| Message ID | 20221019201957.34607-1-mike.kravetz@oracle.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | [v2] hugetlb: fix memory leak associated with vma_lock structure | expand |
On 2022/10/20 4:19, Mike Kravetz wrote: > The hugetlb vma_lock structure hangs off the vm_private_data pointer > of sharable hugetlb vmas. The structure is vma specific and can not > be shared between vmas. At fork and various other times, vmas are > duplicated via vm_area_dup(). When this happens, the pointer in the > newly created vma must be cleared and the structure reallocated. Two > hugetlb specific routines deal with this hugetlb_dup_vma_private and > hugetlb_vm_op_open. Both routines are called for newly created vmas. > hugetlb_dup_vma_private would always clear the pointer and > hugetlb_vm_op_open would allocate the new vms_lock structure. This did > not work in the case of this calling sequence pointed out in [1]. > move_vma > copy_vma > new_vma = vm_area_dup(vma); > new_vma->vm_ops->open(new_vma); --> new_vma has its own vma lock. > is_vm_hugetlb_page(vma) > clear_vma_resv_huge_pages > hugetlb_dup_vma_private --> vma->vm_private_data is set to NULL > When clearing hugetlb_dup_vma_private we actually leak the associated > vma_lock structure. > > The vma_lock structure contains a pointer to the associated vma. This > information can be used in hugetlb_dup_vma_private and hugetlb_vm_op_open > to ensure we only clear the vm_private_data of newly created (copied) > vmas. In such cases, the vma->vma_lock->vma field will not point to the > vma. > > Update hugetlb_dup_vma_private and hugetlb_vm_op_open to not clear > vm_private_data if vma->vma_lock->vma == vma. Also, log a warning if > hugetlb_vm_op_open ever encounters the case where vma_lock has already > been correctly allocated for the vma. > > [1] https://lore.kernel.org/linux-mm/5154292a-4c55-28cd-0935-82441e512fc3@huawei.com/ > > Fixes: 131a79b474e9 ("hugetlb: fix vma lock handling during split vma and range unmapping") > Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com> Thanks for update. This patch looks good to me. Reviewed-by: Miaohe Lin <linmiaohe@huawei.com> Thanks, Miaohe Lin > --- > v2 - Allocate vma_lock in hugetlb_vm_op_open if !vma->vm_private_data > on entry. Thanks Miaohe Lin! > > mm/hugetlb.c | 35 +++++++++++++++++++++++++++-------- > 1 file changed, 27 insertions(+), 8 deletions(-) > > diff --git a/mm/hugetlb.c b/mm/hugetlb.c > index 02f781624fce..ccdffc2fa1ca 100644 > --- a/mm/hugetlb.c > +++ b/mm/hugetlb.c > @@ -1014,15 +1014,23 @@ void hugetlb_dup_vma_private(struct vm_area_struct *vma) > VM_BUG_ON_VMA(!is_vm_hugetlb_page(vma), vma); > /* > * Clear vm_private_data > + * - For shared mappings this is a per-vma semaphore that may be > + * allocated in a subsequent call to hugetlb_vm_op_open. > + * Before clearing, make sure pointer is not associated with vma > + * as this will leak the structure. This is the case when called > + * via clear_vma_resv_huge_pages() and hugetlb_vm_op_open has already > + * been called to allocate a new structure. > * - For MAP_PRIVATE mappings, this is the reserve map which does > * not apply to children. Faults generated by the children are > * not guaranteed to succeed, even if read-only. > - * - For shared mappings this is a per-vma semaphore that may be > - * allocated in a subsequent call to hugetlb_vm_op_open. > */ > - vma->vm_private_data = (void *)0; > - if (!(vma->vm_flags & VM_MAYSHARE)) > - return; > + if (vma->vm_flags & VM_MAYSHARE) { > + struct hugetlb_vma_lock *vma_lock = vma->vm_private_data; > + > + if (vma_lock && vma_lock->vma != vma) > + vma->vm_private_data = NULL; > + } else > + vma->vm_private_data = NULL; > } > > /* > @@ -4601,6 +4609,7 @@ static void hugetlb_vm_op_open(struct vm_area_struct *vma) > struct resv_map *resv = vma_resv_map(vma); > > /* > + * HPAGE_RESV_OWNER indicates a private mapping. > * This new VMA should share its siblings reservation map if present. > * The VMA will only ever have a valid reservation map pointer where > * it is being copied for another still existing VMA. As that VMA > @@ -4615,11 +4624,21 @@ static void hugetlb_vm_op_open(struct vm_area_struct *vma) > > /* > * vma_lock structure for sharable mappings is vma specific. > - * Clear old pointer (if copied via vm_area_dup) and create new. > + * Clear old pointer (if copied via vm_area_dup) and allocate > + * new structure. Before clearing, make sure vma_lock is not > + * for this vma. > */ > if (vma->vm_flags & VM_MAYSHARE) { > - vma->vm_private_data = NULL; > - hugetlb_vma_lock_alloc(vma); > + struct hugetlb_vma_lock *vma_lock = vma->vm_private_data; > + > + if (vma_lock) { > + if (vma_lock->vma != vma) { > + vma->vm_private_data = NULL; > + hugetlb_vma_lock_alloc(vma); > + } else > + pr_warn("HugeTLB: vma_lock already exists in %s.\n", __func__); > + } else > + hugetlb_vma_lock_alloc(vma); > } > } > >
diff --git a/mm/hugetlb.c b/mm/hugetlb.c index 02f781624fce..ccdffc2fa1ca 100644 --- a/mm/hugetlb.c +++ b/mm/hugetlb.c @@ -1014,15 +1014,23 @@ void hugetlb_dup_vma_private(struct vm_area_struct *vma) VM_BUG_ON_VMA(!is_vm_hugetlb_page(vma), vma); /* * Clear vm_private_data + * - For shared mappings this is a per-vma semaphore that may be + * allocated in a subsequent call to hugetlb_vm_op_open. + * Before clearing, make sure pointer is not associated with vma + * as this will leak the structure. This is the case when called + * via clear_vma_resv_huge_pages() and hugetlb_vm_op_open has already + * been called to allocate a new structure. * - For MAP_PRIVATE mappings, this is the reserve map which does * not apply to children. Faults generated by the children are * not guaranteed to succeed, even if read-only. - * - For shared mappings this is a per-vma semaphore that may be - * allocated in a subsequent call to hugetlb_vm_op_open. */ - vma->vm_private_data = (void *)0; - if (!(vma->vm_flags & VM_MAYSHARE)) - return; + if (vma->vm_flags & VM_MAYSHARE) { + struct hugetlb_vma_lock *vma_lock = vma->vm_private_data; + + if (vma_lock && vma_lock->vma != vma) + vma->vm_private_data = NULL; + } else + vma->vm_private_data = NULL; } /* @@ -4601,6 +4609,7 @@ static void hugetlb_vm_op_open(struct vm_area_struct *vma) struct resv_map *resv = vma_resv_map(vma); /* + * HPAGE_RESV_OWNER indicates a private mapping. * This new VMA should share its siblings reservation map if present. * The VMA will only ever have a valid reservation map pointer where * it is being copied for another still existing VMA. As that VMA @@ -4615,11 +4624,21 @@ static void hugetlb_vm_op_open(struct vm_area_struct *vma) /* * vma_lock structure for sharable mappings is vma specific. - * Clear old pointer (if copied via vm_area_dup) and create new. + * Clear old pointer (if copied via vm_area_dup) and allocate + * new structure. Before clearing, make sure vma_lock is not + * for this vma. */ if (vma->vm_flags & VM_MAYSHARE) { - vma->vm_private_data = NULL; - hugetlb_vma_lock_alloc(vma); + struct hugetlb_vma_lock *vma_lock = vma->vm_private_data; + + if (vma_lock) { + if (vma_lock->vma != vma) { + vma->vm_private_data = NULL; + hugetlb_vma_lock_alloc(vma); + } else + pr_warn("HugeTLB: vma_lock already exists in %s.\n", __func__); + } else + hugetlb_vma_lock_alloc(vma); } }
The hugetlb vma_lock structure hangs off the vm_private_data pointer of sharable hugetlb vmas. The structure is vma specific and can not be shared between vmas. At fork and various other times, vmas are duplicated via vm_area_dup(). When this happens, the pointer in the newly created vma must be cleared and the structure reallocated. Two hugetlb specific routines deal with this hugetlb_dup_vma_private and hugetlb_vm_op_open. Both routines are called for newly created vmas. hugetlb_dup_vma_private would always clear the pointer and hugetlb_vm_op_open would allocate the new vms_lock structure. This did not work in the case of this calling sequence pointed out in [1]. move_vma copy_vma new_vma = vm_area_dup(vma); new_vma->vm_ops->open(new_vma); --> new_vma has its own vma lock. is_vm_hugetlb_page(vma) clear_vma_resv_huge_pages hugetlb_dup_vma_private --> vma->vm_private_data is set to NULL When clearing hugetlb_dup_vma_private we actually leak the associated vma_lock structure. The vma_lock structure contains a pointer to the associated vma. This information can be used in hugetlb_dup_vma_private and hugetlb_vm_op_open to ensure we only clear the vm_private_data of newly created (copied) vmas. In such cases, the vma->vma_lock->vma field will not point to the vma. Update hugetlb_dup_vma_private and hugetlb_vm_op_open to not clear vm_private_data if vma->vma_lock->vma == vma. Also, log a warning if hugetlb_vm_op_open ever encounters the case where vma_lock has already been correctly allocated for the vma. [1] https://lore.kernel.org/linux-mm/5154292a-4c55-28cd-0935-82441e512fc3@huawei.com/ Fixes: 131a79b474e9 ("hugetlb: fix vma lock handling during split vma and range unmapping") Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com> --- v2 - Allocate vma_lock in hugetlb_vm_op_open if !vma->vm_private_data on entry. Thanks Miaohe Lin! mm/hugetlb.c | 35 +++++++++++++++++++++++++++-------- 1 file changed, 27 insertions(+), 8 deletions(-)