[RESEND PATCH] xfsrestore: fix fs uuid order check for incremental restores
Rich Johnston
rjohnston at sgi.com
Wed Sep 2 13:49:47 CDT 2015
On 09/02/2015 08:21 AM, Brian Foster wrote:
> On Tue, Sep 01, 2015 at 03:38:29PM -0500, Rich Johnston wrote:
>> Restoring an incremental level 1 dump will fail with the following error
>> if the fs uuid of the most recent level 0 dump in the inventory does not
>> match level 1 dump we are restoring.
>>
>> xfsrestore: ERROR: selected dump not based on previously applied dump
>>
>> This can happen when you have multiple filesystems and you are restoring
>> a level 1 or greater dump of filesystem FS1 but the most recent level 0
>> dump in the inventory was filesystem FS2
>>
>> The fix is to ensure the fs uuid of the inventory entry and the dump to
>> be restored match.
>>
>> Signed-off-by: Rich Johnston <rjohnston at sgi.com>
>> ---
>
> I'm not really familiar with xfsdump, but a few comments below...
>
>> dump/content.c | 8 ++-
>> inventory/inv_api.c | 108 ++++++++++++++++++++++++++++++--------------------
>> inventory/inv_mgr.c | 32 ++++++++++----
>> inventory/inv_priv.h | 7 +--
>> inventory/inventory.h | 5 ++
>> restore/content.c | 17 +++++--
>> 6 files changed, 113 insertions(+), 64 deletions(-)
>>
> ...
>> Index: b/inventory/inv_mgr.c
>> ===================================================================
>> --- a/inventory/inv_mgr.c
>> +++ b/inventory/inv_mgr.c
>> @@ -134,6 +134,7 @@ get_sesstoken( inv_idbtoken_t tok )
>> /*---------------------------------------------------------------------------*/
>> bool_t
>> invmgr_query_all_sessions (
>> + uuid_t *fsidp,
>> void *inarg,
>> void **outarg,
>> search_callback_t func)
>> @@ -169,7 +170,7 @@ invmgr_query_all_sessions (
>> mlog( MLOG_NORMAL | MLOG_INV, _(
>> "INV: Cant get inv-name for uuid\n")
>> );
>> - return BOOL_FALSE;
>> + continue;
>> }
>> strcat( fname, INV_INVINDEX_PREFIX );
>> invfd = open( fname, INV_OFLAG(forwhat) );
>> @@ -178,9 +179,9 @@ invmgr_query_all_sessions (
>> "INV: Open failed on %s\n"),
>> fname
>> );
>> - return BOOL_FALSE;
>> + continue;
>> }
>
> Now that the above two cases don't return false, we can end up returning
> true if these error cases persist throughout the loop.
Yes I missed that case.
> Perhaps we should
> define 'bool ret = false,' return that variable everywhere and only set
> it true when appropriate.
Good suggestion.
>
>> - result = search_invt( invfd, inarg, &objectfound, func );
>> + result = search_invt(fsidp, invfd, inarg, &objectfound, func);
>> close(invfd);
>>
>> /* if error return BOOL_FALSE */
> ...
>> @@ -272,19 +274,31 @@ search_invt(
>> }
>> free ( scnt );
>>
>> - while ( nsess ) {
>> + for (j = nsess - 1; j >= 0; j--) {
>> + invt_session_t ses;
>> +
>> /* fd is kept locked until we return from the
>> callback routine */
>>
>> /* Check to see if this session has been pruned
>> * by xfsinvutil before checking it.
>> */
>> - if ( harr[nsess - 1].sh_pruned ) {
>> - --nsess;
>> + if (harr[j].sh_pruned) {
>> continue;
>> }
>> - found = (* do_chkcriteria ) ( fd, &harr[ --nsess ],
>> - arg, buf );
>> +
>> + /* if we need to check the fs uuid's and they don't
>> + * match or we fail to get the session record,
>> + * then keep looking
>> + */
>> + if (fsidp &&
>> + (GET_REC_NOLOCK(fd, &ses, sizeof(invt_session_t),
>> + harr[j].sh_sess_off) ==
>> + sizeof(invt_session_t)) &&
>> + uuid_compare(ses.s_fsid, *fsidp))
>> + continue ;
>> +
>
> This seems like kind of a loaded check. GET_REC_NOLOCK() looks like it
> returns -1 if the read doesn't return the exact size of the buffer, so
> we probably don't need to check that here. It also seems like we should
> return an error if an error occurs rather than just continue on. How
> about something like this?
Sounds reasonable, would be more readable.
>
> ...
> if (fsidp) {
> ret = GET_REC_NOLOCK(fd, &ses,
> sizeof(invt_session_t),
> harr[j].sh_sess_off);
> if (ret < 0) {
> /* XXX: clean up here */
> return ret;
> }
> ret = uuid_compare(ses.s_fsid, *fsidp);
> if (ret)
> continue;
> }
>
>> + found = (* do_chkcriteria ) (fd, &harr[j], arg, buf);
>> if (! found ) continue;
>>
>> /* we found what we need; just return */
> ...
>> Index: b/restore/content.c
>> ===================================================================
>> --- a/restore/content.c
>> +++ b/restore/content.c
>> @@ -2179,8 +2179,9 @@ content_stream_restore( ix_t thrdix )
>> if ( ! drivep->d_isnamedpipepr
>> &&
>> ! drivep->d_isunnamedpipepr ) {
>> - ok = inv_get_session_byuuid( &grhdrp->gh_dumpid,
>> - &sessp );
>> + ok = inv_get_session_byuuid((uuid_t *)0,
>> + &grhdrp->gh_dumpid,
>> + &sessp);
>> if ( ok && sessp ) {
>> mlog( MLOG_VERBOSE, _(
>> "using online session inventory\n") );
>> @@ -3736,9 +3737,11 @@ Inv_validate_cmdline( void )
>> ok = BOOL_FALSE;
>> sessp = 0;
>> if ( tranp->t_reqdumpidvalpr ) {
>> - ok = inv_get_session_byuuid( &tranp->t_reqdumpid, &sessp );
>> + ok = inv_get_session_byuuid((uuid_t *)0, &tranp->t_reqdumpid,
>> + &sessp );
>> } else if ( tranp->t_reqdumplabvalpr ) {
>> - ok = inv_get_session_bylabel( tranp->t_reqdumplab, &sessp );
>> + ok = inv_get_session_bylabel((uuid_t *)0, tranp->t_reqdumplab,
>> + &sessp );
>
> Can we use NULL instead of 0 in these cases? Not sure the cast is really
> necessary either..?.
I will check that out, if NULL works I will make that change.
Thanks for your time
--Rich
>
> Brian
>
>> }
>> rok = BOOL_FALSE;
>> if ( ok && sessp ) {
>> @@ -6812,11 +6815,13 @@ askinvforbaseof( uuid_t baseid, inv_sess
>> /* get the base session
>> */
>> if ( resumedpr ) {
>> - ok = inv_lastsession_level_equalto( invtok,
>> + ok = inv_lastsession_level_equalto( &sessp->s_fsid,
>> + invtok,
>> ( u_char_t )level,
>> &basesessp );
>> } else {
>> - ok = inv_lastsession_level_lessthan( invtok,
>> + ok = inv_lastsession_level_lessthan( &sessp->s_fsid,
>> + invtok,
>> ( u_char_t )level,
>> &basesessp );
>> }
>>
>> _______________________________________________
>> xfs mailing list
>> xfs at oss.sgi.com
>> http://oss.sgi.com/mailman/listinfo/xfs
More information about the xfs
mailing list