Re: [PATCH ghak124 v3fix] audit: add gfp parameter to audit_log_nfcfg

From: Jones Desougi
Date: Fri Jul 03 2020 - 08:41:12 EST


Doesn't seem entirely consistent now either though. Two cases below.

/Jones

On Sun, Jun 28, 2020 at 5:27 AM Richard Guy Briggs <rgb@xxxxxxxxxx> wrote:
>
> Fixed an inconsistent use of GFP flags in nft_obj_notify() that used
> GFP_KERNEL when a GFP flag was passed in to that function. Given this
> allocated memory was then used in audit_log_nfcfg() it led to an audit
> of all other GFP allocations in net/netfilter/nf_tables_api.c and a
> modification of audit_log_nfcfg() to accept a GFP parameter.
>
> Reported-by: Dan Carptenter <dan.carpenter@xxxxxxxxxx>
> Signed-off-by: Richard Guy Briggs <rgb@xxxxxxxxxx>
> ---
> Passes audit-testsuite.
>
> include/linux/audit.h | 8 ++++----
> kernel/auditsc.c | 4 ++--
> net/bridge/netfilter/ebtables.c | 6 +++---
> net/netfilter/nf_tables_api.c | 33 +++++++++++++++++++++------------
> net/netfilter/x_tables.c | 5 +++--
> 5 files changed, 33 insertions(+), 23 deletions(-)
>
...
> diff --git a/net/netfilter/nf_tables_api.c b/net/netfilter/nf_tables_api.c
> index 164700273947..f7ff91479647 100644
> --- a/net/netfilter/nf_tables_api.c
> +++ b/net/netfilter/nf_tables_api.c
...
> @@ -6071,13 +6077,14 @@ static int nf_tables_getobj(struct net *net, struct sock *nlsk,
> reset = true;
>
> if (reset) {
> - char *buf = kasprintf(GFP_KERNEL, "%s:%llu;?:0",
> + char *buf = kasprintf(GFP_ATOMIC, "%s:%llu;?:0",
> table->name, table->handle);
>
> audit_log_nfcfg(buf,
> family,
> obj->handle,
> - AUDIT_NFT_OP_OBJ_RESET);
> + AUDIT_NFT_OP_OBJ_RESET,
> + GFP_KERNEL);
> kfree(buf);
> }
>

Replaces one GFP_KERNEL (with GFP_ATOMIC) but also adds a new one in
the following statement.
Is that intentional?

> @@ -6156,7 +6163,7 @@ void nft_obj_notify(struct net *net, const struct nft_table *table,
> {
> struct sk_buff *skb;
> int err;
> - char *buf = kasprintf(GFP_KERNEL, "%s:%llu;?:0",
> + char *buf = kasprintf(gfp, "%s:%llu;?:0",
> table->name, table->handle);
>
> audit_log_nfcfg(buf,
> @@ -6164,7 +6171,8 @@ void nft_obj_notify(struct net *net, const struct nft_table *table,
> obj->handle,
> event == NFT_MSG_NEWOBJ ?
> AUDIT_NFT_OP_OBJ_REGISTER :
> - AUDIT_NFT_OP_OBJ_UNREGISTER);
> + AUDIT_NFT_OP_OBJ_UNREGISTER,
> + GFP_KERNEL);
> kfree(buf);
>
> if (!report &&

It would seem these two hunks form a similar discrepancy.

...

> --
> 1.8.3.1
>