[SCSI] fix for fc transport recursion problem.
In the scenario that a link was broken, the devloss timer for each
rport was expire at roughly the same time, causing lots of "delete"
workqueue items being queued. Depth is dependent upon the number of
rports that were on the link.
The rport target remove calls were calling flush_scheduled_work(),
which would interrupt the stream, and start the next workqueue item,
which did the same thing, and so on until recursion depth was large.
This fix stops the recursion in the initial delete path, and pushes it
off to a host-level work item that reaps the dead rports.
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
diff --git a/drivers/scsi/scsi_transport_fc.c b/drivers/scsi/scsi_transport_fc.c
index 6cd5931..2a1a99a 100644
--- a/drivers/scsi/scsi_transport_fc.c
+++ b/drivers/scsi/scsi_transport_fc.c
@@ -105,6 +105,7 @@
{ FC_PORTSTATE_LINKDOWN, "Linkdown" },
{ FC_PORTSTATE_ERROR, "Error" },
{ FC_PORTSTATE_LOOPBACK, "Loopback" },
+ { FC_PORTSTATE_DELETED, "Deleted" },
};
fc_enum_name_search(port_state, fc_port_state, fc_port_state_names)
#define FC_PORTSTATE_MAX_NAMELEN 20
@@ -211,6 +212,7 @@
#define FC_MGMTSRVR_PORTID 0x00000a
+static void fc_shost_remove_rports(void *data);
static void fc_timeout_deleted_rport(void *data);
static void fc_scsi_scan_rport(void *data);
static void fc_rport_terminate(struct fc_rport *rport);
@@ -318,6 +320,8 @@
fc_host_next_rport_number(shost) = 0;
fc_host_next_target_id(shost) = 0;
+ fc_host_flags(shost) = 0;
+ INIT_WORK(&fc_host_rport_del_work(shost), fc_shost_remove_rports, shost);
return 0;
}
@@ -387,6 +391,7 @@
struct fc_internal *i = to_fc_internal(shost->transportt); \
if ((i->f->get_rport_##field) && \
!((rport->port_state == FC_PORTSTATE_BLOCKED) || \
+ (rport->port_state == FC_PORTSTATE_DELETED) || \
(rport->port_state == FC_PORTSTATE_NOTPRESENT))) \
i->f->get_rport_##field(rport); \
return snprintf(buf, sz, format_string, cast rport->field); \
@@ -402,6 +407,7 @@
struct Scsi_Host *shost = rport_to_shost(rport); \
struct fc_internal *i = to_fc_internal(shost->transportt); \
if ((rport->port_state == FC_PORTSTATE_BLOCKED) || \
+ (rport->port_state == FC_PORTSTATE_DELETED) || \
(rport->port_state == FC_PORTSTATE_NOTPRESENT)) \
return -EBUSY; \
val = simple_strtoul(buf, NULL, 0); \
@@ -519,6 +525,7 @@
struct Scsi_Host *shost = rport_to_shost(rport);
struct fc_internal *i = to_fc_internal(shost->transportt);
if ((rport->port_state == FC_PORTSTATE_BLOCKED) ||
+ (rport->port_state == FC_PORTSTATE_DELETED) ||
(rport->port_state == FC_PORTSTATE_NOTPRESENT))
return -EBUSY;
val = simple_strtoul(buf, NULL, 0);
@@ -1769,7 +1776,7 @@
rport->maxframe_size = -1;
rport->supported_classes = FC_COS_UNSPECIFIED;
rport->roles = FC_RPORT_ROLE_UNKNOWN;
- rport->port_state = FC_PORTSTATE_NOTPRESENT;
+ rport->port_state = FC_PORTSTATE_DELETED;
/* remove the identifiers that aren't used in the consisting binding */
switch (fc_host_tgtid_bind_type(shost)) {
@@ -1789,14 +1796,23 @@
break;
}
- spin_unlock_irqrestore(shost->host_lock, flags);
-
/*
* As this only occurs if the remote port (scsi target)
* went away and didn't come back - we'll remove
* all attached scsi devices.
+ *
+ * We'll schedule the shost work item to perform the actual removal
+ * to avoid recursion in the different flush calls if we perform
+ * the removal in each target - and there are lots of targets
+ * whose timeouts fire at the same time.
*/
- fc_rport_tgt_remove(rport);
+
+ if ( !(fc_host_flags(shost) & FC_SHOST_RPORT_DEL_SCHEDULED)) {
+ fc_host_flags(shost) |= FC_SHOST_RPORT_DEL_SCHEDULED;
+ scsi_queue_work(shost, &fc_host_rport_del_work(shost));
+ }
+
+ spin_unlock_irqrestore(shost->host_lock, flags);
}
/**
@@ -1818,6 +1834,41 @@
}
+/**
+ * fc_shost_remove_rports - called to remove all rports that are marked
+ * as in a deleted (not connected) state.
+ *
+ * @data: shost whose rports are to be looked at
+ **/
+static void
+fc_shost_remove_rports(void *data)
+{
+ struct Scsi_Host *shost = (struct Scsi_Host *)data;
+ struct fc_rport *rport, *next_rport;
+ unsigned long flags;
+
+ spin_lock_irqsave(shost->host_lock, flags);
+ while (fc_host_flags(shost) & FC_SHOST_RPORT_DEL_SCHEDULED) {
+
+ fc_host_flags(shost) &= ~FC_SHOST_RPORT_DEL_SCHEDULED;
+
+restart_search:
+ list_for_each_entry_safe(rport, next_rport,
+ &fc_host_rport_bindings(shost), peers) {
+ if (rport->port_state == FC_PORTSTATE_DELETED) {
+ rport->port_state = FC_PORTSTATE_NOTPRESENT;
+ spin_unlock_irqrestore(shost->host_lock, flags);
+ fc_rport_tgt_remove(rport);
+ spin_lock_irqsave(shost->host_lock, flags);
+ goto restart_search;
+ }
+ }
+
+ }
+ spin_unlock_irqrestore(shost->host_lock, flags);
+}
+
+
MODULE_AUTHOR("Martin Hicks");
MODULE_DESCRIPTION("FC Transport Attributes");
MODULE_LICENSE("GPL");
diff --git a/include/scsi/scsi_transport_fc.h b/include/scsi/scsi_transport_fc.h
index fac547d..394f14a 100644
--- a/include/scsi/scsi_transport_fc.h
+++ b/include/scsi/scsi_transport_fc.h
@@ -79,6 +79,7 @@
FC_PORTSTATE_LINKDOWN,
FC_PORTSTATE_ERROR,
FC_PORTSTATE_LOOPBACK,
+ FC_PORTSTATE_DELETED,
};
@@ -325,8 +326,14 @@
struct list_head rport_bindings;
u32 next_rport_number;
u32 next_target_id;
+ u8 flags;
+ struct work_struct rport_del_work;
};
+/* values for struct fc_host_attrs "flags" field: */
+#define FC_SHOST_RPORT_DEL_SCHEDULED 0x01
+
+
#define fc_host_node_name(x) \
(((struct fc_host_attrs *)(x)->shost_data)->node_name)
#define fc_host_port_name(x) \
@@ -365,6 +372,10 @@
(((struct fc_host_attrs *)(x)->shost_data)->next_rport_number)
#define fc_host_next_target_id(x) \
(((struct fc_host_attrs *)(x)->shost_data)->next_target_id)
+#define fc_host_flags(x) \
+ (((struct fc_host_attrs *)(x)->shost_data)->flags)
+#define fc_host_rport_del_work(x) \
+ (((struct fc_host_attrs *)(x)->shost_data)->rport_del_work)
/* The functions by which the transport class and the driver communicate */