Whamcloud - gitweb
LU-8726 osd-ldiskfs: bypass read for benchmarking
[fs/lustre-release.git] / lustre / lmv / lmv_obd.c
index 7acb440..f9ae025 100644 (file)
  *
  * 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.
+ * http://www.gnu.org/licenses/gpl-2.0.html
  *
  * GPL HEADER END
  */
@@ -27,7 +23,7 @@
  * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011, 2014, Intel Corporation.
+ * Copyright (c) 2011, 2016, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -62,6 +58,8 @@
 #include <lustre_kernelcomm.h>
 #include "lmv_internal.h"
 
+static int lmv_check_connect(struct obd_device *obd);
+
 static void lmv_activate_target(struct lmv_obd *lmv,
                                 struct lmv_tgt_desc *tgt,
                                 int activate)
@@ -101,7 +99,7 @@ static int lmv_set_mdc_active(struct lmv_obd *lmv,
                if (tgt == NULL || tgt->ltd_exp == NULL)
                        continue;
 
-               CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
+               CDEBUG(D_INFO, "Target idx %d is %s conn %#llx\n", i,
                       tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
 
                if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
@@ -183,14 +181,7 @@ static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
                 */
                obd->obd_self_export->exp_connect_data = *conn_data;
        }
-#if 0
-        else if (ev == OBD_NOTIFY_DISCON) {
-                /*
-                 * For disconnect event, flush fld cache for failout MDS case.
-                 */
-                fld_client_flush(&lmv->lmv_fld);
-        }
-#endif
+
         /*
          * Pass the notification up the chain.
          */
@@ -200,45 +191,28 @@ static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
         RETURN(rc);
 }
 
-/**
- * This is fake connect function. Its purpose is to initialize lmv and say
- * caller that everything is okay. Real connection will be performed later.
- */
 static int lmv_connect(const struct lu_env *env,
-                       struct obd_export **exp, struct obd_device *obd,
-                       struct obd_uuid *cluuid, struct obd_connect_data *data,
-                       void *localdata)
+                      struct obd_export **pexp, struct obd_device *obd,
+                      struct obd_uuid *cluuid, struct obd_connect_data *data,
+                      void *localdata)
 {
-        struct lmv_obd        *lmv = &obd->u.lmv;
-        struct lustre_handle  conn = { 0 };
-        int                    rc = 0;
-        ENTRY;
-
-        /*
-         * We don't want to actually do the underlying connections more than
-         * once, so keep track.
-         */
-        lmv->refcount++;
-        if (lmv->refcount > 1) {
-                *exp = NULL;
-                RETURN(0);
-        }
-
-        rc = class_connect(&conn, obd, cluuid);
-        if (rc) {
-                CERROR("class_connection() returned %d\n", rc);
-                RETURN(rc);
-        }
+       struct lmv_obd *lmv = &obd->u.lmv;
+       struct lustre_handle conn = { 0 };
+       struct obd_export *exp;
+       int rc;
+       ENTRY;
 
-        *exp = class_conn2export(&conn);
-        class_export_get(*exp);
+       rc = class_connect(&conn, obd, cluuid);
+       if (rc) {
+               CERROR("class_connection() returned %d\n", rc);
+               RETURN(rc);
+       }
 
-        lmv->exp = *exp;
-        lmv->connected = 0;
-        lmv->cluuid = *cluuid;
+       exp = class_conn2export(&conn);
 
-        if (data)
-                lmv->conn_data = *data;
+       lmv->connected = 0;
+       lmv->cluuid = *cluuid;
+       lmv->conn_data = *data;
 
        if (lmv->targets_proc_entry == NULL) {
                lmv->targets_proc_entry = lprocfs_register("target_obds",
@@ -253,18 +227,21 @@ static int lmv_connect(const struct lu_env *env,
                }
        }
 
-        /*
-         * All real clients should perform actual connection right away, because
-         * it is possible, that LMV will not have opportunity to connect targets
-         * and MDC stuff will be called directly, for instance while reading
-         * ../mdc/../kbytesfree procfs file, etc.
-         */
-       if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL))
-                rc = lmv_check_connect(obd);
+       rc = lmv_check_connect(obd);
+       if (rc != 0)
+               GOTO(out_proc, rc);
+
+       *pexp = exp;
 
-       if (rc && lmv->targets_proc_entry != NULL)
-               lprocfs_remove(&lmv->targets_proc_entry);
        RETURN(rc);
+
+out_proc:
+       if (lmv->targets_proc_entry != NULL)
+               lprocfs_remove(&lmv->targets_proc_entry);
+
+       class_disconnect(exp);
+
+       return rc;
 }
 
 static int lmv_init_ea_size(struct obd_export *exp, __u32 easize,
@@ -521,7 +498,7 @@ static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
        RETURN(rc);
 }
 
-int lmv_check_connect(struct obd_device *obd)
+static int lmv_check_connect(struct obd_device *obd)
 {
        struct lmv_obd          *lmv = &obd->u.lmv;
        struct lmv_tgt_desc     *tgt;
@@ -566,7 +543,6 @@ int lmv_check_connect(struct obd_device *obd)
                        GOTO(out_disc, rc);
        }
 
-       class_export_put(lmv->exp);
        lmv->connected = 1;
        easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
        lmv_init_ea_size(obd->obd_self_export, easize, 0);
@@ -590,9 +566,10 @@ int lmv_check_connect(struct obd_device *obd)
                         }
                 }
         }
-        class_disconnect(lmv->exp);
+
        mutex_unlock(&lmv->lmv_init_mutex);
-        RETURN(rc);
+
+       RETURN(rc);
 }
 
 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
@@ -650,13 +627,6 @@ static int lmv_disconnect(struct obd_export *exp)
         if (!lmv->tgts)
                 goto out_local;
 
-        /*
-         * Only disconnect the underlying layers on the final disconnect.
-         */
-        lmv->refcount--;
-        if (lmv->refcount != 0)
-                goto out_local;
-
         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
                if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
                         continue;
@@ -678,26 +648,31 @@ out_local:
         if (!lmv->connected)
                 class_export_put(exp);
         rc = class_disconnect(exp);
-        if (lmv->refcount == 0)
-                lmv->connected = 0;
-        RETURN(rc);
+       lmv->connected = 0;
+
+       RETURN(rc);
 }
 
-static int lmv_fid2path(struct obd_export *exp, int len, void *karg, void *uarg)
+static int lmv_fid2path(struct obd_export *exp, int len, void *karg,
+                       void __user *uarg)
 {
        struct obd_device       *obddev = class_exp2obd(exp);
        struct lmv_obd          *lmv = &obddev->u.lmv;
        struct getinfo_fid2path *gf;
        struct lmv_tgt_desc     *tgt;
        struct getinfo_fid2path *remote_gf = NULL;
+       struct lu_fid           root_fid;
        int                     remote_gf_size = 0;
        int                     rc;
 
-       gf = (struct getinfo_fid2path *)karg;
+       gf = karg;
        tgt = lmv_find_target(lmv, &gf->gf_fid);
        if (IS_ERR(tgt))
                RETURN(PTR_ERR(tgt));
 
+       root_fid = *gf->gf_u.gf_root_fid;
+       LASSERT(fid_is_sane(&root_fid));
+
 repeat_fid2path:
        rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
        if (rc != 0 && rc != -EREMOTE)
@@ -710,23 +685,24 @@ repeat_fid2path:
                char *ptr;
 
                ori_gf = (struct getinfo_fid2path *)karg;
-               if (strlen(ori_gf->gf_path) +
-                   strlen(gf->gf_path) > ori_gf->gf_pathlen)
+               if (strlen(ori_gf->gf_u.gf_path) +
+                   strlen(gf->gf_u.gf_path) > ori_gf->gf_pathlen)
                        GOTO(out_fid2path, rc = -EOVERFLOW);
 
-               ptr = ori_gf->gf_path;
+               ptr = ori_gf->gf_u.gf_path;
 
-               memmove(ptr + strlen(gf->gf_path) + 1, ptr,
-                       strlen(ori_gf->gf_path));
+               memmove(ptr + strlen(gf->gf_u.gf_path) + 1, ptr,
+                       strlen(ori_gf->gf_u.gf_path));
 
-               strncpy(ptr, gf->gf_path, strlen(gf->gf_path));
-               ptr += strlen(gf->gf_path);
+               strncpy(ptr, gf->gf_u.gf_path,
+                       strlen(gf->gf_u.gf_path));
+               ptr += strlen(gf->gf_u.gf_path);
                *ptr = '/';
        }
 
-       CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
+       CDEBUG(D_INFO, "%s: get path %s "DFID" rec: %llu ln: %u\n",
               tgt->ltd_exp->exp_obd->obd_name,
-              gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
+              gf->gf_u.gf_path, PFID(&gf->gf_fid), gf->gf_recno,
               gf->gf_linkno);
 
        if (rc == 0)
@@ -755,7 +731,8 @@ repeat_fid2path:
        remote_gf->gf_fid = gf->gf_fid;
        remote_gf->gf_recno = -1;
        remote_gf->gf_linkno = -1;
-       memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
+       memset(remote_gf->gf_u.gf_path, 0, remote_gf->gf_pathlen);
+       *remote_gf->gf_u.gf_root_fid = root_fid;
        gf = remote_gf;
        goto repeat_fid2path;
 
@@ -776,13 +753,15 @@ static int lmv_hsm_req_count(struct lmv_obd *lmv,
        /* count how many requests must be sent to the given target */
        for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
                curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
+               if (IS_ERR(curr_tgt))
+                       RETURN(PTR_ERR(curr_tgt));
                if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
                        nr++;
        }
        return nr;
 }
 
-static void lmv_hsm_req_build(struct lmv_obd *lmv,
+static int lmv_hsm_req_build(struct lmv_obd *lmv,
                              struct hsm_user_request *hur_in,
                              const struct lmv_tgt_desc *tgt_mds,
                              struct hsm_user_request *hur_out)
@@ -796,6 +775,8 @@ static void lmv_hsm_req_build(struct lmv_obd *lmv,
        for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
                curr_tgt = lmv_find_target(lmv,
                                           &hur_in->hur_user_item[i].hui_fid);
+               if (IS_ERR(curr_tgt))
+                       RETURN(PTR_ERR(curr_tgt));
                if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
                        hur_out->hur_user_item[nr_out] =
                                                hur_in->hur_user_item[i];
@@ -805,10 +786,13 @@ static void lmv_hsm_req_build(struct lmv_obd *lmv,
        hur_out->hur_request.hr_itemcount = nr_out;
        memcpy(hur_data(hur_out), hur_data(hur_in),
               hur_in->hur_request.hr_data_len);
+
+       RETURN(0);
 }
 
 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
-                                struct lustre_kernelcomm *lk, void *uarg)
+                                struct lustre_kernelcomm *lk,
+                                void __user *uarg)
 {
        __u32   i;
        int     rc;
@@ -835,7 +819,7 @@ static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
 }
 
 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
-                              struct lustre_kernelcomm *lk, void *uarg)
+                              struct lustre_kernelcomm *lk, __user void *uarg)
 {
        struct file             *filp;
        __u32                    i, j;
@@ -858,7 +842,7 @@ static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
                                /* permanent error */
                                CERROR("%s: iocontrol MDC %s on MDT"
                                       " idx %d cmd %x: err = %d\n",
-                                      class_exp2obd(lmv->exp)->obd_name,
+                                      lmv2obd_dev(lmv)->obd_name,
                                       tgt->ltd_uuid.uuid, i, cmd, err);
                                rc = err;
                                lk->lk_flags |= LK_FLG_STOP;
@@ -905,7 +889,7 @@ static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
 
 
 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
-                         int len, void *karg, void *uarg)
+                        int len, void *karg, void __user *uarg)
 {
        struct obd_device       *obddev = class_exp2obd(exp);
        struct lmv_obd          *lmv = &obddev->u.lmv;
@@ -1082,8 +1066,8 @@ static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
                } else {
                        /* split fid list to their respective MDS */
                        for (i = 0; i < count; i++) {
-                               unsigned int            nr, reqlen;
-                               int                     rc1;
+                               int nr, rc1;
+                               size_t reqlen;
                                struct hsm_user_request *req;
 
                                tgt = lmv->tgts[i];
@@ -1091,6 +1075,8 @@ static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
                                        continue;
 
                                nr = lmv_hsm_req_count(lmv, hur, tgt);
+                               if (nr < 0)
+                                       RETURN(nr);
                                if (nr == 0) /* nothing for this MDS */
                                        continue;
 
@@ -1101,11 +1087,12 @@ static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
                                OBD_ALLOC_LARGE(req, reqlen);
                                if (req == NULL)
                                        RETURN(-ENOMEM);
-
-                               lmv_hsm_req_build(lmv, hur, tgt, req);
-
+                               rc1 = lmv_hsm_req_build(lmv, hur, tgt, req);
+                               if (rc1 < 0)
+                                       GOTO(hsm_req_err, rc1);
                                rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
                                                    req, uarg);
+hsm_req_err:
                                if (rc1 != 0 && rc == 0)
                                        rc = rc1;
                                OBD_FREE_LARGE(req, reqlen);
@@ -1173,58 +1160,15 @@ static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
         RETURN(rc);
 }
 
-#if 0
-static int lmv_all_chars_policy(int count, const char *name,
-                                int len)
-{
-        unsigned int c = 0;
-
-        while (len > 0)
-                c += name[--len];
-        c = c % count;
-        return c;
-}
-
-static int lmv_nid_policy(struct lmv_obd *lmv)
-{
-        struct obd_import *imp;
-        __u32              id;
-
-        /*
-         * XXX: To get nid we assume that underlying obd device is mdc.
-         */
-        imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
-        id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
-        return id % lmv->desc.ld_tgt_count;
-}
-
-static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
-                          placement_policy_t placement)
-{
-        switch (placement) {
-        case PLACEMENT_CHAR_POLICY:
-                return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
-                                            op_data->op_name,
-                                            op_data->op_namelen);
-        case PLACEMENT_NID_POLICY:
-                return lmv_nid_policy(lmv);
-
-        default:
-                break;
-        }
-
-        CERROR("Unsupported placement policy %x\n", placement);
-        return -EINVAL;
-}
-#endif
-
 /**
  * This is _inode_ placement policy function (not name).
  */
 static int lmv_placement_policy(struct obd_device *obd,
                                struct md_op_data *op_data, u32 *mds)
 {
-       struct lmv_obd          *lmv = &obd->u.lmv;
+       struct lmv_obd     *lmv = &obd->u.lmv;
+       struct lmv_user_md *lum;
+
        ENTRY;
 
        LASSERT(mds != NULL);
@@ -1234,31 +1178,22 @@ static int lmv_placement_policy(struct obd_device *obd,
                RETURN(0);
        }
 
-       if (op_data->op_default_stripe_offset != -1) {
+       lum = op_data->op_data;
+       /* Choose MDS by
+        * 1. See if the stripe offset is specified by lum.
+        * 2. Then check if there is default stripe offset.
+        * 3. Finally choose MDS by name hash if the parent
+        *    is striped directory. (see lmv_locate_mds()). */
+       if (op_data->op_cli_flags & CLI_SET_MEA && lum != NULL &&
+           le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
+               *mds = le32_to_cpu(lum->lum_stripe_offset);
+       } else if (op_data->op_default_stripe_offset != (__u32)-1) {
                *mds = op_data->op_default_stripe_offset;
-               RETURN(0);
-       }
-
-       /**
-        * If stripe_offset is provided during setdirstripe
-        * (setdirstripe -i xx), xx MDS will be choosen.
-        */
-       if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
-               struct lmv_user_md *lum;
-
-               lum = op_data->op_data;
-
-               if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
-                       *mds = le32_to_cpu(lum->lum_stripe_offset);
-               } else {
-                       /* -1 means default, which will be in the same MDT with
-                        * the stripe */
-                       *mds = op_data->op_mds;
-                       lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
-               }
+               op_data->op_mds = *mds;
+               /* Correct the stripe offset in lum */
+               if (lum != NULL)
+                       lum->lum_stripe_offset = cpu_to_le32(*mds);
        } else {
-               /* Allocate new fid on target according to operation type and
-                * parent home mds. */
                *mds = op_data->op_mds;
        }
 
@@ -1356,7 +1291,6 @@ static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
        lmv->desc.ld_active_tgt_count = 0;
        lmv->max_def_easize = 0;
        lmv->max_easize = 0;
-       lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
 
        spin_lock_init(&lmv->lmv_lock);
        mutex_init(&lmv->lmv_init_mutex);
@@ -1445,10 +1379,6 @@ static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
        __u32                    i;
        ENTRY;
 
-        rc = lmv_check_connect(obd);
-        if (rc)
-                RETURN(rc);
-
         OBD_ALLOC(temp, sizeof(*temp));
         if (temp == NULL)
                 RETURN(-ENOMEM);
@@ -1490,18 +1420,15 @@ out_free_temp:
         return rc;
 }
 
-static int lmv_getstatus(struct obd_export *exp, struct lu_fid *fid)
+static int lmv_get_root(struct obd_export *exp, const char *fileset,
+                       struct lu_fid *fid)
 {
         struct obd_device    *obd = exp->exp_obd;
         struct lmv_obd       *lmv = &obd->u.lmv;
         int                   rc;
         ENTRY;
 
-        rc = lmv_check_connect(obd);
-        if (rc)
-                RETURN(rc);
-
-       rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid);
+       rc = md_get_root(lmv->tgts[0]->ltd_exp, fileset, fid);
        RETURN(rc);
 }
 
@@ -1516,10 +1443,6 @@ static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
         int                     rc;
         ENTRY;
 
-        rc = lmv_check_connect(obd);
-        if (rc)
-                RETURN(rc);
-
         tgt = lmv_find_target(lmv, fid);
         if (IS_ERR(tgt))
                 RETURN(PTR_ERR(tgt));
@@ -1542,10 +1465,6 @@ static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
         int                     rc;
         ENTRY;
 
-        rc = lmv_check_connect(obd);
-        if (rc)
-                RETURN(rc);
-
         tgt = lmv_find_target(lmv, fid);
         if (IS_ERR(tgt))
                 RETURN(PTR_ERR(tgt));
@@ -1566,10 +1485,6 @@ static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
         int                      rc;
         ENTRY;
 
-        rc = lmv_check_connect(obd);
-        if (rc)
-                RETURN(rc);
-
         tgt = lmv_find_target(lmv, &op_data->op_fid1);
         if (IS_ERR(tgt))
                 RETURN(PTR_ERR(tgt));
@@ -1589,13 +1504,8 @@ static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
         struct obd_device   *obd = exp->exp_obd;
         struct lmv_obd      *lmv = &obd->u.lmv;
        __u32                i;
-        int                  rc;
         ENTRY;
 
-        rc = lmv_check_connect(obd);
-        if (rc)
-                RETURN(rc);
-
         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
 
        /*
@@ -1612,50 +1522,6 @@ static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
        RETURN(0);
 }
 
-static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
-                           ldlm_iterator_t it, void *data)
-{
-       struct obd_device       *obd = exp->exp_obd;
-       struct lmv_obd          *lmv = &obd->u.lmv;
-       int                     i;
-       int                     tgt;
-       int                     rc;
-       ENTRY;
-
-       rc = lmv_check_connect(obd);
-       if (rc)
-               RETURN(rc);
-
-       CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
-
-       /*
-        * With DNE every object can have two locks in different namespaces:
-        * lookup lock in space of MDT storing direntry and update/open lock in
-        * space of MDT storing inode.  Try the MDT that the FID maps to first,
-        * since this can be easily found, and only try others if that fails.
-        */
-       for (i = 0, tgt = lmv_find_target_index(lmv, fid);
-            i < lmv->desc.ld_tgt_count;
-            i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
-               if (tgt < 0) {
-                       CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
-                              obd->obd_name, PFID(fid), tgt);
-                       tgt = 0;
-               }
-
-               if (lmv->tgts[tgt] == NULL ||
-                   lmv->tgts[tgt]->ltd_exp == NULL)
-                       continue;
-
-               rc = md_find_cbdata(lmv->tgts[tgt]->ltd_exp, fid, it, data);
-               if (rc)
-                       RETURN(rc);
-       }
-
-       RETURN(rc);
-}
-
-
 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
                      struct md_open_data *mod, struct ptlrpc_request **request)
 {
@@ -1665,10 +1531,6 @@ static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
         int                    rc;
         ENTRY;
 
-        rc = lmv_check_connect(obd);
-        if (rc)
-                RETURN(rc);
-
         tgt = lmv_find_target(lmv, &op_data->op_fid1);
         if (IS_ERR(tgt))
                 RETURN(PTR_ERR(tgt));
@@ -1742,26 +1604,27 @@ lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
         * index if the file under striped dir is being restored, see
         * ct_restore(). */
        if (op_data->op_bias & MDS_CREATE_VOLATILE &&
-           (int)op_data->op_mds != -1 && lsm != NULL) {
+           (int)op_data->op_mds != -1) {
                int i;
                tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
                if (IS_ERR(tgt))
                        return tgt;
 
-               /* refill the right parent fid */
-               for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
-                       struct lmv_oinfo *oinfo;
+               if (lsm != NULL) {
+                       /* refill the right parent fid */
+                       for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
+                               struct lmv_oinfo *oinfo;
 
-                       oinfo = &lsm->lsm_md_oinfo[i];
-                       if (oinfo->lmo_mds == op_data->op_mds) {
-                               *fid = oinfo->lmo_fid;
-                               break;
+                               oinfo = &lsm->lsm_md_oinfo[i];
+                               if (oinfo->lmo_mds == op_data->op_mds) {
+                                       *fid = oinfo->lmo_fid;
+                                       break;
+                               }
                        }
-               }
 
-               /* Hmm, can not find the stripe by mdt_index(op_mds) */
-               if (i == lsm->lsm_md_stripe_count)
-                       tgt = ERR_PTR(-EINVAL);
+                       if (i == lsm->lsm_md_stripe_count)
+                               *fid = lsm->lsm_md_oinfo[0].lmo_fid;
+               }
 
                return tgt;
        }
@@ -1791,10 +1654,6 @@ int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
        int                      rc;
        ENTRY;
 
-       rc = lmv_check_connect(obd);
-       if (rc)
-               RETURN(rc);
-
        if (!lmv->desc.ld_active_tgt_count)
                RETURN(-EIO);
 
@@ -1837,8 +1696,7 @@ int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
 
 static int
 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
-           const union ldlm_policy_data *policy,
-           struct lookup_intent *it, struct md_op_data *op_data,
+           const union ldlm_policy_data *policy, struct md_op_data *op_data,
            struct lustre_handle *lockh, __u64 extra_lock_flags)
 {
        struct obd_device        *obd = exp->exp_obd;
@@ -1847,21 +1705,16 @@ lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
        int                       rc;
        ENTRY;
 
-       rc = lmv_check_connect(obd);
-       if (rc)
-               RETURN(rc);
-
-       CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
-              LL_IT2STR(it), PFID(&op_data->op_fid1));
+       CDEBUG(D_INODE, "ENQUEUE on "DFID"\n", PFID(&op_data->op_fid1));
 
        tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
        if (IS_ERR(tgt))
                RETURN(PTR_ERR(tgt));
 
-       CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%u\n",
-              LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
+       CDEBUG(D_INODE, "ENQUEUE on "DFID" -> mds #%u\n",
+              PFID(&op_data->op_fid1), tgt->ltd_idx);
 
-       rc = md_enqueue(tgt->ltd_exp, einfo, policy, it, op_data, lockh,
+       rc = md_enqueue(tgt->ltd_exp, einfo, policy, op_data, lockh,
                        extra_lock_flags);
 
        RETURN(rc);
@@ -1879,10 +1732,6 @@ lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
        int                      rc;
        ENTRY;
 
-       rc = lmv_check_connect(obd);
-       if (rc)
-               RETURN(rc);
-
        tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
        if (IS_ERR(tgt))
                RETURN(PTR_ERR(tgt));
@@ -1977,10 +1826,6 @@ static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
        int                      rc;
        ENTRY;
 
-       rc = lmv_check_connect(obd);
-       if (rc)
-               RETURN(rc);
-
        LASSERT(op_data->op_namelen != 0);
 
        CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
@@ -2042,10 +1887,6 @@ static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
               (int)newlen, new, PFID(&op_data->op_fid2),
               op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
 
-       rc = lmv_check_connect(obd);
-       if (rc)
-               RETURN(rc);
-
        op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
        op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
        op_data->op_cap = cfs_curproc_cap_pack();
@@ -2204,10 +2045,6 @@ static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
        int                      rc = 0;
        ENTRY;
 
-       rc = lmv_check_connect(obd);
-       if (rc)
-               RETURN(rc);
-
        CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
               PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
 
@@ -2230,10 +2067,6 @@ static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
        int                      rc;
        ENTRY;
 
-       rc = lmv_check_connect(obd);
-       if (rc != 0)
-               RETURN(rc);
-
        tgt = lmv_find_target(lmv, fid);
        if (IS_ERR(tgt))
                RETURN(PTR_ERR(tgt));
@@ -2339,7 +2172,7 @@ next:
                        stripe_hash = le64_to_cpu(dp->ldp_hash_end);
 
                        kunmap(page);
-                       page_cache_release(page);
+                       put_page(page);
                        page = NULL;
 
                        /* reach the end of current stripe, go to next stripe */
@@ -2354,12 +2187,12 @@ next:
                            le64_to_cpu(ent->lde_hash)) {
                                min_ent = ent;
                                kunmap(min_page);
-                               page_cache_release(min_page);
+                               put_page(min_page);
                                min_idx = i;
                                min_page = page;
                        } else {
                                kunmap(page);
-                               page_cache_release(page);
+                               put_page(page);
                                page = NULL;
                        }
                } else {
@@ -2372,7 +2205,7 @@ next:
 out:
        if (*ppage != NULL) {
                kunmap(*ppage);
-               page_cache_release(*ppage);
+               put_page(*ppage);
        }
        *stripe_offset = min_idx;
        *entp = min_ent;
@@ -2409,7 +2242,6 @@ static int lmv_read_striped_page(struct obd_export *exp,
                                 struct md_callback *cb_op,
                                 __u64 offset, struct page **ppage)
 {
-       struct obd_device       *obd = exp->exp_obd;
        struct lu_fid           master_fid = op_data->op_fid1;
        struct inode            *master_inode = op_data->op_data;
        __u64                   hash_offset = offset;
@@ -2425,10 +2257,6 @@ static int lmv_read_striped_page(struct obd_export *exp,
        int                     rc;
        ENTRY;
 
-       rc = lmv_check_connect(obd);
-       if (rc)
-               RETURN(rc);
-
        /* Allocate a page and read entries from all of stripes and fill
         * the page by hash order */
        ent_page = alloc_page(GFP_KERNEL);
@@ -2442,7 +2270,7 @@ static int lmv_read_striped_page(struct obd_export *exp,
        dp->ldp_flags |= LDF_COLLIDE;
 
        area = dp + 1;
-       left_bytes = PAGE_CACHE_SIZE - sizeof(*dp);
+       left_bytes = PAGE_SIZE - sizeof(*dp);
        ent = area;
        last_ent = ent;
        do {
@@ -2503,7 +2331,7 @@ static int lmv_read_striped_page(struct obd_export *exp,
 out:
        if (min_ent_page != NULL) {
                kunmap(min_ent_page);
-               page_cache_release(min_ent_page);
+               put_page(min_ent_page);
        }
 
        if (unlikely(rc != 0)) {
@@ -2540,10 +2368,6 @@ int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
        int                     rc;
        ENTRY;
 
-       rc = lmv_check_connect(obd);
-       if (rc != 0)
-               RETURN(rc);
-
        if (unlikely(lsm != NULL)) {
                rc = lmv_read_striped_page(exp, op_data, cb_op, offset, ppage);
                RETURN(rc);
@@ -2595,9 +2419,6 @@ static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
        struct lmv_stripe_md    *lsm = op_data->op_mea1;
        ENTRY;
 
-       rc = lmv_check_connect(obd);
-       if (rc)
-               RETURN(rc);
 retry_unlink:
        /* For striped dir, we need to locate the parent as well */
        if (lsm != NULL) {
@@ -2738,25 +2559,13 @@ try_next_stripe:
        goto retry_unlink;
 }
 
-static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
+static int lmv_precleanup(struct obd_device *obd)
 {
-        struct lmv_obd *lmv = &obd->u.lmv;
-        int rc = 0;
-
-        switch (stage) {
-        case OBD_CLEANUP_EARLY:
-                /* XXX: here should be calling obd_precleanup() down to
-                 * stack. */
-                break;
-        case OBD_CLEANUP_EXPORTS:
-                fld_client_proc_fini(&lmv->lmv_fld);
-                lprocfs_obd_cleanup(obd);
-               lprocfs_free_md_stats(obd);
-                break;
-        default:
-                break;
-        }
-        RETURN(rc);
+       ENTRY;
+       fld_client_proc_fini(&obd->u.lmv.lmv_fld);
+       lprocfs_obd_cleanup(obd);
+       lprocfs_free_md_stats(obd);
+       RETURN(0);
 }
 
 /**
@@ -2785,7 +2594,7 @@ static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
 
         obd = class_exp2obd(exp);
         if (obd == NULL) {
-                CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
+               CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
                        exp->exp_handle.h_cookie);
                 RETURN(-EINVAL);
         }
@@ -2794,10 +2603,6 @@ static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
                 int i;
 
-                rc = lmv_check_connect(obd);
-                if (rc)
-                        RETURN(rc);
-
                 LASSERT(*vallen == sizeof(__u32));
                for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
                        struct lmv_tgt_desc *tgt = lmv->tgts[i];
@@ -2815,10 +2620,6 @@ static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
        } else if (KEY_IS(KEY_MAX_EASIZE) ||
                   KEY_IS(KEY_DEFAULT_EASIZE) ||
                   KEY_IS(KEY_CONN_DATA)) {
-               rc = lmv_check_connect(obd);
-               if (rc)
-                       RETURN(rc);
-
                /*
                 * Forwarding this request to first MDS, it should know LOV
                 * desc.
@@ -2865,7 +2666,7 @@ int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
 
        obd = class_exp2obd(exp);
        if (obd == NULL) {
-               CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
+               CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
                       exp->exp_handle.h_cookie);
                RETURN(-EINVAL);
        }
@@ -3048,8 +2849,9 @@ static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
        RETURN(rc);
 }
 
-int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
-                      __u64 *bits)
+static int lmv_set_lock_data(struct obd_export *exp,
+                            const struct lustre_handle *lockh,
+                            void *data, __u64 *bits)
 {
        struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
        struct lmv_tgt_desc     *tgt = lmv->tgts[0];
@@ -3165,47 +2967,38 @@ int lmv_clear_open_replay_data(struct obd_export *exp,
         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
 }
 
-static int lmv_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
-                              u32 suppgid, struct ptlrpc_request **request)
-{
-        struct obd_device       *obd = exp->exp_obd;
-        struct lmv_obd          *lmv = &obd->u.lmv;
-        struct lmv_tgt_desc     *tgt;
-        int                      rc;
-        ENTRY;
-
-        rc = lmv_check_connect(obd);
-        if (rc)
-                RETURN(rc);
-
-        tgt = lmv_find_target(lmv, fid);
-        if (IS_ERR(tgt))
-                RETURN(PTR_ERR(tgt));
-
-       rc = md_get_remote_perm(tgt->ltd_exp, fid, suppgid, request);
-       RETURN(rc);
-}
-
 int lmv_intent_getattr_async(struct obd_export *exp,
-                             struct md_enqueue_info *minfo,
-                             struct ldlm_enqueue_info *einfo)
+                            struct md_enqueue_info *minfo)
 {
        struct md_op_data       *op_data = &minfo->mi_data;
        struct obd_device       *obd = exp->exp_obd;
        struct lmv_obd          *lmv = &obd->u.lmv;
-       struct lmv_tgt_desc     *tgt = NULL;
+       struct lmv_tgt_desc     *ptgt = NULL;
+       struct lmv_tgt_desc     *ctgt = NULL;
        int                      rc;
        ENTRY;
 
-       rc = lmv_check_connect(obd);
-       if (rc)
-               RETURN(rc);
+       if (!fid_is_sane(&op_data->op_fid2))
+               RETURN(-EINVAL);
 
-       tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
-       if (IS_ERR(tgt))
-               RETURN(PTR_ERR(tgt));
+       ptgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
+       if (IS_ERR(ptgt))
+               RETURN(PTR_ERR(ptgt));
 
-       rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
+       ctgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
+       if (IS_ERR(ctgt))
+               RETURN(PTR_ERR(ctgt));
+
+       /*
+        * if child is on remote MDT, we need 2 async RPCs to fetch both LOOKUP
+        * lock on parent, and UPDATE lock on child MDT, which makes all
+        * complicated. Considering remote dir is rare case, and not supporting
+        * it in statahead won't cause any issue, drop its support for now.
+        */
+       if (ptgt != ctgt)
+               RETURN(-ENOTSUPP);
+
+       rc = md_intent_getattr_async(ptgt->ltd_exp, minfo);
        RETURN(rc);
 }
 
@@ -3218,10 +3011,6 @@ int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
         int                      rc;
         ENTRY;
 
-        rc = lmv_check_connect(obd);
-        if (rc)
-                RETURN(rc);
-
         tgt = lmv_find_target(lmv, fid);
         if (IS_ERR(tgt))
                 RETURN(PTR_ERR(tgt));
@@ -3360,9 +3149,8 @@ struct obd_ops lmv_obd_ops = {
 };
 
 struct md_ops lmv_md_ops = {
-        .m_getstatus            = lmv_getstatus,
+       .m_get_root             = lmv_get_root,
         .m_null_inode          = lmv_null_inode,
-        .m_find_cbdata          = lmv_find_cbdata,
         .m_close                = lmv_close,
         .m_create               = lmv_create,
         .m_enqueue              = lmv_enqueue,
@@ -3386,26 +3174,26 @@ struct md_ops lmv_md_ops = {
        .m_merge_attr           = lmv_merge_attr,
         .m_set_open_replay_data = lmv_set_open_replay_data,
         .m_clear_open_replay_data = lmv_clear_open_replay_data,
-        .m_get_remote_perm      = lmv_get_remote_perm,
         .m_intent_getattr_async = lmv_intent_getattr_async,
        .m_revalidate_lock      = lmv_revalidate_lock,
        .m_get_fid_from_lsm     = lmv_get_fid_from_lsm,
        .m_unpackmd             = lmv_unpackmd,
 };
 
-int __init lmv_init(void)
+static int __init lmv_init(void)
 {
        return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
                                   LUSTRE_LMV_NAME, NULL);
 }
 
-static void lmv_exit(void)
+static void __exit lmv_exit(void)
 {
-        class_unregister_type(LUSTRE_LMV_NAME);
+       class_unregister_type(LUSTRE_LMV_NAME);
 }
 
-MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
-MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
+MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
+MODULE_DESCRIPTION("Lustre Logical Metadata Volume");
+MODULE_VERSION(LUSTRE_VERSION_STRING);
 MODULE_LICENSE("GPL");
 
 module_init(lmv_init);