Skip to contentSkip to conversation

fix memory leak in mm/slab.c::alloc_kmemlist()

PATCH In review Jesper Juhl 3 participants 6 messages Mar 12, 2006 – Mar 13, 2006 Memory management

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 fix memory leak in mm/slab.c::alloc_kmemlist()

The Coverity checker found that we may leak memory in mm/slab.c::alloc_kmemlist()
This should fix the leak and coverity bug #589

---

 mm/slab.c |    4 +++-
 1 files changed, 3 insertions(+), 1 deletion(-)
mm/slab.c +3 -1
@@ -3366,8 +3366,10 @@ static int alloc_kmemlist(struct kmem_ca
 			continue;
 		}
 		if (!(l3 = kmalloc_node(sizeof(struct kmem_list3),
-					GFP_KERNEL, node)))
+					GFP_KERNEL, node))) {
+			kfree(new);
 			goto fail;
+		}
 
 		kmem_list3_init(l3);
 		l3->next_reap = jiffies + REAPTIMEOUT_LIST3 +
#1 lore raw
Quoted 17 lines · Jesper Juhl

The Coverity checker found that we may leak memory in mm/slab.c::alloc_kmemlist()
This should fix the leak and coverity bug #589

---

 mm/slab.c |    4 +++-
 1 files changed, 3 insertions(+), 1 deletion(-)
mm/slab.c +0 -0
@@ -3366,8 +3366,10 @@ static int alloc_kmemlist(struct kmem_ca
 			continue;

mm/slab.c static int alloc_kmemlist(struct kmem_ca · line 3375

Quoted 10 lines
mm/slab.c +3 -1
 		}
 		if (!(l3 = kmalloc_node(sizeof(struct kmem_list3),
-					GFP_KERNEL, node)))
+					GFP_KERNEL, node))) {
+			kfree(new);
 			goto fail;
+		}
 
 		kmem_list3_init(l3);
 		l3->next_reap = jiffies + REAPTIMEOUT_LIST3 +

It's more complicated than that. We can also leak new_alien. And if any allocation in that for_each_online_node() loop fails I guess we need to back out all the allocations we've done thus far, which means another loop. ug.

Patches against rc6-mm1 would be preferred please, that code's changed quite a bit.

#2 lore raw
Quoted 7 lines · Andrew Morton
Quoted 6 lines · Jesper Juhl

The Coverity checker found that we may leak memory in mm/slab.c::alloc_kmemlist()
This should fix the leak and coverity bug #589

[snip]

Quoted 6 lines

It's more complicated than that. We can also leak new_alien. And if any allocation in that for_each_online_node() loop fails I guess we need to back out all the allocations we've done thus far, which means another loop. ug.

Ok, I'll take a second (and much more thorough) look at it tonight. And I'll be sure to describe whatever changes I make and the reasoning behind.

Quoted 3 lines

Patches against rc6-mm1 would be preferred please, that code's changed quite a bit.

Sure thing.

--
Jesper Juhl <[email protected]>
Don't top-post http://www.catb.org/~esr/jargon/html/T/top-post.html
Plain text mails only, please http://www.expita.com/nomime.html

#3 lore raw
Quoted 20 lines · Jesper Juhl
Quoted 7 lines · Andrew Morton
Quoted 6 lines · Jesper Juhl

The Coverity checker found that we may leak memory in mm/slab.c::alloc_kmemlist()
This should fix the leak and coverity bug #589

[snip]

Quoted 6 lines

It's more complicated than that. We can also leak new_alien. And if any allocation in that for_each_online_node() loop fails I guess we need to back out all the allocations we've done thus far, which means another loop. ug.

Ok, I'll take a second (and much more thorough) look at it tonight. And I'll be sure to describe whatever changes I make and the reasoning behind.

Quoted 5 lines
Quoted 3 lines

Patches against rc6-mm1 would be preferred please, that code's changed quite a bit.

Sure thing.

Ok, I've been playing around with a few ways to resolve this, but I'm a bit pressed for time and won't have properly tested patches ready tonight. I will however keep at this, so you'll see patches releatively shortly, just give me another day or two and I'll have this fixed in a nice way (nice little task to work at :) ...

--
Jesper Juhl <[email protected]>
Don't top-post http://www.catb.org/~esr/jargon/html/T/top-post.html
Plain text mails only, please http://www.expita.com/nomime.html

#4 lore raw
Quoted 5 lines · Jesper Juhl

Ok, I've been playing around with a few ways to resolve this, but I'm a bit pressed for time and won't have properly tested patches ready tonight. I will however keep at this, so you'll see patches releatively shortly, just give me another day or two and I'll have this fixed in a nice way (nice little task to work at :) ...

Maybe extract a alloc_kmemlist_node and free_kmemlist_node from alloc_kmemlist()? It gets a bit complicated if all of this is handled within alloc_kmemlist and having these separate may simplify recovery on out of memory.

#5 lore raw
Quoted 8 lines · Christoph Lameter
Quoted 5 lines · Jesper Juhl

Ok, I've been playing around with a few ways to resolve this, but I'm a bit pressed for time and won't have properly tested patches ready tonight. I will however keep at this, so you'll see patches releatively shortly, just give me another day or two and I'll have this fixed in a nice way (nice little task to work at :) ...

Quoted 5 lines

Maybe extract a alloc_kmemlist_node and free_kmemlist_node from alloc_kmemlist()? It gets a bit complicated if all of this is handled within alloc_kmemlist and having these separate may simplify recovery on out of memory.

Nice Idea. So far I've been trying to handle it all inside the function, but maybe sepperating stuff out might simplify it.. anyway, i'll keep working on this and will submit patches in a day or two... It really is a nice little problem to work on and I'll find a solution for it (well, I already have at least one, but it's ugly as hell, want to find a better one) - gimme a few days.

--
Jesper Juhl <[email protected]>
Don't top-post http://www.catb.org/~esr/jargon/html/T/top-post.html
Plain text mails only, please http://www.expita.com/nomime.html