Date: Wed, 11 Jan 2006 06:40:09 GMT From: Bruce Evans <bde@zeta.org.au> To: freebsd-bugs@FreeBSD.org Subject: Re: bin/91622: /bin/cp does not update atime of the source file Message-ID: <200601110640.k0B6e9Iu060779@freefall.freebsd.org>
next in thread | raw e-mail | index | archive | help
The following reply was made to PR bin/91622; it has been noted by GNATS.
From: Bruce Evans <bde@zeta.org.au>
To: joe <freebsd-security@FreeBSD.org>
Cc: freebsd-gnats-submit@FreeBSD.org, freebsd-bugs@FreeBSD.org,
dds@FreeBSD.org
Subject: Re: bin/91622: /bin/cp does not update atime of the source file
Date: Wed, 11 Jan 2006 17:34:58 +1100 (EST)
On Wed, 11 Jan 2006, joe wrote:
>> Environment:
> RELENG_4 20051028
>> Description:
> [cp fails to update atime]
>
> this occurs regardless of whether or not the user has
> rwx permissions on the source file. does not apply of
> course if the source filesystem is mounted read-only.
> they are read-write here.
>
> since the source file is indeed being read, the atime should be updated.
This is because cp(1) doesn't actually read(2) the file. It mmap(2)'s the
file and reads memory.
POSIX doesn't seem to require that cp sets atimes or otherwise acts as
if it reads the file.
Similar for cmp(1) and other utilities that use mmap() to read files
(install(1) is another).
The problem is mostly fixed in -current by "setting" (not actually
setting, but marking for update) atimes in mmap(). I think to be
strictly correct, setting of atimes should be nearer to munmap(2) than
mmap(). No read has occurred from userland's viewpoint at mmap() time.
However, since the OS can't tell if applications read mmapped memory
unless their is a pagefault, the atime would have to be set on every
pagefault to be as correct as possible, and I think mmap() prefaults
some pages so it would have to set the atime whether or not the
application reads the pages. So I think setting atimes in mmap() is
required in practice. This leaves the problem that later pagefaults
never set the atime, so the atime set by mmap() becomes out of date.
I think the atime should be set by mmap() of there is at least one
later pagefault. Setting it for every pagefault may be too inefficient.
Bruce
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200601110640.k0B6e9Iu060779>