On 05/30/2018 09:45 AM, Christoph Hellwig wrote:

Just keep a per-controller buffer of changed namespaces and copy it out
in the get log page implementation.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---

[...]

+static void nvmet_add_to_changed_ns_log(struct nvmet_ctrl *ctrl, __le32 nsid)
+{
+	mutex_lock(&ctrl->lock);
+	if (ctrl->nr_changed_ns < NVME_MAX_CHANGED_NAMESPACES) {

Minor nitpick here: if ctrlr->nr_changed_ns is exactly
NVME_MAX_CHANGED_NAMESPACES and the new nsid is already in the list,
this will skip down to the 0xffffffff case below, even though it could
have just left the list as-is.

I don't know if this is a problem in practice; reporting 0xffffffff is
probably always safe, since the host will presumably treat that as
"rescan everything".

+		u32 i;
+
+		for (i = 0; i < ctrl->nr_changed_ns; i++)
+			if (ctrl->changed_ns_list[i] == nsid)
+				goto out_unlock;
+		ctrl->changed_ns_list[ctrl->nr_changed_ns++] = nsid;
+	} else if (ctrl->nr_changed_ns == NVME_MAX_CHANGED_NAMESPACES) {
+		ctrl->changed_ns_list[0] = cpu_to_le32(0xffffffff);
+		ctrl->nr_changed_ns = U32_MAX;
+	}
+out_unlock:
+	mutex_unlock(&ctrl->lock);
+}

Thanks,
-- Daniel