X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fmdt%2Fmdt_hsm_cdt_agent.c;h=c3d41822211400bb58be7e4d80fbcc4df10cc11c;hb=6b8c570c40829873b01b7388a622ae04b63317aa;hp=f5c60eb0b0d46bffa7bd03cc40ae9f0f3c6041a6;hpb=f24d4c4a3146e2200a642fdb71aa0e11afd36874;p=fs%2Flustre-release.git

diff --git a/lustre/mdt/mdt_hsm_cdt_agent.c b/lustre/mdt/mdt_hsm_cdt_agent.c
index f5c60eb..c3d4182 100644
--- a/lustre/mdt/mdt_hsm_cdt_agent.c
+++ b/lustre/mdt/mdt_hsm_cdt_agent.c
@@ -21,6 +21,8 @@
  */
 /*
  * (C) Copyright 2012 Commissariat a l'energie atomique et aux energies
+ *
+ * Copyright (c) 2016, Intel Corporation.
  *     alternatives
  *
  */
@@ -40,6 +42,7 @@
 #include <lustre_export.h>
 #include <lustre/lustre_user.h>
 #include <lprocfs_status.h>
+#include <lustre_kernelcomm.h>
 #include "mdt_internal.h"
 
 /*
@@ -48,7 +51,7 @@
 
 /*
  * find a hsm_agent by uuid
- * lock cdt_agent_lock needs to be hold by caller
+ * lock cdt_agent_lock needs to be held by caller
  * \param cdt [IN] coordinator
  * \param uuid [IN] agent UUID
  * \retval hsm_agent pointer or NULL if not found
@@ -57,10 +60,8 @@ static struct hsm_agent *mdt_hsm_agent_lookup(struct coordinator *cdt,
 					      const struct obd_uuid *uuid)
 {
 	struct hsm_agent	*ha;
-	cfs_list_t		*pos;
 
-	cfs_list_for_each(pos, &cdt->cdt_agents) {
-		ha = cfs_list_entry(pos, struct hsm_agent, ha_list);
+	list_for_each_entry(ha, &cdt->cdt_agents, ha_list) {
 		if (obd_uuid_equals(&ha->ha_uuid, uuid))
 			return ha;
 	}
@@ -120,7 +121,7 @@ int mdt_hsm_agent_register(struct mdt_thread_info *mti,
 		GOTO(out_free, rc = -EEXIST);
 	}
 
-	cfs_list_add_tail(&ha->ha_list, &cdt->cdt_agents);
+	list_add_tail(&ha->ha_list, &cdt->cdt_agents);
 
 	if (ha->ha_archive_cnt == 0)
 		CDEBUG(D_HSM, "agent %s registered for all archives\n",
@@ -134,7 +135,7 @@ int mdt_hsm_agent_register(struct mdt_thread_info *mti,
 
 out_free:
 
-	if ((ha != NULL) && (ha->ha_archive_id != NULL))
+	if (ha != NULL && ha->ha_archive_id != NULL)
 		OBD_FREE(ha->ha_archive_id,
 			 ha->ha_archive_cnt * sizeof(*ha->ha_archive_id));
 	if (ha != NULL)
@@ -205,7 +206,7 @@ int mdt_hsm_agent_unregister(struct mdt_thread_info *mti,
 
 	ha = mdt_hsm_agent_lookup(cdt, uuid);
 	if (ha != NULL)
-		cfs_list_del_init(&ha->ha_list);
+		list_del_init(&ha->ha_list);
 
 	up_write(&cdt->cdt_agent_lock);
 
@@ -239,14 +240,14 @@ int mdt_hsm_agent_update_statistics(struct coordinator *cdt,
 				    int succ_rq, int fail_rq, int new_rq,
 				    const struct obd_uuid *uuid)
 {
-	struct hsm_agent	*ha, *tmp;
+	struct hsm_agent	*ha;
 	int			 rc;
 	ENTRY;
 
 	down_read(&cdt->cdt_agent_lock);
-	cfs_list_for_each_entry_safe(ha, tmp, &cdt->cdt_agents, ha_list) {
+	list_for_each_entry(ha, &cdt->cdt_agents, ha_list) {
 		if (obd_uuid_equals(&ha->ha_uuid, uuid)) {
-			if ((succ_rq == 0) && (fail_rq == 0) && (new_rq == 0)) {
+			if (succ_rq == 0 && fail_rq == 0 && new_rq == 0) {
 				atomic_set(&ha->ha_success, 0);
 				atomic_set(&ha->ha_failure, 0);
 				atomic_set(&ha->ha_requests, 0);
@@ -279,23 +280,22 @@ int mdt_hsm_find_best_agent(struct coordinator *cdt, __u32 archive,
 			    struct obd_uuid *uuid)
 {
 	int			 rc = -EAGAIN, i, load = -1;
-	struct hsm_agent	*ha, *tmp;
+	struct hsm_agent	*ha;
 	ENTRY;
 
 	/* Choose an export to send a copytool req to */
 	down_read(&cdt->cdt_agent_lock);
-	cfs_list_for_each_entry_safe(ha, tmp, &cdt->cdt_agents, ha_list) {
+	list_for_each_entry(ha, &cdt->cdt_agents, ha_list) {
 		for (i = 0; (i < ha->ha_archive_cnt) &&
 			      (ha->ha_archive_id[i] != archive); i++) {
 			/* nothing to do, just skip unmatching records */
 		}
 
 		/* archive count == 0 means copy tool serves any backend */
-		if ((ha->ha_archive_cnt != 0) && (i == ha->ha_archive_cnt))
+		if (ha->ha_archive_cnt != 0 && i == ha->ha_archive_cnt)
 			continue;
 
-		if ((load == -1) ||
-		    (load > atomic_read(&ha->ha_requests))) {
+		if (load == -1 || load > atomic_read(&ha->ha_requests)) {
 			load = atomic_read(&ha->ha_requests);
 			*uuid = ha->ha_uuid;
 			rc = 0;
@@ -308,6 +308,59 @@ int mdt_hsm_find_best_agent(struct coordinator *cdt, __u32 archive,
 	RETURN(rc);
 }
 
+int mdt_hsm_send_action_to_each_archive(struct mdt_thread_info *mti,
+				    struct hsm_action_item *hai)
+{
+	__u64 compound_id;
+	struct hsm_agent *ha;
+	__u32 archive_mask = 0;
+	struct coordinator *cdt = &mti->mti_mdt->mdt_coordinator;
+	int i;
+	/* return error by default in case all archive_ids have unregistered */
+	int rc = -EAGAIN;
+	ENTRY;
+
+	/* send action to all registered archive_ids */
+	down_read(&cdt->cdt_agent_lock);
+	list_for_each_entry(ha, &cdt->cdt_agents, ha_list) {
+		for (i = 0; (i < ha->ha_archive_cnt); i++) {
+			/* only send once for each archive_id */
+			if ((1 << ha->ha_archive_id[i]) & archive_mask)
+				continue;
+			archive_mask |= (1 << ha->ha_archive_id[i]);
+
+			/* XXX: instead of creating one request record per
+			 * new action, it could make sense to gather
+			 * all for the same archive_id as one compound
+			 * request/id, like in mdt_hsm_add_actions() ?? */
+			compound_id = atomic_inc_return(&cdt->cdt_compound_id);
+			rc = mdt_agent_record_add(mti->mti_env, mti->mti_mdt,
+						  compound_id,
+						  ha->ha_archive_id[i], 0,
+						  hai);
+			if (rc) {
+				CERROR("%s: unable to add HSM remove request "
+				       "for "DFID": rc=%d\n",
+				       mdt_obd_name(mti->mti_mdt),
+				       PFID(&hai->hai_fid), rc);
+				break;
+			} else {
+				CDEBUG(D_HSM, "%s: added HSM remove request "
+				       "for "DFID", archive_id=%d\n",
+				       mdt_obd_name(mti->mti_mdt),
+				       PFID(&hai->hai_fid),
+				       ha->ha_archive_id[i]);
+			}
+		}
+		/* early exit from loop due to error? */
+		if (i != ha->ha_archive_cnt)
+			break;
+	}
+	up_read(&cdt->cdt_agent_lock);
+
+	RETURN(rc);
+}
+
 /**
  * send a compound request to the agent
  * \param mti [IN] context
@@ -336,6 +389,61 @@ int mdt_hsm_agent_send(struct mdt_thread_info *mti,
 	ENTRY;
 
 	rc = mdt_hsm_find_best_agent(cdt, hal->hal_archive_id, &uuid);
+	if (rc && hal->hal_archive_id == 0) {
+		uint notrmcount = 0;
+		int rc2 = 0;
+
+		/* special case of remove requests with no archive_id specified,
+		 * and no agent registered to serve all archives, then create a
+		 * set of new requests, each to be sent to each registered
+		 * archives.
+		 * Todo so, find all HSMA_REMOVE entries, and then :
+		 *     _ set completed status as SUCCESS (or FAIL?)
+		 *     _ create a new LLOG record for each archive_id
+		 *       presently being served by any CT
+		 */
+		hai = hai_first(hal);
+		for (i = 0; i < hal->hal_count; i++,
+		     hai = hai_next(hai)) {
+			/* only removes are concerned */
+			if (hai->hai_action != HSMA_REMOVE) {
+				/* count if other actions than HSMA_REMOVE,
+				 * to return original error/rc */
+				notrmcount++;
+				continue;
+			}
+
+			/* send remove request to all registered archive_ids */
+			rc2 = mdt_hsm_send_action_to_each_archive(mti, hai);
+			if (rc2)
+				break;
+
+			/* only update original request as SUCCEED if it has
+			 * been successfully broadcasted to all available
+			 * archive_ids
+			 * XXX: this should only cause duplicates to be sent,
+			 * unless a method to record already successfully
+			 * reached archive_ids is implemented */
+			rc2 = mdt_agent_record_update(mti->mti_env, mdt,
+						     &hai->hai_cookie,
+						     1, ARS_SUCCEED);
+			if (rc2) {
+				CERROR("%s: mdt_agent_record_update() "
+				      "failed, cannot update "
+				      "status to %s for cookie "
+				      "%#llx: rc = %d\n",
+				      mdt_obd_name(mdt),
+				      agent_req_status2name(ARS_SUCCEED),
+				      hai->hai_cookie, rc2);
+				break;
+			}
+		}
+		/* only remove requests with archive_id=0 */
+		if (notrmcount == 0)
+			RETURN(rc2);
+
+	}
+
 	if (rc) {
 		CERROR("%s: Cannot find agent for archive %d: rc = %d\n",
 		       mdt_obd_name(mdt), hal->hal_archive_id, rc);
@@ -346,51 +454,47 @@ int mdt_hsm_agent_send(struct mdt_thread_info *mti,
 	       hal->hal_archive_id);
 
 	len = hal_size(hal);
-	if (kuc_ispayload(hal)) {
-		/* hal is already a kuc payload
-		 * we do not need to alloc a new one
-		 * this avoid a alloc/memcpy/free
-		 */
-		buf = hal;
-	} else {
-		buf = kuc_alloc(len, KUC_TRANSPORT_HSM, HMT_ACTION_LIST);
-		if (IS_ERR(buf))
-			RETURN(PTR_ERR(buf));
-		memcpy(buf, hal, len);
-	}
+	buf = kuc_alloc(len, KUC_TRANSPORT_HSM, HMT_ACTION_LIST);
+	if (IS_ERR(buf))
+		RETURN(PTR_ERR(buf));
+	memcpy(buf, hal, len);
 
 	/* Check if request is still valid (cf file hsm flags) */
 	fail_request = false;
-	hai = hai_zero(hal);
+	hai = hai_first(hal);
 	for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
 		if (hai->hai_action != HSMA_CANCEL) {
 			struct mdt_object *obj;
 			struct md_hsm hsm;
 
-			obj = mdt_hsm_get_md_hsm(mti, &hai->hai_fid, &hsm,
-						 NULL);
-			if (IS_ERR(obj) && (hai->hai_action == HSMA_REMOVE))
-				continue;
+			obj = mdt_hsm_get_md_hsm(mti, &hai->hai_fid, &hsm);
+			if (!IS_ERR(obj) && obj != NULL) {
+				mdt_object_put(mti->mti_env, obj);
+			} else {
+				if (hai->hai_action == HSMA_REMOVE)
+					continue;
 
-			if (IS_ERR(obj) && (PTR_ERR(obj) == -ENOENT)) {
-				fail_request = true;
-				rc = mdt_agent_record_update(mti->mti_env, mdt,
+				if (obj == NULL) {
+					fail_request = true;
+					rc = mdt_agent_record_update(
+							     mti->mti_env, mdt,
 							     &hai->hai_cookie,
 							     1, ARS_FAILED);
-				if (rc) {
-					CERROR("%s: mdt_agent_record_update() "
-					      "failed, rc=%d, cannot update "
+					if (rc) {
+						CERROR(
+					      "%s: mdt_agent_record_update() "
+					      "failed, cannot update "
 					      "status to %s for cookie "
-					      LPX64": rc = %d\n",
-					      mdt_obd_name(mdt), rc,
+					      "%#llx: rc = %d\n",
+					      mdt_obd_name(mdt),
 					      agent_req_status2name(ARS_FAILED),
 					      hai->hai_cookie, rc);
-					GOTO(out_buf, rc);
+						GOTO(out_buf, rc);
+					}
+					continue;
 				}
-				continue;
-			}
-			if (IS_ERR(obj))
 				GOTO(out_buf, rc = PTR_ERR(obj));
+			}
 
 			if (!mdt_hsm_is_action_compat(hai, hal->hal_archive_id,
 						      hal->hal_flags, &hsm)) {
@@ -404,14 +508,41 @@ int mdt_hsm_agent_send(struct mdt_thread_info *mti,
 							     1, ARS_FAILED);
 				if (rc) {
 					CERROR("%s: mdt_agent_record_update() "
-					      "failed, rc=%d, cannot update "
+					      "failed, cannot update "
 					      "status to %s for cookie "
-					      LPX64": rc = %d\n",
-					      mdt_obd_name(mdt), rc,
+					      "%#llx: rc = %d\n",
+					      mdt_obd_name(mdt),
 					      agent_req_status2name(ARS_FAILED),
 					      hai->hai_cookie, rc);
 					GOTO(out_buf, rc);
 				}
+
+				/* if restore and record status updated, give
+				 * back granted layout lock */
+				if (hai->hai_action == HSMA_RESTORE) {
+					struct cdt_restore_handle *crh = NULL;
+					struct mdt_object *obj = NULL;
+
+					mutex_lock(&cdt->cdt_restore_lock);
+					crh = mdt_hsm_restore_hdl_find(cdt,
+								&hai->hai_fid);
+					if (crh != NULL)
+						list_del(&crh->crh_list);
+					mutex_unlock(&cdt->cdt_restore_lock);
+					obj = mdt_object_find(mti->mti_env,
+							      mti->mti_mdt,
+							      &hai->hai_fid);
+					if (!IS_ERR(obj) && crh != NULL)
+						mdt_object_unlock(mti, obj,
+								  &crh->crh_lh,
+								  1);
+					if (crh != NULL)
+						OBD_SLAB_FREE_PTR(crh,
+							mdt_hsm_cdt_kmem);
+					if (!IS_ERR(obj))
+						mdt_object_put(mti->mti_env,
+							       obj);
+				}
 			}
 		}
 	}
@@ -444,7 +575,9 @@ int mdt_hsm_agent_send(struct mdt_thread_info *mti,
 	 * from a server (MDT) to a client (MDC), backwards of normal comms.
 	 */
 	exp = cfs_hash_lookup(mdt2obd_dev(mdt)->obd_uuid_hash, &uuid);
-	if ((exp == NULL) || (exp->exp_disconnected)) {
+	if (exp == NULL || exp->exp_disconnected) {
+		if (exp != NULL)
+			class_export_put(exp);
 		/* This should clean up agents on evicted exports */
 		rc = -ENOENT;
 		CERROR("%s: agent uuid (%s) not found, unregistering:"
@@ -476,7 +609,7 @@ int mdt_hsm_agent_send(struct mdt_thread_info *mti,
 out:
 	if (rc != 0 && is_registered) {
 		/* in case of error, we have to unregister requests */
-		hai = hai_zero(hal);
+		hai = hai_first(hal);
 		for (i = 0; i < hal->hal_count; i++, hai = hai_next(hai)) {
 			if (hai->hai_action == HSMA_CANCEL)
 				continue;
@@ -485,8 +618,7 @@ out:
 	}
 
 out_buf:
-	if (buf != hal)
-		kuc_free(buf, len);
+	kuc_free(buf, len);
 
 	RETURN(rc);
 }
@@ -504,7 +636,7 @@ int mdt_hsm_coordinator_update(struct mdt_thread_info *mti,
 	int      rc;
 
 	ENTRY;
-	/* ask to coodinator to update request state and
+	/* ask to coordinator to update request state and
 	 * to record on disk the result */
 	rc = mdt_hsm_update_request_state(mti, pgs, 1);
 	RETURN(rc);
@@ -517,20 +649,20 @@ static void *mdt_hsm_agent_proc_start(struct seq_file *s, loff_t *off)
 {
 	struct mdt_device	*mdt = s->private;
 	struct coordinator	*cdt = &mdt->mdt_coordinator;
-	cfs_list_t		*pos;
+	struct list_head	*pos;
 	loff_t			 i;
 	ENTRY;
 
 	down_read(&cdt->cdt_agent_lock);
 
-	if (cfs_list_empty(&cdt->cdt_agents))
+	if (list_empty(&cdt->cdt_agents))
 		RETURN(NULL);
 
 	if (*off == 0)
 		RETURN(SEQ_START_TOKEN);
 
 	i = 0;
-	cfs_list_for_each(pos, &cdt->cdt_agents) {
+	list_for_each(pos, &cdt->cdt_agents) {
 		i++;
 		if (i >= *off)
 			RETURN(pos);
@@ -547,7 +679,7 @@ static void *mdt_hsm_agent_proc_next(struct seq_file *s, void *v, loff_t *p)
 {
 	struct mdt_device	*mdt = s->private;
 	struct coordinator	*cdt = &mdt->mdt_coordinator;
-	cfs_list_t		*pos = v;
+	struct list_head	*pos = v;
 	ENTRY;
 
 	if (pos == SEQ_START_TOKEN)
@@ -566,7 +698,7 @@ static void *mdt_hsm_agent_proc_next(struct seq_file *s, void *v, loff_t *p)
  */
 static int mdt_hsm_agent_proc_show(struct seq_file *s, void *v)
 {
-	cfs_list_t		*pos = v;
+	struct list_head	*pos = v;
 	struct hsm_agent	*ha;
 	int			 i;
 	ENTRY;
@@ -574,16 +706,17 @@ static int mdt_hsm_agent_proc_show(struct seq_file *s, void *v)
 	if (pos == SEQ_START_TOKEN)
 		RETURN(0);
 
-	ha = cfs_list_entry(pos, struct hsm_agent, ha_list);
-	seq_printf(s, "uuid=%s archive#=%d (", ha->ha_uuid.uuid,
-		   ha->ha_archive_cnt);
-	if (ha->ha_archive_cnt == 0)
-		seq_printf(s, "all");
-	else
-		for (i = 0; i < ha->ha_archive_cnt; i++)
-			seq_printf(s, "%d ", ha->ha_archive_id[i]);
+	ha = list_entry(pos, struct hsm_agent, ha_list);
+	seq_printf(s, "uuid=%s archive_id=", ha->ha_uuid.uuid);
+	if (ha->ha_archive_cnt == 0) {
+		seq_printf(s, "ANY");
+	} else {
+		seq_printf(s, "%d", ha->ha_archive_id[0]);
+		for (i = 1; i < ha->ha_archive_cnt; i++)
+			seq_printf(s, ",%d", ha->ha_archive_id[i]);
+	}
 
-	seq_printf(s, ") r=%d s=%d f=%d\n",
+	seq_printf(s, " requests=[current:%d ok:%d errors:%d]\n",
 		   atomic_read(&ha->ha_requests),
 		   atomic_read(&ha->ha_success),
 		   atomic_read(&ha->ha_failure));
@@ -619,16 +752,12 @@ static int lprocfs_open_hsm_agent(struct inode *inode, struct file *file)
 	int		 rc;
 	ENTRY;
 
-	if (LPROCFS_ENTRY_AND_CHECK(PDE(inode)))
-			RETURN(-ENOENT);
-
 	rc = seq_open(file, &mdt_hsm_agent_proc_ops);
-	if (rc) {
-		LPROCFS_EXIT();
+	if (rc)
 		RETURN(rc);
-	}
+
 	s = file->private_data;
-	s->private = PDE(inode)->data;
+	s->private = PDE_DATA(inode);
 
 	RETURN(rc);
 }