hostap_plx: fix CIS verification
Subsystems (3)
PATCH 1/1 hostap_plx: fix CIS verification
hostap_plx: fix CIS verification
The record length for numerical manufacturer ID should be at least 4 bytes (two 16-bit words). The code required 5 bytes, which would break for most (if not all) cards. Reported by [email protected]
---
This is an small fix with significant impact. hostap_plx is currently broken, and this patch fixes it. Please forward this patch to 2.6.x.y kernels.
drivers/net/wireless/hostap/hostap_plx.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)
drivers/net/wireless/hostap/hostap_plx.c +1 -1
@@ -391,7 +391,7 @@ #define CIS_MAX_LEN 256break ;case CISTPL_MANFID : -if (cis[pos +1 ] <5 ) +if (cis[pos +1 ] <4 ) goto cis_error; manfid1 = cis[pos +2 ] + (cis[pos +3 ] <<8 ); manfid2 = cis[pos +4 ] + (cis[pos +5 ] <<8 );
Signature · 2 lines
Regards, Pavel Roskin
drivers/net/wireless/hostap/hostap_plx.c #define CIS_MAX_LEN 256 · line 394
Quoted 7 lines · Pavel Roskin
The record length for numerical manufacturer ID should be at least 4 bytes (two 16-bit words). The code required 5 bytes, which would break for most (if not all) cards. Reported by [email protected]
drivers/net/wireless/hostap/hostap_plx.c +1
-1case CISTPL_MANFID : -if (cis[pos +1 ] <5 ) +if (cis[pos +1 ] <4 )
Hmm.. Interesting. I think this was changed from 4 to 5 due to a potential buffer overflow as reported by Coverity tools.. In addition, I think that I spent long time trying to understand why it could be a buffer overflow and since it was changed, likely finally figured out an example case.. Alas, I don't remember what exactly this was anymore.
It looks like the comparison of the length field to be <5 was incorrect, but in order to avoid re-introducing any potential buffer overflows, that condition could be extended to verify that pos is small enough.. Something like (cis[pos + 1] < 4 && pos + 5 < CIS_MAX_LEN) could be a better fix here. I don't have easy access to PLX cards anymore, so this is untested and I'm too lazy to copy this function into a separate program to run it against CIS dumps.
Signature · 1 line
Jouni Malinen PGP id EFC895FA
Quoted 16 lines · Jouni Malinen
drivers/net/wireless/hostap/hostap_plx.c #define CIS_MAX_LEN 256 · line 394
Quoted 7 lines · Pavel Roskin
The record length for numerical manufacturer ID should be at least 4 bytes (two 16-bit words). The code required 5 bytes, which would break for most (if not all) cards. Reported by [email protected]
drivers/net/wireless/hostap/hostap_plx.c +1
-1case CISTPL_MANFID : -if (cis[pos +1 ] <5 ) +if (cis[pos +1 ] <4 )Hmm.. Interesting. I think this was changed from 4 to 5 due to a potential buffer overflow as reported by Coverity tools.. In addition, I think that I spent long time trying to understand why it could be a buffer overflow and since it was changed, likely finally figured out an example case.. Alas, I don't remember what exactly this was anymore.
Quoted 7 lines
It looks like the comparison of the length field to be <5 was incorrect, but in order to avoid re-introducing any potential buffer overflows, that condition could be extended to verify that pos is small enough.. Something like (cis[pos + 1] < 4 && pos + 5 <
CIS_MAX_LEN) could be a better fix here. I don't have easy access to PLX cards anymore, so this is untested and I'm too lazy to copy this function into a separate program to run it against CIS dumps.
Pavel,
Will you be refactoring this patch? Or do you disagree with Jouni's assessment?
Thanks,
John
Quoted 2 lines · John W. Linville
Will you be refactoring this patch? Or do you disagree with Jouni's assessment?
OK, give me an hour to produce a better patch. My patch has an advantage of being simple and of fixing exactly one thing, but if Jouni feels more comfortable with an additional check, I'll add it.
I don't have Coverity to check, and I think the results of Coverity were misinterpreted. It doesn't know anything about CIS structure. Anyway, let me just check Jouni's suggestion on a real PLX card.
Signature · 2 lines
Regards, Pavel Roskin
Quoted 3 lines · Pavel Roskin
I don't have Coverity to check, and I think the results of Coverity were misinterpreted. It doesn't know anything about CIS structure. Anyway, let me just check Jouni's suggestion on a real PLX card.
It doesn't need to know anything about the CIS structure. The driver must not be allowed to run over a buffer even if given an invalid CIS data.
Signature · 1 line
Jouni Malinen PGP id EFC895FA
Hello, Jouni!
Quoted 6 lines · Jouni Malinen
Quoted 3 lines · Pavel Roskin
The record length for numerical manufacturer ID should be at least 4 bytes (two 16-bit words). The code required 5 bytes, which would break for most (if not all) cards. Reported by [email protected]
Quoted 9 lines
Review
Quoted 3 lines
(quoted diff) +1
-1case CISTPL_MANFID: - if (cis[pos + 1] < 5) + if (cis[pos + 1] < 4)Hmm.. Interesting. I think this was changed from 4 to 5 due to a potential buffer overflow as reported by Coverity tools.. In addition, I think that I spent long time trying to understand why it could be a buffer overflow and since it was changed, likely finally figured out an example case.. Alas, I don't remember what exactly this was anymore.
Coverity has no means to interpret CIS. However, it may understand kmalloc, which allocates CIS_MAX_LEN for the CIS copy.
The value of cis[pos + 1] has no bearing on the validity of the access to cis[pos + 5] from the point of view of a checking tool.
Quoted 3 lines
It looks like the comparison of the length field to be <5 was incorrect, but in order to avoid re-introducing any potential buffer overflows, that condition could be extended to verify that pos is small enough..
pos is already checked in the beginning of the loop to be small enough, but the check is not strong enough. The next tuple starts at (pos + cis[pos + 1] + 2), and we want that to be at most CIS_MAX_LEN.
That's something Coverity could have found.
So, the right fix would be:
drivers/net/wireless/hostap/hostap_plx.c +1 -1
@@ -364,7 +364,7 @@ #define CIS_MAX_LEN 256 pos =0 ;while (pos <CIS_MAX_LEN -1 && cis[pos] !=CISTPL_END ) { -if (pos + cis[pos +1 ] >=CIS_MAX_LEN ) +if (pos +2 + cis[pos +1 ] >CIS_MAX_LEN ) goto cis_error;switch (cis[pos]) {
I'm rewriting this with "<" because it's easier to read. Next tuple at exactly CIS_MEX_LEN is fine - we just stop precessing at that point.
I'm going to combine this with my previous fix and resend.
Signature · 2 lines
Regards, Pavel Roskin
Quoted 5 lines · Pavel Roskin
Coverity has no means to interpret CIS. However, it may understand kmalloc, which allocates
CIS_MAX_LENfor the CIS copy.The value of cis[pos + 1] has no bearing on the validity of the access to cis[pos + 5] from the point of view of a checking tool.
Indeed.
Quoted 5 lines
pos is already checked in the beginning of the loop to be small enough, but the check is not strong enough. The next tuple starts at (pos + cis[pos + 1] + 2), and we want that to be at most
CIS_MAX_LEN.That's something Coverity could have found.
Agreed and I went through the report at some point and I think I found it to be valid..
Quoted 9 lines
So, the right fix would be:
drivers/net/wireless/hostap/hostap_plx.c +1
-1-if (pos + cis[pos +1 ] >=CIS_MAX_LEN ) +if (pos +2 + cis[pos +1 ] >CIS_MAX_LEN ) goto cis_error;I'm rewriting this with "<" because it's easier to read. Next tuple at exactly
CIS_MEX_LENis fine - we just stop precessing at that point.I'm going to combine this with my previous fix and resend.
Great, thanks!
Signature · 1 line
Jouni Malinen PGP id EFC895FA
PATCH 1/1 hostap_plx: fix CIS verification
hostap_plx: fix two related off-by-one errors in CIS parser
From: Pavel Roskin <[email protected]>
The length of the manfid CIS should be at least 4, and it's normally 4. It's incorrect to require it to be at least 5. This breaks support for most (if not all) cards.
The right place to ensure that we don't access beyond the CIS buffer is to strengthen another check. Make sure that the next tuple begins at least at the CIS buffer end (in which case we stop processing) or before that.
Reported by [email protected]
---
I'd like to remind that it's 2.6.x.y material.
drivers/net/wireless/hostap/hostap_plx.c | 4 ++-- 1 files changed, 2 insertions(+), 2 deletions(-)
drivers/net/wireless/hostap/hostap_plx.c +2 -2
@@ -364,7 +364,7 @@ #define CIS_MAX_LEN 256 pos =0 ;while (pos <CIS_MAX_LEN -1 && cis[pos] !=CISTPL_END ) { -if (pos + cis[pos +1 ] >=CIS_MAX_LEN ) +if (pos +2 + cis[pos +1 ] >CIS_MAX_LEN ) goto cis_error;switch (cis[pos]) {
@@ -391,7 +391,7 @@ #define CIS_MAX_LEN 256break ;case CISTPL_MANFID : -if (cis[pos +1 ] <5 ) +if (cis[pos +1 ] <4 ) goto cis_error; manfid1 = cis[pos +2 ] + (cis[pos +3 ] <<8 ); manfid2 = cis[pos +4 ] + (cis[pos +5 ] <<8 );
Signature · 2 lines
Regards, Pavel Roskin