[
https://issues.apache.org/jira/browse/THRIFT-3175?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15056882#comment-15056882
]
Dvir Volk commented on THRIFT-3175:
-----------------------------------
I agree it's not a good solution, although it was a problem not in the JVM but
in python code. I would suggest a simpler heuristic - whether an allocated a
list of the requested size be applicable to the message length. i.e. it's clear
to see that there's no use in allocating 2G elements for a message that's 1K in
length...
> fastbinary.c python deserialize can cause huge allocations from garbage
> -----------------------------------------------------------------------
>
> Key: THRIFT-3175
> URL: https://issues.apache.org/jira/browse/THRIFT-3175
> Project: Thrift
> Issue Type: Bug
> Components: Python - Library
> Reporter: Dvir Volk
> Assignee: Dvir Volk
> Fix For: 0.9.3
>
>
> In the fastbinary python deserializer, allocating a list is done like so:
> {code}
> len = readI32(input);
> if (!check_ssize_t_32(len)) {
> return NULL;
> }
> ret = PyList_New(len);
> {code}
> The only validation of len is that it's under INT_MAX. I've encountered a
> situation where upon receiving garbage input, and having len be read as
> something like 1 billion, the library treated this as a valid input,
> allocated gigs of RAM, and caused a server to crash.
> The quick fix I made was to limit list sizes to a sane value of a few
> thousands that more than suits my personal needs.
> But IMO this should be dealt with properly. One way that comes to mind is not
> pre-allocating the entire list in advance in case it's really big, and
> resizing it in smaller steps while reading the input.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
