[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <a5e2570f-222b-5633-af67-6cf5650a741f@canonical.com>
Date: Fri, 29 Nov 2019 09:03:50 +0000
From: Colin Ian King <colin.king@...onical.com>
To: Jeremy Kerr <jk@...abs.org>, Joel Stanley <joel@....id.au>,
Alistar Popple <alistair@...ple.id.au>,
Eddie James <eajames@...ux.ibm.com>,
Benjamin Herrenschmidt <benh@...nel.crashing.org>,
linux-fsi@...ts.ozlabs.org
Cc: kernel-janitors@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] [PATCH v2] fsi: fix bogus error returns from cfam_read
and cfam_write
On 29/11/2019 03:24, Jeremy Kerr wrote:
> Based on a static analysis report and original patch from Colin Ian King
> <colin.king@...onical.com>.
>
> Currently, we may drop error values from cfam_read and cfam_write. This
> change returns the actual error on failure, but a partial read/write will
> take precedence.
>
> Addresses-Coverity: ("Unused value")
> Fixes: d1dcd6782576 ("fsi: Add cfam char devices")
> Reported-by: Colin Ian King <colin.king@...onical.com>
> Signed-off-by: Jeremy Kerr <jk@...abs.org>
>
> ---
> Colin: thanks for the report and patch. I think this is a more complete
> fix, as we want to preseve any partial read/write status if a failure
> happens mid-way through an operation. Let me know if you (or the
> coverity analysis) have any feedback.
Looks good to me. Thanks Jeremy.
>
> ---
>
> drivers/fsi/fsi-core.c | 32 ++++++++++++++++++++++----------
> 1 file changed, 22 insertions(+), 10 deletions(-)
>
> diff --git a/drivers/fsi/fsi-core.c b/drivers/fsi/fsi-core.c
> index 71c6f9fef648..3158a78c2e94 100644
> --- a/drivers/fsi/fsi-core.c
> +++ b/drivers/fsi/fsi-core.c
> @@ -699,6 +699,8 @@ static ssize_t cfam_read(struct file *filep, char __user *buf, size_t count,
> if (off > 0xffffffff || count > 0xffffffff || off + count > 0xffffffff)
> return -EINVAL;
>
> + rc = 0;
> +
> for (total_len = 0; total_len < count; total_len += read_len) {
> __be32 data;
>
> @@ -707,18 +709,22 @@ static ssize_t cfam_read(struct file *filep, char __user *buf, size_t count,
>
> rc = fsi_slave_read(slave, off, &data, read_len);
> if (rc)
> - goto fail;
> + break;
> rc = copy_to_user(buf + total_len, &data, read_len);
> if (rc) {
> rc = -EFAULT;
> - goto fail;
> + break;
> }
> off += read_len;
> }
> - rc = count;
> - fail:
> +
> + /* if we've read any data, we want that to be returned in
> + * preference to an error state */
> + if (total_len)
> + rc = total_len;
> +
> *offset = off;
> - return count;
> + return rc;
> }
>
> static ssize_t cfam_write(struct file *filep, const char __user *buf,
> @@ -736,6 +742,8 @@ static ssize_t cfam_write(struct file *filep, const char __user *buf,
> if (off > 0xffffffff || count > 0xffffffff || off + count > 0xffffffff)
> return -EINVAL;
>
> + rc = 0;
> +
> for (total_len = 0; total_len < count; total_len += write_len) {
> __be32 data;
>
> @@ -745,17 +753,21 @@ static ssize_t cfam_write(struct file *filep, const char __user *buf,
> rc = copy_from_user(&data, buf + total_len, write_len);
> if (rc) {
> rc = -EFAULT;
> - goto fail;
> + break;
> }
> rc = fsi_slave_write(slave, off, &data, write_len);
> if (rc)
> - goto fail;
> + break;
> off += write_len;
> }
> - rc = count;
> - fail:
> +
> + /* if we've written any data, we want to indicate that partial write
> + * instead of any mid-stream error */
> + if (total_len)
> + rc = total_len;
> +
> *offset = off;
> - return count;
> + return rc;
> }
>
> static loff_t cfam_llseek(struct file *file, loff_t offset, int whence)
>
Powered by blists - more mailing lists