Re: [PATCH] firmware: google: update vpd_decode from upstream

From: Stephen Boyd
Date: Fri Aug 02 2019 - 18:27:59 EST


Quoting Hung-Te Lin (2019-08-02 01:20:31)
> The VPD implementation from Chromium Vital Product Data project has been
> updated so vpd_decode be easily shared by kernel, firmware and the user
> space utility programs. Also improved value range checks to prevent
> kernel crash due to bad VPD data.

Please add a Fixes: tag here to fix the commit that introduces the
problem. It would also be nice to get a description of the problem that
this patch is fixing. For example, explaining why the types change from
signed to unsigned.

>
> Signed-off-by: Hung-Te Lin <hungte@xxxxxxxxxxxx>
> ---
> drivers/firmware/google/vpd.c | 38 +++++++++------
> drivers/firmware/google/vpd_decode.c | 69 +++++++++++++++-------------
> drivers/firmware/google/vpd_decode.h | 17 ++++---
> 3 files changed, 71 insertions(+), 53 deletions(-)
>
> diff --git a/drivers/firmware/google/vpd.c b/drivers/firmware/google/vpd.c
> index 0739f3b70347..ecf217a7db39 100644
> --- a/drivers/firmware/google/vpd.c
> +++ b/drivers/firmware/google/vpd.c
> @@ -73,7 +73,7 @@ static ssize_t vpd_attrib_read(struct file *filp, struct kobject *kobp,
> * exporting them as sysfs attributes. These keys present in old firmwares are
> * ignored.
> *
> - * Returns VPD_OK for a valid key name, VPD_FAIL otherwise.
> + * Returns VPD_DECODE_OK for a valid key name, VPD_DECODE_FAIL otherwise.

Maybe we should convert these things to use linux conventions instead of
VPD error codes?

> *
> * @key: The key name to check
> * @key_len: key name length
> @@ -86,14 +86,14 @@ static int vpd_section_check_key_name(const u8 *key, s32 key_len)
> c = *key++;
>
> if (!isalnum(c) && c != '_')
> - return VPD_FAIL;
> + return VPD_DECODE_FAIL;
> }
>
> - return VPD_OK;
> + return VPD_DECODE_OK;

Can you split this rename out into it's own patch. That way we can
confirm that there are no changes due to the rename of the enum.

> }
>
> -static int vpd_section_attrib_add(const u8 *key, s32 key_len,
> - const u8 *value, s32 value_len,
> +static int vpd_section_attrib_add(const u8 *key, u32 key_len,
> + const u8 *value, u32 value_len,
> void *arg)
> {
> int ret;
> @@ -246,7 +246,7 @@ static int vpd_section_destroy(struct vpd_section *sec)
>
> static int vpd_sections_init(phys_addr_t physaddr)
> {
> - struct vpd_cbmem *temp;
> + struct vpd_cbmem __iomem *temp;
> struct vpd_cbmem header;
> int ret = 0;
>
> @@ -254,7 +254,7 @@ static int vpd_sections_init(phys_addr_t physaddr)
> if (!temp)
> return -ENOMEM;
>
> - memcpy(&header, temp, sizeof(struct vpd_cbmem));
> + memcpy_fromio(&header, temp, sizeof(struct vpd_cbmem));
> memunmap(temp);
>
> if (header.magic != VPD_CBMEM_MAGIC)
> @@ -316,7 +316,19 @@ static struct coreboot_driver vpd_driver = {
> },
> .tag = CB_TAG_VPD,
> };
> -module_coreboot_driver(vpd_driver);
> +
> +static int __init coreboot_vpd_init(void)
> +{
> + return coreboot_driver_register(&vpd_driver);
> +}
> +
> +static void __exit coreboot_vpd_exit(void)
> +{
> + coreboot_driver_unregister(&vpd_driver);
> +}
> +
> +module_init(coreboot_vpd_init);
> +module_exit(coreboot_vpd_exit);
>
> MODULE_AUTHOR("Google, Inc.");
> MODULE_LICENSE("GPL");

The above three hunks should be dropped. They're undoing other patches
that have gone upstream.

> diff --git a/drivers/firmware/google/vpd_decode.c b/drivers/firmware/google/vpd_decode.c
> index 92e3258552fc..5531770e3d58 100644
> --- a/drivers/firmware/google/vpd_decode.c
> +++ b/drivers/firmware/google/vpd_decode.c
> @@ -9,19 +9,19 @@
>
> #include "vpd_decode.h"
>
> -static int vpd_decode_len(const s32 max_len, const u8 *in,
> - s32 *length, s32 *decoded_len)
> +static int vpd_decode_len(const u32 max_len, const u8 *in, u32 *length,

Is there a reason why max_len and length changes to be unsigned?
Presumably to fix something.

> + u32 *decoded_len)
> {
> u8 more;
> int i = 0;
>
> if (!length || !decoded_len)
> - return VPD_FAIL;
> + return VPD_DECODE_FAIL;
>
> *length = 0;
> do {
> if (i >= max_len)
> - return VPD_FAIL;
> + return VPD_DECODE_FAIL;
>
> more = in[i] & 0x80;
> *length <<= 7;
> @@ -30,24 +30,43 @@ static int vpd_decode_len(const s32 max_len, const u8 *in,
> } while (more);
>
> *decoded_len = i;
> + return VPD_DECODE_OK;
> +}
> +
> +static int vpd_decode_entry(const u32 max_len, const u8 *input_buf,
> + u32 *consumed, const u8 **entry, u32 *entry_len)
> +{
> + u32 decoded_len;
> +
> + if (vpd_decode_len(max_len - *consumed, &input_buf[*consumed],

Can you add a local variable for *consumed? So _consumed is passed in
and then u32 consume = *_consumed.

> + entry_len, &decoded_len) != VPD_DECODE_OK)
> + return VPD_DECODE_FAIL;
> + if (max_len - *consumed < decoded_len)
> + return VPD_DECODE_FAIL;
>
> - return VPD_OK;
> + *consumed += decoded_len;
> + *entry = input_buf + *consumed;
> +
> + /* entry_len is untrusted data and must be checked again. */
> + if (max_len - *consumed < *entry_len)
> + return VPD_DECODE_FAIL;

Is consumed supposed to have move forward here on failure? Is entry
supposed to point to something, or should it be pointing to NULL on a
failure?

> +
> + *consumed += *entry_len;
> + return VPD_DECODE_OK;
> }
>
> -int vpd_decode_string(const s32 max_len, const u8 *input_buf, s32 *consumed,
> +int vpd_decode_string(const u32 max_len, const u8 *input_buf, u32 *consumed,
> vpd_decode_callback callback, void *callback_arg)
> {
> int type;
> - int res;
> - s32 key_len;
> - s32 value_len;
> - s32 decoded_len;
> + u32 key_len;
> + u32 value_len;
> const u8 *key;
> const u8 *value;
>
> /* type */
> if (*consumed >= max_len)
> - return VPD_FAIL;
> + return VPD_DECODE_FAIL;
>
> type = input_buf[*consumed];
>
> @@ -56,25 +75,13 @@ int vpd_decode_string(const s32 max_len, const u8 *input_buf, s32 *consumed,
> case VPD_TYPE_STRING:
> (*consumed)++;
>
> - /* key */
> - res = vpd_decode_len(max_len - *consumed, &input_buf[*consumed],
> - &key_len, &decoded_len);
> - if (res != VPD_OK || *consumed + decoded_len >= max_len)
> - return VPD_FAIL;
> -
> - *consumed += decoded_len;
> - key = &input_buf[*consumed];
> - *consumed += key_len;
> -
> - /* value */
> - res = vpd_decode_len(max_len - *consumed, &input_buf[*consumed],
> - &value_len, &decoded_len);
> - if (res != VPD_OK || *consumed + decoded_len > max_len)
> - return VPD_FAIL;
> + if (vpd_decode_entry(max_len, input_buf, consumed, &key,
> + &key_len) != VPD_DECODE_OK)
> + return VPD_DECODE_FAIL;
>
> - *consumed += decoded_len;
> - value = &input_buf[*consumed];
> - *consumed += value_len;
> + if (vpd_decode_entry(max_len, input_buf, consumed, &value,
> + &value_len) != VPD_DECODE_OK)
> + return VPD_DECODE_FAIL;
>
> if (type == VPD_TYPE_STRING)
> return callback(key, key_len, value, value_len,
> @@ -82,8 +89,8 @@ int vpd_decode_string(const s32 max_len, const u8 *input_buf, s32 *consumed,
> break;
>
> default:
> - return VPD_FAIL;
> + return VPD_DECODE_FAIL;
> }
>
> - return VPD_OK;
> + return VPD_DECODE_OK;
> }
> diff --git a/drivers/firmware/google/vpd_decode.h b/drivers/firmware/google/vpd_decode.h
> index cf8c2ace155a..4113ac2f4a70 100644
> --- a/drivers/firmware/google/vpd_decode.h
> +++ b/drivers/firmware/google/vpd_decode.h
> @@ -13,28 +13,27 @@
> #include <linux/types.h>
>
> enum {
> - VPD_OK = 0,
> - VPD_FAIL,
> + VPD_DECODE_OK = 0,
> + VPD_DECODE_FAIL = 1,

I wonder why this is an enum vs. just using the typical kernel error
codes from errno.h.

> };
>
> enum {
> VPD_TYPE_TERMINATOR = 0,
> VPD_TYPE_STRING,
> - VPD_TYPE_INFO = 0xfe,
> + VPD_TYPE_INFO = 0xfe,
> VPD_TYPE_IMPLICIT_TERMINATOR = 0xff,
> };

Please drop this change, it's just unaligning things.

>
> /* Callback for vpd_decode_string to invoke. */
> -typedef int vpd_decode_callback(const u8 *key, s32 key_len,
> - const u8 *value, s32 value_len,
> - void *arg);
> +typedef int vpd_decode_callback(const u8 *key, u32 key_len, const u8 *value,
> + u32 value_len, void *arg);
>
> /*
> * vpd_decode_string
> *
> * Given the encoded string, this function invokes callback with extracted
> - * (key, value). The *consumed will be plused the number of bytes consumed in
> - * this function.
> + * (key, value). The *consumed will be incremented by the number of bytes

Use @consumed to refer to the argument. Split this out into a different
patch as it's just updating the documentation to say the same thing.

> + * consumed in this function.
> *
> * The input_buf points to the first byte of the input buffer.
> *