t/lib-httpd: add cgi-lib.sh for concurrency-safe CGI helpers

The CGI helpers under t/lib-httpd keep state between requests in the shared
HTTPD_ROOT_PATH, on the assumption that the web server hands them one request
at a time. It does not: the server serves requests concurrently, and a single
Git operation can open more than one request to the same endpoint at once. A
partial fetch that receives a REF_DELTA against a missing promisor object
lazily fetches that base while the first response is still being served, and
protocol v2 and HTTP/2 can multiplex further. A helper that reads shared state
with a "test -f" and then acts on it, or that writes to a fixed-name scratch
file, therefore races with a concurrent request; the loser emits a malformed
response and the server answers HTTP 500.

Rather than fix each helper in isolation, add a small library, cgi-lib.sh,
that captures the safe idioms once: cgi_claim atomically claims a one-shot
marker, and cgi_first_request is an atomic first-wins test-and-set. Scratch
files use a documented $$ suffix rather than a helper, since that is all they
need. Install the library into HTTPD_ROOT_PATH alongside the helpers so they
can source it; the following commits build the stateful helpers on top of it.

Signed-off-by: Michael Montalbo <mmontalbo@gmail.com>

Author: mmontalbo

t/lib-httpd.sh

@@ -159,6 +159,7 @@ prepare_httpd() {
 	mkdir -p "$HTTPD_DOCUMENT_ROOT_PATH"
 	cp "$TEST_PATH"/passwd "$HTTPD_ROOT_PATH"
 	cp "$TEST_PATH"/proxy-passwd "$HTTPD_ROOT_PATH"
+	cp "$TEST_PATH"/cgi-lib.sh "$HTTPD_ROOT_PATH"
 	install_script incomplete-length-upload-pack-v2-http.sh
 	install_script incomplete-body-upload-pack-v2-http.sh
 	install_script error-no-report.sh

t/lib-httpd/cgi-lib.sh

@@ -0,0 +1,31 @@
+# Helpers for writing concurrency-safe CGI test scripts.
+#
+# The web server runs requests concurrently, and a single Git operation can
+# open more than one request to the same endpoint at once: a partial fetch
+# that receives a REF_DELTA against a missing promisor object lazily fetches
+# that base while the first response is still being served, and protocol v2
+# and HTTP/2 can multiplex further. A CGI helper that keeps state between
+# requests must therefore make each decision atomically and must not share
+# fixed-name scratch files between requests.
+#
+# For a scratch file, suffix its name with $$ (unique among live processes).
+# For a decision that only one of several concurrent requests may take, use one
+# of the primitives below rather than a "test -f" followed by a separate remove
+# or create, which races.
+
+# cgi_claim NAME
+#	Atomically claim NAME by renaming it aside. Succeeds for exactly one of
+#	any number of concurrent callers; the rest, and callers that arrive
+#	after NAME is gone, fail. Use to consume a one-shot marker that a test
+#	planted.
+cgi_claim () {
+	mv "$1" "$1.$$" 2>/dev/null
+}
+
+# cgi_first_request NAME
+#	Succeed for exactly one of any number of concurrent callers and fail
+#	for the rest, by creating NAME. mkdir(2) is atomic and fails if NAME
+#	exists, which makes the test-and-set a single step.
+cgi_first_request () {
+	mkdir "$1" 2>/dev/null
+}