1 tbd Sun Microsystems, Inc.
4 2.6.16.60-0.23 (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: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
18 : l_getgroups(8), lst(8), routerstat(8)
19 Details : included man pages for llobdstat(8), llstat(8),
20 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
22 Severity : enhancement
24 Description: configure's --enable-quota should check the
25 : kernel .config for CONFIG_QUOTA
26 Details : configure is terminated if --enable-quota is passed but
27 : no quota support is in kernel
31 Description: Remove portals compatibility
32 Details : Remove portals compatibility, not interoperable with releases
37 Description: Resolve device initialization race
38 Details : Prevent proc handler from accessing devices added to the
39 obd_devs array but yet be intialized.
41 Severity : enhancement
43 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
45 Severity : enhancement
47 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
52 Description: Using wrong pointer in osc_brw_prep_request
53 Details : Access to array[-1] can produce panic if kernel compiled with
54 CONFIG_PAGE_ALLOC enabled
58 Description: Client runs out of low memory
59 Details : Consider only lowmem when counting initial number of llap pages
63 Description: *optional* service tags registration
64 Details : if the "service tags" package is installed on a Lustre node
65 When the filesystem is mounted, a local-node service tag will
66 be created. See http://inventory.sun.com/ for more information
67 about the Service Tags asset management system.
71 Description: Kernel BUG tries to release flock
72 Details : Lustre does not destroy flock lock before last reference goes
73 away. So always drop flock locks when client is evicted and
74 perform unlock regardless of successfulness of speaking to MDS.
78 Description: add recount protection for osc callbacks, so avoid panic on shutdown
82 Description: sanity test 65a fails if stripecount of -1 is set
83 Details : handle -1 striping on filesystem in ll_dirstripe_verify
88 Description: ASSERTION(CheckWriteback(page,cmd)) failed
89 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
92 Severity : enhancement
94 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
98 Description: do not process already freed flock
99 Details : flock can possibly be freed by another thread before it reaches
100 to ldlm_flock_completion_ast.
104 Description: LBUG during stress test
105 Details : Need properly lock accesses the flock deadlock detection list.
109 Description: oops in page fault handler
110 Details : kernel page fault handler can return two special 'pages' in error case, don't
111 try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
115 Description: timeout with invalidate import.
116 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which should be
117 handled by ptlrpcd. This produce long age waiting and -ETIMEOUT
118 ptlrpc_invalidate_import and as result LASSERT.
120 Severity : enhancement
122 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
126 Description: enable MGS and MDT services start separately
127 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
128 a co-located MGS without starting the MGS, which is a complement
129 to 'nosvc' mount option.
133 Frequency : after recovery
134 Description: precreate to many object's after del orphan.
135 Details : del orphan st in oscc last_id == next_id and this triger growing
136 count of precreated objects. Set flag LOW to skip increase count
137 of precreated objects.
141 Frequency : rare, on clear nid stats
142 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
143 Details : when clean nid stats sometimes try destroy live entry,
144 and this produce panic in free.
148 Description: Stack overflow during MDS log replay
149 ease stack pressure by using a thread dealing llog_process.
153 Description: wait until IO finished before start new when do lock cancel.
154 Details : VM protocol want old IO finished before start new, in this case
155 need wait until PG_writeback is cleared until check dirty flag and
156 call writepages in lock cancel callback.
158 Severity : enhancement
160 Description: using special macro for print time and cleanup in includes.
164 Description: mds_mfd_close() ASSERTION(rc == 0)
165 Details : In mds_mfd_close(), we need protect inode's writecount change
166 within its orphan write semaphore to prevent possible races.
170 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
174 Frequency : rare, on shutdown ost
175 Description: don't hit live lock with umount ost.
176 Details : shrink_dcache_parent can be in long loop with destroy dentries,
177 use shrink_dcache_sb instead.
181 Description: don't panic with use echo client
182 Details : echo client pass NULL as client nid pointer and this produce null
187 Description: fix build on ppc32
188 Details : compile code with -m64 flag produce wrong object file for ppc32.
192 Description: add message levels for liblustreapi
196 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
197 Details : Move the 'good_osts' check before the 'total_bavail' check. This
198 will result in an -EAGAIN and in the exit call path we call
199 alloc_rr() which will with increasing aggressiveness attempt to
200 aquire precreated objects on the minimum number of required OSCs.
204 Description: Use old size assignment to avoid deadlock
205 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
206 the scheduling while holding a spinlock. We do not need locking
207 for size in ll_update_inode() because size is only updated from
208 the MDS for directories or files without objects, so there is no
209 other place to do the update, and concurrent access to such inodes
210 are protected by the inode lock.
214 Description: resolve "_IOWR redefined" build error on SLES10
218 Description: dump the memory debugging after all modules are unloaded to
219 suppress false negative in conf_sanity test 39
221 Severity : enhancement
223 Description: build kernel-ib packages for OFED 1.3 in our release cycle
228 Description: fix SLES kernel versioning
229 Details : the kernel version for our SLES 10 kernel did not include a "-"
230 before the "smp" at the end. while this was not a problem in
231 general, it did mean that software trying to use the kernel
232 version to try to detect a vendor specific kernel would fail.
233 this was most evident by the OFED build scripts.
237 Description: Don't update lov_desc members until making sure they are valid
238 Details : When updating lov_desc members via proc fs, need fix their
239 validities before doing the real update.
243 Description: don't put request into delay list while invalidate in flight.
244 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
245 import in flight. this produce timeout for invalidate and sometimes
246 can cause stale data.
250 Frequency : on ppc only
251 Description: not convert ost objects for directory because it's not exist.
252 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
256 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
258 Severity : enhancement
260 Description: Update to support 2.6.22.14 vanilla kernel.
264 Frequency : rare, on recovery
265 Description: read procfs can produce deadlock in some situation
266 Details : Holding lprocfs lock which send rpc can produce block for destroy
267 obd objects and this also block reconnect with -EALREADY. This isn't
268 fix all lprocfs bugs - but make it rare.
270 Severity : enhancement
272 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
275 Frequency : frequent on X2 node
277 Description: mdc_set_open_replay_data LBUG
278 Details : Set replay data for requests that are eligible for replay.
282 Description: lustre_mgs: operation 101 on unconnected MGS
283 Details : When MGC is disconnected from MGS long enough, MGS will evict the
284 MGC, and late on MGC cannot successfully connect to MGS and a lot
285 of the error messages complaining that MGS is not connected.
289 Frequency : on network error
290 Description: panic with double free request if network error
291 Details : mdc_finish_enqueue is finish request if any network error ocuring,
292 but it's true only for synchronus enqueue, for async enqueue
293 (via ptlrpcd) this incorrect and ptlrpcd want finish request
296 Severity : enhancement
298 Description: client-side metadata stat-ahead during readdir(directory readahead)
299 Details : perform client-side metadata stat-ahead when the client detects
300 readdir and sequential stat of dir entries therein
303 Frequency : on start mds
305 Description: Implement get_info(last_id) in obdfilter.
308 Frequency : occasional
310 Description: Correctly check stale fid, not start epoch if ost not support SOM
311 Details : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
312 overwrited with new generated fid, but mds can anwer with one of these
313 two fids and both is not stale. setattr incorectly start epoch and
314 assume will be called done_writeting, but without SOM done_writing
318 Frequency : rare, depends on device drivers and load
320 Description: MDS or OSS nodes crash due to stack overflow
321 Details : Code changes in 1.8.0 increased the stack usage of some functions.
322 In some cases, in conjunction with device drivers that use a lot
323 of stack the MDS (or possibly OSS) service threads could overflow
324 the stack. One change which was identified to consume additional
325 stack has been reworked to avoid the extra stack usage.
328 Frequency : occasional
330 Description: Do not fail import if osc_interpret_create gets -EAGAIN
331 Details : If osc_interpret_create got -EAGAIN it immediately exits and
332 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
333 oscc_has_objects and see OSC has no objests and call
334 oscc_internal_create to resend create request.
336 Severity : enhancement
338 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
340 Severity : enhancement
342 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
345 Frequency : very rare
347 Description: Fix replay if there is an un-replied request and open
348 Details : In some cases, older replay request will revert the
349 mcd->mcd_last_xid on MDS which is used to record the client's
352 Severity : enhancement
354 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
356 Severity : enhancement
358 Description: Add rhel5 support to HEAD.
360 Serverity : enhancement
362 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
367 Description: Don't allow skipping OSTs if index has been specified.
368 Details : Don't allow skipping OSTs if index has been specified, make locking
369 in internal create lots better.
373 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
374 Details : access to bitfield in structure is always rounded to long
375 and this produce problem with not atomic change any bit.
379 Description: Lustre make rpms failed.
380 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
381 builds Lustre rpms from tarball.
384 Frequency : rare on shutdown ost
386 Description: If llog cancel was not send before clean_exports phase, this can
387 produce deadlock in llog code.
388 Details : If llog thread has last reference to obd and call class_import_put
389 this produce deadlock because llog_cleanup_commit_master wait when
390 last llog_commit_thread exited, but this never success because was
391 called from llog_commit_thread.
395 Description: allow userland application know is lost one of stripes.
396 Details : fill lvb_blocks with error code on ost and return it to
397 application if error flag found.
401 Description: NULL lov_tgts causing MDS oops
402 Details : more safe checks for NULL lov_tgts for avoid oops.
404 Severity : enhancement
406 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
410 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
411 Details : If an OST is down the MDS will hang indefinitely in
412 obd_statfs_rqset() waiting for the statfs data. While for
413 MDS QOS usage of statfs, it should not stuck in waiting.
415 Severity : enhancement
417 Description: remote_acl support
418 Details : Support ACL-based permission check for remote user.
419 Support setfacl/getfacl for remote user with the utils
420 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
421 the system "{s,g}etfacl" utils.
423 Severity : enhancement
425 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
427 Severity : enhancement
429 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
433 Description: make mgs_setparam() handle fsname containing dash
434 Details : fsname containing a dash does not work with lctl conf_param
436 Severity : enhancement
438 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
440 Severity : enhancement
442 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
446 Frequency : rare, at shutdown
447 Description: access already free / zero obd_namespace.
448 Details : if client_disconnect_export was called without force flag set,
449 and exist connect request in flight, this can produce access to
450 NULL pointer (or already free pointer) when connect_interpret
451 store ocd flags in obd_namespace.
455 Frequency : only at startup
456 Description: not alloc memory with spinlock held.
457 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
458 if any spinlock held.
460 Severity : enhancement
462 Description: make lustre randomly fail allocating memory
463 Details : Make lustre randomly failed allocating memory for testing purpose.
465 Severity : enhancement
467 Description: lost problems with lov objid file
468 Details : Fixes some scability and access to not inited memory problems
469 in work with lov objdid file.
474 Description: lfs find does not continue on file error
475 Details : Continue other files processing when a file/dir is absent.
479 Description: Inconsistent usage of lustre_pack_reply()
480 Details : Standardize the usage of lustre_pack_reply() such that it
481 always generate a CERROR on failure.
484 Frequency : occasional
486 Description: MDS hang or stay in waiting lock
487 Details : If client receive lock with CBPENDING flag ldlm need send lock
488 cancel as separate rpc, to avoid situation when cancel request
489 can't processed due all i/o threads stay in wait lock.
493 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
497 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
501 Description: Update to SLES9 kernel-2.6.5-7.287.3.
505 Description: Update to RHEL4 latest kernel.
507 Severity : enhancement
509 Description: Build SLES10 patchless client fails
510 Details : The configure was broken by run ./configure with
511 --with-linux-obj=.... argument for patchless client. When the
512 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
513 can't search header adequately. Use absolute path such as
514 -I($LINUX)/include instead.
518 Description: interrupt oig_wait produce painc on resend.
519 Details : brw_redo_request can be used for resend requests from ptlrpcd and
520 private set, and this produce situation when rq_ptlrpcd_data not
521 copyed to new allocated request and triggered LBUG on assert
522 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
523 ptlrpcd set if request is changed and can be safety changed to use
528 Description: LASSERT_{REQ,REP}SWAB macros are buggy
529 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
530 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
531 these macros and replace them with their difinitions instead.
535 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
536 Details : Update which_patch & target file for SLES10 latest kernel.
539 Description: Remove client patches from SLES 10 kernel.
540 Details : This causes SLES 10 clients to behave as patchless clients
541 even on a Lustre-patched (server) kernel.
543 Severity : enhancement
545 Description: self-adjustable client's lru lists
546 Details : use adaptive algorithm for managing client cached locks lru
547 lists according to current server load, other client's work
548 pattern, memory activities, etc. Both, server and client
549 side namespaces provide number of proc tunables for controlling
552 Severity : enhancement
554 Description: light-weight GSS support
555 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
556 and reduce performance overhead.
558 Severity : enhancement
560 Description: Linux keyring support
561 Details : Support using service of Linux keyring for Lustre GSS internal
562 context refresh/cache mechanism.
566 Description: Fix errors in lfs documentation
567 Details : Fixes man pages
571 Description: don't use GFP_* in generic Lustre code.
572 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
576 Description: obdclass is limited by single OBD_ALLOC(idarray)
577 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
581 Description: warnings with build patchless client with vanila 2.6.19 and up
582 Details : change old ctl_table style and replace ctl_table/ctl_table_header
583 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
587 Frequency : Only for SLES
588 Description: sanity_quota fail test_1
589 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
590 file, which makes TSTID[2] inunique.
594 Frequency : Only for SLES9
595 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
596 Details : Update target/ChangeLog/which_patch .
600 Frequency : Only for test_10 in sanity-quota.sh
601 Description: a bug in quota test code
602 Details : checking a test flag in a wrong place
606 Frequency : Only for test_13 in sanity-quota.sh
607 Description: a bug in quota test script
608 Details : shouldn't check overfull allocated quota
612 Description: jbd statistics
613 Details : Port older jbd statistics patch for sles10
615 Severity : enhancement
617 Description: Add printing inode info into message about error in writepage.
622 Description: reply_lock_interpret crash due to race with it and lock cancel.
623 Details : Do not replay locks that are being cancelled. Do not reference
624 locks by their address during replay, just by their handle.
628 Frequency : When flocks are used.
629 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
630 Details : Flock locks might destroy just granted lock if it could be merged
631 with another existing flock, this is done in completion handler,
632 so teach ldlm_cli_enquque_fini that this is a valid case for
638 Description: Oops in read and write path when failing to allocate lock.
639 Details : Check if lock allocation failed and return error back.
643 Description: lstripe command fails for valid OST index
644 Details : The stripe offset is compared to lov->desc.ld_tgt_count
645 instead of lov->desc.ld_active_tgt_count.
649 Description: sanity.sh failed test 103
650 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
655 Description: df doesn't work properly if diskfs blocksize != 4K
656 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
660 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
661 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
663 Severity : enhancement
665 Description: Adaptive timeouts
666 Details : RPC timeouts adapt to changing server load and network
667 conditions to reduce resend attempts and improve recovery time.
671 Description: llapi_file_create() does not allow some changes
672 Details : add llapi_file_open() that allows specifying the mode and
673 open flags, and also returns an open file handle.
677 Description: merge and cleanup kernel patches.
678 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
682 Description: Add journal checksum support.(Kernel part)
683 Details : The journal checksum feature adds two new flags i.e
684 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
685 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
686 indicates that the commit block contains the checksum for
687 the blocks described by the descriptor blocks. Now commit
688 record can be sent to disk without waiting for descriptor
689 blocks to be written to disk. This behavior is controlled
690 using JBD2_FEATURE_ASYNC_COMMIT flag.
694 Description: OSS needs mutliple precreate threads
695 Details : Add ability to start more than one create thread per OSS.
699 Description: Sanity.sh test_65k failed: setstripe should have succeeded
700 Details : execute lfs setstripe on client
704 Description: mds_obd_create error creating tmp object
705 Details : When the user sets quota on root, llog will be affected and can't
706 create files and write files.
709 Frequency : Always on ia64 patchless client, and possibly others.
711 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
712 Details : This allows the patchless client to be loaded on architectures
717 Description: RedHat Update kernel for RHEL5
718 Details : Add the kernel config file for RHEL5.
722 Description: RedHat Update kernel for RHEL5
723 Details : Modify the kernel config file more closer RHEL5.
727 Description: Build failure against Centos5 (RHEL5)
728 Details : Define PAGE_SIZE when it isn't present.
732 Description: test_7 on sanity-quota.sh can't be used on separate machines
733 Details : add facet to handle it
737 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
738 Details : with NFS, the anon dentry's parent was set to itself in
739 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
740 corresponding dentry other than use rec->ur_name.
742 Severity : enhancement
744 Description: lfs setstripe enhancement
745 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
749 Description: enable data checksumming by default
750 Details : enable checksum by default, allow --disable-checksum
751 configure option and "-o nochecksum" mount option. Checksums
752 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
756 Description: lustre support for RHEL5
757 Details : Add support for RHEL5.
761 Description: Client eviction due to failover config
762 Details : after a connection loss, the lustre client should attempt to
763 reconnect to the last active server first before trying the
764 other potential connections.
768 Description: when mds and osts use different quota unit(32bit and 64bit),
769 quota will be released repeatly.
770 Details : void sending multiple quota reqs to mds, which will keep the
771 status between the reqs.
775 Description: osts not allocated evenly to files
776 Details : change the condition to increase offset_idx
779 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
781 Description: Data corruption for OSTs that are formatted larger than 2TB
783 Details : When generating the bio request for lustre file writes the
784 sector number would overflow a temporary variable before being
785 used for the IO. The data reads correctly from Lustre (which
786 will overflow in a similar manner) but other file data or
787 filesystem metadata may be corrupted in some cases.
791 Description: Tune the kernel for good SCSI performance.
792 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
793 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
798 Description: rewrite ext2-derived code in obdclass/uuid.c
799 Details : rewrite inherited code (uuid parsing code from ext2 utils)
800 from scratch preserving functionality.
804 Description: "lfs find -obd UUID" prints directories
805 Details : "lfs find -obd UUID" will return all directory names instead
806 of just file names. It is incorrect because the directories
807 do not reside on the OSTs.
811 Description: use i_size_read and i_size_write in 2.6 port
812 Details : replace inode->i_size access with i_size_read/write()
815 Frequency : when removing large files
817 Description: scheduling issue during removal of large Lustre files
818 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
819 It causes scheduling issues when removing large files (17TB in the
823 Frequency : only with liblustre clients on XT3
825 Description: evictions taking too long
826 Details : allow llrd to evict clients directly on OSTs
829 Frequency : only on ppc
831 Description: /proc/fs/lustre/devices broken on ppc
832 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
833 We should make sure the type of 'index' is loff_t and then cast
834 explicitly as needed below. Do not assign an explicitly cast
838 Frequency : only for rhel5
840 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
841 Details : Modify the target file & which_kernel.
843 Severity : enhancement
845 Description: omit set fsid for export NFS
846 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
847 as FS_REQUIRES_DEV for avoid problems with generate fsid.
851 Frequency : Always, for kernels after 2.6.16
852 Description: Fix warning idr_remove called for id=.. which is not allocated.
853 Details : Last kernels save old s_dev before kill super and not allow
854 to restore from callback - restore it before call kill_anon_super.
858 Description: buffer overruns could theoretically occur
859 Details : llapi_semantic_traverse() modifies the "path" argument by
860 appending values to the end of the origin string, and a
861 overrun may occur. Adding buffer overrun check in liblustreapi.
865 Description: Fix error on 'ls .' at the top of the Lustre mount.
866 Details : Don't revalidate dentry if it is a root dentry.
870 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
871 Details : Modify vm-tunables-rhel4.patch.
875 Description: Kernel config for 2.6.18-vanilla.
876 Details : Modify targets/2.6-vanilla.target.in.
877 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
878 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
879 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
880 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
884 Description: improve handling recoverable errors
885 Details : if request processig with error which can be recoverable on server
886 request should be resend, otherwise page released from cache and
891 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
892 Details : Modify target file & which_patch.
893 A flaw was found in the IA32 system call emulation provided
894 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
895 value could be stored in the %RAX register, which could trigger an
896 out-of-bounds system call table access. An untrusted local user
897 could exploit this flaw to run code in the kernel
898 (ie a root privilege escalation). (CVE-2007-4573).
902 Description: O_DIRECT bypasses client statistics.
903 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
904 read_ahead_stats were updated. copied the stats section from
905 osc_send_oap_rpc() into async_internal().
909 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
913 Description: change order of libsysio includes
914 Details : '#include sysio.h' should always come before '#include xtio.h'
918 Description: handle "serious error: objid * already exists" more gracefully
919 Details : If LAST_ID value on disk is smaller than the objects existing in
920 the O/0/d* directories, it indicates disk corruption and causes an
921 LBUG(). If the object is 0-length, then we should use the existing
922 object. This will help to avoid a full fsck in most cases.
924 Severity : enhancement
926 Description: adapt the lustre_config script to support the upgrade case
927 Details : Add "-u" option for lustre_config script to support upgrading 1.4
928 server targets to 1.6 in parallel.
932 Description: To avoid grant space > avaible space when the disk is almost
933 full. Without this patch you might see the error "grant XXXX >
934 available" or some LBUG about grant, when the disk is almost
936 Details : In filter_check_grant, for non_grant cache write, we should
937 check the left space by if (*left > ungranted + bytes), instead
938 of (*left > ungranted), because only we are sure the left space
939 is enough for another "bytes", then the ungrant space should be
940 increase. In client, we should update cl_avail_grant only there
941 is OBD_MD_FLGRANT in the reply.
945 Description: Update RHEL 4 kernel to fix local root privilege escalation.
946 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
947 described in CVE-2007-4573. This problem could allow untrusted
948 local users to gain root access.
951 Frequency : when using O_DIRECT and quotas
953 Description: Incorrect file ownership on O_DIRECT output files
954 Details : block usage reported by 'lfs quota' does not take into account
955 files that have been written with O_DIRECT.
960 Description: touch file failed when fs is not full
961 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
962 otherwise we can get ENOSP while fs is not full.
966 Description: parallel lock callbacks
967 Details : Instead of sending blocking and completion callbacks as separated
968 requests, adding them to a set and sending in parallel.
971 Frequency : only for Cray XT3
972 Bugzilla : 12829/13455
973 Description: Changing primary group doesn't change the group lustre assigns to
975 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
976 overridden with the primary group provided by the group upcall,
977 whereas the supplied fsgid can be trusted if it is in the list of
978 supplementary groups returned by the group upcall.
980 Severity : enhancement
982 Description: Allow masking D_WARNING, D_ERROR messages from console
983 Details : Console messages can now be disabled via lnet.printk.
987 Description: User code with malformed file open parameter crashes client node
988 Details : Before packing join_file req, all the related reference should be
989 checked carefully in case some malformed flags cause fake
990 join_file req on client.
994 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
995 Details : ldlm_completion_ast() assumes that a lock is granted when the req
996 mode is equal to the granted mode. However, it should also check
997 that LDLM_FL_CP_REQD is not set.
1001 Description: Heavy nfs access might result in deadlocks
1002 Details : After ELC code landed, it is now improper to enqueue any mds
1003 locks under och_sem, because enqueue might want to decide to
1004 cancel open locks for same inode we are holding och_sem for.
1008 Description: Client eviction while running blogbench
1009 Details : A lot of unlink operations with concurrent I/O can lead to a
1010 deadlock causing evictions. To address the problem, the number of
1011 oustanding OST_DESTROY requests is now throttled to
1012 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1013 callbacks are processed in priority.
1017 Description: enable ACLs on MDS by default
1018 Details : ACLs must be enabled on MDS by default.
1021 Frequency : PPC/PPC64 only
1023 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1024 Details : fix duplicated definitions between asm-ppc64/types.h and
1025 lustre_types.h on PPC.
1028 Frequency : PPC/PPC64 only
1030 Description: asm-ppc/segment.h does not exist
1031 Details : fix compile issue on PPC.
1035 Description: better handle error messages in extents code
1038 Frequency : RHEL4 only
1040 Description: mkfs is very slow on IA64/RHEL4
1041 Details : A performance regression has been discovered in the MPT Fusion
1042 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1043 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1044 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1046 Severity : enhancement
1048 Description: SNMP support enhancement
1049 Details : Adding total number of sampled request for an MDS node in snmp
1052 Severity : enhancement
1054 Description: Optimize ldlm waiting list processing for PR extent locks
1055 Details : When processing waiting list for read extent lock and meeting read
1056 lock that is same or wider to it that is not contended, skip
1057 processing rest of the list and immediatelly return current
1058 status of conflictness, since we are guaranteed there are no
1059 conflicting locks in the rest of the list.
1063 Description: Time out and refuse to reconnect
1064 Details : When the failover node is the primary node, it is possible
1065 to have two identical connections in imp_conn_list. We must
1066 compare not conn's pointers but NIDs, otherwise we can defeat
1067 connection throttling.
1071 Description: port llog fixes from b1_6 into HEAD
1072 Details : Port llog reference couting and some llog cleanups from b1_6
1073 (bug 10800) into HEAD, for protect from panic and access to already
1074 free llog structures.
1078 Description: Detect stride IO mode in read-ahead
1079 Details : When a client does stride read, read-ahead should detect that and
1080 read-ahead pages according to the detected stride pattern.
1084 Description: data checksumming impacts single node performance
1085 Details : add support for several checksum algorithm. Currently, only CRC32
1086 and Adler-32 are supported. The checksum type can be changed on
1087 the fly via /proc/fs/lustre/osc/*/checksum_type.
1091 Description: use adler32 for page checksums
1092 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1097 Description: build for x2 fails
1098 Details : fix compile issue on Cray systems.
1102 Description: Properly match for duplicate locks
1103 Details : Due to different lock order from skiplists code, we need to
1104 traverse entire list for now
1107 Frequency : only on PPC/SLES10
1109 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1110 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1111 be defined. Add a hack in mkfs_lustre.c to work around this compile
1116 Description: LASSERT on MDS when client holding flock lock dies
1117 Details : ldlm pool logic depends on number of granted locks equal to
1118 number of released locks which is not true for flock locks, so
1119 just exclude such locks from consideration.
1123 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1124 Details : Use fsfilt_sync() for both the whole filesystem sync and
1125 individual file sync to eliminate dangerous inode locking
1126 with I_LOCK that can lead to a deadlock.
1130 Description: performance in 1.6.3
1131 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1132 sure that 1MB requests are not fragmented by the block layer.
1134 Severity : enhancement
1136 Description: organize the server-side client stats on per-nid basis
1137 Details : Change the structure of stats under obdfilter and mds to
1145 The "uuid"s file would list the uuids of _active_ exports.
1146 And the clear entry is to clear all stats and stale nids.
1148 Severity : enhancement
1150 Description: eliminate client locks in face of contention
1151 Details : file contention detection and lockless i/o implementation
1152 for contended files.
1156 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1159 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1161 Description: MDS may lose file striping (and hence file data) in some cases
1162 Details : If there are additional extended attributes stored on the MDS,
1163 in particular ACLs, SELinux, or user attributes (if user_xattr
1164 is specified for the client mount options) then there is a risk
1165 of attribute loss. Additionally, the Lustre file striping
1166 needs to be larger than default (e.g. striped over all OSTs),
1167 and an additional attribute must be stored initially in the
1168 inode and then increase in size enough to be moved to the
1169 external attribute block (e.g. ACL growing in size) for file
1174 Description: skiplist implementation simplification
1175 Details : skiplists are used to group compatible locks on granted list
1176 that was implemented as tracking first and last lock of each lock group
1177 the patch changes that to using doubly linked lists
1181 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1182 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1183 because LASSERT on both the data supplied by a client, and the data
1184 on disk is dangerous and incorrect.
1186 Severity : enhancement
1188 Description: Slow truncate/writes to huge files at high offsets.
1189 Details : Directly associate cached pages to lock that protect those pages,
1190 this allows us to quickly find what pages to write and remove
1191 once lock callback is received.
1195 Description: more ldlm soft lockups
1196 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1197 starve other threads from the resource lock for a long time in
1198 case of long waiting queue, so change the debug level from
1199 D_OTHER to the less frequently used D_INFO.
1201 Severity : enhancement
1203 Description: add -gid, -group, -uid, -user options to lfs find
1207 Description: Hung threads in invalidate_inode_pages2_range
1208 Details : The direct IO path doesn't call check_rpcs to submit a new RPC once
1209 one is completed. As a result, some RPCs are stuck in the queue
1214 Description: filter threads hungs on waiting journal commit
1215 Details : Cleanup filter group llog code, then only filter group llog will
1216 be only created in the MDS/OST syncing process.
1220 Description: Procfs and llog threads access destoryed import sometimes.
1221 Details : Sync the import destoryed process with procfs and llog threads by
1222 the import refcount and semaphore.
1224 Severity : enhancement
1226 Description: openlock cache of b1_6 port to HEAD
1231 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1232 Details : If insertion of an extent fails, then discard the inode
1233 preallocation and free data blocks else it can lead to duplicate
1238 Description: don't always update ctime in ext3_xattr_set_handle()
1239 Details : Current xattr code updates the inode ctime in ext3_xattr_set_handle.
1240 In some cases the ctime should not be updated, for example for
1241 2.0->1.8 compatibility it is necessary to delete an xattr and it
1242 should not update the ctime.
1246 Bugzilla : 15713/16362
1247 Description: Assertion in iopen_connect_dentry in 1.6.3
1248 Details : looking up an inode via iopen with the wrong generation number can
1249 populate the dcache with a disconneced dentry while the inode
1250 number is in the process of being reallocated. This causes an
1251 assertion failure in iopen since the inode's dentry list contains
1252 both a connected and disconnected dentry.
1256 Description: assertion failure in ldlm_handle2lock()
1257 Details : fix a race between class_handle_unhash() and class_handle2object()
1258 introduced in lustre 1.6.5 by bug 13622.
1263 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1264 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1269 Description: oops in sd_iostats_seq_show()
1270 Details : unloading/reloading the scsi low level driver triggers a kernel
1271 bug when trying to access the sd iostat file.
1276 Description: Kernel panics during QLogic driver reload
1277 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1278 patch, causing memory corruption.
1283 Description: journal_dev option does not work in b1_6
1284 Details : pass mount option during pre-mount.
1286 --------------------------------------------------------------------------------
1288 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
1290 * Support for kernels:
1291 2.6.5-7.283 (SLES 9),
1292 2.6.9-55.EL (RHEL 4),
1293 2.6.16.46-0.14 (SLES 10),
1294 2.6.18.8 vanilla (kernel.org)
1295 * Client support for unpatched kernels:
1296 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1297 2.6.16 - 2.6.22 vanilla (kernel.org)
1298 * Due to recently discovered recovery problems, we do not recommend
1299 using patchless RHEL 4 clients with this or any earlier release.
1300 * Recommended e2fsprogs version: 1.39.cfs8
1301 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1302 * Starting with this release, the ldiskfs backing filesystem required
1303 by Lustre is now in its own package, lustre-ldiskfs. This package
1304 should be installed. It is versioned separately from Lustre and
1305 may be released separately in future.
1309 Description: block reactivating mgc import until all deactivates complete
1310 Details : Fix race when failing back MDT/MGS to itself (testing)
1313 Frequency : at statup only
1315 Description: mds_lov_synchronize race leads to various problems
1316 Details : simultaneous MDT->OST connections at startup can cause the
1317 sync to abort, leaving the OSC in a bad state.
1319 Severity : enhancement
1321 Description: add optional extra BUILD_VERSION info
1322 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1323 to override the lustre version.
1326 Frequency : 2.6.18 servers only
1328 Description: ll_kern_mount() doesn't release the module reference
1329 Details : The ldiskfs module reference count never drops down to 0
1330 because ll_kern_mount() doesn't release the module reference.
1335 Description: server LBUG when using old ost_num_threads parameter
1336 Details : Accept the old ost_num_threads parameter but warn that it
1337 is deprecated, and fix an off-by-one error that caused an LBUG.
1342 Description: Transient SCSI error results in persistent IO issue
1343 Details : iobuf->dr_error is not reinitialized to 0 between two
1347 Frequency : sometimes when underlying device returns I/O errors
1349 Description: OSTs not going read-only during write failures
1350 Details : OSTs are not remounted read-only when the journal commit threads
1351 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1352 instead of the ext3 wrappers.
1356 Description: poor connect scaling with increasing client count
1357 Details : Don't run filter_grant_sanity_check for more than 100 exports
1358 to improve scaling for large numbers of clients.
1361 Frequency : SLES10 only
1363 Description: sanity-quota.sh quotacheck failed: rc = -22
1364 Details : Quotas cannot be enabled on SLES10.
1367 Frequency : liblustre clients only
1369 Description: getdirentries does not give error when run on compute nodes
1370 Details : getdirentries does not fail when the size specified as an argument
1371 is too small to contain at least one entry
1373 Severity : enhancement
1375 Description: Add LNET router traceability for debug purposes
1376 Details : If a checksum failure occurs with a router as part of the
1377 IO path, the NID of the last router that forwarded the bulk data
1378 is printed so it can be identified.
1383 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1384 Details : Due to a race condition, liblustre clients were occasionally
1385 evicted incorrectly.
1387 Severity : enhancement
1389 Description: lfs setstripe use optional parameters instead of postional
1392 Severity : enhancement
1394 Description: Nanosecond timestamp support for ldiskfs
1395 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1396 resolution timestamps. There is not yet support for this at
1397 the Lustre filesystem level.
1400 Frequency : during server recovery
1402 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1403 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1407 Frequency : nfs export on patchless client
1409 Description: connectathon hang when test nfs export over patchless client
1410 Details : Disconnected dentry cannot be found with lookup, so we do not need
1411 to unhash it or make it invalid
1414 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1415 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1416 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1421 Description: open req refcounting wrong on reconnect
1422 Details : If reconnect happened between getting open reply from server and
1423 call to mdc_set_replay_data in ll_file_open, we will schedule
1424 replay for unreferenced request that we are about to free.
1425 Subsequent close will crash in variety of ways.
1426 Check that request is still eligible for replay in
1427 mdc_set_replay_data().
1432 Description: disable writes to filesystem when reading health_check file
1433 Details : the default for reading the health_check proc file has changed
1434 to NOT do a journal transaction and write to disk, because this
1435 can cause reads of the /proc file to hang and block HA state
1436 checking on a healthy but otherwise heavily loaded system. It
1437 is possible to return to the previous behaviour during configure
1438 with --enable-health-write.
1440 Severity : enhancement
1442 Description: 64-bit inode version
1443 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1444 to the inode. This will be required for version-based recovery.
1449 Description: MDS fails to start if a duplicate client export is detected
1450 Details : in some rare cases it was possible for a client to connect to
1451 an MDS multiple times. Upon recovery the MDS would detect this
1452 and fail during startup. Handle this more gracefully.
1454 Severity : enhancement
1456 Description: Add -o localflock option to simulate old noflock
1458 Details : This will achieve local-only flock/fcntl locks
1464 Description: log_commit_thread vs filter_destroy race leads to crash
1465 Details : Take import reference before releasing llog record semaphore
1470 Description: Wrong request locking in request set processing
1471 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1472 imp_delayed_list, in this place should be used imp_lock.
1475 Frequency : when reconnection
1477 Description: Grant Leak when osc reconnect to OST
1478 Details : When osc reconnect ost, OST(filter) should check whether it
1479 should grant more space to client by comparing fed_grant and
1480 cl_avail_grant, and return the granted space to client instead
1481 of "new granted" space, because client will call osc_init_grant
1482 to update the client grant space info.
1485 Frequency : when client reconnect to OST
1487 Description: Grant Leak when osc do resend and replay bulk write
1488 Details : When osc reconnect to OST, OST(filter)should clear grant info of
1489 bulk write request, because the grant info while be sync between
1490 OSC and OST when reconnect, and we should ignore the grant info
1491 these of resend/replay write req.
1496 Description: Grant space more than avaiable left space sometimes.
1497 Details : When then OST is about to be full, if two bulk writing from
1498 different clients came to OST. Accord the avaliable space of the
1499 OST, the first req should be permitted, and the second one
1500 should be denied by ENOSPC. But if the seconde arrived before
1501 the first one is commited. The OST might wrongly permit second
1502 writing, which will cause grant space > avaiable space.
1505 Frequency : when client is evicted
1507 Description: Grant might be wrongly erased when osc is evicted by OST
1508 Details : when the import is evicted by server, it will fork another
1509 thread ptlrpc_invalidate_import_thread to invalidate the
1510 import, where the grant will be set to 0. While the original
1511 thread will update the grant it got when connecting. So if
1512 the former happened latter, the grant will be wrongly errased
1513 because of this race.
1518 Description: Checking Stale with correct fid
1519 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1520 to check whether it is stale, because sometimes, we want the
1521 enqueue happened anyway, and op_data.fid2 will not be initialized.
1523 Severity : enhancement
1525 Description: update patchless client
1526 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1529 Frequency : only with 2.4 kernel
1531 Description: random memory corruption
1532 Details : size of struct ll_inode_info is to big for union inode.u and this
1533 can be cause of random memory corruption.
1538 Description: Memory leak in recovery
1539 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1540 It should also check obd_fail before fsfilt_start, otherwise if
1541 fsfilt_start return -EROFS,(failover mds during mds recovery).
1542 then the req will return with repmsg->transno = 0 and rc = EROFS.
1543 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1544 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1545 be freed no matter whether fsfilt_commit success or not.
1548 Frequency : only with huge count clients
1550 Description: Prevents from taking the superblock lock in llap_from_page for
1552 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1553 for llap_from_page call in ll_removepage prevents from taking the
1554 superblock lock for a soon died page.
1559 Description: Not check open intent error before release open handle
1560 Details : in some rare cases, the open intent error is not checked before
1561 release open handle, which may cause
1562 ASSERTION(open_req->rq_transno != 0), because it tries to release
1563 the failed open handle.
1568 Description: Set cat log bitmap only after create log success.
1569 Details : in some rare cases, the cat log bitmap is set too early. and it
1570 should be set only after create log success.
1574 Description: Accessing a block bevice can re-enable I/O when Lustre is
1575 tearing down a device.
1576 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1581 Description: service threads may hog cpus when there are a lot of requests
1582 Details : Insert cond_resched to give other threads a chance to use some CPU
1587 Description: the cat log was not initialized in recovery
1588 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1589 unlink log on mds will not be initialized until mds post
1590 recovery. And also in mds post recovery, the unlink log will
1591 initialization will be done asynchronausly, so there will be race
1592 between add unlink log and unlink log initialization.
1596 Description: brw_stats were being printed incorrectly
1597 Details : brw_stats were being printed as log2 but all of them were not
1598 recorded as log2. Also remove some code duplication arising from
1599 filter_tally_{read,write}.
1603 Frequency : rare, only in recovery.
1604 Description: ASSERTION(req->rq_type != LI_POISON) failed
1605 Details : imp_lock should be held while iterating over imp_sending_list for
1606 prevent destroy request after get timeout in ptlrpc_queue_wait.
1610 Description: replay-single.sh test 52 fails
1611 Details : A lock's skiplist need to be cleanup when it being unlinked
1612 from its resource list.
1616 Description: Short directio read returns full requested size rather than
1618 Details : Direct I/O operations should return actual amount of bytes
1619 transferred rather than requested size.
1621 Severity : enhancement
1623 Description: metadata RPC reduction (e.g. for rm performance)
1624 Details : decrease the amount of synchronous RPC between clients and servers
1625 by canceling conflicing lock before the operation on the client
1626 and packing thier handles into the main operation RPC to server.
1628 Severity : enhancement
1630 Description: Async OSC create to avoid the blocking unnecessarily.
1631 Details : If a OST has no remain object, system will block on the creating
1632 when need to create a new object on this OST. Now, ways use
1633 pre-created objects when available, instead of blocking on an
1634 empty osc while others are not empty. If we must block, we block
1635 for the shortest possible period of time.
1637 Severity : enhancement
1639 Description: refine locking for avoid write wrong info into lov_objid file
1640 Details : fix possible races with add new target and write/update data in
1643 --------------------------------------------------------------------------------
1645 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1650 Frequency : on some architectures
1652 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1653 Details : Uninitialized flags sometimes cause configuration commands to
1657 Frequency : patchless clients only
1659 Description: missing __iget() symbol export
1660 Details : The __iget() symbol export is missing. To avoid the need for
1661 this on patchless clients the deathrow inode reaper is turned
1662 off, and we depend on the VM to clean up old inodes. This
1663 dependency was during via the fix for bug 12181.
1665 --------------------------------------------------------------------------------
1667 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1669 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1670 INTEROPERATE with older versions automatically. In many cases a
1671 special upgrade step is needed. Please read the
1672 user documentation before upgrading any part of a live system.
1673 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1674 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1675 * WARNING: Lustre configuration and startup changes are required with
1676 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1682 Severity : enhancement
1684 Description: MountConf
1685 Details : Lustre configuration is now managed via mkfs and mount
1686 commands instead of lmc and lconf. New obd types (MGS, MGC)
1687 are added for dynamic configuration management. See
1688 https://mail.clusterfs.com/wikis/lustre/MountConf for
1691 Severity : enhancement
1693 Description: dynamic OST addition
1694 Details : OSTs can now be added to a live filesystem
1696 Severity : enhancement
1698 Description: startup order invariance
1699 Details : MDTs and OSTs can be started in any order. Clients only
1700 require the MDT to complete startup.
1702 Severity : enhancement
1704 Description: parallel, asynchronous orphan cleanup
1705 Details : orphan cleanup is now performed in separate threads for each
1706 OST, allowing parallel non-blocking operation.
1708 Severity : enhancement
1710 Description: optimized stripe assignment
1711 Details : stripe assignments are now made based on ost space available,
1712 ost previous usage, and OSS previous usage, in order to try
1713 to optimize storage space and networking resources.
1715 Severity : enhancement
1717 Description: Permanently set tunables
1718 Details : All writable /proc/fs/lustre tunables can now be permanently
1719 set on a per-server basis, at mkfs time or on a live system.
1721 Severity : enhancement
1723 Description: Lustre message v2
1724 Details : Add lustre message format v2.
1726 Severity : enhancement
1728 Description: client OST exclusion list
1729 Details : Clients can be started with a list of OSTs that should be
1730 declared "inactive" for known non-responsive OSTs.
1734 Description: SPEC SFS validation failure on NFS v2 over lustre.
1735 Details : Changes the blocksize for regular files to be 2x RPC size,
1736 and not depend on stripe size.
1738 Severity : enhancement
1740 Description: Multiple MD RPCs in flight.
1741 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1742 To avoid overly-overloading MDS, introduce a limit on number of
1743 MDS RPCs in flight for a single client and add /proc controls
1744 to adjust this limit.
1746 Severity : enhancement
1748 Description: client read/write statistics
1749 Details : Add client read/write call usage stats for performance
1750 analysis of user processes.
1751 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1752 file access. extents_stats shows chunk size distribution.
1753 extents_stats_per_process show chunk size distribution per
1756 Severity : enhancement
1758 Description: mds statistics
1759 Details : Add detailed mds operations statistics in
1760 /proc/fs/lustre/mds/*/stats.
1764 Description: Failure of copying files with lustre special EAs.
1765 Details : Client side always return success for setxattr call for lustre
1766 special xattr (currently only "trusted.lov").
1771 Description: Refcount LNET uuids
1772 Details : The global LNET uuid list grew linearly with every startup;
1773 refcount repeated list entries instead of always adding to
1776 Severity : enhancement
1778 Description: Dynamic service threads
1779 Details : Within a small range, start extra service threads
1780 automatically when the request queue builds up.
1783 Frequency : mixed-endian client/server environments
1785 Description: mixed-endian crashes
1786 Details : The new msg_v2 system had some failures in mixed-endian
1789 Severity : enhancement
1791 Description: Easy OST removal
1792 Details : OSTs can be permanently deactivated with e.g. 'lctl
1793 conf_param lustre-OST0001.osc.active=0'
1795 Severity : enhancement
1797 Description: MGS proc entries
1798 Details : Added basic proc entries for the MGS showing what filesystems
1801 Severity : enhancement
1803 Description: provide MGS failover
1804 Details : Added config lock reacquisition after MGS server failover.
1806 Severity : enhancement
1808 Description: add Linux 2.4 support
1809 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1813 Description: a large application tries to do I/O to the same resource and dies
1814 in the middle of it.
1815 Details : Check the req->rq_arrival time after the call to
1816 ost_brw_lock_get(), but before we do anything about
1817 processing it & sending the BULK transfer request. This
1818 should help move old stale pending locks off the queue as
1819 quickly as obd_timeout.
1822 Frequency : when an incorrect nid is specified during startup
1824 Description: ptlrpc connect to non-existant node causes kernel crash
1825 Details : LNET can't be re-entered from an event callback, which
1826 happened when we expire a message after the export has been
1827 cleaned up. Instead, hand the zombie cleanup off to another
1830 Severity : enhancement
1832 Description: plain/inodebits lock performance improvement
1833 Details : Grouping plain/inodebits in granted list by their request modes
1834 and bits policy, thus improving the performance of search through
1838 Frequency : only if OST filesystem is corrupted
1840 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1841 Details : for a short time RPCs with bulk IO are in the replay list,
1842 but replay of bulk IOs is unimplemented. If the OST filesystem
1843 is corrupted due to disk cache incoherency and then replay is
1844 started it is possible to trip an assertion. Avoid putting
1845 committed RPCs into the replay list at all to avoid this issue.
1848 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1850 Description: System hang on startup
1851 Details : This bug allowed the liblustre (e.g. catamount) client to
1852 return to the app before handling all startup RPCs. This
1853 could leave the node unresponsive to lustre network traffic
1854 and manifested as a server ptllnd timeout.
1856 Severity : enhancement
1858 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1859 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1860 boolean to control whether to print peer debug info when a
1861 client's RPC times out.
1864 Frequency : only for kernels with patches from Lustre below 1.4.3
1866 Description: Remove old rdonly API
1867 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1870 Frequency : only for devices with external journals
1872 Description: Set external device read-only also
1873 Details : During a commanded failover stop, we set the disk device
1874 read-only while the server shuts down. We now also set any
1875 external journal device read-only at the same time.
1878 Frequency : when upgrading from 1.4 while trying to change parameters
1880 Description: The wrong (new) MDC name was used when setting parameters for
1881 upgraded MDT's. Also allows changing of OSC (and MDC)
1882 parameters if --writeconf is specified at tunefs upgrade time.
1885 Frequency : when setting specific ost indicies
1887 Description: QOS code breaks on skipped indicies
1888 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1889 created with --index need not be sequential.
1891 Severity : enhancement
1893 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1894 Details : The uninit_groups feature works in conjunction with the kernel
1895 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1896 up the pass1 processing of e2fsck. This is a read-only feature
1897 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1898 mount filesystems that have had this feature enabled.
1900 Severity : enhancement
1902 Description: Improve multi-block allocation algorithm to avoid fragmentation
1903 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1904 allocation locality and avoid filesystem fragmentation.
1906 ------------------------------------------------------------------------------
1908 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1910 * Support for kernels:
1911 2.4.21-47.0.1.EL (RHEL 3)
1912 2.6.5-7.283 (SLES 9)
1913 2.6.9-42.0.10.EL (RHEL 4)
1914 2.6.12.6 vanilla (kernel.org)
1915 2.6.16.27-0.9 (SLES 10)
1916 * Recommended e2fsprogs version: 1.39.cfs5
1918 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1922 Frequency : occasional, depends on client load and configuration
1923 Bugzilla : 12181, 12203
1924 Description: data loss for recently-modified files
1926 Details : In some cases it is possible that recently written or created
1927 files may not be written to disk in a timely manner (this should
1928 normally be within 30s unless client IO load is very high).
1929 The problem appears as zero-length files or files that are a
1930 multiple of 1MB in size after a client crash or client eviction
1931 that are missing data at the end of the file.
1933 This problem is more likely to be hit on clients where files are
1934 repeatedly created and unlinked in the same directory, clients
1935 have a large amount of RAM, have many CPUs, the filesystem has
1936 many OSTs, the clients are rebooted frequently, and/or the files
1937 are not accessed by other nodes after being written.
1939 The presence of the problem can be detected by looking at
1940 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1941 smaller than the second (nr_unused) then dirty files will not
1942 be flushed automatically to disk. "sync; sleep 10" should be
1943 run several times on the node before unmounting it to update
1944 Lustre (this is also safe to run on nodes without this problem).
1946 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1947 that can cause this same problem, so customers using that kernel
1948 also need to update the kernel in addition to Lustre. In order
1949 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1951 It is normal that files written just before a client crash (less
1952 than 30s) may not yet have been flushed to disk, even for local
1956 Frequency : frequent on thin XT3 nodes
1958 Description: UUID collision on thin XT3 Linux nodes
1959 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1960 randomly, since we relied on an insufficiently-seeded PRNG.
1965 Description: OSS hangs after "All ost request buffers busy"
1966 Details : A deadlock between quota and journal operations caused OSS
1967 hangs after printing "All ost request buffers busy."
1970 Frequency : always on liblustre builds
1972 Description: Cleanup compiler warnings on liblustre
1975 Frequency : always on liblustre builds on XT3
1977 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1981 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1982 > 32000 subdirectories
1983 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1984 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1985 subdirectory count crosses 32000. This will aid e2fsck to
1986 correctly handle more than 32000 subdirectories.
1989 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1991 Description: System hang on startup
1992 Details : This bug allowed the liblustre (e.g. catamount) client to
1993 return to the app before handling all startup RPCs. This
1994 could leave the node unresponsive to lustre network traffic
1995 and manifested as a server ptllnd timeout.
1997 Severity : enhancement
1999 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2000 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2001 boolean to control whether to print peer debug info when a
2002 client's RPC times out.
2007 Description: make O_SYNC working on 2.6 kernels
2008 Details : 2.6 kernels use different method for mark pages for write,
2009 so need add a code to lustre for O_SYNC work.
2014 Description: Failure to close file and release space on NFS
2015 Details : Put inode details into lock acquired in ll_intent_file_open.
2016 Use mdc_intent_lock in ll_intent_open to properly
2017 detect all kind of errors unhandled by mdc_enqueue.
2022 Description: proc file read during shutdown sometimes raced obd removal,
2024 Details : Add lock to prevent obd access after proc file removal.
2027 Frequency : Only for files larger than 4GB on 32-bit clients.
2029 Description: improperly doing page alignment of locks
2030 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2031 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2036 Description: RPCs being resent when they shouldn't be
2037 Details : Some RPCs that should not be resent are being resent. This
2038 can cause inconsistencies in the RPC state machine. Do not
2039 resend such requests.
2042 Frequency : rare, only with NFS export
2044 Description: Crash on NFS re-export node
2045 Details : under very unusual load conditions an assertion is hit in
2046 ll_intent_file_open()
2049 Frequency : only if OST filesystem is corrupted
2051 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2052 Details : for a short time RPCs with bulk IO are in the replay list,
2053 but replay of bulk IOs is unimplemented. If the OST filesystem
2054 is corrupted due to disk cache incoherency and then replay is
2055 started it is possible to trip an assertion. Avoid putting
2056 committed RPCs into the replay list at all to avoid this issue.
2061 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2062 Details : Large single O_DIRECT read and write calls can fail to allocate
2063 a sufficiently large buffer to process the request. In case of
2064 allocation failure the allocation is retried with a smaller
2065 buffer and broken into smaller requests.
2067 Severity : enhancement
2069 Description: Add -o localflock option to simulate old noflock behaviour.
2070 Details : This will achieve local-only flock/fcntl locks coherentness.
2075 Description: versioning check is incomplete
2076 Details : Checking the version difference of client vs. server, report
2077 error if the gap is too big.
2082 Description: add support PG_writeback bit
2083 Details : add support for PG_writeback bit for Lustre, for more carefull
2084 work with page cache in 2.6 kernel. This also fix some deadlocks
2085 and remove hack for work O_SYNC with 2.6 kernel.
2087 Severity : enhancement
2089 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2090 Details : The uninit_groups feature works in conjunction with the kernel
2091 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2092 up the pass1 processing of e2fsck. This is a read-only feature
2093 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2094 mount filesystems that have had this feature enabled.
2096 Severity : enhancement
2098 Description: Improve multi-block allocation algorithm to avoid fragmentation
2099 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2100 allocation locality and avoid filesystem fragmentation.
2102 ------------------------------------------------------------------------------
2104 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2106 * Support for kernels:
2107 2.6.9-42.0.3.EL (RHEL 4)
2108 2.6.5-7.276 (SLES 9)
2109 2.4.21-47.0.1.EL (RHEL 3)
2110 2.6.12.6 vanilla (kernel.org)
2111 2.6.16.21-0.8 (SLES10)
2112 * Recommended e2fsprogs version: 1.39.cfs2-0
2114 * The backwards-compatible /proc/sys/portals symlink has been removed
2115 in this release. Before upgrading, please ensure that you change
2116 any configuration scripts or /etc/sysctl.conf files that access
2117 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2118 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2119 in advance of the upgrade on any system running Lustre 1.4.6 or
2120 newer, since /proc/sys/lnet was added in that version.
2121 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2125 Frequency : only when quota is used
2127 Description: avoid scanning export list for quota master
2128 Details : Change the algorithms to avoid scanning export list in order
2129 to improve the efficiency.
2132 Frequency : MDS failover only, very rarely
2134 Description: "went back in time" messages on mds failover
2135 Details : The greatest transno may be lost when the current operation
2136 finishes with an error (transno==0) and the client's last_rcvd
2137 record is over-written. Save the greatest transno in the
2138 mds_last_transno for this case.
2141 Frequency : always for specific kernels and striping counts
2143 Description: client may get "Matching packet too big" without ACL support
2144 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2145 when trying to access files in certain configurations. The
2146 clients should in fact be denied when mounting because they do
2147 not understand ACLs.
2150 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2152 Description: many clients connecting with IO in progress causes connect timeouts
2153 Details : Avoid synchronous journal commits to avoid delays caused by many
2154 clients connecting/disconnecting when bulk IO is in progress.
2155 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2156 OST_IO_PORTAL to avoid delays behind potentially many pending
2160 Frequency : occasionally with multiple writers to a single file
2162 Description: shared writes to file may result in wrong size reported by stat()
2163 Details : Allow growing of kms when extent lock is cancelled
2166 Frequency : always with random mmap IO to multi-striped file
2168 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2169 Details : Only if the hole is at the end of OST object so that kms is too
2170 small. Fix is to increase kms accordingly in ll_nopage.
2173 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2175 Description: writes to a missing object would leak memory on the OST
2176 Details : If there is an inconsistency between the MDS and OST filesystems,
2177 such that the MDS references an object that doesn't exist, writes
2178 to that object will leak memory due to incorrect cleanup in the
2179 error handling path, eventually running out of memory on the OST.
2184 Description: Creating too long symlink causes lustre errors
2185 Details : Check symlink and name lengths before sending requests to MDS.
2188 Frequency : only if flock is enabled (not on by default)
2190 Description: posix locks not released on fd closure on 2.6.9+
2191 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2192 kernels, this caused such locks not to be released on fd close and
2193 then assertions on fs unmount about still used locks.
2196 Frequency : MDS failover only, very rarely
2198 Description: clients may get ASSERTION(granted_lock != NULL)
2199 Details : When request was taking a long time, and a client was resending
2200 a getattr by name lock request. The were multiple lock requests
2201 with the same client lock handle and
2202 mds_getattr_name->fixup_handle_for_resent_request found one of the
2203 lock handles but later failed with ASSERTION(granted_lock != NULL).
2208 Description: handle->h_buffer_credits > 0, assertion failure
2209 Details : h_buffer_credits is zero after truncate, causing assertion
2210 failure. This patch extends the transaction or creates a new
2214 Frequency : NFS re-export or patchless client
2215 Bugzilla : 11179, 10796
2216 Description: Crash on NFS re-export node (__d_move)
2217 Details : We do not want to hash the dentry if we don't have a lock.
2218 But if this dentry is later used in d_move, we'd hit uninitialised
2219 list head d_hash, so we just do this to init d_hash field but
2220 leave dentry unhashed.
2223 Frequency : NFS re-export or patchless client
2225 Description: NFS exports has problem with symbolic link
2226 Details : lustre client didn't properly install dentry when re-exported
2227 to NFS or running patchless client.
2230 Frequency : NFS re-export or patchless client
2232 Description: Various nfs/patchless fixes.
2233 Details : fixes reuse disconected alias for lookup process - this fixes
2234 warning "find_exported_dentry: npd != pd",
2235 fix permission error with open files at nfs.
2236 fix apply umask when do revalidate.
2239 Frequency : occasional
2241 Description: Crash on NFS re-export node
2242 Details : calling clear_page() on the wrong pointer triggered oops in
2243 generic_mapping_read().
2246 Frequency : rarely, using O_DIRECT IO
2248 Description: unaligned directio crashes client with LASSERT
2249 Details : check for unaligned buffers before trying any requests.
2252 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2254 Description: stale data returned from RAID cache
2255 Details : If only a small amount of IO is done to the RAID device before
2256 reading it again it is possible to get stale data from the RAID
2257 cache instead of reading it from disk.
2260 Frequency : always for sles10 kernel
2262 Description: sles10 support
2263 Details : ll_follow_link: compile fixes and using of nd_set_link
2264 under newer kernels.
2267 Frequency : depends on arch, kernel and compiler version, always on sles10
2270 Description: recursive or deep enough symlinks cause stack overflow
2271 Details : getting rid of large stack-allocated variable in
2275 Frequency : depends on hardware
2277 Description: lustre write performance loss in the SLES10 kernel
2278 Details : the performance loss is caused by using of write barriers in the
2279 ext3 code. The SLES10 kernel turns barrier support on by
2280 default. The fix is to undo that change for ldiskfs.
2282 ------------------------------------------------------------------------------
2284 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2286 * Support for kernels:
2287 2.6.9-42.0.3EL (RHEL 4)
2288 2.6.5-7.276 (SLES 9)
2289 2.4.21-47.0.1.EL (RHEL 3)
2290 2.6.12.6 vanilla (kernel.org)
2294 Frequency : quota enabled and large files being deleted
2296 Description: releasing more than 4GB of quota at once hangs OST
2297 Details : If a user deletes more than 4GB of files on a single OST it
2298 will cause the OST to spin in an infinite loop. Release
2299 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2304 Description: statfs data retrieved from /proc may be stale or zero
2305 Details : When reading per-device statfs data from /proc, in the
2306 {kbytes,files}_{total,free,avail} files, it may appear
2307 as zero or be out of date.
2310 Frequency : systems with MD RAID1 external journal devices
2312 Description: lconf's call to blkid is confused by RAID1 journal devices
2313 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2314 of returning all block devices that match the journal UUID.
2317 Frequency : always, for aggregate stripe size over 4GB
2319 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2320 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2321 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2324 Frequency : always if "lfs find" used on a local file/directory
2326 Description: "lfs find" segfaults if used on a local file/directory
2327 Details : The case where a directory component was not specified wasn't
2328 handled correctly. Handle this properly.
2331 Frequency : always on ppc64
2333 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2334 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2335 perform endianness conversion.
2338 Frequency : rarely (truncate to non-zero file size after write under load)
2339 Bugzilla : 10730, 10687
2340 Description: Files padded with zeros to next 4K multiple
2341 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2342 it is possible that files that are truncated to a non-zero size
2343 immediately after being written are filled with zero bytes beyond
2344 the truncated size. No file data is lost.
2346 Severity : enhancement
2348 Description: Allow recovery/failover for liblustre clients.
2349 Details : liblustre clients were unaware of failover configurations until
2352 Severity : enhancement
2354 Description: user file locks should fail when not mounting with flock option
2355 Details : Set up an error-returning stub in ll_file_operations.lock field
2356 to prevent incorrect behaviour when client is mounted without
2357 flock option. Also, set up properly f_op->flock field for
2361 Frequency : always on ia64
2363 Description: "lfs df" loops on printing out MDS statfs information
2364 Details : The obd_ioctl_data was not initialized and in some systems
2365 this caused a failure during the ioctl that did not return
2366 an error. Initialize the struct and return an error on failure.
2369 Frequency : SLES 9 only
2371 Description: Error of copying files with lustre special EAs as root
2372 Details : Client side always return success for setxattr call for lustre
2373 special xattr (currently only "trusted.lov").
2376 Frequency : rarely on clusters with both ia64+i386 clients
2378 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2379 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2380 clients are doing concurrent writes to the same file it is
2381 possible that the ia64 clients may LASSERT because the OST
2382 extent locks are not PAGE_SIZE aligned. Ensure that grown
2383 locks are always aligned on the request boundary.
2386 Frequency : specific use, occasional
2388 Description: Overwriting in use executable truncates on-disk binary image
2389 Details : If one node attempts to overwrite an executable in use by
2390 another node, we now correctly return ETXTBSY instead of
2391 truncating the file.
2393 Severity : enhancement
2395 Description: Async OSC create to avoid the blocking unnecessarily.
2396 Details : If a OST has no remain object, system will block on the creating
2397 when need to create a new object on this OST. Now, ways use
2398 pre-created objects when available, instead of blocking on an
2399 empty osc while others are not empty. If we must block, we block
2400 for the shortest possible period of time.
2405 Description: chmod on Lustre root is propagated to other clients
2406 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2407 invalid by the follow_mount time.
2410 Frequency : liblustre clients only
2412 Description: Race in 'instant cancel' lock handling could lead to such locks
2413 never to be granted in case of SMP MDS
2414 Details : Do not destroy not yet granted but cbpending locks in
2418 Frequency : replay/resend of open
2420 Description: non null lock assetion failure in mds_intent_policy
2421 Details : Trying to replay/resend lockless open requests resulted in
2422 mds_open() returning 0 with no lock. Now it sets a flag if
2423 a lock is going to be returned.
2425 Severity : enhancement
2427 Description: Checksum enhancements
2428 Details : New checksum enhancements allow for resending RPCs that failed
2431 Severity : enhancement
2433 Description: Tunables on number of dirty pages in cacche
2434 Details : Allow to set limit on number of dirty pages cached.
2439 Description: client crash on unmount - lock still has references
2440 Details : In some error handling cases it was possible to leak a lock
2441 reference on a client while accessing a file. This was not
2442 harmful to the client during operation, but would cause the
2443 client to crash when the filesystem is unmounted.
2446 Frequency : specific case, rare
2448 Description: ETXTBSY on mds though file not in use by client
2449 Details : ETXTBSY is no longer incorrectly returned when attempting to
2450 chmod or chown a directory that the user previously tried to
2451 execute or a currently-executing binary.
2454 Frequency : extremely rare except on liblustre-based clients
2456 Description: Lustre space not freed when files are deleted
2457 Details : Clean up open-unlinked files after client eviction. Previously
2458 the unlink was skipped and the files remained as orphans.
2463 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2464 Details : In some cases it was possible to send a blocking callback to
2465 a client doing a glimpse, even though that client didn't get
2466 a lock granted. When the glimpse lock is cancelled on the OST
2467 the freed lock is left on the waiting list and corrupted the list.
2470 Frequency : all core dumps
2472 Description: Broke core dumps to lustre
2473 Details : Negative dentry may be unhashed if parent does not have UPDATE
2474 lock, but some callers, e.g. do_coredump, expect dentry to be
2475 hashed after successful create, hash it in ll_create_it.
2477 ------------------------------------------------------------------------------
2479 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2481 * Support for kernels:
2482 2.6.9-42.0.2.EL (RHEL 4)
2483 2.6.5-7.276 (SLES 9)
2484 2.4.21-40.EL (RHEL 3)
2485 2.6.12.6 vanilla (kernel.org)
2489 Frequency : always on RHEL 3
2491 Description: Number of open files grows over time
2492 Details : The number of open files grows over time, whether or not
2493 Lustre is started. This was due to a filp leak introduced
2494 by one of our kernel patches.
2496 ------------------------------------------------------------------------------
2498 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2500 * Support for kernels:
2501 2.6.9-42.EL (RHEL 4)
2502 2.6.5-7.276 (SLES 9)
2503 2.4.21-40.EL (RHEL 3)
2504 2.6.12.6 vanilla (kernel.org)
2509 Bugzilla : 5719, 9635, 9792, 9684
2510 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2511 Details : If a server is under heavy load and cannot reply to new
2512 connection requests before the client resends the (re)connect,
2513 the connection handling code can behave badly if two service
2514 threads are concurrently handing separate (re)connections from
2515 the same client. Add better locking to the connection handling
2516 code, and ensure that only a single connection will be processed
2517 for a given client UUID, even if the lock is dropped.
2519 Severity : enhancement
2521 Description: add TCP zero-copy support to kernel
2522 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2523 sends if the hardware supports scatter-gather and checksumming.
2524 This allows socklnd to do client-write and server-read more
2525 efficiently and reduce CPU utilization from skbuf copying.
2528 Frequency : only if NFS exporting from client
2530 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2531 Details : If setting attributes on a file created under NFS that had
2532 never been opened it would be possible to oops the client
2533 if the file had no objects.
2536 Frequency : always for liblustre
2538 Description: liblustre client does MDS+OSTs setattr RPC for each write
2539 Details : When doing a write from a liblustre client, the client
2540 incorrectly issued an RPC to the MDS and each OST the file was
2541 striped over in order to update the timestamps. When writing
2542 with small chunks and many clients this could overwhelm the MDS
2543 with RPCs. In all cases it would slow down the write because
2544 these RPCs are unnecessary.
2546 Severity : enhancement
2548 Description: allow number of MDS service threads to be changed at module load
2549 Details : It is now possible to change the number of MDS service threads
2550 running. Adding "options mds mds_num_threads={N}" to the MDS's
2551 /etc/modprobe.conf will set the number of threads for the next
2552 time Lustre is restarted (assuming the "mds" module is also
2553 reloaded at that time). The default number of threads will
2554 stay the same, 32 for most systems.
2559 Description: OST crash if filesystem is unformatted or corrupt
2560 Details : If an OST is started on a device that has never been formatted
2561 or if the filesystem is corrupt and cannot even mount then the
2562 error handling cleanup routines would dereference a NULL pointer.
2567 Description: NULL pointer deref in llap_from_page.
2568 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2569 incorrect) mapping if the page was truncated/reclaimed while it was
2570 searched for. Check for this condition and skip such pages when
2571 doing readahead. Introduce extra check to llap_from_page() to
2572 verify page->mapping->host is non-NULL (so page is not anonymous).
2575 Frequency : Sometimes when using sys_sendfile
2577 Description: "page not covered by a lock" warnings from ll_readpage
2578 Details : sendfile called ll_readpage without right page locks present.
2579 Now we introduced ll_file_sendfile that does necessary locking
2580 around call to generic_file_sendfile() much like we do in
2584 Frequency : with certain MDS communication failures at client mount time
2586 Description: NULL pointer deref after failed client mount
2587 Details : a client connection request may delayed by the network layer
2588 and not be sent until after the PTLRPC layer has timed out the
2589 request. If the client fails the mount immediately it will try
2590 to clean up before the network times out the request. Add a
2591 reference from the request import to the obd device and delay
2592 the cleanup until the network drops the request.
2595 Frequency : occasionally during client (re)connect
2597 Description: assertion failure during client (re)connect
2598 Details : processing a client connection request may be delayed by the
2599 client or server longer than the client connect timeout. This
2600 causes the client to resend the connection request. If the
2601 original connection request is replied in this interval, the
2602 client may trip an assertion failure in ptlrpc_connect_interpret()
2603 which thought it would be the only running connect process.
2606 Frequency : only with obd_echo servers and clients that are rebooted
2608 Description: kernel BUG accessing uninitialized data structure
2609 Details : When running an obd_echo server it did not start the ping_evictor
2610 thread, and when a client was evicted an uninitialized data
2611 structure was accessed. Start the ping_evictor in the RPC
2612 service startup instead of the OBD startup.
2614 Severity : enhancement
2615 Bugzilla : 10193 (patchless)
2616 Description: Remove dependency on various unexported kernel interfaces.
2617 Details : No longer need reparent_to_init, exit_mm, exit_files,
2618 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2621 Frequency : rare (only users of deprecated and unsupported LDAP config)
2623 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2624 Details : LDAP apparently contains 'lustreName' attributes instead of
2625 'name'. A simple remapping of the name is sufficient.
2628 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2630 Description: waiting_locks_callback trips kernel BUG if client is evicted
2631 Details : Running with the dump_on_timeout debug flag turned on makes
2632 it possible that the waiting_locks_callback() can try to dump
2633 the Lustre kernel debug logs from an interrupt handler. Defer
2634 this log dumping to the expired_lock_main() thread.
2636 Severity : enhancement
2638 Description: Support NFS exporting on 2.6 kernels.
2639 Details : Implement non-rawops metadata methods for NFS server to use without
2640 changing NFS server code.
2643 Frequency : very rare (synthetic metadata workload only)
2645 Description: two racing renames might cause an MDS thread to deadlock
2646 Details : Running the "racer" program may cause one MDS thread to rename
2647 a file from being the source of a rename to being the target of
2648 a rename at exactly the same time that another thread is doing
2649 so, and the second thread has already enqueued these locks after
2650 doing a lookup of the target and is trying to relock them in
2651 order. Ensure that we don't try to re-lock the same resource.
2654 Frequency : only very large systems with liblustre clients
2656 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2657 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2658 all OSTs in parallel. This allows the eviction of stale liblustre
2659 clients to proceed much faster than if they were done in series,
2660 and also offers similar improvements for other set_info RPCs.
2665 Description: excessive CPU usage during initial read phase on client
2666 Details : During the initial read phase on a client, it would agressively
2667 retry readahead on the file, consuming too much CPU and impacting
2668 performance (since 1.4.5.8). Improve the readahead algorithm
2669 to avoid this, and also improve some other common cases (read
2670 of small files in particular, where "small" is files smaller than
2671 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2676 Description: MDS crash when receiving packet with unknown intent.
2677 Details : Do not LBUG in unknown intent case, just return -EFAULT
2679 Severity : enhancement
2680 Bugzilla : 9293, 9385
2681 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2682 Details : Do not serialize getattr (non-intent version) and statfs.
2685 Frequency : occasional, when OST network is overloaded/intermittent
2687 Description: client evicted by OST after bulk IO timeout
2688 Details : If a client sends a bulk IO request (read or write) the OST
2689 may evict the client if it is unresposive to its data GET/PUT
2690 request. This is incorrect if the network is overloaded (takes
2691 too long to transfer the RPC data) or dropped the OST GET/PUT
2692 request. There is no need to evict the client at all, since
2693 the pinger and/or lock callbacks will handle this, and the
2694 client can restart the bulk request.
2697 Frequency : Always when mmapping file with no objects
2699 Description: client crashes when mmapping file with no objects
2700 Details : Check that we actually have objects in a file before doing any
2701 operations on objects in ll_vm_open, ll_vm_close and
2707 Description: Request leak when working with deleted CWD
2708 Details : Introduce advanced request refcount tracking for requests
2709 referenced from lustre intent.
2711 Severity : Enhancement
2713 Description: Cache open file handles on client.
2714 Details : MDS now will return special lock along with openhandle, if
2715 requested and client is allowed to hold openhandle, even if unused,
2716 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2717 closing files for every read/write openration.
2719 Severity : Enhancement
2721 Description: Cache open negative dentries on client when possible.
2722 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2723 negative dentries on lock revocation.
2728 Description: Remounting a client read-only wasn't possible with a zconf mount
2729 Details : It wasn't possible to remount a client read-only with llmount.
2731 Severity : enhancement
2732 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2733 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2734 MPICH is included to add Lustre-specific ADIO interfaces.
2735 This is based closely on the UFS ADIO layer and only differs
2736 in file creation, in order to allow the OST striping to be set.
2737 This is user-contributed code and not supported by CFS.
2742 Description: extended inode attributes (immutable, append-only) work improperly
2743 when 2.4 and 2.6 kernels are used on client/server or vice versa
2744 Details : Introduce kernel-independent values for these flags.
2746 Severity : enhancement
2749 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2750 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2751 granularity. Now, Fractional MB granularity is be supported,
2752 this is very useful for low memory system.
2754 Severity : enhancement
2756 Description: Getattr by fid
2757 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2758 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2759 have locks and avoids getattr rpc requests for it on every stat.
2762 Frequency : Always, for filesystems larger than 2TB
2764 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2765 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2766 filesystems larger than 2TB to be mounted with mballoc enabled.
2769 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2771 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2772 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2773 filesystems larger than 2TB is mounted then the kernel will
2774 silently corrupt the start of the filesystem. CONFIG_LBD is
2775 enabled for all CFS-supported kernels, but the possibility of
2776 this happening with a modified kernel config exists.
2778 Severity : enhancement
2780 Description: add client O_DIRECT support for 2.6 kernels
2781 Details : It is now possible to do O_DIRECT reads and writes to files
2782 in the Lustre client mountpoint on 2.6 kernel clients.
2784 Severity : enhancement
2786 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2787 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2788 parallel, not waiting for response from every OST before sending
2789 a rpc to the next OST.
2794 Description: setattr vs write race when updating file timestamps
2795 Details : Client processes that update a file timestamp into the past
2796 right after writing to the file (e.g. tar) it is possible that
2797 the updated file modification time can be reset to the current
2798 time due to a race between processing the setattr and write RPC.
2800 Severity : enhancement
2802 Description: Bring 'lfs find' closer in line with regular Linux find.
2803 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2804 -print0 options and obtains all the needed info through the lustre
2807 Severity : enhancement
2809 Description: support up to 1024 configured devices on one node
2810 Details : change obd_dev array from statically allocated to dynamically
2811 allocated structs as they are first used to reduce memory usage
2816 Description: Flush dirty partially truncated pages during truncate
2817 Details : Immediatelly flush partially truncated pages in filter_setattr,
2818 this way we completely avoid having any pages in page cache on OST
2819 and can retire ugly workarounds during writes to flush such pages.
2824 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2825 Details : protect inode from truncation within vfs_unlink() context
2826 just take a reference before calling vfs_unlink() and release it
2827 when parent's i_sem is free.
2832 Description: last_id value checked outside lock on OST caused LASSERT failure
2833 Details : If there were multiple MDS->OST object precreate requests in
2834 flight, it was possible that the OST's last object id was checked
2835 outside a lock and incorrectly tripped an assertion. Move checks
2836 inside locks, and discard old precreate requests.
2839 Frequency : always, if extents are used on OSTs
2841 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2842 Details : OSTs using the extents format would not zero the high 16 bits of
2843 the index physical block number. This is not a problem for any
2844 OST filesystems smaller than 16TB, and no kernels support ext3
2845 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2846 new/modified files) and can be fixed for existing filesystems
2847 with e2fsprogs-1.39-cfs1.
2852 Description: import connection selection may be incorrect if timer wraps
2853 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2854 import connections to be ignored if the 32-bit jiffies counter
2855 wraps. Use a 64-bit jiffies counter.
2858 Frequency : very large clusters immediately after boot
2860 Description: LNET request buffers exhausted under heavy short-term load
2861 Details : If a large number of client requests are generated on a service
2862 that has previously never seen so many requests it is possible
2863 that the request buffer growth cannot keep up with the spike in
2864 demand. Instead of dropping incoming requests, they are held in
2865 the LND until the RPC service can accept more requests.
2868 Frequency : Sometimes during replay
2870 Description: Assertion failure in ll_local_open after replay.
2871 Details : If replay happened on an open request reply before we were able
2872 to set replay handler, reply will become not swabbed tripping the
2873 assertion in ll_local_open. Now we set the handler right after
2874 recognising of open request
2877 Frequency : very rare
2879 Description: kernel reports "badness in vsnprintf"
2880 Details : Reading from the "recovery_status" /proc file in small chunks
2881 may cause a negative length in lprocfs_obd_rd_recovery_status()
2882 call to vsnprintf() (which is otherwise harmless). Exit early
2883 if there is no more space in the output buffer.
2885 Severity : enhancement
2887 Description: clear OBD RPC statistics by writing to them
2888 Details : It is now possible to clear the OBD RPC statistics by writing
2889 to the "stats" file.
2894 Description: Client mtime is not the same on different clients after utimes
2895 Details : In some cases, the client was using the utimes() syscall on
2896 a file cached on another node. The clients now validate the
2897 ctime from the MDS + OSTs to determine which one is right.
2902 Description: Inability to activate failout mode
2903 Details : lconf script incorrectly assumed that in pythong string's numeric
2904 value is used in comparisons.
2907 Frequency : always with multiple stripes per file
2909 Description: Inefficient object allocation for mutli-stripe files
2910 Details : When selecting which OSTs to stripe files over, for files with
2911 a stripe count that divides evenly into the number of OSTs,
2912 the MDS is always picking the same starting OST for each file.
2913 Return the OST selection heuristic to the original design.
2918 Description: mount failures may take full timeout to return an error
2919 Details : Under some heavy load conditions it is possible that a
2920 failed mount can wait for the full obd_timeout interval,
2921 possibly several minutes, before reporting an error.
2922 Instead return an error as soon as the status is known.
2924 Frequency : quota enabled and large files being deleted
2926 Description: releasing more than 4GB of quota at once hangs OST
2927 Details : If a user deletes more than 4GB of files on a single OST it
2928 will cause the OST to spin in an infinite loop. Release
2929 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2934 Description: statfs data retrieved from /proc may be stale or zero
2935 Details : When reading per-device statfs data from /proc, in the
2936 {kbytes,files}_{total,free,avail} files, it may appear
2937 as zero or be out of date.
2940 Frequency : systems with MD RAID1 external journal devices
2942 Description: lconf's call to blkid is confused by RAID1 journal devices
2943 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2944 of returning all block devices that match the journal UUID.
2947 Frequency : always, for aggregate stripe size over 4GB
2949 Description: assertion fails when trying to use 4GB stripe size
2950 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2951 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2954 Frequency : always on ppc64
2956 Description: the first write on an ext3 filesystem with mballoc got stuck
2957 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2958 perform endianness conversion.
2960 ------------------------------------------------------------------------------
2962 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2964 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2965 INTEROPERATE with older versions automatically. Please read the
2966 user documentation before upgrading any part of a live system.
2967 * WARNING: Lustre networking configuration changes are required with
2968 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2971 * Support for kernels:
2972 2.6.9-22.0.2.EL (RHEL 4)
2973 2.6.5-7.244 (SLES 9)
2974 2.6.12.6 vanilla (kernel.org)
2977 Severity : enhancement
2978 Bugzilla : 7981/8208
2979 Description: Introduced Lustre Networking (LNET)
2980 Details : LNET is new networking infrastructure for Lustre, it includes
2981 a reorganized network configuration mode (see the user
2982 documentation for full details) as well as support for routing
2983 between different network fabrics. Lustre Networking Devices
2984 (LNDs) for the supported network fabrics have also been
2985 created for this new infrastructure.
2987 Severity : enhancement
2988 Description: Introduced Access control lists
2989 Details : clients can set ACLs on files and directories in order to have
2990 more fine-grained permissions than the standard Unix UGO+RWX.
2991 The MDS must be started with the "-o acl" mount option.
2993 Severity : enhancement
2994 Description: Introduced filesystem quotas
2995 Details : Administrators may now establish per-user quotas on the
2998 Severity : enhancement
3000 Description: Configuration change for the XT3
3001 The PTLLND is now used to run Lustre over Portals on the XT3
3002 The configure option(s) --with-cray-portals are no longer used.
3003 Rather --with-portals=<path-to-portals-includes> is used to
3004 enable building on the XT3. In addition to enable XT3 specific
3005 features the option --enable-cray-xt3 must be used.
3010 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3011 Details : A race between allocating a new llcd and re-getting the llcd_lock
3012 allowed another thread to grab newly-allocated llcd.
3014 Severity : enhancement
3016 Description: 2.6 OST async journal commit and locking fix to improve performance
3017 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3018 2.6 kernels are now async as they already were in 2.4 kernels so
3019 that they can commit concurrently with the network bulk transfer.
3020 For block-allocated files the filter allocation semaphore is held
3021 to avoid filesystem fragmentation during allocation. BKL lock
3022 removed for 2.6 xattr operations where it is no longer needed.
3027 Description: lconf incorrectly determined whether two IP networks could talk
3028 Details : In some more complicated routing and multiple-network
3029 configurations, lconf will avoid trying to make a network
3030 connection to a disjoint part of the IP space. It was doing the
3031 math incorrectly for one set of cases.
3036 Description: Fix for potential infinite loop processing records in an llog.
3037 Details : If an llog record is corrupted/zeroed, it is possible to loop
3038 forever in llog_process(). Validate the llog record length
3039 and skip the remainder of the block on error.
3042 Frequency : occasional (liblustre only)
3044 Description: liblustre could not open files whose last component is a symlink
3045 Details : sysio_path_walk() would incorrectly pass the open intent to
3046 intermediate path components.
3049 Frequency : rare (liblustre only with non-standard tuning)
3050 Bugzilla : 7201 (7350)
3051 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3052 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3053 after it has been released, only do this for referenced locks
3055 Severity : enhancement
3057 Description: specifying an (invalid) directory default stripe_size of -1
3058 would reset the directory default striping
3059 Details : stripe_size -1 was used internally to signal directory stripe
3060 removal, now use "all default" to signal dir stripe removal
3061 as a directory striping of "all default" is not useful
3064 Frequency : common for large clusters running liblustre clients
3066 Description: doing an ls when liblustre clients are running is slow
3067 Details : sending a glimpse AST to a liblustre client waits for every AST
3068 to time out, as liblustre clients will not respond. Since they
3069 cannot cache data we refresh the OST lock LVB from disk instead.
3071 Severity : enhancement
3073 Description: doing an ls at the same time as file IO can be slow
3074 Details : enqueue and other "small" requests can be blocked behind many
3075 large IO requests. Create a new OST IO portal for non-IO
3076 requests so they can be processed faster.
3079 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3081 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3082 created file with mode 000
3083 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3084 allow an ftruncate() as a non-root user to a file with mode 000.
3085 Lustre now respects this flag to disable mode checks when
3086 truncating a file owned by the user
3089 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3092 Description: Revoking locks from clients that went dead or catatonic might take
3094 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3095 cancellation of such locks instant on servers without waiting for
3096 any reply from clients. Clients drops these locks when cancel
3097 notification from server is received without replying.
3100 Frequency : liblustre-only, when liblustre client dies or becomes busy
3102 Description: Doing ls on Linux clients can take a long time with active
3104 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3105 granting such locks to it in the first place if possible. Locks
3106 are taken by proxy on the OST during the read or write and
3107 dropped immediately afterward. Add connect flags handling, do
3108 not grant locks to liblustre clients for glimpse ASTs.
3110 Severity : enhancement
3112 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3113 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3114 sequential reads and behaves badly with applications doing random
3115 reads. Improve it by reading ahead at least read region, and
3116 avoiding excessive large RPC for small reads.
3118 Severity : enhancement
3120 Description: Creating more than 1000 files for a single job may cause a load
3121 imbalance on the OSTs if there are also a large number of OSTs.
3122 Details : qos_prep_create() uses an OST index reseed value that is an
3123 even multiple of the number of available OSTs so that if the
3124 reseed happens in the middle of the object allocation it will
3125 still utilize the OSTs as uniformly as possible.
3130 Description: OST or MDS may oops in ping_evictor_main()
3131 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3132 but doesn't restart at beginning of obd_exports_timed list
3135 Severity : enhancement
3137 Description: improve by-nid export eviction on the MDS and OST
3138 Details : allow multiple exports with the same NID to be evicted at one
3139 time without re-searching the exports list.
3142 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3144 Description: MDS may oops in groups_free()
3145 Details : in rare race conditions a newly allocated group_info struct is
3146 freed again, and this can be NULL. The 2.4 compatibility code
3147 for groups_free() checked for a NULL pointer, but 2.6 did not.
3150 Frequency : common for liblustre clients doing little filesystem IO
3151 Bugzilla : 9352, 7313
3152 Description: server may evict liblustre clients accessing contended locks
3153 Details : if a client is granted a lock or receives a completion AST
3154 with a blocking AST already set it would not reply to the AST
3155 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3158 Frequency : lfs setstripe, only systems with more than 160 OSTs
3160 Description: unable to set striping with a starting offset beyond OST 160
3161 Details : llapi_create_file() incorrectly limited the starting stripe
3162 index to the maximum single-file stripe count.
3165 Frequency : LDAP users only
3167 Description: lconf did not handle in-kernel recovery with LDAP properly
3168 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3170 Severity : enhancement
3172 Description: bind OST threads to NUMA nodes to improve performance
3173 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3174 node and do their allocations there to localize memory access
3176 Severity : enhancement
3178 Description: llmount can determine client NID directly from Myrinet (GM)
3179 Details : the client NID code from gmnalnid was moved directly into
3180 llmount, removing the need to use this or specifying the
3181 client NID explicitly when mounting GM clients with zeroconf
3184 Frequency : if client is started with down MDS
3186 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3187 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3188 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3189 ptlrpc_import_delayed_req() because we didn't check if the
3190 request was interrupted, and we also didn't break out of the
3191 event loop if there was no timeout
3196 Description: data loss during non-page-aligned writes to a single file from
3197 both multiple nodes and multiple threads on one node at same time
3198 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3199 inconsistency led to false short-reads, that were cached and later
3200 used by ->prepare_write() to fill in partially written page,
3201 leading to data loss.
3204 Frequency : always, if lconf --abort_recovery used
3206 Description: lconf --abort_recovery fails with 'Operation not supported'
3207 Details : lconf was attempting to abort recovery on the MDT device and not
3210 Severity : enhancement
3212 Description: remove cleanup logs
3213 Details : replace lconf-generated cleanup logs with lustre internal
3214 cleanup routines. Eliminates the need for client-cleanup and
3217 Severity : enhancement
3219 Description: add support for EAs (user and system) on lustre filesystems
3220 Details : it is now possible to store extended attributes in the Lustre
3221 client filesystem, and with the user_xattr mount option it
3222 is possible to allow users to store EAs on their files also
3224 Severity : enhancement
3226 Description: Add possibility (config option) to show minimal available OST free
3228 Details : When compiled with --enable-mindf configure option, statfs(2)
3229 (and so, df) will return least minimal free space available from
3230 all OSTs as amount of free space on FS, instead of summary of
3231 free spaces of all OSTs.
3233 Severity : enhancement
3235 Description: do not expand extent locks acquired on OST-side
3236 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3237 by server: they are not cached anyway.
3240 Frequency : when mmap is used/binaries executed from Lustre
3242 Description: Unmmap pages before throwing them away from read cache.
3243 Details : llap_shrink cache now attempts to unmap pages before discarding
3244 them (if unmapping failed - do not discard). SLES9 kernel has
3245 extra checks that trigger if this unmapping is not done first.
3250 Description: lconf didn't resolve symlinks before checking to see whether a
3251 given mountpoint was already in use
3254 Frequency : when migrating failover services
3255 Bugzilla : 6395, 9514
3256 Description: When migrating a subset of services from a node (e.g. failback
3257 from a failover service node) the remaining services would
3258 time out and evict clients.
3259 Details : lconf --force (implied by --failover) sets the global obd_timeout
3260 to 5 seconds in order to quickly disconnect, but this caused
3261 other RPCs to time out too quickly. Do not change the global
3262 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3264 Severity : enhancement
3265 Frequency : if MDS is started with down OST
3266 Bugzilla : 9439,5706
3267 Description: Allow startup/shutdown of an MDS without depending on the
3268 availability of the OSTs.
3269 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3270 Add appropriate locking and lov-osc refcounts for safe
3271 cleaning. Add osc abort_inflight calls in case the
3272 synchronize never started.
3275 Frequency : occasional (Cray XT3 only)
3277 Description: root not authorized to access files in CRAY_PORTALS environment
3278 Details : The client process capabilities were not honoured on the MDS in
3279 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3280 been accessed by an authorized user then root was able to access
3281 the file on the local client also. The root user capabilities
3282 are now allowed on the MDS, as this environment has secure UID.
3285 Frequency : occasional
3287 Description: ldiskfs "too long searching" message happens too often
3288 Details : A debugging message (otherwise harmless) prints too often on
3289 the OST console. This has been reduced to only happen when
3290 there are fragmentation problems on the filesystem.
3295 Description: Division by zero in statfs when all OSCs are inactive
3296 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3297 lov_statfs divides by value returned by lov_get_stripecnt().
3301 Bugzilla : 9489, 3273
3302 Description: First write from each client to each OST was only 4kB in size,
3303 to initialize client writeback cache, which caused sub-optimal
3304 RPCs and poor layout on disk for the first writen file.
3305 Details : Clients now request an initial cache grant at (re)connect time
3306 and so that they can start streaming writes to the cache right
3307 away and always do full-sized RPCs if there is enough data.
3308 If the OST is rebooted the client also re-establishes its grant
3309 so that client cached writes will be honoured under the grant.
3314 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3315 Details : Now I/O RPCs go to different portal number and (presumably) fast
3316 lock requests (and glimses) and other RPCs get their own service
3317 threads pool that should be able to service those RPCs
3320 Severity : enhancement
3322 Description: Ability to exchange lustre version between client and servers and
3323 issue warnings at client side if client is too old. Also for
3324 liblustre clients there is ability to refuse connection of too old
3326 Details : New 'version' field is added to connect data structure that is
3327 filled with version info. That info is later checked by server and
3331 Frequency : rare, liblustre only.
3332 Bugzilla : 9296, 9581
3333 Description: Two simultaneous writes from liblustre at offset within same page
3334 might proceed at the same time overwriting eachother with stale
3336 Details : I/O lock withing llu_file_prwv was released too early, before data
3337 actually was hitting the wire. Extended lock-holding time until
3338 server acknowledges receiving data.
3341 Frequency : extremely rare. Never observed in practice.
3343 Description: avoid generating lustre_handle cookie of 0.
3344 Details : class_handle_hash() generates handle cookies by incrementing
3345 global counter, and can hit 0 occasionaly (this is unlikely, but
3346 not impossible, because initial value of cookie counter is
3347 selected randonly). Value of 0 is used as a sentinel meaning
3348 "unassigned handle" --- avoid it. Also coalesce two critical
3349 sections in this function into one.
3351 Severity : enhancement
3353 Description: allow liblustre clients to delegate truncate locking to OST
3354 Details : To avoid overhead of locking, liblustre client instructs OST to
3355 take extent lock in ost_punch() on client's behalf. New connection
3356 flag is added to handle backward compatibility.
3358 Severity : enhancement
3359 Bugzilla : 4928, 7341, 9758
3360 Description: allow number of OST service threads to be specified
3361 Details : a module parameter allows the number of OST service threads
3362 to be specified via "options ost ost_num_threads={N}" in the
3363 OSS's /etc/modules.conf or /etc/modprobe.conf.
3367 Bugzilla : 6146, 9635, 9895
3368 Description: servers crash with bad pointer in target_handle_connect()
3369 Details : In rare cases when a client is reconnecting it was possible that
3370 the connection request was the last reference for that export.
3371 We would temporarily drop the export reference and get a new
3372 one, but this may have been the last reference and the export
3373 was just destroyed. Get new reference before dropping old one.
3375 Severity : enhancement
3376 Frequency : if client is started with failover MDS
3378 Description: Allow multiple MDS hostnames in the mount command
3379 Details : Try to read the configuration from all specified MDS
3380 hostnames during a client mount in case the "primary"
3383 Severity : enhancement
3385 Description: Stop sending data to evicted clients as soon as possible.
3386 Details : Check if the client we are about to send or are sending data to
3387 was evicted already. (Check is done every second of waiting,
3388 for which l_wait_event interface was extended to allow checking
3389 of exit condition at specified intervals).
3392 Frequency : rare, normally only when NFS exporting is done from client
3394 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3396 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3399 Frequency : always, if srand() is called before liblustre initialization
3401 Description: Liblustre uses system PRNG disturbing its usage by user application
3402 Details : Introduce internal to lustre fast and high-quality PRNG for
3403 lustre usage and make liblustre and some other places in generic
3404 lustre code to use it.
3406 Severity : enhancement
3407 Bugzilla : 9477, 9557, 9870
3408 Description: Verify that the MDS configuration logs are updated when xml is
3409 Details : Check if the .xml configuration logs are newer than the config
3410 logs stored on the MDS and report an error if this is the case.
3411 Request --write-conf, or allow starting with --old_conf.
3413 Severity : enhancement
3415 Description: Handle symlinks in the path when checking if Lustre is mounted.
3416 Details : Resolve intermediate symlinks when checking if a client has
3417 mounted a filesystem to avoid duplicate client mounts.
3422 Description: lconf can hit an error exception but still return success.
3423 Details : The lconf command catches the Command error exception at the top
3424 level script context and will exit with the associated exit
3425 status, but doesn't ensure that this exit status is non-zero.
3430 Description: failure of ptlrpc thread startup can cause oops
3431 Details : Starting a ptlrpc service thread can fail if there are a large
3432 number of threads or the server memory is very fragmented.
3433 Handle this without oopsing.
3436 Frequency : always, only if liblustre and non-default acceptor port was used
3438 Description: liblustre cannot connect to servers with non-default acceptor port
3439 Details : tcpnal_set_default_params() was not called and was therefore
3440 ignoring the environment varaible TCPNAL_PORT, as well as other
3441 TCPNAL_ environment variables
3446 Description: two objects could be created on the same OST for a single file
3447 Details : If an OST is down, in some cases it was possible to create two
3448 objects on a single OST for a single file. No problems other
3449 than potential performance impact and spurious error messages.
3453 Bugzilla : 5681, 9562
3454 Description: Client may oops in ll_unhash_aliases
3455 Details : Client dcache may become inconsistent in race condition.
3456 In some cases "getcwd" can fail if the current directory is
3462 Description: Inode refcounting problems in NFS export code
3463 Details : link_raw functions used to call d_instantiate without obtaining
3464 extra inode reference first.
3468 Bugzilla : 9942, 9903
3469 Description: Referencing freed requests leading to crash, memleaks with NFS.
3470 Details : We used to require that call to ll_revalidate_it was always
3471 followed by ll_lookup_it. Also with revalidate_special() it is
3472 possible to call ll_revalidate_it() twice for the same dentry
3473 even if first occurence returned success. This fix changes semantic
3474 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3475 reference on a request referred from the intent.
3476 ll_intent_release() then releases such a request.
3479 Frequency : rare, normally benchmark loads only
3481 Description: unlinked inodes were kept in memory on the client
3482 Details : If a client is repeatedly creating and unlinking files it
3483 can accumulate a lot of stale inodes in the inode slab cache.
3484 If there is no other client load running this can cause the
3485 client node to run out of memory. Instead flush old inodes
3486 from client cache that have the same inode number as a new inode.
3489 Frequency : SLES9 2.6.5 kernel and long filenames only
3490 Bugzilla : 9969, 10379
3491 Description: utime reports stale NFS file handle
3492 Details : SLES9 uses out-of-dentry names in some cases, which confused
3493 the lustre dentry revalidation. Change it to always use the
3497 Frequency : rare, unless heavy write-truncate concurrency is continuous
3498 Bugzilla : 4180, 6984, 7171, 9963, 9331
3499 Description: OST becomes very slow and/or deadlocked during object unlink
3500 Details : filter_destroy() was holding onto the parent directory lock
3501 while truncating+unlinking objects. For very large objects this
3502 may block other threads for a long time and slow overall OST
3503 responsiveness. It may also be possible to get a lock ordering
3504 deadlock in this case, or run out of journal credits because of
3505 the combined truncate+unlink. Solution is to do object truncate
3506 first in one transaction without parent lock, and then do the
3507 final unlink in a new transaction with the parent lock. This
3508 reduces the lock hold time dramatically.
3511 Frequency : rare, 2.4 kernels only
3513 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3514 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3515 before unmounting their filesystem in order to avoid deadlock.
3516 The kernel_locked() function in 2.4 kernels only checks whether
3517 the kernel lock is held, not whether it is this process that is
3518 holding it as 2.6 kernels do.
3523 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3524 Details : The client ptlrpc code may be trying to reconnect to a down
3525 server before a previous connection attempt has timed out.
3526 Increase the reconnect interval to be longer than the connection
3527 timeout interval to avoid sending duplicate connections to
3531 Frequency : echo_client brw_test command
3533 Description: fix echo_client to work with OST preallocated code
3534 Details : OST preallocation code (5137) didn't take echo_client IO path
3535 into account: echo_client calls filter methods outside of any
3536 OST thread and, hence, there is no per-thread preallocated
3537 pages and buffers to use. Solution: hijack pga pages for IO. As
3538 a byproduct, this avoids unnecessary data copying.
3542 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3543 Description: Client can oops in mdc_commit_close() after open replay
3544 Details : It was possible for the MDS to return an open request with no
3545 transaction number in mds_finish_transno() if the client was
3546 evicted, but without actually returning an error. Clients
3547 would later try to replay that open and may trip an assertion
3548 Simplify the client close codepath, and always return an error
3549 from the MDS in case the open is not successful.
3552 Frequency : rare, 2.6 OSTs only
3554 Description: OST may deadlock under high load on fragmented files
3555 Details : If there was a heavy load and highly-fragmented OST filesystems
3556 it was possible to have all the OST threads deadlock waiting on
3557 allocation of biovecs, because the biovecs were not released
3558 until the entire RPC IO was completed. Instead, release biovecs
3559 as soon as they are complete to ensure forward IO progress.
3561 Severity : enhancement
3563 Description: Support for specifying external journal device at mount
3564 Details : If an OST or MDS device is formatted with an external journal
3565 device, this device major/minor is stored in the ext3 superblock
3566 and may not be valid for failover. Allow detecting and
3567 specifying the external journal at mount time.
3572 Description: Mounting an MDS with pending unlinked files may cause oops
3573 Details : target_finish_recovery() calls mds_postrecov() which returned
3574 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3575 considers this an error and immediately begins cleaning up the
3576 lov, just after starting the mds_lov process
3578 Severity : enhancement
3580 Description: Implement 'lfs df' to report actual free space on per-OST basis
3581 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3582 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3583 usage of objects; '-h' to report in human readable format.
3585 ------------------------------------------------------------------------------
3587 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3594 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3595 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3596 at startup during orphan unlinking. Instead, skip buddy update
3597 before setup, as it will be regenerated after recovery is complete.
3602 Description: If an OST is inactive, its locks might reference stale inodes.
3603 Details : lov_change_cbdata() must iterate over all namespaces, even if
3604 they are inactive to clear inode references from the lock.
3606 Severity : enhancement
3607 Frequency : occasional, if non-standard max_dirty_mb used
3609 Description: Client will block write RPCs if not enough grant
3610 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3611 then the client will block writes while waiting for another RPC
3612 to complete instead of consuming its dirty limit. With change
3613 we get improved performance when max_dirty_mb is small.
3615 Severity : enhancement
3616 Bugzilla : 3389, 6253
3617 Description: Add support for supplementary groups on the MDS.
3618 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3619 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3620 lookups for user supplementary groups into a cache.
3624 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3625 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3626 for liblustre O_EXCL is enforced by MDS.
3630 Description: readdir never returns NULL in liblustre.
3631 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3632 dentry in struct dirent.
3636 Frequency : liblustre only, depends on application IO pattern
3637 Description: liblustre clients evicted if not contacting servers
3638 Details : Don't put liblustre clients into the ping_evictor list, so
3639 they will not be evicted by the pinger ever.
3641 Severity : enhancement
3643 Description: Add ability to evict clients by NID from MDS.
3644 Details : By echoing "nid:$NID" string into
3645 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3646 $NID would be instantly evicted from this MDS and from all active
3647 OSTs connected to it.
3651 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3652 Details : lookup_it_finish() used to query file size from OSTs that was not
3657 Description: service threads change working directory to that of init
3658 Details : Starting lustre service threads may pin the working directory
3659 of the parent thread, making that filesystem busy. Threads
3660 now change to the working directory of init to avoid this.
3664 Frequency : during shutdown only
3665 Description: shutdown with a failed MDS or OST can cause unmount to hang
3666 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3671 Frequency : 2.6 only
3672 Description: chmod/chown may include an extra supplementary group
3673 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3674 supplementary group and if none is specified this is used.
3677 Bugzilla : 5479 (6816)
3679 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3680 Details : If lookup is in progress on a file that is unlinked we might try
3681 to revalidate the inode and fail in revalidate after lookup is
3682 complete and ll_file_open() enqueues the open again but
3683 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3686 Frequency : always, if lconf --abort_recovery used
3688 Description: lconf --abort_recovery fails with 'Operation not supported'
3689 Details : lconf was attempting to abort recovery on the MDT device and not
3692 ------------------------------------------------------------------------------
3694 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3699 Frequency : rare (only unsupported configurations with a node running as an
3701 Bugzilla : 6514, 5137
3702 Description: Mounting a Lustre file system on a node running as an OST could
3704 Details : OSTs now preallocates memory needed to write out data at
3705 startup, instead of when needed, to avoid having to
3706 allocate memory in possibly low memory situations.
3707 Specifically, if the file system is mounted on on OST,
3708 memory pressure could force it to try to write out data,
3709 which it needed to allocate memory to do. Due to the low
3710 memory, it would be unable to do so and the node would
3711 become unresponsive.
3713 Severity : enhancement
3715 Description: Addition of lconf --service command line option
3716 Details : lconf now accepts a '--service <arg>' option, which is
3717 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3719 Severity : enhancement
3721 Description: Failover mode is now the default for OSTs.
3722 Details : By default, OSTs will now run in failover mode. To return to
3723 the old behaviour, add '--failout' to the lmc line for OSTs.
3725 Severity : enhancement
3727 Description: Health checks are now provided for MDS and OSTs
3728 Details : Additional detailed health check information on MSD and OSTs
3729 is now provided through the procfs health_check value.
3732 Frequency : occasional, depends on IO load
3734 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3735 after numerous create/delete cycles
3736 Details : The ext3 inode allocation policy would not allocate new inodes
3737 very well on the OSTs because there are no new directories
3738 being created. Instead we look for groups with free space if
3739 the parent directories are nearly full.
3744 Description: Network or server problems during mount may cause partially
3745 mounted clients instead of returning an error.
3746 Details : The config llog parsing code may overwrite the error return
3747 code during mount error handling, returning success instead
3753 Description: MDS can fail to allocate large reply buffers
3754 Details : After long uptimes the MDS can fail to allocate large reply
3755 buffers (e.g. zconf client mount config records) due to memory
3756 fragmentation or consumption by the buffer cache. Preallocate
3757 some large reply buffers so that these replies can be sent even
3758 under memory pressure.
3762 Frequency : rare (liblustre)
3763 Description: fsx running with liblustre complained that using truncate() to
3764 extend the file doesn't work. This patch corrects that issue.
3765 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3766 ATTR_SIZE and lsm use in llu_setattr_raw.
3770 Frequency : rare, only 2.6 kernels
3771 Description: Unusual file access patterns on the MDS may result in inode
3772 data being lost in very rare circumstances.
3773 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3774 code in the 2.6 kernel caused the inode and/or EA data not to
3775 be read from disk, causing single-file corruption.
3779 Frequency : rare, only 2.6 filesystems using extents
3780 Description: Heavy concurrent write and delete load may cause data corruption.
3781 Details : It was possible under high-load situations to have an extent
3782 metadata block in the block device cache from a just-unlinked
3783 file overwrite a newly-allocated data block. We now unmap any
3784 metadata buffers that alias just-allocated data blocks.
3788 Frequency : filesystems with default stripe_count larger than 77
3789 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3790 Details : lconf specifies an inode size of 4096 bytes when the default
3791 stripe_count is larger than 77. This conflicts with the default
3792 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3793 this case to avoid pinning too much memory for large EAs.
3795 ------------------------------------------------------------------------------
3797 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3802 Frequency : rare (extremely heavy IO load with hundreds of clients)
3804 Description: Client is evicted, gets IO error writing to file
3805 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3806 caused clients to be evicted for AST timeouts. The fixes in
3807 bug 5192 mean we no longer need to have such short AST timeouts
3808 so ldlm_timeout has been increased.
3811 Frequency : occasional during --force or --failover shutdown under load
3812 Bugzilla : 5949, 4834
3813 Description: Server oops/LBUG if stopped with --force or --failover under load
3814 Details : a collection of import/export refcount and cleanup ordering
3815 issues fixed for safer force cleanup
3818 Frequency : only filesystems larger than 120 OSTs
3819 Bugzilla : 5990, 6223
3820 Description: lfs getstripe would oops on a very large filesystem
3821 Details : lov_getconfig used kfree on vmalloc'd memory
3824 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3825 Bugzilla : 6242, 6243
3826 Description: reading from files that had been truncated to a non-zero size
3827 but never opened returned no data
3828 Details : ll_file_read() reads zeros from no-object files to EOF
3833 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3834 Details : There's little harm in aborting MDS/OSS recovery and letting it
3835 try again, so I removed the LASSERT and return an error instead.
3837 Severity : enhancement
3839 Description: New debugging infrastructure for tracking down data corruption
3840 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3841 (b) cover more of the client-side code path, and (c) try to narrow
3842 down where problems occurred
3846 Bugzilla : 3819, 4364, 4397, 6313
3847 Description: Racing close and eviction MDS could cause assertion in mds_close
3848 Details : It was possible to get multiple mfd references during close and
3849 client eviction, leading to one thread referencing a freed mfd.
3851 Severity: : enhancement
3852 Bugzilla : 3262, 6359
3853 Description: Attempts to reconnect to servers are now more aggressive.
3854 Details : This builds on the enhanced upcall-less recovery that was added
3855 in 1.4.2. When trying to reconnect to servers, clients will
3856 now try each server in the failover group every 10 seconds. By
3857 default, clients would previously try one server every 25 seconds.
3862 Description: After recovery, certain operations trigger a failed
3863 assertion on a client.
3864 Details : Failing over an mds, using lconf -d --failover, while a
3865 client was doing a readdir() call would cause the client to
3866 LBUG after recovery completed and the readdir() was resent.
3868 Severity : enhancement
3870 Description: Default groups are now added by lconf
3871 Details : You can now run lconf --group <servicename> without having to
3872 manually add groups with lmc.
3875 Frequency : occasional
3877 Description: Nodes with an elan id of 0 trigger a failed assertion
3880 Frequency : always when accessing e.g. tty/console device nodes
3882 Description: tty and some other devices nodes cannot be used on lustre
3883 Details : file's private_data field is used by device data and lustre
3884 values in there got lost. New field was added to struct file to
3885 store fs-specific private data.
3888 Frequency : when exporting Lustre via NFS
3890 Description: NFSD failed occasionally when looking up a path component
3891 Details : NFSD is looking up ".." which was broken in ext3 directories
3892 that had grown large enough to become hashed.
3895 Frequency : Clusters with multiple interfaces not on the same subnet
3897 Description: Nodes will repeatedly try to reconnect to an interface which it
3898 cannot reach and report an error to the log.
3899 Details : Extra peer list entries will be created by lconf with some peers
3900 unreachable. lconf now validates the peer before adding it.
3903 Frequency : Only if a default stripe is set on the filesystem root.
3905 Description: Setting a default stripe on the filesystem root prevented the
3906 filesystem from being remounted.
3907 Details : The client was sending extra request flags in the root getattr
3908 request and did not allocate a reply buffer for the dir EA.
3911 Frequency : occasional, higher if lots of files are accessed by one client
3912 Bugzilla : 6159, 6097
3913 Description: Client trips assertion regarding lsm mismatch/magic
3914 Details : While revalidating inodes the VFS looks up inodes with ifind()
3915 and in rare cases can find an inode that is being freed.
3916 The ll_test_inode() code will free the lsm during ifind()
3917 when it finds an existing inode and then the VFS later attaches
3918 this free lsm to a new inode.
3922 Bugzilla : 6422, 7030
3923 Description: MDS deadlock between mkdir and client eviction
3924 Details : Creating a new file via mkdir or mknod (starting a transaction
3925 and getting the ns lock) can deadlock with client eviction
3926 (gets ns lock and trying to finish a synchronous transaction).
3929 Frequency : occasional
3930 Description: While starting a server, the fsfilt_ext3 module could not be
3932 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3933 kernels. Previously, lconf would still use the ext3 name
3934 when trying to load modules. Now, it will correctly use
3935 ext3 on 2.4 and ldiskfs on 2.6.
3937 Severity : enhancement
3938 Description: The default stripe count has been changed to 1
3939 Details : The interpretation of the default stripe count (0, to lfs
3940 or lmc) has been changed to mean striping across a single
3941 OST, rather than all available. For general usage we have
3942 found a stripe count of 1 or 2 works best.
3944 Severity : enhancement
3945 Description: Add support for compiling against Cray portals.
3946 Details : Conditional compiling for some areas that are different
3950 Frequency : occasional
3951 Bugzilla : 6409, 6834
3952 Description: Creating files with an explicit stripe count may lead to
3953 a failed assertion on the MDS
3954 Details : If some OSTs are full or unavailable, creating files may
3955 trigger a failed assertion on the MDS. Now, Lustre will
3956 try to use other servers or return an error to the
3960 Frequency : occasional
3962 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3963 caused lock timeout during readahead (since 1.4.2).
3964 Details : Processes doing readahead might match a lock that hasn't been
3965 granted yet if there are overlapping and conflicting lock
3966 requests. The readahead process waits on ungranted lock
3967 (original lock is CBPENDING), while OST waits for that process
3968 to cancel CBPENDING read lock and eventually evicts client.
3970 Severity : enhancement
3972 Description: Initial enabling of flock support for clients
3973 Details : Implements fcntl advisory locking and file status functions.
3974 This feature is provided as an optional mount flag (default
3975 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3976 locking are implemented yet, and those that are are not guaranteed
3977 to be completely correct in production environments.
3978 mount -t lustre -o [flock|noflock] ...
3981 Frequency : occasional
3983 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3984 assertion in the ext3 JBD (journaling) layer.
3985 Details : The b_committed_data struct is protected by the big kernel lock
3986 in 2.4 kernels, serializing journal_commit_transaction() and
3987 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3988 access to this struct. In 2.6 kernels there is finer grained
3989 locking to improve SMP performance of the JBD layer.
3993 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3998 Description: merge and cleanup kernel patches.
4001 -----------------------------------------------------------------------------
4003 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4005 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4006 versions of Lustre. Please update all servers and clients to
4007 version 1.4.2 or later at the same time. You must also run
4008 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4009 to update the configuration logs.
4011 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4012 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4013 - swab llog records of type '0' so we get proper header size/idx (5861)
4014 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4015 - fix rename of one directory over another leaking an inode (5953)
4016 - avoid SetPageDirty on 2.6 (5981)
4017 - don't re-add just-being-destroyed locks to the waiting list (5653)
4018 - when creating new directories, inherit the parent's custom
4019 striping settings if present parent (3048)
4020 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4021 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4022 - don't hold client locks on temporary worklist from l_lru (5666)
4023 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4024 - automatically evict dead clients (5921)
4025 - Update file size properly in create+truncate+fstat case (6196)
4026 - Do not unhash mountpoint dentries, do not allow removal of
4028 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4029 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4030 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4031 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4032 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4033 - Correctly handle reads of files with no objects (6243)
4034 - lctl recover will also mark a device active if deactivate used (5933)
4036 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
4037 - allow --write-conf on an MDS with different nettype than client (5619)
4038 - don't write config llogs to MDS for mounts not from that MDS (5617)
4039 - lconf should create multiple TCP connections from a client (5201)
4040 - init scripts are now turned off by default; run chkconfig --on
4041 lustre and chkconfig --on lustrefs to use them
4042 - upcalls are no longer needed for clients to recover to failover
4044 - add --abort-recovery option to lconf to abort recovery on device
4046 - add support for an arbitrary number of OSTs (3026)
4047 - Quota support protocol changes.
4048 - forward compatibility changes to wire structs (6007)
4049 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
4050 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
4051 - improved "lustre status" script
4052 - initialize blocksize for non-regular files (6062)
4053 - added --disable-server and --disable-client configure options (5782)
4054 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
4055 - Vanilla 2.4.29 support
4056 - increase maximum number of obd devices to 520 (6242)
4057 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4058 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4059 SLES 9 2.6.5 kernels
4060 - limit stripes per file to 160 (the maximum EA size) (6093)
4062 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4065 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4066 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4067 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4068 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4069 - small ext3 extents cleanups and fixes (5733)
4070 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4071 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4072 - avoid needless client->OST connect, fix handle mismatch (5317)
4073 - fix DLM error path that led to out-of-sync client, long delays (5779)
4074 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4075 - fix several locking issues related to i_size (5492,5624,5654,5672)
4076 - don't move pending lock onto export if it is already evicted (5683)
4077 - fix kernel oops when creating .foo in unlinked directory (5548)
4078 - fix deadlock in obdfilter statistics vs. object create (5811)
4079 - use time_{before,after} to avoid timer jiffies wrap (5882)
4080 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4081 - Do not leak request if server was not able to process it (5154)
4082 - If mds_open unable to find parent dir, make that negative lookup(5154)
4083 - don't create new directories with extent-mapping (5909, 5936)
4085 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
4086 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
4087 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
4088 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
4089 - support for RHEL4 kernel on i686 (5773)
4090 - provide error messages when incompatible logs are encountered (5898)
4092 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
4093 * version 1.4.0.10 (1.4.1 release candidate 1)
4095 - don't keep a lock reference when lock is not granted (4238)
4096 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4097 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
4098 - drop import inflight refcount on signal_completed_replay error (5255)
4099 - unlock page after async write error during send (3677)
4100 - handle missing objects in filter_preprw_read properly (5265)
4101 - no transno return for symlink open, don't save no-trasno open (3440)
4102 - don't try to complete elan receive that already failed (4012)
4103 - free RPC server reply state on error (5406)
4104 - clean up thread from ptlrpc_start_thread() on error (5160)
4105 - readahead could read extra page into cache that wasn't ejected (5388)
4106 - prevent races in class_attach/setup/cleanup/detach (5260)
4107 - don't dereference de->d_inode after l_dput of de (5458)
4108 - use "int" for stripe value returned from lock_to_stripe (5544)
4109 - mballoc allocation and error-checking fixes in 2.6 (5504)
4110 - block device patches to fix I/O request sizes in 2.6 (5482)
4111 - look up hostnames for IB nals (5602)
4112 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
4113 - don't start multiple acceptors for the same port (5277)
4114 - fix incorrect LASSERT in mds_getattr_name (5635)
4115 - export a proc file for general "ping" checking (5628)
4116 - fix "lfs check" to not block when the MDS is down (5628)
4118 - service request history (4965)
4119 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4120 - create an "evict_client" /proc entry on OSTs, like the MDS has
4121 - fix mount usage message, return errors per mount(8) (5168)
4122 - change grep [] to grep "[]" in tests so they work in more UMLs
4123 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4124 - remove ancient LOV_MAGIC_V0 EA support (5047)
4125 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4126 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4127 - allow readahead allocations to fail when low on memory (5383)
4128 - mmap locking landed again, after considerable improvement (2828)
4129 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4131 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4134 - send OST transaction number in read/write reply to free req (4966)
4135 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4136 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4137 - the watchdog thread now runs as interruptible (5246)
4138 - flock/lockf fixes (but it's still disabled, pending 5135)
4139 - don't use EXT3 constants in llite code (5094)
4140 - memory shortage at startup could cause assertion (5176)
4142 - reorganization of lov code
4143 - single portals codebase
4145 - add extents/mballoc support (5025)
4146 - direct I/O reads in the obdfilter (4048)
4147 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4149 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4152 - send OST transaction number in read/write reply to free req (4966)
4153 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4154 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4155 - the watchdog thread now runs as interruptible (5246)
4156 - handle missing objects in filter_preprw_read properly (5265)
4157 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4158 - drop import inflight refcount on signal_completed_replay error (5255)
4159 - unlock page after async write error during send (3677)
4160 - return original error code on reconstructed replies (3761)
4161 - no transno return for symlink open, don't save no-trasno open (3440)
4163 - add pid to ldlm debugging output (4922)
4164 - bump the watchdog timeouts -- we can't handle 30sec yet
4165 - extra debugging for orphan dentry/inode bug (5259)
4167 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
4170 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4171 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4172 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
4173 - fix NULL dereference in /proc/sys/portals/routes (4827)
4174 - allow failed mdc_close() operations to be interrupted (4561)
4175 - stop precreate on OST before MDS would time out on it (4778)
4176 - don't send partial-page writes before EOF from client (4410)
4177 - discard client grant for sub-page writes on large-page clients (4520)
4178 - don't free dentries not owned by NFS code, check generation (4806)
4179 - fix lsm leak if mds_create_objects() fails (4801)
4180 - limit debug_daemon file size, always print CERROR messages (4789)
4181 - use transno after validating reply (3892)
4182 - process timed out requests if import state changes (3754)
4183 - update mtime on OST during writes, return in glimpse (4829)
4184 - add mkfsoptions to LDAP (4679)
4185 - use ->max_readahead method instead of zapping global ra (5039)
4186 - don't interrupt __l_wait_event() during strace
4188 - add software watchdogs to catch hung threads quickly (4941)
4189 - make lustrefs init script start after nfs is mounted
4190 - fix CWARN/ERROR duplication (4930)
4191 - return async write errors to application if possible (2248)
4192 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4193 - print NAL number in %x format (4645)
4194 - update barely-supported suse-2.4.21-171 series (4842)
4195 - support for sles 9 %post scripts
4196 - support for building 2.6 kernel-source packages
4197 - support for sles km_* packages
4199 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4202 - ignore -ENOENT errors in osc_destroy (3639)
4203 - notify osc create thread that OSC is being cleaned up (4600)
4204 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4205 - reconstruct ost_handle() like mds_handle() (4657)
4206 - create a new thread to do import eviction to avoid deadlock (3969)
4207 - let lconf resolve symlinked-to devices (4629)
4208 - don't unlink "objects" from directory with default EA (4554)
4209 - hold socknal file ref over connect in case target is down (4394)
4210 - allow more than 32000 subdirectories in a single directory (3244)
4211 - fix blocks count for O_DIRECT writes (3751)
4212 - OST returns ENOSPC from object create when no space left (4539)
4213 - don't send truncate RPC if file size isn't changing (4410)
4214 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4215 - bind to privileged port in socknal and tcpnal (3689)
4217 - rate limit CERROR/CWARN console message to avoid overload (4519)
4218 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4219 - basic mmap support (3918)
4220 - kernel patch series update from b1_4 (4711)
4222 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4225 - avoid crash during MDS cleanup with OST shut down (2775)
4226 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4227 - don't use bad inodes on the MDS (3744)
4228 - dynamic object preallocation to improve recovery speed (4236)
4229 - don't hold spinlock over lock dumping or change debug flags (4401)
4230 - don't zero obd_dev when it is force cleaned (3651)
4231 - print grants to console if they go negative (4431)
4232 - "lctl deactivate" will stop automatic recovery attempts (3406)
4233 - look for existing locks in ldlm_handle_enqueue() (3764)
4234 - don't resolve lock handle twice in recovery avoiding race (4401)
4235 - revalidate should check working dir is a directory (4134)
4237 - don't always mark "slow" obdfilter messages as errors (4418)
4239 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4242 - don't close LustreDB during write_conf until it is done (3860)
4243 - fix typo in lconf for_each_profile (3821)
4244 - allow dumping logs from multiple threads at one time (3820)
4245 - don't allow multiple threads in OSC recovery (3812)
4246 - fix debug_size parameters (3864)
4247 - fix mds_postrecov to initialize import for llog ctxt (3121)
4248 - replace config semaphore with spinlock (3306)
4249 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4250 - don't allow enqueue to complete on a destroyed export (3822)
4251 - down write_lock before checking llog header bitmap (3825)
4252 - recover from lock replay timeout (3764)
4253 - up llog sem before sending rpc (3652)
4254 - reduce ns lock hold times when setting kms (3267)
4255 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4256 - fix NULL deref and obd_dev leak on setup error (3312)
4257 - replace some LBUG about llog ops with error handling (3841)
4258 - don't match INVALID dentries from d_lookup and spin (3784)
4259 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4260 - fix invalid assertion in ptlrpc_set_wait (3880)
4262 - add libwrap support for the TCP acceptor (3996)
4263 - add /proc/sys/portals/routes for non-root route listing (3994)
4264 - allow setting MDS UUID in .xml (2580)
4265 - print the stack of a process that LBUGs (4228)
4267 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4270 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4271 - make sure to unset replay flag from failed open requests (3440)
4272 - if default stripe count is 0, use OST count for inode size (3636)
4273 - update parent mtime/ctime on client for create/unlink (2611)
4274 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4275 - free recovery state on server during a forced cleanup (3571)
4276 - unregister_reply for resent reqs (3063)
4277 - loop back devices mounting and status check on 2.6 (3563)
4278 - fix resource-creation race that can provoke i_size == 0 (3513)
4279 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4280 - more debugging for page-accounting assertion (3746)
4281 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4282 - avoid deadlock after precreation failure (3758)
4283 - fix race and lock order deadlock in orphan handling (3450, 3750)
4284 - add validity checks when grabbing inodes from l_ast_data (3599)
4286 - add /proc/.../recovery_status to obdfilter (3428)
4287 - lightweight CDEBUG infrastructure, debug daemon (3668)
4288 - change default OSC RPC parameters to be better on small clusters
4289 - turn off OST read cache for files smaller than 32MB
4290 - install man pages and include them in rpms (3100)
4291 - add new init script for (un)mounting lustre filesystems (2593)
4292 - run chkconfig in %post for init scripts (3701)
4293 - drop scimac NAL (unmaintained)
4295 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4298 - clean kiobufs before and after use (3485)
4299 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4300 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4301 - change init script to fail more clearly if not run as root (1528)
4302 - allow clients to reconnect during replay (1742)
4303 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4304 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4305 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4306 - fix cond_resched (3554)
4307 - only evict unfinished clients after recovery (3515)
4308 - allow bulk resend, prevent data loss (3570)
4309 - dynamic ptlrpc request buffer allocation (2102)
4310 - don't allow unlinking open directory if it isn't empty (2904)
4311 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4312 - remove extraneous obd dereference causing LASSERT failure (3334)
4313 - don't use get_cycles() when creating temp. files on the mds (3156)
4314 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4315 - handle EEXIST for set-stripe, set proper directory name (3336)
4317 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4318 - fix ksocknal_fmb_callback() error messages (2918)
4320 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4323 - don't copy lvb into (possibly NULL) reply on error (2983)
4324 - don't deref dentry after dput, don't free lvb on error (2922)
4325 - use the kms to determine writeback rpc length (2947)
4326 - increment oti_logcookies when osc is inactive (2948)
4327 - update client's i_blocks count via lvb messages (2543)
4328 - handle intent open/close of special files properly (1557)
4329 - mount MDS with errors=remount-ro, like obdfilter (2009)
4330 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4331 - don't use cancelling-locks' kms values (2947)
4332 - use highest lock extent for kms, not last one (2925)
4333 - don't dereference ERR_PTR() dentry in error handling path (3107)
4334 - fix thread race in portals_debug_dumplog() (3122)
4335 - create lprocfs device entries at setup instead of at attach (1519)
4336 - common AST error handler, don't evict client on completion race (3145)
4337 - zero nameidata in detach_mnt in 2.6 (3118)
4338 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4339 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4340 - initialize RPC timeout timer earlier for 2.6 (3219)
4341 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4342 - print nal/nid for unknown nid (3258)
4343 - additional checks for oscc recovery before doing precreate (3284)
4344 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4345 - don't crash in mdc_close for bad permissions on open (3285)
4346 - zero i_rdev for non-device files (3147)
4347 - clear page->private before handing to FS, better assertion (3119)
4348 - tune the read pipeline (3236)
4349 - fix incorrect decref of invalidated dentry (2350)
4350 - provide read-ahead stats and refine rpc in flight stats (3328)
4351 - don't hold journal transaction open across create RPC (3313)
4352 - update atime on MDS at close time (3265)
4353 - close LDAP connection when recovering to avoid server load (3315)
4354 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4355 - don't leak open file on MDS after open resend (3325)
4356 - serialize filter_precreate and filter_destroy_precreated (3329)
4357 - loop device shouldn't call sync_dev() for nul device (3092)
4358 - clear page cache after eviction (2766)
4359 - resynchronize MDS->OST in background (2824)
4360 - refuse to mount the same filesystem twice on same mountpoint (3394)
4361 - allow llmount to create routes for mounting behind routers (3320)
4362 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4363 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4364 - fix rare problem with rename on htree directories (3417)
4366 - allow default OST striping configuration per directory (1414)
4367 - fix compilation for qswnal for 2.6 kernels (3125)
4368 - increase maximum number of MDS request buffers for large systems
4369 - change liblustreapi to be useful for external progs like lfsck (3098)
4370 - increase local configuration timeout for slow disks (3353)
4371 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4373 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4376 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4377 - don't overwrite extent policy data in reply if lock was blocked (2901)
4378 - drop filter export grants atomically with removal from device (2663)
4379 - del obd_self_export from work_list in class_disconnect_exports (2908)
4380 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4381 - swab reply message in mdc_close, other PPC fixes (2464)
4382 - fix destroying of named logs (2325)
4383 - overwrite old logs when running lconf --write_conf (2264)
4384 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4385 - fix race in target_handle_connect (2898)
4386 - mds_reint_create() should take same inode create lock (2926)
4387 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4388 - don't close files for self_export to avoid uninitialized obd (2936)
4389 - allow MDS with the same name as client node (2939)
4390 - hold dentry reference for closed log files for unlink (2325)
4391 - reserve space for all logs during transactions (2059)
4392 - don't evict page beyond end of stripe extent (2925)
4393 - don't oops on a deleted current working directory (2399)
4394 - handle hard links to targets without a parent properly (2517)
4395 - don't dereference NULL lock when racing during eviction (2867)
4396 - don't grow lock extents when lots of conflicting locks (2919)
4398 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4401 - account for cache space usage on clients to avoid data loss (974)
4402 - lfsck support in lustre kernel code (2349)
4403 - reduce journal credits needed for BRW writes (2370)
4404 - orphan handling to avoid losing space on client/server crashes
4405 - ptlrpcd can be blocked, stopping ALL progress (2477)
4406 - use lock value blocks to assist in proper KMS, faster stat (1021)
4407 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4408 - recovery for initial connections (2355)
4409 - fixes for mds_cleanup_orphans (1934)
4410 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4411 - block all file creations until orphan recovery completes (1901)
4412 - client remove rq_connection from request struct (2423)
4413 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4414 - recovery timer race (2670)
4415 - mdc_close recovey bug (2532)
4416 - ptlrpc cleanup bug (2710)
4417 - mds timeout on local locks (2588)
4418 - namespace lock held during RPCs (2431)
4419 - handle interrupted sync write properly (2503)
4420 - don't try to handle a message that hasn't been replied to (2699)
4421 - client assert failure during cleanup after abort recovery (2701)
4422 - leak mdc device after failed mount (2712)
4423 - ptlrpc_check_set allows timedout requests to complete (2714)
4424 - wait for inflight reqs when ptlrpcd finishes (2710)
4425 - make sure unregistered services are removed from the srv_list
4426 - reset bulk XID's when resending them (caught by 1138 test)
4427 - unregister_bulk after timeout
4428 - fix lconf error (2694)
4429 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4430 - readahead locks pages, leaves pending causing memory pressure (2673)
4431 - increase OST request buffers to 4096 on large machines (2729)
4432 - fix up permission of existing directories in simple_mkdir (2661)
4433 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4434 - don't assign transno to errored transactions (2742)
4435 - don't delete objects on OST if given a bogus objid from MDS (2751)
4436 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4437 - if rq_no_resend, then timeout request after recovery (2432)
4438 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4439 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4440 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4441 - MDS recovery shouldn't delete objects that it has given out (2730)
4442 - if enqueue arrives after completion, don't clobber LVB (2819)
4443 - don't unlock pages twice when trigger_group_io returns error (2814)
4444 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4445 - don't write pages to disk if there was an error (1450)
4446 - don't ping imports that have recovery disabled (2676)
4447 - take buffered bytes into account when balancing socknal conn (2817)
4448 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4449 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4450 - remove little-endian swabbing of llog records (1987)
4451 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4452 - retry reposting request buffers if they fail (1191)
4453 - grow extent at grant time to avoid granting a revoked lock (2809)
4454 - lock revoke doesn't evict page if covered by a second lock (2765)
4455 - disable VM readahead to avoid reading outside lock extents (2805)
4457 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4458 - updated kernel patches for hp-2.4.20 kernel (2681)
4460 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4463 - fix truncated write corruption (2366)
4464 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4466 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4467 - ensure len(uuid) < 37 in lmc (1171)
4468 - fix ia64 OOPS in llog_test (2255)
4469 - zero end of page at obdfilter for partial page writes (2648)
4470 - don't leave stale dentries around after renames (bug 2428)
4471 - fix timeouts when evicting a client with a single lock held (2642)
4472 - set deadline for the initial HELLO message to drain (2634)
4473 - print out dotted-quad IP addresses in the socknal (2302)
4475 - additional debugging for MDS client eviction problem (2443)
4476 - fix mkfsoptions support for osts (2603, 2604)
4478 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4481 - add series for the vanilla 2.6.0 kernel
4482 - add series for the vanilla 2.4.24 kernel
4483 - add series for a cray x86/64 UL kernel drop
4484 - fix xattr patches for the vanilla 2.4.19 series
4486 - generate true UUIDs in lmc (1171)
4487 - have portals stack dumping break in UML (2466)
4488 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4489 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4490 - wake callback waiting threads on client eviction (2460)
4491 - Add --ptldebug and --subsystem to lmc (1719)
4492 - update assertion to allow safe interrupt allocation
4493 - set rq_no_resend for cancel requests (2432)
4494 - recalculate ptlrpcd timeout after resend (2494)
4495 - call vfs_rmdir when removing pending directories (2368)
4496 - fix renaming a file to itself (2429)
4497 - lmc creates a default one-stripe lov (2454)
4498 - expand procfs space to handle large clusters (2326)
4499 - increase UML stack to avoid overflow
4500 - update lconf's list of debug and subsystem masks
4501 - fix lfs find --obd (2510)
4502 - /proc tunable for disabling filter read caching (2591)
4503 - stop rpm packages from altering slapd.conf (2301)
4504 - disable nagle in the socknal under 0conf (2578)
4505 - choose mds inode size based on stripe count (2572)
4506 - fix kernel-source rpm problems (2516)
4508 - add --disable-doc to avoid pdf generation (2421)
4509 - update documentation, tests, type-os, comments
4510 - avoid format warnings on ia64
4511 - remove the TOE NAL
4512 - tiny code cleanups by removing unused fields
4514 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4517 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4518 - remove the most blatant lies from BUILDING file (2371)
4519 - change default debug level to reasonable production setting
4520 - reduce client side cache size to reduce cache flush time
4521 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4522 - make TCP zerocopy and pinger support enabled by default (2476)
4523 - sync writes completed after process exits caused crashes (2319)
4524 - maintain correct mount count on the MDS (2356)
4525 - backout 1557, because 2316 wasn't really fixed
4526 - better file I/O statistics gathering in /proc
4527 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4528 - another kernel patch to fix zero-copy TCP function export
4529 - don't take duplicate lock when processing re-sent getattr (2420)
4530 - lctl uses obd_self_export instead of creating new conn (2353)
4531 - MDS/OST recovery case which requires object creation asserted (2425)
4532 - move lfs from /usr/sbin to /usr/bin in packages
4533 - fix race between mds_client_add and mds_client_free (2417)
4534 - use kmalloc instead of slabs in portals (2430)
4535 - don't create duplicate records when a failover MDS is present (2442)
4536 - remove unnecessary mount age check (2332)
4537 - don't remove directory inodes from locks prematurely (2451)
4538 - don't break if MDS service name is the same as hostname (2103)
4539 - fix races in client write RPC generation when cache full (2482)
4541 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4544 - remove now-unused request->rq_obd (278)
4545 - if an allocation fails, print out how much memory we've used (1933)
4546 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4547 - add the "configurable stack size" patch to most series files (1256)
4548 - ability to write large log records, for 100+ OST configs (2306)
4549 - fix NULL deref when filter_prep fails (2314)
4550 - fix operator precedence error in filter_sync
4551 - dynamic allocation of socknal TX descriptors (2315)
4552 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4553 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4554 - ensure that CWARN messages in Portals always get to the syslog
4555 - __init/__exit are not for prototype decls (ldlm_init/exit)
4556 - x86-64 compile warning fixes
4557 - fix gateway LMC keyword conflict (2318)
4558 - fix MDS lock inversions in getattr/reint paths (1844)
4559 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4560 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4561 - DLM race condition prevented some lock evictions (2328)
4562 - ENOMEM detection and retry on socknal sends (2230)
4563 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4564 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4565 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4566 - better error messages when a client is rejected during recovery (1505)
4567 - avoid cancelling locks which were never granted, after failure (2330)
4568 - fix i_sem/journal inversion in mds_client_add (2333)
4569 - fix truncate/getattr lock cycle deadlock (2334)
4570 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4571 - fix two rare exit paths which could leak an l_lock() ref (2321)
4572 - fencepost error in MDS/OST orphan recovery (2226)
4573 - make log record alignment 8 bytes (1988)
4574 - lstripe now fails when requested offset > ost_count (2237)
4575 - ensure that all kernel series have a complete list.h (1607)
4576 - fix crashes in special-file operations (2316)
4577 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4579 - allow configurable automake binary, for testing new versions
4580 - small update to the lfs documentation
4582 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4584 * fix negative export reference count in fsfilt_sync (2312)
4586 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4587 * release candidate 0.9.1
4589 - orphans are moved into the PENDING directory for possible recovery
4590 - replayed opens now open by fid for orphan/rename safety (1042)
4591 - last close of an orphan inode generates a transno (683)
4592 - chdir() and mount() now pin the directory entry (1020)
4593 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4594 - discard very old requests without processing them (1502)
4595 - remove some common, well-understood CERRORs (1505)
4596 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4597 - clear "grant" flags in OST replies until OST grant code lands (1644)
4598 - fix read performance by not clobbering i_blksize on client (1598)
4599 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4600 - make LRU size a /proc tunable, clears locks when reduced (707)
4601 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4602 - server locks take references on exports now (1558)
4603 - build fixes for 2.4.20-rh trees (1663)
4604 - return an error from lov_create if all OSCs are inactive (1751)
4605 - fix import levels when a reconnect happens without a timeout (1597)
4606 - exit early from mds_open if we get a lookup error (1749)
4607 - partial page read at EOF wouldn't wait for disk before sending (1642)
4608 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4609 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4610 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4611 - write/truncate lock inversion (1639)
4612 - Don't auto-load obdclass, portals modules during cleanup (1495)
4613 - fix timestamps from jumping to "now" (1763)
4614 - extra journal assertions (1648)
4615 - add an extra multiunlink test (1771)
4616 - fix read_record/write_record API (1776)
4617 - fix leak of offset_extent, possible incorrect i_size later (1772)
4618 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4619 - Debugging for the kqswnal_get_idle_tx problems (1820)
4620 - Allow recovery to be attempted multiple times (1536)
4621 - Write out MDS last_rcvd file after it is first created (1600)
4622 - Fix tx_descriptor leak in failed transmit situations (1827)
4623 - ext3 journaling fixes for assertion failure after IO error (1871)
4624 - class_export_put() on freed export after completion AST error (1896)
4625 - Fix revalidate looping in VFS (1322)
4626 - Don't access a freed export during MDS_REINT timeout (1521)
4627 - Add open-unlink recovery support on the MDS (1673,1764)
4628 - Return an error if no MDS data was read from last_rcvd (1946)
4629 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4630 - Don't setup a disk device that doesn't match exported UUID (317)
4631 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4632 - avoid committing NULL handle in force close
4633 - local.sh is now a one-stripe LOV configuration
4634 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4635 - fix ext3 htree duplicate directory entry corruption (1516)
4636 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4637 - update to new LOV EA format (2097)
4638 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4640 2003-06-15 Phil Schwan <phil@clusterfs.com>
4643 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4644 - per-import/export recovery handling (958, 931, 959)
4645 - multiple last-rcvd slots, for serving multiple FSes (949)
4646 - connections are again shared between multiple imp/exports (963, 964)
4647 - "umount -f" would hang if any requests needed to be sent (393, 978)
4648 - avoid pinning large req buffer by copying for queued messages (989)
4649 - add "uuid" to "lctl device" command to help upcalls (991)
4650 - "open" RPCs with transnos would confuse recovery counters (1037)
4651 - do proper endian conversion of all wire messages (288, 340, 891)
4652 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4653 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4654 - fix ptlrpc_connection leak in target_handle_connect (1174)
4655 - fix import refcounting bug in OST and MDS cleanup (1134)
4656 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4657 - fix very unlikely obd_types race condition (501)
4658 - remove osc_open hack for echo_client (1187)
4659 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4660 - a failure in read_inode2 leads to deadlock (1139)
4661 - cancel ack-locks as soon as transaction is committed (1072)
4662 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4663 - make sure to commitrw after any preprw to avoid deadlock (1162)
4664 - failing to execute a file in a lustre FS would lock inode (1203)
4665 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4666 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4667 - obd_uuid2tgt would walk off the end of the list (1255)
4668 - on IA64 the osc would give portals incorrect bulk size (1258)
4669 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4670 - fix lock inversion caused by new llite matching code (1282)
4671 - limit the number of dirty pages on a client to 10MB (1286)
4672 - timed out locks were not being corrected cancelled (1289)
4673 - fix O_DIRECT above 4GB on IA-32 (1292)
4674 * major user-visible changes
4675 - fail out/fail over policy now controlled by the upcall (993)
4677 - add OBD_PING to check server availability and failure (954)
4678 - lustre messages are now sent in sending host order (288, 340, 891)
4679 - add eadatalen to MDS getattr reply (340)
4680 - OST read replies may contain second buffer, with per-page status (593)
4682 2003-03-11 Phil Schwan <phil@clusterfs.com>
4685 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4686 - failed open()s could cause deadlock; fixed (867, 869)
4687 - stop cancelling OST locks when files are closed (481)
4688 - overlapping XID spaces caused network corruption (851, 853)
4689 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4690 - setattr_raw added, to do single-RPC, server-side setattrs
4691 - lmc/lconf syntax change for OST UUIDs
4692 - fix crashy race condition between ptlrpc_free_req and osc_close
4693 - don't use request in mdc_enqueue if we hit a timeout (889)
4694 - don't set the inode i_size for regular files from the MDS (896)
4695 - handle out of order completion AST (842)
4696 - don't LBUG if a lock request times out after receiving AST (913)
4697 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4698 - if a bad lock AST arrives, send an error instead of dropping entirely
4699 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4700 - fix leak in bulk IO when only partially completed (899, 900, 926)
4701 - fix O_DIRECT for ia64 (55)
4702 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4703 - C-z after timeout could hang a process forever; fixed (977)
4705 - client-side I/O cache (678, 924, 929, 941, 970)
4707 - READPAGE and SETATTRs which don't take server-side locks get
4710 2003-02-11 Phil Schwan <phil@clusterfs.com>
4713 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4714 - Fix fsfilt_extN_readpage() to read a full page of directory
4715 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4716 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4717 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4718 - Most intent operations were removed, in favour of a new RPC mode
4719 that does a single RPC to the server and bypasses most of the VFS
4720 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4721 - Aggressively cancel local locks on DLM servers
4722 - mds_reint_unlink sends EA to the client if it's the last nlink.
4723 client uses that EA to unlink OST objects.
4724 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4725 - recursive symlinks were fixed (439)
4726 - fixed NULL deref in DEBUG_REQ
4727 - filter_update_lastobjid no longer calls sync, which annoyed extN
4728 - fixed multi-client small-writes to a single file problem (445)
4729 - fixed mtime updates during file writes (607)
4730 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4731 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4732 - fixed Linux OST size reporting problem (444, 656)
4733 - OST now updates object mtime with writes or setattr (607, 619)
4734 - client verifies file size before zeroing page past EOF (445)
4735 - OST now writes last allocated objid to disk with allocation (108)
4736 - LOV on echo now works (409)
4738 - mds_reint_unlink sends a new buffer, with the EA included. this
4739 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4740 set if a regular file was being unlinked, and it was the last link
4741 - use PtlGet from the target for bulk writes (315)
4742 - OST now updates object mtime with writes or setattr (607, 619)
4743 - LDLM now has a grant-time callback to revalidate locked items, if
4745 - Many MDS operations were reorganized to combat race conditions
4747 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4748 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4750 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4753 - Fully reactivate OST imports after reconnection (512, others)
4754 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4755 - More graceful error handling for truncating on dead OST (515)
4756 - Don't error out unless we're actually accessing dead stripes (474)
4757 - Fix garbage sizes when stripes are missing (410)
4758 - LRU counters were broken, causing constant lock purge (433, 432)
4759 - garbage on read from stripes with failed OSTs (441)
4760 - mark OSCs as active before reconnecting during recovery (438)
4761 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4762 - lfind did not preserve OST order in output (443)
4763 - symlinks cause hung clients, incorrect data (439)
4764 - stop dereferencing request after dropping refcount (457)
4765 - don't LASSERT(spin_is_locked) on non-SMP (455)
4766 - fixes for many rename() bugs
4767 - fstat didn't correctly synchronize attributes (399)
4768 - server must handle lock cancellation during blocking AST prep (487)
4769 - bulk descriptors were free()d too soon (511)
4770 - fix paths in lconf, which would load incorrect modules (451, 507)
4771 - fix confusing lconf 'host not found' error message (386)
4772 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4773 - fix race condition in mdc_blocking_ast() for inode access (526)
4774 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4775 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4776 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4777 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4778 - fix lstripe in several ways (kernel side) (527)
4779 - fix request leak in ldlm_cli_enqueue (262)
4780 - incorrect OSC was marked inactive after OST failure
4781 - call mds_fs_cleanup before unmounting filesystem (524)
4782 - fix races between taking ns_lock and ldlm_lock_change_resource
4783 - fix races updating LOV export open file list
4784 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4785 - fix recovery NULL deref in ldlm_cli_cancel_unused
4786 - fix some DLM races by using new hash table for lock handles (419)
4787 - permit the client to specify desired inodes, at replay
4788 - duplicate requests when we queue them for replay reintegration
4789 - fix last_rcvd offset calculation
4790 - sync after each recovered transaction, so we always make progress
4791 - never, not always, ERESTART requests without transnos
4792 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4794 - skip replay if the MDS says that the client is already connected
4795 - don't check for a recovery-enabled export to match lctl's UUID
4796 - don't INC_USE_COUNT for phantom exports
4797 - don't crash when cleaning up phantom exports (567)
4798 - don't double-finish or set replay data for errored mdc_open requests
4799 - abort requests when they time out, so we don't get old replies
4800 - send/receive replies for AST messages again
4801 - if the client says that it doesn't have the lock, cancel it on the
4803 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4804 mark it as destroyed, it will all work out when decref is called
4805 - fix module use counts (22, 581)
4807 - ASTs now expect a reply (server cancels lock on error reply)
4809 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4812 - fix many simultaneous client startup (392)
4813 - fix dentry->d_it clobbering
4814 - credentials weren't being shipped for readdir/getattr operations
4815 - remove invalid assertions triggered during some concurrent MD
4817 - proper Lustre versions added (336, 389)
4818 - fix memory leak for create error case (398)
4819 - fix LOV locking bug that would get cli/srv out of sync
4820 - fix echo client over LOV (409)
4821 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4822 - fix double-O_EXCL intent crash (424)
4823 - avoid sending multiple lock CANCELs (352)
4825 - MDS can do multi-client recovery (modulo bugs in new code)
4827 - many updates, edits, cleanups
4829 2002-11-18 Phil Schwan <phil@clusterfs.com>
4832 - fix null d_it dereference (346)
4833 - fix full OST/dbench hang (333)
4834 - fix permission problem with file removal (286)
4835 - fix removal of OSCs from LOV when they fail
4836 - fix NULL deref during bulk timeout (214)
4837 - fix problems related to multiple filesystems on one MDS (241)
4838 - fixed serious subtle metadata locking bugs
4839 - free locks on clients when inodes are removed due to memory
4841 - fix inode pointer in lock data (285)
4842 - partial support for multiple MDS on a single host (241)
4843 - data locks weren't cancelled at clear_inode time (290, 311)
4844 - intent locks could lead to unbounded lock growth (205)
4845 - added a maximum lock count, an LRU list, and a flusher
4846 - fix multiple rename (365)
4847 - properly abstracted the echo client
4848 - OSC locked 1 byte too many; fixed
4849 - rewrote brw callback code:
4850 - fixed recovery bugs related to LOVs (306)
4851 - fixed too-many-pages-in-one-write crash (191)
4852 - fixed (again) crash in sync_io_timeout (214)
4853 - probably fixed callback-related race (385)
4855 - Add capability to MDS protocol
4856 - LDLM cancellations and callbacks on different portals
4858 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4861 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4862 - defer open object destruction to close time (601981, 138)
4863 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4864 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4865 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4866 - don't call obd_setattr in ll_file_release for destroyed objects
4867 * protocol change to lustre_msg: move |version| and add |flags|
4868 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4869 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4870 * added replay of create, unlink, link and rename operations during
4871 MDS failover; recovery should be much more robust now
4872 * remove failed OSCs from LOVs (only lov_create uses this so far)
4873 * the lustre-HOWTO was brought (more) up to date (582544)
4875 2002-10-23 Phil Schwan <phil@clusterfs.com>
4878 - in-use dentries weren't being reused properly (617851)
4879 - prevent multiple LDLM setup (599178)
4880 - fix LOV size calculations for truncate (617853)
4881 - fix client handling of MDS intent errors (POSIX)
4882 - fix permission bug in lovstripe.c test (624321)
4883 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4884 - truncate past end of file could corrupt data
4885 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4886 - a race in recovery could return ETIMEDOUT to apps (623947)
4887 - building outside the source directory was fixed
4888 * the lustre-HOWTO was brought (more) up to date (582544)
4889 * major progress was made on recovery functionality
4891 2002-10-10 Phil Schwan <phil@clusterfs.com>
4894 - recovery deadlock fix
4895 - rm -rf causes LBUG fix (617817)
4896 - file open by multiple tasks fix (618962)
4897 - directory permissions bugs (602707 and 620007)
4898 - journal_stop fixed with locking (611313)
4899 - O_APPEND failures resolved (618273, perhaps 614459)
4900 - lconf PATH fix (619770)
4901 - IA64 build fix (621450)
4902 - RPC buffer sizes scale with amount of memory
4904 2002-10-01 Phil Schwan <phil@clusterfs.com>
4907 - locks would be cancelled without throwing away data pages,
4908 resulting in inconsistent data (605627)
4909 - inode attributes were not always being refreshed (605627, 612449)
4910 - lconf now continues to cleanup after lctl reports an error
4911 - MDS now enforces user permissions (602707)
4912 - lprocfs cleanup fixed, but not yet enabled (614157)
4913 - fixed infinite server hang, should a client not respond to an AST
4914 - avoid going into recovery if user calls readlink() with a buffer
4915 that's too small (613941)
4916 - AST RPCs no longer require replies (614867) -- this may be changed
4917 - don't crash server if client sends an IOV that's too big (611336)
4918 - fixed lock conversion deadlock (611892)
4919 - fixed the following of symlinks (614622)
4920 * recovery: the server can remove locks from a client that dies, other
4921 clients can make progress
4922 * more extN patch fixes
4923 * compile-time configurable ptlrpc buffer allocations
4925 - collaborative read cache document
4926 - Lustre Lite Performance CDR document-in-progress
4928 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4931 - fix typo in patch-2.4.18
4933 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4936 - clear ptlrpc request each time in handle_incoming_request()
4937 - unlink of files now destroys the object on the OST
4939 2002-09-19 Peter Braam <braam@clusterfs.com>
4941 * add hard link support
4942 * change obdfile creation method
4943 * kernel patch changed
4945 2002-09-19 Peter Braam <braam@clusterfs.com>
4948 - stack overflow bug in extN fixed
4950 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4952 * documentation updates
4953 - add man pages for config tools
4954 - update tests/README to describe testing with new config tools
4955 - finish metadata API descriptions
4956 * bug fixes and cleanups
4957 - statfs workaround for 16TB limit
4958 - LOV stripe allocation improved, can stripe on subset of OSTs
4959 - LOV file size/IO offset was wrong for files > 4GB in size
4960 - object EA data was being dropped, caused files to be unreadable
4961 - memory overflow with non-LOV OST caused memory corruption
4962 - fixed regression tests to work with new config tools, obdfilter
4963 - fixed bug when directory size became larger than 1 block
4964 - fixed bug (for single client case) when PWD was deleted
4965 - invalidate local directory pages when doing intent-based ops
4966 - avoid LDLM oops when lock callback contained bad data
4968 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4970 * documentation updates
4971 * bug fixes and cleanups
4972 - configuration tools
4975 - 64-bit compile warnings
4976 - 64-bit internal statfs data
4978 * test_brw on persistent OST devices
4980 * lprocfs (disabled)
4982 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4984 * documentation updates
4985 * bug fixes and cleanups
4986 * configuration tools
4988 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4991 * many small fixes to 0_5_4
4992 * io/network handling
4993 * thinkos in MDS operations
4995 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4998 * crucial basic fixes to 0.5.3
4999 * IOR, Iozone work over Elan
5002 2002-08-07 Phil Schwan <phil@clusterfs.com>
5003 * version 0_5_3, our first alpha
5004 * we use the new Portals iovs
5005 * documentation updates
5006 * bug fixes and cleanups
5007 * small changes in the DLM wire protocol
5009 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5010 * version 0_5_1 with some initial stability,
5011 * locking on MD and file I/O.
5012 * documentation updates
5013 * several bug fixes since 0.5.0
5014 * small changes in wire protocol
5016 2002-07-18 Phil Schwan <phil@clusterfs.com>
5018 * delivered as Lustre Light Alpha
5019 * fixed a crash after handling invalid MDS requests
5020 * fixed directory pages for architectures with non-4k pages sizes
5022 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5023 * release version v0_4_4
5024 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5026 * Fixes a number of interruption problems with OST operations.
5027 * Update documentation for portals header changes
5028 * Move all wire protocol structs/defines to lustre_idl.h
5029 * Fixes symlink length bug.
5030 * Add tcpdump to repository.
5032 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5033 * release version v0_4_3
5034 * Fixes statfs for inodes on extN.
5035 * Fixes bug in runtests which would delete /etc/hosts.
5036 * Use 64-bit object IDs wherever possible (not into VFS though)
5037 Remove ost_get_info, which is unused by lustre, and out of date.
5039 2002-07-03 Peter Braam <braam@clusterfs.com>
5040 * release version v0_4_2 Fixes a lookup error (type not passed)
5041 * move forward to head of Portals
5042 * move forward to latest Lustre kernel
5044 2002-06-25 Peter Braam <braam@clusterfs.com>
5045 * release version v0_4_1. Hopefully stable on single node use.