1 tbd Sun Microsystems, Inc.
4 2.6.16.54-0.2.5 (SLES 10),
5 2.6.18-53.1.21.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: 1.40.7-sun3
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
17 Description: add recount protection for osc callbacks, so avoid panic on shutdown
21 Description: sanity test 65a fails if stripecount of -1 is set
22 Details : handle -1 striping on filesystem in ll_dirstripe_verify
27 Description: ASSERTION(CheckWriteback(page,cmd)) failed
28 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
31 Severity : enhancement
33 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
37 Description: do not process already freed flock
38 Details : flock can possibly be freed by another thread before it reaches
39 to ldlm_flock_completion_ast.
43 Description: LBUG during stress test
44 Details : Need properly lock accesses the flock deadlock detection list.
48 Description: oops in page fault handler
49 Details : kernel page fault handler can return two special 'pages' in error case, don't
50 try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
54 Description: timeout with invalidate import.
55 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which should be
56 handled by ptlrpcd. This produce long age waiting and -ETIMEOUT
57 ptlrpc_invalidate_import and as result LASSERT.
59 Severity : enhancement
61 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
65 Description: enable MGS and MDT services start separately
66 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
67 a co-located MGS without starting the MGS, which is a complement
68 to 'nosvc' mount option.
72 Frequency : after recovery
73 Description: precreate to many object's after del orphan.
74 Details : del orphan st in oscc last_id == next_id and this triger growing
75 count of precreated objects. Set flag LOW to skip increase count
76 of precreated objects.
80 Frequency : rare, on clear nid stats
81 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
82 Details : when clean nid stats sometimes try destroy live entry,
83 and this produce panic in free.
87 Description: Stack overflow during MDS log replay
88 ease stack pressure by using a thread dealing llog_process.
92 Description: wait until IO finished before start new when do lock cancel.
93 Details : VM protocol want old IO finished before start new, in this case
94 need wait until PG_writeback is cleared until check dirty flag and
95 call writepages in lock cancel callback.
97 Severity : enhancement
99 Description: using special macro for print time and cleanup in includes.
103 Description: mds_mfd_close() ASSERTION(rc == 0)
104 Details : In mds_mfd_close(), we need protect inode's writecount change
105 within its orphan write semaphore to prevent possible races.
109 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
113 Frequency : rare, on shutdown ost
114 Description: don't hit live lock with umount ost.
115 Details : shrink_dcache_parent can be in long loop with destroy dentries,
116 use shrink_dcache_sb instead.
120 Description: don't panic with use echo client
121 Details : echo client pass NULL as client nid pointer and this produce null
126 Description: fix build on ppc32
127 Details : compile code with -m64 flag produce wrong object file for ppc32.
131 Description: add message levels for liblustreapi
135 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
136 Details : Move the 'good_osts' check before the 'total_bavail' check. This
137 will result in an -EAGAIN and in the exit call path we call
138 alloc_rr() which will with increasing aggressiveness attempt to
139 aquire precreated objects on the minimum number of required OSCs.
143 Description: Use old size assignment to avoid deadlock
144 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
145 the scheduling while holding a spinlock. We do not need locking
146 for size in ll_update_inode() because size is only updated from
147 the MDS for directories or files without objects, so there is no
148 other place to do the update, and concurrent access to such inodes
149 are protected by the inode lock.
153 Description: resolve "_IOWR redefined" build error on SLES10
157 Description: dump the memory debugging after all modules are unloaded to
158 suppress false negative in conf_sanity test 39
160 Severity : enhancement
162 Description: build kernel-ib packages for OFED 1.3 in our release cycle
167 Description: fix SLES kernel versioning
168 Details : the kernel version for our SLES 10 kernel did not include a "-"
169 before the "smp" at the end. while this was not a problem in
170 general, it did mean that software trying to use the kernel
171 version to try to detect a vendor specific kernel would fail.
172 this was most evident by the OFED build scripts.
176 Description: Don't update lov_desc members until making sure they are valid
177 Details : When updating lov_desc members via proc fs, need fix their
178 validities before doing the real update.
182 Description: don't put request into delay list while invalidate in flight.
183 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
184 import in flight. this produce timeout for invalidate and sometimes
185 can cause stale data.
189 Frequency : on ppc only
190 Description: not convert ost objects for directory because it's not exist.
191 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
195 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
197 Severity : enhancement
199 Description: Update to support 2.6.22.14 vanilla kernel.
203 Frequency : rare, on recovery
204 Description: read procfs can produce deadlock in some situation
205 Details : Holding lprocfs lock which send rpc can produce block for destroy
206 obd objects and this also block reconnect with -EALREADY. This isn't
207 fix all lprocfs bugs - but make it rare.
209 Severity : enhancement
211 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
214 Frequency : frequent on X2 node
216 Description: mdc_set_open_replay_data LBUG
217 Details : Set replay data for requests that are eligible for replay.
221 Description: lustre_mgs: operation 101 on unconnected MGS
222 Details : When MGC is disconnected from MGS long enough, MGS will evict the
223 MGC, and late on MGC cannot successfully connect to MGS and a lot
224 of the error messages complaining that MGS is not connected.
228 Frequency : on network error
229 Description: panic with double free request if network error
230 Details : mdc_finish_enqueue is finish request if any network error ocuring,
231 but it's true only for synchronus enqueue, for async enqueue
232 (via ptlrpcd) this incorrect and ptlrpcd want finish request
235 Severity : enhancement
237 Description: client-side metadata stat-ahead during readdir(directory readahead)
238 Details : perform client-side metadata stat-ahead when the client detects
239 readdir and sequential stat of dir entries therein
242 Frequency : on start mds
244 Description: Implement get_info(last_id) in obdfilter.
247 Frequency : occasional
249 Description: Correctly check stale fid, not start epoch if ost not support SOM
250 Details : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
251 overwrited with new generated fid, but mds can anwer with one of these
252 two fids and both is not stale. setattr incorectly start epoch and
253 assume will be called done_writeting, but without SOM done_writing
257 Frequency : rare, depends on device drivers and load
259 Description: MDS or OSS nodes crash due to stack overflow
260 Details : Code changes in 1.8.0 increased the stack usage of some functions.
261 In some cases, in conjunction with device drivers that use a lot
262 of stack the MDS (or possibly OSS) service threads could overflow
263 the stack. One change which was identified to consume additional
264 stack has been reworked to avoid the extra stack usage.
267 Frequency : occasional
269 Description: Do not fail import if osc_interpret_create gets -EAGAIN
270 Details : If osc_interpret_create got -EAGAIN it immediately exits and
271 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
272 oscc_has_objects and see OSC has no objests and call
273 oscc_internal_create to resend create request.
275 Severity : enhancement
277 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
279 Severity : enhancement
281 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
284 Frequency : very rare
286 Description: Fix replay if there is an un-replied request and open
287 Details : In some cases, older replay request will revert the
288 mcd->mcd_last_xid on MDS which is used to record the client's
291 Severity : enhancement
293 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
295 Severity : enhancement
297 Description: Add rhel5 support to HEAD.
299 Serverity : enhancement
301 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
306 Description: Don't allow skipping OSTs if index has been specified.
307 Details : Don't allow skipping OSTs if index has been specified, make locking
308 in internal create lots better.
312 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
313 Details : access to bitfield in structure is always rounded to long
314 and this produce problem with not atomic change any bit.
318 Description: Lustre make rpms failed.
319 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
320 builds Lustre rpms from tarball.
323 Frequency : rare on shutdown ost
325 Description: If llog cancel was not send before clean_exports phase, this can
326 produce deadlock in llog code.
327 Details : If llog thread has last reference to obd and call class_import_put
328 this produce deadlock because llog_cleanup_commit_master wait when
329 last llog_commit_thread exited, but this never success because was
330 called from llog_commit_thread.
334 Description: allow userland application know is lost one of stripes.
335 Details : fill lvb_blocks with error code on ost and return it to
336 application if error flag found.
340 Description: NULL lov_tgts causing MDS oops
341 Details : more safe checks for NULL lov_tgts for avoid oops.
343 Severity : enhancement
345 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
349 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
350 Details : If an OST is down the MDS will hang indefinitely in
351 obd_statfs_rqset() waiting for the statfs data. While for
352 MDS QOS usage of statfs, it should not stuck in waiting.
354 Severity : enhancement
356 Description: remote_acl support
357 Details : Support ACL-based permission check for remote user.
358 Support setfacl/getfacl for remote user with the utils
359 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
360 the system "{s,g}etfacl" utils.
362 Severity : enhancement
364 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
366 Severity : enhancement
368 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
372 Description: make mgs_setparam() handle fsname containing dash
373 Details : fsname containing a dash does not work with lctl conf_param
375 Severity : enhancement
377 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
379 Severity : enhancement
381 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
385 Frequency : rare, at shutdown
386 Description: access already free / zero obd_namespace.
387 Details : if client_disconnect_export was called without force flag set,
388 and exist connect request in flight, this can produce access to
389 NULL pointer (or already free pointer) when connect_interpret
390 store ocd flags in obd_namespace.
394 Frequency : only at startup
395 Description: not alloc memory with spinlock held.
396 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
397 if any spinlock held.
399 Severity : enhancement
401 Description: make lustre randomly fail allocating memory
402 Details : Make lustre randomly failed allocating memory for testing purpose.
404 Severity : enhancement
406 Description: lost problems with lov objid file
407 Details : Fixes some scability and access to not inited memory problems
408 in work with lov objdid file.
413 Description: lfs find does not continue on file error
414 Details : Continue other files processing when a file/dir is absent.
418 Description: Inconsistent usage of lustre_pack_reply()
419 Details : Standardize the usage of lustre_pack_reply() such that it
420 always generate a CERROR on failure.
423 Frequency : occasional
425 Description: MDS hang or stay in waiting lock
426 Details : If client receive lock with CBPENDING flag ldlm need send lock
427 cancel as separate rpc, to avoid situation when cancel request
428 can't processed due all i/o threads stay in wait lock.
432 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
436 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
440 Description: Update to SLES9 kernel-2.6.5-7.287.3.
444 Description: Update to RHEL4 latest kernel.
446 Severity : enhancement
448 Description: Build SLES10 patchless client fails
449 Details : The configure was broken by run ./configure with
450 --with-linux-obj=.... argument for patchless client. When the
451 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
452 can't search header adequately. Use absolute path such as
453 -I($LINUX)/include instead.
457 Description: interrupt oig_wait produce painc on resend.
458 Details : brw_redo_request can be used for resend requests from ptlrpcd and
459 private set, and this produce situation when rq_ptlrpcd_data not
460 copyed to new allocated request and triggered LBUG on assert
461 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
462 ptlrpcd set if request is changed and can be safety changed to use
467 Description: LASSERT_{REQ,REP}SWAB macros are buggy
468 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
469 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
470 these macros and replace them with their difinitions instead.
474 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
475 Details : Update which_patch & target file for SLES10 latest kernel.
478 Description: Remove client patches from SLES 10 kernel.
479 Details : This causes SLES 10 clients to behave as patchless clients
480 even on a Lustre-patched (server) kernel.
482 Severity : enhancement
484 Description: self-adjustable client's lru lists
485 Details : use adaptive algorithm for managing client cached locks lru
486 lists according to current server load, other client's work
487 pattern, memory activities, etc. Both, server and client
488 side namespaces provide number of proc tunables for controlling
491 Severity : enhancement
493 Description: light-weight GSS support
494 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
495 and reduce performance overhead.
497 Severity : enhancement
499 Description: Linux keyring support
500 Details : Support using service of Linux keyring for Lustre GSS internal
501 context refresh/cache mechanism.
505 Description: Fix errors in lfs documentation
506 Details : Fixes man pages
510 Description: don't use GFP_* in generic Lustre code.
511 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
515 Description: obdclass is limited by single OBD_ALLOC(idarray)
516 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
520 Description: warnings with build patchless client with vanila 2.6.19 and up
521 Details : change old ctl_table style and replace ctl_table/ctl_table_header
522 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
526 Frequency : Only for SLES
527 Description: sanity_quota fail test_1
528 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
529 file, which makes TSTID[2] inunique.
533 Frequency : Only for SLES9
534 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
535 Details : Update target/ChangeLog/which_patch .
539 Frequency : Only for test_10 in sanity-quota.sh
540 Description: a bug in quota test code
541 Details : checking a test flag in a wrong place
545 Frequency : Only for test_13 in sanity-quota.sh
546 Description: a bug in quota test script
547 Details : shouldn't check overfull allocated quota
551 Description: jbd statistics
552 Details : Port older jbd statistics patch for sles10
554 Severity : enhancement
556 Description: Add printing inode info into message about error in writepage.
561 Description: reply_lock_interpret crash due to race with it and lock cancel.
562 Details : Do not replay locks that are being cancelled. Do not reference
563 locks by their address during replay, just by their handle.
567 Frequency : When flocks are used.
568 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
569 Details : Flock locks might destroy just granted lock if it could be merged
570 with another existing flock, this is done in completion handler,
571 so teach ldlm_cli_enquque_fini that this is a valid case for
577 Description: Oops in read and write path when failing to allocate lock.
578 Details : Check if lock allocation failed and return error back.
582 Description: lstripe command fails for valid OST index
583 Details : The stripe offset is compared to lov->desc.ld_tgt_count
584 instead of lov->desc.ld_active_tgt_count.
588 Description: sanity.sh failed test 103
589 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
594 Description: df doesn't work properly if diskfs blocksize != 4K
595 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
599 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
600 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
602 Severity : enhancement
604 Description: Adaptive timeouts
605 Details : RPC timeouts adapt to changing server load and network
606 conditions to reduce resend attempts and improve recovery time.
610 Description: llapi_file_create() does not allow some changes
611 Details : add llapi_file_open() that allows specifying the mode and
612 open flags, and also returns an open file handle.
616 Description: merge and cleanup kernel patches.
617 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
621 Description: Add journal checksum support.(Kernel part)
622 Details : The journal checksum feature adds two new flags i.e
623 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
624 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
625 indicates that the commit block contains the checksum for
626 the blocks described by the descriptor blocks. Now commit
627 record can be sent to disk without waiting for descriptor
628 blocks to be written to disk. This behavior is controlled
629 using JBD2_FEATURE_ASYNC_COMMIT flag.
633 Description: OSS needs mutliple precreate threads
634 Details : Add ability to start more than one create thread per OSS.
638 Description: Sanity.sh test_65k failed: setstripe should have succeeded
639 Details : execute lfs setstripe on client
643 Description: mds_obd_create error creating tmp object
644 Details : When the user sets quota on root, llog will be affected and can't
645 create files and write files.
648 Frequency : Always on ia64 patchless client, and possibly others.
650 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
651 Details : This allows the patchless client to be loaded on architectures
656 Description: RedHat Update kernel for RHEL5
657 Details : Add the kernel config file for RHEL5.
661 Description: RedHat Update kernel for RHEL5
662 Details : Modify the kernel config file more closer RHEL5.
666 Description: Build failure against Centos5 (RHEL5)
667 Details : Define PAGE_SIZE when it isn't present.
671 Description: test_7 on sanity-quota.sh can't be used on separate machines
672 Details : add facet to handle it
676 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
677 Details : with NFS, the anon dentry's parent was set to itself in
678 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
679 corresponding dentry other than use rec->ur_name.
681 Severity : enhancement
683 Description: lfs setstripe enhancement
684 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
688 Description: enable data checksumming by default
689 Details : enable checksum by default, allow --disable-checksum
690 configure option and "-o nochecksum" mount option. Checksums
691 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
695 Description: lustre support for RHEL5
696 Details : Add support for RHEL5.
700 Description: Client eviction due to failover config
701 Details : after a connection loss, the lustre client should attempt to
702 reconnect to the last active server first before trying the
703 other potential connections.
707 Description: when mds and osts use different quota unit(32bit and 64bit),
708 quota will be released repeatly.
709 Details : void sending multiple quota reqs to mds, which will keep the
710 status between the reqs.
714 Description: osts not allocated evenly to files
715 Details : change the condition to increase offset_idx
718 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
720 Description: Data corruption for OSTs that are formatted larger than 2TB
722 Details : When generating the bio request for lustre file writes the
723 sector number would overflow a temporary variable before being
724 used for the IO. The data reads correctly from Lustre (which
725 will overflow in a similar manner) but other file data or
726 filesystem metadata may be corrupted in some cases.
730 Description: Tune the kernel for good SCSI performance.
731 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
732 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
737 Description: rewrite ext2-derived code in obdclass/uuid.c
738 Details : rewrite inherited code (uuid parsing code from ext2 utils)
739 from scratch preserving functionality.
743 Description: "lfs find -obd UUID" prints directories
744 Details : "lfs find -obd UUID" will return all directory names instead
745 of just file names. It is incorrect because the directories
746 do not reside on the OSTs.
750 Description: use i_size_read and i_size_write in 2.6 port
751 Details : replace inode->i_size access with i_size_read/write()
754 Frequency : when removing large files
756 Description: scheduling issue during removal of large Lustre files
757 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
758 It causes scheduling issues when removing large files (17TB in the
762 Frequency : only with liblustre clients on XT3
764 Description: evictions taking too long
765 Details : allow llrd to evict clients directly on OSTs
768 Frequency : only on ppc
770 Description: /proc/fs/lustre/devices broken on ppc
771 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
772 We should make sure the type of 'index' is loff_t and then cast
773 explicitly as needed below. Do not assign an explicitly cast
777 Frequency : only for rhel5
779 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
780 Details : Modify the target file & which_kernel.
782 Severity : enhancement
784 Description: omit set fsid for export NFS
785 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
786 as FS_REQUIRES_DEV for avoid problems with generate fsid.
790 Frequency : Always, for kernels after 2.6.16
791 Description: Fix warning idr_remove called for id=.. which is not allocated.
792 Details : Last kernels save old s_dev before kill super and not allow
793 to restore from callback - restore it before call kill_anon_super.
797 Description: buffer overruns could theoretically occur
798 Details : llapi_semantic_traverse() modifies the "path" argument by
799 appending values to the end of the origin string, and a
800 overrun may occur. Adding buffer overrun check in liblustreapi.
804 Description: Fix error on 'ls .' at the top of the Lustre mount.
805 Details : Don't revalidate dentry if it is a root dentry.
809 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
810 Details : Modify vm-tunables-rhel4.patch.
814 Description: Kernel config for 2.6.18-vanilla.
815 Details : Modify targets/2.6-vanilla.target.in.
816 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
817 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
818 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
819 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
823 Description: improve handling recoverable errors
824 Details : if request processig with error which can be recoverable on server
825 request should be resend, otherwise page released from cache and
830 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
831 Details : Modify target file & which_patch.
832 A flaw was found in the IA32 system call emulation provided
833 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
834 value could be stored in the %RAX register, which could trigger an
835 out-of-bounds system call table access. An untrusted local user
836 could exploit this flaw to run code in the kernel
837 (ie a root privilege escalation). (CVE-2007-4573).
841 Description: O_DIRECT bypasses client statistics.
842 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
843 read_ahead_stats were updated. copied the stats section from
844 osc_send_oap_rpc() into async_internal().
848 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
852 Description: change order of libsysio includes
853 Details : '#include sysio.h' should always come before '#include xtio.h'
857 Description: handle "serious error: objid * already exists" more gracefully
858 Details : If LAST_ID value on disk is smaller than the objects existing in
859 the O/0/d* directories, it indicates disk corruption and causes an
860 LBUG(). If the object is 0-length, then we should use the existing
861 object. This will help to avoid a full fsck in most cases.
863 Severity : enhancement
865 Description: adapt the lustre_config script to support the upgrade case
866 Details : Add "-u" option for lustre_config script to support upgrading 1.4
867 server targets to 1.6 in parallel.
871 Description: To avoid grant space > avaible space when the disk is almost
872 full. Without this patch you might see the error "grant XXXX >
873 available" or some LBUG about grant, when the disk is almost
875 Details : In filter_check_grant, for non_grant cache write, we should
876 check the left space by if (*left > ungranted + bytes), instead
877 of (*left > ungranted), because only we are sure the left space
878 is enough for another "bytes", then the ungrant space should be
879 increase. In client, we should update cl_avail_grant only there
880 is OBD_MD_FLGRANT in the reply.
884 Description: Update RHEL 4 kernel to fix local root privilege escalation.
885 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
886 described in CVE-2007-4573. This problem could allow untrusted
887 local users to gain root access.
890 Frequency : when using O_DIRECT and quotas
892 Description: Incorrect file ownership on O_DIRECT output files
893 Details : block usage reported by 'lfs quota' does not take into account
894 files that have been written with O_DIRECT.
899 Description: touch file failed when fs is not full
900 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
901 otherwise we can get ENOSP while fs is not full.
905 Description: parallel lock callbacks
906 Details : Instead of sending blocking and completion callbacks as separated
907 requests, adding them to a set and sending in parallel.
910 Frequency : only for Cray XT3
911 Bugzilla : 12829/13455
912 Description: Changing primary group doesn't change the group lustre assigns to
914 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
915 overridden with the primary group provided by the group upcall,
916 whereas the supplied fsgid can be trusted if it is in the list of
917 supplementary groups returned by the group upcall.
919 Severity : enhancement
921 Description: Allow masking D_WARNING, D_ERROR messages from console
922 Details : Console messages can now be disabled via lnet.printk.
926 Description: User code with malformed file open parameter crashes client node
927 Details : Before packing join_file req, all the related reference should be
928 checked carefully in case some malformed flags cause fake
929 join_file req on client.
933 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
934 Details : ldlm_completion_ast() assumes that a lock is granted when the req
935 mode is equal to the granted mode. However, it should also check
936 that LDLM_FL_CP_REQD is not set.
940 Description: Heavy nfs access might result in deadlocks
941 Details : After ELC code landed, it is now improper to enqueue any mds
942 locks under och_sem, because enqueue might want to decide to
943 cancel open locks for same inode we are holding och_sem for.
947 Description: Client eviction while running blogbench
948 Details : A lot of unlink operations with concurrent I/O can lead to a
949 deadlock causing evictions. To address the problem, the number of
950 oustanding OST_DESTROY requests is now throttled to
951 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
952 callbacks are processed in priority.
956 Description: enable ACLs on MDS by default
957 Details : ACLs must be enabled on MDS by default.
960 Frequency : PPC/PPC64 only
962 Description: conflicts between asm-ppc64/types.h and lustre_types.h
963 Details : fix duplicated definitions between asm-ppc64/types.h and
964 lustre_types.h on PPC.
967 Frequency : PPC/PPC64 only
969 Description: asm-ppc/segment.h does not exist
970 Details : fix compile issue on PPC.
974 Description: better handle error messages in extents code
977 Frequency : RHEL4 only
979 Description: mkfs is very slow on IA64/RHEL4
980 Details : A performance regression has been discovered in the MPT Fusion
981 driver between versions 3.02.73rh and 3.02.99.00rh. As a
982 consequence, we have downgraded the MPT Fusion driver in the RHEL4
983 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
985 Severity : enhancement
987 Description: SNMP support enhancement
988 Details : Adding total number of sampled request for an MDS node in snmp
991 Severity : enhancement
993 Description: Optimize ldlm waiting list processing for PR extent locks
994 Details : When processing waiting list for read extent lock and meeting read
995 lock that is same or wider to it that is not contended, skip
996 processing rest of the list and immediatelly return current
997 status of conflictness, since we are guaranteed there are no
998 conflicting locks in the rest of the list.
1002 Description: Time out and refuse to reconnect
1003 Details : When the failover node is the primary node, it is possible
1004 to have two identical connections in imp_conn_list. We must
1005 compare not conn's pointers but NIDs, otherwise we can defeat
1006 connection throttling.
1010 Description: port llog fixes from b1_6 into HEAD
1011 Details : Port llog reference couting and some llog cleanups from b1_6
1012 (bug 10800) into HEAD, for protect from panic and access to already
1013 free llog structures.
1017 Description: Detect stride IO mode in read-ahead
1018 Details : When a client does stride read, read-ahead should detect that and
1019 read-ahead pages according to the detected stride pattern.
1023 Description: data checksumming impacts single node performance
1024 Details : add support for several checksum algorithm. Currently, only CRC32
1025 and Adler-32 are supported. The checksum type can be changed on
1026 the fly via /proc/fs/lustre/osc/*/checksum_type.
1030 Description: use adler32 for page checksums
1031 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1036 Description: build for x2 fails
1037 Details : fix compile issue on Cray systems.
1041 Description: Properly match for duplicate locks
1042 Details : Due to different lock order from skiplists code, we need to
1043 traverse entire list for now
1046 Frequency : only on PPC/SLES10
1048 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1049 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1050 be defined. Add a hack in mkfs_lustre.c to work around this compile
1055 Description: LASSERT on MDS when client holding flock lock dies
1056 Details : ldlm pool logic depends on number of granted locks equal to
1057 number of released locks which is not true for flock locks, so
1058 just exclude such locks from consideration.
1062 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1063 Details : Use fsfilt_sync() for both the whole filesystem sync and
1064 individual file sync to eliminate dangerous inode locking
1065 with I_LOCK that can lead to a deadlock.
1069 Description: performance in 1.6.3
1070 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1071 sure that 1MB requests are not fragmented by the block layer.
1073 Severity : enhancement
1075 Description: organize the server-side client stats on per-nid basis
1076 Details : Change the structure of stats under obdfilter and mds to
1084 The "uuid"s file would list the uuids of _active_ exports.
1085 And the clear entry is to clear all stats and stale nids.
1087 Severity : enhancement
1089 Description: eliminate client locks in face of contention
1090 Details : file contention detection and lockless i/o implementation
1091 for contended files.
1095 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1098 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1100 Description: MDS may lose file striping (and hence file data) in some cases
1101 Details : If there are additional extended attributes stored on the MDS,
1102 in particular ACLs, SELinux, or user attributes (if user_xattr
1103 is specified for the client mount options) then there is a risk
1104 of attribute loss. Additionally, the Lustre file striping
1105 needs to be larger than default (e.g. striped over all OSTs),
1106 and an additional attribute must be stored initially in the
1107 inode and then increase in size enough to be moved to the
1108 external attribute block (e.g. ACL growing in size) for file
1113 Description: skiplist implementation simplification
1114 Details : skiplists are used to group compatible locks on granted list
1115 that was implemented as tracking first and last lock of each lock group
1116 the patch changes that to using doubly linked lists
1120 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1121 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1122 because LASSERT on both the data supplied by a client, and the data
1123 on disk is dangerous and incorrect.
1125 Severity : enhancement
1127 Description: Slow truncate/writes to huge files at high offsets.
1128 Details : Directly associate cached pages to lock that protect those pages,
1129 this allows us to quickly find what pages to write and remove
1130 once lock callback is received.
1134 Description: more ldlm soft lockups
1135 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1136 starve other threads from the resource lock for a long time in
1137 case of long waiting queue, so change the debug level from
1138 D_OTHER to the less frequently used D_INFO.
1140 Severity : enhancement
1142 Description: add -gid, -group, -uid, -user options to lfs find
1146 Description: Hung threads in invalidate_inode_pages2_range
1147 Details : The direct IO path doesn't call check_rpcs to submit a new RPC once
1148 one is completed. As a result, some RPCs are stuck in the queue
1153 Description: filter threads hungs on waiting journal commit
1154 Details : Cleanup filter group llog code, then only filter group llog will
1155 be only created in the MDS/OST syncing process.
1159 Description: Procfs and llog threads access destoryed import sometimes.
1160 Details : Sync the import destoryed process with procfs and llog threads by
1161 the import refcount and semaphore.
1163 Severity : enhancement
1165 Description: openlock cache of b1_6 port to HEAD
1167 --------------------------------------------------------------------------------
1169 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
1171 * Support for kernels:
1172 2.6.5-7.283 (SLES 9),
1173 2.6.9-55.EL (RHEL 4),
1174 2.6.16.46-0.14 (SLES 10),
1175 2.6.18.8 vanilla (kernel.org)
1176 * Client support for unpatched kernels:
1177 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1178 2.6.16 - 2.6.22 vanilla (kernel.org)
1179 * Due to recently discovered recovery problems, we do not recommend
1180 using patchless RHEL 4 clients with this or any earlier release.
1181 * Recommended e2fsprogs version: 1.39.cfs8
1182 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1183 * Starting with this release, the ldiskfs backing filesystem required
1184 by Lustre is now in its own package, lustre-ldiskfs. This package
1185 should be installed. It is versioned separately from Lustre and
1186 may be released separately in future.
1190 Description: block reactivating mgc import until all deactivates complete
1191 Details : Fix race when failing back MDT/MGS to itself (testing)
1194 Frequency : at statup only
1196 Description: mds_lov_synchronize race leads to various problems
1197 Details : simultaneous MDT->OST connections at startup can cause the
1198 sync to abort, leaving the OSC in a bad state.
1200 Severity : enhancement
1202 Description: add optional extra BUILD_VERSION info
1203 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1204 to override the lustre version.
1207 Frequency : 2.6.18 servers only
1209 Description: ll_kern_mount() doesn't release the module reference
1210 Details : The ldiskfs module reference count never drops down to 0
1211 because ll_kern_mount() doesn't release the module reference.
1216 Description: server LBUG when using old ost_num_threads parameter
1217 Details : Accept the old ost_num_threads parameter but warn that it
1218 is deprecated, and fix an off-by-one error that caused an LBUG.
1223 Description: Transient SCSI error results in persistent IO issue
1224 Details : iobuf->dr_error is not reinitialized to 0 between two
1228 Frequency : sometimes when underlying device returns I/O errors
1230 Description: OSTs not going read-only during write failures
1231 Details : OSTs are not remounted read-only when the journal commit threads
1232 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1233 instead of the ext3 wrappers.
1237 Description: poor connect scaling with increasing client count
1238 Details : Don't run filter_grant_sanity_check for more than 100 exports
1239 to improve scaling for large numbers of clients.
1242 Frequency : SLES10 only
1244 Description: sanity-quota.sh quotacheck failed: rc = -22
1245 Details : Quotas cannot be enabled on SLES10.
1248 Frequency : liblustre clients only
1250 Description: getdirentries does not give error when run on compute nodes
1251 Details : getdirentries does not fail when the size specified as an argument
1252 is too small to contain at least one entry
1254 Severity : enhancement
1256 Description: Add LNET router traceability for debug purposes
1257 Details : If a checksum failure occurs with a router as part of the
1258 IO path, the NID of the last router that forwarded the bulk data
1259 is printed so it can be identified.
1264 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1265 Details : Due to a race condition, liblustre clients were occasionally
1266 evicted incorrectly.
1268 Severity : enhancement
1270 Description: lfs setstripe use optional parameters instead of postional
1273 Severity : enhancement
1275 Description: Nanosecond timestamp support for ldiskfs
1276 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1277 resolution timestamps. There is not yet support for this at
1278 the Lustre filesystem level.
1281 Frequency : during server recovery
1283 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1284 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1288 Frequency : nfs export on patchless client
1290 Description: connectathon hang when test nfs export over patchless client
1291 Details : Disconnected dentry cannot be found with lookup, so we do not need
1292 to unhash it or make it invalid
1295 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1296 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1297 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1302 Description: open req refcounting wrong on reconnect
1303 Details : If reconnect happened between getting open reply from server and
1304 call to mdc_set_replay_data in ll_file_open, we will schedule
1305 replay for unreferenced request that we are about to free.
1306 Subsequent close will crash in variety of ways.
1307 Check that request is still eligible for replay in
1308 mdc_set_replay_data().
1313 Description: disable writes to filesystem when reading health_check file
1314 Details : the default for reading the health_check proc file has changed
1315 to NOT do a journal transaction and write to disk, because this
1316 can cause reads of the /proc file to hang and block HA state
1317 checking on a healthy but otherwise heavily loaded system. It
1318 is possible to return to the previous behaviour during configure
1319 with --enable-health-write.
1321 Severity : enhancement
1323 Description: 64-bit inode version
1324 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1325 to the inode. This will be required for version-based recovery.
1330 Description: MDS fails to start if a duplicate client export is detected
1331 Details : in some rare cases it was possible for a client to connect to
1332 an MDS multiple times. Upon recovery the MDS would detect this
1333 and fail during startup. Handle this more gracefully.
1335 Severity : enhancement
1337 Description: Add -o localflock option to simulate old noflock
1339 Details : This will achieve local-only flock/fcntl locks
1345 Description: log_commit_thread vs filter_destroy race leads to crash
1346 Details : Take import reference before releasing llog record semaphore
1351 Description: Wrong request locking in request set processing
1352 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1353 imp_delayed_list, in this place should be used imp_lock.
1356 Frequency : when reconnection
1358 Description: Grant Leak when osc reconnect to OST
1359 Details : When osc reconnect ost, OST(filter) should check whether it
1360 should grant more space to client by comparing fed_grant and
1361 cl_avail_grant, and return the granted space to client instead
1362 of "new granted" space, because client will call osc_init_grant
1363 to update the client grant space info.
1366 Frequency : when client reconnect to OST
1368 Description: Grant Leak when osc do resend and replay bulk write
1369 Details : When osc reconnect to OST, OST(filter)should clear grant info of
1370 bulk write request, because the grant info while be sync between
1371 OSC and OST when reconnect, and we should ignore the grant info
1372 these of resend/replay write req.
1377 Description: Grant space more than avaiable left space sometimes.
1378 Details : When then OST is about to be full, if two bulk writing from
1379 different clients came to OST. Accord the avaliable space of the
1380 OST, the first req should be permitted, and the second one
1381 should be denied by ENOSPC. But if the seconde arrived before
1382 the first one is commited. The OST might wrongly permit second
1383 writing, which will cause grant space > avaiable space.
1386 Frequency : when client is evicted
1388 Description: Grant might be wrongly erased when osc is evicted by OST
1389 Details : when the import is evicted by server, it will fork another
1390 thread ptlrpc_invalidate_import_thread to invalidate the
1391 import, where the grant will be set to 0. While the original
1392 thread will update the grant it got when connecting. So if
1393 the former happened latter, the grant will be wrongly errased
1394 because of this race.
1399 Description: Checking Stale with correct fid
1400 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1401 to check whether it is stale, because sometimes, we want the
1402 enqueue happened anyway, and op_data.fid2 will not be initialized.
1404 Severity : enhancement
1406 Description: update patchless client
1407 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1410 Frequency : only with 2.4 kernel
1412 Description: random memory corruption
1413 Details : size of struct ll_inode_info is to big for union inode.u and this
1414 can be cause of random memory corruption.
1419 Description: Memory leak in recovery
1420 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1421 It should also check obd_fail before fsfilt_start, otherwise if
1422 fsfilt_start return -EROFS,(failover mds during mds recovery).
1423 then the req will return with repmsg->transno = 0 and rc = EROFS.
1424 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1425 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1426 be freed no matter whether fsfilt_commit success or not.
1429 Frequency : only with huge count clients
1431 Description: Prevents from taking the superblock lock in llap_from_page for
1433 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1434 for llap_from_page call in ll_removepage prevents from taking the
1435 superblock lock for a soon died page.
1440 Description: Not check open intent error before release open handle
1441 Details : in some rare cases, the open intent error is not checked before
1442 release open handle, which may cause
1443 ASSERTION(open_req->rq_transno != 0), because it tries to release
1444 the failed open handle.
1449 Description: Set cat log bitmap only after create log success.
1450 Details : in some rare cases, the cat log bitmap is set too early. and it
1451 should be set only after create log success.
1455 Description: Accessing a block bevice can re-enable I/O when Lustre is
1456 tearing down a device.
1457 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1462 Description: service threads may hog cpus when there are a lot of requests
1463 Details : Insert cond_resched to give other threads a chance to use some CPU
1468 Description: the cat log was not initialized in recovery
1469 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1470 unlink log on mds will not be initialized until mds post
1471 recovery. And also in mds post recovery, the unlink log will
1472 initialization will be done asynchronausly, so there will be race
1473 between add unlink log and unlink log initialization.
1477 Description: brw_stats were being printed incorrectly
1478 Details : brw_stats were being printed as log2 but all of them were not
1479 recorded as log2. Also remove some code duplication arising from
1480 filter_tally_{read,write}.
1484 Frequency : rare, only in recovery.
1485 Description: ASSERTION(req->rq_type != LI_POISON) failed
1486 Details : imp_lock should be held while iterating over imp_sending_list for
1487 prevent destroy request after get timeout in ptlrpc_queue_wait.
1491 Description: replay-single.sh test 52 fails
1492 Details : A lock's skiplist need to be cleanup when it being unlinked
1493 from its resource list.
1497 Description: Short directio read returns full requested size rather than
1499 Details : Direct I/O operations should return actual amount of bytes
1500 transferred rather than requested size.
1502 Severity : enhancement
1504 Description: metadata RPC reduction (e.g. for rm performance)
1505 Details : decrease the amount of synchronous RPC between clients and servers
1506 by canceling conflicing lock before the operation on the client
1507 and packing thier handles into the main operation RPC to server.
1509 Severity : enhancement
1511 Description: Async OSC create to avoid the blocking unnecessarily.
1512 Details : If a OST has no remain object, system will block on the creating
1513 when need to create a new object on this OST. Now, ways use
1514 pre-created objects when available, instead of blocking on an
1515 empty osc while others are not empty. If we must block, we block
1516 for the shortest possible period of time.
1518 Severity : enhancement
1520 Description: refine locking for avoid write wrong info into lov_objid file
1521 Details : fix possible races with add new target and write/update data in
1524 --------------------------------------------------------------------------------
1526 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1531 Frequency : on some architectures
1533 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1534 Details : Uninitialized flags sometimes cause configuration commands to
1538 Frequency : patchless clients only
1540 Description: missing __iget() symbol export
1541 Details : The __iget() symbol export is missing. To avoid the need for
1542 this on patchless clients the deathrow inode reaper is turned
1543 off, and we depend on the VM to clean up old inodes. This
1544 dependency was during via the fix for bug 12181.
1546 --------------------------------------------------------------------------------
1548 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1550 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1551 INTEROPERATE with older versions automatically. In many cases a
1552 special upgrade step is needed. Please read the
1553 user documentation before upgrading any part of a live system.
1554 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1555 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1556 * WARNING: Lustre configuration and startup changes are required with
1557 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1563 Severity : enhancement
1565 Description: MountConf
1566 Details : Lustre configuration is now managed via mkfs and mount
1567 commands instead of lmc and lconf. New obd types (MGS, MGC)
1568 are added for dynamic configuration management. See
1569 https://mail.clusterfs.com/wikis/lustre/MountConf for
1572 Severity : enhancement
1574 Description: dynamic OST addition
1575 Details : OSTs can now be added to a live filesystem
1577 Severity : enhancement
1579 Description: startup order invariance
1580 Details : MDTs and OSTs can be started in any order. Clients only
1581 require the MDT to complete startup.
1583 Severity : enhancement
1585 Description: parallel, asynchronous orphan cleanup
1586 Details : orphan cleanup is now performed in separate threads for each
1587 OST, allowing parallel non-blocking operation.
1589 Severity : enhancement
1591 Description: optimized stripe assignment
1592 Details : stripe assignments are now made based on ost space available,
1593 ost previous usage, and OSS previous usage, in order to try
1594 to optimize storage space and networking resources.
1596 Severity : enhancement
1598 Description: Permanently set tunables
1599 Details : All writable /proc/fs/lustre tunables can now be permanently
1600 set on a per-server basis, at mkfs time or on a live system.
1602 Severity : enhancement
1604 Description: Lustre message v2
1605 Details : Add lustre message format v2.
1607 Severity : enhancement
1609 Description: client OST exclusion list
1610 Details : Clients can be started with a list of OSTs that should be
1611 declared "inactive" for known non-responsive OSTs.
1615 Description: SPEC SFS validation failure on NFS v2 over lustre.
1616 Details : Changes the blocksize for regular files to be 2x RPC size,
1617 and not depend on stripe size.
1619 Severity : enhancement
1621 Description: Multiple MD RPCs in flight.
1622 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1623 To avoid overly-overloading MDS, introduce a limit on number of
1624 MDS RPCs in flight for a single client and add /proc controls
1625 to adjust this limit.
1627 Severity : enhancement
1629 Description: client read/write statistics
1630 Details : Add client read/write call usage stats for performance
1631 analysis of user processes.
1632 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1633 file access. extents_stats shows chunk size distribution.
1634 extents_stats_per_process show chunk size distribution per
1637 Severity : enhancement
1639 Description: mds statistics
1640 Details : Add detailed mds operations statistics in
1641 /proc/fs/lustre/mds/*/stats.
1645 Description: Failure of copying files with lustre special EAs.
1646 Details : Client side always return success for setxattr call for lustre
1647 special xattr (currently only "trusted.lov").
1652 Description: Refcount LNET uuids
1653 Details : The global LNET uuid list grew linearly with every startup;
1654 refcount repeated list entries instead of always adding to
1657 Severity : enhancement
1659 Description: Dynamic service threads
1660 Details : Within a small range, start extra service threads
1661 automatically when the request queue builds up.
1664 Frequency : mixed-endian client/server environments
1666 Description: mixed-endian crashes
1667 Details : The new msg_v2 system had some failures in mixed-endian
1670 Severity : enhancement
1672 Description: Easy OST removal
1673 Details : OSTs can be permanently deactivated with e.g. 'lctl
1674 conf_param lustre-OST0001.osc.active=0'
1676 Severity : enhancement
1678 Description: MGS proc entries
1679 Details : Added basic proc entries for the MGS showing what filesystems
1682 Severity : enhancement
1684 Description: provide MGS failover
1685 Details : Added config lock reacquisition after MGS server failover.
1687 Severity : enhancement
1689 Description: add Linux 2.4 support
1690 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1694 Description: a large application tries to do I/O to the same resource and dies
1695 in the middle of it.
1696 Details : Check the req->rq_arrival time after the call to
1697 ost_brw_lock_get(), but before we do anything about
1698 processing it & sending the BULK transfer request. This
1699 should help move old stale pending locks off the queue as
1700 quickly as obd_timeout.
1703 Frequency : when an incorrect nid is specified during startup
1705 Description: ptlrpc connect to non-existant node causes kernel crash
1706 Details : LNET can't be re-entered from an event callback, which
1707 happened when we expire a message after the export has been
1708 cleaned up. Instead, hand the zombie cleanup off to another
1711 Severity : enhancement
1713 Description: plain/inodebits lock performance improvement
1714 Details : Grouping plain/inodebits in granted list by their request modes
1715 and bits policy, thus improving the performance of search through
1719 Frequency : only if OST filesystem is corrupted
1721 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1722 Details : for a short time RPCs with bulk IO are in the replay list,
1723 but replay of bulk IOs is unimplemented. If the OST filesystem
1724 is corrupted due to disk cache incoherency and then replay is
1725 started it is possible to trip an assertion. Avoid putting
1726 committed RPCs into the replay list at all to avoid this issue.
1729 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1731 Description: System hang on startup
1732 Details : This bug allowed the liblustre (e.g. catamount) client to
1733 return to the app before handling all startup RPCs. This
1734 could leave the node unresponsive to lustre network traffic
1735 and manifested as a server ptllnd timeout.
1737 Severity : enhancement
1739 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1740 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1741 boolean to control whether to print peer debug info when a
1742 client's RPC times out.
1745 Frequency : only for kernels with patches from Lustre below 1.4.3
1747 Description: Remove old rdonly API
1748 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1751 Frequency : only for devices with external journals
1753 Description: Set external device read-only also
1754 Details : During a commanded failover stop, we set the disk device
1755 read-only while the server shuts down. We now also set any
1756 external journal device read-only at the same time.
1759 Frequency : when upgrading from 1.4 while trying to change parameters
1761 Description: The wrong (new) MDC name was used when setting parameters for
1762 upgraded MDT's. Also allows changing of OSC (and MDC)
1763 parameters if --writeconf is specified at tunefs upgrade time.
1766 Frequency : when setting specific ost indicies
1768 Description: QOS code breaks on skipped indicies
1769 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1770 created with --index need not be sequential.
1772 Severity : enhancement
1774 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1775 Details : The uninit_groups feature works in conjunction with the kernel
1776 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1777 up the pass1 processing of e2fsck. This is a read-only feature
1778 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1779 mount filesystems that have had this feature enabled.
1781 Severity : enhancement
1783 Description: Improve multi-block allocation algorithm to avoid fragmentation
1784 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1785 allocation locality and avoid filesystem fragmentation.
1787 ------------------------------------------------------------------------------
1789 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1791 * Support for kernels:
1792 2.4.21-47.0.1.EL (RHEL 3)
1793 2.6.5-7.283 (SLES 9)
1794 2.6.9-42.0.10.EL (RHEL 4)
1795 2.6.12.6 vanilla (kernel.org)
1796 2.6.16.27-0.9 (SLES 10)
1797 * Recommended e2fsprogs version: 1.39.cfs5
1799 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1803 Frequency : occasional, depends on client load and configuration
1804 Bugzilla : 12181, 12203
1805 Description: data loss for recently-modified files
1807 Details : In some cases it is possible that recently written or created
1808 files may not be written to disk in a timely manner (this should
1809 normally be within 30s unless client IO load is very high).
1810 The problem appears as zero-length files or files that are a
1811 multiple of 1MB in size after a client crash or client eviction
1812 that are missing data at the end of the file.
1814 This problem is more likely to be hit on clients where files are
1815 repeatedly created and unlinked in the same directory, clients
1816 have a large amount of RAM, have many CPUs, the filesystem has
1817 many OSTs, the clients are rebooted frequently, and/or the files
1818 are not accessed by other nodes after being written.
1820 The presence of the problem can be detected by looking at
1821 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1822 smaller than the second (nr_unused) then dirty files will not
1823 be flushed automatically to disk. "sync; sleep 10" should be
1824 run several times on the node before unmounting it to update
1825 Lustre (this is also safe to run on nodes without this problem).
1827 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1828 that can cause this same problem, so customers using that kernel
1829 also need to update the kernel in addition to Lustre. In order
1830 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1832 It is normal that files written just before a client crash (less
1833 than 30s) may not yet have been flushed to disk, even for local
1837 Frequency : frequent on thin XT3 nodes
1839 Description: UUID collision on thin XT3 Linux nodes
1840 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1841 randomly, since we relied on an insufficiently-seeded PRNG.
1846 Description: OSS hangs after "All ost request buffers busy"
1847 Details : A deadlock between quota and journal operations caused OSS
1848 hangs after printing "All ost request buffers busy."
1851 Frequency : always on liblustre builds
1853 Description: Cleanup compiler warnings on liblustre
1856 Frequency : always on liblustre builds on XT3
1858 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1862 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1863 > 32000 subdirectories
1864 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1865 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1866 subdirectory count crosses 32000. This will aid e2fsck to
1867 correctly handle more than 32000 subdirectories.
1870 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1872 Description: System hang on startup
1873 Details : This bug allowed the liblustre (e.g. catamount) client to
1874 return to the app before handling all startup RPCs. This
1875 could leave the node unresponsive to lustre network traffic
1876 and manifested as a server ptllnd timeout.
1878 Severity : enhancement
1880 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1881 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1882 boolean to control whether to print peer debug info when a
1883 client's RPC times out.
1888 Description: make O_SYNC working on 2.6 kernels
1889 Details : 2.6 kernels use different method for mark pages for write,
1890 so need add a code to lustre for O_SYNC work.
1895 Description: Failure to close file and release space on NFS
1896 Details : Put inode details into lock acquired in ll_intent_file_open.
1897 Use mdc_intent_lock in ll_intent_open to properly
1898 detect all kind of errors unhandled by mdc_enqueue.
1903 Description: proc file read during shutdown sometimes raced obd removal,
1905 Details : Add lock to prevent obd access after proc file removal.
1908 Frequency : Only for files larger than 4GB on 32-bit clients.
1910 Description: improperly doing page alignment of locks
1911 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1912 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1917 Description: RPCs being resent when they shouldn't be
1918 Details : Some RPCs that should not be resent are being resent. This
1919 can cause inconsistencies in the RPC state machine. Do not
1920 resend such requests.
1923 Frequency : rare, only with NFS export
1925 Description: Crash on NFS re-export node
1926 Details : under very unusual load conditions an assertion is hit in
1927 ll_intent_file_open()
1930 Frequency : only if OST filesystem is corrupted
1932 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1933 Details : for a short time RPCs with bulk IO are in the replay list,
1934 but replay of bulk IOs is unimplemented. If the OST filesystem
1935 is corrupted due to disk cache incoherency and then replay is
1936 started it is possible to trip an assertion. Avoid putting
1937 committed RPCs into the replay list at all to avoid this issue.
1942 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1943 Details : Large single O_DIRECT read and write calls can fail to allocate
1944 a sufficiently large buffer to process the request. In case of
1945 allocation failure the allocation is retried with a smaller
1946 buffer and broken into smaller requests.
1948 Severity : enhancement
1950 Description: Add -o localflock option to simulate old noflock behaviour.
1951 Details : This will achieve local-only flock/fcntl locks coherentness.
1956 Description: versioning check is incomplete
1957 Details : Checking the version difference of client vs. server, report
1958 error if the gap is too big.
1963 Description: add support PG_writeback bit
1964 Details : add support for PG_writeback bit for Lustre, for more carefull
1965 work with page cache in 2.6 kernel. This also fix some deadlocks
1966 and remove hack for work O_SYNC with 2.6 kernel.
1968 Severity : enhancement
1970 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1971 Details : The uninit_groups feature works in conjunction with the kernel
1972 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1973 up the pass1 processing of e2fsck. This is a read-only feature
1974 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1975 mount filesystems that have had this feature enabled.
1977 Severity : enhancement
1979 Description: Improve multi-block allocation algorithm to avoid fragmentation
1980 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1981 allocation locality and avoid filesystem fragmentation.
1983 ------------------------------------------------------------------------------
1985 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1987 * Support for kernels:
1988 2.6.9-42.0.3.EL (RHEL 4)
1989 2.6.5-7.276 (SLES 9)
1990 2.4.21-47.0.1.EL (RHEL 3)
1991 2.6.12.6 vanilla (kernel.org)
1992 2.6.16.21-0.8 (SLES10)
1993 * Recommended e2fsprogs version: 1.39.cfs2-0
1995 * The backwards-compatible /proc/sys/portals symlink has been removed
1996 in this release. Before upgrading, please ensure that you change
1997 any configuration scripts or /etc/sysctl.conf files that access
1998 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1999 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2000 in advance of the upgrade on any system running Lustre 1.4.6 or
2001 newer, since /proc/sys/lnet was added in that version.
2002 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2006 Frequency : only when quota is used
2008 Description: avoid scanning export list for quota master
2009 Details : Change the algorithms to avoid scanning export list in order
2010 to improve the efficiency.
2013 Frequency : MDS failover only, very rarely
2015 Description: "went back in time" messages on mds failover
2016 Details : The greatest transno may be lost when the current operation
2017 finishes with an error (transno==0) and the client's last_rcvd
2018 record is over-written. Save the greatest transno in the
2019 mds_last_transno for this case.
2022 Frequency : always for specific kernels and striping counts
2024 Description: client may get "Matching packet too big" without ACL support
2025 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2026 when trying to access files in certain configurations. The
2027 clients should in fact be denied when mounting because they do
2028 not understand ACLs.
2031 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2033 Description: many clients connecting with IO in progress causes connect timeouts
2034 Details : Avoid synchronous journal commits to avoid delays caused by many
2035 clients connecting/disconnecting when bulk IO is in progress.
2036 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2037 OST_IO_PORTAL to avoid delays behind potentially many pending
2041 Frequency : occasionally with multiple writers to a single file
2043 Description: shared writes to file may result in wrong size reported by stat()
2044 Details : Allow growing of kms when extent lock is cancelled
2047 Frequency : always with random mmap IO to multi-striped file
2049 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2050 Details : Only if the hole is at the end of OST object so that kms is too
2051 small. Fix is to increase kms accordingly in ll_nopage.
2054 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2056 Description: writes to a missing object would leak memory on the OST
2057 Details : If there is an inconsistency between the MDS and OST filesystems,
2058 such that the MDS references an object that doesn't exist, writes
2059 to that object will leak memory due to incorrect cleanup in the
2060 error handling path, eventually running out of memory on the OST.
2065 Description: Creating too long symlink causes lustre errors
2066 Details : Check symlink and name lengths before sending requests to MDS.
2069 Frequency : only if flock is enabled (not on by default)
2071 Description: posix locks not released on fd closure on 2.6.9+
2072 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2073 kernels, this caused such locks not to be released on fd close and
2074 then assertions on fs unmount about still used locks.
2077 Frequency : MDS failover only, very rarely
2079 Description: clients may get ASSERTION(granted_lock != NULL)
2080 Details : When request was taking a long time, and a client was resending
2081 a getattr by name lock request. The were multiple lock requests
2082 with the same client lock handle and
2083 mds_getattr_name->fixup_handle_for_resent_request found one of the
2084 lock handles but later failed with ASSERTION(granted_lock != NULL).
2089 Description: handle->h_buffer_credits > 0, assertion failure
2090 Details : h_buffer_credits is zero after truncate, causing assertion
2091 failure. This patch extends the transaction or creates a new
2095 Frequency : NFS re-export or patchless client
2096 Bugzilla : 11179, 10796
2097 Description: Crash on NFS re-export node (__d_move)
2098 Details : We do not want to hash the dentry if we don't have a lock.
2099 But if this dentry is later used in d_move, we'd hit uninitialised
2100 list head d_hash, so we just do this to init d_hash field but
2101 leave dentry unhashed.
2104 Frequency : NFS re-export or patchless client
2106 Description: NFS exports has problem with symbolic link
2107 Details : lustre client didn't properly install dentry when re-exported
2108 to NFS or running patchless client.
2111 Frequency : NFS re-export or patchless client
2113 Description: Various nfs/patchless fixes.
2114 Details : fixes reuse disconected alias for lookup process - this fixes
2115 warning "find_exported_dentry: npd != pd",
2116 fix permission error with open files at nfs.
2117 fix apply umask when do revalidate.
2120 Frequency : occasional
2122 Description: Crash on NFS re-export node
2123 Details : calling clear_page() on the wrong pointer triggered oops in
2124 generic_mapping_read().
2127 Frequency : rarely, using O_DIRECT IO
2129 Description: unaligned directio crashes client with LASSERT
2130 Details : check for unaligned buffers before trying any requests.
2133 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2135 Description: stale data returned from RAID cache
2136 Details : If only a small amount of IO is done to the RAID device before
2137 reading it again it is possible to get stale data from the RAID
2138 cache instead of reading it from disk.
2141 Frequency : always for sles10 kernel
2143 Description: sles10 support
2144 Details : ll_follow_link: compile fixes and using of nd_set_link
2145 under newer kernels.
2148 Frequency : depends on arch, kernel and compiler version, always on sles10
2151 Description: recursive or deep enough symlinks cause stack overflow
2152 Details : getting rid of large stack-allocated variable in
2156 Frequency : depends on hardware
2158 Description: lustre write performance loss in the SLES10 kernel
2159 Details : the performance loss is caused by using of write barriers in the
2160 ext3 code. The SLES10 kernel turns barrier support on by
2161 default. The fix is to undo that change for ldiskfs.
2163 ------------------------------------------------------------------------------
2165 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2167 * Support for kernels:
2168 2.6.9-42.0.3EL (RHEL 4)
2169 2.6.5-7.276 (SLES 9)
2170 2.4.21-47.0.1.EL (RHEL 3)
2171 2.6.12.6 vanilla (kernel.org)
2175 Frequency : quota enabled and large files being deleted
2177 Description: releasing more than 4GB of quota at once hangs OST
2178 Details : If a user deletes more than 4GB of files on a single OST it
2179 will cause the OST to spin in an infinite loop. Release
2180 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2185 Description: statfs data retrieved from /proc may be stale or zero
2186 Details : When reading per-device statfs data from /proc, in the
2187 {kbytes,files}_{total,free,avail} files, it may appear
2188 as zero or be out of date.
2191 Frequency : systems with MD RAID1 external journal devices
2193 Description: lconf's call to blkid is confused by RAID1 journal devices
2194 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2195 of returning all block devices that match the journal UUID.
2198 Frequency : always, for aggregate stripe size over 4GB
2200 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2201 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2202 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2205 Frequency : always if "lfs find" used on a local file/directory
2207 Description: "lfs find" segfaults if used on a local file/directory
2208 Details : The case where a directory component was not specified wasn't
2209 handled correctly. Handle this properly.
2212 Frequency : always on ppc64
2214 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2215 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2216 perform endianness conversion.
2219 Frequency : rarely (truncate to non-zero file size after write under load)
2220 Bugzilla : 10730, 10687
2221 Description: Files padded with zeros to next 4K multiple
2222 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2223 it is possible that files that are truncated to a non-zero size
2224 immediately after being written are filled with zero bytes beyond
2225 the truncated size. No file data is lost.
2227 Severity : enhancement
2229 Description: Allow recovery/failover for liblustre clients.
2230 Details : liblustre clients were unaware of failover configurations until
2233 Severity : enhancement
2235 Description: user file locks should fail when not mounting with flock option
2236 Details : Set up an error-returning stub in ll_file_operations.lock field
2237 to prevent incorrect behaviour when client is mounted without
2238 flock option. Also, set up properly f_op->flock field for
2242 Frequency : always on ia64
2244 Description: "lfs df" loops on printing out MDS statfs information
2245 Details : The obd_ioctl_data was not initialized and in some systems
2246 this caused a failure during the ioctl that did not return
2247 an error. Initialize the struct and return an error on failure.
2250 Frequency : SLES 9 only
2252 Description: Error of copying files with lustre special EAs as root
2253 Details : Client side always return success for setxattr call for lustre
2254 special xattr (currently only "trusted.lov").
2257 Frequency : rarely on clusters with both ia64+i386 clients
2259 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2260 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2261 clients are doing concurrent writes to the same file it is
2262 possible that the ia64 clients may LASSERT because the OST
2263 extent locks are not PAGE_SIZE aligned. Ensure that grown
2264 locks are always aligned on the request boundary.
2267 Frequency : specific use, occasional
2269 Description: Overwriting in use executable truncates on-disk binary image
2270 Details : If one node attempts to overwrite an executable in use by
2271 another node, we now correctly return ETXTBSY instead of
2272 truncating the file.
2274 Severity : enhancement
2276 Description: Async OSC create to avoid the blocking unnecessarily.
2277 Details : If a OST has no remain object, system will block on the creating
2278 when need to create a new object on this OST. Now, ways use
2279 pre-created objects when available, instead of blocking on an
2280 empty osc while others are not empty. If we must block, we block
2281 for the shortest possible period of time.
2286 Description: chmod on Lustre root is propagated to other clients
2287 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2288 invalid by the follow_mount time.
2291 Frequency : liblustre clients only
2293 Description: Race in 'instant cancel' lock handling could lead to such locks
2294 never to be granted in case of SMP MDS
2295 Details : Do not destroy not yet granted but cbpending locks in
2299 Frequency : replay/resend of open
2301 Description: non null lock assetion failure in mds_intent_policy
2302 Details : Trying to replay/resend lockless open requests resulted in
2303 mds_open() returning 0 with no lock. Now it sets a flag if
2304 a lock is going to be returned.
2306 Severity : enhancement
2308 Description: Checksum enhancements
2309 Details : New checksum enhancements allow for resending RPCs that failed
2312 Severity : enhancement
2314 Description: Tunables on number of dirty pages in cacche
2315 Details : Allow to set limit on number of dirty pages cached.
2320 Description: client crash on unmount - lock still has references
2321 Details : In some error handling cases it was possible to leak a lock
2322 reference on a client while accessing a file. This was not
2323 harmful to the client during operation, but would cause the
2324 client to crash when the filesystem is unmounted.
2327 Frequency : specific case, rare
2329 Description: ETXTBSY on mds though file not in use by client
2330 Details : ETXTBSY is no longer incorrectly returned when attempting to
2331 chmod or chown a directory that the user previously tried to
2332 execute or a currently-executing binary.
2335 Frequency : extremely rare except on liblustre-based clients
2337 Description: Lustre space not freed when files are deleted
2338 Details : Clean up open-unlinked files after client eviction. Previously
2339 the unlink was skipped and the files remained as orphans.
2344 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2345 Details : In some cases it was possible to send a blocking callback to
2346 a client doing a glimpse, even though that client didn't get
2347 a lock granted. When the glimpse lock is cancelled on the OST
2348 the freed lock is left on the waiting list and corrupted the list.
2351 Frequency : all core dumps
2353 Description: Broke core dumps to lustre
2354 Details : Negative dentry may be unhashed if parent does not have UPDATE
2355 lock, but some callers, e.g. do_coredump, expect dentry to be
2356 hashed after successful create, hash it in ll_create_it.
2358 ------------------------------------------------------------------------------
2360 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2362 * Support for kernels:
2363 2.6.9-42.0.2.EL (RHEL 4)
2364 2.6.5-7.276 (SLES 9)
2365 2.4.21-40.EL (RHEL 3)
2366 2.6.12.6 vanilla (kernel.org)
2370 Frequency : always on RHEL 3
2372 Description: Number of open files grows over time
2373 Details : The number of open files grows over time, whether or not
2374 Lustre is started. This was due to a filp leak introduced
2375 by one of our kernel patches.
2377 ------------------------------------------------------------------------------
2379 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2381 * Support for kernels:
2382 2.6.9-42.EL (RHEL 4)
2383 2.6.5-7.276 (SLES 9)
2384 2.4.21-40.EL (RHEL 3)
2385 2.6.12.6 vanilla (kernel.org)
2390 Bugzilla : 5719, 9635, 9792, 9684
2391 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2392 Details : If a server is under heavy load and cannot reply to new
2393 connection requests before the client resends the (re)connect,
2394 the connection handling code can behave badly if two service
2395 threads are concurrently handing separate (re)connections from
2396 the same client. Add better locking to the connection handling
2397 code, and ensure that only a single connection will be processed
2398 for a given client UUID, even if the lock is dropped.
2400 Severity : enhancement
2402 Description: add TCP zero-copy support to kernel
2403 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2404 sends if the hardware supports scatter-gather and checksumming.
2405 This allows socklnd to do client-write and server-read more
2406 efficiently and reduce CPU utilization from skbuf copying.
2409 Frequency : only if NFS exporting from client
2411 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2412 Details : If setting attributes on a file created under NFS that had
2413 never been opened it would be possible to oops the client
2414 if the file had no objects.
2417 Frequency : always for liblustre
2419 Description: liblustre client does MDS+OSTs setattr RPC for each write
2420 Details : When doing a write from a liblustre client, the client
2421 incorrectly issued an RPC to the MDS and each OST the file was
2422 striped over in order to update the timestamps. When writing
2423 with small chunks and many clients this could overwhelm the MDS
2424 with RPCs. In all cases it would slow down the write because
2425 these RPCs are unnecessary.
2427 Severity : enhancement
2429 Description: allow number of MDS service threads to be changed at module load
2430 Details : It is now possible to change the number of MDS service threads
2431 running. Adding "options mds mds_num_threads={N}" to the MDS's
2432 /etc/modprobe.conf will set the number of threads for the next
2433 time Lustre is restarted (assuming the "mds" module is also
2434 reloaded at that time). The default number of threads will
2435 stay the same, 32 for most systems.
2440 Description: OST crash if filesystem is unformatted or corrupt
2441 Details : If an OST is started on a device that has never been formatted
2442 or if the filesystem is corrupt and cannot even mount then the
2443 error handling cleanup routines would dereference a NULL pointer.
2448 Description: NULL pointer deref in llap_from_page.
2449 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2450 incorrect) mapping if the page was truncated/reclaimed while it was
2451 searched for. Check for this condition and skip such pages when
2452 doing readahead. Introduce extra check to llap_from_page() to
2453 verify page->mapping->host is non-NULL (so page is not anonymous).
2456 Frequency : Sometimes when using sys_sendfile
2458 Description: "page not covered by a lock" warnings from ll_readpage
2459 Details : sendfile called ll_readpage without right page locks present.
2460 Now we introduced ll_file_sendfile that does necessary locking
2461 around call to generic_file_sendfile() much like we do in
2465 Frequency : with certain MDS communication failures at client mount time
2467 Description: NULL pointer deref after failed client mount
2468 Details : a client connection request may delayed by the network layer
2469 and not be sent until after the PTLRPC layer has timed out the
2470 request. If the client fails the mount immediately it will try
2471 to clean up before the network times out the request. Add a
2472 reference from the request import to the obd device and delay
2473 the cleanup until the network drops the request.
2476 Frequency : occasionally during client (re)connect
2478 Description: assertion failure during client (re)connect
2479 Details : processing a client connection request may be delayed by the
2480 client or server longer than the client connect timeout. This
2481 causes the client to resend the connection request. If the
2482 original connection request is replied in this interval, the
2483 client may trip an assertion failure in ptlrpc_connect_interpret()
2484 which thought it would be the only running connect process.
2487 Frequency : only with obd_echo servers and clients that are rebooted
2489 Description: kernel BUG accessing uninitialized data structure
2490 Details : When running an obd_echo server it did not start the ping_evictor
2491 thread, and when a client was evicted an uninitialized data
2492 structure was accessed. Start the ping_evictor in the RPC
2493 service startup instead of the OBD startup.
2495 Severity : enhancement
2496 Bugzilla : 10193 (patchless)
2497 Description: Remove dependency on various unexported kernel interfaces.
2498 Details : No longer need reparent_to_init, exit_mm, exit_files,
2499 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2502 Frequency : rare (only users of deprecated and unsupported LDAP config)
2504 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2505 Details : LDAP apparently contains 'lustreName' attributes instead of
2506 'name'. A simple remapping of the name is sufficient.
2509 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2511 Description: waiting_locks_callback trips kernel BUG if client is evicted
2512 Details : Running with the dump_on_timeout debug flag turned on makes
2513 it possible that the waiting_locks_callback() can try to dump
2514 the Lustre kernel debug logs from an interrupt handler. Defer
2515 this log dumping to the expired_lock_main() thread.
2517 Severity : enhancement
2519 Description: Support NFS exporting on 2.6 kernels.
2520 Details : Implement non-rawops metadata methods for NFS server to use without
2521 changing NFS server code.
2524 Frequency : very rare (synthetic metadata workload only)
2526 Description: two racing renames might cause an MDS thread to deadlock
2527 Details : Running the "racer" program may cause one MDS thread to rename
2528 a file from being the source of a rename to being the target of
2529 a rename at exactly the same time that another thread is doing
2530 so, and the second thread has already enqueued these locks after
2531 doing a lookup of the target and is trying to relock them in
2532 order. Ensure that we don't try to re-lock the same resource.
2535 Frequency : only very large systems with liblustre clients
2537 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2538 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2539 all OSTs in parallel. This allows the eviction of stale liblustre
2540 clients to proceed much faster than if they were done in series,
2541 and also offers similar improvements for other set_info RPCs.
2546 Description: excessive CPU usage during initial read phase on client
2547 Details : During the initial read phase on a client, it would agressively
2548 retry readahead on the file, consuming too much CPU and impacting
2549 performance (since 1.4.5.8). Improve the readahead algorithm
2550 to avoid this, and also improve some other common cases (read
2551 of small files in particular, where "small" is files smaller than
2552 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2557 Description: MDS crash when receiving packet with unknown intent.
2558 Details : Do not LBUG in unknown intent case, just return -EFAULT
2560 Severity : enhancement
2561 Bugzilla : 9293, 9385
2562 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2563 Details : Do not serialize getattr (non-intent version) and statfs.
2566 Frequency : occasional, when OST network is overloaded/intermittent
2568 Description: client evicted by OST after bulk IO timeout
2569 Details : If a client sends a bulk IO request (read or write) the OST
2570 may evict the client if it is unresposive to its data GET/PUT
2571 request. This is incorrect if the network is overloaded (takes
2572 too long to transfer the RPC data) or dropped the OST GET/PUT
2573 request. There is no need to evict the client at all, since
2574 the pinger and/or lock callbacks will handle this, and the
2575 client can restart the bulk request.
2578 Frequency : Always when mmapping file with no objects
2580 Description: client crashes when mmapping file with no objects
2581 Details : Check that we actually have objects in a file before doing any
2582 operations on objects in ll_vm_open, ll_vm_close and
2588 Description: Request leak when working with deleted CWD
2589 Details : Introduce advanced request refcount tracking for requests
2590 referenced from lustre intent.
2592 Severity : Enhancement
2594 Description: Cache open file handles on client.
2595 Details : MDS now will return special lock along with openhandle, if
2596 requested and client is allowed to hold openhandle, even if unused,
2597 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2598 closing files for every read/write openration.
2600 Severity : Enhancement
2602 Description: Cache open negative dentries on client when possible.
2603 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2604 negative dentries on lock revocation.
2609 Description: Remounting a client read-only wasn't possible with a zconf mount
2610 Details : It wasn't possible to remount a client read-only with llmount.
2612 Severity : enhancement
2613 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2614 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2615 MPICH is included to add Lustre-specific ADIO interfaces.
2616 This is based closely on the UFS ADIO layer and only differs
2617 in file creation, in order to allow the OST striping to be set.
2618 This is user-contributed code and not supported by CFS.
2623 Description: extended inode attributes (immutable, append-only) work improperly
2624 when 2.4 and 2.6 kernels are used on client/server or vice versa
2625 Details : Introduce kernel-independent values for these flags.
2627 Severity : enhancement
2630 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2631 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2632 granularity. Now, Fractional MB granularity is be supported,
2633 this is very useful for low memory system.
2635 Severity : enhancement
2637 Description: Getattr by fid
2638 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2639 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2640 have locks and avoids getattr rpc requests for it on every stat.
2643 Frequency : Always, for filesystems larger than 2TB
2645 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2646 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2647 filesystems larger than 2TB to be mounted with mballoc enabled.
2650 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2652 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2653 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2654 filesystems larger than 2TB is mounted then the kernel will
2655 silently corrupt the start of the filesystem. CONFIG_LBD is
2656 enabled for all CFS-supported kernels, but the possibility of
2657 this happening with a modified kernel config exists.
2659 Severity : enhancement
2661 Description: add client O_DIRECT support for 2.6 kernels
2662 Details : It is now possible to do O_DIRECT reads and writes to files
2663 in the Lustre client mountpoint on 2.6 kernel clients.
2665 Severity : enhancement
2667 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2668 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2669 parallel, not waiting for response from every OST before sending
2670 a rpc to the next OST.
2675 Description: setattr vs write race when updating file timestamps
2676 Details : Client processes that update a file timestamp into the past
2677 right after writing to the file (e.g. tar) it is possible that
2678 the updated file modification time can be reset to the current
2679 time due to a race between processing the setattr and write RPC.
2681 Severity : enhancement
2683 Description: Bring 'lfs find' closer in line with regular Linux find.
2684 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2685 -print0 options and obtains all the needed info through the lustre
2688 Severity : enhancement
2690 Description: support up to 1024 configured devices on one node
2691 Details : change obd_dev array from statically allocated to dynamically
2692 allocated structs as they are first used to reduce memory usage
2697 Description: Flush dirty partially truncated pages during truncate
2698 Details : Immediatelly flush partially truncated pages in filter_setattr,
2699 this way we completely avoid having any pages in page cache on OST
2700 and can retire ugly workarounds during writes to flush such pages.
2705 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2706 Details : protect inode from truncation within vfs_unlink() context
2707 just take a reference before calling vfs_unlink() and release it
2708 when parent's i_sem is free.
2713 Description: last_id value checked outside lock on OST caused LASSERT failure
2714 Details : If there were multiple MDS->OST object precreate requests in
2715 flight, it was possible that the OST's last object id was checked
2716 outside a lock and incorrectly tripped an assertion. Move checks
2717 inside locks, and discard old precreate requests.
2720 Frequency : always, if extents are used on OSTs
2722 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2723 Details : OSTs using the extents format would not zero the high 16 bits of
2724 the index physical block number. This is not a problem for any
2725 OST filesystems smaller than 16TB, and no kernels support ext3
2726 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2727 new/modified files) and can be fixed for existing filesystems
2728 with e2fsprogs-1.39-cfs1.
2733 Description: import connection selection may be incorrect if timer wraps
2734 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2735 import connections to be ignored if the 32-bit jiffies counter
2736 wraps. Use a 64-bit jiffies counter.
2739 Frequency : very large clusters immediately after boot
2741 Description: LNET request buffers exhausted under heavy short-term load
2742 Details : If a large number of client requests are generated on a service
2743 that has previously never seen so many requests it is possible
2744 that the request buffer growth cannot keep up with the spike in
2745 demand. Instead of dropping incoming requests, they are held in
2746 the LND until the RPC service can accept more requests.
2749 Frequency : Sometimes during replay
2751 Description: Assertion failure in ll_local_open after replay.
2752 Details : If replay happened on an open request reply before we were able
2753 to set replay handler, reply will become not swabbed tripping the
2754 assertion in ll_local_open. Now we set the handler right after
2755 recognising of open request
2758 Frequency : very rare
2760 Description: kernel reports "badness in vsnprintf"
2761 Details : Reading from the "recovery_status" /proc file in small chunks
2762 may cause a negative length in lprocfs_obd_rd_recovery_status()
2763 call to vsnprintf() (which is otherwise harmless). Exit early
2764 if there is no more space in the output buffer.
2766 Severity : enhancement
2768 Description: clear OBD RPC statistics by writing to them
2769 Details : It is now possible to clear the OBD RPC statistics by writing
2770 to the "stats" file.
2775 Description: Client mtime is not the same on different clients after utimes
2776 Details : In some cases, the client was using the utimes() syscall on
2777 a file cached on another node. The clients now validate the
2778 ctime from the MDS + OSTs to determine which one is right.
2783 Description: Inability to activate failout mode
2784 Details : lconf script incorrectly assumed that in pythong string's numeric
2785 value is used in comparisons.
2788 Frequency : always with multiple stripes per file
2790 Description: Inefficient object allocation for mutli-stripe files
2791 Details : When selecting which OSTs to stripe files over, for files with
2792 a stripe count that divides evenly into the number of OSTs,
2793 the MDS is always picking the same starting OST for each file.
2794 Return the OST selection heuristic to the original design.
2799 Description: mount failures may take full timeout to return an error
2800 Details : Under some heavy load conditions it is possible that a
2801 failed mount can wait for the full obd_timeout interval,
2802 possibly several minutes, before reporting an error.
2803 Instead return an error as soon as the status is known.
2805 Frequency : quota enabled and large files being deleted
2807 Description: releasing more than 4GB of quota at once hangs OST
2808 Details : If a user deletes more than 4GB of files on a single OST it
2809 will cause the OST to spin in an infinite loop. Release
2810 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2815 Description: statfs data retrieved from /proc may be stale or zero
2816 Details : When reading per-device statfs data from /proc, in the
2817 {kbytes,files}_{total,free,avail} files, it may appear
2818 as zero or be out of date.
2821 Frequency : systems with MD RAID1 external journal devices
2823 Description: lconf's call to blkid is confused by RAID1 journal devices
2824 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2825 of returning all block devices that match the journal UUID.
2828 Frequency : always, for aggregate stripe size over 4GB
2830 Description: assertion fails when trying to use 4GB stripe size
2831 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2832 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2835 Frequency : always on ppc64
2837 Description: the first write on an ext3 filesystem with mballoc got stuck
2838 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2839 perform endianness conversion.
2841 ------------------------------------------------------------------------------
2843 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2845 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2846 INTEROPERATE with older versions automatically. Please read the
2847 user documentation before upgrading any part of a live system.
2848 * WARNING: Lustre networking configuration changes are required with
2849 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2852 * Support for kernels:
2853 2.6.9-22.0.2.EL (RHEL 4)
2854 2.6.5-7.244 (SLES 9)
2855 2.6.12.6 vanilla (kernel.org)
2858 Severity : enhancement
2859 Bugzilla : 7981/8208
2860 Description: Introduced Lustre Networking (LNET)
2861 Details : LNET is new networking infrastructure for Lustre, it includes
2862 a reorganized network configuration mode (see the user
2863 documentation for full details) as well as support for routing
2864 between different network fabrics. Lustre Networking Devices
2865 (LNDs) for the supported network fabrics have also been
2866 created for this new infrastructure.
2868 Severity : enhancement
2869 Description: Introduced Access control lists
2870 Details : clients can set ACLs on files and directories in order to have
2871 more fine-grained permissions than the standard Unix UGO+RWX.
2872 The MDS must be started with the "-o acl" mount option.
2874 Severity : enhancement
2875 Description: Introduced filesystem quotas
2876 Details : Administrators may now establish per-user quotas on the
2879 Severity : enhancement
2881 Description: Configuration change for the XT3
2882 The PTLLND is now used to run Lustre over Portals on the XT3
2883 The configure option(s) --with-cray-portals are no longer used.
2884 Rather --with-portals=<path-to-portals-includes> is used to
2885 enable building on the XT3. In addition to enable XT3 specific
2886 features the option --enable-cray-xt3 must be used.
2891 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2892 Details : A race between allocating a new llcd and re-getting the llcd_lock
2893 allowed another thread to grab newly-allocated llcd.
2895 Severity : enhancement
2897 Description: 2.6 OST async journal commit and locking fix to improve performance
2898 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2899 2.6 kernels are now async as they already were in 2.4 kernels so
2900 that they can commit concurrently with the network bulk transfer.
2901 For block-allocated files the filter allocation semaphore is held
2902 to avoid filesystem fragmentation during allocation. BKL lock
2903 removed for 2.6 xattr operations where it is no longer needed.
2908 Description: lconf incorrectly determined whether two IP networks could talk
2909 Details : In some more complicated routing and multiple-network
2910 configurations, lconf will avoid trying to make a network
2911 connection to a disjoint part of the IP space. It was doing the
2912 math incorrectly for one set of cases.
2917 Description: Fix for potential infinite loop processing records in an llog.
2918 Details : If an llog record is corrupted/zeroed, it is possible to loop
2919 forever in llog_process(). Validate the llog record length
2920 and skip the remainder of the block on error.
2923 Frequency : occasional (liblustre only)
2925 Description: liblustre could not open files whose last component is a symlink
2926 Details : sysio_path_walk() would incorrectly pass the open intent to
2927 intermediate path components.
2930 Frequency : rare (liblustre only with non-standard tuning)
2931 Bugzilla : 7201 (7350)
2932 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2933 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2934 after it has been released, only do this for referenced locks
2936 Severity : enhancement
2938 Description: specifying an (invalid) directory default stripe_size of -1
2939 would reset the directory default striping
2940 Details : stripe_size -1 was used internally to signal directory stripe
2941 removal, now use "all default" to signal dir stripe removal
2942 as a directory striping of "all default" is not useful
2945 Frequency : common for large clusters running liblustre clients
2947 Description: doing an ls when liblustre clients are running is slow
2948 Details : sending a glimpse AST to a liblustre client waits for every AST
2949 to time out, as liblustre clients will not respond. Since they
2950 cannot cache data we refresh the OST lock LVB from disk instead.
2952 Severity : enhancement
2954 Description: doing an ls at the same time as file IO can be slow
2955 Details : enqueue and other "small" requests can be blocked behind many
2956 large IO requests. Create a new OST IO portal for non-IO
2957 requests so they can be processed faster.
2960 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2962 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2963 created file with mode 000
2964 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2965 allow an ftruncate() as a non-root user to a file with mode 000.
2966 Lustre now respects this flag to disable mode checks when
2967 truncating a file owned by the user
2970 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2973 Description: Revoking locks from clients that went dead or catatonic might take
2975 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2976 cancellation of such locks instant on servers without waiting for
2977 any reply from clients. Clients drops these locks when cancel
2978 notification from server is received without replying.
2981 Frequency : liblustre-only, when liblustre client dies or becomes busy
2983 Description: Doing ls on Linux clients can take a long time with active
2985 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2986 granting such locks to it in the first place if possible. Locks
2987 are taken by proxy on the OST during the read or write and
2988 dropped immediately afterward. Add connect flags handling, do
2989 not grant locks to liblustre clients for glimpse ASTs.
2991 Severity : enhancement
2993 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2994 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2995 sequential reads and behaves badly with applications doing random
2996 reads. Improve it by reading ahead at least read region, and
2997 avoiding excessive large RPC for small reads.
2999 Severity : enhancement
3001 Description: Creating more than 1000 files for a single job may cause a load
3002 imbalance on the OSTs if there are also a large number of OSTs.
3003 Details : qos_prep_create() uses an OST index reseed value that is an
3004 even multiple of the number of available OSTs so that if the
3005 reseed happens in the middle of the object allocation it will
3006 still utilize the OSTs as uniformly as possible.
3011 Description: OST or MDS may oops in ping_evictor_main()
3012 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3013 but doesn't restart at beginning of obd_exports_timed list
3016 Severity : enhancement
3018 Description: improve by-nid export eviction on the MDS and OST
3019 Details : allow multiple exports with the same NID to be evicted at one
3020 time without re-searching the exports list.
3023 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3025 Description: MDS may oops in groups_free()
3026 Details : in rare race conditions a newly allocated group_info struct is
3027 freed again, and this can be NULL. The 2.4 compatibility code
3028 for groups_free() checked for a NULL pointer, but 2.6 did not.
3031 Frequency : common for liblustre clients doing little filesystem IO
3032 Bugzilla : 9352, 7313
3033 Description: server may evict liblustre clients accessing contended locks
3034 Details : if a client is granted a lock or receives a completion AST
3035 with a blocking AST already set it would not reply to the AST
3036 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3039 Frequency : lfs setstripe, only systems with more than 160 OSTs
3041 Description: unable to set striping with a starting offset beyond OST 160
3042 Details : llapi_create_file() incorrectly limited the starting stripe
3043 index to the maximum single-file stripe count.
3046 Frequency : LDAP users only
3048 Description: lconf did not handle in-kernel recovery with LDAP properly
3049 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3051 Severity : enhancement
3053 Description: bind OST threads to NUMA nodes to improve performance
3054 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3055 node and do their allocations there to localize memory access
3057 Severity : enhancement
3059 Description: llmount can determine client NID directly from Myrinet (GM)
3060 Details : the client NID code from gmnalnid was moved directly into
3061 llmount, removing the need to use this or specifying the
3062 client NID explicitly when mounting GM clients with zeroconf
3065 Frequency : if client is started with down MDS
3067 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3068 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3069 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3070 ptlrpc_import_delayed_req() because we didn't check if the
3071 request was interrupted, and we also didn't break out of the
3072 event loop if there was no timeout
3077 Description: data loss during non-page-aligned writes to a single file from
3078 both multiple nodes and multiple threads on one node at same time
3079 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3080 inconsistency led to false short-reads, that were cached and later
3081 used by ->prepare_write() to fill in partially written page,
3082 leading to data loss.
3085 Frequency : always, if lconf --abort_recovery used
3087 Description: lconf --abort_recovery fails with 'Operation not supported'
3088 Details : lconf was attempting to abort recovery on the MDT device and not
3091 Severity : enhancement
3093 Description: remove cleanup logs
3094 Details : replace lconf-generated cleanup logs with lustre internal
3095 cleanup routines. Eliminates the need for client-cleanup and
3098 Severity : enhancement
3100 Description: add support for EAs (user and system) on lustre filesystems
3101 Details : it is now possible to store extended attributes in the Lustre
3102 client filesystem, and with the user_xattr mount option it
3103 is possible to allow users to store EAs on their files also
3105 Severity : enhancement
3107 Description: Add possibility (config option) to show minimal available OST free
3109 Details : When compiled with --enable-mindf configure option, statfs(2)
3110 (and so, df) will return least minimal free space available from
3111 all OSTs as amount of free space on FS, instead of summary of
3112 free spaces of all OSTs.
3114 Severity : enhancement
3116 Description: do not expand extent locks acquired on OST-side
3117 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3118 by server: they are not cached anyway.
3121 Frequency : when mmap is used/binaries executed from Lustre
3123 Description: Unmmap pages before throwing them away from read cache.
3124 Details : llap_shrink cache now attempts to unmap pages before discarding
3125 them (if unmapping failed - do not discard). SLES9 kernel has
3126 extra checks that trigger if this unmapping is not done first.
3131 Description: lconf didn't resolve symlinks before checking to see whether a
3132 given mountpoint was already in use
3135 Frequency : when migrating failover services
3136 Bugzilla : 6395, 9514
3137 Description: When migrating a subset of services from a node (e.g. failback
3138 from a failover service node) the remaining services would
3139 time out and evict clients.
3140 Details : lconf --force (implied by --failover) sets the global obd_timeout
3141 to 5 seconds in order to quickly disconnect, but this caused
3142 other RPCs to time out too quickly. Do not change the global
3143 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3145 Severity : enhancement
3146 Frequency : if MDS is started with down OST
3147 Bugzilla : 9439,5706
3148 Description: Allow startup/shutdown of an MDS without depending on the
3149 availability of the OSTs.
3150 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3151 Add appropriate locking and lov-osc refcounts for safe
3152 cleaning. Add osc abort_inflight calls in case the
3153 synchronize never started.
3156 Frequency : occasional (Cray XT3 only)
3158 Description: root not authorized to access files in CRAY_PORTALS environment
3159 Details : The client process capabilities were not honoured on the MDS in
3160 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3161 been accessed by an authorized user then root was able to access
3162 the file on the local client also. The root user capabilities
3163 are now allowed on the MDS, as this environment has secure UID.
3166 Frequency : occasional
3168 Description: ldiskfs "too long searching" message happens too often
3169 Details : A debugging message (otherwise harmless) prints too often on
3170 the OST console. This has been reduced to only happen when
3171 there are fragmentation problems on the filesystem.
3176 Description: Division by zero in statfs when all OSCs are inactive
3177 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3178 lov_statfs divides by value returned by lov_get_stripecnt().
3182 Bugzilla : 9489, 3273
3183 Description: First write from each client to each OST was only 4kB in size,
3184 to initialize client writeback cache, which caused sub-optimal
3185 RPCs and poor layout on disk for the first writen file.
3186 Details : Clients now request an initial cache grant at (re)connect time
3187 and so that they can start streaming writes to the cache right
3188 away and always do full-sized RPCs if there is enough data.
3189 If the OST is rebooted the client also re-establishes its grant
3190 so that client cached writes will be honoured under the grant.
3195 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3196 Details : Now I/O RPCs go to different portal number and (presumably) fast
3197 lock requests (and glimses) and other RPCs get their own service
3198 threads pool that should be able to service those RPCs
3201 Severity : enhancement
3203 Description: Ability to exchange lustre version between client and servers and
3204 issue warnings at client side if client is too old. Also for
3205 liblustre clients there is ability to refuse connection of too old
3207 Details : New 'version' field is added to connect data structure that is
3208 filled with version info. That info is later checked by server and
3212 Frequency : rare, liblustre only.
3213 Bugzilla : 9296, 9581
3214 Description: Two simultaneous writes from liblustre at offset within same page
3215 might proceed at the same time overwriting eachother with stale
3217 Details : I/O lock withing llu_file_prwv was released too early, before data
3218 actually was hitting the wire. Extended lock-holding time until
3219 server acknowledges receiving data.
3222 Frequency : extremely rare. Never observed in practice.
3224 Description: avoid generating lustre_handle cookie of 0.
3225 Details : class_handle_hash() generates handle cookies by incrementing
3226 global counter, and can hit 0 occasionaly (this is unlikely, but
3227 not impossible, because initial value of cookie counter is
3228 selected randonly). Value of 0 is used as a sentinel meaning
3229 "unassigned handle" --- avoid it. Also coalesce two critical
3230 sections in this function into one.
3232 Severity : enhancement
3234 Description: allow liblustre clients to delegate truncate locking to OST
3235 Details : To avoid overhead of locking, liblustre client instructs OST to
3236 take extent lock in ost_punch() on client's behalf. New connection
3237 flag is added to handle backward compatibility.
3239 Severity : enhancement
3240 Bugzilla : 4928, 7341, 9758
3241 Description: allow number of OST service threads to be specified
3242 Details : a module parameter allows the number of OST service threads
3243 to be specified via "options ost ost_num_threads={N}" in the
3244 OSS's /etc/modules.conf or /etc/modprobe.conf.
3248 Bugzilla : 6146, 9635, 9895
3249 Description: servers crash with bad pointer in target_handle_connect()
3250 Details : In rare cases when a client is reconnecting it was possible that
3251 the connection request was the last reference for that export.
3252 We would temporarily drop the export reference and get a new
3253 one, but this may have been the last reference and the export
3254 was just destroyed. Get new reference before dropping old one.
3256 Severity : enhancement
3257 Frequency : if client is started with failover MDS
3259 Description: Allow multiple MDS hostnames in the mount command
3260 Details : Try to read the configuration from all specified MDS
3261 hostnames during a client mount in case the "primary"
3264 Severity : enhancement
3266 Description: Stop sending data to evicted clients as soon as possible.
3267 Details : Check if the client we are about to send or are sending data to
3268 was evicted already. (Check is done every second of waiting,
3269 for which l_wait_event interface was extended to allow checking
3270 of exit condition at specified intervals).
3273 Frequency : rare, normally only when NFS exporting is done from client
3275 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3277 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3280 Frequency : always, if srand() is called before liblustre initialization
3282 Description: Liblustre uses system PRNG disturbing its usage by user application
3283 Details : Introduce internal to lustre fast and high-quality PRNG for
3284 lustre usage and make liblustre and some other places in generic
3285 lustre code to use it.
3287 Severity : enhancement
3288 Bugzilla : 9477, 9557, 9870
3289 Description: Verify that the MDS configuration logs are updated when xml is
3290 Details : Check if the .xml configuration logs are newer than the config
3291 logs stored on the MDS and report an error if this is the case.
3292 Request --write-conf, or allow starting with --old_conf.
3294 Severity : enhancement
3296 Description: Handle symlinks in the path when checking if Lustre is mounted.
3297 Details : Resolve intermediate symlinks when checking if a client has
3298 mounted a filesystem to avoid duplicate client mounts.
3303 Description: lconf can hit an error exception but still return success.
3304 Details : The lconf command catches the Command error exception at the top
3305 level script context and will exit with the associated exit
3306 status, but doesn't ensure that this exit status is non-zero.
3311 Description: failure of ptlrpc thread startup can cause oops
3312 Details : Starting a ptlrpc service thread can fail if there are a large
3313 number of threads or the server memory is very fragmented.
3314 Handle this without oopsing.
3317 Frequency : always, only if liblustre and non-default acceptor port was used
3319 Description: liblustre cannot connect to servers with non-default acceptor port
3320 Details : tcpnal_set_default_params() was not called and was therefore
3321 ignoring the environment varaible TCPNAL_PORT, as well as other
3322 TCPNAL_ environment variables
3327 Description: two objects could be created on the same OST for a single file
3328 Details : If an OST is down, in some cases it was possible to create two
3329 objects on a single OST for a single file. No problems other
3330 than potential performance impact and spurious error messages.
3334 Bugzilla : 5681, 9562
3335 Description: Client may oops in ll_unhash_aliases
3336 Details : Client dcache may become inconsistent in race condition.
3337 In some cases "getcwd" can fail if the current directory is
3343 Description: Inode refcounting problems in NFS export code
3344 Details : link_raw functions used to call d_instantiate without obtaining
3345 extra inode reference first.
3349 Bugzilla : 9942, 9903
3350 Description: Referencing freed requests leading to crash, memleaks with NFS.
3351 Details : We used to require that call to ll_revalidate_it was always
3352 followed by ll_lookup_it. Also with revalidate_special() it is
3353 possible to call ll_revalidate_it() twice for the same dentry
3354 even if first occurence returned success. This fix changes semantic
3355 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3356 reference on a request referred from the intent.
3357 ll_intent_release() then releases such a request.
3360 Frequency : rare, normally benchmark loads only
3362 Description: unlinked inodes were kept in memory on the client
3363 Details : If a client is repeatedly creating and unlinking files it
3364 can accumulate a lot of stale inodes in the inode slab cache.
3365 If there is no other client load running this can cause the
3366 client node to run out of memory. Instead flush old inodes
3367 from client cache that have the same inode number as a new inode.
3370 Frequency : SLES9 2.6.5 kernel and long filenames only
3371 Bugzilla : 9969, 10379
3372 Description: utime reports stale NFS file handle
3373 Details : SLES9 uses out-of-dentry names in some cases, which confused
3374 the lustre dentry revalidation. Change it to always use the
3378 Frequency : rare, unless heavy write-truncate concurrency is continuous
3379 Bugzilla : 4180, 6984, 7171, 9963, 9331
3380 Description: OST becomes very slow and/or deadlocked during object unlink
3381 Details : filter_destroy() was holding onto the parent directory lock
3382 while truncating+unlinking objects. For very large objects this
3383 may block other threads for a long time and slow overall OST
3384 responsiveness. It may also be possible to get a lock ordering
3385 deadlock in this case, or run out of journal credits because of
3386 the combined truncate+unlink. Solution is to do object truncate
3387 first in one transaction without parent lock, and then do the
3388 final unlink in a new transaction with the parent lock. This
3389 reduces the lock hold time dramatically.
3392 Frequency : rare, 2.4 kernels only
3394 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3395 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3396 before unmounting their filesystem in order to avoid deadlock.
3397 The kernel_locked() function in 2.4 kernels only checks whether
3398 the kernel lock is held, not whether it is this process that is
3399 holding it as 2.6 kernels do.
3404 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3405 Details : The client ptlrpc code may be trying to reconnect to a down
3406 server before a previous connection attempt has timed out.
3407 Increase the reconnect interval to be longer than the connection
3408 timeout interval to avoid sending duplicate connections to
3412 Frequency : echo_client brw_test command
3414 Description: fix echo_client to work with OST preallocated code
3415 Details : OST preallocation code (5137) didn't take echo_client IO path
3416 into account: echo_client calls filter methods outside of any
3417 OST thread and, hence, there is no per-thread preallocated
3418 pages and buffers to use. Solution: hijack pga pages for IO. As
3419 a byproduct, this avoids unnecessary data copying.
3423 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3424 Description: Client can oops in mdc_commit_close() after open replay
3425 Details : It was possible for the MDS to return an open request with no
3426 transaction number in mds_finish_transno() if the client was
3427 evicted, but without actually returning an error. Clients
3428 would later try to replay that open and may trip an assertion
3429 Simplify the client close codepath, and always return an error
3430 from the MDS in case the open is not successful.
3433 Frequency : rare, 2.6 OSTs only
3435 Description: OST may deadlock under high load on fragmented files
3436 Details : If there was a heavy load and highly-fragmented OST filesystems
3437 it was possible to have all the OST threads deadlock waiting on
3438 allocation of biovecs, because the biovecs were not released
3439 until the entire RPC IO was completed. Instead, release biovecs
3440 as soon as they are complete to ensure forward IO progress.
3442 Severity : enhancement
3444 Description: Support for specifying external journal device at mount
3445 Details : If an OST or MDS device is formatted with an external journal
3446 device, this device major/minor is stored in the ext3 superblock
3447 and may not be valid for failover. Allow detecting and
3448 specifying the external journal at mount time.
3453 Description: Mounting an MDS with pending unlinked files may cause oops
3454 Details : target_finish_recovery() calls mds_postrecov() which returned
3455 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3456 considers this an error and immediately begins cleaning up the
3457 lov, just after starting the mds_lov process
3459 Severity : enhancement
3461 Description: Implement 'lfs df' to report actual free space on per-OST basis
3462 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3463 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3464 usage of objects; '-h' to report in human readable format.
3466 ------------------------------------------------------------------------------
3468 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3475 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3476 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3477 at startup during orphan unlinking. Instead, skip buddy update
3478 before setup, as it will be regenerated after recovery is complete.
3483 Description: If an OST is inactive, its locks might reference stale inodes.
3484 Details : lov_change_cbdata() must iterate over all namespaces, even if
3485 they are inactive to clear inode references from the lock.
3487 Severity : enhancement
3488 Frequency : occasional, if non-standard max_dirty_mb used
3490 Description: Client will block write RPCs if not enough grant
3491 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3492 then the client will block writes while waiting for another RPC
3493 to complete instead of consuming its dirty limit. With change
3494 we get improved performance when max_dirty_mb is small.
3496 Severity : enhancement
3497 Bugzilla : 3389, 6253
3498 Description: Add support for supplementary groups on the MDS.
3499 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3500 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3501 lookups for user supplementary groups into a cache.
3505 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3506 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3507 for liblustre O_EXCL is enforced by MDS.
3511 Description: readdir never returns NULL in liblustre.
3512 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3513 dentry in struct dirent.
3517 Frequency : liblustre only, depends on application IO pattern
3518 Description: liblustre clients evicted if not contacting servers
3519 Details : Don't put liblustre clients into the ping_evictor list, so
3520 they will not be evicted by the pinger ever.
3522 Severity : enhancement
3524 Description: Add ability to evict clients by NID from MDS.
3525 Details : By echoing "nid:$NID" string into
3526 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3527 $NID would be instantly evicted from this MDS and from all active
3528 OSTs connected to it.
3532 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3533 Details : lookup_it_finish() used to query file size from OSTs that was not
3538 Description: service threads change working directory to that of init
3539 Details : Starting lustre service threads may pin the working directory
3540 of the parent thread, making that filesystem busy. Threads
3541 now change to the working directory of init to avoid this.
3545 Frequency : during shutdown only
3546 Description: shutdown with a failed MDS or OST can cause unmount to hang
3547 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3552 Frequency : 2.6 only
3553 Description: chmod/chown may include an extra supplementary group
3554 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3555 supplementary group and if none is specified this is used.
3558 Bugzilla : 5479 (6816)
3560 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3561 Details : If lookup is in progress on a file that is unlinked we might try
3562 to revalidate the inode and fail in revalidate after lookup is
3563 complete and ll_file_open() enqueues the open again but
3564 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3567 Frequency : always, if lconf --abort_recovery used
3569 Description: lconf --abort_recovery fails with 'Operation not supported'
3570 Details : lconf was attempting to abort recovery on the MDT device and not
3573 ------------------------------------------------------------------------------
3575 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3580 Frequency : rare (only unsupported configurations with a node running as an
3582 Bugzilla : 6514, 5137
3583 Description: Mounting a Lustre file system on a node running as an OST could
3585 Details : OSTs now preallocates memory needed to write out data at
3586 startup, instead of when needed, to avoid having to
3587 allocate memory in possibly low memory situations.
3588 Specifically, if the file system is mounted on on OST,
3589 memory pressure could force it to try to write out data,
3590 which it needed to allocate memory to do. Due to the low
3591 memory, it would be unable to do so and the node would
3592 become unresponsive.
3594 Severity : enhancement
3596 Description: Addition of lconf --service command line option
3597 Details : lconf now accepts a '--service <arg>' option, which is
3598 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3600 Severity : enhancement
3602 Description: Failover mode is now the default for OSTs.
3603 Details : By default, OSTs will now run in failover mode. To return to
3604 the old behaviour, add '--failout' to the lmc line for OSTs.
3606 Severity : enhancement
3608 Description: Health checks are now provided for MDS and OSTs
3609 Details : Additional detailed health check information on MSD and OSTs
3610 is now provided through the procfs health_check value.
3613 Frequency : occasional, depends on IO load
3615 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3616 after numerous create/delete cycles
3617 Details : The ext3 inode allocation policy would not allocate new inodes
3618 very well on the OSTs because there are no new directories
3619 being created. Instead we look for groups with free space if
3620 the parent directories are nearly full.
3625 Description: Network or server problems during mount may cause partially
3626 mounted clients instead of returning an error.
3627 Details : The config llog parsing code may overwrite the error return
3628 code during mount error handling, returning success instead
3634 Description: MDS can fail to allocate large reply buffers
3635 Details : After long uptimes the MDS can fail to allocate large reply
3636 buffers (e.g. zconf client mount config records) due to memory
3637 fragmentation or consumption by the buffer cache. Preallocate
3638 some large reply buffers so that these replies can be sent even
3639 under memory pressure.
3643 Frequency : rare (liblustre)
3644 Description: fsx running with liblustre complained that using truncate() to
3645 extend the file doesn't work. This patch corrects that issue.
3646 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3647 ATTR_SIZE and lsm use in llu_setattr_raw.
3651 Frequency : rare, only 2.6 kernels
3652 Description: Unusual file access patterns on the MDS may result in inode
3653 data being lost in very rare circumstances.
3654 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3655 code in the 2.6 kernel caused the inode and/or EA data not to
3656 be read from disk, causing single-file corruption.
3660 Frequency : rare, only 2.6 filesystems using extents
3661 Description: Heavy concurrent write and delete load may cause data corruption.
3662 Details : It was possible under high-load situations to have an extent
3663 metadata block in the block device cache from a just-unlinked
3664 file overwrite a newly-allocated data block. We now unmap any
3665 metadata buffers that alias just-allocated data blocks.
3669 Frequency : filesystems with default stripe_count larger than 77
3670 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3671 Details : lconf specifies an inode size of 4096 bytes when the default
3672 stripe_count is larger than 77. This conflicts with the default
3673 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3674 this case to avoid pinning too much memory for large EAs.
3676 ------------------------------------------------------------------------------
3678 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3683 Frequency : rare (extremely heavy IO load with hundreds of clients)
3685 Description: Client is evicted, gets IO error writing to file
3686 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3687 caused clients to be evicted for AST timeouts. The fixes in
3688 bug 5192 mean we no longer need to have such short AST timeouts
3689 so ldlm_timeout has been increased.
3692 Frequency : occasional during --force or --failover shutdown under load
3693 Bugzilla : 5949, 4834
3694 Description: Server oops/LBUG if stopped with --force or --failover under load
3695 Details : a collection of import/export refcount and cleanup ordering
3696 issues fixed for safer force cleanup
3699 Frequency : only filesystems larger than 120 OSTs
3700 Bugzilla : 5990, 6223
3701 Description: lfs getstripe would oops on a very large filesystem
3702 Details : lov_getconfig used kfree on vmalloc'd memory
3705 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3706 Bugzilla : 6242, 6243
3707 Description: reading from files that had been truncated to a non-zero size
3708 but never opened returned no data
3709 Details : ll_file_read() reads zeros from no-object files to EOF
3714 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3715 Details : There's little harm in aborting MDS/OSS recovery and letting it
3716 try again, so I removed the LASSERT and return an error instead.
3718 Severity : enhancement
3720 Description: New debugging infrastructure for tracking down data corruption
3721 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3722 (b) cover more of the client-side code path, and (c) try to narrow
3723 down where problems occurred
3727 Bugzilla : 3819, 4364, 4397, 6313
3728 Description: Racing close and eviction MDS could cause assertion in mds_close
3729 Details : It was possible to get multiple mfd references during close and
3730 client eviction, leading to one thread referencing a freed mfd.
3732 Severity: : enhancement
3733 Bugzilla : 3262, 6359
3734 Description: Attempts to reconnect to servers are now more aggressive.
3735 Details : This builds on the enhanced upcall-less recovery that was added
3736 in 1.4.2. When trying to reconnect to servers, clients will
3737 now try each server in the failover group every 10 seconds. By
3738 default, clients would previously try one server every 25 seconds.
3743 Description: After recovery, certain operations trigger a failed
3744 assertion on a client.
3745 Details : Failing over an mds, using lconf -d --failover, while a
3746 client was doing a readdir() call would cause the client to
3747 LBUG after recovery completed and the readdir() was resent.
3749 Severity : enhancement
3751 Description: Default groups are now added by lconf
3752 Details : You can now run lconf --group <servicename> without having to
3753 manually add groups with lmc.
3756 Frequency : occasional
3758 Description: Nodes with an elan id of 0 trigger a failed assertion
3761 Frequency : always when accessing e.g. tty/console device nodes
3763 Description: tty and some other devices nodes cannot be used on lustre
3764 Details : file's private_data field is used by device data and lustre
3765 values in there got lost. New field was added to struct file to
3766 store fs-specific private data.
3769 Frequency : when exporting Lustre via NFS
3771 Description: NFSD failed occasionally when looking up a path component
3772 Details : NFSD is looking up ".." which was broken in ext3 directories
3773 that had grown large enough to become hashed.
3776 Frequency : Clusters with multiple interfaces not on the same subnet
3778 Description: Nodes will repeatedly try to reconnect to an interface which it
3779 cannot reach and report an error to the log.
3780 Details : Extra peer list entries will be created by lconf with some peers
3781 unreachable. lconf now validates the peer before adding it.
3784 Frequency : Only if a default stripe is set on the filesystem root.
3786 Description: Setting a default stripe on the filesystem root prevented the
3787 filesystem from being remounted.
3788 Details : The client was sending extra request flags in the root getattr
3789 request and did not allocate a reply buffer for the dir EA.
3792 Frequency : occasional, higher if lots of files are accessed by one client
3793 Bugzilla : 6159, 6097
3794 Description: Client trips assertion regarding lsm mismatch/magic
3795 Details : While revalidating inodes the VFS looks up inodes with ifind()
3796 and in rare cases can find an inode that is being freed.
3797 The ll_test_inode() code will free the lsm during ifind()
3798 when it finds an existing inode and then the VFS later attaches
3799 this free lsm to a new inode.
3803 Bugzilla : 6422, 7030
3804 Description: MDS deadlock between mkdir and client eviction
3805 Details : Creating a new file via mkdir or mknod (starting a transaction
3806 and getting the ns lock) can deadlock with client eviction
3807 (gets ns lock and trying to finish a synchronous transaction).
3810 Frequency : occasional
3811 Description: While starting a server, the fsfilt_ext3 module could not be
3813 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3814 kernels. Previously, lconf would still use the ext3 name
3815 when trying to load modules. Now, it will correctly use
3816 ext3 on 2.4 and ldiskfs on 2.6.
3818 Severity : enhancement
3819 Description: The default stripe count has been changed to 1
3820 Details : The interpretation of the default stripe count (0, to lfs
3821 or lmc) has been changed to mean striping across a single
3822 OST, rather than all available. For general usage we have
3823 found a stripe count of 1 or 2 works best.
3825 Severity : enhancement
3826 Description: Add support for compiling against Cray portals.
3827 Details : Conditional compiling for some areas that are different
3831 Frequency : occasional
3832 Bugzilla : 6409, 6834
3833 Description: Creating files with an explicit stripe count may lead to
3834 a failed assertion on the MDS
3835 Details : If some OSTs are full or unavailable, creating files may
3836 trigger a failed assertion on the MDS. Now, Lustre will
3837 try to use other servers or return an error to the
3841 Frequency : occasional
3843 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3844 caused lock timeout during readahead (since 1.4.2).
3845 Details : Processes doing readahead might match a lock that hasn't been
3846 granted yet if there are overlapping and conflicting lock
3847 requests. The readahead process waits on ungranted lock
3848 (original lock is CBPENDING), while OST waits for that process
3849 to cancel CBPENDING read lock and eventually evicts client.
3851 Severity : enhancement
3853 Description: Initial enabling of flock support for clients
3854 Details : Implements fcntl advisory locking and file status functions.
3855 This feature is provided as an optional mount flag (default
3856 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3857 locking are implemented yet, and those that are are not guaranteed
3858 to be completely correct in production environments.
3859 mount -t lustre -o [flock|noflock] ...
3862 Frequency : occasional
3864 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3865 assertion in the ext3 JBD (journaling) layer.
3866 Details : The b_committed_data struct is protected by the big kernel lock
3867 in 2.4 kernels, serializing journal_commit_transaction() and
3868 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3869 access to this struct. In 2.6 kernels there is finer grained
3870 locking to improve SMP performance of the JBD layer.
3874 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3879 Description: merge and cleanup kernel patches.
3882 -----------------------------------------------------------------------------
3884 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3886 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3887 versions of Lustre. Please update all servers and clients to
3888 version 1.4.2 or later at the same time. You must also run
3889 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3890 to update the configuration logs.
3892 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3893 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3894 - swab llog records of type '0' so we get proper header size/idx (5861)
3895 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3896 - fix rename of one directory over another leaking an inode (5953)
3897 - avoid SetPageDirty on 2.6 (5981)
3898 - don't re-add just-being-destroyed locks to the waiting list (5653)
3899 - when creating new directories, inherit the parent's custom
3900 striping settings if present parent (3048)
3901 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3902 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3903 - don't hold client locks on temporary worklist from l_lru (5666)
3904 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3905 - automatically evict dead clients (5921)
3906 - Update file size properly in create+truncate+fstat case (6196)
3907 - Do not unhash mountpoint dentries, do not allow removal of
3909 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3910 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3911 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3912 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3913 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3914 - Correctly handle reads of files with no objects (6243)
3915 - lctl recover will also mark a device active if deactivate used (5933)
3917 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3918 - allow --write-conf on an MDS with different nettype than client (5619)
3919 - don't write config llogs to MDS for mounts not from that MDS (5617)
3920 - lconf should create multiple TCP connections from a client (5201)
3921 - init scripts are now turned off by default; run chkconfig --on
3922 lustre and chkconfig --on lustrefs to use them
3923 - upcalls are no longer needed for clients to recover to failover
3925 - add --abort-recovery option to lconf to abort recovery on device
3927 - add support for an arbitrary number of OSTs (3026)
3928 - Quota support protocol changes.
3929 - forward compatibility changes to wire structs (6007)
3930 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3931 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3932 - improved "lustre status" script
3933 - initialize blocksize for non-regular files (6062)
3934 - added --disable-server and --disable-client configure options (5782)
3935 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3936 - Vanilla 2.4.29 support
3937 - increase maximum number of obd devices to 520 (6242)
3938 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3939 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3940 SLES 9 2.6.5 kernels
3941 - limit stripes per file to 160 (the maximum EA size) (6093)
3943 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3946 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3947 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3948 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3949 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3950 - small ext3 extents cleanups and fixes (5733)
3951 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3952 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3953 - avoid needless client->OST connect, fix handle mismatch (5317)
3954 - fix DLM error path that led to out-of-sync client, long delays (5779)
3955 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3956 - fix several locking issues related to i_size (5492,5624,5654,5672)
3957 - don't move pending lock onto export if it is already evicted (5683)
3958 - fix kernel oops when creating .foo in unlinked directory (5548)
3959 - fix deadlock in obdfilter statistics vs. object create (5811)
3960 - use time_{before,after} to avoid timer jiffies wrap (5882)
3961 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3962 - Do not leak request if server was not able to process it (5154)
3963 - If mds_open unable to find parent dir, make that negative lookup(5154)
3964 - don't create new directories with extent-mapping (5909, 5936)
3966 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3967 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3968 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3969 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3970 - support for RHEL4 kernel on i686 (5773)
3971 - provide error messages when incompatible logs are encountered (5898)
3973 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3974 * version 1.4.0.10 (1.4.1 release candidate 1)
3976 - don't keep a lock reference when lock is not granted (4238)
3977 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3978 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3979 - drop import inflight refcount on signal_completed_replay error (5255)
3980 - unlock page after async write error during send (3677)
3981 - handle missing objects in filter_preprw_read properly (5265)
3982 - no transno return for symlink open, don't save no-trasno open (3440)
3983 - don't try to complete elan receive that already failed (4012)
3984 - free RPC server reply state on error (5406)
3985 - clean up thread from ptlrpc_start_thread() on error (5160)
3986 - readahead could read extra page into cache that wasn't ejected (5388)
3987 - prevent races in class_attach/setup/cleanup/detach (5260)
3988 - don't dereference de->d_inode after l_dput of de (5458)
3989 - use "int" for stripe value returned from lock_to_stripe (5544)
3990 - mballoc allocation and error-checking fixes in 2.6 (5504)
3991 - block device patches to fix I/O request sizes in 2.6 (5482)
3992 - look up hostnames for IB nals (5602)
3993 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3994 - don't start multiple acceptors for the same port (5277)
3995 - fix incorrect LASSERT in mds_getattr_name (5635)
3996 - export a proc file for general "ping" checking (5628)
3997 - fix "lfs check" to not block when the MDS is down (5628)
3999 - service request history (4965)
4000 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4001 - create an "evict_client" /proc entry on OSTs, like the MDS has
4002 - fix mount usage message, return errors per mount(8) (5168)
4003 - change grep [] to grep "[]" in tests so they work in more UMLs
4004 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4005 - remove ancient LOV_MAGIC_V0 EA support (5047)
4006 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4007 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4008 - allow readahead allocations to fail when low on memory (5383)
4009 - mmap locking landed again, after considerable improvement (2828)
4010 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4012 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4015 - send OST transaction number in read/write reply to free req (4966)
4016 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4017 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4018 - the watchdog thread now runs as interruptible (5246)
4019 - flock/lockf fixes (but it's still disabled, pending 5135)
4020 - don't use EXT3 constants in llite code (5094)
4021 - memory shortage at startup could cause assertion (5176)
4023 - reorganization of lov code
4024 - single portals codebase
4026 - add extents/mballoc support (5025)
4027 - direct I/O reads in the obdfilter (4048)
4028 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4030 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4033 - send OST transaction number in read/write reply to free req (4966)
4034 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4035 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4036 - the watchdog thread now runs as interruptible (5246)
4037 - handle missing objects in filter_preprw_read properly (5265)
4038 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4039 - drop import inflight refcount on signal_completed_replay error (5255)
4040 - unlock page after async write error during send (3677)
4041 - return original error code on reconstructed replies (3761)
4042 - no transno return for symlink open, don't save no-trasno open (3440)
4044 - add pid to ldlm debugging output (4922)
4045 - bump the watchdog timeouts -- we can't handle 30sec yet
4046 - extra debugging for orphan dentry/inode bug (5259)
4048 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
4051 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4052 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4053 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
4054 - fix NULL dereference in /proc/sys/portals/routes (4827)
4055 - allow failed mdc_close() operations to be interrupted (4561)
4056 - stop precreate on OST before MDS would time out on it (4778)
4057 - don't send partial-page writes before EOF from client (4410)
4058 - discard client grant for sub-page writes on large-page clients (4520)
4059 - don't free dentries not owned by NFS code, check generation (4806)
4060 - fix lsm leak if mds_create_objects() fails (4801)
4061 - limit debug_daemon file size, always print CERROR messages (4789)
4062 - use transno after validating reply (3892)
4063 - process timed out requests if import state changes (3754)
4064 - update mtime on OST during writes, return in glimpse (4829)
4065 - add mkfsoptions to LDAP (4679)
4066 - use ->max_readahead method instead of zapping global ra (5039)
4067 - don't interrupt __l_wait_event() during strace
4069 - add software watchdogs to catch hung threads quickly (4941)
4070 - make lustrefs init script start after nfs is mounted
4071 - fix CWARN/ERROR duplication (4930)
4072 - return async write errors to application if possible (2248)
4073 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4074 - print NAL number in %x format (4645)
4075 - update barely-supported suse-2.4.21-171 series (4842)
4076 - support for sles 9 %post scripts
4077 - support for building 2.6 kernel-source packages
4078 - support for sles km_* packages
4080 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4083 - ignore -ENOENT errors in osc_destroy (3639)
4084 - notify osc create thread that OSC is being cleaned up (4600)
4085 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4086 - reconstruct ost_handle() like mds_handle() (4657)
4087 - create a new thread to do import eviction to avoid deadlock (3969)
4088 - let lconf resolve symlinked-to devices (4629)
4089 - don't unlink "objects" from directory with default EA (4554)
4090 - hold socknal file ref over connect in case target is down (4394)
4091 - allow more than 32000 subdirectories in a single directory (3244)
4092 - fix blocks count for O_DIRECT writes (3751)
4093 - OST returns ENOSPC from object create when no space left (4539)
4094 - don't send truncate RPC if file size isn't changing (4410)
4095 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4096 - bind to privileged port in socknal and tcpnal (3689)
4098 - rate limit CERROR/CWARN console message to avoid overload (4519)
4099 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4100 - basic mmap support (3918)
4101 - kernel patch series update from b1_4 (4711)
4103 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4106 - avoid crash during MDS cleanup with OST shut down (2775)
4107 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4108 - don't use bad inodes on the MDS (3744)
4109 - dynamic object preallocation to improve recovery speed (4236)
4110 - don't hold spinlock over lock dumping or change debug flags (4401)
4111 - don't zero obd_dev when it is force cleaned (3651)
4112 - print grants to console if they go negative (4431)
4113 - "lctl deactivate" will stop automatic recovery attempts (3406)
4114 - look for existing locks in ldlm_handle_enqueue() (3764)
4115 - don't resolve lock handle twice in recovery avoiding race (4401)
4116 - revalidate should check working dir is a directory (4134)
4118 - don't always mark "slow" obdfilter messages as errors (4418)
4120 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4123 - don't close LustreDB during write_conf until it is done (3860)
4124 - fix typo in lconf for_each_profile (3821)
4125 - allow dumping logs from multiple threads at one time (3820)
4126 - don't allow multiple threads in OSC recovery (3812)
4127 - fix debug_size parameters (3864)
4128 - fix mds_postrecov to initialize import for llog ctxt (3121)
4129 - replace config semaphore with spinlock (3306)
4130 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4131 - don't allow enqueue to complete on a destroyed export (3822)
4132 - down write_lock before checking llog header bitmap (3825)
4133 - recover from lock replay timeout (3764)
4134 - up llog sem before sending rpc (3652)
4135 - reduce ns lock hold times when setting kms (3267)
4136 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4137 - fix NULL deref and obd_dev leak on setup error (3312)
4138 - replace some LBUG about llog ops with error handling (3841)
4139 - don't match INVALID dentries from d_lookup and spin (3784)
4140 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4141 - fix invalid assertion in ptlrpc_set_wait (3880)
4143 - add libwrap support for the TCP acceptor (3996)
4144 - add /proc/sys/portals/routes for non-root route listing (3994)
4145 - allow setting MDS UUID in .xml (2580)
4146 - print the stack of a process that LBUGs (4228)
4148 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4151 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4152 - make sure to unset replay flag from failed open requests (3440)
4153 - if default stripe count is 0, use OST count for inode size (3636)
4154 - update parent mtime/ctime on client for create/unlink (2611)
4155 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4156 - free recovery state on server during a forced cleanup (3571)
4157 - unregister_reply for resent reqs (3063)
4158 - loop back devices mounting and status check on 2.6 (3563)
4159 - fix resource-creation race that can provoke i_size == 0 (3513)
4160 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4161 - more debugging for page-accounting assertion (3746)
4162 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4163 - avoid deadlock after precreation failure (3758)
4164 - fix race and lock order deadlock in orphan handling (3450, 3750)
4165 - add validity checks when grabbing inodes from l_ast_data (3599)
4167 - add /proc/.../recovery_status to obdfilter (3428)
4168 - lightweight CDEBUG infrastructure, debug daemon (3668)
4169 - change default OSC RPC parameters to be better on small clusters
4170 - turn off OST read cache for files smaller than 32MB
4171 - install man pages and include them in rpms (3100)
4172 - add new init script for (un)mounting lustre filesystems (2593)
4173 - run chkconfig in %post for init scripts (3701)
4174 - drop scimac NAL (unmaintained)
4176 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4179 - clean kiobufs before and after use (3485)
4180 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4181 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4182 - change init script to fail more clearly if not run as root (1528)
4183 - allow clients to reconnect during replay (1742)
4184 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4185 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4186 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4187 - fix cond_resched (3554)
4188 - only evict unfinished clients after recovery (3515)
4189 - allow bulk resend, prevent data loss (3570)
4190 - dynamic ptlrpc request buffer allocation (2102)
4191 - don't allow unlinking open directory if it isn't empty (2904)
4192 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4193 - remove extraneous obd dereference causing LASSERT failure (3334)
4194 - don't use get_cycles() when creating temp. files on the mds (3156)
4195 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4196 - handle EEXIST for set-stripe, set proper directory name (3336)
4198 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4199 - fix ksocknal_fmb_callback() error messages (2918)
4201 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4204 - don't copy lvb into (possibly NULL) reply on error (2983)
4205 - don't deref dentry after dput, don't free lvb on error (2922)
4206 - use the kms to determine writeback rpc length (2947)
4207 - increment oti_logcookies when osc is inactive (2948)
4208 - update client's i_blocks count via lvb messages (2543)
4209 - handle intent open/close of special files properly (1557)
4210 - mount MDS with errors=remount-ro, like obdfilter (2009)
4211 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4212 - don't use cancelling-locks' kms values (2947)
4213 - use highest lock extent for kms, not last one (2925)
4214 - don't dereference ERR_PTR() dentry in error handling path (3107)
4215 - fix thread race in portals_debug_dumplog() (3122)
4216 - create lprocfs device entries at setup instead of at attach (1519)
4217 - common AST error handler, don't evict client on completion race (3145)
4218 - zero nameidata in detach_mnt in 2.6 (3118)
4219 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4220 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4221 - initialize RPC timeout timer earlier for 2.6 (3219)
4222 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4223 - print nal/nid for unknown nid (3258)
4224 - additional checks for oscc recovery before doing precreate (3284)
4225 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4226 - don't crash in mdc_close for bad permissions on open (3285)
4227 - zero i_rdev for non-device files (3147)
4228 - clear page->private before handing to FS, better assertion (3119)
4229 - tune the read pipeline (3236)
4230 - fix incorrect decref of invalidated dentry (2350)
4231 - provide read-ahead stats and refine rpc in flight stats (3328)
4232 - don't hold journal transaction open across create RPC (3313)
4233 - update atime on MDS at close time (3265)
4234 - close LDAP connection when recovering to avoid server load (3315)
4235 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4236 - don't leak open file on MDS after open resend (3325)
4237 - serialize filter_precreate and filter_destroy_precreated (3329)
4238 - loop device shouldn't call sync_dev() for nul device (3092)
4239 - clear page cache after eviction (2766)
4240 - resynchronize MDS->OST in background (2824)
4241 - refuse to mount the same filesystem twice on same mountpoint (3394)
4242 - allow llmount to create routes for mounting behind routers (3320)
4243 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4244 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4245 - fix rare problem with rename on htree directories (3417)
4247 - allow default OST striping configuration per directory (1414)
4248 - fix compilation for qswnal for 2.6 kernels (3125)
4249 - increase maximum number of MDS request buffers for large systems
4250 - change liblustreapi to be useful for external progs like lfsck (3098)
4251 - increase local configuration timeout for slow disks (3353)
4252 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4254 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4257 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4258 - don't overwrite extent policy data in reply if lock was blocked (2901)
4259 - drop filter export grants atomically with removal from device (2663)
4260 - del obd_self_export from work_list in class_disconnect_exports (2908)
4261 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4262 - swab reply message in mdc_close, other PPC fixes (2464)
4263 - fix destroying of named logs (2325)
4264 - overwrite old logs when running lconf --write_conf (2264)
4265 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4266 - fix race in target_handle_connect (2898)
4267 - mds_reint_create() should take same inode create lock (2926)
4268 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4269 - don't close files for self_export to avoid uninitialized obd (2936)
4270 - allow MDS with the same name as client node (2939)
4271 - hold dentry reference for closed log files for unlink (2325)
4272 - reserve space for all logs during transactions (2059)
4273 - don't evict page beyond end of stripe extent (2925)
4274 - don't oops on a deleted current working directory (2399)
4275 - handle hard links to targets without a parent properly (2517)
4276 - don't dereference NULL lock when racing during eviction (2867)
4277 - don't grow lock extents when lots of conflicting locks (2919)
4279 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4282 - account for cache space usage on clients to avoid data loss (974)
4283 - lfsck support in lustre kernel code (2349)
4284 - reduce journal credits needed for BRW writes (2370)
4285 - orphan handling to avoid losing space on client/server crashes
4286 - ptlrpcd can be blocked, stopping ALL progress (2477)
4287 - use lock value blocks to assist in proper KMS, faster stat (1021)
4288 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4289 - recovery for initial connections (2355)
4290 - fixes for mds_cleanup_orphans (1934)
4291 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4292 - block all file creations until orphan recovery completes (1901)
4293 - client remove rq_connection from request struct (2423)
4294 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4295 - recovery timer race (2670)
4296 - mdc_close recovey bug (2532)
4297 - ptlrpc cleanup bug (2710)
4298 - mds timeout on local locks (2588)
4299 - namespace lock held during RPCs (2431)
4300 - handle interrupted sync write properly (2503)
4301 - don't try to handle a message that hasn't been replied to (2699)
4302 - client assert failure during cleanup after abort recovery (2701)
4303 - leak mdc device after failed mount (2712)
4304 - ptlrpc_check_set allows timedout requests to complete (2714)
4305 - wait for inflight reqs when ptlrpcd finishes (2710)
4306 - make sure unregistered services are removed from the srv_list
4307 - reset bulk XID's when resending them (caught by 1138 test)
4308 - unregister_bulk after timeout
4309 - fix lconf error (2694)
4310 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4311 - readahead locks pages, leaves pending causing memory pressure (2673)
4312 - increase OST request buffers to 4096 on large machines (2729)
4313 - fix up permission of existing directories in simple_mkdir (2661)
4314 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4315 - don't assign transno to errored transactions (2742)
4316 - don't delete objects on OST if given a bogus objid from MDS (2751)
4317 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4318 - if rq_no_resend, then timeout request after recovery (2432)
4319 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4320 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4321 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4322 - MDS recovery shouldn't delete objects that it has given out (2730)
4323 - if enqueue arrives after completion, don't clobber LVB (2819)
4324 - don't unlock pages twice when trigger_group_io returns error (2814)
4325 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4326 - don't write pages to disk if there was an error (1450)
4327 - don't ping imports that have recovery disabled (2676)
4328 - take buffered bytes into account when balancing socknal conn (2817)
4329 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4330 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4331 - remove little-endian swabbing of llog records (1987)
4332 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4333 - retry reposting request buffers if they fail (1191)
4334 - grow extent at grant time to avoid granting a revoked lock (2809)
4335 - lock revoke doesn't evict page if covered by a second lock (2765)
4336 - disable VM readahead to avoid reading outside lock extents (2805)
4338 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4339 - updated kernel patches for hp-2.4.20 kernel (2681)
4341 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4344 - fix truncated write corruption (2366)
4345 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4347 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4348 - ensure len(uuid) < 37 in lmc (1171)
4349 - fix ia64 OOPS in llog_test (2255)
4350 - zero end of page at obdfilter for partial page writes (2648)
4351 - don't leave stale dentries around after renames (bug 2428)
4352 - fix timeouts when evicting a client with a single lock held (2642)
4353 - set deadline for the initial HELLO message to drain (2634)
4354 - print out dotted-quad IP addresses in the socknal (2302)
4356 - additional debugging for MDS client eviction problem (2443)
4357 - fix mkfsoptions support for osts (2603, 2604)
4359 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4362 - add series for the vanilla 2.6.0 kernel
4363 - add series for the vanilla 2.4.24 kernel
4364 - add series for a cray x86/64 UL kernel drop
4365 - fix xattr patches for the vanilla 2.4.19 series
4367 - generate true UUIDs in lmc (1171)
4368 - have portals stack dumping break in UML (2466)
4369 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4370 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4371 - wake callback waiting threads on client eviction (2460)
4372 - Add --ptldebug and --subsystem to lmc (1719)
4373 - update assertion to allow safe interrupt allocation
4374 - set rq_no_resend for cancel requests (2432)
4375 - recalculate ptlrpcd timeout after resend (2494)
4376 - call vfs_rmdir when removing pending directories (2368)
4377 - fix renaming a file to itself (2429)
4378 - lmc creates a default one-stripe lov (2454)
4379 - expand procfs space to handle large clusters (2326)
4380 - increase UML stack to avoid overflow
4381 - update lconf's list of debug and subsystem masks
4382 - fix lfs find --obd (2510)
4383 - /proc tunable for disabling filter read caching (2591)
4384 - stop rpm packages from altering slapd.conf (2301)
4385 - disable nagle in the socknal under 0conf (2578)
4386 - choose mds inode size based on stripe count (2572)
4387 - fix kernel-source rpm problems (2516)
4389 - add --disable-doc to avoid pdf generation (2421)
4390 - update documentation, tests, type-os, comments
4391 - avoid format warnings on ia64
4392 - remove the TOE NAL
4393 - tiny code cleanups by removing unused fields
4395 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4398 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4399 - remove the most blatant lies from BUILDING file (2371)
4400 - change default debug level to reasonable production setting
4401 - reduce client side cache size to reduce cache flush time
4402 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4403 - make TCP zerocopy and pinger support enabled by default (2476)
4404 - sync writes completed after process exits caused crashes (2319)
4405 - maintain correct mount count on the MDS (2356)
4406 - backout 1557, because 2316 wasn't really fixed
4407 - better file I/O statistics gathering in /proc
4408 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4409 - another kernel patch to fix zero-copy TCP function export
4410 - don't take duplicate lock when processing re-sent getattr (2420)
4411 - lctl uses obd_self_export instead of creating new conn (2353)
4412 - MDS/OST recovery case which requires object creation asserted (2425)
4413 - move lfs from /usr/sbin to /usr/bin in packages
4414 - fix race between mds_client_add and mds_client_free (2417)
4415 - use kmalloc instead of slabs in portals (2430)
4416 - don't create duplicate records when a failover MDS is present (2442)
4417 - remove unnecessary mount age check (2332)
4418 - don't remove directory inodes from locks prematurely (2451)
4419 - don't break if MDS service name is the same as hostname (2103)
4420 - fix races in client write RPC generation when cache full (2482)
4422 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4425 - remove now-unused request->rq_obd (278)
4426 - if an allocation fails, print out how much memory we've used (1933)
4427 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4428 - add the "configurable stack size" patch to most series files (1256)
4429 - ability to write large log records, for 100+ OST configs (2306)
4430 - fix NULL deref when filter_prep fails (2314)
4431 - fix operator precedence error in filter_sync
4432 - dynamic allocation of socknal TX descriptors (2315)
4433 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4434 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4435 - ensure that CWARN messages in Portals always get to the syslog
4436 - __init/__exit are not for prototype decls (ldlm_init/exit)
4437 - x86-64 compile warning fixes
4438 - fix gateway LMC keyword conflict (2318)
4439 - fix MDS lock inversions in getattr/reint paths (1844)
4440 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4441 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4442 - DLM race condition prevented some lock evictions (2328)
4443 - ENOMEM detection and retry on socknal sends (2230)
4444 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4445 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4446 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4447 - better error messages when a client is rejected during recovery (1505)
4448 - avoid cancelling locks which were never granted, after failure (2330)
4449 - fix i_sem/journal inversion in mds_client_add (2333)
4450 - fix truncate/getattr lock cycle deadlock (2334)
4451 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4452 - fix two rare exit paths which could leak an l_lock() ref (2321)
4453 - fencepost error in MDS/OST orphan recovery (2226)
4454 - make log record alignment 8 bytes (1988)
4455 - lstripe now fails when requested offset > ost_count (2237)
4456 - ensure that all kernel series have a complete list.h (1607)
4457 - fix crashes in special-file operations (2316)
4458 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4460 - allow configurable automake binary, for testing new versions
4461 - small update to the lfs documentation
4463 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4465 * fix negative export reference count in fsfilt_sync (2312)
4467 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4468 * release candidate 0.9.1
4470 - orphans are moved into the PENDING directory for possible recovery
4471 - replayed opens now open by fid for orphan/rename safety (1042)
4472 - last close of an orphan inode generates a transno (683)
4473 - chdir() and mount() now pin the directory entry (1020)
4474 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4475 - discard very old requests without processing them (1502)
4476 - remove some common, well-understood CERRORs (1505)
4477 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4478 - clear "grant" flags in OST replies until OST grant code lands (1644)
4479 - fix read performance by not clobbering i_blksize on client (1598)
4480 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4481 - make LRU size a /proc tunable, clears locks when reduced (707)
4482 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4483 - server locks take references on exports now (1558)
4484 - build fixes for 2.4.20-rh trees (1663)
4485 - return an error from lov_create if all OSCs are inactive (1751)
4486 - fix import levels when a reconnect happens without a timeout (1597)
4487 - exit early from mds_open if we get a lookup error (1749)
4488 - partial page read at EOF wouldn't wait for disk before sending (1642)
4489 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4490 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4491 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4492 - write/truncate lock inversion (1639)
4493 - Don't auto-load obdclass, portals modules during cleanup (1495)
4494 - fix timestamps from jumping to "now" (1763)
4495 - extra journal assertions (1648)
4496 - add an extra multiunlink test (1771)
4497 - fix read_record/write_record API (1776)
4498 - fix leak of offset_extent, possible incorrect i_size later (1772)
4499 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4500 - Debugging for the kqswnal_get_idle_tx problems (1820)
4501 - Allow recovery to be attempted multiple times (1536)
4502 - Write out MDS last_rcvd file after it is first created (1600)
4503 - Fix tx_descriptor leak in failed transmit situations (1827)
4504 - ext3 journaling fixes for assertion failure after IO error (1871)
4505 - class_export_put() on freed export after completion AST error (1896)
4506 - Fix revalidate looping in VFS (1322)
4507 - Don't access a freed export during MDS_REINT timeout (1521)
4508 - Add open-unlink recovery support on the MDS (1673,1764)
4509 - Return an error if no MDS data was read from last_rcvd (1946)
4510 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4511 - Don't setup a disk device that doesn't match exported UUID (317)
4512 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4513 - avoid committing NULL handle in force close
4514 - local.sh is now a one-stripe LOV configuration
4515 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4516 - fix ext3 htree duplicate directory entry corruption (1516)
4517 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4518 - update to new LOV EA format (2097)
4519 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4521 2003-06-15 Phil Schwan <phil@clusterfs.com>
4524 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4525 - per-import/export recovery handling (958, 931, 959)
4526 - multiple last-rcvd slots, for serving multiple FSes (949)
4527 - connections are again shared between multiple imp/exports (963, 964)
4528 - "umount -f" would hang if any requests needed to be sent (393, 978)
4529 - avoid pinning large req buffer by copying for queued messages (989)
4530 - add "uuid" to "lctl device" command to help upcalls (991)
4531 - "open" RPCs with transnos would confuse recovery counters (1037)
4532 - do proper endian conversion of all wire messages (288, 340, 891)
4533 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4534 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4535 - fix ptlrpc_connection leak in target_handle_connect (1174)
4536 - fix import refcounting bug in OST and MDS cleanup (1134)
4537 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4538 - fix very unlikely obd_types race condition (501)
4539 - remove osc_open hack for echo_client (1187)
4540 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4541 - a failure in read_inode2 leads to deadlock (1139)
4542 - cancel ack-locks as soon as transaction is committed (1072)
4543 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4544 - make sure to commitrw after any preprw to avoid deadlock (1162)
4545 - failing to execute a file in a lustre FS would lock inode (1203)
4546 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4547 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4548 - obd_uuid2tgt would walk off the end of the list (1255)
4549 - on IA64 the osc would give portals incorrect bulk size (1258)
4550 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4551 - fix lock inversion caused by new llite matching code (1282)
4552 - limit the number of dirty pages on a client to 10MB (1286)
4553 - timed out locks were not being corrected cancelled (1289)
4554 - fix O_DIRECT above 4GB on IA-32 (1292)
4555 * major user-visible changes
4556 - fail out/fail over policy now controlled by the upcall (993)
4558 - add OBD_PING to check server availability and failure (954)
4559 - lustre messages are now sent in sending host order (288, 340, 891)
4560 - add eadatalen to MDS getattr reply (340)
4561 - OST read replies may contain second buffer, with per-page status (593)
4563 2003-03-11 Phil Schwan <phil@clusterfs.com>
4566 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4567 - failed open()s could cause deadlock; fixed (867, 869)
4568 - stop cancelling OST locks when files are closed (481)
4569 - overlapping XID spaces caused network corruption (851, 853)
4570 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4571 - setattr_raw added, to do single-RPC, server-side setattrs
4572 - lmc/lconf syntax change for OST UUIDs
4573 - fix crashy race condition between ptlrpc_free_req and osc_close
4574 - don't use request in mdc_enqueue if we hit a timeout (889)
4575 - don't set the inode i_size for regular files from the MDS (896)
4576 - handle out of order completion AST (842)
4577 - don't LBUG if a lock request times out after receiving AST (913)
4578 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4579 - if a bad lock AST arrives, send an error instead of dropping entirely
4580 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4581 - fix leak in bulk IO when only partially completed (899, 900, 926)
4582 - fix O_DIRECT for ia64 (55)
4583 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4584 - C-z after timeout could hang a process forever; fixed (977)
4586 - client-side I/O cache (678, 924, 929, 941, 970)
4588 - READPAGE and SETATTRs which don't take server-side locks get
4591 2003-02-11 Phil Schwan <phil@clusterfs.com>
4594 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4595 - Fix fsfilt_extN_readpage() to read a full page of directory
4596 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4597 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4598 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4599 - Most intent operations were removed, in favour of a new RPC mode
4600 that does a single RPC to the server and bypasses most of the VFS
4601 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4602 - Aggressively cancel local locks on DLM servers
4603 - mds_reint_unlink sends EA to the client if it's the last nlink.
4604 client uses that EA to unlink OST objects.
4605 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4606 - recursive symlinks were fixed (439)
4607 - fixed NULL deref in DEBUG_REQ
4608 - filter_update_lastobjid no longer calls sync, which annoyed extN
4609 - fixed multi-client small-writes to a single file problem (445)
4610 - fixed mtime updates during file writes (607)
4611 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4612 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4613 - fixed Linux OST size reporting problem (444, 656)
4614 - OST now updates object mtime with writes or setattr (607, 619)
4615 - client verifies file size before zeroing page past EOF (445)
4616 - OST now writes last allocated objid to disk with allocation (108)
4617 - LOV on echo now works (409)
4619 - mds_reint_unlink sends a new buffer, with the EA included. this
4620 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4621 set if a regular file was being unlinked, and it was the last link
4622 - use PtlGet from the target for bulk writes (315)
4623 - OST now updates object mtime with writes or setattr (607, 619)
4624 - LDLM now has a grant-time callback to revalidate locked items, if
4626 - Many MDS operations were reorganized to combat race conditions
4628 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4629 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4631 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4634 - Fully reactivate OST imports after reconnection (512, others)
4635 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4636 - More graceful error handling for truncating on dead OST (515)
4637 - Don't error out unless we're actually accessing dead stripes (474)
4638 - Fix garbage sizes when stripes are missing (410)
4639 - LRU counters were broken, causing constant lock purge (433, 432)
4640 - garbage on read from stripes with failed OSTs (441)
4641 - mark OSCs as active before reconnecting during recovery (438)
4642 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4643 - lfind did not preserve OST order in output (443)
4644 - symlinks cause hung clients, incorrect data (439)
4645 - stop dereferencing request after dropping refcount (457)
4646 - don't LASSERT(spin_is_locked) on non-SMP (455)
4647 - fixes for many rename() bugs
4648 - fstat didn't correctly synchronize attributes (399)
4649 - server must handle lock cancellation during blocking AST prep (487)
4650 - bulk descriptors were free()d too soon (511)
4651 - fix paths in lconf, which would load incorrect modules (451, 507)
4652 - fix confusing lconf 'host not found' error message (386)
4653 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4654 - fix race condition in mdc_blocking_ast() for inode access (526)
4655 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4656 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4657 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4658 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4659 - fix lstripe in several ways (kernel side) (527)
4660 - fix request leak in ldlm_cli_enqueue (262)
4661 - incorrect OSC was marked inactive after OST failure
4662 - call mds_fs_cleanup before unmounting filesystem (524)
4663 - fix races between taking ns_lock and ldlm_lock_change_resource
4664 - fix races updating LOV export open file list
4665 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4666 - fix recovery NULL deref in ldlm_cli_cancel_unused
4667 - fix some DLM races by using new hash table for lock handles (419)
4668 - permit the client to specify desired inodes, at replay
4669 - duplicate requests when we queue them for replay reintegration
4670 - fix last_rcvd offset calculation
4671 - sync after each recovered transaction, so we always make progress
4672 - never, not always, ERESTART requests without transnos
4673 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4675 - skip replay if the MDS says that the client is already connected
4676 - don't check for a recovery-enabled export to match lctl's UUID
4677 - don't INC_USE_COUNT for phantom exports
4678 - don't crash when cleaning up phantom exports (567)
4679 - don't double-finish or set replay data for errored mdc_open requests
4680 - abort requests when they time out, so we don't get old replies
4681 - send/receive replies for AST messages again
4682 - if the client says that it doesn't have the lock, cancel it on the
4684 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4685 mark it as destroyed, it will all work out when decref is called
4686 - fix module use counts (22, 581)
4688 - ASTs now expect a reply (server cancels lock on error reply)
4690 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4693 - fix many simultaneous client startup (392)
4694 - fix dentry->d_it clobbering
4695 - credentials weren't being shipped for readdir/getattr operations
4696 - remove invalid assertions triggered during some concurrent MD
4698 - proper Lustre versions added (336, 389)
4699 - fix memory leak for create error case (398)
4700 - fix LOV locking bug that would get cli/srv out of sync
4701 - fix echo client over LOV (409)
4702 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4703 - fix double-O_EXCL intent crash (424)
4704 - avoid sending multiple lock CANCELs (352)
4706 - MDS can do multi-client recovery (modulo bugs in new code)
4708 - many updates, edits, cleanups
4710 2002-11-18 Phil Schwan <phil@clusterfs.com>
4713 - fix null d_it dereference (346)
4714 - fix full OST/dbench hang (333)
4715 - fix permission problem with file removal (286)
4716 - fix removal of OSCs from LOV when they fail
4717 - fix NULL deref during bulk timeout (214)
4718 - fix problems related to multiple filesystems on one MDS (241)
4719 - fixed serious subtle metadata locking bugs
4720 - free locks on clients when inodes are removed due to memory
4722 - fix inode pointer in lock data (285)
4723 - partial support for multiple MDS on a single host (241)
4724 - data locks weren't cancelled at clear_inode time (290, 311)
4725 - intent locks could lead to unbounded lock growth (205)
4726 - added a maximum lock count, an LRU list, and a flusher
4727 - fix multiple rename (365)
4728 - properly abstracted the echo client
4729 - OSC locked 1 byte too many; fixed
4730 - rewrote brw callback code:
4731 - fixed recovery bugs related to LOVs (306)
4732 - fixed too-many-pages-in-one-write crash (191)
4733 - fixed (again) crash in sync_io_timeout (214)
4734 - probably fixed callback-related race (385)
4736 - Add capability to MDS protocol
4737 - LDLM cancellations and callbacks on different portals
4739 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4742 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4743 - defer open object destruction to close time (601981, 138)
4744 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4745 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4746 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4747 - don't call obd_setattr in ll_file_release for destroyed objects
4748 * protocol change to lustre_msg: move |version| and add |flags|
4749 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4750 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4751 * added replay of create, unlink, link and rename operations during
4752 MDS failover; recovery should be much more robust now
4753 * remove failed OSCs from LOVs (only lov_create uses this so far)
4754 * the lustre-HOWTO was brought (more) up to date (582544)
4756 2002-10-23 Phil Schwan <phil@clusterfs.com>
4759 - in-use dentries weren't being reused properly (617851)
4760 - prevent multiple LDLM setup (599178)
4761 - fix LOV size calculations for truncate (617853)
4762 - fix client handling of MDS intent errors (POSIX)
4763 - fix permission bug in lovstripe.c test (624321)
4764 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4765 - truncate past end of file could corrupt data
4766 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4767 - a race in recovery could return ETIMEDOUT to apps (623947)
4768 - building outside the source directory was fixed
4769 * the lustre-HOWTO was brought (more) up to date (582544)
4770 * major progress was made on recovery functionality
4772 2002-10-10 Phil Schwan <phil@clusterfs.com>
4775 - recovery deadlock fix
4776 - rm -rf causes LBUG fix (617817)
4777 - file open by multiple tasks fix (618962)
4778 - directory permissions bugs (602707 and 620007)
4779 - journal_stop fixed with locking (611313)
4780 - O_APPEND failures resolved (618273, perhaps 614459)
4781 - lconf PATH fix (619770)
4782 - IA64 build fix (621450)
4783 - RPC buffer sizes scale with amount of memory
4785 2002-10-01 Phil Schwan <phil@clusterfs.com>
4788 - locks would be cancelled without throwing away data pages,
4789 resulting in inconsistent data (605627)
4790 - inode attributes were not always being refreshed (605627, 612449)
4791 - lconf now continues to cleanup after lctl reports an error
4792 - MDS now enforces user permissions (602707)
4793 - lprocfs cleanup fixed, but not yet enabled (614157)
4794 - fixed infinite server hang, should a client not respond to an AST
4795 - avoid going into recovery if user calls readlink() with a buffer
4796 that's too small (613941)
4797 - AST RPCs no longer require replies (614867) -- this may be changed
4798 - don't crash server if client sends an IOV that's too big (611336)
4799 - fixed lock conversion deadlock (611892)
4800 - fixed the following of symlinks (614622)
4801 * recovery: the server can remove locks from a client that dies, other
4802 clients can make progress
4803 * more extN patch fixes
4804 * compile-time configurable ptlrpc buffer allocations
4806 - collaborative read cache document
4807 - Lustre Lite Performance CDR document-in-progress
4809 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4812 - fix typo in patch-2.4.18
4814 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4817 - clear ptlrpc request each time in handle_incoming_request()
4818 - unlink of files now destroys the object on the OST
4820 2002-09-19 Peter Braam <braam@clusterfs.com>
4822 * add hard link support
4823 * change obdfile creation method
4824 * kernel patch changed
4826 2002-09-19 Peter Braam <braam@clusterfs.com>
4829 - stack overflow bug in extN fixed
4831 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4833 * documentation updates
4834 - add man pages for config tools
4835 - update tests/README to describe testing with new config tools
4836 - finish metadata API descriptions
4837 * bug fixes and cleanups
4838 - statfs workaround for 16TB limit
4839 - LOV stripe allocation improved, can stripe on subset of OSTs
4840 - LOV file size/IO offset was wrong for files > 4GB in size
4841 - object EA data was being dropped, caused files to be unreadable
4842 - memory overflow with non-LOV OST caused memory corruption
4843 - fixed regression tests to work with new config tools, obdfilter
4844 - fixed bug when directory size became larger than 1 block
4845 - fixed bug (for single client case) when PWD was deleted
4846 - invalidate local directory pages when doing intent-based ops
4847 - avoid LDLM oops when lock callback contained bad data
4849 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4851 * documentation updates
4852 * bug fixes and cleanups
4853 - configuration tools
4856 - 64-bit compile warnings
4857 - 64-bit internal statfs data
4859 * test_brw on persistent OST devices
4861 * lprocfs (disabled)
4863 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4865 * documentation updates
4866 * bug fixes and cleanups
4867 * configuration tools
4869 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4872 * many small fixes to 0_5_4
4873 * io/network handling
4874 * thinkos in MDS operations
4876 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4879 * crucial basic fixes to 0.5.3
4880 * IOR, Iozone work over Elan
4883 2002-08-07 Phil Schwan <phil@clusterfs.com>
4884 * version 0_5_3, our first alpha
4885 * we use the new Portals iovs
4886 * documentation updates
4887 * bug fixes and cleanups
4888 * small changes in the DLM wire protocol
4890 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4891 * version 0_5_1 with some initial stability,
4892 * locking on MD and file I/O.
4893 * documentation updates
4894 * several bug fixes since 0.5.0
4895 * small changes in wire protocol
4897 2002-07-18 Phil Schwan <phil@clusterfs.com>
4899 * delivered as Lustre Light Alpha
4900 * fixed a crash after handling invalid MDS requests
4901 * fixed directory pages for architectures with non-4k pages sizes
4903 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4904 * release version v0_4_4
4905 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4907 * Fixes a number of interruption problems with OST operations.
4908 * Update documentation for portals header changes
4909 * Move all wire protocol structs/defines to lustre_idl.h
4910 * Fixes symlink length bug.
4911 * Add tcpdump to repository.
4913 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4914 * release version v0_4_3
4915 * Fixes statfs for inodes on extN.
4916 * Fixes bug in runtests which would delete /etc/hosts.
4917 * Use 64-bit object IDs wherever possible (not into VFS though)
4918 Remove ost_get_info, which is unused by lustre, and out of date.
4920 2002-07-03 Peter Braam <braam@clusterfs.com>
4921 * release version v0_4_2 Fixes a lookup error (type not passed)
4922 * move forward to head of Portals
4923 * move forward to latest Lustre kernel
4925 2002-06-25 Peter Braam <braam@clusterfs.com>
4926 * release version v0_4_1. Hopefully stable on single node use.