1 tbd Sun Microsystems, Inc.
4 2.6.16.60-0.27 (SLES 10),
5 2.6.18-92.1.10.el5 (RHEL 5),
6 2.6.22.14 vanilla (kernel.org).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.40.11-sun1
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
17 Frequency : only patchless client
18 Description: add workaround for race between add/remove dentry from hash
20 Severity : enhancement
22 Description: Allow OST glimpses to return PW locks
26 Description: LBUG when llog conf file is full
27 Details : When llog bitmap is full, ENOSPC should be returned for plain
32 Description: Prevent import from entering FULL state when server in recovery
36 Description: service mount cannot take device name with ":"
37 Details : Only when device name contains ":/" will mount treat it as
43 Description: replace ptlrpcd with the statahead thread to interpret the async
44 statahead RPC callback
48 Frequency : on recovery
49 Description: I/O failures after umount during fail back
50 Details : if client reconnected to restarted server we need join to recovery
51 instead of find server handler is changed and process self eviction
52 with cancel all locks.
54 Severity : enhancement
56 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
58 Severity : enhancement
60 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
62 Severity : enhancement
64 Description: Upcall on Lustre log has been dumped
65 Details : Allow for a user mode script to be called once a Lustre log has
66 been dumped. It passes the filename of the dumped log to the
67 script, the location of the script can be specified via
68 /proc/sys/lnet/debug_log_upcall.
73 Description: avoid idr_remove called for id which is not allocated.
74 Details : Move assigment s_dev for clustered nfs to end of initialization,
75 to avoid problem with error handling.
80 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
81 Details : When connection is reused this not moved from CONN_UNUSED_HASH
82 into CONN_USED_HASH and this prodice warning when put connection
85 Severity : enhancement
87 Description: Export bytes_read/bytes_write count on OSC/OST.
91 Description: Early reply size mismatch, MGC loses connection
92 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
93 the connect flags are properly negotiated.
97 Description: Properly propagate oinfo flags from lov to osc for statfs
98 Details : restore missing copy oi_flags to lov requests.
100 Severity : enhancement
102 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
103 : l_getgroups(8), lst(8), routerstat(8)
104 Details : included man pages for llobdstat(8), llstat(8),
105 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
107 Severity : enhancement
109 Description: configure's --enable-quota should check the
110 : kernel .config for CONFIG_QUOTA
111 Details : configure is terminated if --enable-quota is passed but
112 : no quota support is in kernel
116 Description: Remove portals compatibility
117 Details : Remove portals compatibility, not interoperable with releases
122 Description: Resolve device initialization race
123 Details : Prevent proc handler from accessing devices added to the
124 obd_devs array but yet be intialized.
126 Severity : enhancement
128 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
130 Severity : enhancement
132 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
137 Description: Using wrong pointer in osc_brw_prep_request
138 Details : Access to array[-1] can produce panic if kernel compiled with
139 CONFIG_PAGE_ALLOC enabled
143 Description: Client runs out of low memory
144 Details : Consider only lowmem when counting initial number of llap pages
148 Description: *optional* service tags registration
149 Details : if the "service tags" package is installed on a Lustre node
150 When the filesystem is mounted, a local-node service tag will
151 be created. See http://inventory.sun.com/ for more information
152 about the Service Tags asset management system.
156 Description: Kernel BUG tries to release flock
157 Details : Lustre does not destroy flock lock before last reference goes
158 away. So always drop flock locks when client is evicted and
159 perform unlock regardless of successfulness of speaking to MDS.
163 Description: add recount protection for osc callbacks, so avoid panic on shutdown
167 Description: sanity test 65a fails if stripecount of -1 is set
168 Details : handle -1 striping on filesystem in ll_dirstripe_verify
173 Description: ASSERTION(CheckWriteback(page,cmd)) failed
174 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
177 Severity : enhancement
179 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
183 Description: do not process already freed flock
184 Details : flock can possibly be freed by another thread before it reaches
185 to ldlm_flock_completion_ast.
189 Description: LBUG during stress test
190 Details : Need properly lock accesses the flock deadlock detection list.
194 Description: oops in page fault handler
195 Details : kernel page fault handler can return two special 'pages' in error case, don't
196 try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
200 Description: timeout with invalidate import.
201 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which should be
202 handled by ptlrpcd. This produce long age waiting and -ETIMEOUT
203 ptlrpc_invalidate_import and as result LASSERT.
205 Severity : enhancement
207 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
211 Description: enable MGS and MDT services start separately
212 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
213 a co-located MGS without starting the MGS, which is a complement
214 to 'nosvc' mount option.
218 Frequency : after recovery
219 Description: precreate to many object's after del orphan.
220 Details : del orphan st in oscc last_id == next_id and this triger growing
221 count of precreated objects. Set flag LOW to skip increase count
222 of precreated objects.
226 Frequency : rare, on clear nid stats
227 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
228 Details : when clean nid stats sometimes try destroy live entry,
229 and this produce panic in free.
233 Description: Stack overflow during MDS log replay
234 ease stack pressure by using a thread dealing llog_process.
238 Description: wait until IO finished before start new when do lock cancel.
239 Details : VM protocol want old IO finished before start new, in this case
240 need wait until PG_writeback is cleared until check dirty flag and
241 call writepages in lock cancel callback.
243 Severity : enhancement
245 Description: using special macro for print time and cleanup in includes.
249 Description: mds_mfd_close() ASSERTION(rc == 0)
250 Details : In mds_mfd_close(), we need protect inode's writecount change
251 within its orphan write semaphore to prevent possible races.
255 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
259 Frequency : rare, on shutdown ost
260 Description: don't hit live lock with umount ost.
261 Details : shrink_dcache_parent can be in long loop with destroy dentries,
262 use shrink_dcache_sb instead.
266 Description: don't panic with use echo client
267 Details : echo client pass NULL as client nid pointer and this produce null
272 Description: fix build on ppc32
273 Details : compile code with -m64 flag produce wrong object file for ppc32.
277 Description: add message levels for liblustreapi
281 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
282 Details : Move the 'good_osts' check before the 'total_bavail' check. This
283 will result in an -EAGAIN and in the exit call path we call
284 alloc_rr() which will with increasing aggressiveness attempt to
285 aquire precreated objects on the minimum number of required OSCs.
289 Description: Use old size assignment to avoid deadlock
290 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
291 the scheduling while holding a spinlock. We do not need locking
292 for size in ll_update_inode() because size is only updated from
293 the MDS for directories or files without objects, so there is no
294 other place to do the update, and concurrent access to such inodes
295 are protected by the inode lock.
299 Description: resolve "_IOWR redefined" build error on SLES10
303 Description: dump the memory debugging after all modules are unloaded to
304 suppress false negative in conf_sanity test 39
306 Severity : enhancement
308 Description: build kernel-ib packages for OFED 1.3 in our release cycle
313 Description: fix SLES kernel versioning
314 Details : the kernel version for our SLES 10 kernel did not include a "-"
315 before the "smp" at the end. while this was not a problem in
316 general, it did mean that software trying to use the kernel
317 version to try to detect a vendor specific kernel would fail.
318 this was most evident by the OFED build scripts.
322 Description: Don't update lov_desc members until making sure they are valid
323 Details : When updating lov_desc members via proc fs, need fix their
324 validities before doing the real update.
328 Description: don't put request into delay list while invalidate in flight.
329 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
330 import in flight. this produce timeout for invalidate and sometimes
331 can cause stale data.
335 Frequency : on ppc only
336 Description: not convert ost objects for directory because it's not exist.
337 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
341 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
343 Severity : enhancement
345 Description: Update to support 2.6.22.14 vanilla kernel.
349 Frequency : rare, on recovery
350 Description: read procfs can produce deadlock in some situation
351 Details : Holding lprocfs lock which send rpc can produce block for destroy
352 obd objects and this also block reconnect with -EALREADY. This isn't
353 fix all lprocfs bugs - but make it rare.
355 Severity : enhancement
357 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
360 Frequency : frequent on X2 node
362 Description: mdc_set_open_replay_data LBUG
363 Details : Set replay data for requests that are eligible for replay.
367 Description: lustre_mgs: operation 101 on unconnected MGS
368 Details : When MGC is disconnected from MGS long enough, MGS will evict the
369 MGC, and late on MGC cannot successfully connect to MGS and a lot
370 of the error messages complaining that MGS is not connected.
374 Frequency : on network error
375 Description: panic with double free request if network error
376 Details : mdc_finish_enqueue is finish request if any network error ocuring,
377 but it's true only for synchronus enqueue, for async enqueue
378 (via ptlrpcd) this incorrect and ptlrpcd want finish request
381 Severity : enhancement
383 Description: client-side metadata stat-ahead during readdir(directory readahead)
384 Details : perform client-side metadata stat-ahead when the client detects
385 readdir and sequential stat of dir entries therein
388 Frequency : on start mds
390 Description: Implement get_info(last_id) in obdfilter.
393 Frequency : occasional
395 Description: Correctly check stale fid, not start epoch if ost not support SOM
396 Details : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
397 overwrited with new generated fid, but mds can anwer with one of these
398 two fids and both is not stale. setattr incorectly start epoch and
399 assume will be called done_writeting, but without SOM done_writing
403 Frequency : rare, depends on device drivers and load
405 Description: MDS or OSS nodes crash due to stack overflow
406 Details : Code changes in 1.8.0 increased the stack usage of some functions.
407 In some cases, in conjunction with device drivers that use a lot
408 of stack the MDS (or possibly OSS) service threads could overflow
409 the stack. One change which was identified to consume additional
410 stack has been reworked to avoid the extra stack usage.
413 Frequency : occasional
415 Description: Do not fail import if osc_interpret_create gets -EAGAIN
416 Details : If osc_interpret_create got -EAGAIN it immediately exits and
417 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
418 oscc_has_objects and see OSC has no objests and call
419 oscc_internal_create to resend create request.
421 Severity : enhancement
423 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
425 Severity : enhancement
427 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
430 Frequency : very rare
432 Description: Fix replay if there is an un-replied request and open
433 Details : In some cases, older replay request will revert the
434 mcd->mcd_last_xid on MDS which is used to record the client's
437 Severity : enhancement
439 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
441 Severity : enhancement
443 Description: Add rhel5 support to HEAD.
445 Serverity : enhancement
447 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
452 Description: Don't allow skipping OSTs if index has been specified.
453 Details : Don't allow skipping OSTs if index has been specified, make locking
454 in internal create lots better.
458 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
459 Details : access to bitfield in structure is always rounded to long
460 and this produce problem with not atomic change any bit.
464 Description: Lustre make rpms failed.
465 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
466 builds Lustre rpms from tarball.
469 Frequency : rare on shutdown ost
471 Description: If llog cancel was not send before clean_exports phase, this can
472 produce deadlock in llog code.
473 Details : If llog thread has last reference to obd and call class_import_put
474 this produce deadlock because llog_cleanup_commit_master wait when
475 last llog_commit_thread exited, but this never success because was
476 called from llog_commit_thread.
480 Description: allow userland application know is lost one of stripes.
481 Details : fill lvb_blocks with error code on ost and return it to
482 application if error flag found.
486 Description: NULL lov_tgts causing MDS oops
487 Details : more safe checks for NULL lov_tgts for avoid oops.
489 Severity : enhancement
491 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
495 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
496 Details : If an OST is down the MDS will hang indefinitely in
497 obd_statfs_rqset() waiting for the statfs data. While for
498 MDS QOS usage of statfs, it should not stuck in waiting.
500 Severity : enhancement
502 Description: remote_acl support
503 Details : Support ACL-based permission check for remote user.
504 Support setfacl/getfacl for remote user with the utils
505 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
506 the system "{s,g}etfacl" utils.
508 Severity : enhancement
510 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
512 Severity : enhancement
514 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
518 Description: make mgs_setparam() handle fsname containing dash
519 Details : fsname containing a dash does not work with lctl conf_param
521 Severity : enhancement
523 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
525 Severity : enhancement
527 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
531 Frequency : rare, at shutdown
532 Description: access already free / zero obd_namespace.
533 Details : if client_disconnect_export was called without force flag set,
534 and exist connect request in flight, this can produce access to
535 NULL pointer (or already free pointer) when connect_interpret
536 store ocd flags in obd_namespace.
540 Frequency : only at startup
541 Description: not alloc memory with spinlock held.
542 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
543 if any spinlock held.
545 Severity : enhancement
547 Description: make lustre randomly fail allocating memory
548 Details : Make lustre randomly failed allocating memory for testing purpose.
550 Severity : enhancement
552 Description: lost problems with lov objid file
553 Details : Fixes some scability and access to not inited memory problems
554 in work with lov objdid file.
559 Description: lfs find does not continue on file error
560 Details : Continue other files processing when a file/dir is absent.
564 Description: Inconsistent usage of lustre_pack_reply()
565 Details : Standardize the usage of lustre_pack_reply() such that it
566 always generate a CERROR on failure.
569 Frequency : occasional
571 Description: MDS hang or stay in waiting lock
572 Details : If client receive lock with CBPENDING flag ldlm need send lock
573 cancel as separate rpc, to avoid situation when cancel request
574 can't processed due all i/o threads stay in wait lock.
578 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
582 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
586 Description: Update to SLES9 kernel-2.6.5-7.287.3.
590 Description: Update to RHEL4 latest kernel.
592 Severity : enhancement
594 Description: Build SLES10 patchless client fails
595 Details : The configure was broken by run ./configure with
596 --with-linux-obj=.... argument for patchless client. When the
597 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
598 can't search header adequately. Use absolute path such as
599 -I($LINUX)/include instead.
603 Description: interrupt oig_wait produce painc on resend.
604 Details : brw_redo_request can be used for resend requests from ptlrpcd and
605 private set, and this produce situation when rq_ptlrpcd_data not
606 copyed to new allocated request and triggered LBUG on assert
607 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
608 ptlrpcd set if request is changed and can be safety changed to use
613 Description: LASSERT_{REQ,REP}SWAB macros are buggy
614 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
615 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
616 these macros and replace them with their difinitions instead.
620 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
621 Details : Update which_patch & target file for SLES10 latest kernel.
624 Description: Remove client patches from SLES 10 kernel.
625 Details : This causes SLES 10 clients to behave as patchless clients
626 even on a Lustre-patched (server) kernel.
628 Severity : enhancement
630 Description: self-adjustable client's lru lists
631 Details : use adaptive algorithm for managing client cached locks lru
632 lists according to current server load, other client's work
633 pattern, memory activities, etc. Both, server and client
634 side namespaces provide number of proc tunables for controlling
637 Severity : enhancement
639 Description: light-weight GSS support
640 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
641 and reduce performance overhead.
643 Severity : enhancement
645 Description: Linux keyring support
646 Details : Support using service of Linux keyring for Lustre GSS internal
647 context refresh/cache mechanism.
651 Description: Fix errors in lfs documentation
652 Details : Fixes man pages
656 Description: don't use GFP_* in generic Lustre code.
657 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
661 Description: obdclass is limited by single OBD_ALLOC(idarray)
662 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
666 Description: warnings with build patchless client with vanila 2.6.19 and up
667 Details : change old ctl_table style and replace ctl_table/ctl_table_header
668 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
672 Frequency : Only for SLES
673 Description: sanity_quota fail test_1
674 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
675 file, which makes TSTID[2] inunique.
679 Frequency : Only for SLES9
680 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
681 Details : Update target/ChangeLog/which_patch .
685 Frequency : Only for test_10 in sanity-quota.sh
686 Description: a bug in quota test code
687 Details : checking a test flag in a wrong place
691 Frequency : Only for test_13 in sanity-quota.sh
692 Description: a bug in quota test script
693 Details : shouldn't check overfull allocated quota
697 Description: jbd statistics
698 Details : Port older jbd statistics patch for sles10
700 Severity : enhancement
702 Description: Add printing inode info into message about error in writepage.
707 Description: reply_lock_interpret crash due to race with it and lock cancel.
708 Details : Do not replay locks that are being cancelled. Do not reference
709 locks by their address during replay, just by their handle.
713 Frequency : When flocks are used.
714 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
715 Details : Flock locks might destroy just granted lock if it could be merged
716 with another existing flock, this is done in completion handler,
717 so teach ldlm_cli_enquque_fini that this is a valid case for
723 Description: Oops in read and write path when failing to allocate lock.
724 Details : Check if lock allocation failed and return error back.
728 Description: lstripe command fails for valid OST index
729 Details : The stripe offset is compared to lov->desc.ld_tgt_count
730 instead of lov->desc.ld_active_tgt_count.
734 Description: sanity.sh failed test 103
735 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
740 Description: df doesn't work properly if diskfs blocksize != 4K
741 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
745 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
746 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
748 Severity : enhancement
750 Description: Adaptive timeouts
751 Details : RPC timeouts adapt to changing server load and network
752 conditions to reduce resend attempts and improve recovery time.
756 Description: llapi_file_create() does not allow some changes
757 Details : add llapi_file_open() that allows specifying the mode and
758 open flags, and also returns an open file handle.
762 Description: merge and cleanup kernel patches.
763 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
767 Description: Add journal checksum support.(Kernel part)
768 Details : The journal checksum feature adds two new flags i.e
769 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
770 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
771 indicates that the commit block contains the checksum for
772 the blocks described by the descriptor blocks. Now commit
773 record can be sent to disk without waiting for descriptor
774 blocks to be written to disk. This behavior is controlled
775 using JBD2_FEATURE_ASYNC_COMMIT flag.
779 Description: OSS needs mutliple precreate threads
780 Details : Add ability to start more than one create thread per OSS.
784 Description: Sanity.sh test_65k failed: setstripe should have succeeded
785 Details : execute lfs setstripe on client
789 Description: mds_obd_create error creating tmp object
790 Details : When the user sets quota on root, llog will be affected and can't
791 create files and write files.
794 Frequency : Always on ia64 patchless client, and possibly others.
796 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
797 Details : This allows the patchless client to be loaded on architectures
802 Description: RedHat Update kernel for RHEL5
803 Details : Add the kernel config file for RHEL5.
807 Description: RedHat Update kernel for RHEL5
808 Details : Modify the kernel config file more closer RHEL5.
812 Description: Build failure against Centos5 (RHEL5)
813 Details : Define PAGE_SIZE when it isn't present.
817 Description: test_7 on sanity-quota.sh can't be used on separate machines
818 Details : add facet to handle it
822 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
823 Details : with NFS, the anon dentry's parent was set to itself in
824 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
825 corresponding dentry other than use rec->ur_name.
827 Severity : enhancement
829 Description: lfs setstripe enhancement
830 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
834 Description: enable data checksumming by default
835 Details : enable checksum by default, allow --disable-checksum
836 configure option and "-o nochecksum" mount option. Checksums
837 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
841 Description: lustre support for RHEL5
842 Details : Add support for RHEL5.
846 Description: Client eviction due to failover config
847 Details : after a connection loss, the lustre client should attempt to
848 reconnect to the last active server first before trying the
849 other potential connections.
853 Description: when mds and osts use different quota unit(32bit and 64bit),
854 quota will be released repeatly.
855 Details : void sending multiple quota reqs to mds, which will keep the
856 status between the reqs.
860 Description: osts not allocated evenly to files
861 Details : change the condition to increase offset_idx
864 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
866 Description: Data corruption for OSTs that are formatted larger than 2TB
868 Details : When generating the bio request for lustre file writes the
869 sector number would overflow a temporary variable before being
870 used for the IO. The data reads correctly from Lustre (which
871 will overflow in a similar manner) but other file data or
872 filesystem metadata may be corrupted in some cases.
876 Description: Tune the kernel for good SCSI performance.
877 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
878 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
883 Description: rewrite ext2-derived code in obdclass/uuid.c
884 Details : rewrite inherited code (uuid parsing code from ext2 utils)
885 from scratch preserving functionality.
889 Description: "lfs find -obd UUID" prints directories
890 Details : "lfs find -obd UUID" will return all directory names instead
891 of just file names. It is incorrect because the directories
892 do not reside on the OSTs.
896 Description: use i_size_read and i_size_write in 2.6 port
897 Details : replace inode->i_size access with i_size_read/write()
900 Frequency : when removing large files
902 Description: scheduling issue during removal of large Lustre files
903 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
904 It causes scheduling issues when removing large files (17TB in the
908 Frequency : only with liblustre clients on XT3
910 Description: evictions taking too long
911 Details : allow llrd to evict clients directly on OSTs
914 Frequency : only on ppc
916 Description: /proc/fs/lustre/devices broken on ppc
917 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
918 We should make sure the type of 'index' is loff_t and then cast
919 explicitly as needed below. Do not assign an explicitly cast
923 Frequency : only for rhel5
925 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
926 Details : Modify the target file & which_kernel.
928 Severity : enhancement
930 Description: omit set fsid for export NFS
931 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
932 as FS_REQUIRES_DEV for avoid problems with generate fsid.
936 Frequency : Always, for kernels after 2.6.16
937 Description: Fix warning idr_remove called for id=.. which is not allocated.
938 Details : Last kernels save old s_dev before kill super and not allow
939 to restore from callback - restore it before call kill_anon_super.
943 Description: buffer overruns could theoretically occur
944 Details : llapi_semantic_traverse() modifies the "path" argument by
945 appending values to the end of the origin string, and a
946 overrun may occur. Adding buffer overrun check in liblustreapi.
950 Description: Fix error on 'ls .' at the top of the Lustre mount.
951 Details : Don't revalidate dentry if it is a root dentry.
955 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
956 Details : Modify vm-tunables-rhel4.patch.
960 Description: Kernel config for 2.6.18-vanilla.
961 Details : Modify targets/2.6-vanilla.target.in.
962 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
963 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
964 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
965 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
969 Description: improve handling recoverable errors
970 Details : if request processig with error which can be recoverable on server
971 request should be resend, otherwise page released from cache and
976 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
977 Details : Modify target file & which_patch.
978 A flaw was found in the IA32 system call emulation provided
979 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
980 value could be stored in the %RAX register, which could trigger an
981 out-of-bounds system call table access. An untrusted local user
982 could exploit this flaw to run code in the kernel
983 (ie a root privilege escalation). (CVE-2007-4573).
987 Description: O_DIRECT bypasses client statistics.
988 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
989 read_ahead_stats were updated. copied the stats section from
990 osc_send_oap_rpc() into async_internal().
994 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
998 Description: change order of libsysio includes
999 Details : '#include sysio.h' should always come before '#include xtio.h'
1003 Description: handle "serious error: objid * already exists" more gracefully
1004 Details : If LAST_ID value on disk is smaller than the objects existing in
1005 the O/0/d* directories, it indicates disk corruption and causes an
1006 LBUG(). If the object is 0-length, then we should use the existing
1007 object. This will help to avoid a full fsck in most cases.
1009 Severity : enhancement
1011 Description: adapt the lustre_config script to support the upgrade case
1012 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1013 server targets to 1.6 in parallel.
1017 Description: To avoid grant space > avaible space when the disk is almost
1018 full. Without this patch you might see the error "grant XXXX >
1019 available" or some LBUG about grant, when the disk is almost
1021 Details : In filter_check_grant, for non_grant cache write, we should
1022 check the left space by if (*left > ungranted + bytes), instead
1023 of (*left > ungranted), because only we are sure the left space
1024 is enough for another "bytes", then the ungrant space should be
1025 increase. In client, we should update cl_avail_grant only there
1026 is OBD_MD_FLGRANT in the reply.
1030 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1031 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1032 described in CVE-2007-4573. This problem could allow untrusted
1033 local users to gain root access.
1036 Frequency : when using O_DIRECT and quotas
1038 Description: Incorrect file ownership on O_DIRECT output files
1039 Details : block usage reported by 'lfs quota' does not take into account
1040 files that have been written with O_DIRECT.
1045 Description: touch file failed when fs is not full
1046 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1047 otherwise we can get ENOSP while fs is not full.
1051 Description: parallel lock callbacks
1052 Details : Instead of sending blocking and completion callbacks as separated
1053 requests, adding them to a set and sending in parallel.
1056 Frequency : only for Cray XT3
1057 Bugzilla : 12829/13455
1058 Description: Changing primary group doesn't change the group lustre assigns to
1060 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1061 overridden with the primary group provided by the group upcall,
1062 whereas the supplied fsgid can be trusted if it is in the list of
1063 supplementary groups returned by the group upcall.
1065 Severity : enhancement
1067 Description: Allow masking D_WARNING, D_ERROR messages from console
1068 Details : Console messages can now be disabled via lnet.printk.
1072 Description: User code with malformed file open parameter crashes client node
1073 Details : Before packing join_file req, all the related reference should be
1074 checked carefully in case some malformed flags cause fake
1075 join_file req on client.
1079 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1080 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1081 mode is equal to the granted mode. However, it should also check
1082 that LDLM_FL_CP_REQD is not set.
1086 Description: Heavy nfs access might result in deadlocks
1087 Details : After ELC code landed, it is now improper to enqueue any mds
1088 locks under och_sem, because enqueue might want to decide to
1089 cancel open locks for same inode we are holding och_sem for.
1093 Description: Client eviction while running blogbench
1094 Details : A lot of unlink operations with concurrent I/O can lead to a
1095 deadlock causing evictions. To address the problem, the number of
1096 oustanding OST_DESTROY requests is now throttled to
1097 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1098 callbacks are processed in priority.
1102 Description: enable ACLs on MDS by default
1103 Details : ACLs must be enabled on MDS by default.
1106 Frequency : PPC/PPC64 only
1108 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1109 Details : fix duplicated definitions between asm-ppc64/types.h and
1110 lustre_types.h on PPC.
1113 Frequency : PPC/PPC64 only
1115 Description: asm-ppc/segment.h does not exist
1116 Details : fix compile issue on PPC.
1120 Description: better handle error messages in extents code
1123 Frequency : RHEL4 only
1125 Description: mkfs is very slow on IA64/RHEL4
1126 Details : A performance regression has been discovered in the MPT Fusion
1127 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1128 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1129 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1131 Severity : enhancement
1133 Description: SNMP support enhancement
1134 Details : Adding total number of sampled request for an MDS node in snmp
1137 Severity : enhancement
1139 Description: Optimize ldlm waiting list processing for PR extent locks
1140 Details : When processing waiting list for read extent lock and meeting read
1141 lock that is same or wider to it that is not contended, skip
1142 processing rest of the list and immediatelly return current
1143 status of conflictness, since we are guaranteed there are no
1144 conflicting locks in the rest of the list.
1148 Description: Time out and refuse to reconnect
1149 Details : When the failover node is the primary node, it is possible
1150 to have two identical connections in imp_conn_list. We must
1151 compare not conn's pointers but NIDs, otherwise we can defeat
1152 connection throttling.
1156 Description: port llog fixes from b1_6 into HEAD
1157 Details : Port llog reference couting and some llog cleanups from b1_6
1158 (bug 10800) into HEAD, for protect from panic and access to already
1159 free llog structures.
1163 Description: Detect stride IO mode in read-ahead
1164 Details : When a client does stride read, read-ahead should detect that and
1165 read-ahead pages according to the detected stride pattern.
1169 Description: data checksumming impacts single node performance
1170 Details : add support for several checksum algorithm. Currently, only CRC32
1171 and Adler-32 are supported. The checksum type can be changed on
1172 the fly via /proc/fs/lustre/osc/*/checksum_type.
1176 Description: use adler32 for page checksums
1177 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1182 Description: build for x2 fails
1183 Details : fix compile issue on Cray systems.
1187 Description: Properly match for duplicate locks
1188 Details : Due to different lock order from skiplists code, we need to
1189 traverse entire list for now
1192 Frequency : only on PPC/SLES10
1194 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1195 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1196 be defined. Add a hack in mkfs_lustre.c to work around this compile
1201 Description: LASSERT on MDS when client holding flock lock dies
1202 Details : ldlm pool logic depends on number of granted locks equal to
1203 number of released locks which is not true for flock locks, so
1204 just exclude such locks from consideration.
1208 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1209 Details : Use fsfilt_sync() for both the whole filesystem sync and
1210 individual file sync to eliminate dangerous inode locking
1211 with I_LOCK that can lead to a deadlock.
1215 Description: performance in 1.6.3
1216 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1217 sure that 1MB requests are not fragmented by the block layer.
1219 Severity : enhancement
1221 Description: organize the server-side client stats on per-nid basis
1222 Details : Change the structure of stats under obdfilter and mds to
1230 The "uuid"s file would list the uuids of _active_ exports.
1231 And the clear entry is to clear all stats and stale nids.
1233 Severity : enhancement
1235 Description: eliminate client locks in face of contention
1236 Details : file contention detection and lockless i/o implementation
1237 for contended files.
1241 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1244 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1246 Description: MDS may lose file striping (and hence file data) in some cases
1247 Details : If there are additional extended attributes stored on the MDS,
1248 in particular ACLs, SELinux, or user attributes (if user_xattr
1249 is specified for the client mount options) then there is a risk
1250 of attribute loss. Additionally, the Lustre file striping
1251 needs to be larger than default (e.g. striped over all OSTs),
1252 and an additional attribute must be stored initially in the
1253 inode and then increase in size enough to be moved to the
1254 external attribute block (e.g. ACL growing in size) for file
1259 Description: skiplist implementation simplification
1260 Details : skiplists are used to group compatible locks on granted list
1261 that was implemented as tracking first and last lock of each
1262 lock group the patch changes that to using doubly linked lists
1266 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1267 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1268 because LASSERT on both the data supplied by a client, and the
1269 data on disk is dangerous and incorrect.
1271 Severity : enhancement
1273 Description: Slow truncate/writes to huge files at high offsets.
1274 Details : Directly associate cached pages to lock that protect those pages,
1275 this allows us to quickly find what pages to write and remove
1276 once lock callback is received.
1280 Description: more ldlm soft lockups
1281 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1282 starve other threads from the resource lock for a long time in
1283 case of long waiting queue, so change the debug level from
1284 D_OTHER to the less frequently used D_INFO.
1286 Severity : enhancement
1288 Description: add -gid, -group, -uid, -user options to lfs find
1292 Description: Hung threads in invalidate_inode_pages2_range
1293 Details : The direct IO path doesn't call check_rpcs to submit a new RPC once
1294 one is completed. As a result, some RPCs are stuck in the queue
1299 Description: filter threads hungs on waiting journal commit
1300 Details : Cleanup filter group llog code, then only filter group llog will
1301 be only created in the MDS/OST syncing process.
1305 Description: Procfs and llog threads access destoryed import sometimes.
1306 Details : Sync the import destoryed process with procfs and llog threads by
1307 the import refcount and semaphore.
1309 Severity : enhancement
1311 Description: openlock cache of b1_6 port to HEAD
1316 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1317 Details : If insertion of an extent fails, then discard the inode
1318 preallocation and free data blocks else it can lead to duplicate
1323 Description: don't always update ctime in ext3_xattr_set_handle()
1324 Details : Current xattr code updates the inode ctime in ext3_xattr_set_handle.
1325 In some cases the ctime should not be updated, for example for
1326 2.0->1.8 compatibility it is necessary to delete an xattr and it
1327 should not update the ctime.
1331 Bugzilla : 15713/16362
1332 Description: Assertion in iopen_connect_dentry in 1.6.3
1333 Details : looking up an inode via iopen with the wrong generation number can
1334 populate the dcache with a disconneced dentry while the inode
1335 number is in the process of being reallocated. This causes an
1336 assertion failure in iopen since the inode's dentry list contains
1337 both a connected and disconnected dentry.
1341 Description: assertion failure in ldlm_handle2lock()
1342 Details : fix a race between class_handle_unhash() and class_handle2object()
1343 introduced in lustre 1.6.5 by bug 13622.
1348 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1349 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1354 Description: oops in sd_iostats_seq_show()
1355 Details : unloading/reloading the scsi low level driver triggers a kernel
1356 bug when trying to access the sd iostat file.
1361 Description: Kernel panics during QLogic driver reload
1362 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1363 patch, causing memory corruption.
1368 Description: journal_dev option does not work in b1_6
1369 Details : pass mount option during pre-mount.
1371 Severity : enhancement
1373 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1374 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1375 extent information of a file. It can be used to map logical blocks
1376 in a file to physical blocks in the block device.
1380 Description: LDLM soft lockups - improvement
1381 Details : It is be possible to send the lock handle along with each read
1382 or write request because the client is already doing a lock match
1383 itself so there isn't any reason the OST should have to re-do that
1389 Description: X2 build failures
1390 Details : fix build failures on Cray X2.
1392 --------------------------------------------------------------------------------
1394 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
1396 * Support for kernels:
1397 2.6.5-7.283 (SLES 9),
1398 2.6.9-55.EL (RHEL 4),
1399 2.6.16.46-0.14 (SLES 10),
1400 2.6.18.8 vanilla (kernel.org)
1401 * Client support for unpatched kernels:
1402 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1403 2.6.16 - 2.6.22 vanilla (kernel.org)
1404 * Due to recently discovered recovery problems, we do not recommend
1405 using patchless RHEL 4 clients with this or any earlier release.
1406 * Recommended e2fsprogs version: 1.39.cfs8
1407 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1408 * Starting with this release, the ldiskfs backing filesystem required
1409 by Lustre is now in its own package, lustre-ldiskfs. This package
1410 should be installed. It is versioned separately from Lustre and
1411 may be released separately in future.
1415 Description: block reactivating mgc import until all deactivates complete
1416 Details : Fix race when failing back MDT/MGS to itself (testing)
1419 Frequency : at statup only
1421 Description: mds_lov_synchronize race leads to various problems
1422 Details : simultaneous MDT->OST connections at startup can cause the
1423 sync to abort, leaving the OSC in a bad state.
1425 Severity : enhancement
1427 Description: add optional extra BUILD_VERSION info
1428 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1429 to override the lustre version.
1432 Frequency : 2.6.18 servers only
1434 Description: ll_kern_mount() doesn't release the module reference
1435 Details : The ldiskfs module reference count never drops down to 0
1436 because ll_kern_mount() doesn't release the module reference.
1441 Description: server LBUG when using old ost_num_threads parameter
1442 Details : Accept the old ost_num_threads parameter but warn that it
1443 is deprecated, and fix an off-by-one error that caused an LBUG.
1448 Description: Transient SCSI error results in persistent IO issue
1449 Details : iobuf->dr_error is not reinitialized to 0 between two
1453 Frequency : sometimes when underlying device returns I/O errors
1455 Description: OSTs not going read-only during write failures
1456 Details : OSTs are not remounted read-only when the journal commit threads
1457 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1458 instead of the ext3 wrappers.
1462 Description: poor connect scaling with increasing client count
1463 Details : Don't run filter_grant_sanity_check for more than 100 exports
1464 to improve scaling for large numbers of clients.
1467 Frequency : SLES10 only
1469 Description: sanity-quota.sh quotacheck failed: rc = -22
1470 Details : Quotas cannot be enabled on SLES10.
1473 Frequency : liblustre clients only
1475 Description: getdirentries does not give error when run on compute nodes
1476 Details : getdirentries does not fail when the size specified as an argument
1477 is too small to contain at least one entry
1479 Severity : enhancement
1481 Description: Add LNET router traceability for debug purposes
1482 Details : If a checksum failure occurs with a router as part of the
1483 IO path, the NID of the last router that forwarded the bulk data
1484 is printed so it can be identified.
1489 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1490 Details : Due to a race condition, liblustre clients were occasionally
1491 evicted incorrectly.
1493 Severity : enhancement
1495 Description: lfs setstripe use optional parameters instead of postional
1498 Severity : enhancement
1500 Description: Nanosecond timestamp support for ldiskfs
1501 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1502 resolution timestamps. There is not yet support for this at
1503 the Lustre filesystem level.
1506 Frequency : during server recovery
1508 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1509 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1513 Frequency : nfs export on patchless client
1515 Description: connectathon hang when test nfs export over patchless client
1516 Details : Disconnected dentry cannot be found with lookup, so we do not need
1517 to unhash it or make it invalid
1520 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1521 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1522 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1527 Description: open req refcounting wrong on reconnect
1528 Details : If reconnect happened between getting open reply from server and
1529 call to mdc_set_replay_data in ll_file_open, we will schedule
1530 replay for unreferenced request that we are about to free.
1531 Subsequent close will crash in variety of ways.
1532 Check that request is still eligible for replay in
1533 mdc_set_replay_data().
1538 Description: disable writes to filesystem when reading health_check file
1539 Details : the default for reading the health_check proc file has changed
1540 to NOT do a journal transaction and write to disk, because this
1541 can cause reads of the /proc file to hang and block HA state
1542 checking on a healthy but otherwise heavily loaded system. It
1543 is possible to return to the previous behaviour during configure
1544 with --enable-health-write.
1546 Severity : enhancement
1548 Description: 64-bit inode version
1549 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1550 to the inode. This will be required for version-based recovery.
1555 Description: MDS fails to start if a duplicate client export is detected
1556 Details : in some rare cases it was possible for a client to connect to
1557 an MDS multiple times. Upon recovery the MDS would detect this
1558 and fail during startup. Handle this more gracefully.
1560 Severity : enhancement
1562 Description: Add -o localflock option to simulate old noflock
1564 Details : This will achieve local-only flock/fcntl locks
1570 Description: log_commit_thread vs filter_destroy race leads to crash
1571 Details : Take import reference before releasing llog record semaphore
1576 Description: Wrong request locking in request set processing
1577 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1578 imp_delayed_list, in this place should be used imp_lock.
1581 Frequency : when reconnection
1583 Description: Grant Leak when osc reconnect to OST
1584 Details : When osc reconnect ost, OST(filter) should check whether it
1585 should grant more space to client by comparing fed_grant and
1586 cl_avail_grant, and return the granted space to client instead
1587 of "new granted" space, because client will call osc_init_grant
1588 to update the client grant space info.
1591 Frequency : when client reconnect to OST
1593 Description: Grant Leak when osc do resend and replay bulk write
1594 Details : When osc reconnect to OST, OST(filter)should clear grant info of
1595 bulk write request, because the grant info while be sync between
1596 OSC and OST when reconnect, and we should ignore the grant info
1597 these of resend/replay write req.
1602 Description: Grant space more than avaiable left space sometimes.
1603 Details : When then OST is about to be full, if two bulk writing from
1604 different clients came to OST. Accord the avaliable space of the
1605 OST, the first req should be permitted, and the second one
1606 should be denied by ENOSPC. But if the seconde arrived before
1607 the first one is commited. The OST might wrongly permit second
1608 writing, which will cause grant space > avaiable space.
1611 Frequency : when client is evicted
1613 Description: Grant might be wrongly erased when osc is evicted by OST
1614 Details : when the import is evicted by server, it will fork another
1615 thread ptlrpc_invalidate_import_thread to invalidate the
1616 import, where the grant will be set to 0. While the original
1617 thread will update the grant it got when connecting. So if
1618 the former happened latter, the grant will be wrongly errased
1619 because of this race.
1624 Description: Checking Stale with correct fid
1625 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1626 to check whether it is stale, because sometimes, we want the
1627 enqueue happened anyway, and op_data.fid2 will not be initialized.
1629 Severity : enhancement
1631 Description: update patchless client
1632 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1635 Frequency : only with 2.4 kernel
1637 Description: random memory corruption
1638 Details : size of struct ll_inode_info is to big for union inode.u and this
1639 can be cause of random memory corruption.
1644 Description: Memory leak in recovery
1645 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1646 It should also check obd_fail before fsfilt_start, otherwise if
1647 fsfilt_start return -EROFS,(failover mds during mds recovery).
1648 then the req will return with repmsg->transno = 0 and rc = EROFS.
1649 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1650 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1651 be freed no matter whether fsfilt_commit success or not.
1654 Frequency : only with huge count clients
1656 Description: Prevents from taking the superblock lock in llap_from_page for
1658 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1659 for llap_from_page call in ll_removepage prevents from taking the
1660 superblock lock for a soon died page.
1665 Description: Not check open intent error before release open handle
1666 Details : in some rare cases, the open intent error is not checked before
1667 release open handle, which may cause
1668 ASSERTION(open_req->rq_transno != 0), because it tries to release
1669 the failed open handle.
1674 Description: Set cat log bitmap only after create log success.
1675 Details : in some rare cases, the cat log bitmap is set too early. and it
1676 should be set only after create log success.
1680 Description: Accessing a block bevice can re-enable I/O when Lustre is
1681 tearing down a device.
1682 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1687 Description: service threads may hog cpus when there are a lot of requests
1688 Details : Insert cond_resched to give other threads a chance to use some CPU
1693 Description: the cat log was not initialized in recovery
1694 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1695 unlink log on mds will not be initialized until mds post
1696 recovery. And also in mds post recovery, the unlink log will
1697 initialization will be done asynchronausly, so there will be race
1698 between add unlink log and unlink log initialization.
1702 Description: brw_stats were being printed incorrectly
1703 Details : brw_stats were being printed as log2 but all of them were not
1704 recorded as log2. Also remove some code duplication arising from
1705 filter_tally_{read,write}.
1709 Frequency : rare, only in recovery.
1710 Description: ASSERTION(req->rq_type != LI_POISON) failed
1711 Details : imp_lock should be held while iterating over imp_sending_list for
1712 prevent destroy request after get timeout in ptlrpc_queue_wait.
1716 Description: replay-single.sh test 52 fails
1717 Details : A lock's skiplist need to be cleanup when it being unlinked
1718 from its resource list.
1722 Description: Short directio read returns full requested size rather than
1724 Details : Direct I/O operations should return actual amount of bytes
1725 transferred rather than requested size.
1727 Severity : enhancement
1729 Description: metadata RPC reduction (e.g. for rm performance)
1730 Details : decrease the amount of synchronous RPC between clients and servers
1731 by canceling conflicing lock before the operation on the client
1732 and packing thier handles into the main operation RPC to server.
1734 Severity : enhancement
1736 Description: Async OSC create to avoid the blocking unnecessarily.
1737 Details : If a OST has no remain object, system will block on the creating
1738 when need to create a new object on this OST. Now, ways use
1739 pre-created objects when available, instead of blocking on an
1740 empty osc while others are not empty. If we must block, we block
1741 for the shortest possible period of time.
1745 Description: improve handling recoverable errors
1746 Details : if request processig with error which can be recoverable on server
1747 request should be resend, otherwise page released from cache and
1750 Severity : enhancement
1752 Description: refine locking for avoid write wrong info into lov_objid file
1753 Details : fix possible races with add new target and write/update data in
1756 --------------------------------------------------------------------------------
1758 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1763 Frequency : on some architectures
1765 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1766 Details : Uninitialized flags sometimes cause configuration commands to
1770 Frequency : patchless clients only
1772 Description: missing __iget() symbol export
1773 Details : The __iget() symbol export is missing. To avoid the need for
1774 this on patchless clients the deathrow inode reaper is turned
1775 off, and we depend on the VM to clean up old inodes. This
1776 dependency was during via the fix for bug 12181.
1778 --------------------------------------------------------------------------------
1780 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1782 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1783 INTEROPERATE with older versions automatically. In many cases a
1784 special upgrade step is needed. Please read the
1785 user documentation before upgrading any part of a live system.
1786 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1787 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1788 * WARNING: Lustre configuration and startup changes are required with
1789 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1795 Severity : enhancement
1797 Description: MountConf
1798 Details : Lustre configuration is now managed via mkfs and mount
1799 commands instead of lmc and lconf. New obd types (MGS, MGC)
1800 are added for dynamic configuration management. See
1801 https://mail.clusterfs.com/wikis/lustre/MountConf for
1804 Severity : enhancement
1806 Description: dynamic OST addition
1807 Details : OSTs can now be added to a live filesystem
1809 Severity : enhancement
1811 Description: startup order invariance
1812 Details : MDTs and OSTs can be started in any order. Clients only
1813 require the MDT to complete startup.
1815 Severity : enhancement
1817 Description: parallel, asynchronous orphan cleanup
1818 Details : orphan cleanup is now performed in separate threads for each
1819 OST, allowing parallel non-blocking operation.
1821 Severity : enhancement
1823 Description: optimized stripe assignment
1824 Details : stripe assignments are now made based on ost space available,
1825 ost previous usage, and OSS previous usage, in order to try
1826 to optimize storage space and networking resources.
1828 Severity : enhancement
1830 Description: Permanently set tunables
1831 Details : All writable /proc/fs/lustre tunables can now be permanently
1832 set on a per-server basis, at mkfs time or on a live system.
1834 Severity : enhancement
1836 Description: Lustre message v2
1837 Details : Add lustre message format v2.
1839 Severity : enhancement
1841 Description: client OST exclusion list
1842 Details : Clients can be started with a list of OSTs that should be
1843 declared "inactive" for known non-responsive OSTs.
1847 Description: SPEC SFS validation failure on NFS v2 over lustre.
1848 Details : Changes the blocksize for regular files to be 2x RPC size,
1849 and not depend on stripe size.
1851 Severity : enhancement
1853 Description: Multiple MD RPCs in flight.
1854 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1855 To avoid overly-overloading MDS, introduce a limit on number of
1856 MDS RPCs in flight for a single client and add /proc controls
1857 to adjust this limit.
1859 Severity : enhancement
1861 Description: client read/write statistics
1862 Details : Add client read/write call usage stats for performance
1863 analysis of user processes.
1864 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1865 file access. extents_stats shows chunk size distribution.
1866 extents_stats_per_process show chunk size distribution per
1869 Severity : enhancement
1871 Description: mds statistics
1872 Details : Add detailed mds operations statistics in
1873 /proc/fs/lustre/mds/*/stats.
1877 Description: Failure of copying files with lustre special EAs.
1878 Details : Client side always return success for setxattr call for lustre
1879 special xattr (currently only "trusted.lov").
1884 Description: Refcount LNET uuids
1885 Details : The global LNET uuid list grew linearly with every startup;
1886 refcount repeated list entries instead of always adding to
1889 Severity : enhancement
1891 Description: Dynamic service threads
1892 Details : Within a small range, start extra service threads
1893 automatically when the request queue builds up.
1896 Frequency : mixed-endian client/server environments
1898 Description: mixed-endian crashes
1899 Details : The new msg_v2 system had some failures in mixed-endian
1902 Severity : enhancement
1904 Description: Easy OST removal
1905 Details : OSTs can be permanently deactivated with e.g. 'lctl
1906 conf_param lustre-OST0001.osc.active=0'
1908 Severity : enhancement
1910 Description: MGS proc entries
1911 Details : Added basic proc entries for the MGS showing what filesystems
1914 Severity : enhancement
1916 Description: provide MGS failover
1917 Details : Added config lock reacquisition after MGS server failover.
1919 Severity : enhancement
1921 Description: add Linux 2.4 support
1922 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1926 Description: a large application tries to do I/O to the same resource and dies
1927 in the middle of it.
1928 Details : Check the req->rq_arrival time after the call to
1929 ost_brw_lock_get(), but before we do anything about
1930 processing it & sending the BULK transfer request. This
1931 should help move old stale pending locks off the queue as
1932 quickly as obd_timeout.
1935 Frequency : when an incorrect nid is specified during startup
1937 Description: ptlrpc connect to non-existant node causes kernel crash
1938 Details : LNET can't be re-entered from an event callback, which
1939 happened when we expire a message after the export has been
1940 cleaned up. Instead, hand the zombie cleanup off to another
1943 Severity : enhancement
1945 Description: plain/inodebits lock performance improvement
1946 Details : Grouping plain/inodebits in granted list by their request modes
1947 and bits policy, thus improving the performance of search through
1951 Frequency : only if OST filesystem is corrupted
1953 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1954 Details : for a short time RPCs with bulk IO are in the replay list,
1955 but replay of bulk IOs is unimplemented. If the OST filesystem
1956 is corrupted due to disk cache incoherency and then replay is
1957 started it is possible to trip an assertion. Avoid putting
1958 committed RPCs into the replay list at all to avoid this issue.
1961 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1963 Description: System hang on startup
1964 Details : This bug allowed the liblustre (e.g. catamount) client to
1965 return to the app before handling all startup RPCs. This
1966 could leave the node unresponsive to lustre network traffic
1967 and manifested as a server ptllnd timeout.
1969 Severity : enhancement
1971 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1972 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1973 boolean to control whether to print peer debug info when a
1974 client's RPC times out.
1977 Frequency : only for kernels with patches from Lustre below 1.4.3
1979 Description: Remove old rdonly API
1980 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1983 Frequency : only for devices with external journals
1985 Description: Set external device read-only also
1986 Details : During a commanded failover stop, we set the disk device
1987 read-only while the server shuts down. We now also set any
1988 external journal device read-only at the same time.
1991 Frequency : when upgrading from 1.4 while trying to change parameters
1993 Description: The wrong (new) MDC name was used when setting parameters for
1994 upgraded MDT's. Also allows changing of OSC (and MDC)
1995 parameters if --writeconf is specified at tunefs upgrade time.
1998 Frequency : when setting specific ost indicies
2000 Description: QOS code breaks on skipped indicies
2001 Details : Add checks for missing OST indicies in the QOS code, so OSTs
2002 created with --index need not be sequential.
2004 Severity : enhancement
2006 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2007 Details : The uninit_groups feature works in conjunction with the kernel
2008 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2009 up the pass1 processing of e2fsck. This is a read-only feature
2010 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2011 mount filesystems that have had this feature enabled.
2013 Severity : enhancement
2015 Description: Improve multi-block allocation algorithm to avoid fragmentation
2016 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2017 allocation locality and avoid filesystem fragmentation.
2019 ------------------------------------------------------------------------------
2021 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
2023 * Support for kernels:
2024 2.4.21-47.0.1.EL (RHEL 3)
2025 2.6.5-7.283 (SLES 9)
2026 2.6.9-42.0.10.EL (RHEL 4)
2027 2.6.12.6 vanilla (kernel.org)
2028 2.6.16.27-0.9 (SLES 10)
2029 * Recommended e2fsprogs version: 1.39.cfs5
2031 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2035 Frequency : occasional, depends on client load and configuration
2036 Bugzilla : 12181, 12203
2037 Description: data loss for recently-modified files
2039 Details : In some cases it is possible that recently written or created
2040 files may not be written to disk in a timely manner (this should
2041 normally be within 30s unless client IO load is very high).
2042 The problem appears as zero-length files or files that are a
2043 multiple of 1MB in size after a client crash or client eviction
2044 that are missing data at the end of the file.
2046 This problem is more likely to be hit on clients where files are
2047 repeatedly created and unlinked in the same directory, clients
2048 have a large amount of RAM, have many CPUs, the filesystem has
2049 many OSTs, the clients are rebooted frequently, and/or the files
2050 are not accessed by other nodes after being written.
2052 The presence of the problem can be detected by looking at
2053 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2054 smaller than the second (nr_unused) then dirty files will not
2055 be flushed automatically to disk. "sync; sleep 10" should be
2056 run several times on the node before unmounting it to update
2057 Lustre (this is also safe to run on nodes without this problem).
2059 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2060 that can cause this same problem, so customers using that kernel
2061 also need to update the kernel in addition to Lustre. In order
2062 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2064 It is normal that files written just before a client crash (less
2065 than 30s) may not yet have been flushed to disk, even for local
2069 Frequency : frequent on thin XT3 nodes
2071 Description: UUID collision on thin XT3 Linux nodes
2072 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2073 randomly, since we relied on an insufficiently-seeded PRNG.
2078 Description: OSS hangs after "All ost request buffers busy"
2079 Details : A deadlock between quota and journal operations caused OSS
2080 hangs after printing "All ost request buffers busy."
2083 Frequency : always on liblustre builds
2085 Description: Cleanup compiler warnings on liblustre
2088 Frequency : always on liblustre builds on XT3
2090 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2094 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2095 > 32000 subdirectories
2096 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2097 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2098 subdirectory count crosses 32000. This will aid e2fsck to
2099 correctly handle more than 32000 subdirectories.
2102 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2104 Description: System hang on startup
2105 Details : This bug allowed the liblustre (e.g. catamount) client to
2106 return to the app before handling all startup RPCs. This
2107 could leave the node unresponsive to lustre network traffic
2108 and manifested as a server ptllnd timeout.
2110 Severity : enhancement
2112 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2113 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2114 boolean to control whether to print peer debug info when a
2115 client's RPC times out.
2120 Description: make O_SYNC working on 2.6 kernels
2121 Details : 2.6 kernels use different method for mark pages for write,
2122 so need add a code to lustre for O_SYNC work.
2127 Description: Failure to close file and release space on NFS
2128 Details : Put inode details into lock acquired in ll_intent_file_open.
2129 Use mdc_intent_lock in ll_intent_open to properly
2130 detect all kind of errors unhandled by mdc_enqueue.
2135 Description: proc file read during shutdown sometimes raced obd removal,
2137 Details : Add lock to prevent obd access after proc file removal.
2140 Frequency : Only for files larger than 4GB on 32-bit clients.
2142 Description: improperly doing page alignment of locks
2143 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2144 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2149 Description: RPCs being resent when they shouldn't be
2150 Details : Some RPCs that should not be resent are being resent. This
2151 can cause inconsistencies in the RPC state machine. Do not
2152 resend such requests.
2155 Frequency : rare, only with NFS export
2157 Description: Crash on NFS re-export node
2158 Details : under very unusual load conditions an assertion is hit in
2159 ll_intent_file_open()
2162 Frequency : only if OST filesystem is corrupted
2164 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2165 Details : for a short time RPCs with bulk IO are in the replay list,
2166 but replay of bulk IOs is unimplemented. If the OST filesystem
2167 is corrupted due to disk cache incoherency and then replay is
2168 started it is possible to trip an assertion. Avoid putting
2169 committed RPCs into the replay list at all to avoid this issue.
2174 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2175 Details : Large single O_DIRECT read and write calls can fail to allocate
2176 a sufficiently large buffer to process the request. In case of
2177 allocation failure the allocation is retried with a smaller
2178 buffer and broken into smaller requests.
2180 Severity : enhancement
2182 Description: Add -o localflock option to simulate old noflock behaviour.
2183 Details : This will achieve local-only flock/fcntl locks coherentness.
2188 Description: versioning check is incomplete
2189 Details : Checking the version difference of client vs. server, report
2190 error if the gap is too big.
2195 Description: add support PG_writeback bit
2196 Details : add support for PG_writeback bit for Lustre, for more carefull
2197 work with page cache in 2.6 kernel. This also fix some deadlocks
2198 and remove hack for work O_SYNC with 2.6 kernel.
2200 Severity : enhancement
2202 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2203 Details : The uninit_groups feature works in conjunction with the kernel
2204 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2205 up the pass1 processing of e2fsck. This is a read-only feature
2206 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2207 mount filesystems that have had this feature enabled.
2209 Severity : enhancement
2211 Description: Improve multi-block allocation algorithm to avoid fragmentation
2212 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2213 allocation locality and avoid filesystem fragmentation.
2215 ------------------------------------------------------------------------------
2217 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2219 * Support for kernels:
2220 2.6.9-42.0.3.EL (RHEL 4)
2221 2.6.5-7.276 (SLES 9)
2222 2.4.21-47.0.1.EL (RHEL 3)
2223 2.6.12.6 vanilla (kernel.org)
2224 2.6.16.21-0.8 (SLES10)
2225 * Recommended e2fsprogs version: 1.39.cfs2-0
2227 * The backwards-compatible /proc/sys/portals symlink has been removed
2228 in this release. Before upgrading, please ensure that you change
2229 any configuration scripts or /etc/sysctl.conf files that access
2230 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2231 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2232 in advance of the upgrade on any system running Lustre 1.4.6 or
2233 newer, since /proc/sys/lnet was added in that version.
2234 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2238 Frequency : only when quota is used
2240 Description: avoid scanning export list for quota master
2241 Details : Change the algorithms to avoid scanning export list in order
2242 to improve the efficiency.
2245 Frequency : MDS failover only, very rarely
2247 Description: "went back in time" messages on mds failover
2248 Details : The greatest transno may be lost when the current operation
2249 finishes with an error (transno==0) and the client's last_rcvd
2250 record is over-written. Save the greatest transno in the
2251 mds_last_transno for this case.
2254 Frequency : always for specific kernels and striping counts
2256 Description: client may get "Matching packet too big" without ACL support
2257 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2258 when trying to access files in certain configurations. The
2259 clients should in fact be denied when mounting because they do
2260 not understand ACLs.
2263 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2265 Description: many clients connecting with IO in progress causes connect timeouts
2266 Details : Avoid synchronous journal commits to avoid delays caused by many
2267 clients connecting/disconnecting when bulk IO is in progress.
2268 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2269 OST_IO_PORTAL to avoid delays behind potentially many pending
2273 Frequency : occasionally with multiple writers to a single file
2275 Description: shared writes to file may result in wrong size reported by stat()
2276 Details : Allow growing of kms when extent lock is cancelled
2279 Frequency : always with random mmap IO to multi-striped file
2281 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2282 Details : Only if the hole is at the end of OST object so that kms is too
2283 small. Fix is to increase kms accordingly in ll_nopage.
2286 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2288 Description: writes to a missing object would leak memory on the OST
2289 Details : If there is an inconsistency between the MDS and OST filesystems,
2290 such that the MDS references an object that doesn't exist, writes
2291 to that object will leak memory due to incorrect cleanup in the
2292 error handling path, eventually running out of memory on the OST.
2297 Description: Creating too long symlink causes lustre errors
2298 Details : Check symlink and name lengths before sending requests to MDS.
2301 Frequency : only if flock is enabled (not on by default)
2303 Description: posix locks not released on fd closure on 2.6.9+
2304 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2305 kernels, this caused such locks not to be released on fd close and
2306 then assertions on fs unmount about still used locks.
2309 Frequency : MDS failover only, very rarely
2311 Description: clients may get ASSERTION(granted_lock != NULL)
2312 Details : When request was taking a long time, and a client was resending
2313 a getattr by name lock request. The were multiple lock requests
2314 with the same client lock handle and
2315 mds_getattr_name->fixup_handle_for_resent_request found one of the
2316 lock handles but later failed with ASSERTION(granted_lock != NULL).
2321 Description: handle->h_buffer_credits > 0, assertion failure
2322 Details : h_buffer_credits is zero after truncate, causing assertion
2323 failure. This patch extends the transaction or creates a new
2327 Frequency : NFS re-export or patchless client
2328 Bugzilla : 11179, 10796
2329 Description: Crash on NFS re-export node (__d_move)
2330 Details : We do not want to hash the dentry if we don't have a lock.
2331 But if this dentry is later used in d_move, we'd hit uninitialised
2332 list head d_hash, so we just do this to init d_hash field but
2333 leave dentry unhashed.
2336 Frequency : NFS re-export or patchless client
2338 Description: NFS exports has problem with symbolic link
2339 Details : lustre client didn't properly install dentry when re-exported
2340 to NFS or running patchless client.
2343 Frequency : NFS re-export or patchless client
2345 Description: Various nfs/patchless fixes.
2346 Details : fixes reuse disconected alias for lookup process - this fixes
2347 warning "find_exported_dentry: npd != pd",
2348 fix permission error with open files at nfs.
2349 fix apply umask when do revalidate.
2352 Frequency : occasional
2354 Description: Crash on NFS re-export node
2355 Details : calling clear_page() on the wrong pointer triggered oops in
2356 generic_mapping_read().
2359 Frequency : rarely, using O_DIRECT IO
2361 Description: unaligned directio crashes client with LASSERT
2362 Details : check for unaligned buffers before trying any requests.
2365 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2367 Description: stale data returned from RAID cache
2368 Details : If only a small amount of IO is done to the RAID device before
2369 reading it again it is possible to get stale data from the RAID
2370 cache instead of reading it from disk.
2373 Frequency : always for sles10 kernel
2375 Description: sles10 support
2376 Details : ll_follow_link: compile fixes and using of nd_set_link
2377 under newer kernels.
2380 Frequency : depends on arch, kernel and compiler version, always on sles10
2383 Description: recursive or deep enough symlinks cause stack overflow
2384 Details : getting rid of large stack-allocated variable in
2388 Frequency : depends on hardware
2390 Description: lustre write performance loss in the SLES10 kernel
2391 Details : the performance loss is caused by using of write barriers in the
2392 ext3 code. The SLES10 kernel turns barrier support on by
2393 default. The fix is to undo that change for ldiskfs.
2395 ------------------------------------------------------------------------------
2397 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2399 * Support for kernels:
2400 2.6.9-42.0.3EL (RHEL 4)
2401 2.6.5-7.276 (SLES 9)
2402 2.4.21-47.0.1.EL (RHEL 3)
2403 2.6.12.6 vanilla (kernel.org)
2407 Frequency : quota enabled and large files being deleted
2409 Description: releasing more than 4GB of quota at once hangs OST
2410 Details : If a user deletes more than 4GB of files on a single OST it
2411 will cause the OST to spin in an infinite loop. Release
2412 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2417 Description: statfs data retrieved from /proc may be stale or zero
2418 Details : When reading per-device statfs data from /proc, in the
2419 {kbytes,files}_{total,free,avail} files, it may appear
2420 as zero or be out of date.
2423 Frequency : systems with MD RAID1 external journal devices
2425 Description: lconf's call to blkid is confused by RAID1 journal devices
2426 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2427 of returning all block devices that match the journal UUID.
2430 Frequency : always, for aggregate stripe size over 4GB
2432 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2433 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2434 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2437 Frequency : always if "lfs find" used on a local file/directory
2439 Description: "lfs find" segfaults if used on a local file/directory
2440 Details : The case where a directory component was not specified wasn't
2441 handled correctly. Handle this properly.
2444 Frequency : always on ppc64
2446 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2447 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2448 perform endianness conversion.
2451 Frequency : rarely (truncate to non-zero file size after write under load)
2452 Bugzilla : 10730, 10687
2453 Description: Files padded with zeros to next 4K multiple
2454 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2455 it is possible that files that are truncated to a non-zero size
2456 immediately after being written are filled with zero bytes beyond
2457 the truncated size. No file data is lost.
2459 Severity : enhancement
2461 Description: Allow recovery/failover for liblustre clients.
2462 Details : liblustre clients were unaware of failover configurations until
2465 Severity : enhancement
2467 Description: user file locks should fail when not mounting with flock option
2468 Details : Set up an error-returning stub in ll_file_operations.lock field
2469 to prevent incorrect behaviour when client is mounted without
2470 flock option. Also, set up properly f_op->flock field for
2474 Frequency : always on ia64
2476 Description: "lfs df" loops on printing out MDS statfs information
2477 Details : The obd_ioctl_data was not initialized and in some systems
2478 this caused a failure during the ioctl that did not return
2479 an error. Initialize the struct and return an error on failure.
2482 Frequency : SLES 9 only
2484 Description: Error of copying files with lustre special EAs as root
2485 Details : Client side always return success for setxattr call for lustre
2486 special xattr (currently only "trusted.lov").
2489 Frequency : rarely on clusters with both ia64+i386 clients
2491 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2492 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2493 clients are doing concurrent writes to the same file it is
2494 possible that the ia64 clients may LASSERT because the OST
2495 extent locks are not PAGE_SIZE aligned. Ensure that grown
2496 locks are always aligned on the request boundary.
2499 Frequency : specific use, occasional
2501 Description: Overwriting in use executable truncates on-disk binary image
2502 Details : If one node attempts to overwrite an executable in use by
2503 another node, we now correctly return ETXTBSY instead of
2504 truncating the file.
2506 Severity : enhancement
2508 Description: Async OSC create to avoid the blocking unnecessarily.
2509 Details : If a OST has no remain object, system will block on the creating
2510 when need to create a new object on this OST. Now, ways use
2511 pre-created objects when available, instead of blocking on an
2512 empty osc while others are not empty. If we must block, we block
2513 for the shortest possible period of time.
2518 Description: chmod on Lustre root is propagated to other clients
2519 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2520 invalid by the follow_mount time.
2523 Frequency : liblustre clients only
2525 Description: Race in 'instant cancel' lock handling could lead to such locks
2526 never to be granted in case of SMP MDS
2527 Details : Do not destroy not yet granted but cbpending locks in
2531 Frequency : replay/resend of open
2533 Description: non null lock assetion failure in mds_intent_policy
2534 Details : Trying to replay/resend lockless open requests resulted in
2535 mds_open() returning 0 with no lock. Now it sets a flag if
2536 a lock is going to be returned.
2538 Severity : enhancement
2540 Description: Checksum enhancements
2541 Details : New checksum enhancements allow for resending RPCs that failed
2544 Severity : enhancement
2546 Description: Tunables on number of dirty pages in cacche
2547 Details : Allow to set limit on number of dirty pages cached.
2552 Description: client crash on unmount - lock still has references
2553 Details : In some error handling cases it was possible to leak a lock
2554 reference on a client while accessing a file. This was not
2555 harmful to the client during operation, but would cause the
2556 client to crash when the filesystem is unmounted.
2559 Frequency : specific case, rare
2561 Description: ETXTBSY on mds though file not in use by client
2562 Details : ETXTBSY is no longer incorrectly returned when attempting to
2563 chmod or chown a directory that the user previously tried to
2564 execute or a currently-executing binary.
2567 Frequency : extremely rare except on liblustre-based clients
2569 Description: Lustre space not freed when files are deleted
2570 Details : Clean up open-unlinked files after client eviction. Previously
2571 the unlink was skipped and the files remained as orphans.
2576 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2577 Details : In some cases it was possible to send a blocking callback to
2578 a client doing a glimpse, even though that client didn't get
2579 a lock granted. When the glimpse lock is cancelled on the OST
2580 the freed lock is left on the waiting list and corrupted the list.
2583 Frequency : all core dumps
2585 Description: Broke core dumps to lustre
2586 Details : Negative dentry may be unhashed if parent does not have UPDATE
2587 lock, but some callers, e.g. do_coredump, expect dentry to be
2588 hashed after successful create, hash it in ll_create_it.
2590 ------------------------------------------------------------------------------
2592 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2594 * Support for kernels:
2595 2.6.9-42.0.2.EL (RHEL 4)
2596 2.6.5-7.276 (SLES 9)
2597 2.4.21-40.EL (RHEL 3)
2598 2.6.12.6 vanilla (kernel.org)
2602 Frequency : always on RHEL 3
2604 Description: Number of open files grows over time
2605 Details : The number of open files grows over time, whether or not
2606 Lustre is started. This was due to a filp leak introduced
2607 by one of our kernel patches.
2609 ------------------------------------------------------------------------------
2611 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2613 * Support for kernels:
2614 2.6.9-42.EL (RHEL 4)
2615 2.6.5-7.276 (SLES 9)
2616 2.4.21-40.EL (RHEL 3)
2617 2.6.12.6 vanilla (kernel.org)
2622 Bugzilla : 5719, 9635, 9792, 9684
2623 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2624 Details : If a server is under heavy load and cannot reply to new
2625 connection requests before the client resends the (re)connect,
2626 the connection handling code can behave badly if two service
2627 threads are concurrently handing separate (re)connections from
2628 the same client. Add better locking to the connection handling
2629 code, and ensure that only a single connection will be processed
2630 for a given client UUID, even if the lock is dropped.
2632 Severity : enhancement
2634 Description: add TCP zero-copy support to kernel
2635 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2636 sends if the hardware supports scatter-gather and checksumming.
2637 This allows socklnd to do client-write and server-read more
2638 efficiently and reduce CPU utilization from skbuf copying.
2641 Frequency : only if NFS exporting from client
2643 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2644 Details : If setting attributes on a file created under NFS that had
2645 never been opened it would be possible to oops the client
2646 if the file had no objects.
2649 Frequency : always for liblustre
2651 Description: liblustre client does MDS+OSTs setattr RPC for each write
2652 Details : When doing a write from a liblustre client, the client
2653 incorrectly issued an RPC to the MDS and each OST the file was
2654 striped over in order to update the timestamps. When writing
2655 with small chunks and many clients this could overwhelm the MDS
2656 with RPCs. In all cases it would slow down the write because
2657 these RPCs are unnecessary.
2659 Severity : enhancement
2661 Description: allow number of MDS service threads to be changed at module load
2662 Details : It is now possible to change the number of MDS service threads
2663 running. Adding "options mds mds_num_threads={N}" to the MDS's
2664 /etc/modprobe.conf will set the number of threads for the next
2665 time Lustre is restarted (assuming the "mds" module is also
2666 reloaded at that time). The default number of threads will
2667 stay the same, 32 for most systems.
2672 Description: OST crash if filesystem is unformatted or corrupt
2673 Details : If an OST is started on a device that has never been formatted
2674 or if the filesystem is corrupt and cannot even mount then the
2675 error handling cleanup routines would dereference a NULL pointer.
2680 Description: NULL pointer deref in llap_from_page.
2681 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2682 incorrect) mapping if the page was truncated/reclaimed while it was
2683 searched for. Check for this condition and skip such pages when
2684 doing readahead. Introduce extra check to llap_from_page() to
2685 verify page->mapping->host is non-NULL (so page is not anonymous).
2688 Frequency : Sometimes when using sys_sendfile
2690 Description: "page not covered by a lock" warnings from ll_readpage
2691 Details : sendfile called ll_readpage without right page locks present.
2692 Now we introduced ll_file_sendfile that does necessary locking
2693 around call to generic_file_sendfile() much like we do in
2697 Frequency : with certain MDS communication failures at client mount time
2699 Description: NULL pointer deref after failed client mount
2700 Details : a client connection request may delayed by the network layer
2701 and not be sent until after the PTLRPC layer has timed out the
2702 request. If the client fails the mount immediately it will try
2703 to clean up before the network times out the request. Add a
2704 reference from the request import to the obd device and delay
2705 the cleanup until the network drops the request.
2708 Frequency : occasionally during client (re)connect
2710 Description: assertion failure during client (re)connect
2711 Details : processing a client connection request may be delayed by the
2712 client or server longer than the client connect timeout. This
2713 causes the client to resend the connection request. If the
2714 original connection request is replied in this interval, the
2715 client may trip an assertion failure in ptlrpc_connect_interpret()
2716 which thought it would be the only running connect process.
2719 Frequency : only with obd_echo servers and clients that are rebooted
2721 Description: kernel BUG accessing uninitialized data structure
2722 Details : When running an obd_echo server it did not start the ping_evictor
2723 thread, and when a client was evicted an uninitialized data
2724 structure was accessed. Start the ping_evictor in the RPC
2725 service startup instead of the OBD startup.
2727 Severity : enhancement
2728 Bugzilla : 10193 (patchless)
2729 Description: Remove dependency on various unexported kernel interfaces.
2730 Details : No longer need reparent_to_init, exit_mm, exit_files,
2731 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2734 Frequency : rare (only users of deprecated and unsupported LDAP config)
2736 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2737 Details : LDAP apparently contains 'lustreName' attributes instead of
2738 'name'. A simple remapping of the name is sufficient.
2741 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2743 Description: waiting_locks_callback trips kernel BUG if client is evicted
2744 Details : Running with the dump_on_timeout debug flag turned on makes
2745 it possible that the waiting_locks_callback() can try to dump
2746 the Lustre kernel debug logs from an interrupt handler. Defer
2747 this log dumping to the expired_lock_main() thread.
2749 Severity : enhancement
2751 Description: Support NFS exporting on 2.6 kernels.
2752 Details : Implement non-rawops metadata methods for NFS server to use without
2753 changing NFS server code.
2756 Frequency : very rare (synthetic metadata workload only)
2758 Description: two racing renames might cause an MDS thread to deadlock
2759 Details : Running the "racer" program may cause one MDS thread to rename
2760 a file from being the source of a rename to being the target of
2761 a rename at exactly the same time that another thread is doing
2762 so, and the second thread has already enqueued these locks after
2763 doing a lookup of the target and is trying to relock them in
2764 order. Ensure that we don't try to re-lock the same resource.
2767 Frequency : only very large systems with liblustre clients
2769 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2770 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2771 all OSTs in parallel. This allows the eviction of stale liblustre
2772 clients to proceed much faster than if they were done in series,
2773 and also offers similar improvements for other set_info RPCs.
2778 Description: excessive CPU usage during initial read phase on client
2779 Details : During the initial read phase on a client, it would agressively
2780 retry readahead on the file, consuming too much CPU and impacting
2781 performance (since 1.4.5.8). Improve the readahead algorithm
2782 to avoid this, and also improve some other common cases (read
2783 of small files in particular, where "small" is files smaller than
2784 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2789 Description: MDS crash when receiving packet with unknown intent.
2790 Details : Do not LBUG in unknown intent case, just return -EFAULT
2792 Severity : enhancement
2793 Bugzilla : 9293, 9385
2794 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2795 Details : Do not serialize getattr (non-intent version) and statfs.
2798 Frequency : occasional, when OST network is overloaded/intermittent
2800 Description: client evicted by OST after bulk IO timeout
2801 Details : If a client sends a bulk IO request (read or write) the OST
2802 may evict the client if it is unresposive to its data GET/PUT
2803 request. This is incorrect if the network is overloaded (takes
2804 too long to transfer the RPC data) or dropped the OST GET/PUT
2805 request. There is no need to evict the client at all, since
2806 the pinger and/or lock callbacks will handle this, and the
2807 client can restart the bulk request.
2810 Frequency : Always when mmapping file with no objects
2812 Description: client crashes when mmapping file with no objects
2813 Details : Check that we actually have objects in a file before doing any
2814 operations on objects in ll_vm_open, ll_vm_close and
2820 Description: Request leak when working with deleted CWD
2821 Details : Introduce advanced request refcount tracking for requests
2822 referenced from lustre intent.
2824 Severity : Enhancement
2826 Description: Cache open file handles on client.
2827 Details : MDS now will return special lock along with openhandle, if
2828 requested and client is allowed to hold openhandle, even if unused,
2829 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2830 closing files for every read/write openration.
2832 Severity : Enhancement
2834 Description: Cache open negative dentries on client when possible.
2835 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2836 negative dentries on lock revocation.
2841 Description: Remounting a client read-only wasn't possible with a zconf mount
2842 Details : It wasn't possible to remount a client read-only with llmount.
2844 Severity : enhancement
2845 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2846 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2847 MPICH is included to add Lustre-specific ADIO interfaces.
2848 This is based closely on the UFS ADIO layer and only differs
2849 in file creation, in order to allow the OST striping to be set.
2850 This is user-contributed code and not supported by CFS.
2855 Description: extended inode attributes (immutable, append-only) work improperly
2856 when 2.4 and 2.6 kernels are used on client/server or vice versa
2857 Details : Introduce kernel-independent values for these flags.
2859 Severity : enhancement
2862 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2863 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2864 granularity. Now, Fractional MB granularity is be supported,
2865 this is very useful for low memory system.
2867 Severity : enhancement
2869 Description: Getattr by fid
2870 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2871 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2872 have locks and avoids getattr rpc requests for it on every stat.
2875 Frequency : Always, for filesystems larger than 2TB
2877 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2878 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2879 filesystems larger than 2TB to be mounted with mballoc enabled.
2882 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2884 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2885 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2886 filesystems larger than 2TB is mounted then the kernel will
2887 silently corrupt the start of the filesystem. CONFIG_LBD is
2888 enabled for all CFS-supported kernels, but the possibility of
2889 this happening with a modified kernel config exists.
2891 Severity : enhancement
2893 Description: add client O_DIRECT support for 2.6 kernels
2894 Details : It is now possible to do O_DIRECT reads and writes to files
2895 in the Lustre client mountpoint on 2.6 kernel clients.
2897 Severity : enhancement
2899 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2900 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2901 parallel, not waiting for response from every OST before sending
2902 a rpc to the next OST.
2907 Description: setattr vs write race when updating file timestamps
2908 Details : Client processes that update a file timestamp into the past
2909 right after writing to the file (e.g. tar) it is possible that
2910 the updated file modification time can be reset to the current
2911 time due to a race between processing the setattr and write RPC.
2913 Severity : enhancement
2915 Description: Bring 'lfs find' closer in line with regular Linux find.
2916 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2917 -print0 options and obtains all the needed info through the lustre
2920 Severity : enhancement
2922 Description: support up to 1024 configured devices on one node
2923 Details : change obd_dev array from statically allocated to dynamically
2924 allocated structs as they are first used to reduce memory usage
2929 Description: Flush dirty partially truncated pages during truncate
2930 Details : Immediatelly flush partially truncated pages in filter_setattr,
2931 this way we completely avoid having any pages in page cache on OST
2932 and can retire ugly workarounds during writes to flush such pages.
2937 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2938 Details : protect inode from truncation within vfs_unlink() context
2939 just take a reference before calling vfs_unlink() and release it
2940 when parent's i_sem is free.
2945 Description: last_id value checked outside lock on OST caused LASSERT failure
2946 Details : If there were multiple MDS->OST object precreate requests in
2947 flight, it was possible that the OST's last object id was checked
2948 outside a lock and incorrectly tripped an assertion. Move checks
2949 inside locks, and discard old precreate requests.
2952 Frequency : always, if extents are used on OSTs
2954 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2955 Details : OSTs using the extents format would not zero the high 16 bits of
2956 the index physical block number. This is not a problem for any
2957 OST filesystems smaller than 16TB, and no kernels support ext3
2958 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2959 new/modified files) and can be fixed for existing filesystems
2960 with e2fsprogs-1.39-cfs1.
2965 Description: import connection selection may be incorrect if timer wraps
2966 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2967 import connections to be ignored if the 32-bit jiffies counter
2968 wraps. Use a 64-bit jiffies counter.
2971 Frequency : very large clusters immediately after boot
2973 Description: LNET request buffers exhausted under heavy short-term load
2974 Details : If a large number of client requests are generated on a service
2975 that has previously never seen so many requests it is possible
2976 that the request buffer growth cannot keep up with the spike in
2977 demand. Instead of dropping incoming requests, they are held in
2978 the LND until the RPC service can accept more requests.
2981 Frequency : Sometimes during replay
2983 Description: Assertion failure in ll_local_open after replay.
2984 Details : If replay happened on an open request reply before we were able
2985 to set replay handler, reply will become not swabbed tripping the
2986 assertion in ll_local_open. Now we set the handler right after
2987 recognising of open request
2990 Frequency : very rare
2992 Description: kernel reports "badness in vsnprintf"
2993 Details : Reading from the "recovery_status" /proc file in small chunks
2994 may cause a negative length in lprocfs_obd_rd_recovery_status()
2995 call to vsnprintf() (which is otherwise harmless). Exit early
2996 if there is no more space in the output buffer.
2998 Severity : enhancement
3000 Description: clear OBD RPC statistics by writing to them
3001 Details : It is now possible to clear the OBD RPC statistics by writing
3002 to the "stats" file.
3007 Description: Client mtime is not the same on different clients after utimes
3008 Details : In some cases, the client was using the utimes() syscall on
3009 a file cached on another node. The clients now validate the
3010 ctime from the MDS + OSTs to determine which one is right.
3015 Description: Inability to activate failout mode
3016 Details : lconf script incorrectly assumed that in pythong string's numeric
3017 value is used in comparisons.
3020 Frequency : always with multiple stripes per file
3022 Description: Inefficient object allocation for mutli-stripe files
3023 Details : When selecting which OSTs to stripe files over, for files with
3024 a stripe count that divides evenly into the number of OSTs,
3025 the MDS is always picking the same starting OST for each file.
3026 Return the OST selection heuristic to the original design.
3031 Description: mount failures may take full timeout to return an error
3032 Details : Under some heavy load conditions it is possible that a
3033 failed mount can wait for the full obd_timeout interval,
3034 possibly several minutes, before reporting an error.
3035 Instead return an error as soon as the status is known.
3037 Frequency : quota enabled and large files being deleted
3039 Description: releasing more than 4GB of quota at once hangs OST
3040 Details : If a user deletes more than 4GB of files on a single OST it
3041 will cause the OST to spin in an infinite loop. Release
3042 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3047 Description: statfs data retrieved from /proc may be stale or zero
3048 Details : When reading per-device statfs data from /proc, in the
3049 {kbytes,files}_{total,free,avail} files, it may appear
3050 as zero or be out of date.
3053 Frequency : systems with MD RAID1 external journal devices
3055 Description: lconf's call to blkid is confused by RAID1 journal devices
3056 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3057 of returning all block devices that match the journal UUID.
3060 Frequency : always, for aggregate stripe size over 4GB
3062 Description: assertion fails when trying to use 4GB stripe size
3063 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
3064 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
3067 Frequency : always on ppc64
3069 Description: the first write on an ext3 filesystem with mballoc got stuck
3070 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3071 perform endianness conversion.
3073 ------------------------------------------------------------------------------
3075 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3077 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3078 INTEROPERATE with older versions automatically. Please read the
3079 user documentation before upgrading any part of a live system.
3080 * WARNING: Lustre networking configuration changes are required with
3081 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3084 * Support for kernels:
3085 2.6.9-22.0.2.EL (RHEL 4)
3086 2.6.5-7.244 (SLES 9)
3087 2.6.12.6 vanilla (kernel.org)
3090 Severity : enhancement
3091 Bugzilla : 7981/8208
3092 Description: Introduced Lustre Networking (LNET)
3093 Details : LNET is new networking infrastructure for Lustre, it includes
3094 a reorganized network configuration mode (see the user
3095 documentation for full details) as well as support for routing
3096 between different network fabrics. Lustre Networking Devices
3097 (LNDs) for the supported network fabrics have also been
3098 created for this new infrastructure.
3100 Severity : enhancement
3101 Description: Introduced Access control lists
3102 Details : clients can set ACLs on files and directories in order to have
3103 more fine-grained permissions than the standard Unix UGO+RWX.
3104 The MDS must be started with the "-o acl" mount option.
3106 Severity : enhancement
3107 Description: Introduced filesystem quotas
3108 Details : Administrators may now establish per-user quotas on the
3111 Severity : enhancement
3113 Description: Configuration change for the XT3
3114 The PTLLND is now used to run Lustre over Portals on the XT3
3115 The configure option(s) --with-cray-portals are no longer used.
3116 Rather --with-portals=<path-to-portals-includes> is used to
3117 enable building on the XT3. In addition to enable XT3 specific
3118 features the option --enable-cray-xt3 must be used.
3123 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3124 Details : A race between allocating a new llcd and re-getting the llcd_lock
3125 allowed another thread to grab newly-allocated llcd.
3127 Severity : enhancement
3129 Description: 2.6 OST async journal commit and locking fix to improve performance
3130 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3131 2.6 kernels are now async as they already were in 2.4 kernels so
3132 that they can commit concurrently with the network bulk transfer.
3133 For block-allocated files the filter allocation semaphore is held
3134 to avoid filesystem fragmentation during allocation. BKL lock
3135 removed for 2.6 xattr operations where it is no longer needed.
3140 Description: lconf incorrectly determined whether two IP networks could talk
3141 Details : In some more complicated routing and multiple-network
3142 configurations, lconf will avoid trying to make a network
3143 connection to a disjoint part of the IP space. It was doing the
3144 math incorrectly for one set of cases.
3149 Description: Fix for potential infinite loop processing records in an llog.
3150 Details : If an llog record is corrupted/zeroed, it is possible to loop
3151 forever in llog_process(). Validate the llog record length
3152 and skip the remainder of the block on error.
3155 Frequency : occasional (liblustre only)
3157 Description: liblustre could not open files whose last component is a symlink
3158 Details : sysio_path_walk() would incorrectly pass the open intent to
3159 intermediate path components.
3162 Frequency : rare (liblustre only with non-standard tuning)
3163 Bugzilla : 7201 (7350)
3164 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3165 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3166 after it has been released, only do this for referenced locks
3168 Severity : enhancement
3170 Description: specifying an (invalid) directory default stripe_size of -1
3171 would reset the directory default striping
3172 Details : stripe_size -1 was used internally to signal directory stripe
3173 removal, now use "all default" to signal dir stripe removal
3174 as a directory striping of "all default" is not useful
3177 Frequency : common for large clusters running liblustre clients
3179 Description: doing an ls when liblustre clients are running is slow
3180 Details : sending a glimpse AST to a liblustre client waits for every AST
3181 to time out, as liblustre clients will not respond. Since they
3182 cannot cache data we refresh the OST lock LVB from disk instead.
3184 Severity : enhancement
3186 Description: doing an ls at the same time as file IO can be slow
3187 Details : enqueue and other "small" requests can be blocked behind many
3188 large IO requests. Create a new OST IO portal for non-IO
3189 requests so they can be processed faster.
3192 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3194 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3195 created file with mode 000
3196 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3197 allow an ftruncate() as a non-root user to a file with mode 000.
3198 Lustre now respects this flag to disable mode checks when
3199 truncating a file owned by the user
3202 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3205 Description: Revoking locks from clients that went dead or catatonic might take
3207 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3208 cancellation of such locks instant on servers without waiting for
3209 any reply from clients. Clients drops these locks when cancel
3210 notification from server is received without replying.
3213 Frequency : liblustre-only, when liblustre client dies or becomes busy
3215 Description: Doing ls on Linux clients can take a long time with active
3217 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3218 granting such locks to it in the first place if possible. Locks
3219 are taken by proxy on the OST during the read or write and
3220 dropped immediately afterward. Add connect flags handling, do
3221 not grant locks to liblustre clients for glimpse ASTs.
3223 Severity : enhancement
3225 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3226 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3227 sequential reads and behaves badly with applications doing random
3228 reads. Improve it by reading ahead at least read region, and
3229 avoiding excessive large RPC for small reads.
3231 Severity : enhancement
3233 Description: Creating more than 1000 files for a single job may cause a load
3234 imbalance on the OSTs if there are also a large number of OSTs.
3235 Details : qos_prep_create() uses an OST index reseed value that is an
3236 even multiple of the number of available OSTs so that if the
3237 reseed happens in the middle of the object allocation it will
3238 still utilize the OSTs as uniformly as possible.
3243 Description: OST or MDS may oops in ping_evictor_main()
3244 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3245 but doesn't restart at beginning of obd_exports_timed list
3248 Severity : enhancement
3250 Description: improve by-nid export eviction on the MDS and OST
3251 Details : allow multiple exports with the same NID to be evicted at one
3252 time without re-searching the exports list.
3255 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3257 Description: MDS may oops in groups_free()
3258 Details : in rare race conditions a newly allocated group_info struct is
3259 freed again, and this can be NULL. The 2.4 compatibility code
3260 for groups_free() checked for a NULL pointer, but 2.6 did not.
3263 Frequency : common for liblustre clients doing little filesystem IO
3264 Bugzilla : 9352, 7313
3265 Description: server may evict liblustre clients accessing contended locks
3266 Details : if a client is granted a lock or receives a completion AST
3267 with a blocking AST already set it would not reply to the AST
3268 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3271 Frequency : lfs setstripe, only systems with more than 160 OSTs
3273 Description: unable to set striping with a starting offset beyond OST 160
3274 Details : llapi_create_file() incorrectly limited the starting stripe
3275 index to the maximum single-file stripe count.
3278 Frequency : LDAP users only
3280 Description: lconf did not handle in-kernel recovery with LDAP properly
3281 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3283 Severity : enhancement
3285 Description: bind OST threads to NUMA nodes to improve performance
3286 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3287 node and do their allocations there to localize memory access
3289 Severity : enhancement
3291 Description: llmount can determine client NID directly from Myrinet (GM)
3292 Details : the client NID code from gmnalnid was moved directly into
3293 llmount, removing the need to use this or specifying the
3294 client NID explicitly when mounting GM clients with zeroconf
3297 Frequency : if client is started with down MDS
3299 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3300 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3301 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3302 ptlrpc_import_delayed_req() because we didn't check if the
3303 request was interrupted, and we also didn't break out of the
3304 event loop if there was no timeout
3309 Description: data loss during non-page-aligned writes to a single file from
3310 both multiple nodes and multiple threads on one node at same time
3311 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3312 inconsistency led to false short-reads, that were cached and later
3313 used by ->prepare_write() to fill in partially written page,
3314 leading to data loss.
3317 Frequency : always, if lconf --abort_recovery used
3319 Description: lconf --abort_recovery fails with 'Operation not supported'
3320 Details : lconf was attempting to abort recovery on the MDT device and not
3323 Severity : enhancement
3325 Description: remove cleanup logs
3326 Details : replace lconf-generated cleanup logs with lustre internal
3327 cleanup routines. Eliminates the need for client-cleanup and
3330 Severity : enhancement
3332 Description: add support for EAs (user and system) on lustre filesystems
3333 Details : it is now possible to store extended attributes in the Lustre
3334 client filesystem, and with the user_xattr mount option it
3335 is possible to allow users to store EAs on their files also
3337 Severity : enhancement
3339 Description: Add possibility (config option) to show minimal available OST free
3341 Details : When compiled with --enable-mindf configure option, statfs(2)
3342 (and so, df) will return least minimal free space available from
3343 all OSTs as amount of free space on FS, instead of summary of
3344 free spaces of all OSTs.
3346 Severity : enhancement
3348 Description: do not expand extent locks acquired on OST-side
3349 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3350 by server: they are not cached anyway.
3353 Frequency : when mmap is used/binaries executed from Lustre
3355 Description: Unmmap pages before throwing them away from read cache.
3356 Details : llap_shrink cache now attempts to unmap pages before discarding
3357 them (if unmapping failed - do not discard). SLES9 kernel has
3358 extra checks that trigger if this unmapping is not done first.
3363 Description: lconf didn't resolve symlinks before checking to see whether a
3364 given mountpoint was already in use
3367 Frequency : when migrating failover services
3368 Bugzilla : 6395, 9514
3369 Description: When migrating a subset of services from a node (e.g. failback
3370 from a failover service node) the remaining services would
3371 time out and evict clients.
3372 Details : lconf --force (implied by --failover) sets the global obd_timeout
3373 to 5 seconds in order to quickly disconnect, but this caused
3374 other RPCs to time out too quickly. Do not change the global
3375 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3377 Severity : enhancement
3378 Frequency : if MDS is started with down OST
3379 Bugzilla : 9439,5706
3380 Description: Allow startup/shutdown of an MDS without depending on the
3381 availability of the OSTs.
3382 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3383 Add appropriate locking and lov-osc refcounts for safe
3384 cleaning. Add osc abort_inflight calls in case the
3385 synchronize never started.
3388 Frequency : occasional (Cray XT3 only)
3390 Description: root not authorized to access files in CRAY_PORTALS environment
3391 Details : The client process capabilities were not honoured on the MDS in
3392 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3393 been accessed by an authorized user then root was able to access
3394 the file on the local client also. The root user capabilities
3395 are now allowed on the MDS, as this environment has secure UID.
3398 Frequency : occasional
3400 Description: ldiskfs "too long searching" message happens too often
3401 Details : A debugging message (otherwise harmless) prints too often on
3402 the OST console. This has been reduced to only happen when
3403 there are fragmentation problems on the filesystem.
3408 Description: Division by zero in statfs when all OSCs are inactive
3409 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3410 lov_statfs divides by value returned by lov_get_stripecnt().
3414 Bugzilla : 9489, 3273
3415 Description: First write from each client to each OST was only 4kB in size,
3416 to initialize client writeback cache, which caused sub-optimal
3417 RPCs and poor layout on disk for the first writen file.
3418 Details : Clients now request an initial cache grant at (re)connect time
3419 and so that they can start streaming writes to the cache right
3420 away and always do full-sized RPCs if there is enough data.
3421 If the OST is rebooted the client also re-establishes its grant
3422 so that client cached writes will be honoured under the grant.
3427 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3428 Details : Now I/O RPCs go to different portal number and (presumably) fast
3429 lock requests (and glimses) and other RPCs get their own service
3430 threads pool that should be able to service those RPCs
3433 Severity : enhancement
3435 Description: Ability to exchange lustre version between client and servers and
3436 issue warnings at client side if client is too old. Also for
3437 liblustre clients there is ability to refuse connection of too old
3439 Details : New 'version' field is added to connect data structure that is
3440 filled with version info. That info is later checked by server and
3444 Frequency : rare, liblustre only.
3445 Bugzilla : 9296, 9581
3446 Description: Two simultaneous writes from liblustre at offset within same page
3447 might proceed at the same time overwriting eachother with stale
3449 Details : I/O lock withing llu_file_prwv was released too early, before data
3450 actually was hitting the wire. Extended lock-holding time until
3451 server acknowledges receiving data.
3454 Frequency : extremely rare. Never observed in practice.
3456 Description: avoid generating lustre_handle cookie of 0.
3457 Details : class_handle_hash() generates handle cookies by incrementing
3458 global counter, and can hit 0 occasionaly (this is unlikely, but
3459 not impossible, because initial value of cookie counter is
3460 selected randonly). Value of 0 is used as a sentinel meaning
3461 "unassigned handle" --- avoid it. Also coalesce two critical
3462 sections in this function into one.
3464 Severity : enhancement
3466 Description: allow liblustre clients to delegate truncate locking to OST
3467 Details : To avoid overhead of locking, liblustre client instructs OST to
3468 take extent lock in ost_punch() on client's behalf. New connection
3469 flag is added to handle backward compatibility.
3471 Severity : enhancement
3472 Bugzilla : 4928, 7341, 9758
3473 Description: allow number of OST service threads to be specified
3474 Details : a module parameter allows the number of OST service threads
3475 to be specified via "options ost ost_num_threads={N}" in the
3476 OSS's /etc/modules.conf or /etc/modprobe.conf.
3480 Bugzilla : 6146, 9635, 9895
3481 Description: servers crash with bad pointer in target_handle_connect()
3482 Details : In rare cases when a client is reconnecting it was possible that
3483 the connection request was the last reference for that export.
3484 We would temporarily drop the export reference and get a new
3485 one, but this may have been the last reference and the export
3486 was just destroyed. Get new reference before dropping old one.
3488 Severity : enhancement
3489 Frequency : if client is started with failover MDS
3491 Description: Allow multiple MDS hostnames in the mount command
3492 Details : Try to read the configuration from all specified MDS
3493 hostnames during a client mount in case the "primary"
3496 Severity : enhancement
3498 Description: Stop sending data to evicted clients as soon as possible.
3499 Details : Check if the client we are about to send or are sending data to
3500 was evicted already. (Check is done every second of waiting,
3501 for which l_wait_event interface was extended to allow checking
3502 of exit condition at specified intervals).
3505 Frequency : rare, normally only when NFS exporting is done from client
3507 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3509 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3512 Frequency : always, if srand() is called before liblustre initialization
3514 Description: Liblustre uses system PRNG disturbing its usage by user application
3515 Details : Introduce internal to lustre fast and high-quality PRNG for
3516 lustre usage and make liblustre and some other places in generic
3517 lustre code to use it.
3519 Severity : enhancement
3520 Bugzilla : 9477, 9557, 9870
3521 Description: Verify that the MDS configuration logs are updated when xml is
3522 Details : Check if the .xml configuration logs are newer than the config
3523 logs stored on the MDS and report an error if this is the case.
3524 Request --write-conf, or allow starting with --old_conf.
3526 Severity : enhancement
3528 Description: Handle symlinks in the path when checking if Lustre is mounted.
3529 Details : Resolve intermediate symlinks when checking if a client has
3530 mounted a filesystem to avoid duplicate client mounts.
3535 Description: lconf can hit an error exception but still return success.
3536 Details : The lconf command catches the Command error exception at the top
3537 level script context and will exit with the associated exit
3538 status, but doesn't ensure that this exit status is non-zero.
3543 Description: failure of ptlrpc thread startup can cause oops
3544 Details : Starting a ptlrpc service thread can fail if there are a large
3545 number of threads or the server memory is very fragmented.
3546 Handle this without oopsing.
3549 Frequency : always, only if liblustre and non-default acceptor port was used
3551 Description: liblustre cannot connect to servers with non-default acceptor port
3552 Details : tcpnal_set_default_params() was not called and was therefore
3553 ignoring the environment varaible TCPNAL_PORT, as well as other
3554 TCPNAL_ environment variables
3559 Description: two objects could be created on the same OST for a single file
3560 Details : If an OST is down, in some cases it was possible to create two
3561 objects on a single OST for a single file. No problems other
3562 than potential performance impact and spurious error messages.
3566 Bugzilla : 5681, 9562
3567 Description: Client may oops in ll_unhash_aliases
3568 Details : Client dcache may become inconsistent in race condition.
3569 In some cases "getcwd" can fail if the current directory is
3575 Description: Inode refcounting problems in NFS export code
3576 Details : link_raw functions used to call d_instantiate without obtaining
3577 extra inode reference first.
3581 Bugzilla : 9942, 9903
3582 Description: Referencing freed requests leading to crash, memleaks with NFS.
3583 Details : We used to require that call to ll_revalidate_it was always
3584 followed by ll_lookup_it. Also with revalidate_special() it is
3585 possible to call ll_revalidate_it() twice for the same dentry
3586 even if first occurence returned success. This fix changes semantic
3587 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3588 reference on a request referred from the intent.
3589 ll_intent_release() then releases such a request.
3592 Frequency : rare, normally benchmark loads only
3594 Description: unlinked inodes were kept in memory on the client
3595 Details : If a client is repeatedly creating and unlinking files it
3596 can accumulate a lot of stale inodes in the inode slab cache.
3597 If there is no other client load running this can cause the
3598 client node to run out of memory. Instead flush old inodes
3599 from client cache that have the same inode number as a new inode.
3602 Frequency : SLES9 2.6.5 kernel and long filenames only
3603 Bugzilla : 9969, 10379
3604 Description: utime reports stale NFS file handle
3605 Details : SLES9 uses out-of-dentry names in some cases, which confused
3606 the lustre dentry revalidation. Change it to always use the
3610 Frequency : rare, unless heavy write-truncate concurrency is continuous
3611 Bugzilla : 4180, 6984, 7171, 9963, 9331
3612 Description: OST becomes very slow and/or deadlocked during object unlink
3613 Details : filter_destroy() was holding onto the parent directory lock
3614 while truncating+unlinking objects. For very large objects this
3615 may block other threads for a long time and slow overall OST
3616 responsiveness. It may also be possible to get a lock ordering
3617 deadlock in this case, or run out of journal credits because of
3618 the combined truncate+unlink. Solution is to do object truncate
3619 first in one transaction without parent lock, and then do the
3620 final unlink in a new transaction with the parent lock. This
3621 reduces the lock hold time dramatically.
3624 Frequency : rare, 2.4 kernels only
3626 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3627 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3628 before unmounting their filesystem in order to avoid deadlock.
3629 The kernel_locked() function in 2.4 kernels only checks whether
3630 the kernel lock is held, not whether it is this process that is
3631 holding it as 2.6 kernels do.
3636 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3637 Details : The client ptlrpc code may be trying to reconnect to a down
3638 server before a previous connection attempt has timed out.
3639 Increase the reconnect interval to be longer than the connection
3640 timeout interval to avoid sending duplicate connections to
3644 Frequency : echo_client brw_test command
3646 Description: fix echo_client to work with OST preallocated code
3647 Details : OST preallocation code (5137) didn't take echo_client IO path
3648 into account: echo_client calls filter methods outside of any
3649 OST thread and, hence, there is no per-thread preallocated
3650 pages and buffers to use. Solution: hijack pga pages for IO. As
3651 a byproduct, this avoids unnecessary data copying.
3655 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3656 Description: Client can oops in mdc_commit_close() after open replay
3657 Details : It was possible for the MDS to return an open request with no
3658 transaction number in mds_finish_transno() if the client was
3659 evicted, but without actually returning an error. Clients
3660 would later try to replay that open and may trip an assertion
3661 Simplify the client close codepath, and always return an error
3662 from the MDS in case the open is not successful.
3665 Frequency : rare, 2.6 OSTs only
3667 Description: OST may deadlock under high load on fragmented files
3668 Details : If there was a heavy load and highly-fragmented OST filesystems
3669 it was possible to have all the OST threads deadlock waiting on
3670 allocation of biovecs, because the biovecs were not released
3671 until the entire RPC IO was completed. Instead, release biovecs
3672 as soon as they are complete to ensure forward IO progress.
3674 Severity : enhancement
3676 Description: Support for specifying external journal device at mount
3677 Details : If an OST or MDS device is formatted with an external journal
3678 device, this device major/minor is stored in the ext3 superblock
3679 and may not be valid for failover. Allow detecting and
3680 specifying the external journal at mount time.
3685 Description: Mounting an MDS with pending unlinked files may cause oops
3686 Details : target_finish_recovery() calls mds_postrecov() which returned
3687 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3688 considers this an error and immediately begins cleaning up the
3689 lov, just after starting the mds_lov process
3691 Severity : enhancement
3693 Description: Implement 'lfs df' to report actual free space on per-OST basis
3694 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3695 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3696 usage of objects; '-h' to report in human readable format.
3698 ------------------------------------------------------------------------------
3700 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3707 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3708 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3709 at startup during orphan unlinking. Instead, skip buddy update
3710 before setup, as it will be regenerated after recovery is complete.
3715 Description: If an OST is inactive, its locks might reference stale inodes.
3716 Details : lov_change_cbdata() must iterate over all namespaces, even if
3717 they are inactive to clear inode references from the lock.
3719 Severity : enhancement
3720 Frequency : occasional, if non-standard max_dirty_mb used
3722 Description: Client will block write RPCs if not enough grant
3723 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3724 then the client will block writes while waiting for another RPC
3725 to complete instead of consuming its dirty limit. With change
3726 we get improved performance when max_dirty_mb is small.
3728 Severity : enhancement
3729 Bugzilla : 3389, 6253
3730 Description: Add support for supplementary groups on the MDS.
3731 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3732 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3733 lookups for user supplementary groups into a cache.
3737 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3738 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3739 for liblustre O_EXCL is enforced by MDS.
3743 Description: readdir never returns NULL in liblustre.
3744 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3745 dentry in struct dirent.
3749 Frequency : liblustre only, depends on application IO pattern
3750 Description: liblustre clients evicted if not contacting servers
3751 Details : Don't put liblustre clients into the ping_evictor list, so
3752 they will not be evicted by the pinger ever.
3754 Severity : enhancement
3756 Description: Add ability to evict clients by NID from MDS.
3757 Details : By echoing "nid:$NID" string into
3758 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3759 $NID would be instantly evicted from this MDS and from all active
3760 OSTs connected to it.
3764 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3765 Details : lookup_it_finish() used to query file size from OSTs that was not
3770 Description: service threads change working directory to that of init
3771 Details : Starting lustre service threads may pin the working directory
3772 of the parent thread, making that filesystem busy. Threads
3773 now change to the working directory of init to avoid this.
3777 Frequency : during shutdown only
3778 Description: shutdown with a failed MDS or OST can cause unmount to hang
3779 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3784 Frequency : 2.6 only
3785 Description: chmod/chown may include an extra supplementary group
3786 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3787 supplementary group and if none is specified this is used.
3790 Bugzilla : 5479 (6816)
3792 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3793 Details : If lookup is in progress on a file that is unlinked we might try
3794 to revalidate the inode and fail in revalidate after lookup is
3795 complete and ll_file_open() enqueues the open again but
3796 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3799 Frequency : always, if lconf --abort_recovery used
3801 Description: lconf --abort_recovery fails with 'Operation not supported'
3802 Details : lconf was attempting to abort recovery on the MDT device and not
3805 ------------------------------------------------------------------------------
3807 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3812 Frequency : rare (only unsupported configurations with a node running as an
3814 Bugzilla : 6514, 5137
3815 Description: Mounting a Lustre file system on a node running as an OST could
3817 Details : OSTs now preallocates memory needed to write out data at
3818 startup, instead of when needed, to avoid having to
3819 allocate memory in possibly low memory situations.
3820 Specifically, if the file system is mounted on on OST,
3821 memory pressure could force it to try to write out data,
3822 which it needed to allocate memory to do. Due to the low
3823 memory, it would be unable to do so and the node would
3824 become unresponsive.
3826 Severity : enhancement
3828 Description: Addition of lconf --service command line option
3829 Details : lconf now accepts a '--service <arg>' option, which is
3830 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3832 Severity : enhancement
3834 Description: Failover mode is now the default for OSTs.
3835 Details : By default, OSTs will now run in failover mode. To return to
3836 the old behaviour, add '--failout' to the lmc line for OSTs.
3838 Severity : enhancement
3840 Description: Health checks are now provided for MDS and OSTs
3841 Details : Additional detailed health check information on MSD and OSTs
3842 is now provided through the procfs health_check value.
3845 Frequency : occasional, depends on IO load
3847 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3848 after numerous create/delete cycles
3849 Details : The ext3 inode allocation policy would not allocate new inodes
3850 very well on the OSTs because there are no new directories
3851 being created. Instead we look for groups with free space if
3852 the parent directories are nearly full.
3857 Description: Network or server problems during mount may cause partially
3858 mounted clients instead of returning an error.
3859 Details : The config llog parsing code may overwrite the error return
3860 code during mount error handling, returning success instead
3866 Description: MDS can fail to allocate large reply buffers
3867 Details : After long uptimes the MDS can fail to allocate large reply
3868 buffers (e.g. zconf client mount config records) due to memory
3869 fragmentation or consumption by the buffer cache. Preallocate
3870 some large reply buffers so that these replies can be sent even
3871 under memory pressure.
3875 Frequency : rare (liblustre)
3876 Description: fsx running with liblustre complained that using truncate() to
3877 extend the file doesn't work. This patch corrects that issue.
3878 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3879 ATTR_SIZE and lsm use in llu_setattr_raw.
3883 Frequency : rare, only 2.6 kernels
3884 Description: Unusual file access patterns on the MDS may result in inode
3885 data being lost in very rare circumstances.
3886 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3887 code in the 2.6 kernel caused the inode and/or EA data not to
3888 be read from disk, causing single-file corruption.
3892 Frequency : rare, only 2.6 filesystems using extents
3893 Description: Heavy concurrent write and delete load may cause data corruption.
3894 Details : It was possible under high-load situations to have an extent
3895 metadata block in the block device cache from a just-unlinked
3896 file overwrite a newly-allocated data block. We now unmap any
3897 metadata buffers that alias just-allocated data blocks.
3901 Frequency : filesystems with default stripe_count larger than 77
3902 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3903 Details : lconf specifies an inode size of 4096 bytes when the default
3904 stripe_count is larger than 77. This conflicts with the default
3905 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3906 this case to avoid pinning too much memory for large EAs.
3908 ------------------------------------------------------------------------------
3910 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3915 Frequency : rare (extremely heavy IO load with hundreds of clients)
3917 Description: Client is evicted, gets IO error writing to file
3918 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3919 caused clients to be evicted for AST timeouts. The fixes in
3920 bug 5192 mean we no longer need to have such short AST timeouts
3921 so ldlm_timeout has been increased.
3924 Frequency : occasional during --force or --failover shutdown under load
3925 Bugzilla : 5949, 4834
3926 Description: Server oops/LBUG if stopped with --force or --failover under load
3927 Details : a collection of import/export refcount and cleanup ordering
3928 issues fixed for safer force cleanup
3931 Frequency : only filesystems larger than 120 OSTs
3932 Bugzilla : 5990, 6223
3933 Description: lfs getstripe would oops on a very large filesystem
3934 Details : lov_getconfig used kfree on vmalloc'd memory
3937 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3938 Bugzilla : 6242, 6243
3939 Description: reading from files that had been truncated to a non-zero size
3940 but never opened returned no data
3941 Details : ll_file_read() reads zeros from no-object files to EOF
3946 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3947 Details : There's little harm in aborting MDS/OSS recovery and letting it
3948 try again, so I removed the LASSERT and return an error instead.
3950 Severity : enhancement
3952 Description: New debugging infrastructure for tracking down data corruption
3953 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3954 (b) cover more of the client-side code path, and (c) try to narrow
3955 down where problems occurred
3959 Bugzilla : 3819, 4364, 4397, 6313
3960 Description: Racing close and eviction MDS could cause assertion in mds_close
3961 Details : It was possible to get multiple mfd references during close and
3962 client eviction, leading to one thread referencing a freed mfd.
3964 Severity: : enhancement
3965 Bugzilla : 3262, 6359
3966 Description: Attempts to reconnect to servers are now more aggressive.
3967 Details : This builds on the enhanced upcall-less recovery that was added
3968 in 1.4.2. When trying to reconnect to servers, clients will
3969 now try each server in the failover group every 10 seconds. By
3970 default, clients would previously try one server every 25 seconds.
3975 Description: After recovery, certain operations trigger a failed
3976 assertion on a client.
3977 Details : Failing over an mds, using lconf -d --failover, while a
3978 client was doing a readdir() call would cause the client to
3979 LBUG after recovery completed and the readdir() was resent.
3981 Severity : enhancement
3983 Description: Default groups are now added by lconf
3984 Details : You can now run lconf --group <servicename> without having to
3985 manually add groups with lmc.
3988 Frequency : occasional
3990 Description: Nodes with an elan id of 0 trigger a failed assertion
3993 Frequency : always when accessing e.g. tty/console device nodes
3995 Description: tty and some other devices nodes cannot be used on lustre
3996 Details : file's private_data field is used by device data and lustre
3997 values in there got lost. New field was added to struct file to
3998 store fs-specific private data.
4001 Frequency : when exporting Lustre via NFS
4003 Description: NFSD failed occasionally when looking up a path component
4004 Details : NFSD is looking up ".." which was broken in ext3 directories
4005 that had grown large enough to become hashed.
4008 Frequency : Clusters with multiple interfaces not on the same subnet
4010 Description: Nodes will repeatedly try to reconnect to an interface which it
4011 cannot reach and report an error to the log.
4012 Details : Extra peer list entries will be created by lconf with some peers
4013 unreachable. lconf now validates the peer before adding it.
4016 Frequency : Only if a default stripe is set on the filesystem root.
4018 Description: Setting a default stripe on the filesystem root prevented the
4019 filesystem from being remounted.
4020 Details : The client was sending extra request flags in the root getattr
4021 request and did not allocate a reply buffer for the dir EA.
4024 Frequency : occasional, higher if lots of files are accessed by one client
4025 Bugzilla : 6159, 6097
4026 Description: Client trips assertion regarding lsm mismatch/magic
4027 Details : While revalidating inodes the VFS looks up inodes with ifind()
4028 and in rare cases can find an inode that is being freed.
4029 The ll_test_inode() code will free the lsm during ifind()
4030 when it finds an existing inode and then the VFS later attaches
4031 this free lsm to a new inode.
4035 Bugzilla : 6422, 7030
4036 Description: MDS deadlock between mkdir and client eviction
4037 Details : Creating a new file via mkdir or mknod (starting a transaction
4038 and getting the ns lock) can deadlock with client eviction
4039 (gets ns lock and trying to finish a synchronous transaction).
4042 Frequency : occasional
4043 Description: While starting a server, the fsfilt_ext3 module could not be
4045 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
4046 kernels. Previously, lconf would still use the ext3 name
4047 when trying to load modules. Now, it will correctly use
4048 ext3 on 2.4 and ldiskfs on 2.6.
4050 Severity : enhancement
4051 Description: The default stripe count has been changed to 1
4052 Details : The interpretation of the default stripe count (0, to lfs
4053 or lmc) has been changed to mean striping across a single
4054 OST, rather than all available. For general usage we have
4055 found a stripe count of 1 or 2 works best.
4057 Severity : enhancement
4058 Description: Add support for compiling against Cray portals.
4059 Details : Conditional compiling for some areas that are different
4063 Frequency : occasional
4064 Bugzilla : 6409, 6834
4065 Description: Creating files with an explicit stripe count may lead to
4066 a failed assertion on the MDS
4067 Details : If some OSTs are full or unavailable, creating files may
4068 trigger a failed assertion on the MDS. Now, Lustre will
4069 try to use other servers or return an error to the
4073 Frequency : occasional
4075 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4076 caused lock timeout during readahead (since 1.4.2).
4077 Details : Processes doing readahead might match a lock that hasn't been
4078 granted yet if there are overlapping and conflicting lock
4079 requests. The readahead process waits on ungranted lock
4080 (original lock is CBPENDING), while OST waits for that process
4081 to cancel CBPENDING read lock and eventually evicts client.
4083 Severity : enhancement
4085 Description: Initial enabling of flock support for clients
4086 Details : Implements fcntl advisory locking and file status functions.
4087 This feature is provided as an optional mount flag (default
4088 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4089 locking are implemented yet, and those that are are not guaranteed
4090 to be completely correct in production environments.
4091 mount -t lustre -o [flock|noflock] ...
4094 Frequency : occasional
4096 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4097 assertion in the ext3 JBD (journaling) layer.
4098 Details : The b_committed_data struct is protected by the big kernel lock
4099 in 2.4 kernels, serializing journal_commit_transaction() and
4100 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4101 access to this struct. In 2.6 kernels there is finer grained
4102 locking to improve SMP performance of the JBD layer.
4106 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4111 Description: merge and cleanup kernel patches.
4114 -----------------------------------------------------------------------------
4116 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4118 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4119 versions of Lustre. Please update all servers and clients to
4120 version 1.4.2 or later at the same time. You must also run
4121 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4122 to update the configuration logs.
4124 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4125 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4126 - swab llog records of type '0' so we get proper header size/idx (5861)
4127 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4128 - fix rename of one directory over another leaking an inode (5953)
4129 - avoid SetPageDirty on 2.6 (5981)
4130 - don't re-add just-being-destroyed locks to the waiting list (5653)
4131 - when creating new directories, inherit the parent's custom
4132 striping settings if present parent (3048)
4133 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4134 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4135 - don't hold client locks on temporary worklist from l_lru (5666)
4136 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4137 - automatically evict dead clients (5921)
4138 - Update file size properly in create+truncate+fstat case (6196)
4139 - Do not unhash mountpoint dentries, do not allow removal of
4141 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4142 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4143 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4144 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4145 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4146 - Correctly handle reads of files with no objects (6243)
4147 - lctl recover will also mark a device active if deactivate used (5933)
4149 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
4150 - allow --write-conf on an MDS with different nettype than client (5619)
4151 - don't write config llogs to MDS for mounts not from that MDS (5617)
4152 - lconf should create multiple TCP connections from a client (5201)
4153 - init scripts are now turned off by default; run chkconfig --on
4154 lustre and chkconfig --on lustrefs to use them
4155 - upcalls are no longer needed for clients to recover to failover
4157 - add --abort-recovery option to lconf to abort recovery on device
4159 - add support for an arbitrary number of OSTs (3026)
4160 - Quota support protocol changes.
4161 - forward compatibility changes to wire structs (6007)
4162 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
4163 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
4164 - improved "lustre status" script
4165 - initialize blocksize for non-regular files (6062)
4166 - added --disable-server and --disable-client configure options (5782)
4167 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
4168 - Vanilla 2.4.29 support
4169 - increase maximum number of obd devices to 520 (6242)
4170 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4171 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4172 SLES 9 2.6.5 kernels
4173 - limit stripes per file to 160 (the maximum EA size) (6093)
4175 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4178 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4179 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4180 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4181 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4182 - small ext3 extents cleanups and fixes (5733)
4183 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4184 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4185 - avoid needless client->OST connect, fix handle mismatch (5317)
4186 - fix DLM error path that led to out-of-sync client, long delays (5779)
4187 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4188 - fix several locking issues related to i_size (5492,5624,5654,5672)
4189 - don't move pending lock onto export if it is already evicted (5683)
4190 - fix kernel oops when creating .foo in unlinked directory (5548)
4191 - fix deadlock in obdfilter statistics vs. object create (5811)
4192 - use time_{before,after} to avoid timer jiffies wrap (5882)
4193 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4194 - Do not leak request if server was not able to process it (5154)
4195 - If mds_open unable to find parent dir, make that negative lookup(5154)
4196 - don't create new directories with extent-mapping (5909, 5936)
4198 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
4199 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
4200 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
4201 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
4202 - support for RHEL4 kernel on i686 (5773)
4203 - provide error messages when incompatible logs are encountered (5898)
4205 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
4206 * version 1.4.0.10 (1.4.1 release candidate 1)
4208 - don't keep a lock reference when lock is not granted (4238)
4209 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4210 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
4211 - drop import inflight refcount on signal_completed_replay error (5255)
4212 - unlock page after async write error during send (3677)
4213 - handle missing objects in filter_preprw_read properly (5265)
4214 - no transno return for symlink open, don't save no-trasno open (3440)
4215 - don't try to complete elan receive that already failed (4012)
4216 - free RPC server reply state on error (5406)
4217 - clean up thread from ptlrpc_start_thread() on error (5160)
4218 - readahead could read extra page into cache that wasn't ejected (5388)
4219 - prevent races in class_attach/setup/cleanup/detach (5260)
4220 - don't dereference de->d_inode after l_dput of de (5458)
4221 - use "int" for stripe value returned from lock_to_stripe (5544)
4222 - mballoc allocation and error-checking fixes in 2.6 (5504)
4223 - block device patches to fix I/O request sizes in 2.6 (5482)
4224 - look up hostnames for IB nals (5602)
4225 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
4226 - don't start multiple acceptors for the same port (5277)
4227 - fix incorrect LASSERT in mds_getattr_name (5635)
4228 - export a proc file for general "ping" checking (5628)
4229 - fix "lfs check" to not block when the MDS is down (5628)
4231 - service request history (4965)
4232 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4233 - create an "evict_client" /proc entry on OSTs, like the MDS has
4234 - fix mount usage message, return errors per mount(8) (5168)
4235 - change grep [] to grep "[]" in tests so they work in more UMLs
4236 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4237 - remove ancient LOV_MAGIC_V0 EA support (5047)
4238 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4239 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4240 - allow readahead allocations to fail when low on memory (5383)
4241 - mmap locking landed again, after considerable improvement (2828)
4242 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4244 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4247 - send OST transaction number in read/write reply to free req (4966)
4248 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4249 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4250 - the watchdog thread now runs as interruptible (5246)
4251 - flock/lockf fixes (but it's still disabled, pending 5135)
4252 - don't use EXT3 constants in llite code (5094)
4253 - memory shortage at startup could cause assertion (5176)
4255 - reorganization of lov code
4256 - single portals codebase
4258 - add extents/mballoc support (5025)
4259 - direct I/O reads in the obdfilter (4048)
4260 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4262 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4265 - send OST transaction number in read/write reply to free req (4966)
4266 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4267 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4268 - the watchdog thread now runs as interruptible (5246)
4269 - handle missing objects in filter_preprw_read properly (5265)
4270 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4271 - drop import inflight refcount on signal_completed_replay error (5255)
4272 - unlock page after async write error during send (3677)
4273 - return original error code on reconstructed replies (3761)
4274 - no transno return for symlink open, don't save no-trasno open (3440)
4276 - add pid to ldlm debugging output (4922)
4277 - bump the watchdog timeouts -- we can't handle 30sec yet
4278 - extra debugging for orphan dentry/inode bug (5259)
4280 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
4283 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4284 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4285 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
4286 - fix NULL dereference in /proc/sys/portals/routes (4827)
4287 - allow failed mdc_close() operations to be interrupted (4561)
4288 - stop precreate on OST before MDS would time out on it (4778)
4289 - don't send partial-page writes before EOF from client (4410)
4290 - discard client grant for sub-page writes on large-page clients (4520)
4291 - don't free dentries not owned by NFS code, check generation (4806)
4292 - fix lsm leak if mds_create_objects() fails (4801)
4293 - limit debug_daemon file size, always print CERROR messages (4789)
4294 - use transno after validating reply (3892)
4295 - process timed out requests if import state changes (3754)
4296 - update mtime on OST during writes, return in glimpse (4829)
4297 - add mkfsoptions to LDAP (4679)
4298 - use ->max_readahead method instead of zapping global ra (5039)
4299 - don't interrupt __l_wait_event() during strace
4301 - add software watchdogs to catch hung threads quickly (4941)
4302 - make lustrefs init script start after nfs is mounted
4303 - fix CWARN/ERROR duplication (4930)
4304 - return async write errors to application if possible (2248)
4305 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4306 - print NAL number in %x format (4645)
4307 - update barely-supported suse-2.4.21-171 series (4842)
4308 - support for sles 9 %post scripts
4309 - support for building 2.6 kernel-source packages
4310 - support for sles km_* packages
4312 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4315 - ignore -ENOENT errors in osc_destroy (3639)
4316 - notify osc create thread that OSC is being cleaned up (4600)
4317 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4318 - reconstruct ost_handle() like mds_handle() (4657)
4319 - create a new thread to do import eviction to avoid deadlock (3969)
4320 - let lconf resolve symlinked-to devices (4629)
4321 - don't unlink "objects" from directory with default EA (4554)
4322 - hold socknal file ref over connect in case target is down (4394)
4323 - allow more than 32000 subdirectories in a single directory (3244)
4324 - fix blocks count for O_DIRECT writes (3751)
4325 - OST returns ENOSPC from object create when no space left (4539)
4326 - don't send truncate RPC if file size isn't changing (4410)
4327 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4328 - bind to privileged port in socknal and tcpnal (3689)
4330 - rate limit CERROR/CWARN console message to avoid overload (4519)
4331 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4332 - basic mmap support (3918)
4333 - kernel patch series update from b1_4 (4711)
4335 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4338 - avoid crash during MDS cleanup with OST shut down (2775)
4339 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4340 - don't use bad inodes on the MDS (3744)
4341 - dynamic object preallocation to improve recovery speed (4236)
4342 - don't hold spinlock over lock dumping or change debug flags (4401)
4343 - don't zero obd_dev when it is force cleaned (3651)
4344 - print grants to console if they go negative (4431)
4345 - "lctl deactivate" will stop automatic recovery attempts (3406)
4346 - look for existing locks in ldlm_handle_enqueue() (3764)
4347 - don't resolve lock handle twice in recovery avoiding race (4401)
4348 - revalidate should check working dir is a directory (4134)
4350 - don't always mark "slow" obdfilter messages as errors (4418)
4352 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4355 - don't close LustreDB during write_conf until it is done (3860)
4356 - fix typo in lconf for_each_profile (3821)
4357 - allow dumping logs from multiple threads at one time (3820)
4358 - don't allow multiple threads in OSC recovery (3812)
4359 - fix debug_size parameters (3864)
4360 - fix mds_postrecov to initialize import for llog ctxt (3121)
4361 - replace config semaphore with spinlock (3306)
4362 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4363 - don't allow enqueue to complete on a destroyed export (3822)
4364 - down write_lock before checking llog header bitmap (3825)
4365 - recover from lock replay timeout (3764)
4366 - up llog sem before sending rpc (3652)
4367 - reduce ns lock hold times when setting kms (3267)
4368 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4369 - fix NULL deref and obd_dev leak on setup error (3312)
4370 - replace some LBUG about llog ops with error handling (3841)
4371 - don't match INVALID dentries from d_lookup and spin (3784)
4372 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4373 - fix invalid assertion in ptlrpc_set_wait (3880)
4375 - add libwrap support for the TCP acceptor (3996)
4376 - add /proc/sys/portals/routes for non-root route listing (3994)
4377 - allow setting MDS UUID in .xml (2580)
4378 - print the stack of a process that LBUGs (4228)
4380 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4383 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4384 - make sure to unset replay flag from failed open requests (3440)
4385 - if default stripe count is 0, use OST count for inode size (3636)
4386 - update parent mtime/ctime on client for create/unlink (2611)
4387 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4388 - free recovery state on server during a forced cleanup (3571)
4389 - unregister_reply for resent reqs (3063)
4390 - loop back devices mounting and status check on 2.6 (3563)
4391 - fix resource-creation race that can provoke i_size == 0 (3513)
4392 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4393 - more debugging for page-accounting assertion (3746)
4394 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4395 - avoid deadlock after precreation failure (3758)
4396 - fix race and lock order deadlock in orphan handling (3450, 3750)
4397 - add validity checks when grabbing inodes from l_ast_data (3599)
4399 - add /proc/.../recovery_status to obdfilter (3428)
4400 - lightweight CDEBUG infrastructure, debug daemon (3668)
4401 - change default OSC RPC parameters to be better on small clusters
4402 - turn off OST read cache for files smaller than 32MB
4403 - install man pages and include them in rpms (3100)
4404 - add new init script for (un)mounting lustre filesystems (2593)
4405 - run chkconfig in %post for init scripts (3701)
4406 - drop scimac NAL (unmaintained)
4408 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4411 - clean kiobufs before and after use (3485)
4412 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4413 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4414 - change init script to fail more clearly if not run as root (1528)
4415 - allow clients to reconnect during replay (1742)
4416 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4417 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4418 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4419 - fix cond_resched (3554)
4420 - only evict unfinished clients after recovery (3515)
4421 - allow bulk resend, prevent data loss (3570)
4422 - dynamic ptlrpc request buffer allocation (2102)
4423 - don't allow unlinking open directory if it isn't empty (2904)
4424 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4425 - remove extraneous obd dereference causing LASSERT failure (3334)
4426 - don't use get_cycles() when creating temp. files on the mds (3156)
4427 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4428 - handle EEXIST for set-stripe, set proper directory name (3336)
4430 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4431 - fix ksocknal_fmb_callback() error messages (2918)
4433 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4436 - don't copy lvb into (possibly NULL) reply on error (2983)
4437 - don't deref dentry after dput, don't free lvb on error (2922)
4438 - use the kms to determine writeback rpc length (2947)
4439 - increment oti_logcookies when osc is inactive (2948)
4440 - update client's i_blocks count via lvb messages (2543)
4441 - handle intent open/close of special files properly (1557)
4442 - mount MDS with errors=remount-ro, like obdfilter (2009)
4443 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4444 - don't use cancelling-locks' kms values (2947)
4445 - use highest lock extent for kms, not last one (2925)
4446 - don't dereference ERR_PTR() dentry in error handling path (3107)
4447 - fix thread race in portals_debug_dumplog() (3122)
4448 - create lprocfs device entries at setup instead of at attach (1519)
4449 - common AST error handler, don't evict client on completion race (3145)
4450 - zero nameidata in detach_mnt in 2.6 (3118)
4451 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4452 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4453 - initialize RPC timeout timer earlier for 2.6 (3219)
4454 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4455 - print nal/nid for unknown nid (3258)
4456 - additional checks for oscc recovery before doing precreate (3284)
4457 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4458 - don't crash in mdc_close for bad permissions on open (3285)
4459 - zero i_rdev for non-device files (3147)
4460 - clear page->private before handing to FS, better assertion (3119)
4461 - tune the read pipeline (3236)
4462 - fix incorrect decref of invalidated dentry (2350)
4463 - provide read-ahead stats and refine rpc in flight stats (3328)
4464 - don't hold journal transaction open across create RPC (3313)
4465 - update atime on MDS at close time (3265)
4466 - close LDAP connection when recovering to avoid server load (3315)
4467 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4468 - don't leak open file on MDS after open resend (3325)
4469 - serialize filter_precreate and filter_destroy_precreated (3329)
4470 - loop device shouldn't call sync_dev() for nul device (3092)
4471 - clear page cache after eviction (2766)
4472 - resynchronize MDS->OST in background (2824)
4473 - refuse to mount the same filesystem twice on same mountpoint (3394)
4474 - allow llmount to create routes for mounting behind routers (3320)
4475 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4476 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4477 - fix rare problem with rename on htree directories (3417)
4479 - allow default OST striping configuration per directory (1414)
4480 - fix compilation for qswnal for 2.6 kernels (3125)
4481 - increase maximum number of MDS request buffers for large systems
4482 - change liblustreapi to be useful for external progs like lfsck (3098)
4483 - increase local configuration timeout for slow disks (3353)
4484 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4486 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4489 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4490 - don't overwrite extent policy data in reply if lock was blocked (2901)
4491 - drop filter export grants atomically with removal from device (2663)
4492 - del obd_self_export from work_list in class_disconnect_exports (2908)
4493 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4494 - swab reply message in mdc_close, other PPC fixes (2464)
4495 - fix destroying of named logs (2325)
4496 - overwrite old logs when running lconf --write_conf (2264)
4497 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4498 - fix race in target_handle_connect (2898)
4499 - mds_reint_create() should take same inode create lock (2926)
4500 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4501 - don't close files for self_export to avoid uninitialized obd (2936)
4502 - allow MDS with the same name as client node (2939)
4503 - hold dentry reference for closed log files for unlink (2325)
4504 - reserve space for all logs during transactions (2059)
4505 - don't evict page beyond end of stripe extent (2925)
4506 - don't oops on a deleted current working directory (2399)
4507 - handle hard links to targets without a parent properly (2517)
4508 - don't dereference NULL lock when racing during eviction (2867)
4509 - don't grow lock extents when lots of conflicting locks (2919)
4511 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4514 - account for cache space usage on clients to avoid data loss (974)
4515 - lfsck support in lustre kernel code (2349)
4516 - reduce journal credits needed for BRW writes (2370)
4517 - orphan handling to avoid losing space on client/server crashes
4518 - ptlrpcd can be blocked, stopping ALL progress (2477)
4519 - use lock value blocks to assist in proper KMS, faster stat (1021)
4520 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4521 - recovery for initial connections (2355)
4522 - fixes for mds_cleanup_orphans (1934)
4523 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4524 - block all file creations until orphan recovery completes (1901)
4525 - client remove rq_connection from request struct (2423)
4526 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4527 - recovery timer race (2670)
4528 - mdc_close recovey bug (2532)
4529 - ptlrpc cleanup bug (2710)
4530 - mds timeout on local locks (2588)
4531 - namespace lock held during RPCs (2431)
4532 - handle interrupted sync write properly (2503)
4533 - don't try to handle a message that hasn't been replied to (2699)
4534 - client assert failure during cleanup after abort recovery (2701)
4535 - leak mdc device after failed mount (2712)
4536 - ptlrpc_check_set allows timedout requests to complete (2714)
4537 - wait for inflight reqs when ptlrpcd finishes (2710)
4538 - make sure unregistered services are removed from the srv_list
4539 - reset bulk XID's when resending them (caught by 1138 test)
4540 - unregister_bulk after timeout
4541 - fix lconf error (2694)
4542 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4543 - readahead locks pages, leaves pending causing memory pressure (2673)
4544 - increase OST request buffers to 4096 on large machines (2729)
4545 - fix up permission of existing directories in simple_mkdir (2661)
4546 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4547 - don't assign transno to errored transactions (2742)
4548 - don't delete objects on OST if given a bogus objid from MDS (2751)
4549 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4550 - if rq_no_resend, then timeout request after recovery (2432)
4551 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4552 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4553 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4554 - MDS recovery shouldn't delete objects that it has given out (2730)
4555 - if enqueue arrives after completion, don't clobber LVB (2819)
4556 - don't unlock pages twice when trigger_group_io returns error (2814)
4557 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4558 - don't write pages to disk if there was an error (1450)
4559 - don't ping imports that have recovery disabled (2676)
4560 - take buffered bytes into account when balancing socknal conn (2817)
4561 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4562 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4563 - remove little-endian swabbing of llog records (1987)
4564 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4565 - retry reposting request buffers if they fail (1191)
4566 - grow extent at grant time to avoid granting a revoked lock (2809)
4567 - lock revoke doesn't evict page if covered by a second lock (2765)
4568 - disable VM readahead to avoid reading outside lock extents (2805)
4570 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4571 - updated kernel patches for hp-2.4.20 kernel (2681)
4573 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4576 - fix truncated write corruption (2366)
4577 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4579 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4580 - ensure len(uuid) < 37 in lmc (1171)
4581 - fix ia64 OOPS in llog_test (2255)
4582 - zero end of page at obdfilter for partial page writes (2648)
4583 - don't leave stale dentries around after renames (bug 2428)
4584 - fix timeouts when evicting a client with a single lock held (2642)
4585 - set deadline for the initial HELLO message to drain (2634)
4586 - print out dotted-quad IP addresses in the socknal (2302)
4588 - additional debugging for MDS client eviction problem (2443)
4589 - fix mkfsoptions support for osts (2603, 2604)
4591 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4594 - add series for the vanilla 2.6.0 kernel
4595 - add series for the vanilla 2.4.24 kernel
4596 - add series for a cray x86/64 UL kernel drop
4597 - fix xattr patches for the vanilla 2.4.19 series
4599 - generate true UUIDs in lmc (1171)
4600 - have portals stack dumping break in UML (2466)
4601 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4602 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4603 - wake callback waiting threads on client eviction (2460)
4604 - Add --ptldebug and --subsystem to lmc (1719)
4605 - update assertion to allow safe interrupt allocation
4606 - set rq_no_resend for cancel requests (2432)
4607 - recalculate ptlrpcd timeout after resend (2494)
4608 - call vfs_rmdir when removing pending directories (2368)
4609 - fix renaming a file to itself (2429)
4610 - lmc creates a default one-stripe lov (2454)
4611 - expand procfs space to handle large clusters (2326)
4612 - increase UML stack to avoid overflow
4613 - update lconf's list of debug and subsystem masks
4614 - fix lfs find --obd (2510)
4615 - /proc tunable for disabling filter read caching (2591)
4616 - stop rpm packages from altering slapd.conf (2301)
4617 - disable nagle in the socknal under 0conf (2578)
4618 - choose mds inode size based on stripe count (2572)
4619 - fix kernel-source rpm problems (2516)
4621 - add --disable-doc to avoid pdf generation (2421)
4622 - update documentation, tests, type-os, comments
4623 - avoid format warnings on ia64
4624 - remove the TOE NAL
4625 - tiny code cleanups by removing unused fields
4627 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4630 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4631 - remove the most blatant lies from BUILDING file (2371)
4632 - change default debug level to reasonable production setting
4633 - reduce client side cache size to reduce cache flush time
4634 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4635 - make TCP zerocopy and pinger support enabled by default (2476)
4636 - sync writes completed after process exits caused crashes (2319)
4637 - maintain correct mount count on the MDS (2356)
4638 - backout 1557, because 2316 wasn't really fixed
4639 - better file I/O statistics gathering in /proc
4640 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4641 - another kernel patch to fix zero-copy TCP function export
4642 - don't take duplicate lock when processing re-sent getattr (2420)
4643 - lctl uses obd_self_export instead of creating new conn (2353)
4644 - MDS/OST recovery case which requires object creation asserted (2425)
4645 - move lfs from /usr/sbin to /usr/bin in packages
4646 - fix race between mds_client_add and mds_client_free (2417)
4647 - use kmalloc instead of slabs in portals (2430)
4648 - don't create duplicate records when a failover MDS is present (2442)
4649 - remove unnecessary mount age check (2332)
4650 - don't remove directory inodes from locks prematurely (2451)
4651 - don't break if MDS service name is the same as hostname (2103)
4652 - fix races in client write RPC generation when cache full (2482)
4654 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4657 - remove now-unused request->rq_obd (278)
4658 - if an allocation fails, print out how much memory we've used (1933)
4659 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4660 - add the "configurable stack size" patch to most series files (1256)
4661 - ability to write large log records, for 100+ OST configs (2306)
4662 - fix NULL deref when filter_prep fails (2314)
4663 - fix operator precedence error in filter_sync
4664 - dynamic allocation of socknal TX descriptors (2315)
4665 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4666 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4667 - ensure that CWARN messages in Portals always get to the syslog
4668 - __init/__exit are not for prototype decls (ldlm_init/exit)
4669 - x86-64 compile warning fixes
4670 - fix gateway LMC keyword conflict (2318)
4671 - fix MDS lock inversions in getattr/reint paths (1844)
4672 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4673 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4674 - DLM race condition prevented some lock evictions (2328)
4675 - ENOMEM detection and retry on socknal sends (2230)
4676 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4677 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4678 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4679 - better error messages when a client is rejected during recovery (1505)
4680 - avoid cancelling locks which were never granted, after failure (2330)
4681 - fix i_sem/journal inversion in mds_client_add (2333)
4682 - fix truncate/getattr lock cycle deadlock (2334)
4683 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4684 - fix two rare exit paths which could leak an l_lock() ref (2321)
4685 - fencepost error in MDS/OST orphan recovery (2226)
4686 - make log record alignment 8 bytes (1988)
4687 - lstripe now fails when requested offset > ost_count (2237)
4688 - ensure that all kernel series have a complete list.h (1607)
4689 - fix crashes in special-file operations (2316)
4690 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4692 - allow configurable automake binary, for testing new versions
4693 - small update to the lfs documentation
4695 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4697 * fix negative export reference count in fsfilt_sync (2312)
4699 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4700 * release candidate 0.9.1
4702 - orphans are moved into the PENDING directory for possible recovery
4703 - replayed opens now open by fid for orphan/rename safety (1042)
4704 - last close of an orphan inode generates a transno (683)
4705 - chdir() and mount() now pin the directory entry (1020)
4706 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4707 - discard very old requests without processing them (1502)
4708 - remove some common, well-understood CERRORs (1505)
4709 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4710 - clear "grant" flags in OST replies until OST grant code lands (1644)
4711 - fix read performance by not clobbering i_blksize on client (1598)
4712 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4713 - make LRU size a /proc tunable, clears locks when reduced (707)
4714 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4715 - server locks take references on exports now (1558)
4716 - build fixes for 2.4.20-rh trees (1663)
4717 - return an error from lov_create if all OSCs are inactive (1751)
4718 - fix import levels when a reconnect happens without a timeout (1597)
4719 - exit early from mds_open if we get a lookup error (1749)
4720 - partial page read at EOF wouldn't wait for disk before sending (1642)
4721 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4722 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4723 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4724 - write/truncate lock inversion (1639)
4725 - Don't auto-load obdclass, portals modules during cleanup (1495)
4726 - fix timestamps from jumping to "now" (1763)
4727 - extra journal assertions (1648)
4728 - add an extra multiunlink test (1771)
4729 - fix read_record/write_record API (1776)
4730 - fix leak of offset_extent, possible incorrect i_size later (1772)
4731 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4732 - Debugging for the kqswnal_get_idle_tx problems (1820)
4733 - Allow recovery to be attempted multiple times (1536)
4734 - Write out MDS last_rcvd file after it is first created (1600)
4735 - Fix tx_descriptor leak in failed transmit situations (1827)
4736 - ext3 journaling fixes for assertion failure after IO error (1871)
4737 - class_export_put() on freed export after completion AST error (1896)
4738 - Fix revalidate looping in VFS (1322)
4739 - Don't access a freed export during MDS_REINT timeout (1521)
4740 - Add open-unlink recovery support on the MDS (1673,1764)
4741 - Return an error if no MDS data was read from last_rcvd (1946)
4742 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4743 - Don't setup a disk device that doesn't match exported UUID (317)
4744 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4745 - avoid committing NULL handle in force close
4746 - local.sh is now a one-stripe LOV configuration
4747 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4748 - fix ext3 htree duplicate directory entry corruption (1516)
4749 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4750 - update to new LOV EA format (2097)
4751 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4753 2003-06-15 Phil Schwan <phil@clusterfs.com>
4756 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4757 - per-import/export recovery handling (958, 931, 959)
4758 - multiple last-rcvd slots, for serving multiple FSes (949)
4759 - connections are again shared between multiple imp/exports (963, 964)
4760 - "umount -f" would hang if any requests needed to be sent (393, 978)
4761 - avoid pinning large req buffer by copying for queued messages (989)
4762 - add "uuid" to "lctl device" command to help upcalls (991)
4763 - "open" RPCs with transnos would confuse recovery counters (1037)
4764 - do proper endian conversion of all wire messages (288, 340, 891)
4765 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4766 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4767 - fix ptlrpc_connection leak in target_handle_connect (1174)
4768 - fix import refcounting bug in OST and MDS cleanup (1134)
4769 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4770 - fix very unlikely obd_types race condition (501)
4771 - remove osc_open hack for echo_client (1187)
4772 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4773 - a failure in read_inode2 leads to deadlock (1139)
4774 - cancel ack-locks as soon as transaction is committed (1072)
4775 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4776 - make sure to commitrw after any preprw to avoid deadlock (1162)
4777 - failing to execute a file in a lustre FS would lock inode (1203)
4778 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4779 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4780 - obd_uuid2tgt would walk off the end of the list (1255)
4781 - on IA64 the osc would give portals incorrect bulk size (1258)
4782 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4783 - fix lock inversion caused by new llite matching code (1282)
4784 - limit the number of dirty pages on a client to 10MB (1286)
4785 - timed out locks were not being corrected cancelled (1289)
4786 - fix O_DIRECT above 4GB on IA-32 (1292)
4787 * major user-visible changes
4788 - fail out/fail over policy now controlled by the upcall (993)
4790 - add OBD_PING to check server availability and failure (954)
4791 - lustre messages are now sent in sending host order (288, 340, 891)
4792 - add eadatalen to MDS getattr reply (340)
4793 - OST read replies may contain second buffer, with per-page status (593)
4795 2003-03-11 Phil Schwan <phil@clusterfs.com>
4798 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4799 - failed open()s could cause deadlock; fixed (867, 869)
4800 - stop cancelling OST locks when files are closed (481)
4801 - overlapping XID spaces caused network corruption (851, 853)
4802 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4803 - setattr_raw added, to do single-RPC, server-side setattrs
4804 - lmc/lconf syntax change for OST UUIDs
4805 - fix crashy race condition between ptlrpc_free_req and osc_close
4806 - don't use request in mdc_enqueue if we hit a timeout (889)
4807 - don't set the inode i_size for regular files from the MDS (896)
4808 - handle out of order completion AST (842)
4809 - don't LBUG if a lock request times out after receiving AST (913)
4810 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4811 - if a bad lock AST arrives, send an error instead of dropping entirely
4812 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4813 - fix leak in bulk IO when only partially completed (899, 900, 926)
4814 - fix O_DIRECT for ia64 (55)
4815 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4816 - C-z after timeout could hang a process forever; fixed (977)
4818 - client-side I/O cache (678, 924, 929, 941, 970)
4820 - READPAGE and SETATTRs which don't take server-side locks get
4823 2003-02-11 Phil Schwan <phil@clusterfs.com>
4826 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4827 - Fix fsfilt_extN_readpage() to read a full page of directory
4828 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4829 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4830 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4831 - Most intent operations were removed, in favour of a new RPC mode
4832 that does a single RPC to the server and bypasses most of the VFS
4833 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4834 - Aggressively cancel local locks on DLM servers
4835 - mds_reint_unlink sends EA to the client if it's the last nlink.
4836 client uses that EA to unlink OST objects.
4837 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4838 - recursive symlinks were fixed (439)
4839 - fixed NULL deref in DEBUG_REQ
4840 - filter_update_lastobjid no longer calls sync, which annoyed extN
4841 - fixed multi-client small-writes to a single file problem (445)
4842 - fixed mtime updates during file writes (607)
4843 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4844 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4845 - fixed Linux OST size reporting problem (444, 656)
4846 - OST now updates object mtime with writes or setattr (607, 619)
4847 - client verifies file size before zeroing page past EOF (445)
4848 - OST now writes last allocated objid to disk with allocation (108)
4849 - LOV on echo now works (409)
4851 - mds_reint_unlink sends a new buffer, with the EA included. this
4852 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4853 set if a regular file was being unlinked, and it was the last link
4854 - use PtlGet from the target for bulk writes (315)
4855 - OST now updates object mtime with writes or setattr (607, 619)
4856 - LDLM now has a grant-time callback to revalidate locked items, if
4858 - Many MDS operations were reorganized to combat race conditions
4860 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4861 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4863 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4866 - Fully reactivate OST imports after reconnection (512, others)
4867 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4868 - More graceful error handling for truncating on dead OST (515)
4869 - Don't error out unless we're actually accessing dead stripes (474)
4870 - Fix garbage sizes when stripes are missing (410)
4871 - LRU counters were broken, causing constant lock purge (433, 432)
4872 - garbage on read from stripes with failed OSTs (441)
4873 - mark OSCs as active before reconnecting during recovery (438)
4874 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4875 - lfind did not preserve OST order in output (443)
4876 - symlinks cause hung clients, incorrect data (439)
4877 - stop dereferencing request after dropping refcount (457)
4878 - don't LASSERT(spin_is_locked) on non-SMP (455)
4879 - fixes for many rename() bugs
4880 - fstat didn't correctly synchronize attributes (399)
4881 - server must handle lock cancellation during blocking AST prep (487)
4882 - bulk descriptors were free()d too soon (511)
4883 - fix paths in lconf, which would load incorrect modules (451, 507)
4884 - fix confusing lconf 'host not found' error message (386)
4885 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4886 - fix race condition in mdc_blocking_ast() for inode access (526)
4887 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4888 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4889 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4890 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4891 - fix lstripe in several ways (kernel side) (527)
4892 - fix request leak in ldlm_cli_enqueue (262)
4893 - incorrect OSC was marked inactive after OST failure
4894 - call mds_fs_cleanup before unmounting filesystem (524)
4895 - fix races between taking ns_lock and ldlm_lock_change_resource
4896 - fix races updating LOV export open file list
4897 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4898 - fix recovery NULL deref in ldlm_cli_cancel_unused
4899 - fix some DLM races by using new hash table for lock handles (419)
4900 - permit the client to specify desired inodes, at replay
4901 - duplicate requests when we queue them for replay reintegration
4902 - fix last_rcvd offset calculation
4903 - sync after each recovered transaction, so we always make progress
4904 - never, not always, ERESTART requests without transnos
4905 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4907 - skip replay if the MDS says that the client is already connected
4908 - don't check for a recovery-enabled export to match lctl's UUID
4909 - don't INC_USE_COUNT for phantom exports
4910 - don't crash when cleaning up phantom exports (567)
4911 - don't double-finish or set replay data for errored mdc_open requests
4912 - abort requests when they time out, so we don't get old replies
4913 - send/receive replies for AST messages again
4914 - if the client says that it doesn't have the lock, cancel it on the
4916 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4917 mark it as destroyed, it will all work out when decref is called
4918 - fix module use counts (22, 581)
4920 - ASTs now expect a reply (server cancels lock on error reply)
4922 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4925 - fix many simultaneous client startup (392)
4926 - fix dentry->d_it clobbering
4927 - credentials weren't being shipped for readdir/getattr operations
4928 - remove invalid assertions triggered during some concurrent MD
4930 - proper Lustre versions added (336, 389)
4931 - fix memory leak for create error case (398)
4932 - fix LOV locking bug that would get cli/srv out of sync
4933 - fix echo client over LOV (409)
4934 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4935 - fix double-O_EXCL intent crash (424)
4936 - avoid sending multiple lock CANCELs (352)
4938 - MDS can do multi-client recovery (modulo bugs in new code)
4940 - many updates, edits, cleanups
4942 2002-11-18 Phil Schwan <phil@clusterfs.com>
4945 - fix null d_it dereference (346)
4946 - fix full OST/dbench hang (333)
4947 - fix permission problem with file removal (286)
4948 - fix removal of OSCs from LOV when they fail
4949 - fix NULL deref during bulk timeout (214)
4950 - fix problems related to multiple filesystems on one MDS (241)
4951 - fixed serious subtle metadata locking bugs
4952 - free locks on clients when inodes are removed due to memory
4954 - fix inode pointer in lock data (285)
4955 - partial support for multiple MDS on a single host (241)
4956 - data locks weren't cancelled at clear_inode time (290, 311)
4957 - intent locks could lead to unbounded lock growth (205)
4958 - added a maximum lock count, an LRU list, and a flusher
4959 - fix multiple rename (365)
4960 - properly abstracted the echo client
4961 - OSC locked 1 byte too many; fixed
4962 - rewrote brw callback code:
4963 - fixed recovery bugs related to LOVs (306)
4964 - fixed too-many-pages-in-one-write crash (191)
4965 - fixed (again) crash in sync_io_timeout (214)
4966 - probably fixed callback-related race (385)
4968 - Add capability to MDS protocol
4969 - LDLM cancellations and callbacks on different portals
4971 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4974 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4975 - defer open object destruction to close time (601981, 138)
4976 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4977 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4978 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4979 - don't call obd_setattr in ll_file_release for destroyed objects
4980 * protocol change to lustre_msg: move |version| and add |flags|
4981 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4982 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4983 * added replay of create, unlink, link and rename operations during
4984 MDS failover; recovery should be much more robust now
4985 * remove failed OSCs from LOVs (only lov_create uses this so far)
4986 * the lustre-HOWTO was brought (more) up to date (582544)
4988 2002-10-23 Phil Schwan <phil@clusterfs.com>
4991 - in-use dentries weren't being reused properly (617851)
4992 - prevent multiple LDLM setup (599178)
4993 - fix LOV size calculations for truncate (617853)
4994 - fix client handling of MDS intent errors (POSIX)
4995 - fix permission bug in lovstripe.c test (624321)
4996 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4997 - truncate past end of file could corrupt data
4998 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4999 - a race in recovery could return ETIMEDOUT to apps (623947)
5000 - building outside the source directory was fixed
5001 * the lustre-HOWTO was brought (more) up to date (582544)
5002 * major progress was made on recovery functionality
5004 2002-10-10 Phil Schwan <phil@clusterfs.com>
5007 - recovery deadlock fix
5008 - rm -rf causes LBUG fix (617817)
5009 - file open by multiple tasks fix (618962)
5010 - directory permissions bugs (602707 and 620007)
5011 - journal_stop fixed with locking (611313)
5012 - O_APPEND failures resolved (618273, perhaps 614459)
5013 - lconf PATH fix (619770)
5014 - IA64 build fix (621450)
5015 - RPC buffer sizes scale with amount of memory
5017 2002-10-01 Phil Schwan <phil@clusterfs.com>
5020 - locks would be cancelled without throwing away data pages,
5021 resulting in inconsistent data (605627)
5022 - inode attributes were not always being refreshed (605627, 612449)
5023 - lconf now continues to cleanup after lctl reports an error
5024 - MDS now enforces user permissions (602707)
5025 - lprocfs cleanup fixed, but not yet enabled (614157)
5026 - fixed infinite server hang, should a client not respond to an AST
5027 - avoid going into recovery if user calls readlink() with a buffer
5028 that's too small (613941)
5029 - AST RPCs no longer require replies (614867) -- this may be changed
5030 - don't crash server if client sends an IOV that's too big (611336)
5031 - fixed lock conversion deadlock (611892)
5032 - fixed the following of symlinks (614622)
5033 * recovery: the server can remove locks from a client that dies, other
5034 clients can make progress
5035 * more extN patch fixes
5036 * compile-time configurable ptlrpc buffer allocations
5038 - collaborative read cache document
5039 - Lustre Lite Performance CDR document-in-progress
5041 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5044 - fix typo in patch-2.4.18
5046 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5049 - clear ptlrpc request each time in handle_incoming_request()
5050 - unlink of files now destroys the object on the OST
5052 2002-09-19 Peter Braam <braam@clusterfs.com>
5054 * add hard link support
5055 * change obdfile creation method
5056 * kernel patch changed
5058 2002-09-19 Peter Braam <braam@clusterfs.com>
5061 - stack overflow bug in extN fixed
5063 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
5065 * documentation updates
5066 - add man pages for config tools
5067 - update tests/README to describe testing with new config tools
5068 - finish metadata API descriptions
5069 * bug fixes and cleanups
5070 - statfs workaround for 16TB limit
5071 - LOV stripe allocation improved, can stripe on subset of OSTs
5072 - LOV file size/IO offset was wrong for files > 4GB in size
5073 - object EA data was being dropped, caused files to be unreadable
5074 - memory overflow with non-LOV OST caused memory corruption
5075 - fixed regression tests to work with new config tools, obdfilter
5076 - fixed bug when directory size became larger than 1 block
5077 - fixed bug (for single client case) when PWD was deleted
5078 - invalidate local directory pages when doing intent-based ops
5079 - avoid LDLM oops when lock callback contained bad data
5081 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5083 * documentation updates
5084 * bug fixes and cleanups
5085 - configuration tools
5088 - 64-bit compile warnings
5089 - 64-bit internal statfs data
5091 * test_brw on persistent OST devices
5093 * lprocfs (disabled)
5095 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5097 * documentation updates
5098 * bug fixes and cleanups
5099 * configuration tools
5101 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5104 * many small fixes to 0_5_4
5105 * io/network handling
5106 * thinkos in MDS operations
5108 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5111 * crucial basic fixes to 0.5.3
5112 * IOR, Iozone work over Elan
5115 2002-08-07 Phil Schwan <phil@clusterfs.com>
5116 * version 0_5_3, our first alpha
5117 * we use the new Portals iovs
5118 * documentation updates
5119 * bug fixes and cleanups
5120 * small changes in the DLM wire protocol
5122 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5123 * version 0_5_1 with some initial stability,
5124 * locking on MD and file I/O.
5125 * documentation updates
5126 * several bug fixes since 0.5.0
5127 * small changes in wire protocol
5129 2002-07-18 Phil Schwan <phil@clusterfs.com>
5131 * delivered as Lustre Light Alpha
5132 * fixed a crash after handling invalid MDS requests
5133 * fixed directory pages for architectures with non-4k pages sizes
5135 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5136 * release version v0_4_4
5137 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5139 * Fixes a number of interruption problems with OST operations.
5140 * Update documentation for portals header changes
5141 * Move all wire protocol structs/defines to lustre_idl.h
5142 * Fixes symlink length bug.
5143 * Add tcpdump to repository.
5145 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5146 * release version v0_4_3
5147 * Fixes statfs for inodes on extN.
5148 * Fixes bug in runtests which would delete /etc/hosts.
5149 * Use 64-bit object IDs wherever possible (not into VFS though)
5150 Remove ost_get_info, which is unused by lustre, and out of date.
5152 2002-07-03 Peter Braam <braam@clusterfs.com>
5153 * release version v0_4_2 Fixes a lookup error (type not passed)
5154 * move forward to head of Portals
5155 * move forward to latest Lustre kernel
5157 2002-06-25 Peter Braam <braam@clusterfs.com>
5158 * release version v0_4_1. Hopefully stable on single node use.