out of bounds writes in net/hsr/
Subsystems (2)
I found this in coverity, and I think it's a real bug..
hsr_register_frame_in does a check that dev_idx is between 0 and 2, therefore, a dev_idx of 2 is possible when it gets to the array writes at the end of the function. The arrays are defined such..
26 struct node_entry {
...
33 unsigned long time_in[HSR_MAX_SLAVE]; 34 bool time_in_stale[HSR_MAX_SLAVE];
and HSR_MAX_SLAVE is...
139 enum hsr_dev_idx {
140 HSR_DEV_NONE = -1, 141 HSR_DEV_SLAVE_A = 0, 142 HSR_DEV_SLAVE_B, 143 HSR_DEV_MASTER,
144 };
145 #define HSR_MAX_SLAVE (HSR_DEV_SLAVE_B + 1)
So we have arrays of 2 bytes, and we can try to write to the 3rd byte.
The problem seems to be that the checking in hsr_register_frame is on HSR_MAX_DEV which is defined as..
#define HSR_MAX_DEV (HSR_DEV_MASTER + 1)
The + 1 seems odd, and looking at the other uses of HSR_MAX_DEV, I can't figure out why it's there.
Dave
Quoted 5 lines · Dave Jones
I found this in coverity, and I think it's a real bug..
hsr_register_frame_indoes a check thatdev_idxis between 0 and 2, therefore, adev_idxof 2 is possible when it gets to the array writes at the end of the function.
Thanks for finding this; it is a bug (although I don't think it has actually lead to any out of bound accesses).
However, I think you are a bit late - I believe this was fixed in a patch from Dan Carpenter just a few days ago. See
http://www.spinics.net/lists/netdev/msg272815.html
Quoted 7 lines
#define
HSR_MAX_DEV(HSR_DEV_MASTER+ 1)The + 1 seems odd, and looking at the other uses of
HSR_MAX_DEV, I can't figure out why it's there.Dave
Yes, maybe the names are a bit misleading, and they should be called something like HSR_DEVS and HSR_SLAVES instead. I.e.:
some-type array-name[HSR_DEVS];
... where the last element is accessed by array-name[HSR_MAX_DEV].
Signature · 3 lines
Arvid Brodin | Consultant (Linux) ALTEN | Knarrarnäsgatan 7 | SE-164 40 Kista | Sweden [email protected] | www.alten.se/en/
Quoted 7 lines · Arvid Brodin
Quoted 5 lines · Dave Jones
I found this in coverity, and I think it's a real bug..
hsr_register_frame_indoes a check thatdev_idxis between 0 and 2, therefore, adev_idxof 2 is possible when it gets to the array writes at the end of the function.
Quoted 7 lines
Thanks for finding this; it is a bug (although I don't think it has actually lead to any out of bound accesses).
However, I think you are a bit late - I believe this was fixed in a patch from Dan Carpenter just a few days ago. See
excellent, thanks for checking.
Dave