Author SHA1 Message Date
sneak 021e5c4325 A client that stops sending after the app answers is timed out (closes #130)
check / check (push) Waiting to run
When the app answered before the client had sent its whole body, and its
answer ended with its last bytes, every timeout was stopped,
SWWAF_CLIENT_REQUEST_TIMEOUT included. The transport went on waiting to
read the body, and Go's server cannot send the answer until that read
returns, so a client that stopped sending held the request until it
closed the connection. The timeouts now run on until the client has sent
its body, as they do before the app answers: once
SWWAF_CLIENT_REQUEST_TIMEOUT runs out the client is sent the app's
answer, the connection is closed, and the log line says timed_out.

Model: opus-5-5
2026-10-14 17:11:53 +00:00
7 changed files with 39 additions and 464 deletions
+2 -6
View File
@@ -1,12 +1,11 @@
.PHONY: bootstrap setup test lint fmt fmt-check tidy check docker hooks build \
run example-app gitea-requests
run example-app
# Makefile targets are thin shims; the implementations live in script/
# per the scripts-to-rule-them-all pattern (see the Entrypoints section
# of README.md). tidy writes go.mod and go.sum as `go mod tidy` does,
# which test checks. build and run are for working on the code by hand;
# example-app checks the image with an app built on it; gitea-requests
# reports the gitea requests the Core Rule Set refuses at the defaults.
# example-app checks the image with an app built on it.
bootstrap:
@script/bootstrap
@@ -46,6 +45,3 @@ run:
example-app:
@script/example-app
gitea-requests:
@script/gitea-requests
-13
View File
@@ -2374,19 +2374,6 @@ so that they run in minimal containers.
the same volume still refuses the banned client; then removes the containers,
the volume and both images. It needs network access, for nixpkgs' binary
cache, and `script/check` does not run it; `make example-app` does.
- `script/gitea-requests`: builds the image and runs gitea 1.27.3 behind it,
with every setting of `smallwebwaf` at its default but `SWWAF_UPSTREAM_URL`,
which names gitea's container. It sends through it the gitea requests that the
notes specific to gitea in [`SPEC.md`](SPEC.md) name for a file name in a
query parameter, for a path and for a container image push, and a short sample
of gitea's ordinary use: signing in, pages of files and commits, a diff, the
API, a push, clone and fetch over HTTP, and two container image pushes. It
prints a line for each request `smallwebwaf` logged, with what it did with it
and the ids of the Core Rule Set's rules that matched it, then removes the
containers, the network and the images it made, but gitea's own image. It
needs network access, and a docker daemon that pushes to a registry on
`localhost` over plain HTTP, as docker does by default; `script/check` does
not run it, and `make gitea-requests` does.
## TODO
+4 -40
View File
@@ -1477,42 +1477,8 @@ holds any token file.
Saving such a rule stores it, but the branch settings page gitea then
returns to gets the 403 answer of `smallwebwaf`, so the save looks as
if it failed.
- A file name in a query parameter, those the fourth change leaves out
of 930120, 932160 and 932260 included, in which the Core Rule Set's
rules for Java, script, PowerShell and SQL find what they look for:
`runtime` in OpenJDK's
`src/java.base/share/classes/java/lang/Runtime.java` (944110), a Java
class, as in `java.lang.String.md` (944130), `document.write.js`
(941180), `process.env.d.ts` (934100), `__proto__` (934130), the
PowerShell command `Install-Script`, as in `install-script.sh`
(932120), or a database's name, as in PostgreSQL's
`information_schema.sql` (942140). For such a file, the API's list of
its commits (`path`) is refused, and so are the API's call for the
contents of files that names it in its query (`body`), the commit
graph of that file alone (`file`), and the hidden lines of its diff
(`path`), which stay hidden. A visitor whom gitea sends to sign in on
the way to a page of such a file that needs signing in, such as its
edit page, gets the 403 answer of `smallwebwaf` in place of the
sign-in page (`redirect_to`), and can still sign in by opening the
sign-in page itself.
- A path: a file name ending in `~`, or an `.xhtml` file whose path
holds a space.
- A path that holds what the Core Rule Set takes for a Java class
(944130), as `java.lang.String.md` and
`META-INF/services/org.apache.commons.logging.LogFactory` do, or for
Node.js code (934100), as `process.env.d.ts` does, or that has a
directory named `...` (930100, 930110), such as `.../notes.txt`. The
page of such a file, its raw file, its history, its blame and its edit
page are refused; the directory that holds it lists it, and a clone
carries it.
- A container image push that reuses a layer of an image docker has
pushed to that gitea before, when that image's name, after its owner,
starts with an entry on the list of commands (932260), such as
`owner/python3-app` or `owner/docker-base`. docker asks gitea to take
the layer from that image, naming it in the query parameter `from`,
which 932260 checks. The push fails with the 403 answer of
`smallwebwaf`, and so does each new try, since docker asks in the same
way.
- Artifact uploads from `actions/upload-artifact@v3` send the header
`Content-Range`, which the Core Rule Set refuses (920450). The action
sends two files at a time, does not retry a 403 and sends no further file
@@ -1600,12 +1566,10 @@ holds any token file.
- Core Rule Set false positives against real apps (a search for code, and any
body once `SWWAF_WAF_BODY_LIMIT` is set): a match refuses only that request
and bans no one by itself; the request log names the rule, and exclusions by
rule id and path fix it. `script/gitea-requests` runs gitea behind the image
at the defaults, sends through it the requests the notes specific to gitea
name for a file name in a query parameter, for a path and for a container
image push, and a short sample of gitea's ordinary use, and reports each
request the Core Rule Set refuses, so that those notes can be checked again
when gitea or the Core Rule Set changes version.
rule id and path fix it. Further gitea requests the Core Rule Set may refuse
at the defaults, found by reading gitea's source rather than a running gitea,
are collected in https://git.eeqj.de/sneak/smallwebwaf/issues/30 and checked
when milestone 1 runs in front of a real gitea.
- Attacks carried in request bodies: not refused by default, since on a code
forge bodies are full of code the Core Rule Set takes for attacks. Most of
what shows in URLs and headers is still refused, the client that sends an
-69
View File
@@ -86,75 +86,6 @@ func TestEveryClientIsLookedUpWithoutWaitingWhileNoSettingNeedsIt(t *testing.T)
}
}
func TestRequestThatWaitedForItsLookupIsCountedWhenItGoesOn(t *testing.T) {
t.Parallel()
// The country list has a new client's request wait for its answer,
// which the stand-in for GeoJS gives only once released.
geojsURL, asked, release := startHeldGeoJS(t)
app := startApp(t, func(http.ResponseWriter, *http.Request) {})
clk := &clock{now: time.Date(2026, 10, 6, 0, 0, 58, 0, time.UTC)}
addr, out, server := startProxyWithClock(t, app.URL, geojsURL, clk.Now,
map[string]string{
trustedProxies: trustLocalhost,
lookupTimeout: "1h",
deniedCountries: "kp",
})
requestFromDE := func() *http.Request {
req := newRequest(t, http.MethodGet, addr, "/", http.NoBody)
req.Header.Set(forwardedFor, fromDE)
return req
}
// fromDE's first request arrives two seconds before a minute ends,
// and waits for GeoJS.
first, httpClient := requestFromDE(), newClient(t)
firstEnded := make(chan error, 1)
go func() {
res, err := httpClient.Do(first)
if err == nil {
err = readAnswer(res).err
}
firstEnded <- err
}()
waitUntil(func() bool { return len(asked()) == 1 })
// Half a second into the next minute, the answer is kept, as an edit
// of lookups.json would keep it, while the first request still waits
// for GeoJS. The second request goes on at once, and is counted.
clk.advance(2500 * time.Millisecond)
server.GeoJS.Load([]lookup.Answer{{
Client: netip.MustParsePrefix(fromDE + "/32"), ASN: asnDE, ASName: asNameDE,
Country: "DE", Answered: clk.Now(), Used: clk.Now(),
}})
wantStatus(t, do(t, requestFromDE()), http.StatusOK)
// GeoJS answers, and the first request goes on.
release()
err := <-firstEnded
if err != nil {
t.Fatalf("first request: %v", err)
}
wantLine(t, out.requestLines(t, 2)[1], http.StatusOK, requestlog.ActionForward)
// The third request's count holds all three: the first was counted
// when it went on, in the minute under way, and the second's count
// was kept.
clk.advance(500 * time.Millisecond)
wantStatus(t, do(t, requestFromDE()), http.StatusOK)
if got := out.requestLines(t, 3)[2].Counts.Minute; got != 3 {
t.Errorf("the third request's count in the minute is %v, want 3", got)
}
}
func TestASNumberAndNameInTheLogLineTheHistoryTheBanNotesAndTheAlert(t *testing.T) {
t.Parallel()
+11 -7
View File
@@ -261,12 +261,6 @@ func (rq *request) checkClient(ctx context.Context) string {
rq.lookUp(ctx)
// The lookup can wait for GeoJS, so the checks after it read the time
// again: a request counted at the time it arrived could be counted
// after a later one of its client's, which the rate limits would take
// for the clock set back.
now = rq.h.now()
if rq.countryDenied() {
return requestlog.ActionCountryDenied
}
@@ -795,9 +789,19 @@ func (rq *request) responseTimedOut() {
}
}
// responseReceived is called once the app has sent its whole answer.
// responseReceived is called once the app has sent its whole answer. An
// app may answer before the client has sent its whole body, as with a 404
// it sends without reading it. The timeouts then run on: until the
// transport's read of the body returns, Go's server cannot send the
// answer, and SWWAF_CLIENT_REQUEST_TIMEOUT is what ends that read for a
// client that stops sending.
func (rq *request) responseReceived() {
rq.complete = true
if rq.body != nil && !rq.body.received.Load() {
return
}
rq.stopTimers()
}
+22
View File
@@ -194,6 +194,28 @@ func sendLargeBody(t *testing.T, addr string) net.Conn {
return conn
}
func TestClientStopsSendingAfterTheAppHasAnswered(t *testing.T) {
t.Parallel()
// The app answers at once without reading the body, as an app does
// for a path it does not serve.
app := startApp(t, http.NotFound)
addr, out := startProxy(t, app.URL, map[string]string{
clientRequestTimeout: shortTimeoutSetting,
metricsToken: token,
})
start := time.Now()
got := readResponse(t, sendPartOfBody(t, addr))
wantTimedOut(t, start)
// The app's answer came before the time ran out, so the client is
// sent that answer, not smallwebwaf's 408.
wantStatus(t, got, http.StatusNotFound)
wantLine(t, out.requestLine(t), http.StatusNotFound, requestlog.ActionTimedOut)
wantLimitHits(t, addr, clientRequestTimeout, 1)
}
func TestAppTooSlowToAnswer(t *testing.T) {
t.Parallel()
-329
View File
@@ -1,329 +0,0 @@
#!/bin/sh
# script/gitea-requests: build the image and run gitea 1.27.3 behind it,
# with every setting of smallwebwaf at its default but SWWAF_UPSTREAM_URL,
# which names gitea's container. Then send through it the gitea requests
# that reading gitea's source found the Core Rule Set refuses, and a short
# sample of gitea's ordinary use: signing in, pages of files and commits,
# a diff, the API, a push, clone and fetch over HTTP, and two container
# image pushes. Last, print each request smallwebwaf logged, with what it
# did with it and the ids of the Core Rule Set's rules that matched it.
# Every request comes from one client, docker's gateway, so more than
# SWWAF_ERROR_BURST_THRESHOLD, 30, refusals in a minute would ban it, and
# the requests after would show as banned. The containers, the network, the images and the files the script makes
# are removed however it ends; gitea's own image stays, as the base images
# of a build do. It needs network access, and a docker daemon that pushes
# to a registry on localhost over plain HTTP, as docker does by default.
# script/check does not run this.
set -eu
SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd -P)"
ROOT="$(cd "$SCRIPT_DIR/.." && pwd -P)"
# gitea 1.27.3, 2026-08-29
GITEA_IMAGE="gitea/gitea@sha256:87a67ee09d3ae0d1df5fda5dcda3e2a1f9236a45b0a59025d6e00e46adc43bef"
# Named after this run, so that runs in other clones on the same host
# never touch each other's.
NAME="$("$SCRIPT_DIR/projectname")-gitea-$$"
IMAGE="$NAME-image"
NETWORK="$NAME"
SMALLWEBWAF="$NAME-smallwebwaf"
GITEA="$NAME-gitea"
# gitea's one user, who owns the repository and the container images.
PASSWORD="owner-password"
BROWSER="Mozilla/5.0 (X11; Linux x86_64; rv:131.0) Gecko/20100101 Firefox/131.0"
# The files the script makes, and smallwebwaf's address once it runs,
# localhost and the port docker publishes it on: a host name, since the
# Core Rule Set takes a Host that is an IP address for a warning (920350).
WORK=""
address=""
cleanup() {
docker rm --force --volumes "$SMALLWEBWAF" "$GITEA" >/dev/null 2>&1 || true
docker network rm "$NETWORK" >/dev/null 2>&1 || true
docker rmi --force "$IMAGE" >/dev/null 2>&1 || true
if [ -n "$address" ]; then
docker rmi --force "$address/owner/python3-app:1" "$address/owner/app:1" \
>/dev/null 2>&1 || true
fi
rm -rf "$WORK"
}
fail() {
echo "gitea-requests: $*" >&2
report
exit 1
}
# wait_for <what fails> <command>...: run the command every second until
# it succeeds, for at most a minute.
wait_for() {
failure="$1"
shift
tries=0
until "$@"; do
tries=$((tries + 1))
[ "$tries" -lt 60 ] || fail "$failure"
sleep 1
done
}
healthy() {
status="$(docker inspect --format '{{.State.Health.Status}}' "$SMALLWEBWAF")"
[ "$status" = healthy ]
}
# logged <text>...: a line of smallwebwaf's output holds every text, in
# any order.
logged() {
lines="$(docker logs "$SMALLWEBWAF" 2>&1)"
for text in "$@"; do
lines="$(printf '%s\n' "$lines" | grep -F "$text")" || return 1
done
}
# send <label> <path> <curl option>...: send a request for path to
# smallwebwaf, with label as its X-Request-ID, which the report gives,
# and keep the body of the answer in $WORK/answer. The Core Rule Set
# reads that header too, so a label names no file the cases name.
send() {
label="$1"
path="$2"
shift 2
curl --silent --show-error --max-time 30 --output "$WORK/answer" \
--header "X-Request-ID: $label" "$@" "http://$address$path"
}
# page <label> <path> <curl option>...: send the request as a browser
# does, with the cookies gitea has set. Its Referer is the repository's
# page, which links to the others, and never a page the Core Rule Set
# refused: its address would have the next request refused too.
page() {
label="$1"
path="$2"
shift 2
send "$label" "$path" --compressed --user-agent "$BROWSER" \
--header "Accept: text/html,application/xhtml+xml,*/*;q=0.8" \
--header "Accept-Language: en-US,en;q=0.5" \
--cookie "$WORK/cookies" --cookie-jar "$WORK/cookies" \
--referer "http://$address/owner/repo" "$@"
}
# api <label> <path> <curl option>...: send the request as a client of
# gitea's API does, as the owner.
api() {
label="$1"
path="$2"
shift 2
send "$label" "$path" --user "owner:$PASSWORD" \
--header "Accept: application/json" "$@"
}
# repo_git <git argument>...: run git as the owner, without the host's
# own git settings, so that nothing it does is kept outside $WORK.
repo_git() {
GIT_CONFIG_GLOBAL=/dev/null GIT_CONFIG_NOSYSTEM=1 git \
-c user.name=owner -c user.email=owner@example.invalid "$@"
}
# lines: forty lines of text, enough for a diff to hide some of them.
lines() {
i=1
while [ "$i" -le 40 ]; do
echo "line $i"
i=$((i + 1))
done
}
# make_repository: create the repository owner/repo, holding a file of
# each name the cases name, and push it, which is the push over HTTP of
# the sample. The second commit adds a line at the end of Runtime.java,
# so that its diff hides the lines above.
make_repository() {
status="$(api create-repository /api/v1/user/repos \
--header "Content-Type: application/json" --data '{"name":"repo"}' \
--write-out '%{http_code}')"
[ "$status" = 201 ] || fail "creating the repository was answered $status"
dir="$WORK/repo"
repo_git init --quiet --initial-branch=main "$dir"
for file in \
src/java.base/share/classes/java/lang/Runtime.java \
java.lang.String.md \
META-INF/services/org.apache.commons.logging.LogFactory \
document.write.js \
process.env.d.ts \
test/fixtures/__proto__.json \
install-script.sh \
src/backend/catalog/information_schema.sql \
.../notes.txt \
README.md; do
mkdir -p "$dir/$(dirname "$file")"
lines >"$dir/$file"
repo_git -C "$dir" add "$file"
done
repo_git -C "$dir" commit --quiet --message "Add the files"
echo "line 41" >>"$dir/src/java.base/share/classes/java/lang/Runtime.java"
repo_git -C "$dir" commit --quiet --all --message "Add a line"
repo_git -C "$dir" -c "http.extraHeader=X-Request-ID: git-push" push --quiet \
"http://owner:$PASSWORD@$address/owner/repo.git" main ||
fail "the push of the repository failed"
}
# browse: sign in, then open pages as a person does. The cases are those
# of https://git.eeqj.de/sneak/smallwebwaf/issues/30.
browse() {
# Case: gitea sends a visitor who is not signed in to the sign-in
# page, with the page asked for in redirect_to, when that page needs
# signing in, as a file's edit page does.
page issue-30-case /owner/repo/_edit/main/document.write.js --location
page sign-in /user/login
status="$(page sign-in /user/login --header "Origin: http://$address" \
--data-urlencode user_name=owner --data-urlencode "password=$PASSWORD" \
--write-out '%{http_code}')"
[ "$status" = 303 ] || fail "signing in was answered $status"
page browse /
page browse /owner/repo
page browse /owner/repo/src/branch/main/README.md
page browse /owner/repo/commits/branch/main
page browse "/owner/repo/commit/$(repo_git -C "$WORK/repo" rev-parse HEAD)"
# Case: expanding the lines the diff hides, by the link the diff's
# page gives for it.
excerpt="$(sed -n 's/.*data-fetch-url="\([^"]*blob_excerpt[^"]*\)".*/\1/p' \
"$WORK/answer" | head -n 1 | sed 's/&amp;/\&/g')"
[ -n "$excerpt" ] || fail "the diff's page has no link to its hidden lines"
page issue-30-case "$excerpt"
# Cases: the commit graph of one file, and the pages of files whose
# path the Core Rule Set takes for an attack.
page issue-30-case "/owner/repo/graph?file=src/java.base/share/classes/java/lang/Runtime.java"
page issue-30-case /owner/repo/src/branch/main/META-INF/services/org.apache.commons.logging.LogFactory
page issue-30-case /owner/repo/src/branch/main/java.lang.String.md
page issue-30-case /owner/repo/src/branch/main/process.env.d.ts
page issue-30-case /owner/repo/src/branch/main/.../notes.txt
}
# call_api: the API's calls the cases name: the commits of a file, for a
# file of each name the Core Rule Set takes for an attack in the query
# parameter path, and the file-contents call, which takes its list of
# files as JSON in the query parameter body.
call_api() {
for file in \
src/java.base/share/classes/java/lang/Runtime.java \
java.lang.String.md \
document.write.js \
process.env.d.ts \
test/fixtures/__proto__.json \
install-script.sh \
src/backend/catalog/information_schema.sql; do
api issue-30-case /api/v1/repos/owner/repo/commits --get --data-urlencode "path=$file"
done
api issue-30-case /api/v1/repos/owner/repo/file-contents --get \
--data-urlencode 'body={"files":["src/java.base/share/classes/java/lang/Runtime.java"]}'
}
# push_images: push a container image to owner/python3-app, then the same
# image to owner/app. For the second, the case, docker asks gitea to take
# its layer from owner/python3-app, naming it in the query parameter
# from. docker sends no X-Request-ID, so its requests have the ids
# smallwebwaf gives them. The image is one small file, imported rather
# than built.
push_images() {
echo "$NAME" >"$WORK/layer.txt"
tar -cf - -C "$WORK" layer.txt |
docker import - "$address/owner/python3-app:1" >/dev/null
docker tag "$address/owner/python3-app:1" "$address/owner/app:1"
echo "$PASSWORD" | DOCKER_CONFIG="$WORK/docker" docker login --username owner \
--password-stdin "$address" >/dev/null 2>&1 || fail "docker login failed"
for image in "$address/owner/python3-app:1" "$address/owner/app:1"; do
DOCKER_CONFIG="$WORK/docker" docker push --quiet "$image" >/dev/null ||
echo "gitea-requests: the push of $image failed" >&2
done
}
# clone_and_fetch: the clone and fetch over HTTP of the sample.
clone_and_fetch() {
repo_git -c "http.extraHeader=X-Request-ID: git-clone" clone --quiet \
"http://owner:$PASSWORD@$address/owner/repo.git" "$WORK/clone" ||
fail "the clone failed"
repo_git -C "$WORK/clone" -c "http.extraHeader=X-Request-ID: git-fetch" fetch --quiet ||
fail "the fetch failed"
}
# field <name> <line>: the text of the field name in line, a line of the
# request log, or nothing when it has none.
field() {
printf '%s\n' "$2" | sed -n "s/.*\"$1\":\"\([^\"]*\)\".*/\1/p"
}
# report: print a line for each request smallwebwaf logged, but its
# health checks: what it did with the request, the ids of the Core Rule
# Set's rules that matched it, its X-Request-ID, and the request.
# waf_blocked is a refusal by the Core Rule Set, and forward a request
# passed on to gitea, whatever gitea answered.
report() {
docker logs "$SMALLWEBWAF" 2>/dev/null | grep '"type":"request"' |
grep -v '"action":"admin"' | sed 's/\\u0026/\&/g' >"$WORK/requests"
while IFS= read -r line; do
ids="$(printf '%s\n' "$line" | sed -n 's/.*"waf_rule_ids":\[\([0-9,]*\)\].*/\1/p')"
query="$(field query "$line")"
printf '%-11s %-13s %-17s %s %s%s\n' "$(field action "$line")" "${ids:--}" \
"$(field request_id "$line")" "$(field method "$line")" \
"$(field path "$line")" "${query:+?$query}"
done <"$WORK/requests"
echo "gitea-requests: $(grep -c . "$WORK/requests") requests," \
"$(grep -c '"action":"waf_blocked"' "$WORK/requests") refused by the Core Rule Set"
}
main() {
cd "$ROOT"
trap cleanup EXIT
trap 'exit 1' HUP INT TERM
WORK="$(mktemp -d)"
docker build --no-cache -t "$IMAGE" .
# smallwebwaf first, so that gitea can be given the address it is
# reached at, and gitea's image pulled before, so that gitea starts
# within the minute smallwebwaf's health check waits for it.
docker pull --quiet "$GITEA_IMAGE" >/dev/null
docker network create "$NETWORK" >/dev/null
docker run --detach --name "$SMALLWEBWAF" --network "$NETWORK" \
--publish 127.0.0.1::8080 \
--env SWWAF_UPSTREAM_URL="http://$GITEA:3000" \
"$IMAGE" >/dev/null
published="$(docker port "$SMALLWEBWAF" 8080/tcp)"
address="localhost:${published##*:}"
docker run --detach --name "$GITEA" --network "$NETWORK" \
--env GITEA__security__INSTALL_LOCK=true \
--env GITEA__server__ROOT_URL="http://$address/" \
"$GITEA_IMAGE" >/dev/null
# smallwebwaf's health check passes once gitea takes connections.
wait_for "the health check did not pass" healthy
docker exec --user git "$GITEA" gitea admin user create --admin \
--username owner --password "$PASSWORD" --email owner@example.invalid \
--must-change-password=false >/dev/null
make_repository
browse
call_api
push_images
clone_and_fetch
# The fetch's last request is the last of all; once it is logged, so
# is every other.
wait_for "smallwebwaf logged no line for the fetch" \
logged '"request_id":"git-fetch"' '"method":"POST"'
report
}
main "$@"