 recovery. And also in mds post recovery, the unlink log will
1652 initialization will be done asynchronausly, so there will be race
1653 between add unlink log and unlink log initialization.
1657 Description: brw_stats were being printed incorrectly
1658 Details : brw_stats were being printed as log2 but all of them were not
1659 recorded as log2. Also remove some code duplication arising from
1660 filter_tally_{read,write}.
1664 Frequency : rare, only in recovery.
1665 Description: ASSERTION(req->rq_type != LI_POISON) failed
1666 Details : imp_lock should be held while iterating over imp_sending_list for
1667 prevent destroy request after get timeout in ptlrpc_queue_wait.
1671 Description: replay-single.sh test 52 fails
1672 Details : A lock's skiplist need to be cleanup when it being unlinked
1673 from its resource list.
1677 Description: Short directio read returns full requested size rather than
1679 Details : Direct I/O operations should return actual amount of bytes
1680 transferred rather than requested size.
1682 Severity : enhancement
1684 Description: metadata RPC reduction (e.g. for rm performance)
1685 Details : decrease the amount of synchronous RPC between clients and servers
1686 by canceling conflicing lock before the operation on the client
1687 and packing thier handles into the main operation RPC to server.
1689 Severity : enhancement
1691 Description: Async OSC create to avoid the blocking unnecessarily.
1692 Details : If a OST has no remain object, system will block on the creating
1693 when need to create a new object on this OST. Now, ways use
1694 pre-created objects when available, instead of blocking on an
1695 empty osc while others are not empty. If we must block, we block
1696 for the shortest possible period of time.
1698 Severity : enhancement
1700 Description: refine locking for avoid write wrong info into lov_objid file
1701 Details : fix possible races with add new target and write/update data in
1704 --------------------------------------------------------------------------------
1706 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1711 Frequency : on some architectures
1713 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1714 Details : Uninitialized flags sometimes cause configuration commands to
1718 Frequency : patchless clients only
1720 Description: missing __iget() symbol export
1721 Details : The __iget() symbol export is missing. To avoid the need for
1722 this on patchless clients the deathrow inode reaper is turned
1723 off, and we depend on the VM to clean up old inodes. This
1724 dependency was during via the fix for bug 12181.
1726 --------------------------------------------------------------------------------
1728 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1730 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1731 INTEROPERATE with older versions automatically. In many cases a
1732 special upgrade step is needed. Please read the
1733 user documentation before upgrading any part of a live system.
1734 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1735 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1736 * WARNING: Lustre configuration and startup changes are required with
1737 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1743 Severity : enhancement
1745 Description: MountConf
1746 Details : Lustre configuration is now managed via mkfs and mount
1747 commands instead of lmc and lconf. New obd types (MGS, MGC)
1748 are added for dynamic configuration management. See
1749 https://mail.clusterfs.com/wikis/lustre/MountConf for
1752 Severity : enhancement
1754 Description: dynamic OST addition
1755 Details : OSTs can now be added to a live filesystem
1757 Severity : enhancement
1759 Description: startup order invariance
1760 Details : MDTs and OSTs can be started in any order. Clients only
1761 require the MDT to complete startup.
1763 Severity : enhancement
1765 Description: parallel, asynchronous orphan cleanup
1766 Details : orphan cleanup is now performed in separate threads for each
1767 OST, allowing parallel non-blocking operation.
1769 Severity : enhancement
1771 Description: optimized stripe assignment
1772 Details : stripe assignments are now made based on ost space available,
1773 ost previous usage, and OSS previous usage, in order to try
1774 to optimize storage space and networking resources.
1776 Severity : enhancement
1778 Description: Permanently set tunables
1779 Details : All writable /proc/fs/lustre tunables can now be permanently
1780 set on a per-server basis, at mkfs time or on a live system.
1782 Severity : enhancement
1784 Description: Lustre message v2
1785 Details : Add lustre message format v2.
1787 Severity : enhancement
1789 Description: client OST exclusion list
1790 Details : Clients can be started with a list of OSTs that should be
1791 declared "inactive" for known non-responsive OSTs.
1795 Description: SPEC SFS validation failure on NFS v2 over lustre.
1796 Details : Changes the blocksize for regular files to be 2x RPC size,
1797 and not depend on stripe size.
1799 Severity : enhancement
1801 Description: Multiple MD RPCs in flight.
1802 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1803 To avoid overly-overloading MDS, introduce a limit on number of
1804 MDS RPCs in flight for a single client and add /proc controls
1805 to adjust this limit.
1807 Severity : enhancement
1809 Description: client read/write statistics
1810 Details : Add client read/write call usage stats for performance
1811 analysis of user processes.
1812 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1813 file access. extents_stats shows chunk size distribution.
1814 extents_stats_per_process show chunk size distribution per
1817 Severity : enhancement
1819 Description: mds statistics
1820 Details : Add detailed mds operations statistics in
1821 /proc/fs/lustre/mds/*/stats.
1825 Description: Failure of copying files with lustre special EAs.
1826 Details : Client side always return success for setxattr call for lustre
1827 special xattr (currently only "trusted.lov").
1832 Description: Refcount LNET uuids
1833 Details : The global LNET uuid list grew linearly with every startup;
1834 refcount repeated list entries instead of always adding to
1837 Severity : enhancement
1839 Description: Dynamic service threads
1840 Details : Within a small range, start extra service threads
1841 automatically when the request queue builds up.
1844 Frequency : mixed-endian client/server environments
1846 Description: mixed-endian crashes
1847 Details : The new msg_v2 system had some failures in mixed-endian
1850 Severity : enhancement
1852 Description: Easy OST removal
1853 Details : OSTs can be permanently deactivated with e.g. 'lctl
1854 conf_param lustre-OST0001.osc.active=0'
1856 Severity : enhancement
1858 Description: MGS proc entries
1859 Details : Added basic proc entries for the MGS showing what filesystems
1862 Severity : enhancement
1864 Description: provide MGS failover
1865 Details : Added config lock reacquisition after MGS server failover.
1867 Severity : enhancement
1869 Description: add Linux 2.4 support
1870 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1874 Description: a large application tries to do I/O to the same resource and dies
1875 in the middle of it.
1876 Details : Check the req->rq_arrival time after the call to
1877 ost_brw_lock_get(), but before we do anything about
1878 processing it & sending the BULK transfer request. This
1879 should help move old stale pending locks off the queue as
1880 quickly as obd_timeout.
1883 Frequency : when an incorrect nid is specified during startup
1885 Description: ptlrpc connect to non-existant node causes kernel crash
1886 Details : LNET can't be re-entered from an event callback, which
1887 happened when we expire a message after the export has been
1888 cleaned up. Instead, hand the zombie cleanup off to another
1891 Severity : enhancement
1893 Description: plain/inodebits lock performance improvement
1894 Details : Grouping plain/inodebits in granted list by their request modes
1895 and bits policy, thus improving the performance of search through
1899 Frequency : only if OST filesystem is corrupted
1901 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1902 Details : for a short time RPCs with bulk IO are in the replay list,
1903 but replay of bulk IOs is unimplemented. If the OST filesystem
1904 is corrupted due to disk cache incoherency and then replay is
1905 started it is possible to trip an assertion. Avoid putting
1906 committed RPCs into the replay list at all to avoid this issue.
1909 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1911 Description: System hang on startup
1912 Details : This bug allowed the liblustre (e.g. catamount) client to
1913 return to the app before handling all startup RPCs. This
1914 could leave the node unresponsive to lustre network traffic
1915 and manifested as a server ptllnd timeout.
1917 Severity : enhancement
1919 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1920 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1921 boolean to control whether to print peer debug info when a
1922 client's RPC times out.
1925 Frequency : only for kernels with patches from Lustre below 1.4.3
1927 Description: Remove old rdonly API
1928 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1931 Frequency : only for devices with external journals
1933 Description: Set external device read-only also
1934 Details : During a commanded failover stop, we set the disk device
1935 read-only while the server shuts down. We now also set any
1936 external journal device read-only at the same time.
1939 Frequency : when upgrading from 1.4 while trying to change parameters
1941 Description: The wrong (new) MDC name was used when setting parameters for
1942 upgraded MDT's. Also allows changing of OSC (and MDC)
1943 parameters if --writeconf is specified at tunefs upgrade time.
1946 Frequency : when setting specific ost indicies
1948 Description: QOS code breaks on skipped indicies
1949 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1950 created with --index need not be sequential.
1952 Severity : enhancement
1954 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1955 Details : The uninit_groups feature works in conjunction with the kernel
1956 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1957 up the pass1 processing of e2fsck. This is a read-only feature
1958 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1959 mount filesystems that have had this feature enabled.
1961 Severity : enhancement
1963 Description: Improve multi-block allocation algorithm to avoid fragmentation
1964 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1965 allocation locality and avoid filesystem fragmentation.
1967 ------------------------------------------------------------------------------
1969 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1971 * Support for kernels:
1972 2.4.21-47.0.1.EL (RHEL 3)
1973 2.6.5-7.283 (SLES 9)
1974 2.6.9-42.0.10.EL (RHEL 4)
1975 2.6.12.6 vanilla (kernel.org)
1976 2.6.16.27-0.9 (SLES 10)
1977 * Recommended e2fsprogs version: 1.39.cfs5
1979 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1983 Frequency : occasional, depends on client load and configuration
1984 Bugzilla : 12181, 12203
1985 Description: data loss for recently-modified files
1987 Details : In some cases it is possible that recently written or created
1988 files may not be written to disk in a timely manner (this should
1989 normally be within 30s unless client IO load is very high).
1990 The problem appears as zero-length files or files that are a
1991 multiple of 1MB in size after a client crash or client eviction
1992 that are missing data at the end of the file.
1994 This problem is more likely to be hit on clients where files are
1995 repeatedly created and unlinked in the same directory, clients
1996 have a large amount of RAM, have many CPUs, the filesystem has
1997 many OSTs, the clients are rebooted frequently, and/or the files
1998 are not accessed by other nodes after being written.
2000 The presence of the problem can be detected by looking at
2001 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2002 smaller than the second (nr_unused) then dirty files will not
2003 be flushed automatically to disk. "sync; sleep 10" should be
2004 run several times on the node before unmounting it to update
2005 Lustre (this is also safe to run on nodes without this problem).
2007 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2008 that can cause this same problem, so customers using that kernel
2009 also need to update the kernel in addition to Lustre. In order
2010 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2012 It is normal that files written just before a client crash (less
2013 than 30s) may not yet have been flushed to disk, even for local
2017 Frequency : frequent on thin XT3 nodes
2019 Description: UUID collision on thin XT3 Linux nodes
2020 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2021 randomly, since we relied on an insufficiently-seeded PRNG.
2026 Description: OSS hangs after "All ost request buffers busy"
2027 Details : A deadlock between quota and journal operations caused OSS
2028 hangs after printing "All ost request buffers busy."
2031 Frequency : always on liblustre builds
2033 Description: Cleanup compiler warnings on liblustre
2036 Frequency : always on liblustre builds on XT3
2038 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2042 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2043 > 32000 subdirectories
2044 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2045 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2046 subdirectory count crosses 32000. This will aid e2fsck to
2047 correctly handle more than 32000 subdirectories.
2050 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2052 Description: System hang on startup
2053 Details : This bug allowed the liblustre (e.g. catamount) client to
2054 return to the app before handling all startup RPCs. This
2055 could leave the node unresponsive to lustre network traffic
2056 and manifested as a server ptllnd timeout.
2058 Severity : enhancement
2060 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2061 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2062 boolean to control whether to print peer debug info when a
2063 client's RPC times out.
2068 Description: make O_SYNC working on 2.6 kernels
2069 Details : 2.6 kernels use different method for mark pages for write,
2070 so need add a code to lustre for O_SYNC work.
2075 Description: Failure to close file and release space on NFS
2076 Details : Put inode details into lock acquired in ll_intent_file_open.
2077 Use mdc_intent_lock in ll_intent_open to properly
2078 detect all kind of errors unhandled by mdc_enqueue.
2083 Description: proc file read during shutdown sometimes raced obd removal,
2085 Details : Add lock to prevent obd access after proc file removal.
2088 Frequency : Only for files larger than 4GB on 32-bit clients.
2090 Description: improperly doing page alignment of locks
2091 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2092 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2097 Description: RPCs being resent when they shouldn't be
2098 Details : Some RPCs that should not be resent are being resent. This
2099 can cause inconsistencies in the RPC state machine. Do not
2100 resend such requests.
2103 Frequency : rare, only with NFS export
2105 Description: Crash on NFS re-export node
2106 Details : under very unusual load conditions an assertion is hit in
2107 ll_intent_file_open()
2110 Frequency : only if OST filesystem is corrupted
2112 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2113 Details : for a short time RPCs with bulk IO are in the replay list,
2114 but replay of bulk IOs is unimplemented. If the OST filesystem
2115 is corrupted due to disk cache incoherency and then replay is
2116 started it is possible to trip an assertion. Avoid putting
2117 committed RPCs into the replay list at all to avoid this issue.
2122 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2123 Details : Large single O_DIRECT read and write calls can fail to allocate
2124 a sufficiently large buffer to process the request. In case of
2125 allocation failure the allocation is retried with a smaller
2126 buffer and broken into smaller requests.
2128 Severity : enhancement
2130 Description: Add -o localflock option to simulate old noflock behaviour.
2131 Details : This will achieve local-only flock/fcntl locks coherentness.
2136 Description: versioning check is incomplete
2137 Details : Checking the version difference of client vs. server, report
2138 error if the gap is too big.
2143 Description: add support PG_writeback bit
2144 Details : add support for PG_writeback bit for Lustre, for more carefull
2145 work with page cache in 2.6 kernel. This also fix some deadlocks
2146 and remove hack for work O_SYNC with 2.6 kernel.
2148 Severity : enhancement
2150 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2151 Details : The uninit_groups feature works in conjunction with the kernel
2152 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2153 up the pass1 processing of e2fsck. This is a read-only feature
2154 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2155 mount filesystems that have had this feature enabled.
2157 Severity : enhancement
2159 Description: Improve multi-block allocation algorithm to avoid fragmentation
2160 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2161 allocation locality and avoid filesystem fragmentation.
2163 ------------------------------------------------------------------------------
2165 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2167 * Support for kernels:
2168 2.6.9-42.0.3.EL (RHEL 4)
2169 2.6.5-7.276 (SLES 9)
2170 2.4.21-47.0.1.EL (RHEL 3)
2171 2.6.12.6 vanilla (kernel.org)
2172 2.6.16.21-0.8 (SLES10)
2173 * Recommended e2fsprogs version: 1.39.cfs2-0
2175 * The backwards-compatible /proc/sys/portals symlink has been removed
2176 in this release. Before upgrading, please ensure that you change
2177 any configuration scripts or /etc/sysctl.conf files that access
2178 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2179 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2180 in advance of the upgrade on any system running Lustre 1.4.6 or
2181 newer, since /proc/sys/lnet was added in that version.
2182 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2186 Frequency : only when quota is used
2188 Description: avoid scanning export list for quota master
2189 Details : Change the algorithms to avoid scanning export list in order
2190 to improve the efficiency.
2193 Frequency : MDS failover only, very rarely
2195 Description: "went back in time" messages on mds failover
2196 Details : The greatest transno may be lost when the current operation
2197 finishes with an error (transno==0) and the client's last_rcvd
2198 record is over-written. Save the greatest transno in the
2199 mds_last_transno for this case.
2202 Frequency : always for specific kernels and striping counts
2204 Description: client may get "Matching packet too big" without ACL support
2205 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2206 when trying to access files in certain configurations. The
2207 clients should in fact be denied when mounting because they do
2208 not understand ACLs.
2211 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2213 Description: many clients connecting with IO in progress causes connect timeouts
2214 Details : Avoid synchronous journal commits to avoid delays caused by many
2215 clients connecting/disconnecting when bulk IO is in progress.
2216 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2217 OST_IO_PORTAL to avoid delays behind potentially many pending
2221 Frequency : occasionally with multiple writers to a single file
2223 Description: shared writes to file may result in wrong size reported by stat()
2224 Details : Allow growing of kms when extent lock is cancelled
2227 Frequency : always with random mmap IO to multi-striped file
2229 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2230 Details : Only if the hole is at the end of OST object so that kms is too
2231 small. Fix is to increase kms accordingly in ll_nopage.
2234 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2236 Description: writes to a missing object would leak memory on the OST
2237 Details : If there is an inconsistency between the MDS and OST filesystems,
2238 such that the MDS references an object that doesn't exist, writes
2239 to that object will leak memory due to incorrect cleanup in the
2240 error handling path, eventually running out of memory on the OST.
2245 Description: Creating too long symlink causes lustre errors
2246 Details : Check symlink and name lengths before sending requests to MDS.
2249 Frequency : only if flock is enabled (not on by default)
2251 Description: posix locks not released on fd closure on 2.6.9+
2252 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2253 kernels, this caused such locks not to be released on fd close and
2254 then assertions on fs unmount about still used locks.
2257 Frequency : MDS failover only, very rarely
2259 Description: clients may get ASSERTION(granted_lock != NULL)
2260 Details : When request was taking a long time, and a client was resending
2261 a getattr by name lock request. The were multiple lock requests
2262 with the same client lock handle and
2263 mds_getattr_name->fixup_handle_for_resent_request found one of the
2264 lock handles but later failed with ASSERTION(granted_lock != NULL).
2269 Description: handle->h_buffer_credits > 0, assertion failure
2270 Details : h_buffer_credits is zero after truncate, causing assertion
2271 failure. This patch extends the transaction or creates a new
2275 Frequency : NFS re-export or patchless client
2276 Bugzilla : 11179, 10796
2277 Description: Crash on NFS re-export node (__d_move)
2278 Details : We do not want to hash the dentry if we don't have a lock.
2279 But if this dentry is later used in d_move, we'd hit uninitialised
2280 list head d_hash, so we just do this to init d_hash field but
2281 leave dentry unhashed.
2284 Frequency : NFS re-export or patchless client
2286 Description: NFS exports has problem with symbolic link
2287 Details : lustre client didn't properly install dentry when re-exported
2288 to NFS or running patchless client.
2291 Frequency : NFS re-export or patchless client
2293 Description: Various nfs/patchless fixes.
2294 Details : fixes reuse disconected alias for lookup process - this fixes
2295 warning "find_exported_dentry: npd != pd",
2296 fix permission error with open files at nfs.
2297 fix apply umask when do revalidate.
2300 Frequency : occasional
2302 Description: Crash on NFS re-export node
2303 Details : calling clear_page() on the wrong pointer triggered oops in
2304 generic_mapping_read().
2307 Frequency : rarely, using O_DIRECT IO
2309 Description: unaligned directio crashes client with LASSERT
2310 Details : check for unaligned buffers before trying any requests.
2313 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2315 Description: stale data returned from RAID cache
2316 Details : If only a small amount of IO is done to the RAID device before
2317 reading it again it is possible to get stale data from the RAID
2318 cache instead of reading it from disk.
2321 Frequency : always for sles10 kernel
2323 Description: sles10 support
2324 Details : ll_follow_link: compile fixes and using of nd_set_link
2325 under newer kernels.
2328 Frequency : depends on arch, kernel and compiler version, always on sles10
2331 Description: recursive or deep enough symlinks cause stack overflow
2332 Details : getting rid of large stack-allocated variable in
2336 Frequency : depends on hardware
2338 Description: lustre write performance loss in the SLES10 kernel
2339 Details : the performance loss is caused by using of write barriers in the
2340 ext3 code. The SLES10 kernel turns barrier support on by
2341 default. The fix is to undo that change for ldiskfs.
2343 ------------------------------------------------------------------------------
2345 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2347 * Support for kernels:
2348 2.6.9-42.0.3EL (RHEL 4)
2349 2.6.5-7.276 (SLES 9)
2350 2.4.21-47.0.1.EL (RHEL 3)
2351 2.6.12.6 vanilla (kernel.org)
2355 Frequency : quota enabled and large files being deleted
2357 Description: releasing more than 4GB of quota at once hangs OST
2358 Details : If a user deletes more than 4GB of files on a single OST it
2359 will cause the OST to spin in an infinite loop. Release
2360 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2365 Description: statfs data retrieved from /proc may be stale or zero
2366 Details : When reading per-device statfs data from /proc, in the
2367 {kbytes,files}_{total,free,avail} files, it may appear
2368 as zero or be out of date.
2371 Frequency : systems with MD RAID1 external journal devices
2373 Description: lconf's call to blkid is confused by RAID1 journal devices
2374 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2375 of returning all block devices that match the journal UUID.
2378 Frequency : always, for aggregate stripe size over 4GB
2380 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2381 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2382 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2385 Frequency : always if "lfs find" used on a local file/directory
2387 Description: "lfs find" segfaults if used on a local file/directory
2388 Details : The case where a directory component was not specified wasn't
2389 handled correctly. Handle this properly.
2392 Frequency : always on ppc64
2394 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2395 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2396 perform endianness conversion.
2399 Frequency : rarely (truncate to non-zero file size after write under load)
2400 Bugzilla : 10730, 10687
2401 Description: Files padded with zeros to next 4K multiple
2402 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2403 it is possible that files that are truncated to a non-zero size
2404 immediately after being written are filled with zero bytes beyond
2405 the truncated size. No file data is lost.
2407 Severity : enhancement
2409 Description: Allow recovery/failover for liblustre clients.
2410 Details : liblustre clients were unaware of failover configurations until
2413 Severity : enhancement
2415 Description: user file locks should fail when not mounting with flock option
2416 Details : Set up an error-returning stub in ll_file_operations.lock field
2417 to prevent incorrect behaviour when client is mounted without
2418 flock option. Also, set up properly f_op->flock field for
2422 Frequency : always on ia64
2424 Description: "lfs df" loops on printing out MDS statfs information
2425 Details : The obd_ioctl_data was not initialized and in some systems
2426 this caused a failure during the ioctl that did not return
2427 an error. Initialize the struct and return an error on failure.
2430 Frequency : SLES 9 only
2432 Description: Error of copying files with lustre special EAs as root
2433 Details : Client side always return success for setxattr call for lustre
2434 special xattr (currently only "trusted.lov").
2437 Frequency : rarely on clusters with both ia64+i386 clients
2439 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2440 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2441 clients are doing concurrent writes to the same file it is
2442 possible that the ia64 clients may LASSERT because the OST
2443 extent locks are not PAGE_SIZE aligned. Ensure that grown
2444 locks are always aligned on the request boundary.
2447 Frequency : specific use, occasional
2449 Description: Overwriting in use executable truncates on-disk binary image
2450 Details : If one node attempts to overwrite an executable in use by
2451 another node, we now correctly return ETXTBSY instead of
2452 truncating the file.
2454 Severity : enhancement
2456 Description: Async OSC create to avoid the blocking unnecessarily.
2457 Details : If a OST has no remain object, system will block on the creating
2458 when need to create a new object on this OST. Now, ways use
2459 pre-created objects when available, instead of blocking on an
2460 empty osc while others are not empty. If we must block, we block
2461 for the shortest possible period of time.
2466 Description: chmod on Lustre root is propagated to other clients
2467 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2468 invalid by the follow_mount time.
2471 Frequency : liblustre clients only
2473 Description: Race in 'instant cancel' lock handling could lead to such locks
2474 never to be granted in case of SMP MDS
2475 Details : Do not destroy not yet granted but cbpending locks in
2479 Frequency : replay/resend of open
2481 Description: non null lock assetion failure in mds_intent_policy
2482 Details : Trying to replay/resend lockless open requests resulted in
2483 mds_open() returning 0 with no lock. Now it sets a flag if
2484 a lock is going to be returned.
2486 Severity : enhancement
2488 Description: Checksum enhancements
2489 Details : New checksum enhancements allow for resending RPCs that failed
2492 Severity : enhancement
2494 Description: Tunables on number of dirty pages in cacche
2495 Details : Allow to set limit on number of dirty pages cached.
2500 Description: client crash on unmount - lock still has references
2501 Details : In some error handling cases it was possible to leak a lock
2502 reference on a client while accessing a file. This was not
2503 harmful to the client during operation, but would cause the
2504 client to crash when the filesystem is unmounted.
2507 Frequency : specific case, rare
2509 Description: ETXTBSY on mds though file not in use by client
2510 Details : ETXTBSY is no longer incorrectly returned when attempting to
2511 chmod or chown a directory that the user previously tried to
2512 execute or a currently-executing binary.
2515 Frequency : extremely rare except on liblustre-based clients
2517 Description: Lustre space not freed when files are deleted
2518 Details : Clean up open-unlinked files after client eviction. Previously
2519 the unlink was skipped and the files remained as orphans.
2524 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2525 Details : In some cases it was possible to send a blocking callback to
2526 a client doing a glimpse, even though that client didn't get
2527 a lock granted. When the glimpse lock is cancelled on the OST
2528 the freed lock is left on the waiting list and corrupted the list.
2531 Frequency : all core dumps
2533 Description: Broke core dumps to lustre
2534 Details : Negative dentry may be unhashed if parent does not have UPDATE
2535 lock, but some callers, e.g. do_coredump, expect dentry to be
2536 hashed after successful create, hash it in ll_create_it.
2538 ------------------------------------------------------------------------------
2540 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2542 * Support for kernels:
2543 2.6.9-42.0.2.EL (RHEL 4)
2544 2.6.5-7.276 (SLES 9)
2545 2.4.21-40.EL (RHEL 3)
2546 2.6.12.6 vanilla (kernel.org)
2550 Frequency : always on RHEL 3
2552 Description: Number of open files grows over time
2553 Details : The number of open files grows over time, whether or not
2554 Lustre is started. This was due to a filp leak introduced
2555 by one of our kernel patches.
2557 ------------------------------------------------------------------------------
2559 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2561 * Support for kernels:
2562 2.6.9-42.EL (RHEL 4)
2563 2.6.5-7.276 (SLES 9)
2564 2.4.21-40.EL (RHEL 3)
2565 2.6.12.6 vanilla (kernel.org)
2570 Bugzilla : 5719, 9635, 9792, 9684
2571 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2572 Details : If a server is under heavy load and cannot reply to new
2573 connection requests before the client resends the (re)connect,
2574 the connection handling code can behave badly if two service
2575 threads are concurrently handing separate (re)connections from
2576 the same client. Add better locking to the connection handling
2577 code, and ensure that only a single connection will be processed
2578 for a given client UUID, even if the lock is dropped.
2580 Severity : enhancement
2582 Description: add TCP zero-copy support to kernel
2583 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2584 sends if the hardware supports scatter-gather and checksumming.
2585 This allows socklnd to do client-write and server-read more
2586 efficiently and reduce CPU utilization from skbuf copying.
2589 Frequency : only if NFS exporting from client
2591 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2592 Details : If setting attributes on a file created under NFS that had
2593 never been opened it would be possible to oops the client
2594 if the file had no objects.
2597 Frequency : always for liblustre
2599 Description: liblustre client does MDS+OSTs setattr RPC for each write
2600 Details : When doing a write from a liblustre client, the client
2601 incorrectly issued an RPC to the MDS and each OST the file was
2602 striped over in order to update the timestamps. When writing
2603 with small chunks and many clients this could overwhelm the MDS
2604 with RPCs. In all cases it would slow down the write because
2605 these RPCs are unnecessary.
2607 Severity : enhancement
2609 Description: allow number of MDS service threads to be changed at module load
2610 Details : It is now possible to change the number of MDS service threads
2611 running. Adding "options mds mds_num_threads={N}" to the MDS's
2612 /etc/modprobe.conf will set the number of threads for the next
2613 time Lustre is restarted (assuming the "mds" module is also
2614 reloaded at that time). The default number of threads will
2615 stay the same, 32 for most systems.
2620 Description: OST crash if filesystem is unformatted or corrupt
2621 Details : If an OST is started on a device that has never been formatted
2622 or if the filesystem is corrupt and cannot even mount then the
2623 error handling cleanup routines would dereference a NULL pointer.
2628 Description: NULL pointer deref in llap_from_page.
2629 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2630 incorrect) mapping if the page was truncated/reclaimed while it was
2631 searched for. Check for this condition and skip such pages when
2632 doing readahead. Introduce extra check to llap_from_page() to
2633 verify page->mapping->host is non-NULL (so page is not anonymous).
2636 Frequency : Sometimes when using sys_sendfile
2638 Description: "page not covered by a lock" warnings from ll_readpage
2639 Details : sendfile called ll_readpage without right page locks present.
2640 Now we introduced ll_file_sendfile that does necessary locking
2641 around call to generic_file_sendfile() much like we do in
2645 Frequency : with certain MDS communication failures at client mount time
2647 Description: NULL pointer deref after failed client mount
2648 Details : a client connection request may delayed by the network layer
2649 and not be sent until after the PTLRPC layer has timed out the
2650 request. If the client fails the mount immediately it will try
2651 to clean up before the network times out the request. Add a
2652 reference from the request import to the obd device and delay
2653 the cleanup until the network drops the request.
2656 Frequency : occasionally during client (re)connect
2658 Description: assertion failure during client (re)connect
2659 Details : processing a client connection request may be delayed by the
2660 client or server longer than the client connect timeout. This
2661 causes the client to resend the connection request. If the
2662 original connection request is replied in this interval, the
2663 client may trip an assertion failure in ptlrpc_connect_interpret()
2664 which thought it would be the only running connect process.
2667 Frequency : only with obd_echo servers and clients that are rebooted
2669 Description: kernel BUG accessing uninitialized data structure
2670 Details : When running an obd_echo server it did not start the ping_evictor
2671 thread, and when a client was evicted an uninitialized data
2672 structure was accessed. Start the ping_evictor in the RPC
2673 service startup instead of the OBD startup.
2675 Severity : enhancement
2676 Bugzilla : 10193 (patchless)
2677 Description: Remove dependency on various unexported kernel interfaces.
2678 Details : No longer need reparent_to_init, exit_mm, exit_files,
2679 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2682 Frequency : rare (only users of deprecated and unsupported LDAP config)
2684 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2685 Details : LDAP apparently contains 'lustreName' attributes instead of
2686 'name'. A simple remapping of the name is sufficient.
2689 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2691 Description: waiting_locks_callback trips kernel BUG if client is evicted
2692 Details : Running with the dump_on_timeout debug flag turned on makes
2693 it possible that the waiting_locks_callback() can try to dump
2694 the Lustre kernel debug logs from an interrupt handler. Defer
2695 this log dumping to the expired_lock_main() thread.
2697 Severity : enhancement
2699 Description: Support NFS exporting on 2.6 kernels.
2700 Details : Implement non-rawops metadata methods for NFS server to use without
2701 changing NFS server code.
2704 Frequency : very rare (synthetic metadata workload only)
2706 Description: two racing renames might cause an MDS thread to deadlock
2707 Details : Running the "racer" program may cause one MDS thread to rename
2708 a file from being the source of a rename to being the target of
2709 a rename at exactly the same time that another thread is doing
2710 so, and the second thread has already enqueued these locks after
2711 doing a lookup of the target and is trying to relock them in
2712 order. Ensure that we don't try to re-lock the same resource.
2715 Frequency : only very large systems with liblustre clients
2717 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2718 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2719 all OSTs in parallel. This allows the eviction of stale liblustre
2720 clients to proceed much faster than if they were done in series,
2721 and also offers similar improvements for other set_info RPCs.
2726 Description: excessive CPU usage during initial read phase on client
2727 Details : During the initial read phase on a client, it would agressively
2728 retry readahead on the file, consuming too much CPU and impacting
2729 performance (since 1.4.5.8). Improve the readahead algorithm
2730 to avoid this, and also improve some other common cases (read
2731 of small files in particular, where "small" is files smaller than
2732 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2737 Description: MDS crash when receiving packet with unknown intent.
2738 Details : Do not LBUG in unknown intent case, just return -EFAULT
2740 Severity : enhancement
2741 Bugzilla : 9293, 9385
2742 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2743 Details : Do not serialize getattr (non-intent version) and statfs.
2746 Frequency : occasional, when OST network is overloaded/intermittent
2748 Description: client evicted by OST after bulk IO timeout
2749 Details : If a client sends a bulk IO request (read or write) the OST
2750 may evict the client if it is unresposive to its data GET/PUT
2751 request. This is incorrect if the network is overloaded (takes
2752 too long to transfer the RPC data) or dropped the OST GET/PUT
2753 request. There is no need to evict the client at all, since
2754 the pinger and/or lock callbacks will handle this, and the
2755 client can restart the bulk request.
2758 Frequency : Always when mmapping file with no objects
2760 Description: client crashes when mmapping file with no objects
2761 Details : Check that we actually have objects in a file before doing any
2762 operations on objects in ll_vm_open, ll_vm_close and
2768 Description: Request leak when working with deleted CWD
2769 Details : Introduce advanced request refcount tracking for requests
2770 referenced from lustre intent.
2772 Severity : Enhancement
2774 Description: Cache open file handles on client.
2775 Details : MDS now will return special lock along with openhandle, if
2776 requested and client is allowed to hold openhandle, even if unused,
2777 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2778 closing files for every read/write openration.
2780 Severity : Enhancement
2782 Description: Cache open negative dentries on client when possible.
2783 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2784 negative dentries on lock revocation.
2789 Description: Remounting a client read-only wasn't possible with a zconf mount
2790 Details : It wasn't possible to remount a client read-only with llmount.
2792 Severity : enhancement
2793 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2794 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2795 MPICH is included to add Lustre-specific ADIO interfaces.
2796 This is based closely on the UFS ADIO layer and only differs
2797 in file creation, in order to allow the OST striping to be set.
2798 This is user-contributed code and not supported by CFS.
2803 Description: extended inode attributes (immutable, append-only) work improperly
2804 when 2.4 and 2.6 kernels are used on client/server or vice versa
2805 Details : Introduce kernel-independent values for these flags.
2807 Severity : enhancement
2810 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2811 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2812 granularity. Now, Fractional MB granularity is be supported,
2813 this is very useful for low memory system.
2815 Severity : enhancement
2817 Description: Getattr by fid
2818 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2819 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2820 have locks and avoids getattr rpc requests for it on every stat.
2823 Frequency : Always, for filesystems larger than 2TB
2825 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2826 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2827 filesystems larger than 2TB to be mounted with mballoc enabled.
2830 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2832 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2833 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2834 filesystems larger than 2TB is mounted then the kernel will
2835 silently corrupt the start of the filesystem. CONFIG_LBD is
2836 enabled for all CFS-supported kernels, but the possibility of
2837 this happening with a modified kernel config exists.
2839 Severity : enhancement
2841 Description: add client O_DIRECT support for 2.6 kernels
2842 Details : It is now possible to do O_DIRECT reads and writes to files
2843 in the Lustre client mountpoint on 2.6 kernel clients.
2845 Severity : enhancement
2847 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2848 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2849 parallel, not waiting for response from every OST before sending
2850 a rpc to the next OST.
2855 Description: setattr vs write race when updating file timestamps
2856 Details : Client processes that update a file timestamp into the past
2857 right after writing to the file (e.g. tar) it is possible that
2858 the updated file modification time can be reset to the current
2859 time due to a race between processing the setattr and write RPC.
2861 Severity : enhancement
2863 Description: Bring 'lfs find' closer in line with regular Linux find.
2864 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2865 -print0 options and obtains all the needed info through the lustre
2868 Severity : enhancement
2870 Description: support up to 1024 configured devices on one node
2871 Details : change obd_dev array from statically allocated to dynamically
2872 allocated structs as they are first used to reduce memory usage
2877 Description: Flush dirty partially truncated pages during truncate
2878 Details : Immediatelly flush partially truncated pages in filter_setattr,
2879 this way we completely avoid having any pages in page cache on OST
2880 and can retire ugly workarounds during writes to flush such pages.
2885 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2886 Details : protect inode from truncation within vfs_unlink() context
2887 just take a reference before calling vfs_unlink() and release it
2888 when parent's i_sem is free.
2893 Description: last_id value checked outside lock on OST caused LASSERT failure
2894 Details : If there were multiple MDS->OST object precreate requests in
2895 flight, it was possible that the OST's last object id was checked
2896 outside a lock and incorrectly tripped an assertion. Move checks
2897 inside locks, and discard old precreate requests.
2900 Frequency : always, if extents are used on OSTs
2902 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2903 Details : OSTs using the extents format would not zero the high 16 bits of
2904 the index physical block number. This is not a problem for any
2905 OST filesystems smaller than 16TB, and no kernels support ext3
2906 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2907 new/modified files) and can be fixed for existing filesystems
2908 with e2fsprogs-1.39-cfs1.
2913 Description: import connection selection may be incorrect if timer wraps
2914 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2915 import connections to be ignored if the 32-bit jiffies counter
2916 wraps. Use a 64-bit jiffies counter.
2919 Frequency : very large clusters immediately after boot
2921 Description: LNET request buffers exhausted under heavy short-term load
2922 Details : If a large number of client requests are generated on a service
2923 that has previously never seen so many requests it is possible
2924 that the request buffer growth cannot keep up with the spike in
2925 demand. Instead of dropping incoming requests, they are held in
2926 the LND until the RPC service can accept more requests.
2929 Frequency : Sometimes during replay
2931 Description: Assertion failure in ll_local_open after replay.
2932 Details : If replay happened on an open request reply before we were able
2933 to set replay handler, reply will become not swabbed tripping the
2934 assertion in ll_local_open. Now we set the handler right after
2935 recognising of open request
2938 Frequency : very rare
2940 Description: kernel reports "badness in vsnprintf"
2941 Details : Reading from the "recovery_status" /proc file in small chunks
2942 may cause a negative length in lprocfs_obd_rd_recovery_status()
2943 call to vsnprintf() (which is otherwise harmless). Exit early
2944 if there is no more space in the output buffer.
2946 Severity : enhancement
2948 Description: clear OBD RPC statistics by writing to them
2949 Details : It is now possible to clear the OBD RPC statistics by writing
2950 to the "stats" file.
2955 Description: Client mtime is not the same on different clients after utimes
2956 Details : In some cases, the client was using the utimes() syscall on
2957 a file cached on another node. The clients now validate the
2958 ctime from the MDS + OSTs to determine which one is right.
2963 Description: Inability to activate failout mode
2964 Details : lconf script incorrectly assumed that in pythong string's numeric
2965 value is used in comparisons.
2968 Frequency : always with multiple stripes per file
2970 Description: Inefficient object allocation for mutli-stripe files
2971 Details : When selecting which OSTs to stripe files over, for files with
2972 a stripe count that divides evenly into the number of OSTs,
2973 the MDS is always picking the same starting OST for each file.
2974 Return the OST selection heuristic to the original design.
2979 Description: mount failures may take full timeout to return an error
2980 Details : Under some heavy load conditions it is possible that a
2981 failed mount can wait for the full obd_timeout interval,
2982 possibly several minutes, before reporting an error.
2983 Instead return an error as soon as the status is known.
2985 Frequency : quota enabled and large files being deleted
2987 Description: releasing more than 4GB of quota at once hangs OST
2988 Details : If a user deletes more than 4GB of files on a single OST it
2989 will cause the OST to spin in an infinite loop. Release
2990 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2995 Description: statfs data retrieved from /proc may be stale or zero
2996 Details : When reading per-device statfs data from /proc, in the
2997 {kbytes,files}_{total,free,avail} files, it may appear
2998 as zero or be out of date.
3001 Frequency : systems with MD RAID1 external journal devices
3003 Description: lconf's call to blkid is confused by RAID1 journal devices
3004 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3005 of returning all block devices that match the journal UUID.
3008 Frequency : always, for aggregate stripe size over 4GB
3010 Description: assertion fails when trying to use 4GB stripe size
3011 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
3012 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
3015 Frequency : always on ppc64
3017 Description: the first write on an ext3 filesystem with mballoc got stuck
3018 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3019 perform endianness conversion.
3021 ------------------------------------------------------------------------------
3023 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3025 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3026 INTEROPERATE with older versions automatically. Please read the
3027 user documentation before upgrading any part of a live system.
3028 * WARNING: Lustre networking configuration changes are required with
3029 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3032 * Support for kernels:
3033 2.6.9-22.0.2.EL (RHEL 4)
3034 2.6.5-7.244 (SLES 9)
3035 2.6.12.6 vanilla (kernel.org)
3038 Severity : enhancement
3039 Bugzilla : 7981/8208
3040 Description: Introduced Lustre Networking (LNET)
3041 Details : LNET is new networking infrastructure for Lustre, it includes
3042 a reorganized network configuration mode (see the user
3043 documentation for full details) as well as support for routing
3044 between different network fabrics. Lustre Networking Devices
3045 (LNDs) for the supported network fabrics have also been
3046 created for this new infrastructure.
3048 Severity : enhancement
3049 Description: Introduced Access control lists
3050 Details : clients can set ACLs on files and directories in order to have
3051 more fine-grained permissions than the standard Unix UGO+RWX.
3052 The MDS must be started with the "-o acl" mount option.
3054 Severity : enhancement
3055 Description: Introduced filesystem quotas
3056 Details : Administrators may now establish per-user quotas on the
3059 Severity : enhancement
3061 Description: Configuration change for the XT3
3062 The PTLLND is now used to run Lustre over Portals on the XT3
3063 The configure option(s) --with-cray-portals are no longer used.
3064 Rather --with-portals=<path-to-portals-includes> is used to
3065 enable building on the XT3. In addition to enable XT3 specific
3066 features the option --enable-cray-xt3 must be used.
3071 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3072 Details : A race between allocating a new llcd and re-getting the llcd_lock
3073 allowed another thread to grab newly-allocated llcd.
3075 Severity : enhancement
3077 Description: 2.6 OST async journal commit and locking fix to improve performance
3078 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3079 2.6 kernels are now async as they already were in 2.4 kernels so
3080 that they can commit concurrently with the network bulk transfer.
3081 For block-allocated files the filter allocation semaphore is held
3082 to avoid filesystem fragmentation during allocation. BKL lock
3083 removed for 2.6 xattr operations where it is no longer needed.
3088 Description: lconf incorrectly determined whether two IP networks could talk
3089 Details : In some more complicated routing and multiple-network
3090 configurations, lconf will avoid trying to make a network
3091 connection to a disjoint part of the IP space. It was doing the
3092 math incorrectly for one set of cases.
3097 Description: Fix for potential infinite loop processing records in an llog.
3098 Details : If an llog record is corrupted/zeroed, it is possible to loop
3099 forever in llog_process(). Validate the llog record length
3100 and skip the remainder of the block on error.
3103 Frequency : occasional (liblustre only)
3105 Description: liblustre could not open files whose last component is a symlink
3106 Details : sysio_path_walk() would incorrectly pass the open intent to
3107 intermediate path components.
3110 Frequency : rare (liblustre only with non-standard tuning)
3111 Bugzilla : 7201 (7350)
3112 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3113 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3114 after it has been released, only do this for referenced locks
3116 Severity : enhancement
3118 Description: specifying an (invalid) directory default stripe_size of -1
3119 would reset the directory default striping
3120 Details : stripe_size -1 was used internally to signal directory stripe
3121 removal, now use "all default" to signal dir stripe removal
3122 as a directory striping of "all default" is not useful
3125 Frequency : common for large clusters running liblustre clients
3127 Description: doing an ls when liblustre clients are running is slow
3128 Details : sending a glimpse AST to a liblustre client waits for every AST
3129 to time out, as liblustre clients will not respond. Since they
3130 cannot cache data we refresh the OST lock LVB from disk instead.
3132 Severity : enhancement
3134 Description: doing an ls at the same time as file IO can be slow
3135 Details : enqueue and other "small" requests can be blocked behind many
3136 large IO requests. Create a new OST IO portal for non-IO
3137 requests so they can be processed faster.
3140 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3142 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3143 created file with mode 000
3144 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3145 allow an ftruncate() as a non-root user to a file with mode 000.
3146 Lustre now respects this flag to disable mode checks when
3147 truncating a file owned by the user
3150 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3153 Description: Revoking locks from clients that went dead or catatonic might take
3155 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3156 cancellation of such locks instant on servers without waiting for
3157 any reply from clients. Clients drops these locks when cancel
3158 notification from server is received without replying.
3161 Frequency : liblustre-only, when liblustre client dies or becomes busy
3163 Description: Doing ls on Linux clients can take a long time with active
3165 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3166 granting such locks to it in the first place if possible. Locks
3167 are taken by proxy on the OST during the read or write and
3168 dropped immediately afterward. Add connect flags handling, do
3169 not grant locks to liblustre clients for glimpse ASTs.
3171 Severity : enhancement
3173 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3174 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3175 sequential reads and behaves badly with applications doing random
3176 reads. Improve it by reading ahead at least read region, and
3177 avoiding excessive large RPC for small reads.
3179 Severity : enhancement
3181 Description: Creating more than 1000 files for a single job may cause a load
3182 imbalance on the OSTs if there are also a large number of OSTs.
3183 Details : qos_prep_create() uses an OST index reseed value that is an
3184 even multiple of the number of available OSTs so that if the
3185 reseed happens in the middle of the object allocation it will
3186 still utilize the OSTs as uniformly as possible.
3191 Description: OST or MDS may oops in ping_evictor_main()
3192 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3193 but doesn't restart at beginning of obd_exports_timed list
3196 Severity : enhancement
3198 Description: improve by-nid export eviction on the MDS and OST
3199 Details : allow multiple exports with the same NID to be evicted at one
3200 time without re-searching the exports list.
3203 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3205 Description: MDS may oops in groups_free()
3206 Details : in rare race conditions a newly allocated group_info struct is
3207 freed again, and this can be NULL. The 2.4 compatibility code
3208 for groups_free() checked for a NULL pointer, but 2.6 did not.
3211 Frequency : common for liblustre clients doing little filesystem IO
3212 Bugzilla : 9352, 7313
3213 Description: server may evict liblustre clients accessing contended locks
3214 Details : if a client is granted a lock or receives a completion AST
3215 with a blocking AST already set it would not reply to the AST
3216 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3219 Frequency : lfs setstripe, only systems with more than 160 OSTs
3221 Description: unable to set striping with a starting offset beyond OST 160
3222 Details : llapi_create_file() incorrectly limited the starting stripe
3223 index to the maximum single-file stripe count.
3226 Frequency : LDAP users only
3228 Description: lconf did not handle in-kernel recovery with LDAP properly
3229 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3231 Severity : enhancement
3233 Description: bind OST threads to NUMA nodes to improve performance
3234 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3235 node and do their allocations there to localize memory access
3237 Severity : enhancement
3239 Description: llmount can determine client NID directly from Myrinet (GM)
3240 Details : the client NID code from gmnalnid was moved directly into
3241 llmount, removing the need to use this or specifying the
3242 client NID explicitly when mounting GM clients with zeroconf
3245 Frequency : if client is started with down MDS
3247 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3248 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3249 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3250 ptlrpc_import_delayed_req() because we didn't check if the
3251 request was interrupted, and we also didn't break out of the
3252 event loop if there was no timeout
3257 Description: data loss during non-page-aligned writes to a single file from
3258 both multiple nodes and multiple threads on one node at same time
3259 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3260 inconsistency led to false short-reads, that were cached and later
3261 used by ->prepare_write() to fill in partially written page,
3262 leading to data loss.
3265 Frequency : always, if lconf --abort_recovery used
3267 Description: lconf --abort_recovery fails with 'Operation not supported'
3268 Details : lconf was attempting to abort recovery on the MDT device and not
3271 Severity : enhancement
3273 Description: remove cleanup logs
3274 Details : replace lconf-generated cleanup logs with lustre internal
3275 cleanup routines. Eliminates the need for client-cleanup and
3278 Severity : enhancement
3280 Description: add support for EAs (user and system) on lustre filesystems
3281 Details : it is now possible to store extended attributes in the Lustre
3282 client filesystem, and with the user_xattr mount option it
3283 is possible to allow users to store EAs on their files also
3285 Severity : enhancement
3287 Description: Add possibility (config option) to show minimal available OST free
3289 Details : When compiled with --enable-mindf configure option, statfs(2)
3290 (and so, df) will return least minimal free space available from
3291 all OSTs as amount of free space on FS, instead of summary of
3292 free spaces of all OSTs.
3294 Severity : enhancement
3296 Description: do not expand extent locks acquired on OST-side
3297 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3298 by server: they are not cached anyway.
3301 Frequency : when mmap is used/binaries executed from Lustre
3303 Description: Unmmap pages before throwing them away from read cache.
3304 Details : llap_shrink cache now attempts to unmap pages before discarding
3305 them (if unmapping failed - do not discard). SLES9 kernel has
3306 extra checks that trigger if this unmapping is not done first.
3311 Description: lconf didn't resolve symlinks before checking to see whether a
3312 given mountpoint was already in use
3315 Frequency : when migrating failover services
3316 Bugzilla : 6395, 9514
3317 Description: When migrating a subset of services from a node (e.g. failback
3318 from a failover service node) the remaining services would
3319 time out and evict clients.
3320 Details : lconf --force (implied by --failover) sets the global obd_timeout
3321 to 5 seconds in order to quickly disconnect, but this caused
3322 other RPCs to time out too quickly. Do not change the global
3323 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3325 Severity : enhancement
3326 Frequency : if MDS is started with down OST
3327 Bugzilla : 9439,5706
3328 Description: Allow startup/shutdown of an MDS without depending on the
3329 availability of the OSTs.
3330 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3331 Add appropriate locking and lov-osc refcounts for safe
3332 cleaning. Add osc abort_inflight calls in case the
3333 synchronize never started.
3336 Frequency : occasional (Cray XT3 only)
3338 Description: root not authorized to access files in CRAY_PORTALS environment
3339 Details : The client process capabilities were not honoured on the MDS in
3340 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3341 been accessed by an authorized user then root was able to access
3342 the file on the local client also. The root user capabilities
3343 are now allowed on the MDS, as this environment has secure UID.
3346 Frequency : occasional
3348 Description: ldiskfs "too long searching" message happens too often
3349 Details : A debugging message (otherwise harmless) prints too often on
3350 the OST console. This has been reduced to only happen when
3351 there are fragmentation problems on the filesystem.
3356 Description: Division by zero in statfs when all OSCs are inactive
3357 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3358 lov_statfs divides by value returned by lov_get_stripecnt().
3362 Bugzilla : 9489, 3273
3363 Description: First write from each client to each OST was only 4kB in size,
3364 to initialize client writeback cache, which caused sub-optimal
3365 RPCs and poor layout on disk for the first writen file.
3366 Details : Clients now request an initial cache grant at (re)connect time
3367 and so that they can start streaming writes to the cache right
3368 away and always do full-sized RPCs if there is enough data.
3369 If the OST is rebooted the client also re-establishes its grant
3370 so that client cached writes will be honoured under the grant.
3375 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3376 Details : Now I/O RPCs go to different portal number and (presumably) fast
3377 lock requests (and glimses) and other RPCs get their own service
3378 threads pool that should be able to service those RPCs
3381 Severity : enhancement
3383 Description: Ability to exchange lustre version between client and servers and
3384 issue warnings at client side if client is too old. Also for
3385 liblustre clients there is ability to refuse connection of too old
3387 Details : New 'version' field is added to connect data structure that is
3388 filled with version info. That info is later checked by server and
3392 Frequency : rare, liblustre only.
3393 Bugzilla : 9296, 9581
3394 Description: Two simultaneous writes from liblustre at offset within same page
3395 might proceed at the same time overwriting eachother with stale
3397 Details : I/O lock withing llu_file_prwv was released too early, before data
3398 actually was hitting the wire. Extended lock-holding time until
3399 server acknowledges receiving data.
3402 Frequency : extremely rare. Never observed in practice.
3404 Description: avoid generating lustre_handle cookie of 0.
3405 Details : class_handle_hash() generates handle cookies by incrementing
3406 global counter, and can hit 0 occasionaly (this is unlikely, but
3407 not impossible, because initial value of cookie counter is
3408 selected randonly). Value of 0 is used as a sentinel meaning
3409 "unassigned handle" --- avoid it. Also coalesce two critical
3410 sections in this function into one.
3412 Severity : enhancement
3414 Description: allow liblustre clients to delegate truncate locking to OST
3415 Details : To avoid overhead of locking, liblustre client instructs OST to
3416 take extent lock in ost_punch() on client's behalf. New connection
3417 flag is added to handle backward compatibility.
3419 Severity : enhancement
3420 Bugzilla : 4928, 7341, 9758
3421 Description: allow number of OST service threads to be specified
3422 Details : a module parameter allows the number of OST service threads
3423 to be specified via "options ost ost_num_threads={N}" in the
3424 OSS's /etc/modules.conf or /etc/modprobe.conf.
3428 Bugzilla : 6146, 9635, 9895
3429 Description: servers crash with bad pointer in target_handle_connect()
3430 Details : In rare cases when a client is reconnecting it was possible that
3431 the connection request was the last reference for that export.
3432 We would temporarily drop the export reference and get a new
3433 one, but this may have been the last reference and the export
3434 was just destroyed. Get new reference before dropping old one.
3436 Severity : enhancement
3437 Frequency : if client is started with failover MDS
3439 Description: Allow multiple MDS hostnames in the mount command
3440 Details : Try to read the configuration from all specified MDS
3441 hostnames during a client mount in case the "primary"
3444 Severity : enhancement
3446 Description: Stop sending data to evicted clients as soon as possible.
3447 Details : Check if the client we are about to send or are sending data to
3448 was evicted already. (Check is done every second of waiting,
3449 for which l_wait_event interface was extended to allow checking
3450 of exit condition at specified intervals).
3453 Frequency : rare, normally only when NFS exporting is done from client
3455 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3457 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3460 Frequency : always, if srand() is called before liblustre initialization
3462 Description: Liblustre uses system PRNG disturbing its usage by user application
3463 Details : Introduce internal to lustre fast and high-quality PRNG for
3464 lustre usage and make liblustre and some other places in generic
3465 lustre code to use it.
3467 Severity : enhancement
3468 Bugzilla : 9477, 9557, 9870
3469 Description: Verify that the MDS configuration logs are updated when xml is
3470 Details : Check if the .xml configuration logs are newer than the config
3471 logs stored on the MDS and report an error if this is the case.
3472 Request --write-conf, or allow starting with --old_conf.
3474 Severity : enhancement
3476 Description: Handle symlinks in the path when checking if Lustre is mounted.
3477 Details : Resolve intermediate symlinks when checking if a client has
3478 mounted a filesystem to avoid duplicate client mounts.
3483 Description: lconf can hit an error exception but still return success.
3484 Details : The lconf command catches the Command error exception at the top
3485 level script context and will exit with the associated exit
3486 status, but doesn't ensure that this exit status is non-zero.
3491 Description: failure of ptlrpc thread startup can cause oops
3492 Details : Starting a ptlrpc service thread can fail if there are a large
3493 number of threads or the server memory is very fragmented.
3494 Handle this without oopsing.
3497 Frequency : always, only if liblustre and non-default acceptor port was used
3499 Description: liblustre cannot connect to servers with non-default acceptor port
3500 Details : tcpnal_set_default_params() was not called and was therefore
3501 ignoring the environment varaible TCPNAL_PORT, as well as other
3502 TCPNAL_ environment variables
3507 Description: two objects could be created on the same OST for a single file
3508 Details : If an OST is down, in some cases it was possible to create two
3509 objects on a single OST for a single file. No problems other
3510 than potential performance impact and spurious error messages.
3514 Bugzilla : 5681, 9562
3515 Description: Client may oops in ll_unhash_aliases
3516 Details : Client dcache may become inconsistent in race condition.
3517 In some cases "getcwd" can fail if the current directory is
3523 Description: Inode refcounting problems in NFS export code
3524 Details : link_raw functions used to call d_instantiate without obtaining
3525 extra inode reference first.
3529 Bugzilla : 9942, 9903
3530 Description: Referencing freed requests leading to crash, memleaks with NFS.
3531 Details : We used to require that call to ll_revalidate_it was always
3532 followed by ll_lookup_it. Also with revalidate_special() it is
3533 possible to call ll_revalidate_it() twice for the same dentry
3534 even if first occurence returned success. This fix changes semantic
3535 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3536 reference on a request referred from the intent.
3537 ll_intent_release() then releases such a request.
3540 Frequency : rare, normally benchmark loads only
3542 Description: unlinked inodes were kept in memory on the client
3543 Details : If a client is repeatedly creating and unlinking files it
3544 can accumulate a lot of stale inodes in the inode slab cache.
3545 If there is no other client load running this can cause the
3546 client node to run out of memory. Instead flush old inodes
3547 from client cache that have the same inode number as a new inode.
3550 Frequency : SLES9 2.6.5 kernel and long filenames only
3551 Bugzilla : 9969, 10379
3552 Description: utime reports stale NFS file handle
3553 Details : SLES9 uses out-of-dentry names in some cases, which confused
3554 the lustre dentry revalidation. Change it to always use the
3558 Frequency : rare, unless heavy write-truncate concurrency is continuous
3559 Bugzilla : 4180, 6984, 7171, 9963, 9331
3560 Description: OST becomes very slow and/or deadlocked during object unlink
3561 Details : filter_destroy() was holding onto the parent directory lock
3562 while truncating+unlinking objects. For very large objects this
3563 may block other threads for a long time and slow overall OST
3564 responsiveness. It may also be possible to get a lock ordering
3565 deadlock in this case, or run out of journal credits because of
3566 the combined truncate+unlink. Solution is to do object truncate
3567 first in one transaction without parent lock, and then do the
3568 final unlink in a new transaction with the parent lock. This
3569 reduces the lock hold time dramatically.
3572 Frequency : rare, 2.4 kernels only
3574 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3575 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3576 before unmounting their filesystem in order to avoid deadlock.
3577 The kernel_locked() function in 2.4 kernels only checks whether
3578 the kernel lock is held, not whether it is this process that is
3579 holding it as 2.6 kernels do.
3584 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3585 Details : The client ptlrpc code may be trying to reconnect to a down
3586 server before a previous connection attempt has timed out.
3587 Increase the reconnect interval to be longer than the connection
3588 timeout interval to avoid sending duplicate connections to
3592 Frequency : echo_client brw_test command
3594 Description: fix echo_client to work with OST preallocated code
3595 Details : OST preallocation code (5137) didn't take echo_client IO path
3596 into account: echo_client calls filter methods outside of any
3597 OST thread and, hence, there is no per-thread preallocated
3598 pages and buffers to use. Solution: hijack pga pages for IO. As
3599 a byproduct, this avoids unnecessary data copying.
3603 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3604 Description: Client can oops in mdc_commit_close() after open replay
3605 Details : It was possible for the MDS to return an open request with no
3606 transaction number in mds_finish_transno() if the client was
3607 evicted, but without actually returning an error. Clients
3608 would later try to replay that open and may trip an assertion
3609 Simplify the client close codepath, and always return an error
3610 from the MDS in case the open is not successful.
3613 Frequency : rare, 2.6 OSTs only
3615 Description: OST may deadlock under high load on fragmented files
3616 Details : If there was a heavy load and highly-fragmented OST filesystems
3617 it was possible to have all the OST threads deadlock waiting on
3618 allocation of biovecs, because the biovecs were not released
3619 until the entire RPC IO was completed. Instead, release biovecs
3620 as soon as they are complete to ensure forward IO progress.
3622 Severity : enhancement
3624 Description: Support for specifying external journal device at mount
3625 Details : If an OST or MDS device is formatted with an external journal
3626 device, this device major/minor is stored in the ext3 superblock
3627 and may not be valid for failover. Allow detecting and
3628 specifying the external journal at mount time.
3633 Description: Mounting an MDS with pending unlinked files may cause oops
3634 Details : target_finish_recovery() calls mds_postrecov() which returned
3635 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3636 considers this an error and immediately begins cleaning up the
3637 lov, just after starting the mds_lov process
3639 Severity : enhancement
3641 Description: Implement 'lfs df' to report actual free space on per-OST basis
3642 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3643 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3644 usage of objects; '-h' to report in human readable format.
3646 ------------------------------------------------------------------------------
3648 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3655 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3656 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3657 at startup during orphan unlinking. Instead, skip buddy update
3658 before setup, as it will be regenerated after recovery is complete.
3663 Description: If an OST is inactive, its locks might reference stale inodes.
3664 Details : lov_change_cbdata() must iterate over all namespaces, even if
3665 they are inactive to clear inode references from the lock.
3667 Severity : enhancement
3668 Frequency : occasional, if non-standard max_dirty_mb used
3670 Description: Client will block write RPCs if not enough grant
3671 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3672 then the client will block writes while waiting for another RPC
3673 to complete instead of consuming its dirty limit. With change
3674 we get improved performance when max_dirty_mb is small.
3676 Severity : enhancement
3677 Bugzilla : 3389, 6253
3678 Description: Add support for supplementary groups on the MDS.
3679 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3680 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3681 lookups for user supplementary groups into a cache.
3685 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3686 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3687 for liblustre O_EXCL is enforced by MDS.
3691 Description: readdir never returns NULL in liblustre.
3692 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3693 dentry in struct dirent.
3697 Frequency : liblustre only, depends on application IO pattern
3698 Description: liblustre clients evicted if not contacting servers
3699 Details : Don't put liblustre clients into the ping_evictor list, so
3700 they will not be evicted by the pinger ever.
3702 Severity : enhancement
3704 Description: Add ability to evict clients by NID from MDS.
3705 Details : By echoing "nid:$NID" string into
3706 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3707 $NID would be instantly evicted from this MDS and from all active
3708 OSTs connected to it.
3712 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3713 Details : lookup_it_finish() used to query file size from OSTs that was not
3718 Description: service threads change working directory to that of init
3719 Details : Starting lustre service threads may pin the working directory
3720 of the parent thread, making that filesystem busy. Threads
3721 now change to the working directory of init to avoid this.
3725 Frequency : during shutdown only
3726 Description: shutdown with a failed MDS or OST can cause unmount to hang
3727 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3732 Frequency : 2.6 only
3733 Description: chmod/chown may include an extra supplementary group
3734 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3735 supplementary group and if none is specified this is used.
3738 Bugzilla : 5479 (6816)
3740 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3741 Details : If lookup is in progress on a file that is unlinked we might try
3742 to revalidate the inode and fail in revalidate after lookup is
3743 complete and ll_file_open() enqueues the open again but
3744 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3747 Frequency : always, if lconf --abort_recovery used
3749 Description: lconf --abort_recovery fails with 'Operation not supported'
3750 Details : lconf was attempting to abort recovery on the MDT device and not
3753 ------------------------------------------------------------------------------
3755 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3760 Frequency : rare (only unsupported configurations with a node running as an
3762 Bugzilla : 6514, 5137
3763 Description: Mounting a Lustre file system on a node running as an OST could
3765 Details : OSTs now preallocates memory needed to write out data at
3766 startup, instead of when needed, to avoid having to
3767 allocate memory in possibly low memory situations.
3768 Specifically, if the file system is mounted on on OST,
3769 memory pressure could force it to try to write out data,
3770 which it needed to allocate memory to do. Due to the low
3771 memory, it would be unable to do so and the node would
3772 become unresponsive.
3774 Severity : enhancement
3776 Description: Addition of lconf --service command line option
3777 Details : lconf now accepts a '--service <arg>' option, which is
3778 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3780 Severity : enhancement
3782 Description: Failover mode is now the default for OSTs.
3783 Details : By default, OSTs will now run in failover mode. To return to
3784 the old behaviour, add '--failout' to the lmc line for OSTs.
3786 Severity : enhancement
3788 Description: Health checks are now provided for MDS and OSTs
3789 Details : Additional detailed health check information on MSD and OSTs
3790 is now provided through the procfs health_check value.
3793 Frequency : occasional, depends on IO load
3795 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3796 after numerous create/delete cycles
3797 Details : The ext3 inode allocation policy would not allocate new inodes
3798 very well on the OSTs because there are no new directories
3799 being created. Instead we look for groups with free space if
3800 the parent directories are nearly full.
3805 Description: Network or server problems during mount may cause partially
3806 mounted clients instead of returning an error.
3807 Details : The config llog parsing code may overwrite the error return
3808 code during mount error handling, returning success instead
3814 Description: MDS can fail to allocate large reply buffers
3815 Details : After long uptimes the MDS can fail to allocate large reply
3816 buffers (e.g. zconf client mount config records) due to memory
3817 fragmentation or consumption by the buffer cache. Preallocate
3818 some large reply buffers so that these replies can be sent even
3819 under memory pressure.
3823 Frequency : rare (liblustre)
3824 Description: fsx running with liblustre complained that using truncate() to
3825 extend the file doesn't work. This patch corrects that issue.
3826 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3827 ATTR_SIZE and lsm use in llu_setattr_raw.
3831 Frequency : rare, only 2.6 kernels
3832 Description: Unusual file access patterns on the MDS may result in inode
3833 data being lost in very rare circumstances.
3834 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3835 code in the 2.6 kernel caused the inode and/or EA data not to
3836 be read from disk, causing single-file corruption.
3840 Frequency : rare, only 2.6 filesystems using extents
3841 Description: Heavy concurrent write and delete load may cause data corruption.
3842 Details : It was possible under high-load situations to have an extent
3843 metadata block in the block device cache from a just-unlinked
3844 file overwrite a newly-allocated data block. We now unmap any
3845 metadata buffers that alias just-allocated data blocks.
3849 Frequency : filesystems with default stripe_count larger than 77
3850 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3851 Details : lconf specifies an inode size of 4096 bytes when the default
3852 stripe_count is larger than 77. This conflicts with the default
3853 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3854 this case to avoid pinning too much memory for large EAs.
3856 ------------------------------------------------------------------------------
3858 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3863 Frequency : rare (extremely heavy IO load with hundreds of clients)
3865 Description: Client is evicted, gets IO error writing to file
3866 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3867 caused clients to be evicted for AST timeouts. The fixes in
3868 bug 5192 mean we no longer need to have such short AST timeouts
3869 so ldlm_timeout has been increased.
3872 Frequency : occasional during --force or --failover shutdown under load
3873 Bugzilla : 5949, 4834
3874 Description: Server oops/LBUG if stopped with --force or --failover under load
3875 Details : a collection of import/export refcount and cleanup ordering
3876 issues fixed for safer force cleanup
3879 Frequency : only filesystems larger than 120 OSTs
3880 Bugzilla : 5990, 6223
3881 Description: lfs getstripe would oops on a very large filesystem
3882 Details : lov_getconfig used kfree on vmalloc'd memory
3885 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3886 Bugzilla : 6242, 6243
3887 Description: reading from files that had been truncated to a non-zero size
3888 but never opened returned no data
3889 Details : ll_file_read() reads zeros from no-object files to EOF
3894 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3895 Details : There's little harm in aborting MDS/OSS recovery and letting it
3896 try again, so I removed the LASSERT and return an error instead.
3898 Severity : enhancement
3900 Description: New debugging infrastructure for tracking down data corruption
3901 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3902 (b) cover more of the client-side code path, and (c) try to narrow
3903 down where problems occurred
3907 Bugzilla : 3819, 4364, 4397, 6313
3908 Description: Racing close and eviction MDS could cause assertion in mds_close
3909 Details : It was possible to get multiple mfd references during close and
3910 client eviction, leading to one thread referencing a freed mfd.
3912 Severity: : enhancement
3913 Bugzilla : 3262, 6359
3914 Description: Attempts to reconnect to servers are now more aggressive.
3915 Details : This builds on the enhanced upcall-less recovery that was added
3916 in 1.4.2. When trying to reconnect to servers, clients will
3917 now try each server in the failover group every 10 seconds. By
3918 default, clients would previously try one server every 25 seconds.
3923 Description: After recovery, certain operations trigger a failed
3924 assertion on a client.
3925 Details : Failing over an mds, using lconf -d --failover, while a
3926 client was doing a readdir() call would cause the client to
3927 LBUG after recovery completed and the readdir() was resent.
3929 Severity : enhancement
3931 Description: Default groups are now added by lconf
3932 Details : You can now run lconf --group <servicename> without having to
3933 manually add groups with lmc.
3936 Frequency : occasional
3938 Description: Nodes with an elan id of 0 trigger a failed assertion
3941 Frequency : always when accessing e.g. tty/console device nodes
3943 Description: tty and some other devices nodes cannot be used on lustre
3944 Details : file's private_data field is used by device data and lustre
3945 values in there got lost. New field was added to struct file to
3946 store fs-specific private data.
3949 Frequency : when exporting Lustre via NFS
3951 Description: NFSD failed occasionally when looking up a path component
3952 Details : NFSD is looking up ".." which was broken in ext3 directories
3953 that had grown large enough to become hashed.
3956 Frequency : Clusters with multiple interfaces not on the same subnet
3958 Description: Nodes will repeatedly try to reconnect to an interface which it
3959 cannot reach and report an error to the log.
3960 Details : Extra peer list entries will be created by lconf with some peers
3961 unreachable. lconf now validates the peer before adding it.
3964 Frequency : Only if a default stripe is set on the filesystem root.
3966 Description: Setting a default stripe on the filesystem root prevented the
3967 filesystem from being remounted.
3968 Details : The client was sending extra request flags in the root getattr
3969 request and did not allocate a reply buffer for the dir EA.
3972 Frequency : occasional, higher if lots of files are accessed by one client
3973 Bugzilla : 6159, 6097
3974 Description: Client trips assertion regarding lsm mismatch/magic
3975 Details : While revalidating inodes the VFS looks up inodes with ifind()
3976 and in rare cases can find an inode that is being freed.
3977 The ll_test_inode() code will free the lsm during ifind()
3978 when it finds an existing inode and then the VFS later attaches
3979 this free lsm to a new inode.
3983 Bugzilla : 6422, 7030
3984 Description: MDS deadlock between mkdir and client eviction
3985 Details : Creating a new file via mkdir or mknod (starting a transaction
3986 and getting the ns lock) can deadlock with client eviction
3987 (gets ns lock and trying to finish a synchronous transaction).
3990 Frequency : occasional
3991 Description: While starting a server, the fsfilt_ext3 module could not be
3993 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3994 kernels. Previously, lconf would still use the ext3 name
3995 when trying to load modules. Now, it will correctly use
3996 ext3 on 2.4 and ldiskfs on 2.6.
3998 Severity : enhancement
3999 Description: The default stripe count has been changed to 1
4000 Details : The interpretation of the default stripe count (0, to lfs
4001 or lmc) has been changed to mean striping across a single
4002 OST, rather than all available. For general usage we have
4003 found a stripe count of 1 or 2 works best.
4005 Severity : enhancement
4006 Description: Add support for compiling against Cray portals.
4007 Details : Conditional compiling for some areas that are different
4011 Frequency : occasional
4012 Bugzilla : 6409, 6834
4013 Description: Creating files with an explicit stripe count may lead to
4014 a failed assertion on the MDS
4015 Details : If some OSTs are full or unavailable, creating files may
4016 trigger a failed assertion on the MDS. Now, Lustre will
4017 try to use other servers or return an error to the
4021 Frequency : occasional
4023 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4024 caused lock timeout during readahead (since 1.4.2).
4025 Details : Processes doing readahead might match a lock that hasn't been
4026 granted yet if there are overlapping and conflicting lock
4027 requests. The readahead process waits on ungranted lock
4028 (original lock is CBPENDING), while OST waits for that process
4029 to cancel CBPENDING read lock and eventually evicts client.
4031 Severity : enhancement
4033 Description: Initial enabling of flock support for clients
4034 Details : Implements fcntl advisory locking and file status functions.
4035 This feature is provided as an optional mount flag (default
4036 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4037 locking are implemented yet, and those that are are not guaranteed
4038 to be completely correct in production environments.
4039 mount -t lustre -o [flock|noflock] ...
4042 Frequency : occasional
4044 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4045 assertion in the ext3 JBD (journaling) layer.
4046 Details : The b_committed_data struct is protected by the big kernel lock
4047 in 2.4 kernels, serializing journal_commit_transaction() and
4048 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4049 access to this struct. In 2.6 kernels there is finer grained
4050 locking to improve SMP performance of the JBD layer.
4054 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4059 Description: merge and cleanup kernel patches.
4062 -----------------------------------------------------------------------------
4064 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4066 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4067 versions of Lustre. Please update all servers and clients to
4068 version 1.4.2 or later at the same time. You must also run
4069 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4070 to update the configuration logs.
4072 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4073 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4074 - swab llog records of type '0' so we get proper header size/idx (5861)
4075 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4076 - fix rename of one directory over another leaking an inode (5953)
4077 - avoid SetPageDirty on 2.6 (5981)
4078 - don't re-add just-being-destroyed locks to the waiting list (5653)
4079 - when creating new directories, inherit the parent's custom
4080 striping settings if present parent (3048)
4081 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4082 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4083 - don't hold client locks on temporary worklist from l_lru (5666)
4084 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4085 - automatically evict dead clients (5921)
4086 - Update file size properly in create+truncate+fstat case (6196)
4087 - Do not unhash mountpoint dentries, do not allow removal of
4089 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4090 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4091 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4092 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4093 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4094 - Correctly handle reads of files with no objects (6243)
4095 - lctl recover will also mark a device active if deactivate used (5933)
4097 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
4098 - allow --write-conf on an MDS with different nettype than client (5619)
4099 - don't write config llogs to MDS for mounts not from that MDS (5617)
4100 - lconf should create multiple TCP connections from a client (5201)
4101 - init scripts are now turned off by default; run chkconfig --on
4102 lustre and chkconfig --on lustrefs to use them
4103 - upcalls are no longer needed for clients to recover to failover
4105 - add --abort-recovery option to lconf to abort recovery on device
4107 - add support for an arbitrary number of OSTs (3026)
4108 - Quota support protocol changes.
4109 - forward compatibility changes to wire structs (6007)
4110 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
4111 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
4112 - improved "lustre status" script
4113 - initialize blocksize for non-regular files (6062)
4114 - added --disable-server and --disable-client configure options (5782)
4115 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
4116 - Vanilla 2.4.29 support
4117 - increase maximum number of obd devices to 520 (6242)
4118 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4119 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4120 SLES 9 2.6.5 kernels
4121 - limit stripes per file to 160 (the maximum EA size) (6093)
4123 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4126 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4127 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4128 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4129 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4130 - small ext3 extents cleanups and fixes (5733)
4131 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4132 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4133 - avoid needless client->OST connect, fix handle mismatch (5317)
4134 - fix DLM error path that led to out-of-sync client, long delays (5779)
4135 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4136 - fix several locking issues related to i_size (5492,5624,5654,5672)
4137 - don't move pending lock onto export if it is already evicted (5683)
4138 - fix kernel oops when creating .foo in unlinked directory (5548)
4139 - fix deadlock in obdfilter statistics vs. object create (5811)
4140 - use time_{before,after} to avoid timer jiffies wrap (5882)
4141 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4142 - Do not leak request if server was not able to process it (5154)
4143 - If mds_open unable to find parent dir, make that negative lookup(5154)
4144 - don't create new directories with extent-mapping (5909, 5936)
4146 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
4147 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
4148 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
4149 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
4150 - support for RHEL4 kernel on i686 (5773)
4151 - provide error messages when incompatible logs are encountered (5898)
4153 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
4154 * version 1.4.0.10 (1.4.1 release candidate 1)
4156 - don't keep a lock reference when lock is not granted (4238)
4157 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4158 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
4159 - drop import inflight refcount on signal_completed_replay error (5255)
4160 - unlock page after async write error during send (3677)
4161 - handle missing objects in filter_preprw_read properly (5265)
4162 - no transno return for symlink open, don't save no-trasno open (3440)
4163 - don't try to complete elan receive that already failed (4012)
4164 - free RPC server reply state on error (5406)
4165 - clean up thread from ptlrpc_start_thread() on error (5160)
4166 - readahead could read extra page into cache that wasn't ejected (5388)
4167 - prevent races in class_attach/setup/cleanup/detach (5260)
4168 - don't dereference de->d_inode after l_dput of de (5458)
4169 - use "int" for stripe value returned from lock_to_stripe (5544)
4170 - mballoc allocation and error-checking fixes in 2.6 (5504)
4171 - block device patches to fix I/O request sizes in 2.6 (5482)
4172 - look up hostnames for IB nals (5602)
4173 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
4174 - don't start multiple acceptors for the same port (5277)
4175 - fix incorrect LASSERT in mds_getattr_name (5635)
4176 - export a proc file for general "ping" checking (5628)
4177 - fix "lfs check" to not block when the MDS is down (5628)
4179 - service request history (4965)
4180 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4181 - create an "evict_client" /proc entry on OSTs, like the MDS has
4182 - fix mount usage message, return errors per mount(8) (5168)
4183 - change grep [] to grep "[]" in tests so they work in more UMLs
4184 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4185 - remove ancient LOV_MAGIC_V0 EA support (5047)
4186 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4187 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4188 - allow readahead allocations to fail when low on memory (5383)
4189 - mmap locking landed again, after considerable improvement (2828)
4190 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4192 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4195 - send OST transaction number in read/write reply to free req (4966)
4196 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4197 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4198 - the watchdog thread now runs as interruptible (5246)
4199 - flock/lockf fixes (but it's still disabled, pending 5135)
4200 - don't use EXT3 constants in llite code (5094)
4201 - memory shortage at startup could cause assertion (5176)
4203 - reorganization of lov code
4204 - single portals codebase
4206 - add extents/mballoc support (5025)
4207 - direct I/O reads in the obdfilter (4048)
4208 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4210 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4213 - send OST transaction number in read/write reply to free req (4966)
4214 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4215 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4216 - the watchdog thread now runs as interruptible (5246)
4217 - handle missing objects in filter_preprw_read properly (5265)
4218 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4219 - drop import inflight refcount on signal_completed_replay error (5255)
4220 - unlock page after async write error during send (3677)
4221 - return original error code on reconstructed replies (3761)
4222 - no transno return for symlink open, don't save no-trasno open (3440)
4224 - add pid to ldlm debugging output (4922)
4225 - bump the watchdog timeouts -- we can't handle 30sec yet
4226 - extra debugging for orphan dentry/inode bug (5259)
4228 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
4231 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4232 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4233 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
4234 - fix NULL dereference in /proc/sys/portals/routes (4827)
4235 - allow failed mdc_close() operations to be interrupted (4561)
4236 - stop precreate on OST before MDS would time out on it (4778)
4237 - don't send partial-page writes before EOF from client (4410)
4238 - discard client grant for sub-page writes on large-page clients (4520)
4239 - don't free dentries not owned by NFS code, check generation (4806)
4240 - fix lsm leak if mds_create_objects() fails (4801)
4241 - limit debug_daemon file size, always print CERROR messages (4789)
4242 - use transno after validating reply (3892)
4243 - process timed out requests if import state changes (3754)
4244 - update mtime on OST during writes, return in glimpse (4829)
4245 - add mkfsoptions to LDAP (4679)
4246 - use ->max_readahead method instead of zapping global ra (5039)
4247 - don't interrupt __l_wait_event() during strace
4249 - add software watchdogs to catch hung threads quickly (4941)
4250 - make lustrefs init script start after nfs is mounted
4251 - fix CWARN/ERROR duplication (4930)
4252 - return async write errors to application if possible (2248)
4253 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4254 - print NAL number in %x format (4645)
4255 - update barely-supported suse-2.4.21-171 series (4842)
4256 - support for sles 9 %post scripts
4257 - support for building 2.6 kernel-source packages
4258 - support for sles km_* packages
4260 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4263 - ignore -ENOENT errors in osc_destroy (3639)
4264 - notify osc create thread that OSC is being cleaned up (4600)
4265 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4266 - reconstruct ost_handle() like mds_handle() (4657)
4267 - create a new thread to do import eviction to avoid deadlock (3969)
4268 - let lconf resolve symlinked-to devices (4629)
4269 - don't unlink "objects" from directory with default EA (4554)
4270 - hold socknal file ref over connect in case target is down (4394)
4271 - allow more than 32000 subdirectories in a single directory (3244)
4272 - fix blocks count for O_DIRECT writes (3751)
4273 - OST returns ENOSPC from object create when no space left (4539)
4274 - don't send truncate RPC if file size isn't changing (4410)
4275 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4276 - bind to privileged port in socknal and tcpnal (3689)
4278 - rate limit CERROR/CWARN console message to avoid overload (4519)
4279 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4280 - basic mmap support (3918)
4281 - kernel patch series update from b1_4 (4711)
4283 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4286 - avoid crash during MDS cleanup with OST shut down (2775)
4287 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4288 - don't use bad inodes on the MDS (3744)
4289 - dynamic object preallocation to improve recovery speed (4236)
4290 - don't hold spinlock over lock dumping or change debug flags (4401)
4291 - don't zero obd_dev when it is force cleaned (3651)
4292 - print grants to console if they go negative (4431)
4293 - "lctl deactivate" will stop automatic recovery attempts (3406)
4294 - look for existing locks in ldlm_handle_enqueue() (3764)
4295 - don't resolve lock handle twice in recovery avoiding race (4401)
4296 - revalidate should check working dir is a directory (4134)
4298 - don't always mark "slow" obdfilter messages as errors (4418)
4300 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4303 - don't close LustreDB during write_conf until it is done (3860)
4304 - fix typo in lconf for_each_profile (3821)
4305 - allow dumping logs from multiple threads at one time (3820)
4306 - don't allow multiple threads in OSC recovery (3812)
4307 - fix debug_size parameters (3864)
4308 - fix mds_postrecov to initialize import for llog ctxt (3121)
4309 - replace config semaphore with spinlock (3306)
4310 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4311 - don't allow enqueue to complete on a destroyed export (3822)
4312 - down write_lock before checking llog header bitmap (3825)
4313 - recover from lock replay timeout (3764)
4314 - up llog sem before sending rpc (3652)
4315 - reduce ns lock hold times when setting kms (3267)
4316 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4317 - fix NULL deref and obd_dev leak on setup error (3312)
4318 - replace some LBUG about llog ops with error handling (3841)
4319 - don't match INVALID dentries from d_lookup and spin (3784)
4320 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4321 - fix invalid assertion in ptlrpc_set_wait (3880)
4323 - add libwrap support for the TCP acceptor (3996)
4324 - add /proc/sys/portals/routes for non-root route listing (3994)
4325 - allow setting MDS UUID in .xml (2580)
4326 - print the stack of a process that LBUGs (4228)
4328 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4331 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4332 - make sure to unset replay flag from failed open requests (3440)
4333 - if default stripe count is 0, use OST count for inode size (3636)
4334 - update parent mtime/ctime on client for create/unlink (2611)
4335 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4336 - free recovery state on server during a forced cleanup (3571)
4337 - unregister_reply for resent reqs (3063)
4338 - loop back devices mounting and status check on 2.6 (3563)
4339 - fix resource-creation race that can provoke i_size == 0 (3513)
4340 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4341 - more debugging for page-accounting assertion (3746)
4342 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4343 - avoid deadlock after precreation failure (3758)
4344 - fix race and lock order deadlock in orphan handling (3450, 3750)
4345 - add validity checks when grabbing inodes from l_ast_data (3599)
4347 - add /proc/.../recovery_status to obdfilter (3428)
4348 - lightweight CDEBUG infrastructure, debug daemon (3668)
4349 - change default OSC RPC parameters to be better on small clusters
4350 - turn off OST read cache for files smaller than 32MB
4351 - install man pages and include them in rpms (3100)
4352 - add new init script for (un)mounting lustre filesystems (2593)
4353 - run chkconfig in %post for init scripts (3701)
4354 - drop scimac NAL (unmaintained)
4356 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4359 - clean kiobufs before and after use (3485)
4360 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4361 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4362 - change init script to fail more clearly if not run as root (1528)
4363 - allow clients to reconnect during replay (1742)
4364 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4365 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4366 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4367 - fix cond_resched (3554)
4368 - only evict unfinished clients after recovery (3515)
4369 - allow bulk resend, prevent data loss (3570)
4370 - dynamic ptlrpc request buffer allocation (2102)
4371 - don't allow unlinking open directory if it isn't empty (2904)
4372 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4373 - remove extraneous obd dereference causing LASSERT failure (3334)
4374 - don't use get_cycles() when creating temp. files on the mds (3156)
4375 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4376 - handle EEXIST for set-stripe, set proper directory name (3336)
4378 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4379 - fix ksocknal_fmb_callback() error messages (2918)
4381 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4384 - don't copy lvb into (possibly NULL) reply on error (2983)
4385 - don't deref dentry after dput, don't free lvb on error (2922)
4386 - use the kms to determine writeback rpc length (2947)
4387 - increment oti_logcookies when osc is inactive (2948)
4388 - update client's i_blocks count via lvb messages (2543)
4389 - handle intent open/close of special files properly (1557)
4390 - mount MDS with errors=remount-ro, like obdfilter (2009)
4391 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4392 - don't use cancelling-locks' kms values (2947)
4393 - use highest lock extent for kms, not last one (2925)
4394 - don't dereference ERR_PTR() dentry in error handling path (3107)
4395 - fix thread race in portals_debug_dumplog() (3122)
4396 - create lprocfs device entries at setup instead of at attach (1519)
4397 - common AST error handler, don't evict client on completion race (3145)
4398 - zero nameidata in detach_mnt in 2.6 (3118)
4399 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4400 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4401 - initialize RPC timeout timer earlier for 2.6 (3219)
4402 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4403 - print nal/nid for unknown nid (3258)
4404 - additional checks for oscc recovery before doing precreate (3284)
4405 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4406 - don't crash in mdc_close for bad permissions on open (3285)
4407 - zero i_rdev for non-device files (3147)
4408 - clear page->private before handing to FS, better assertion (3119)
4409 - tune the read pipeline (3236)
4410 - fix incorrect decref of invalidated dentry (2350)
4411 - provide read-ahead stats and refine rpc in flight stats (3328)
4412 - don't hold journal transaction open across create RPC (3313)
4413 - update atime on MDS at close time (3265)
4414 - close LDAP connection when recovering to avoid server load (3315)
4415 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4416 - don't leak open file on MDS after open resend (3325)
4417 - serialize filter_precreate and filter_destroy_precreated (3329)
4418 - loop device shouldn't call sync_dev() for nul device (3092)
4419 - clear page cache after eviction (2766)
4420 - resynchronize MDS->OST in background (2824)
4421 - refuse to mount the same filesystem twice on same mountpoint (3394)
4422 - allow llmount to create routes for mounting behind routers (3320)
4423 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4424 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4425 - fix rare problem with rename on htree directories (3417)
4427 - allow default OST striping configuration per directory (1414)
4428 - fix compilation for qswnal for 2.6 kernels (3125)
4429 - increase maximum number of MDS request buffers for large systems
4430 - change liblustreapi to be useful for external progs like lfsck (3098)
4431 - increase local configuration timeout for slow disks (3353)
4432 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4434 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4437 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4438 - don't overwrite extent policy data in reply if lock was blocked (2901)
4439 - drop filter export grants atomically with removal from device (2663)
4440 - del obd_self_export from work_list in class_disconnect_exports (2908)
4441 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4442 - swab reply message in mdc_close, other PPC fixes (2464)
4443 - fix destroying of named logs (2325)
4444 - overwrite old logs when running lconf --write_conf (2264)
4445 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4446 - fix race in target_handle_connect (2898)
4447 - mds_reint_create() should take same inode create lock (2926)
4448 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4449 - don't close files for self_export to avoid uninitialized obd (2936)
4450 - allow MDS with the same name as client node (2939)
4451 - hold dentry reference for closed log files for unlink (2325)
4452 - reserve space for all logs during transactions (2059)
4453 - don't evict page beyond end of stripe extent (2925)
4454 - don't oops on a deleted current working directory (2399)
4455 - handle hard links to targets without a parent properly (2517)
4456 - don't dereference NULL lock when racing during eviction (2867)
4457 - don't grow lock extents when lots of conflicting locks (2919)
4459 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4462 - account for cache space usage on clients to avoid data loss (974)
4463 - lfsck support in lustre kernel code (2349)
4464 - reduce journal credits needed for BRW writes (2370)
4465 - orphan handling to avoid losing space on client/server crashes
4466 - ptlrpcd can be blocked, stopping ALL progress (2477)
4467 - use lock value blocks to assist in proper KMS, faster stat (1021)
4468 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4469 - recovery for initial connections (2355)
4470 - fixes for mds_cleanup_orphans (1934)
4471 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4472 - block all file creations until orphan recovery completes (1901)
4473 - client remove rq_connection from request struct (2423)
4474 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4475 - recovery timer race (2670)
4476 - mdc_close recovey bug (2532)
4477 - ptlrpc cleanup bug (2710)
4478 - mds timeout on local locks (2588)
4479 - namespace lock held during RPCs (2431)
4480 - handle interrupted sync write properly (2503)
4481 - don't try to handle a message that hasn't been replied to (2699)
4482 - client assert failure during cleanup after abort recovery (2701)
4483 - leak mdc device after failed mount (2712)
4484 - ptlrpc_check_set allows timedout requests to complete (2714)
4485 - wait for inflight reqs when ptlrpcd finishes (2710)
4486 - make sure unregistered services are removed from the srv_list
4487 - reset bulk XID's when resending them (caught by 1138 test)
4488 - unregister_bulk after timeout
4489 - fix lconf error (2694)
4490 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4491 - readahead locks pages, leaves pending causing memory pressure (2673)
4492 - increase OST request buffers to 4096 on large machines (2729)
4493 - fix up permission of existing directories in simple_mkdir (2661)
4494 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4495 - don't assign transno to errored transactions (2742)
4496 - don't delete objects on OST if given a bogus objid from MDS (2751)
4497 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4498 - if rq_no_resend, then timeout request after recovery (2432)
4499 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4500 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4501 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4502 - MDS recovery shouldn't delete objects that it has given out (2730)
4503 - if enqueue arrives after completion, don't clobber LVB (2819)
4504 - don't unlock pages twice when trigger_group_io returns error (2814)
4505 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4506 - don't write pages to disk if there was an error (1450)
4507 - don't ping imports that have recovery disabled (2676)
4508 - take buffered bytes into account when balancing socknal conn (2817)
4509 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4510 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4511 - remove little-endian swabbing of llog records (1987)
4512 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4513 - retry reposting request buffers if they fail (1191)
4514 - grow extent at grant time to avoid granting a revoked lock (2809)
4515 - lock revoke doesn't evict page if covered by a second lock (2765)
4516 - disable VM readahead to avoid reading outside lock extents (2805)
4518 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4519 - updated kernel patches for hp-2.4.20 kernel (2681)
4521 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4524 - fix truncated write corruption (2366)
4525 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4527 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4528 - ensure len(uuid) < 37 in lmc (1171)
4529 - fix ia64 OOPS in llog_test (2255)
4530 - zero end of page at obdfilter for partial page writes (2648)
4531 - don't leave stale dentries around after renames (bug 2428)
4532 - fix timeouts when evicting a client with a single lock held (2642)
4533 - set deadline for the initial HELLO message to drain (2634)
4534 - print out dotted-quad IP addresses in the socknal (2302)
4536 - additional debugging for MDS client eviction problem (2443)
4537 - fix mkfsoptions support for osts (2603, 2604)
4539 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4542 - add series for the vanilla 2.6.0 kernel
4543 - add series for the vanilla 2.4.24 kernel
4544 - add series for a cray x86/64 UL kernel drop
4545 - fix xattr patches for the vanilla 2.4.19 series
4547 - generate true UUIDs in lmc (1171)
4548 - have portals stack dumping break in UML (2466)
4549 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4550 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4551 - wake callback waiting threads on client eviction (2460)
4552 - Add --ptldebug and --subsystem to lmc (1719)
4553 - update assertion to allow safe interrupt allocation
4554 - set rq_no_resend for cancel requests (2432)
4555 - recalculate ptlrpcd timeout after resend (2494)
4556 - call vfs_rmdir when removing pending directories (2368)
4557 - fix renaming a file to itself (2429)
4558 - lmc creates a default one-stripe lov (2454)
4559 - expand procfs space to handle large clusters (2326)
4560 - increase UML stack to avoid overflow
4561 - update lconf's list of debug and subsystem masks
4562 - fix lfs find --obd (2510)
4563 - /proc tunable for disabling filter read caching (2591)
4564 - stop rpm packages from altering slapd.conf (2301)
4565 - disable nagle in the socknal under 0conf (2578)
4566 - choose mds inode size based on stripe count (2572)
4567 - fix kernel-source rpm problems (2516)
4569 - add --disable-doc to avoid pdf generation (2421)
4570 - update documentation, tests, type-os, comments
4571 - avoid format warnings on ia64
4572 - remove the TOE NAL
4573 - tiny code cleanups by removing unused fields
4575 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4578 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4579 - remove the most blatant lies from BUILDING file (2371)
4580 - change default debug level to reasonable production setting
4581 - reduce client side cache size to reduce cache flush time
4582 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4583 - make TCP zerocopy and pinger support enabled by default (2476)
4584 - sync writes completed after process exits caused crashes (2319)
4585 - maintain correct mount count on the MDS (2356)
4586 - backout 1557, because 2316 wasn't really fixed
4587 - better file I/O statistics gathering in /proc
4588 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4589 - another kernel patch to fix zero-copy TCP function export
4590 - don't take duplicate lock when processing re-sent getattr (2420)
4591 - lctl uses obd_self_export instead of creating new conn (2353)
4592 - MDS/OST recovery case which requires object creation asserted (2425)
4593 - move lfs from /usr/sbin to /usr/bin in packages
4594 - fix race between mds_client_add and mds_client_free (2417)
4595 - use kmalloc instead of slabs in portals (2430)
4596 - don't create duplicate records when a failover MDS is present (2442)
4597 - remove unnecessary mount age check (2332)
4598 - don't remove directory inodes from locks prematurely (2451)
4599 - don't break if MDS service name is the same as hostname (2103)
4600 - fix races in client write RPC generation when cache full (2482)
4602 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4605 - remove now-unused request->rq_obd (278)
4606 - if an allocation fails, print out how much memory we've used (1933)
4607 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4608 - add the "configurable stack size" patch to most series files (1256)
4609 - ability to write large log records, for 100+ OST configs (2306)
4610 - fix NULL deref when filter_prep fails (2314)
4611 - fix operator precedence error in filter_sync
4612 - dynamic allocation of socknal TX descriptors (2315)
4613 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4614 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4615 - ensure that CWARN messages in Portals always get to the syslog
4616 - __init/__exit are not for prototype decls (ldlm_init/exit)
4617 - x86-64 compile warning fixes
4618 - fix gateway LMC keyword conflict (2318)
4619 - fix MDS lock inversions in getattr/reint paths (1844)
4620 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4621 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4622 - DLM race condition prevented some lock evictions (2328)
4623 - ENOMEM detection and retry on socknal sends (2230)
4624 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4625 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4626 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4627 - better error messages when a client is rejected during recovery (1505)
4628 - avoid cancelling locks which were never granted, after failure (2330)
4629 - fix i_sem/journal inversion in mds_client_add (2333)
4630 - fix truncate/getattr lock cycle deadlock (2334)
4631 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4632 - fix two rare exit paths which could leak an l_lock() ref (2321)
4633 - fencepost error in MDS/OST orphan recovery (2226)
4634 - make log record alignment 8 bytes (1988)
4635 - lstripe now fails when requested offset > ost_count (2237)
4636 - ensure that all kernel series have a complete list.h (1607)
4637 - fix crashes in special-file operations (2316)
4638 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4640 - allow configurable automake binary, for testing new versions
4641 - small update to the lfs documentation
4643 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4645 * fix negative export reference count in fsfilt_sync (2312)
4647 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4648 * release candidate 0.9.1
4650 - orphans are moved into the PENDING directory for possible recovery
4651 - replayed opens now open by fid for orphan/rename safety (1042)
4652 - last close of an orphan inode generates a transno (683)
4653 - chdir() and mount() now pin the directory entry (1020)
4654 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4655 - discard very old requests without processing them (1502)
4656 - remove some common, well-understood CERRORs (1505)
4657 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4658 - clear "grant" flags in OST replies until OST grant code lands (1644)
4659 - fix read performance by not clobbering i_blksize on client (1598)
4660 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4661 - make LRU size a /proc tunable, clears locks when reduced (707)
4662 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4663 - server locks take references on exports now (1558)
4664 - build fixes for 2.4.20-rh trees (1663)
4665 - return an error from lov_create if all OSCs are inactive (1751)
4666 - fix import levels when a reconnect happens without a timeout (1597)
4667 - exit early from mds_open if we get a lookup error (1749)
4668 - partial page read at EOF wouldn't wait for disk before sending (1642)
4669 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4670 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4671 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4672 - write/truncate lock inversion (1639)
4673 - Don't auto-load obdclass, portals modules during cleanup (1495)
4674 - fix timestamps from jumping to "now" (1763)
4675 - extra journal assertions (1648)
4676 - add an extra multiunlink test (1771)
4677 - fix read_record/write_record API (1776)
4678 - fix leak of offset_extent, possible incorrect i_size later (1772)
4679 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4680 - Debugging for the kqswnal_get_idle_tx problems (1820)
4681 - Allow recovery to be attempted multiple times (1536)
4682 - Write out MDS last_rcvd file after it is first created (1600)
4683 - Fix tx_descriptor leak in failed transmit situations (1827)
4684 - ext3 journaling fixes for assertion failure after IO error (1871)
4685 - class_export_put() on freed export after completion AST error (1896)
4686 - Fix revalidate looping in VFS (1322)
4687 - Don't access a freed export during MDS_REINT timeout (1521)
4688 - Add open-unlink recovery support on the MDS (1673,1764)
4689 - Return an error if no MDS data was read from last_rcvd (1946)
4690 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4691 - Don't setup a disk device that doesn't match exported UUID (317)
4692 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4693 - avoid committing NULL handle in force close
4694 - local.sh is now a one-stripe LOV configuration
4695 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4696 - fix ext3 htree duplicate directory entry corruption (1516)
4697 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4698 - update to new LOV EA format (2097)
4699 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4701 2003-06-15 Phil Schwan <phil@clusterfs.com>
4704 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4705 - per-import/export recovery handling (958, 931, 959)
4706 - multiple last-rcvd slots, for serving multiple FSes (949)
4707 - connections are again shared between multiple imp/exports (963, 964)
4708 - "umount -f" would hang if any requests needed to be sent (393, 978)
4709 - avoid pinning large req buffer by copying for queued messages (989)
4710 - add "uuid" to "lctl device" command to help upcalls (991)
4711 - "open" RPCs with transnos would confuse recovery counters (1037)
4712 - do proper endian conversion of all wire messages (288, 340, 891)
4713 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4714 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4715 - fix ptlrpc_connection leak in target_handle_connect (1174)
4716 - fix import refcounting bug in OST and MDS cleanup (1134)
4717 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4718 - fix very unlikely obd_types race condition (501)
4719 - remove osc_open hack for echo_client (1187)
4720 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4721 - a failure in read_inode2 leads to deadlock (1139)
4722 - cancel ack-locks as soon as transaction is committed (1072)
4723 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4724 - make sure to commitrw after any preprw to avoid deadlock (1162)
4725 - failing to execute a file in a lustre FS would lock inode (1203)
4726 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4727 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4728 - obd_uuid2tgt would walk off the end of the list (1255)
4729 - on IA64 the osc would give portals incorrect bulk size (1258)
4730 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4731 - fix lock inversion caused by new llite matching code (1282)
4732 - limit the number of dirty pages on a client to 10MB (1286)
4733 - timed out locks were not being corrected cancelled (1289)
4734 - fix O_DIRECT above 4GB on IA-32 (1292)
4735 * major user-visible changes
4736 - fail out/fail over policy now controlled by the upcall (993)
4738 - add OBD_PING to check server availability and failure (954)
4739 - lustre messages are now sent in sending host order (288, 340, 891)
4740 - add eadatalen to MDS getattr reply (340)
4741 - OST read replies may contain second buffer, with per-page status (593)
4743 2003-03-11 Phil Schwan <phil@clusterfs.com>
4746 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4747 - failed open()s could cause deadlock; fixed (867, 869)
4748 - stop cancelling OST locks when files are closed (481)
4749 - overlapping XID spaces caused network corruption (851, 853)
4750 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4751 - setattr_raw added, to do single-RPC, server-side setattrs
4752 - lmc/lconf syntax change for OST UUIDs
4753 - fix crashy race condition between ptlrpc_free_req and osc_close
4754 - don't use request in mdc_enqueue if we hit a timeout (889)
4755 - don't set the inode i_size for regular files from the MDS (896)
4756 - handle out of order completion AST (842)
4757 - don't LBUG if a lock request times out after receiving AST (913)
4758 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4759 - if a bad lock AST arrives, send an error instead of dropping entirely
4760 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4761 - fix leak in bulk IO when only partially completed (899, 900, 926)
4762 - fix O_DIRECT for ia64 (55)
4763 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4764 - C-z after timeout could hang a process forever; fixed (977)
4766 - client-side I/O cache (678, 924, 929, 941, 970)
4768 - READPAGE and SETATTRs which don't take server-side locks get
4771 2003-02-11 Phil Schwan <phil@clusterfs.com>
4774 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4775 - Fix fsfilt_extN_readpage() to read a full page of directory
4776 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4777 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4778 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4779 - Most intent operations were removed, in favour of a new RPC mode
4780 that does a single RPC to the server and bypasses most of the VFS
4781 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4782 - Aggressively cancel local locks on DLM servers
4783 - mds_reint_unlink sends EA to the client if it's the last nlink.
4784 client uses that EA to unlink OST objects.
4785 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4786 - recursive symlinks were fixed (439)
4787 - fixed NULL deref in DEBUG_REQ
4788 - filter_update_lastobjid no longer calls sync, which annoyed extN
4789 - fixed multi-client small-writes to a single file problem (445)
4790 - fixed mtime updates during file writes (607)
4791 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4792 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4793 - fixed Linux OST size reporting problem (444, 656)
4794 - OST now updates object mtime with writes or setattr (607, 619)
4795 - client verifies file size before zeroing page past EOF (445)
4796 - OST now writes last allocated objid to disk with allocation (108)
4797 - LOV on echo now works (409)
4799 - mds_reint_unlink sends a new buffer, with the EA included. this
4800 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4801 set if a regular file was being unlinked, and it was the last link
4802 - use PtlGet from the target for bulk writes (315)
4803 - OST now updates object mtime with writes or setattr (607, 619)
4804 - LDLM now has a grant-time callback to revalidate locked items, if
4806 - Many MDS operations were reorganized to combat race conditions
4808 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4809 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4811 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4814 - Fully reactivate OST imports after reconnection (512, others)
4815 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4816 - More graceful error handling for truncating on dead OST (515)
4817 - Don't error out unless we're actually accessing dead stripes (474)
4818 - Fix garbage sizes when stripes are missing (410)
4819 - LRU counters were broken, causing constant lock purge (433, 432)
4820 - garbage on read from stripes with failed OSTs (441)
4821 - mark OSCs as active before reconnecting during recovery (438)
4822 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4823 - lfind did not preserve OST order in output (443)
4824 - symlinks cause hung clients, incorrect data (439)
4825 - stop dereferencing request after dropping refcount (457)
4826 - don't LASSERT(spin_is_locked) on non-SMP (455)
4827 - fixes for many rename() bugs
4828 - fstat didn't correctly synchronize attributes (399)
4829 - server must handle lock cancellation during blocking AST prep (487)
4830 - bulk descriptors were free()d too soon (511)
4831 - fix paths in lconf, which would load incorrect modules (451, 507)
4832 - fix confusing lconf 'host not found' error message (386)
4833 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4834 - fix race condition in mdc_blocking_ast() for inode access (526)
4835 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4836 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4837 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4838 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4839 - fix lstripe in several ways (kernel side) (527)
4840 - fix request leak in ldlm_cli_enqueue (262)
4841 - incorrect OSC was marked inactive after OST failure
4842 - call mds_fs_cleanup before unmounting filesystem (524)
4843 - fix races between taking ns_lock and ldlm_lock_change_resource
4844 - fix races updating LOV export open file list
4845 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4846 - fix recovery NULL deref in ldlm_cli_cancel_unused
4847 - fix some DLM races by using new hash table for lock handles (419)
4848 - permit the client to specify desired inodes, at replay
4849 - duplicate requests when we queue them for replay reintegration
4850 - fix last_rcvd offset calculation
4851 - sync after each recovered transaction, so we always make progress
4852 - never, not always, ERESTART requests without transnos
4853 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4855 - skip replay if the MDS says that the client is already connected
4856 - don't check for a recovery-enabled export to match lctl's UUID
4857 - don't INC_USE_COUNT for phantom exports
4858 - don't crash when cleaning up phantom exports (567)
4859 - don't double-finish or set replay data for errored mdc_open requests
4860 - abort requests when they time out, so we don't get old replies
4861 - send/receive replies for AST messages again
4862 - if the client says that it doesn't have the lock, cancel it on the
4864 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4865 mark it as destroyed, it will all work out when decref is called
4866 - fix module use counts (22, 581)
4868 - ASTs now expect a reply (server cancels lock on error reply)
4870 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4873 - fix many simultaneous client startup (392)
4874 - fix dentry->d_it clobbering
4875 - credentials weren't being shipped for readdir/getattr operations
4876 - remove invalid assertions triggered during some concurrent MD
4878 - proper Lustre versions added (336, 389)
4879 - fix memory leak for create error case (398)
4880 - fix LOV locking bug that would get cli/srv out of sync
4881 - fix echo client over LOV (409)
4882 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4883 - fix double-O_EXCL intent crash (424)
4884 - avoid sending multiple lock CANCELs (352)
4886 - MDS can do multi-client recovery (modulo bugs in new code)
4888 - many updates, edits, cleanups
4890 2002-11-18 Phil Schwan <phil@clusterfs.com>
4893 - fix null d_it dereference (346)
4894 - fix full OST/dbench hang (333)
4895 - fix permission problem with file removal (286)
4896 - fix removal of OSCs from LOV when they fail
4897 - fix NULL deref during bulk timeout (214)
4898 - fix problems related to multiple filesystems on one MDS (241)
4899 - fixed serious subtle metadata locking bugs
4900 - free locks on clients when inodes are removed due to memory
4902 - fix inode pointer in lock data (285)
4903 - partial support for multiple MDS on a single host (241)
4904 - data locks weren't cancelled at clear_inode time (290, 311)
4905 - intent locks could lead to unbounded lock growth (205)
4906 - added a maximum lock count, an LRU list, and a flusher
4907 - fix multiple rename (365)
4908 - properly abstracted the echo client
4909 - OSC locked 1 byte too many; fixed
4910 - rewrote brw callback code:
4911 - fixed recovery bugs related to LOVs (306)
4912 - fixed too-many-pages-in-one-write crash (191)
4913 - fixed (again) crash in sync_io_timeout (214)
4914 - probably fixed callback-related race (385)
4916 - Add capability to MDS protocol
4917 - LDLM cancellations and callbacks on different portals
4919 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4922 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4923 - defer open object destruction to close time (601981, 138)
4924 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4925 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4926 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4927 - don't call obd_setattr in ll_file_release for destroyed objects
4928 * protocol change to lustre_msg: move |version| and add |flags|
4929 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4930 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4931 * added replay of create, unlink, link and rename operations during
4932 MDS failover; recovery should be much more robust now
4933 * remove failed OSCs from LOVs (only lov_create uses this so far)
4934 * the lustre-HOWTO was brought (more) up to date (582544)
4936 2002-10-23 Phil Schwan <phil@clusterfs.com>
4939 - in-use dentries weren't being reused properly (617851)
4940 - prevent multiple LDLM setup (599178)
4941 - fix LOV size calculations for truncate (617853)
4942 - fix client handling of MDS intent errors (POSIX)
4943 - fix permission bug in lovstripe.c test (624321)
4944 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4945 - truncate past end of file could corrupt data
4946 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4947 - a race in recovery could return ETIMEDOUT to apps (623947)
4948 - building outside the source directory was fixed
4949 * the lustre-HOWTO was brought (more) up to date (582544)
4950 * major progress was made on recovery functionality
4952 2002-10-10 Phil Schwan <phil@clusterfs.com>
4955 - recovery deadlock fix
4956 - rm -rf causes LBUG fix (617817)
4957 - file open by multiple tasks fix (618962)
4958 - directory permissions bugs (602707 and 620007)
4959 - journal_stop fixed with locking (611313)
4960 - O_APPEND failures resolved (618273, perhaps 614459)
4961 - lconf PATH fix (619770)
4962 - IA64 build fix (621450)
4963 - RPC buffer sizes scale with amount of memory
4965 2002-10-01 Phil Schwan <phil@clusterfs.com>
4968 - locks would be cancelled without throwing away data pages,
4969 resulting in inconsistent data (605627)
4970 - inode attributes were not always being refreshed (605627, 612449)
4971 - lconf now continues to cleanup after lctl reports an error
4972 - MDS now enforces user permissions (602707)
4973 - lprocfs cleanup fixed, but not yet enabled (614157)
4974 - fixed infinite server hang, should a client not respond to an AST
4975 - avoid going into recovery if user calls readlink() with a buffer
4976 that's too small (613941)
4977 - AST RPCs no longer require replies (614867) -- this may be changed
4978 - don't crash server if client sends an IOV that's too big (611336)
4979 - fixed lock conversion deadlock (611892)
4980 - fixed the following of symlinks (614622)
4981 * recovery: the server can remove locks from a client that dies, other
4982 clients can make progress
4983 * more extN patch fixes
4984 * compile-time configurable ptlrpc buffer allocations
4986 - collaborative read cache document
4987 - Lustre Lite Performance CDR document-in-progress
4989 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4992 - fix typo in patch-2.4.18
4994 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4997 - clear ptlrpc request each time in handle_incoming_request()
4998 - unlink of files now destroys the object on the OST
5000 2002-09-19 Peter Braam <braam@clusterfs.com>
5002 * add hard link support
5003 * change obdfile creation method
5004 * kernel patch changed
5006 2002-09-19 Peter Braam <braam@clusterfs.com>
5009 - stack overflow bug in extN fixed
5011 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
5013 * documentation updates
5014 - add man pages for config tools
5015 - update tests/README to describe testing with new config tools
5016 - finish metadata API descriptions
5017 * bug fixes and cleanups
5018 - statfs workaround for 16TB limit
5019 - LOV stripe allocation improved, can stripe on subset of OSTs
5020 - LOV file size/IO offset was wrong for files > 4GB in size
5021 - object EA data was being dropped, caused files to be unreadable
5022 - memory overflow with non-LOV OST caused memory corruption
5023 - fixed regression tests to work with new config tools, obdfilter
5024 - fixed bug when directory size became larger than 1 block
5025 - fixed bug (for single client case) when PWD was deleted
5026 - invalidate local directory pages when doing intent-based ops
5027 - avoid LDLM oops when lock callback contained bad data
5029 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5031 * documentation updates
5032 * bug fixes and cleanups
5033 - configuration tools
5036 - 64-bit compile warnings
5037 - 64-bit internal statfs data
5039 * test_brw on persistent OST devices
5041 * lprocfs (disabled)
5043 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5045 * documentation updates
5046 * bug fixes and cleanups
5047 * configuration tools
5049 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5052 * many small fixes to 0_5_4
5053 * io/network handling
5054 * thinkos in MDS operations
5056 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5059 * crucial basic fixes to 0.5.3
5060 * IOR, Iozone work over Elan
5063 2002-08-07 Phil Schwan <phil@clusterfs.com>
5064 * version 0_5_3, our first alpha
5065 * we use the new Portals iovs
5066 * documentation updates
5067 * bug fixes and cleanups
5068 * small changes in the DLM wire protocol
5070 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5071 * version 0_5_1 with some initial stability,
5072 * locking on MD and file I/O.
5073 * documentation updates
5074 * several bug fixes since 0.5.0
5075 * small changes in wire protocol
5077 2002-07-18 Phil Schwan <phil@clusterfs.com>
5079 * delivered as Lustre Light Alpha
5080 * fixed a crash after handling invalid MDS requests
5081 * fixed directory pages for architectures with non-4k pages sizes
5083 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5084 * release version v0_4_4
5085 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5087 * Fixes a number of interruption problems with OST operations.
5088 * Update documentation for portals header changes
5089 * Move all wire protocol structs/defines to lustre_idl.h
5090 * Fixes symlink length bug.
5091 * Add tcpdump to repository.
5093 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5094 * release version v0_4_3
5095 * Fixes statfs for inodes on extN.
5096 * Fixes bug in runtests which would delete /etc/hosts.
5097 * Use 64-bit object IDs wherever possible (not into VFS though)
5098 Remove ost_get_info, which is unused by lustre, and out of date.
5100 2002-07-03 Peter Braam <braam@clusterfs.com>
5101 * release version v0_4_2 Fixes a lookup error (type not passed)
5102 * move forward to head of Portals
5103 * move forward to latest Lustre kernel
5105 2002-06-25 Peter Braam <braam@clusterfs.com>
5106 * release version v0_4_1. Hopefully stable on single node use.