nvmf: delay remove subsystem cb until no qpairs remain

We cannot solely rely on the qpair_ctx->count reaching
0, because qpairs that are in process of being
disconnected will immediately invoke the qpair
disconnect cb.

Instead, we need to wait until the poll group
no longer has any qpairs remaining on the subsystem.

Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I977747d367d14a4bf60f66a1147b3d75679e5179
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/7870
Community-CI: Broadcom CI
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: <dongx.yi@intel.com>
This commit is contained in:
Jim Harris 2021-05-13 15:31:54 +00:00 committed by Tomasz Zawadzki
parent d92a67be47
commit da9766336e

View File

@ -1364,6 +1364,8 @@ fini:
}
}
static void nvmf_poll_group_remove_subsystem_msg(void *ctx);
static void
remove_subsystem_qpair_cb(void *ctx)
{
@ -1372,7 +1374,12 @@ remove_subsystem_qpair_cb(void *ctx)
assert(qpair_ctx->count > 0);
qpair_ctx->count--;
if (qpair_ctx->count == 0) {
_nvmf_poll_group_remove_subsystem_cb(ctx, 0);
/* All of the asynchronous callbacks for this context have been
* completed. Call nvmf_poll_group_remove_subsystem_msg() again
* to check if all associated qpairs for this subsystem have
* been removed from the poll group.
*/
nvmf_poll_group_remove_subsystem_msg(ctx);
}
}
@ -1383,6 +1390,7 @@ nvmf_poll_group_remove_subsystem_msg(void *ctx)
struct spdk_nvmf_subsystem *subsystem;
struct spdk_nvmf_poll_group *group;
struct nvmf_qpair_disconnect_many_ctx *qpair_ctx = ctx;
bool qpairs_found = false;
int rc = 0;
group = qpair_ctx->group;
@ -1390,12 +1398,13 @@ nvmf_poll_group_remove_subsystem_msg(void *ctx)
/* Initialize count to 1. This acts like a ref count, to ensure that if spdk_nvmf_qpair_disconnect
* immediately invokes the callback (i.e. the qpairs is already in process of being disconnected)
* that we don't prematurely call _nvmf_poll_group_remove_subsystem_cb() before we've
* iterated the full list of qpairs.
* that we don't recursively call nvmf_poll_group_remove_subsystem_msg before we've iterated the
* full list of qpairs.
*/
qpair_ctx->count = 1;
TAILQ_FOREACH_SAFE(qpair, &group->qpairs, link, qpair_tmp) {
if ((qpair->ctrlr != NULL) && (qpair->ctrlr->subsys == subsystem)) {
qpairs_found = true;
qpair_ctx->count++;
rc = spdk_nvmf_qpair_disconnect(qpair, remove_subsystem_qpair_cb, ctx);
if (rc) {
@ -1405,8 +1414,19 @@ nvmf_poll_group_remove_subsystem_msg(void *ctx)
}
qpair_ctx->count--;
if (!qpairs_found) {
_nvmf_poll_group_remove_subsystem_cb(ctx, 0);
return;
}
if (qpair_ctx->count == 0 || rc) {
_nvmf_poll_group_remove_subsystem_cb(ctx, rc);
/* If count == 0, it means there were some qpairs in the poll group but they
* were already in process of being disconnected. So we send a message to this
* same thread so that this function executes again later. We won't actually
* invoke the remove_subsystem_cb until all of the qpairs are actually removed
* from the poll group.
*/
spdk_thread_send_msg(spdk_get_thread(), nvmf_poll_group_remove_subsystem_msg, ctx);
}
}