Date: Wed, 13 Nov 2024 20:10:53 +0000 From: Jessica Clarke <jrtc27@freebsd.org> To: John Baldwin <jhb@FreeBSD.org> Cc: Andrew Turner <andrew@FreeBSD.org>, "src-committers@freebsd.org" <src-committers@FreeBSD.org>, "dev-commits-src-all@freebsd.org" <dev-commits-src-all@FreeBSD.org>, "dev-commits-src-main@freebsd.org" <dev-commits-src-main@FreeBSD.org> Subject: Re: git: e85eaa930862 - main - Have rtld query the page size from the kernel Message-ID: <768D45F9-2F02-4BA1-BFB7-51685486CFCC@freebsd.org> In-Reply-To: <92a05dfe-683c-43d9-bd29-3110e89be275@FreeBSD.org> References: <202204071438.237Ecn2A012737@gitrepo.freebsd.org> <92a05dfe-683c-43d9-bd29-3110e89be275@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On 13 Nov 2024, at 19:44, John Baldwin <jhb@FreeBSD.org> wrote: >=20 > On 4/7/22 07:38, Andrew Turner wrote: >> The branch main has been updated by andrew: >> URL: = https://cgit.FreeBSD.org/src/commit/?id=3De85eaa930862d5b4dc917bc31e8d7254= a693635d >> commit e85eaa930862d5b4dc917bc31e8d7254a693635d >> Author: Andrew Turner <andrew@FreeBSD.org> >> AuthorDate: 2022-04-04 15:05:40 +0000 >> Commit: Andrew Turner <andrew@FreeBSD.org> >> CommitDate: 2022-04-07 14:37:37 +0000 >> Have rtld query the page size from the kernel >> To allow for a dynamic page size on arm64 have the runtime = linker >> query the kernel for the currentl page size. >> Reviewed by: kib >> Sponsored by: The FreeBSD Foundation >> Differential Revision: https://reviews.freebsd.org/D34765 >=20 > This broke relro handling for rtld. The reason is that = init_pagesizes() is > called after parsing the program headers for rltd in init_rtld(). As = a result, > page_size is 0 when rtld_round_page() is called so the relro_size is = 0. The > RTLD_INIT_EARLY_PAGESIZES case was for ia64, and in the early case = it's probably > not safe to call sysctl? If it is safe to call sysctl, we could just = always > init pagesizes early? It looks like there are a few things going on: 1. relocate_object calls obj_enforce_relro if !obj->mainprog, so will try to enforce RELRO for RTLD itself whilst page_size is 0 2. init_rtld later calls obj_enforce_relro for obj_rtld, after page_size has been initialised 3. init_rtld is careful to avoid using global variables until it=E2=80=99s= called relocate_objects for RTLD itself, but by hiding accesses to page_size away in rtld_*_page that=E2=80=99s no longer true (definitely = not true in the case of text relocations, for example, though whether it also occurs for other cases we care more about I don=E2=80=99t know) So I think there are a couple of things to fix: 1. Stop accessing page_size prior to relocate_objects returning for RTLD itself 2. Stop enforcing RELRO twice for RTLD (e.g. add && obj !=3D rtldobj to relocate_object=E2=80=99s case) At least, that=E2=80=99s what I=E2=80=99ve inferred from reading the = code. Though, to be honest, things might be rather nicer if we just made .rtld_start responsible for relocating RTLD itself prior to calling init_rtld, that=E2=80=99s what we have to do for CHERI, as do arm, = powerpc and powerpc64, and it means you can use globals from the start in init_rtld. Jess
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?768D45F9-2F02-4BA1-BFB7-51685486CFCC>