RDMA/umem: Fix potential addition overflow
authorDoug Ledford <dledford@redhat.com>
Fri, 21 Sep 2018 15:30:13 +0000 (11:30 -0400)
committerJason Gunthorpe <jgg@mellanox.com>
Tue, 25 Sep 2018 21:19:06 +0000 (15:19 -0600)
Given a large enough memory allocation, it is possible to wrap the
pinned_vm counter.  Check for addition overflow to prevent such
eventualities.

Fixes: 40ddacf2dda9 ("RDMA/umem: Don't hold mmap_sem for too long")
Reported-by: Jason Gunthorpe <jgg@ziepe.ca>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
drivers/infiniband/core/umem.c

index 1886d77..8da1cf2 100644 (file)
@@ -85,6 +85,7 @@ struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr,
        struct page **page_list;
        struct vm_area_struct **vma_list;
        unsigned long lock_limit;
+       unsigned long new_pinned;
        unsigned long cur_base;
        struct mm_struct *mm;
        unsigned long npages;
@@ -160,12 +161,13 @@ struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr,
        lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 
        down_write(&mm->mmap_sem);
-       mm->pinned_vm += npages;
-       if ((mm->pinned_vm > lock_limit) && !capable(CAP_IPC_LOCK)) {
+       if (check_add_overflow(mm->pinned_vm, npages, &new_pinned) ||
+           (new_pinned > lock_limit && !capable(CAP_IPC_LOCK))) {
                up_write(&mm->mmap_sem);
                ret = -ENOMEM;
-               goto vma;
+               goto out;
        }
+       mm->pinned_vm = new_pinned;
        up_write(&mm->mmap_sem);
 
        cur_base = addr & PAGE_MASK;