When IO abort times out during eh_abort or a flush operation is performed while
abort is pending, the driver is not cleaning up the IO and thus not reducing
the IO reference count. With this change, as part of explicit logout, the IO is
cleaned up.
Signed-off-by: Bhanu Prakash Gollapudi <bprakash@xxxxxxxxxxxx>
---
drivers/scsi/bnx2fc/bnx2fc_io.c | 106 +++++++++++++++++++++----------------
drivers/scsi/bnx2fc/bnx2fc_tgt.c | 11 ++++-
2 files changed, 70 insertions(+), 47 deletions(-)
diff --git a/drivers/scsi/bnx2fc/bnx2fc_io.c b/drivers/scsi/bnx2fc/bnx2fc_io.c
index 84a78af..4843b42 100644
--- a/drivers/scsi/bnx2fc/bnx2fc_io.c
+++ b/drivers/scsi/bnx2fc/bnx2fc_io.c
@@ -1089,6 +1089,48 @@ int bnx2fc_eh_device_reset(struct scsi_cmnd *sc_cmd)
return bnx2fc_initiate_tmf(sc_cmd, FCP_TMF_LUN_RESET);
}
+int bnx2fc_expl_logo(struct fc_lport *lport, struct bnx2fc_cmd *io_req)
+{
+ struct bnx2fc_rport *tgt = io_req->tgt;
+ struct fc_rport_priv *rdata = tgt->rdata;
+ int logo_issued;
+ int rc = SUCCESS;
+ int wait_cnt = 0;
+
+ BNX2FC_IO_DBG(io_req, "Expl logo - tgt flags = 0x%lx\n",
+ tgt->flags);
+ logo_issued = test_and_set_bit(BNX2FC_FLAG_EXPL_LOGO,
+ &tgt->flags);
+ io_req->wait_for_comp = 1;
+ bnx2fc_initiate_cleanup(io_req);
+
+ spin_unlock_bh(&tgt->tgt_lock);
+
+ wait_for_completion(&io_req->tm_done);
+
+ io_req->wait_for_comp = 0;
+ /*
+ * release the reference taken in eh_abort to allow the
+ * target to re-login after flushing IOs
+ */
+ kref_put(&io_req->refcount, bnx2fc_cmd_release);
+
+ if (!logo_issued) {
+ clear_bit(BNX2FC_FLAG_SESSION_READY, &tgt->flags);
+ mutex_lock(&lport->disc.disc_mutex);
+ lport->tt.rport_logoff(rdata);
+ mutex_unlock(&lport->disc.disc_mutex);
+ do {
+ msleep(BNX2FC_RELOGIN_WAIT_TIME);
+ if (wait_cnt++ > BNX2FC_RELOGIN_WAIT_CNT) {
+ rc = FAILED;
+ break;
+ }
+ } while (!test_bit(BNX2FC_FLAG_SESSION_READY, &tgt->flags));
+ }
+ spin_lock_bh(&tgt->tgt_lock);
+ return rc;
+}
/**
* bnx2fc_eh_abort - eh_abort_handler api to abort an outstanding
* SCSI command
@@ -1103,10 +1145,7 @@ int bnx2fc_eh_abort(struct scsi_cmnd *sc_cmd)
struct fc_rport_libfc_priv *rp = rport->dd_data;
struct bnx2fc_cmd *io_req;
struct fc_lport *lport;
- struct fc_rport_priv *rdata;
struct bnx2fc_rport *tgt;
- int logo_issued;
- int wait_cnt = 0;
int rc = FAILED;
@@ -1183,58 +1222,31 @@ int bnx2fc_eh_abort(struct scsi_cmnd *sc_cmd)
list_add_tail(&io_req->link, &tgt->io_retire_queue);
init_completion(&io_req->tm_done);
- io_req->wait_for_comp = 1;
- if (!test_and_set_bit(BNX2FC_FLAG_ISSUE_ABTS, &io_req->req_flags)) {
- /* Cancel the current timer running on this io_req */
- if (cancel_delayed_work(&io_req->timeout_work))
- kref_put(&io_req->refcount,
- bnx2fc_cmd_release); /* drop timer hold */
- set_bit(BNX2FC_FLAG_EH_ABORT, &io_req->req_flags);
- rc = bnx2fc_initiate_abts(io_req);
- } else {
+ if (test_and_set_bit(BNX2FC_FLAG_ISSUE_ABTS, &io_req->req_flags)) {
printk(KERN_ERR PFX "eh_abort: io_req (xid = 0x%x) "
"already in abts processing\n", io_req->xid);
if (cancel_delayed_work(&io_req->timeout_work))
kref_put(&io_req->refcount,
bnx2fc_cmd_release); /* drop timer hold */
- bnx2fc_initiate_cleanup(io_req);
+ rc = bnx2fc_expl_logo(lport, io_req);
+ goto out;
+ }
+ /* Cancel the current timer running on this io_req */
+ if (cancel_delayed_work(&io_req->timeout_work))
+ kref_put(&io_req->refcount,
+ bnx2fc_cmd_release); /* drop timer hold */
+ set_bit(BNX2FC_FLAG_EH_ABORT, &io_req->req_flags);
+ io_req->wait_for_comp = 1;
+ rc = bnx2fc_initiate_abts(io_req);
+ if (rc == FAILED) {
+ bnx2fc_initiate_cleanup(io_req);
spin_unlock_bh(&tgt->tgt_lock);
-
wait_for_completion(&io_req->tm_done);
-
spin_lock_bh(&tgt->tgt_lock);
io_req->wait_for_comp = 0;
- rdata = io_req->tgt->rdata;
- logo_issued = test_and_set_bit(BNX2FC_FLAG_EXPL_LOGO,
- &tgt->flags);
- kref_put(&io_req->refcount, bnx2fc_cmd_release);
- spin_unlock_bh(&tgt->tgt_lock);
-
- if (!logo_issued) {
- BNX2FC_IO_DBG(io_req, "Expl logo - tgt flags = 0x%lx\n",
- tgt->flags);
- mutex_lock(&lport->disc.disc_mutex);
- lport->tt.rport_logoff(rdata);
- mutex_unlock(&lport->disc.disc_mutex);
- do {
- msleep(BNX2FC_RELOGIN_WAIT_TIME);
- /*
- * If session not recovered, let SCSI-ml
- * escalate error recovery.
- */
- if (wait_cnt++ > BNX2FC_RELOGIN_WAIT_CNT)
- return FAILED;
- } while (!test_bit(BNX2FC_FLAG_SESSION_READY,
- &tgt->flags));
- }
- return SUCCESS;
- }
- if (rc == FAILED) {
- kref_put(&io_req->refcount, bnx2fc_cmd_release);
- spin_unlock_bh(&tgt->tgt_lock);
- return rc;
+ goto done;
}
spin_unlock_bh(&tgt->tgt_lock);
@@ -1247,7 +1259,8 @@ int bnx2fc_eh_abort(struct scsi_cmnd *sc_cmd)
/* Let the scsi-ml try to recover this command */
printk(KERN_ERR PFX "abort failed, xid = 0x%x\n",
io_req->xid);
- rc = FAILED;
+ rc = bnx2fc_expl_logo(lport, io_req);
+ goto out;
} else {
/*
* We come here even when there was a race condition
@@ -1259,9 +1272,10 @@ int bnx2fc_eh_abort(struct scsi_cmnd *sc_cmd)
bnx2fc_scsi_done(io_req, DID_ABORT);
kref_put(&io_req->refcount, bnx2fc_cmd_release);
}
-
+done:
/* release the reference taken in eh_abort */
kref_put(&io_req->refcount, bnx2fc_cmd_release);
+out:
spin_unlock_bh(&tgt->tgt_lock);
return rc;
}
diff --git a/drivers/scsi/bnx2fc/bnx2fc_tgt.c b/drivers/scsi/bnx2fc/bnx2fc_tgt.c
index c1800b5..d3ee231 100644
--- a/drivers/scsi/bnx2fc/bnx2fc_tgt.c
+++ b/drivers/scsi/bnx2fc/bnx2fc_tgt.c
@@ -213,8 +213,17 @@ void bnx2fc_flush_active_ios(struct bnx2fc_rport *tgt)
BNX2FC_IO_DBG(io_req, "retire_queue flush\n");
- if (cancel_delayed_work(&io_req->timeout_work))
+ if (cancel_delayed_work(&io_req->timeout_work)) {
+ if (test_and_clear_bit(BNX2FC_FLAG_EH_ABORT,
+ &io_req->req_flags)) {
+ /* Handle eh_abort timeout */
+ BNX2FC_IO_DBG(io_req, "eh_abort for IO "
+ "in retire_q\n");
+ if (io_req->wait_for_comp)
+ complete(&io_req->tm_done);
+ }
kref_put(&io_req->refcount, bnx2fc_cmd_release);
+ }
clear_bit(BNX2FC_FLAG_ISSUE_RRQ, &io_req->req_flags);
}
--
1.7.0.6
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
