Re: Doubts about mmap(2)

[Oskari Pirhonen <xxc3ncoredxx@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 2049 bytes --]

On Thu, May 18, 2023 at 11:03:14 +0200, Alex Colomar wrote:
> On 5/18/23 10:53, Alex Colomar wrote:
> > Hi!
> > 
> > I believe mmap(2) is confusing (see the emboldened parts):
> > 
> > DESCRIPTION
> >         mmap()  creates  a  new mapping in the virtual address space of
> >         the calling process.  **The starting address for the new  mapping
> >         is specified in addr.**  The length argument specifies the length
> >         of the mapping (which must be greater than 0).
> > 
> >         If addr is NULL, then the kernel chooses the (page‐aligned) ad‐
> >         dress at which to create the mapping; this is the most portable
> >         method  of  creating  a new mapping.  **If addr is not NULL, then
> >         the kernel takes it as a hint about where to place the mapping;
> >         on Linux, the kernel will pick a nearby page boundary** (but  al‐
> >         ways    above    or   equal   to   the   value   specified   by
> >         /proc/sys/vm/mmap_min_addr) and attempt to create  the  mapping
> >         there.   If  another  mapping  already exists there, the kernel
> >         picks a new address that may or may not  depend  on  the  hint.
> >         The address of the new mapping is returned as the result of the
> >         call.
> > 
> > 
> > The first emboldened sentence contradicts the second.  It's not the
> > starting address, but just a hint to find a nearby (>=) starting
> 
> I was a bit wrong there.  It's >= mmap_min_addr, not >= hint.  But the
> suggestion below holds.
> 
> > address.  How about saying this?:
> > 
> >      A hint for the starting address for the new mapping is specified in
> >      addr.
> > 

I think describing the default behavior first is a good idea. If a user
needs an exact address they can read further for MAP_FIXED/_NOREPLACE
(and the potential footguns involved).

- Oskari

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]