Re: [RFC PATCH V3 net-next 1/5] ethtool: Allow network drivers to dump arbitrary EEPROM data

On Thu, Mar 18, 2021 at 02:03:02PM +0100, Andrew Lunn wrote:

On Mon, Mar 15, 2021 at 07:12:39PM +0200, Moshe Shemesh wrote:

quoted

 
+EEPROM_DATA
+===========
+
+Fetch module EEPROM data dump.
+
+Request contents:
+
+  =====================================  ======  ==========================
+  ``ETHTOOL_A_EEPROM_DATA_HEADER``       nested  request header
+  ``ETHTOOL_A_EEPROM_DATA_OFFSET``       u32     offset within a page
+  ``ETHTOOL_A_EEPROM_DATA_LENGTH``       u32     amount of bytes to read

I wonder if offset and length should be u8. At most, we should only be
returning a 1/2 page, so 128 bytes. We don't need a u32.

There is no actual gain using NLA_U8 due to padding. Out of the
interfaces used here, kernel-userspace API is the least flexible so
I would generally prefer NLA_U32, except for bools or enumerated values
where it's absolutely obvious the number of possible values cannot grow
too much. In this case, I can't really say it's impossible we could have
devices with bigger pages in something like 20 years.

quoted

 Request translation
 ===================
 

@@ -1357,8 +1387,8 @@ are netlink only.
   ``ETHTOOL_GET_DUMP_FLAG``           n/a
   ``ETHTOOL_GET_DUMP_DATA``           n/a
   ``ETHTOOL_GET_TS_INFO``             ``ETHTOOL_MSG_TSINFO_GET``
-  ``ETHTOOL_GMODULEINFO``             n/a
-  ``ETHTOOL_GMODULEEEPROM``           n/a
+  ``ETHTOOL_GMODULEINFO``             ``ETHTOOL_MSG_MODULE_EEPROM_GET``
+  ``ETHTOOL_GMODULEEEPROM``           ``ETHTOOL_MSG_MODULE_EEPROM_GET``
   ``ETHTOOL_GEEE``                    ``ETHTOOL_MSG_EEE_GET``
   ``ETHTOOL_SEEE``                    ``ETHTOOL_MSG_EEE_SET``
   ``ETHTOOL_GRSSH``                   n/a

We should check with Michal about this. It is not a direct replacement
of the old IOCTL API, it is new API. He may want it documented
differently.

This table is meant to give a hint in the sense "for what you used
ioctl command in left column, use now netlink request in the right".
So IMHO it's appropriate. Perhaps it would deserve a comment explaining
this.

quoted

+	request->offset = nla_get_u32(tb[ETHTOOL_A_EEPROM_DATA_OFFSET]);
+	request->length = nla_get_u32(tb[ETHTOOL_A_EEPROM_DATA_LENGTH]);
+	if (tb[ETHTOOL_A_EEPROM_DATA_PAGE] &&
+	    dev->ethtool_ops->get_module_eeprom_data_by_page &&
+	    request->offset + request->length > ETH_MODULE_EEPROM_PAGE_LEN)
+		return -EINVAL;

You need to watch out for overflows here. 0xfffffff0 + 0x20 is less
than ETH_MODULE_EEPROM_PAGE_LEN when it wraps around, but will cause
bad things to happen.

BtW, the ioctl code also suffers from this problem and we recently had
a report from customer (IIRC the effect was ethtool trying to allocate
~4GB of memory), upstream fix should follow soon.

Michal