From: Quentin Casasnovas <quentin.casasnovas@oracle.com>
To: Quentin Casasnovas <quentin.casasnovas@oracle.com>,
Rusty Russell <rusty@rustcorp.com.au>,
Michal Marek <mmarek@suse.cz>
Cc: lkml <linux-kernel@vger.kernel.org>,
Oleg Nesterov <oleg@redhat.com>, Borislav Petkov <bp@alien8.de>,
Linus Torvalds <torvalds@linux-foundation.org>
Subject: Re: [PATCH 6/7] scripts: add check_extable.sh script.
Date: Wed, 18 Mar 2015 10:09:25 +0100 [thread overview]
Message-ID: <20150318090925.GK19131@chrystal.uk.oracle.com> (raw)
In-Reply-To: <1426596002-26128-7-git-send-email-quentin.casasnovas@oracle.com>
Adding Rusty and Michal to CC.
On Tue, Mar 17, 2015 at 01:40:01PM +0100, Quentin Casasnovas wrote:
> This shell script can be used to sanity check the __ex_table section on an
> object file, making sure the relocations in there are pointing to valid
> executable sections. If it finds some suspicious relocations, it'll use
> addr2line to try and dump where this is coming from.
>
> This works best with CONFIG_DEBUG_INFO.
>
> Signed-off-by: Quentin Casasnovas <quentin.casasnovas@oracle.com>
> ---
> scripts/check_extable.sh | 146 +++++++++++++++++++++++++++++++++++++++++++++++
> 1 file changed, 146 insertions(+)
> create mode 100755 scripts/check_extable.sh
>
> diff --git a/scripts/check_extable.sh b/scripts/check_extable.sh
> new file mode 100755
> index 0000000..0fb6b1c
> --- /dev/null
> +++ b/scripts/check_extable.sh
> @@ -0,0 +1,146 @@
> +#! /bin/bash
> +# (c) 2015, Quentin Casasnovas <quentin.casasnovas@oracle.com>
> +
> +obj=$1
> +
> +file ${obj} | grep -q ELF || (echo "${obj} is not and ELF file." 1>&2 ; exit 0)
> +
> +# Bail out early if there isn't an __ex_table section in this object file.
> +objdump -hj __ex_table ${obj} 2> /dev/null > /dev/null
> +[ $? -ne 0 ] && exit 0
> +
> +white_list=.text,.fixup
> +
> +suspicious_relocs=$(objdump -rj __ex_table ${obj} | tail -n +6 |
> + grep -v $(eval echo -e{${white_list}}) | awk '{print $3}')
> +
> +# No suspicious relocs in __ex_table, jobs a good'un
> +[ -z "${suspicious_relocs}" ] && exit 0
> +
> +
> +# After this point, something is seriously wrong since we just found out we
> +# have some relocations in __ex_table which point to sections which aren't
> +# white listed. If you're adding a new section in the Linux kernel, and
> +# you're expecting this section to contain code which can fault (i.e. the
> +# __ex_table relocation to your new section is expected), simply add your
> +# new section to the white_list variable above. If not, you're probably
> +# doing something wrong and the rest of this code is just trying to print
> +# you more information about it.
> +
> +function find_section_offset_from_symbol()
> +{
> + eval $(objdump -t ${obj} | grep ${1} | sed 's/\([0-9a-f]\+\) .\{7\} \([^ \t]\+\).*/section="\2"; section_offset="0x\1" /')
> +
> + # addr2line takes addresses in hexadecimal...
> + section_offset=$(printf "0x%016x" $(( ${section_offset} + $2 )) )
> +}
> +
> +function find_symbol_and_offset_from_reloc()
> +{
> + # Extract symbol and offset from the objdump output
> + eval $(echo $reloc | sed 's/\([^+]\+\)+\?\(0x[0-9a-f]\+\)\?/symbol="\1"; symbol_offset="\2"/')
> +
> + # When the relocation points to the begining of a symbol or section, it
> + # won't print the offset since it is zero.
> + if [ -z "${symbol_offset}" ]; then
> + symbol_offset=0x0
> + fi
> +}
> +
> +function find_alt_replacement_target()
> +{
> + # The target of the .altinstr_replacement is the relocation just before
> + # the .altinstr_replacement one.
> + eval $(objdump -rj .altinstructions ${obj} | grep -B1 "${section}+${section_offset}" | head -n1 | awk '{print $3}' |
> + sed 's/\([^+]\+\)+\(0x[0-9a-f]\+\)/alt_target_section="\1"; alt_target_offset="\2"/')
> +}
> +
> +function handle_alt_replacement_reloc()
> +{
> + # This will define alt_target_section and alt_target_section_offset
> + find_alt_replacement_target ${section} ${section_offset}
> +
> + echo "Error: found a reference to .altinstr_replacement in __ex_table:"
> + addr2line -fip -j ${alt_target_section} -e ${obj} ${alt_target_offset} | awk '{print "\t" $0}'
> +
> + error=true
> +}
> +
> +function is_executable_section()
> +{
> + objdump -hwj ${section} ${obj} | grep -q CODE
> + return $?
> +}
> +
> +function handle_suspicious_generic_reloc()
> +{
> + if is_executable_section ${section}; then
> + # We've got a relocation to a non white listed _executable_
> + # section, print a warning so the developper adds the section to
> + # the white list or fix his code. We try to pretty-print the file
> + # and line number where that relocation was added.
> + echo "Warning: found a reference to section \"${section}\" in __ex_table:"
> + addr2line -fip -j ${section} -e ${obj} ${section_offset} | awk '{print "\t" $0}'
> + else
> + # Something is definitively wrong here since we've got a relocation
> + # to a non-executable section, there's no way this would ever be
> + # running in the kernel.
> + echo "Error: found a reference to non-executable section \"${section}\" in __ex_table at offset ${section_offset}"
> + error=true
> + fi
> +}
> +
> +function handle_suspicious_reloc()
> +{
> + case "${section}" in
> + ".altinstr_replacement")
> + handle_alt_replacement_reloc ${section} ${section_offset}
> + ;;
> + *)
> + handle_suspicious_generic_reloc ${section} ${section_offset}
> + ;;
> + esac
> +}
> +
> +function diagnose()
> +{
> +
> + for reloc in ${suspicious_relocs}; do
> + # Let's find out where the target of the relocation in __ex_table
> + # is, this will define ${symbol} and ${symbol_offset}
> + find_symbol_and_offset_from_reloc ${reloc}
> +
> + # When there's a global symbol at the place of the relocation,
> + # objdump will use it instead of giving us a section+offset, so
> + # let's find out which section is this symbol in and the total
> + # offset withing that section.
> + find_section_offset_from_symbol ${symbol} ${symbol_offset}
> +
> + # In this case objdump was presenting us with a reloc to a symbol
> + # rather than a section. Now that we've got the actual section,
> + # we can skip it if it's in the white_list.
> + if [ -z "$( echo $section | grep -v $(eval echo -e{${white_list}}))" ]; then
> + continue;
> + fi
> +
> + # Will either print a warning if the relocation happens to be in a
> + # section we do not know but has executable bit set, or error out.
> + handle_suspicious_reloc
> + done
> +}
> +
> +function check_debug_info() {
> + objdump -hj .debug_info ${obj} 2> /dev/null > /dev/null ||
> + echo -e "${obj} does not contain debug information, the addr2line output will be limited.\n" \
> + "Recompile ${obj} with CONFIG_DEBUG_INFO to get a more useful output."
> +}
> +
> +check_debug_info
> +
> +diagnose
> +
> +if [ "${error}" ]; then
> + exit 1
> +fi
> +
> +exit 0
> --
> 2.0.5
>
next prev parent reply other threads:[~2015-03-18 9:07 UTC|newest]
Thread overview: 26+ messages / expand[flat|nested] mbox.gz Atom feed top
2015-03-17 12:39 [PATCH 0/7] Detect future mis-uses of __ex_table section Quentin Casasnovas
2015-03-17 12:39 ` [PATCH 1/7] modpost: add strict white-listing when referencing sections Quentin Casasnovas
2015-03-17 16:25 ` Linus Torvalds
2015-03-18 9:14 ` Quentin Casasnovas
2015-03-20 1:29 ` Rusty Russell
2015-04-13 9:04 ` Quentin Casasnovas
2015-04-13 11:19 ` Rusty Russell
2015-04-13 11:24 ` Rusty Russell
2015-03-17 12:39 ` [PATCH 2/7] modpost: add .sched.text and .kprobes.text to the TEXT_SECTIONS list Quentin Casasnovas
2015-03-18 9:08 ` Quentin Casasnovas
2015-03-17 12:39 ` [PATCH 3/7] modpost: add handler function pointer to sectioncheck Quentin Casasnovas
2015-03-18 9:08 ` Quentin Casasnovas
2015-03-17 12:39 ` [PATCH 4/7] modpost: factorize symbol pretty print in get_pretty_name() Quentin Casasnovas
2015-03-18 9:08 ` Quentin Casasnovas
2015-03-17 12:40 ` [PATCH 5/7] modpost: mismatch_handler: retrieve tosym information only when needed Quentin Casasnovas
2015-03-18 9:09 ` Quentin Casasnovas
2015-03-17 12:40 ` [PATCH 6/7] scripts: add check_extable.sh script Quentin Casasnovas
2015-03-18 9:09 ` Quentin Casasnovas [this message]
2015-03-17 12:40 ` [PATCH 7/7] modpost: handle relocations mismatch in __ex_table Quentin Casasnovas
2015-03-18 9:09 ` Quentin Casasnovas
2015-04-13 11:18 ` Rusty Russell
2015-04-13 13:33 ` Quentin Casasnovas
2015-04-14 12:14 ` Thierry Reding
2015-04-14 12:35 ` Quentin Casasnovas
2015-04-15 3:27 ` Rusty Russell
2015-04-15 8:35 ` Quentin Casasnovas
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20150318090925.GK19131@chrystal.uk.oracle.com \
--to=quentin.casasnovas@oracle.com \
--cc=bp@alien8.de \
--cc=linux-kernel@vger.kernel.org \
--cc=mmarek@suse.cz \
--cc=oleg@redhat.com \
--cc=rusty@rustcorp.com.au \
--cc=torvalds@linux-foundation.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®