Re: [Libguestfs] [PATCH nbdkit] filters: cache, cow: Handle bitmap overflow on 32 bit architectures.
On 01/26/2018 07:53 AM, Eric Blake wrote:
On 01/26/2018 07:17 AM, Richard W.M. Jones wrote:
> When compiling on armv7 it becomes clear from the compiler warnings
> that the current code is wrong.
>
> The bitmap has to be allocated in virtual memory, so use size_t to
> describe the length of the bitmap. When changing the length of the
> bitmap, compute the new size as an unsigned 64 bit int, and then check
> whether or not it is too large to fit into size_t before casting it.
Indeed, size_t is a sane limit (the only way to get a larger map would
be using an mmap'd file on disk, so that we can use 64-bit off_t instead
of 32-bit size_t).
and using a second layer of caching to swap in windows of that file as
needed, as the entire file won't fit in memory at once. That's a lot of
bookkeeping overhead, for something that is less and less useful as more
people use 64-bit machines.
--
Eric Blake, Principal Software Engineer
Red Hat, Inc. +1-919-301-3266
Virtualization: qemu.org | libvirt.org
Attachments:
- signature.asc (application/pgp-signature — 619 bytes)