[Bug-tar] extracting multivol archives: no chance to correct sequence of volumes

[Joerg Weilbier]

Hi,
while extracting multivol archives in a violation of right sequence of volumes 
will cause first a warning (... is  not continued on this volume). That's ok, 
but if you now take the right volume, you can't reach the right way anymore.

I used tar 1.15.91
Older versions (e.g. 1.15.1) will not fail!

I used this commands (Gnu/Linux (Debian 3.1, SuSE9.0), bash):
   #TAR="/home/joerg/tar/tar-1.13.94/src/tar"
   #TAR="/home/joerg/tar/tar-1.14.91/src/tar"
   #TAR="/home/joerg/tar/tar1.15.1/src/tar"
   TAR="/home/joerg/Linux/tar/Fehleranalyse/tar-1.15.91.erw/src/tar"
   
   rm z*.tar 2> /dev/null
   rm $File1 $File2 2> /dev/null
   
   #create files:
   dd bs=1024 count=250 if=/dev/urandom of=File1
   dd bs=1024 count=250 if=/dev/urandom of=File2
   
   #create archive:
   $TAR --create -v --multi-volume --tape-length=100 --file=z1.tar \
   --file=z2.tar --file=z3.tar --file=z4.tar --file=z5.tar --file=z6.tar \
   --file=z7.tar File1 File2
   
   mkdir -p bak/
   mv File1 File2 bak/
   
   #extract with false sequence, but then corrected:
   $TAR --extract -v --multi-volume --file=z1.tar --file=z2.tar \
   --file=z4.tar --file=z3.tar --file=z4.tar --file=z5.tar --file=z6.tar 
   
   #compare extracted files and backup-copies
   cmp File1 bak/File1
   cmp File2 bak/File2


The problem is in buffer.c, try_new_volume():

In case of wrong volume (while extracting) in "switch 
(header->header.typeflag) ... case GNUTPYPE_MULTIVOL" the global variable 
current_block will increment with a blocksize (inside of read_header0
(&dummy)).

In immediate following next call of try_new_volume() therefore the 
current_block points to data _behind_ the (multivol-) header of next volume.

I suggest this patch (inserting of initialization of current_block in every 
run of try_new_volume()):

--- buffer.c.orig       2006-08-24 11:59:17.000000000 +0200
+++ buffer.c    2006-10-01 15:37:14.000000000 +0200
@@ -1143,6 +1143,11 @@
       break;
     }

+/*if try_new_volume() was ended with false, current_block is still 
record_start + BLOCKSIZE (cause of read_header0()-call)
+  We first must put it back
+*/
+  current_block = record_start;
+
   while ((status = rmtread (archive, record_start->buffer, record_size))
         == SAFE_READ_ERROR)
     archive_read_error ();



-- 
Joerg