On Mon, Feb 7, 2022 at 11:15 PM Dan Carpenter <dan.carpenter@xxxxxxxxxx> wrote:
>
> Hello Eric Dumazet,
>
> The patch 28f922213886: "net/smc: fix ref_tracker issue in
> smc_pnet_add()" from Feb 5, 2022, leads to the following Smatch
> static checker warning:
>
> net/smc/smc_pnet.c:384 smc_pnet_add_eth()
> warn: sleeping in atomic context
>
> net/smc/smc_pnet.c
> 342 static int smc_pnet_add_eth(struct smc_pnettable *pnettable, struct net *net,
> 343 char *eth_name, char *pnet_name)
> 344 {
> 345 struct smc_pnetentry *tmp_pe, *new_pe;
> 346 struct net_device *ndev, *base_ndev;
> 347 u8 ndev_pnetid[SMC_MAX_PNETID_LEN];
> 348 bool new_netdev;
> 349 int rc;
> 350
> 351 /* check if (base) netdev already has a pnetid. If there is one, we do
> 352 * not want to add a pnet table entry
> 353 */
> 354 rc = -EEXIST;
> 355 ndev = dev_get_by_name(net, eth_name); /* dev_hold() */
> 356 if (ndev) {
> 357 base_ndev = pnet_find_base_ndev(ndev);
> 358 if (!smc_pnetid_by_dev_port(base_ndev->dev.parent,
> 359 base_ndev->dev_port, ndev_pnetid))
> 360 goto out_put;
> 361 }
> 362
> 363 /* add a new netdev entry to the pnet table if there isn't one */
> 364 rc = -ENOMEM;
> 365 new_pe = kzalloc(sizeof(*new_pe), GFP_KERNEL);
> 366 if (!new_pe)
> 367 goto out_put;
> 368 new_pe->type = SMC_PNET_ETH;
> 369 memcpy(new_pe->pnet_name, pnet_name, SMC_MAX_PNETID_LEN);
> 370 strncpy(new_pe->eth_name, eth_name, IFNAMSIZ);
> 371 rc = -EEXIST;
> 372 new_netdev = true;
> 373 write_lock(&pnettable->lock);
> 374 list_for_each_entry(tmp_pe, &pnettable->pnetlist, list) {
> 375 if (tmp_pe->type == SMC_PNET_ETH &&
> 376 !strncmp(tmp_pe->eth_name, eth_name, IFNAMSIZ)) {
> 377 new_netdev = false;
> 378 break;
> 379 }
> 380 }
> 381 if (new_netdev) {
> 382 if (ndev) {
> 383 new_pe->ndev = ndev;
> --> 384 netdev_tracker_alloc(ndev, &new_pe->dev_tracker,
> 385 GFP_KERNEL);
>
> We can't do a GFP_KERNEL allocation here
>
> 386 }
> 387 list_add_tail(&new_pe->list, &pnettable->pnetlist);
> 388 write_unlock(&pnettable->lock);
>
> because we're holding this write lock
>
Sure, the fix is already in net tree, and should reach your tree eventually.
Thanks.
commit 94fdd7c02a56d0316d20e417a1141b71a8dcee82 (HEAD -> net-1, net/master)
Author: Eric Dumazet <edumazet@xxxxxxxxxx>
Date: Sun Feb 6 06:33:48 2022 -0800
net/smc: use GFP_ATOMIC allocation in smc_pnet_add_eth()
My last patch moved the netdev_tracker_alloc() call to a section
protected by a write_lock().
I should have replaced GFP_KERNEL with GFP_ATOMIC to avoid the infamous:
BUG: sleeping function called from invalid context at
include/linux/sched/mm.h:256
Fixes: 28f922213886 ("net/smc: fix ref_tracker issue in smc_pnet_add()")
Signed-off-by: Eric Dumazet <edumazet@xxxxxxxxxx>
Signed-off-by: David S. Miller <davem@xxxxxxxxxxxxx>
diff --git a/net/smc/smc_pnet.c b/net/smc/smc_pnet.c
index fb6331d97185a5db9b4539e7f081e9fa469bc44b..0599246c037690b4b01813956e4af74519277bea
100644
--- a/net/smc/smc_pnet.c
+++ b/net/smc/smc_pnet.c
@@ -382,7 +382,7 @@ static int smc_pnet_add_eth(struct smc_pnettable
*pnettable, struct net *net,
if (ndev) {
new_pe->ndev = ndev;
netdev_tracker_alloc(ndev, &new_pe->dev_tracker,
- GFP_KERNEL);
+ GFP_ATOMIC);
}
list_add_tail(&new_pe->list, &pnettable->pnetlist);
write_unlock(&pnettable->lock);
