X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ffld%2Ffld_request.c;h=8d7539d0aff4ef24e5efc0d01f65d59fa14bb7b1;hb=refs%2Fchanges%2F40%2F7140%2F10;hp=536b244a370476f042f696c2430d90c9e8214b8a;hpb=0f8dca08a4f68cba82c2c822998ecc309d3b7aaf;p=fs%2Flustre-release.git

diff --git a/lustre/fld/fld_request.c b/lustre/fld/fld_request.c
index 536b244..8d7539d 100644
--- a/lustre/fld/fld_request.c
+++ b/lustre/fld/fld_request.c
@@ -1,6 +1,4 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
  * GPL HEADER START
  *
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
@@ -28,6 +26,8 @@
 /*
  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
+ *
+ * Copyright (c) 2011, 2013, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -40,15 +40,11 @@
  * Author: Yury Umanets <umka@clusterfs.com>
  */
 
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
 #define DEBUG_SUBSYSTEM S_FLD
 
 #ifdef __KERNEL__
 # include <libcfs/libcfs.h>
 # include <linux/module.h>
-# include <linux/jbd.h>
 # include <asm/div64.h>
 #else /* __KERNEL__ */
 # include <liblustre.h>
@@ -57,12 +53,8 @@
 
 #include <obd.h>
 #include <obd_class.h>
-#include <lustre_ver.h>
 #include <obd_support.h>
 #include <lprocfs_status.h>
-
-#include <dt_object.h>
-#include <md_object.h>
 #include <lustre_req_layout.h>
 #include <lustre_fld.h>
 #include <lustre_mdc.h>
@@ -82,41 +74,41 @@ static int fld_req_avail(struct client_obd *cli, struct mdc_cache_waiter *mcw)
 
 static void fld_enter_request(struct client_obd *cli)
 {
-        struct mdc_cache_waiter mcw;
-        struct l_wait_info lwi = { 0 };
-
-        client_obd_list_lock(&cli->cl_loi_list_lock);
-        if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
-                cfs_list_add_tail(&mcw.mcw_entry, &cli->cl_cache_waiters);
-                cfs_waitq_init(&mcw.mcw_waitq);
-                client_obd_list_unlock(&cli->cl_loi_list_lock);
-                l_wait_event(mcw.mcw_waitq, fld_req_avail(cli, &mcw), &lwi);
-        } else {
-                cli->cl_r_in_flight++;
-                client_obd_list_unlock(&cli->cl_loi_list_lock);
-        }
+	struct mdc_cache_waiter mcw;
+	struct l_wait_info lwi = { 0 };
+
+	client_obd_list_lock(&cli->cl_loi_list_lock);
+	if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
+		cfs_list_add_tail(&mcw.mcw_entry, &cli->cl_cache_waiters);
+		init_waitqueue_head(&mcw.mcw_waitq);
+		client_obd_list_unlock(&cli->cl_loi_list_lock);
+		l_wait_event(mcw.mcw_waitq, fld_req_avail(cli, &mcw), &lwi);
+	} else {
+		cli->cl_r_in_flight++;
+		client_obd_list_unlock(&cli->cl_loi_list_lock);
+	}
 }
 
 static void fld_exit_request(struct client_obd *cli)
 {
-        cfs_list_t *l, *tmp;
-        struct mdc_cache_waiter *mcw;
-
-        client_obd_list_lock(&cli->cl_loi_list_lock);
-        cli->cl_r_in_flight--;
-        cfs_list_for_each_safe(l, tmp, &cli->cl_cache_waiters) {
-
-                if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
-                        /* No free request slots anymore */
-                        break;
-                }
-
-                mcw = cfs_list_entry(l, struct mdc_cache_waiter, mcw_entry);
-                cfs_list_del_init(&mcw->mcw_entry);
-                cli->cl_r_in_flight++;
-                cfs_waitq_signal(&mcw->mcw_waitq);
-        }
-        client_obd_list_unlock(&cli->cl_loi_list_lock);
+	cfs_list_t *l, *tmp;
+	struct mdc_cache_waiter *mcw;
+
+	client_obd_list_lock(&cli->cl_loi_list_lock);
+	cli->cl_r_in_flight--;
+	cfs_list_for_each_safe(l, tmp, &cli->cl_cache_waiters) {
+
+		if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
+			/* No free request slots anymore */
+			break;
+		}
+
+		mcw = cfs_list_entry(l, struct mdc_cache_waiter, mcw_entry);
+		cfs_list_del_init(&mcw->mcw_entry);
+		cli->cl_r_in_flight++;
+		wake_up(&mcw->mcw_waitq);
+	}
+	client_obd_list_unlock(&cli->cl_loi_list_lock);
 }
 
 static int fld_rrb_hash(struct lu_client_fld *fld,
@@ -133,7 +125,15 @@ fld_rrb_scan(struct lu_client_fld *fld, seqno_t seq)
         int hash;
         ENTRY;
 
-        hash = fld_rrb_hash(fld, seq);
+	/* Because almost all of special sequence located in MDT0,
+	 * it should go to index 0 directly, instead of calculating
+	 * hash again, and also if other MDTs is not being connected,
+	 * the fld lookup requests(for seq on MDT0) should not be
+	 * blocked because of other MDTs */
+	if (fid_seq_is_norm(seq))
+		hash = fld_rrb_hash(fld, seq);
+	else
+		hash = 0;
 
         cfs_list_for_each_entry(target, &fld->lcf_targets, ft_chain) {
                 if (target->ft_idx == hash)
@@ -176,17 +176,16 @@ struct lu_fld_hash fld_hash[] = {
 };
 
 static struct lu_fld_target *
-fld_client_get_target(struct lu_client_fld *fld,
-                      seqno_t seq)
+fld_client_get_target(struct lu_client_fld *fld, seqno_t seq)
 {
-        struct lu_fld_target *target;
-        ENTRY;
+	struct lu_fld_target *target;
+	ENTRY;
 
-        LASSERT(fld->lcf_hash != NULL);
+	LASSERT(fld->lcf_hash != NULL);
 
-        cfs_spin_lock(&fld->lcf_lock);
-        target = fld->lcf_hash->fh_scan_func(fld, seq);
-        cfs_spin_unlock(&fld->lcf_lock);
+	spin_lock(&fld->lcf_lock);
+	target = fld->lcf_hash->fh_scan_func(fld, seq);
+	spin_unlock(&fld->lcf_lock);
 
         if (target != NULL) {
                 CDEBUG(D_INFO, "%s: Found target (idx "LPU64
@@ -204,11 +203,12 @@ fld_client_get_target(struct lu_client_fld *fld,
 int fld_client_add_target(struct lu_client_fld *fld,
                           struct lu_fld_target *tar)
 {
-        const char *name = fld_target_name(tar);
+	const char *name;
         struct lu_fld_target *target, *tmp;
         ENTRY;
 
         LASSERT(tar != NULL);
+	name = fld_target_name(tar);
         LASSERT(name != NULL);
         LASSERT(tar->ft_srv != NULL || tar->ft_exp != NULL);
 
@@ -226,10 +226,10 @@ int fld_client_add_target(struct lu_client_fld *fld,
         if (target == NULL)
                 RETURN(-ENOMEM);
 
-        cfs_spin_lock(&fld->lcf_lock);
-        cfs_list_for_each_entry(tmp, &fld->lcf_targets, ft_chain) {
-                if (tmp->ft_idx == tar->ft_idx) {
-                        cfs_spin_unlock(&fld->lcf_lock);
+	spin_lock(&fld->lcf_lock);
+	cfs_list_for_each_entry(tmp, &fld->lcf_targets, ft_chain) {
+		if (tmp->ft_idx == tar->ft_idx) {
+			spin_unlock(&fld->lcf_lock);
                         OBD_FREE_PTR(target);
                         CERROR("Target %s exists in FLD and known as %s:#"LPU64"\n",
                                name, fld_target_name(tmp), tmp->ft_idx);
@@ -247,26 +247,25 @@ int fld_client_add_target(struct lu_client_fld *fld,
                           &fld->lcf_targets);
 
         fld->lcf_count++;
-        cfs_spin_unlock(&fld->lcf_lock);
+	spin_unlock(&fld->lcf_lock);
 
-        RETURN(0);
+	RETURN(0);
 }
 EXPORT_SYMBOL(fld_client_add_target);
 
 /* Remove export from FLD */
-int fld_client_del_target(struct lu_client_fld *fld,
-                          __u64 idx)
+int fld_client_del_target(struct lu_client_fld *fld, __u64 idx)
 {
-        struct lu_fld_target *target, *tmp;
-        ENTRY;
+	struct lu_fld_target *target, *tmp;
+	ENTRY;
 
-        cfs_spin_lock(&fld->lcf_lock);
-        cfs_list_for_each_entry_safe(target, tmp,
-                                     &fld->lcf_targets, ft_chain) {
-                if (target->ft_idx == idx) {
-                        fld->lcf_count--;
-                        cfs_list_del(&target->ft_chain);
-                        cfs_spin_unlock(&fld->lcf_lock);
+	spin_lock(&fld->lcf_lock);
+	cfs_list_for_each_entry_safe(target, tmp,
+				     &fld->lcf_targets, ft_chain) {
+		if (target->ft_idx == idx) {
+			fld->lcf_count--;
+			cfs_list_del(&target->ft_chain);
+			spin_unlock(&fld->lcf_lock);
 
                         if (target->ft_exp != NULL)
                                 class_export_put(target->ft_exp);
@@ -275,46 +274,43 @@ int fld_client_del_target(struct lu_client_fld *fld,
                         RETURN(0);
                 }
         }
-        cfs_spin_unlock(&fld->lcf_lock);
-        RETURN(-ENOENT);
+	spin_unlock(&fld->lcf_lock);
+	RETURN(-ENOENT);
 }
 EXPORT_SYMBOL(fld_client_del_target);
 
-static void fld_client_proc_fini(struct lu_client_fld *fld);
-
 #ifdef LPROCFS
 static int fld_client_proc_init(struct lu_client_fld *fld)
 {
-        int rc;
-        ENTRY;
-
-        fld->lcf_proc_dir = lprocfs_register(fld->lcf_name,
-                                             fld_type_proc_dir,
-                                             NULL, NULL);
-
-        if (IS_ERR(fld->lcf_proc_dir)) {
-                CERROR("%s: LProcFS failed in fld-init\n",
-                       fld->lcf_name);
-                rc = PTR_ERR(fld->lcf_proc_dir);
-                RETURN(rc);
-        }
-
-        rc = lprocfs_add_vars(fld->lcf_proc_dir,
-                              fld_client_proc_list, fld);
-        if (rc) {
-                CERROR("%s: Can't init FLD proc, rc %d\n",
-                       fld->lcf_name, rc);
-                GOTO(out_cleanup, rc);
-        }
-
-        RETURN(0);
+	int rc;
+	ENTRY;
+
+	fld->lcf_proc_dir = lprocfs_seq_register(fld->lcf_name,
+						 fld_type_proc_dir,
+						 NULL, NULL);
+	if (IS_ERR(fld->lcf_proc_dir)) {
+		CERROR("%s: LProcFS failed in fld-init\n",
+		       fld->lcf_name);
+		rc = PTR_ERR(fld->lcf_proc_dir);
+		RETURN(rc);
+	}
+
+	rc = lprocfs_seq_add_vars(fld->lcf_proc_dir,
+				  fld_client_proc_list, fld);
+	if (rc) {
+		CERROR("%s: Can't init FLD proc, rc %d\n",
+		       fld->lcf_name, rc);
+		GOTO(out_cleanup, rc);
+	}
+
+	RETURN(0);
 
 out_cleanup:
-        fld_client_proc_fini(fld);
-        return rc;
+	fld_client_proc_fini(fld);
+	return rc;
 }
 
-static void fld_client_proc_fini(struct lu_client_fld *fld)
+void fld_client_proc_fini(struct lu_client_fld *fld)
 {
         ENTRY;
         if (fld->lcf_proc_dir) {
@@ -330,12 +326,14 @@ static int fld_client_proc_init(struct lu_client_fld *fld)
         return 0;
 }
 
-static void fld_client_proc_fini(struct lu_client_fld *fld)
+void fld_client_proc_fini(struct lu_client_fld *fld)
 {
         return;
 }
 #endif
 
+EXPORT_SYMBOL(fld_client_proc_fini);
+
 static inline int hash_is_sane(int hash)
 {
         return (hash >= 0 && hash < ARRAY_SIZE(fld_hash));
@@ -360,7 +358,7 @@ int fld_client_init(struct lu_client_fld *fld,
         }
 
         fld->lcf_count = 0;
-        cfs_spin_lock_init(&fld->lcf_lock);
+	spin_lock_init(&fld->lcf_lock);
         fld->lcf_hash = &fld_hash[hash];
         fld->lcf_flags = LUSTRE_FLD_INIT;
         CFS_INIT_LIST_HEAD(&fld->lcf_targets);
@@ -395,12 +393,10 @@ EXPORT_SYMBOL(fld_client_init);
 
 void fld_client_fini(struct lu_client_fld *fld)
 {
-        struct lu_fld_target *target, *tmp;
-        ENTRY;
-
-        fld_client_proc_fini(fld);
+	struct lu_fld_target *target, *tmp;
+	ENTRY;
 
-        cfs_spin_lock(&fld->lcf_lock);
+	spin_lock(&fld->lcf_lock);
         cfs_list_for_each_entry_safe(target, tmp,
                                      &fld->lcf_targets, ft_chain) {
                 fld->lcf_count--;
@@ -409,7 +405,7 @@ void fld_client_fini(struct lu_client_fld *fld)
                         class_export_put(target->ft_exp);
                 OBD_FREE_PTR(target);
         }
-        cfs_spin_unlock(&fld->lcf_lock);
+	spin_unlock(&fld->lcf_lock);
 
         if (fld->lcf_cache != NULL) {
                 if (!IS_ERR(fld->lcf_cache))
@@ -422,65 +418,98 @@ void fld_client_fini(struct lu_client_fld *fld)
 EXPORT_SYMBOL(fld_client_fini);
 
 int fld_client_rpc(struct obd_export *exp,
-                   struct lu_seq_range *range, __u32 fld_op)
+		   struct lu_seq_range *range, __u32 fld_op,
+		   struct ptlrpc_request **reqp)
 {
-        struct ptlrpc_request *req;
-        struct lu_seq_range      *prange;
-        __u32                 *op;
-        int                    rc;
-        ENTRY;
-
-        LASSERT(exp != NULL);
-
-        req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_FLD_QUERY,
-                                        LUSTRE_MDS_VERSION, FLD_QUERY);
-        if (req == NULL)
-                RETURN(-ENOMEM);
-
-        op = req_capsule_client_get(&req->rq_pill, &RMF_FLD_OPC);
-        *op = fld_op;
-
-        prange = req_capsule_client_get(&req->rq_pill, &RMF_FLD_MDFLD);
-        *prange = *range;
-
-        ptlrpc_request_set_replen(req);
+	struct ptlrpc_request *req = NULL;
+	struct lu_seq_range   *prange;
+	__u32                 *op;
+	int                    rc = 0;
+	struct obd_import     *imp;
+	ENTRY;
+
+	LASSERT(exp != NULL);
+
+	imp = class_exp2cliimp(exp);
+	switch (fld_op) {
+	case FLD_QUERY:
+		req = ptlrpc_request_alloc_pack(imp, &RQF_FLD_QUERY,
+						LUSTRE_MDS_VERSION, FLD_QUERY);
+		if (req == NULL)
+			RETURN(-ENOMEM);
+
+		/* XXX: only needed when talking to old server(< 2.6), it should
+		 * be removed when < 2.6 server is not supported */
+		op = req_capsule_client_get(&req->rq_pill, &RMF_FLD_OPC);
+		*op = FLD_LOOKUP;
+
+		if (imp->imp_connect_flags_orig & OBD_CONNECT_MDS_MDS)
+			req->rq_allow_replay = 1;
+		break;
+	case FLD_READ:
+		req = ptlrpc_request_alloc_pack(imp, &RQF_FLD_READ,
+						LUSTRE_MDS_VERSION, FLD_READ);
+		if (req == NULL)
+			RETURN(-ENOMEM);
+
+		req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA,
+				     RCL_SERVER, PAGE_CACHE_SIZE);
+		break;
+	default:
+		rc = -EINVAL;
+		break;
+	}
+
+	if (rc != 0)
+		RETURN(rc);
+
+	prange = req_capsule_client_get(&req->rq_pill, &RMF_FLD_MDFLD);
+	*prange = *range;
+	ptlrpc_request_set_replen(req);
         req->rq_request_portal = FLD_REQUEST_PORTAL;
+	req->rq_reply_portal = MDC_REPLY_PORTAL;
         ptlrpc_at_set_req_timeout(req);
 
-        if (fld_op != FLD_LOOKUP)
-                mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
-        fld_enter_request(&exp->exp_obd->u.cli);
-        rc = ptlrpc_queue_wait(req);
-        fld_exit_request(&exp->exp_obd->u.cli);
-        if (fld_op != FLD_LOOKUP)
-                mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
-        if (rc)
-                GOTO(out_req, rc);
-
-        prange = req_capsule_server_get(&req->rq_pill, &RMF_FLD_MDFLD);
-        if (prange == NULL)
-                GOTO(out_req, rc = -EFAULT);
-        *range = *prange;
-        EXIT;
+	fld_enter_request(&exp->exp_obd->u.cli);
+	rc = ptlrpc_queue_wait(req);
+	fld_exit_request(&exp->exp_obd->u.cli);
+	if (rc)
+		GOTO(out_req, rc);
+
+	if (fld_op == FLD_QUERY) {
+		prange = req_capsule_server_get(&req->rq_pill,
+						&RMF_FLD_MDFLD);
+		if (prange == NULL)
+			GOTO(out_req, rc = -EFAULT);
+		*range = *prange;
+	}
+
+	EXIT;
 out_req:
-        ptlrpc_req_finished(req);
-        return rc;
+	if (rc != 0 || reqp == NULL) {
+		ptlrpc_req_finished(req);
+		req = NULL;
+	}
+
+	if (reqp != NULL)
+		*reqp = req;
+
+	return rc;
 }
 
-int fld_client_lookup(struct lu_client_fld *fld,
-                      seqno_t seq, mdsno_t *mds,
-                      const struct lu_env *env)
+int fld_client_lookup(struct lu_client_fld *fld, seqno_t seq, mdsno_t *mds,
+		      __u32 flags, const struct lu_env *env)
 {
-        struct lu_seq_range res;
-        struct lu_fld_target *target;
-        int rc;
-        ENTRY;
+	struct lu_seq_range res = { 0 };
+	struct lu_fld_target *target;
+	int rc;
+	ENTRY;
 
         fld->lcf_flags |= LUSTRE_FLD_RUN;
 
         rc = fld_cache_lookup(fld->lcf_cache, seq, &res);
         if (rc == 0) {
-                *mds = res.lsr_mdt;
+                *mds = res.lsr_index;
                 RETURN(0);
         }
 
@@ -492,26 +521,25 @@ int fld_client_lookup(struct lu_client_fld *fld,
                "target %s (idx "LPU64")\n", fld->lcf_name, seq,
                fld_target_name(target), target->ft_idx);
 
-        res.lsr_start = seq;
-#ifdef __KERNEL__
-        if (target->ft_srv != NULL) {
-                LASSERT(env != NULL);
-                rc = fld_server_lookup(target->ft_srv,
-                                       env, seq, &res);
-        } else {
-#endif
-                rc = fld_client_rpc(target->ft_exp,
-                                    &res, FLD_LOOKUP);
-#ifdef __KERNEL__
-        }
+	res.lsr_start = seq;
+	fld_range_set_type(&res, flags);
+
+#if defined(__KERNEL__) && defined(HAVE_SERVER_SUPPORT)
+	if (target->ft_srv != NULL) {
+		LASSERT(env != NULL);
+		rc = fld_server_lookup(env, target->ft_srv, seq, &res);
+	} else
 #endif
+	{
+		rc = fld_client_rpc(target->ft_exp, &res, FLD_QUERY, NULL);
+	}
 
-        if (rc == 0) {
-                *mds = res.lsr_mdt;
+	if (rc == 0) {
+		*mds = res.lsr_index;
+		fld_cache_insert(fld->lcf_cache, &res);
+	}
 
-                fld_cache_insert(fld->lcf_cache, &res);
-        }
-        RETURN(rc);
+	RETURN(rc);
 }
 EXPORT_SYMBOL(fld_client_lookup);
 
@@ -520,3 +548,41 @@ void fld_client_flush(struct lu_client_fld *fld)
         fld_cache_flush(fld->lcf_cache);
 }
 EXPORT_SYMBOL(fld_client_flush);
+
+#ifdef __KERNEL__
+
+struct proc_dir_entry *fld_type_proc_dir;
+
+static int __init fld_mod_init(void)
+{
+	fld_type_proc_dir = lprocfs_seq_register(LUSTRE_FLD_NAME,
+						 proc_lustre_root,
+						 NULL, NULL);
+	if (IS_ERR(fld_type_proc_dir))
+		return PTR_ERR(fld_type_proc_dir);
+
+#ifdef HAVE_SERVER_SUPPORT
+	fld_server_mod_init();
+#endif
+
+	return 0;
+}
+
+static void __exit fld_mod_exit(void)
+{
+#ifdef HAVE_SERVER_SUPPORT
+	fld_server_mod_exit();
+#endif
+
+	if (fld_type_proc_dir != NULL && !IS_ERR(fld_type_proc_dir)) {
+		lprocfs_remove(&fld_type_proc_dir);
+		fld_type_proc_dir = NULL;
+	}
+}
+
+MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
+MODULE_DESCRIPTION("Lustre FLD");
+MODULE_LICENSE("GPL");
+
+cfs_module(mdd, LUSTRE_VERSION_STRING, fld_mod_init, fld_mod_exit);
+#endif /* __KERNEL__ */