Whamcloud - gitweb
LU-1347 style: removes obsolete EXPORT_SYMTAB macros
[fs/lustre-release.git] / lustre / mdc / mdc_request.c
index 74afc27..a0e392a 100644 (file)
@@ -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.
  * GPL HEADER END
  */
 /*
- * Copyright  2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
+ *
+ * Copyright (c) 2011, 2012, Whamcloud, Inc.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
  * Lustre is a trademark of Sun Microsystems, Inc.
  */
 
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
 #define DEBUG_SUBSYSTEM S_MDC
 
 #ifdef __KERNEL__
 
 #define REQUEST_MINOR 244
 
-static quota_interface_t *quota_interface;
-extern quota_interface_t mdc_quota_interface;
+struct mdc_renew_capa_args {
+        struct obd_capa        *ra_oc;
+        renew_capa_cb_t         ra_cb;
+};
 
 static int mdc_cleanup(struct obd_device *obd);
 
@@ -767,7 +766,14 @@ int mdc_clear_open_replay_data(struct obd_export *exp,
         struct md_open_data *mod = och->och_mod;
         ENTRY;
 
-        LASSERT(mod != LP_POISON && mod != NULL);
+        /**
+         * It is possible to not have \var mod in a case of eviction between
+         * lookup and ll_file_open().
+         **/
+        if (mod == NULL)
+                RETURN(0);
+
+        LASSERT(mod != LP_POISON);
 
         mod->mod_och = NULL;
         och->och_mod = NULL;
@@ -875,6 +881,7 @@ int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
                  * exists and return no error in that case
                  */
                 if (mod) {
+                        DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
                         LASSERT(mod->mod_open_req != NULL);
                         if (mod->mod_open_req->rq_committed)
                                 rc = 0;
@@ -1006,21 +1013,27 @@ out:
 EXPORT_SYMBOL(mdc_sendpage);
 #endif
 
-int mdc_readpage(struct obd_export *exp, const struct lu_fid *fid,
-                 struct obd_capa *oc, __u64 offset, struct page *page,
-                 struct ptlrpc_request **request)
+int mdc_readpage(struct obd_export *exp, struct md_op_data *op_data,
+                 struct page **pages, struct ptlrpc_request **request)
 {
         struct ptlrpc_request   *req;
         struct ptlrpc_bulk_desc *desc;
+        int                      i;
+        cfs_waitq_t              waitq;
+        int                      resends = 0;
+        struct l_wait_info       lwi;
         int                      rc;
         ENTRY;
 
         *request = NULL;
+        cfs_waitq_init(&waitq);
+
+restart_bulk:
         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
         if (req == NULL)
                 RETURN(-ENOMEM);
 
-        mdc_set_capa_size(req, &RMF_CAPA1, oc);
+        mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 
         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
         if (rc) {
@@ -1031,21 +1044,37 @@ int mdc_readpage(struct obd_export *exp, const struct lu_fid *fid,
         req->rq_request_portal = MDS_READPAGE_PORTAL;
         ptlrpc_at_set_req_timeout(req);
 
-        desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
+        desc = ptlrpc_prep_bulk_imp(req, op_data->op_npages, BULK_PUT_SINK,
+                                    MDS_BULK_PORTAL);
         if (desc == NULL) {
                 ptlrpc_request_free(req);
                 RETURN(-ENOMEM);
         }
 
         /* NB req now owns desc and will free it when it gets freed */
-        ptlrpc_prep_bulk_page(desc, page, 0, CFS_PAGE_SIZE);
-        mdc_readdir_pack(req, offset, CFS_PAGE_SIZE, fid, oc);
+        for (i = 0; i < op_data->op_npages; i++)
+                ptlrpc_prep_bulk_page(desc, pages[i], 0, CFS_PAGE_SIZE);
+
+        mdc_readdir_pack(req, op_data->op_offset,
+                         CFS_PAGE_SIZE * op_data->op_npages,
+                         &op_data->op_fid1, op_data->op_capa1);
 
         ptlrpc_request_set_replen(req);
         rc = ptlrpc_queue_wait(req);
         if (rc) {
                 ptlrpc_req_finished(req);
-                RETURN(rc);
+                if (rc != -ETIMEDOUT)
+                        RETURN(rc);
+
+                resends++;
+                if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
+                        CERROR("too many resend retries, returning error\n");
+                        RETURN(-EIO);
+                }
+                lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL, NULL);
+                l_wait_event(waitq, 0, &lwi);
+
+                goto restart_bulk;
         }
 
         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
@@ -1055,9 +1084,10 @@ int mdc_readpage(struct obd_export *exp, const struct lu_fid *fid,
                 RETURN(rc);
         }
 
-        if (req->rq_bulk->bd_nob_transferred != CFS_PAGE_SIZE) {
+        if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
                 CERROR("Unexpected # bytes transferred: %d (%ld expected)\n",
-                        req->rq_bulk->bd_nob_transferred, CFS_PAGE_SIZE);
+                        req->rq_bulk->bd_nob_transferred,
+                        CFS_PAGE_SIZE * op_data->op_npages);
                 ptlrpc_req_finished(req);
                 RETURN(-EPROTO);
         }
@@ -1066,9 +1096,11 @@ int mdc_readpage(struct obd_export *exp, const struct lu_fid *fid,
         RETURN(0);
 }
 
-static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
+static int mdc_statfs(const struct lu_env *env,
+                      struct obd_export *exp, struct obd_statfs *osfs,
                       __u64 max_age, __u32 flags)
 {
+        struct obd_device     *obd = class_exp2obd(exp);
         struct ptlrpc_request *req;
         struct obd_statfs     *msfs;
         struct obd_import     *imp = NULL;
@@ -1148,7 +1180,7 @@ static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
         /* Val is struct getinfo_fid2path result plus path */
         vallen = sizeof(*gf) + gf->gf_pathlen;
 
-        rc = obd_get_info(exp, keylen, key, &vallen, gf, NULL);
+        rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf, NULL);
         if (rc)
                 GOTO(out, rc);
 
@@ -1165,6 +1197,249 @@ out:
         return rc;
 }
 
+static struct kuc_hdr *changelog_kuc_hdr(char *buf, int len, int flags)
+{
+        struct kuc_hdr *lh = (struct kuc_hdr *)buf;
+
+        LASSERT(len <= CR_MAXSIZE);
+
+        lh->kuc_magic = KUC_MAGIC;
+        lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
+        lh->kuc_flags = flags;
+        lh->kuc_msgtype = CL_RECORD;
+        lh->kuc_msglen = len;
+        return lh;
+}
+
+#define D_CHANGELOG 0
+
+struct changelog_show {
+        __u64       cs_startrec;
+        __u32       cs_flags;
+        cfs_file_t *cs_fp;
+        char       *cs_buf;
+        struct obd_device *cs_obd;
+};
+
+static int changelog_show_cb(struct llog_handle *llh, struct llog_rec_hdr *hdr,
+                             void *data)
+{
+        struct changelog_show *cs = data;
+        struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
+        struct kuc_hdr *lh;
+        int len, rc;
+        ENTRY;
+
+        if ((rec->cr_hdr.lrh_type != CHANGELOG_REC) ||
+            (rec->cr.cr_type >= CL_LAST)) {
+                CERROR("Not a changelog rec %d/%d\n", rec->cr_hdr.lrh_type,
+                       rec->cr.cr_type);
+                RETURN(-EINVAL);
+        }
+
+        if (rec->cr.cr_index < cs->cs_startrec) {
+                /* Skip entries earlier than what we are interested in */
+                CDEBUG(D_CHANGELOG, "rec="LPU64" start="LPU64"\n",
+                       rec->cr.cr_index, cs->cs_startrec);
+                RETURN(0);
+        }
+
+        CDEBUG(D_CHANGELOG, LPU64" %02d%-5s "LPU64" 0x%x t="DFID" p="DFID
+               " %.*s\n", rec->cr.cr_index, rec->cr.cr_type,
+               changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
+               rec->cr.cr_flags & CLF_FLAGMASK,
+               PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
+               rec->cr.cr_namelen, rec->cr.cr_name);
+
+        len = sizeof(*lh) + sizeof(rec->cr) + rec->cr.cr_namelen;
+
+        /* Set up the message */
+        lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
+        memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
+
+        rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
+        CDEBUG(D_CHANGELOG, "kucmsg fp %p len %d rc %d\n", cs->cs_fp, len,rc);
+
+        RETURN(rc);
+}
+
+static int mdc_changelog_send_thread(void *csdata)
+{
+        struct changelog_show *cs = csdata;
+        struct llog_ctxt *ctxt = NULL;
+        struct llog_handle *llh = NULL;
+        struct kuc_hdr *kuch;
+        int rc;
+
+        CDEBUG(D_CHANGELOG, "changelog to fp=%p start "LPU64"\n",
+               cs->cs_fp, cs->cs_startrec);
+
+        /*
+         * It's important to daemonize here to close unused FDs.
+         * The write fd from pipe is already opened by the caller,
+         * so it's fine to clear all files here
+         */
+        cfs_daemonize("mdc_clg_send_thread");
+
+        OBD_ALLOC(cs->cs_buf, CR_MAXSIZE);
+        if (cs->cs_buf == NULL)
+                GOTO(out, rc = -ENOMEM);
+
+        /* Set up the remote catalog handle */
+        ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
+        if (ctxt == NULL)
+                GOTO(out, rc = -ENOENT);
+        rc = llog_create(ctxt, &llh, NULL, CHANGELOG_CATALOG);
+        if (rc) {
+                CERROR("llog_create() failed %d\n", rc);
+                GOTO(out, rc);
+        }
+        rc = llog_init_handle(llh, LLOG_F_IS_CAT, NULL);
+        if (rc) {
+                CERROR("llog_init_handle failed %d\n", rc);
+                GOTO(out, rc);
+        }
+
+        rc = llog_cat_process_flags(llh, changelog_show_cb, cs, 0, 0, 0);
+
+        /* Send EOF no matter what our result */
+        if ((kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch),
+                                      cs->cs_flags))) {
+                kuch->kuc_msgtype = CL_EOF;
+                libcfs_kkuc_msg_put(cs->cs_fp, kuch);
+        }
+
+out:
+        cfs_put_file(cs->cs_fp);
+        if (llh)
+                llog_cat_put(llh);
+        if (ctxt)
+                llog_ctxt_put(ctxt);
+        if (cs->cs_buf)
+                OBD_FREE(cs->cs_buf, CR_MAXSIZE);
+        OBD_FREE_PTR(cs);
+        /* detach from parent process so we get cleaned up */
+        cfs_daemonize("cl_send");
+        return rc;
+}
+
+static int mdc_ioc_changelog_send(struct obd_device *obd,
+                                  struct ioc_changelog *icc)
+{
+        struct changelog_show *cs;
+        int rc;
+
+        /* Freed in mdc_changelog_send_thread */
+        OBD_ALLOC_PTR(cs);
+        if (!cs)
+                return -ENOMEM;
+
+        cs->cs_obd = obd;
+        cs->cs_startrec = icc->icc_recno;
+        /* matching cfs_put_file in mdc_changelog_send_thread */
+        cs->cs_fp = cfs_get_fd(icc->icc_id);
+        cs->cs_flags = icc->icc_flags;
+
+        /* New thread because we should return to user app before
+           writing into our pipe */
+        rc = cfs_create_thread(mdc_changelog_send_thread, cs, CFS_DAEMON_FLAGS);
+        if (rc >= 0) {
+                CDEBUG(D_CHANGELOG, "start changelog thread: %d\n", rc);
+                return 0;
+        }
+
+        CERROR("Failed to start changelog thread: %d\n", rc);
+        OBD_FREE_PTR(cs);
+        return rc;
+}
+
+static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
+                                struct lustre_kernelcomm *lk);
+
+static int mdc_quotacheck(struct obd_device *unused, struct obd_export *exp,
+                          struct obd_quotactl *oqctl)
+{
+        struct client_obd       *cli = &exp->exp_obd->u.cli;
+        struct ptlrpc_request   *req;
+        struct obd_quotactl     *body;
+        int                      rc;
+        ENTRY;
+
+        req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
+                                        &RQF_MDS_QUOTACHECK, LUSTRE_MDS_VERSION,
+                                        MDS_QUOTACHECK);
+        if (req == NULL)
+                RETURN(-ENOMEM);
+
+        body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
+        *body = *oqctl;
+
+        ptlrpc_request_set_replen(req);
+
+        /* the next poll will find -ENODATA, that means quotacheck is
+         * going on */
+        cli->cl_qchk_stat = -ENODATA;
+        rc = ptlrpc_queue_wait(req);
+        if (rc)
+                cli->cl_qchk_stat = rc;
+        ptlrpc_req_finished(req);
+        RETURN(rc);
+}
+
+static int mdc_quota_poll_check(struct obd_export *exp,
+                                struct if_quotacheck *qchk)
+{
+        struct client_obd *cli = &exp->exp_obd->u.cli;
+        int rc;
+        ENTRY;
+
+        qchk->obd_uuid = cli->cl_target_uuid;
+        memcpy(qchk->obd_type, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME));
+
+        rc = cli->cl_qchk_stat;
+        /* the client is not the previous one */
+        if (rc == CL_NOT_QUOTACHECKED)
+                rc = -EINTR;
+        RETURN(rc);
+}
+
+static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
+                        struct obd_quotactl *oqctl)
+{
+        struct ptlrpc_request   *req;
+        struct obd_quotactl     *oqc;
+        int                      rc;
+        ENTRY;
+
+        req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
+                                        &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
+                                        MDS_QUOTACTL);
+        if (req == NULL)
+                RETURN(-ENOMEM);
+
+        oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
+        *oqc = *oqctl;
+
+        ptlrpc_request_set_replen(req);
+        ptlrpc_at_set_req_timeout(req);
+        req->rq_no_resend = 1;
+
+        rc = ptlrpc_queue_wait(req);
+        if (rc)
+                CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
+
+        if (req->rq_repmsg &&
+            (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
+                *oqctl = *oqc;
+        } else if (!rc) {
+                CERROR ("Can't unpack obd_quotactl\n");
+                rc = -EPROTO;
+        }
+        ptlrpc_req_finished(req);
+
+        RETURN(rc);
+}
+
 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
                          void *karg, void *uarg)
 {
@@ -1180,11 +1455,17 @@ static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
                 return -EINVAL;
         }
         switch (cmd) {
+        case LL_IOC_HSM_CT_START:
+                rc = mdc_ioc_hsm_ct_start(exp, karg);
+                GOTO(out, rc);
+        case OBD_IOC_CHANGELOG_SEND:
+                rc = mdc_ioc_changelog_send(obd, karg);
+                GOTO(out, rc);
         case OBD_IOC_CHANGELOG_CLEAR: {
-                struct ioc_changelog_clear *icc = karg;
+                struct ioc_changelog *icc = karg;
                 struct changelog_setinfo cs =
-                        {icc->icc_recno, icc->icc_id};
-                rc = obd_set_info_async(exp, strlen(KEY_CHANGELOG_CLEAR),
+                        {.cs_recno = icc->icc_recno, .cs_id = icc->icc_id};
+                rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
                                         KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
                                         NULL);
                 GOTO(out, rc);
@@ -1194,7 +1475,7 @@ static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
                 GOTO(out, rc);
         }
         case OBD_IOC_CLIENT_RECOVER:
-                rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
+                rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
                 if (rc < 0)
                         GOTO(out, rc);
                 GOTO(out, rc = 0);
@@ -1217,17 +1498,17 @@ static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
         }
 #endif
         case OBD_IOC_POLL_QUOTACHECK:
-                rc = lquota_poll_check(quota_interface, exp,
-                                       (struct if_quotacheck *)karg);
+                rc = mdc_quota_poll_check(exp, (struct if_quotacheck *)karg);
                 GOTO(out, rc);
         case OBD_IOC_PING_TARGET:
                 rc = ptlrpc_obd_ping(obd);
                 GOTO(out, rc);
         /*
-         * Normally IOC_OBD_STATFS iocontrol is handled by LMV instead of MDC.
-         * But when the cluster is upgraded from 1.8, there'd be no LMV layer
-         * thus we might be called here. Eventually this code should be removed.
-         * bz20731.
+         * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
+         * LMV instead of MDC. But when the cluster is upgraded from 1.8,
+         * there'd be no LMV layer thus we might be called here. Eventually
+         * this code should be removed.
+         * bz20731, LU-592.
          */
         case IOC_OBD_STATFS: {
                 struct obd_statfs stat_buf = {0};
@@ -1235,20 +1516,50 @@ static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
                 if (*((__u32 *) data->ioc_inlbuf2) != 0)
                         GOTO(out, rc = -ENODEV);
 
-                rc = mdc_statfs(obd, &stat_buf,
-                                cfs_time_current_64() - CFS_HZ, 0);
+                /* copy UUID */
+                if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
+                                     min((int) data->ioc_plen2,
+                                         (int) sizeof(struct obd_uuid))))
+                        GOTO(out, rc = -EFAULT);
+
+                rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
+                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
+                                0);
                 if (rc != 0)
                         GOTO(out, rc);
 
                 if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
-                                     data->ioc_plen1))
-                        GOTO(out, rc = -EFAULT);
-                if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
-                                     data->ioc_plen2))
+                                     min((int) data->ioc_plen1,
+                                         (int) sizeof(stat_buf))))
                         GOTO(out, rc = -EFAULT);
 
                 GOTO(out, rc = 0);
         }
+        case OBD_IOC_QUOTACTL: {
+                struct if_quotactl *qctl = karg;
+                struct obd_quotactl *oqctl;
+
+                OBD_ALLOC_PTR(oqctl);
+                if (!oqctl)
+                        RETURN(-ENOMEM);
+
+                QCTL_COPY(oqctl, qctl);
+                rc = obd_quotactl(exp, oqctl);
+                if (rc == 0) {
+                        QCTL_COPY(qctl, oqctl);
+                        qctl->qc_valid = QC_MDTIDX;
+                        qctl->obd_uuid = obd->u.cli.cl_target_uuid;
+                }
+                OBD_FREE_PTR(oqctl);
+                break;
+        }
+        case LL_IOC_GET_CONNECT_FLAGS: {
+                if (cfs_copy_to_user(uarg, &exp->exp_connect_flags,
+                                     sizeof(__u64)))
+                        GOTO(out, rc = -EFAULT);
+                else
+                        GOTO(out, rc = 0);
+        }
         default:
                 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
                 GOTO(out, rc = -ENOTTY);
@@ -1314,8 +1625,8 @@ static void lustre_swab_hai(struct hsm_action_item *h)
         __swab32s(&h->hai_action);
         lustre_swab_lu_fid(&h->hai_fid);
         __swab64s(&h->hai_cookie);
-        __swab64s(&h->hai_extent_start);
-        __swab64s(&h->hai_extent_end);
+        __swab64s(&h->hai_extent.offset);
+        __swab64s(&h->hai_extent.length);
         __swab64s(&h->hai_gid);
 }
 
@@ -1334,14 +1645,52 @@ static void lustre_swab_hal(struct hsm_action_list *h)
         }
 }
 
+static void lustre_swab_kuch(struct kuc_hdr *l)
+{
+        __swab16s(&l->kuc_magic);
+        /* __u8 l->kuc_transport */
+        __swab16s(&l->kuc_msgtype);
+        __swab16s(&l->kuc_msglen);
+}
+
+static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
+                                struct lustre_kernelcomm *lk)
+{
+        int rc = 0;
+
+        if (lk->lk_group != KUC_GRP_HSM) {
+                CERROR("Bad copytool group %d\n", lk->lk_group);
+                return -EINVAL;
+        }
+
+        CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
+               lk->lk_uid, lk->lk_group, lk->lk_flags);
+
+        if (lk->lk_flags & LK_FLG_STOP)
+                rc = libcfs_kkuc_group_rem(lk->lk_uid,lk->lk_group);
+        else {
+                cfs_file_t *fp = cfs_get_fd(lk->lk_wfd);
+                rc = libcfs_kkuc_group_add(fp, lk->lk_uid,lk->lk_group,
+                                           lk->lk_data);
+                if (rc && fp)
+                        cfs_put_file(fp);
+        }
+
+        /* lk_data is archive number mask */
+        /* TODO: register archive num with mdt so coordinator can choose
+           correct agent. */
+
+        return rc;
+}
+
 /**
- * Send a message to any listening copytools, nonblocking
- * @param val LNL message (lnl_hdr + hsm_action_list)
+ * Send a message to any listening copytools
+ * @param val KUC message (kuc_hdr + hsm_action_list)
  * @param len total length of message
  */
 static int mdc_hsm_copytool_send(int len, void *val)
 {
-        struct lnl_hdr *lh = (struct lnl_hdr *)val;
+        struct kuc_hdr *lh = (struct kuc_hdr *)val;
         struct hsm_action_list *hal = (struct hsm_action_list *)(lh + 1);
         int rc;
         ENTRY;
@@ -1351,25 +1700,26 @@ static int mdc_hsm_copytool_send(int len, void *val)
                       (int) (sizeof(*lh) + sizeof(*hal)));
                 RETURN(-EPROTO);
         }
-        if (lh->lnl_magic == __swab16(LNL_MAGIC)) {
-                lustre_swab_lnlh(lh);
+        if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
+                lustre_swab_kuch(lh);
                 lustre_swab_hal(hal);
-        } else if (lh->lnl_magic != LNL_MAGIC) {
-                CERROR("Bad magic %x!=%x\n", lh->lnl_magic, LNL_MAGIC);
+        } else if (lh->kuc_magic != KUC_MAGIC) {
+                CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
                 RETURN(-EPROTO);
         }
 
-        CDEBUG(D_IOCTL, " Received message mg=%x t=%d m=%d l=%d actions=%d\n",
-               lh->lnl_magic, lh->lnl_transport, lh->lnl_msgtype,
-               lh->lnl_msglen, hal->hal_count);
+        CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d\n",
+               lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
+               lh->kuc_msglen, hal->hal_count);
 
         /* Broadcast to HSM listeners */
-        rc = libcfs_klnl_msg_put(0, LNL_GRP_HSM, lh);
+        rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
 
         RETURN(rc);
 }
 
-int mdc_set_info_async(struct obd_export *exp,
+int mdc_set_info_async(const struct lu_env *env,
+                       struct obd_export *exp,
                        obd_count keylen, void *key,
                        obd_count vallen, void *val,
                        struct ptlrpc_request_set *set)
@@ -1378,29 +1728,6 @@ int mdc_set_info_async(struct obd_export *exp,
         int                rc = -EINVAL;
         ENTRY;
 
-        if (KEY_IS(KEY_INIT_RECOV)) {
-                if (vallen != sizeof(int))
-                        RETURN(-EINVAL);
-                cfs_spin_lock(&imp->imp_lock);
-                imp->imp_initial_recov = *(int *)val;
-                cfs_spin_unlock(&imp->imp_lock);
-                CDEBUG(D_HA, "%s: set imp_initial_recov = %d\n",
-                       exp->exp_obd->obd_name, imp->imp_initial_recov);
-                RETURN(0);
-        }
-        /* Turn off initial_recov after we try all backup servers once */
-        if (KEY_IS(KEY_INIT_RECOV_BACKUP)) {
-                if (vallen != sizeof(int))
-                        RETURN(-EINVAL);
-                cfs_spin_lock(&imp->imp_lock);
-                imp->imp_initial_recov_bk = *(int *)val;
-                if (imp->imp_initial_recov_bk)
-                        imp->imp_initial_recov = 1;
-                cfs_spin_unlock(&imp->imp_lock);
-                CDEBUG(D_HA, "%s: set imp_initial_recov_bk = %d\n",
-                       exp->exp_obd->obd_name, imp->imp_initial_recov_bk);
-                RETURN(0);
-        }
         if (KEY_IS(KEY_READ_ONLY)) {
                 if (vallen != sizeof(int))
                         RETURN(-EINVAL);
@@ -1415,8 +1742,8 @@ int mdc_set_info_async(struct obd_export *exp,
                 }
                 cfs_spin_unlock(&imp->imp_lock);
 
-                rc = target_set_info_rpc(imp, MDS_SET_INFO,
-                                         keylen, key, vallen, val, set);
+                rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
+                                       keylen, key, vallen, val, set);
                 RETURN(rc);
         }
         if (KEY_IS(KEY_SPTLRPC_CONF)) {
@@ -1437,8 +1764,8 @@ int mdc_set_info_async(struct obd_export *exp,
                 RETURN(0);
         }
         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
-                rc = target_set_info_rpc(imp, MDS_SET_INFO,
-                                         keylen, key, vallen, val, set);
+                rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
+                                       keylen, key, vallen, val, set);
                 RETURN(rc);
         }
         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
@@ -1449,8 +1776,9 @@ int mdc_set_info_async(struct obd_export *exp,
         RETURN(rc);
 }
 
-int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
-                 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
+int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
+                 __u32 keylen, void *key, __u32 *vallen, void *val,
+                 struct lov_stripe_md *lsm)
 {
         int rc = -EINVAL;
 
@@ -1465,8 +1793,7 @@ int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
                 max_easize = val;
                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
                 RETURN(0);
-        }
-        if (KEY_IS(KEY_CONN_DATA)) {
+        } else if (KEY_IS(KEY_CONN_DATA)) {
                 struct obd_import *imp = class_exp2cliimp(exp);
                 struct obd_connect_data *data = val;
 
@@ -1475,6 +1802,9 @@ int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
 
                 *data = imp->imp_connect_data;
                 RETURN(0);
+        } else if (KEY_IS(KEY_TGT_COUNT)) {
+                *((int *)val) = 1;
+                RETURN(0);
         }
 
         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
@@ -1621,12 +1951,9 @@ static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
                 /*
                  * Flush current sequence to make client obtain new one
                  * from server in case of disconnect/reconnect.
-                 * If range is already empty then no need to flush it.
                  */
-                if (cli->cl_seq != NULL &&
-                    !range_is_exhausted(&cli->cl_seq->lcs_space)) {
+                if (cli->cl_seq != NULL)
                         seq_client_flush(cli->cl_seq);
-                }
 
                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
                 break;
@@ -1645,7 +1972,9 @@ static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
         case IMP_EVENT_OCD:
                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
                 break;
-
+        case IMP_EVENT_DEACTIVATE:
+        case IMP_EVENT_ACTIVATE:
+                break;
         default:
                 CERROR("Unknown import event %x\n", event);
                 LBUG();
@@ -1706,7 +2035,7 @@ int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
         struct client_obd *cli = &exp->exp_obd->u.cli;
         struct lu_client_seq *seq = cli->cl_seq;
         ENTRY;
-        RETURN(seq_client_alloc_fid(seq, fid));
+        RETURN(seq_client_alloc_fid(NULL, seq, fid));
 }
 
 /* XXX This method is used only to clear current fid seq
@@ -1724,6 +2053,25 @@ struct obd_uuid *mdc_get_uuid(struct obd_export *exp) {
         return &cli->cl_target_uuid;
 }
 
+/**
+ * Determine whether the lock can be canceled before replaying it during
+ * recovery, non zero value will be return if the lock can be canceled,
+ * or zero returned for not
+ */
+static int mdc_cancel_for_recovery(struct ldlm_lock *lock)
+{
+        if (lock->l_resource->lr_type != LDLM_IBITS)
+                RETURN(0);
+
+        /* FIXME: if we ever get into a situation where there are too many
+         * opened files with open locks on a single node, then we really
+         * should replay these open locks to reget it */
+        if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
+                RETURN(0);
+
+        RETURN(1);
+}
+
 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
 {
         struct client_obd *cli = &obd->u.cli;
@@ -1738,14 +2086,9 @@ static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
 
         ptlrpcd_addref();
 
-        OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
-        if (!cli->cl_setattr_lock)
-                GOTO(err_rpc_lock, rc = -ENOMEM);
-        mdc_init_rpc_lock(cli->cl_setattr_lock);
-
         OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
         if (!cli->cl_close_lock)
-                GOTO(err_setattr_lock, rc = -ENOMEM);
+                GOTO(err_rpc_lock, rc = -ENOMEM);
         mdc_init_rpc_lock(cli->cl_close_lock);
 
         rc = client_obd_setup(obd, cfg);
@@ -1756,22 +2099,18 @@ static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
         sptlrpc_lprocfs_cliobd_attach(obd);
         ptlrpc_lprocfs_register_obd(obd);
 
+        ns_register_cancel(obd->obd_namespace, mdc_cancel_for_recovery);
+
         rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
         if (rc) {
                 mdc_cleanup(obd);
                 CERROR("failed to setup llogging subsystems\n");
         }
 
-        /* ignore errors */
-        libcfs_klnl_start(LNL_TRANSPORT_HSM);
-        libcfs_klnl_start(LNL_TRANSPORT_CHANGELOG);
-
         RETURN(rc);
 
 err_close_lock:
         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
-err_setattr_lock:
-        OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
 err_rpc_lock:
         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
         ptlrpcd_decref();
@@ -1808,19 +2147,16 @@ static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
 
         switch (stage) {
         case OBD_CLEANUP_EARLY:
+                break;
         case OBD_CLEANUP_EXPORTS:
-                /* If we set up but never connected, the
-                   client import will not have been cleaned. */
-                if (obd->u.cli.cl_import) {
-                        struct obd_import *imp;
-                        cfs_down_write(&obd->u.cli.cl_sem);
-                        imp = obd->u.cli.cl_import;
-                        CERROR("client import never connected\n");
-                        ptlrpc_invalidate_import(imp);
-                        class_destroy_import(imp);
-                        cfs_up_write(&obd->u.cli.cl_sem);
-                        obd->u.cli.cl_import = NULL;
-                }
+                /* Failsafe, ok if racy */
+                if (obd->obd_type->typ_refcnt <= 1)
+                        libcfs_kkuc_group_rem(0, KUC_GRP_HSM);
+
+                obd_cleanup_client_import(obd);
+                ptlrpc_lprocfs_unregister_obd(obd);
+                lprocfs_obd_cleanup(obd);
+
                 rc = obd_llog_finish(obd, 0);
                 if (rc != 0)
                         CERROR("failed to cleanup llogging subsystems\n");
@@ -1833,15 +2169,9 @@ static int mdc_cleanup(struct obd_device *obd)
 {
         struct client_obd *cli = &obd->u.cli;
 
-        libcfs_klnl_stop(LNL_TRANSPORT_HSM, LNL_GRP_HSM);
-        libcfs_klnl_stop(LNL_TRANSPORT_CHANGELOG, 0);
-
         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
-        OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
 
-        ptlrpc_lprocfs_unregister_obd(obd);
-        lprocfs_obd_cleanup(obd);
         ptlrpcd_decref();
 
         return client_obd_cleanup(obd);
@@ -1953,11 +2283,10 @@ int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
 }
 
 static int mdc_interpret_renew_capa(const struct lu_env *env,
-                                    struct ptlrpc_request *req, void *unused,
+                                    struct ptlrpc_request *req, void *args,
                                     int status)
 {
-        struct obd_capa *oc = req->rq_async_args.pointer_arg[0];
-        renew_capa_cb_t cb = req->rq_async_args.pointer_arg[1];
+        struct mdc_renew_capa_args *ra = args;
         struct mdt_body *body = NULL;
         struct lustre_capa *capa;
         ENTRY;
@@ -1977,7 +2306,7 @@ static int mdc_interpret_renew_capa(const struct lu_env *env,
                 GOTO(out, capa = ERR_PTR(-EFAULT));
         EXIT;
 out:
-        cb(oc, capa);
+        ra->ra_cb(ra->ra_oc, capa);
         return 0;
 }
 
@@ -1985,6 +2314,7 @@ static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
                           renew_capa_cb_t cb)
 {
         struct ptlrpc_request *req;
+        struct mdc_renew_capa_args *ra;
         ENTRY;
 
         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
@@ -1998,10 +2328,12 @@ static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
         mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
         ptlrpc_request_set_replen(req);
 
-        req->rq_async_args.pointer_arg[0] = oc;
-        req->rq_async_args.pointer_arg[1] = cb;
+        CLASSERT(sizeof(*ra) <= sizeof(req->rq_async_args));
+        ra = ptlrpc_req_async_args(req);
+        ra->ra_oc = oc;
+        ra->ra_cb = cb;
         req->rq_interpret_reply = mdc_interpret_renew_capa;
-        ptlrpcd_add_req(req, PSCOPE_OTHER);
+        ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
         RETURN(0);
 }
 
@@ -2050,11 +2382,14 @@ struct obd_ops mdc_obd_ops = {
         .o_get_info         = mdc_get_info,
         .o_process_config   = mdc_process_config,
         .o_get_uuid         = mdc_get_uuid,
+        .o_quotactl         = mdc_quotactl,
+        .o_quotacheck       = mdc_quotacheck
 };
 
 struct md_ops mdc_md_ops = {
         .m_getstatus        = mdc_getstatus,
         .m_change_cbdata    = mdc_change_cbdata,
+        .m_find_cbdata      = mdc_find_cbdata,
         .m_close            = mdc_close,
         .m_create           = mdc_create,
         .m_done_writing     = mdc_done_writing,
@@ -2092,24 +2427,14 @@ int __init mdc_init(void)
         struct lprocfs_static_vars lvars = { 0 };
         lprocfs_mdc_init_vars(&lvars);
 
-        cfs_request_module("lquota");
-        quota_interface = PORTAL_SYMBOL_GET(mdc_quota_interface);
-        init_obd_quota_ops(quota_interface, &mdc_obd_ops);
-
         rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
                                  LUSTRE_MDC_NAME, NULL);
-        if (rc && quota_interface)
-                PORTAL_SYMBOL_PUT(mdc_quota_interface);
-
         RETURN(rc);
 }
 
 #ifdef __KERNEL__
 static void /*__exit*/ mdc_exit(void)
 {
-        if (quota_interface)
-                PORTAL_SYMBOL_PUT(mdc_quota_interface);
-
         class_unregister_type(LUSTRE_MDC_NAME);
 }