Re: [PATCH net-next v4 02/15] gve: refactor initialization with helper functions

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 3 potential issue(s) to consider:

Pre-existing issues:
- [Critical] Disabling XDP triggers an out-of-bounds array access due to the old BPF program remaining active while the TX arrays are shrunk.
- [High] Device resets permanently break AF_XDP sockets because the driver zeroes the `xsk_pools` bitmask.
- [High] `gve_xsk_pool_enable` causes a data race by modifying the XDP memory model concurrently with the NAPI polling context.
--

--- Patch [2]: [PATCH net-next v4 02/15] gve: refactor initialization with helper functions ---
commit f4310d54d17a7a878691c693aea2b3ce23d39802
Author: Harshitha Ramamurthy <hramamurthy@google.com>

gve: refactor initialization with helper functions

In the interest of commonizing code, refactor gve_probe()
and gve_init_priv() with a few helper functions that can
be expanded and utilized in upcoming patches that add the
mailbox ABI to the driver.

> diff --git a/drivers/net/ethernet/google/gve/gve_main.c b/drivers/net/ethernet/google/gve/gve_main.c
> index 30bf6df4ebc58..37e6205e2eefd 100644
> --- a/drivers/net/ethernet/google/gve/gve_main.c
> +++ b/drivers/net/ethernet/google/gve/gve_main.c
> @@ -2400,7 +2400,6 @@ static const struct xdp_metadata_ops gve_xdp_metadata_ops = {
>  
>  static int gve_init_priv(struct gve_priv *priv, bool skip_describe_device)
>  {

[Severity: High]
This is a pre-existing issue, but could device resets permanently break AF_XDP
sockets because the driver zeroes the xsk_pools bitmask?

When a device reset occurs (e.g., due to timeout, adminq failure, or MTU
change), it triggers gve_teardown_priv_resources() which frees the bitmask:

    bitmap_free(priv->xsk_pools);
    priv->xsk_pools = NULL;

During recovery, the initialization path reallocates a fresh zeroed
xsk_pools bitmap (e.g., via priv->xsk_pools = bitmap_zalloc(...)),
permanently losing the previous state. Because the driver never restores the
xsk_pools state from the core net_device, AF_XDP sockets silently stop
working after a reset.

[Severity: Critical]
This is a pre-existing issue, but is there a risk of an out-of-bounds
array access when disabling XDP?

In gve_set_xdp(), the xdp_prog pointer is updated after the queues have been
restarted with a shrunken TX array:

    err = gve_configure_rings_xdp(priv, 0);
    if (err)
        goto out;

    WRITE_ONCE(priv->xdp_prog, prog);

If a local privileged user disables XDP, gve_set_xdp() calls
gve_configure_rings_xdp(priv, 0), which reallocates priv->tx without XDP
queues and restarts NAPI polling. Since priv->xdp_prog is updated only after
NAPI restarts, the active NAPI poll can execute the old BPF program. If the
program returns XDP_TX, it accesses priv->tx[tx_qid] where tx_qid exceeds
the new, smaller array size, leading to memory corruption.

[Severity: High]
This is a pre-existing issue, but does gve_xsk_pool_enable() cause a data
race by modifying the XDP memory model concurrently with the NAPI polling
context?

In gve_xsk_pool_enable(), it registers an AF_XDP pool by modifying the XDP
memory model and rx->xsk_pool without first disabling NAPI polling:

    err = xdp_rxq_info_reg_mem_model(&rx->xdp_rxq, MEM_TYPE_XSK_BUFF_POOL,
                                     pool);
    if (err) {
        ...
    }

    rx->xsk_pool = pool;

If NAPI receives a packet in the race window, it may initialize the XDP
buffer with MEM_TYPE_XSK_BUFF_POOL but allocate memory from the standard page
pool because rx->xsk_pool was observed as NULL. When freed, xsk_buff_free()
would be incorrectly called on a standard page, causing type confusion and
memory corruption.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260814021406.3044324-1-hramamurthy@google.com?part=2