Whamcloud - gitweb
LU-1866 lfsck: enhance otable-based iteration
[fs/lustre-release.git] / lustre / fid / fid_request.c
index bc0f46b..6facae1 100644 (file)
@@ -27,7 +27,7 @@
  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011, 2012, Whamcloud, Inc.
+ * Copyright (c) 2011, 2012, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -87,34 +87,39 @@ static int seq_client_rpc(struct lu_client_seq *seq,
 
        ptlrpc_request_set_replen(req);
 
-       if (seq->lcs_type == LUSTRE_SEQ_METADATA) {
-               req->rq_request_portal = SEQ_METADATA_PORTAL;
+       in->lsr_index = seq->lcs_space.lsr_index;
+       if (seq->lcs_type == LUSTRE_SEQ_METADATA)
                in->lsr_flags = LU_SEQ_RANGE_MDT;
-       } else {
-               LASSERTF(seq->lcs_type == LUSTRE_SEQ_DATA,
-                        "unknown lcs_type %u\n", seq->lcs_type);
-               req->rq_request_portal = SEQ_DATA_PORTAL;
+       else
                in->lsr_flags = LU_SEQ_RANGE_OST;
-       }
 
        if (opc == SEQ_ALLOC_SUPER) {
-               /* Update index field of *in, it is required for
-                * FLD update on super sequence allocator node. */
-               in->lsr_index = seq->lcs_space.lsr_index;
                req->rq_request_portal = SEQ_CONTROLLER_PORTAL;
+               req->rq_reply_portal = MDC_REPLY_PORTAL;
+               /* During allocating super sequence for data object,
+                * the current thread might hold the export of MDT0(MDT0
+                * precreating objects on this OST), and it will send the
+                * request to MDT0 here, so we can not keep resending the
+                * request here, otherwise if MDT0 is failed(umounted),
+                * it can not release the export of MDT0 */
+               if (seq->lcs_type == LUSTRE_SEQ_DATA)
+                       req->rq_no_delay = req->rq_no_resend = 1;
                debug_mask = D_CONSOLE;
        } else {
+               if (seq->lcs_type == LUSTRE_SEQ_METADATA)
+                       req->rq_request_portal = SEQ_METADATA_PORTAL;
+               else
+                       req->rq_request_portal = SEQ_DATA_PORTAL;
                debug_mask = D_INFO;
-               LASSERTF(opc == SEQ_ALLOC_META,
-                        "unknown opcode %u\n, opc", opc);
        }
 
        ptlrpc_at_set_req_timeout(req);
 
-       mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
+       if (seq->lcs_type == LUSTRE_SEQ_METADATA)
+               mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
        rc = ptlrpc_queue_wait(req);
-       mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
-
+       if (seq->lcs_type == LUSTRE_SEQ_METADATA)
+               mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
        if (rc)
                GOTO(out_req, rc);
 
@@ -149,7 +154,7 @@ int seq_client_alloc_super(struct lu_client_seq *seq,
         int rc;
         ENTRY;
 
-        cfs_mutex_lock(&seq->lcs_mutex);
+       mutex_lock(&seq->lcs_mutex);
 
 #ifdef __KERNEL__
         if (seq->lcs_srv) {
@@ -158,12 +163,19 @@ int seq_client_alloc_super(struct lu_client_seq *seq,
                                             env);
         } else {
 #endif
-                rc = seq_client_rpc(seq, &seq->lcs_space,
+               /* Check whether the connection to seq controller has been
+                * setup (lcs_exp != NULL) */
+               if (seq->lcs_exp == NULL) {
+                       mutex_unlock(&seq->lcs_mutex);
+                       RETURN(-EINPROGRESS);
+               }
+
+               rc = seq_client_rpc(seq, &seq->lcs_space,
                                     SEQ_ALLOC_SUPER, "super");
 #ifdef __KERNEL__
         }
 #endif
-        cfs_mutex_unlock(&seq->lcs_mutex);
+       mutex_unlock(&seq->lcs_mutex);
         RETURN(rc);
 }
 
@@ -180,8 +192,14 @@ static int seq_client_alloc_meta(const struct lu_env *env,
                 rc = seq_server_alloc_meta(seq->lcs_srv, &seq->lcs_space, env);
         } else {
 #endif
-                rc = seq_client_rpc(seq, &seq->lcs_space,
-                                    SEQ_ALLOC_META, "meta");
+               do {
+                       /* If meta server return -EINPROGRESS or EAGAIN,
+                        * it means meta server might not be ready to
+                        * allocate super sequence from sequence controller
+                        * (MDT0)yet */
+                       rc = seq_client_rpc(seq, &seq->lcs_space,
+                                           SEQ_ALLOC_META, "meta");
+               } while (rc == -EINPROGRESS || rc == -EAGAIN);
 #ifdef __KERNEL__
         }
 #endif
@@ -227,29 +245,31 @@ static int seq_fid_alloc_prep(struct lu_client_seq *seq,
         if (seq->lcs_update) {
                 cfs_waitq_add(&seq->lcs_waitq, link);
                 cfs_set_current_state(CFS_TASK_UNINT);
-                cfs_mutex_unlock(&seq->lcs_mutex);
+               mutex_unlock(&seq->lcs_mutex);
 
                 cfs_waitq_wait(link, CFS_TASK_UNINT);
 
-                cfs_mutex_lock(&seq->lcs_mutex);
+               mutex_lock(&seq->lcs_mutex);
                 cfs_waitq_del(&seq->lcs_waitq, link);
                 cfs_set_current_state(CFS_TASK_RUNNING);
                 return -EAGAIN;
         }
         ++seq->lcs_update;
-        cfs_mutex_unlock(&seq->lcs_mutex);
+       mutex_unlock(&seq->lcs_mutex);
         return 0;
 }
 
 static void seq_fid_alloc_fini(struct lu_client_seq *seq)
 {
         LASSERT(seq->lcs_update == 1);
-        cfs_mutex_lock(&seq->lcs_mutex);
+       mutex_lock(&seq->lcs_mutex);
         --seq->lcs_update;
         cfs_waitq_signal(&seq->lcs_waitq);
 }
 
-/* Allocate the whole seq to the caller*/
+/**
+ * Allocate the whole seq to the caller.
+ **/
 int seq_client_get_seq(const struct lu_env *env,
                        struct lu_client_seq *seq, seqno_t *seqnr)
 {
@@ -257,7 +277,7 @@ int seq_client_get_seq(const struct lu_env *env,
         int rc;
 
         LASSERT(seqnr != NULL);
-        cfs_mutex_lock(&seq->lcs_mutex);
+       mutex_lock(&seq->lcs_mutex);
         cfs_waitlink_init(&link);
 
         while (1) {
@@ -271,25 +291,28 @@ int seq_client_get_seq(const struct lu_env *env,
                 CERROR("%s: Can't allocate new sequence, "
                        "rc %d\n", seq->lcs_name, rc);
                 seq_fid_alloc_fini(seq);
-                cfs_mutex_unlock(&seq->lcs_mutex);
+               mutex_unlock(&seq->lcs_mutex);
                 return rc;
         }
 
         CDEBUG(D_INFO, "%s: allocate sequence "
                "[0x%16.16"LPF64"x]\n", seq->lcs_name, *seqnr);
 
-        /*Since the caller require the whole seq,
-         *so marked this seq to be used*/
-        seq->lcs_fid.f_oid = LUSTRE_SEQ_MAX_WIDTH;
-        seq->lcs_fid.f_seq = *seqnr;
-        seq->lcs_fid.f_ver = 0;
+       /* Since the caller require the whole seq,
+        * so marked this seq to be used */
+       if (seq->lcs_type == LUSTRE_SEQ_METADATA)
+               seq->lcs_fid.f_oid = LUSTRE_METADATA_SEQ_MAX_WIDTH;
+       else
+               seq->lcs_fid.f_oid = LUSTRE_DATA_SEQ_MAX_WIDTH;
 
+       seq->lcs_fid.f_seq = *seqnr;
+       seq->lcs_fid.f_ver = 0;
         /*
          * Inform caller that sequence switch is performed to allow it
          * to setup FLD for it.
          */
         seq_fid_alloc_fini(seq);
-        cfs_mutex_unlock(&seq->lcs_mutex);
+       mutex_unlock(&seq->lcs_mutex);
 
         return rc;
 }
@@ -307,7 +330,7 @@ int seq_client_alloc_fid(const struct lu_env *env,
         LASSERT(fid != NULL);
 
         cfs_waitlink_init(&link);
-        cfs_mutex_lock(&seq->lcs_mutex);
+       mutex_lock(&seq->lcs_mutex);
 
        if (OBD_FAIL_CHECK(OBD_FAIL_SEQ_EXHAUST))
                seq->lcs_fid.f_oid = seq->lcs_width;
@@ -332,7 +355,7 @@ int seq_client_alloc_fid(const struct lu_env *env,
                         CERROR("%s: Can't allocate new sequence, "
                                "rc %d\n", seq->lcs_name, rc);
                         seq_fid_alloc_fini(seq);
-                        cfs_mutex_unlock(&seq->lcs_mutex);
+                       mutex_unlock(&seq->lcs_mutex);
                         RETURN(rc);
                 }
 
@@ -354,7 +377,7 @@ int seq_client_alloc_fid(const struct lu_env *env,
         }
 
         *fid = seq->lcs_fid;
-        cfs_mutex_unlock(&seq->lcs_mutex);
+       mutex_unlock(&seq->lcs_mutex);
 
         CDEBUG(D_INFO, "%s: Allocated FID "DFID"\n", seq->lcs_name,  PFID(fid));
         RETURN(rc);
@@ -371,16 +394,16 @@ void seq_client_flush(struct lu_client_seq *seq)
 
         LASSERT(seq != NULL);
         cfs_waitlink_init(&link);
-        cfs_mutex_lock(&seq->lcs_mutex);
+       mutex_lock(&seq->lcs_mutex);
 
         while (seq->lcs_update) {
                 cfs_waitq_add(&seq->lcs_waitq, &link);
                 cfs_set_current_state(CFS_TASK_UNINT);
-                cfs_mutex_unlock(&seq->lcs_mutex);
+               mutex_unlock(&seq->lcs_mutex);
 
                 cfs_waitq_wait(&link, CFS_TASK_UNINT);
 
-                cfs_mutex_lock(&seq->lcs_mutex);
+               mutex_lock(&seq->lcs_mutex);
                 cfs_waitq_del(&seq->lcs_waitq, &link);
                 cfs_set_current_state(CFS_TASK_RUNNING);
         }
@@ -394,7 +417,7 @@ void seq_client_flush(struct lu_client_seq *seq)
         seq->lcs_space.lsr_index = -1;
 
         range_init(&seq->lcs_space);
-        cfs_mutex_unlock(&seq->lcs_mutex);
+       mutex_unlock(&seq->lcs_mutex);
 }
 EXPORT_SYMBOL(seq_client_flush);
 
@@ -460,36 +483,37 @@ int seq_client_init(struct lu_client_seq *seq,
                     const char *prefix,
                     struct lu_server_seq *srv)
 {
-        int rc;
-        ENTRY;
+       int rc;
+       ENTRY;
 
-        LASSERT(seq != NULL);
-        LASSERT(prefix != NULL);
+       LASSERT(seq != NULL);
+       LASSERT(prefix != NULL);
 
-        seq->lcs_exp = exp;
-        seq->lcs_srv = srv;
-        seq->lcs_type = type;
-        cfs_mutex_init(&seq->lcs_mutex);
-        seq->lcs_width = LUSTRE_SEQ_MAX_WIDTH;
-        cfs_waitq_init(&seq->lcs_waitq);
+       seq->lcs_srv = srv;
+       seq->lcs_type = type;
 
-        /* Make sure that things are clear before work is started. */
-        seq_client_flush(seq);
+       mutex_init(&seq->lcs_mutex);
+       if (type == LUSTRE_SEQ_METADATA)
+               seq->lcs_width = LUSTRE_METADATA_SEQ_MAX_WIDTH;
+       else
+               seq->lcs_width = LUSTRE_DATA_SEQ_MAX_WIDTH;
 
-        if (exp == NULL) {
-                LASSERT(seq->lcs_srv != NULL);
-        } else {
-                LASSERT(seq->lcs_exp != NULL);
-                seq->lcs_exp = class_export_get(seq->lcs_exp);
-        }
+       cfs_waitq_init(&seq->lcs_waitq);
+       /* Make sure that things are clear before work is started. */
+       seq_client_flush(seq);
 
-        snprintf(seq->lcs_name, sizeof(seq->lcs_name),
-                 "cli-%s", prefix);
+       if (exp != NULL)
+               seq->lcs_exp = class_export_get(exp);
+       else if (type == LUSTRE_SEQ_METADATA)
+               LASSERT(seq->lcs_srv != NULL);
 
-        rc = seq_client_proc_init(seq);
-        if (rc)
-                seq_client_fini(seq);
-        RETURN(rc);
+       snprintf(seq->lcs_name, sizeof(seq->lcs_name),
+                "cli-%s", prefix);
+
+       rc = seq_client_proc_init(seq);
+       if (rc)
+               seq_client_fini(seq);
+       RETURN(rc);
 }
 EXPORT_SYMBOL(seq_client_init);