201547 - (zen7) Buffer overrun in mimetric.cpp

Status: VERIFIED FIXED

(MailNews Core :: Security, defect)

Description

[Mitchell Stoltz (not reading bugmail)]

from [e-mail]:

Issue details:
MimeRichtextConvert() in mailnews/mime/src/mimetric.cpp allocates a buffer that
is smaller than is needed.

http://lxr.mozilla.org/mozilla/source/mailnews/mime/src/mimetric.cpp#106 

The buffer allocated is 4 times the length of the input, however the '&'
character expands to "&" (5 chars).

A fix could be to change 
106   desired_size = length * 4;
into
106   desired_size = (length * 5) + 1; 
which would handle the worst case of an input consisting of all '&' characters.

In order to exploit this vulnerability someone would need to seen a specially
crafted email, and it would ned to be viewed.

Comment 1

[Mitchell Stoltz (not reading bugmail)]

Attachment: C source file for demonstration

Comment 2

[georgi - hopefully not receiving bugspam]

Does "view message body as plain text" stop this?

Comment 3

[Mitchell Stoltz (not reading bugmail)]

Stack trace when the overrun happens (thanks Darin):

==3531==    at 0x4046E384: PL_strcpy
(/builds/trunk/mozilla/nsprpub/lib/libc/src/strcpy.c:45)
==3531==    by 0x49D3528C: MimeRichtextConvert(char*, int, int (*)(char*, int,
void*), void*, char**, int*, int)
(/builds/trunk/mozilla/mailnews/mime/src/mimetric.cpp:165)
==3531==    by 0x49D35AA3: MimeInlineTextRichtext_parse_line(char*, int,
MimeObject*) (/builds/trunk/mozilla/mailnews/mime/src/mimetric.cpp:336)
==3531==    by 0x49D2EF5A: MimeInlineText_convert_and_parse_line(char*, int,
MimeObject*) (/builds/trunk/mozilla/mailnews/mime/src/mimetext.cpp:412)
==3531==    Address 0x4FA944C4 is 0 bytes after a block of size 3972 alloc'd
==3531==    at 0x4003BA4E: malloc (vg_clientfuncs.c:100)
==3531==    by 0x40489512: PR_Malloc
(/builds/trunk/mozilla/nsprpub/pr/src/malloc/prmem.c:433)
==3531==    by 0x49D36B77: mime_GrowBuffer
(/builds/trunk/mozilla/mailnews/mime/src/mimebuf.cpp:49)
==3531==    by 0x49D350C1: MimeRichtextConvert(char*, int, int (*)(char*, int,
void*), void*, char**, int*, int)
(/builds/trunk/mozilla/mailnews/mime/src/mimetric.cpp:108)

Comment 4

[Mitchell Stoltz (not reading bugmail)]

Attachment: Patch - zen's suggested fix

Comment 5

[Mitchell Stoltz (not reading bugmail)]

Comment on attachment 121225 [details] [diff] [review]
Patch - zen's suggested fix

r=mstoltz, needs an sr

Comment 6

[Heikki Toivonen (remove -bugzilla when emailing directly)]

Comment on attachment 121225 [details] [diff] [review]
Patch - zen's suggested fix

Should fix the ampersand at least, so sr=heikki.

Then I got scared that the stuff below that in the function might result in
bigger expansion than 1->5, but it does not seem so.

So this brief moment of panic made me realize we really need some comments.
When setting the desired_length, how about adding a comment like this for
example:
// The code below must never expand anything by more than the factor specified
here; if it does this value must be changed too

Comment 7

[Cavin Song]

It is possible that the switch statement in the function eats up more space than
the buffer was allocated at the beginning of the function. I added the following
line to the attached C demonstration program (which must be compiled on linux
and run) and the switch statement was indeed invoked.

strcat(buf,"<BIGGER> xx </BIGGER>");  // this is what I added
strcat(buf,"AAAABBBBCCCCDDDDEEEEFFFFGGGG");

In this case, we are replacing:

  <BIGGER> xx </BIGGER>

with

  <FONT SIZE="+1"> xx </FONT>

so we eat up 6 more chars than what was allocated. But in other cases we are
actually saving some room. For example,

  <BOLD> xx </BOLD>  --->  <B> xx </B> 

we are saving 6 chars in this case. So it depends on what's in the body text we
may or may not overrun the allocated buffer.

Comment 8

[Mitchell Stoltz (not reading bugmail)]

Cavin and I discussed converting this function to use the string classes, so we
won't have to deal with allocation. Cavin is going to look into this.

In the meantime, I already checked in the above change, so we're a little better
off.

Comment 9

[georgi - hopefully not receiving bugspam]

Is there another testcase to check whether "view message body as plaintext"
stops this attack?
The original testcase just makes mozilla unresponsive for a long time, though
ltrace shows it does some memory relocation.

Comment 10

[Heikki Toivonen (remove -bugzilla when emailing directly)]

I haven't actually debugged this, but replacing "<BIGGER> xx </BIGGER>" with
"<FONT SIZE="+1"> xx </FONT>" means an expansion factor 1.27, and that should be
safe. Compare to "&" -> "&amp;", which is expansion factor 5. Since the
calculation for the new string is now 5*original_length+1 we should be ok - none
of the tag stuff seems to expand the string bigger than that. Or am I missing
something?

Comment 11

[Cavin Song]

Good point. We should be fine with *5.

Comment 12

[Mitchell Stoltz (not reading bugmail)]

The 5x+1 fix has been checked in, but I'd like to get some verification that
none of the other expansions are more than 5x. Georgi, could you take a look?

Comment 13

[Asa Dotzler [:asa]]

mstolts, can you land this on the MOZILLA_1_3_BRANCH as well? 

Comment 14

[(not reading, please use [e-mail] instead)]

I've landed the fix on the 1.3 branch, per asa's request.

Comment 15

[georgi - hopefully not receiving bugspam]

I shall look into it.
I am concerned also about the following: in text/plain email *word* becomes bold
when rendered. What file/class is responsible for this conversion to check it
for similar issues?

Comment 16

[Heikki Toivonen (remove -bugzilla when emailing directly)]

I'm marking this fixed, please reopen if you find a case where the landed fix
was not sufficient.

Comment 17

[John Unruh]

V

Comment 18

[Daniel Veditz [:dveditz]]

Bugs published on the Known-vulnerabilities page long ago, removing confidential
flag.