1 tbd Sun Microsystems, Inc.
4 2.6.16.60-0.27 (SLES 10),
5 2.6.18-92.1.10.el5 (RHEL 5),
6 2.6.22.14 vanilla (kernel.org).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.40.11-sun1
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
17 Description: service mount cannot take device name with ":"
18 Details : Only when device name contains ":/" will mount treat it as
23 Description: Hitting mdc_commit_close() ASSERTION
24 Details : Properly handle request reference release in
25 ll_release_openhandle().
30 Description: replace ptlrpcd with the statahead thread to interpret the async
31 statahead RPC callback
35 Frequency : on recovery
36 Description: I/O failures after umount during fail back
37 Details : if client reconnected to restarted server we need join to recovery
38 instead of find server handler is changed and process self eviction
39 with cancel all locks.
41 Severity : enhancement
43 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
45 Severity : enhancement
47 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
49 Severity : enhancement
51 Description: Upcall on Lustre log has been dumped
52 Details : Allow for a user mode script to be called once a Lustre log has
53 been dumped. It passes the filename of the dumped log to the
54 script, the location of the script can be specified via
55 /proc/sys/lnet/debug_log_upcall.
60 Description: avoid messages about idr_remove called for id which is not allocated.
61 Details : Move assigment s_dev for clustered nfs to end of initialization, for
62 avoid problem with error handling.
67 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
68 Details : When connection is reused this not moved from CONN_UNUSED_HASH into
69 CONN_USED_HASH and this prodice warning when put connection again
72 Severity : enhancement
74 Description: Export bytes_read/bytes_write count on OSC/OST.
78 Description: Early reply size mismatch, MGC loses connection
79 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
80 the connect flags are properly negotiated.
84 Description: Properly propagate oinfo flags from lov to osc for statfs
85 Details : restore missing copy oi_flags to lov requests.
87 Severity : enhancement
89 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
90 : l_getgroups(8), lst(8), routerstat(8)
91 Details : included man pages for llobdstat(8), llstat(8),
92 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
94 Severity : enhancement
96 Description: configure's --enable-quota should check the
97 : kernel .config for CONFIG_QUOTA
98 Details : configure is terminated if --enable-quota is passed but
99 : no quota support is in kernel
103 Description: Remove portals compatibility
104 Details : Remove portals compatibility, not interoperable with releases
109 Description: Resolve device initialization race
110 Details : Prevent proc handler from accessing devices added to the
111 obd_devs array but yet be intialized.
113 Severity : enhancement
115 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
117 Severity : enhancement
119 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
124 Description: Using wrong pointer in osc_brw_prep_request
125 Details : Access to array[-1] can produce panic if kernel compiled with
126 CONFIG_PAGE_ALLOC enabled
130 Description: Client runs out of low memory
131 Details : Consider only lowmem when counting initial number of llap pages
135 Description: *optional* service tags registration
136 Details : if the "service tags" package is installed on a Lustre node
137 When the filesystem is mounted, a local-node service tag will
138 be created. See http://inventory.sun.com/ for more information
139 about the Service Tags asset management system.
143 Description: Kernel BUG tries to release flock
144 Details : Lustre does not destroy flock lock before last reference goes
145 away. So always drop flock locks when client is evicted and
146 perform unlock regardless of successfulness of speaking to MDS.
150 Description: add recount protection for osc callbacks, so avoid panic on shutdown
154 Description: sanity test 65a fails if stripecount of -1 is set
155 Details : handle -1 striping on filesystem in ll_dirstripe_verify
160 Description: ASSERTION(CheckWriteback(page,cmd)) failed
161 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
164 Severity : enhancement
166 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
170 Description: do not process already freed flock
171 Details : flock can possibly be freed by another thread before it reaches
172 to ldlm_flock_completion_ast.
176 Description: LBUG during stress test
177 Details : Need properly lock accesses the flock deadlock detection list.
181 Description: oops in page fault handler
182 Details : kernel page fault handler can return two special 'pages' in error case, don't
183 try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
187 Description: timeout with invalidate import.
188 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which should be
189 handled by ptlrpcd. This produce long age waiting and -ETIMEOUT
190 ptlrpc_invalidate_import and as result LASSERT.
192 Severity : enhancement
194 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
198 Description: enable MGS and MDT services start separately
199 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
200 a co-located MGS without starting the MGS, which is a complement
201 to 'nosvc' mount option.
205 Frequency : after recovery
206 Description: precreate to many object's after del orphan.
207 Details : del orphan st in oscc last_id == next_id and this triger growing
208 count of precreated objects. Set flag LOW to skip increase count
209 of precreated objects.
213 Frequency : rare, on clear nid stats
214 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
215 Details : when clean nid stats sometimes try destroy live entry,
216 and this produce panic in free.
220 Description: Stack overflow during MDS log replay
221 ease stack pressure by using a thread dealing llog_process.
225 Description: wait until IO finished before start new when do lock cancel.
226 Details : VM protocol want old IO finished before start new, in this case
227 need wait until PG_writeback is cleared until check dirty flag and
228 call writepages in lock cancel callback.
230 Severity : enhancement
232 Description: using special macro for print time and cleanup in includes.
236 Description: mds_mfd_close() ASSERTION(rc == 0)
237 Details : In mds_mfd_close(), we need protect inode's writecount change
238 within its orphan write semaphore to prevent possible races.
242 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
246 Frequency : rare, on shutdown ost
247 Description: don't hit live lock with umount ost.
248 Details : shrink_dcache_parent can be in long loop with destroy dentries,
249 use shrink_dcache_sb instead.
253 Description: don't panic with use echo client
254 Details : echo client pass NULL as client nid pointer and this produce null
259 Description: fix build on ppc32
260 Details : compile code with -m64 flag produce wrong object file for ppc32.
264 Description: add message levels for liblustreapi
268 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
269 Details : Move the 'good_osts' check before the 'total_bavail' check. This
270 will result in an -EAGAIN and in the exit call path we call
271 alloc_rr() which will with increasing aggressiveness attempt to
272 aquire precreated objects on the minimum number of required OSCs.
276 Description: Use old size assignment to avoid deadlock
277 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
278 the scheduling while holding a spinlock. We do not need locking
279 for size in ll_update_inode() because size is only updated from
280 the MDS for directories or files without objects, so there is no
281 other place to do the update, and concurrent access to such inodes
282 are protected by the inode lock.
286 Description: resolve "_IOWR redefined" build error on SLES10
290 Description: dump the memory debugging after all modules are unloaded to
291 suppress false negative in conf_sanity test 39
293 Severity : enhancement
295 Description: build kernel-ib packages for OFED 1.3 in our release cycle
300 Description: fix SLES kernel versioning
301 Details : the kernel version for our SLES 10 kernel did not include a "-"
302 before the "smp" at the end. while this was not a problem in
303 general, it did mean that software trying to use the kernel
304 version to try to detect a vendor specific kernel would fail.
305 this was most evident by the OFED build scripts.
309 Description: Don't update lov_desc members until making sure they are valid
310 Details : When updating lov_desc members via proc fs, need fix their
311 validities before doing the real update.
315 Description: don't put request into delay list while invalidate in flight.
316 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
317 import in flight. this produce timeout for invalidate and sometimes
318 can cause stale data.
322 Frequency : on ppc only
323 Description: not convert ost objects for directory because it's not exist.
324 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
328 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
330 Severity : enhancement
332 Description: Update to support 2.6.22.14 vanilla kernel.
336 Frequency : rare, on recovery
337 Description: read procfs can produce deadlock in some situation
338 Details : Holding lprocfs lock which send rpc can produce block for destroy
339 obd objects and this also block reconnect with -EALREADY. This isn't
340 fix all lprocfs bugs - but make it rare.
342 Severity : enhancement
344 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
347 Frequency : frequent on X2 node
349 Description: mdc_set_open_replay_data LBUG
350 Details : Set replay data for requests that are eligible for replay.
354 Description: lustre_mgs: operation 101 on unconnected MGS
355 Details : When MGC is disconnected from MGS long enough, MGS will evict the
356 MGC, and late on MGC cannot successfully connect to MGS and a lot
357 of the error messages complaining that MGS is not connected.
361 Frequency : on network error
362 Description: panic with double free request if network error
363 Details : mdc_finish_enqueue is finish request if any network error ocuring,
364 but it's true only for synchronus enqueue, for async enqueue
365 (via ptlrpcd) this incorrect and ptlrpcd want finish request
368 Severity : enhancement
370 Description: client-side metadata stat-ahead during readdir(directory readahead)
371 Details : perform client-side metadata stat-ahead when the client detects
372 readdir and sequential stat of dir entries therein
375 Frequency : on start mds
377 Description: Implement get_info(last_id) in obdfilter.
380 Frequency : occasional
382 Description: Correctly check stale fid, not start epoch if ost not support SOM
383 Details : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
384 overwrited with new generated fid, but mds can anwer with one of these
385 two fids and both is not stale. setattr incorectly start epoch and
386 assume will be called done_writeting, but without SOM done_writing
390 Frequency : rare, depends on device drivers and load
392 Description: MDS or OSS nodes crash due to stack overflow
393 Details : Code changes in 1.8.0 increased the stack usage of some functions.
394 In some cases, in conjunction with device drivers that use a lot
395 of stack the MDS (or possibly OSS) service threads could overflow
396 the stack. One change which was identified to consume additional
397 stack has been reworked to avoid the extra stack usage.
400 Frequency : occasional
402 Description: Do not fail import if osc_interpret_create gets -EAGAIN
403 Details : If osc_interpret_create got -EAGAIN it immediately exits and
404 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
405 oscc_has_objects and see OSC has no objests and call
406 oscc_internal_create to resend create request.
408 Severity : enhancement
410 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
412 Severity : enhancement
414 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
417 Frequency : very rare
419 Description: Fix replay if there is an un-replied request and open
420 Details : In some cases, older replay request will revert the
421 mcd->mcd_last_xid on MDS which is used to record the client's
424 Severity : enhancement
426 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
428 Severity : enhancement
430 Description: Add rhel5 support to HEAD.
432 Serverity : enhancement
434 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
439 Description: Don't allow skipping OSTs if index has been specified.
440 Details : Don't allow skipping OSTs if index has been specified, make locking
441 in internal create lots better.
445 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
446 Details : access to bitfield in structure is always rounded to long
447 and this produce problem with not atomic change any bit.
451 Description: Lustre make rpms failed.
452 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
453 builds Lustre rpms from tarball.
456 Frequency : rare on shutdown ost
458 Description: If llog cancel was not send before clean_exports phase, this can
459 produce deadlock in llog code.
460 Details : If llog thread has last reference to obd and call class_import_put
461 this produce deadlock because llog_cleanup_commit_master wait when
462 last llog_commit_thread exited, but this never success because was
463 called from llog_commit_thread.
467 Description: allow userland application know is lost one of stripes.
468 Details : fill lvb_blocks with error code on ost and return it to
469 application if error flag found.
473 Description: NULL lov_tgts causing MDS oops
474 Details : more safe checks for NULL lov_tgts for avoid oops.
476 Severity : enhancement
478 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
482 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
483 Details : If an OST is down the MDS will hang indefinitely in
484 obd_statfs_rqset() waiting for the statfs data. While for
485 MDS QOS usage of statfs, it should not stuck in waiting.
487 Severity : enhancement
489 Description: remote_acl support
490 Details : Support ACL-based permission check for remote user.
491 Support setfacl/getfacl for remote user with the utils
492 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
493 the system "{s,g}etfacl" utils.
495 Severity : enhancement
497 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
499 Severity : enhancement
501 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
505 Description: make mgs_setparam() handle fsname containing dash
506 Details : fsname containing a dash does not work with lctl conf_param
508 Severity : enhancement
510 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
512 Severity : enhancement
514 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
518 Frequency : rare, at shutdown
519 Description: access already free / zero obd_namespace.
520 Details : if client_disconnect_export was called without force flag set,
521 and exist connect request in flight, this can produce access to
522 NULL pointer (or already free pointer) when connect_interpret
523 store ocd flags in obd_namespace.
527 Frequency : only at startup
528 Description: not alloc memory with spinlock held.
529 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
530 if any spinlock held.
532 Severity : enhancement
534 Description: make lustre randomly fail allocating memory
535 Details : Make lustre randomly failed allocating memory for testing purpose.
537 Severity : enhancement
539 Description: lost problems with lov objid file
540 Details : Fixes some scability and access to not inited memory problems
541 in work with lov objdid file.
546 Description: lfs find does not continue on file error
547 Details : Continue other files processing when a file/dir is absent.
551 Description: Inconsistent usage of lustre_pack_reply()
552 Details : Standardize the usage of lustre_pack_reply() such that it
553 always generate a CERROR on failure.
556 Frequency : occasional
558 Description: MDS hang or stay in waiting lock
559 Details : If client receive lock with CBPENDING flag ldlm need send lock
560 cancel as separate rpc, to avoid situation when cancel request
561 can't processed due all i/o threads stay in wait lock.
565 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
569 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
573 Description: Update to SLES9 kernel-2.6.5-7.287.3.
577 Description: Update to RHEL4 latest kernel.
579 Severity : enhancement
581 Description: Build SLES10 patchless client fails
582 Details : The configure was broken by run ./configure with
583 --with-linux-obj=.... argument for patchless client. When the
584 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
585 can't search header adequately. Use absolute path such as
586 -I($LINUX)/include instead.
590 Description: interrupt oig_wait produce painc on resend.
591 Details : brw_redo_request can be used for resend requests from ptlrpcd and
592 private set, and this produce situation when rq_ptlrpcd_data not
593 copyed to new allocated request and triggered LBUG on assert
594 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
595 ptlrpcd set if request is changed and can be safety changed to use
600 Description: LASSERT_{REQ,REP}SWAB macros are buggy
601 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
602 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
603 these macros and replace them with their difinitions instead.
607 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
608 Details : Update which_patch & target file for SLES10 latest kernel.
611 Description: Remove client patches from SLES 10 kernel.
612 Details : This causes SLES 10 clients to behave as patchless clients
613 even on a Lustre-patched (server) kernel.
615 Severity : enhancement
617 Description: self-adjustable client's lru lists
618 Details : use adaptive algorithm for managing client cached locks lru
619 lists according to current server load, other client's work
620 pattern, memory activities, etc. Both, server and client
621 side namespaces provide number of proc tunables for controlling
624 Severity : enhancement
626 Description: light-weight GSS support
627 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
628 and reduce performance overhead.
630 Severity : enhancement
632 Description: Linux keyring support
633 Details : Support using service of Linux keyring for Lustre GSS internal
634 context refresh/cache mechanism.
638 Description: Fix errors in lfs documentation
639 Details : Fixes man pages
643 Description: don't use GFP_* in generic Lustre code.
644 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
648 Description: obdclass is limited by single OBD_ALLOC(idarray)
649 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
653 Description: warnings with build patchless client with vanila 2.6.19 and up
654 Details : change old ctl_table style and replace ctl_table/ctl_table_header
655 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
659 Frequency : Only for SLES
660 Description: sanity_quota fail test_1
661 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
662 file, which makes TSTID[2] inunique.
666 Frequency : Only for SLES9
667 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
668 Details : Update target/ChangeLog/which_patch .
672 Frequency : Only for test_10 in sanity-quota.sh
673 Description: a bug in quota test code
674 Details : checking a test flag in a wrong place
678 Frequency : Only for test_13 in sanity-quota.sh
679 Description: a bug in quota test script
680 Details : shouldn't check overfull allocated quota
684 Description: jbd statistics
685 Details : Port older jbd statistics patch for sles10
687 Severity : enhancement
689 Description: Add printing inode info into message about error in writepage.
694 Description: reply_lock_interpret crash due to race with it and lock cancel.
695 Details : Do not replay locks that are being cancelled. Do not reference
696 locks by their address during replay, just by their handle.
700 Frequency : When flocks are used.
701 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
702 Details : Flock locks might destroy just granted lock if it could be merged
703 with another existing flock, this is done in completion handler,
704 so teach ldlm_cli_enquque_fini that this is a valid case for
710 Description: Oops in read and write path when failing to allocate lock.
711 Details : Check if lock allocation failed and return error back.
715 Description: lstripe command fails for valid OST index
716 Details : The stripe offset is compared to lov->desc.ld_tgt_count
717 instead of lov->desc.ld_active_tgt_count.
721 Description: sanity.sh failed test 103
722 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
727 Description: df doesn't work properly if diskfs blocksize != 4K
728 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
732 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
733 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
735 Severity : enhancement
737 Description: Adaptive timeouts
738 Details : RPC timeouts adapt to changing server load and network
739 conditions to reduce resend attempts and improve recovery time.
743 Description: llapi_file_create() does not allow some changes
744 Details : add llapi_file_open() that allows specifying the mode and
745 open flags, and also returns an open file handle.
749 Description: merge and cleanup kernel patches.
750 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
754 Description: Add journal checksum support.(Kernel part)
755 Details : The journal checksum feature adds two new flags i.e
756 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
757 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
758 indicates that the commit block contains the checksum for
759 the blocks described by the descriptor blocks. Now commit
760 record can be sent to disk without waiting for descriptor
761 blocks to be written to disk. This behavior is controlled
762 using JBD2_FEATURE_ASYNC_COMMIT flag.
766 Description: OSS needs mutliple precreate threads
767 Details : Add ability to start more than one create thread per OSS.
771 Description: Sanity.sh test_65k failed: setstripe should have succeeded
772 Details : execute lfs setstripe on client
776 Description: mds_obd_create error creating tmp object
777 Details : When the user sets quota on root, llog will be affected and can't
778 create files and write files.
781 Frequency : Always on ia64 patchless client, and possibly others.
783 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
784 Details : This allows the patchless client to be loaded on architectures
789 Description: RedHat Update kernel for RHEL5
790 Details : Add the kernel config file for RHEL5.
794 Description: RedHat Update kernel for RHEL5
795 Details : Modify the kernel config file more closer RHEL5.
799 Description: Build failure against Centos5 (RHEL5)
800 Details : Define PAGE_SIZE when it isn't present.
804 Description: test_7 on sanity-quota.sh can't be used on separate machines
805 Details : add facet to handle it
809 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
810 Details : with NFS, the anon dentry's parent was set to itself in
811 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
812 corresponding dentry other than use rec->ur_name.
814 Severity : enhancement
816 Description: lfs setstripe enhancement
817 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
821 Description: enable data checksumming by default
822 Details : enable checksum by default, allow --disable-checksum
823 configure option and "-o nochecksum" mount option. Checksums
824 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
828 Description: lustre support for RHEL5
829 Details : Add support for RHEL5.
833 Description: Client eviction due to failover config
834 Details : after a connection loss, the lustre client should attempt to
835 reconnect to the last active server first before trying the
836 other potential connections.
840 Description: when mds and osts use different quota unit(32bit and 64bit),
841 quota will be released repeatly.
842 Details : void sending multiple quota reqs to mds, which will keep the
843 status between the reqs.
847 Description: osts not allocated evenly to files
848 Details : change the condition to increase offset_idx
851 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
853 Description: Data corruption for OSTs that are formatted larger than 2TB
855 Details : When generating the bio request for lustre file writes the
856 sector number would overflow a temporary variable before being
857 used for the IO. The data reads correctly from Lustre (which
858 will overflow in a similar manner) but other file data or
859 filesystem metadata may be corrupted in some cases.
863 Description: Tune the kernel for good SCSI performance.
864 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
865 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
870 Description: rewrite ext2-derived code in obdclass/uuid.c
871 Details : rewrite inherited code (uuid parsing code from ext2 utils)
872 from scratch preserving functionality.
876 Description: "lfs find -obd UUID" prints directories
877 Details : "lfs find -obd UUID" will return all directory names instead
878 of just file names. It is incorrect because the directories
879 do not reside on the OSTs.
883 Description: use i_size_read and i_size_write in 2.6 port
884 Details : replace inode->i_size access with i_size_read/write()
887 Frequency : when removing large files
889 Description: scheduling issue during removal of large Lustre files
890 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
891 It causes scheduling issues when removing large files (17TB in the
895 Frequency : only with liblustre clients on XT3
897 Description: evictions taking too long
898 Details : allow llrd to evict clients directly on OSTs
901 Frequency : only on ppc
903 Description: /proc/fs/lustre/devices broken on ppc
904 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
905 We should make sure the type of 'index' is loff_t and then cast
906 explicitly as needed below. Do not assign an explicitly cast
910 Frequency : only for rhel5
912 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
913 Details : Modify the target file & which_kernel.
915 Severity : enhancement
917 Description: omit set fsid for export NFS
918 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
919 as FS_REQUIRES_DEV for avoid problems with generate fsid.
923 Frequency : Always, for kernels after 2.6.16
924 Description: Fix warning idr_remove called for id=.. which is not allocated.
925 Details : Last kernels save old s_dev before kill super and not allow
926 to restore from callback - restore it before call kill_anon_super.
930 Description: buffer overruns could theoretically occur
931 Details : llapi_semantic_traverse() modifies the "path" argument by
932 appending values to the end of the origin string, and a
933 overrun may occur. Adding buffer overrun check in liblustreapi.
937 Description: Fix error on 'ls .' at the top of the Lustre mount.
938 Details : Don't revalidate dentry if it is a root dentry.
942 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
943 Details : Modify vm-tunables-rhel4.patch.
947 Description: Kernel config for 2.6.18-vanilla.
948 Details : Modify targets/2.6-vanilla.target.in.
949 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
950 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
951 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
952 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
956 Description: improve handling recoverable errors
957 Details : if request processig with error which can be recoverable on server
958 request should be resend, otherwise page released from cache and
963 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
964 Details : Modify target file & which_patch.
965 A flaw was found in the IA32 system call emulation provided
966 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
967 value could be stored in the %RAX register, which could trigger an
968 out-of-bounds system call table access. An untrusted local user
969 could exploit this flaw to run code in the kernel
970 (ie a root privilege escalation). (CVE-2007-4573).
974 Description: O_DIRECT bypasses client statistics.
975 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
976 read_ahead_stats were updated. copied the stats section from
977 osc_send_oap_rpc() into async_internal().
981 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
985 Description: change order of libsysio includes
986 Details : '#include sysio.h' should always come before '#include xtio.h'
990 Description: handle "serious error: objid * already exists" more gracefully
991 Details : If LAST_ID value on disk is smaller than the objects existing in
992 the O/0/d* directories, it indicates disk corruption and causes an
993 LBUG(). If the object is 0-length, then we should use the existing
994 object. This will help to avoid a full fsck in most cases.
996 Severity : enhancement
998 Description: adapt the lustre_config script to support the upgrade case
999 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1000 server targets to 1.6 in parallel.
1004 Description: To avoid grant space > avaible space when the disk is almost
1005 full. Without this patch you might see the error "grant XXXX >
1006 available" or some LBUG about grant, when the disk is almost
1008 Details : In filter_check_grant, for non_grant cache write, we should
1009 check the left space by if (*left > ungranted + bytes), instead
1010 of (*left > ungranted), because only we are sure the left space
1011 is enough for another "bytes", then the ungrant space should be
1012 increase. In client, we should update cl_avail_grant only there
1013 is OBD_MD_FLGRANT in the reply.
1017 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1018 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1019 described in CVE-2007-4573. This problem could allow untrusted
1020 local users to gain root access.
1023 Frequency : when using O_DIRECT and quotas
1025 Description: Incorrect file ownership on O_DIRECT output files
1026 Details : block usage reported by 'lfs quota' does not take into account
1027 files that have been written with O_DIRECT.
1032 Description: touch file failed when fs is not full
1033 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1034 otherwise we can get ENOSP while fs is not full.
1038 Description: parallel lock callbacks
1039 Details : Instead of sending blocking and completion callbacks as separated
1040 requests, adding them to a set and sending in parallel.
1043 Frequency : only for Cray XT3
1044 Bugzilla : 12829/13455
1045 Description: Changing primary group doesn't change the group lustre assigns to
1047 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1048 overridden with the primary group provided by the group upcall,
1049 whereas the supplied fsgid can be trusted if it is in the list of
1050 supplementary groups returned by the group upcall.
1052 Severity : enhancement
1054 Description: Allow masking D_WARNING, D_ERROR messages from console
1055 Details : Console messages can now be disabled via lnet.printk.
1059 Description: User code with malformed file open parameter crashes client node
1060 Details : Before packing join_file req, all the related reference should be
1061 checked carefully in case some malformed flags cause fake
1062 join_file req on client.
1066 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1067 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1068 mode is equal to the granted mode. However, it should also check
1069 that LDLM_FL_CP_REQD is not set.
1073 Description: Heavy nfs access might result in deadlocks
1074 Details : After ELC code landed, it is now improper to enqueue any mds
1075 locks under och_sem, because enqueue might want to decide to
1076 cancel open locks for same inode we are holding och_sem for.
1080 Description: Client eviction while running blogbench
1081 Details : A lot of unlink operations with concurrent I/O can lead to a
1082 deadlock causing evictions. To address the problem, the number of
1083 oustanding OST_DESTROY requests is now throttled to
1084 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1085 callbacks are processed in priority.
1089 Description: enable ACLs on MDS by default
1090 Details : ACLs must be enabled on MDS by default.
1093 Frequency : PPC/PPC64 only
1095 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1096 Details : fix duplicated definitions between asm-ppc64/types.h and
1097 lustre_types.h on PPC.
1100 Frequency : PPC/PPC64 only
1102 Description: asm-ppc/segment.h does not exist
1103 Details : fix compile issue on PPC.
1107 Description: better handle error messages in extents code
1110 Frequency : RHEL4 only
1112 Description: mkfs is very slow on IA64/RHEL4
1113 Details : A performance regression has been discovered in the MPT Fusion
1114 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1115 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1116 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1118 Severity : enhancement
1120 Description: SNMP support enhancement
1121 Details : Adding total number of sampled request for an MDS node in snmp
1124 Severity : enhancement
1126 Description: Optimize ldlm waiting list processing for PR extent locks
1127 Details : When processing waiting list for read extent lock and meeting read
1128 lock that is same or wider to it that is not contended, skip
1129 processing rest of the list and immediatelly return current
1130 status of conflictness, since we are guaranteed there are no
1131 conflicting locks in the rest of the list.
1135 Description: Time out and refuse to reconnect
1136 Details : When the failover node is the primary node, it is possible
1137 to have two identical connections in imp_conn_list. We must
1138 compare not conn's pointers but NIDs, otherwise we can defeat
1139 connection throttling.
1143 Description: port llog fixes from b1_6 into HEAD
1144 Details : Port llog reference couting and some llog cleanups from b1_6
1145 (bug 10800) into HEAD, for protect from panic and access to already
1146 free llog structures.
1150 Description: Detect stride IO mode in read-ahead
1151 Details : When a client does stride read, read-ahead should detect that and
1152 read-ahead pages according to the detected stride pattern.
1156 Description: data checksumming impacts single node performance
1157 Details : add support for several checksum algorithm. Currently, only CRC32
1158 and Adler-32 are supported. The checksum type can be changed on
1159 the fly via /proc/fs/lustre/osc/*/checksum_type.
1163 Description: use adler32 for page checksums
1164 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1169 Description: build for x2 fails
1170 Details : fix compile issue on Cray systems.
1174 Description: Properly match for duplicate locks
1175 Details : Due to different lock order from skiplists code, we need to
1176 traverse entire list for now
1179 Frequency : only on PPC/SLES10
1181 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1182 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1183 be defined. Add a hack in mkfs_lustre.c to work around this compile
1188 Description: LASSERT on MDS when client holding flock lock dies
1189 Details : ldlm pool logic depends on number of granted locks equal to
1190 number of released locks which is not true for flock locks, so
1191 just exclude such locks from consideration.
1195 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1196 Details : Use fsfilt_sync() for both the whole filesystem sync and
1197 individual file sync to eliminate dangerous inode locking
1198 with I_LOCK that can lead to a deadlock.
1202 Description: performance in 1.6.3
1203 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1204 sure that 1MB requests are not fragmented by the block layer.
1206 Severity : enhancement
1208 Description: organize the server-side client stats on per-nid basis
1209 Details : Change the structure of stats under obdfilter and mds to
1217 The "uuid"s file would list the uuids of _active_ exports.
1218 And the clear entry is to clear all stats and stale nids.
1220 Severity : enhancement
1222 Description: eliminate client locks in face of contention
1223 Details : file contention detection and lockless i/o implementation
1224 for contended files.
1228 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1231 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1233 Description: MDS may lose file striping (and hence file data) in some cases
1234 Details : If there are additional extended attributes stored on the MDS,
1235 in particular ACLs, SELinux, or user attributes (if user_xattr
1236 is specified for the client mount options) then there is a risk
1237 of attribute loss. Additionally, the Lustre file striping
1238 needs to be larger than default (e.g. striped over all OSTs),
1239 and an additional attribute must be stored initially in the
1240 inode and then increase in size enough to be moved to the
1241 external attribute block (e.g. ACL growing in size) for file
1246 Description: skiplist implementation simplification
1247 Details : skiplists are used to group compatible locks on granted list
1248 that was implemented as tracking first and last lock of each lock group
1249 the patch changes that to using doubly linked lists
1253 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1254 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1255 because LASSERT on both the data supplied by a client, and the data
1256 on disk is dangerous and incorrect.
1258 Severity : enhancement
1260 Description: Slow truncate/writes to huge files at high offsets.
1261 Details : Directly associate cached pages to lock that protect those pages,
1262 this allows us to quickly find what pages to write and remove
1263 once lock callback is received.
1267 Description: more ldlm soft lockups
1268 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1269 starve other threads from the resource lock for a long time in
1270 case of long waiting queue, so change the debug level from
1271 D_OTHER to the less frequently used D_INFO.
1273 Severity : enhancement
1275 Description: add -gid, -group, -uid, -user options to lfs find
1279 Description: Hung threads in invalidate_inode_pages2_range
1280 Details : The direct IO path doesn't call check_rpcs to submit a new RPC once
1281 one is completed. As a result, some RPCs are stuck in the queue
1286 Description: filter threads hungs on waiting journal commit
1287 Details : Cleanup filter group llog code, then only filter group llog will
1288 be only created in the MDS/OST syncing process.
1292 Description: Procfs and llog threads access destoryed import sometimes.
1293 Details : Sync the import destoryed process with procfs and llog threads by
1294 the import refcount and semaphore.
1296 Severity : enhancement
1298 Description: openlock cache of b1_6 port to HEAD
1303 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1304 Details : If insertion of an extent fails, then discard the inode
1305 preallocation and free data blocks else it can lead to duplicate
1310 Description: don't always update ctime in ext3_xattr_set_handle()
1311 Details : Current xattr code updates the inode ctime in ext3_xattr_set_handle.
1312 In some cases the ctime should not be updated, for example for
1313 2.0->1.8 compatibility it is necessary to delete an xattr and it
1314 should not update the ctime.
1318 Bugzilla : 15713/16362
1319 Description: Assertion in iopen_connect_dentry in 1.6.3
1320 Details : looking up an inode via iopen with the wrong generation number can
1321 populate the dcache with a disconneced dentry while the inode
1322 number is in the process of being reallocated. This causes an
1323 assertion failure in iopen since the inode's dentry list contains
1324 both a connected and disconnected dentry.
1328 Description: assertion failure in ldlm_handle2lock()
1329 Details : fix a race between class_handle_unhash() and class_handle2object()
1330 introduced in lustre 1.6.5 by bug 13622.
1335 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1336 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1341 Description: oops in sd_iostats_seq_show()
1342 Details : unloading/reloading the scsi low level driver triggers a kernel
1343 bug when trying to access the sd iostat file.
1348 Description: Kernel panics during QLogic driver reload
1349 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1350 patch, causing memory corruption.
1355 Description: journal_dev option does not work in b1_6
1356 Details : pass mount option during pre-mount.
1358 Severity : enhancement
1360 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1361 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1362 extent information of a file. It can be used to map logical blocks
1363 in a file to physical blocks in the block device.
1367 Description: LDLM soft lockups - improvement
1368 Details : It is be possible to send the lock handle along with each read
1369 or write request because the client is already doing a lock match
1370 itself so there isn't any reason the OST should have to re-do that
1373 --------------------------------------------------------------------------------
1375 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
1377 * Support for kernels:
1378 2.6.5-7.283 (SLES 9),
1379 2.6.9-55.EL (RHEL 4),
1380 2.6.16.46-0.14 (SLES 10),
1381 2.6.18.8 vanilla (kernel.org)
1382 * Client support for unpatched kernels:
1383 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1384 2.6.16 - 2.6.22 vanilla (kernel.org)
1385 * Due to recently discovered recovery problems, we do not recommend
1386 using patchless RHEL 4 clients with this or any earlier release.
1387 * Recommended e2fsprogs version: 1.39.cfs8
1388 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1389 * Starting with this release, the ldiskfs backing filesystem required
1390 by Lustre is now in its own package, lustre-ldiskfs. This package
1391 should be installed. It is versioned separately from Lustre and
1392 may be released separately in future.
1396 Description: block reactivating mgc import until all deactivates complete
1397 Details : Fix race when failing back MDT/MGS to itself (testing)
1400 Frequency : at statup only
1402 Description: mds_lov_synchronize race leads to various problems
1403 Details : simultaneous MDT->OST connections at startup can cause the
1404 sync to abort, leaving the OSC in a bad state.
1406 Severity : enhancement
1408 Description: add optional extra BUILD_VERSION info
1409 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1410 to override the lustre version.
1413 Frequency : 2.6.18 servers only
1415 Description: ll_kern_mount() doesn't release the module reference
1416 Details : The ldiskfs module reference count never drops down to 0
1417 because ll_kern_mount() doesn't release the module reference.
1422 Description: server LBUG when using old ost_num_threads parameter
1423 Details : Accept the old ost_num_threads parameter but warn that it
1424 is deprecated, and fix an off-by-one error that caused an LBUG.
1429 Description: Transient SCSI error results in persistent IO issue
1430 Details : iobuf->dr_error is not reinitialized to 0 between two
1434 Frequency : sometimes when underlying device returns I/O errors
1436 Description: OSTs not going read-only during write failures
1437 Details : OSTs are not remounted read-only when the journal commit threads
1438 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1439 instead of the ext3 wrappers.
1443 Description: poor connect scaling with increasing client count
1444 Details : Don't run filter_grant_sanity_check for more than 100 exports
1445 to improve scaling for large numbers of clients.
1448 Frequency : SLES10 only
1450 Description: sanity-quota.sh quotacheck failed: rc = -22
1451 Details : Quotas cannot be enabled on SLES10.
1454 Frequency : liblustre clients only
1456 Description: getdirentries does not give error when run on compute nodes
1457 Details : getdirentries does not fail when the size specified as an argument
1458 is too small to contain at least one entry
1460 Severity : enhancement
1462 Description: Add LNET router traceability for debug purposes
1463 Details : If a checksum failure occurs with a router as part of the
1464 IO path, the NID of the last router that forwarded the bulk data
1465 is printed so it can be identified.
1470 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1471 Details : Due to a race condition, liblustre clients were occasionally
1472 evicted incorrectly.
1474 Severity : enhancement
1476 Description: lfs setstripe use optional parameters instead of postional
1479 Severity : enhancement
1481 Description: Nanosecond timestamp support for ldiskfs
1482 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1483 resolution timestamps. There is not yet support for this at
1484 the Lustre filesystem level.
1487 Frequency : during server recovery
1489 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1490 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1494 Frequency : nfs export on patchless client
1496 Description: connectathon hang when test nfs export over patchless client
1497 Details : Disconnected dentry cannot be found with lookup, so we do not need
1498 to unhash it or make it invalid
1501 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1502 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1503 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1508 Description: open req refcounting wrong on reconnect
1509 Details : If reconnect happened between getting open reply from server and
1510 call to mdc_set_replay_data in ll_file_open, we will schedule
1511 replay for unreferenced request that we are about to free.
1512 Subsequent close will crash in variety of ways.
1513 Check that request is still eligible for replay in
1514 mdc_set_replay_data().
1519 Description: disable writes to filesystem when reading health_check file
1520 Details : the default for reading the health_check proc file has changed
1521 to NOT do a journal transaction and write to disk, because this
1522 can cause reads of the /proc file to hang and block HA state
1523 checking on a healthy but otherwise heavily loaded system. It
1524 is possible to return to the previous behaviour during configure
1525 with --enable-health-write.
1527 Severity : enhancement
1529 Description: 64-bit inode version
1530 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1531 to the inode. This will be required for version-based recovery.
1536 Description: MDS fails to start if a duplicate client export is detected
1537 Details : in some rare cases it was possible for a client to connect to
1538 an MDS multiple times. Upon recovery the MDS would detect this
1539 and fail during startup. Handle this more gracefully.
1541 Severity : enhancement
1543 Description: Add -o localflock option to simulate old noflock
1545 Details : This will achieve local-only flock/fcntl locks
1551 Description: log_commit_thread vs filter_destroy race leads to crash
1552 Details : Take import reference before releasing llog record semaphore
1557 Description: Wrong request locking in request set processing
1558 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1559 imp_delayed_list, in this place should be used imp_lock.
1562 Frequency : when reconnection
1564 Description: Grant Leak when osc reconnect to OST
1565 Details : When osc reconnect ost, OST(filter) should check whether it
1566 should grant more space to client by comparing fed_grant and
1567 cl_avail_grant, and return the granted space to client instead
1568 of "new granted" space, because client will call osc_init_grant
1569 to update the client grant space info.
1572 Frequency : when client reconnect to OST
1574 Description: Grant Leak when osc do resend and replay bulk write
1575 Details : When osc reconnect to OST, OST(filter)should clear grant info of
1576 bulk write request, because the grant info while be sync between
1577 OSC and OST when reconnect, and we should ignore the grant info
1578 these of resend/replay write req.
1583 Description: Grant space more than avaiable left space sometimes.
1584 Details : When then OST is about to be full, if two bulk writing from
1585 different clients came to OST. Accord the avaliable space of the
1586 OST, the first req should be permitted, and the second one
1587 should be denied by ENOSPC. But if the seconde arrived before
1588 the first one is commited. The OST might wrongly permit second
1589 writing, which will cause grant space > avaiable space.
1592 Frequency : when client is evicted
1594 Description: Grant might be wrongly erased when osc is evicted by OST
1595 Details : when the import is evicted by server, it will fork another
1596 thread ptlrpc_invalidate_import_thread to invalidate the
1597 import, where the grant will be set to 0. While the original
1598 thread will update the grant it got when connecting. So if
1599 the former happened latter, the grant will be wrongly errased
1600 because of this race.
1605 Description: Checking Stale with correct fid
1606 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1607 to check whether it is stale, because sometimes, we want the
1608 enqueue happened anyway, and op_data.fid2 will not be initialized.
1610 Severity : enhancement
1612 Description: update patchless client
1613 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1616 Frequency : only with 2.4 kernel
1618 Description: random memory corruption
1619 Details : size of struct ll_inode_info is to big for union inode.u and this
1620 can be cause of random memory corruption.
1625 Description: Memory leak in recovery
1626 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1627 It should also check obd_fail before fsfilt_start, otherwise if
1628 fsfilt_start return -EROFS,(failover mds during mds recovery).
1629 then the req will return with repmsg->transno = 0 and rc = EROFS.
1630 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1631 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1632 be freed no matter whether fsfilt_commit success or not.
1635 Frequency : only with huge count clients
1637 Description: Prevents from taking the superblock lock in llap_from_page for
1639 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1640 for llap_from_page call in ll_removepage prevents from taking the
1641 superblock lock for a soon died page.
1646 Description: Not check open intent error before release open handle
1647 Details : in some rare cases, the open intent error is not checked before
1648 release open handle, which may cause
1649 ASSERTION(open_req->rq_transno != 0), because it tries to release
1650 the failed open handle.
1655 Description: Set cat log bitmap only after create log success.
1656 Details : in some rare cases, the cat log bitmap is set too early. and it
1657 should be set only after create log success.
1661 Description: Accessing a block bevice can re-enable I/O when Lustre is
1662 tearing down a device.
1663 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1668 Description: service threads may hog cpus when there are a lot of requests
1669 Details : Insert cond_resched to give other threads a chance to use some CPU
1674 Description: the cat log was not initialized in recovery
1675 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1676 unlink log on mds will not be initialized until mds post
1677 recovery. And also in mds post recovery, the unlink log will
1678 initialization will be done asynchronausly, so there will be race
1679 between add unlink log and unlink log initialization.
1683 Description: brw_stats were being printed incorrectly
1684 Details : brw_stats were being printed as log2 but all of them were not
1685 recorded as log2. Also remove some code duplication arising from
1686 filter_tally_{read,write}.
1690 Frequency : rare, only in recovery.
1691 Description: ASSERTION(req->rq_type != LI_POISON) failed
1692 Details : imp_lock should be held while iterating over imp_sending_list for
1693 prevent destroy request after get timeout in ptlrpc_queue_wait.
1697 Description: replay-single.sh test 52 fails
1698 Details : A lock's skiplist need to be cleanup when it being unlinked
1699 from its resource list.
1703 Description: Short directio read returns full requested size rather than
1705 Details : Direct I/O operations should return actual amount of bytes
1706 transferred rather than requested size.
1708 Severity : enhancement
1710 Description: metadata RPC reduction (e.g. for rm performance)
1711 Details : decrease the amount of synchronous RPC between clients and servers
1712 by canceling conflicing lock before the operation on the client
1713 and packing thier handles into the main operation RPC to server.
1715 Severity : enhancement
1717 Description: Async OSC create to avoid the blocking unnecessarily.
1718 Details : If a OST has no remain object, system will block on the creating
1719 when need to create a new object on this OST. Now, ways use
1720 pre-created objects when available, instead of blocking on an
1721 empty osc while others are not empty. If we must block, we block
1722 for the shortest possible period of time.
1724 Severity : enhancement
1726 Description: refine locking for avoid write wrong info into lov_objid file
1727 Details : fix possible races with add new target and write/update data in
1730 --------------------------------------------------------------------------------
1732 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1737 Frequency : on some architectures
1739 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1740 Details : Uninitialized flags sometimes cause configuration commands to
1744 Frequency : patchless clients only
1746 Description: missing __iget() symbol export
1747 Details : The __iget() symbol export is missing. To avoid the need for
1748 this on patchless clients the deathrow inode reaper is turned
1749 off, and we depend on the VM to clean up old inodes. This
1750 dependency was during via the fix for bug 12181.
1752 --------------------------------------------------------------------------------
1754 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1756 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1757 INTEROPERATE with older versions automatically. In many cases a
1758 special upgrade step is needed. Please read the
1759 user documentation before upgrading any part of a live system.
1760 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1761 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1762 * WARNING: Lustre configuration and startup changes are required with
1763 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1769 Severity : enhancement
1771 Description: MountConf
1772 Details : Lustre configuration is now managed via mkfs and mount
1773 commands instead of lmc and lconf. New obd types (MGS, MGC)
1774 are added for dynamic configuration management. See
1775 https://mail.clusterfs.com/wikis/lustre/MountConf for
1778 Severity : enhancement
1780 Description: dynamic OST addition
1781 Details : OSTs can now be added to a live filesystem
1783 Severity : enhancement
1785 Description: startup order invariance
1786 Details : MDTs and OSTs can be started in any order. Clients only
1787 require the MDT to complete startup.
1789 Severity : enhancement
1791 Description: parallel, asynchronous orphan cleanup
1792 Details : orphan cleanup is now performed in separate threads for each
1793 OST, allowing parallel non-blocking operation.
1795 Severity : enhancement
1797 Description: optimized stripe assignment
1798 Details : stripe assignments are now made based on ost space available,
1799 ost previous usage, and OSS previous usage, in order to try
1800 to optimize storage space and networking resources.
1802 Severity : enhancement
1804 Description: Permanently set tunables
1805 Details : All writable /proc/fs/lustre tunables can now be permanently
1806 set on a per-server basis, at mkfs time or on a live system.
1808 Severity : enhancement
1810 Description: Lustre message v2
1811 Details : Add lustre message format v2.
1813 Severity : enhancement
1815 Description: client OST exclusion list
1816 Details : Clients can be started with a list of OSTs that should be
1817 declared "inactive" for known non-responsive OSTs.
1821 Description: SPEC SFS validation failure on NFS v2 over lustre.
1822 Details : Changes the blocksize for regular files to be 2x RPC size,
1823 and not depend on stripe size.
1825 Severity : enhancement
1827 Description: Multiple MD RPCs in flight.
1828 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1829 To avoid overly-overloading MDS, introduce a limit on number of
1830 MDS RPCs in flight for a single client and add /proc controls
1831 to adjust this limit.
1833 Severity : enhancement
1835 Description: client read/write statistics
1836 Details : Add client read/write call usage stats for performance
1837 analysis of user processes.
1838 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1839 file access. extents_stats shows chunk size distribution.
1840 extents_stats_per_process show chunk size distribution per
1843 Severity : enhancement
1845 Description: mds statistics
1846 Details : Add detailed mds operations statistics in
1847 /proc/fs/lustre/mds/*/stats.
1851 Description: Failure of copying files with lustre special EAs.
1852 Details : Client side always return success for setxattr call for lustre
1853 special xattr (currently only "trusted.lov").
1858 Description: Refcount LNET uuids
1859 Details : The global LNET uuid list grew linearly with every startup;
1860 refcount repeated list entries instead of always adding to
1863 Severity : enhancement
1865 Description: Dynamic service threads
1866 Details : Within a small range, start extra service threads
1867 automatically when the request queue builds up.
1870 Frequency : mixed-endian client/server environments
1872 Description: mixed-endian crashes
1873 Details : The new msg_v2 system had some failures in mixed-endian
1876 Severity : enhancement
1878 Description: Easy OST removal
1879 Details : OSTs can be permanently deactivated with e.g. 'lctl
1880 conf_param lustre-OST0001.osc.active=0'
1882 Severity : enhancement
1884 Description: MGS proc entries
1885 Details : Added basic proc entries for the MGS showing what filesystems
1888 Severity : enhancement
1890 Description: provide MGS failover
1891 Details : Added config lock reacquisition after MGS server failover.
1893 Severity : enhancement
1895 Description: add Linux 2.4 support
1896 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1900 Description: a large application tries to do I/O to the same resource and dies
1901 in the middle of it.
1902 Details : Check the req->rq_arrival time after the call to
1903 ost_brw_lock_get(), but before we do anything about
1904 processing it & sending the BULK transfer request. This
1905 should help move old stale pending locks off the queue as
1906 quickly as obd_timeout.
1909 Frequency : when an incorrect nid is specified during startup
1911 Description: ptlrpc connect to non-existant node causes kernel crash
1912 Details : LNET can't be re-entered from an event callback, which
1913 happened when we expire a message after the export has been
1914 cleaned up. Instead, hand the zombie cleanup off to another
1917 Severity : enhancement
1919 Description: plain/inodebits lock performance improvement
1920 Details : Grouping plain/inodebits in granted list by their request modes
1921 and bits policy, thus improving the performance of search through
1925 Frequency : only if OST filesystem is corrupted
1927 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1928 Details : for a short time RPCs with bulk IO are in the replay list,
1929 but replay of bulk IOs is unimplemented. If the OST filesystem
1930 is corrupted due to disk cache incoherency and then replay is
1931 started it is possible to trip an assertion. Avoid putting
1932 committed RPCs into the replay list at all to avoid this issue.
1935 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1937 Description: System hang on startup
1938 Details : This bug allowed the liblustre (e.g. catamount) client to
1939 return to the app before handling all startup RPCs. This
1940 could leave the node unresponsive to lustre network traffic
1941 and manifested as a server ptllnd timeout.
1943 Severity : enhancement
1945 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1946 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1947 boolean to control whether to print peer debug info when a
1948 client's RPC times out.
1951 Frequency : only for kernels with patches from Lustre below 1.4.3
1953 Description: Remove old rdonly API
1954 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1957 Frequency : only for devices with external journals
1959 Description: Set external device read-only also
1960 Details : During a commanded failover stop, we set the disk device
1961 read-only while the server shuts down. We now also set any
1962 external journal device read-only at the same time.
1965 Frequency : when upgrading from 1.4 while trying to change parameters
1967 Description: The wrong (new) MDC name was used when setting parameters for
1968 upgraded MDT's. Also allows changing of OSC (and MDC)
1969 parameters if --writeconf is specified at tunefs upgrade time.
1972 Frequency : when setting specific ost indicies
1974 Description: QOS code breaks on skipped indicies
1975 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1976 created with --index need not be sequential.
1978 Severity : enhancement
1980 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1981 Details : The uninit_groups feature works in conjunction with the kernel
1982 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1983 up the pass1 processing of e2fsck. This is a read-only feature
1984 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1985 mount filesystems that have had this feature enabled.
1987 Severity : enhancement
1989 Description: Improve multi-block allocation algorithm to avoid fragmentation
1990 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1991 allocation locality and avoid filesystem fragmentation.
1993 ------------------------------------------------------------------------------
1995 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1997 * Support for kernels:
1998 2.4.21-47.0.1.EL (RHEL 3)
1999 2.6.5-7.283 (SLES 9)
2000 2.6.9-42.0.10.EL (RHEL 4)
2001 2.6.12.6 vanilla (kernel.org)
2002 2.6.16.27-0.9 (SLES 10)
2003 * Recommended e2fsprogs version: 1.39.cfs5
2005 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2009 Frequency : occasional, depends on client load and configuration
2010 Bugzilla : 12181, 12203
2011 Description: data loss for recently-modified files
2013 Details : In some cases it is possible that recently written or created
2014 files may not be written to disk in a timely manner (this should
2015 normally be within 30s unless client IO load is very high).
2016 The problem appears as zero-length files or files that are a
2017 multiple of 1MB in size after a client crash or client eviction
2018 that are missing data at the end of the file.
2020 This problem is more likely to be hit on clients where files are
2021 repeatedly created and unlinked in the same directory, clients
2022 have a large amount of RAM, have many CPUs, the filesystem has
2023 many OSTs, the clients are rebooted frequently, and/or the files
2024 are not accessed by other nodes after being written.
2026 The presence of the problem can be detected by looking at
2027 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2028 smaller than the second (nr_unused) then dirty files will not
2029 be flushed automatically to disk. "sync; sleep 10" should be
2030 run several times on the node before unmounting it to update
2031 Lustre (this is also safe to run on nodes without this problem).
2033 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2034 that can cause this same problem, so customers using that kernel
2035 also need to update the kernel in addition to Lustre. In order
2036 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2038 It is normal that files written just before a client crash (less
2039 than 30s) may not yet have been flushed to disk, even for local
2043 Frequency : frequent on thin XT3 nodes
2045 Description: UUID collision on thin XT3 Linux nodes
2046 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2047 randomly, since we relied on an insufficiently-seeded PRNG.
2052 Description: OSS hangs after "All ost request buffers busy"
2053 Details : A deadlock between quota and journal operations caused OSS
2054 hangs after printing "All ost request buffers busy."
2057 Frequency : always on liblustre builds
2059 Description: Cleanup compiler warnings on liblustre
2062 Frequency : always on liblustre builds on XT3
2064 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2068 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2069 > 32000 subdirectories
2070 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2071 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2072 subdirectory count crosses 32000. This will aid e2fsck to
2073 correctly handle more than 32000 subdirectories.
2076 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2078 Description: System hang on startup
2079 Details : This bug allowed the liblustre (e.g. catamount) client to
2080 return to the app before handling all startup RPCs. This
2081 could leave the node unresponsive to lustre network traffic
2082 and manifested as a server ptllnd timeout.
2084 Severity : enhancement
2086 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2087 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2088 boolean to control whether to print peer debug info when a
2089 client's RPC times out.
2094 Description: make O_SYNC working on 2.6 kernels
2095 Details : 2.6 kernels use different method for mark pages for write,
2096 so need add a code to lustre for O_SYNC work.
2101 Description: Failure to close file and release space on NFS
2102 Details : Put inode details into lock acquired in ll_intent_file_open.
2103 Use mdc_intent_lock in ll_intent_open to properly
2104 detect all kind of errors unhandled by mdc_enqueue.
2109 Description: proc file read during shutdown sometimes raced obd removal,
2111 Details : Add lock to prevent obd access after proc file removal.
2114 Frequency : Only for files larger than 4GB on 32-bit clients.
2116 Description: improperly doing page alignment of locks
2117 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2118 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2123 Description: RPCs being resent when they shouldn't be
2124 Details : Some RPCs that should not be resent are being resent. This
2125 can cause inconsistencies in the RPC state machine. Do not
2126 resend such requests.
2129 Frequency : rare, only with NFS export
2131 Description: Crash on NFS re-export node
2132 Details : under very unusual load conditions an assertion is hit in
2133 ll_intent_file_open()
2136 Frequency : only if OST filesystem is corrupted
2138 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2139 Details : for a short time RPCs with bulk IO are in the replay list,
2140 but replay of bulk IOs is unimplemented. If the OST filesystem
2141 is corrupted due to disk cache incoherency and then replay is
2142 started it is possible to trip an assertion. Avoid putting
2143 committed RPCs into the replay list at all to avoid this issue.
2148 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2149 Details : Large single O_DIRECT read and write calls can fail to allocate
2150 a sufficiently large buffer to process the request. In case of
2151 allocation failure the allocation is retried with a smaller
2152 buffer and broken into smaller requests.
2154 Severity : enhancement
2156 Description: Add -o localflock option to simulate old noflock behaviour.
2157 Details : This will achieve local-only flock/fcntl locks coherentness.
2162 Description: versioning check is incomplete
2163 Details : Checking the version difference of client vs. server, report
2164 error if the gap is too big.
2169 Description: add support PG_writeback bit
2170 Details : add support for PG_writeback bit for Lustre, for more carefull
2171 work with page cache in 2.6 kernel. This also fix some deadlocks
2172 and remove hack for work O_SYNC with 2.6 kernel.
2174 Severity : enhancement
2176 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2177 Details : The uninit_groups feature works in conjunction with the kernel
2178 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2179 up the pass1 processing of e2fsck. This is a read-only feature
2180 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2181 mount filesystems that have had this feature enabled.
2183 Severity : enhancement
2185 Description: Improve multi-block allocation algorithm to avoid fragmentation
2186 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2187 allocation locality and avoid filesystem fragmentation.
2189 ------------------------------------------------------------------------------
2191 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2193 * Support for kernels:
2194 2.6.9-42.0.3.EL (RHEL 4)
2195 2.6.5-7.276 (SLES 9)
2196 2.4.21-47.0.1.EL (RHEL 3)
2197 2.6.12.6 vanilla (kernel.org)
2198 2.6.16.21-0.8 (SLES10)
2199 * Recommended e2fsprogs version: 1.39.cfs2-0
2201 * The backwards-compatible /proc/sys/portals symlink has been removed
2202 in this release. Before upgrading, please ensure that you change
2203 any configuration scripts or /etc/sysctl.conf files that access
2204 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2205 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2206 in advance of the upgrade on any system running Lustre 1.4.6 or
2207 newer, since /proc/sys/lnet was added in that version.
2208 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2212 Frequency : only when quota is used
2214 Description: avoid scanning export list for quota master
2215 Details : Change the algorithms to avoid scanning export list in order
2216 to improve the efficiency.
2219 Frequency : MDS failover only, very rarely
2221 Description: "went back in time" messages on mds failover
2222 Details : The greatest transno may be lost when the current operation
2223 finishes with an error (transno==0) and the client's last_rcvd
2224 record is over-written. Save the greatest transno in the
2225 mds_last_transno for this case.
2228 Frequency : always for specific kernels and striping counts
2230 Description: client may get "Matching packet too big" without ACL support
2231 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2232 when trying to access files in certain configurations. The
2233 clients should in fact be denied when mounting because they do
2234 not understand ACLs.
2237 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2239 Description: many clients connecting with IO in progress causes connect timeouts
2240 Details : Avoid synchronous journal commits to avoid delays caused by many
2241 clients connecting/disconnecting when bulk IO is in progress.
2242 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2243 OST_IO_PORTAL to avoid delays behind potentially many pending
2247 Frequency : occasionally with multiple writers to a single file
2249 Description: shared writes to file may result in wrong size reported by stat()
2250 Details : Allow growing of kms when extent lock is cancelled
2253 Frequency : always with random mmap IO to multi-striped file
2255 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2256 Details : Only if the hole is at the end of OST object so that kms is too
2257 small. Fix is to increase kms accordingly in ll_nopage.
2260 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2262 Description: writes to a missing object would leak memory on the OST
2263 Details : If there is an inconsistency between the MDS and OST filesystems,
2264 such that the MDS references an object that doesn't exist, writes
2265 to that object will leak memory due to incorrect cleanup in the
2266 error handling path, eventually running out of memory on the OST.
2271 Description: Creating too long symlink causes lustre errors
2272 Details : Check symlink and name lengths before sending requests to MDS.
2275 Frequency : only if flock is enabled (not on by default)
2277 Description: posix locks not released on fd closure on 2.6.9+
2278 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2279 kernels, this caused such locks not to be released on fd close and
2280 then assertions on fs unmount about still used locks.
2283 Frequency : MDS failover only, very rarely
2285 Description: clients may get ASSERTION(granted_lock != NULL)
2286 Details : When request was taking a long time, and a client was resending
2287 a getattr by name lock request. The were multiple lock requests
2288 with the same client lock handle and
2289 mds_getattr_name->fixup_handle_for_resent_request found one of the
2290 lock handles but later failed with ASSERTION(granted_lock != NULL).
2295 Description: handle->h_buffer_credits > 0, assertion failure
2296 Details : h_buffer_credits is zero after truncate, causing assertion
2297 failure. This patch extends the transaction or creates a new
2301 Frequency : NFS re-export or patchless client
2302 Bugzilla : 11179, 10796
2303 Description: Crash on NFS re-export node (__d_move)
2304 Details : We do not want to hash the dentry if we don't have a lock.
2305 But if this dentry is later used in d_move, we'd hit uninitialised
2306 list head d_hash, so we just do this to init d_hash field but
2307 leave dentry unhashed.
2310 Frequency : NFS re-export or patchless client
2312 Description: NFS exports has problem with symbolic link
2313 Details : lustre client didn't properly install dentry when re-exported
2314 to NFS or running patchless client.
2317 Frequency : NFS re-export or patchless client
2319 Description: Various nfs/patchless fixes.
2320 Details : fixes reuse disconected alias for lookup process - this fixes
2321 warning "find_exported_dentry: npd != pd",
2322 fix permission error with open files at nfs.
2323 fix apply umask when do revalidate.
2326 Frequency : occasional
2328 Description: Crash on NFS re-export node
2329 Details : calling clear_page() on the wrong pointer triggered oops in
2330 generic_mapping_read().
2333 Frequency : rarely, using O_DIRECT IO
2335 Description: unaligned directio crashes client with LASSERT
2336 Details : check for unaligned buffers before trying any requests.
2339 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2341 Description: stale data returned from RAID cache
2342 Details : If only a small amount of IO is done to the RAID device before
2343 reading it again it is possible to get stale data from the RAID
2344 cache instead of reading it from disk.
2347 Frequency : always for sles10 kernel
2349 Description: sles10 support
2350 Details : ll_follow_link: compile fixes and using of nd_set_link
2351 under newer kernels.
2354 Frequency : depends on arch, kernel and compiler version, always on sles10
2357 Description: recursive or deep enough symlinks cause stack overflow
2358 Details : getting rid of large stack-allocated variable in
2362 Frequency : depends on hardware
2364 Description: lustre write performance loss in the SLES10 kernel
2365 Details : the performance loss is caused by using of write barriers in the
2366 ext3 code. The SLES10 kernel turns barrier support on by
2367 default. The fix is to undo that change for ldiskfs.
2369 ------------------------------------------------------------------------------
2371 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2373 * Support for kernels:
2374 2.6.9-42.0.3EL (RHEL 4)
2375 2.6.5-7.276 (SLES 9)
2376 2.4.21-47.0.1.EL (RHEL 3)
2377 2.6.12.6 vanilla (kernel.org)
2381 Frequency : quota enabled and large files being deleted
2383 Description: releasing more than 4GB of quota at once hangs OST
2384 Details : If a user deletes more than 4GB of files on a single OST it
2385 will cause the OST to spin in an infinite loop. Release
2386 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2391 Description: statfs data retrieved from /proc may be stale or zero
2392 Details : When reading per-device statfs data from /proc, in the
2393 {kbytes,files}_{total,free,avail} files, it may appear
2394 as zero or be out of date.
2397 Frequency : systems with MD RAID1 external journal devices
2399 Description: lconf's call to blkid is confused by RAID1 journal devices
2400 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2401 of returning all block devices that match the journal UUID.
2404 Frequency : always, for aggregate stripe size over 4GB
2406 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2407 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2408 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2411 Frequency : always if "lfs find" used on a local file/directory
2413 Description: "lfs find" segfaults if used on a local file/directory
2414 Details : The case where a directory component was not specified wasn't
2415 handled correctly. Handle this properly.
2418 Frequency : always on ppc64
2420 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2421 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2422 perform endianness conversion.
2425 Frequency : rarely (truncate to non-zero file size after write under load)
2426 Bugzilla : 10730, 10687
2427 Description: Files padded with zeros to next 4K multiple
2428 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2429 it is possible that files that are truncated to a non-zero size
2430 immediately after being written are filled with zero bytes beyond
2431 the truncated size. No file data is lost.
2433 Severity : enhancement
2435 Description: Allow recovery/failover for liblustre clients.
2436 Details : liblustre clients were unaware of failover configurations until
2439 Severity : enhancement
2441 Description: user file locks should fail when not mounting with flock option
2442 Details : Set up an error-returning stub in ll_file_operations.lock field
2443 to prevent incorrect behaviour when client is mounted without
2444 flock option. Also, set up properly f_op->flock field for
2448 Frequency : always on ia64
2450 Description: "lfs df" loops on printing out MDS statfs information
2451 Details : The obd_ioctl_data was not initialized and in some systems
2452 this caused a failure during the ioctl that did not return
2453 an error. Initialize the struct and return an error on failure.
2456 Frequency : SLES 9 only
2458 Description: Error of copying files with lustre special EAs as root
2459 Details : Client side always return success for setxattr call for lustre
2460 special xattr (currently only "trusted.lov").
2463 Frequency : rarely on clusters with both ia64+i386 clients
2465 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2466 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2467 clients are doing concurrent writes to the same file it is
2468 possible that the ia64 clients may LASSERT because the OST
2469 extent locks are not PAGE_SIZE aligned. Ensure that grown
2470 locks are always aligned on the request boundary.
2473 Frequency : specific use, occasional
2475 Description: Overwriting in use executable truncates on-disk binary image
2476 Details : If one node attempts to overwrite an executable in use by
2477 another node, we now correctly return ETXTBSY instead of
2478 truncating the file.
2480 Severity : enhancement
2482 Description: Async OSC create to avoid the blocking unnecessarily.
2483 Details : If a OST has no remain object, system will block on the creating
2484 when need to create a new object on this OST. Now, ways use
2485 pre-created objects when available, instead of blocking on an
2486 empty osc while others are not empty. If we must block, we block
2487 for the shortest possible period of time.
2492 Description: chmod on Lustre root is propagated to other clients
2493 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2494 invalid by the follow_mount time.
2497 Frequency : liblustre clients only
2499 Description: Race in 'instant cancel' lock handling could lead to such locks
2500 never to be granted in case of SMP MDS
2501 Details : Do not destroy not yet granted but cbpending locks in
2505 Frequency : replay/resend of open
2507 Description: non null lock assetion failure in mds_intent_policy
2508 Details : Trying to replay/resend lockless open requests resulted in
2509 mds_open() returning 0 with no lock. Now it sets a flag if
2510 a lock is going to be returned.
2512 Severity : enhancement
2514 Description: Checksum enhancements
2515 Details : New checksum enhancements allow for resending RPCs that failed
2518 Severity : enhancement
2520 Description: Tunables on number of dirty pages in cacche
2521 Details : Allow to set limit on number of dirty pages cached.
2526 Description: client crash on unmount - lock still has references
2527 Details : In some error handling cases it was possible to leak a lock
2528 reference on a client while accessing a file. This was not
2529 harmful to the client during operation, but would cause the
2530 client to crash when the filesystem is unmounted.
2533 Frequency : specific case, rare
2535 Description: ETXTBSY on mds though file not in use by client
2536 Details : ETXTBSY is no longer incorrectly returned when attempting to
2537 chmod or chown a directory that the user previously tried to
2538 execute or a currently-executing binary.
2541 Frequency : extremely rare except on liblustre-based clients
2543 Description: Lustre space not freed when files are deleted
2544 Details : Clean up open-unlinked files after client eviction. Previously
2545 the unlink was skipped and the files remained as orphans.
2550 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2551 Details : In some cases it was possible to send a blocking callback to
2552 a client doing a glimpse, even though that client didn't get
2553 a lock granted. When the glimpse lock is cancelled on the OST
2554 the freed lock is left on the waiting list and corrupted the list.
2557 Frequency : all core dumps
2559 Description: Broke core dumps to lustre
2560 Details : Negative dentry may be unhashed if parent does not have UPDATE
2561 lock, but some callers, e.g. do_coredump, expect dentry to be
2562 hashed after successful create, hash it in ll_create_it.
2564 ------------------------------------------------------------------------------
2566 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2568 * Support for kernels:
2569 2.6.9-42.0.2.EL (RHEL 4)
2570 2.6.5-7.276 (SLES 9)
2571 2.4.21-40.EL (RHEL 3)
2572 2.6.12.6 vanilla (kernel.org)
2576 Frequency : always on RHEL 3
2578 Description: Number of open files grows over time
2579 Details : The number of open files grows over time, whether or not
2580 Lustre is started. This was due to a filp leak introduced
2581 by one of our kernel patches.
2583 ------------------------------------------------------------------------------
2585 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2587 * Support for kernels:
2588 2.6.9-42.EL (RHEL 4)
2589 2.6.5-7.276 (SLES 9)
2590 2.4.21-40.EL (RHEL 3)
2591 2.6.12.6 vanilla (kernel.org)
2596 Bugzilla : 5719, 9635, 9792, 9684
2597 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2598 Details : If a server is under heavy load and cannot reply to new
2599 connection requests before the client resends the (re)connect,
2600 the connection handling code can behave badly if two service
2601 threads are concurrently handing separate (re)connections from
2602 the same client. Add better locking to the connection handling
2603 code, and ensure that only a single connection will be processed
2604 for a given client UUID, even if the lock is dropped.
2606 Severity : enhancement
2608 Description: add TCP zero-copy support to kernel
2609 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2610 sends if the hardware supports scatter-gather and checksumming.
2611 This allows socklnd to do client-write and server-read more
2612 efficiently and reduce CPU utilization from skbuf copying.
2615 Frequency : only if NFS exporting from client
2617 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2618 Details : If setting attributes on a file created under NFS that had
2619 never been opened it would be possible to oops the client
2620 if the file had no objects.
2623 Frequency : always for liblustre
2625 Description: liblustre client does MDS+OSTs setattr RPC for each write
2626 Details : When doing a write from a liblustre client, the client
2627 incorrectly issued an RPC to the MDS and each OST the file was
2628 striped over in order to update the timestamps. When writing
2629 with small chunks and many clients this could overwhelm the MDS
2630 with RPCs. In all cases it would slow down the write because
2631 these RPCs are unnecessary.
2633 Severity : enhancement
2635 Description: allow number of MDS service threads to be changed at module load
2636 Details : It is now possible to change the number of MDS service threads
2637 running. Adding "options mds mds_num_threads={N}" to the MDS's
2638 /etc/modprobe.conf will set the number of threads for the next
2639 time Lustre is restarted (assuming the "mds" module is also
2640 reloaded at that time). The default number of threads will
2641 stay the same, 32 for most systems.
2646 Description: OST crash if filesystem is unformatted or corrupt
2647 Details : If an OST is started on a device that has never been formatted
2648 or if the filesystem is corrupt and cannot even mount then the
2649 error handling cleanup routines would dereference a NULL pointer.
2654 Description: NULL pointer deref in llap_from_page.
2655 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2656 incorrect) mapping if the page was truncated/reclaimed while it was
2657 searched for. Check for this condition and skip such pages when
2658 doing readahead. Introduce extra check to llap_from_page() to
2659 verify page->mapping->host is non-NULL (so page is not anonymous).
2662 Frequency : Sometimes when using sys_sendfile
2664 Description: "page not covered by a lock" warnings from ll_readpage
2665 Details : sendfile called ll_readpage without right page locks present.
2666 Now we introduced ll_file_sendfile that does necessary locking
2667 around call to generic_file_sendfile() much like we do in
2671 Frequency : with certain MDS communication failures at client mount time
2673 Description: NULL pointer deref after failed client mount
2674 Details : a client connection request may delayed by the network layer
2675 and not be sent until after the PTLRPC layer has timed out the
2676 request. If the client fails the mount immediately it will try
2677 to clean up before the network times out the request. Add a
2678 reference from the request import to the obd device and delay
2679 the cleanup until the network drops the request.
2682 Frequency : occasionally during client (re)connect
2684 Description: assertion failure during client (re)connect
2685 Details : processing a client connection request may be delayed by the
2686 client or server longer than the client connect timeout. This
2687 causes the client to resend the connection request. If the
2688 original connection request is replied in this interval, the
2689 client may trip an assertion failure in ptlrpc_connect_interpret()
2690 which thought it would be the only running connect process.
2693 Frequency : only with obd_echo servers and clients that are rebooted
2695 Description: kernel BUG accessing uninitialized data structure
2696 Details : When running an obd_echo server it did not start the ping_evictor
2697 thread, and when a client was evicted an uninitialized data
2698 structure was accessed. Start the ping_evictor in the RPC
2699 service startup instead of the OBD startup.
2701 Severity : enhancement
2702 Bugzilla : 10193 (patchless)
2703 Description: Remove dependency on various unexported kernel interfaces.
2704 Details : No longer need reparent_to_init, exit_mm, exit_files,
2705 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2708 Frequency : rare (only users of deprecated and unsupported LDAP config)
2710 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2711 Details : LDAP apparently contains 'lustreName' attributes instead of
2712 'name'. A simple remapping of the name is sufficient.
2715 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2717 Description: waiting_locks_callback trips kernel BUG if client is evicted
2718 Details : Running with the dump_on_timeout debug flag turned on makes
2719 it possible that the waiting_locks_callback() can try to dump
2720 the Lustre kernel debug logs from an interrupt handler. Defer
2721 this log dumping to the expired_lock_main() thread.
2723 Severity : enhancement
2725 Description: Support NFS exporting on 2.6 kernels.
2726 Details : Implement non-rawops metadata methods for NFS server to use without
2727 changing NFS server code.
2730 Frequency : very rare (synthetic metadata workload only)
2732 Description: two racing renames might cause an MDS thread to deadlock
2733 Details : Running the "racer" program may cause one MDS thread to rename
2734 a file from being the source of a rename to being the target of
2735 a rename at exactly the same time that another thread is doing
2736 so, and the second thread has already enqueued these locks after
2737 doing a lookup of the target and is trying to relock them in
2738 order. Ensure that we don't try to re-lock the same resource.
2741 Frequency : only very large systems with liblustre clients
2743 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2744 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2745 all OSTs in parallel. This allows the eviction of stale liblustre
2746 clients to proceed much faster than if they were done in series,
2747 and also offers similar improvements for other set_info RPCs.
2752 Description: excessive CPU usage during initial read phase on client
2753 Details : During the initial read phase on a client, it would agressively
2754 retry readahead on the file, consuming too much CPU and impacting
2755 performance (since 1.4.5.8). Improve the readahead algorithm
2756 to avoid this, and also improve some other common cases (read
2757 of small files in particular, where "small" is files smaller than
2758 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2763 Description: MDS crash when receiving packet with unknown intent.
2764 Details : Do not LBUG in unknown intent case, just return -EFAULT
2766 Severity : enhancement
2767 Bugzilla : 9293, 9385
2768 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2769 Details : Do not serialize getattr (non-intent version) and statfs.
2772 Frequency : occasional, when OST network is overloaded/intermittent
2774 Description: client evicted by OST after bulk IO timeout
2775 Details : If a client sends a bulk IO request (read or write) the OST
2776 may evict the client if it is unresposive to its data GET/PUT
2777 request. This is incorrect if the network is overloaded (takes
2778 too long to transfer the RPC data) or dropped the OST GET/PUT
2779 request. There is no need to evict the client at all, since
2780 the pinger and/or lock callbacks will handle this, and the
2781 client can restart the bulk request.
2784 Frequency : Always when mmapping file with no objects
2786 Description: client crashes when mmapping file with no objects
2787 Details : Check that we actually have objects in a file before doing any
2788 operations on objects in ll_vm_open, ll_vm_close and
2794 Description: Request leak when working with deleted CWD
2795 Details : Introduce advanced request refcount tracking for requests
2796 referenced from lustre intent.
2798 Severity : Enhancement
2800 Description: Cache open file handles on client.
2801 Details : MDS now will return special lock along with openhandle, if
2802 requested and client is allowed to hold openhandle, even if unused,
2803 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2804 closing files for every read/write openration.
2806 Severity : Enhancement
2808 Description: Cache open negative dentries on client when possible.
2809 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2810 negative dentries on lock revocation.
2815 Description: Remounting a client read-only wasn't possible with a zconf mount
2816 Details : It wasn't possible to remount a client read-only with llmount.
2818 Severity : enhancement
2819 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2820 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2821 MPICH is included to add Lustre-specific ADIO interfaces.
2822 This is based closely on the UFS ADIO layer and only differs
2823 in file creation, in order to allow the OST striping to be set.
2824 This is user-contributed code and not supported by CFS.
2829 Description: extended inode attributes (immutable, append-only) work improperly
2830 when 2.4 and 2.6 kernels are used on client/server or vice versa
2831 Details : Introduce kernel-independent values for these flags.
2833 Severity : enhancement
2836 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2837 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2838 granularity. Now, Fractional MB granularity is be supported,
2839 this is very useful for low memory system.
2841 Severity : enhancement
2843 Description: Getattr by fid
2844 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2845 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2846 have locks and avoids getattr rpc requests for it on every stat.
2849 Frequency : Always, for filesystems larger than 2TB
2851 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2852 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2853 filesystems larger than 2TB to be mounted with mballoc enabled.
2856 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2858 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2859 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2860 filesystems larger than 2TB is mounted then the kernel will
2861 silently corrupt the start of the filesystem. CONFIG_LBD is
2862 enabled for all CFS-supported kernels, but the possibility of
2863 this happening with a modified kernel config exists.
2865 Severity : enhancement
2867 Description: add client O_DIRECT support for 2.6 kernels
2868 Details : It is now possible to do O_DIRECT reads and writes to files
2869 in the Lustre client mountpoint on 2.6 kernel clients.
2871 Severity : enhancement
2873 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2874 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2875 parallel, not waiting for response from every OST before sending
2876 a rpc to the next OST.
2881 Description: setattr vs write race when updating file timestamps
2882 Details : Client processes that update a file timestamp into the past
2883 right after writing to the file (e.g. tar) it is possible that
2884 the updated file modification time can be reset to the current
2885 time due to a race between processing the setattr and write RPC.
2887 Severity : enhancement
2889 Description: Bring 'lfs find' closer in line with regular Linux find.
2890 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2891 -print0 options and obtains all the needed info through the lustre
2894 Severity : enhancement
2896 Description: support up to 1024 configured devices on one node
2897 Details : change obd_dev array from statically allocated to dynamically
2898 allocated structs as they are first used to reduce memory usage
2903 Description: Flush dirty partially truncated pages during truncate
2904 Details : Immediatelly flush partially truncated pages in filter_setattr,
2905 this way we completely avoid having any pages in page cache on OST
2906 and can retire ugly workarounds during writes to flush such pages.
2911 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2912 Details : protect inode from truncation within vfs_unlink() context
2913 just take a reference before calling vfs_unlink() and release it
2914 when parent's i_sem is free.
2919 Description: last_id value checked outside lock on OST caused LASSERT failure
2920 Details : If there were multiple MDS->OST object precreate requests in
2921 flight, it was possible that the OST's last object id was checked
2922 outside a lock and incorrectly tripped an assertion. Move checks
2923 inside locks, and discard old precreate requests.
2926 Frequency : always, if extents are used on OSTs
2928 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2929 Details : OSTs using the extents format would not zero the high 16 bits of
2930 the index physical block number. This is not a problem for any
2931 OST filesystems smaller than 16TB, and no kernels support ext3
2932 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2933 new/modified files) and can be fixed for existing filesystems
2934 with e2fsprogs-1.39-cfs1.
2939 Description: import connection selection may be incorrect if timer wraps
2940 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2941 import connections to be ignored if the 32-bit jiffies counter
2942 wraps. Use a 64-bit jiffies counter.
2945 Frequency : very large clusters immediately after boot
2947 Description: LNET request buffers exhausted under heavy short-term load
2948 Details : If a large number of client requests are generated on a service
2949 that has previously never seen so many requests it is possible
2950 that the request buffer growth cannot keep up with the spike in
2951 demand. Instead of dropping incoming requests, they are held in
2952 the LND until the RPC service can accept more requests.
2955 Frequency : Sometimes during replay
2957 Description: Assertion failure in ll_local_open after replay.
2958 Details : If replay happened on an open request reply before we were able
2959 to set replay handler, reply will become not swabbed tripping the
2960 assertion in ll_local_open. Now we set the handler right after
2961 recognising of open request
2964 Frequency : very rare
2966 Description: kernel reports "badness in vsnprintf"
2967 Details : Reading from the "recovery_status" /proc file in small chunks
2968 may cause a negative length in lprocfs_obd_rd_recovery_status()
2969 call to vsnprintf() (which is otherwise harmless). Exit early
2970 if there is no more space in the output buffer.
2972 Severity : enhancement
2974 Description: clear OBD RPC statistics by writing to them
2975 Details : It is now possible to clear the OBD RPC statistics by writing
2976 to the "stats" file.
2981 Description: Client mtime is not the same on different clients after utimes
2982 Details : In some cases, the client was using the utimes() syscall on
2983 a file cached on another node. The clients now validate the
2984 ctime from the MDS + OSTs to determine which one is right.
2989 Description: Inability to activate failout mode
2990 Details : lconf script incorrectly assumed that in pythong string's numeric
2991 value is used in comparisons.
2994 Frequency : always with multiple stripes per file
2996 Description: Inefficient object allocation for mutli-stripe files
2997 Details : When selecting which OSTs to stripe files over, for files with
2998 a stripe count that divides evenly into the number of OSTs,
2999 the MDS is always picking the same starting OST for each file.
3000 Return the OST selection heuristic to the original design.
3005 Description: mount failures may take full timeout to return an error
3006 Details : Under some heavy load conditions it is possible that a
3007 failed mount can wait for the full obd_timeout interval,
3008 possibly several minutes, before reporting an error.
3009 Instead return an error as soon as the status is known.
3011 Frequency : quota enabled and large files being deleted
3013 Description: releasing more than 4GB of quota at once hangs OST
3014 Details : If a user deletes more than 4GB of files on a single OST it
3015 will cause the OST to spin in an infinite loop. Release
3016 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3021 Description: statfs data retrieved from /proc may be stale or zero
3022 Details : When reading per-device statfs data from /proc, in the
3023 {kbytes,files}_{total,free,avail} files, it may appear
3024 as zero or be out of date.
3027 Frequency : systems with MD RAID1 external journal devices
3029 Description: lconf's call to blkid is confused by RAID1 journal devices
3030 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3031 of returning all block devices that match the journal UUID.
3034 Frequency : always, for aggregate stripe size over 4GB
3036 Description: assertion fails when trying to use 4GB stripe size
3037 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
3038 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
3041 Frequency : always on ppc64
3043 Description: the first write on an ext3 filesystem with mballoc got stuck
3044 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3045 perform endianness conversion.
3047 ------------------------------------------------------------------------------
3049 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3051 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3052 INTEROPERATE with older versions automatically. Please read the
3053 user documentation before upgrading any part of a live system.
3054 * WARNING: Lustre networking configuration changes are required with
3055 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3058 * Support for kernels:
3059 2.6.9-22.0.2.EL (RHEL 4)
3060 2.6.5-7.244 (SLES 9)
3061 2.6.12.6 vanilla (kernel.org)
3064 Severity : enhancement
3065 Bugzilla : 7981/8208
3066 Description: Introduced Lustre Networking (LNET)
3067 Details : LNET is new networking infrastructure for Lustre, it includes
3068 a reorganized network configuration mode (see the user
3069 documentation for full details) as well as support for routing
3070 between different network fabrics. Lustre Networking Devices
3071 (LNDs) for the supported network fabrics have also been
3072 created for this new infrastructure.
3074 Severity : enhancement
3075 Description: Introduced Access control lists
3076 Details : clients can set ACLs on files and directories in order to have
3077 more fine-grained permissions than the standard Unix UGO+RWX.
3078 The MDS must be started with the "-o acl" mount option.
3080 Severity : enhancement
3081 Description: Introduced filesystem quotas
3082 Details : Administrators may now establish per-user quotas on the
3085 Severity : enhancement
3087 Description: Configuration change for the XT3
3088 The PTLLND is now used to run Lustre over Portals on the XT3
3089 The configure option(s) --with-cray-portals are no longer used.
3090 Rather --with-portals=<path-to-portals-includes> is used to
3091 enable building on the XT3. In addition to enable XT3 specific
3092 features the option --enable-cray-xt3 must be used.
3097 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3098 Details : A race between allocating a new llcd and re-getting the llcd_lock
3099 allowed another thread to grab newly-allocated llcd.
3101 Severity : enhancement
3103 Description: 2.6 OST async journal commit and locking fix to improve performance
3104 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3105 2.6 kernels are now async as they already were in 2.4 kernels so
3106 that they can commit concurrently with the network bulk transfer.
3107 For block-allocated files the filter allocation semaphore is held
3108 to avoid filesystem fragmentation during allocation. BKL lock
3109 removed for 2.6 xattr operations where it is no longer needed.
3114 Description: lconf incorrectly determined whether two IP networks could talk
3115 Details : In some more complicated routing and multiple-network
3116 configurations, lconf will avoid trying to make a network
3117 connection to a disjoint part of the IP space. It was doing the
3118 math incorrectly for one set of cases.
3123 Description: Fix for potential infinite loop processing records in an llog.
3124 Details : If an llog record is corrupted/zeroed, it is possible to loop
3125 forever in llog_process(). Validate the llog record length
3126 and skip the remainder of the block on error.
3129 Frequency : occasional (liblustre only)
3131 Description: liblustre could not open files whose last component is a symlink
3132 Details : sysio_path_walk() would incorrectly pass the open intent to
3133 intermediate path components.
3136 Frequency : rare (liblustre only with non-standard tuning)
3137 Bugzilla : 7201 (7350)
3138 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3139 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3140 after it has been released, only do this for referenced locks
3142 Severity : enhancement
3144 Description: specifying an (invalid) directory default stripe_size of -1
3145 would reset the directory default striping
3146 Details : stripe_size -1 was used internally to signal directory stripe
3147 removal, now use "all default" to signal dir stripe removal
3148 as a directory striping of "all default" is not useful
3151 Frequency : common for large clusters running liblustre clients
3153 Description: doing an ls when liblustre clients are running is slow
3154 Details : sending a glimpse AST to a liblustre client waits for every AST
3155 to time out, as liblustre clients will not respond. Since they
3156 cannot cache data we refresh the OST lock LVB from disk instead.
3158 Severity : enhancement
3160 Description: doing an ls at the same time as file IO can be slow
3161 Details : enqueue and other "small" requests can be blocked behind many
3162 large IO requests. Create a new OST IO portal for non-IO
3163 requests so they can be processed faster.
3166 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3168 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3169 created file with mode 000
3170 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3171 allow an ftruncate() as a non-root user to a file with mode 000.
3172 Lustre now respects this flag to disable mode checks when
3173 truncating a file owned by the user
3176 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3179 Description: Revoking locks from clients that went dead or catatonic might take
3181 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3182 cancellation of such locks instant on servers without waiting for
3183 any reply from clients. Clients drops these locks when cancel
3184 notification from server is received without replying.
3187 Frequency : liblustre-only, when liblustre client dies or becomes busy
3189 Description: Doing ls on Linux clients can take a long time with active
3191 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3192 granting such locks to it in the first place if possible. Locks
3193 are taken by proxy on the OST during the read or write and
3194 dropped immediately afterward. Add connect flags handling, do
3195 not grant locks to liblustre clients for glimpse ASTs.
3197 Severity : enhancement
3199 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3200 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3201 sequential reads and behaves badly with applications doing random
3202 reads. Improve it by reading ahead at least read region, and
3203 avoiding excessive large RPC for small reads.
3205 Severity : enhancement
3207 Description: Creating more than 1000 files for a single job may cause a load
3208 imbalance on the OSTs if there are also a large number of OSTs.
3209 Details : qos_prep_create() uses an OST index reseed value that is an
3210 even multiple of the number of available OSTs so that if the
3211 reseed happens in the middle of the object allocation it will
3212 still utilize the OSTs as uniformly as possible.
3217 Description: OST or MDS may oops in ping_evictor_main()
3218 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3219 but doesn't restart at beginning of obd_exports_timed list
3222 Severity : enhancement
3224 Description: improve by-nid export eviction on the MDS and OST
3225 Details : allow multiple exports with the same NID to be evicted at one
3226 time without re-searching the exports list.
3229 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3231 Description: MDS may oops in groups_free()
3232 Details : in rare race conditions a newly allocated group_info struct is
3233 freed again, and this can be NULL. The 2.4 compatibility code
3234 for groups_free() checked for a NULL pointer, but 2.6 did not.
3237 Frequency : common for liblustre clients doing little filesystem IO
3238 Bugzilla : 9352, 7313
3239 Description: server may evict liblustre clients accessing contended locks
3240 Details : if a client is granted a lock or receives a completion AST
3241 with a blocking AST already set it would not reply to the AST
3242 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3245 Frequency : lfs setstripe, only systems with more than 160 OSTs
3247 Description: unable to set striping with a starting offset beyond OST 160
3248 Details : llapi_create_file() incorrectly limited the starting stripe
3249 index to the maximum single-file stripe count.
3252 Frequency : LDAP users only
3254 Description: lconf did not handle in-kernel recovery with LDAP properly
3255 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3257 Severity : enhancement
3259 Description: bind OST threads to NUMA nodes to improve performance
3260 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3261 node and do their allocations there to localize memory access
3263 Severity : enhancement
3265 Description: llmount can determine client NID directly from Myrinet (GM)
3266 Details : the client NID code from gmnalnid was moved directly into
3267 llmount, removing the need to use this or specifying the
3268 client NID explicitly when mounting GM clients with zeroconf
3271 Frequency : if client is started with down MDS
3273 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3274 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3275 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3276 ptlrpc_import_delayed_req() because we didn't check if the
3277 request was interrupted, and we also didn't break out of the
3278 event loop if there was no timeout
3283 Description: data loss during non-page-aligned writes to a single file from
3284 both multiple nodes and multiple threads on one node at same time
3285 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3286 inconsistency led to false short-reads, that were cached and later
3287 used by ->prepare_write() to fill in partially written page,
3288 leading to data loss.
3291 Frequency : always, if lconf --abort_recovery used
3293 Description: lconf --abort_recovery fails with 'Operation not supported'
3294 Details : lconf was attempting to abort recovery on the MDT device and not
3297 Severity : enhancement
3299 Description: remove cleanup logs
3300 Details : replace lconf-generated cleanup logs with lustre internal
3301 cleanup routines. Eliminates the need for client-cleanup and
3304 Severity : enhancement
3306 Description: add support for EAs (user and system) on lustre filesystems
3307 Details : it is now possible to store extended attributes in the Lustre
3308 client filesystem, and with the user_xattr mount option it
3309 is possible to allow users to store EAs on their files also
3311 Severity : enhancement
3313 Description: Add possibility (config option) to show minimal available OST free
3315 Details : When compiled with --enable-mindf configure option, statfs(2)
3316 (and so, df) will return least minimal free space available from
3317 all OSTs as amount of free space on FS, instead of summary of
3318 free spaces of all OSTs.
3320 Severity : enhancement
3322 Description: do not expand extent locks acquired on OST-side
3323 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3324 by server: they are not cached anyway.
3327 Frequency : when mmap is used/binaries executed from Lustre
3329 Description: Unmmap pages before throwing them away from read cache.
3330 Details : llap_shrink cache now attempts to unmap pages before discarding
3331 them (if unmapping failed - do not discard). SLES9 kernel has
3332 extra checks that trigger if this unmapping is not done first.
3337 Description: lconf didn't resolve symlinks before checking to see whether a
3338 given mountpoint was already in use
3341 Frequency : when migrating failover services
3342 Bugzilla : 6395, 9514
3343 Description: When migrating a subset of services from a node (e.g. failback
3344 from a failover service node) the remaining services would
3345 time out and evict clients.
3346 Details : lconf --force (implied by --failover) sets the global obd_timeout
3347 to 5 seconds in order to quickly disconnect, but this caused
3348 other RPCs to time out too quickly. Do not change the global
3349 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3351 Severity : enhancement
3352 Frequency : if MDS is started with down OST
3353 Bugzilla : 9439,5706
3354 Description: Allow startup/shutdown of an MDS without depending on the
3355 availability of the OSTs.
3356 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3357 Add appropriate locking and lov-osc refcounts for safe
3358 cleaning. Add osc abort_inflight calls in case the
3359 synchronize never started.
3362 Frequency : occasional (Cray XT3 only)
3364 Description: root not authorized to access files in CRAY_PORTALS environment
3365 Details : The client process capabilities were not honoured on the MDS in
3366 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3367 been accessed by an authorized user then root was able to access
3368 the file on the local client also. The root user capabilities
3369 are now allowed on the MDS, as this environment has secure UID.
3372 Frequency : occasional
3374 Description: ldiskfs "too long searching" message happens too often
3375 Details : A debugging message (otherwise harmless) prints too often on
3376 the OST console. This has been reduced to only happen when
3377 there are fragmentation problems on the filesystem.
3382 Description: Division by zero in statfs when all OSCs are inactive
3383 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3384 lov_statfs divides by value returned by lov_get_stripecnt().
3388 Bugzilla : 9489, 3273
3389 Description: First write from each client to each OST was only 4kB in size,
3390 to initialize client writeback cache, which caused sub-optimal
3391 RPCs and poor layout on disk for the first writen file.
3392 Details : Clients now request an initial cache grant at (re)connect time
3393 and so that they can start streaming writes to the cache right
3394 away and always do full-sized RPCs if there is enough data.
3395 If the OST is rebooted the client also re-establishes its grant
3396 so that client cached writes will be honoured under the grant.
3401 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3402 Details : Now I/O RPCs go to different portal number and (presumably) fast
3403 lock requests (and glimses) and other RPCs get their own service
3404 threads pool that should be able to service those RPCs
3407 Severity : enhancement
3409 Description: Ability to exchange lustre version between client and servers and
3410 issue warnings at client side if client is too old. Also for
3411 liblustre clients there is ability to refuse connection of too old
3413 Details : New 'version' field is added to connect data structure that is
3414 filled with version info. That info is later checked by server and
3418 Frequency : rare, liblustre only.
3419 Bugzilla : 9296, 9581
3420 Description: Two simultaneous writes from liblustre at offset within same page
3421 might proceed at the same time overwriting eachother with stale
3423 Details : I/O lock withing llu_file_prwv was released too early, before data
3424 actually was hitting the wire. Extended lock-holding time until
3425 server acknowledges receiving data.
3428 Frequency : extremely rare. Never observed in practice.
3430 Description: avoid generating lustre_handle cookie of 0.
3431 Details : class_handle_hash() generates handle cookies by incrementing
3432 global counter, and can hit 0 occasionaly (this is unlikely, but
3433 not impossible, because initial value of cookie counter is
3434 selected randonly). Value of 0 is used as a sentinel meaning
3435 "unassigned handle" --- avoid it. Also coalesce two critical
3436 sections in this function into one.
3438 Severity : enhancement
3440 Description: allow liblustre clients to delegate truncate locking to OST
3441 Details : To avoid overhead of locking, liblustre client instructs OST to
3442 take extent lock in ost_punch() on client's behalf. New connection
3443 flag is added to handle backward compatibility.
3445 Severity : enhancement
3446 Bugzilla : 4928, 7341, 9758
3447 Description: allow number of OST service threads to be specified
3448 Details : a module parameter allows the number of OST service threads
3449 to be specified via "options ost ost_num_threads={N}" in the
3450 OSS's /etc/modules.conf or /etc/modprobe.conf.
3454 Bugzilla : 6146, 9635, 9895
3455 Description: servers crash with bad pointer in target_handle_connect()
3456 Details : In rare cases when a client is reconnecting it was possible that
3457 the connection request was the last reference for that export.
3458 We would temporarily drop the export reference and get a new
3459 one, but this may have been the last reference and the export
3460 was just destroyed. Get new reference before dropping old one.
3462 Severity : enhancement
3463 Frequency : if client is started with failover MDS
3465 Description: Allow multiple MDS hostnames in the mount command
3466 Details : Try to read the configuration from all specified MDS
3467 hostnames during a client mount in case the "primary"
3470 Severity : enhancement
3472 Description: Stop sending data to evicted clients as soon as possible.
3473 Details : Check if the client we are about to send or are sending data to
3474 was evicted already. (Check is done every second of waiting,
3475 for which l_wait_event interface was extended to allow checking
3476 of exit condition at specified intervals).
3479 Frequency : rare, normally only when NFS exporting is done from client
3481 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3483 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3486 Frequency : always, if srand() is called before liblustre initialization
3488 Description: Liblustre uses system PRNG disturbing its usage by user application
3489 Details : Introduce internal to lustre fast and high-quality PRNG for
3490 lustre usage and make liblustre and some other places in generic
3491 lustre code to use it.
3493 Severity : enhancement
3494 Bugzilla : 9477, 9557, 9870
3495 Description: Verify that the MDS configuration logs are updated when xml is
3496 Details : Check if the .xml configuration logs are newer than the config
3497 logs stored on the MDS and report an error if this is the case.
3498 Request --write-conf, or allow starting with --old_conf.
3500 Severity : enhancement
3502 Description: Handle symlinks in the path when checking if Lustre is mounted.
3503 Details : Resolve intermediate symlinks when checking if a client has
3504 mounted a filesystem to avoid duplicate client mounts.
3509 Description: lconf can hit an error exception but still return success.
3510 Details : The lconf command catches the Command error exception at the top
3511 level script context and will exit with the associated exit
3512 status, but doesn't ensure that this exit status is non-zero.
3517 Description: failure of ptlrpc thread startup can cause oops
3518 Details : Starting a ptlrpc service thread can fail if there are a large
3519 number of threads or the server memory is very fragmented.
3520 Handle this without oopsing.
3523 Frequency : always, only if liblustre and non-default acceptor port was used
3525 Description: liblustre cannot connect to servers with non-default acceptor port
3526 Details : tcpnal_set_default_params() was not called and was therefore
3527 ignoring the environment varaible TCPNAL_PORT, as well as other
3528 TCPNAL_ environment variables
3533 Description: two objects could be created on the same OST for a single file
3534 Details : If an OST is down, in some cases it was possible to create two
3535 objects on a single OST for a single file. No problems other
3536 than potential performance impact and spurious error messages.
3540 Bugzilla : 5681, 9562
3541 Description: Client may oops in ll_unhash_aliases
3542 Details : Client dcache may become inconsistent in race condition.
3543 In some cases "getcwd" can fail if the current directory is
3549 Description: Inode refcounting problems in NFS export code
3550 Details : link_raw functions used to call d_instantiate without obtaining
3551 extra inode reference first.
3555 Bugzilla : 9942, 9903
3556 Description: Referencing freed requests leading to crash, memleaks with NFS.
3557 Details : We used to require that call to ll_revalidate_it was always
3558 followed by ll_lookup_it. Also with revalidate_special() it is
3559 possible to call ll_revalidate_it() twice for the same dentry
3560 even if first occurence returned success. This fix changes semantic
3561 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3562 reference on a request referred from the intent.
3563 ll_intent_release() then releases such a request.
3566 Frequency : rare, normally benchmark loads only
3568 Description: unlinked inodes were kept in memory on the client
3569 Details : If a client is repeatedly creating and unlinking files it
3570 can accumulate a lot of stale inodes in the inode slab cache.
3571 If there is no other client load running this can cause the
3572 client node to run out of memory. Instead flush old inodes
3573 from client cache that have the same inode number as a new inode.
3576 Frequency : SLES9 2.6.5 kernel and long filenames only
3577 Bugzilla : 9969, 10379
3578 Description: utime reports stale NFS file handle
3579 Details : SLES9 uses out-of-dentry names in some cases, which confused
3580 the lustre dentry revalidation. Change it to always use the
3584 Frequency : rare, unless heavy write-truncate concurrency is continuous
3585 Bugzilla : 4180, 6984, 7171, 9963, 9331
3586 Description: OST becomes very slow and/or deadlocked during object unlink
3587 Details : filter_destroy() was holding onto the parent directory lock
3588 while truncating+unlinking objects. For very large objects this
3589 may block other threads for a long time and slow overall OST
3590 responsiveness. It may also be possible to get a lock ordering
3591 deadlock in this case, or run out of journal credits because of
3592 the combined truncate+unlink. Solution is to do object truncate
3593 first in one transaction without parent lock, and then do the
3594 final unlink in a new transaction with the parent lock. This
3595 reduces the lock hold time dramatically.
3598 Frequency : rare, 2.4 kernels only
3600 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3601 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3602 before unmounting their filesystem in order to avoid deadlock.
3603 The kernel_locked() function in 2.4 kernels only checks whether
3604 the kernel lock is held, not whether it is this process that is
3605 holding it as 2.6 kernels do.
3610 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3611 Details : The client ptlrpc code may be trying to reconnect to a down
3612 server before a previous connection attempt has timed out.
3613 Increase the reconnect interval to be longer than the connection
3614 timeout interval to avoid sending duplicate connections to
3618 Frequency : echo_client brw_test command
3620 Description: fix echo_client to work with OST preallocated code
3621 Details : OST preallocation code (5137) didn't take echo_client IO path
3622 into account: echo_client calls filter methods outside of any
3623 OST thread and, hence, there is no per-thread preallocated
3624 pages and buffers to use. Solution: hijack pga pages for IO. As
3625 a byproduct, this avoids unnecessary data copying.
3629 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3630 Description: Client can oops in mdc_commit_close() after open replay
3631 Details : It was possible for the MDS to return an open request with no
3632 transaction number in mds_finish_transno() if the client was
3633 evicted, but without actually returning an error. Clients
3634 would later try to replay that open and may trip an assertion
3635 Simplify the client close codepath, and always return an error
3636 from the MDS in case the open is not successful.
3639 Frequency : rare, 2.6 OSTs only
3641 Description: OST may deadlock under high load on fragmented files
3642 Details : If there was a heavy load and highly-fragmented OST filesystems
3643 it was possible to have all the OST threads deadlock waiting on
3644 allocation of biovecs, because the biovecs were not released
3645 until the entire RPC IO was completed. Instead, release biovecs
3646 as soon as they are complete to ensure forward IO progress.
3648 Severity : enhancement
3650 Description: Support for specifying external journal device at mount
3651 Details : If an OST or MDS device is formatted with an external journal
3652 device, this device major/minor is stored in the ext3 superblock
3653 and may not be valid for failover. Allow detecting and
3654 specifying the external journal at mount time.
3659 Description: Mounting an MDS with pending unlinked files may cause oops
3660 Details : target_finish_recovery() calls mds_postrecov() which returned
3661 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3662 considers this an error and immediately begins cleaning up the
3663 lov, just after starting the mds_lov process
3665 Severity : enhancement
3667 Description: Implement 'lfs df' to report actual free space on per-OST basis
3668 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3669 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3670 usage of objects; '-h' to report in human readable format.
3672 ------------------------------------------------------------------------------
3674 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3681 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3682 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3683 at startup during orphan unlinking. Instead, skip buddy update
3684 before setup, as it will be regenerated after recovery is complete.
3689 Description: If an OST is inactive, its locks might reference stale inodes.
3690 Details : lov_change_cbdata() must iterate over all namespaces, even if
3691 they are inactive to clear inode references from the lock.
3693 Severity : enhancement
3694 Frequency : occasional, if non-standard max_dirty_mb used
3696 Description: Client will block write RPCs if not enough grant
3697 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3698 then the client will block writes while waiting for another RPC
3699 to complete instead of consuming its dirty limit. With change
3700 we get improved performance when max_dirty_mb is small.
3702 Severity : enhancement
3703 Bugzilla : 3389, 6253
3704 Description: Add support for supplementary groups on the MDS.
3705 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3706 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3707 lookups for user supplementary groups into a cache.
3711 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3712 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3713 for liblustre O_EXCL is enforced by MDS.
3717 Description: readdir never returns NULL in liblustre.
3718 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3719 dentry in struct dirent.
3723 Frequency : liblustre only, depends on application IO pattern
3724 Description: liblustre clients evicted if not contacting servers
3725 Details : Don't put liblustre clients into the ping_evictor list, so
3726 they will not be evicted by the pinger ever.
3728 Severity : enhancement
3730 Description: Add ability to evict clients by NID from MDS.
3731 Details : By echoing "nid:$NID" string into
3732 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3733 $NID would be instantly evicted from this MDS and from all active
3734 OSTs connected to it.
3738 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3739 Details : lookup_it_finish() used to query file size from OSTs that was not
3744 Description: service threads change working directory to that of init
3745 Details : Starting lustre service threads may pin the working directory
3746 of the parent thread, making that filesystem busy. Threads
3747 now change to the working directory of init to avoid this.
3751 Frequency : during shutdown only
3752 Description: shutdown with a failed MDS or OST can cause unmount to hang
3753 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3758 Frequency : 2.6 only
3759 Description: chmod/chown may include an extra supplementary group
3760 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3761 supplementary group and if none is specified this is used.
3764 Bugzilla : 5479 (6816)
3766 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3767 Details : If lookup is in progress on a file that is unlinked we might try
3768 to revalidate the inode and fail in revalidate after lookup is
3769 complete and ll_file_open() enqueues the open again but
3770 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3773 Frequency : always, if lconf --abort_recovery used
3775 Description: lconf --abort_recovery fails with 'Operation not supported'
3776 Details : lconf was attempting to abort recovery on the MDT device and not
3779 ------------------------------------------------------------------------------
3781 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3786 Frequency : rare (only unsupported configurations with a node running as an
3788 Bugzilla : 6514, 5137
3789 Description: Mounting a Lustre file system on a node running as an OST could
3791 Details : OSTs now preallocates memory needed to write out data at
3792 startup, instead of when needed, to avoid having to
3793 allocate memory in possibly low memory situations.
3794 Specifically, if the file system is mounted on on OST,
3795 memory pressure could force it to try to write out data,
3796 which it needed to allocate memory to do. Due to the low
3797 memory, it would be unable to do so and the node would
3798 become unresponsive.
3800 Severity : enhancement
3802 Description: Addition of lconf --service command line option
3803 Details : lconf now accepts a '--service <arg>' option, which is
3804 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3806 Severity : enhancement
3808 Description: Failover mode is now the default for OSTs.
3809 Details : By default, OSTs will now run in failover mode. To return to
3810 the old behaviour, add '--failout' to the lmc line for OSTs.
3812 Severity : enhancement
3814 Description: Health checks are now provided for MDS and OSTs
3815 Details : Additional detailed health check information on MSD and OSTs
3816 is now provided through the procfs health_check value.
3819 Frequency : occasional, depends on IO load
3821 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3822 after numerous create/delete cycles
3823 Details : The ext3 inode allocation policy would not allocate new inodes
3824 very well on the OSTs because there are no new directories
3825 being created. Instead we look for groups with free space if
3826 the parent directories are nearly full.
3831 Description: Network or server problems during mount may cause partially
3832 mounted clients instead of returning an error.
3833 Details : The config llog parsing code may overwrite the error return
3834 code during mount error handling, returning success instead
3840 Description: MDS can fail to allocate large reply buffers
3841 Details : After long uptimes the MDS can fail to allocate large reply
3842 buffers (e.g. zconf client mount config records) due to memory
3843 fragmentation or consumption by the buffer cache. Preallocate
3844 some large reply buffers so that these replies can be sent even
3845 under memory pressure.
3849 Frequency : rare (liblustre)
3850 Description: fsx running with liblustre complained that using truncate() to
3851 extend the file doesn't work. This patch corrects that issue.
3852 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3853 ATTR_SIZE and lsm use in llu_setattr_raw.
3857 Frequency : rare, only 2.6 kernels
3858 Description: Unusual file access patterns on the MDS may result in inode
3859 data being lost in very rare circumstances.
3860 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3861 code in the 2.6 kernel caused the inode and/or EA data not to
3862 be read from disk, causing single-file corruption.
3866 Frequency : rare, only 2.6 filesystems using extents
3867 Description: Heavy concurrent write and delete load may cause data corruption.
3868 Details : It was possible under high-load situations to have an extent
3869 metadata block in the block device cache from a just-unlinked
3870 file overwrite a newly-allocated data block. We now unmap any
3871 metadata buffers that alias just-allocated data blocks.
3875 Frequency : filesystems with default stripe_count larger than 77
3876 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3877 Details : lconf specifies an inode size of 4096 bytes when the default
3878 stripe_count is larger than 77. This conflicts with the default
3879 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3880 this case to avoid pinning too much memory for large EAs.
3882 ------------------------------------------------------------------------------
3884 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3889 Frequency : rare (extremely heavy IO load with hundreds of clients)
3891 Description: Client is evicted, gets IO error writing to file
3892 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3893 caused clients to be evicted for AST timeouts. The fixes in
3894 bug 5192 mean we no longer need to have such short AST timeouts
3895 so ldlm_timeout has been increased.
3898 Frequency : occasional during --force or --failover shutdown under load
3899 Bugzilla : 5949, 4834
3900 Description: Server oops/LBUG if stopped with --force or --failover under load
3901 Details : a collection of import/export refcount and cleanup ordering
3902 issues fixed for safer force cleanup
3905 Frequency : only filesystems larger than 120 OSTs
3906 Bugzilla : 5990, 6223
3907 Description: lfs getstripe would oops on a very large filesystem
3908 Details : lov_getconfig used kfree on vmalloc'd memory
3911 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3912 Bugzilla : 6242, 6243
3913 Description: reading from files that had been truncated to a non-zero size
3914 but never opened returned no data
3915 Details : ll_file_read() reads zeros from no-object files to EOF
3920 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3921 Details : There's little harm in aborting MDS/OSS recovery and letting it
3922 try again, so I removed the LASSERT and return an error instead.
3924 Severity : enhancement
3926 Description: New debugging infrastructure for tracking down data corruption
3927 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3928 (b) cover more of the client-side code path, and (c) try to narrow
3929 down where problems occurred
3933 Bugzilla : 3819, 4364, 4397, 6313
3934 Description: Racing close and eviction MDS could cause assertion in mds_close
3935 Details : It was possible to get multiple mfd references during close and
3936 client eviction, leading to one thread referencing a freed mfd.
3938 Severity: : enhancement
3939 Bugzilla : 3262, 6359
3940 Description: Attempts to reconnect to servers are now more aggressive.
3941 Details : This builds on the enhanced upcall-less recovery that was added
3942 in 1.4.2. When trying to reconnect to servers, clients will
3943 now try each server in the failover group every 10 seconds. By
3944 default, clients would previously try one server every 25 seconds.
3949 Description: After recovery, certain operations trigger a failed
3950 assertion on a client.
3951 Details : Failing over an mds, using lconf -d --failover, while a
3952 client was doing a readdir() call would cause the client to
3953 LBUG after recovery completed and the readdir() was resent.
3955 Severity : enhancement
3957 Description: Default groups are now added by lconf
3958 Details : You can now run lconf --group <servicename> without having to
3959 manually add groups with lmc.
3962 Frequency : occasional
3964 Description: Nodes with an elan id of 0 trigger a failed assertion
3967 Frequency : always when accessing e.g. tty/console device nodes
3969 Description: tty and some other devices nodes cannot be used on lustre
3970 Details : file's private_data field is used by device data and lustre
3971 values in there got lost. New field was added to struct file to
3972 store fs-specific private data.
3975 Frequency : when exporting Lustre via NFS
3977 Description: NFSD failed occasionally when looking up a path component
3978 Details : NFSD is looking up ".." which was broken in ext3 directories
3979 that had grown large enough to become hashed.
3982 Frequency : Clusters with multiple interfaces not on the same subnet
3984 Description: Nodes will repeatedly try to reconnect to an interface which it
3985 cannot reach and report an error to the log.
3986 Details : Extra peer list entries will be created by lconf with some peers
3987 unreachable. lconf now validates the peer before adding it.
3990 Frequency : Only if a default stripe is set on the filesystem root.
3992 Description: Setting a default stripe on the filesystem root prevented the
3993 filesystem from being remounted.
3994 Details : The client was sending extra request flags in the root getattr
3995 request and did not allocate a reply buffer for the dir EA.
3998 Frequency : occasional, higher if lots of files are accessed by one client
3999 Bugzilla : 6159, 6097
4000 Description: Client trips assertion regarding lsm mismatch/magic
4001 Details : While revalidating inodes the VFS looks up inodes with ifind()
4002 and in rare cases can find an inode that is being freed.
4003 The ll_test_inode() code will free the lsm during ifind()
4004 when it finds an existing inode and then the VFS later attaches
4005 this free lsm to a new inode.
4009 Bugzilla : 6422, 7030
4010 Description: MDS deadlock between mkdir and client eviction
4011 Details : Creating a new file via mkdir or mknod (starting a transaction
4012 and getting the ns lock) can deadlock with client eviction
4013 (gets ns lock and trying to finish a synchronous transaction).
4016 Frequency : occasional
4017 Description: While starting a server, the fsfilt_ext3 module could not be
4019 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
4020 kernels. Previously, lconf would still use the ext3 name
4021 when trying to load modules. Now, it will correctly use
4022 ext3 on 2.4 and ldiskfs on 2.6.
4024 Severity : enhancement
4025 Description: The default stripe count has been changed to 1
4026 Details : The interpretation of the default stripe count (0, to lfs
4027 or lmc) has been changed to mean striping across a single
4028 OST, rather than all available. For general usage we have
4029 found a stripe count of 1 or 2 works best.
4031 Severity : enhancement
4032 Description: Add support for compiling against Cray portals.
4033 Details : Conditional compiling for some areas that are different
4037 Frequency : occasional
4038 Bugzilla : 6409, 6834
4039 Description: Creating files with an explicit stripe count may lead to
4040 a failed assertion on the MDS
4041 Details : If some OSTs are full or unavailable, creating files may
4042 trigger a failed assertion on the MDS. Now, Lustre will
4043 try to use other servers or return an error to the
4047 Frequency : occasional
4049 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4050 caused lock timeout during readahead (since 1.4.2).
4051 Details : Processes doing readahead might match a lock that hasn't been
4052 granted yet if there are overlapping and conflicting lock
4053 requests. The readahead process waits on ungranted lock
4054 (original lock is CBPENDING), while OST waits for that process
4055 to cancel CBPENDING read lock and eventually evicts client.
4057 Severity : enhancement
4059 Description: Initial enabling of flock support for clients
4060 Details : Implements fcntl advisory locking and file status functions.
4061 This feature is provided as an optional mount flag (default
4062 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4063 locking are implemented yet, and those that are are not guaranteed
4064 to be completely correct in production environments.
4065 mount -t lustre -o [flock|noflock] ...
4068 Frequency : occasional
4070 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4071 assertion in the ext3 JBD (journaling) layer.
4072 Details : The b_committed_data struct is protected by the big kernel lock
4073 in 2.4 kernels, serializing journal_commit_transaction() and
4074 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4075 access to this struct. In 2.6 kernels there is finer grained
4076 locking to improve SMP performance of the JBD layer.
4080 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4085 Description: merge and cleanup kernel patches.
4088 -----------------------------------------------------------------------------
4090 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4092 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4093 versions of Lustre. Please update all servers and clients to
4094 version 1.4.2 or later at the same time. You must also run
4095 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4096 to update the configuration logs.
4098 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4099 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4100 - swab llog records of type '0' so we get proper header size/idx (5861)
4101 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4102 - fix rename of one directory over another leaking an inode (5953)
4103 - avoid SetPageDirty on 2.6 (5981)
4104 - don't re-add just-being-destroyed locks to the waiting list (5653)
4105 - when creating new directories, inherit the parent's custom
4106 striping settings if present parent (3048)
4107 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4108 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4109 - don't hold client locks on temporary worklist from l_lru (5666)
4110 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4111 - automatically evict dead clients (5921)
4112 - Update file size properly in create+truncate+fstat case (6196)
4113 - Do not unhash mountpoint dentries, do not allow removal of
4115 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4116 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4117 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4118 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4119 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4120 - Correctly handle reads of files with no objects (6243)
4121 - lctl recover will also mark a device active if deactivate used (5933)
4123 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
4124 - allow --write-conf on an MDS with different nettype than client (5619)
4125 - don't write config llogs to MDS for mounts not from that MDS (5617)
4126 - lconf should create multiple TCP connections from a client (5201)
4127 - init scripts are now turned off by default; run chkconfig --on
4128 lustre and chkconfig --on lustrefs to use them
4129 - upcalls are no longer needed for clients to recover to failover
4131 - add --abort-recovery option to lconf to abort recovery on device
4133 - add support for an arbitrary number of OSTs (3026)
4134 - Quota support protocol changes.
4135 - forward compatibility changes to wire structs (6007)
4136 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
4137 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
4138 - improved "lustre status" script
4139 - initialize blocksize for non-regular files (6062)
4140 - added --disable-server and --disable-client configure options (5782)
4141 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
4142 - Vanilla 2.4.29 support
4143 - increase maximum number of obd devices to 520 (6242)
4144 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4145 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4146 SLES 9 2.6.5 kernels
4147 - limit stripes per file to 160 (the maximum EA size) (6093)
4149 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4152 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4153 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4154 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4155 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4156 - small ext3 extents cleanups and fixes (5733)
4157 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4158 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4159 - avoid needless client->OST connect, fix handle mismatch (5317)
4160 - fix DLM error path that led to out-of-sync client, long delays (5779)
4161 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4162 - fix several locking issues related to i_size (5492,5624,5654,5672)
4163 - don't move pending lock onto export if it is already evicted (5683)
4164 - fix kernel oops when creating .foo in unlinked directory (5548)
4165 - fix deadlock in obdfilter statistics vs. object create (5811)
4166 - use time_{before,after} to avoid timer jiffies wrap (5882)
4167 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4168 - Do not leak request if server was not able to process it (5154)
4169 - If mds_open unable to find parent dir, make that negative lookup(5154)
4170 - don't create new directories with extent-mapping (5909, 5936)
4172 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
4173 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
4174 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
4175 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
4176 - support for RHEL4 kernel on i686 (5773)
4177 - provide error messages when incompatible logs are encountered (5898)
4179 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
4180 * version 1.4.0.10 (1.4.1 release candidate 1)
4182 - don't keep a lock reference when lock is not granted (4238)
4183 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4184 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
4185 - drop import inflight refcount on signal_completed_replay error (5255)
4186 - unlock page after async write error during send (3677)
4187 - handle missing objects in filter_preprw_read properly (5265)
4188 - no transno return for symlink open, don't save no-trasno open (3440)
4189 - don't try to complete elan receive that already failed (4012)
4190 - free RPC server reply state on error (5406)
4191 - clean up thread from ptlrpc_start_thread() on error (5160)
4192 - readahead could read extra page into cache that wasn't ejected (5388)
4193 - prevent races in class_attach/setup/cleanup/detach (5260)
4194 - don't dereference de->d_inode after l_dput of de (5458)
4195 - use "int" for stripe value returned from lock_to_stripe (5544)
4196 - mballoc allocation and error-checking fixes in 2.6 (5504)
4197 - block device patches to fix I/O request sizes in 2.6 (5482)
4198 - look up hostnames for IB nals (5602)
4199 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
4200 - don't start multiple acceptors for the same port (5277)
4201 - fix incorrect LASSERT in mds_getattr_name (5635)
4202 - export a proc file for general "ping" checking (5628)
4203 - fix "lfs check" to not block when the MDS is down (5628)
4205 - service request history (4965)
4206 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4207 - create an "evict_client" /proc entry on OSTs, like the MDS has
4208 - fix mount usage message, return errors per mount(8) (5168)
4209 - change grep [] to grep "[]" in tests so they work in more UMLs
4210 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4211 - remove ancient LOV_MAGIC_V0 EA support (5047)
4212 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4213 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4214 - allow readahead allocations to fail when low on memory (5383)
4215 - mmap locking landed again, after considerable improvement (2828)
4216 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4218 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4221 - send OST transaction number in read/write reply to free req (4966)
4222 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4223 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4224 - the watchdog thread now runs as interruptible (5246)
4225 - flock/lockf fixes (but it's still disabled, pending 5135)
4226 - don't use EXT3 constants in llite code (5094)
4227 - memory shortage at startup could cause assertion (5176)
4229 - reorganization of lov code
4230 - single portals codebase
4232 - add extents/mballoc support (5025)
4233 - direct I/O reads in the obdfilter (4048)
4234 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4236 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4239 - send OST transaction number in read/write reply to free req (4966)
4240 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4241 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4242 - the watchdog thread now runs as interruptible (5246)
4243 - handle missing objects in filter_preprw_read properly (5265)
4244 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4245 - drop import inflight refcount on signal_completed_replay error (5255)
4246 - unlock page after async write error during send (3677)
4247 - return original error code on reconstructed replies (3761)
4248 - no transno return for symlink open, don't save no-trasno open (3440)
4250 - add pid to ldlm debugging output (4922)
4251 - bump the watchdog timeouts -- we can't handle 30sec yet
4252 - extra debugging for orphan dentry/inode bug (5259)
4254 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
4257 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4258 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4259 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
4260 - fix NULL dereference in /proc/sys/portals/routes (4827)
4261 - allow failed mdc_close() operations to be interrupted (4561)
4262 - stop precreate on OST before MDS would time out on it (4778)
4263 - don't send partial-page writes before EOF from client (4410)
4264 - discard client grant for sub-page writes on large-page clients (4520)
4265 - don't free dentries not owned by NFS code, check generation (4806)
4266 - fix lsm leak if mds_create_objects() fails (4801)
4267 - limit debug_daemon file size, always print CERROR messages (4789)
4268 - use transno after validating reply (3892)
4269 - process timed out requests if import state changes (3754)
4270 - update mtime on OST during writes, return in glimpse (4829)
4271 - add mkfsoptions to LDAP (4679)
4272 - use ->max_readahead method instead of zapping global ra (5039)
4273 - don't interrupt __l_wait_event() during strace
4275 - add software watchdogs to catch hung threads quickly (4941)
4276 - make lustrefs init script start after nfs is mounted
4277 - fix CWARN/ERROR duplication (4930)
4278 - return async write errors to application if possible (2248)
4279 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4280 - print NAL number in %x format (4645)
4281 - update barely-supported suse-2.4.21-171 series (4842)
4282 - support for sles 9 %post scripts
4283 - support for building 2.6 kernel-source packages
4284 - support for sles km_* packages
4286 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4289 - ignore -ENOENT errors in osc_destroy (3639)
4290 - notify osc create thread that OSC is being cleaned up (4600)
4291 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4292 - reconstruct ost_handle() like mds_handle() (4657)
4293 - create a new thread to do import eviction to avoid deadlock (3969)
4294 - let lconf resolve symlinked-to devices (4629)
4295 - don't unlink "objects" from directory with default EA (4554)
4296 - hold socknal file ref over connect in case target is down (4394)
4297 - allow more than 32000 subdirectories in a single directory (3244)
4298 - fix blocks count for O_DIRECT writes (3751)
4299 - OST returns ENOSPC from object create when no space left (4539)
4300 - don't send truncate RPC if file size isn't changing (4410)
4301 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4302 - bind to privileged port in socknal and tcpnal (3689)
4304 - rate limit CERROR/CWARN console message to avoid overload (4519)
4305 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4306 - basic mmap support (3918)
4307 - kernel patch series update from b1_4 (4711)
4309 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4312 - avoid crash during MDS cleanup with OST shut down (2775)
4313 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4314 - don't use bad inodes on the MDS (3744)
4315 - dynamic object preallocation to improve recovery speed (4236)
4316 - don't hold spinlock over lock dumping or change debug flags (4401)
4317 - don't zero obd_dev when it is force cleaned (3651)
4318 - print grants to console if they go negative (4431)
4319 - "lctl deactivate" will stop automatic recovery attempts (3406)
4320 - look for existing locks in ldlm_handle_enqueue() (3764)
4321 - don't resolve lock handle twice in recovery avoiding race (4401)
4322 - revalidate should check working dir is a directory (4134)
4324 - don't always mark "slow" obdfilter messages as errors (4418)
4326 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4329 - don't close LustreDB during write_conf until it is done (3860)
4330 - fix typo in lconf for_each_profile (3821)
4331 - allow dumping logs from multiple threads at one time (3820)
4332 - don't allow multiple threads in OSC recovery (3812)
4333 - fix debug_size parameters (3864)
4334 - fix mds_postrecov to initialize import for llog ctxt (3121)
4335 - replace config semaphore with spinlock (3306)
4336 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4337 - don't allow enqueue to complete on a destroyed export (3822)
4338 - down write_lock before checking llog header bitmap (3825)
4339 - recover from lock replay timeout (3764)
4340 - up llog sem before sending rpc (3652)
4341 - reduce ns lock hold times when setting kms (3267)
4342 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4343 - fix NULL deref and obd_dev leak on setup error (3312)
4344 - replace some LBUG about llog ops with error handling (3841)
4345 - don't match INVALID dentries from d_lookup and spin (3784)
4346 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4347 - fix invalid assertion in ptlrpc_set_wait (3880)
4349 - add libwrap support for the TCP acceptor (3996)
4350 - add /proc/sys/portals/routes for non-root route listing (3994)
4351 - allow setting MDS UUID in .xml (2580)
4352 - print the stack of a process that LBUGs (4228)
4354 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4357 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4358 - make sure to unset replay flag from failed open requests (3440)
4359 - if default stripe count is 0, use OST count for inode size (3636)
4360 - update parent mtime/ctime on client for create/unlink (2611)
4361 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4362 - free recovery state on server during a forced cleanup (3571)
4363 - unregister_reply for resent reqs (3063)
4364 - loop back devices mounting and status check on 2.6 (3563)
4365 - fix resource-creation race that can provoke i_size == 0 (3513)
4366 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4367 - more debugging for page-accounting assertion (3746)
4368 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4369 - avoid deadlock after precreation failure (3758)
4370 - fix race and lock order deadlock in orphan handling (3450, 3750)
4371 - add validity checks when grabbing inodes from l_ast_data (3599)
4373 - add /proc/.../recovery_status to obdfilter (3428)
4374 - lightweight CDEBUG infrastructure, debug daemon (3668)
4375 - change default OSC RPC parameters to be better on small clusters
4376 - turn off OST read cache for files smaller than 32MB
4377 - install man pages and include them in rpms (3100)
4378 - add new init script for (un)mounting lustre filesystems (2593)
4379 - run chkconfig in %post for init scripts (3701)
4380 - drop scimac NAL (unmaintained)
4382 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4385 - clean kiobufs before and after use (3485)
4386 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4387 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4388 - change init script to fail more clearly if not run as root (1528)
4389 - allow clients to reconnect during replay (1742)
4390 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4391 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4392 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4393 - fix cond_resched (3554)
4394 - only evict unfinished clients after recovery (3515)
4395 - allow bulk resend, prevent data loss (3570)
4396 - dynamic ptlrpc request buffer allocation (2102)
4397 - don't allow unlinking open directory if it isn't empty (2904)
4398 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4399 - remove extraneous obd dereference causing LASSERT failure (3334)
4400 - don't use get_cycles() when creating temp. files on the mds (3156)
4401 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4402 - handle EEXIST for set-stripe, set proper directory name (3336)
4404 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4405 - fix ksocknal_fmb_callback() error messages (2918)
4407 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4410 - don't copy lvb into (possibly NULL) reply on error (2983)
4411 - don't deref dentry after dput, don't free lvb on error (2922)
4412 - use the kms to determine writeback rpc length (2947)
4413 - increment oti_logcookies when osc is inactive (2948)
4414 - update client's i_blocks count via lvb messages (2543)
4415 - handle intent open/close of special files properly (1557)
4416 - mount MDS with errors=remount-ro, like obdfilter (2009)
4417 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4418 - don't use cancelling-locks' kms values (2947)
4419 - use highest lock extent for kms, not last one (2925)
4420 - don't dereference ERR_PTR() dentry in error handling path (3107)
4421 - fix thread race in portals_debug_dumplog() (3122)
4422 - create lprocfs device entries at setup instead of at attach (1519)
4423 - common AST error handler, don't evict client on completion race (3145)
4424 - zero nameidata in detach_mnt in 2.6 (3118)
4425 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4426 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4427 - initialize RPC timeout timer earlier for 2.6 (3219)
4428 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4429 - print nal/nid for unknown nid (3258)
4430 - additional checks for oscc recovery before doing precreate (3284)
4431 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4432 - don't crash in mdc_close for bad permissions on open (3285)
4433 - zero i_rdev for non-device files (3147)
4434 - clear page->private before handing to FS, better assertion (3119)
4435 - tune the read pipeline (3236)
4436 - fix incorrect decref of invalidated dentry (2350)
4437 - provide read-ahead stats and refine rpc in flight stats (3328)
4438 - don't hold journal transaction open across create RPC (3313)
4439 - update atime on MDS at close time (3265)
4440 - close LDAP connection when recovering to avoid server load (3315)
4441 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4442 - don't leak open file on MDS after open resend (3325)
4443 - serialize filter_precreate and filter_destroy_precreated (3329)
4444 - loop device shouldn't call sync_dev() for nul device (3092)
4445 - clear page cache after eviction (2766)
4446 - resynchronize MDS->OST in background (2824)
4447 - refuse to mount the same filesystem twice on same mountpoint (3394)
4448 - allow llmount to create routes for mounting behind routers (3320)
4449 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4450 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4451 - fix rare problem with rename on htree directories (3417)
4453 - allow default OST striping configuration per directory (1414)
4454 - fix compilation for qswnal for 2.6 kernels (3125)
4455 - increase maximum number of MDS request buffers for large systems
4456 - change liblustreapi to be useful for external progs like lfsck (3098)
4457 - increase local configuration timeout for slow disks (3353)
4458 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4460 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4463 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4464 - don't overwrite extent policy data in reply if lock was blocked (2901)
4465 - drop filter export grants atomically with removal from device (2663)
4466 - del obd_self_export from work_list in class_disconnect_exports (2908)
4467 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4468 - swab reply message in mdc_close, other PPC fixes (2464)
4469 - fix destroying of named logs (2325)
4470 - overwrite old logs when running lconf --write_conf (2264)
4471 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4472 - fix race in target_handle_connect (2898)
4473 - mds_reint_create() should take same inode create lock (2926)
4474 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4475 - don't close files for self_export to avoid uninitialized obd (2936)
4476 - allow MDS with the same name as client node (2939)
4477 - hold dentry reference for closed log files for unlink (2325)
4478 - reserve space for all logs during transactions (2059)
4479 - don't evict page beyond end of stripe extent (2925)
4480 - don't oops on a deleted current working directory (2399)
4481 - handle hard links to targets without a parent properly (2517)
4482 - don't dereference NULL lock when racing during eviction (2867)
4483 - don't grow lock extents when lots of conflicting locks (2919)
4485 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4488 - account for cache space usage on clients to avoid data loss (974)
4489 - lfsck support in lustre kernel code (2349)
4490 - reduce journal credits needed for BRW writes (2370)
4491 - orphan handling to avoid losing space on client/server crashes
4492 - ptlrpcd can be blocked, stopping ALL progress (2477)
4493 - use lock value blocks to assist in proper KMS, faster stat (1021)
4494 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4495 - recovery for initial connections (2355)
4496 - fixes for mds_cleanup_orphans (1934)
4497 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4498 - block all file creations until orphan recovery completes (1901)
4499 - client remove rq_connection from request struct (2423)
4500 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4501 - recovery timer race (2670)
4502 - mdc_close recovey bug (2532)
4503 - ptlrpc cleanup bug (2710)
4504 - mds timeout on local locks (2588)
4505 - namespace lock held during RPCs (2431)
4506 - handle interrupted sync write properly (2503)
4507 - don't try to handle a message that hasn't been replied to (2699)
4508 - client assert failure during cleanup after abort recovery (2701)
4509 - leak mdc device after failed mount (2712)
4510 - ptlrpc_check_set allows timedout requests to complete (2714)
4511 - wait for inflight reqs when ptlrpcd finishes (2710)
4512 - make sure unregistered services are removed from the srv_list
4513 - reset bulk XID's when resending them (caught by 1138 test)
4514 - unregister_bulk after timeout
4515 - fix lconf error (2694)
4516 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4517 - readahead locks pages, leaves pending causing memory pressure (2673)
4518 - increase OST request buffers to 4096 on large machines (2729)
4519 - fix up permission of existing directories in simple_mkdir (2661)
4520 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4521 - don't assign transno to errored transactions (2742)
4522 - don't delete objects on OST if given a bogus objid from MDS (2751)
4523 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4524 - if rq_no_resend, then timeout request after recovery (2432)
4525 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4526 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4527 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4528 - MDS recovery shouldn't delete objects that it has given out (2730)
4529 - if enqueue arrives after completion, don't clobber LVB (2819)
4530 - don't unlock pages twice when trigger_group_io returns error (2814)
4531 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4532 - don't write pages to disk if there was an error (1450)
4533 - don't ping imports that have recovery disabled (2676)
4534 - take buffered bytes into account when balancing socknal conn (2817)
4535 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4536 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4537 - remove little-endian swabbing of llog records (1987)
4538 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4539 - retry reposting request buffers if they fail (1191)
4540 - grow extent at grant time to avoid granting a revoked lock (2809)
4541 - lock revoke doesn't evict page if covered by a second lock (2765)
4542 - disable VM readahead to avoid reading outside lock extents (2805)
4544 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4545 - updated kernel patches for hp-2.4.20 kernel (2681)
4547 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4550 - fix truncated write corruption (2366)
4551 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4553 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4554 - ensure len(uuid) < 37 in lmc (1171)
4555 - fix ia64 OOPS in llog_test (2255)
4556 - zero end of page at obdfilter for partial page writes (2648)
4557 - don't leave stale dentries around after renames (bug 2428)
4558 - fix timeouts when evicting a client with a single lock held (2642)
4559 - set deadline for the initial HELLO message to drain (2634)
4560 - print out dotted-quad IP addresses in the socknal (2302)
4562 - additional debugging for MDS client eviction problem (2443)
4563 - fix mkfsoptions support for osts (2603, 2604)
4565 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4568 - add series for the vanilla 2.6.0 kernel
4569 - add series for the vanilla 2.4.24 kernel
4570 - add series for a cray x86/64 UL kernel drop
4571 - fix xattr patches for the vanilla 2.4.19 series
4573 - generate true UUIDs in lmc (1171)
4574 - have portals stack dumping break in UML (2466)
4575 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4576 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4577 - wake callback waiting threads on client eviction (2460)
4578 - Add --ptldebug and --subsystem to lmc (1719)
4579 - update assertion to allow safe interrupt allocation
4580 - set rq_no_resend for cancel requests (2432)
4581 - recalculate ptlrpcd timeout after resend (2494)
4582 - call vfs_rmdir when removing pending directories (2368)
4583 - fix renaming a file to itself (2429)
4584 - lmc creates a default one-stripe lov (2454)
4585 - expand procfs space to handle large clusters (2326)
4586 - increase UML stack to avoid overflow
4587 - update lconf's list of debug and subsystem masks
4588 - fix lfs find --obd (2510)
4589 - /proc tunable for disabling filter read caching (2591)
4590 - stop rpm packages from altering slapd.conf (2301)
4591 - disable nagle in the socknal under 0conf (2578)
4592 - choose mds inode size based on stripe count (2572)
4593 - fix kernel-source rpm problems (2516)
4595 - add --disable-doc to avoid pdf generation (2421)
4596 - update documentation, tests, type-os, comments
4597 - avoid format warnings on ia64
4598 - remove the TOE NAL
4599 - tiny code cleanups by removing unused fields
4601 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4604 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4605 - remove the most blatant lies from BUILDING file (2371)
4606 - change default debug level to reasonable production setting
4607 - reduce client side cache size to reduce cache flush time
4608 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4609 - make TCP zerocopy and pinger support enabled by default (2476)
4610 - sync writes completed after process exits caused crashes (2319)
4611 - maintain correct mount count on the MDS (2356)
4612 - backout 1557, because 2316 wasn't really fixed
4613 - better file I/O statistics gathering in /proc
4614 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4615 - another kernel patch to fix zero-copy TCP function export
4616 - don't take duplicate lock when processing re-sent getattr (2420)
4617 - lctl uses obd_self_export instead of creating new conn (2353)
4618 - MDS/OST recovery case which requires object creation asserted (2425)
4619 - move lfs from /usr/sbin to /usr/bin in packages
4620 - fix race between mds_client_add and mds_client_free (2417)
4621 - use kmalloc instead of slabs in portals (2430)
4622 - don't create duplicate records when a failover MDS is present (2442)
4623 - remove unnecessary mount age check (2332)
4624 - don't remove directory inodes from locks prematurely (2451)
4625 - don't break if MDS service name is the same as hostname (2103)
4626 - fix races in client write RPC generation when cache full (2482)
4628 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4631 - remove now-unused request->rq_obd (278)
4632 - if an allocation fails, print out how much memory we've used (1933)
4633 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4634 - add the "configurable stack size" patch to most series files (1256)
4635 - ability to write large log records, for 100+ OST configs (2306)
4636 - fix NULL deref when filter_prep fails (2314)
4637 - fix operator precedence error in filter_sync
4638 - dynamic allocation of socknal TX descriptors (2315)
4639 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4640 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4641 - ensure that CWARN messages in Portals always get to the syslog
4642 - __init/__exit are not for prototype decls (ldlm_init/exit)
4643 - x86-64 compile warning fixes
4644 - fix gateway LMC keyword conflict (2318)
4645 - fix MDS lock inversions in getattr/reint paths (1844)
4646 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4647 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4648 - DLM race condition prevented some lock evictions (2328)
4649 - ENOMEM detection and retry on socknal sends (2230)
4650 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4651 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4652 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4653 - better error messages when a client is rejected during recovery (1505)
4654 - avoid cancelling locks which were never granted, after failure (2330)
4655 - fix i_sem/journal inversion in mds_client_add (2333)
4656 - fix truncate/getattr lock cycle deadlock (2334)
4657 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4658 - fix two rare exit paths which could leak an l_lock() ref (2321)
4659 - fencepost error in MDS/OST orphan recovery (2226)
4660 - make log record alignment 8 bytes (1988)
4661 - lstripe now fails when requested offset > ost_count (2237)
4662 - ensure that all kernel series have a complete list.h (1607)
4663 - fix crashes in special-file operations (2316)
4664 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4666 - allow configurable automake binary, for testing new versions
4667 - small update to the lfs documentation
4669 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4671 * fix negative export reference count in fsfilt_sync (2312)
4673 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4674 * release candidate 0.9.1
4676 - orphans are moved into the PENDING directory for possible recovery
4677 - replayed opens now open by fid for orphan/rename safety (1042)
4678 - last close of an orphan inode generates a transno (683)
4679 - chdir() and mount() now pin the directory entry (1020)
4680 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4681 - discard very old requests without processing them (1502)
4682 - remove some common, well-understood CERRORs (1505)
4683 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4684 - clear "grant" flags in OST replies until OST grant code lands (1644)
4685 - fix read performance by not clobbering i_blksize on client (1598)
4686 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4687 - make LRU size a /proc tunable, clears locks when reduced (707)
4688 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4689 - server locks take references on exports now (1558)
4690 - build fixes for 2.4.20-rh trees (1663)
4691 - return an error from lov_create if all OSCs are inactive (1751)
4692 - fix import levels when a reconnect happens without a timeout (1597)
4693 - exit early from mds_open if we get a lookup error (1749)
4694 - partial page read at EOF wouldn't wait for disk before sending (1642)
4695 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4696 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4697 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4698 - write/truncate lock inversion (1639)
4699 - Don't auto-load obdclass, portals modules during cleanup (1495)
4700 - fix timestamps from jumping to "now" (1763)
4701 - extra journal assertions (1648)
4702 - add an extra multiunlink test (1771)
4703 - fix read_record/write_record API (1776)
4704 - fix leak of offset_extent, possible incorrect i_size later (1772)
4705 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4706 - Debugging for the kqswnal_get_idle_tx problems (1820)
4707 - Allow recovery to be attempted multiple times (1536)
4708 - Write out MDS last_rcvd file after it is first created (1600)
4709 - Fix tx_descriptor leak in failed transmit situations (1827)
4710 - ext3 journaling fixes for assertion failure after IO error (1871)
4711 - class_export_put() on freed export after completion AST error (1896)
4712 - Fix revalidate looping in VFS (1322)
4713 - Don't access a freed export during MDS_REINT timeout (1521)
4714 - Add open-unlink recovery support on the MDS (1673,1764)
4715 - Return an error if no MDS data was read from last_rcvd (1946)
4716 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4717 - Don't setup a disk device that doesn't match exported UUID (317)
4718 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4719 - avoid committing NULL handle in force close
4720 - local.sh is now a one-stripe LOV configuration
4721 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4722 - fix ext3 htree duplicate directory entry corruption (1516)
4723 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4724 - update to new LOV EA format (2097)
4725 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4727 2003-06-15 Phil Schwan <phil@clusterfs.com>
4730 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4731 - per-import/export recovery handling (958, 931, 959)
4732 - multiple last-rcvd slots, for serving multiple FSes (949)
4733 - connections are again shared between multiple imp/exports (963, 964)
4734 - "umount -f" would hang if any requests needed to be sent (393, 978)
4735 - avoid pinning large req buffer by copying for queued messages (989)
4736 - add "uuid" to "lctl device" command to help upcalls (991)
4737 - "open" RPCs with transnos would confuse recovery counters (1037)
4738 - do proper endian conversion of all wire messages (288, 340, 891)
4739 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4740 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4741 - fix ptlrpc_connection leak in target_handle_connect (1174)
4742 - fix import refcounting bug in OST and MDS cleanup (1134)
4743 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4744 - fix very unlikely obd_types race condition (501)
4745 - remove osc_open hack for echo_client (1187)
4746 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4747 - a failure in read_inode2 leads to deadlock (1139)
4748 - cancel ack-locks as soon as transaction is committed (1072)
4749 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4750 - make sure to commitrw after any preprw to avoid deadlock (1162)
4751 - failing to execute a file in a lustre FS would lock inode (1203)
4752 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4753 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4754 - obd_uuid2tgt would walk off the end of the list (1255)
4755 - on IA64 the osc would give portals incorrect bulk size (1258)
4756 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4757 - fix lock inversion caused by new llite matching code (1282)
4758 - limit the number of dirty pages on a client to 10MB (1286)
4759 - timed out locks were not being corrected cancelled (1289)
4760 - fix O_DIRECT above 4GB on IA-32 (1292)
4761 * major user-visible changes
4762 - fail out/fail over policy now controlled by the upcall (993)
4764 - add OBD_PING to check server availability and failure (954)
4765 - lustre messages are now sent in sending host order (288, 340, 891)
4766 - add eadatalen to MDS getattr reply (340)
4767 - OST read replies may contain second buffer, with per-page status (593)
4769 2003-03-11 Phil Schwan <phil@clusterfs.com>
4772 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4773 - failed open()s could cause deadlock; fixed (867, 869)
4774 - stop cancelling OST locks when files are closed (481)
4775 - overlapping XID spaces caused network corruption (851, 853)
4776 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4777 - setattr_raw added, to do single-RPC, server-side setattrs
4778 - lmc/lconf syntax change for OST UUIDs
4779 - fix crashy race condition between ptlrpc_free_req and osc_close
4780 - don't use request in mdc_enqueue if we hit a timeout (889)
4781 - don't set the inode i_size for regular files from the MDS (896)
4782 - handle out of order completion AST (842)
4783 - don't LBUG if a lock request times out after receiving AST (913)
4784 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4785 - if a bad lock AST arrives, send an error instead of dropping entirely
4786 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4787 - fix leak in bulk IO when only partially completed (899, 900, 926)
4788 - fix O_DIRECT for ia64 (55)
4789 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4790 - C-z after timeout could hang a process forever; fixed (977)
4792 - client-side I/O cache (678, 924, 929, 941, 970)
4794 - READPAGE and SETATTRs which don't take server-side locks get
4797 2003-02-11 Phil Schwan <phil@clusterfs.com>
4800 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4801 - Fix fsfilt_extN_readpage() to read a full page of directory
4802 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4803 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4804 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4805 - Most intent operations were removed, in favour of a new RPC mode
4806 that does a single RPC to the server and bypasses most of the VFS
4807 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4808 - Aggressively cancel local locks on DLM servers
4809 - mds_reint_unlink sends EA to the client if it's the last nlink.
4810 client uses that EA to unlink OST objects.
4811 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4812 - recursive symlinks were fixed (439)
4813 - fixed NULL deref in DEBUG_REQ
4814 - filter_update_lastobjid no longer calls sync, which annoyed extN
4815 - fixed multi-client small-writes to a single file problem (445)
4816 - fixed mtime updates during file writes (607)
4817 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4818 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4819 - fixed Linux OST size reporting problem (444, 656)
4820 - OST now updates object mtime with writes or setattr (607, 619)
4821 - client verifies file size before zeroing page past EOF (445)
4822 - OST now writes last allocated objid to disk with allocation (108)
4823 - LOV on echo now works (409)
4825 - mds_reint_unlink sends a new buffer, with the EA included. this
4826 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4827 set if a regular file was being unlinked, and it was the last link
4828 - use PtlGet from the target for bulk writes (315)
4829 - OST now updates object mtime with writes or setattr (607, 619)
4830 - LDLM now has a grant-time callback to revalidate locked items, if
4832 - Many MDS operations were reorganized to combat race conditions
4834 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4835 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4837 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4840 - Fully reactivate OST imports after reconnection (512, others)
4841 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4842 - More graceful error handling for truncating on dead OST (515)
4843 - Don't error out unless we're actually accessing dead stripes (474)
4844 - Fix garbage sizes when stripes are missing (410)
4845 - LRU counters were broken, causing constant lock purge (433, 432)
4846 - garbage on read from stripes with failed OSTs (441)
4847 - mark OSCs as active before reconnecting during recovery (438)
4848 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4849 - lfind did not preserve OST order in output (443)
4850 - symlinks cause hung clients, incorrect data (439)
4851 - stop dereferencing request after dropping refcount (457)
4852 - don't LASSERT(spin_is_locked) on non-SMP (455)
4853 - fixes for many rename() bugs
4854 - fstat didn't correctly synchronize attributes (399)
4855 - server must handle lock cancellation during blocking AST prep (487)
4856 - bulk descriptors were free()d too soon (511)
4857 - fix paths in lconf, which would load incorrect modules (451, 507)
4858 - fix confusing lconf 'host not found' error message (386)
4859 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4860 - fix race condition in mdc_blocking_ast() for inode access (526)
4861 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4862 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4863 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4864 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4865 - fix lstripe in several ways (kernel side) (527)
4866 - fix request leak in ldlm_cli_enqueue (262)
4867 - incorrect OSC was marked inactive after OST failure
4868 - call mds_fs_cleanup before unmounting filesystem (524)
4869 - fix races between taking ns_lock and ldlm_lock_change_resource
4870 - fix races updating LOV export open file list
4871 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4872 - fix recovery NULL deref in ldlm_cli_cancel_unused
4873 - fix some DLM races by using new hash table for lock handles (419)
4874 - permit the client to specify desired inodes, at replay
4875 - duplicate requests when we queue them for replay reintegration
4876 - fix last_rcvd offset calculation
4877 - sync after each recovered transaction, so we always make progress
4878 - never, not always, ERESTART requests without transnos
4879 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4881 - skip replay if the MDS says that the client is already connected
4882 - don't check for a recovery-enabled export to match lctl's UUID
4883 - don't INC_USE_COUNT for phantom exports
4884 - don't crash when cleaning up phantom exports (567)
4885 - don't double-finish or set replay data for errored mdc_open requests
4886 - abort requests when they time out, so we don't get old replies
4887 - send/receive replies for AST messages again
4888 - if the client says that it doesn't have the lock, cancel it on the
4890 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4891 mark it as destroyed, it will all work out when decref is called
4892 - fix module use counts (22, 581)
4894 - ASTs now expect a reply (server cancels lock on error reply)
4896 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4899 - fix many simultaneous client startup (392)
4900 - fix dentry->d_it clobbering
4901 - credentials weren't being shipped for readdir/getattr operations
4902 - remove invalid assertions triggered during some concurrent MD
4904 - proper Lustre versions added (336, 389)
4905 - fix memory leak for create error case (398)
4906 - fix LOV locking bug that would get cli/srv out of sync
4907 - fix echo client over LOV (409)
4908 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4909 - fix double-O_EXCL intent crash (424)
4910 - avoid sending multiple lock CANCELs (352)
4912 - MDS can do multi-client recovery (modulo bugs in new code)
4914 - many updates, edits, cleanups
4916 2002-11-18 Phil Schwan <phil@clusterfs.com>
4919 - fix null d_it dereference (346)
4920 - fix full OST/dbench hang (333)
4921 - fix permission problem with file removal (286)
4922 - fix removal of OSCs from LOV when they fail
4923 - fix NULL deref during bulk timeout (214)
4924 - fix problems related to multiple filesystems on one MDS (241)
4925 - fixed serious subtle metadata locking bugs
4926 - free locks on clients when inodes are removed due to memory
4928 - fix inode pointer in lock data (285)
4929 - partial support for multiple MDS on a single host (241)
4930 - data locks weren't cancelled at clear_inode time (290, 311)
4931 - intent locks could lead to unbounded lock growth (205)
4932 - added a maximum lock count, an LRU list, and a flusher
4933 - fix multiple rename (365)
4934 - properly abstracted the echo client
4935 - OSC locked 1 byte too many; fixed
4936 - rewrote brw callback code:
4937 - fixed recovery bugs related to LOVs (306)
4938 - fixed too-many-pages-in-one-write crash (191)
4939 - fixed (again) crash in sync_io_timeout (214)
4940 - probably fixed callback-related race (385)
4942 - Add capability to MDS protocol
4943 - LDLM cancellations and callbacks on different portals
4945 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4948 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4949 - defer open object destruction to close time (601981, 138)
4950 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4951 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4952 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4953 - don't call obd_setattr in ll_file_release for destroyed objects
4954 * protocol change to lustre_msg: move |version| and add |flags|
4955 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4956 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4957 * added replay of create, unlink, link and rename operations during
4958 MDS failover; recovery should be much more robust now
4959 * remove failed OSCs from LOVs (only lov_create uses this so far)
4960 * the lustre-HOWTO was brought (more) up to date (582544)
4962 2002-10-23 Phil Schwan <phil@clusterfs.com>
4965 - in-use dentries weren't being reused properly (617851)
4966 - prevent multiple LDLM setup (599178)
4967 - fix LOV size calculations for truncate (617853)
4968 - fix client handling of MDS intent errors (POSIX)
4969 - fix permission bug in lovstripe.c test (624321)
4970 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4971 - truncate past end of file could corrupt data
4972 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4973 - a race in recovery could return ETIMEDOUT to apps (623947)
4974 - building outside the source directory was fixed
4975 * the lustre-HOWTO was brought (more) up to date (582544)
4976 * major progress was made on recovery functionality
4978 2002-10-10 Phil Schwan <phil@clusterfs.com>
4981 - recovery deadlock fix
4982 - rm -rf causes LBUG fix (617817)
4983 - file open by multiple tasks fix (618962)
4984 - directory permissions bugs (602707 and 620007)
4985 - journal_stop fixed with locking (611313)
4986 - O_APPEND failures resolved (618273, perhaps 614459)
4987 - lconf PATH fix (619770)
4988 - IA64 build fix (621450)
4989 - RPC buffer sizes scale with amount of memory
4991 2002-10-01 Phil Schwan <phil@clusterfs.com>
4994 - locks would be cancelled without throwing away data pages,
4995 resulting in inconsistent data (605627)
4996 - inode attributes were not always being refreshed (605627, 612449)
4997 - lconf now continues to cleanup after lctl reports an error
4998 - MDS now enforces user permissions (602707)
4999 - lprocfs cleanup fixed, but not yet enabled (614157)
5000 - fixed infinite server hang, should a client not respond to an AST
5001 - avoid going into recovery if user calls readlink() with a buffer
5002 that's too small (613941)
5003 - AST RPCs no longer require replies (614867) -- this may be changed
5004 - don't crash server if client sends an IOV that's too big (611336)
5005 - fixed lock conversion deadlock (611892)
5006 - fixed the following of symlinks (614622)
5007 * recovery: the server can remove locks from a client that dies, other
5008 clients can make progress
5009 * more extN patch fixes
5010 * compile-time configurable ptlrpc buffer allocations
5012 - collaborative read cache document
5013 - Lustre Lite Performance CDR document-in-progress
5015 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5018 - fix typo in patch-2.4.18
5020 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5023 - clear ptlrpc request each time in handle_incoming_request()
5024 - unlink of files now destroys the object on the OST
5026 2002-09-19 Peter Braam <braam@clusterfs.com>
5028 * add hard link support
5029 * change obdfile creation method
5030 * kernel patch changed
5032 2002-09-19 Peter Braam <braam@clusterfs.com>
5035 - stack overflow bug in extN fixed
5037 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
5039 * documentation updates
5040 - add man pages for config tools
5041 - update tests/README to describe testing with new config tools
5042 - finish metadata API descriptions
5043 * bug fixes and cleanups
5044 - statfs workaround for 16TB limit
5045 - LOV stripe allocation improved, can stripe on subset of OSTs
5046 - LOV file size/IO offset was wrong for files > 4GB in size
5047 - object EA data was being dropped, caused files to be unreadable
5048 - memory overflow with non-LOV OST caused memory corruption
5049 - fixed regression tests to work with new config tools, obdfilter
5050 - fixed bug when directory size became larger than 1 block
5051 - fixed bug (for single client case) when PWD was deleted
5052 - invalidate local directory pages when doing intent-based ops
5053 - avoid LDLM oops when lock callback contained bad data
5055 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5057 * documentation updates
5058 * bug fixes and cleanups
5059 - configuration tools
5062 - 64-bit compile warnings
5063 - 64-bit internal statfs data
5065 * test_brw on persistent OST devices
5067 * lprocfs (disabled)
5069 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5071 * documentation updates
5072 * bug fixes and cleanups
5073 * configuration tools
5075 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5078 * many small fixes to 0_5_4
5079 * io/network handling
5080 * thinkos in MDS operations
5082 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5085 * crucial basic fixes to 0.5.3
5086 * IOR, Iozone work over Elan
5089 2002-08-07 Phil Schwan <phil@clusterfs.com>
5090 * version 0_5_3, our first alpha
5091 * we use the new Portals iovs
5092 * documentation updates
5093 * bug fixes and cleanups
5094 * small changes in the DLM wire protocol
5096 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5097 * version 0_5_1 with some initial stability,
5098 * locking on MD and file I/O.
5099 * documentation updates
5100 * several bug fixes since 0.5.0
5101 * small changes in wire protocol
5103 2002-07-18 Phil Schwan <phil@clusterfs.com>
5105 * delivered as Lustre Light Alpha
5106 * fixed a crash after handling invalid MDS requests
5107 * fixed directory pages for architectures with non-4k pages sizes
5109 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5110 * release version v0_4_4
5111 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5113 * Fixes a number of interruption problems with OST operations.
5114 * Update documentation for portals header changes
5115 * Move all wire protocol structs/defines to lustre_idl.h
5116 * Fixes symlink length bug.
5117 * Add tcpdump to repository.
5119 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5120 * release version v0_4_3
5121 * Fixes statfs for inodes on extN.
5122 * Fixes bug in runtests which would delete /etc/hosts.
5123 * Use 64-bit object IDs wherever possible (not into VFS though)
5124 Remove ost_get_info, which is unused by lustre, and out of date.
5126 2002-07-03 Peter Braam <braam@clusterfs.com>
5127 * release version v0_4_2 Fixes a lookup error (type not passed)
5128 * move forward to head of Portals
5129 * move forward to latest Lustre kernel
5131 2002-06-25 Peter Braam <braam@clusterfs.com>
5132 * release version v0_4_1. Hopefully stable on single node use.