Re: [RESEND RFC PATCH 4/5] mm/remote_mapping: use a pidfd to access memory belonging to unrelated process

On Mon, 7 Sep 2020 16:30:08 +0200, Oleg Nesterov [off-list ref] wrote:

it seems that nobody is going to review this patch ;)

So I tried to read mirror_vm_fault() and the usage of mmap_sem doesn't
look right to me. But let me repeat, this is not my area I can be easily
wrong, please correct me.

On 09/04, Adalbert Lazăr wrote:

quoted

+static vm_fault_t mirror_vm_fault(struct vm_fault *vmf)
+{
+	struct vm_area_struct *vma = vmf->vma;
+	struct mm_struct *mm = vma->vm_mm;
+	struct remote_vma_context *ctx = vma->vm_private_data;
+	struct remote_view *view = ctx->view;
+	struct file *file = vma->vm_file;
+	struct remote_file_context *fctx = file->private_data;
+	unsigned long req_addr;
+	unsigned int gup_flags;
+	struct page *req_page;
+	vm_fault_t result = VM_FAULT_SIGBUS;
+	struct mm_struct *src_mm = fctx->mm;
+	unsigned long seq;
+	int idx;
+
+fault_retry:
+	seq = mmu_interval_read_begin(&view->mmin);
+
+	idx = srcu_read_lock(&fctx->fault_srcu);
+
+	/* check if view was invalidated */
+	if (unlikely(!READ_ONCE(view->valid))) {
+		pr_debug("%s: region [%lx-%lx) was invalidated!!\n", __func__,
+			view->offset, view->offset + view->size);
+		goto out_invalid;		/* VM_FAULT_SIGBUS */
+	}
+
+	/* drop current mm semapchore */
+	up_read(&current->mm->mmap_sem);

Please use mmap_read_lock/unlock(mm) instead of down/up_read(mmap_sem).

This patch series is based on 5.7-rc2.
The cover letter has base-commit: present, but I forgot to mention this
explicitly, sorry.

Adalbert