Whamcloud - gitweb
Branch HEAD
[fs/lustre-release.git] / lustre / lov / lov_request.c
index 5d80fcb..f0eee27 100644 (file)
@@ -1,25 +1,37 @@
 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
  * vim:expandtab:shiftwidth=8:tabstop=8:
  *
- * Copyright (C) 2002, 2003 Cluster File Systems, Inc.
+ * GPL HEADER START
  *
- *   This file is part of the Lustre file system, http://www.lustre.org
- *   Lustre is a trademark of Cluster File Systems, Inc.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
- *   You may have signed or agreed to another license before downloading
- *   this software.  If so, you are bound by the terms and conditions
- *   of that agreement, and the following does not apply to you.  See the
- *   LICENSE file included with this distribution for more information.
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
  *
- *   If you did not agree to a different license, then this copy of Lustre
- *   is open source software; you can redistribute it and/or modify it
- *   under the terms of version 2 of the GNU General Public License as
- *   published by the Free Software Foundation.
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
  *
- *   In either case, Lustre is distributed in the hope that it will be
- *   useful, but WITHOUT ANY WARRANTY; without even the implied warranty
- *   of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- *   license text for more details.
+ * You should have received a copy of the GNU General Public License
+ * version 2 along with this program; If not, see
+ * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
+ *
+ * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
+ * CA 95054 USA or visit www.sun.com if you need additional information or
+ * have any questions.
+ *
+ * GPL HEADER END
+ */
+/*
+ * Copyright  2008 Sun Microsystems, Inc. All rights reserved
+ * Use is subject to license terms.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
  */
 
 #ifndef EXPORT_SYMTAB
@@ -47,6 +59,7 @@ static void lov_init_set(struct lov_request_set *set)
         set->set_cookies = 0;
         CFS_INIT_LIST_HEAD(&set->set_list);
         atomic_set(&set->set_refcount, 1);
+        cfs_waitq_init(&set->set_waitq);
 }
 
 static void lov_finish_set(struct lov_request_set *set)
@@ -81,6 +94,14 @@ static void lov_finish_set(struct lov_request_set *set)
         EXIT;
 }
 
+int lov_finished_set(struct lov_request_set *set)
+{
+        CDEBUG(D_INFO, "check set %d/%d\n", set->set_completes,
+               set->set_count);
+        return set->set_completes == set->set_count;
+}
+
+
 void lov_update_set(struct lov_request_set *set,
                     struct lov_request *req, int rc)
 {
@@ -90,6 +111,8 @@ void lov_update_set(struct lov_request_set *set,
         set->set_completes++;
         if (rc == 0)
                 set->set_success++;
+
+        cfs_waitq_signal(&set->set_waitq);
 }
 
 int lov_update_common_set(struct lov_request_set *set,
@@ -101,7 +124,7 @@ int lov_update_common_set(struct lov_request_set *set,
         lov_update_set(set, req, rc);
 
         /* grace error on inactive ost */
-        if (rc && !(lov->lov_tgts[req->rq_idx] && 
+        if (rc && !(lov->lov_tgts[req->rq_idx] &&
                     lov->lov_tgts[req->rq_idx]->ltd_active))
                 rc = 0;
 
@@ -113,20 +136,47 @@ void lov_set_add_req(struct lov_request *req, struct lov_request_set *set)
 {
         list_add_tail(&req->rq_link, &set->set_list);
         set->set_count++;
+        req->rq_rqset = set;
+}
+
+extern void osc_update_enqueue(struct lustre_handle *lov_lockhp,
+                               struct lov_oinfo *loi, int flags,
+                               struct ost_lvb *lvb, __u32 mode, int rc);
+
+static int lov_update_enqueue_lov(struct obd_export *exp,
+                                  struct lustre_handle *lov_lockhp,
+                                  struct lov_oinfo *loi, int flags, int idx,
+                                  __u64 oid, int rc)
+{
+        struct lov_obd *lov = &exp->exp_obd->u.lov;
+
+        if (rc != ELDLM_OK &&
+            !(rc == ELDLM_LOCK_ABORTED && (flags & LDLM_FL_HAS_INTENT))) {
+                memset(lov_lockhp, 0, sizeof(*lov_lockhp));
+                if (lov->lov_tgts[idx] && lov->lov_tgts[idx]->ltd_active) {
+                        /* -EUSERS used by OST to report file contention */
+                        if (rc != -EINTR && rc != -EUSERS)
+                                CERROR("enqueue objid "LPX64" subobj "
+                                       LPX64" on OST idx %d: rc %d\n",
+                                       oid, loi->loi_id, loi->loi_ost_idx, rc);
+                } else
+                        rc = ELDLM_OK;
+        }
+        return rc;
 }
 
 int lov_update_enqueue_set(struct lov_request *req, __u32 mode, int rc)
 {
         struct lov_request_set *set = req->rq_rqset;
         struct lustre_handle *lov_lockhp;
+        struct obd_info *oi = set->set_oi;
         struct lov_oinfo *loi;
         ENTRY;
 
-        LASSERT(set != NULL);
-        LASSERT(set->set_oi != NULL);
+        LASSERT(oi != NULL);
 
         lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
-        loi = set->set_oi->oi_md->lsm_oinfo[req->rq_stripe];
+        loi = oi->oi_md->lsm_oinfo[req->rq_stripe];
 
         /* XXX LOV STACKING: OSC gets a copy, created in lov_prep_enqueue_set
          * and that copy can be arbitrarily out of date.
@@ -134,64 +184,22 @@ int lov_update_enqueue_set(struct lov_request *req, __u32 mode, int rc)
          * The LOV API is due for a serious rewriting anyways, and this
          * can be addressed then. */
 
-        if (rc == ELDLM_OK) {
-                struct ldlm_lock *lock = ldlm_handle2lock(lov_lockhp);
-                __u64 tmp;
-
-                LASSERT(lock != NULL);
-                lov_stripe_lock(set->set_oi->oi_md);
-                loi->loi_lvb = req->rq_oi.oi_md->lsm_oinfo[0]->loi_lvb;
-                tmp = loi->loi_lvb.lvb_size;
-                /* Extend KMS up to the end of this lock and no further
-                 * A lock on [x,y] means a KMS of up to y + 1 bytes! */
-                if (tmp > lock->l_policy_data.l_extent.end)
-                        tmp = lock->l_policy_data.l_extent.end + 1;
-                if (tmp >= loi->loi_kms) {
-                        LDLM_DEBUG(lock, "lock acquired, setting rss="LPU64
-                                   ", kms="LPU64, loi->loi_lvb.lvb_size, tmp);
-                        loi->loi_kms = tmp;
-                        loi->loi_kms_valid = 1;
-                } else {
-                        LDLM_DEBUG(lock, "lock acquired, setting rss="
-                                   LPU64"; leaving kms="LPU64", end="LPU64,
-                                   loi->loi_lvb.lvb_size, loi->loi_kms,
-                                   lock->l_policy_data.l_extent.end);
-                }
-                lov_stripe_unlock(set->set_oi->oi_md);
-                ldlm_lock_allow_match(lock);
-                LDLM_LOCK_PUT(lock);
-        } else if ((rc == ELDLM_LOCK_ABORTED) &&
-                   (set->set_oi->oi_flags & LDLM_FL_HAS_INTENT)) {
-                memset(lov_lockhp, 0, sizeof(*lov_lockhp));
-                lov_stripe_lock(set->set_oi->oi_md);
-                loi->loi_lvb = req->rq_oi.oi_md->lsm_oinfo[0]->loi_lvb;
-                lov_stripe_unlock(set->set_oi->oi_md);
-                CDEBUG(D_INODE, "glimpsed, setting rss="LPU64"; leaving"
-                       " kms="LPU64"\n", loi->loi_lvb.lvb_size, loi->loi_kms);
-                rc = ELDLM_OK;
-        } else {
-                struct obd_export *exp = set->set_exp;
-                struct lov_obd *lov = &exp->exp_obd->u.lov;
-
-                memset(lov_lockhp, 0, sizeof(*lov_lockhp));
-                if (lov->lov_tgts[req->rq_idx] && 
-                    lov->lov_tgts[req->rq_idx]->ltd_active) {
-                        if (rc != -EINTR)
-                                CERROR("enqueue objid "LPX64" subobj "
-                                       LPX64" on OST idx %d: rc %d\n",
-                                       set->set_oi->oi_md->lsm_object_id,
-                                       loi->loi_id, loi->loi_ost_idx, rc);
-                } else {
-                        rc = ELDLM_OK;
-                }
-        }
+        lov_stripe_lock(oi->oi_md);
+        osc_update_enqueue(lov_lockhp, loi, oi->oi_flags,
+                           &req->rq_oi.oi_md->lsm_oinfo[0]->loi_lvb, mode, rc);
+        if (rc == ELDLM_LOCK_ABORTED && (oi->oi_flags & LDLM_FL_HAS_INTENT))
+                memset(lov_lockhp, 0, sizeof *lov_lockhp);
+        rc = lov_update_enqueue_lov(set->set_exp, lov_lockhp, loi, oi->oi_flags,
+                                    req->rq_idx, oi->oi_md->lsm_object_id, rc);
+        lov_stripe_unlock(oi->oi_md);
         lov_update_set(set, req, rc);
         RETURN(rc);
 }
 
 /* The callback for osc_enqueue that updates lov info for every OSC request. */
-static int cb_update_enqueue(struct obd_info *oinfo, int rc)
+static int cb_update_enqueue(void *cookie, int rc)
 {
+        struct obd_info *oinfo = cookie;
         struct ldlm_enqueue_info *einfo;
         struct lov_request *lovreq;
 
@@ -317,8 +325,6 @@ int lov_prep_enqueue_set(struct obd_export *exp, struct obd_info *oinfo,
                         ((void *)req->rq_oi.oi_md) + sizeof(*req->rq_oi.oi_md) +
                         sizeof(struct lov_oinfo *);
 
-
-                req->rq_rqset = set;
                 /* Set lov request specific parameters. */
                 req->rq_oi.oi_lockh = set->set_lockh->llh_handles + i;
                 req->rq_oi.oi_cb_up = cb_update_enqueue;
@@ -570,9 +576,6 @@ static int create_done(struct obd_export *exp, struct lov_request_set *set,
 
                         rc = qos_remedy_create(set, req);
                         lov_update_create_set(set, req, rc);
-
-                        if (rc)
-                                break;
                 }
         }
 
@@ -580,11 +583,7 @@ static int create_done(struct obd_export *exp, struct lov_request_set *set,
         if (set->set_success == 0)
                 GOTO(cleanup, rc);
 
-        /* If there was an explicit stripe set, fail.  Otherwise, we
-         * got some objects and that's not bad. */
         if (set->set_count != set->set_success) {
-                if (*lsmp)
-                        GOTO(cleanup, rc);
                 set->set_count = set->set_success;
                 qos_shrink_lsm(set);
         }
@@ -624,7 +623,8 @@ cleanup:
                         continue;
 
                 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
-                err = obd_destroy(sub_exp, req->rq_oi.oi_oa, NULL, oti, NULL);
+                err = obd_destroy(sub_exp, req->rq_oi.oi_oa, NULL, oti, NULL,
+                                  NULL);
                 if (err)
                         CERROR("Failed to uncreate objid "LPX64" subobj "
                                LPX64" on OST idx %d: rc = %d\n",
@@ -705,6 +705,16 @@ int lov_update_create_set(struct lov_request_set *set,
         RETURN(0);
 }
 
+int cb_create_update(void *cookie, int rc)
+{
+        struct obd_info *oinfo = cookie;
+        struct lov_request *lovreq;
+
+        lovreq = container_of(oinfo, struct lov_request, rq_oi);
+        return lov_update_create_set(lovreq->rq_rqset, lovreq, rc);
+}
+
+
 int lov_prep_create_set(struct obd_export *exp, struct obd_info *oinfo,
                         struct lov_stripe_md **lsmp, struct obdo *src_oa,
                         struct obd_trans_info *oti,
@@ -726,6 +736,8 @@ int lov_prep_create_set(struct obd_export *exp, struct obd_info *oinfo,
         set->set_oti = oti;
 
         rc = qos_prep_create(exp, set);
+        /* qos_shrink_lsm() may have allocated a new lsm */
+        *lsmp = oinfo->oi_md;
         if (rc)
                 lov_fini_create_set(set, lsmp);
         else
@@ -864,9 +876,9 @@ int lov_prep_brw_set(struct obd_export *exp, struct obd_info *oinfo,
 
                 if (info[i].count == 0)
                         continue;
-                
+
                 loi = oinfo->oi_md->lsm_oinfo[i];
-                if (!lov->lov_tgts[loi->loi_ost_idx] || 
+                if (!lov->lov_tgts[loi->loi_ost_idx] ||
                     !lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
                         GOTO(out, rc = -EIO);
@@ -958,9 +970,10 @@ int lov_fini_getattr_set(struct lov_request_set *set)
 }
 
 /* The callback for osc_getattr_async that finilizes a request info when a
- * response is recieved. */
-static int cb_getattr_update(struct obd_info *oinfo, int rc)
+ * response is received. */
+static int cb_getattr_update(void *cookie, int rc)
 {
+        struct obd_info *oinfo = cookie;
         struct lov_request *lovreq;
         lovreq = container_of(oinfo, struct lov_request, rq_oi);
         return lov_update_common_set(lovreq->rq_rqset, lovreq, rc);
@@ -1010,7 +1023,6 @@ int lov_prep_getattr_set(struct obd_export *exp, struct obd_info *oinfo,
                 req->rq_oi.oi_oa->o_id = loi->loi_id;
                 req->rq_oi.oi_cb_up = cb_getattr_update;
                 req->rq_oi.oi_capa = oinfo->oi_capa;
-                req->rq_rqset = set;
 
                 lov_set_add_req(req, set);
         }
@@ -1068,7 +1080,7 @@ int lov_prep_destroy_set(struct obd_export *exp, struct obd_info *oinfo,
                 struct lov_request *req;
 
                 loi = lsm->lsm_oinfo[i];
-                if (!lov->lov_tgts[loi->loi_ost_idx] || 
+                if (!lov->lov_tgts[loi->loi_ost_idx] ||
                     !lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
                         continue;
@@ -1127,7 +1139,7 @@ int lov_update_setattr_set(struct lov_request_set *set,
         lov_update_set(set, req, rc);
 
         /* grace error on inactive ost */
-        if (rc && !(lov->lov_tgts[req->rq_idx] && 
+        if (rc && !(lov->lov_tgts[req->rq_idx] &&
                     lov->lov_tgts[req->rq_idx]->ltd_active))
                 rc = 0;
 
@@ -1147,9 +1159,10 @@ int lov_update_setattr_set(struct lov_request_set *set,
 }
 
 /* The callback for osc_setattr_async that finilizes a request info when a
- * response is recieved. */
-static int cb_setattr_update(struct obd_info *oinfo, int rc)
+ * response is received. */
+static int cb_setattr_update(void *cookie, int rc)
 {
+        struct obd_info *oinfo = cookie;
         struct lov_request *lovreq;
         lovreq = container_of(oinfo, struct lov_request, rq_oi);
         return lov_update_setattr_set(lovreq->rq_rqset, lovreq, rc);
@@ -1199,12 +1212,14 @@ int lov_prep_setattr_set(struct obd_export *exp, struct obd_info *oinfo,
                 memcpy(req->rq_oi.oi_oa, oinfo->oi_oa,
                        sizeof(*req->rq_oi.oi_oa));
                 req->rq_oi.oi_oa->o_id = loi->loi_id;
-                LASSERT(!(req->rq_oi.oi_oa->o_valid & OBD_MD_FLGROUP) 
-                                || req->rq_oi.oi_oa->o_gr>0);
+                LASSERTF(!(req->rq_oi.oi_oa->o_valid & OBD_MD_FLGROUP) ||
+                         CHECK_MDS_GROUP(req->rq_oi.oi_oa->o_gr),
+                         "req->rq_oi.oi_oa->o_valid="LPX64" "
+                         "req->rq_oi.oi_oa->o_gr="LPU64"\n",
+                         req->rq_oi.oi_oa->o_valid, req->rq_oi.oi_oa->o_gr);
                 req->rq_oi.oi_oa->o_stripe_idx = i;
                 req->rq_oi.oi_cb_up = cb_setattr_update;
                 req->rq_oi.oi_capa = oinfo->oi_capa;
-                req->rq_rqset = set;
 
                 if (oinfo->oi_oa->o_valid & OBD_MD_FLSIZE) {
                         int off = lov_stripe_offset(oinfo->oi_md,
@@ -1279,9 +1294,10 @@ int lov_update_punch_set(struct lov_request_set *set,
 }
 
 /* The callback for osc_punch that finilizes a request info when a response
- * is recieved. */
-static int cb_update_punch(struct obd_info *oinfo, int rc)
+ * is received. */
+static int cb_update_punch(void *cookie, int rc)
 {
+        struct obd_info *oinfo = cookie;
         struct lov_request *lovreq;
         lovreq = container_of(oinfo, struct lov_request, rq_oi);
         return lov_update_punch_set(lovreq->rq_rqset, lovreq, rc);
@@ -1309,18 +1325,18 @@ int lov_prep_punch_set(struct obd_export *exp, struct obd_info *oinfo,
                 struct lov_request *req;
                 obd_off rs, re;
 
-                if (!lov->lov_tgts[loi->loi_ost_idx] ||
-                    !lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
-                        CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
-                        continue;
-                }
-
                 if (!lov_stripe_intersects(oinfo->oi_md, i,
                                            oinfo->oi_policy.l_extent.start,
                                            oinfo->oi_policy.l_extent.end,
                                            &rs, &re))
                         continue;
 
+                if (!lov->lov_tgts[loi->loi_ost_idx] ||
+                    !lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
+                        CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
+                        GOTO(out_set, rc = -EIO);
+                }
+
                 OBD_ALLOC(req, sizeof(*req));
                 if (req == NULL)
                         GOTO(out_set, rc = -ENOMEM);
@@ -1340,7 +1356,6 @@ int lov_prep_punch_set(struct obd_export *exp, struct obd_info *oinfo,
 
                 req->rq_oi.oi_oa->o_stripe_idx = i;
                 req->rq_oi.oi_cb_up = cb_update_punch;
-                req->rq_rqset = set;
 
                 req->rq_oi.oi_policy.l_extent.start = rs;
                 req->rq_oi.oi_policy.l_extent.end = re;
@@ -1466,7 +1481,7 @@ int lov_fini_statfs(struct obd_device *obd, struct obd_statfs *osfs,int success)
 
                 spin_lock(&obd->obd_osfs_lock);
                 memcpy(&obd->obd_osfs, osfs, sizeof(*osfs));
-                obd->obd_osfs_age = get_jiffies_64();
+                obd->obd_osfs_age = cfs_time_current_64();
                 spin_unlock(&obd->obd_osfs_lock);
                 RETURN(0);
         }
@@ -1493,15 +1508,11 @@ int lov_fini_statfs_set(struct lov_request_set *set)
         RETURN(rc);
 }
 
-void lov_update_statfs(struct obd_device *obd, struct obd_statfs *osfs,
-                       struct obd_statfs *lov_sfs, int success)
+void lov_update_statfs(struct obd_statfs *osfs, struct obd_statfs *lov_sfs,
+                       int success)
 {
         int shift = 0, quit = 0;
         __u64 tmp;
-        spin_lock(&obd->obd_osfs_lock);
-        memcpy(&obd->obd_osfs, lov_sfs, sizeof(*lov_sfs));
-        obd->obd_osfs_age = get_jiffies_64();
-        spin_unlock(&obd->obd_osfs_lock);
 
         if (success == 0) {
                 memcpy(osfs, lov_sfs, sizeof(*lov_sfs));
@@ -1566,9 +1577,10 @@ void lov_update_statfs(struct obd_device *obd, struct obd_statfs *osfs,
 }
 
 /* The callback for osc_statfs_async that finilizes a request info when a
- * response is recieved. */
-static int cb_statfs_update(struct obd_info *oinfo, int rc)
+ * response is received. */
+static int cb_statfs_update(void *cookie, int rc)
 {
+        struct obd_info *oinfo = cookie;
         struct lov_request *lovreq;
         struct obd_statfs *osfs, *lov_sfs;
         struct obd_device *obd;
@@ -1584,19 +1596,33 @@ static int cb_statfs_update(struct obd_info *oinfo, int rc)
         lov_sfs = oinfo->oi_osfs;
 
         success = lovreq->rq_rqset->set_success;
-
         /* XXX: the same is done in lov_update_common_set, however
            lovset->set_exp is not initialized. */
         lov_update_set(lovreq->rq_rqset, lovreq, rc);
         if (rc) {
+                /* XXX ignore error for disconnected ost ? */
                 if (rc && !(lov->lov_tgts[lovreq->rq_idx] &&
                             lov->lov_tgts[lovreq->rq_idx]->ltd_active))
                         rc = 0;
-                RETURN(rc);
+                GOTO(out, rc);
         }
 
-        lov_update_statfs(obd, osfs, lov_sfs, success);
+        spin_lock(&obd->obd_osfs_lock);
+        memcpy(&obd->obd_osfs, lov_sfs, sizeof(*lov_sfs));
+        if ((oinfo->oi_flags & OBD_STATFS_FROM_CACHE) == 0)
+                obd->obd_osfs_age = cfs_time_current_64();
+        spin_unlock(&obd->obd_osfs_lock);
+
+        lov_update_statfs(osfs, lov_sfs, success);
         qos_update(lov);
+out:
+        if (lovreq->rq_rqset->set_oi->oi_flags & OBD_STATFS_PTLRPCD &&
+            lov_finished_set(lovreq->rq_rqset)) {
+               lov_statfs_interpret(NULL, lovreq->rq_rqset,
+                                    lovreq->rq_rqset->set_success !=
+                                                  lovreq->rq_rqset->set_count);
+               qos_statfs_done(lov);
+        }
 
         RETURN(0);
 }
@@ -1621,7 +1647,8 @@ int lov_prep_statfs_set(struct obd_device *obd, struct obd_info *oinfo,
         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
                 struct lov_request *req;
 
-                if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active) {
+                if (!lov->lov_tgts[i] || (!lov->lov_tgts[i]->ltd_active
+                                          && (oinfo->oi_flags & OBD_STATFS_NODELAY))) {
                         CDEBUG(D_HA, "lov idx %d inactive\n", i);
                         continue;
                 }
@@ -1638,7 +1665,7 @@ int lov_prep_statfs_set(struct obd_device *obd, struct obd_info *oinfo,
 
                 req->rq_idx = i;
                 req->rq_oi.oi_cb_up = cb_statfs_update;
-                req->rq_rqset = set;
+                req->rq_oi.oi_flags = oinfo->oi_flags;
 
                 lov_set_add_req(req, set);
         }