Re: [PATCH net] batman-adv: bound BLA claim and backbone gateway table growth

[Sven Eckelmann <sven@narfation.org>]

[-- Attachment #1: Type: text/plain, Size: 5661 bytes --]

On Friday, 10 July 2026 18:52:24 CEST Ibrahim Hashimov wrote:
> batadv_bla_add_claim() and batadv_bla_get_backbone_gw() each kzalloc a
> new hash entry (struct batadv_bla_claim / struct batadv_bla_backbone_gw)
> for every distinct (mac, vid) / (orig, vid) pair carried in a
> BLA-group-conforming CLAIM frame, and insert it into
> bat_priv->bla.claim_hash / backbone_hash. There is no maximum-entry
> cap on either table -- entries are only ever removed by the
> timeout-driven periodic purge (batadv_bla_purge_claims() /
> batadv_bla_purge_backbone_gw(), BATADV_BLA_CLAIM_TIMEOUT /
> BATADV_BLA_BACKBONE_TIMEOUT, on the order of 100 s).
> 
> The BLA group a frame must carry is htons(crc16(primary_hardif_mac)),
> which is trivially observable/derivable by any node already on the
> mesh soft-interface (batadv_check_claim_group()). A single on-mesh
> sender can therefore emit CLAIM frames with an incrementing source MAC
> and force one kzalloc(GFP_ATOMIC) per frame on both hot paths, growing
> kernel memory without bound for as long as the attacker keeps sending
> -- uncontrolled resource allocation. Both allocations are
> GFP_ATOMIC with a NULL check, so this is a graceful memory-pressure
> DoS, not a crash: there is no OOB access.
> 
> batman-adv already has an established pattern for capping an
> attacker-/peer-influenced, unbounded-growth per-mesh-interface
> resource: the TP-meter session list bounds concurrent sessions with a
> fixed ceiling and an atomic_add_unless() admission check that rejects
> new allocations once the cap is hit, logging and freeing/decrementing
> on the abort path (net/batman-adv/tp_meter.c, BATADV_TP_MAX_NUM,
> bat_priv->tp_num, "Meter: too many ongoing sessions, aborting").
> 
> Apply the same pattern to BLA: add two atomic_t counters,
> bat_priv->bla.num_claims and bat_priv->bla.num_backbone_gws, each
> capped at a new BATADV_BLA_MAX_CLAIMS / BATADV_BLA_MAX_BACKBONE_GW
> limit (4096 / 256 -- generous for any real bridged LAN/VLAN
> population, several orders of magnitude below what would need to be
> sprayed to threaten memory availability). batadv_bla_add_claim() and
> batadv_bla_get_backbone_gw() reserve a slot with atomic_add_unless()
> before allocating; on cap-hit the frame is dropped (matching existing
> "drop silently, let the sender resync/backoff" BLA behaviour) instead
> of allocating. The reservation is released on every existing early-out
> (kzalloc failure, hash_add failure) and in the kref release paths
> (batadv_claim_release(), batadv_backbone_gw_release()), where the
> counters are decremented right before the objects are freed. No
> locking changes are needed: the counters are only ever touched via
> atomic ops, mirroring tp_num.
> 
> This does not change on-the-wire behaviour, hash table sizing, or
> timeout-based purging; it only stops a single on-mesh peer from
> growing the tables past a bounded ceiling.
> 
> Verified by code review rather than by driving either counter to its
> cap at runtime: the atomic_add_unless()/atomic_dec() pairing was
> checked against every existing early-out (kzalloc failure, hash_add
> failure) and against both kref release callbacks, confirming exactly
> one reservation and one release per entry, mirroring the same
> tp_num accounting in tp_meter.c. A loopback CLAIM-frame reproducer
> was used earlier to confirm the pre-fix unbounded growth itself
> (distinct claim_hash/backbone_hash entries scale linearly with the
> number of distinct (mac, vid) pairs sent), but reaching the new
> 4096 / 256 caps with that same reproducer is impractical: entries
> age out via the existing timeout-driven purge faster than a
> single-host reproducer can accumulate enough distinct pairs to hit
> the ceiling, so the cap-hit and slot-release paths were exercised
> by inspection, not by a live saturation run.
> 
> Fixes: 23721387c409 ("batman-adv: add basic bridge loop avoidance code")
> Cc: stable@vger.kernel.org
> Signed-off-by: Ibrahim Hashimov <security@auditcode.ai>
> Assisted-by: AuditCode-AI:2026.07
> ---
>  net/batman-adv/bridge_loop_avoidance.c | 38 ++++++++++++++++++++++++--
>  net/batman-adv/main.h                  |  3 ++
>  net/batman-adv/types.h                 |  6 ++++
>  3 files changed, 45 insertions(+), 2 deletions(-)


Consider this rejected:

* the target tree should be batadv and not "net"
* patch doesn't even apply to batadv.git

  $ git am 20260710165224.39411-1-security@auditcode.ai.mbx
  Applying: batman-adv: bound BLA claim and backbone gateway table growth
  error: patch failed: net/batman-adv/bridge_loop_avoidance.c:179
  error: net/batman-adv/bridge_loop_avoidance.c: patch does not apply
  Patch failed at 0001 batman-adv: bound BLA claim and backbone gateway table growth
  hint: Use 'git am --show-current-patch=diff' to see the failed patch
  hint: When you have resolved this problem, run "git am --continue".
  hint: If you prefer to skip this patch, run "git am --skip" instead.
  hint: To restore the original branch and stop patching, run "git am --abort".
  hint: Disable this message with "git config advice.mergeConflict false"

* there are already patches for these for review [1]
* breaks existing setups and doesn't allow users to adjust limits according to 
  their requirements
* own backbones must never be prevented
* I start to get tired by all these overlong AI generated patch summaries 
  which meander around completely irrelevant detail information - or 
  information which are not even relevant here

Regards,
	Sven

[1] https://patchwork.open-mesh.org/project/b.a.t.m.a.n./list/?series=726
    https://git.open-mesh.org/pub/ecsv/batman-adv.git/log/?h=b4/resource-limit

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 228 bytes --]