vmapbuf: don't smuggle user address via b_data

[brooks]

Summary

Instead, add a uaddr argument to vmapbuf. Since this argument is
always a pointer use a type of void * and cast to vm_offset_t in
vmapbuf. (In CheriBSD we've altered vm_fault_quick_hold_pages to
take a pointer and check its bounds.)

In no other situtation does b_data contain a user pointer and vmapbuf
replaces b_data with the actual mapping.

[brooks]

CI run: https://cirrus-ci.com/task/5059292485124096

[jhb]

I know I didn't say this earlier in CheriBSD, but looking at the patch again, I think it might be cleaner to pass in the size explicitly as well and have vmapbuf set bp->bufsize rather than than smuggling that in as well. Together, b_data, b_offset, and b_bufsize describe the memory buffer, and passing in the length would let vmapbuf be responsible for setting all of those.

sys/ufs/ffs/ffs_rawread.c

250 ↗

(On Diff #78238)

Unrelated bug: vmapbuf() overwrites bp->b_offset. This should only be setting bp->b_iooffset.

[brooks]

• Pass size rather than smuggling via b_bufsize.

[brooks]

https://cirrus-ci.com/task/5974398819631104

[jhb]

Thanks.

sys/kern/vfs_bio.c

4915 ↗

(On Diff #78430)

This check is now dead code that can be removed I think since size_t is unsigned?

[brooks]

• GC a dead check.

[imp]

This is good... only caveat is that MFC code near this stuff might be a pain... A minor inconvenience at best, so this is a LGTM!