DAOS-18929 cart: add crt_bulk_free_common() and fix refcounting

Signed-off-by: Jerome Soumagne <jerome.soumagne@hpe.com>

Author: soumagne

src/cart/crt_bulk.c

@@ -184,19 +184,12 @@ crt_bulk_addref(crt_bulk_t crt_bulk)
 {
 	struct crt_bulk *bulk = crt_bulk;
 	int              rc   = -DER_SUCCESS;
-	hg_return_t      hg_ret;
 
 	if (bulk == NULL) {
 		D_ERROR("invalid parameter, NULL bulk\n");
 		D_GOTO(out, rc = -DER_INVAL);
 	}
 
-	hg_ret = HG_Bulk_ref_incr(bulk->hg_bulk_hdl);
-	if (hg_ret != HG_SUCCESS) {
-		D_ERROR("HG_Bulk_ref_incr failed, hg_ret: %d.\n", hg_ret);
-		rc = crt_hgret_2_der(hg_ret);
-	}
-
 	atomic_fetch_add(&bulk->refcount, 1);
 out:
 	return rc;
@@ -206,43 +199,43 @@ int
 crt_bulk_free(crt_bulk_t crt_bulk)
 {
 	struct crt_bulk *bulk = crt_bulk;
-	int              rc   = -DER_SUCCESS;
-	hg_return_t      hg_ret;
 
 	if (bulk == NULL) {
 		D_ERROR("invalid parameter, NULL bulk\n");
-		D_GOTO(out, rc = -DER_INVAL);
-	}
-
-	/* This can happen if D_QUOTA_BULKS is enabled on a client */
-	if (bulk->hg_bulk_hdl == HG_BULK_NULL) {
-		if (bulk->deferred) {
-			/* Treat as success */
-			D_GOTO(out, rc = DER_SUCCESS);
-		} else {
-			D_ASSERTF(0, "Bulk handle should not be NULL\n");
-		}
-	}
-
-	hg_ret = HG_Bulk_free(bulk->hg_bulk_hdl);
-	if (hg_ret != HG_SUCCESS) {
-		D_ERROR("HG_Bulk_free failed, hg_ret: %d.\n", hg_ret);
-		rc = crt_hgret_2_der(hg_ret);
+		return -DER_INVAL;
 	}
 
 	if (atomic_fetch_sub(&bulk->refcount, 1) > 1)
 		return DER_SUCCESS;
 
+	crt_bulk_free_common(bulk);
+
+	return DER_SUCCESS;
+}
+
+void
+crt_bulk_free_common(struct crt_bulk *bulk)
+{
+	hg_return_t hg_ret;
+
+	D_ASSERT(bulk != NULL);
+
+	/* Decrement refcount on bulk handle itself */
+	if (bulk->hg_bulk_hdl != HG_BULK_NULL) {
+		hg_ret = HG_Bulk_free(bulk->hg_bulk_hdl);
+		if (hg_ret != HG_SUCCESS) {
+			D_ERROR("HG_Bulk_free() failed (%s)\n", HG_Error_to_string(hg_ret));
+			/* Ignore the error, as we are already in a cleanup path */
+		}
+	}
+
 	/* decoded bulks are not counted towards quota; such bulks have crt_ctx set to NULL */
-	if (bulk->crt_ctx)
+	if (!bulk->deferred && bulk->crt_ctx != NULL)
 		put_quota_resource(bulk->crt_ctx, CRT_QUOTA_BULKS);
-out:
-	if (bulk != NULL) {
-		if (bulk->iovs)
-			D_FREE(bulk->iovs);
-		D_FREE(bulk);
-	}
-	return rc;
+
+	if (bulk->iovs != NULL)
+		D_FREE(bulk->iovs);
+	D_FREE(bulk);
 }
 
 /* Helper function to check for bulk expiration */

src/cart/crt_hg_proc.c

@@ -134,12 +134,13 @@ crt_proc_crt_bulk_t_deffered(struct crt_bulk *bulk)
 	if (bulk->bound) {
 		rc = crt_hg_bulk_bind(bulk->hg_bulk_hdl, &ctx->cc_hg_ctx);
 		if (rc != 0) {
-			D_ERROR("Failed to bind bulk during proc\n");
+			DL_ERROR(rc, "Failed to bind bulk during proc");
 			/* free will return quota resource */
 			crt_bulk_free(bulk->hg_bulk_hdl);
 			return rc;
 		}
 	}
+	/* Mark as no longer deferred once allocation is complete */
 	bulk->deferred = false;
 
 	return 0;
@@ -165,7 +166,7 @@ crt_proc_crt_bulk_t(crt_proc_t proc, crt_proc_op_t proc_op, crt_bulk_t *pcrt_bul
 		/* Deferred allocation as a result of D_QUOTA_BULKS limit */
 		if (bulk != CRT_BULK_NULL) {
 			if (bulk->deferred && (rc = crt_proc_crt_bulk_t_deffered(bulk)) != 0) {
-				D_ERROR("Failed to do deferred bulk allocation during proc\n");
+				DL_ERROR(rc, "Failed to do deferred bulk allocation during proc");
 				return rc;
 			}
 			hg_bulk = bulk->hg_bulk_hdl;
@@ -192,12 +193,10 @@ crt_proc_crt_bulk_t(crt_proc_t proc, crt_proc_op_t proc_op, crt_bulk_t *pcrt_bul
 
 		/* Allocate space for a wrapper struct */
 		D_ALLOC_PTR(bulk);
-		if (!bulk)
+		if (bulk == NULL)
 			return -DER_NOMEM;
 
 		bulk->hg_bulk_hdl = hg_bulk;
-		bulk->deferred    = false;
-		bulk->crt_ctx     = NULL;
 		bulk->refcount    = 1;
 
 		*pcrt_bulk = bulk;
@@ -209,21 +208,25 @@ crt_proc_crt_bulk_t(crt_proc_t proc, crt_proc_op_t proc_op, crt_bulk_t *pcrt_bul
 		if (bulk == NULL)
 			return 0;
 
-		/* Prevent HG proc from assigning NULL if recount is not zero */
+		/**
+		 * Prevent HG proc from assigning NULL if refcount is not zero and keep reference on
+		 * HG bulk, we'll free it ourselves. hg_proc_hg_bulk_t() will decrement refcount on
+		 * the HG bulk.
+		 */
 		hg_bulk = bulk->hg_bulk_hdl;
-		hg_ret  = hg_proc_hg_bulk_t(proc, &hg_bulk);
+		HG_Bulk_ref_incr(hg_bulk);
+		hg_ret = hg_proc_hg_bulk_t(proc, &hg_bulk);
 		if (hg_ret != HG_SUCCESS) {
-			D_ERROR("Failed to free bulk during proc\n");
+			D_ERROR("Failed to free bulk during proc (%s)\n",
+				HG_Error_to_string(hg_ret));
 			return -DER_HG;
 		}
 
 		if (atomic_fetch_sub(&bulk->refcount, 1) > 1)
 			return 0;
 
-		/* Free the wrapper struct */
-		if (bulk->iovs)
-			D_FREE(bulk->iovs);
-		D_FREE(bulk);
+		/* This is the real free */
+		crt_bulk_free_common(bulk);
 		*pcrt_bulk = NULL;
 		return 0;
 	}

src/cart/crt_internal_fns.h

@@ -1,6 +1,6 @@
 /*
  * (C) Copyright 2016-2024 Intel Corporation.
- * (C) Copyright 2025 Hewlett Packard Enterprise Development LP
+ * (C) Copyright 2025-2026 Hewlett Packard Enterprise Development LP
  *
  * SPDX-License-Identifier: BSD-2-Clause-Patent
  */
@@ -62,6 +62,9 @@ crt_bulk_desc_dup(struct crt_bulk_desc *bulk_desc_new,
 	*bulk_desc_new = *bulk_desc;
 }
 
+void
+crt_bulk_free_common(struct crt_bulk *bulk);
+
 void
 crt_hdlr_proto_query(crt_rpc_t *rpc_req);