William Schumann wrote: > RE: http://defect.opensolaris.org/bz/show_bug.cgi?id=6854 > Automated Installer can provide FAT formatting > > We want to offer automatic formatting of new DOS partitions created through > AI. > > The current solution relies upon mkfs_pcfs(1M), which requires either a > logical drive letter ('c'-based) or number (zero-based). > > AI has all information for the new fdisk format table, yet none of that > information seems to lead to an absolute correlation of the > drive letter for a given DOS partition, the consequence of which can be that > the wrong partition is formatted. > > This becomes evident when different DOS partition types are mixed. For > example, I have been testing with partitions (in this order): > 1 FAT32 <2G, /dev/rdsk/cxtxdxp1 > 1 DOS16. /dev/rdsk/cxtxdxp2 > 1 FAT32 extended, /dev/rdsk/cxtxdxp3 > 1 Solaris, /dev/rdsk/cxtxdxp4 > > From the man page, it might be assumed that the logical drive letter > assignments would be as follows: > C: FAT32 <2G > D: DOS16 > E: FAT32 extended > > So that to format the third partition, the following command might be issued: > mkfs -F pcfs -o fat=32 /dev/rdsk/cxtxdxp0:e > --where 'e' is selected because it is the 3rd DOS partition in the table (or > alternatively, '2') > > But there appears to be nothing that guarantees that this drive letter > matches the second DOS partition. I have seen several > examples where the ordering was not followed. Furthermore, I looked at the > sector number ordering and found no reliable correlation > between logical drive letter and the sector ordering, either. > > The actual logical drive letter assignment is: > C: FAT32 <2G > E: DOS16 > D: FAT32 extended > > Assignment of the logical drive letters appears to be internally consistent > and reliable, but that does not help with finding the > logical drive letter based on the fdisk format table information. > > It appears that, since the logical drive assignment does not occur as > documented, this is a bug in the code that performs the > assignment. > > Possible workarounds: > - AI looks at the raw partition data and determines the type, whether it is > formatted, and determines the logical drive letter from > the raw partition data before using mkfs_pcfs to format > - AI uses the same library calls as mount(1m) and mkfs(1m) to determine the > logical drive letter > > Does anyone have any ideas on the best approach to this for the Automated > Installer?
I would suggest that the lack of determinism in the underlying device mapping is the real problem, and this functionality has to be deferred until that is fixed. This is in no way a blocker for the upcoming release, and I think its risk level is too high with these issues. Dave
