From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-pg1-x536.google.com (mail-pg1-x536.google.com [IPv6:2607:f8b0:4864:20::536]) by sourceware.org (Postfix) with ESMTPS id 21D6C3858400; Mon, 18 Oct 2021 18:27:58 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org 21D6C3858400 Received: by mail-pg1-x536.google.com with SMTP id g184so17036089pgc.6; Mon, 18 Oct 2021 11:27:58 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=x-gm-message-state:mime-version:references:in-reply-to:from:date :message-id:subject:to:cc:content-transfer-encoding; bh=xB9sOtrT4Q03SmmIP9RmyYHPPX3LcMU1TKD8bzRGouY=; b=b97iBwkVb3t99G1Xx5gLijTgRkpLqiD7YakKNRlY8ZOvHkiRdRJJxIlp/tpJTvEexz LKwEcjJZqZEIWG0WpOvb8AFP5qh/k5ar+wY0ei5C7cXIfuGrxzbrZdoSGi3fEFpZXeb0 2O8W/Lo7JWg+RzzzCsAUCO13L65JPlMCGA6pqSWZaIW3T7Prv8JM9winQlvY8MG5HM9F YsgWaNozRuwrE8vI9LSEZzxfVtVZh7+UNhF4FqqA8gRCQxNWD2BkaV7OlsPyEB540yts eJZsURhKE0rkseAZMNkphYgQyU0EGsPlErULWGIkUwKh/TiW+6zKd/IAkgbVR6/Tz/XT Im7A== X-Gm-Message-State: AOAM533e1TvuDJZCWpCwbpDUnqBSjbkyNRlsFTZFi/UjFUXxfcsOa+jD y16bI99YdZ80VeHJtflf8U2l01BgxeYpJntA+zdeVYeERSE= X-Google-Smtp-Source: ABdhPJzY2adkwnARW0Iz41NH8vtlCwkQ3W9Wd8NZJP8wpR44RDo2tVosn4e5P83gjz0xqSnsf8PI3vn+zqTdVOuSKzs= X-Received: by 2002:a63:6116:: with SMTP id v22mr24884215pgb.225.1634581676901; Mon, 18 Oct 2021 11:27:56 -0700 (PDT) MIME-Version: 1.0 References: <20211017005020.2645717-1-maskray@google.com> <20211018181530.yz62n64dgn2kd4oe@google.com> In-Reply-To: <20211018181530.yz62n64dgn2kd4oe@google.com> From: "H.J. Lu" Date: Mon, 18 Oct 2021 11:27:20 -0700 Message-ID: Subject: Re: [PATCH v2] elf: Support DT_RELR relative relocation format [BZ #27924] To: =?UTF-8?B?RsSBbmctcnXDrCBTw7JuZw==?= Cc: GNU C Library , Binutils Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable X-Spam-Status: No, score=-3030.0 required=5.0 tests=BAYES_00, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, FREEMAIL_FROM, GIT_PATCH_0, RCVD_IN_DNSWL_NONE, SPF_HELO_NONE, SPF_PASS, TXREP autolearn=ham autolearn_force=no version=3.4.4 X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on server2.sourceware.org X-BeenThere: libc-alpha@sourceware.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Libc-alpha mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Mon, 18 Oct 2021 18:28:00 -0000 On Mon, Oct 18, 2021 at 11:15 AM F=C4=81ng-ru=C3=AC S=C3=B2ng wrote: > > > On 2021-10-18, H.J. Lu wrote: > >On Mon, Oct 18, 2021 at 9:17 AM F=C4=81ng-ru=C3=AC S=C3=B2ng wrote: > >> > >> On Mon, Oct 18, 2021 at 7:42 AM H.J. Lu wrote: > >> > > >> > On Sat, Oct 16, 2021 at 5:50 PM Fangrui Song via Binutils > >> > wrote: > >> > > > >> > > PIE and shared objects usually have many relative relocations. In > >> > > 2017/2018, SHT_RELR/DT_RELR was proposed on > >> > > https://groups.google.com/g/generic-abi/c/bX460iggiKg/m/GxjM0L-PBA= AJ > >> > > ("Proposal for a new section type SHT_RELR") and is a pre-standard= . RELR > >> > > usually takes 3% or smaller space than R_*_RELATIVE relocations. T= he > >> > > virtual memory size of a mostly statically linked PIE is typically= 5~10% > >> > > smaller. > >> > > > >> > > --- > >> > > > >> > > Notes I will not include in the submitted commit: > >> > > > >> > > Available on https://sourceware.org/git/?p=3Dglibc.git;a=3Dshortlo= g;h=3Drefs/heads/maskray/relr > >> > > > >> > > "pre-standard": even Solaris folks are happy with the refined gene= ric-abi > >> > > proposal. Cary Coutant will apply the change > >> > > https://sourceware.org/pipermail/libc-alpha/2021-October/131781.ht= ml > >> > > > >> > > This patch is simpler than Chrome OS's glibc patch and makes ELF_D= YNAMIC_DO_RELR > >> > > available to all ports. I don't think the current glibc implementa= tion > >> > > supports ia64 in an ELFCLASS32 container. That said, the style I u= sed is > >> > > works with an ELFCLASS32 container for 64-bit machine if ElfW(Addr= ) is > >> > > 64-bit. > >> > > > >> > > * Chrome OS folks have carried a local patch since 2018 (latest ve= rsion: > >> > > https://chromium.googlesource.com/chromiumos/overlays/chromiumos= -overlay/+/refs/heads/main/sys-libs/glibc/files/local/glibc-2.32). > >> > > I.e. this feature has been battle tested. > >> > > * Android bionic supports 2018 and switched to DT_RELR=3D=3D36 in = 2020. > >> > > * The Linux kernel has supported CONFIG_RELR since 2019-08 > >> > > (https://git.kernel.org/linus/5cf896fb6be3effd9aea455b22213e27be= 8bdb1d). > >> > > * A musl patch (by me) exists but is not applied: > >> > > https://www.openwall.com/lists/musl/2019/03/06/3 > >> > > * rtld-elf from FreeBSD 14 will support DT_RELR. > >> > > > >> > > I believe upstream glibc should support DT_RELR to benefit all Lin= ux > >> > > distributions. I filed some feature requests to get their attentio= n: > >> > > > >> > > * Gentoo: https://bugs.gentoo.org/818376 > >> > > * Arch Linux: https://bugs.archlinux.org/task/72433 > >> > > * Debian https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=3D99659= 8 > >> > > * Fedora https://bugzilla.redhat.com/show_bug.cgi?id=3D2014699 > >> > > > >> > > As of linker support (to the best of my knowledge): > >> > > > >> > > * LLD support DT_RELR. > >> > > * https://chromium.googlesource.com/chromiumos/overlays/chromiumos= -overlay/+/refs/heads/main/sys-devel/binutils/files/ > >> > > has a gold patch. > >> > > * GNU ld feature request https://sourceware.org/bugzilla/show_bug.= cgi?id=3D27923 > >> > > > >> > > I wish that GNU ld and gold maintainers can implement the feature = as well :) > >> > > > >> > > Tested on aarch64 and x86_64. > >> > > > >> > > Changes from v1 (https://sourceware.org/pipermail/libc-alpha/2021-= October/131768.html) > >> > > * Fix style, simplify code > >> > > * Improve test > >> > > --- > >> > > configure | 31 +++++++++++++++++++++++++++++++ > >> > > configure.ac | 4 ++++ > >> > > elf/Makefile | 4 ++++ > >> > > elf/dynamic-link.h | 28 ++++++++++++++++++++++++++++ > >> > > elf/elf.h | 13 +++++++++++-- > >> > > elf/get-dynamic-info.h | 3 +++ > >> > > elf/tst-relr.c | 27 +++++++++++++++++++++++++++ > >> > > 7 files changed, 108 insertions(+), 2 deletions(-) > >> > > create mode 100644 elf/tst-relr.c > >> > > > >> > > diff --git a/configure b/configure > >> > > index 3227e434d3..fdab6a97ef 100755 > >> > > --- a/configure > >> > > +++ b/configure > >> > > @@ -6067,6 +6067,37 @@ $as_echo "$libc_linker_feature" >&6; } > >> > > config_vars=3D"$config_vars > >> > > have-depaudit =3D $libc_cv_depaudit" > >> > > > >> > > +{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for linker that= supports --pack-dyn-relocs=3Drelr" >&5 > >> > > +$as_echo_n "checking for linker that supports --pack-dyn-relocs= =3Drelr... " >&6; } > >> > > +libc_linker_feature=3Dno > >> > > +if test x"$gnu_ld" =3D x"yes"; then > >> > > + cat > conftest.c < >> > > +int _start (void) { return 42; } > >> > > +EOF > >> > > + if { ac_try=3D'${CC-cc} $CFLAGS $CPPFLAGS $LDFLAGS $no_ssp > >> > > + -Wl,--pack-dyn-relocs=3Drelr -nostdlib -nostar= tfiles > >> > > + -fPIC -shared -o conftest.so conftest.c > >> > > + 1>&5' > >> > > + { { eval echo "\"\$as_me\":${as_lineno-$LINENO}: \"$ac_try\""; = } >&5 > >> > > + (eval $ac_try) 2>&5 > >> > > + ac_status=3D$? > >> > > + $as_echo "$as_me:${as_lineno-$LINENO}: \$? =3D $ac_status" >&5 > >> > > + test $ac_status =3D 0; }; } > >> > > + then > >> > > + libc_linker_feature=3Dyes > >> > > + fi > >> > > + rm -f conftest* > >> > > +fi > >> > > +if test $libc_linker_feature =3D yes; then > >> > > + libc_cv_relr=3Dyes > >> > > +else > >> > > + libc_cv_relr=3Dno > >> > > +fi > >> > > +{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $libc_linker_fea= ture" >&5 > >> > > +$as_echo "$libc_linker_feature" >&6; } > >> > > +config_vars=3D"$config_vars > >> > > +have-relr =3D $libc_cv_relr" > >> > > + > >> > > { $as_echo "$as_me:${as_lineno-$LINENO}: checking for linker that= supports --no-dynamic-linker" >&5 > >> > > $as_echo_n "checking for linker that supports --no-dynamic-linker= ... " >&6; } > >> > > libc_linker_feature=3Dno > >> > > diff --git a/configure.ac b/configure.ac > >> > > index 00f49f09f7..96110f9d7d 100644 > >> > > --- a/configure.ac > >> > > +++ b/configure.ac > >> > > @@ -1354,6 +1354,10 @@ LIBC_LINKER_FEATURE([--depaudit], [-Wl,--de= paudit,x], > >> > > [libc_cv_depaudit=3Dyes], [libc_cv_depaudit=3D= no]) > >> > > LIBC_CONFIG_VAR([have-depaudit], [$libc_cv_depaudit]) > >> > > > >> > > +LIBC_LINKER_FEATURE([--pack-dyn-relocs=3Drelr], [-Wl,--pack-dyn-r= elocs=3Drelr], > >> > > + [libc_cv_relr=3Dyes], [libc_cv_relr=3Dno]) > >> > > +LIBC_CONFIG_VAR([have-relr], [$libc_cv_relr]) > >> > > + > >> > > LIBC_LINKER_FEATURE([--no-dynamic-linker], > >> > > [-Wl,--no-dynamic-linker], > >> > > [libc_cv_no_dynamic_linker=3Dyes], > >> > > diff --git a/elf/Makefile b/elf/Makefile > >> > > index bf45d8ee24..2c4cdfac68 100644 > >> > > --- a/elf/Makefile > >> > > +++ b/elf/Makefile > >> > > @@ -245,6 +245,10 @@ tests-special +=3D $(objpfx)tst-audit14-cmp.o= ut $(objpfx)tst-audit15-cmp.out \ > >> > > $(objpfx)tst-audit16-cmp.out > >> > > endif > >> > > endif > >> > > +ifeq ($(have-relr),yes) > >> > > +tests +=3D tst-relr > >> > > +LDFLAGS-tst-relr +=3D -Wl,--pack-dyn-relocs=3Drelr > >> > > +endif > >> > > endif > >> > > >> > Is DT_RELR only generated for PIE? If yes, you need to add it > >> > to tests-pie and compile it as PIE. > >> > >> PIE and shared objects. PDE doesn't need relative relocations. > >> It is useful to ensure that -Wl,--pack-dyn-relocs=3Drelr doesn't cause > >> breakage to PDE. > > > >Then you need to add a PIE test to verify that > > > >1. It has DT_RELR. > >2. It runs correctly. > > --- a/elf/Makefile > +++ b/elf/Makefile > @@ -246,4 +246,8 @@ tests-special +=3D $(objpfx)tst-audit14-cmp.out $(obj= pfx)tst-audit15-cmp.out \ > endif > endif > +ifeq ($(have-relr),yes) > +tests +=3D tst-relr > +LDFLAGS-tst-relr +=3D -Wl,--pack-dyn-relocs=3Drelr > +endif > endif > tests +=3D $(tests-execstack-$(have-z-execstack)) > > I added this chunk to my test (https://sourceware.org/git/?p=3Dglibc.git;= a=3Dshortlog;h=3Drefs/heads/maskray/relr): > > ElfW(Dyn) *d =3D _DYNAMIC; > if (d) > { > bool has_relr =3D false; > for (; d->d_tag !=3D DT_NULL; d++) > if (d->d_tag =3D=3D DT_RELR) > has_relr =3D true; > if (!has_relr) > { > fprintf (stderr, "no DT_RELR\n"); > return 1; > } > } We need 2 tests from the same source: 1. The DT_RELR linker option doesn't break PDE. 2. The DT_RELR linker option works for PIE. You only have one test. > >> > [hjl@gnu-cfl-2 tmp]$ gcc -pie -fPIE -O2 tst-relr.c > >> > -Wl,--pack-dyn-relocs=3Drelr -fuse-ld=3Dlld > >> > [hjl@gnu-cfl-2 tmp]$ ./a.out > >> > Segmentation fault (core dumped) > >> > [hjl@gnu-cfl-2 tmp]$ > >> > > >> > Given that the current lld implementation generates broken > >> > binaries for existing glibc without any warning at run-time, > >> > we should use a different linker command line option to > >> > implement it properly so that the new binary will fail to > >> > run on glibc without DT_RELR support at run-time. > >> > >> I don't think so. LLD's design is to be machine agnostic and NOT make > >> decisions which would vary on different machines. > >> --pack-dyn-relocs=3Drelr is not the default. The user tells LLD to use > >> DT_RELR and the user is responsible for making sure target ld.so > >> supports DT_RELR. > >> LLD is often used as a cross linker. The host ld.so doesn't support > >> LLD doesn't mean that LLD should disable the format. > >> For example, it is totally fine to link a FreeBSD executable on a Linu= x machine. > > > >For Linux targets, we need the glibc version dependency for > >DT_RELR support at run-time. > > I know that having an executable crash right away (SIGSEGV, if ld.so does= n't > resolve relative relocations) without a diagnostic is arguably poor user = experience, > and you might come from this angle and therefore suggest a version check. Correct. We need to resolve this before DT_RELR can be accepted. > However, this tight libc coupling is not LLD's usual design decision. > LLD doesn't want to understand what libc it needs to work with. > > The user experience problem is related to the ELF spirit that we don't wa= nt to > unnecessarily error for unknown sections/tags/etc. In this case, the evol= vement > of a foundamental feature (relocation) has to be incompatible with older = loaders. > That compatibility check could be implemented in many ways, with their ow= n pros and cons. > For example, allocating a range of dynamic tags saying that unknown tags = in this range > are hard errors. However, such a foundamental evolution is rare. I tend t= o think we just > accept the status quo and do nothing on it. --pack-dyn-relocs=3Drelr user= s, at least > from the time being, understand that they need new loaders for the relati= vely new feature. It may be OK for some use cases. I don't think it is acceptable for all u= se cases which glibc cares about. > > The problem can be moot when users use --pack-dyn-relocs=3Drelr with new = glibc. > The linked binary will get a high version requirement and running with ol= der glibc > will be detected right away. I just think it's not linker's responsibilit= y to synthesize > a symbol targeting a specific glibc version. > > -z relr=3Dglibc targeting glibc and -z relr=3Dnone targeting non-glibc? T= hat'd just > be unneeded additional complexity. > If GNU ld finally decides to proceed with the -z relr=3D* road, I can con= sider > adding a -z relr compatibility option to LLD as well, but I may not synth= esize a symbol. > It's complexity with relatively little benefit IMHO. --=20 H.J.