When called for anonymous (non-shared) mappings,
hugetlb_reserve_pages() does a resv_map_alloc(). It depends on
code in hugetlbfs's vm_ops->close() to release that allocation.
However, in the mmap() failure path, we do a plain unmap_region()
without the remove_vma() which actually calls vm_ops->close().
This is a decent fix. This leak could get reintroduced if
new code (say, after hugetlb_reserve_pages() in
hugetlbfs_file_mmap()) decides to return an error. But, I think
it would have to unroll the reservation anyway.
This hasn't been extensively tested. Pretty much compile and
boot tested along with Christoph's test case.
Comments?
Signed-off-by: Dave Hansen <[email protected]>
---
linux-2.6.git-dave/mm/hugetlb.c | 28 ++++++++++++++++++++++------
1 file changed, 22 insertions(+), 6 deletions(-)
diff -puN mm/hugetlb.c~hugetlb-fix-leak mm/hugetlb.c
--- linux-2.6.git/mm/hugetlb.c~hugetlb-fix-leak 2012-05-18 11:45:50.355089708 -0700
+++ linux-2.6.git-dave/mm/hugetlb.c 2012-05-18 11:45:50.363089800 -0700
@@ -2157,6 +2157,15 @@ static void hugetlb_vm_op_open(struct vm
kref_get(&reservations->refs);
}
+static void resv_map_put(struct vm_area_struct *vma)
+{
+ struct resv_map *reservations = vma_resv_map(vma);
+
+ if (!reservations)
+ return;
+ kref_put(&reservations->refs, resv_map_release);
+}
+
static void hugetlb_vm_op_close(struct vm_area_struct *vma)
{
struct hstate *h = hstate_vma(vma);
@@ -2173,7 +2182,7 @@ static void hugetlb_vm_op_close(struct v
reserve = (end - start) -
region_count(&reservations->regions, start, end);
- kref_put(&reservations->refs, resv_map_release);
+ resv_map_put(vma);
if (reserve) {
hugetlb_acct_memory(h, -reserve);
@@ -2990,12 +2999,16 @@ int hugetlb_reserve_pages(struct inode *
set_vma_resv_flags(vma, HPAGE_RESV_OWNER);
}
- if (chg < 0)
- return chg;
+ if (chg < 0) {
+ ret = chg;
+ goto out_err;
+ }
/* There must be enough pages in the subpool for the mapping */
- if (hugepage_subpool_get_pages(spool, chg))
- return -ENOSPC;
+ if (hugepage_subpool_get_pages(spool, chg)) {
+ ret = -ENOSPC;
+ goto out_err;
+ }
/*
* Check enough hugepages are available for the reservation.
@@ -3004,7 +3017,7 @@ int hugetlb_reserve_pages(struct inode *
ret = hugetlb_acct_memory(h, chg);
if (ret < 0) {
hugepage_subpool_put_pages(spool, chg);
- return ret;
+ goto out_err;
}
/*
@@ -3021,6 +3034,9 @@ int hugetlb_reserve_pages(struct inode *
if (!vma || vma->vm_flags & VM_MAYSHARE)
region_add(&inode->i_mapping->private_list, from, to);
return 0;
+out_err:
+ resv_map_put(vma);
+ return ret;
}
void hugetlb_unreserve_pages(struct inode *inode, long offset, long freed)
_
On Fri, May 18, 2012 at 11:46:30AM -0700, Dave Hansen wrote:
>
> When called for anonymous (non-shared) mappings,
> hugetlb_reserve_pages() does a resv_map_alloc(). It depends on
> code in hugetlbfs's vm_ops->close() to release that allocation.
>
> However, in the mmap() failure path, we do a plain unmap_region()
> without the remove_vma() which actually calls vm_ops->close().
>
> This is a decent fix. This leak could get reintroduced if
> new code (say, after hugetlb_reserve_pages() in
> hugetlbfs_file_mmap()) decides to return an error. But, I think
> it would have to unroll the reservation anyway.
>
> This hasn't been extensively tested. Pretty much compile and
> boot tested along with Christoph's test case.
>
> Comments?
>
> Signed-off-by: Dave Hansen <[email protected]>
Acked-by: Mel Gorman <[email protected]>
--
Mel Gorman
SUSE Labs
On Mon, 21 May 2012, Mel Gorman wrote:
> > Signed-off-by: Dave Hansen <[email protected]>
>
> Acked-by: Mel Gorman <[email protected]>
Reported/tested-by: Christoph Lameter <[email protected]>
On 5/18/2012 2:46 PM, Dave Hansen wrote:
> When called for anonymous (non-shared) mappings,
> hugetlb_reserve_pages() does a resv_map_alloc(). It depends on
> code in hugetlbfs's vm_ops->close() to release that allocation.
>
> However, in the mmap() failure path, we do a plain unmap_region()
> without the remove_vma() which actually calls vm_ops->close().
>
> This is a decent fix. This leak could get reintroduced if
> new code (say, after hugetlb_reserve_pages() in
> hugetlbfs_file_mmap()) decides to return an error. But, I think
> it would have to unroll the reservation anyway.
>
> This hasn't been extensively tested. Pretty much compile and
> boot tested along with Christoph's test case.
>
> Comments?
>
>
> Signed-off-by: Dave Hansen <[email protected]>
> ---
I don't think this is cleaner fix. but I also think we should fix the leak
asap. so Let's simple fix first.
Acked-by: KOSAKI Motohiro <[email protected]>