This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH v2 2/2] elf: Correct absolute (SHN_ABS) symbol run-time calculation [BZ #19818]
- From: Adhemerval Zanella <adhemerval dot zanella at linaro dot org>
- To: libc-alpha at sourceware dot org
- Cc: Andreas Schwab <schwab at suse dot de>, Florian Weimer <fweimer at redhat dot com>
- Date: Wed, 7 Mar 2018 17:25:20 -0300
- Subject: Re: [PATCH v2 2/2] elf: Correct absolute (SHN_ABS) symbol run-time calculation [BZ #19818]
- Authentication-results: sourceware.org; auth=none
- References: <alpine.DEB.2.00.1802201804140.3553@tp.orcam.me.uk> <alpine.DEB.2.00.1802201815060.3553@tp.orcam.me.uk>
On 20/02/2018 15:38, Maciej W. Rozycki wrote:
> Do not relocate absolute symbols by the base address. Such symbols have
> SHN_ABS as the section index and their value is not supposed to be
> affected by relocation as per the ELF gABI[1]:
>
> "SHN_ABS
> The symbol has an absolute value that will not change because of
> relocation."
>
> The reason for our non-conformance here seems to be an old SysV linker
> bug causing symbols like _DYNAMIC to be incorrectly emitted as absolute
> symbols[2]. However in a previous discussion it was pointed that this
> is seriously flawed by preventing the lone purpose of the existence of
> absolute symbols from being used[3]:
>
> "On the contrary, the only interpretation that makes sense to me is that
> it will not change because of relocation at link time or at load time.
> Absolute symbols, from the days of the earliest linking loaders, have
> been used to represent addresses that are outside the address space of
> the module (e.g., memory-mapped addresses or kernel gateway pages).
> They've even been used to represent true symbolic constants (e.g.,
> system entry point numbers, sizes, version numbers). There's no other
> way to represent a true absolute symbol, while the meaning you seek is
> easily represented by giving the symbol a non-negative st_shndx value."
>
> and we ought to stop supporting our current broken interpretation.
>
> Update processing for dladdr(3) and dladdr1(3) so that SHN_ABS symbols
> are ignored, because under the corrected interpretation they do not
> represent addresses within a mapped file and therefore are not supposed
> to be considered.
>
> References:
>
> [1] "System V Application Binary Interface - DRAFT - 19 October 2010",
> The SCO Group, Section "Symbol Table",
> <http://www.sco.com/developers/gabi/2012-12-31/ch4.symtab.html>
>
> [2] Alan Modra, "Absolute symbols"
> <https://sourceware.org/ml/binutils/2012-05/msg00019.html>
>
> [3] Cary Coutant, "Re: Absolute symbols"
> <https://sourceware.org/ml/binutils/2012-05/msg00020.html>
>
> [BZ #19818]
> * sysdeps/generic/ldsodefs.h (SYMBOL_ADDRESS): Handle SHN_ABS
> symbols.
> * elf/dl-addr.c (determine_info): Ignore SHN_ABS symbols.
> * elf/tst-absolute-sym.c: New file.
> * elf/tst-absolute-sym-lib.c: New file.
> * elf/tst-absolute-sym-lib.lds: New file.
> * elf/Makefile (tests): Add `tst-absolute-sym'.
> (modules-names): Add `tst-absolute-sym-lib'.
> (LDLIBS-tst-absolute-sym-lib.so): New variable.
> ($(objpfx)tst-absolute-sym-lib.so): New dependency.
> ($(objpfx)tst-absolute-sym): New dependency.
The patch and rationale looks reasonable, Andrea and Florian do you still have
objections or doubts about the patch?