[Bug 1881982] Re: memory exhaustion in parse_report()
Seong-Joong Kim
1881982 at bugs.launchpad.net
Tue Jun 16 02:13:31 UTC 2020
** Description changed:
Hi,
I have found a security issue on whoopsie 0.2.69 and earlier.
## Vulnerability in whoopsie
- - whoopsie 0.2.69 and earlier have a memory leak vulnerability.
- - An attacker can cause a denial of service (application crash) via a crafted .crash file.
+ - It was discovered that whoopsie incorrectly handled certain malformed crash files. If a user using whoopsie were tricked into parsing and uploading a specially crafted crash file, an attacker could exploit this to cause a denial of service.
## Basic
When a program has been crashed, Linux system tries to create a '.crash' file on '/var/crash/' directory with python script located in '/usr/share/apport/apport'.
The file contains a series of system crash information including core dump, syslog, stack trace, memory map info, etc.
A user is given read and write permission to the file.
After then, whoopsie parses key-value pairs in ‘.crash’ file and encodes it into binary json (bson) format.
Lastly, whoopsie forwards the data to a remotely connected Ubuntu error report system.
## Vulnerability
We have found a memory leak vulnerability during the parsing the crash file, when a collision occurs on GHashTable through g_hash_table_insert().
According to [1], if the key already exists in the GHashTable, its current value is replaced with the new value.
If 'key_destory_func' and 'value_destroy_func' are supplied when creating the table, the old value and the passed key are freed using that function.
Unfortunately, whoopsie does not handle the old value and the passed key when collision happens.
If a crash file contains same repetitive key-value pairs, it leads to memory leak as much as the amount of repetition and results in denial-of-service.
## Attack
- 1) Create a fake.crash file
- memory_leak_poc.py script measures an available memory and generates a malicious crash file that contains same repetitive key-value pairs as much as 20% of the available memory size; 'ProblemType: Crash'.
- 20% indicates arbitrary amount of the memory leakage.
- 2) Before the attack, the script checks memory usage of whoopsie process with psutil
- 3) It triggers the whoopsie to read the fake.crash file
- 4) Then, it measures the memory usage of whoopsie process
- 5) It results in denial-of-service and then other users can no longer report crash to the Ubuntu error report system.
+ 1) Generates a certain malformed crash file that contains same repetitive key-value pairs.
+ 2) Trigger the whoopsie to read the generated crash file.
+ 3) After then, the whoopsie process has been killed.
## Mitigation
We should use g_hash_table_new_full() with ‘key_destroy_func’ and ‘value_destroy_func’ functions instead of g_hash_table_new().
Otherwise, before g_hash_table_insert(), we should check the collision via g_hash_table_lookup_extended() and obtain pointer to the old value and remove it.
Sincerely,
[1] https://developer.gnome.org/glib/stable/glib-Hash-Tables.html#g
-hash-table-insert
--
You received this bug notification because you are a member of Ubuntu
Foundations Bugs, which is subscribed to whoopsie in Ubuntu.
https://bugs.launchpad.net/bugs/1881982
Title:
memory exhaustion in parse_report()
Status in whoopsie package in Ubuntu:
New
Bug description:
Hi,
I have found a security issue on whoopsie 0.2.69 and earlier.
## Vulnerability in whoopsie
- It was discovered that whoopsie incorrectly handled certain malformed crash files. If a user using whoopsie were tricked into parsing and uploading a specially crafted crash file, an attacker could exploit this to cause a denial of service.
## Basic
When a program has been crashed, Linux system tries to create a '.crash' file on '/var/crash/' directory with python script located in '/usr/share/apport/apport'.
The file contains a series of system crash information including core dump, syslog, stack trace, memory map info, etc.
A user is given read and write permission to the file.
After then, whoopsie parses key-value pairs in ‘.crash’ file and encodes it into binary json (bson) format.
Lastly, whoopsie forwards the data to a remotely connected Ubuntu error report system.
## Vulnerability
We have found a memory leak vulnerability during the parsing the crash file, when a collision occurs on GHashTable through g_hash_table_insert().
According to [1], if the key already exists in the GHashTable, its current value is replaced with the new value.
If 'key_destory_func' and 'value_destroy_func' are supplied when creating the table, the old value and the passed key are freed using that function.
Unfortunately, whoopsie does not handle the old value and the passed key when collision happens.
If a crash file contains same repetitive key-value pairs, it leads to memory leak as much as the amount of repetition and results in denial-of-service.
## Attack
1) Generates a certain malformed crash file that contains same repetitive key-value pairs.
2) Trigger the whoopsie to read the generated crash file.
3) After then, the whoopsie process has been killed.
## Mitigation
We should use g_hash_table_new_full() with ‘key_destroy_func’ and ‘value_destroy_func’ functions instead of g_hash_table_new().
Otherwise, before g_hash_table_insert(), we should check the collision via g_hash_table_lookup_extended() and obtain pointer to the old value and remove it.
Sincerely,
[1] https://developer.gnome.org/glib/stable/glib-Hash-Tables.html#g
-hash-table-insert
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/whoopsie/+bug/1881982/+subscriptions
More information about the foundations-bugs
mailing list