Two processes doing mmap

[dimakuv]

How do we plan to tackle the case where two processes:

1. Access the same file
2. At least one of them uses mmap()
3. At least one of them has write permission

Do we consider this case unlikely and thus fail on it? Like, when the second file tries to do an mmap() on the already-opened file (by another process), we return an error?

[pwmarcz]

This is assuming we emulate mmap by loading the whole file to memory, and saving it on close? (if the map is writable and shared)

I don't have an idea for anything other than (a) pretending to support this, and not updating the map when the file changes, (b) detecting such cases (by way of some common state stored on server) and disallowing them. If we consider this case unlikely, maybe there's no need to even detect it?

[mkow]

I'd prefer to disallow such shared mappings, but: gramineproject/graphene#2147 (review).
But maybe it would work if we allowed such mappings, but enforced that they are not actually shared by anyone.