From patchwork Tue Nov 19 12:51:36 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Ross Burton X-Patchwork-Id: 52739 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from aws-us-west-2-korg-lkml-1.web.codeaurora.org (localhost.localdomain [127.0.0.1]) by smtp.lore.kernel.org (Postfix) with ESMTP id 282FCD44155 for ; Tue, 19 Nov 2024 12:51:50 +0000 (UTC) Received: from foss.arm.com (foss.arm.com [217.140.110.172]) by mx.groups.io with SMTP id smtpd.web11.20352.1732020703352095698 for ; Tue, 19 Nov 2024 04:51:43 -0800 Authentication-Results: mx.groups.io; dkim=none (message not signed); spf=pass (domain: arm.com, ip: 217.140.110.172, mailfrom: ross.burton@arm.com) Received: from usa-sjc-imap-foss1.foss.arm.com (unknown [10.121.207.14]) by usa-sjc-mx-foss1.foss.arm.com (Postfix) with ESMTP id EA578FEC for ; Tue, 19 Nov 2024 04:52:12 -0800 (PST) Received: from cesw-amp-gbt-1s-m12830-04.oss.cambridge.arm.com (usa-sjc-imap-foss1.foss.arm.com [10.121.207.14]) by usa-sjc-imap-foss1.foss.arm.com (Postfix) with ESMTPA id 848A43F5A1 for ; Tue, 19 Nov 2024 04:51:42 -0800 (PST) From: Ross Burton To: openembedded-core@lists.openembedded.org Subject: [PATCH] systemd: extract dependencies from .note.dlopen ELF segments Date: Tue, 19 Nov 2024 12:51:36 +0000 Message-Id: <20241119125136.2507561-1-ross.burton@arm.com> X-Mailer: git-send-email 2.34.1 MIME-Version: 1.0 List-Id: X-Webhook-Received: from li982-79.members.linode.com [45.33.32.79] by aws-us-west-2-korg-lkml-1.web.codeaurora.org with HTTPS for ; Tue, 19 Nov 2024 12:51:50 -0000 X-Groupsio-URL: https://lists.openembedded.org/g/openembedded-core/message/207388 First, this is likely not the final implementation, but a RFC and prototype. Some binaries don't dynamically link to libraries, but instead at runtime dlopen() them. This means extra work for distributions as the dependencies are not detected automatically, so libraries may be missing. systemd is one such project which does this, and in an attempt to solve the packaging problem it also embeds the names of the libraries that can potentially be opened at runtime into ELF notes. These can be read to generate package dependencies. For example: packages/cortexa57-poky-linux/systemd/libsystemd-shared: RRECOMMENDS: added "libkmod (['>= 33']) libzstd (['>= 1.5.6'])" packages/cortexa57-poky-linux/systemd/libsystemd: RRECOMMENDS: added "libzstd (['>= 1.5.6'])" I expect this code to be changed before merging. Whilst systemd is the main user of his approach right now, I expect to see it used in more places in the future so there's a reasonably good argument to merge it into the core shlibs code. Also it currently manually extracts and parses the data, whereas maybe we should incorporate pyelftools into meta/lib/oe and use that to parse ELF files across all of OE. Signed-off-by: Ross Burton --- meta/recipes-core/systemd/dlopen-deps.inc | 73 ++++++++++++++++++++++ meta/recipes-core/systemd/systemd_256.7.bb | 2 + 2 files changed, 75 insertions(+) create mode 100644 meta/recipes-core/systemd/dlopen-deps.inc diff --git a/meta/recipes-core/systemd/dlopen-deps.inc b/meta/recipes-core/systemd/dlopen-deps.inc new file mode 100644 index 00000000000..eaf6ca1f79a --- /dev/null +++ b/meta/recipes-core/systemd/dlopen-deps.inc @@ -0,0 +1,73 @@ +PACKAGEFUNCS =+ "package_generate_dlopen_deps" + +python package_generate_dlopen_deps() { + # https://systemd.io/ELF_DLOPEN_METADATA/ + + import struct, json + + def extract_segment(filename, segment): + """ + Return the named segment from the ELF. + """ + import tempfile, subprocess + + with tempfile.NamedTemporaryFile() as f: + cmd = [d.getVar("OBJCOPY"), "--dump-section", f"{segment}={f.name}", filename] + subprocess.run(cmd, check=True) + return f.read() + + def parse(buffer, is_little): + deps = [] + offset = 0 + while offset < len(buffer): + format = f"{'<' if is_little else '>'}iii" + name_size, desc_size, note_type = struct.unpack_from(format, buffer, offset) + offset += struct.calcsize(format) + + format = f"{name_size}s0i{desc_size}s0i" + if note_type == 0x407c0c0a: + name_b, desc_b = struct.unpack_from(format, buffer, offset) + name = name_b.strip(b"\x00").decode("ascii") + if name == "FDO": + desc = desc_b.strip(b"\x00").decode("utf-8") + deps.append(*json.loads(desc)) + offset += struct.calcsize(format) + return deps + + dep_map = { + "required": "RDEPENDS", + "recommended": "RRECOMMENDS", + "suggested": "RSUGGESTS" + } + + shlibs = oe.package.read_shlib_providers(d) + + for pkg, files in pkgfiles.items(): + # Skip -dbg packages as we won't need to generate dependencies for those + # but scanning can take time + if pkg.endswith("-dbg"): + continue + + for f in files: + # Skip symlinks, just look for real libraries + if cpath.islink(f): + continue + + if ".so." in f or f.endswith(".so"): + try: + elf = oe.qa.ELFFile(f) + elf.open() + for dep in parse(extract_segment(f, ".note.dlopen"), elf.isLittleEndian()): + for soname in dep["soname"]: + if soname in shlibs: + # TODO assumes the first match is good + package, version = list(shlibs[soname].values())[0] + dependency = dep_map[dep["priority"]] + bb.note(f"{pkg}: adding {dependency} on {package} via .note.dlopen") + d.appendVar(f"{dependency}:{pkg}", f" {package} (>= {version})") + else: + bb.warn(f"cannot find {soname}") + except oe.qa.NotELFFileError as e: + bb.note(f"Cannot extract ELF notes: {e}") + pass +} diff --git a/meta/recipes-core/systemd/systemd_256.7.bb b/meta/recipes-core/systemd/systemd_256.7.bb index f3af4ac44dc..4942d9518aa 100644 --- a/meta/recipes-core/systemd/systemd_256.7.bb +++ b/meta/recipes-core/systemd/systemd_256.7.bb @@ -909,3 +909,5 @@ pkg_postinst:udev-hwdb () { pkg_prerm:udev-hwdb () { rm -f $D${sysconfdir}/udev/hwdb.bin } + +require dlopen-deps.inc