ceph/win32_build.sh

294 lines
10 KiB
Bash
Raw Normal View History

#!/usr/bin/env bash
set -e
set -o pipefail
SCRIPT_DIR="$(dirname "$BASH_SOURCE")"
SCRIPT_DIR="$(realpath "$SCRIPT_DIR")"
num_vcpus=$(nproc)
CEPH_DIR="${CEPH_DIR:-$SCRIPT_DIR}"
BUILD_DIR="${BUILD_DIR:-${CEPH_DIR}/build}"
DEPS_DIR="${DEPS_DIR:-$CEPH_DIR/build.deps}"
ZIP_DEST="${ZIP_DEST:-$BUILD_DIR/ceph.zip}"
CLEAN_BUILD=${CLEAN_BUILD:-}
SKIP_BUILD=${SKIP_BUILD:-}
# Usefull when packaging existing binaries.
SKIP_CMAKE=${SKIP_CMAKE:-}
SKIP_DLL_COPY=${SKIP_DLL_COPY:-}
SKIP_TESTS=${SKIP_TESTS:-}
SKIP_BINDIR_CLEAN=${SKIP_BINDIR_CLEAN:-}
# Use Ninja's default, it might be useful when having few cores.
NUM_WORKERS_DEFAULT=$(( $num_vcpus + 2 ))
NUM_WORKERS=${NUM_WORKERS:-$NUM_WORKERS_DEFAULT}
DEV_BUILD=${DEV_BUILD:-}
# Unless SKIP_ZIP is set, we're preparing an archive that contains the Ceph
# binaries, debug symbols as well as the required DLLs.
SKIP_ZIP=${SKIP_ZIP:-}
# By default, we'll move the debug symbols to separate files located in the
# ".debug" directory. If "EMBEDDED_DBG_SYM" is set, the debug symbols will
# remain embedded in the binaries.
#
# Unfortunately we cannot use pdb symbols when cross compiling. cv2pdb
# well as llvm rely on mspdb*.dll in order to support this proprietary format.
EMBEDDED_DBG_SYM=${EMBEDDED_DBG_SYM:-}
# Allow for OS specific customizations through the OS flag.
# Valid options are currently "ubuntu", "suse", and "rhel".
OS=${OS}
if [[ -z $OS ]]; then
source /etc/os-release
case "$ID" in
opensuse*|suse|sles)
OS="suse"
;;
rhel|centos)
OS="rhel"
;;
ubuntu)
OS="ubuntu"
;;
*)
echo "Unsupported Linux distro $ID."
echo "only SUSE, Ubuntu and RHEL are supported."
echo "Set the OS environment variable to override."
exit 1
;;
esac
fi
# The main advantages of mingw-llvm:
# * not affected by the libstdc++/winpthread rw lock bugs
# * can generate pdb debug symbols, which are compatible with WinDBG
TOOLCHAIN=${TOOLCHAIN:-"mingw-llvm"}
case "$TOOLCHAIN" in
mingw-llvm)
echo "Using mingw-llvm."
USE_MINGW_LLVM=1
;;
mingw-gcc)
echo "Using mingw-gcc"
;;
*)
echo "Unsupported toolchain: $TOOLCHAIN."
echo "Allowed toolchains: mingw-llvm or mingw-gcc."
esac
# We'll have to be explicit here since auto-detecting doesn't work
# properly when cross compiling.
ALLOCATOR=${ALLOCATOR:-libc}
# Debug builds don't work with MINGW for the time being, failing with
# can't close <file>: File too big
# -Wa,-mbig-obj does not help.
CMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE:-}
if [[ -z $CMAKE_BUILD_TYPE ]]; then
# By default, we're building release binaries with minimal debug information.
export CFLAGS="$CFLAGS -g1"
export CXXFLAGS="$CXXFLAGS -g1"
CMAKE_BUILD_TYPE=Release
fi
ENABLE_SHARED=${ENABLE_SHARED:-ON}
binDir="$BUILD_DIR/bin"
strippedBinDir="$BUILD_DIR/bin_stripped"
# GDB will look for this directory by default.
dbgDirname=".debug"
dbgSymbolDir="$strippedBinDir/${dbgDirname}"
depsSrcDir="$DEPS_DIR/src"
depsToolsetDir="$DEPS_DIR/mingw"
cmakeGenerator="Ninja"
lz4Dir="${depsToolsetDir}/lz4"
sslDir="${depsToolsetDir}/openssl"
boostDir="${depsToolsetDir}/boost"
zlibDir="${depsToolsetDir}/zlib"
backtraceDir="${depsToolsetDir}/libbacktrace"
snappyDir="${depsToolsetDir}/snappy"
winLibDir="${depsToolsetDir}/windows/lib"
rbd: allow mounting images on Windows This change will allow mapping rbd images on Windows, leveraging the WNBD[1] Virtual Storport Miniport driver [2]. The behavior and CLI is similar to the Linux rbd-nbd, with a few notable differences: * device paths cannot be requested. The disk number and path will be picked by Windows. If a device path is provided by the user when mapping an image, it will be used as an identifier, which can also be used when unmapping the image. * the "show" command was added, which describes a specific mapping. This can be used for retrieving the disk path. * the "service" command was added, allowing rbd-wnbd to run as a Windows service. All mappings are currently perisistent, being recreated when the service stops, unless explicitly unmapped. The service disconnects the mappings when being stopped. * the "list" command also includes a "status" column. The purpose of the "service" mode is to ensure that mappings survive reboots and that the Windows service start order can be adjusted so that rbd images can be mapped before starting services that may depend on it, such as VMMS. The mapped images can either be consumed by the host directly or exposed to Hyper-V VMs. While at it, we'll skip building rbd-mirror as it's quite unlikely that this daemon is going to be used on Windows for now. [1] https://github.com/cloudbase/wnbd [2] https://docs.microsoft.com/en-us/windows-hardware/drivers/storage/overview-of-storage-virtual-miniport-drivers Signed-off-by: Lucian Petrut <lpetrut@cloudbasesolutions.com> Signed-off-by: Alin Gabriel Serdean <aserdean@cloudbasesolutions.com>
2020-07-30 11:40:31 +00:00
wnbdSrcDir="${depsSrcDir}/wnbd"
wnbdLibDir="${depsToolsetDir}/wnbd/lib"
cephfs: Add ceph-dokan, providing Windows support In order to expose ceph filesystems to Windows hosts, we propose including ceph-dokan[1][2] in the Ceph tree, while updating it to work with the latest CephFS and Dokany APIs. Dokany is a well maintained project (fork of the original Dokan project), allowing filesystems to be implemented in userspace, even providing a Fuse compatibility layer. One reason for not using the FUSE compatibility layer is that it's only covering the high level API while Ceph is using the low level FUSE API, which among other things is inode centric. Changes made by this patch compared to the upstream ceph-dokan: * support latest stable Dokany API. The upstream version relies on the legacy unmaintained Dokan API * return proper error codes, converting standard errno.h values to NTSTATUS * minor changes to support latest cephfs API * drop duplicated ceph code, no longer needed if we're to include it in tree. This makes it much easier to maintain. * drop redundant permission checks, leaving it up to libcephfs * use ceph argparse helpers * use ceph logging and daemon initialization * fixed unicode handling * switched to ceph coding style * made ceph.conf param optional, using the default path if available * enabled setting file timestamps * append support * configurable timeouts set once per mount * ensure that the error code is always logged * various cleanups (removed unused entry points, checks that have been moved to dokany, simplified conditional statements, unnecessary conversions in the hot path, etc). [1] https://github.com/ketor/ceph-dokan [2] https://github.com/ceph/ceph-dokan Signed-off-by: Lucian Petrut <lpetrut@cloudbasesolutions.com>
2020-08-07 12:07:20 +00:00
dokanSrcDir="${depsSrcDir}/dokany"
dokanLibDir="${depsToolsetDir}/dokany/lib"
depsDirs="$lz4Dir;$sslDir;$boostDir;$zlibDir;$backtraceDir;$snappyDir"
depsDirs+=";$winLibDir"
# Cmake recommends using CMAKE_PREFIX_PATH instead of link_directories.
# Still, some library dependencies may not include the full path (e.g. Boost
# sets the "-lz" flag through INTERFACE_LINK_LIBRARIES), which is why
# we have to ensure that the linker will be able to locate them.
linkDirs="$zlibDir/lib"
lz4Lib="${lz4Dir}/lib/dll/liblz4-1.dll"
lz4Include="${lz4Dir}/lib"
if [[ -n $CLEAN_BUILD ]]; then
echo "Cleaning up build dir: $BUILD_DIR"
rm -rf $BUILD_DIR
rm -rf $DEPS_DIR
fi
if [[ -z $SKIP_BINDIR_CLEAN ]]; then
echo "Cleaning up bin dir: $binDir"
rm -rf $binDir
fi
mkdir -p $BUILD_DIR
cd $BUILD_DIR
if [[ ! -f ${depsToolsetDir}/completed ]]; then
echo "Preparing dependencies: $DEPS_DIR. Log: ${BUILD_DIR}/build_deps.log"
NUM_WORKERS=$NUM_WORKERS \
DEPS_DIR=$DEPS_DIR \
OS="$OS" \
ENABLE_SHARED=$ENABLE_SHARED \
USE_MINGW_LLVM=$USE_MINGW_LLVM \
"$SCRIPT_DIR/win32_deps_build.sh" | tee "${BUILD_DIR}/build_deps.log"
fi
# Due to distribution specific mingw settings, the mingw.cmake file
# must be built prior to running cmake.
MINGW_CMAKE_FILE="$BUILD_DIR/mingw32.cmake"
MINGW_POSIX_FLAGS=1
source "$SCRIPT_DIR/mingw_conf.sh"
if [[ -z $SKIP_CMAKE ]]; then
# We'll need to cross compile Boost.Python before enabling
# "WITH_MGR".
echo "Generating solution. Log: ${BUILD_DIR}/cmake.log"
# This isn't propagated to some of the subprojects, we'll use an env variable
# for now.
export CMAKE_PREFIX_PATH=$depsDirs
if [[ -n $DEV_BUILD ]]; then
echo "Dev build enabled."
echo "Git versioning will be disabled."
ENABLE_GIT_VERSION="OFF"
WITH_CEPH_DEBUG_MUTEX="ON"
else
ENABLE_GIT_VERSION="ON"
WITH_CEPH_DEBUG_MUTEX="OFF"
fi
# As opposed to Linux, Windows shared libraries can't have unresolved
# symbols. Until we fix the dependencies (which are either unspecified
# or circular), we'll have to stick to static linking.
cmake -D CMAKE_PREFIX_PATH=$depsDirs \
-D MINGW_LINK_DIRECTORIES="$linkDirs" \
-D CMAKE_TOOLCHAIN_FILE="$MINGW_CMAKE_FILE" \
-D WITH_FMT_HEADER_ONLY=ON \
-D WITH_LIBCEPHSQLITE=OFF \
-D WITH_QATLIB=OFF -D WITH_QATZIP=OFF \
-D WITH_RDMA=OFF -D WITH_OPENLDAP=OFF \
cephfs: Add ceph-dokan, providing Windows support In order to expose ceph filesystems to Windows hosts, we propose including ceph-dokan[1][2] in the Ceph tree, while updating it to work with the latest CephFS and Dokany APIs. Dokany is a well maintained project (fork of the original Dokan project), allowing filesystems to be implemented in userspace, even providing a Fuse compatibility layer. One reason for not using the FUSE compatibility layer is that it's only covering the high level API while Ceph is using the low level FUSE API, which among other things is inode centric. Changes made by this patch compared to the upstream ceph-dokan: * support latest stable Dokany API. The upstream version relies on the legacy unmaintained Dokan API * return proper error codes, converting standard errno.h values to NTSTATUS * minor changes to support latest cephfs API * drop duplicated ceph code, no longer needed if we're to include it in tree. This makes it much easier to maintain. * drop redundant permission checks, leaving it up to libcephfs * use ceph argparse helpers * use ceph logging and daemon initialization * fixed unicode handling * switched to ceph coding style * made ceph.conf param optional, using the default path if available * enabled setting file timestamps * append support * configurable timeouts set once per mount * ensure that the error code is always logged * various cleanups (removed unused entry points, checks that have been moved to dokany, simplified conditional statements, unnecessary conversions in the hot path, etc). [1] https://github.com/ketor/ceph-dokan [2] https://github.com/ceph/ceph-dokan Signed-off-by: Lucian Petrut <lpetrut@cloudbasesolutions.com>
2020-08-07 12:07:20 +00:00
-D WITH_GSSAPI=OFF -D WITH_XFS=OFF \
-D WITH_FUSE=OFF -D WITH_DOKAN=ON \
-D WITH_BLUESTORE=OFF -D WITH_LEVELDB=OFF \
-D WITH_LTTNG=OFF -D WITH_BABELTRACE=OFF -D WITH_JAEGER=OFF \
-D WITH_SYSTEM_BOOST=ON -D WITH_MGR=OFF -D WITH_KVS=OFF \
cephfs: Add ceph-dokan, providing Windows support In order to expose ceph filesystems to Windows hosts, we propose including ceph-dokan[1][2] in the Ceph tree, while updating it to work with the latest CephFS and Dokany APIs. Dokany is a well maintained project (fork of the original Dokan project), allowing filesystems to be implemented in userspace, even providing a Fuse compatibility layer. One reason for not using the FUSE compatibility layer is that it's only covering the high level API while Ceph is using the low level FUSE API, which among other things is inode centric. Changes made by this patch compared to the upstream ceph-dokan: * support latest stable Dokany API. The upstream version relies on the legacy unmaintained Dokan API * return proper error codes, converting standard errno.h values to NTSTATUS * minor changes to support latest cephfs API * drop duplicated ceph code, no longer needed if we're to include it in tree. This makes it much easier to maintain. * drop redundant permission checks, leaving it up to libcephfs * use ceph argparse helpers * use ceph logging and daemon initialization * fixed unicode handling * switched to ceph coding style * made ceph.conf param optional, using the default path if available * enabled setting file timestamps * append support * configurable timeouts set once per mount * ensure that the error code is always logged * various cleanups (removed unused entry points, checks that have been moved to dokany, simplified conditional statements, unnecessary conversions in the hot path, etc). [1] https://github.com/ketor/ceph-dokan [2] https://github.com/ceph/ceph-dokan Signed-off-by: Lucian Petrut <lpetrut@cloudbasesolutions.com>
2020-08-07 12:07:20 +00:00
-D WITH_LIBCEPHFS=ON -D WITH_KRBD=OFF -D WITH_RADOSGW=OFF \
-D ENABLE_SHARED=$ENABLE_SHARED -D WITH_RBD=ON -D BUILD_GMOCK=ON \
-D WITH_CEPHFS=OFF -D WITH_MANPAGE=OFF \
-D WITH_MGR_DASHBOARD_FRONTEND=OFF -D WITH_SYSTEMD=OFF -D WITH_TESTS=ON \
-D LZ4_INCLUDE_DIR=$lz4Include -D LZ4_LIBRARY=$lz4Lib \
-D Backtrace_INCLUDE_DIR="$backtraceDir/include" \
-D Backtrace_LIBRARY="$backtraceDir/lib/libbacktrace.a" \
-D ENABLE_GIT_VERSION=$ENABLE_GIT_VERSION \
-D ALLOCATOR="$ALLOCATOR" -D CMAKE_BUILD_TYPE=$CMAKE_BUILD_TYPE \
rbd: allow mounting images on Windows This change will allow mapping rbd images on Windows, leveraging the WNBD[1] Virtual Storport Miniport driver [2]. The behavior and CLI is similar to the Linux rbd-nbd, with a few notable differences: * device paths cannot be requested. The disk number and path will be picked by Windows. If a device path is provided by the user when mapping an image, it will be used as an identifier, which can also be used when unmapping the image. * the "show" command was added, which describes a specific mapping. This can be used for retrieving the disk path. * the "service" command was added, allowing rbd-wnbd to run as a Windows service. All mappings are currently perisistent, being recreated when the service stops, unless explicitly unmapped. The service disconnects the mappings when being stopped. * the "list" command also includes a "status" column. The purpose of the "service" mode is to ensure that mappings survive reboots and that the Windows service start order can be adjusted so that rbd images can be mapped before starting services that may depend on it, such as VMMS. The mapped images can either be consumed by the host directly or exposed to Hyper-V VMs. While at it, we'll skip building rbd-mirror as it's quite unlikely that this daemon is going to be used on Windows for now. [1] https://github.com/cloudbase/wnbd [2] https://docs.microsoft.com/en-us/windows-hardware/drivers/storage/overview-of-storage-virtual-miniport-drivers Signed-off-by: Lucian Petrut <lpetrut@cloudbasesolutions.com> Signed-off-by: Alin Gabriel Serdean <aserdean@cloudbasesolutions.com>
2020-07-30 11:40:31 +00:00
-D WNBD_INCLUDE_DIRS="$wnbdSrcDir/include" \
-D WNBD_LIBRARIES="$wnbdLibDir/libwnbd.a" \
-D WITH_CEPH_DEBUG_MUTEX=$WITH_CEPH_DEBUG_MUTEX \
cephfs: Add ceph-dokan, providing Windows support In order to expose ceph filesystems to Windows hosts, we propose including ceph-dokan[1][2] in the Ceph tree, while updating it to work with the latest CephFS and Dokany APIs. Dokany is a well maintained project (fork of the original Dokan project), allowing filesystems to be implemented in userspace, even providing a Fuse compatibility layer. One reason for not using the FUSE compatibility layer is that it's only covering the high level API while Ceph is using the low level FUSE API, which among other things is inode centric. Changes made by this patch compared to the upstream ceph-dokan: * support latest stable Dokany API. The upstream version relies on the legacy unmaintained Dokan API * return proper error codes, converting standard errno.h values to NTSTATUS * minor changes to support latest cephfs API * drop duplicated ceph code, no longer needed if we're to include it in tree. This makes it much easier to maintain. * drop redundant permission checks, leaving it up to libcephfs * use ceph argparse helpers * use ceph logging and daemon initialization * fixed unicode handling * switched to ceph coding style * made ceph.conf param optional, using the default path if available * enabled setting file timestamps * append support * configurable timeouts set once per mount * ensure that the error code is always logged * various cleanups (removed unused entry points, checks that have been moved to dokany, simplified conditional statements, unnecessary conversions in the hot path, etc). [1] https://github.com/ketor/ceph-dokan [2] https://github.com/ceph/ceph-dokan Signed-off-by: Lucian Petrut <lpetrut@cloudbasesolutions.com>
2020-08-07 12:07:20 +00:00
-D DOKAN_INCLUDE_DIRS="$dokanSrcDir/dokan" \
-D DOKAN_LIBRARIES="$dokanLibDir/libdokan.a" \
-G "$cmakeGenerator" \
$CEPH_DIR 2>&1 | tee "${BUILD_DIR}/cmake.log"
fi # [[ -z $SKIP_CMAKE ]]
if [[ -z $SKIP_BUILD ]]; then
echo "Building using $NUM_WORKERS workers. Log: ${BUILD_DIR}/build.log"
echo "" > "${BUILD_DIR}/build.log"
cd $BUILD_DIR
ninja_targets="rados rbd rbd-wnbd "
ninja_targets+=" ceph-conf ceph-immutable-object-cache"
ninja_targets+=" cephfs ceph-dokan"
# TODO: do we actually need the ceph compression libs?
ninja_targets+=" compressor ceph_lz4 ceph_snappy ceph_zlib ceph_zstd"
if [[ -z $SKIP_TESTS ]]; then
ninja_targets+=" tests ceph_radosacl ceph_scratchtool "
ninja_targets+=`ninja -t targets | grep ceph_test | cut -d ":" -f 1 | grep -v exe`
fi
ninja -v $ninja_targets 2>&1 | tee "${BUILD_DIR}/build.log"
fi
if [[ -z $SKIP_DLL_COPY ]]; then
# To adjust mingw paths, see 'mingw_conf.sh'.
required_dlls=(
$zlibDir/zlib1.dll
$lz4Dir/lib/dll/liblz4-1.dll
$sslDir/bin/libcrypto-1_1-x64.dll
$sslDir/bin/libssl-1_1-x64.dll
$mingwLibpthreadDir/libwinpthread-1.dll)
if [[ $ENABLE_SHARED == "ON" ]]; then
required_dlls+=(
$boostDir/lib/*.dll
)
fi
if [[ -n $USE_MINGW_LLVM ]]; then
required_dlls+=(
$mingwTargetLibDir/libc++.dll
$mingwTargetLibDir/libunwind.dll)
else
required_dlls+=(
$mingwTargetLibDir/libstdc++-6.dll
$mingwTargetLibDir/libssp*.dll
$mingwTargetLibDir/libgcc_s_seh-1.dll)
fi
echo "Copying required dlls to $binDir."
cp ${required_dlls[@]} $binDir
fi
if [[ -z $SKIP_ZIP ]]; then
# Use a temp directory, in order to create a clean zip file
ZIP_TMPDIR=$(mktemp -d win_binaries.XXXXX)
if [[ -z $EMBEDDED_DBG_SYM ]]; then
echo "Extracting debug symbols from binaries."
rm -rf $strippedBinDir; mkdir $strippedBinDir
rm -rf $dbgSymbolDir; mkdir $dbgSymbolDir
# Strip files individually, to save time and space
for file in $binDir/*.exe $binDir/*.dll; do
dbgFilename=$(basename $file).debug
dbgFile="$dbgSymbolDir/$dbgFilename"
strippedFile="$strippedBinDir/$(basename $file)"
echo "Copying debug symbols: $dbgFile"
$MINGW_OBJCOPY --only-keep-debug $file $dbgFile
$MINGW_STRIP --strip-debug --strip-unneeded -o $strippedFile $file
$MINGW_OBJCOPY --remove-section .gnu_debuglink $strippedFile
$MINGW_OBJCOPY --add-gnu-debuglink=$dbgFile $strippedFile
done
# Copy any remaining files to the stripped directory
for file in $binDir/*; do
[[ ! -f $strippedBinDir/$(basename $file) ]] && \
cp $file $strippedBinDir
done
ln -s $strippedBinDir $ZIP_TMPDIR/ceph
else
ln -s $binDir $ZIP_TMPDIR/ceph
fi
echo "Building zip archive $ZIP_DEST."
# Include the README file in the archive
ln -s $CEPH_DIR/README.windows.rst $ZIP_TMPDIR/ceph/README.windows.rst
cd $ZIP_TMPDIR
[[ -f $ZIP_DEST ]] && rm $ZIP_DEST
zip -r $ZIP_DEST ceph
cd -
rm -rf $ZIP_TMPDIR/ceph/README.windows.rst $ZIP_TMPDIR
echo -e '\n WIN32 files zipped to: '$ZIP_DEST'\n'
fi