mirror of
https://github.com/gentoo-mirror/gentoo.git
synced 2026-04-30 16:57:29 -07:00
b70df51dad
In be04841726cae8da2f7589e1d8b30f6f5eab94d0, I naively didn't refetch the tarballs from upstream so didn't notice that the renamed repo affected the existing autogenerated tarballs from GH. I've diffed them and indeed it's only the directory name which changed. Closes: https://bugs.gentoo.org/971212 Fixes: be04841726cae8da2f7589e1d8b30f6f5eab94d0 Signed-off-by: Sam James <sam@gentoo.org>
234 lines
6.2 KiB
Bash
234 lines
6.2 KiB
Bash
# Copyright 2024-2026 Gentoo Authors
|
|
# Distributed under the terms of the GNU General Public License v2
|
|
|
|
EAPI=8
|
|
|
|
inherit edo flag-o-matic linux-info multilib systemd toolchain-funcs udev
|
|
|
|
DESCRIPTION="Dynamic BPF-based system-wide tracing tool"
|
|
HOMEPAGE="https://github.com/oracle/dtrace https://wiki.gentoo.org/wiki/DTrace"
|
|
|
|
if [[ ${PV} == 9999 ]]; then
|
|
EGIT_BRANCH="devel"
|
|
EGIT_REPO_URI="https://github.com/oracle/dtrace"
|
|
inherit git-r3
|
|
else
|
|
SRC_URI="https://github.com/oracle/dtrace/archive/refs/tags/${PV}.tar.gz -> ${P}.tar.gz"
|
|
|
|
KEYWORDS="-* ~amd64 ~arm64"
|
|
fi
|
|
|
|
LICENSE="UPL-1.0"
|
|
SLOT="0"
|
|
IUSE="test-install valgrind"
|
|
|
|
# TODO: can we make the wireshark dep conditional?
|
|
DEPEND="
|
|
dev-libs/elfutils
|
|
dev-libs/libbpf
|
|
dev-libs/libpfm:=
|
|
net-analyzer/wireshark[dumpcap]
|
|
net-libs/libpcap
|
|
>=sys-fs/fuse-3.2.0:3=
|
|
>=sys-libs/binutils-libs-2.42:=
|
|
virtual/zlib:=
|
|
"
|
|
RDEPEND="
|
|
${DEPEND}
|
|
!dev-debug/systemtap[dtrace-symlink(+)]
|
|
net-analyzer/wireshark
|
|
test-install? (
|
|
app-alternatives/bc
|
|
app-editors/vim-core
|
|
dev-build/make
|
|
dev-lang/perl
|
|
dev-util/perf
|
|
net-fs/nfs-utils
|
|
sys-apps/coreutils
|
|
sys-fs/xfsprogs
|
|
sys-process/time
|
|
virtual/jdk
|
|
)
|
|
"
|
|
BDEPEND="
|
|
dev-build/make
|
|
sys-apps/gawk
|
|
sys-devel/bison
|
|
>=sys-devel/bpf-toolchain-14.1.0
|
|
sys-devel/flex
|
|
"
|
|
DEPEND+=" valgrind? ( dev-debug/valgrind )"
|
|
|
|
QA_PRESTRIPPED="
|
|
usr/.*/dtrace/testsuite/test/triggers/.*
|
|
"
|
|
QA_FLAGS_IGNORED="
|
|
usr/.*/dtrace/testsuite/test/triggers/.*
|
|
"
|
|
|
|
# TODO: report upstream (bug #938221) although it seems like it's
|
|
# not relevant given it's a BPF object.
|
|
QA_EXECSTACK="
|
|
usr/*/dtrace/bpf_dlib.*
|
|
"
|
|
|
|
pkg_pretend() {
|
|
# TODO: optional kernel patches
|
|
|
|
# Basics for debugging information, BPF
|
|
local CONFIG_CHECK="~BPF ~DEBUG_INFO_BTF ~KALLSYMS_ALL"
|
|
|
|
CONFIG_CHECK+=" ~CUSE"
|
|
|
|
# Tracing
|
|
CONFIG_CHECK+=" ~TRACING"
|
|
CONFIG_CHECK+=" ~UPROBES ~UPROBE_EVENTS"
|
|
CONFIG_CHECK+=" ~FTRACE ~FTRACE_SYSCALLS ~DYNAMIC_FTRACE ~FUNCTION_TRACER"
|
|
CONFIG_CHECK+=" ~FPROBE"
|
|
CONFIG_CHECK+=" ~BPF_KPROBE_OVERRIDE ~FUNCTION_ERROR_INJECTION"
|
|
# DTrace can fallback to kprobes for fbt but people often want them off
|
|
# for security and newer kernels work fine with BPF for that, so
|
|
# let's omit it. kprobes are slower and scale poorly.
|
|
|
|
# https://gcc.gnu.org/PR84052
|
|
CONFIG_CHECK+=" !GCC_PLUGIN_RANDSTRUCT"
|
|
|
|
if use test-install ; then
|
|
# See test/modules
|
|
CONFIG_CHECK+=" ~EXT4_FS ~ISO9660_FS ~NFS_FS ~RDS ~TUN"
|
|
fi
|
|
|
|
check_extra_config
|
|
}
|
|
|
|
pkg_setup() {
|
|
eval unset ${!LC_*} LANG
|
|
}
|
|
|
|
src_configure() {
|
|
if tc-is-cross-compiler; then
|
|
die "DTrace does not yet support cross-compilation."
|
|
fi
|
|
|
|
tc-export CC
|
|
|
|
# lld does this by default, so fix that, although lld fails anyway...
|
|
# 'LIBDTRACE_1.0' to symbol 'dtrace_provider_modules' failed: symbol not defined
|
|
append-ldflags $(test-flags-CCLD -Wl,--undefined-version)
|
|
# mold and lld can't cope with some relocation types used, e.g.
|
|
# 'test-triggers--usdt-tst-forker-prov.o:(.SUNW_dof): unknown relocation: R_X86_64_GLOB_DAT'
|
|
tc-ld-force-bfd
|
|
|
|
# -fno-semantic-interposition seems to lead to a broken dtrace
|
|
# that can't actually obtain results from probes, even trivial examples
|
|
# just hang.
|
|
filter-flags -fno-semantic-interposition
|
|
# While it builds as of 2025-06-08, it's broken at runtime
|
|
# in the same way as -fno-semantic-interposition (hangs, no probes fire).
|
|
filter-lto
|
|
|
|
local confargs=(
|
|
--disable-dependency-tracking
|
|
# TODO: Maybe we should set the UNPRIV_UID to something? -3 is a bit... kludgy
|
|
--prefix="${EPREFIX}"/usr
|
|
--mandir="${EPREFIX}"/usr/share/man
|
|
--docdir="${EPREFIX}"/usr/share/doc/${PF}
|
|
--with-systemd
|
|
HAVE_LIBCTF=yes
|
|
HAVE_BPFV3=yes
|
|
HAVE_VALGRIND=$(usex valgrind)
|
|
)
|
|
|
|
edo ./configure "${confargs[@]}"
|
|
}
|
|
|
|
src_compile() {
|
|
local myemakeargs=(
|
|
verbose=1
|
|
$(usev !test-install TRIGGERS='')
|
|
)
|
|
|
|
if use amd64 ; then
|
|
! has_multilib_profile && myemakeargs+=( NATIVE_BITNESS_ONLY=1 )
|
|
fi
|
|
|
|
# -j1: https://github.com/oracle/dtrace/issues/82
|
|
emake -j1 "${myemakeargs[@]}"
|
|
}
|
|
|
|
src_test() {
|
|
# Needs root and is also very time-consuming
|
|
:;
|
|
}
|
|
|
|
src_install() {
|
|
emake DESTDIR="${D}" -j1 install $(usev test-install install-test)
|
|
|
|
# We want to strip neither the BPF libraries nor libdtrace.so itself
|
|
# as probes attach to some symbols that would get removed otherwise.
|
|
dostrip -x "/usr/$(get_libdir)"
|
|
|
|
# It's a binary (TODO: move it?)
|
|
docompress -x /usr/share/doc/${PF}/showUSDT
|
|
|
|
newinitd "${FILESDIR}"/dtprobed.init dtprobed
|
|
}
|
|
|
|
pkg_postinst() {
|
|
# We need a udev reload to pick up the CUSE device node rules.
|
|
udev_reload
|
|
|
|
if [[ -n ${REPLACING_VERSIONS} ]]; then
|
|
# TODO: Make this more intelligent wrt comparison
|
|
# One option for this is to detect when it's needed (DOF stash layout changes)
|
|
# and then e.g. sleep and restart for the user.
|
|
if systemd_is_booted ; then
|
|
einfo "Restart the DTrace 'dtprobed' service after upgrades once all dtraces are stopped with:"
|
|
einfo " systemctl try-restart dtprobed"
|
|
else
|
|
einfo "Restart the DTrace 'dtprobed' service after upgrades once all dtraces are stopped with:"
|
|
einfo " /etc/init.d/dtprobed restart"
|
|
fi
|
|
else
|
|
einfo "See https://wiki.gentoo.org/wiki/DTrace for getting started."
|
|
|
|
# We can't do magic for people with ROOT=.
|
|
if [[ -n ${ROOT} ]] ; then
|
|
einfo "Enable and start the DTrace 'dtprobed' service for systemd with:"
|
|
einfo " systemctl enable --now dtprobed"
|
|
einfo
|
|
einfo "Enable and start the DTrace 'dtprobed' service for OpenRC with:"
|
|
einfo " rc-update add dtprobed"
|
|
einfo " /etc/init.d/dtprobed start"
|
|
return
|
|
fi
|
|
|
|
# For first installs, we enable the service and start it.
|
|
#
|
|
# This is unusual, but the behaviour without dtprobed running
|
|
# is untested/unsupported. It's not a network service, it
|
|
# has no configuration, reads a single device node, and
|
|
# does all parsing within a seccomp jail. It also leads
|
|
# to hard-to-diagnose issues because USDT probes won't
|
|
# be registered and an application might have already
|
|
# started up which needs to be traced.
|
|
if systemd_is_booted ; then
|
|
ebegin "Enabling & starting DTrace 'dtprobed' service"
|
|
systemctl enable --now dtprobed
|
|
eend $?
|
|
else
|
|
ebegin "Enabling DTrace 'dtprobed' service"
|
|
rc-update add dtprobed
|
|
eend $?
|
|
|
|
ebegin "Starting DTrace 'dtprobed' service"
|
|
rc-service dtprobed start
|
|
eend $?
|
|
fi
|
|
fi
|
|
}
|
|
|
|
pkg_postrm() {
|
|
udev_reload
|
|
}
|