These changes are the raw update to linux-4.4.6-rt14. Kernel sources
[kvmfornfv.git] / kernel / drivers / target / target_core_tpg.c
index 47f0644..5fb9dd7 100644 (file)
 #include <linux/export.h>
 #include <net/sock.h>
 #include <net/tcp.h>
-#include <scsi/scsi.h>
-#include <scsi/scsi_cmnd.h>
+#include <scsi/scsi_proto.h>
 
 #include <target/target_core_base.h>
 #include <target/target_core_backend.h>
 #include <target/target_core_fabric.h>
 
 #include "target_core_internal.h"
+#include "target_core_alua.h"
 #include "target_core_pr.h"
+#include "target_core_ua.h"
 
 extern struct se_device *g_lun0_dev;
 
 static DEFINE_SPINLOCK(tpg_lock);
 static LIST_HEAD(tpg_list);
 
-/*     core_clear_initiator_node_from_tpg():
- *
- *
- */
-static void core_clear_initiator_node_from_tpg(
-       struct se_node_acl *nacl,
-       struct se_portal_group *tpg)
-{
-       int i;
-       struct se_dev_entry *deve;
-       struct se_lun *lun;
-
-       spin_lock_irq(&nacl->device_list_lock);
-       for (i = 0; i < TRANSPORT_MAX_LUNS_PER_TPG; i++) {
-               deve = nacl->device_list[i];
-
-               if (!(deve->lun_flags & TRANSPORT_LUNFLAGS_INITIATOR_ACCESS))
-                       continue;
-
-               if (!deve->se_lun) {
-                       pr_err("%s device entries device pointer is"
-                               " NULL, but Initiator has access.\n",
-                               tpg->se_tpg_tfo->get_fabric_name());
-                       continue;
-               }
-
-               lun = deve->se_lun;
-               spin_unlock_irq(&nacl->device_list_lock);
-               core_disable_device_list_for_node(lun, NULL, deve->mapped_lun,
-                       TRANSPORT_LUNFLAGS_NO_ACCESS, nacl, tpg);
-
-               spin_lock_irq(&nacl->device_list_lock);
-       }
-       spin_unlock_irq(&nacl->device_list_lock);
-}
-
 /*     __core_tpg_get_initiator_node_acl():
  *
- *     spin_lock_bh(&tpg->acl_node_lock); must be held when calling
+ *     mutex_lock(&tpg->acl_node_mutex); must be held when calling
  */
 struct se_node_acl *__core_tpg_get_initiator_node_acl(
        struct se_portal_group *tpg,
@@ -111,36 +76,50 @@ struct se_node_acl *core_tpg_get_initiator_node_acl(
 {
        struct se_node_acl *acl;
 
-       spin_lock_irq(&tpg->acl_node_lock);
+       mutex_lock(&tpg->acl_node_mutex);
        acl = __core_tpg_get_initiator_node_acl(tpg, initiatorname);
-       spin_unlock_irq(&tpg->acl_node_lock);
+       mutex_unlock(&tpg->acl_node_mutex);
 
        return acl;
 }
 EXPORT_SYMBOL(core_tpg_get_initiator_node_acl);
 
+void core_allocate_nexus_loss_ua(
+       struct se_node_acl *nacl)
+{
+       struct se_dev_entry *deve;
+
+       if (!nacl)
+               return;
+
+       rcu_read_lock();
+       hlist_for_each_entry_rcu(deve, &nacl->lun_entry_hlist, link)
+               core_scsi3_ua_allocate(deve, 0x29,
+                       ASCQ_29H_NEXUS_LOSS_OCCURRED);
+       rcu_read_unlock();
+}
+EXPORT_SYMBOL(core_allocate_nexus_loss_ua);
+
 /*     core_tpg_add_node_to_devs():
  *
  *
  */
 void core_tpg_add_node_to_devs(
        struct se_node_acl *acl,
-       struct se_portal_group *tpg)
+       struct se_portal_group *tpg,
+       struct se_lun *lun_orig)
 {
-       int i = 0;
        u32 lun_access = 0;
        struct se_lun *lun;
        struct se_device *dev;
 
-       spin_lock(&tpg->tpg_lun_lock);
-       for (i = 0; i < TRANSPORT_MAX_LUNS_PER_TPG; i++) {
-               lun = tpg->tpg_lun_list[i];
-               if (lun->lun_status != TRANSPORT_LUN_STATUS_ACTIVE)
+       mutex_lock(&tpg->tpg_lun_mutex);
+       hlist_for_each_entry_rcu(lun, &tpg->tpg_lun_hlist, link) {
+               if (lun_orig && lun != lun_orig)
                        continue;
 
-               spin_unlock(&tpg->tpg_lun_lock);
-
-               dev = lun->lun_se_dev;
+               dev = rcu_dereference_check(lun->lun_se_dev,
+                                           lockdep_is_held(&tpg->tpg_lun_mutex));
                /*
                 * By default in LIO-Target $FABRIC_MOD,
                 * demo_mode_write_protect is ON, or READ_ONLY;
@@ -158,7 +137,7 @@ void core_tpg_add_node_to_devs(
                                lun_access = TRANSPORT_LUNFLAGS_READ_WRITE;
                }
 
-               pr_debug("TARGET_CORE[%s]->TPG[%u]_LUN[%u] - Adding %s"
+               pr_debug("TARGET_CORE[%s]->TPG[%u]_LUN[%llu] - Adding %s"
                        " access for LUN in Demo Mode\n",
                        tpg->se_tpg_tfo->get_fabric_name(),
                        tpg->se_tpg_tfo->tpg_get_tag(tpg), lun->unpacked_lun,
@@ -166,7 +145,7 @@ void core_tpg_add_node_to_devs(
                        "READ-WRITE" : "READ-ONLY");
 
                core_enable_device_list_for_node(lun, NULL, lun->unpacked_lun,
-                               lun_access, acl, tpg);
+                                                lun_access, acl, tpg);
                /*
                 * Check to see if there are any existing persistent reservation
                 * APTPL pre-registrations that need to be enabled for this dynamic
@@ -174,9 +153,8 @@ void core_tpg_add_node_to_devs(
                 */
                core_scsi3_check_aptpl_registration(dev, tpg, lun, acl,
                                                    lun->unpacked_lun);
-               spin_lock(&tpg->tpg_lun_lock);
        }
-       spin_unlock(&tpg->tpg_lun_lock);
+       mutex_unlock(&tpg->tpg_lun_mutex);
 }
 
 /*      core_set_queue_depth_for_node():
@@ -197,67 +175,63 @@ static int core_set_queue_depth_for_node(
        return 0;
 }
 
-void array_free(void *array, int n)
+static struct se_node_acl *target_alloc_node_acl(struct se_portal_group *tpg,
+               const unsigned char *initiatorname)
 {
-       void **a = array;
-       int i;
+       struct se_node_acl *acl;
 
-       for (i = 0; i < n; i++)
-               kfree(a[i]);
-       kfree(a);
-}
+       acl = kzalloc(max(sizeof(*acl), tpg->se_tpg_tfo->node_acl_size),
+                       GFP_KERNEL);
+       if (!acl)
+               return NULL;
 
-static void *array_zalloc(int n, size_t size, gfp_t flags)
-{
-       void **a;
-       int i;
+       INIT_LIST_HEAD(&acl->acl_list);
+       INIT_LIST_HEAD(&acl->acl_sess_list);
+       INIT_HLIST_HEAD(&acl->lun_entry_hlist);
+       kref_init(&acl->acl_kref);
+       init_completion(&acl->acl_free_comp);
+       spin_lock_init(&acl->nacl_sess_lock);
+       mutex_init(&acl->lun_entry_mutex);
+       atomic_set(&acl->acl_pr_ref_count, 0);
+       if (tpg->se_tpg_tfo->tpg_get_default_depth)
+               acl->queue_depth = tpg->se_tpg_tfo->tpg_get_default_depth(tpg);
+       else
+               acl->queue_depth = 1;
+       snprintf(acl->initiatorname, TRANSPORT_IQN_LEN, "%s", initiatorname);
+       acl->se_tpg = tpg;
+       acl->acl_index = scsi_get_new_index(SCSI_AUTH_INTR_INDEX);
 
-       a = kzalloc(n * sizeof(void*), flags);
-       if (!a)
-               return NULL;
-       for (i = 0; i < n; i++) {
-               a[i] = kzalloc(size, flags);
-               if (!a[i]) {
-                       array_free(a, n);
-                       return NULL;
-               }
-       }
-       return a;
+       tpg->se_tpg_tfo->set_default_node_attributes(acl);
+
+       if (core_set_queue_depth_for_node(tpg, acl) < 0)
+               goto out_free_acl;
+
+       return acl;
+
+out_free_acl:
+       kfree(acl);
+       return NULL;
 }
 
-/*      core_create_device_list_for_node():
- *
- *
- */
-static int core_create_device_list_for_node(struct se_node_acl *nacl)
+static void target_add_node_acl(struct se_node_acl *acl)
 {
-       struct se_dev_entry *deve;
-       int i;
-
-       nacl->device_list = array_zalloc(TRANSPORT_MAX_LUNS_PER_TPG,
-                       sizeof(struct se_dev_entry), GFP_KERNEL);
-       if (!nacl->device_list) {
-               pr_err("Unable to allocate memory for"
-                       " struct se_node_acl->device_list\n");
-               return -ENOMEM;
-       }
-       for (i = 0; i < TRANSPORT_MAX_LUNS_PER_TPG; i++) {
-               deve = nacl->device_list[i];
-
-               atomic_set(&deve->ua_count, 0);
-               atomic_set(&deve->pr_ref_count, 0);
-               spin_lock_init(&deve->ua_lock);
-               INIT_LIST_HEAD(&deve->alua_port_list);
-               INIT_LIST_HEAD(&deve->ua_list);
-       }
+       struct se_portal_group *tpg = acl->se_tpg;
 
-       return 0;
+       mutex_lock(&tpg->acl_node_mutex);
+       list_add_tail(&acl->acl_list, &tpg->acl_node_list);
+       tpg->num_node_acls++;
+       mutex_unlock(&tpg->acl_node_mutex);
+
+       pr_debug("%s_TPG[%hu] - Added %s ACL with TCQ Depth: %d for %s"
+               " Initiator Node: %s\n",
+               tpg->se_tpg_tfo->get_fabric_name(),
+               tpg->se_tpg_tfo->tpg_get_tag(tpg),
+               acl->dynamic_node_acl ? "DYNAMIC" : "",
+               acl->queue_depth,
+               tpg->se_tpg_tfo->get_fabric_name(),
+               acl->initiatorname);
 }
 
-/*     core_tpg_check_initiator_node_acl()
- *
- *
- */
 struct se_node_acl *core_tpg_check_initiator_node_acl(
        struct se_portal_group *tpg,
        unsigned char *initiatorname)
@@ -271,35 +245,11 @@ struct se_node_acl *core_tpg_check_initiator_node_acl(
        if (!tpg->se_tpg_tfo->tpg_check_demo_mode(tpg))
                return NULL;
 
-       acl =  tpg->se_tpg_tfo->tpg_alloc_fabric_acl(tpg);
+       acl = target_alloc_node_acl(tpg, initiatorname);
        if (!acl)
                return NULL;
-
-       INIT_LIST_HEAD(&acl->acl_list);
-       INIT_LIST_HEAD(&acl->acl_sess_list);
-       kref_init(&acl->acl_kref);
-       init_completion(&acl->acl_free_comp);
-       spin_lock_init(&acl->device_list_lock);
-       spin_lock_init(&acl->nacl_sess_lock);
-       atomic_set(&acl->acl_pr_ref_count, 0);
-       acl->queue_depth = tpg->se_tpg_tfo->tpg_get_default_depth(tpg);
-       snprintf(acl->initiatorname, TRANSPORT_IQN_LEN, "%s", initiatorname);
-       acl->se_tpg = tpg;
-       acl->acl_index = scsi_get_new_index(SCSI_AUTH_INTR_INDEX);
        acl->dynamic_node_acl = 1;
 
-       tpg->se_tpg_tfo->set_default_node_attributes(acl);
-
-       if (core_create_device_list_for_node(acl) < 0) {
-               tpg->se_tpg_tfo->tpg_release_fabric_acl(tpg, acl);
-               return NULL;
-       }
-
-       if (core_set_queue_depth_for_node(tpg, acl) < 0) {
-               core_free_device_list_for_node(acl, tpg);
-               tpg->se_tpg_tfo->tpg_release_fabric_acl(tpg, acl);
-               return NULL;
-       }
        /*
         * Here we only create demo-mode MappedLUNs from the active
         * TPG LUNs if the fabric is not explicitly asking for
@@ -307,18 +257,9 @@ struct se_node_acl *core_tpg_check_initiator_node_acl(
         */
        if ((tpg->se_tpg_tfo->tpg_check_demo_mode_login_only == NULL) ||
            (tpg->se_tpg_tfo->tpg_check_demo_mode_login_only(tpg) != 1))
-               core_tpg_add_node_to_devs(acl, tpg);
-
-       spin_lock_irq(&tpg->acl_node_lock);
-       list_add_tail(&acl->acl_list, &tpg->acl_node_list);
-       tpg->num_node_acls++;
-       spin_unlock_irq(&tpg->acl_node_lock);
-
-       pr_debug("%s_TPG[%u] - Added DYNAMIC ACL with TCQ Depth: %d for %s"
-               " Initiator Node: %s\n", tpg->se_tpg_tfo->get_fabric_name(),
-               tpg->se_tpg_tfo->tpg_get_tag(tpg), acl->queue_depth,
-               tpg->se_tpg_tfo->get_fabric_name(), initiatorname);
+               core_tpg_add_node_to_devs(acl, tpg, NULL);
 
+       target_add_node_acl(acl);
        return acl;
 }
 EXPORT_SYMBOL(core_tpg_check_initiator_node_acl);
@@ -329,40 +270,13 @@ void core_tpg_wait_for_nacl_pr_ref(struct se_node_acl *nacl)
                cpu_relax();
 }
 
-void core_tpg_clear_object_luns(struct se_portal_group *tpg)
-{
-       int i;
-       struct se_lun *lun;
-
-       spin_lock(&tpg->tpg_lun_lock);
-       for (i = 0; i < TRANSPORT_MAX_LUNS_PER_TPG; i++) {
-               lun = tpg->tpg_lun_list[i];
-
-               if ((lun->lun_status != TRANSPORT_LUN_STATUS_ACTIVE) ||
-                   (lun->lun_se_dev == NULL))
-                       continue;
-
-               spin_unlock(&tpg->tpg_lun_lock);
-               core_dev_del_lun(tpg, lun);
-               spin_lock(&tpg->tpg_lun_lock);
-       }
-       spin_unlock(&tpg->tpg_lun_lock);
-}
-EXPORT_SYMBOL(core_tpg_clear_object_luns);
-
-/*     core_tpg_add_initiator_node_acl():
- *
- *
- */
 struct se_node_acl *core_tpg_add_initiator_node_acl(
        struct se_portal_group *tpg,
-       struct se_node_acl *se_nacl,
-       const char *initiatorname,
-       u32 queue_depth)
+       const char *initiatorname)
 {
-       struct se_node_acl *acl = NULL;
+       struct se_node_acl *acl;
 
-       spin_lock_irq(&tpg->acl_node_lock);
+       mutex_lock(&tpg->acl_node_mutex);
        acl = __core_tpg_get_initiator_node_acl(tpg, initiatorname);
        if (acl) {
                if (acl->dynamic_node_acl) {
@@ -370,99 +284,42 @@ struct se_node_acl *core_tpg_add_initiator_node_acl(
                        pr_debug("%s_TPG[%u] - Replacing dynamic ACL"
                                " for %s\n", tpg->se_tpg_tfo->get_fabric_name(),
                                tpg->se_tpg_tfo->tpg_get_tag(tpg), initiatorname);
-                       spin_unlock_irq(&tpg->acl_node_lock);
-                       /*
-                        * Release the locally allocated struct se_node_acl
-                        * because * core_tpg_add_initiator_node_acl() returned
-                        * a pointer to an existing demo mode node ACL.
-                        */
-                       if (se_nacl)
-                               tpg->se_tpg_tfo->tpg_release_fabric_acl(tpg,
-                                                       se_nacl);
-                       goto done;
+                       mutex_unlock(&tpg->acl_node_mutex);
+                       return acl;
                }
 
                pr_err("ACL entry for %s Initiator"
                        " Node %s already exists for TPG %u, ignoring"
                        " request.\n",  tpg->se_tpg_tfo->get_fabric_name(),
                        initiatorname, tpg->se_tpg_tfo->tpg_get_tag(tpg));
-               spin_unlock_irq(&tpg->acl_node_lock);
+               mutex_unlock(&tpg->acl_node_mutex);
                return ERR_PTR(-EEXIST);
        }
-       spin_unlock_irq(&tpg->acl_node_lock);
+       mutex_unlock(&tpg->acl_node_mutex);
 
-       if (!se_nacl) {
-               pr_err("struct se_node_acl pointer is NULL\n");
-               return ERR_PTR(-EINVAL);
-       }
-       /*
-        * For v4.x logic the se_node_acl_s is hanging off a fabric
-        * dependent structure allocated via
-        * struct target_core_fabric_ops->fabric_make_nodeacl()
-        */
-       acl = se_nacl;
-
-       INIT_LIST_HEAD(&acl->acl_list);
-       INIT_LIST_HEAD(&acl->acl_sess_list);
-       kref_init(&acl->acl_kref);
-       init_completion(&acl->acl_free_comp);
-       spin_lock_init(&acl->device_list_lock);
-       spin_lock_init(&acl->nacl_sess_lock);
-       atomic_set(&acl->acl_pr_ref_count, 0);
-       acl->queue_depth = queue_depth;
-       snprintf(acl->initiatorname, TRANSPORT_IQN_LEN, "%s", initiatorname);
-       acl->se_tpg = tpg;
-       acl->acl_index = scsi_get_new_index(SCSI_AUTH_INTR_INDEX);
-
-       tpg->se_tpg_tfo->set_default_node_attributes(acl);
-
-       if (core_create_device_list_for_node(acl) < 0) {
-               tpg->se_tpg_tfo->tpg_release_fabric_acl(tpg, acl);
+       acl = target_alloc_node_acl(tpg, initiatorname);
+       if (!acl)
                return ERR_PTR(-ENOMEM);
-       }
-
-       if (core_set_queue_depth_for_node(tpg, acl) < 0) {
-               core_free_device_list_for_node(acl, tpg);
-               tpg->se_tpg_tfo->tpg_release_fabric_acl(tpg, acl);
-               return ERR_PTR(-EINVAL);
-       }
-
-       spin_lock_irq(&tpg->acl_node_lock);
-       list_add_tail(&acl->acl_list, &tpg->acl_node_list);
-       tpg->num_node_acls++;
-       spin_unlock_irq(&tpg->acl_node_lock);
-
-done:
-       pr_debug("%s_TPG[%hu] - Added ACL with TCQ Depth: %d for %s"
-               " Initiator Node: %s\n", tpg->se_tpg_tfo->get_fabric_name(),
-               tpg->se_tpg_tfo->tpg_get_tag(tpg), acl->queue_depth,
-               tpg->se_tpg_tfo->get_fabric_name(), initiatorname);
 
+       target_add_node_acl(acl);
        return acl;
 }
-EXPORT_SYMBOL(core_tpg_add_initiator_node_acl);
 
-/*     core_tpg_del_initiator_node_acl():
- *
- *
- */
-int core_tpg_del_initiator_node_acl(
-       struct se_portal_group *tpg,
-       struct se_node_acl *acl,
-       int force)
+void core_tpg_del_initiator_node_acl(struct se_node_acl *acl)
 {
+       struct se_portal_group *tpg = acl->se_tpg;
        LIST_HEAD(sess_list);
        struct se_session *sess, *sess_tmp;
        unsigned long flags;
        int rc;
 
-       spin_lock_irq(&tpg->acl_node_lock);
+       mutex_lock(&tpg->acl_node_mutex);
        if (acl->dynamic_node_acl) {
                acl->dynamic_node_acl = 0;
        }
        list_del(&acl->acl_list);
        tpg->num_node_acls--;
-       spin_unlock_irq(&tpg->acl_node_lock);
+       mutex_unlock(&tpg->acl_node_mutex);
 
        spin_lock_irqsave(&acl->nacl_sess_lock, flags);
        acl->acl_stop = 1;
@@ -494,7 +351,6 @@ int core_tpg_del_initiator_node_acl(
        wait_for_completion(&acl->acl_free_comp);
 
        core_tpg_wait_for_nacl_pr_ref(acl);
-       core_clear_initiator_node_from_tpg(acl, tpg);
        core_free_device_list_for_node(acl, tpg);
 
        pr_debug("%s_TPG[%hu] - Deleted ACL with TCQ Depth: %d for %s"
@@ -502,9 +358,8 @@ int core_tpg_del_initiator_node_acl(
                tpg->se_tpg_tfo->tpg_get_tag(tpg), acl->queue_depth,
                tpg->se_tpg_tfo->get_fabric_name(), acl->initiatorname);
 
-       return 0;
+       kfree(acl);
 }
-EXPORT_SYMBOL(core_tpg_del_initiator_node_acl);
 
 /*     core_tpg_set_initiator_node_queue_depth():
  *
@@ -521,21 +376,21 @@ int core_tpg_set_initiator_node_queue_depth(
        unsigned long flags;
        int dynamic_acl = 0;
 
-       spin_lock_irq(&tpg->acl_node_lock);
+       mutex_lock(&tpg->acl_node_mutex);
        acl = __core_tpg_get_initiator_node_acl(tpg, initiatorname);
        if (!acl) {
                pr_err("Access Control List entry for %s Initiator"
                        " Node %s does not exists for TPG %hu, ignoring"
                        " request.\n", tpg->se_tpg_tfo->get_fabric_name(),
                        initiatorname, tpg->se_tpg_tfo->tpg_get_tag(tpg));
-               spin_unlock_irq(&tpg->acl_node_lock);
+               mutex_unlock(&tpg->acl_node_mutex);
                return -ENODEV;
        }
        if (acl->dynamic_node_acl) {
                acl->dynamic_node_acl = 0;
                dynamic_acl = 1;
        }
-       spin_unlock_irq(&tpg->acl_node_lock);
+       mutex_unlock(&tpg->acl_node_mutex);
 
        spin_lock_irqsave(&tpg->session_lock, flags);
        list_for_each_entry(sess, &tpg->tpg_sess_list, sess_list) {
@@ -551,10 +406,10 @@ int core_tpg_set_initiator_node_queue_depth(
                                tpg->se_tpg_tfo->get_fabric_name(), initiatorname);
                        spin_unlock_irqrestore(&tpg->session_lock, flags);
 
-                       spin_lock_irq(&tpg->acl_node_lock);
+                       mutex_lock(&tpg->acl_node_mutex);
                        if (dynamic_acl)
                                acl->dynamic_node_acl = 1;
-                       spin_unlock_irq(&tpg->acl_node_lock);
+                       mutex_unlock(&tpg->acl_node_mutex);
                        return -EEXIST;
                }
                /*
@@ -589,10 +444,10 @@ int core_tpg_set_initiator_node_queue_depth(
                if (init_sess)
                        tpg->se_tpg_tfo->close_session(init_sess);
 
-               spin_lock_irq(&tpg->acl_node_lock);
+               mutex_lock(&tpg->acl_node_mutex);
                if (dynamic_acl)
                        acl->dynamic_node_acl = 1;
-               spin_unlock_irq(&tpg->acl_node_lock);
+               mutex_unlock(&tpg->acl_node_mutex);
                return -EINVAL;
        }
        spin_unlock_irqrestore(&tpg->session_lock, flags);
@@ -608,10 +463,10 @@ int core_tpg_set_initiator_node_queue_depth(
                initiatorname, tpg->se_tpg_tfo->get_fabric_name(),
                tpg->se_tpg_tfo->tpg_get_tag(tpg));
 
-       spin_lock_irq(&tpg->acl_node_lock);
+       mutex_lock(&tpg->acl_node_mutex);
        if (dynamic_acl)
                acl->dynamic_node_acl = 1;
-       spin_unlock_irq(&tpg->acl_node_lock);
+       mutex_unlock(&tpg->acl_node_mutex);
 
        return 0;
 }
@@ -647,78 +502,54 @@ static void core_tpg_lun_ref_release(struct percpu_ref *ref)
        complete(&lun->lun_ref_comp);
 }
 
-static int core_tpg_setup_virtual_lun0(struct se_portal_group *se_tpg)
-{
-       /* Set in core_dev_setup_virtual_lun0() */
-       struct se_device *dev = g_lun0_dev;
-       struct se_lun *lun = &se_tpg->tpg_virt_lun0;
-       u32 lun_access = TRANSPORT_LUNFLAGS_READ_ONLY;
-       int ret;
-
-       lun->unpacked_lun = 0;
-       lun->lun_status = TRANSPORT_LUN_STATUS_FREE;
-       atomic_set(&lun->lun_acl_count, 0);
-       init_completion(&lun->lun_shutdown_comp);
-       INIT_LIST_HEAD(&lun->lun_acl_list);
-       spin_lock_init(&lun->lun_acl_lock);
-       spin_lock_init(&lun->lun_sep_lock);
-       init_completion(&lun->lun_ref_comp);
-
-       ret = core_tpg_add_lun(se_tpg, lun, lun_access, dev);
-       if (ret < 0)
-               return ret;
-
-       return 0;
-}
-
 int core_tpg_register(
-       const struct target_core_fabric_ops *tfo,
        struct se_wwn *se_wwn,
        struct se_portal_group *se_tpg,
-       void *tpg_fabric_ptr,
-       int se_tpg_type)
+       int proto_id)
 {
-       struct se_lun *lun;
-       u32 i;
-
-       se_tpg->tpg_lun_list = array_zalloc(TRANSPORT_MAX_LUNS_PER_TPG,
-                       sizeof(struct se_lun), GFP_KERNEL);
-       if (!se_tpg->tpg_lun_list) {
-               pr_err("Unable to allocate struct se_portal_group->"
-                               "tpg_lun_list\n");
-               return -ENOMEM;
-       }
+       int ret;
 
-       for (i = 0; i < TRANSPORT_MAX_LUNS_PER_TPG; i++) {
-               lun = se_tpg->tpg_lun_list[i];
-               lun->unpacked_lun = i;
-               lun->lun_link_magic = SE_LUN_LINK_MAGIC;
-               lun->lun_status = TRANSPORT_LUN_STATUS_FREE;
-               atomic_set(&lun->lun_acl_count, 0);
-               init_completion(&lun->lun_shutdown_comp);
-               INIT_LIST_HEAD(&lun->lun_acl_list);
-               spin_lock_init(&lun->lun_acl_lock);
-               spin_lock_init(&lun->lun_sep_lock);
-               init_completion(&lun->lun_ref_comp);
+       if (!se_tpg)
+               return -EINVAL;
+       /*
+        * For the typical case where core_tpg_register() is called by a
+        * fabric driver from target_core_fabric_ops->fabric_make_tpg()
+        * configfs context, use the original tf_ops pointer already saved
+        * by target-core in target_fabric_make_wwn().
+        *
+        * Otherwise, for special cases like iscsi-target discovery TPGs
+        * the caller is responsible for setting ->se_tpg_tfo ahead of
+        * calling core_tpg_register().
+        */
+       if (se_wwn)
+               se_tpg->se_tpg_tfo = se_wwn->wwn_tf->tf_ops;
+
+       if (!se_tpg->se_tpg_tfo) {
+               pr_err("Unable to locate se_tpg->se_tpg_tfo pointer\n");
+               return -EINVAL;
        }
 
-       se_tpg->se_tpg_type = se_tpg_type;
-       se_tpg->se_tpg_fabric_ptr = tpg_fabric_ptr;
-       se_tpg->se_tpg_tfo = tfo;
+       INIT_HLIST_HEAD(&se_tpg->tpg_lun_hlist);
+       se_tpg->proto_id = proto_id;
        se_tpg->se_tpg_wwn = se_wwn;
        atomic_set(&se_tpg->tpg_pr_ref_count, 0);
        INIT_LIST_HEAD(&se_tpg->acl_node_list);
        INIT_LIST_HEAD(&se_tpg->se_tpg_node);
        INIT_LIST_HEAD(&se_tpg->tpg_sess_list);
-       spin_lock_init(&se_tpg->acl_node_lock);
        spin_lock_init(&se_tpg->session_lock);
-       spin_lock_init(&se_tpg->tpg_lun_lock);
-
-       if (se_tpg->se_tpg_type == TRANSPORT_TPG_TYPE_NORMAL) {
-               if (core_tpg_setup_virtual_lun0(se_tpg) < 0) {
-                       array_free(se_tpg->tpg_lun_list,
-                                  TRANSPORT_MAX_LUNS_PER_TPG);
-                       return -ENOMEM;
+       mutex_init(&se_tpg->tpg_lun_mutex);
+       mutex_init(&se_tpg->acl_node_mutex);
+
+       if (se_tpg->proto_id >= 0) {
+               se_tpg->tpg_virt_lun0 = core_tpg_alloc_lun(se_tpg, 0);
+               if (IS_ERR(se_tpg->tpg_virt_lun0))
+                       return PTR_ERR(se_tpg->tpg_virt_lun0);
+
+               ret = core_tpg_add_lun(se_tpg, se_tpg->tpg_virt_lun0,
+                               TRANSPORT_LUNFLAGS_READ_ONLY, g_lun0_dev);
+               if (ret < 0) {
+                       kfree(se_tpg->tpg_virt_lun0);
+                       return ret;
                }
        }
 
@@ -726,11 +557,11 @@ int core_tpg_register(
        list_add_tail(&se_tpg->se_tpg_node, &tpg_list);
        spin_unlock_bh(&tpg_lock);
 
-       pr_debug("TARGET_CORE[%s]: Allocated %s struct se_portal_group for"
-               " endpoint: %s, Portal Tag: %u\n", tfo->get_fabric_name(),
-               (se_tpg->se_tpg_type == TRANSPORT_TPG_TYPE_NORMAL) ?
-               "Normal" : "Discovery", (tfo->tpg_get_wwn(se_tpg) == NULL) ?
-               "None" : tfo->tpg_get_wwn(se_tpg), tfo->tpg_get_tag(se_tpg));
+       pr_debug("TARGET_CORE[%s]: Allocated portal_group for endpoint: %s, "
+                "Proto: %d, Portal Tag: %u\n", se_tpg->se_tpg_tfo->get_fabric_name(),
+               se_tpg->se_tpg_tfo->tpg_get_wwn(se_tpg) ?
+               se_tpg->se_tpg_tfo->tpg_get_wwn(se_tpg) : NULL,
+               se_tpg->proto_id, se_tpg->se_tpg_tfo->tpg_get_tag(se_tpg));
 
        return 0;
 }
@@ -738,14 +569,14 @@ EXPORT_SYMBOL(core_tpg_register);
 
 int core_tpg_deregister(struct se_portal_group *se_tpg)
 {
+       const struct target_core_fabric_ops *tfo = se_tpg->se_tpg_tfo;
        struct se_node_acl *nacl, *nacl_tmp;
+       LIST_HEAD(node_list);
 
-       pr_debug("TARGET_CORE[%s]: Deallocating %s struct se_portal_group"
-               " for endpoint: %s Portal Tag %u\n",
-               (se_tpg->se_tpg_type == TRANSPORT_TPG_TYPE_NORMAL) ?
-               "Normal" : "Discovery", se_tpg->se_tpg_tfo->get_fabric_name(),
-               se_tpg->se_tpg_tfo->tpg_get_wwn(se_tpg),
-               se_tpg->se_tpg_tfo->tpg_get_tag(se_tpg));
+       pr_debug("TARGET_CORE[%s]: Deallocating portal_group for endpoint: %s, "
+                "Proto: %d, Portal Tag: %u\n", tfo->get_fabric_name(),
+               tfo->tpg_get_wwn(se_tpg) ? tfo->tpg_get_wwn(se_tpg) : NULL,
+               se_tpg->proto_id, tfo->tpg_get_tag(se_tpg));
 
        spin_lock_bh(&tpg_lock);
        list_del(&se_tpg->se_tpg_node);
@@ -753,61 +584,56 @@ int core_tpg_deregister(struct se_portal_group *se_tpg)
 
        while (atomic_read(&se_tpg->tpg_pr_ref_count) != 0)
                cpu_relax();
+
+       mutex_lock(&se_tpg->acl_node_mutex);
+       list_splice_init(&se_tpg->acl_node_list, &node_list);
+       mutex_unlock(&se_tpg->acl_node_mutex);
        /*
         * Release any remaining demo-mode generated se_node_acl that have
         * not been released because of TFO->tpg_check_demo_mode_cache() == 1
         * in transport_deregister_session().
         */
-       spin_lock_irq(&se_tpg->acl_node_lock);
-       list_for_each_entry_safe(nacl, nacl_tmp, &se_tpg->acl_node_list,
-                       acl_list) {
+       list_for_each_entry_safe(nacl, nacl_tmp, &node_list, acl_list) {
                list_del(&nacl->acl_list);
                se_tpg->num_node_acls--;
-               spin_unlock_irq(&se_tpg->acl_node_lock);
 
                core_tpg_wait_for_nacl_pr_ref(nacl);
                core_free_device_list_for_node(nacl, se_tpg);
-               se_tpg->se_tpg_tfo->tpg_release_fabric_acl(se_tpg, nacl);
-
-               spin_lock_irq(&se_tpg->acl_node_lock);
+               kfree(nacl);
        }
-       spin_unlock_irq(&se_tpg->acl_node_lock);
 
-       if (se_tpg->se_tpg_type == TRANSPORT_TPG_TYPE_NORMAL)
-               core_tpg_remove_lun(se_tpg, &se_tpg->tpg_virt_lun0);
+       if (se_tpg->proto_id >= 0) {
+               core_tpg_remove_lun(se_tpg, se_tpg->tpg_virt_lun0);
+               kfree_rcu(se_tpg->tpg_virt_lun0, rcu_head);
+       }
 
-       se_tpg->se_tpg_fabric_ptr = NULL;
-       array_free(se_tpg->tpg_lun_list, TRANSPORT_MAX_LUNS_PER_TPG);
        return 0;
 }
 EXPORT_SYMBOL(core_tpg_deregister);
 
 struct se_lun *core_tpg_alloc_lun(
        struct se_portal_group *tpg,
-       u32 unpacked_lun)
+       u64 unpacked_lun)
 {
        struct se_lun *lun;
 
-       if (unpacked_lun > (TRANSPORT_MAX_LUNS_PER_TPG-1)) {
-               pr_err("%s LUN: %u exceeds TRANSPORT_MAX_LUNS_PER_TPG"
-                       "-1: %u for Target Portal Group: %u\n",
-                       tpg->se_tpg_tfo->get_fabric_name(),
-                       unpacked_lun, TRANSPORT_MAX_LUNS_PER_TPG-1,
-                       tpg->se_tpg_tfo->tpg_get_tag(tpg));
-               return ERR_PTR(-EOVERFLOW);
-       }
-
-       spin_lock(&tpg->tpg_lun_lock);
-       lun = tpg->tpg_lun_list[unpacked_lun];
-       if (lun->lun_status == TRANSPORT_LUN_STATUS_ACTIVE) {
-               pr_err("TPG Logical Unit Number: %u is already active"
-                       " on %s Target Portal Group: %u, ignoring request.\n",
-                       unpacked_lun, tpg->se_tpg_tfo->get_fabric_name(),
-                       tpg->se_tpg_tfo->tpg_get_tag(tpg));
-               spin_unlock(&tpg->tpg_lun_lock);
-               return ERR_PTR(-EINVAL);
+       lun = kzalloc(sizeof(*lun), GFP_KERNEL);
+       if (!lun) {
+               pr_err("Unable to allocate se_lun memory\n");
+               return ERR_PTR(-ENOMEM);
        }
-       spin_unlock(&tpg->tpg_lun_lock);
+       lun->unpacked_lun = unpacked_lun;
+       lun->lun_link_magic = SE_LUN_LINK_MAGIC;
+       atomic_set(&lun->lun_acl_count, 0);
+       init_completion(&lun->lun_ref_comp);
+       INIT_LIST_HEAD(&lun->lun_deve_list);
+       INIT_LIST_HEAD(&lun->lun_dev_link);
+       atomic_set(&lun->lun_tg_pt_secondary_offline, 0);
+       spin_lock_init(&lun->lun_deve_lock);
+       mutex_init(&lun->lun_tg_pt_md_mutex);
+       INIT_LIST_HEAD(&lun->lun_tg_pt_gp_link);
+       spin_lock_init(&lun->lun_tg_pt_gp_lock);
+       lun->lun_tpg = tpg;
 
        return lun;
 }
@@ -823,34 +649,73 @@ int core_tpg_add_lun(
        ret = percpu_ref_init(&lun->lun_ref, core_tpg_lun_ref_release, 0,
                              GFP_KERNEL);
        if (ret < 0)
-               return ret;
+               goto out;
 
-       ret = core_dev_export(dev, tpg, lun);
-       if (ret < 0) {
-               percpu_ref_exit(&lun->lun_ref);
-               return ret;
-       }
+       ret = core_alloc_rtpi(lun, dev);
+       if (ret)
+               goto out_kill_ref;
 
-       spin_lock(&tpg->tpg_lun_lock);
-       lun->lun_access = lun_access;
-       lun->lun_status = TRANSPORT_LUN_STATUS_ACTIVE;
-       spin_unlock(&tpg->tpg_lun_lock);
+       if (!(dev->transport->transport_flags & TRANSPORT_FLAG_PASSTHROUGH) &&
+           !(dev->se_hba->hba_flags & HBA_FLAGS_INTERNAL_USE))
+               target_attach_tg_pt_gp(lun, dev->t10_alua.default_tg_pt_gp);
+
+       mutex_lock(&tpg->tpg_lun_mutex);
+
+       spin_lock(&dev->se_port_lock);
+       lun->lun_index = dev->dev_index;
+       rcu_assign_pointer(lun->lun_se_dev, dev);
+       dev->export_count++;
+       list_add_tail(&lun->lun_dev_link, &dev->dev_sep_list);
+       spin_unlock(&dev->se_port_lock);
+
+       if (dev->dev_flags & DF_READ_ONLY)
+               lun->lun_access = TRANSPORT_LUNFLAGS_READ_ONLY;
+       else
+               lun->lun_access = lun_access;
+       if (!(dev->se_hba->hba_flags & HBA_FLAGS_INTERNAL_USE))
+               hlist_add_head_rcu(&lun->link, &tpg->tpg_lun_hlist);
+       mutex_unlock(&tpg->tpg_lun_mutex);
 
        return 0;
+
+out_kill_ref:
+       percpu_ref_exit(&lun->lun_ref);
+out:
+       return ret;
 }
 
 void core_tpg_remove_lun(
        struct se_portal_group *tpg,
        struct se_lun *lun)
 {
+       /*
+        * rcu_dereference_raw protected by se_lun->lun_group symlink
+        * reference to se_device->dev_group.
+        */
+       struct se_device *dev = rcu_dereference_raw(lun->lun_se_dev);
+
        core_clear_lun_from_tpg(lun, tpg);
+       /*
+        * Wait for any active I/O references to percpu se_lun->lun_ref to
+        * be released.  Also, se_lun->lun_ref is now used by PR and ALUA
+        * logic when referencing a remote target port during ALL_TGT_PT=1
+        * and generating UNIT_ATTENTIONs for ALUA access state transition.
+        */
        transport_clear_lun_ref(lun);
 
-       core_dev_unexport(lun->lun_se_dev, tpg, lun);
+       mutex_lock(&tpg->tpg_lun_mutex);
+       if (lun->lun_se_dev) {
+               target_detach_tg_pt_gp(lun);
 
-       spin_lock(&tpg->tpg_lun_lock);
-       lun->lun_status = TRANSPORT_LUN_STATUS_FREE;
-       spin_unlock(&tpg->tpg_lun_lock);
+               spin_lock(&dev->se_port_lock);
+               list_del(&lun->lun_dev_link);
+               dev->export_count--;
+               rcu_assign_pointer(lun->lun_se_dev, NULL);
+               spin_unlock(&dev->se_port_lock);
+       }
+       if (!(dev->se_hba->hba_flags & HBA_FLAGS_INTERNAL_USE))
+               hlist_del_rcu(&lun->link);
+       mutex_unlock(&tpg->tpg_lun_mutex);
 
        percpu_ref_exit(&lun->lun_ref);
 }