diff options
author | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-09 13:08:37 +0000 |
---|---|---|
committer | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-09 13:08:37 +0000 |
commit | 971e619d8602fa52b1bfcb3ea65b7ab96be85318 (patch) | |
tree | 26feb2498c72b796e07b86349d17f544046de279 /tests/shell/helpers/test-wrapper.sh | |
parent | Initial commit. (diff) | |
download | nftables-971e619d8602fa52b1bfcb3ea65b7ab96be85318.tar.xz nftables-971e619d8602fa52b1bfcb3ea65b7ab96be85318.zip |
Adding upstream version 1.0.9.upstream/1.0.9upstream
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'tests/shell/helpers/test-wrapper.sh')
-rwxr-xr-x | tests/shell/helpers/test-wrapper.sh | 223 |
1 files changed, 223 insertions, 0 deletions
diff --git a/tests/shell/helpers/test-wrapper.sh b/tests/shell/helpers/test-wrapper.sh new file mode 100755 index 0000000..13b918f --- /dev/null +++ b/tests/shell/helpers/test-wrapper.sh @@ -0,0 +1,223 @@ +#!/bin/bash -e + +# This wrapper wraps the invocation of the test. It is called by run-tests.sh, +# and already in the unshared namespace. +# +# For some printf debugging, you can also patch this file. + +array_contains() { + local needle="$1" + local a + shift + for a; do + [ "$a" = "$needle" ] && return 0 + done + return 1 +} + +TEST="$1" +TESTBASE="$(basename "$TEST")" +TESTDIR="$(dirname "$TEST")" + +START_TIME="$(cut -d ' ' -f1 /proc/uptime)" + +export TMPDIR="$NFT_TEST_TESTTMPDIR" + +CLEANUP_UMOUNT_VAR_RUN=n + +cleanup() { + if [ "$CLEANUP_UMOUNT_VAR_RUN" = y ] ; then + umount "/var/run" &>/dev/null || : + fi +} + +trap cleanup EXIT + +printf '%s\n' "$TEST" > "$NFT_TEST_TESTTMPDIR/name" + +read tainted_before < /proc/sys/kernel/tainted + +if [ "$NFT_TEST_HAS_UNSHARED_MOUNT" = y ] ; then + # We have a private mount namespace. We will mount /var/run/ as a tmpfs. + # + # The main purpose is so that we can create /var/run/netns, which is + # required for `ip netns add` to work. When running as rootless, this + # is necessary to get such tests to pass. When running rootful, it's + # still useful to not touch the "real" /var/run/netns of the system. + # + # Note that this also hides everything that might reside in /var/run. + # That is desirable, as tests should not depend on content there (or if + # they do, we need to explicitly handle it as appropriate). + if mount -t tmpfs --make-private "/var/run" ; then + CLEANUP_UMOUNT_VAR_RUN=y + fi + mkdir -p /var/run/netns +fi + +TEST_TAGS_PARSED=0 +ensure_TEST_TAGS() { + if [ "$TEST_TAGS_PARSED" = 0 ] ; then + TEST_TAGS_PARSED=1 + TEST_TAGS=( $(sed -n '1,10 { s/^.*\<\(NFT_TEST_REQUIRES\|NFT_TEST_SKIP\)\>\s*(\s*\(NFT_TEST_SKIP_[a-zA-Z0-9_]\+\|NFT_TEST_HAVE_[a-zA-Z0-9_]\+\)\s*).*$/\1(\2)/p }' "$1" 2>/dev/null || : ) ) + fi +} + +rc_test=0 + +if [ "$rc_test" -eq 0 ] ; then + for KEY in $(compgen -v | grep '^NFT_TEST_HAVE_') ; do + if [ "${!KEY}" != n ]; then + continue + fi + ensure_TEST_TAGS "$TEST" + if array_contains "NFT_TEST_REQUIRES($KEY)" "${TEST_TAGS[@]}" ; then + echo "Test skipped due to $KEY=n (test has \"NFT_TEST_REQUIRES($KEY)\" tag)" >> "$NFT_TEST_TESTTMPDIR/testout.log" + rc_test=77 + break + fi + done +fi + +if [ "$rc_test" -eq 0 ] ; then + for KEY in $(compgen -v | grep '^NFT_TEST_SKIP_') ; do + if [ "${!KEY}" != y ]; then + continue + fi + ensure_TEST_TAGS "$TEST" + if array_contains "NFT_TEST_SKIP($KEY)" "${TEST_TAGS[@]}" ; then + echo "Test skipped due to $KEY=y (test has \"NFT_TEST_SKIP($KEY)\" tag)" >> "$NFT_TEST_TESTTMPDIR/testout.log" + rc_test=77 + break + fi + done +fi + +if [ "$rc_test" -eq 0 ] ; then + "$TEST" &>> "$NFT_TEST_TESTTMPDIR/testout.log" || rc_test=$? +fi + +$NFT list ruleset > "$NFT_TEST_TESTTMPDIR/ruleset-after" + +read tainted_after < /proc/sys/kernel/tainted + +DUMPPATH="$TESTDIR/dumps" +DUMPFILE="$DUMPPATH/$TESTBASE.nft" +NODUMPFILE="$DUMPPATH/$TESTBASE.nodump" + +dump_written= + +# The caller can request a re-geneating of the dumps, by setting +# DUMPGEN=y. +# +# This only will happen if the command completed with success. +# +# It also will only happen for tests, that have a "$DUMPPATH" directory. There +# might be tests, that don't want to have dumps created. The existence of the +# directory controls that. Tests that have a "$NODUMPFILE" file, don't get a dump generated. +if [ "$rc_test" -eq 0 -a "$DUMPGEN" = y -a -d "$DUMPPATH" -a ! -f "$NODUMPFILE" ] ; then + dump_written=y + if [ ! -f "$DUMPFILE" ] ; then + # No dumpfile exists yet. We generate both a .nft and a .nodump + # file. The user can pick which one to commit to git. + : > "$NODUMPFILE" + fi + cat "$NFT_TEST_TESTTMPDIR/ruleset-after" > "$DUMPFILE" +fi + +rc_dump=0 +if [ "$rc_test" -ne 77 -a -f "$DUMPFILE" ] ; then + if [ "$dump_written" != y ] ; then + if ! $DIFF -u "$DUMPFILE" "$NFT_TEST_TESTTMPDIR/ruleset-after" &> "$NFT_TEST_TESTTMPDIR/ruleset-diff" ; then + rc_dump=1 + else + rm -f "$NFT_TEST_TESTTMPDIR/ruleset-diff" + fi + fi +fi +if [ "$rc_dump" -ne 0 ] ; then + echo "$DUMPFILE" > "$NFT_TEST_TESTTMPDIR/rc-failed-dump" +fi + +rc_chkdump=0 +# check that a flush after the test succeeds. We anyway need a clean ruleset +# for the `nft --check` next. +$NFT flush ruleset &> "$NFT_TEST_TESTTMPDIR/rc-failed-chkdump" || rc_chkdump=1 +if [ -f "$DUMPFILE" ] ; then + # We have a dumpfile. Call `nft --check` to possibly cover new code + # paths. + if [ "$rc_test" -eq 77 ] ; then + # The test was skipped. Possibly we don't have the required + # features to process this file. Ignore any output and exit + # code, but still call the program (for valgrind or sanitizer + # issue we hope to find). + $NFT --check -f "$DUMPFILE" &>/dev/null || : + else + $NFT --check -f "$DUMPFILE" &>> "$NFT_TEST_TESTTMPDIR/rc-failed-chkdump" || rc_chkdump=1 + fi +fi +if [ -s "$NFT_TEST_TESTTMPDIR/rc-failed-chkdump" ] ; then + # Non-empty output? That is wrong. + rc_chkdump=1 +elif [ "$rc_chkdump" -eq 0 ] ; then + rm -rf "$NFT_TEST_TESTTMPDIR/rc-failed-chkdump" +fi +if [ "$rc_chkdump" -ne 0 ] ; then + # Ensure we don't have empty output files. Always write something, so + # that `grep ^ -R` lists the file. + echo -e "<<<<<\n\nCalling \`nft --check\` (or \`nft flush ruleset\`) failed for \"$DUMPFILE\"" >> "$NFT_TEST_TESTTMPDIR/rc-failed-chkdump" +fi + +rc_valgrind=0 +[ -f "$NFT_TEST_TESTTMPDIR/rc-failed-valgrind" ] && rc_valgrind=1 + +rc_tainted=0 +if [ "$tainted_before" != "$tainted_after" ] ; then + echo "$tainted_after" > "$NFT_TEST_TESTTMPDIR/rc-failed-tainted" + rc_tainted=1 +fi + +if [ "$rc_valgrind" -ne 0 ] ; then + rc_exit=122 +elif [ "$rc_tainted" -ne 0 ] ; then + rc_exit=123 +elif [ "$rc_test" -ge 118 -a "$rc_test" -le 124 ] ; then + # Special exit codes are reserved. Coerce them. + rc_exit=125 +elif [ "$rc_test" -ne 0 ] ; then + rc_exit="$rc_test" +elif [ "$rc_dump" -ne 0 ] ; then + rc_exit=124 +elif [ "$rc_chkdump" -ne 0 ] ; then + rc_exit=121 +else + rc_exit=0 +fi + + +# We always write the real exit code of the test ($rc_test) to one of the files +# rc-{ok,skipped,failed}, depending on which it is. +# +# Note that there might be other rc-failed-{dump,tainted,valgrind} files with +# additional errors. Note that if such files exist, the overall state will +# always be failed too (and an "rc-failed" file exists). +# +# On failure, we also write the combined "$rc_exit" code from "test-wrapper.sh" +# to "rc-failed-exit" file. +# +# This means, failed tests will have a "rc-failed" file, and additional +# "rc-failed-*" files exist for further information. +if [ "$rc_exit" -eq 0 ] ; then + RC_FILENAME="rc-ok" +elif [ "$rc_exit" -eq 77 ] ; then + RC_FILENAME="rc-skipped" +else + RC_FILENAME="rc-failed" + echo "$rc_exit" > "$NFT_TEST_TESTTMPDIR/rc-failed-exit" +fi +echo "$rc_test" > "$NFT_TEST_TESTTMPDIR/$RC_FILENAME" + +END_TIME="$(cut -d ' ' -f1 /proc/uptime)" +WALL_TIME="$(awk -v start="$START_TIME" -v end="$END_TIME" "BEGIN { print(end - start) }")" +printf "%s\n" "$WALL_TIME" "$START_TIME" "$END_TIME" > "$NFT_TEST_TESTTMPDIR/times" + +exit "$rc_exit" |