Le 09/04/2026 à 1:06 PM, Greg Kroah-Hartman a écrit :
FCGI content_length is a 16-bit field but fcgi_set_record_size()
is called with size_t/uint32_t arguments. With tune.bufsize >= 65544
(legal; cfgparse-global.c only enforces <= INT_MAX-16), a single
HTX DATA block or accumulated outbuf can exceed 65535 bytes. The
implicit conversion to uint16_t silently truncates the length field
while b_add(mbuf, outbuf.data) writes the full body.

A client posting ~99000 bytes can craft the body so that bytes
after the truncated length are parsed by PHP-FPM as fresh FCGI
records on the connection: a smuggled BEGIN_REQUEST + PARAMS with
arbitrary SCRIPT_FILENAME / PHP_VALUE bypasses all haproxy ACLs.

Fix the zero-copy path by refusing it when the block exceeds 65535
bytes (falls through to copy). Fix the copy path by capping
outbuf.size to 65535 + header so the data-fill loop naturally
stops at the FCGI maximum and emits the rest in a subsequent record.

The PARAMS path at line 2084 is similarly affected but harder to
trigger (requires combined header+param size > 65535) and is
covered by the same outbuf.size cap pattern if applied there;
left for a follow-up since the STDIN path is the practical attack.
---
  src/mux_fcgi.c | 14 +++++++++++++-
  1 file changed, 13 insertions(+), 1 deletion(-)

diff --git a/src/mux_fcgi.c b/src/mux_fcgi.c
index acb06a7637b3..f57e1a5fa99c 100644
--- a/src/mux_fcgi.c
+++ b/src/mux_fcgi.c
@@ -2153,7 +2153,14 @@ static size_t fcgi_strm_send_stdin(struct fcgi_conn 
*fconn, struct fcgi_strm *fs
                goto end;
        type = htx_get_blk_type(blk);
        size = htx_get_blksz(blk);
-       if (unlikely(size == count && b_size(mbuf) == b_size(buf) &&
+       /* FCGI content_len is uint16_t. With tune.bufsize >= 65544 a single
+        * HTX block can exceed 65535 bytes; the implicit truncation in
+        * fcgi_set_record_size() would then desynchronize the record
+        * stream and let the client smuggle a forged FCGI request to the
+        * backend. Refuse zero-copy in that case and let the copy path
+        * split the data across multiple records.
+        */
+       if (unlikely(size <= 0xFFFF && size == count && b_size(mbuf) == b_size(buf) 
&&
                     htx_nbblks(htx) == 1 && type == HTX_BLK_DATA)) {
                void *old_area = mbuf->area;
                int eom = (htx->flags & HTX_FL_EOM);
@@ -2212,6 +2219,11 @@ static size_t fcgi_strm_send_stdin(struct fcgi_conn 
*fconn, struct fcgi_strm *fs
        if (outbuf.size < FCGI_RECORD_HEADER_SZ + extra_bytes)
                goto full;
+ /* FCGI content_len is uint16_t; cap output to avoid truncation in
+        * fcgi_set_record_size() when tune.bufsize is large. */
+       if (outbuf.size > 0xFFFF + FCGI_RECORD_HEADER_SZ)
+               outbuf.size = 0xFFFF + FCGI_RECORD_HEADER_SZ;
+
        /* vsn: 1(FCGI_VERSION), type: (5)FCGI_STDIN, id: fstrm->id,
         *  len: 0x0000 (fill later), padding: 0x00, rsv: 0x00 */
        memcpy(outbuf.area, "\x01\x05\x00\x00\x00\x00\x00\x00", 
FCGI_RECORD_HEADER_SZ);

This one is valid too, but after a quick review, I guess fcgi_strm_send_params() must also be fixed. If you agree, I can amend the patch. But if you prefer to take a look and resubmit a new patch, I'm ok too. Let me know what you prefer.

--
Christopher Faulet



Reply via email to