[PATCH 3/3] nvme/multipath: fix stale ana state for namespaces just added by scan work

Anton Eidelman anton at lightbitslabs.com
Sat Sep 11 18:07:31 PDT 2021


Scan work initially adds new namespaces to ctrl->namespaces TAIL.
They make the list unordered temporarily until nvme_scan_work()
finally sorts the list.

In case nvme_update_ana_state() runs while the list is unsorted,
the recently added namespaces are missed and their ana state
may remain not updated forever if timing between scan work and ana work
is unfortunate, e.g.
Initial state: namespaces = {2, 3}
scan_work: adds nsid=1: namespaces = {2, 3, 1}
scan_work: finds nsid=1 is still Inaccessible
ana_work: log page has nsids = {1, 2, 3, 4}, all Optimized.
ana_work: updates nsids {2, 3} but fails to find nsid=1 in namespaces.
scan_work: adds nsid=4: namespaces = {2, 3, 1, 4}
scan_work: finds nsid=4 is Optimized: sets it live.
scan_work: completes an sorts namespaces = {1, 2, 3, 4}
Result: nsid=1 will remain in Inaccessible state.

Solution:
In order to preserve the way ctrl->namespaces is updated and sorted,
make nvme_update_ana_state() deal with the case where ctrl->namespaces
is not fully sorted and has new namespaces appended with potentially
lower nsids.
nvme_update_ana_state() keeps track of the nsid seen in the list,
detects the unsorted case (rare), and restarts scanning of desc->nsids.

Signed-off-by: Anton Eidelman <anton at lightbitslabs.com>
---
 drivers/nvme/host/multipath.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/drivers/nvme/host/multipath.c b/drivers/nvme/host/multipath.c
index a51561d67b93..1ad8dc8adb86 100644
--- a/drivers/nvme/host/multipath.c
+++ b/drivers/nvme/host/multipath.c
@@ -587,6 +587,7 @@ static int nvme_update_ana_state(struct nvme_ctrl *ctrl,
 	u32 nr_nsids = le32_to_cpu(desc->nnsids), n = 0;
 	unsigned *nr_change_groups = data;
 	struct nvme_ns *ns;
+	unsigned int last_ns_nsid = 0;
 
 	dev_dbg(ctrl->device, "ANA group %d: %s.\n",
 			le32_to_cpu(desc->grpid),
@@ -603,6 +604,11 @@ static int nvme_update_ana_state(struct nvme_ctrl *ctrl,
 		unsigned int nsid;
 		unsigned int ns_nsid = ns->head->ns_id;
 
+		if (ns_nsid < last_ns_nsid) {
+			/* Detected unsorted ctrl->namespaces: re-scan desc->nsids */
+			last_ns_nsid = ns_nsid;
+			n = 0;
+		}
 again:
 		nsid = le32_to_cpu(desc->nsids[n]);
 		if (ns_nsid < nsid)
-- 
2.25.1


-- 


*Lightbits Labs**
*Lead the cloud-native data center
transformation by 
delivering *scalable *and *efficient *software
defined storage that is 
*easy *to consume.



*This message is sent in confidence for the addressee 
only.  It
may contain legally privileged information. The contents are not 
to be
disclosed to anyone other than the addressee. Unauthorized recipients 
are
requested to preserve this confidentiality, advise the sender 
immediately of
any error in transmission and delete the email from their 
systems.*




More information about the Linux-nvme mailing list