Re: [PATCH v2 net] nfp: cast sizeof() to int when comparing with error code

From: Julia Lawall
Date: Tue Jun 26 2018 - 03:22:00 EST




On Mon, 25 Jun 2018, Joe Perches wrote:

> On Tue, 2018-06-26 at 09:16 +0800, Chengguang Xu wrote:
> > sizeof() will return unsigned value so in the error check
> > negative error code will be always larger than sizeof().
>
> This looks like a general class of error in the kernel
> where a signed result that could be returning a -errno
> is tested against < or <= sizeof()
>
> A couple examples:
>
> drivers/input/mouse/elan_i2c_smbus.c:
>
> len = i2c_smbus_read_block_data(client,
> ETP_SMBUS_IAP_PASSWORD_READ,
> val);
> if (len < sizeof(u16)) {
>
> i2c_smbus_read_block_data can return a negative errno
>
>
> net/smc/smc_clc.c:
>
> len = kernel_sendmsg(smc->clcsock, &msg, &vec, 1,
> sizeof(struct smc_clc_msg_decline));
> if (len < sizeof(struct smc_clc_msg_decline))
>
> where kernel_sendmsg can return a negative errno
>
> There are probably others, I didn't look hard.
>
> Perhaps a cocci script to find these could be generated?

Currently there is a rule for comparison of unsigneds to 0. It would be
reasonable to extend it for sizes. I will see what it gives.

julia