Hi David, On 5/14/22 16:32, David Kahurani wrote:
Reads that are lesser than the requested size lead to uninit-value bugs.
In this particular case a variable which was supposed to be initialized
after a read is left uninitialized after a partial read.
Qualify such reads as errors and handle them correctly and while at it
convert the reader functions to return zero on success for easier error
handling.
Fixes: e2ca90c276e1 ("ax88179_178a: ASIX AX88179_178A USB 3.0/2.0 to gigabit ethernet adapter driver")
Signed-off-by: David Kahurani <k.kahurani@xxxxxxxxx>
Reported-and-tested-by: syzbot+d3dbdf31fbe9d8f5f311@xxxxxxxxxxxxxxxxxxxxxxxxx
---
<--- here (*)
drivers/net/usb/ax88179_178a.c | 281 ++++++++++++++++++++++++++------- 1 file changed, 227 insertions(+), 54 deletions(-)
I don't see any error in that patch, but I had to find previous versions of that patch in my inbox.
Usually new versions of single patches are linked in one thread and have a version number in a title. You can generate patch with version using -v option of git format-patch like:
$ git format-patch -v2 HEAD~And you can send new version as reply using --in-reply= option of git send-email. It helps a lot with finding previous version, since all version are linked in one thread
And all updates from version to version should be put under --- (*), since it's hard to remember why previous version was rejected.
jtimeout = jiffies + delay; do { - ax88179_read_cmd(dev, AX_ACCESS_MAC, AX_SROM_CMD, - 1, 1, &buf); + ret = ax88179_read_cmd(dev, AX_ACCESS_MAC, AX_SROM_CMD, + 1, 1, &buf); + if (ret) { + netdev_dbg(dev->net, + "Failed to read SROM_CMD: %d\n", + ret); + return ret; + }if (time_after(jiffies, jtimeout))return -EINVAL;} while (buf & EEP_BUSY);
I think, this change might be dangerous. Maybe it should be done in the same way as in asix driver [1]. Code polls for some register after a write and maybe non-fatal read error might occur here.
Just my thoughts, I don't know anything about that device :)
+ ret = ax88179_read_cmd(dev, AX_ACCESS_MAC, AX_NODE_ID, ETH_ALEN, + ETH_ALEN, mac); + + if (ret) + netdev_dbg(dev->net, "Failed to read NODE_ID: %d", ret); + else + netif_dbg(dev, ifup, dev->net, + "MAC address read from ASIX chip");
Maybe also use `netif_dbg` here?... There should be a reason why it was used here in the first place. Or should not :)
Anyway, if someone will say that bailing out from while loop on any error is OK feel free to add
Reviewed-by: Pavel Skripkin <paskripkin@xxxxxxxxx>[1] https://elixir.bootlin.com/linux/latest/source/drivers/net/usb/asix_common.c#L78
With regards, Pavel Skripkin
Attachment:
OpenPGP_signature
Description: OpenPGP digital signature
