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).
18 Description: replace ptlrpcd with the statahead thread to interpret the async
19 statahead RPC callback
23 Frequency : on recovery
24 Description: I/O failures after umount during fail back
25 Details : if client reconnected to restarted server we need join to recovery
26 instead of find server handler is changed and process self eviction
27 with cancel all locks.
29 Severity : enhancement
31 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
33 Severity : enhancement
35 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
37 Severity : enhancement
39 Description: Upcall on Lustre log has been dumped
40 Details : Allow for a user mode script to be called once a Lustre log has
41 been dumped. It passes the filename of the dumped log to the
42 script, the location of the script can be specified via
43 /proc/sys/lnet/debug_log_upcall.
48 Description: avoid messages about idr_remove called for id which is not allocated.
49 Details : Move assigment s_dev for clustered nfs to end of initialization, for
50 avoid problem with error handling.
55 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
56 Details : When connection is reused this not moved from CONN_UNUSED_HASH into
57 CONN_USED_HASH and this prodice warning when put connection again
60 Severity : enhancement
62 Description: Export bytes_read/bytes_write count on OSC/OST.
66 Description: Early reply size mismatch, MGC loses connection
67 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
68 the connect flags are properly negotiated.
72 Description: Properly propagate oinfo flags from lov to osc for statfs
73 Details : restore missing copy oi_flags to lov requests.
75 Severity : enhancement
77 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
78 : l_getgroups(8), lst(8), routerstat(8)
79 Details : included man pages for llobdstat(8), llstat(8),
80 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
82 Severity : enhancement
84 Description: configure's --enable-quota should check the
85 : kernel .config for CONFIG_QUOTA
86 Details : configure is terminated if --enable-quota is passed but
87 : no quota support is in kernel
91 Description: Remove portals compatibility
92 Details : Remove portals compatibility, not interoperable with releases
97 Description: Resolve device initialization race
98 Details : Prevent proc handler from accessing devices added to the
99 obd_devs array but yet be intialized.
101 Severity : enhancement
103 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
105 Severity : enhancement
107 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
112 Description: Using wrong pointer in osc_brw_prep_request
113 Details : Access to array[-1] can produce panic if kernel compiled with
114 CONFIG_PAGE_ALLOC enabled
118 Description: Client runs out of low memory
119 Details : Consider only lowmem when counting initial number of llap pages
123 Description: *optional* service tags registration
124 Details : if the "service tags" package is installed on a Lustre node
125 When the filesystem is mounted, a local-node service tag will
126 be created. See http://inventory.sun.com/ for more information
127 about the Service Tags asset management system.
131 Description: Kernel BUG tries to release flock
132 Details : Lustre does not destroy flock lock before last reference goes
133 away. So always drop flock locks when client is evicted and
134 perform unlock regardless of successfulness of speaking to MDS.
138 Description: add recount protection for osc callbacks, so avoid panic on shutdown
142 Description: sanity test 65a fails if stripecount of -1 is set
143 Details : handle -1 striping on filesystem in ll_dirstripe_verify
148 Description: ASSERTION(CheckWriteback(page,cmd)) failed
149 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
152 Severity : enhancement
154 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
158 Description: do not process already freed flock
159 Details : flock can possibly be freed by another thread before it reaches
160 to ldlm_flock_completion_ast.
164 Description: LBUG during stress test
165 Details : Need properly lock accesses the flock deadlock detection list.
169 Description: oops in page fault handler
170 Details : kernel page fault handler can return two special 'pages' in error case, don't
171 try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
175 Description: timeout with invalidate import.
176 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which should be
177 handled by ptlrpcd. This produce long age waiting and -ETIMEOUT
178 ptlrpc_invalidate_import and as result LASSERT.
180 Severity : enhancement
182 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
186 Description: enable MGS and MDT services start separately
187 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
188 a co-located MGS without starting the MGS, which is a complement
189 to 'nosvc' mount option.
193 Frequency : after recovery
194 Description: precreate to many object's after del orphan.
195 Details : del orphan st in oscc last_id == next_id and this triger growing
196 count of precreated objects. Set flag LOW to skip increase count
197 of precreated objects.
201 Frequency : rare, on clear nid stats
202 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
203 Details : when clean nid stats sometimes try destroy live entry,
204 and this produce panic in free.
208 Description: Stack overflow during MDS log replay
209 ease stack pressure by using a thread dealing llog_process.
213 Description: wait until IO finished before start new when do lock cancel.
214 Details : VM protocol want old IO finished before start new, in this case
215 need wait until PG_writeback is cleared until check dirty flag and
216 call writepages in lock cancel callback.
218 Severity : enhancement
220 Description: using special macro for print time and cleanup in includes.
224 Description: mds_mfd_close() ASSERTION(rc == 0)
225 Details : In mds_mfd_close(), we need protect inode's writecount change
226 within its orphan write semaphore to prevent possible races.
230 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
234 Frequency : rare, on shutdown ost
235 Description: don't hit live lock with umount ost.
236 Details : shrink_dcache_parent can be in long loop with destroy dentries,
237 use shrink_dcache_sb instead.
241 Description: don't panic with use echo client
242 Details : echo client pass NULL as client nid pointer and this produce null
247 Description: fix build on ppc32
248 Details : compile code with -m64 flag produce wrong object file for ppc32.
252 Description: add message levels for liblustreapi
256 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
257 Details : Move the 'good_osts' check before the 'total_bavail' check. This
258 will result in an -EAGAIN and in the exit call path we call
259 alloc_rr() which will with increasing aggressiveness attempt to
260 aquire precreated objects on the minimum number of required OSCs.
264 Description: Use old size assignment to avoid deadlock
265 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
266 the scheduling while holding a spinlock. We do not need locking
267 for size in ll_update_inode() because size is only updated from
268 the MDS for directories or files without objects, so there is no
269 other place to do the update, and concurrent access to such inodes
270 are protected by the inode lock.
274 Description: resolve "_IOWR redefined" build error on SLES10
278 Description: dump the memory debugging after all modules are unloaded to
279 suppress false negative in conf_sanity test 39
281 Severity : enhancement
283 Description: build kernel-ib packages for OFED 1.3 in our release cycle
288 Description: fix SLES kernel versioning
289 Details : the kernel version for our SLES 10 kernel did not include a "-"
290 before the "smp" at the end. while this was not a problem in
291 general, it did mean that software trying to use the kernel
292 version to try to detect a vendor specific kernel would fail.
293 this was most evident by the OFED build scripts.
297 Description: Don't update lov_desc members until making sure they are valid
298 Details : When updating lov_desc members via proc fs, need fix their
299 validities before doing the real update.
303 Description: don't put request into delay list while invalidate in flight.
304 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
305 import in flight. this produce timeout for invalidate and sometimes
306 can cause stale data.
310 Frequency : on ppc only
311 Description: not convert ost objects for directory because it's not exist.
312 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
316 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
318 Severity : enhancement
320 Description: Update to support 2.6.22.14 vanilla kernel.
324 Frequency : rare, on recovery
325 Description: read procfs can produce deadlock in some situation
326 Details : Holding lprocfs lock which send rpc can produce block for destroy
327 obd objects and this also block reconnect with -EALREADY. This isn't
328 fix all lprocfs bugs - but make it rare.
330 Severity : enhancement
332 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
335 Frequency : frequent on X2 node
337 Description: mdc_set_open_replay_data LBUG
338 Details : Set replay data for requests that are eligible for replay.
342 Description: lustre_mgs: operation 101 on unconnected MGS
343 Details : When MGC is disconnected from MGS long enough, MGS will evict the
344 MGC, and late on MGC cannot successfully connect to MGS and a lot
345 of the error messages complaining that MGS is not connected.
349 Frequency : on network error
350 Description: panic with double free request if network error
351 Details : mdc_finish_enqueue is finish request if any network error ocuring,
352 but it's true only for synchronus enqueue, for async enqueue
353 (via ptlrpcd) this incorrect and ptlrpcd want finish request
356 Severity : enhancement
358 Description: client-side metadata stat-ahead during readdir(directory readahead)
359 Details : perform client-side metadata stat-ahead when the client detects
360 readdir and sequential stat of dir entries therein
363 Frequency : on start mds
365 Description: Implement get_info(last_id) in obdfilter.
368 Frequency : occasional
370 Description: Correctly check stale fid, not start epoch if ost not support SOM
371 Details : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
372 overwrited with new generated fid, but mds can anwer with one of these
373 two fids and both is not stale. setattr incorectly start epoch and
374 assume will be called done_writeting, but without SOM done_writing
378 Frequency : rare, depends on device drivers and load
380 Description: MDS or OSS nodes crash due to stack overflow
381 Details : Code changes in 1.8.0 increased the stack usage of some functions.
382 In some cases, in conjunction with device drivers that use a lot
383 of stack the MDS (or possibly OSS) service threads could overflow
384 the stack. One change which was identified to consume additional
385 stack has been reworked to avoid the extra stack usage.
388 Frequency : occasional
390 Description: Do not fail import if osc_interpret_create gets -EAGAIN
391 Details : If osc_interpret_create got -EAGAIN it immediately exits and
392 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
393 oscc_has_objects and see OSC has no objests and call
394 oscc_internal_create to resend create request.
396 Severity : enhancement
398 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
400 Severity : enhancement
402 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
405 Frequency : very rare
407 Description: Fix replay if there is an un-replied request and open
408 Details : In some cases, older replay request will revert the
409 mcd->mcd_last_xid on MDS which is used to record the client's
412 Severity : enhancement
414 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
416 Severity : enhancement
418 Description: Add rhel5 support to HEAD.
420 Serverity : enhancement
422 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
427 Description: Don't allow skipping OSTs if index has been specified.
428 Details : Don't allow skipping OSTs if index has been specified, make locking
429 in internal create lots better.
433 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
434 Details : access to bitfield in structure is always rounded to long
435 and this produce problem with not atomic change any bit.
439 Description: Lustre make rpms failed.
440 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
441 builds Lustre rpms from tarball.
444 Frequency : rare on shutdown ost
446 Description: If llog cancel was not send before clean_exports phase, this can
447 produce deadlock in llog code.
448 Details : If llog thread has last reference to obd and call class_import_put
449 this produce deadlock because llog_cleanup_commit_master wait when
450 last llog_commit_thread exited, but this never success because was
451 called from llog_commit_thread.
455 Description: allow userland application know is lost one of stripes.
456 Details : fill lvb_blocks with error code on ost and return it to
457 application if error flag found.
461 Description: NULL lov_tgts causing MDS oops
462 Details : more safe checks for NULL lov_tgts for avoid oops.
464 Severity : enhancement
466 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
470 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
471 Details : If an OST is down the MDS will hang indefinitely in
472 obd_statfs_rqset() waiting for the statfs data. While for
473 MDS QOS usage of statfs, it should not stuck in waiting.
475 Severity : enhancement
477 Description: remote_acl support
478 Details : Support ACL-based permission check for remote user.
479 Support setfacl/getfacl for remote user with the utils
480 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
481 the system "{s,g}etfacl" utils.
483 Severity : enhancement
485 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
487 Severity : enhancement
489 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
493 Description: make mgs_setparam() handle fsname containing dash
494 Details : fsname containing a dash does not work with lctl conf_param
496 Severity : enhancement
498 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
500 Severity : enhancement
502 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
506 Frequency : rare, at shutdown
507 Description: access already free / zero obd_namespace.
508 Details : if client_disconnect_export was called without force flag set,
509 and exist connect request in flight, this can produce access to
510 NULL pointer (or already free pointer) when connect_interpret
511 store ocd flags in obd_namespace.
515 Frequency : only at startup
516 Description: not alloc memory with spinlock held.
517 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
518 if any spinlock held.
520 Severity : enhancement
522 Description: make lustre randomly fail allocating memory
523 Details : Make lustre randomly failed allocating memory for testing purpose.
525 Severity : enhancement
527 Description: lost problems with lov objid file
528 Details : Fixes some scability and access to not inited memory problems
529 in work with lov objdid file.
534 Description: lfs find does not continue on file error
535 Details : Continue other files processing when a file/dir is absent.
539 Description: Inconsistent usage of lustre_pack_reply()
540 Details : Standardize the usage of lustre_pack_reply() such that it
541 always generate a CERROR on failure.
544 Frequency : occasional
546 Description: MDS hang or stay in waiting lock
547 Details : If client receive lock with CBPENDING flag ldlm need send lock
548 cancel as separate rpc, to avoid situation when cancel request
549 can't processed due all i/o threads stay in wait lock.
553 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
557 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
561 Description: Update to SLES9 kernel-2.6.5-7.287.3.
565 Description: Update to RHEL4 latest kernel.
567 Severity : enhancement
569 Description: Build SLES10 patchless client fails
570 Details : The configure was broken by run ./configure with
571 --with-linux-obj=.... argument for patchless client. When the
572 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
573 can't search header adequately. Use absolute path such as
574 -I($LINUX)/include instead.
578 Description: interrupt oig_wait produce painc on resend.
579 Details : brw_redo_request can be used for resend requests from ptlrpcd and
580 private set, and this produce situation when rq_ptlrpcd_data not
581 copyed to new allocated request and triggered LBUG on assert
582 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
583 ptlrpcd set if request is changed and can be safety changed to use
588 Description: LASSERT_{REQ,REP}SWAB macros are buggy
589 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
590 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
591 these macros and replace them with their difinitions instead.
595 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
596 Details : Update which_patch & target file for SLES10 latest kernel.
599 Description: Remove client patches from SLES 10 kernel.
600 Details : This causes SLES 10 clients to behave as patchless clients
601 even on a Lustre-patched (server) kernel.
603 Severity : enhancement
605 Description: self-adjustable client's lru lists
606 Details : use adaptive algorithm for managing client cached locks lru
607 lists according to current server load, other client's work
608 pattern, memory activities, etc. Both, server and client
609 side namespaces provide number of proc tunables for controlling
612 Severity : enhancement
614 Description: light-weight GSS support
615 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
616 and reduce performance overhead.
618 Severity : enhancement
620 Description: Linux keyring support
621 Details : Support using service of Linux keyring for Lustre GSS internal
622 context refresh/cache mechanism.
626 Description: Fix errors in lfs documentation
627 Details : Fixes man pages
631 Description: don't use GFP_* in generic Lustre code.
632 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
636 Description: obdclass is limited by single OBD_ALLOC(idarray)
637 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
641 Description: warnings with build patchless client with vanila 2.6.19 and up
642 Details : change old ctl_table style and replace ctl_table/ctl_table_header
643 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
647 Frequency : Only for SLES
648 Description: sanity_quota fail test_1
649 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
650 file, which makes TSTID[2] inunique.
654 Frequency : Only for SLES9
655 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
656 Details : Update target/ChangeLog/which_patch .
660 Frequency : Only for test_10 in sanity-quota.sh
661 Description: a bug in quota test code
662 Details : checking a test flag in a wrong place
666 Frequency : Only for test_13 in sanity-quota.sh
667 Description: a bug in quota test script
668 Details : shouldn't check overfull allocated quota
672 Description: jbd statistics
673 Details : Port older jbd statistics patch for sles10
675 Severity : enhancement
677 Description: Add printing inode info into message about error in writepage.
682 Description: reply_lock_interpret crash due to race with it and lock cancel.
683 Details : Do not replay locks that are being cancelled. Do not reference
684 locks by their address during replay, just by their handle.
688 Frequency : When flocks are used.
689 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
690 Details : Flock locks might destroy just granted lock if it could be merged
691 with another existing flock, this is done in completion handler,
692 so teach ldlm_cli_enquque_fini that this is a valid case for
698 Description: Oops in read and write path when failing to allocate lock.
699 Details : Check if lock allocation failed and return error back.
703 Description: lstripe command fails for valid OST index
704 Details : The stripe offset is compared to lov->desc.ld_tgt_count
705 instead of lov->desc.ld_active_tgt_count.
709 Description: sanity.sh failed test 103
710 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
715 Description: df doesn't work properly if diskfs blocksize != 4K
716 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
720 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
721 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
723 Severity : enhancement
725 Description: Adaptive timeouts
726 Details : RPC timeouts adapt to changing server load and network
727 conditions to reduce resend attempts and improve recovery time.
731 Description: llapi_file_create() does not allow some changes
732 Details : add llapi_file_open() that allows specifying the mode and
733 open flags, and also returns an open file handle.
737 Description: merge and cleanup kernel patches.
738 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
742 Description: Add journal checksum support.(Kernel part)
743 Details : The journal checksum feature adds two new flags i.e
744 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
745 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
746 indicates that the commit block contains the checksum for
747 the blocks described by the descriptor blocks. Now commit
748 record can be sent to disk without waiting for descriptor
749 blocks to be written to disk. This behavior is controlled
750 using JBD2_FEATURE_ASYNC_COMMIT flag.
754 Description: OSS needs mutliple precreate threads
755 Details : Add ability to start more than one create thread per OSS.
759 Description: Sanity.sh test_65k failed: setstripe should have succeeded
760 Details : execute lfs setstripe on client
764 Description: mds_obd_create error creating tmp object
765 Details : When the user sets quota on root, llog will be affected and can't
766 create files and write files.
769 Frequency : Always on ia64 patchless client, and possibly others.
771 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
772 Details : This allows the patchless client to be loaded on architectures
777 Description: RedHat Update kernel for RHEL5
778 Details : Add the kernel config file for RHEL5.
782 Description: RedHat Update kernel for RHEL5
783 Details : Modify the kernel config file more closer RHEL5.
787 Description: Build failure against Centos5 (RHEL5)
788 Details : Define PAGE_SIZE when it isn't present.
792 Description: test_7 on sanity-quota.sh can't be used on separate machines
793 Details : add facet to handle it
797 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
798 Details : with NFS, the anon dentry's parent was set to itself in
799 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
800 corresponding dentry other than use rec->ur_name.
802 Severity : enhancement
804 Description: lfs setstripe enhancement
805 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
809 Description: enable data checksumming by default
810 Details : enable checksum by default, allow --disable-checksum
811 configure option and "-o nochecksum" mount option. Checksums
812 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
816 Description: lustre support for RHEL5
817 Details : Add support for RHEL5.
821 Description: Client eviction due to failover config
822 Details : after a connection loss, the lustre client should attempt to
823 reconnect to the last active server first before trying the
824 other potential connections.
828 Description: when mds and osts use different quota unit(32bit and 64bit),
829 quota will be released repeatly.
830 Details : void sending multiple quota reqs to mds, which will keep the
831 status between the reqs.
835 Description: osts not allocated evenly to files
836 Details : change the condition to increase offset_idx
839 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
841 Description: Data corruption for OSTs that are formatted larger than 2TB
843 Details : When generating the bio request for lustre file writes the
844 sector number would overflow a temporary variable before being
845 used for the IO. The data reads correctly from Lustre (which
846 will overflow in a similar manner) but other file data or
847 filesystem metadata may be corrupted in some cases.
851 Description: Tune the kernel for good SCSI performance.
852 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
853 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
858 Description: rewrite ext2-derived code in obdclass/uuid.c
859 Details : rewrite inherited code (uuid parsing code from ext2 utils)
860 from scratch preserving functionality.
864 Description: "lfs find -obd UUID" prints directories
865 Details : "lfs find -obd UUID" will return all directory names instead
866 of just file names. It is incorrect because the directories
867 do not reside on the OSTs.
871 Description: use i_size_read and i_size_write in 2.6 port
872 Details : replace inode->i_size access with i_size_read/write()
875 Frequency : when removing large files
877 Description: scheduling issue during removal of large Lustre files
878 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
879 It causes scheduling issues when removing large files (17TB in the
883 Frequency : only with liblustre clients on XT3
885 Description: evictions taking too long
886 Details : allow llrd to evict clients directly on OSTs
889 Frequency : only on ppc
891 Description: /proc/fs/lustre/devices broken on ppc
892 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
893 We should make sure the type of 'index' is loff_t and then cast
894 explicitly as needed below. Do not assign an explicitly cast
898 Frequency : only for rhel5
900 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
901 Details : Modify the target file & which_kernel.
903 Severity : enhancement
905 Description: omit set fsid for export NFS
906 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
907 as FS_REQUIRES_DEV for avoid problems with generate fsid.
911 Frequency : Always, for kernels after 2.6.16
912 Description: Fix warning idr_remove called for id=.. which is not allocated.
913 Details : Last kernels save old s_dev before kill super and not allow
914 to restore from callback - restore it before call kill_anon_super.
918 Description: buffer overruns could theoretically occur
919 Details : llapi_semantic_traverse() modifies the "path" argument by
920 appending values to the end of the origin string, and a
921 overrun may occur. Adding buffer overrun check in liblustreapi.
925 Description: Fix error on 'ls .' at the top of the Lustre mount.
926 Details : Don't revalidate dentry if it is a root dentry.
930 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
931 Details : Modify vm-tunables-rhel4.patch.
935 Description: Kernel config for 2.6.18-vanilla.
936 Details : Modify targets/2.6-vanilla.target.in.
937 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
938 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
939 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
940 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
944 Description: improve handling recoverable errors
945 Details : if request processig with error which can be recoverable on server
946 request should be resend, otherwise page released from cache and
951 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
952 Details : Modify target file & which_patch.
953 A flaw was found in the IA32 system call emulation provided
954 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
955 value could be stored in the %RAX register, which could trigger an
956 out-of-bounds system call table access. An untrusted local user
957 could exploit this flaw to run code in the kernel
958 (ie a root privilege escalation). (CVE-2007-4573).
962 Description: O_DIRECT bypasses client statistics.
963 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
964 read_ahead_stats were updated. copied the stats section from
965 osc_send_oap_rpc() into async_internal().
969 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
973 Description: change order of libsysio includes
974 Details : '#include sysio.h' should always come before '#include xtio.h'
978 Description: handle "serious error: objid * already exists" more gracefully
979 Details : If LAST_ID value on disk is smaller than the objects existing in
980 the O/0/d* directories, it indicates disk corruption and causes an
981 LBUG(). If the object is 0-length, then we should use the existing
982 object. This will help to avoid a full fsck in most cases.
984 Severity : enhancement
986 Description: adapt the lustre_config script to support the upgrade case
987 Details : Add "-u" option for lustre_config script to support upgrading 1.4
988 server targets to 1.6 in parallel.
992 Description: To avoid grant space > avaible space when the disk is almost
993 full. Without this patch you might see the error "grant XXXX >
994 available" or some LBUG about grant, when the disk is almost
996 Details : In filter_check_grant, for non_grant cache write, we should
997 check the left space by if (*left > ungranted + bytes), instead
998 of (*left > ungranted), because only we are sure the left space
999 is enough for another "bytes", then the ungrant space should be
1000 increase. In client, we should update cl_avail_grant only there
1001 is OBD_MD_FLGRANT in the reply.
1005 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1006 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1007 described in CVE-2007-4573. This problem could allow untrusted
1008 local users to gain root access.
1011 Frequency : when using O_DIRECT and quotas
1013 Description: Incorrect file ownership on O_DIRECT output files
1014 Details : block usage reported by 'lfs quota' does not take into account
1015 files that have been written with O_DIRECT.
1020 Description: touch file failed when fs is not full
1021 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1022 otherwise we can get ENOSP while fs is not full.
1026 Description: parallel lock callbacks
1027 Details : Instead of sending blocking and completion callbacks as separated
1028 requests, adding them to a set and sending in parallel.
1031 Frequency : only for Cray XT3
1032 Bugzilla : 12829/13455
1033 Description: Changing primary group doesn't change the group lustre assigns to
1035 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1036 overridden with the primary group provided by the group upcall,
1037 whereas the supplied fsgid can be trusted if it is in the list of
1038 supplementary groups returned by the group upcall.
1040 Severity : enhancement
1042 Description: Allow masking D_WARNING, D_ERROR messages from console
1043 Details : Console messages can now be disabled via lnet.printk.
1047 Description: User code with malformed file open parameter crashes client node
1048 Details : Before packing join_file req, all the related reference should be
1049 checked carefully in case some malformed flags cause fake
1050 join_file req on client.
1054 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1055 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1056 mode is equal to the granted mode. However, it should also check
1057 that LDLM_FL_CP_REQD is not set.
1061 Description: Heavy nfs access might result in deadlocks
1062 Details : After ELC code landed, it is now improper to enqueue any mds
1063 locks under och_sem, because enqueue might want to decide to
1064 cancel open locks for same inode we are holding och_sem for.
1068 Description: Client eviction while running blogbench
1069 Details : A lot of unlink operations with concurrent I/O can lead to a
1070 deadlock causing evictions. To address the problem, the number of
1071 oustanding OST_DESTROY requests is now throttled to
1072 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1073 callbacks are processed in priority.
1077 Description: enable ACLs on MDS by default
1078 Details : ACLs must be enabled on MDS by default.
1081 Frequency : PPC/PPC64 only
1083 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1084 Details : fix duplicated definitions between asm-ppc64/types.h and
1085 lustre_types.h on PPC.
1088 Frequency : PPC/PPC64 only
1090 Description: asm-ppc/segment.h does not exist
1091 Details : fix compile issue on PPC.
1095 Description: better handle error messages in extents code
1098 Frequency : RHEL4 only
1100 Description: mkfs is very slow on IA64/RHEL4
1101 Details : A performance regression has been discovered in the MPT Fusion
1102 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1103 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1104 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1106 Severity : enhancement
1108 Description: SNMP support enhancement
1109 Details : Adding total number of sampled request for an MDS node in snmp
1112 Severity : enhancement
1114 Description: Optimize ldlm waiting list processing for PR extent locks
1115 Details : When processing waiting list for read extent lock and meeting read
1116 lock that is same or wider to it that is not contended, skip
1117 processing rest of the list and immediatelly return current
1118 status of conflictness, since we are guaranteed there are no
1119 conflicting locks in the rest of the list.
1123 Description: Time out and refuse to reconnect
1124 Details : When the failover node is the primary node, it is possible
1125 to have two identical connections in imp_conn_list. We must
1126 compare not conn's pointers but NIDs, otherwise we can defeat
1127 connection throttling.
1131 Description: port llog fixes from b1_6 into HEAD
1132 Details : Port llog reference couting and some llog cleanups from b1_6
1133 (bug 10800) into HEAD, for protect from panic and access to already
1134 free llog structures.
1138 Description: Detect stride IO mode in read-ahead
1139 Details : When a client does stride read, read-ahead should detect that and
1140 read-ahead pages according to the detected stride pattern.
1144 Description: data checksumming impacts single node performance
1145 Details : add support for several checksum algorithm. Currently, only CRC32
1146 and Adler-32 are supported. The checksum type can be changed on
1147 the fly via /proc/fs/lustre/osc/*/checksum_type.
1151 Description: use adler32 for page checksums
1152 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1157 Description: build for x2 fails
1158 Details : fix compile issue on Cray systems.
1162 Description: Properly match for duplicate locks
1163 Details : Due to different lock order from skiplists code, we need to
1164 traverse entire list for now
1167 Frequency : only on PPC/SLES10
1169 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1170 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1171 be defined. Add a hack in mkfs_lustre.c to work around this compile
1176 Description: LASSERT on MDS when client holding flock lock dies
1177 Details : ldlm pool logic depends on number of granted locks equal to
1178 number of released locks which is not true for flock locks, so
1179 just exclude such locks from consideration.
1183 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1184 Details : Use fsfilt_sync() for both the whole filesystem sync and
1185 individual file sync to eliminate dangerous inode locking
1186 with I_LOCK that can lead to a deadlock.
1190 Description: performance in 1.6.3
1191 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1192 sure that 1MB requests are not fragmented by the block layer.
1194 Severity : enhancement
1196 Description: organize the server-side client stats on per-nid basis
1197 Details : Change the structure of stats under obdfilter and mds to
1205 The "uuid"s file would list the uuids of _active_ exports.
1206 And the clear entry is to clear all stats and stale nids.
1208 Severity : enhancement
1210 Description: eliminate client locks in face of contention
1211 Details : file contention detection and lockless i/o implementation
1212 for contended files.
1216 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1219 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1221 Description: MDS may lose file striping (and hence file data) in some cases
1222 Details : If there are additional extended attributes stored on the MDS,
1223 in particular ACLs, SELinux, or user attributes (if user_xattr
1224 is specified for the client mount options) then there is a risk
1225 of attribute loss. Additionally, the Lustre file striping
1226 needs to be larger than default (e.g. striped over all OSTs),
1227 and an additional attribute must be stored initially in the
1228 inode and then increase in size enough to be moved to the
1229 external attribute block (e.g. ACL growing in size) for file
1234 Description: skiplist implementation simplification
1235 Details : skiplists are used to group compatible locks on granted list
1236 that was implemented as tracking first and last lock of each lock group
1237 the patch changes that to using doubly linked lists
1241 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1242 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1243 because LASSERT on both the data supplied by a client, and the data
1244 on disk is dangerous and incorrect.
1246 Severity : enhancement
1248 Description: Slow truncate/writes to huge files at high offsets.
1249 Details : Directly associate cached pages to lock that protect those pages,
1250 this allows us to quickly find what pages to write and remove
1251 once lock callback is received.
1255 Description: more ldlm soft lockups
1256 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1257 starve other threads from the resource lock for a long time in
1258 case of long waiting queue, so change the debug level from
1259 D_OTHER to the less frequently used D_INFO.
1261 Severity : enhancement
1263 Description: add -gid, -group, -uid, -user options to lfs find
1267 Description: Hung threads in invalidate_inode_pages2_range
1268 Details : The direct IO path doesn't call check_rpcs to submit a new RPC once
1269 one is completed. As a result, some RPCs are stuck in the queue
1274 Description: filter threads hungs on waiting journal commit
1275 Details : Cleanup filter group llog code, then only filter group llog will
1276 be only created in the MDS/OST syncing process.
1280 Description: Procfs and llog threads access destoryed import sometimes.
1281 Details : Sync the import destoryed process with procfs and llog threads by
1282 the import refcount and semaphore.
1284 Severity : enhancement
1286 Description: openlock cache of b1_6 port to HEAD
1291 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1292 Details : If insertion of an extent fails, then discard the inode
1293 preallocation and free data blocks else it can lead to duplicate
1298 Description: don't always update ctime in ext3_xattr_set_handle()
1299 Details : Current xattr code updates the inode ctime in ext3_xattr_set_handle.
1300 In some cases the ctime should not be updated, for example for
1301 2.0->1.8 compatibility it is necessary to delete an xattr and it
1302 should not update the ctime.
1306 Bugzilla : 15713/16362
1307 Description: Assertion in iopen_connect_dentry in 1.6.3
1308 Details : looking up an inode via iopen with the wrong generation number can
1309 populate the dcache with a disconneced dentry while the inode
1310 number is in the process of being reallocated. This causes an
1311 assertion failure in iopen since the inode's dentry list contains
1312 both a connected and disconnected dentry.
1316 Description: assertion failure in ldlm_handle2lock()
1317 Details : fix a race between class_handle_unhash() and class_handle2object()
1318 introduced in lustre 1.6.5 by bug 13622.
1323 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1324 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1329 Description: oops in sd_iostats_seq_show()
1330 Details : unloading/reloading the scsi low level driver triggers a kernel
1331 bug when trying to access the sd iostat file.
1336 Description: Kernel panics during QLogic driver reload
1337 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1338 patch, causing memory corruption.
1343 Description: journal_dev option does not work in b1_6
1344 Details : pass mount option during pre-mount.
1346 Severity : enhancement
1348 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1349 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1350 extent information of a file. It can be used to map logical blocks
1351 in a file to physical blocks in the block device.
1355 Description: LDLM soft lockups - improvement
1356 Details : It is be possible to send the lock handle along with each read
1357 or write request because the client is already doing a lock match
1358 itself so there isn't any reason the OST should have to re-do that
1361 --------------------------------------------------------------------------------
1363 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
1365 * Support for kernels:
1366 2.6.5-7.283 (SLES 9),
1367 2.6.9-55.EL (RHEL 4),
1368 2.6.16.46-0.14 (SLES 10),
1369 2.6.18.8 vanilla (kernel.org)
1370 * Client support for unpatched kernels:
1371 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1372 2.6.16 - 2.6.22 vanilla (kernel.org)
1373 * Due to recently discovered recovery problems, we do not recommend
1374 using patchless RHEL 4 clients with this or any earlier release.
1375 * Recommended e2fsprogs version: 1.39.cfs8
1376 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1377 * Starting with this release, the ldiskfs backing filesystem required
1378 by Lustre is now in its own package, lustre-ldiskfs. This package
1379 should be installed. It is versioned separately from Lustre and
1380 may be released separately in future.
1384 Description: block reactivating mgc import until all deactivates complete
1385 Details : Fix race when failing back MDT/MGS to itself (testing)
1388 Frequency : at statup only
1390 Description: mds_lov_synchronize race leads to various problems
1391 Details : simultaneous MDT->OST connections at startup can cause the
1392 sync to abort, leaving the OSC in a bad state.
1394 Severity : enhancement
1396 Description: add optional extra BUILD_VERSION info
1397 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1398 to override the lustre version.
1401 Frequency : 2.6.18 servers only
1403 Description: ll_kern_mount() doesn't release the module reference
1404 Details : The ldiskfs module reference count never drops down to 0
1405 because ll_kern_mount() doesn't release the module reference.
1410 Description: server LBUG when using old ost_num_threads parameter
1411 Details : Accept the old ost_num_threads parameter but warn that it
1412 is deprecated, and fix an off-by-one error that caused an LBUG.
1417 Description: Transient SCSI error results in persistent IO issue
1418 Details : iobuf->dr_error is not reinitialized to 0 between two
1422 Frequency : sometimes when underlying device returns I/O errors
1424 Description: OSTs not going read-only during write failures
1425 Details : OSTs are not remounted read-only when the journal commit threads
1426 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1427 instead of the ext3 wrappers.
1431 Description: poor connect scaling with increasing client count
1432 Details : Don't run filter_grant_sanity_check for more than 100 exports
1433 to improve scaling for large numbers of clients.
1436 Frequency : SLES10 only
1438 Description: sanity-quota.sh quotacheck failed: rc = -22
1439 Details : Quotas cannot be enabled on SLES10.
1442 Frequency : liblustre clients only
1444 Description: getdirentries does not give error when run on compute nodes
1445 Details : getdirentries does not fail when the size specified as an argument
1446 is too small to contain at least one entry
1448 Severity : enhancement
1450 Description: Add LNET router traceability for debug purposes
1451 Details : If a checksum failure occurs with a router as part of the
1452 IO path, the NID of the last router that forwarded the bulk data
1453 is printed so it can be identified.
1458 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1459 Details : Due to a race condition, liblustre clients were occasionally
1460 evicted incorrectly.
1462 Severity : enhancement
1464 Description: lfs setstripe use optional parameters instead of postional
1467 Severity : enhancement
1469 Description: Nanosecond timestamp support for ldiskfs
1470 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1471 resolution timestamps. There is not yet support for this at
1472 the Lustre filesystem level.
1475 Frequency : during server recovery
1477 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1478 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1482 Frequency : nfs export on patchless client
1484 Description: connectathon hang when test nfs export over patchless client
1485 Details : Disconnected dentry cannot be found with lookup, so we do not need
1486 to unhash it or make it invalid
1489 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1490 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1491 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1496 Description: open req refcounting wrong on reconnect
1497 Details : If reconnect happened between getting open reply from server and
1498 call to mdc_set_replay_data in ll_file_open, we will schedule
1499 replay for unreferenced request that we are about to free.
1500 Subsequent close will crash in variety of ways.
1501 Check that request is still eligible for replay in
1502 mdc_set_replay_data().
1507 Description: disable writes to filesystem when reading health_check file
1508 Details : the default for reading the health_check proc file has changed
1509 to NOT do a journal transaction and write to disk, because this
1510 can cause reads of the /proc file to hang and block HA state
1511 checking on a healthy but otherwise heavily loaded system. It
1512 is possible to return to the previous behaviour during configure
1513 with --enable-health-write.
1515 Severity : enhancement
1517 Description: 64-bit inode version
1518 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1519 to the inode. This will be required for version-based recovery.
1524 Description: MDS fails to start if a duplicate client export is detected
1525 Details : in some rare cases it was possible for a client to connect to
1526 an MDS multiple times. Upon recovery the MDS would detect this
1527 and fail during startup. Handle this more gracefully.
1529 Severity : enhancement
1531 Description: Add -o localflock option to simulate old noflock
1533 Details : This will achieve local-only flock/fcntl locks
1539 Description: log_commit_thread vs filter_destroy race leads to crash
1540 Details : Take import reference before releasing llog record semaphore
1545 Description: Wrong request locking in request set processing
1546 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1547 imp_delayed_list, in this place should be used imp_lock.
1550 Frequency : when reconnection
1552 Description: Grant Leak when osc reconnect to OST
1553 Details : When osc reconnect ost, OST(filter) should check whether it
1554 should grant more space to client by comparing fed_grant and
1555 cl_avail_grant, and return the granted space to client instead
1556 of "new granted" space, because client will call osc_init_grant
1557 to update the client grant space info.
1560 Frequency : when client reconnect to OST
1562 Description: Grant Leak when osc do resend and replay bulk write
1563 Details : When osc reconnect to OST, OST(filter)should clear grant info of
1564 bulk write request, because the grant info while be sync between
1565 OSC and OST when reconnect, and we should ignore the grant info
1566 these of resend/replay write req.
1571 Description: Grant space more than avaiable left space sometimes.
1572 Details : When then OST is about to be full, if two bulk writing from
1573 different clients came to OST. Accord the avaliable space of the
1574 OST, the first req should be permitted, and the second one
1575 should be denied by ENOSPC. But if the seconde arrived before
1576 the first one is commited. The OST might wrongly permit second
1577 writing, which will cause grant space > avaiable space.
1580 Frequency : when client is evicted
1582 Description: Grant might be wrongly erased when osc is evicted by OST
1583 Details : when the import is evicted by server, it will fork another
1584 thread ptlrpc_invalidate_import_thread to invalidate the
1585 import, where the grant will be set to 0. While the original
1586 thread will update the grant it got when connecting. So if
1587 the former happened latter, the grant will be wrongly errased
1588 because of this race.
1593 Description: Checking Stale with correct fid
1594 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1595 to check whether it is stale, because sometimes, we want the
1596 enqueue happened anyway, and op_data.fid2 will not be initialized.
1598 Severity : enhancement
1600 Description: update patchless client
1601 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1604 Frequency : only with 2.4 kernel
1606 Description: random memory corruption
1607 Details : size of struct ll_inode_info is to big for union inode.u and this
1608 can be cause of random memory corruption.
1613 Description: Memory leak in recovery
1614 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1615 It should also check obd_fail before fsfilt_start, otherwise if
1616 fsfilt_start return -EROFS,(failover mds during mds recovery).
1617 then the req will return with repmsg->transno = 0 and rc = EROFS.
1618 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1619 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1620 be freed no matter whether fsfilt_commit success or not.
1623 Frequency : only with huge count clients
1625 Description: Prevents from taking the superblock lock in llap_from_page for
1627 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1628 for llap_from_page call in ll_removepage prevents from taking the
1629 superblock lock for a soon died page.
1634 Description: Not check open intent error before release open handle
1635 Details : in some rare cases, the open intent error is not checked before
1636 release open handle, which may cause
1637 ASSERTION(open_req->rq_transno != 0), because it tries to release
1638 the failed open handle.
1643 Description: Set cat log bitmap only after create log success.
1644 Details : in some rare cases, the cat log bitmap is set too early. and it
1645 should be set only after create log success.
1649 Description: Accessing a block bevice can re-enable I/O when Lustre is
1650 tearing down a device.
1651 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1656 Description: service threads may hog cpus when there are a lot of requests
1657 Details : Insert cond_resched to give other threads a chance to use some CPU
1662 Description: the cat log was not initialized in recovery
1663 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1664 unlink log on mds will not be initialized until mds post
1665 recovery. And also in mds post recovery, the unlink log will
1666 initialization will be done asynchronausly, so there will be race
1667 between add unlink log and unlink log initialization.
1671 Description: brw_stats were being printed incorrectly
1672 Details : brw_stats were being printed as log2 but all of them were not
1673 recorded as log2. Also remove some code duplication arising from
1674 filter_tally_{read,write}.
1678 Frequency : rare, only in recovery.
1679 Description: ASSERTION(req->rq_type != LI_POISON) failed
1680 Details : imp_lock should be held while iterating over imp_sending_list for
1681 prevent destroy request after get timeout in ptlrpc_queue_wait.
1685 Description: replay-single.sh test 52 fails
1686 Details : A lock's skiplist need to be cleanup when it being unlinked
1687 from its resource list.
1691 Description: Short directio read returns full requested size rather than
1693 Details : Direct I/O operations should return actual amount of bytes
1694 transferred rather than requested size.
1696 Severity : enhancement
1698 Description: metadata RPC reduction (e.g. for rm performance)
1699 Details : decrease the amount of synchronous RPC between clients and servers
1700 by canceling conflicing lock before the operation on the client
1701 and packing thier handles into the main operation RPC to server.
1703 Severity : enhancement
1705 Description: Async OSC create to avoid the blocking unnecessarily.
1706 Details : If a OST has no remain object, system will block on the creating
1707 when need to create a new object on this OST. Now, ways use
1708 pre-created objects when available, instead of blocking on an
1709 empty osc while others are not empty. If we must block, we block
1710 for the shortest possible period of time.
1712 Severity : enhancement
1714 Description: refine locking for avoid write wrong info into lov_objid file
1715 Details : fix possible races with add new target and write/update data in
1718 --------------------------------------------------------------------------------
1720 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1725 Frequency : on some architectures
1727 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1728 Details : Uninitialized flags sometimes cause configuration commands to
1732 Frequency : patchless clients only
1734 Description: missing __iget() symbol export
1735 Details : The __iget() symbol export is missing. To avoid the need for
1736 this on patchless clients the deathrow inode reaper is turned
1737 off, and we depend on the VM to clean up old inodes. This
1738 dependency was during via the fix for bug 12181.
1740 --------------------------------------------------------------------------------
1742 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1744 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1745 INTEROPERATE with older versions automatically. In many cases a
1746 special upgrade step is needed. Please read the
1747 user documentation before upgrading any part of a live system.
1748 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1749 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1750 * WARNING: Lustre configuration and startup changes are required with
1751 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1757 Severity : enhancement
1759 Description: MountConf
1760 Details : Lustre configuration is now managed via mkfs and mount
1761 commands instead of lmc and lconf. New obd types (MGS, MGC)
1762 are added for dynamic configuration management. See
1763 https://mail.clusterfs.com/wikis/lustre/MountConf for
1766 Severity : enhancement
1768 Description: dynamic OST addition
1769 Details : OSTs can now be added to a live filesystem
1771 Severity : enhancement
1773 Description: startup order invariance
1774 Details : MDTs and OSTs can be started in any order. Clients only
1775 require the MDT to complete startup.
1777 Severity : enhancement
1779 Description: parallel, asynchronous orphan cleanup
1780 Details : orphan cleanup is now performed in separate threads for each
1781 OST, allowing parallel non-blocking operation.
1783 Severity : enhancement
1785 Description: optimized stripe assignment
1786 Details : stripe assignments are now made based on ost space available,
1787 ost previous usage, and OSS previous usage, in order to try
1788 to optimize storage space and networking resources.
1790 Severity : enhancement
1792 Description: Permanently set tunables
1793 Details : All writable /proc/fs/lustre tunables can now be permanently
1794 set on a per-server basis, at mkfs time or on a live system.
1796 Severity : enhancement
1798 Description: Lustre message v2
1799 Details : Add lustre message format v2.
1801 Severity : enhancement
1803 Description: client OST exclusion list
1804 Details : Clients can be started with a list of OSTs that should be
1805 declared "inactive" for known non-responsive OSTs.
1809 Description: SPEC SFS validation failure on NFS v2 over lustre.
1810 Details : Changes the blocksize for regular files to be 2x RPC size,
1811 and not depend on stripe size.
1813 Severity : enhancement
1815 Description: Multiple MD RPCs in flight.
1816 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1817 To avoid overly-overloading MDS, introduce a limit on number of
1818 MDS RPCs in flight for a single client and add /proc controls
1819 to adjust this limit.
1821 Severity : enhancement
1823 Description: client read/write statistics
1824 Details : Add client read/write call usage stats for performance
1825 analysis of user processes.
1826 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1827 file access. extents_stats shows chunk size distribution.
1828 extents_stats_per_process show chunk size distribution per
1831 Severity : enhancement
1833 Description: mds statistics
1834 Details : Add detailed mds operations statistics in
1835 /proc/fs/lustre/mds/*/stats.
1839 Description: Failure of copying files with lustre special EAs.
1840 Details : Client side always return success for setxattr call for lustre
1841 special xattr (currently only "trusted.lov").
1846 Description: Refcount LNET uuids
1847 Details : The global LNET uuid list grew linearly with every startup;
1848 refcount repeated list entries instead of always adding to
1851 Severity : enhancement
1853 Description: Dynamic service threads
1854 Details : Within a small range, start extra service threads
1855 automatically when the request queue builds up.
1858 Frequency : mixed-endian client/server environments
1860 Description: mixed-endian crashes
1861 Details : The new msg_v2 system had some failures in mixed-endian
1864 Severity : enhancement
1866 Description: Easy OST removal
1867 Details : OSTs can be permanently deactivated with e.g. 'lctl
1868 conf_param lustre-OST0001.osc.active=0'
1870 Severity : enhancement
1872 Description: MGS proc entries
1873 Details : Added basic proc entries for the MGS showing what filesystems
1876 Severity : enhancement
1878 Description: provide MGS failover
1879 Details : Added config lock reacquisition after MGS server failover.
1881 Severity : enhancement
1883 Description: add Linux 2.4 support
1884 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1888 Description: a large application tries to do I/O to the same resource and dies
1889 in the middle of it.
1890 Details : Check the req->rq_arrival time after the call to
1891 ost_brw_lock_get(), but before we do anything about
1892 processing it & sending the BULK transfer request. This
1893 should help move old stale pending locks off the queue as
1894 quickly as obd_timeout.
1897 Frequency : when an incorrect nid is specified during startup
1899 Description: ptlrpc connect to non-existant node causes kernel crash
1900 Details : LNET can't be re-entered from an event callback, which
1901 happened when we expire a message after the export has been
1902 cleaned up. Instead, hand the zombie cleanup off to another
1905 Severity : enhancement
1907 Description: plain/inodebits lock performance improvement
1908 Details : Grouping plain/inodebits in granted list by their request modes
1909 and bits policy, thus improving the performance of search through
1913 Frequency : only if OST filesystem is corrupted
1915 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1916 Details : for a short time RPCs with bulk IO are in the replay list,
1917 but replay of bulk IOs is unimplemented. If the OST filesystem
1918 is corrupted due to disk cache incoherency and then replay is
1919 started it is possible to trip an assertion. Avoid putting
1920 committed RPCs into the replay list at all to avoid this issue.
1923 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1925 Description: System hang on startup
1926 Details : This bug allowed the liblustre (e.g. catamount) client to
1927 return to the app before handling all startup RPCs. This
1928 could leave the node unresponsive to lustre network traffic
1929 and manifested as a server ptllnd timeout.
1931 Severity : enhancement
1933 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1934 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1935 boolean to control whether to print peer debug info when a
1936 client's RPC times out.
1939 Frequency : only for kernels with patches from Lustre below 1.4.3
1941 Description: Remove old rdonly API
1942 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1945 Frequency : only for devices with external journals
1947 Description: Set external device read-only also
1948 Details : During a commanded failover stop, we set the disk device
1949 read-only while the server shuts down. We now also set any
1950 external journal device read-only at the same time.
1953 Frequency : when upgrading from 1.4 while trying to change parameters
1955 Description: The wrong (new) MDC name was used when setting parameters for
1956 upgraded MDT's. Also allows changing of OSC (and MDC)
1957 parameters if --writeconf is specified at tunefs upgrade time.
1960 Frequency : when setting specific ost indicies
1962 Description: QOS code breaks on skipped indicies
1963 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1964 created with --index need not be sequential.
1966 Severity : enhancement
1968 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1969 Details : The uninit_groups feature works in conjunction with the kernel
1970 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1971 up the pass1 processing of e2fsck. This is a read-only feature
1972 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1973 mount filesystems that have had this feature enabled.
1975 Severity : enhancement
1977 Description: Improve multi-block allocation algorithm to avoid fragmentation
1978 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1979 allocation locality and avoid filesystem fragmentation.
1981 ------------------------------------------------------------------------------
1983 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1985 * Support for kernels:
1986 2.4.21-47.0.1.EL (RHEL 3)
1987 2.6.5-7.283 (SLES 9)
1988 2.6.9-42.0.10.EL (RHEL 4)
1989 2.6.12.6 vanilla (kernel.org)
1990 2.6.16.27-0.9 (SLES 10)
1991 * Recommended e2fsprogs version: 1.39.cfs5
1993 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1997 Frequency : occasional, depends on client load and configuration
1998 Bugzilla : 12181, 12203
1999 Description: data loss for recently-modified files
2001 Details : In some cases it is possible that recently written or created
2002 files may not be written to disk in a timely manner (this should
2003 normally be within 30s unless client IO load is very high).
2004 The problem appears as zero-length files or files that are a
2005 multiple of 1MB in size after a client crash or client eviction
2006 that are missing data at the end of the file.
2008 This problem is more likely to be hit on clients where files are
2009 repeatedly created and unlinked in the same directory, clients
2010 have a large amount of RAM, have many CPUs, the filesystem has
2011 many OSTs, the clients are rebooted frequently, and/or the files
2012 are not accessed by other nodes after being written.
2014 The presence of the problem can be detected by looking at
2015 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2016 smaller than the second (nr_unused) then dirty files will not
2017 be flushed automatically to disk. "sync; sleep 10" should be
2018 run several times on the node before unmounting it to update
2019 Lustre (this is also safe to run on nodes without this problem).
2021 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2022 that can cause this same problem, so customers using that kernel
2023 also need to update the kernel in addition to Lustre. In order
2024 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2026 It is normal that files written just before a client crash (less
2027 than 30s) may not yet have been flushed to disk, even for local
2031 Frequency : frequent on thin XT3 nodes
2033 Description: UUID collision on thin XT3 Linux nodes
2034 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2035 randomly, since we relied on an insufficiently-seeded PRNG.
2040 Description: OSS hangs after "All ost request buffers busy"
2041 Details : A deadlock between quota and journal operations caused OSS
2042 hangs after printing "All ost request buffers busy."
2045 Frequency : always on liblustre builds
2047 Description: Cleanup compiler warnings on liblustre
2050 Frequency : always on liblustre builds on XT3
2052 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2056 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2057 > 32000 subdirectories
2058 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2059 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2060 subdirectory count crosses 32000. This will aid e2fsck to
2061 correctly handle more than 32000 subdirectories.
2064 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2066 Description: System hang on startup
2067 Details : This bug allowed the liblustre (e.g. catamount) client to
2068 return to the app before handling all startup RPCs. This
2069 could leave the node unresponsive to lustre network traffic
2070 and manifested as a server ptllnd timeout.
2072 Severity : enhancement
2074 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2075 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2076 boolean to control whether to print peer debug info when a
2077 client's RPC times out.
2082 Description: make O_SYNC working on 2.6 kernels
2083 Details : 2.6 kernels use different method for mark pages for write,
2084 so need add a code to lustre for O_SYNC work.
2089 Description: Failure to close file and release space on NFS
2090 Details : Put inode details into lock acquired in ll_intent_file_open.
2091 Use mdc_intent_lock in ll_intent_open to properly
2092 detect all kind of errors unhandled by mdc_enqueue.
2097 Description: proc file read during shutdown sometimes raced obd removal,
2099 Details : Add lock to prevent obd access after proc file removal.
2102 Frequency : Only for files larger than 4GB on 32-bit clients.
2104 Description: improperly doing page alignment of locks
2105 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2106 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2111 Description: RPCs being resent when they shouldn't be
2112 Details : Some RPCs that should not be resent are being resent. This
2113 can cause inconsistencies in the RPC state machine. Do not
2114 resend such requests.
2117 Frequency : rare, only with NFS export
2119 Description: Crash on NFS re-export node
2120 Details : under very unusual load conditions an assertion is hit in
2121 ll_intent_file_open()
2124 Frequency : only if OST filesystem is corrupted
2126 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2127 Details : for a short time RPCs with bulk IO are in the replay list,
2128 but replay of bulk IOs is unimplemented. If the OST filesystem
2129 is corrupted due to disk cache incoherency and then replay is
2130 started it is possible to trip an assertion. Avoid putting
2131 committed RPCs into the replay list at all to avoid this issue.
2136 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2137 Details : Large single O_DIRECT read and write calls can fail to allocate
2138 a sufficiently large buffer to process the request. In case of
2139 allocation failure the allocation is retried with a smaller
2140 buffer and broken into smaller requests.
2142 Severity : enhancement
2144 Description: Add -o localflock option to simulate old noflock behaviour.
2145 Details : This will achieve local-only flock/fcntl locks coherentness.
2150 Description: versioning check is incomplete
2151 Details : Checking the version difference of client vs. server, report
2152 error if the gap is too big.
2157 Description: add support PG_writeback bit
2158 Details : add support for PG_writeback bit for Lustre, for more carefull
2159 work with page cache in 2.6 kernel. This also fix some deadlocks
2160 and remove hack for work O_SYNC with 2.6 kernel.
2162 Severity : enhancement
2164 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2165 Details : The uninit_groups feature works in conjunction with the kernel
2166 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2167 up the pass1 processing of e2fsck. This is a read-only feature
2168 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2169 mount filesystems that have had this feature enabled.
2171 Severity : enhancement
2173 Description: Improve multi-block allocation algorithm to avoid fragmentation
2174 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2175 allocation locality and avoid filesystem fragmentation.
2177 ------------------------------------------------------------------------------
2179 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2181 * Support for kernels:
2182 2.6.9-42.0.3.EL (RHEL 4)
2183 2.6.5-7.276 (SLES 9)
2184 2.4.21-47.0.1.EL (RHEL 3)
2185 2.6.12.6 vanilla (kernel.org)
2186 2.6.16.21-0.8 (SLES10)
2187 * Recommended e2fsprogs version: 1.39.cfs2-0
2189 * The backwards-compatible /proc/sys/portals symlink has been removed
2190 in this release. Before upgrading, please ensure that you change
2191 any configuration scripts or /etc/sysctl.conf files that access
2192 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2193 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2194 in advance of the upgrade on any system running Lustre 1.4.6 or
2195 newer, since /proc/sys/lnet was added in that version.
2196 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2200 Frequency : only when quota is used
2202 Description: avoid scanning export list for quota master
2203 Details : Change the algorithms to avoid scanning export list in order
2204 to improve the efficiency.
2207 Frequency : MDS failover only, very rarely
2209 Description: "went back in time" messages on mds failover
2210 Details : The greatest transno may be lost when the current operation
2211 finishes with an error (transno==0) and the client's last_rcvd
2212 record is over-written. Save the greatest transno in the
2213 mds_last_transno for this case.
2216 Frequency : always for specific kernels and striping counts
2218 Description: client may get "Matching packet too big" without ACL support
2219 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2220 when trying to access files in certain configurations. The
2221 clients should in fact be denied when mounting because they do
2222 not understand ACLs.
2225 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2227 Description: many clients connecting with IO in progress causes connect timeouts
2228 Details : Avoid synchronous journal commits to avoid delays caused by many
2229 clients connecting/disconnecting when bulk IO is in progress.
2230 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2231 OST_IO_PORTAL to avoid delays behind potentially many pending
2235 Frequency : occasionally with multiple writers to a single file
2237 Description: shared writes to file may result in wrong size reported by stat()
2238 Details : Allow growing of kms when extent lock is cancelled
2241 Frequency : always with random mmap IO to multi-striped file
2243 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2244 Details : Only if the hole is at the end of OST object so that kms is too
2245 small. Fix is to increase kms accordingly in ll_nopage.
2248 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2250 Description: writes to a missing object would leak memory on the OST
2251 Details : If there is an inconsistency between the MDS and OST filesystems,
2252 such that the MDS references an object that doesn't exist, writes
2253 to that object will leak memory due to incorrect cleanup in the
2254 error handling path, eventually running out of memory on the OST.
2259 Description: Creating too long symlink causes lustre errors
2260 Details : Check symlink and name lengths before sending requests to MDS.
2263 Frequency : only if flock is enabled (not on by default)
2265 Description: posix locks not released on fd closure on 2.6.9+
2266 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2267 kernels, this caused such locks not to be released on fd close and
2268 then assertions on fs unmount about still used locks.
2271 Frequency : MDS failover only, very rarely
2273 Description: clients may get ASSERTION(granted_lock != NULL)
2274 Details : When request was taking a long time, and a client was resending
2275 a getattr by name lock request. The were multiple lock requests
2276 with the same client lock handle and
2277 mds_getattr_name->fixup_handle_for_resent_request found one of the
2278 lock handles but later failed with ASSERTION(granted_lock != NULL).
2283 Description: handle->h_buffer_credits > 0, assertion failure
2284 Details : h_buffer_credits is zero after truncate, causing assertion
2285 failure. This patch extends the transaction or creates a new
2289 Frequency : NFS re-export or patchless client
2290 Bugzilla : 11179, 10796
2291 Description: Crash on NFS re-export node (__d_move)
2292 Details : We do not want to hash the dentry if we don't have a lock.
2293 But if this dentry is later used in d_move, we'd hit uninitialised
2294 list head d_hash, so we just do this to init d_hash field but
2295 leave dentry unhashed.
2298 Frequency : NFS re-export or patchless client
2300 Description: NFS exports has problem with symbolic link
2301 Details : lustre client didn't properly install dentry when re-exported
2302 to NFS or running patchless client.
2305 Frequency : NFS re-export or patchless client
2307 Description: Various nfs/patchless fixes.
2308 Details : fixes reuse disconected alias for lookup process - this fixes
2309 warning "find_exported_dentry: npd != pd",
2310 fix permission error with open files at nfs.
2311 fix apply umask when do revalidate.
2314 Frequency : occasional
2316 Description: Crash on NFS re-export node
2317 Details : calling clear_page() on the wrong pointer triggered oops in
2318 generic_mapping_read().
2321 Frequency : rarely, using O_DIRECT IO
2323 Description: unaligned directio crashes client with LASSERT
2324 Details : check for unaligned buffers before trying any requests.
2327 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2329 Description: stale data returned from RAID cache
2330 Details : If only a small amount of IO is done to the RAID device before
2331 reading it again it is possible to get stale data from the RAID
2332 cache instead of reading it from disk.
2335 Frequency : always for sles10 kernel
2337 Description: sles10 support
2338 Details : ll_follow_link: compile fixes and using of nd_set_link
2339 under newer kernels.
2342 Frequency : depends on arch, kernel and compiler version, always on sles10
2345 Description: recursive or deep enough symlinks cause stack overflow
2346 Details : getting rid of large stack-allocated variable in
2350 Frequency : depends on hardware
2352 Description: lustre write performance loss in the SLES10 kernel
2353 Details : the performance loss is caused by using of write barriers in the
2354 ext3 code. The SLES10 kernel turns barrier support on by
2355 default. The fix is to undo that change for ldiskfs.
2357 ------------------------------------------------------------------------------
2359 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2361 * Support for kernels:
2362 2.6.9-42.0.3EL (RHEL 4)
2363 2.6.5-7.276 (SLES 9)
2364 2.4.21-47.0.1.EL (RHEL 3)
2365 2.6.12.6 vanilla (kernel.org)
2369 Frequency : quota enabled and large files being deleted
2371 Description: releasing more than 4GB of quota at once hangs OST
2372 Details : If a user deletes more than 4GB of files on a single OST it
2373 will cause the OST to spin in an infinite loop. Release
2374 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2379 Description: statfs data retrieved from /proc may be stale or zero
2380 Details : When reading per-device statfs data from /proc, in the
2381 {kbytes,files}_{total,free,avail} files, it may appear
2382 as zero or be out of date.
2385 Frequency : systems with MD RAID1 external journal devices
2387 Description: lconf's call to blkid is confused by RAID1 journal devices
2388 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2389 of returning all block devices that match the journal UUID.
2392 Frequency : always, for aggregate stripe size over 4GB
2394 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2395 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2396 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2399 Frequency : always if "lfs find" used on a local file/directory
2401 Description: "lfs find" segfaults if used on a local file/directory
2402 Details : The case where a directory component was not specified wasn't
2403 handled correctly. Handle this properly.
2406 Frequency : always on ppc64
2408 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2409 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2410 perform endianness conversion.
2413 Frequency : rarely (truncate to non-zero file size after write under load)
2414 Bugzilla : 10730, 10687
2415 Description: Files padded with zeros to next 4K multiple
2416 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2417 it is possible that files that are truncated to a non-zero size
2418 immediately after being written are filled with zero bytes beyond
2419 the truncated size. No file data is lost.
2421 Severity : enhancement
2423 Description: Allow recovery/failover for liblustre clients.
2424 Details : liblustre clients were unaware of failover configurations until
2427 Severity : enhancement
2429 Description: user file locks should fail when not mounting with flock option
2430 Details : Set up an error-returning stub in ll_file_operations.lock field
2431 to prevent incorrect behaviour when client is mounted without
2432 flock option. Also, set up properly f_op->flock field for
2436 Frequency : always on ia64
2438 Description: "lfs df" loops on printing out MDS statfs information
2439 Details : The obd_ioctl_data was not initialized and in some systems
2440 this caused a failure during the ioctl that did not return
2441 an error. Initialize the struct and return an error on failure.
2444 Frequency : SLES 9 only
2446 Description: Error of copying files with lustre special EAs as root
2447 Details : Client side always return success for setxattr call for lustre
2448 special xattr (currently only "trusted.lov").
2451 Frequency : rarely on clusters with both ia64+i386 clients
2453 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2454 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2455 clients are doing concurrent writes to the same file it is
2456 possible that the ia64 clients may LASSERT because the OST
2457 extent locks are not PAGE_SIZE aligned. Ensure that grown
2458 locks are always aligned on the request boundary.
2461 Frequency : specific use, occasional
2463 Description: Overwriting in use executable truncates on-disk binary image
2464 Details : If one node attempts to overwrite an executable in use by
2465 another node, we now correctly return ETXTBSY instead of
2466 truncating the file.
2468 Severity : enhancement
2470 Description: Async OSC create to avoid the blocking unnecessarily.
2471 Details : If a OST has no remain object, system will block on the creating
2472 when need to create a new object on this OST. Now, ways use
2473 pre-created objects when available, instead of blocking on an
2474 empty osc while others are not empty. If we must block, we block
2475 for the shortest possible period of time.
2480 Description: chmod on Lustre root is propagated to other clients
2481 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2482 invalid by the follow_mount time.
2485 Frequency : liblustre clients only
2487 Description: Race in 'instant cancel' lock handling could lead to such locks
2488 never to be granted in case of SMP MDS
2489 Details : Do not destroy not yet granted but cbpending locks in
2493 Frequency : replay/resend of open
2495 Description: non null lock assetion failure in mds_intent_policy
2496 Details : Trying to replay/resend lockless open requests resulted in
2497 mds_open() returning 0 with no lock. Now it sets a flag if
2498 a lock is going to be returned.
2500 Severity : enhancement
2502 Description: Checksum enhancements
2503 Details : New checksum enhancements allow for resending RPCs that failed
2506 Severity : enhancement
2508 Description: Tunables on number of dirty pages in cacche
2509 Details : Allow to set limit on number of dirty pages cached.
2514 Description: client crash on unmount - lock still has references
2515 Details : In some error handling cases it was possible to leak a lock
2516 reference on a client while accessing a file. This was not
2517 harmful to the client during operation, but would cause the
2518 client to crash when the filesystem is unmounted.
2521 Frequency : specific case, rare
2523 Description: ETXTBSY on mds though file not in use by client
2524 Details : ETXTBSY is no longer incorrectly returned when attempting to
2525 chmod or chown a directory that the user previously tried to
2526 execute or a currently-executing binary.
2529 Frequency : extremely rare except on liblustre-based clients
2531 Description: Lustre space not freed when files are deleted
2532 Details : Clean up open-unlinked files after client eviction. Previously
2533 the unlink was skipped and the files remained as orphans.
2538 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2539 Details : In some cases it was possible to send a blocking callback to
2540 a client doing a glimpse, even though that client didn't get
2541 a lock granted. When the glimpse lock is cancelled on the OST
2542 the freed lock is left on the waiting list and corrupted the list.
2545 Frequency : all core dumps
2547 Description: Broke core dumps to lustre
2548 Details : Negative dentry may be unhashed if parent does not have UPDATE
2549 lock, but some callers, e.g. do_coredump, expect dentry to be
2550 hashed after successful create, hash it in ll_create_it.
2552 ------------------------------------------------------------------------------
2554 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2556 * Support for kernels:
2557 2.6.9-42.0.2.EL (RHEL 4)
2558 2.6.5-7.276 (SLES 9)
2559 2.4.21-40.EL (RHEL 3)
2560 2.6.12.6 vanilla (kernel.org)
2564 Frequency : always on RHEL 3
2566 Description: Number of open files grows over time
2567 Details : The number of open files grows over time, whether or not
2568 Lustre is started. This was due to a filp leak introduced
2569 by one of our kernel patches.
2571 ------------------------------------------------------------------------------
2573 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2575 * Support for kernels:
2576 2.6.9-42.EL (RHEL 4)
2577 2.6.5-7.276 (SLES 9)
2578 2.4.21-40.EL (RHEL 3)
2579 2.6.12.6 vanilla (kernel.org)
2584 Bugzilla : 5719, 9635, 9792, 9684
2585 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2586 Details : If a server is under heavy load and cannot reply to new
2587 connection requests before the client resends the (re)connect,
2588 the connection handling code can behave badly if two service
2589 threads are concurrently handing separate (re)connections from
2590 the same client. Add better locking to the connection handling
2591 code, and ensure that only a single connection will be processed
2592 for a given client UUID, even if the lock is dropped.
2594 Severity : enhancement
2596 Description: add TCP zero-copy support to kernel
2597 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2598 sends if the hardware supports scatter-gather and checksumming.
2599 This allows socklnd to do client-write and server-read more
2600 efficiently and reduce CPU utilization from skbuf copying.
2603 Frequency : only if NFS exporting from client
2605 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2606 Details : If setting attributes on a file created under NFS that had
2607 never been opened it would be possible to oops the client
2608 if the file had no objects.
2611 Frequency : always for liblustre
2613 Description: liblustre client does MDS+OSTs setattr RPC for each write
2614 Details : When doing a write from a liblustre client, the client
2615 incorrectly issued an RPC to the MDS and each OST the file was
2616 striped over in order to update the timestamps. When writing
2617 with small chunks and many clients this could overwhelm the MDS
2618 with RPCs. In all cases it would slow down the write because
2619 these RPCs are unnecessary.
2621 Severity : enhancement
2623 Description: allow number of MDS service threads to be changed at module load
2624 Details : It is now possible to change the number of MDS service threads
2625 running. Adding "options mds mds_num_threads={N}" to the MDS's
2626 /etc/modprobe.conf will set the number of threads for the next
2627 time Lustre is restarted (assuming the "mds" module is also
2628 reloaded at that time). The default number of threads will
2629 stay the same, 32 for most systems.
2634 Description: OST crash if filesystem is unformatted or corrupt
2635 Details : If an OST is started on a device that has never been formatted
2636 or if the filesystem is corrupt and cannot even mount then the
2637 error handling cleanup routines would dereference a NULL pointer.
2642 Description: NULL pointer deref in llap_from_page.
2643 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2644 incorrect) mapping if the page was truncated/reclaimed while it was
2645 searched for. Check for this condition and skip such pages when
2646 doing readahead. Introduce extra check to llap_from_page() to
2647 verify page->mapping->host is non-NULL (so page is not anonymous).
2650 Frequency : Sometimes when using sys_sendfile
2652 Description: "page not covered by a lock" warnings from ll_readpage
2653 Details : sendfile called ll_readpage without right page locks present.
2654 Now we introduced ll_file_sendfile that does necessary locking
2655 around call to generic_file_sendfile() much like we do in
2659 Frequency : with certain MDS communication failures at client mount time
2661 Description: NULL pointer deref after failed client mount
2662 Details : a client connection request may delayed by the network layer
2663 and not be sent until after the PTLRPC layer has timed out the
2664 request. If the client fails the mount immediately it will try
2665 to clean up before the network times out the request. Add a
2666 reference from the request import to the obd device and delay
2667 the cleanup until the network drops the request.
2670 Frequency : occasionally during client (re)connect
2672 Description: assertion failure during client (re)connect
2673 Details : processing a client connection request may be delayed by the
2674 client or server longer than the client connect timeout. This
2675 causes the client to resend the connection request. If the
2676 original connection request is replied in this interval, the
2677 client may trip an assertion failure in ptlrpc_connect_interpret()
2678 which thought it would be the only running connect process.
2681 Frequency : only with obd_echo servers and clients that are rebooted
2683 Description: kernel BUG accessing uninitialized data structure
2684 Details : When running an obd_echo server it did not start the ping_evictor
2685 thread, and when a client was evicted an uninitialized data
2686 structure was accessed. Start the ping_evictor in the RPC
2687 service startup instead of the OBD startup.
2689 Severity : enhancement
2690 Bugzilla : 10193 (patchless)
2691 Description: Remove dependency on various unexported kernel interfaces.
2692 Details : No longer need reparent_to_init, exit_mm, exit_files,
2693 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2696 Frequency : rare (only users of deprecated and unsupported LDAP config)
2698 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2699 Details : LDAP apparently contains 'lustreName' attributes instead of
2700 'name'. A simple remapping of the name is sufficient.
2703 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2705 Description: waiting_locks_callback trips kernel BUG if client is evicted
2706 Details : Running with the dump_on_timeout debug flag turned on makes
2707 it possible that the waiting_locks_callback() can try to dump
2708 the Lustre kernel debug logs from an interrupt handler. Defer
2709 this log dumping to the expired_lock_main() thread.
2711 Severity : enhancement
2713 Description: Support NFS exporting on 2.6 kernels.
2714 Details : Implement non-rawops metadata methods for NFS server to use without
2715 changing NFS server code.
2718 Frequency : very rare (synthetic metadata workload only)
2720 Description: two racing renames might cause an MDS thread to deadlock
2721 Details : Running the "racer" program may cause one MDS thread to rename
2722 a file from being the source of a rename to being the target of
2723 a rename at exactly the same time that another thread is doing
2724 so, and the second thread has already enqueued these locks after
2725 doing a lookup of the target and is trying to relock them in
2726 order. Ensure that we don't try to re-lock the same resource.
2729 Frequency : only very large systems with liblustre clients
2731 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2732 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2733 all OSTs in parallel. This allows the eviction of stale liblustre
2734 clients to proceed much faster than if they were done in series,
2735 and also offers similar improvements for other set_info RPCs.
2740 Description: excessive CPU usage during initial read phase on client
2741 Details : During the initial read phase on a client, it would agressively
2742 retry readahead on the file, consuming too much CPU and impacting
2743 performance (since 1.4.5.8). Improve the readahead algorithm
2744 to avoid this, and also improve some other common cases (read
2745 of small files in particular, where "small" is files smaller than
2746 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2751 Description: MDS crash when receiving packet with unknown intent.
2752 Details : Do not LBUG in unknown intent case, just return -EFAULT
2754 Severity : enhancement
2755 Bugzilla : 9293, 9385
2756 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2757 Details : Do not serialize getattr (non-intent version) and statfs.
2760 Frequency : occasional, when OST network is overloaded/intermittent
2762 Description: client evicted by OST after bulk IO timeout
2763 Details : If a client sends a bulk IO request (read or write) the OST
2764 may evict the client if it is unresposive to its data GET/PUT
2765 request. This is incorrect if the network is overloaded (takes
2766 too long to transfer the RPC data) or dropped the OST GET/PUT
2767 request. There is no need to evict the client at all, since
2768 the pinger and/or lock callbacks will handle this, and the
2769 client can restart the bulk request.
2772 Frequency : Always when mmapping file with no objects
2774 Description: client crashes when mmapping file with no objects
2775 Details : Check that we actually have objects in a file before doing any
2776 operations on objects in ll_vm_open, ll_vm_close and
2782 Description: Request leak when working with deleted CWD
2783 Details : Introduce advanced request refcount tracking for requests
2784 referenced from lustre intent.
2786 Severity : Enhancement
2788 Description: Cache open file handles on client.
2789 Details : MDS now will return special lock along with openhandle, if
2790 requested and client is allowed to hold openhandle, even if unused,
2791 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2792 closing files for every read/write openration.
2794 Severity : Enhancement
2796 Description: Cache open negative dentries on client when possible.
2797 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2798 negative dentries on lock revocation.
2803 Description: Remounting a client read-only wasn't possible with a zconf mount
2804 Details : It wasn't possible to remount a client read-only with llmount.
2806 Severity : enhancement
2807 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2808 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2809 MPICH is included to add Lustre-specific ADIO interfaces.
2810 This is based closely on the UFS ADIO layer and only differs
2811 in file creation, in order to allow the OST striping to be set.
2812 This is user-contributed code and not supported by CFS.
2817 Description: extended inode attributes (immutable, append-only) work improperly
2818 when 2.4 and 2.6 kernels are used on client/server or vice versa
2819 Details : Introduce kernel-independent values for these flags.
2821 Severity : enhancement
2824 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2825 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2826 granularity. Now, Fractional MB granularity is be supported,
2827 this is very useful for low memory system.
2829 Severity : enhancement
2831 Description: Getattr by fid
2832 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2833 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2834 have locks and avoids getattr rpc requests for it on every stat.
2837 Frequency : Always, for filesystems larger than 2TB
2839 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2840 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2841 filesystems larger than 2TB to be mounted with mballoc enabled.
2844 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2846 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2847 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2848 filesystems larger than 2TB is mounted then the kernel will
2849 silently corrupt the start of the filesystem. CONFIG_LBD is
2850 enabled for all CFS-supported kernels, but the possibility of
2851 this happening with a modified kernel config exists.
2853 Severity : enhancement
2855 Description: add client O_DIRECT support for 2.6 kernels
2856 Details : It is now possible to do O_DIRECT reads and writes to files
2857 in the Lustre client mountpoint on 2.6 kernel clients.
2859 Severity : enhancement
2861 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2862 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2863 parallel, not waiting for response from every OST before sending
2864 a rpc to the next OST.
2869 Description: setattr vs write race when updating file timestamps
2870 Details : Client processes that update a file timestamp into the past
2871 right after writing to the file (e.g. tar) it is possible that
2872 the updated file modification time can be reset to the current
2873 time due to a race between processing the setattr and write RPC.
2875 Severity : enhancement
2877 Description: Bring 'lfs find' closer in line with regular Linux find.
2878 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2879 -print0 options and obtains all the needed info through the lustre
2882 Severity : enhancement
2884 Description: support up to 1024 configured devices on one node
2885 Details : change obd_dev array from statically allocated to dynamically
2886 allocated structs as they are first used to reduce memory usage
2891 Description: Flush dirty partially truncated pages during truncate
2892 Details : Immediatelly flush partially truncated pages in filter_setattr,
2893 this way we completely avoid having any pages in page cache on OST
2894 and can retire ugly workarounds during writes to flush such pages.
2899 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2900 Details : protect inode from truncation within vfs_unlink() context
2901 just take a reference before calling vfs_unlink() and release it
2902 when parent's i_sem is free.
2907 Description: last_id value checked outside lock on OST caused LASSERT failure
2908 Details : If there were multiple MDS->OST object precreate requests in
2909 flight, it was possible that the OST's last object id was checked
2910 outside a lock and incorrectly tripped an assertion. Move checks
2911 inside locks, and discard old precreate requests.
2914 Frequency : always, if extents are used on OSTs
2916 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2917 Details : OSTs using the extents format would not zero the high 16 bits of
2918 the index physical block number. This is not a problem for any
2919 OST filesystems smaller than 16TB, and no kernels support ext3
2920 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2921 new/modified files) and can be fixed for existing filesystems
2922 with e2fsprogs-1.39-cfs1.
2927 Description: import connection selection may be incorrect if timer wraps
2928 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2929 import connections to be ignored if the 32-bit jiffies counter
2930 wraps. Use a 64-bit jiffies counter.
2933 Frequency : very large clusters immediately after boot
2935 Description: LNET request buffers exhausted under heavy short-term load
2936 Details : If a large number of client requests are generated on a service
2937 that has previously never seen so many requests it is possible
2938 that the request buffer growth cannot keep up with the spike in
2939 demand. Instead of dropping incoming requests, they are held in
2940 the LND until the RPC service can accept more requests.
2943 Frequency : Sometimes during replay
2945 Description: Assertion failure in ll_local_open after replay.
2946 Details : If replay happened on an open request reply before we were able
2947 to set replay handler, reply will become not swabbed tripping the
2948 assertion in ll_local_open. Now we set the handler right after
2949 recognising of open request
2952 Frequency : very rare
2954 Description: kernel reports "badness in vsnprintf"
2955 Details : Reading from the "recovery_status" /proc file in small chunks
2956 may cause a negative length in lprocfs_obd_rd_recovery_status()
2957 call to vsnprintf() (which is otherwise harmless). Exit early
2958 if there is no more space in the output buffer.
2960 Severity : enhancement
2962 Description: clear OBD RPC statistics by writing to them
2963 Details : It is now possible to clear the OBD RPC statistics by writing
2964 to the "stats" file.
2969 Description: Client mtime is not the same on different clients after utimes
2970 Details : In some cases, the client was using the utimes() syscall on
2971 a file cached on another node. The clients now validate the
2972 ctime from the MDS + OSTs to determine which one is right.
2977 Description: Inability to activate failout mode
2978 Details : lconf script incorrectly assumed that in pythong string's numeric
2979 value is used in comparisons.
2982 Frequency : always with multiple stripes per file
2984 Description: Inefficient object allocation for mutli-stripe files
2985 Details : When selecting which OSTs to stripe files over, for files with
2986 a stripe count that divides evenly into the number of OSTs,
2987 the MDS is always picking the same starting OST for each file.
2988 Return the OST selection heuristic to the original design.
2993 Description: mount failures may take full timeout to return an error
2994 Details : Under some heavy load conditions it is possible that a
2995 failed mount can wait for the full obd_timeout interval,
2996 possibly several minutes, before reporting an error.
2997 Instead return an error as soon as the status is known.
2999 Frequency : quota enabled and large files being deleted
3001 Description: releasing more than 4GB of quota at once hangs OST
3002 Details : If a user deletes more than 4GB of files on a single OST it
3003 will cause the OST to spin in an infinite loop. Release
3004 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3009 Description: statfs data retrieved from /proc may be stale or zero
3010 Details : When reading per-device statfs data from /proc, in the
3011 {kbytes,files}_{total,free,avail} files, it may appear
3012 as zero or be out of date.
3015 Frequency : systems with MD RAID1 external journal devices
3017 Description: lconf's call to blkid is confused by RAID1 journal devices
3018 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3019 of returning all block devices that match the journal UUID.
3022 Frequency : always, for aggregate stripe size over 4GB
3024 Description: assertion fails when trying to use 4GB stripe size
3025 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
3026 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
3029 Frequency : always on ppc64
3031 Description: the first write on an ext3 filesystem with mballoc got stuck
3032 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3033 perform endianness conversion.
3035 ------------------------------------------------------------------------------
3037 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3039 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3040 INTEROPERATE with older versions automatically. Please read the
3041 user documentation before upgrading any part of a live system.
3042 * WARNING: Lustre networking configuration changes are required with
3043 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3046 * Support for kernels:
3047 2.6.9-22.0.2.EL (RHEL 4)
3048 2.6.5-7.244 (SLES 9)
3049 2.6.12.6 vanilla (kernel.org)
3052 Severity : enhancement
3053 Bugzilla : 7981/8208
3054 Description: Introduced Lustre Networking (LNET)
3055 Details : LNET is new networking infrastructure for Lustre, it includes
3056 a reorganized network configuration mode (see the user
3057 documentation for full details) as well as support for routing
3058 between different network fabrics. Lustre Networking Devices
3059 (LNDs) for the supported network fabrics have also been
3060 created for this new infrastructure.
3062 Severity : enhancement
3063 Description: Introduced Access control lists
3064 Details : clients can set ACLs on files and directories in order to have
3065 more fine-grained permissions than the standard Unix UGO+RWX.
3066 The MDS must be started with the "-o acl" mount option.
3068 Severity : enhancement
3069 Description: Introduced filesystem quotas
3070 Details : Administrators may now establish per-user quotas on the
3073 Severity : enhancement
3075 Description: Configuration change for the XT3
3076 The PTLLND is now used to run Lustre over Portals on the XT3
3077 The configure option(s) --with-cray-portals are no longer used.
3078 Rather --with-portals=<path-to-portals-includes> is used to
3079 enable building on the XT3. In addition to enable XT3 specific
3080 features the option --enable-cray-xt3 must be used.
3085 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3086 Details : A race between allocating a new llcd and re-getting the llcd_lock
3087 allowed another thread to grab newly-allocated llcd.
3089 Severity : enhancement
3091 Description: 2.6 OST async journal commit and locking fix to improve performance
3092 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3093 2.6 kernels are now async as they already were in 2.4 kernels so
3094 that they can commit concurrently with the network bulk transfer.
3095 For block-allocated files the filter allocation semaphore is held
3096 to avoid filesystem fragmentation during allocation. BKL lock
3097 removed for 2.6 xattr operations where it is no longer needed.
3102 Description: lconf incorrectly determined whether two IP networks could talk
3103 Details : In some more complicated routing and multiple-network
3104 configurations, lconf will avoid trying to make a network
3105 connection to a disjoint part of the IP space. It was doing the
3106 math incorrectly for one set of cases.
3111 Description: Fix for potential infinite loop processing records in an llog.
3112 Details : If an llog record is corrupted/zeroed, it is possible to loop
3113 forever in llog_process(). Validate the llog record length
3114 and skip the remainder of the block on error.
3117 Frequency : occasional (liblustre only)
3119 Description: liblustre could not open files whose last component is a symlink
3120 Details : sysio_path_walk() would incorrectly pass the open intent to
3121 intermediate path components.
3124 Frequency : rare (liblustre only with non-standard tuning)
3125 Bugzilla : 7201 (7350)
3126 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3127 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3128 after it has been released, only do this for referenced locks
3130 Severity : enhancement
3132 Description: specifying an (invalid) directory default stripe_size of -1
3133 would reset the directory default striping
3134 Details : stripe_size -1 was used internally to signal directory stripe
3135 removal, now use "all default" to signal dir stripe removal
3136 as a directory striping of "all default" is not useful
3139 Frequency : common for large clusters running liblustre clients
3141 Description: doing an ls when liblustre clients are running is slow
3142 Details : sending a glimpse AST to a liblustre client waits for every AST
3143 to time out, as liblustre clients will not respond. Since they
3144 cannot cache data we refresh the OST lock LVB from disk instead.
3146 Severity : enhancement
3148 Description: doing an ls at the same time as file IO can be slow
3149 Details : enqueue and other "small" requests can be blocked behind many
3150 large IO requests. Create a new OST IO portal for non-IO
3151 requests so they can be processed faster.
3154 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3156 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3157 created file with mode 000
3158 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3159 allow an ftruncate() as a non-root user to a file with mode 000.
3160 Lustre now respects this flag to disable mode checks when
3161 truncating a file owned by the user
3164 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3167 Description: Revoking locks from clients that went dead or catatonic might take
3169 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3170 cancellation of such locks instant on servers without waiting for
3171 any reply from clients. Clients drops these locks when cancel
3172 notification from server is received without replying.
3175 Frequency : liblustre-only, when liblustre client dies or becomes busy
3177 Description: Doing ls on Linux clients can take a long time with active
3179 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3180 granting such locks to it in the first place if possible. Locks
3181 are taken by proxy on the OST during the read or write and
3182 dropped immediately afterward. Add connect flags handling, do
3183 not grant locks to liblustre clients for glimpse ASTs.
3185 Severity : enhancement
3187 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3188 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3189 sequential reads and behaves badly with applications doing random
3190 reads. Improve it by reading ahead at least read region, and
3191 avoiding excessive large RPC for small reads.
3193 Severity : enhancement
3195 Description: Creating more than 1000 files for a single job may cause a load
3196 imbalance on the OSTs if there are also a large number of OSTs.
3197 Details : qos_prep_create() uses an OST index reseed value that is an
3198 even multiple of the number of available OSTs so that if the
3199 reseed happens in the middle of the object allocation it will
3200 still utilize the OSTs as uniformly as possible.
3205 Description: OST or MDS may oops in ping_evictor_main()
3206 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3207 but doesn't restart at beginning of obd_exports_timed list
3210 Severity : enhancement
3212 Description: improve by-nid export eviction on the MDS and OST
3213 Details : allow multiple exports with the same NID to be evicted at one
3214 time without re-searching the exports list.
3217 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3219 Description: MDS may oops in groups_free()
3220 Details : in rare race conditions a newly allocated group_info struct is
3221 freed again, and this can be NULL. The 2.4 compatibility code
3222 for groups_free() checked for a NULL pointer, but 2.6 did not.
3225 Frequency : common for liblustre clients doing little filesystem IO
3226 Bugzilla : 9352, 7313
3227 Description: server may evict liblustre clients accessing contended locks
3228 Details : if a client is granted a lock or receives a completion AST
3229 with a blocking AST already set it would not reply to the AST
3230 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3233 Frequency : lfs setstripe, only systems with more than 160 OSTs
3235 Description: unable to set striping with a starting offset beyond OST 160
3236 Details : llapi_create_file() incorrectly limited the starting stripe
3237 index to the maximum single-file stripe count.
3240 Frequency : LDAP users only
3242 Description: lconf did not handle in-kernel recovery with LDAP properly
3243 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3245 Severity : enhancement
3247 Description: bind OST threads to NUMA nodes to improve performance
3248 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3249 node and do their allocations there to localize memory access
3251 Severity : enhancement
3253 Description: llmount can determine client NID directly from Myrinet (GM)
3254 Details : the client NID code from gmnalnid was moved directly into
3255 llmount, removing the need to use this or specifying the
3256 client NID explicitly when mounting GM clients with zeroconf
3259 Frequency : if client is started with down MDS
3261 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3262 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3263 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3264 ptlrpc_import_delayed_req() because we didn't check if the
3265 request was interrupted, and we also didn't break out of the
3266 event loop if there was no timeout
3271 Description: data loss during non-page-aligned writes to a single file from
3272 both multiple nodes and multiple threads on one node at same time
3273 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3274 inconsistency led to false short-reads, that were cached and later
3275 used by ->prepare_write() to fill in partially written page,
3276 leading to data loss.
3279 Frequency : always, if lconf --abort_recovery used
3281 Description: lconf --abort_recovery fails with 'Operation not supported'
3282 Details : lconf was attempting to abort recovery on the MDT device and not
3285 Severity : enhancement
3287 Description: remove cleanup logs
3288 Details : replace lconf-generated cleanup logs with lustre internal
3289 cleanup routines. Eliminates the need for client-cleanup and
3292 Severity : enhancement
3294 Description: add support for EAs (user and system) on lustre filesystems
3295 Details : it is now possible to store extended attributes in the Lustre
3296 client filesystem, and with the user_xattr mount option it
3297 is possible to allow users to store EAs on their files also
3299 Severity : enhancement
3301 Description: Add possibility (config option) to show minimal available OST free
3303 Details : When compiled with --enable-mindf configure option, statfs(2)
3304 (and so, df) will return least minimal free space available from
3305 all OSTs as amount of free space on FS, instead of summary of
3306 free spaces of all OSTs.
3308 Severity : enhancement
3310 Description: do not expand extent locks acquired on OST-side
3311 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3312 by server: they are not cached anyway.
3315 Frequency : when mmap is used/binaries executed from Lustre
3317 Description: Unmmap pages before throwing them away from read cache.
3318 Details : llap_shrink cache now attempts to unmap pages before discarding
3319 them (if unmapping failed - do not discard). SLES9 kernel has
3320 extra checks that trigger if this unmapping is not done first.
3325 Description: lconf didn't resolve symlinks before checking to see whether a
3326 given mountpoint was already in use
3329 Frequency : when migrating failover services
3330 Bugzilla : 6395, 9514
3331 Description: When migrating a subset of services from a node (e.g. failback
3332 from a failover service node) the remaining services would
3333 time out and evict clients.
3334 Details : lconf --force (implied by --failover) sets the global obd_timeout
3335 to 5 seconds in order to quickly disconnect, but this caused
3336 other RPCs to time out too quickly. Do not change the global
3337 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3339 Severity : enhancement
3340 Frequency : if MDS is started with down OST
3341 Bugzilla : 9439,5706
3342 Description: Allow startup/shutdown of an MDS without depending on the
3343 availability of the OSTs.
3344 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3345 Add appropriate locking and lov-osc refcounts for safe
3346 cleaning. Add osc abort_inflight calls in case the
3347 synchronize never started.
3350 Frequency : occasional (Cray XT3 only)
3352 Description: root not authorized to access files in CRAY_PORTALS environment
3353 Details : The client process capabilities were not honoured on the MDS in
3354 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3355 been accessed by an authorized user then root was able to access
3356 the file on the local client also. The root user capabilities
3357 are now allowed on the MDS, as this environment has secure UID.
3360 Frequency : occasional
3362 Description: ldiskfs "too long searching" message happens too often
3363 Details : A debugging message (otherwise harmless) prints too often on
3364 the OST console. This has been reduced to only happen when
3365 there are fragmentation problems on the filesystem.
3370 Description: Division by zero in statfs when all OSCs are inactive
3371 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3372 lov_statfs divides by value returned by lov_get_stripecnt().
3376 Bugzilla : 9489, 3273
3377 Description: First write from each client to each OST was only 4kB in size,
3378 to initialize client writeback cache, which caused sub-optimal
3379 RPCs and poor layout on disk for the first writen file.
3380 Details : Clients now request an initial cache grant at (re)connect time
3381 and so that they can start streaming writes to the cache right
3382 away and always do full-sized RPCs if there is enough data.
3383 If the OST is rebooted the client also re-establishes its grant
3384 so that client cached writes will be honoured under the grant.
3389 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3390 Details : Now I/O RPCs go to different portal number and (presumably) fast
3391 lock requests (and glimses) and other RPCs get their own service
3392 threads pool that should be able to service those RPCs
3395 Severity : enhancement
3397 Description: Ability to exchange lustre version between client and servers and
3398 issue warnings at client side if client is too old. Also for
3399 liblustre clients there is ability to refuse connection of too old
3401 Details : New 'version' field is added to connect data structure that is
3402 filled with version info. That info is later checked by server and
3406 Frequency : rare, liblustre only.
3407 Bugzilla : 9296, 9581
3408 Description: Two simultaneous writes from liblustre at offset within same page
3409 might proceed at the same time overwriting eachother with stale
3411 Details : I/O lock withing llu_file_prwv was released too early, before data
3412 actually was hitting the wire. Extended lock-holding time until
3413 server acknowledges receiving data.
3416 Frequency : extremely rare. Never observed in practice.
3418 Description: avoid generating lustre_handle cookie of 0.
3419 Details : class_handle_hash() generates handle cookies by incrementing
3420 global counter, and can hit 0 occasionaly (this is unlikely, but
3421 not impossible, because initial value of cookie counter is
3422 selected randonly). Value of 0 is used as a sentinel meaning
3423 "unassigned handle" --- avoid it. Also coalesce two critical
3424 sections in this function into one.
3426 Severity : enhancement
3428 Description: allow liblustre clients to delegate truncate locking to OST
3429 Details : To avoid overhead of locking, liblustre client instructs OST to
3430 take extent lock in ost_punch() on client's behalf. New connection
3431 flag is added to handle backward compatibility.
3433 Severity : enhancement
3434 Bugzilla : 4928, 7341, 9758
3435 Description: allow number of OST service threads to be specified
3436 Details : a module parameter allows the number of OST service threads
3437 to be specified via "options ost ost_num_threads={N}" in the
3438 OSS's /etc/modules.conf or /etc/modprobe.conf.
3442 Bugzilla : 6146, 9635, 9895
3443 Description: servers crash with bad pointer in target_handle_connect()
3444 Details : In rare cases when a client is reconnecting it was possible that
3445 the connection request was the last reference for that export.
3446 We would temporarily drop the export reference and get a new
3447 one, but this may have been the last reference and the export
3448 was just destroyed. Get new reference before dropping old one.
3450 Severity : enhancement
3451 Frequency : if client is started with failover MDS
3453 Description: Allow multiple MDS hostnames in the mount command
3454 Details : Try to read the configuration from all specified MDS
3455 hostnames during a client mount in case the "primary"
3458 Severity : enhancement
3460 Description: Stop sending data to evicted clients as soon as possible.
3461 Details : Check if the client we are about to send or are sending data to
3462 was evicted already. (Check is done every second of waiting,
3463 for which l_wait_event interface was extended to allow checking
3464 of exit condition at specified intervals).
3467 Frequency : rare, normally only when NFS exporting is done from client
3469 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3471 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3474 Frequency : always, if srand() is called before liblustre initialization
3476 Description: Liblustre uses system PRNG disturbing its usage by user application
3477 Details : Introduce internal to lustre fast and high-quality PRNG for
3478 lustre usage and make liblustre and some other places in generic
3479 lustre code to use it.
3481 Severity : enhancement
3482 Bugzilla : 9477, 9557, 9870
3483 Description: Verify that the MDS configuration logs are updated when xml is
3484 Details : Check if the .xml configuration logs are newer than the config
3485 logs stored on the MDS and report an error if this is the case.
3486 Request --write-conf, or allow starting with --old_conf.
3488 Severity : enhancement
3490 Description: Handle symlinks in the path when checking if Lustre is mounted.
3491 Details : Resolve intermediate symlinks when checking if a client has
3492 mounted a filesystem to avoid duplicate client mounts.
3497 Description: lconf can hit an error exception but still return success.
3498 Details : The lconf command catches the Command error exception at the top
3499 level script context and will exit with the associated exit
3500 status, but doesn't ensure that this exit status is non-zero.
3505 Description: failure of ptlrpc thread startup can cause oops
3506 Details : Starting a ptlrpc service thread can fail if there are a large
3507 number of threads or the server memory is very fragmented.
3508 Handle this without oopsing.
3511 Frequency : always, only if liblustre and non-default acceptor port was used
3513 Description: liblustre cannot connect to servers with non-default acceptor port
3514 Details : tcpnal_set_default_params() was not called and was therefore
3515 ignoring the environment varaible TCPNAL_PORT, as well as other
3516 TCPNAL_ environment variables
3521 Description: two objects could be created on the same OST for a single file
3522 Details : If an OST is down, in some cases it was possible to create two
3523 objects on a single OST for a single file. No problems other
3524 than potential performance impact and spurious error messages.
3528 Bugzilla : 5681, 9562
3529 Description: Client may oops in ll_unhash_aliases
3530 Details : Client dcache may become inconsistent in race condition.
3531 In some cases "getcwd" can fail if the current directory is
3537 Description: Inode refcounting problems in NFS export code
3538 Details : link_raw functions used to call d_instantiate without obtaining
3539 extra inode reference first.
3543 Bugzilla : 9942, 9903
3544 Description: Referencing freed requests leading to crash, memleaks with NFS.
3545 Details : We used to require that call to ll_revalidate_it was always
3546 followed by ll_lookup_it. Also with revalidate_special() it is
3547 possible to call ll_revalidate_it() twice for the same dentry
3548 even if first occurence returned success. This fix changes semantic
3549 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3550 reference on a request referred from the intent.
3551 ll_intent_release() then releases such a request.
3554 Frequency : rare, normally benchmark loads only
3556 Description: unlinked inodes were kept in memory on the client
3557 Details : If a client is repeatedly creating and unlinking files it
3558 can accumulate a lot of stale inodes in the inode slab cache.
3559 If there is no other client load running this can cause the
3560 client node to run out of memory. Instead flush old inodes
3561 from client cache that have the same inode number as a new inode.
3564 Frequency : SLES9 2.6.5 kernel and long filenames only
3565 Bugzilla : 9969, 10379
3566 Description: utime reports stale NFS file handle
3567 Details : SLES9 uses out-of-dentry names in some cases, which confused
3568 the lustre dentry revalidation. Change it to always use the
3572 Frequency : rare, unless heavy write-truncate concurrency is continuous
3573 Bugzilla : 4180, 6984, 7171, 9963, 9331
3574 Description: OST becomes very slow and/or deadlocked during object unlink
3575 Details : filter_destroy() was holding onto the parent directory lock
3576 while truncating+unlinking objects. For very large objects this
3577 may block other threads for a long time and slow overall OST
3578 responsiveness. It may also be possible to get a lock ordering
3579 deadlock in this case, or run out of journal credits because of
3580 the combined truncate+unlink. Solution is to do object truncate
3581 first in one transaction without parent lock, and then do the
3582 final unlink in a new transaction with the parent lock. This
3583 reduces the lock hold time dramatically.
3586 Frequency : rare, 2.4 kernels only
3588 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3589 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3590 before unmounting their filesystem in order to avoid deadlock.
3591 The kernel_locked() function in 2.4 kernels only checks whether
3592 the kernel lock is held, not whether it is this process that is
3593 holding it as 2.6 kernels do.
3598 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3599 Details : The client ptlrpc code may be trying to reconnect to a down
3600 server before a previous connection attempt has timed out.
3601 Increase the reconnect interval to be longer than the connection
3602 timeout interval to avoid sending duplicate connections to
3606 Frequency : echo_client brw_test command
3608 Description: fix echo_client to work with OST preallocated code
3609 Details : OST preallocation code (5137) didn't take echo_client IO path
3610 into account: echo_client calls filter methods outside of any
3611 OST thread and, hence, there is no per-thread preallocated
3612 pages and buffers to use. Solution: hijack pga pages for IO. As
3613 a byproduct, this avoids unnecessary data copying.
3617 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3618 Description: Client can oops in mdc_commit_close() after open replay
3619 Details : It was possible for the MDS to return an open request with no
3620 transaction number in mds_finish_transno() if the client was
3621 evicted, but without actually returning an error. Clients
3622 would later try to replay that open and may trip an assertion
3623 Simplify the client close codepath, and always return an error
3624 from the MDS in case the open is not successful.
3627 Frequency : rare, 2.6 OSTs only
3629 Description: OST may deadlock under high load on fragmented files
3630 Details : If there was a heavy load and highly-fragmented OST filesystems
3631 it was possible to have all the OST threads deadlock waiting on
3632 allocation of biovecs, because the biovecs were not released
3633 until the entire RPC IO was completed. Instead, release biovecs
3634 as soon as they are complete to ensure forward IO progress.
3636 Severity : enhancement
3638 Description: Support for specifying external journal device at mount
3639 Details : If an OST or MDS device is formatted with an external journal
3640 device, this device major/minor is stored in the ext3 superblock
3641 and may not be valid for failover. Allow detecting and
3642 specifying the external journal at mount time.
3647 Description: Mounting an MDS with pending unlinked files may cause oops
3648 Details : target_finish_recovery() calls mds_postrecov() which returned
3649 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3650 considers this an error and immediately begins cleaning up the
3651 lov, just after starting the mds_lov process
3653 Severity : enhancement
3655 Description: Implement 'lfs df' to report actual free space on per-OST basis
3656 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3657 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3658 usage of objects; '-h' to report in human readable format.
3660 ------------------------------------------------------------------------------
3662 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3669 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3670 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3671 at startup during orphan unlinking. Instead, skip buddy update
3672 before setup, as it will be regenerated after recovery is complete.
3677 Description: If an OST is inactive, its locks might reference stale inodes.
3678 Details : lov_change_cbdata() must iterate over all namespaces, even if
3679 they are inactive to clear inode references from the lock.
3681 Severity : enhancement
3682 Frequency : occasional, if non-standard max_dirty_mb used
3684 Description: Client will block write RPCs if not enough grant
3685 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3686 then the client will block writes while waiting for another RPC
3687 to complete instead of consuming its dirty limit. With change
3688 we get improved performance when max_dirty_mb is small.
3690 Severity : enhancement
3691 Bugzilla : 3389, 6253
3692 Description: Add support for supplementary groups on the MDS.
3693 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3694 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3695 lookups for user supplementary groups into a cache.
3699 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3700 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3701 for liblustre O_EXCL is enforced by MDS.
3705 Description: readdir never returns NULL in liblustre.
3706 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3707 dentry in struct dirent.
3711 Frequency : liblustre only, depends on application IO pattern
3712 Description: liblustre clients evicted if not contacting servers
3713 Details : Don't put liblustre clients into the ping_evictor list, so
3714 they will not be evicted by the pinger ever.
3716 Severity : enhancement
3718 Description: Add ability to evict clients by NID from MDS.
3719 Details : By echoing "nid:$NID" string into
3720 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3721 $NID would be instantly evicted from this MDS and from all active
3722 OSTs connected to it.
3726 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3727 Details : lookup_it_finish() used to query file size from OSTs that was not
3732 Description: service threads change working directory to that of init
3733 Details : Starting lustre service threads may pin the working directory
3734 of the parent thread, making that filesystem busy. Threads
3735 now change to the working directory of init to avoid this.
3739 Frequency : during shutdown only
3740 Description: shutdown with a failed MDS or OST can cause unmount to hang
3741 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3746 Frequency : 2.6 only
3747 Description: chmod/chown may include an extra supplementary group
3748 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3749 supplementary group and if none is specified this is used.
3752 Bugzilla : 5479 (6816)
3754 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3755 Details : If lookup is in progress on a file that is unlinked we might try
3756 to revalidate the inode and fail in revalidate after lookup is
3757 complete and ll_file_open() enqueues the open again but
3758 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3761 Frequency : always, if lconf --abort_recovery used
3763 Description: lconf --abort_recovery fails with 'Operation not supported'
3764 Details : lconf was attempting to abort recovery on the MDT device and not
3767 ------------------------------------------------------------------------------
3769 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3774 Frequency : rare (only unsupported configurations with a node running as an
3776 Bugzilla : 6514, 5137
3777 Description: Mounting a Lustre file system on a node running as an OST could
3779 Details : OSTs now preallocates memory needed to write out data at
3780 startup, instead of when needed, to avoid having to
3781 allocate memory in possibly low memory situations.
3782 Specifically, if the file system is mounted on on OST,
3783 memory pressure could force it to try to write out data,
3784 which it needed to allocate memory to do. Due to the low
3785 memory, it would be unable to do so and the node would
3786 become unresponsive.
3788 Severity : enhancement
3790 Description: Addition of lconf --service command line option
3791 Details : lconf now accepts a '--service <arg>' option, which is
3792 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3794 Severity : enhancement
3796 Description: Failover mode is now the default for OSTs.
3797 Details : By default, OSTs will now run in failover mode. To return to
3798 the old behaviour, add '--failout' to the lmc line for OSTs.
3800 Severity : enhancement
3802 Description: Health checks are now provided for MDS and OSTs
3803 Details : Additional detailed health check information on MSD and OSTs
3804 is now provided through the procfs health_check value.
3807 Frequency : occasional, depends on IO load
3809 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3810 after numerous create/delete cycles
3811 Details : The ext3 inode allocation policy would not allocate new inodes
3812 very well on the OSTs because there are no new directories
3813 being created. Instead we look for groups with free space if
3814 the parent directories are nearly full.
3819 Description: Network or server problems during mount may cause partially
3820 mounted clients instead of returning an error.
3821 Details : The config llog parsing code may overwrite the error return
3822 code during mount error handling, returning success instead
3828 Description: MDS can fail to allocate large reply buffers
3829 Details : After long uptimes the MDS can fail to allocate large reply
3830 buffers (e.g. zconf client mount config records) due to memory
3831 fragmentation or consumption by the buffer cache. Preallocate
3832 some large reply buffers so that these replies can be sent even
3833 under memory pressure.
3837 Frequency : rare (liblustre)
3838 Description: fsx running with liblustre complained that using truncate() to
3839 extend the file doesn't work. This patch corrects that issue.
3840 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3841 ATTR_SIZE and lsm use in llu_setattr_raw.
3845 Frequency : rare, only 2.6 kernels
3846 Description: Unusual file access patterns on the MDS may result in inode
3847 data being lost in very rare circumstances.
3848 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3849 code in the 2.6 kernel caused the inode and/or EA data not to
3850 be read from disk, causing single-file corruption.
3854 Frequency : rare, only 2.6 filesystems using extents
3855 Description: Heavy concurrent write and delete load may cause data corruption.
3856 Details : It was possible under high-load situations to have an extent
3857 metadata block in the block device cache from a just-unlinked
3858 file overwrite a newly-allocated data block. We now unmap any
3859 metadata buffers that alias just-allocated data blocks.
3863 Frequency : filesystems with default stripe_count larger than 77
3864 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3865 Details : lconf specifies an inode size of 4096 bytes when the default
3866 stripe_count is larger than 77. This conflicts with the default
3867 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3868 this case to avoid pinning too much memory for large EAs.
3870 ------------------------------------------------------------------------------
3872 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3877 Frequency : rare (extremely heavy IO load with hundreds of clients)
3879 Description: Client is evicted, gets IO error writing to file
3880 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3881 caused clients to be evicted for AST timeouts. The fixes in
3882 bug 5192 mean we no longer need to have such short AST timeouts
3883 so ldlm_timeout has been increased.
3886 Frequency : occasional during --force or --failover shutdown under load
3887 Bugzilla : 5949, 4834
3888 Description: Server oops/LBUG if stopped with --force or --failover under load
3889 Details : a collection of import/export refcount and cleanup ordering
3890 issues fixed for safer force cleanup
3893 Frequency : only filesystems larger than 120 OSTs
3894 Bugzilla : 5990, 6223
3895 Description: lfs getstripe would oops on a very large filesystem
3896 Details : lov_getconfig used kfree on vmalloc'd memory
3899 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3900 Bugzilla : 6242, 6243
3901 Description: reading from files that had been truncated to a non-zero size
3902 but never opened returned no data
3903 Details : ll_file_read() reads zeros from no-object files to EOF
3908 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3909 Details : There's little harm in aborting MDS/OSS recovery and letting it
3910 try again, so I removed the LASSERT and return an error instead.
3912 Severity : enhancement
3914 Description: New debugging infrastructure for tracking down data corruption
3915 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3916 (b) cover more of the client-side code path, and (c) try to narrow
3917 down where problems occurred
3921 Bugzilla : 3819, 4364, 4397, 6313
3922 Description: Racing close and eviction MDS could cause assertion in mds_close
3923 Details : It was possible to get multiple mfd references during close and
3924 client eviction, leading to one thread referencing a freed mfd.
3926 Severity: : enhancement
3927 Bugzilla : 3262, 6359
3928 Description: Attempts to reconnect to servers are now more aggressive.
3929 Details : This builds on the enhanced upcall-less recovery that was added
3930 in 1.4.2. When trying to reconnect to servers, clients will
3931 now try each server in the failover group every 10 seconds. By
3932 default, clients would previously try one server every 25 seconds.
3937 Description: After recovery, certain operations trigger a failed
3938 assertion on a client.
3939 Details : Failing over an mds, using lconf -d --failover, while a
3940 client was doing a readdir() call would cause the client to
3941 LBUG after recovery completed and the readdir() was resent.
3943 Severity : enhancement
3945 Description: Default groups are now added by lconf
3946 Details : You can now run lconf --group <servicename> without having to
3947 manually add groups with lmc.
3950 Frequency : occasional
3952 Description: Nodes with an elan id of 0 trigger a failed assertion
3955 Frequency : always when accessing e.g. tty/console device nodes
3957 Description: tty and some other devices nodes cannot be used on lustre
3958 Details : file's private_data field is used by device data and lustre
3959 values in there got lost. New field was added to struct file to
3960 store fs-specific private data.
3963 Frequency : when exporting Lustre via NFS
3965 Description: NFSD failed occasionally when looking up a path component
3966 Details : NFSD is looking up ".." which was broken in ext3 directories
3967 that had grown large enough to become hashed.
3970 Frequency : Clusters with multiple interfaces not on the same subnet
3972 Description: Nodes will repeatedly try to reconnect to an interface which it
3973 cannot reach and report an error to the log.
3974 Details : Extra peer list entries will be created by lconf with some peers
3975 unreachable. lconf now validates the peer before adding it.
3978 Frequency : Only if a default stripe is set on the filesystem root.
3980 Description: Setting a default stripe on the filesystem root prevented the
3981 filesystem from being remounted.
3982 Details : The client was sending extra request flags in the root getattr
3983 request and did not allocate a reply buffer for the dir EA.
3986 Frequency : occasional, higher if lots of files are accessed by one client
3987 Bugzilla : 6159, 6097
3988 Description: Client trips assertion regarding lsm mismatch/magic
3989 Details : While revalidating inodes the VFS looks up inodes with ifind()
3990 and in rare cases can find an inode that is being freed.
3991 The ll_test_inode() code will free the lsm during ifind()
3992 when it finds an existing inode and then the VFS later attaches
3993 this free lsm to a new inode.
3997 Bugzilla : 6422, 7030
3998 Description: MDS deadlock between mkdir and client eviction
3999 Details : Creating a new file via mkdir or mknod (starting a transaction
4000 and getting the ns lock) can deadlock with client eviction
4001 (gets ns lock and trying to finish a synchronous transaction).
4004 Frequency : occasional
4005 Description: While starting a server, the fsfilt_ext3 module could not be
4007 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
4008 kernels. Previously, lconf would still use the ext3 name
4009 when trying to load modules. Now, it will correctly use
4010 ext3 on 2.4 and ldiskfs on 2.6.
4012 Severity : enhancement
4013 Description: The default stripe count has been changed to 1
4014 Details : The interpretation of the default stripe count (0, to lfs
4015 or lmc) has been changed to mean striping across a single
4016 OST, rather than all available. For general usage we have
4017 found a stripe count of 1 or 2 works best.
4019 Severity : enhancement
4020 Description: Add support for compiling against Cray portals.
4021 Details : Conditional compiling for some areas that are different
4025 Frequency : occasional
4026 Bugzilla : 6409, 6834
4027 Description: Creating files with an explicit stripe count may lead to
4028 a failed assertion on the MDS
4029 Details : If some OSTs are full or unavailable, creating files may
4030 trigger a failed assertion on the MDS. Now, Lustre will
4031 try to use other servers or return an error to the
4035 Frequency : occasional
4037 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4038 caused lock timeout during readahead (since 1.4.2).
4039 Details : Processes doing readahead might match a lock that hasn't been
4040 granted yet if there are overlapping and conflicting lock
4041 requests. The readahead process waits on ungranted lock
4042 (original lock is CBPENDING), while OST waits for that process
4043 to cancel CBPENDING read lock and eventually evicts client.
4045 Severity : enhancement
4047 Description: Initial enabling of flock support for clients
4048 Details : Implements fcntl advisory locking and file status functions.
4049 This feature is provided as an optional mount flag (default
4050 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4051 locking are implemented yet, and those that are are not guaranteed
4052 to be completely correct in production environments.
4053 mount -t lustre -o [flock|noflock] ...
4056 Frequency : occasional
4058 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4059 assertion in the ext3 JBD (journaling) layer.
4060 Details : The b_committed_data struct is protected by the big kernel lock
4061 in 2.4 kernels, serializing journal_commit_transaction() and
4062 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4063 access to this struct. In 2.6 kernels there is finer grained
4064 locking to improve SMP performance of the JBD layer.
4068 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4073 Description: merge and cleanup kernel patches.
4076 -----------------------------------------------------------------------------
4078 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4080 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4081 versions of Lustre. Please update all servers and clients to
4082 version 1.4.2 or later at the same time. You must also run
4083 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4084 to update the configuration logs.
4086 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4087 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4088 - swab llog records of type '0' so we get proper header size/idx (5861)
4089 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4090 - fix rename of one directory over another leaking an inode (5953)
4091 - avoid SetPageDirty on 2.6 (5981)
4092 - don't re-add just-being-destroyed locks to the waiting list (5653)
4093 - when creating new directories, inherit the parent's custom
4094 striping settings if present parent (3048)
4095 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4096 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4097 - don't hold client locks on temporary worklist from l_lru (5666)
4098 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4099 - automatically evict dead clients (5921)
4100 - Update file size properly in create+truncate+fstat case (6196)
4101 - Do not unhash mountpoint dentries, do not allow removal of
4103 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4104 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4105 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4106 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4107 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4108 - Correctly handle reads of files with no objects (6243)
4109 - lctl recover will also mark a device active if deactivate used (5933)
4111 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
4112 - allow --write-conf on an MDS with different nettype than client (5619)
4113 - don't write config llogs to MDS for mounts not from that MDS (5617)
4114 - lconf should create multiple TCP connections from a client (5201)
4115 - init scripts are now turned off by default; run chkconfig --on
4116 lustre and chkconfig --on lustrefs to use them
4117 - upcalls are no longer needed for clients to recover to failover
4119 - add --abort-recovery option to lconf to abort recovery on device
4121 - add support for an arbitrary number of OSTs (3026)
4122 - Quota support protocol changes.
4123 - forward compatibility changes to wire structs (6007)
4124 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
4125 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
4126 - improved "lustre status" script
4127 - initialize blocksize for non-regular files (6062)
4128 - added --disable-server and --disable-client configure options (5782)
4129 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
4130 - Vanilla 2.4.29 support
4131 - increase maximum number of obd devices to 520 (6242)
4132 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4133 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4134 SLES 9 2.6.5 kernels
4135 - limit stripes per file to 160 (the maximum EA size) (6093)
4137 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4140 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4141 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4142 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4143 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4144 - small ext3 extents cleanups and fixes (5733)
4145 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4146 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4147 - avoid needless client->OST connect, fix handle mismatch (5317)
4148 - fix DLM error path that led to out-of-sync client, long delays (5779)
4149 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4150 - fix several locking issues related to i_size (5492,5624,5654,5672)
4151 - don't move pending lock onto export if it is already evicted (5683)
4152 - fix kernel oops when creating .foo in unlinked directory (5548)
4153 - fix deadlock in obdfilter statistics vs. object create (5811)
4154 - use time_{before,after} to avoid timer jiffies wrap (5882)
4155 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4156 - Do not leak request if server was not able to process it (5154)
4157 - If mds_open unable to find parent dir, make that negative lookup(5154)
4158 - don't create new directories with extent-mapping (5909, 5936)
4160 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
4161 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
4162 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
4163 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
4164 - support for RHEL4 kernel on i686 (5773)
4165 - provide error messages when incompatible logs are encountered (5898)
4167 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
4168 * version 1.4.0.10 (1.4.1 release candidate 1)
4170 - don't keep a lock reference when lock is not granted (4238)
4171 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4172 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
4173 - drop import inflight refcount on signal_completed_replay error (5255)
4174 - unlock page after async write error during send (3677)
4175 - handle missing objects in filter_preprw_read properly (5265)
4176 - no transno return for symlink open, don't save no-trasno open (3440)
4177 - don't try to complete elan receive that already failed (4012)
4178 - free RPC server reply state on error (5406)
4179 - clean up thread from ptlrpc_start_thread() on error (5160)
4180 - readahead could read extra page into cache that wasn't ejected (5388)
4181 - prevent races in class_attach/setup/cleanup/detach (5260)
4182 - don't dereference de->d_inode after l_dput of de (5458)
4183 - use "int" for stripe value returned from lock_to_stripe (5544)
4184 - mballoc allocation and error-checking fixes in 2.6 (5504)
4185 - block device patches to fix I/O request sizes in 2.6 (5482)
4186 - look up hostnames for IB nals (5602)
4187 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
4188 - don't start multiple acceptors for the same port (5277)
4189 - fix incorrect LASSERT in mds_getattr_name (5635)
4190 - export a proc file for general "ping" checking (5628)
4191 - fix "lfs check" to not block when the MDS is down (5628)
4193 - service request history (4965)
4194 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4195 - create an "evict_client" /proc entry on OSTs, like the MDS has
4196 - fix mount usage message, return errors per mount(8) (5168)
4197 - change grep [] to grep "[]" in tests so they work in more UMLs
4198 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4199 - remove ancient LOV_MAGIC_V0 EA support (5047)
4200 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4201 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4202 - allow readahead allocations to fail when low on memory (5383)
4203 - mmap locking landed again, after considerable improvement (2828)
4204 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4206 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4209 - send OST transaction number in read/write reply to free req (4966)
4210 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4211 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4212 - the watchdog thread now runs as interruptible (5246)
4213 - flock/lockf fixes (but it's still disabled, pending 5135)
4214 - don't use EXT3 constants in llite code (5094)
4215 - memory shortage at startup could cause assertion (5176)
4217 - reorganization of lov code
4218 - single portals codebase
4220 - add extents/mballoc support (5025)
4221 - direct I/O reads in the obdfilter (4048)
4222 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4224 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4227 - send OST transaction number in read/write reply to free req (4966)
4228 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4229 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4230 - the watchdog thread now runs as interruptible (5246)
4231 - handle missing objects in filter_preprw_read properly (5265)
4232 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4233 - drop import inflight refcount on signal_completed_replay error (5255)
4234 - unlock page after async write error during send (3677)
4235 - return original error code on reconstructed replies (3761)
4236 - no transno return for symlink open, don't save no-trasno open (3440)
4238 - add pid to ldlm debugging output (4922)
4239 - bump the watchdog timeouts -- we can't handle 30sec yet
4240 - extra debugging for orphan dentry/inode bug (5259)
4242 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
4245 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4246 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4247 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
4248 - fix NULL dereference in /proc/sys/portals/routes (4827)
4249 - allow failed mdc_close() operations to be interrupted (4561)
4250 - stop precreate on OST before MDS would time out on it (4778)
4251 - don't send partial-page writes before EOF from client (4410)
4252 - discard client grant for sub-page writes on large-page clients (4520)
4253 - don't free dentries not owned by NFS code, check generation (4806)
4254 - fix lsm leak if mds_create_objects() fails (4801)
4255 - limit debug_daemon file size, always print CERROR messages (4789)
4256 - use transno after validating reply (3892)
4257 - process timed out requests if import state changes (3754)
4258 - update mtime on OST during writes, return in glimpse (4829)
4259 - add mkfsoptions to LDAP (4679)
4260 - use ->max_readahead method instead of zapping global ra (5039)
4261 - don't interrupt __l_wait_event() during strace
4263 - add software watchdogs to catch hung threads quickly (4941)
4264 - make lustrefs init script start after nfs is mounted
4265 - fix CWARN/ERROR duplication (4930)
4266 - return async write errors to application if possible (2248)
4267 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4268 - print NAL number in %x format (4645)
4269 - update barely-supported suse-2.4.21-171 series (4842)
4270 - support for sles 9 %post scripts
4271 - support for building 2.6 kernel-source packages
4272 - support for sles km_* packages
4274 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4277 - ignore -ENOENT errors in osc_destroy (3639)
4278 - notify osc create thread that OSC is being cleaned up (4600)
4279 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4280 - reconstruct ost_handle() like mds_handle() (4657)
4281 - create a new thread to do import eviction to avoid deadlock (3969)
4282 - let lconf resolve symlinked-to devices (4629)
4283 - don't unlink "objects" from directory with default EA (4554)
4284 - hold socknal file ref over connect in case target is down (4394)
4285 - allow more than 32000 subdirectories in a single directory (3244)
4286 - fix blocks count for O_DIRECT writes (3751)
4287 - OST returns ENOSPC from object create when no space left (4539)
4288 - don't send truncate RPC if file size isn't changing (4410)
4289 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4290 - bind to privileged port in socknal and tcpnal (3689)
4292 - rate limit CERROR/CWARN console message to avoid overload (4519)
4293 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4294 - basic mmap support (3918)
4295 - kernel patch series update from b1_4 (4711)
4297 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4300 - avoid crash during MDS cleanup with OST shut down (2775)
4301 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4302 - don't use bad inodes on the MDS (3744)
4303 - dynamic object preallocation to improve recovery speed (4236)
4304 - don't hold spinlock over lock dumping or change debug flags (4401)
4305 - don't zero obd_dev when it is force cleaned (3651)
4306 - print grants to console if they go negative (4431)
4307 - "lctl deactivate" will stop automatic recovery attempts (3406)
4308 - look for existing locks in ldlm_handle_enqueue() (3764)
4309 - don't resolve lock handle twice in recovery avoiding race (4401)
4310 - revalidate should check working dir is a directory (4134)
4312 - don't always mark "slow" obdfilter messages as errors (4418)
4314 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4317 - don't close LustreDB during write_conf until it is done (3860)
4318 - fix typo in lconf for_each_profile (3821)
4319 - allow dumping logs from multiple threads at one time (3820)
4320 - don't allow multiple threads in OSC recovery (3812)
4321 - fix debug_size parameters (3864)
4322 - fix mds_postrecov to initialize import for llog ctxt (3121)
4323 - replace config semaphore with spinlock (3306)
4324 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4325 - don't allow enqueue to complete on a destroyed export (3822)
4326 - down write_lock before checking llog header bitmap (3825)
4327 - recover from lock replay timeout (3764)
4328 - up llog sem before sending rpc (3652)
4329 - reduce ns lock hold times when setting kms (3267)
4330 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4331 - fix NULL deref and obd_dev leak on setup error (3312)
4332 - replace some LBUG about llog ops with error handling (3841)
4333 - don't match INVALID dentries from d_lookup and spin (3784)
4334 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4335 - fix invalid assertion in ptlrpc_set_wait (3880)
4337 - add libwrap support for the TCP acceptor (3996)
4338 - add /proc/sys/portals/routes for non-root route listing (3994)
4339 - allow setting MDS UUID in .xml (2580)
4340 - print the stack of a process that LBUGs (4228)
4342 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4345 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4346 - make sure to unset replay flag from failed open requests (3440)
4347 - if default stripe count is 0, use OST count for inode size (3636)
4348 - update parent mtime/ctime on client for create/unlink (2611)
4349 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4350 - free recovery state on server during a forced cleanup (3571)
4351 - unregister_reply for resent reqs (3063)
4352 - loop back devices mounting and status check on 2.6 (3563)
4353 - fix resource-creation race that can provoke i_size == 0 (3513)
4354 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4355 - more debugging for page-accounting assertion (3746)
4356 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4357 - avoid deadlock after precreation failure (3758)
4358 - fix race and lock order deadlock in orphan handling (3450, 3750)
4359 - add validity checks when grabbing inodes from l_ast_data (3599)
4361 - add /proc/.../recovery_status to obdfilter (3428)
4362 - lightweight CDEBUG infrastructure, debug daemon (3668)
4363 - change default OSC RPC parameters to be better on small clusters
4364 - turn off OST read cache for files smaller than 32MB
4365 - install man pages and include them in rpms (3100)
4366 - add new init script for (un)mounting lustre filesystems (2593)
4367 - run chkconfig in %post for init scripts (3701)
4368 - drop scimac NAL (unmaintained)
4370 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4373 - clean kiobufs before and after use (3485)
4374 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4375 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4376 - change init script to fail more clearly if not run as root (1528)
4377 - allow clients to reconnect during replay (1742)
4378 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4379 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4380 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4381 - fix cond_resched (3554)
4382 - only evict unfinished clients after recovery (3515)
4383 - allow bulk resend, prevent data loss (3570)
4384 - dynamic ptlrpc request buffer allocation (2102)
4385 - don't allow unlinking open directory if it isn't empty (2904)
4386 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4387 - remove extraneous obd dereference causing LASSERT failure (3334)
4388 - don't use get_cycles() when creating temp. files on the mds (3156)
4389 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4390 - handle EEXIST for set-stripe, set proper directory name (3336)
4392 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4393 - fix ksocknal_fmb_callback() error messages (2918)
4395 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4398 - don't copy lvb into (possibly NULL) reply on error (2983)
4399 - don't deref dentry after dput, don't free lvb on error (2922)
4400 - use the kms to determine writeback rpc length (2947)
4401 - increment oti_logcookies when osc is inactive (2948)
4402 - update client's i_blocks count via lvb messages (2543)
4403 - handle intent open/close of special files properly (1557)
4404 - mount MDS with errors=remount-ro, like obdfilter (2009)
4405 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4406 - don't use cancelling-locks' kms values (2947)
4407 - use highest lock extent for kms, not last one (2925)
4408 - don't dereference ERR_PTR() dentry in error handling path (3107)
4409 - fix thread race in portals_debug_dumplog() (3122)
4410 - create lprocfs device entries at setup instead of at attach (1519)
4411 - common AST error handler, don't evict client on completion race (3145)
4412 - zero nameidata in detach_mnt in 2.6 (3118)
4413 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4414 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4415 - initialize RPC timeout timer earlier for 2.6 (3219)
4416 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4417 - print nal/nid for unknown nid (3258)
4418 - additional checks for oscc recovery before doing precreate (3284)
4419 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4420 - don't crash in mdc_close for bad permissions on open (3285)
4421 - zero i_rdev for non-device files (3147)
4422 - clear page->private before handing to FS, better assertion (3119)
4423 - tune the read pipeline (3236)
4424 - fix incorrect decref of invalidated dentry (2350)
4425 - provide read-ahead stats and refine rpc in flight stats (3328)
4426 - don't hold journal transaction open across create RPC (3313)
4427 - update atime on MDS at close time (3265)
4428 - close LDAP connection when recovering to avoid server load (3315)
4429 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4430 - don't leak open file on MDS after open resend (3325)
4431 - serialize filter_precreate and filter_destroy_precreated (3329)
4432 - loop device shouldn't call sync_dev() for nul device (3092)
4433 - clear page cache after eviction (2766)
4434 - resynchronize MDS->OST in background (2824)
4435 - refuse to mount the same filesystem twice on same mountpoint (3394)
4436 - allow llmount to create routes for mounting behind routers (3320)
4437 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4438 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4439 - fix rare problem with rename on htree directories (3417)
4441 - allow default OST striping configuration per directory (1414)
4442 - fix compilation for qswnal for 2.6 kernels (3125)
4443 - increase maximum number of MDS request buffers for large systems
4444 - change liblustreapi to be useful for external progs like lfsck (3098)
4445 - increase local configuration timeout for slow disks (3353)
4446 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4448 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4451 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4452 - don't overwrite extent policy data in reply if lock was blocked (2901)
4453 - drop filter export grants atomically with removal from device (2663)
4454 - del obd_self_export from work_list in class_disconnect_exports (2908)
4455 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4456 - swab reply message in mdc_close, other PPC fixes (2464)
4457 - fix destroying of named logs (2325)
4458 - overwrite old logs when running lconf --write_conf (2264)
4459 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4460 - fix race in target_handle_connect (2898)
4461 - mds_reint_create() should take same inode create lock (2926)
4462 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4463 - don't close files for self_export to avoid uninitialized obd (2936)
4464 - allow MDS with the same name as client node (2939)
4465 - hold dentry reference for closed log files for unlink (2325)
4466 - reserve space for all logs during transactions (2059)
4467 - don't evict page beyond end of stripe extent (2925)
4468 - don't oops on a deleted current working directory (2399)
4469 - handle hard links to targets without a parent properly (2517)
4470 - don't dereference NULL lock when racing during eviction (2867)
4471 - don't grow lock extents when lots of conflicting locks (2919)
4473 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4476 - account for cache space usage on clients to avoid data loss (974)
4477 - lfsck support in lustre kernel code (2349)
4478 - reduce journal credits needed for BRW writes (2370)
4479 - orphan handling to avoid losing space on client/server crashes
4480 - ptlrpcd can be blocked, stopping ALL progress (2477)
4481 - use lock value blocks to assist in proper KMS, faster stat (1021)
4482 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4483 - recovery for initial connections (2355)
4484 - fixes for mds_cleanup_orphans (1934)
4485 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4486 - block all file creations until orphan recovery completes (1901)
4487 - client remove rq_connection from request struct (2423)
4488 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4489 - recovery timer race (2670)
4490 - mdc_close recovey bug (2532)
4491 - ptlrpc cleanup bug (2710)
4492 - mds timeout on local locks (2588)
4493 - namespace lock held during RPCs (2431)
4494 - handle interrupted sync write properly (2503)
4495 - don't try to handle a message that hasn't been replied to (2699)
4496 - client assert failure during cleanup after abort recovery (2701)
4497 - leak mdc device after failed mount (2712)
4498 - ptlrpc_check_set allows timedout requests to complete (2714)
4499 - wait for inflight reqs when ptlrpcd finishes (2710)
4500 - make sure unregistered services are removed from the srv_list
4501 - reset bulk XID's when resending them (caught by 1138 test)
4502 - unregister_bulk after timeout
4503 - fix lconf error (2694)
4504 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4505 - readahead locks pages, leaves pending causing memory pressure (2673)
4506 - increase OST request buffers to 4096 on large machines (2729)
4507 - fix up permission of existing directories in simple_mkdir (2661)
4508 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4509 - don't assign transno to errored transactions (2742)
4510 - don't delete objects on OST if given a bogus objid from MDS (2751)
4511 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4512 - if rq_no_resend, then timeout request after recovery (2432)
4513 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4514 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4515 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4516 - MDS recovery shouldn't delete objects that it has given out (2730)
4517 - if enqueue arrives after completion, don't clobber LVB (2819)
4518 - don't unlock pages twice when trigger_group_io returns error (2814)
4519 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4520 - don't write pages to disk if there was an error (1450)
4521 - don't ping imports that have recovery disabled (2676)
4522 - take buffered bytes into account when balancing socknal conn (2817)
4523 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4524 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4525 - remove little-endian swabbing of llog records (1987)
4526 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4527 - retry reposting request buffers if they fail (1191)
4528 - grow extent at grant time to avoid granting a revoked lock (2809)
4529 - lock revoke doesn't evict page if covered by a second lock (2765)
4530 - disable VM readahead to avoid reading outside lock extents (2805)
4532 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4533 - updated kernel patches for hp-2.4.20 kernel (2681)
4535 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4538 - fix truncated write corruption (2366)
4539 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4541 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4542 - ensure len(uuid) < 37 in lmc (1171)
4543 - fix ia64 OOPS in llog_test (2255)
4544 - zero end of page at obdfilter for partial page writes (2648)
4545 - don't leave stale dentries around after renames (bug 2428)
4546 - fix timeouts when evicting a client with a single lock held (2642)
4547 - set deadline for the initial HELLO message to drain (2634)
4548 - print out dotted-quad IP addresses in the socknal (2302)
4550 - additional debugging for MDS client eviction problem (2443)
4551 - fix mkfsoptions support for osts (2603, 2604)
4553 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4556 - add series for the vanilla 2.6.0 kernel
4557 - add series for the vanilla 2.4.24 kernel
4558 - add series for a cray x86/64 UL kernel drop
4559 - fix xattr patches for the vanilla 2.4.19 series
4561 - generate true UUIDs in lmc (1171)
4562 - have portals stack dumping break in UML (2466)
4563 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4564 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4565 - wake callback waiting threads on client eviction (2460)
4566 - Add --ptldebug and --subsystem to lmc (1719)
4567 - update assertion to allow safe interrupt allocation
4568 - set rq_no_resend for cancel requests (2432)
4569 - recalculate ptlrpcd timeout after resend (2494)
4570 - call vfs_rmdir when removing pending directories (2368)
4571 - fix renaming a file to itself (2429)
4572 - lmc creates a default one-stripe lov (2454)
4573 - expand procfs space to handle large clusters (2326)
4574 - increase UML stack to avoid overflow
4575 - update lconf's list of debug and subsystem masks
4576 - fix lfs find --obd (2510)
4577 - /proc tunable for disabling filter read caching (2591)
4578 - stop rpm packages from altering slapd.conf (2301)
4579 - disable nagle in the socknal under 0conf (2578)
4580 - choose mds inode size based on stripe count (2572)
4581 - fix kernel-source rpm problems (2516)
4583 - add --disable-doc to avoid pdf generation (2421)
4584 - update documentation, tests, type-os, comments
4585 - avoid format warnings on ia64
4586 - remove the TOE NAL
4587 - tiny code cleanups by removing unused fields
4589 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4592 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4593 - remove the most blatant lies from BUILDING file (2371)
4594 - change default debug level to reasonable production setting
4595 - reduce client side cache size to reduce cache flush time
4596 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4597 - make TCP zerocopy and pinger support enabled by default (2476)
4598 - sync writes completed after process exits caused crashes (2319)
4599 - maintain correct mount count on the MDS (2356)
4600 - backout 1557, because 2316 wasn't really fixed
4601 - better file I/O statistics gathering in /proc
4602 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4603 - another kernel patch to fix zero-copy TCP function export
4604 - don't take duplicate lock when processing re-sent getattr (2420)
4605 - lctl uses obd_self_export instead of creating new conn (2353)
4606 - MDS/OST recovery case which requires object creation asserted (2425)
4607 - move lfs from /usr/sbin to /usr/bin in packages
4608 - fix race between mds_client_add and mds_client_free (2417)
4609 - use kmalloc instead of slabs in portals (2430)
4610 - don't create duplicate records when a failover MDS is present (2442)
4611 - remove unnecessary mount age check (2332)
4612 - don't remove directory inodes from locks prematurely (2451)
4613 - don't break if MDS service name is the same as hostname (2103)
4614 - fix races in client write RPC generation when cache full (2482)
4616 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4619 - remove now-unused request->rq_obd (278)
4620 - if an allocation fails, print out how much memory we've used (1933)
4621 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4622 - add the "configurable stack size" patch to most series files (1256)
4623 - ability to write large log records, for 100+ OST configs (2306)
4624 - fix NULL deref when filter_prep fails (2314)
4625 - fix operator precedence error in filter_sync
4626 - dynamic allocation of socknal TX descriptors (2315)
4627 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4628 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4629 - ensure that CWARN messages in Portals always get to the syslog
4630 - __init/__exit are not for prototype decls (ldlm_init/exit)
4631 - x86-64 compile warning fixes
4632 - fix gateway LMC keyword conflict (2318)
4633 - fix MDS lock inversions in getattr/reint paths (1844)
4634 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4635 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4636 - DLM race condition prevented some lock evictions (2328)
4637 - ENOMEM detection and retry on socknal sends (2230)
4638 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4639 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4640 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4641 - better error messages when a client is rejected during recovery (1505)
4642 - avoid cancelling locks which were never granted, after failure (2330)
4643 - fix i_sem/journal inversion in mds_client_add (2333)
4644 - fix truncate/getattr lock cycle deadlock (2334)
4645 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4646 - fix two rare exit paths which could leak an l_lock() ref (2321)
4647 - fencepost error in MDS/OST orphan recovery (2226)
4648 - make log record alignment 8 bytes (1988)
4649 - lstripe now fails when requested offset > ost_count (2237)
4650 - ensure that all kernel series have a complete list.h (1607)
4651 - fix crashes in special-file operations (2316)
4652 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4654 - allow configurable automake binary, for testing new versions
4655 - small update to the lfs documentation
4657 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4659 * fix negative export reference count in fsfilt_sync (2312)
4661 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4662 * release candidate 0.9.1
4664 - orphans are moved into the PENDING directory for possible recovery
4665 - replayed opens now open by fid for orphan/rename safety (1042)
4666 - last close of an orphan inode generates a transno (683)
4667 - chdir() and mount() now pin the directory entry (1020)
4668 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4669 - discard very old requests without processing them (1502)
4670 - remove some common, well-understood CERRORs (1505)
4671 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4672 - clear "grant" flags in OST replies until OST grant code lands (1644)
4673 - fix read performance by not clobbering i_blksize on client (1598)
4674 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4675 - make LRU size a /proc tunable, clears locks when reduced (707)
4676 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4677 - server locks take references on exports now (1558)
4678 - build fixes for 2.4.20-rh trees (1663)
4679 - return an error from lov_create if all OSCs are inactive (1751)
4680 - fix import levels when a reconnect happens without a timeout (1597)
4681 - exit early from mds_open if we get a lookup error (1749)
4682 - partial page read at EOF wouldn't wait for disk before sending (1642)
4683 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4684 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4685 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4686 - write/truncate lock inversion (1639)
4687 - Don't auto-load obdclass, portals modules during cleanup (1495)
4688 - fix timestamps from jumping to "now" (1763)
4689 - extra journal assertions (1648)
4690 - add an extra multiunlink test (1771)
4691 - fix read_record/write_record API (1776)
4692 - fix leak of offset_extent, possible incorrect i_size later (1772)
4693 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4694 - Debugging for the kqswnal_get_idle_tx problems (1820)
4695 - Allow recovery to be attempted multiple times (1536)
4696 - Write out MDS last_rcvd file after it is first created (1600)
4697 - Fix tx_descriptor leak in failed transmit situations (1827)
4698 - ext3 journaling fixes for assertion failure after IO error (1871)
4699 - class_export_put() on freed export after completion AST error (1896)
4700 - Fix revalidate looping in VFS (1322)
4701 - Don't access a freed export during MDS_REINT timeout (1521)
4702 - Add open-unlink recovery support on the MDS (1673,1764)
4703 - Return an error if no MDS data was read from last_rcvd (1946)
4704 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4705 - Don't setup a disk device that doesn't match exported UUID (317)
4706 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4707 - avoid committing NULL handle in force close
4708 - local.sh is now a one-stripe LOV configuration
4709 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4710 - fix ext3 htree duplicate directory entry corruption (1516)
4711 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4712 - update to new LOV EA format (2097)
4713 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4715 2003-06-15 Phil Schwan <phil@clusterfs.com>
4718 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4719 - per-import/export recovery handling (958, 931, 959)
4720 - multiple last-rcvd slots, for serving multiple FSes (949)
4721 - connections are again shared between multiple imp/exports (963, 964)
4722 - "umount -f" would hang if any requests needed to be sent (393, 978)
4723 - avoid pinning large req buffer by copying for queued messages (989)
4724 - add "uuid" to "lctl device" command to help upcalls (991)
4725 - "open" RPCs with transnos would confuse recovery counters (1037)
4726 - do proper endian conversion of all wire messages (288, 340, 891)
4727 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4728 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4729 - fix ptlrpc_connection leak in target_handle_connect (1174)
4730 - fix import refcounting bug in OST and MDS cleanup (1134)
4731 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4732 - fix very unlikely obd_types race condition (501)
4733 - remove osc_open hack for echo_client (1187)
4734 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4735 - a failure in read_inode2 leads to deadlock (1139)
4736 - cancel ack-locks as soon as transaction is committed (1072)
4737 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4738 - make sure to commitrw after any preprw to avoid deadlock (1162)
4739 - failing to execute a file in a lustre FS would lock inode (1203)
4740 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4741 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4742 - obd_uuid2tgt would walk off the end of the list (1255)
4743 - on IA64 the osc would give portals incorrect bulk size (1258)
4744 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4745 - fix lock inversion caused by new llite matching code (1282)
4746 - limit the number of dirty pages on a client to 10MB (1286)
4747 - timed out locks were not being corrected cancelled (1289)
4748 - fix O_DIRECT above 4GB on IA-32 (1292)
4749 * major user-visible changes
4750 - fail out/fail over policy now controlled by the upcall (993)
4752 - add OBD_PING to check server availability and failure (954)
4753 - lustre messages are now sent in sending host order (288, 340, 891)
4754 - add eadatalen to MDS getattr reply (340)
4755 - OST read replies may contain second buffer, with per-page status (593)
4757 2003-03-11 Phil Schwan <phil@clusterfs.com>
4760 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4761 - failed open()s could cause deadlock; fixed (867, 869)
4762 - stop cancelling OST locks when files are closed (481)
4763 - overlapping XID spaces caused network corruption (851, 853)
4764 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4765 - setattr_raw added, to do single-RPC, server-side setattrs
4766 - lmc/lconf syntax change for OST UUIDs
4767 - fix crashy race condition between ptlrpc_free_req and osc_close
4768 - don't use request in mdc_enqueue if we hit a timeout (889)
4769 - don't set the inode i_size for regular files from the MDS (896)
4770 - handle out of order completion AST (842)
4771 - don't LBUG if a lock request times out after receiving AST (913)
4772 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4773 - if a bad lock AST arrives, send an error instead of dropping entirely
4774 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4775 - fix leak in bulk IO when only partially completed (899, 900, 926)
4776 - fix O_DIRECT for ia64 (55)
4777 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4778 - C-z after timeout could hang a process forever; fixed (977)
4780 - client-side I/O cache (678, 924, 929, 941, 970)
4782 - READPAGE and SETATTRs which don't take server-side locks get
4785 2003-02-11 Phil Schwan <phil@clusterfs.com>
4788 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4789 - Fix fsfilt_extN_readpage() to read a full page of directory
4790 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4791 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4792 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4793 - Most intent operations were removed, in favour of a new RPC mode
4794 that does a single RPC to the server and bypasses most of the VFS
4795 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4796 - Aggressively cancel local locks on DLM servers
4797 - mds_reint_unlink sends EA to the client if it's the last nlink.
4798 client uses that EA to unlink OST objects.
4799 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4800 - recursive symlinks were fixed (439)
4801 - fixed NULL deref in DEBUG_REQ
4802 - filter_update_lastobjid no longer calls sync, which annoyed extN
4803 - fixed multi-client small-writes to a single file problem (445)
4804 - fixed mtime updates during file writes (607)
4805 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4806 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4807 - fixed Linux OST size reporting problem (444, 656)
4808 - OST now updates object mtime with writes or setattr (607, 619)
4809 - client verifies file size before zeroing page past EOF (445)
4810 - OST now writes last allocated objid to disk with allocation (108)
4811 - LOV on echo now works (409)
4813 - mds_reint_unlink sends a new buffer, with the EA included. this
4814 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4815 set if a regular file was being unlinked, and it was the last link
4816 - use PtlGet from the target for bulk writes (315)
4817 - OST now updates object mtime with writes or setattr (607, 619)
4818 - LDLM now has a grant-time callback to revalidate locked items, if
4820 - Many MDS operations were reorganized to combat race conditions
4822 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4823 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4825 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4828 - Fully reactivate OST imports after reconnection (512, others)
4829 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4830 - More graceful error handling for truncating on dead OST (515)
4831 - Don't error out unless we're actually accessing dead stripes (474)
4832 - Fix garbage sizes when stripes are missing (410)
4833 - LRU counters were broken, causing constant lock purge (433, 432)
4834 - garbage on read from stripes with failed OSTs (441)
4835 - mark OSCs as active before reconnecting during recovery (438)
4836 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4837 - lfind did not preserve OST order in output (443)
4838 - symlinks cause hung clients, incorrect data (439)
4839 - stop dereferencing request after dropping refcount (457)
4840 - don't LASSERT(spin_is_locked) on non-SMP (455)
4841 - fixes for many rename() bugs
4842 - fstat didn't correctly synchronize attributes (399)
4843 - server must handle lock cancellation during blocking AST prep (487)
4844 - bulk descriptors were free()d too soon (511)
4845 - fix paths in lconf, which would load incorrect modules (451, 507)
4846 - fix confusing lconf 'host not found' error message (386)
4847 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4848 - fix race condition in mdc_blocking_ast() for inode access (526)
4849 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4850 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4851 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4852 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4853 - fix lstripe in several ways (kernel side) (527)
4854 - fix request leak in ldlm_cli_enqueue (262)
4855 - incorrect OSC was marked inactive after OST failure
4856 - call mds_fs_cleanup before unmounting filesystem (524)
4857 - fix races between taking ns_lock and ldlm_lock_change_resource
4858 - fix races updating LOV export open file list
4859 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4860 - fix recovery NULL deref in ldlm_cli_cancel_unused
4861 - fix some DLM races by using new hash table for lock handles (419)
4862 - permit the client to specify desired inodes, at replay
4863 - duplicate requests when we queue them for replay reintegration
4864 - fix last_rcvd offset calculation
4865 - sync after each recovered transaction, so we always make progress
4866 - never, not always, ERESTART requests without transnos
4867 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4869 - skip replay if the MDS says that the client is already connected
4870 - don't check for a recovery-enabled export to match lctl's UUID
4871 - don't INC_USE_COUNT for phantom exports
4872 - don't crash when cleaning up phantom exports (567)
4873 - don't double-finish or set replay data for errored mdc_open requests
4874 - abort requests when they time out, so we don't get old replies
4875 - send/receive replies for AST messages again
4876 - if the client says that it doesn't have the lock, cancel it on the
4878 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4879 mark it as destroyed, it will all work out when decref is called
4880 - fix module use counts (22, 581)
4882 - ASTs now expect a reply (server cancels lock on error reply)
4884 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4887 - fix many simultaneous client startup (392)
4888 - fix dentry->d_it clobbering
4889 - credentials weren't being shipped for readdir/getattr operations
4890 - remove invalid assertions triggered during some concurrent MD
4892 - proper Lustre versions added (336, 389)
4893 - fix memory leak for create error case (398)
4894 - fix LOV locking bug that would get cli/srv out of sync
4895 - fix echo client over LOV (409)
4896 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4897 - fix double-O_EXCL intent crash (424)
4898 - avoid sending multiple lock CANCELs (352)
4900 - MDS can do multi-client recovery (modulo bugs in new code)
4902 - many updates, edits, cleanups
4904 2002-11-18 Phil Schwan <phil@clusterfs.com>
4907 - fix null d_it dereference (346)
4908 - fix full OST/dbench hang (333)
4909 - fix permission problem with file removal (286)
4910 - fix removal of OSCs from LOV when they fail
4911 - fix NULL deref during bulk timeout (214)
4912 - fix problems related to multiple filesystems on one MDS (241)
4913 - fixed serious subtle metadata locking bugs
4914 - free locks on clients when inodes are removed due to memory
4916 - fix inode pointer in lock data (285)
4917 - partial support for multiple MDS on a single host (241)
4918 - data locks weren't cancelled at clear_inode time (290, 311)
4919 - intent locks could lead to unbounded lock growth (205)
4920 - added a maximum lock count, an LRU list, and a flusher
4921 - fix multiple rename (365)
4922 - properly abstracted the echo client
4923 - OSC locked 1 byte too many; fixed
4924 - rewrote brw callback code:
4925 - fixed recovery bugs related to LOVs (306)
4926 - fixed too-many-pages-in-one-write crash (191)
4927 - fixed (again) crash in sync_io_timeout (214)
4928 - probably fixed callback-related race (385)
4930 - Add capability to MDS protocol
4931 - LDLM cancellations and callbacks on different portals
4933 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4936 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4937 - defer open object destruction to close time (601981, 138)
4938 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4939 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4940 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4941 - don't call obd_setattr in ll_file_release for destroyed objects
4942 * protocol change to lustre_msg: move |version| and add |flags|
4943 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4944 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4945 * added replay of create, unlink, link and rename operations during
4946 MDS failover; recovery should be much more robust now
4947 * remove failed OSCs from LOVs (only lov_create uses this so far)
4948 * the lustre-HOWTO was brought (more) up to date (582544)
4950 2002-10-23 Phil Schwan <phil@clusterfs.com>
4953 - in-use dentries weren't being reused properly (617851)
4954 - prevent multiple LDLM setup (599178)
4955 - fix LOV size calculations for truncate (617853)
4956 - fix client handling of MDS intent errors (POSIX)
4957 - fix permission bug in lovstripe.c test (624321)
4958 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4959 - truncate past end of file could corrupt data
4960 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4961 - a race in recovery could return ETIMEDOUT to apps (623947)
4962 - building outside the source directory was fixed
4963 * the lustre-HOWTO was brought (more) up to date (582544)
4964 * major progress was made on recovery functionality
4966 2002-10-10 Phil Schwan <phil@clusterfs.com>
4969 - recovery deadlock fix
4970 - rm -rf causes LBUG fix (617817)
4971 - file open by multiple tasks fix (618962)
4972 - directory permissions bugs (602707 and 620007)
4973 - journal_stop fixed with locking (611313)
4974 - O_APPEND failures resolved (618273, perhaps 614459)
4975 - lconf PATH fix (619770)
4976 - IA64 build fix (621450)
4977 - RPC buffer sizes scale with amount of memory
4979 2002-10-01 Phil Schwan <phil@clusterfs.com>
4982 - locks would be cancelled without throwing away data pages,
4983 resulting in inconsistent data (605627)
4984 - inode attributes were not always being refreshed (605627, 612449)
4985 - lconf now continues to cleanup after lctl reports an error
4986 - MDS now enforces user permissions (602707)
4987 - lprocfs cleanup fixed, but not yet enabled (614157)
4988 - fixed infinite server hang, should a client not respond to an AST
4989 - avoid going into recovery if user calls readlink() with a buffer
4990 that's too small (613941)
4991 - AST RPCs no longer require replies (614867) -- this may be changed
4992 - don't crash server if client sends an IOV that's too big (611336)
4993 - fixed lock conversion deadlock (611892)
4994 - fixed the following of symlinks (614622)
4995 * recovery: the server can remove locks from a client that dies, other
4996 clients can make progress
4997 * more extN patch fixes
4998 * compile-time configurable ptlrpc buffer allocations
5000 - collaborative read cache document
5001 - Lustre Lite Performance CDR document-in-progress
5003 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5006 - fix typo in patch-2.4.18
5008 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5011 - clear ptlrpc request each time in handle_incoming_request()
5012 - unlink of files now destroys the object on the OST
5014 2002-09-19 Peter Braam <braam@clusterfs.com>
5016 * add hard link support
5017 * change obdfile creation method
5018 * kernel patch changed
5020 2002-09-19 Peter Braam <braam@clusterfs.com>
5023 - stack overflow bug in extN fixed
5025 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
5027 * documentation updates
5028 - add man pages for config tools
5029 - update tests/README to describe testing with new config tools
5030 - finish metadata API descriptions
5031 * bug fixes and cleanups
5032 - statfs workaround for 16TB limit
5033 - LOV stripe allocation improved, can stripe on subset of OSTs
5034 - LOV file size/IO offset was wrong for files > 4GB in size
5035 - object EA data was being dropped, caused files to be unreadable
5036 - memory overflow with non-LOV OST caused memory corruption
5037 - fixed regression tests to work with new config tools, obdfilter
5038 - fixed bug when directory size became larger than 1 block
5039 - fixed bug (for single client case) when PWD was deleted
5040 - invalidate local directory pages when doing intent-based ops
5041 - avoid LDLM oops when lock callback contained bad data
5043 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5045 * documentation updates
5046 * bug fixes and cleanups
5047 - configuration tools
5050 - 64-bit compile warnings
5051 - 64-bit internal statfs data
5053 * test_brw on persistent OST devices
5055 * lprocfs (disabled)
5057 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5059 * documentation updates
5060 * bug fixes and cleanups
5061 * configuration tools
5063 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5066 * many small fixes to 0_5_4
5067 * io/network handling
5068 * thinkos in MDS operations
5070 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5073 * crucial basic fixes to 0.5.3
5074 * IOR, Iozone work over Elan
5077 2002-08-07 Phil Schwan <phil@clusterfs.com>
5078 * version 0_5_3, our first alpha
5079 * we use the new Portals iovs
5080 * documentation updates
5081 * bug fixes and cleanups
5082 * small changes in the DLM wire protocol
5084 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5085 * version 0_5_1 with some initial stability,
5086 * locking on MD and file I/O.
5087 * documentation updates
5088 * several bug fixes since 0.5.0
5089 * small changes in wire protocol
5091 2002-07-18 Phil Schwan <phil@clusterfs.com>
5093 * delivered as Lustre Light Alpha
5094 * fixed a crash after handling invalid MDS requests
5095 * fixed directory pages for architectures with non-4k pages sizes
5097 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5098 * release version v0_4_4
5099 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5101 * Fixes a number of interruption problems with OST operations.
5102 * Update documentation for portals header changes
5103 * Move all wire protocol structs/defines to lustre_idl.h
5104 * Fixes symlink length bug.
5105 * Add tcpdump to repository.
5107 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5108 * release version v0_4_3
5109 * Fixes statfs for inodes on extN.
5110 * Fixes bug in runtests which would delete /etc/hosts.
5111 * Use 64-bit object IDs wherever possible (not into VFS though)
5112 Remove ost_get_info, which is unused by lustre, and out of date.
5114 2002-07-03 Peter Braam <braam@clusterfs.com>
5115 * release version v0_4_2 Fixes a lookup error (type not passed)
5116 * move forward to head of Portals
5117 * move forward to latest Lustre kernel
5119 2002-06-25 Peter Braam <braam@clusterfs.com>
5120 * release version v0_4_1. Hopefully stable on single node use.