Fix for binary_sysctl() memory leak
PATCH 1/1 Fix for binary_sysctl() memory leak
binary_sysctl() calls sysctl_getname() which allocates from names_cache slab usin __getname()
The matching function to free the name is __putname(), and not putname() which should be used only to match getname() allocations.
--- kernel/sysctl_binary.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)
kernel/sysctl_binary.c +1 -1
@@ -1354,7 +1354,7 @@ static ssize_t binary_sysctl(const int *name, int nlen,fput (file); out_putname: -putname (pathname); +__putname (pathname); out:return result; }
Signature · 1 line
1.7.3.1
Quoted 16 lines · On Wed, 14 Dec 2011 18:44:12 -0800 Michel Lespinasse <[email protected]> wrote:
binary_sysctl()callssysctl_getname()which allocates fromnames_cacheslab usin__getname()The matching function to free the name is
__putname(), and notputname()which should be used only to matchgetname()allocations.Signed-off-by Michel Lespinasse <[email protected]>--- kernel/sysctl_binary.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)kernel/sysctl_binary.c +0
-0
kernel/sysctl_binary.c static ssize_t binary_sysctl(const int *name, int nlen, · line 1360
Quoted 8 lines
kernel/sysctl_binary.c +1
-1@@ -1354,7 +1354,7 @@ static ssize_t binary_sysctl(const int *name, int nlen,fput (file); out_putname: -putname (pathname); +__putname (pathname); out:return result; }
I think the patch is correct but the description is misleading?
I see no memory leak here. Calling __putname() directly simply bypasses some audit-related stuff.
Quoted 4 lines · Andrew Morton
I think the patch is correct but the description is misleading?
I see no memory leak here. Calling
__putname()directly simply bypasses some audit-related stuff.
Hmmm, maybe I wasn't explicit enough about it. We are definitely seeing a memory leak without the patch.
When auditing is enabled, putname() calls audit_putname *instead* (not in addition) to __putname(). Then, if a syscall is in progress, audit_putname does not release the name - instead, it expects the name to get released when the syscall completes, but that will happen only if audit_getname() was called previously, i.e. if the name was allocated with getname() rather than the naked __getname(). So, __getname() followed by putname() ends up leaking memory.
Signature · 2 lines
Michel "Walken" Lespinasse A program is never fully debugged until the last user dies.
Quoted 10 lines · On Thu, 15 Dec 2011 14:38:58 -0800 Michel Lespinasse <[email protected]> wrote:
Quoted 4 lines · Andrew Morton
I think the patch is correct but the description is misleading?
I see no memory leak here.
__Calling__putname()directly simply bypasses some audit-related stuff.Hmmm, maybe I wasn't explicit enough about it. We are definitely seeing a memory leak without the patch.
Quoted 8 lines
When auditing is enabled,
putname()callsaudit_putname*instead* (not in addition) to__putname(). Then, if a syscall is in progress,audit_putnamedoes not release the name - instead, it expects the name to get released when the syscall completes, but that will happen only ifaudit_getname()was called previously, i.e. if the name was allocated withgetname()rather than the naked__getname(). So,__getname()followed byputname()ends up leaking memory.
OK. Please resend with a new changelog?
The bug surprises me - it seems that it makes it trivial for userspace to cause leaking of mad amounts of kernel memory, which would cause the bug to be found and fixed quickly.
Is it a recent regression, or does the bug trigger only in weird circumstances, or what?
Quoted 1 line · Andrew Morton
OK. Please resend with a new changelog?
All right. Will do as a reply to this (I'll add the same explanation as in my previous email).
Quoted 6 lines
The bug surprises me - it seems that it makes it trivial for userspace to cause leaking of mad amounts of kernel memory, which would cause the bug to be found and fixed quickly.
Is it a recent regression, or does the bug trigger only in weird circumstances, or what?
As far as I can tell, the bug was introduced in v2.6.33-rc1 by commit 26a7034b40ba80f82f64fa251a2cbf49f9971c6a
It's probably not common because people use sysctl mainly at boot time, and/or don't use the deprecated binary variant of it anymore. But yes, it is trivial to exploit...
Signature · 2 lines
Michel "Walken" Lespinasse A program is never fully debugged until the last user dies.
binary_sysctl() calls sysctl_getname() which allocates from names_cache slab usin __getname()
The matching function to free the name is __putname(), and not putname() which should be used only to match getname() allocations.
This is because when auditing is enabled, putname() calls audit_putname *instead* (not in addition) to __putname(). Then, if a syscall is in progress, audit_putname does not release the name - instead, it expects the name to get released when the syscall completes, but that will happen only if audit_getname() was called previously, i.e. if the name was allocated with getname() rather than the naked __getname(). So, __getname() followed by putname() ends up leaking memory.
--- kernel/sysctl_binary.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)
kernel/sysctl_binary.c +1 -1
@@ -1354,7 +1354,7 @@ static ssize_t binary_sysctl(const int *name, int nlen,fput (file); out_putname: -putname (pathname); +__putname (pathname); out:return result; }
Signature · 5 lines
1.7.3.1 -- Michel "Walken" Lespinasse A program is never fully debugged until the last user dies.
Quoted 22 lines · Andrew Morton
Quoted 10 lines · On Thu, 15 Dec 2011 14:38:58 -0800 Michel Lespinasse <[email protected]> wrote:
Quoted 4 lines · Andrew Morton
I think the patch is correct but the description is misleading?
I see no memory leak here.
__Calling__putname()directly simply bypasses some audit-related stuff.Hmmm, maybe I wasn't explicit enough about it. We are definitely seeing a memory leak without the patch.
Quoted 8 lines
When auditing is enabled,
putname()callsaudit_putname*instead* (not in addition) to__putname(). Then, if a syscall is in progress,audit_putnamedoes not release the name - instead, it expects the name to get released when the syscall completes, but that will happen only ifaudit_getname()was called previously, i.e. if the name was allocated withgetname()rather than the naked__getname(). So,__getname()followed byputname()ends up leaking memory.
Quoted 1 line
OK. Please resend with a new changelog?
This seems like a reasonable change to me. I guess I misunderstood the __getname/putname interaction. I expect I was focusing on the fact you can only call getname if you have your data in userspace.
Quoted 6 lines
The bug surprises me - it seems that it makes it trivial for userspace to cause leaking of mad amounts of kernel memory, which would cause the bug to be found and fixed quickly.
Is it a recent regression, or does the bug trigger only in weird circumstances, or what?
Calling sysctl(2) is very rare. I don't know if it actually happens anywhere with a modern userspace except in regression tests. Effectively we retain sysctl(2) because it doesn't take too much to maintain.
Michel what caused you to discover this bug? If you are using sysctl(2) in production code I am a bit worried.
Eric
Quoted 4 lines · Eric W. Biederman
Calling sysctl(2) is very rare. I don't know if it actually happens anywhere with a modern userspace except in regression tests. Effectively we retain sysctl(2) because it doesn't take too much to maintain.
Agree, there is no good reason one would call sysctl() anymore.
Quoted 2 lines
Michel what caused you to discover this bug? If you are using sysctl(2) in production code I am a bit worried.
Well, somebody hit it :) To be honest, I'm not sure how they came to do that. It wasn't in a datacenter setting.
Signature · 2 lines
Michel "Walken" Lespinasse A program is never fully debugged until the last user dies.