DPDK-dev Archive on lore.kernel.org
 help / color / mirror / Atom feed
* [DPDK/other Bug 2151] ip_frag: output array overflow when fragmenting mbufs over 64 KB
@ 2026-11-20 18:28 bugzilla
  0 siblings, 0 replies; only message in thread
From: bugzilla @ 2026-11-20 18:28 UTC (permalink / raw)
  To: dev

https://bugs.dpdk.org/show_bug.cgi?id=2151

            Bug ID: 2151
           Summary: ip_frag: output array overflow when fragmenting mbufs
                    over 64 KB
           Product: DPDK
           Version: 26.11
          Hardware: All
                OS: All
            Status: UNCONFIRMED
          Severity: major
          Priority: Normal
         Component: other
          Assignee: dev@dpdk.org
          Reporter: stephen@networkplumber.org
  Target Milestone: ---
             Group: security

```
============================================================================
ANT-2026-KAH9I4A9  [high]  CWE-197  memory-safety
ip_frag: output array overflow when fragmenting mbufs over 64 KB
Location: lib/ip_frag/rte_ipv4_fragmentation.c:159  in
rte_ipv4_fragment_packet()
Commit:   9a45341f18a1ab47af7123e63d942b90c1b79f38
============================================================================
```

A vhost guest can make a host DPDK application that fragments its packets write
past the end of the caller's `pkts_out[]` array. The written values are host
mbuf pointers, not guest-chosen data, and the guest needs only a single Tx
descriptor in the default `net_vhost` configuration. The result is stack or
adjacent-table corruption and a crash of the host data-path process. Code
execution has not been shown. Both `rte_ipv4_fragment_packet()` and
`rte_ipv6_fragment_packet()` are affected, in any application that fragments
received packets larger than the egress MTU based on `pkt_len`, as
`examples/ip_fragmentation` does.

The only check that `pkts_out[nb_pkts_out]` can hold every fragment is at
`lib/ip_frag/rte_ipv4_fragmentation.c:158-160` (and
`rte_ipv6_fragmentation.c:100`). It compares against
`(uint16_t)(pkt_in->pkt_len - header_len)`. The loop that follows walks the
whole mbuf chain and never looks at `nb_pkts_out` again. A chain of 65556 bytes
or more therefore passes the check on its low 16 bits. The function then stores
one pointer per fragment past the array and returns a count larger than
`nb_pkts_out` (45 for 16 slots), so the caller also reads those out-of-bounds
entries. The vhost dequeue path puts no upper bound on frame length unless
`linear-buffer=1` is set. A remote sender can reach the same state through a
separate defect: reassembly ignores a failed `rte_pktmbuf_trim()`. That route
also needs jumbo frames, scattered Rx, and an application that reassembles
before it fragments. The attached reproducer has a vhost-user guest send a
65557-byte IPv4 frame to a forwarder modelled on `examples/ip_fragmentation`.
ASan reports a stack-buffer-overflow WRITE past `frags[16]` at
`rte_ipv4_fragmentation.c:258`. On a plain gcc build, larger frames end in a
"stack smashing detected" abort or SIGSEGV. The truncating check dates from
4c38e5532a07 ("ip_frag: refactor IPv4 fragmentation into a proper library").

-- 
You are receiving this mail because:
You are the assignee for the bug.

^ permalink raw reply	[flat|nested] only message in thread

only message in thread, other threads:[~2026-11-20 18:28 UTC | newest]

Thread overview: (only message) (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-20 18:28 [DPDK/other Bug 2151] ip_frag: output array overflow when fragmenting mbufs over 64 KB bugzilla

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox