On Wed, Mar 17, 1999 at 10:13:45PM -0500, Andrew Pimlott wrote:
> open(... | O_SYNC);
Yep - I tried it. It works well. But: Does this really sync the data? (I did
not want to push reset because of fsck ;))
That raises another point: Do we want to have the filesystem mounted readonly?
Then we will have to check it after reboot.
How about the following plan (in context of installation):
1. we need to detect the available disk drives
- ask the user
- if he does not know:
probe the available modules for controllers (scsi, ide, ...) using the
checkpointing I describe at the end writing to /dev/fd0
[ARG! What about CD-ROM install?]
2. we have detected the disks - let the user partition and format the drives
he needs.
3. do a pci scan
[Could we do this before 1? We could learn about mass storage from there]
4. do an isapnp scan if the user likes the idea ;)
[We should checkpoint this as well - the system can crash using pnpdump]
5. check for the rest of the hardware using checkpointing on the disk mounted
as root of the installation (loading modules as we go)
6. configure the isapnp devices
7. load the drivers for the pci and isapnp devices
Checkpointing the Autodetection
-------------------------------
We can not use the normal filesystem operations because we will have to do a
fsck after reboot. The only alternative I can think of:
1. create a file to hold the checkpointing information and fill it with nulls
(maybe 4k for the detection of mass storage controllers, 100k for the full
scan)
2. remount the partition readonly
3. gather the list of blocks used by the file
4. start the detection writing the results directly to this area, syncing
after each test
5. if the test completed make the partition writeable again (for the
installation)
Comments?
cu
Torsten
Attachment:
pgpilNdQOTMjA.pgp
Description: PGP signature