1 tbd Sun Microsystems, Inc.
4 2.6.16.54-0.2.5 (SLES 10),
5 2.6.18-92.1.6.el5 (RHEL 5),
6 2.6.22.14 vanilla (kernel.org).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.40.11-sun1
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
15 Severity : enhancement
17 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
22 Description: Using wrong pointer in osc_brw_prep_request
23 Details : Access to array[-1] can produce panic if kernel compiled with
24 CONFIG_PAGE_ALLOC enabled
28 Description: Client runs out of low memory
29 Details : Consider only lowmem when counting initial number of llap pages
33 Description: *optional* service tags registration
34 Details : if the "service tags" package is installed on a Lustre node
35 When the filesystem is mounted, a local-node service tag will
36 be created. See http://inventory.sun.com/ for more information
37 about the Service Tags asset management system.
41 Description: Kernel BUG tries to release flock
42 Details : Lustre does not destroy flock lock before last reference goes
43 away. So always drop flock locks when client is evicted and
44 perform unlock regardless of successfulness of speaking to MDS.
48 Description: add recount protection for osc callbacks, so avoid panic on shutdown
52 Description: sanity test 65a fails if stripecount of -1 is set
53 Details : handle -1 striping on filesystem in ll_dirstripe_verify
58 Description: ASSERTION(CheckWriteback(page,cmd)) failed
59 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
62 Severity : enhancement
64 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
68 Description: do not process already freed flock
69 Details : flock can possibly be freed by another thread before it reaches
70 to ldlm_flock_completion_ast.
74 Description: LBUG during stress test
75 Details : Need properly lock accesses the flock deadlock detection list.
79 Description: oops in page fault handler
80 Details : kernel page fault handler can return two special 'pages' in error case, don't
81 try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
85 Description: timeout with invalidate import.
86 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which should be
87 handled by ptlrpcd. This produce long age waiting and -ETIMEOUT
88 ptlrpc_invalidate_import and as result LASSERT.
90 Severity : enhancement
92 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
96 Description: enable MGS and MDT services start separately
97 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
98 a co-located MGS without starting the MGS, which is a complement
99 to 'nosvc' mount option.
103 Frequency : after recovery
104 Description: precreate to many object's after del orphan.
105 Details : del orphan st in oscc last_id == next_id and this triger growing
106 count of precreated objects. Set flag LOW to skip increase count
107 of precreated objects.
111 Frequency : rare, on clear nid stats
112 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
113 Details : when clean nid stats sometimes try destroy live entry,
114 and this produce panic in free.
118 Description: Stack overflow during MDS log replay
119 ease stack pressure by using a thread dealing llog_process.
123 Description: wait until IO finished before start new when do lock cancel.
124 Details : VM protocol want old IO finished before start new, in this case
125 need wait until PG_writeback is cleared until check dirty flag and
126 call writepages in lock cancel callback.
128 Severity : enhancement
130 Description: using special macro for print time and cleanup in includes.
134 Description: mds_mfd_close() ASSERTION(rc == 0)
135 Details : In mds_mfd_close(), we need protect inode's writecount change
136 within its orphan write semaphore to prevent possible races.
140 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
144 Frequency : rare, on shutdown ost
145 Description: don't hit live lock with umount ost.
146 Details : shrink_dcache_parent can be in long loop with destroy dentries,
147 use shrink_dcache_sb instead.
151 Description: don't panic with use echo client
152 Details : echo client pass NULL as client nid pointer and this produce null
157 Description: fix build on ppc32
158 Details : compile code with -m64 flag produce wrong object file for ppc32.
162 Description: add message levels for liblustreapi
166 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
167 Details : Move the 'good_osts' check before the 'total_bavail' check. This
168 will result in an -EAGAIN and in the exit call path we call
169 alloc_rr() which will with increasing aggressiveness attempt to
170 aquire precreated objects on the minimum number of required OSCs.
174 Description: Use old size assignment to avoid deadlock
175 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
176 the scheduling while holding a spinlock. We do not need locking
177 for size in ll_update_inode() because size is only updated from
178 the MDS for directories or files without objects, so there is no
179 other place to do the update, and concurrent access to such inodes
180 are protected by the inode lock.
184 Description: resolve "_IOWR redefined" build error on SLES10
188 Description: dump the memory debugging after all modules are unloaded to
189 suppress false negative in conf_sanity test 39
191 Severity : enhancement
193 Description: build kernel-ib packages for OFED 1.3 in our release cycle
198 Description: fix SLES kernel versioning
199 Details : the kernel version for our SLES 10 kernel did not include a "-"
200 before the "smp" at the end. while this was not a problem in
201 general, it did mean that software trying to use the kernel
202 version to try to detect a vendor specific kernel would fail.
203 this was most evident by the OFED build scripts.
207 Description: Don't update lov_desc members until making sure they are valid
208 Details : When updating lov_desc members via proc fs, need fix their
209 validities before doing the real update.
213 Description: don't put request into delay list while invalidate in flight.
214 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
215 import in flight. this produce timeout for invalidate and sometimes
216 can cause stale data.
220 Frequency : on ppc only
221 Description: not convert ost objects for directory because it's not exist.
222 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
226 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
228 Severity : enhancement
230 Description: Update to support 2.6.22.14 vanilla kernel.
234 Frequency : rare, on recovery
235 Description: read procfs can produce deadlock in some situation
236 Details : Holding lprocfs lock which send rpc can produce block for destroy
237 obd objects and this also block reconnect with -EALREADY. This isn't
238 fix all lprocfs bugs - but make it rare.
240 Severity : enhancement
242 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
245 Frequency : frequent on X2 node
247 Description: mdc_set_open_replay_data LBUG
248 Details : Set replay data for requests that are eligible for replay.
252 Description: lustre_mgs: operation 101 on unconnected MGS
253 Details : When MGC is disconnected from MGS long enough, MGS will evict the
254 MGC, and late on MGC cannot successfully connect to MGS and a lot
255 of the error messages complaining that MGS is not connected.
259 Frequency : on network error
260 Description: panic with double free request if network error
261 Details : mdc_finish_enqueue is finish request if any network error ocuring,
262 but it's true only for synchronus enqueue, for async enqueue
263 (via ptlrpcd) this incorrect and ptlrpcd want finish request
266 Severity : enhancement
268 Description: client-side metadata stat-ahead during readdir(directory readahead)
269 Details : perform client-side metadata stat-ahead when the client detects
270 readdir and sequential stat of dir entries therein
273 Frequency : on start mds
275 Description: Implement get_info(last_id) in obdfilter.
278 Frequency : occasional
280 Description: Correctly check stale fid, not start epoch if ost not support SOM
281 Details : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
282 overwrited with new generated fid, but mds can anwer with one of these
283 two fids and both is not stale. setattr incorectly start epoch and
284 assume will be called done_writeting, but without SOM done_writing
288 Frequency : rare, depends on device drivers and load
290 Description: MDS or OSS nodes crash due to stack overflow
291 Details : Code changes in 1.8.0 increased the stack usage of some functions.
292 In some cases, in conjunction with device drivers that use a lot
293 of stack the MDS (or possibly OSS) service threads could overflow
294 the stack. One change which was identified to consume additional
295 stack has been reworked to avoid the extra stack usage.
298 Frequency : occasional
300 Description: Do not fail import if osc_interpret_create gets -EAGAIN
301 Details : If osc_interpret_create got -EAGAIN it immediately exits and
302 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
303 oscc_has_objects and see OSC has no objests and call
304 oscc_internal_create to resend create request.
306 Severity : enhancement
308 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
310 Severity : enhancement
312 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
315 Frequency : very rare
317 Description: Fix replay if there is an un-replied request and open
318 Details : In some cases, older replay request will revert the
319 mcd->mcd_last_xid on MDS which is used to record the client's
322 Severity : enhancement
324 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
326 Severity : enhancement
328 Description: Add rhel5 support to HEAD.
330 Serverity : enhancement
332 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
337 Description: Don't allow skipping OSTs if index has been specified.
338 Details : Don't allow skipping OSTs if index has been specified, make locking
339 in internal create lots better.
343 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
344 Details : access to bitfield in structure is always rounded to long
345 and this produce problem with not atomic change any bit.
349 Description: Lustre make rpms failed.
350 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
351 builds Lustre rpms from tarball.
354 Frequency : rare on shutdown ost
356 Description: If llog cancel was not send before clean_exports phase, this can
357 produce deadlock in llog code.
358 Details : If llog thread has last reference to obd and call class_import_put
359 this produce deadlock because llog_cleanup_commit_master wait when
360 last llog_commit_thread exited, but this never success because was
361 called from llog_commit_thread.
365 Description: allow userland application know is lost one of stripes.
366 Details : fill lvb_blocks with error code on ost and return it to
367 application if error flag found.
371 Description: NULL lov_tgts causing MDS oops
372 Details : more safe checks for NULL lov_tgts for avoid oops.
374 Severity : enhancement
376 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
380 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
381 Details : If an OST is down the MDS will hang indefinitely in
382 obd_statfs_rqset() waiting for the statfs data. While for
383 MDS QOS usage of statfs, it should not stuck in waiting.
385 Severity : enhancement
387 Description: remote_acl support
388 Details : Support ACL-based permission check for remote user.
389 Support setfacl/getfacl for remote user with the utils
390 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
391 the system "{s,g}etfacl" utils.
393 Severity : enhancement
395 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
397 Severity : enhancement
399 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
403 Description: make mgs_setparam() handle fsname containing dash
404 Details : fsname containing a dash does not work with lctl conf_param
406 Severity : enhancement
408 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
410 Severity : enhancement
412 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
416 Frequency : rare, at shutdown
417 Description: access already free / zero obd_namespace.
418 Details : if client_disconnect_export was called without force flag set,
419 and exist connect request in flight, this can produce access to
420 NULL pointer (or already free pointer) when connect_interpret
421 store ocd flags in obd_namespace.
425 Frequency : only at startup
426 Description: not alloc memory with spinlock held.
427 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
428 if any spinlock held.
430 Severity : enhancement
432 Description: make lustre randomly fail allocating memory
433 Details : Make lustre randomly failed allocating memory for testing purpose.
435 Severity : enhancement
437 Description: lost problems with lov objid file
438 Details : Fixes some scability and access to not inited memory problems
439 in work with lov objdid file.
444 Description: lfs find does not continue on file error
445 Details : Continue other files processing when a file/dir is absent.
449 Description: Inconsistent usage of lustre_pack_reply()
450 Details : Standardize the usage of lustre_pack_reply() such that it
451 always generate a CERROR on failure.
454 Frequency : occasional
456 Description: MDS hang or stay in waiting lock
457 Details : If client receive lock with CBPENDING flag ldlm need send lock
458 cancel as separate rpc, to avoid situation when cancel request
459 can't processed due all i/o threads stay in wait lock.
463 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
467 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
471 Description: Update to SLES9 kernel-2.6.5-7.287.3.
475 Description: Update to RHEL4 latest kernel.
477 Severity : enhancement
479 Description: Build SLES10 patchless client fails
480 Details : The configure was broken by run ./configure with
481 --with-linux-obj=.... argument for patchless client. When the
482 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
483 can't search header adequately. Use absolute path such as
484 -I($LINUX)/include instead.
488 Description: interrupt oig_wait produce painc on resend.
489 Details : brw_redo_request can be used for resend requests from ptlrpcd and
490 private set, and this produce situation when rq_ptlrpcd_data not
491 copyed to new allocated request and triggered LBUG on assert
492 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
493 ptlrpcd set if request is changed and can be safety changed to use
498 Description: LASSERT_{REQ,REP}SWAB macros are buggy
499 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
500 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
501 these macros and replace them with their difinitions instead.
505 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
506 Details : Update which_patch & target file for SLES10 latest kernel.
509 Description: Remove client patches from SLES 10 kernel.
510 Details : This causes SLES 10 clients to behave as patchless clients
511 even on a Lustre-patched (server) kernel.
513 Severity : enhancement
515 Description: self-adjustable client's lru lists
516 Details : use adaptive algorithm for managing client cached locks lru
517 lists according to current server load, other client's work
518 pattern, memory activities, etc. Both, server and client
519 side namespaces provide number of proc tunables for controlling
522 Severity : enhancement
524 Description: light-weight GSS support
525 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
526 and reduce performance overhead.
528 Severity : enhancement
530 Description: Linux keyring support
531 Details : Support using service of Linux keyring for Lustre GSS internal
532 context refresh/cache mechanism.
536 Description: Fix errors in lfs documentation
537 Details : Fixes man pages
541 Description: don't use GFP_* in generic Lustre code.
542 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
546 Description: obdclass is limited by single OBD_ALLOC(idarray)
547 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
551 Description: warnings with build patchless client with vanila 2.6.19 and up
552 Details : change old ctl_table style and replace ctl_table/ctl_table_header
553 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
557 Frequency : Only for SLES
558 Description: sanity_quota fail test_1
559 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
560 file, which makes TSTID[2] inunique.
564 Frequency : Only for SLES9
565 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
566 Details : Update target/ChangeLog/which_patch .
570 Frequency : Only for test_10 in sanity-quota.sh
571 Description: a bug in quota test code
572 Details : checking a test flag in a wrong place
576 Frequency : Only for test_13 in sanity-quota.sh
577 Description: a bug in quota test script
578 Details : shouldn't check overfull allocated quota
582 Description: jbd statistics
583 Details : Port older jbd statistics patch for sles10
585 Severity : enhancement
587 Description: Add printing inode info into message about error in writepage.
592 Description: reply_lock_interpret crash due to race with it and lock cancel.
593 Details : Do not replay locks that are being cancelled. Do not reference
594 locks by their address during replay, just by their handle.
598 Frequency : When flocks are used.
599 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
600 Details : Flock locks might destroy just granted lock if it could be merged
601 with another existing flock, this is done in completion handler,
602 so teach ldlm_cli_enquque_fini that this is a valid case for
608 Description: Oops in read and write path when failing to allocate lock.
609 Details : Check if lock allocation failed and return error back.
613 Description: lstripe command fails for valid OST index
614 Details : The stripe offset is compared to lov->desc.ld_tgt_count
615 instead of lov->desc.ld_active_tgt_count.
619 Description: sanity.sh failed test 103
620 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
625 Description: df doesn't work properly if diskfs blocksize != 4K
626 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
630 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
631 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
633 Severity : enhancement
635 Description: Adaptive timeouts
636 Details : RPC timeouts adapt to changing server load and network
637 conditions to reduce resend attempts and improve recovery time.
641 Description: llapi_file_create() does not allow some changes
642 Details : add llapi_file_open() that allows specifying the mode and
643 open flags, and also returns an open file handle.
647 Description: merge and cleanup kernel patches.
648 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
652 Description: Add journal checksum support.(Kernel part)
653 Details : The journal checksum feature adds two new flags i.e
654 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
655 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
656 indicates that the commit block contains the checksum for
657 the blocks described by the descriptor blocks. Now commit
658 record can be sent to disk without waiting for descriptor
659 blocks to be written to disk. This behavior is controlled
660 using JBD2_FEATURE_ASYNC_COMMIT flag.
664 Description: OSS needs mutliple precreate threads
665 Details : Add ability to start more than one create thread per OSS.
669 Description: Sanity.sh test_65k failed: setstripe should have succeeded
670 Details : execute lfs setstripe on client
674 Description: mds_obd_create error creating tmp object
675 Details : When the user sets quota on root, llog will be affected and can't
676 create files and write files.
679 Frequency : Always on ia64 patchless client, and possibly others.
681 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
682 Details : This allows the patchless client to be loaded on architectures
687 Description: RedHat Update kernel for RHEL5
688 Details : Add the kernel config file for RHEL5.
692 Description: RedHat Update kernel for RHEL5
693 Details : Modify the kernel config file more closer RHEL5.
697 Description: Build failure against Centos5 (RHEL5)
698 Details : Define PAGE_SIZE when it isn't present.
702 Description: test_7 on sanity-quota.sh can't be used on separate machines
703 Details : add facet to handle it
707 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
708 Details : with NFS, the anon dentry's parent was set to itself in
709 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
710 corresponding dentry other than use rec->ur_name.
712 Severity : enhancement
714 Description: lfs setstripe enhancement
715 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
719 Description: enable data checksumming by default
720 Details : enable checksum by default, allow --disable-checksum
721 configure option and "-o nochecksum" mount option. Checksums
722 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
726 Description: lustre support for RHEL5
727 Details : Add support for RHEL5.
731 Description: Client eviction due to failover config
732 Details : after a connection loss, the lustre client should attempt to
733 reconnect to the last active server first before trying the
734 other potential connections.
738 Description: when mds and osts use different quota unit(32bit and 64bit),
739 quota will be released repeatly.
740 Details : void sending multiple quota reqs to mds, which will keep the
741 status between the reqs.
745 Description: osts not allocated evenly to files
746 Details : change the condition to increase offset_idx
749 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
751 Description: Data corruption for OSTs that are formatted larger than 2TB
753 Details : When generating the bio request for lustre file writes the
754 sector number would overflow a temporary variable before being
755 used for the IO. The data reads correctly from Lustre (which
756 will overflow in a similar manner) but other file data or
757 filesystem metadata may be corrupted in some cases.
761 Description: Tune the kernel for good SCSI performance.
762 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
763 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
768 Description: rewrite ext2-derived code in obdclass/uuid.c
769 Details : rewrite inherited code (uuid parsing code from ext2 utils)
770 from scratch preserving functionality.
774 Description: "lfs find -obd UUID" prints directories
775 Details : "lfs find -obd UUID" will return all directory names instead
776 of just file names. It is incorrect because the directories
777 do not reside on the OSTs.
781 Description: use i_size_read and i_size_write in 2.6 port
782 Details : replace inode->i_size access with i_size_read/write()
785 Frequency : when removing large files
787 Description: scheduling issue during removal of large Lustre files
788 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
789 It causes scheduling issues when removing large files (17TB in the
793 Frequency : only with liblustre clients on XT3
795 Description: evictions taking too long
796 Details : allow llrd to evict clients directly on OSTs
799 Frequency : only on ppc
801 Description: /proc/fs/lustre/devices broken on ppc
802 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
803 We should make sure the type of 'index' is loff_t and then cast
804 explicitly as needed below. Do not assign an explicitly cast
808 Frequency : only for rhel5
810 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
811 Details : Modify the target file & which_kernel.
813 Severity : enhancement
815 Description: omit set fsid for export NFS
816 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
817 as FS_REQUIRES_DEV for avoid problems with generate fsid.
821 Frequency : Always, for kernels after 2.6.16
822 Description: Fix warning idr_remove called for id=.. which is not allocated.
823 Details : Last kernels save old s_dev before kill super and not allow
824 to restore from callback - restore it before call kill_anon_super.
828 Description: buffer overruns could theoretically occur
829 Details : llapi_semantic_traverse() modifies the "path" argument by
830 appending values to the end of the origin string, and a
831 overrun may occur. Adding buffer overrun check in liblustreapi.
835 Description: Fix error on 'ls .' at the top of the Lustre mount.
836 Details : Don't revalidate dentry if it is a root dentry.
840 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
841 Details : Modify vm-tunables-rhel4.patch.
845 Description: Kernel config for 2.6.18-vanilla.
846 Details : Modify targets/2.6-vanilla.target.in.
847 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
848 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
849 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
850 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
854 Description: improve handling recoverable errors
855 Details : if request processig with error which can be recoverable on server
856 request should be resend, otherwise page released from cache and
861 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
862 Details : Modify target file & which_patch.
863 A flaw was found in the IA32 system call emulation provided
864 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
865 value could be stored in the %RAX register, which could trigger an
866 out-of-bounds system call table access. An untrusted local user
867 could exploit this flaw to run code in the kernel
868 (ie a root privilege escalation). (CVE-2007-4573).
872 Description: O_DIRECT bypasses client statistics.
873 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
874 read_ahead_stats were updated. copied the stats section from
875 osc_send_oap_rpc() into async_internal().
879 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
883 Description: change order of libsysio includes
884 Details : '#include sysio.h' should always come before '#include xtio.h'
888 Description: handle "serious error: objid * already exists" more gracefully
889 Details : If LAST_ID value on disk is smaller than the objects existing in
890 the O/0/d* directories, it indicates disk corruption and causes an
891 LBUG(). If the object is 0-length, then we should use the existing
892 object. This will help to avoid a full fsck in most cases.
894 Severity : enhancement
896 Description: adapt the lustre_config script to support the upgrade case
897 Details : Add "-u" option for lustre_config script to support upgrading 1.4
898 server targets to 1.6 in parallel.
902 Description: To avoid grant space > avaible space when the disk is almost
903 full. Without this patch you might see the error "grant XXXX >
904 available" or some LBUG about grant, when the disk is almost
906 Details : In filter_check_grant, for non_grant cache write, we should
907 check the left space by if (*left > ungranted + bytes), instead
908 of (*left > ungranted), because only we are sure the left space
909 is enough for another "bytes", then the ungrant space should be
910 increase. In client, we should update cl_avail_grant only there
911 is OBD_MD_FLGRANT in the reply.
915 Description: Update RHEL 4 kernel to fix local root privilege escalation.
916 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
917 described in CVE-2007-4573. This problem could allow untrusted
918 local users to gain root access.
921 Frequency : when using O_DIRECT and quotas
923 Description: Incorrect file ownership on O_DIRECT output files
924 Details : block usage reported by 'lfs quota' does not take into account
925 files that have been written with O_DIRECT.
930 Description: touch file failed when fs is not full
931 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
932 otherwise we can get ENOSP while fs is not full.
936 Description: parallel lock callbacks
937 Details : Instead of sending blocking and completion callbacks as separated
938 requests, adding them to a set and sending in parallel.
941 Frequency : only for Cray XT3
942 Bugzilla : 12829/13455
943 Description: Changing primary group doesn't change the group lustre assigns to
945 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
946 overridden with the primary group provided by the group upcall,
947 whereas the supplied fsgid can be trusted if it is in the list of
948 supplementary groups returned by the group upcall.
950 Severity : enhancement
952 Description: Allow masking D_WARNING, D_ERROR messages from console
953 Details : Console messages can now be disabled via lnet.printk.
957 Description: User code with malformed file open parameter crashes client node
958 Details : Before packing join_file req, all the related reference should be
959 checked carefully in case some malformed flags cause fake
960 join_file req on client.
964 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
965 Details : ldlm_completion_ast() assumes that a lock is granted when the req
966 mode is equal to the granted mode. However, it should also check
967 that LDLM_FL_CP_REQD is not set.
971 Description: Heavy nfs access might result in deadlocks
972 Details : After ELC code landed, it is now improper to enqueue any mds
973 locks under och_sem, because enqueue might want to decide to
974 cancel open locks for same inode we are holding och_sem for.
978 Description: Client eviction while running blogbench
979 Details : A lot of unlink operations with concurrent I/O can lead to a
980 deadlock causing evictions. To address the problem, the number of
981 oustanding OST_DESTROY requests is now throttled to
982 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
983 callbacks are processed in priority.
987 Description: enable ACLs on MDS by default
988 Details : ACLs must be enabled on MDS by default.
991 Frequency : PPC/PPC64 only
993 Description: conflicts between asm-ppc64/types.h and lustre_types.h
994 Details : fix duplicated definitions between asm-ppc64/types.h and
995 lustre_types.h on PPC.
998 Frequency : PPC/PPC64 only
1000 Description: asm-ppc/segment.h does not exist
1001 Details : fix compile issue on PPC.
1005 Description: better handle error messages in extents code
1008 Frequency : RHEL4 only
1010 Description: mkfs is very slow on IA64/RHEL4
1011 Details : A performance regression has been discovered in the MPT Fusion
1012 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1013 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1014 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1016 Severity : enhancement
1018 Description: SNMP support enhancement
1019 Details : Adding total number of sampled request for an MDS node in snmp
1022 Severity : enhancement
1024 Description: Optimize ldlm waiting list processing for PR extent locks
1025 Details : When processing waiting list for read extent lock and meeting read
1026 lock that is same or wider to it that is not contended, skip
1027 processing rest of the list and immediatelly return current
1028 status of conflictness, since we are guaranteed there are no
1029 conflicting locks in the rest of the list.
1033 Description: Time out and refuse to reconnect
1034 Details : When the failover node is the primary node, it is possible
1035 to have two identical connections in imp_conn_list. We must
1036 compare not conn's pointers but NIDs, otherwise we can defeat
1037 connection throttling.
1041 Description: port llog fixes from b1_6 into HEAD
1042 Details : Port llog reference couting and some llog cleanups from b1_6
1043 (bug 10800) into HEAD, for protect from panic and access to already
1044 free llog structures.
1048 Description: Detect stride IO mode in read-ahead
1049 Details : When a client does stride read, read-ahead should detect that and
1050 read-ahead pages according to the detected stride pattern.
1054 Description: data checksumming impacts single node performance
1055 Details : add support for several checksum algorithm. Currently, only CRC32
1056 and Adler-32 are supported. The checksum type can be changed on
1057 the fly via /proc/fs/lustre/osc/*/checksum_type.
1061 Description: use adler32 for page checksums
1062 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1067 Description: build for x2 fails
1068 Details : fix compile issue on Cray systems.
1072 Description: Properly match for duplicate locks
1073 Details : Due to different lock order from skiplists code, we need to
1074 traverse entire list for now
1077 Frequency : only on PPC/SLES10
1079 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1080 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1081 be defined. Add a hack in mkfs_lustre.c to work around this compile
1086 Description: LASSERT on MDS when client holding flock lock dies
1087 Details : ldlm pool logic depends on number of granted locks equal to
1088 number of released locks which is not true for flock locks, so
1089 just exclude such locks from consideration.
1093 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1094 Details : Use fsfilt_sync() for both the whole filesystem sync and
1095 individual file sync to eliminate dangerous inode locking
1096 with I_LOCK that can lead to a deadlock.
1100 Description: performance in 1.6.3
1101 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1102 sure that 1MB requests are not fragmented by the block layer.
1104 Severity : enhancement
1106 Description: organize the server-side client stats on per-nid basis
1107 Details : Change the structure of stats under obdfilter and mds to
1115 The "uuid"s file would list the uuids of _active_ exports.
1116 And the clear entry is to clear all stats and stale nids.
1118 Severity : enhancement
1120 Description: eliminate client locks in face of contention
1121 Details : file contention detection and lockless i/o implementation
1122 for contended files.
1126 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1129 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1131 Description: MDS may lose file striping (and hence file data) in some cases
1132 Details : If there are additional extended attributes stored on the MDS,
1133 in particular ACLs, SELinux, or user attributes (if user_xattr
1134 is specified for the client mount options) then there is a risk
1135 of attribute loss. Additionally, the Lustre file striping
1136 needs to be larger than default (e.g. striped over all OSTs),
1137 and an additional attribute must be stored initially in the
1138 inode and then increase in size enough to be moved to the
1139 external attribute block (e.g. ACL growing in size) for file
1144 Description: skiplist implementation simplification
1145 Details : skiplists are used to group compatible locks on granted list
1146 that was implemented as tracking first and last lock of each lock group
1147 the patch changes that to using doubly linked lists
1151 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1152 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1153 because LASSERT on both the data supplied by a client, and the data
1154 on disk is dangerous and incorrect.
1156 Severity : enhancement
1158 Description: Slow truncate/writes to huge files at high offsets.
1159 Details : Directly associate cached pages to lock that protect those pages,
1160 this allows us to quickly find what pages to write and remove
1161 once lock callback is received.
1165 Description: more ldlm soft lockups
1166 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1167 starve other threads from the resource lock for a long time in
1168 case of long waiting queue, so change the debug level from
1169 D_OTHER to the less frequently used D_INFO.
1171 Severity : enhancement
1173 Description: add -gid, -group, -uid, -user options to lfs find
1177 Description: Hung threads in invalidate_inode_pages2_range
1178 Details : The direct IO path doesn't call check_rpcs to submit a new RPC once
1179 one is completed. As a result, some RPCs are stuck in the queue
1184 Description: filter threads hungs on waiting journal commit
1185 Details : Cleanup filter group llog code, then only filter group llog will
1186 be only created in the MDS/OST syncing process.
1190 Description: Procfs and llog threads access destoryed import sometimes.
1191 Details : Sync the import destoryed process with procfs and llog threads by
1192 the import refcount and semaphore.
1194 Severity : enhancement
1196 Description: openlock cache of b1_6 port to HEAD
1201 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1202 Details : If insertion of an extent fails, then discard the inode
1203 preallocation and free data blocks else it can lead to duplicate
1208 Description: don't always update ctime in ext3_xattr_set_handle()
1209 Details : Current xattr code updates the inode ctime in ext3_xattr_set_handle.
1210 In some cases the ctime should not be updated, for example for
1211 2.0->1.8 compatibility it is necessary to delete an xattr and it
1212 should not update the ctime.
1216 Bugzilla : 15713/16362
1217 Description: Assertion in iopen_connect_dentry in 1.6.3
1218 Details : looking up an inode via iopen with the wrong generation number can
1219 populate the dcache with a disconneced dentry while the inode
1220 number is in the process of being reallocated. This causes an
1221 assertion failure in iopen since the inode's dentry list contains
1222 both a connected and disconnected dentry.
1224 --------------------------------------------------------------------------------
1226 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
1228 * Support for kernels:
1229 2.6.5-7.283 (SLES 9),
1230 2.6.9-55.EL (RHEL 4),
1231 2.6.16.46-0.14 (SLES 10),
1232 2.6.18.8 vanilla (kernel.org)
1233 * Client support for unpatched kernels:
1234 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1235 2.6.16 - 2.6.22 vanilla (kernel.org)
1236 * Due to recently discovered recovery problems, we do not recommend
1237 using patchless RHEL 4 clients with this or any earlier release.
1238 * Recommended e2fsprogs version: 1.39.cfs8
1239 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1240 * Starting with this release, the ldiskfs backing filesystem required
1241 by Lustre is now in its own package, lustre-ldiskfs. This package
1242 should be installed. It is versioned separately from Lustre and
1243 may be released separately in future.
1247 Description: block reactivating mgc import until all deactivates complete
1248 Details : Fix race when failing back MDT/MGS to itself (testing)
1251 Frequency : at statup only
1253 Description: mds_lov_synchronize race leads to various problems
1254 Details : simultaneous MDT->OST connections at startup can cause the
1255 sync to abort, leaving the OSC in a bad state.
1257 Severity : enhancement
1259 Description: add optional extra BUILD_VERSION info
1260 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1261 to override the lustre version.
1264 Frequency : 2.6.18 servers only
1266 Description: ll_kern_mount() doesn't release the module reference
1267 Details : The ldiskfs module reference count never drops down to 0
1268 because ll_kern_mount() doesn't release the module reference.
1273 Description: server LBUG when using old ost_num_threads parameter
1274 Details : Accept the old ost_num_threads parameter but warn that it
1275 is deprecated, and fix an off-by-one error that caused an LBUG.
1280 Description: Transient SCSI error results in persistent IO issue
1281 Details : iobuf->dr_error is not reinitialized to 0 between two
1285 Frequency : sometimes when underlying device returns I/O errors
1287 Description: OSTs not going read-only during write failures
1288 Details : OSTs are not remounted read-only when the journal commit threads
1289 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1290 instead of the ext3 wrappers.
1294 Description: poor connect scaling with increasing client count
1295 Details : Don't run filter_grant_sanity_check for more than 100 exports
1296 to improve scaling for large numbers of clients.
1299 Frequency : SLES10 only
1301 Description: sanity-quota.sh quotacheck failed: rc = -22
1302 Details : Quotas cannot be enabled on SLES10.
1305 Frequency : liblustre clients only
1307 Description: getdirentries does not give error when run on compute nodes
1308 Details : getdirentries does not fail when the size specified as an argument
1309 is too small to contain at least one entry
1311 Severity : enhancement
1313 Description: Add LNET router traceability for debug purposes
1314 Details : If a checksum failure occurs with a router as part of the
1315 IO path, the NID of the last router that forwarded the bulk data
1316 is printed so it can be identified.
1321 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1322 Details : Due to a race condition, liblustre clients were occasionally
1323 evicted incorrectly.
1325 Severity : enhancement
1327 Description: lfs setstripe use optional parameters instead of postional
1330 Severity : enhancement
1332 Description: Nanosecond timestamp support for ldiskfs
1333 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1334 resolution timestamps. There is not yet support for this at
1335 the Lustre filesystem level.
1338 Frequency : during server recovery
1340 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1341 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1345 Frequency : nfs export on patchless client
1347 Description: connectathon hang when test nfs export over patchless client
1348 Details : Disconnected dentry cannot be found with lookup, so we do not need
1349 to unhash it or make it invalid
1352 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1353 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1354 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1359 Description: open req refcounting wrong on reconnect
1360 Details : If reconnect happened between getting open reply from server and
1361 call to mdc_set_replay_data in ll_file_open, we will schedule
1362 replay for unreferenced request that we are about to free.
1363 Subsequent close will crash in variety of ways.
1364 Check that request is still eligible for replay in
1365 mdc_set_replay_data().
1370 Description: disable writes to filesystem when reading health_check file
1371 Details : the default for reading the health_check proc file has changed
1372 to NOT do a journal transaction and write to disk, because this
1373 can cause reads of the /proc file to hang and block HA state
1374 checking on a healthy but otherwise heavily loaded system. It
1375 is possible to return to the previous behaviour during configure
1376 with --enable-health-write.
1378 Severity : enhancement
1380 Description: 64-bit inode version
1381 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1382 to the inode. This will be required for version-based recovery.
1387 Description: MDS fails to start if a duplicate client export is detected
1388 Details : in some rare cases it was possible for a client to connect to
1389 an MDS multiple times. Upon recovery the MDS would detect this
1390 and fail during startup. Handle this more gracefully.
1392 Severity : enhancement
1394 Description: Add -o localflock option to simulate old noflock
1396 Details : This will achieve local-only flock/fcntl locks
1402 Description: log_commit_thread vs filter_destroy race leads to crash
1403 Details : Take import reference before releasing llog record semaphore
1408 Description: Wrong request locking in request set processing
1409 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1410 imp_delayed_list, in this place should be used imp_lock.
1413 Frequency : when reconnection
1415 Description: Grant Leak when osc reconnect to OST
1416 Details : When osc reconnect ost, OST(filter) should check whether it
1417 should grant more space to client by comparing fed_grant and
1418 cl_avail_grant, and return the granted space to client instead
1419 of "new granted" space, because client will call osc_init_grant
1420 to update the client grant space info.
1423 Frequency : when client reconnect to OST
1425 Description: Grant Leak when osc do resend and replay bulk write
1426 Details : When osc reconnect to OST, OST(filter)should clear grant info of
1427 bulk write request, because the grant info while be sync between
1428 OSC and OST when reconnect, and we should ignore the grant info
1429 these of resend/replay write req.
1434 Description: Grant space more than avaiable left space sometimes.
1435 Details : When then OST is about to be full, if two bulk writing from
1436 different clients came to OST. Accord the avaliable space of the
1437 OST, the first req should be permitted, and the second one
1438 should be denied by ENOSPC. But if the seconde arrived before
1439 the first one is commited. The OST might wrongly permit second
1440 writing, which will cause grant space > avaiable space.
1443 Frequency : when client is evicted
1445 Description: Grant might be wrongly erased when osc is evicted by OST
1446 Details : when the import is evicted by server, it will fork another
1447 thread ptlrpc_invalidate_import_thread to invalidate the
1448 import, where the grant will be set to 0. While the original
1449 thread will update the grant it got when connecting. So if
1450 the former happened latter, the grant will be wrongly errased
1451 because of this race.
1456 Description: Checking Stale with correct fid
1457 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1458 to check whether it is stale, because sometimes, we want the
1459 enqueue happened anyway, and op_data.fid2 will not be initialized.
1461 Severity : enhancement
1463 Description: update patchless client
1464 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1467 Frequency : only with 2.4 kernel
1469 Description: random memory corruption
1470 Details : size of struct ll_inode_info is to big for union inode.u and this
1471 can be cause of random memory corruption.
1476 Description: Memory leak in recovery
1477 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1478 It should also check obd_fail before fsfilt_start, otherwise if
1479 fsfilt_start return -EROFS,(failover mds during mds recovery).
1480 then the req will return with repmsg->transno = 0 and rc = EROFS.
1481 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1482 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1483 be freed no matter whether fsfilt_commit success or not.
1486 Frequency : only with huge count clients
1488 Description: Prevents from taking the superblock lock in llap_from_page for
1490 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1491 for llap_from_page call in ll_removepage prevents from taking the
1492 superblock lock for a soon died page.
1497 Description: Not check open intent error before release open handle
1498 Details : in some rare cases, the open intent error is not checked before
1499 release open handle, which may cause
1500 ASSERTION(open_req->rq_transno != 0), because it tries to release
1501 the failed open handle.
1506 Description: Set cat log bitmap only after create log success.
1507 Details : in some rare cases, the cat log bitmap is set too early. and it
1508 should be set only after create log success.
1512 Description: Accessing a block bevice can re-enable I/O when Lustre is
1513 tearing down a device.
1514 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1519 Description: service threads may hog cpus when there are a lot of requests
1520 Details : Insert cond_resched to give other threads a chance to use some CPU
1525 Description: the cat log was not initialized in recovery
1526 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1527 unlink log on mds will not be initialized until mds post
1528 recovery. And also in mds post recovery, the unlink log will
1529 initialization will be done asynchronausly, so there will be race
1530 between add unlink log and unlink log initialization.
1534 Description: brw_stats were being printed incorrectly
1535 Details : brw_stats were being printed as log2 but all of them were not
1536 recorded as log2. Also remove some code duplication arising from
1537 filter_tally_{read,write}.
1541 Frequency : rare, only in recovery.
1542 Description: ASSERTION(req->rq_type != LI_POISON) failed
1543 Details : imp_lock should be held while iterating over imp_sending_list for
1544 prevent destroy request after get timeout in ptlrpc_queue_wait.
1548 Description: replay-single.sh test 52 fails
1549 Details : A lock's skiplist need to be cleanup when it being unlinked
1550 from its resource list.
1554 Description: Short directio read returns full requested size rather than
1556 Details : Direct I/O operations should return actual amount of bytes
1557 transferred rather than requested size.
1559 Severity : enhancement
1561 Description: metadata RPC reduction (e.g. for rm performance)
1562 Details : decrease the amount of synchronous RPC between clients and servers
1563 by canceling conflicing lock before the operation on the client
1564 and packing thier handles into the main operation RPC to server.
1566 Severity : enhancement
1568 Description: Async OSC create to avoid the blocking unnecessarily.
1569 Details : If a OST has no remain object, system will block on the creating
1570 when need to create a new object on this OST. Now, ways use
1571 pre-created objects when available, instead of blocking on an
1572 empty osc while others are not empty. If we must block, we block
1573 for the shortest possible period of time.
1575 Severity : enhancement
1577 Description: refine locking for avoid write wrong info into lov_objid file
1578 Details : fix possible races with add new target and write/update data in
1581 --------------------------------------------------------------------------------
1583 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1588 Frequency : on some architectures
1590 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1591 Details : Uninitialized flags sometimes cause configuration commands to
1595 Frequency : patchless clients only
1597 Description: missing __iget() symbol export
1598 Details : The __iget() symbol export is missing. To avoid the need for
1599 this on patchless clients the deathrow inode reaper is turned
1600 off, and we depend on the VM to clean up old inodes. This
1601 dependency was during via the fix for bug 12181.
1603 --------------------------------------------------------------------------------
1605 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1607 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1608 INTEROPERATE with older versions automatically. In many cases a
1609 special upgrade step is needed. Please read the
1610 user documentation before upgrading any part of a live system.
1611 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1612 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1613 * WARNING: Lustre configuration and startup changes are required with
1614 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1620 Severity : enhancement
1622 Description: MountConf
1623 Details : Lustre configuration is now managed via mkfs and mount
1624 commands instead of lmc and lconf. New obd types (MGS, MGC)
1625 are added for dynamic configuration management. See
1626 https://mail.clusterfs.com/wikis/lustre/MountConf for
1629 Severity : enhancement
1631 Description: dynamic OST addition
1632 Details : OSTs can now be added to a live filesystem
1634 Severity : enhancement
1636 Description: startup order invariance
1637 Details : MDTs and OSTs can be started in any order. Clients only
1638 require the MDT to complete startup.
1640 Severity : enhancement
1642 Description: parallel, asynchronous orphan cleanup
1643 Details : orphan cleanup is now performed in separate threads for each
1644 OST, allowing parallel non-blocking operation.
1646 Severity : enhancement
1648 Description: optimized stripe assignment
1649 Details : stripe assignments are now made based on ost space available,
1650 ost previous usage, and OSS previous usage, in order to try
1651 to optimize storage space and networking resources.
1653 Severity : enhancement
1655 Description: Permanently set tunables
1656 Details : All writable /proc/fs/lustre tunables can now be permanently
1657 set on a per-server basis, at mkfs time or on a live system.
1659 Severity : enhancement
1661 Description: Lustre message v2
1662 Details : Add lustre message format v2.
1664 Severity : enhancement
1666 Description: client OST exclusion list
1667 Details : Clients can be started with a list of OSTs that should be
1668 declared "inactive" for known non-responsive OSTs.
1672 Description: SPEC SFS validation failure on NFS v2 over lustre.
1673 Details : Changes the blocksize for regular files to be 2x RPC size,
1674 and not depend on stripe size.
1676 Severity : enhancement
1678 Description: Multiple MD RPCs in flight.
1679 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1680 To avoid overly-overloading MDS, introduce a limit on number of
1681 MDS RPCs in flight for a single client and add /proc controls
1682 to adjust this limit.
1684 Severity : enhancement
1686 Description: client read/write statistics
1687 Details : Add client read/write call usage stats for performance
1688 analysis of user processes.
1689 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1690 file access. extents_stats shows chunk size distribution.
1691 extents_stats_per_process show chunk size distribution per
1694 Severity : enhancement
1696 Description: mds statistics
1697 Details : Add detailed mds operations statistics in
1698 /proc/fs/lustre/mds/*/stats.
1702 Description: Failure of copying files with lustre special EAs.
1703 Details : Client side always return success for setxattr call for lustre
1704 special xattr (currently only "trusted.lov").
1709 Description: Refcount LNET uuids
1710 Details : The global LNET uuid list grew linearly with every startup;
1711 refcount repeated list entries instead of always adding to
1714 Severity : enhancement
1716 Description: Dynamic service threads
1717 Details : Within a small range, start extra service threads
1718 automatically when the request queue builds up.
1721 Frequency : mixed-endian client/server environments
1723 Description: mixed-endian crashes
1724 Details : The new msg_v2 system had some failures in mixed-endian
1727 Severity : enhancement
1729 Description: Easy OST removal
1730 Details : OSTs can be permanently deactivated with e.g. 'lctl
1731 conf_param lustre-OST0001.osc.active=0'
1733 Severity : enhancement
1735 Description: MGS proc entries
1736 Details : Added basic proc entries for the MGS showing what filesystems
1739 Severity : enhancement
1741 Description: provide MGS failover
1742 Details : Added config lock reacquisition after MGS server failover.
1744 Severity : enhancement
1746 Description: add Linux 2.4 support
1747 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1751 Description: a large application tries to do I/O to the same resource and dies
1752 in the middle of it.
1753 Details : Check the req->rq_arrival time after the call to
1754 ost_brw_lock_get(), but before we do anything about
1755 processing it & sending the BULK transfer request. This
1756 should help move old stale pending locks off the queue as
1757 quickly as obd_timeout.
1760 Frequency : when an incorrect nid is specified during startup
1762 Description: ptlrpc connect to non-existant node causes kernel crash
1763 Details : LNET can't be re-entered from an event callback, which
1764 happened when we expire a message after the export has been
1765 cleaned up. Instead, hand the zombie cleanup off to another
1768 Severity : enhancement
1770 Description: plain/inodebits lock performance improvement
1771 Details : Grouping plain/inodebits in granted list by their request modes
1772 and bits policy, thus improving the performance of search through
1776 Frequency : only if OST filesystem is corrupted
1778 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1779 Details : for a short time RPCs with bulk IO are in the replay list,
1780 but replay of bulk IOs is unimplemented. If the OST filesystem
1781 is corrupted due to disk cache incoherency and then replay is
1782 started it is possible to trip an assertion. Avoid putting
1783 committed RPCs into the replay list at all to avoid this issue.
1786 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1788 Description: System hang on startup
1789 Details : This bug allowed the liblustre (e.g. catamount) client to
1790 return to the app before handling all startup RPCs. This
1791 could leave the node unresponsive to lustre network traffic
1792 and manifested as a server ptllnd timeout.
1794 Severity : enhancement
1796 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1797 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1798 boolean to control whether to print peer debug info when a
1799 client's RPC times out.
1802 Frequency : only for kernels with patches from Lustre below 1.4.3
1804 Description: Remove old rdonly API
1805 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1808 Frequency : only for devices with external journals
1810 Description: Set external device read-only also
1811 Details : During a commanded failover stop, we set the disk device
1812 read-only while the server shuts down. We now also set any
1813 external journal device read-only at the same time.
1816 Frequency : when upgrading from 1.4 while trying to change parameters
1818 Description: The wrong (new) MDC name was used when setting parameters for
1819 upgraded MDT's. Also allows changing of OSC (and MDC)
1820 parameters if --writeconf is specified at tunefs upgrade time.
1823 Frequency : when setting specific ost indicies
1825 Description: QOS code breaks on skipped indicies
1826 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1827 created with --index need not be sequential.
1829 Severity : enhancement
1831 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1832 Details : The uninit_groups feature works in conjunction with the kernel
1833 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1834 up the pass1 processing of e2fsck. This is a read-only feature
1835 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1836 mount filesystems that have had this feature enabled.
1838 Severity : enhancement
1840 Description: Improve multi-block allocation algorithm to avoid fragmentation
1841 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1842 allocation locality and avoid filesystem fragmentation.
1844 ------------------------------------------------------------------------------
1846 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1848 * Support for kernels:
1849 2.4.21-47.0.1.EL (RHEL 3)
1850 2.6.5-7.283 (SLES 9)
1851 2.6.9-42.0.10.EL (RHEL 4)
1852 2.6.12.6 vanilla (kernel.org)
1853 2.6.16.27-0.9 (SLES 10)
1854 * Recommended e2fsprogs version: 1.39.cfs5
1856 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1860 Frequency : occasional, depends on client load and configuration
1861 Bugzilla : 12181, 12203
1862 Description: data loss for recently-modified files
1864 Details : In some cases it is possible that recently written or created
1865 files may not be written to disk in a timely manner (this should
1866 normally be within 30s unless client IO load is very high).
1867 The problem appears as zero-length files or files that are a
1868 multiple of 1MB in size after a client crash or client eviction
1869 that are missing data at the end of the file.
1871 This problem is more likely to be hit on clients where files are
1872 repeatedly created and unlinked in the same directory, clients
1873 have a large amount of RAM, have many CPUs, the filesystem has
1874 many OSTs, the clients are rebooted frequently, and/or the files
1875 are not accessed by other nodes after being written.
1877 The presence of the problem can be detected by looking at
1878 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1879 smaller than the second (nr_unused) then dirty files will not
1880 be flushed automatically to disk. "sync; sleep 10" should be
1881 run several times on the node before unmounting it to update
1882 Lustre (this is also safe to run on nodes without this problem).
1884 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1885 that can cause this same problem, so customers using that kernel
1886 also need to update the kernel in addition to Lustre. In order
1887 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1889 It is normal that files written just before a client crash (less
1890 than 30s) may not yet have been flushed to disk, even for local
1894 Frequency : frequent on thin XT3 nodes
1896 Description: UUID collision on thin XT3 Linux nodes
1897 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1898 randomly, since we relied on an insufficiently-seeded PRNG.
1903 Description: OSS hangs after "All ost request buffers busy"
1904 Details : A deadlock between quota and journal operations caused OSS
1905 hangs after printing "All ost request buffers busy."
1908 Frequency : always on liblustre builds
1910 Description: Cleanup compiler warnings on liblustre
1913 Frequency : always on liblustre builds on XT3
1915 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1919 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1920 > 32000 subdirectories
1921 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1922 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1923 subdirectory count crosses 32000. This will aid e2fsck to
1924 correctly handle more than 32000 subdirectories.
1927 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1929 Description: System hang on startup
1930 Details : This bug allowed the liblustre (e.g. catamount) client to
1931 return to the app before handling all startup RPCs. This
1932 could leave the node unresponsive to lustre network traffic
1933 and manifested as a server ptllnd timeout.
1935 Severity : enhancement
1937 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1938 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1939 boolean to control whether to print peer debug info when a
1940 client's RPC times out.
1945 Description: make O_SYNC working on 2.6 kernels
1946 Details : 2.6 kernels use different method for mark pages for write,
1947 so need add a code to lustre for O_SYNC work.
1952 Description: Failure to close file and release space on NFS
1953 Details : Put inode details into lock acquired in ll_intent_file_open.
1954 Use mdc_intent_lock in ll_intent_open to properly
1955 detect all kind of errors unhandled by mdc_enqueue.
1960 Description: proc file read during shutdown sometimes raced obd removal,
1962 Details : Add lock to prevent obd access after proc file removal.
1965 Frequency : Only for files larger than 4GB on 32-bit clients.
1967 Description: improperly doing page alignment of locks
1968 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1969 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1974 Description: RPCs being resent when they shouldn't be
1975 Details : Some RPCs that should not be resent are being resent. This
1976 can cause inconsistencies in the RPC state machine. Do not
1977 resend such requests.
1980 Frequency : rare, only with NFS export
1982 Description: Crash on NFS re-export node
1983 Details : under very unusual load conditions an assertion is hit in
1984 ll_intent_file_open()
1987 Frequency : only if OST filesystem is corrupted
1989 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1990 Details : for a short time RPCs with bulk IO are in the replay list,
1991 but replay of bulk IOs is unimplemented. If the OST filesystem
1992 is corrupted due to disk cache incoherency and then replay is
1993 started it is possible to trip an assertion. Avoid putting
1994 committed RPCs into the replay list at all to avoid this issue.
1999 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2000 Details : Large single O_DIRECT read and write calls can fail to allocate
2001 a sufficiently large buffer to process the request. In case of
2002 allocation failure the allocation is retried with a smaller
2003 buffer and broken into smaller requests.
2005 Severity : enhancement
2007 Description: Add -o localflock option to simulate old noflock behaviour.
2008 Details : This will achieve local-only flock/fcntl locks coherentness.
2013 Description: versioning check is incomplete
2014 Details : Checking the version difference of client vs. server, report
2015 error if the gap is too big.
2020 Description: add support PG_writeback bit
2021 Details : add support for PG_writeback bit for Lustre, for more carefull
2022 work with page cache in 2.6 kernel. This also fix some deadlocks
2023 and remove hack for work O_SYNC with 2.6 kernel.
2025 Severity : enhancement
2027 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2028 Details : The uninit_groups feature works in conjunction with the kernel
2029 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2030 up the pass1 processing of e2fsck. This is a read-only feature
2031 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2032 mount filesystems that have had this feature enabled.
2034 Severity : enhancement
2036 Description: Improve multi-block allocation algorithm to avoid fragmentation
2037 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2038 allocation locality and avoid filesystem fragmentation.
2040 ------------------------------------------------------------------------------
2042 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2044 * Support for kernels:
2045 2.6.9-42.0.3.EL (RHEL 4)
2046 2.6.5-7.276 (SLES 9)
2047 2.4.21-47.0.1.EL (RHEL 3)
2048 2.6.12.6 vanilla (kernel.org)
2049 2.6.16.21-0.8 (SLES10)
2050 * Recommended e2fsprogs version: 1.39.cfs2-0
2052 * The backwards-compatible /proc/sys/portals symlink has been removed
2053 in this release. Before upgrading, please ensure that you change
2054 any configuration scripts or /etc/sysctl.conf files that access
2055 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2056 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2057 in advance of the upgrade on any system running Lustre 1.4.6 or
2058 newer, since /proc/sys/lnet was added in that version.
2059 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2063 Frequency : only when quota is used
2065 Description: avoid scanning export list for quota master
2066 Details : Change the algorithms to avoid scanning export list in order
2067 to improve the efficiency.
2070 Frequency : MDS failover only, very rarely
2072 Description: "went back in time" messages on mds failover
2073 Details : The greatest transno may be lost when the current operation
2074 finishes with an error (transno==0) and the client's last_rcvd
2075 record is over-written. Save the greatest transno in the
2076 mds_last_transno for this case.
2079 Frequency : always for specific kernels and striping counts
2081 Description: client may get "Matching packet too big" without ACL support
2082 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2083 when trying to access files in certain configurations. The
2084 clients should in fact be denied when mounting because they do
2085 not understand ACLs.
2088 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2090 Description: many clients connecting with IO in progress causes connect timeouts
2091 Details : Avoid synchronous journal commits to avoid delays caused by many
2092 clients connecting/disconnecting when bulk IO is in progress.
2093 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2094 OST_IO_PORTAL to avoid delays behind potentially many pending
2098 Frequency : occasionally with multiple writers to a single file
2100 Description: shared writes to file may result in wrong size reported by stat()
2101 Details : Allow growing of kms when extent lock is cancelled
2104 Frequency : always with random mmap IO to multi-striped file
2106 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2107 Details : Only if the hole is at the end of OST object so that kms is too
2108 small. Fix is to increase kms accordingly in ll_nopage.
2111 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2113 Description: writes to a missing object would leak memory on the OST
2114 Details : If there is an inconsistency between the MDS and OST filesystems,
2115 such that the MDS references an object that doesn't exist, writes
2116 to that object will leak memory due to incorrect cleanup in the
2117 error handling path, eventually running out of memory on the OST.
2122 Description: Creating too long symlink causes lustre errors
2123 Details : Check symlink and name lengths before sending requests to MDS.
2126 Frequency : only if flock is enabled (not on by default)
2128 Description: posix locks not released on fd closure on 2.6.9+
2129 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2130 kernels, this caused such locks not to be released on fd close and
2131 then assertions on fs unmount about still used locks.
2134 Frequency : MDS failover only, very rarely
2136 Description: clients may get ASSERTION(granted_lock != NULL)
2137 Details : When request was taking a long time, and a client was resending
2138 a getattr by name lock request. The were multiple lock requests
2139 with the same client lock handle and
2140 mds_getattr_name->fixup_handle_for_resent_request found one of the
2141 lock handles but later failed with ASSERTION(granted_lock != NULL).
2146 Description: handle->h_buffer_credits > 0, assertion failure
2147 Details : h_buffer_credits is zero after truncate, causing assertion
2148 failure. This patch extends the transaction or creates a new
2152 Frequency : NFS re-export or patchless client
2153 Bugzilla : 11179, 10796
2154 Description: Crash on NFS re-export node (__d_move)
2155 Details : We do not want to hash the dentry if we don't have a lock.
2156 But if this dentry is later used in d_move, we'd hit uninitialised
2157 list head d_hash, so we just do this to init d_hash field but
2158 leave dentry unhashed.
2161 Frequency : NFS re-export or patchless client
2163 Description: NFS exports has problem with symbolic link
2164 Details : lustre client didn't properly install dentry when re-exported
2165 to NFS or running patchless client.
2168 Frequency : NFS re-export or patchless client
2170 Description: Various nfs/patchless fixes.
2171 Details : fixes reuse disconected alias for lookup process - this fixes
2172 warning "find_exported_dentry: npd != pd",
2173 fix permission error with open files at nfs.
2174 fix apply umask when do revalidate.
2177 Frequency : occasional
2179 Description: Crash on NFS re-export node
2180 Details : calling clear_page() on the wrong pointer triggered oops in
2181 generic_mapping_read().
2184 Frequency : rarely, using O_DIRECT IO
2186 Description: unaligned directio crashes client with LASSERT
2187 Details : check for unaligned buffers before trying any requests.
2190 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2192 Description: stale data returned from RAID cache
2193 Details : If only a small amount of IO is done to the RAID device before
2194 reading it again it is possible to get stale data from the RAID
2195 cache instead of reading it from disk.
2198 Frequency : always for sles10 kernel
2200 Description: sles10 support
2201 Details : ll_follow_link: compile fixes and using of nd_set_link
2202 under newer kernels.
2205 Frequency : depends on arch, kernel and compiler version, always on sles10
2208 Description: recursive or deep enough symlinks cause stack overflow
2209 Details : getting rid of large stack-allocated variable in
2213 Frequency : depends on hardware
2215 Description: lustre write performance loss in the SLES10 kernel
2216 Details : the performance loss is caused by using of write barriers in the
2217 ext3 code. The SLES10 kernel turns barrier support on by
2218 default. The fix is to undo that change for ldiskfs.
2220 ------------------------------------------------------------------------------
2222 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2224 * Support for kernels:
2225 2.6.9-42.0.3EL (RHEL 4)
2226 2.6.5-7.276 (SLES 9)
2227 2.4.21-47.0.1.EL (RHEL 3)
2228 2.6.12.6 vanilla (kernel.org)
2232 Frequency : quota enabled and large files being deleted
2234 Description: releasing more than 4GB of quota at once hangs OST
2235 Details : If a user deletes more than 4GB of files on a single OST it
2236 will cause the OST to spin in an infinite loop. Release
2237 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2242 Description: statfs data retrieved from /proc may be stale or zero
2243 Details : When reading per-device statfs data from /proc, in the
2244 {kbytes,files}_{total,free,avail} files, it may appear
2245 as zero or be out of date.
2248 Frequency : systems with MD RAID1 external journal devices
2250 Description: lconf's call to blkid is confused by RAID1 journal devices
2251 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2252 of returning all block devices that match the journal UUID.
2255 Frequency : always, for aggregate stripe size over 4GB
2257 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2258 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2259 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2262 Frequency : always if "lfs find" used on a local file/directory
2264 Description: "lfs find" segfaults if used on a local file/directory
2265 Details : The case where a directory component was not specified wasn't
2266 handled correctly. Handle this properly.
2269 Frequency : always on ppc64
2271 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2272 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2273 perform endianness conversion.
2276 Frequency : rarely (truncate to non-zero file size after write under load)
2277 Bugzilla : 10730, 10687
2278 Description: Files padded with zeros to next 4K multiple
2279 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2280 it is possible that files that are truncated to a non-zero size
2281 immediately after being written are filled with zero bytes beyond
2282 the truncated size. No file data is lost.
2284 Severity : enhancement
2286 Description: Allow recovery/failover for liblustre clients.
2287 Details : liblustre clients were unaware of failover configurations until
2290 Severity : enhancement
2292 Description: user file locks should fail when not mounting with flock option
2293 Details : Set up an error-returning stub in ll_file_operations.lock field
2294 to prevent incorrect behaviour when client is mounted without
2295 flock option. Also, set up properly f_op->flock field for
2299 Frequency : always on ia64
2301 Description: "lfs df" loops on printing out MDS statfs information
2302 Details : The obd_ioctl_data was not initialized and in some systems
2303 this caused a failure during the ioctl that did not return
2304 an error. Initialize the struct and return an error on failure.
2307 Frequency : SLES 9 only
2309 Description: Error of copying files with lustre special EAs as root
2310 Details : Client side always return success for setxattr call for lustre
2311 special xattr (currently only "trusted.lov").
2314 Frequency : rarely on clusters with both ia64+i386 clients
2316 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2317 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2318 clients are doing concurrent writes to the same file it is
2319 possible that the ia64 clients may LASSERT because the OST
2320 extent locks are not PAGE_SIZE aligned. Ensure that grown
2321 locks are always aligned on the request boundary.
2324 Frequency : specific use, occasional
2326 Description: Overwriting in use executable truncates on-disk binary image
2327 Details : If one node attempts to overwrite an executable in use by
2328 another node, we now correctly return ETXTBSY instead of
2329 truncating the file.
2331 Severity : enhancement
2333 Description: Async OSC create to avoid the blocking unnecessarily.
2334 Details : If a OST has no remain object, system will block on the creating
2335 when need to create a new object on this OST. Now, ways use
2336 pre-created objects when available, instead of blocking on an
2337 empty osc while others are not empty. If we must block, we block
2338 for the shortest possible period of time.
2343 Description: chmod on Lustre root is propagated to other clients
2344 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2345 invalid by the follow_mount time.
2348 Frequency : liblustre clients only
2350 Description: Race in 'instant cancel' lock handling could lead to such locks
2351 never to be granted in case of SMP MDS
2352 Details : Do not destroy not yet granted but cbpending locks in
2356 Frequency : replay/resend of open
2358 Description: non null lock assetion failure in mds_intent_policy
2359 Details : Trying to replay/resend lockless open requests resulted in
2360 mds_open() returning 0 with no lock. Now it sets a flag if
2361 a lock is going to be returned.
2363 Severity : enhancement
2365 Description: Checksum enhancements
2366 Details : New checksum enhancements allow for resending RPCs that failed
2369 Severity : enhancement
2371 Description: Tunables on number of dirty pages in cacche
2372 Details : Allow to set limit on number of dirty pages cached.
2377 Description: client crash on unmount - lock still has references
2378 Details : In some error handling cases it was possible to leak a lock
2379 reference on a client while accessing a file. This was not
2380 harmful to the client during operation, but would cause the
2381 client to crash when the filesystem is unmounted.
2384 Frequency : specific case, rare
2386 Description: ETXTBSY on mds though file not in use by client
2387 Details : ETXTBSY is no longer incorrectly returned when attempting to
2388 chmod or chown a directory that the user previously tried to
2389 execute or a currently-executing binary.
2392 Frequency : extremely rare except on liblustre-based clients
2394 Description: Lustre space not freed when files are deleted
2395 Details : Clean up open-unlinked files after client eviction. Previously
2396 the unlink was skipped and the files remained as orphans.
2401 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2402 Details : In some cases it was possible to send a blocking callback to
2403 a client doing a glimpse, even though that client didn't get
2404 a lock granted. When the glimpse lock is cancelled on the OST
2405 the freed lock is left on the waiting list and corrupted the list.
2408 Frequency : all core dumps
2410 Description: Broke core dumps to lustre
2411 Details : Negative dentry may be unhashed if parent does not have UPDATE
2412 lock, but some callers, e.g. do_coredump, expect dentry to be
2413 hashed after successful create, hash it in ll_create_it.
2415 ------------------------------------------------------------------------------
2417 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2419 * Support for kernels:
2420 2.6.9-42.0.2.EL (RHEL 4)
2421 2.6.5-7.276 (SLES 9)
2422 2.4.21-40.EL (RHEL 3)
2423 2.6.12.6 vanilla (kernel.org)
2427 Frequency : always on RHEL 3
2429 Description: Number of open files grows over time
2430 Details : The number of open files grows over time, whether or not
2431 Lustre is started. This was due to a filp leak introduced
2432 by one of our kernel patches.
2434 ------------------------------------------------------------------------------
2436 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2438 * Support for kernels:
2439 2.6.9-42.EL (RHEL 4)
2440 2.6.5-7.276 (SLES 9)
2441 2.4.21-40.EL (RHEL 3)
2442 2.6.12.6 vanilla (kernel.org)
2447 Bugzilla : 5719, 9635, 9792, 9684
2448 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2449 Details : If a server is under heavy load and cannot reply to new
2450 connection requests before the client resends the (re)connect,
2451 the connection handling code can behave badly if two service
2452 threads are concurrently handing separate (re)connections from
2453 the same client. Add better locking to the connection handling
2454 code, and ensure that only a single connection will be processed
2455 for a given client UUID, even if the lock is dropped.
2457 Severity : enhancement
2459 Description: add TCP zero-copy support to kernel
2460 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2461 sends if the hardware supports scatter-gather and checksumming.
2462 This allows socklnd to do client-write and server-read more
2463 efficiently and reduce CPU utilization from skbuf copying.
2466 Frequency : only if NFS exporting from client
2468 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2469 Details : If setting attributes on a file created under NFS that had
2470 never been opened it would be possible to oops the client
2471 if the file had no objects.
2474 Frequency : always for liblustre
2476 Description: liblustre client does MDS+OSTs setattr RPC for each write
2477 Details : When doing a write from a liblustre client, the client
2478 incorrectly issued an RPC to the MDS and each OST the file was
2479 striped over in order to update the timestamps. When writing
2480 with small chunks and many clients this could overwhelm the MDS
2481 with RPCs. In all cases it would slow down the write because
2482 these RPCs are unnecessary.
2484 Severity : enhancement
2486 Description: allow number of MDS service threads to be changed at module load
2487 Details : It is now possible to change the number of MDS service threads
2488 running. Adding "options mds mds_num_threads={N}" to the MDS's
2489 /etc/modprobe.conf will set the number of threads for the next
2490 time Lustre is restarted (assuming the "mds" module is also
2491 reloaded at that time). The default number of threads will
2492 stay the same, 32 for most systems.
2497 Description: OST crash if filesystem is unformatted or corrupt
2498 Details : If an OST is started on a device that has never been formatted
2499 or if the filesystem is corrupt and cannot even mount then the
2500 error handling cleanup routines would dereference a NULL pointer.
2505 Description: NULL pointer deref in llap_from_page.
2506 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2507 incorrect) mapping if the page was truncated/reclaimed while it was
2508 searched for. Check for this condition and skip such pages when
2509 doing readahead. Introduce extra check to llap_from_page() to
2510 verify page->mapping->host is non-NULL (so page is not anonymous).
2513 Frequency : Sometimes when using sys_sendfile
2515 Description: "page not covered by a lock" warnings from ll_readpage
2516 Details : sendfile called ll_readpage without right page locks present.
2517 Now we introduced ll_file_sendfile that does necessary locking
2518 around call to generic_file_sendfile() much like we do in
2522 Frequency : with certain MDS communication failures at client mount time
2524 Description: NULL pointer deref after failed client mount
2525 Details : a client connection request may delayed by the network layer
2526 and not be sent until after the PTLRPC layer has timed out the
2527 request. If the client fails the mount immediately it will try
2528 to clean up before the network times out the request. Add a
2529 reference from the request import to the obd device and delay
2530 the cleanup until the network drops the request.
2533 Frequency : occasionally during client (re)connect
2535 Description: assertion failure during client (re)connect
2536 Details : processing a client connection request may be delayed by the
2537 client or server longer than the client connect timeout. This
2538 causes the client to resend the connection request. If the
2539 original connection request is replied in this interval, the
2540 client may trip an assertion failure in ptlrpc_connect_interpret()
2541 which thought it would be the only running connect process.
2544 Frequency : only with obd_echo servers and clients that are rebooted
2546 Description: kernel BUG accessing uninitialized data structure
2547 Details : When running an obd_echo server it did not start the ping_evictor
2548 thread, and when a client was evicted an uninitialized data
2549 structure was accessed. Start the ping_evictor in the RPC
2550 service startup instead of the OBD startup.
2552 Severity : enhancement
2553 Bugzilla : 10193 (patchless)
2554 Description: Remove dependency on various unexported kernel interfaces.
2555 Details : No longer need reparent_to_init, exit_mm, exit_files,
2556 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2559 Frequency : rare (only users of deprecated and unsupported LDAP config)
2561 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2562 Details : LDAP apparently contains 'lustreName' attributes instead of
2563 'name'. A simple remapping of the name is sufficient.
2566 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2568 Description: waiting_locks_callback trips kernel BUG if client is evicted
2569 Details : Running with the dump_on_timeout debug flag turned on makes
2570 it possible that the waiting_locks_callback() can try to dump
2571 the Lustre kernel debug logs from an interrupt handler. Defer
2572 this log dumping to the expired_lock_main() thread.
2574 Severity : enhancement
2576 Description: Support NFS exporting on 2.6 kernels.
2577 Details : Implement non-rawops metadata methods for NFS server to use without
2578 changing NFS server code.
2581 Frequency : very rare (synthetic metadata workload only)
2583 Description: two racing renames might cause an MDS thread to deadlock
2584 Details : Running the "racer" program may cause one MDS thread to rename
2585 a file from being the source of a rename to being the target of
2586 a rename at exactly the same time that another thread is doing
2587 so, and the second thread has already enqueued these locks after
2588 doing a lookup of the target and is trying to relock them in
2589 order. Ensure that we don't try to re-lock the same resource.
2592 Frequency : only very large systems with liblustre clients
2594 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2595 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2596 all OSTs in parallel. This allows the eviction of stale liblustre
2597 clients to proceed much faster than if they were done in series,
2598 and also offers similar improvements for other set_info RPCs.
2603 Description: excessive CPU usage during initial read phase on client
2604 Details : During the initial read phase on a client, it would agressively
2605 retry readahead on the file, consuming too much CPU and impacting
2606 performance (since 1.4.5.8). Improve the readahead algorithm
2607 to avoid this, and also improve some other common cases (read
2608 of small files in particular, where "small" is files smaller than
2609 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2614 Description: MDS crash when receiving packet with unknown intent.
2615 Details : Do not LBUG in unknown intent case, just return -EFAULT
2617 Severity : enhancement
2618 Bugzilla : 9293, 9385
2619 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2620 Details : Do not serialize getattr (non-intent version) and statfs.
2623 Frequency : occasional, when OST network is overloaded/intermittent
2625 Description: client evicted by OST after bulk IO timeout
2626 Details : If a client sends a bulk IO request (read or write) the OST
2627 may evict the client if it is unresposive to its data GET/PUT
2628 request. This is incorrect if the network is overloaded (takes
2629 too long to transfer the RPC data) or dropped the OST GET/PUT
2630 request. There is no need to evict the client at all, since
2631 the pinger and/or lock callbacks will handle this, and the
2632 client can restart the bulk request.
2635 Frequency : Always when mmapping file with no objects
2637 Description: client crashes when mmapping file with no objects
2638 Details : Check that we actually have objects in a file before doing any
2639 operations on objects in ll_vm_open, ll_vm_close and
2645 Description: Request leak when working with deleted CWD
2646 Details : Introduce advanced request refcount tracking for requests
2647 referenced from lustre intent.
2649 Severity : Enhancement
2651 Description: Cache open file handles on client.
2652 Details : MDS now will return special lock along with openhandle, if
2653 requested and client is allowed to hold openhandle, even if unused,
2654 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2655 closing files for every read/write openration.
2657 Severity : Enhancement
2659 Description: Cache open negative dentries on client when possible.
2660 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2661 negative dentries on lock revocation.
2666 Description: Remounting a client read-only wasn't possible with a zconf mount
2667 Details : It wasn't possible to remount a client read-only with llmount.
2669 Severity : enhancement
2670 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2671 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2672 MPICH is included to add Lustre-specific ADIO interfaces.
2673 This is based closely on the UFS ADIO layer and only differs
2674 in file creation, in order to allow the OST striping to be set.
2675 This is user-contributed code and not supported by CFS.
2680 Description: extended inode attributes (immutable, append-only) work improperly
2681 when 2.4 and 2.6 kernels are used on client/server or vice versa
2682 Details : Introduce kernel-independent values for these flags.
2684 Severity : enhancement
2687 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2688 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2689 granularity. Now, Fractional MB granularity is be supported,
2690 this is very useful for low memory system.
2692 Severity : enhancement
2694 Description: Getattr by fid
2695 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2696 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2697 have locks and avoids getattr rpc requests for it on every stat.
2700 Frequency : Always, for filesystems larger than 2TB
2702 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2703 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2704 filesystems larger than 2TB to be mounted with mballoc enabled.
2707 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2709 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2710 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2711 filesystems larger than 2TB is mounted then the kernel will
2712 silently corrupt the start of the filesystem. CONFIG_LBD is
2713 enabled for all CFS-supported kernels, but the possibility of
2714 this happening with a modified kernel config exists.
2716 Severity : enhancement
2718 Description: add client O_DIRECT support for 2.6 kernels
2719 Details : It is now possible to do O_DIRECT reads and writes to files
2720 in the Lustre client mountpoint on 2.6 kernel clients.
2722 Severity : enhancement
2724 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2725 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2726 parallel, not waiting for response from every OST before sending
2727 a rpc to the next OST.
2732 Description: setattr vs write race when updating file timestamps
2733 Details : Client processes that update a file timestamp into the past
2734 right after writing to the file (e.g. tar) it is possible that
2735 the updated file modification time can be reset to the current
2736 time due to a race between processing the setattr and write RPC.
2738 Severity : enhancement
2740 Description: Bring 'lfs find' closer in line with regular Linux find.
2741 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2742 -print0 options and obtains all the needed info through the lustre
2745 Severity : enhancement
2747 Description: support up to 1024 configured devices on one node
2748 Details : change obd_dev array from statically allocated to dynamically
2749 allocated structs as they are first used to reduce memory usage
2754 Description: Flush dirty partially truncated pages during truncate
2755 Details : Immediatelly flush partially truncated pages in filter_setattr,
2756 this way we completely avoid having any pages in page cache on OST
2757 and can retire ugly workarounds during writes to flush such pages.
2762 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2763 Details : protect inode from truncation within vfs_unlink() context
2764 just take a reference before calling vfs_unlink() and release it
2765 when parent's i_sem is free.
2770 Description: last_id value checked outside lock on OST caused LASSERT failure
2771 Details : If there were multiple MDS->OST object precreate requests in
2772 flight, it was possible that the OST's last object id was checked
2773 outside a lock and incorrectly tripped an assertion. Move checks
2774 inside locks, and discard old precreate requests.
2777 Frequency : always, if extents are used on OSTs
2779 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2780 Details : OSTs using the extents format would not zero the high 16 bits of
2781 the index physical block number. This is not a problem for any
2782 OST filesystems smaller than 16TB, and no kernels support ext3
2783 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2784 new/modified files) and can be fixed for existing filesystems
2785 with e2fsprogs-1.39-cfs1.
2790 Description: import connection selection may be incorrect if timer wraps
2791 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2792 import connections to be ignored if the 32-bit jiffies counter
2793 wraps. Use a 64-bit jiffies counter.
2796 Frequency : very large clusters immediately after boot
2798 Description: LNET request buffers exhausted under heavy short-term load
2799 Details : If a large number of client requests are generated on a service
2800 that has previously never seen so many requests it is possible
2801 that the request buffer growth cannot keep up with the spike in
2802 demand. Instead of dropping incoming requests, they are held in
2803 the LND until the RPC service can accept more requests.
2806 Frequency : Sometimes during replay
2808 Description: Assertion failure in ll_local_open after replay.
2809 Details : If replay happened on an open request reply before we were able
2810 to set replay handler, reply will become not swabbed tripping the
2811 assertion in ll_local_open. Now we set the handler right after
2812 recognising of open request
2815 Frequency : very rare
2817 Description: kernel reports "badness in vsnprintf"
2818 Details : Reading from the "recovery_status" /proc file in small chunks
2819 may cause a negative length in lprocfs_obd_rd_recovery_status()
2820 call to vsnprintf() (which is otherwise harmless). Exit early
2821 if there is no more space in the output buffer.
2823 Severity : enhancement
2825 Description: clear OBD RPC statistics by writing to them
2826 Details : It is now possible to clear the OBD RPC statistics by writing
2827 to the "stats" file.
2832 Description: Client mtime is not the same on different clients after utimes
2833 Details : In some cases, the client was using the utimes() syscall on
2834 a file cached on another node. The clients now validate the
2835 ctime from the MDS + OSTs to determine which one is right.
2840 Description: Inability to activate failout mode
2841 Details : lconf script incorrectly assumed that in pythong string's numeric
2842 value is used in comparisons.
2845 Frequency : always with multiple stripes per file
2847 Description: Inefficient object allocation for mutli-stripe files
2848 Details : When selecting which OSTs to stripe files over, for files with
2849 a stripe count that divides evenly into the number of OSTs,
2850 the MDS is always picking the same starting OST for each file.
2851 Return the OST selection heuristic to the original design.
2856 Description: mount failures may take full timeout to return an error
2857 Details : Under some heavy load conditions it is possible that a
2858 failed mount can wait for the full obd_timeout interval,
2859 possibly several minutes, before reporting an error.
2860 Instead return an error as soon as the status is known.
2862 Frequency : quota enabled and large files being deleted
2864 Description: releasing more than 4GB of quota at once hangs OST
2865 Details : If a user deletes more than 4GB of files on a single OST it
2866 will cause the OST to spin in an infinite loop. Release
2867 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2872 Description: statfs data retrieved from /proc may be stale or zero
2873 Details : When reading per-device statfs data from /proc, in the
2874 {kbytes,files}_{total,free,avail} files, it may appear
2875 as zero or be out of date.
2878 Frequency : systems with MD RAID1 external journal devices
2880 Description: lconf's call to blkid is confused by RAID1 journal devices
2881 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2882 of returning all block devices that match the journal UUID.
2885 Frequency : always, for aggregate stripe size over 4GB
2887 Description: assertion fails when trying to use 4GB stripe size
2888 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2889 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2892 Frequency : always on ppc64
2894 Description: the first write on an ext3 filesystem with mballoc got stuck
2895 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2896 perform endianness conversion.
2898 ------------------------------------------------------------------------------
2900 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2902 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2903 INTEROPERATE with older versions automatically. Please read the
2904 user documentation before upgrading any part of a live system.
2905 * WARNING: Lustre networking configuration changes are required with
2906 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2909 * Support for kernels:
2910 2.6.9-22.0.2.EL (RHEL 4)
2911 2.6.5-7.244 (SLES 9)
2912 2.6.12.6 vanilla (kernel.org)
2915 Severity : enhancement
2916 Bugzilla : 7981/8208
2917 Description: Introduced Lustre Networking (LNET)
2918 Details : LNET is new networking infrastructure for Lustre, it includes
2919 a reorganized network configuration mode (see the user
2920 documentation for full details) as well as support for routing
2921 between different network fabrics. Lustre Networking Devices
2922 (LNDs) for the supported network fabrics have also been
2923 created for this new infrastructure.
2925 Severity : enhancement
2926 Description: Introduced Access control lists
2927 Details : clients can set ACLs on files and directories in order to have
2928 more fine-grained permissions than the standard Unix UGO+RWX.
2929 The MDS must be started with the "-o acl" mount option.
2931 Severity : enhancement
2932 Description: Introduced filesystem quotas
2933 Details : Administrators may now establish per-user quotas on the
2936 Severity : enhancement
2938 Description: Configuration change for the XT3
2939 The PTLLND is now used to run Lustre over Portals on the XT3
2940 The configure option(s) --with-cray-portals are no longer used.
2941 Rather --with-portals=<path-to-portals-includes> is used to
2942 enable building on the XT3. In addition to enable XT3 specific
2943 features the option --enable-cray-xt3 must be used.
2948 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2949 Details : A race between allocating a new llcd and re-getting the llcd_lock
2950 allowed another thread to grab newly-allocated llcd.
2952 Severity : enhancement
2954 Description: 2.6 OST async journal commit and locking fix to improve performance
2955 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2956 2.6 kernels are now async as they already were in 2.4 kernels so
2957 that they can commit concurrently with the network bulk transfer.
2958 For block-allocated files the filter allocation semaphore is held
2959 to avoid filesystem fragmentation during allocation. BKL lock
2960 removed for 2.6 xattr operations where it is no longer needed.
2965 Description: lconf incorrectly determined whether two IP networks could talk
2966 Details : In some more complicated routing and multiple-network
2967 configurations, lconf will avoid trying to make a network
2968 connection to a disjoint part of the IP space. It was doing the
2969 math incorrectly for one set of cases.
2974 Description: Fix for potential infinite loop processing records in an llog.
2975 Details : If an llog record is corrupted/zeroed, it is possible to loop
2976 forever in llog_process(). Validate the llog record length
2977 and skip the remainder of the block on error.
2980 Frequency : occasional (liblustre only)
2982 Description: liblustre could not open files whose last component is a symlink
2983 Details : sysio_path_walk() would incorrectly pass the open intent to
2984 intermediate path components.
2987 Frequency : rare (liblustre only with non-standard tuning)
2988 Bugzilla : 7201 (7350)
2989 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2990 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2991 after it has been released, only do this for referenced locks
2993 Severity : enhancement
2995 Description: specifying an (invalid) directory default stripe_size of -1
2996 would reset the directory default striping
2997 Details : stripe_size -1 was used internally to signal directory stripe
2998 removal, now use "all default" to signal dir stripe removal
2999 as a directory striping of "all default" is not useful
3002 Frequency : common for large clusters running liblustre clients
3004 Description: doing an ls when liblustre clients are running is slow
3005 Details : sending a glimpse AST to a liblustre client waits for every AST
3006 to time out, as liblustre clients will not respond. Since they
3007 cannot cache data we refresh the OST lock LVB from disk instead.
3009 Severity : enhancement
3011 Description: doing an ls at the same time as file IO can be slow
3012 Details : enqueue and other "small" requests can be blocked behind many
3013 large IO requests. Create a new OST IO portal for non-IO
3014 requests so they can be processed faster.
3017 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3019 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3020 created file with mode 000
3021 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3022 allow an ftruncate() as a non-root user to a file with mode 000.
3023 Lustre now respects this flag to disable mode checks when
3024 truncating a file owned by the user
3027 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3030 Description: Revoking locks from clients that went dead or catatonic might take
3032 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3033 cancellation of such locks instant on servers without waiting for
3034 any reply from clients. Clients drops these locks when cancel
3035 notification from server is received without replying.
3038 Frequency : liblustre-only, when liblustre client dies or becomes busy
3040 Description: Doing ls on Linux clients can take a long time with active
3042 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3043 granting such locks to it in the first place if possible. Locks
3044 are taken by proxy on the OST during the read or write and
3045 dropped immediately afterward. Add connect flags handling, do
3046 not grant locks to liblustre clients for glimpse ASTs.
3048 Severity : enhancement
3050 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3051 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3052 sequential reads and behaves badly with applications doing random
3053 reads. Improve it by reading ahead at least read region, and
3054 avoiding excessive large RPC for small reads.
3056 Severity : enhancement
3058 Description: Creating more than 1000 files for a single job may cause a load
3059 imbalance on the OSTs if there are also a large number of OSTs.
3060 Details : qos_prep_create() uses an OST index reseed value that is an
3061 even multiple of the number of available OSTs so that if the
3062 reseed happens in the middle of the object allocation it will
3063 still utilize the OSTs as uniformly as possible.
3068 Description: OST or MDS may oops in ping_evictor_main()
3069 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3070 but doesn't restart at beginning of obd_exports_timed list
3073 Severity : enhancement
3075 Description: improve by-nid export eviction on the MDS and OST
3076 Details : allow multiple exports with the same NID to be evicted at one
3077 time without re-searching the exports list.
3080 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3082 Description: MDS may oops in groups_free()
3083 Details : in rare race conditions a newly allocated group_info struct is
3084 freed again, and this can be NULL. The 2.4 compatibility code
3085 for groups_free() checked for a NULL pointer, but 2.6 did not.
3088 Frequency : common for liblustre clients doing little filesystem IO
3089 Bugzilla : 9352, 7313
3090 Description: server may evict liblustre clients accessing contended locks
3091 Details : if a client is granted a lock or receives a completion AST
3092 with a blocking AST already set it would not reply to the AST
3093 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3096 Frequency : lfs setstripe, only systems with more than 160 OSTs
3098 Description: unable to set striping with a starting offset beyond OST 160
3099 Details : llapi_create_file() incorrectly limited the starting stripe
3100 index to the maximum single-file stripe count.
3103 Frequency : LDAP users only
3105 Description: lconf did not handle in-kernel recovery with LDAP properly
3106 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3108 Severity : enhancement
3110 Description: bind OST threads to NUMA nodes to improve performance
3111 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3112 node and do their allocations there to localize memory access
3114 Severity : enhancement
3116 Description: llmount can determine client NID directly from Myrinet (GM)
3117 Details : the client NID code from gmnalnid was moved directly into
3118 llmount, removing the need to use this or specifying the
3119 client NID explicitly when mounting GM clients with zeroconf
3122 Frequency : if client is started with down MDS
3124 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3125 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3126 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3127 ptlrpc_import_delayed_req() because we didn't check if the
3128 request was interrupted, and we also didn't break out of the
3129 event loop if there was no timeout
3134 Description: data loss during non-page-aligned writes to a single file from
3135 both multiple nodes and multiple threads on one node at same time
3136 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3137 inconsistency led to false short-reads, that were cached and later
3138 used by ->prepare_write() to fill in partially written page,
3139 leading to data loss.
3142 Frequency : always, if lconf --abort_recovery used
3144 Description: lconf --abort_recovery fails with 'Operation not supported'
3145 Details : lconf was attempting to abort recovery on the MDT device and not
3148 Severity : enhancement
3150 Description: remove cleanup logs
3151 Details : replace lconf-generated cleanup logs with lustre internal
3152 cleanup routines. Eliminates the need for client-cleanup and
3155 Severity : enhancement
3157 Description: add support for EAs (user and system) on lustre filesystems
3158 Details : it is now possible to store extended attributes in the Lustre
3159 client filesystem, and with the user_xattr mount option it
3160 is possible to allow users to store EAs on their files also
3162 Severity : enhancement
3164 Description: Add possibility (config option) to show minimal available OST free
3166 Details : When compiled with --enable-mindf configure option, statfs(2)
3167 (and so, df) will return least minimal free space available from
3168 all OSTs as amount of free space on FS, instead of summary of
3169 free spaces of all OSTs.
3171 Severity : enhancement
3173 Description: do not expand extent locks acquired on OST-side
3174 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3175 by server: they are not cached anyway.
3178 Frequency : when mmap is used/binaries executed from Lustre
3180 Description: Unmmap pages before throwing them away from read cache.
3181 Details : llap_shrink cache now attempts to unmap pages before discarding
3182 them (if unmapping failed - do not discard). SLES9 kernel has
3183 extra checks that trigger if this unmapping is not done first.
3188 Description: lconf didn't resolve symlinks before checking to see whether a
3189 given mountpoint was already in use
3192 Frequency : when migrating failover services
3193 Bugzilla : 6395, 9514
3194 Description: When migrating a subset of services from a node (e.g. failback
3195 from a failover service node) the remaining services would
3196 time out and evict clients.
3197 Details : lconf --force (implied by --failover) sets the global obd_timeout
3198 to 5 seconds in order to quickly disconnect, but this caused
3199 other RPCs to time out too quickly. Do not change the global
3200 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3202 Severity : enhancement
3203 Frequency : if MDS is started with down OST
3204 Bugzilla : 9439,5706
3205 Description: Allow startup/shutdown of an MDS without depending on the
3206 availability of the OSTs.
3207 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3208 Add appropriate locking and lov-osc refcounts for safe
3209 cleaning. Add osc abort_inflight calls in case the
3210 synchronize never started.
3213 Frequency : occasional (Cray XT3 only)
3215 Description: root not authorized to access files in CRAY_PORTALS environment
3216 Details : The client process capabilities were not honoured on the MDS in
3217 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3218 been accessed by an authorized user then root was able to access
3219 the file on the local client also. The root user capabilities
3220 are now allowed on the MDS, as this environment has secure UID.
3223 Frequency : occasional
3225 Description: ldiskfs "too long searching" message happens too often
3226 Details : A debugging message (otherwise harmless) prints too often on
3227 the OST console. This has been reduced to only happen when
3228 there are fragmentation problems on the filesystem.
3233 Description: Division by zero in statfs when all OSCs are inactive
3234 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3235 lov_statfs divides by value returned by lov_get_stripecnt().
3239 Bugzilla : 9489, 3273
3240 Description: First write from each client to each OST was only 4kB in size,
3241 to initialize client writeback cache, which caused sub-optimal
3242 RPCs and poor layout on disk for the first writen file.
3243 Details : Clients now request an initial cache grant at (re)connect time
3244 and so that they can start streaming writes to the cache right
3245 away and always do full-sized RPCs if there is enough data.
3246 If the OST is rebooted the client also re-establishes its grant
3247 so that client cached writes will be honoured under the grant.
3252 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3253 Details : Now I/O RPCs go to different portal number and (presumably) fast
3254 lock requests (and glimses) and other RPCs get their own service
3255 threads pool that should be able to service those RPCs
3258 Severity : enhancement
3260 Description: Ability to exchange lustre version between client and servers and
3261 issue warnings at client side if client is too old. Also for
3262 liblustre clients there is ability to refuse connection of too old
3264 Details : New 'version' field is added to connect data structure that is
3265 filled with version info. That info is later checked by server and
3269 Frequency : rare, liblustre only.
3270 Bugzilla : 9296, 9581
3271 Description: Two simultaneous writes from liblustre at offset within same page
3272 might proceed at the same time overwriting eachother with stale
3274 Details : I/O lock withing llu_file_prwv was released too early, before data
3275 actually was hitting the wire. Extended lock-holding time until
3276 server acknowledges receiving data.
3279 Frequency : extremely rare. Never observed in practice.
3281 Description: avoid generating lustre_handle cookie of 0.
3282 Details : class_handle_hash() generates handle cookies by incrementing
3283 global counter, and can hit 0 occasionaly (this is unlikely, but
3284 not impossible, because initial value of cookie counter is
3285 selected randonly). Value of 0 is used as a sentinel meaning
3286 "unassigned handle" --- avoid it. Also coalesce two critical
3287 sections in this function into one.
3289 Severity : enhancement
3291 Description: allow liblustre clients to delegate truncate locking to OST
3292 Details : To avoid overhead of locking, liblustre client instructs OST to
3293 take extent lock in ost_punch() on client's behalf. New connection
3294 flag is added to handle backward compatibility.
3296 Severity : enhancement
3297 Bugzilla : 4928, 7341, 9758
3298 Description: allow number of OST service threads to be specified
3299 Details : a module parameter allows the number of OST service threads
3300 to be specified via "options ost ost_num_threads={N}" in the
3301 OSS's /etc/modules.conf or /etc/modprobe.conf.
3305 Bugzilla : 6146, 9635, 9895
3306 Description: servers crash with bad pointer in target_handle_connect()
3307 Details : In rare cases when a client is reconnecting it was possible that
3308 the connection request was the last reference for that export.
3309 We would temporarily drop the export reference and get a new
3310 one, but this may have been the last reference and the export
3311 was just destroyed. Get new reference before dropping old one.
3313 Severity : enhancement
3314 Frequency : if client is started with failover MDS
3316 Description: Allow multiple MDS hostnames in the mount command
3317 Details : Try to read the configuration from all specified MDS
3318 hostnames during a client mount in case the "primary"
3321 Severity : enhancement
3323 Description: Stop sending data to evicted clients as soon as possible.
3324 Details : Check if the client we are about to send or are sending data to
3325 was evicted already. (Check is done every second of waiting,
3326 for which l_wait_event interface was extended to allow checking
3327 of exit condition at specified intervals).
3330 Frequency : rare, normally only when NFS exporting is done from client
3332 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3334 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3337 Frequency : always, if srand() is called before liblustre initialization
3339 Description: Liblustre uses system PRNG disturbing its usage by user application
3340 Details : Introduce internal to lustre fast and high-quality PRNG for
3341 lustre usage and make liblustre and some other places in generic
3342 lustre code to use it.
3344 Severity : enhancement
3345 Bugzilla : 9477, 9557, 9870
3346 Description: Verify that the MDS configuration logs are updated when xml is
3347 Details : Check if the .xml configuration logs are newer than the config
3348 logs stored on the MDS and report an error if this is the case.
3349 Request --write-conf, or allow starting with --old_conf.
3351 Severity : enhancement
3353 Description: Handle symlinks in the path when checking if Lustre is mounted.
3354 Details : Resolve intermediate symlinks when checking if a client has
3355 mounted a filesystem to avoid duplicate client mounts.
3360 Description: lconf can hit an error exception but still return success.
3361 Details : The lconf command catches the Command error exception at the top
3362 level script context and will exit with the associated exit
3363 status, but doesn't ensure that this exit status is non-zero.
3368 Description: failure of ptlrpc thread startup can cause oops
3369 Details : Starting a ptlrpc service thread can fail if there are a large
3370 number of threads or the server memory is very fragmented.
3371 Handle this without oopsing.
3374 Frequency : always, only if liblustre and non-default acceptor port was used
3376 Description: liblustre cannot connect to servers with non-default acceptor port
3377 Details : tcpnal_set_default_params() was not called and was therefore
3378 ignoring the environment varaible TCPNAL_PORT, as well as other
3379 TCPNAL_ environment variables
3384 Description: two objects could be created on the same OST for a single file
3385 Details : If an OST is down, in some cases it was possible to create two
3386 objects on a single OST for a single file. No problems other
3387 than potential performance impact and spurious error messages.
3391 Bugzilla : 5681, 9562
3392 Description: Client may oops in ll_unhash_aliases
3393 Details : Client dcache may become inconsistent in race condition.
3394 In some cases "getcwd" can fail if the current directory is
3400 Description: Inode refcounting problems in NFS export code
3401 Details : link_raw functions used to call d_instantiate without obtaining
3402 extra inode reference first.
3406 Bugzilla : 9942, 9903
3407 Description: Referencing freed requests leading to crash, memleaks with NFS.
3408 Details : We used to require that call to ll_revalidate_it was always
3409 followed by ll_lookup_it. Also with revalidate_special() it is
3410 possible to call ll_revalidate_it() twice for the same dentry
3411 even if first occurence returned success. This fix changes semantic
3412 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3413 reference on a request referred from the intent.
3414 ll_intent_release() then releases such a request.
3417 Frequency : rare, normally benchmark loads only
3419 Description: unlinked inodes were kept in memory on the client
3420 Details : If a client is repeatedly creating and unlinking files it
3421 can accumulate a lot of stale inodes in the inode slab cache.
3422 If there is no other client load running this can cause the
3423 client node to run out of memory. Instead flush old inodes
3424 from client cache that have the same inode number as a new inode.
3427 Frequency : SLES9 2.6.5 kernel and long filenames only
3428 Bugzilla : 9969, 10379
3429 Description: utime reports stale NFS file handle
3430 Details : SLES9 uses out-of-dentry names in some cases, which confused
3431 the lustre dentry revalidation. Change it to always use the
3435 Frequency : rare, unless heavy write-truncate concurrency is continuous
3436 Bugzilla : 4180, 6984, 7171, 9963, 9331
3437 Description: OST becomes very slow and/or deadlocked during object unlink
3438 Details : filter_destroy() was holding onto the parent directory lock
3439 while truncating+unlinking objects. For very large objects this
3440 may block other threads for a long time and slow overall OST
3441 responsiveness. It may also be possible to get a lock ordering
3442 deadlock in this case, or run out of journal credits because of
3443 the combined truncate+unlink. Solution is to do object truncate
3444 first in one transaction without parent lock, and then do the
3445 final unlink in a new transaction with the parent lock. This
3446 reduces the lock hold time dramatically.
3449 Frequency : rare, 2.4 kernels only
3451 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3452 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3453 before unmounting their filesystem in order to avoid deadlock.
3454 The kernel_locked() function in 2.4 kernels only checks whether
3455 the kernel lock is held, not whether it is this process that is
3456 holding it as 2.6 kernels do.
3461 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3462 Details : The client ptlrpc code may be trying to reconnect to a down
3463 server before a previous connection attempt has timed out.
3464 Increase the reconnect interval to be longer than the connection
3465 timeout interval to avoid sending duplicate connections to
3469 Frequency : echo_client brw_test command
3471 Description: fix echo_client to work with OST preallocated code
3472 Details : OST preallocation code (5137) didn't take echo_client IO path
3473 into account: echo_client calls filter methods outside of any
3474 OST thread and, hence, there is no per-thread preallocated
3475 pages and buffers to use. Solution: hijack pga pages for IO. As
3476 a byproduct, this avoids unnecessary data copying.
3480 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3481 Description: Client can oops in mdc_commit_close() after open replay
3482 Details : It was possible for the MDS to return an open request with no
3483 transaction number in mds_finish_transno() if the client was
3484 evicted, but without actually returning an error. Clients
3485 would later try to replay that open and may trip an assertion
3486 Simplify the client close codepath, and always return an error
3487 from the MDS in case the open is not successful.
3490 Frequency : rare, 2.6 OSTs only
3492 Description: OST may deadlock under high load on fragmented files
3493 Details : If there was a heavy load and highly-fragmented OST filesystems
3494 it was possible to have all the OST threads deadlock waiting on
3495 allocation of biovecs, because the biovecs were not released
3496 until the entire RPC IO was completed. Instead, release biovecs
3497 as soon as they are complete to ensure forward IO progress.
3499 Severity : enhancement
3501 Description: Support for specifying external journal device at mount
3502 Details : If an OST or MDS device is formatted with an external journal
3503 device, this device major/minor is stored in the ext3 superblock
3504 and may not be valid for failover. Allow detecting and
3505 specifying the external journal at mount time.
3510 Description: Mounting an MDS with pending unlinked files may cause oops
3511 Details : target_finish_recovery() calls mds_postrecov() which returned
3512 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3513 considers this an error and immediately begins cleaning up the
3514 lov, just after starting the mds_lov process
3516 Severity : enhancement
3518 Description: Implement 'lfs df' to report actual free space on per-OST basis
3519 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3520 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3521 usage of objects; '-h' to report in human readable format.
3523 ------------------------------------------------------------------------------
3525 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3532 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3533 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3534 at startup during orphan unlinking. Instead, skip buddy update
3535 before setup, as it will be regenerated after recovery is complete.
3540 Description: If an OST is inactive, its locks might reference stale inodes.
3541 Details : lov_change_cbdata() must iterate over all namespaces, even if
3542 they are inactive to clear inode references from the lock.
3544 Severity : enhancement
3545 Frequency : occasional, if non-standard max_dirty_mb used
3547 Description: Client will block write RPCs if not enough grant
3548 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3549 then the client will block writes while waiting for another RPC
3550 to complete instead of consuming its dirty limit. With change
3551 we get improved performance when max_dirty_mb is small.
3553 Severity : enhancement
3554 Bugzilla : 3389, 6253
3555 Description: Add support for supplementary groups on the MDS.
3556 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3557 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3558 lookups for user supplementary groups into a cache.
3562 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3563 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3564 for liblustre O_EXCL is enforced by MDS.
3568 Description: readdir never returns NULL in liblustre.
3569 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3570 dentry in struct dirent.
3574 Frequency : liblustre only, depends on application IO pattern
3575 Description: liblustre clients evicted if not contacting servers
3576 Details : Don't put liblustre clients into the ping_evictor list, so
3577 they will not be evicted by the pinger ever.
3579 Severity : enhancement
3581 Description: Add ability to evict clients by NID from MDS.
3582 Details : By echoing "nid:$NID" string into
3583 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3584 $NID would be instantly evicted from this MDS and from all active
3585 OSTs connected to it.
3589 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3590 Details : lookup_it_finish() used to query file size from OSTs that was not
3595 Description: service threads change working directory to that of init
3596 Details : Starting lustre service threads may pin the working directory
3597 of the parent thread, making that filesystem busy. Threads
3598 now change to the working directory of init to avoid this.
3602 Frequency : during shutdown only
3603 Description: shutdown with a failed MDS or OST can cause unmount to hang
3604 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3609 Frequency : 2.6 only
3610 Description: chmod/chown may include an extra supplementary group
3611 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3612 supplementary group and if none is specified this is used.
3615 Bugzilla : 5479 (6816)
3617 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3618 Details : If lookup is in progress on a file that is unlinked we might try
3619 to revalidate the inode and fail in revalidate after lookup is
3620 complete and ll_file_open() enqueues the open again but
3621 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3624 Frequency : always, if lconf --abort_recovery used
3626 Description: lconf --abort_recovery fails with 'Operation not supported'
3627 Details : lconf was attempting to abort recovery on the MDT device and not
3630 ------------------------------------------------------------------------------
3632 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3637 Frequency : rare (only unsupported configurations with a node running as an
3639 Bugzilla : 6514, 5137
3640 Description: Mounting a Lustre file system on a node running as an OST could
3642 Details : OSTs now preallocates memory needed to write out data at
3643 startup, instead of when needed, to avoid having to
3644 allocate memory in possibly low memory situations.
3645 Specifically, if the file system is mounted on on OST,
3646 memory pressure could force it to try to write out data,
3647 which it needed to allocate memory to do. Due to the low
3648 memory, it would be unable to do so and the node would
3649 become unresponsive.
3651 Severity : enhancement
3653 Description: Addition of lconf --service command line option
3654 Details : lconf now accepts a '--service <arg>' option, which is
3655 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3657 Severity : enhancement
3659 Description: Failover mode is now the default for OSTs.
3660 Details : By default, OSTs will now run in failover mode. To return to
3661 the old behaviour, add '--failout' to the lmc line for OSTs.
3663 Severity : enhancement
3665 Description: Health checks are now provided for MDS and OSTs
3666 Details : Additional detailed health check information on MSD and OSTs
3667 is now provided through the procfs health_check value.
3670 Frequency : occasional, depends on IO load
3672 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3673 after numerous create/delete cycles
3674 Details : The ext3 inode allocation policy would not allocate new inodes
3675 very well on the OSTs because there are no new directories
3676 being created. Instead we look for groups with free space if
3677 the parent directories are nearly full.
3682 Description: Network or server problems during mount may cause partially
3683 mounted clients instead of returning an error.
3684 Details : The config llog parsing code may overwrite the error return
3685 code during mount error handling, returning success instead
3691 Description: MDS can fail to allocate large reply buffers
3692 Details : After long uptimes the MDS can fail to allocate large reply
3693 buffers (e.g. zconf client mount config records) due to memory
3694 fragmentation or consumption by the buffer cache. Preallocate
3695 some large reply buffers so that these replies can be sent even
3696 under memory pressure.
3700 Frequency : rare (liblustre)
3701 Description: fsx running with liblustre complained that using truncate() to
3702 extend the file doesn't work. This patch corrects that issue.
3703 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3704 ATTR_SIZE and lsm use in llu_setattr_raw.
3708 Frequency : rare, only 2.6 kernels
3709 Description: Unusual file access patterns on the MDS may result in inode
3710 data being lost in very rare circumstances.
3711 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3712 code in the 2.6 kernel caused the inode and/or EA data not to
3713 be read from disk, causing single-file corruption.
3717 Frequency : rare, only 2.6 filesystems using extents
3718 Description: Heavy concurrent write and delete load may cause data corruption.
3719 Details : It was possible under high-load situations to have an extent
3720 metadata block in the block device cache from a just-unlinked
3721 file overwrite a newly-allocated data block. We now unmap any
3722 metadata buffers that alias just-allocated data blocks.
3726 Frequency : filesystems with default stripe_count larger than 77
3727 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3728 Details : lconf specifies an inode size of 4096 bytes when the default
3729 stripe_count is larger than 77. This conflicts with the default
3730 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3731 this case to avoid pinning too much memory for large EAs.
3733 ------------------------------------------------------------------------------
3735 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3740 Frequency : rare (extremely heavy IO load with hundreds of clients)
3742 Description: Client is evicted, gets IO error writing to file
3743 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3744 caused clients to be evicted for AST timeouts. The fixes in
3745 bug 5192 mean we no longer need to have such short AST timeouts
3746 so ldlm_timeout has been increased.
3749 Frequency : occasional during --force or --failover shutdown under load
3750 Bugzilla : 5949, 4834
3751 Description: Server oops/LBUG if stopped with --force or --failover under load
3752 Details : a collection of import/export refcount and cleanup ordering
3753 issues fixed for safer force cleanup
3756 Frequency : only filesystems larger than 120 OSTs
3757 Bugzilla : 5990, 6223
3758 Description: lfs getstripe would oops on a very large filesystem
3759 Details : lov_getconfig used kfree on vmalloc'd memory
3762 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3763 Bugzilla : 6242, 6243
3764 Description: reading from files that had been truncated to a non-zero size
3765 but never opened returned no data
3766 Details : ll_file_read() reads zeros from no-object files to EOF
3771 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3772 Details : There's little harm in aborting MDS/OSS recovery and letting it
3773 try again, so I removed the LASSERT and return an error instead.
3775 Severity : enhancement
3777 Description: New debugging infrastructure for tracking down data corruption
3778 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3779 (b) cover more of the client-side code path, and (c) try to narrow
3780 down where problems occurred
3784 Bugzilla : 3819, 4364, 4397, 6313
3785 Description: Racing close and eviction MDS could cause assertion in mds_close
3786 Details : It was possible to get multiple mfd references during close and
3787 client eviction, leading to one thread referencing a freed mfd.
3789 Severity: : enhancement
3790 Bugzilla : 3262, 6359
3791 Description: Attempts to reconnect to servers are now more aggressive.
3792 Details : This builds on the enhanced upcall-less recovery that was added
3793 in 1.4.2. When trying to reconnect to servers, clients will
3794 now try each server in the failover group every 10 seconds. By
3795 default, clients would previously try one server every 25 seconds.
3800 Description: After recovery, certain operations trigger a failed
3801 assertion on a client.
3802 Details : Failing over an mds, using lconf -d --failover, while a
3803 client was doing a readdir() call would cause the client to
3804 LBUG after recovery completed and the readdir() was resent.
3806 Severity : enhancement
3808 Description: Default groups are now added by lconf
3809 Details : You can now run lconf --group <servicename> without having to
3810 manually add groups with lmc.
3813 Frequency : occasional
3815 Description: Nodes with an elan id of 0 trigger a failed assertion
3818 Frequency : always when accessing e.g. tty/console device nodes
3820 Description: tty and some other devices nodes cannot be used on lustre
3821 Details : file's private_data field is used by device data and lustre
3822 values in there got lost. New field was added to struct file to
3823 store fs-specific private data.
3826 Frequency : when exporting Lustre via NFS
3828 Description: NFSD failed occasionally when looking up a path component
3829 Details : NFSD is looking up ".." which was broken in ext3 directories
3830 that had grown large enough to become hashed.
3833 Frequency : Clusters with multiple interfaces not on the same subnet
3835 Description: Nodes will repeatedly try to reconnect to an interface which it
3836 cannot reach and report an error to the log.
3837 Details : Extra peer list entries will be created by lconf with some peers
3838 unreachable. lconf now validates the peer before adding it.
3841 Frequency : Only if a default stripe is set on the filesystem root.
3843 Description: Setting a default stripe on the filesystem root prevented the
3844 filesystem from being remounted.
3845 Details : The client was sending extra request flags in the root getattr
3846 request and did not allocate a reply buffer for the dir EA.
3849 Frequency : occasional, higher if lots of files are accessed by one client
3850 Bugzilla : 6159, 6097
3851 Description: Client trips assertion regarding lsm mismatch/magic
3852 Details : While revalidating inodes the VFS looks up inodes with ifind()
3853 and in rare cases can find an inode that is being freed.
3854 The ll_test_inode() code will free the lsm during ifind()
3855 when it finds an existing inode and then the VFS later attaches
3856 this free lsm to a new inode.
3860 Bugzilla : 6422, 7030
3861 Description: MDS deadlock between mkdir and client eviction
3862 Details : Creating a new file via mkdir or mknod (starting a transaction
3863 and getting the ns lock) can deadlock with client eviction
3864 (gets ns lock and trying to finish a synchronous transaction).
3867 Frequency : occasional
3868 Description: While starting a server, the fsfilt_ext3 module could not be
3870 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3871 kernels. Previously, lconf would still use the ext3 name
3872 when trying to load modules. Now, it will correctly use
3873 ext3 on 2.4 and ldiskfs on 2.6.
3875 Severity : enhancement
3876 Description: The default stripe count has been changed to 1
3877 Details : The interpretation of the default stripe count (0, to lfs
3878 or lmc) has been changed to mean striping across a single
3879 OST, rather than all available. For general usage we have
3880 found a stripe count of 1 or 2 works best.
3882 Severity : enhancement
3883 Description: Add support for compiling against Cray portals.
3884 Details : Conditional compiling for some areas that are different
3888 Frequency : occasional
3889 Bugzilla : 6409, 6834
3890 Description: Creating files with an explicit stripe count may lead to
3891 a failed assertion on the MDS
3892 Details : If some OSTs are full or unavailable, creating files may
3893 trigger a failed assertion on the MDS. Now, Lustre will
3894 try to use other servers or return an error to the
3898 Frequency : occasional
3900 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3901 caused lock timeout during readahead (since 1.4.2).
3902 Details : Processes doing readahead might match a lock that hasn't been
3903 granted yet if there are overlapping and conflicting lock
3904 requests. The readahead process waits on ungranted lock
3905 (original lock is CBPENDING), while OST waits for that process
3906 to cancel CBPENDING read lock and eventually evicts client.
3908 Severity : enhancement
3910 Description: Initial enabling of flock support for clients
3911 Details : Implements fcntl advisory locking and file status functions.
3912 This feature is provided as an optional mount flag (default
3913 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3914 locking are implemented yet, and those that are are not guaranteed
3915 to be completely correct in production environments.
3916 mount -t lustre -o [flock|noflock] ...
3919 Frequency : occasional
3921 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3922 assertion in the ext3 JBD (journaling) layer.
3923 Details : The b_committed_data struct is protected by the big kernel lock
3924 in 2.4 kernels, serializing journal_commit_transaction() and
3925 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3926 access to this struct. In 2.6 kernels there is finer grained
3927 locking to improve SMP performance of the JBD layer.
3931 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3936 Description: merge and cleanup kernel patches.
3939 -----------------------------------------------------------------------------
3941 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3943 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3944 versions of Lustre. Please update all servers and clients to
3945 version 1.4.2 or later at the same time. You must also run
3946 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3947 to update the configuration logs.
3949 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3950 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3951 - swab llog records of type '0' so we get proper header size/idx (5861)
3952 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3953 - fix rename of one directory over another leaking an inode (5953)
3954 - avoid SetPageDirty on 2.6 (5981)
3955 - don't re-add just-being-destroyed locks to the waiting list (5653)
3956 - when creating new directories, inherit the parent's custom
3957 striping settings if present parent (3048)
3958 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3959 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3960 - don't hold client locks on temporary worklist from l_lru (5666)
3961 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3962 - automatically evict dead clients (5921)
3963 - Update file size properly in create+truncate+fstat case (6196)
3964 - Do not unhash mountpoint dentries, do not allow removal of
3966 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3967 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3968 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3969 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3970 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3971 - Correctly handle reads of files with no objects (6243)
3972 - lctl recover will also mark a device active if deactivate used (5933)
3974 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3975 - allow --write-conf on an MDS with different nettype than client (5619)
3976 - don't write config llogs to MDS for mounts not from that MDS (5617)
3977 - lconf should create multiple TCP connections from a client (5201)
3978 - init scripts are now turned off by default; run chkconfig --on
3979 lustre and chkconfig --on lustrefs to use them
3980 - upcalls are no longer needed for clients to recover to failover
3982 - add --abort-recovery option to lconf to abort recovery on device
3984 - add support for an arbitrary number of OSTs (3026)
3985 - Quota support protocol changes.
3986 - forward compatibility changes to wire structs (6007)
3987 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3988 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3989 - improved "lustre status" script
3990 - initialize blocksize for non-regular files (6062)
3991 - added --disable-server and --disable-client configure options (5782)
3992 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3993 - Vanilla 2.4.29 support
3994 - increase maximum number of obd devices to 520 (6242)
3995 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3996 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3997 SLES 9 2.6.5 kernels
3998 - limit stripes per file to 160 (the maximum EA size) (6093)
4000 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4003 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4004 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4005 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4006 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4007 - small ext3 extents cleanups and fixes (5733)
4008 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4009 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4010 - avoid needless client->OST connect, fix handle mismatch (5317)
4011 - fix DLM error path that led to out-of-sync client, long delays (5779)
4012 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4013 - fix several locking issues related to i_size (5492,5624,5654,5672)
4014 - don't move pending lock onto export if it is already evicted (5683)
4015 - fix kernel oops when creating .foo in unlinked directory (5548)
4016 - fix deadlock in obdfilter statistics vs. object create (5811)
4017 - use time_{before,after} to avoid timer jiffies wrap (5882)
4018 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4019 - Do not leak request if server was not able to process it (5154)
4020 - If mds_open unable to find parent dir, make that negative lookup(5154)
git://git.whamcloud.com / fs / lustre-release.git / blob