Whamcloud - gitweb
LU-10370 ofd: truncate does not update blocks count on client
[fs/lustre-release.git] / lustre / ofd / ofd_dev.c
index 4cf5e53..c2cda39 100644 (file)
@@ -23,7 +23,7 @@
  * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2012, 2016, Intel Corporation.
+ * Copyright (c) 2012, 2017, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -245,9 +245,14 @@ static void ofd_stack_fini(const struct lu_env *env, struct ofd_device *m,
        top->ld_ops->ldo_process_config(env, top, lcfg);
        OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
 
+       if (m->ofd_los != NULL) {
+               local_oid_storage_fini(env, m->ofd_los);
+               m->ofd_los = NULL;
+       }
+
        lu_site_purge(env, top->ld_site, ~0);
        if (!cfs_hash_is_empty(top->ld_site->ls_obj_hash)) {
-               LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
+               LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_OTHER, NULL);
                lu_site_print(env, top->ld_site, &msgdata, lu_cdebug_printer);
        }
 
@@ -257,6 +262,30 @@ static void ofd_stack_fini(const struct lu_env *env, struct ofd_device *m,
        EXIT;
 }
 
+static void ofd_stack_pre_fini(const struct lu_env *env, struct ofd_device *m,
+                              struct lu_device *top)
+{
+       struct lustre_cfg_bufs bufs;
+       struct lustre_cfg *lcfg;
+       ENTRY;
+
+       LASSERT(top);
+
+       lustre_cfg_bufs_reset(&bufs, ofd_name(m));
+       lustre_cfg_bufs_set_string(&bufs, 1, NULL);
+       OBD_ALLOC(lcfg, lustre_cfg_len(bufs.lcfg_bufcount, bufs.lcfg_buflen));
+       if (!lcfg) {
+               CERROR("%s: failed to trigger LCFG_PRE_CLEANUP\n", ofd_name(m));
+       } else {
+               lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, &bufs);
+               top->ld_ops->ldo_process_config(env, top, lcfg);
+               OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount,
+                                             lcfg->lcfg_buflens));
+       }
+
+       EXIT;
+}
+
 /* For interoperability, see mdt_interop_param[]. */
 static struct cfg_interop_param ofd_interop_param[] = {
        { "ost.quota_type",     NULL },
@@ -993,48 +1022,24 @@ int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
        return rc;
 }
 
-struct locked_region {
-       struct list_head        list;
-       struct lustre_handle    lh;
-};
 
-/**
- * Lock single extent and save lock handle in the list.
- *
- * This is supplemental function for lock_zero_regions(). It allocates
- * new locked_region structure and locks it with extent lock, then adds
- * it to the list of all such regions.
- *
- * \param[in] ns       LDLM namespace
- * \param[in] res_id   resource ID
- * \param[in] begin    start of region
- * \param[in] end      end of region
- * \param[in] locked   list head of regions list
- *
- * \retval             0 if successful locking
- * \retval             negative value on error
- */
-static int lock_region(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
-                      unsigned long long begin, unsigned long long end,
-                      struct list_head *locked)
+static int ofd_lock_unlock_region(struct ldlm_namespace *ns,
+                                 struct ldlm_res_id *res_id,
+                                 unsigned long long begin,
+                                 unsigned long long end)
 {
-       struct locked_region    *region = NULL;
        __u64                    flags = 0;
        int                      rc;
+       struct lustre_handle     lh = { 0 };
 
        LASSERT(begin <= end);
-       OBD_ALLOC_PTR(region);
-       if (region == NULL)
-               return -ENOMEM;
 
-       rc = tgt_extent_lock(ns, res_id, begin, end, &region->lh,
-                            LCK_PR, &flags);
+       rc = tgt_extent_lock(ns, res_id, begin, end, &lh, LCK_PR, &flags);
        if (rc != 0)
                return rc;
 
-       CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end,
-              &region->lh);
-       list_add(&region->list, locked);
+       CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end, &lh);
+       tgt_extent_unlock(&lh, LCK_PR);
 
        return 0;
 }
@@ -1059,8 +1064,7 @@ static int lock_region(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
  */
 static int lock_zero_regions(struct ldlm_namespace *ns,
                             struct ldlm_res_id *res_id,
-                            struct fiemap *fiemap,
-                            struct list_head *locked)
+                            struct fiemap *fiemap)
 {
        __u64 begin = fiemap->fm_start;
        unsigned int i;
@@ -1074,8 +1078,8 @@ static int lock_zero_regions(struct ldlm_namespace *ns,
                if (fiemap_start[i].fe_logical > begin) {
                        CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
                               begin, fiemap_start[i].fe_logical);
-                       rc = lock_region(ns, res_id, begin,
-                                        fiemap_start[i].fe_logical, locked);
+                       rc = ofd_lock_unlock_region(ns, res_id, begin,
+                                                   fiemap_start[i].fe_logical);
                        if (rc)
                                RETURN(rc);
                }
@@ -1086,34 +1090,13 @@ static int lock_zero_regions(struct ldlm_namespace *ns,
        if (begin < (fiemap->fm_start + fiemap->fm_length)) {
                CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
                       begin, fiemap->fm_start + fiemap->fm_length);
-               rc = lock_region(ns, res_id, begin,
-                                fiemap->fm_start + fiemap->fm_length, locked);
+               rc = ofd_lock_unlock_region(ns, res_id, begin,
+                               fiemap->fm_start + fiemap->fm_length);
        }
 
        RETURN(rc);
 }
 
-/**
- * Unlock all previously locked sparse areas for given resource.
- *
- * This function goes through list of locked regions, unlocking and freeing
- * them one-by-one.
- *
- * \param[in] ns       LDLM namespace
- * \param[in] locked   list head of regions list
- */
-static void
-unlock_zero_regions(struct ldlm_namespace *ns, struct list_head *locked)
-{
-       struct locked_region *entry, *temp;
-
-       list_for_each_entry_safe(entry, temp, locked, list) {
-               CDEBUG(D_OTHER, "ost unlock lh=%p\n", &entry->lh);
-               tgt_extent_unlock(&entry->lh, LCK_PR);
-               list_del(&entry->list);
-               OBD_FREE_PTR(entry);
-       }
-}
 
 /**
  * OFD request handler for OST_GET_INFO RPC.
@@ -1207,19 +1190,12 @@ static int ofd_get_info_hdl(struct tgt_session_info *tsi)
                 * flushed back from client, then call fiemap again. */
                if (fm_key->lfik_oa.o_valid & OBD_MD_FLFLAGS &&
                    fm_key->lfik_oa.o_flags & OBD_FL_SRVLOCK) {
-                       struct list_head locked;
-
-                       INIT_LIST_HEAD(&locked);
                        ost_fid_build_resid(fid, &fti->fti_resid);
                        rc = lock_zero_regions(ofd->ofd_namespace,
-                                              &fti->fti_resid, fiemap,
-                                              &locked);
-                       if (rc == 0 && !list_empty(&locked)) {
+                                              &fti->fti_resid, fiemap);
+                       if (rc == 0)
                                rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
                                                    fiemap);
-                               unlock_zero_regions(ofd->ofd_namespace,
-                                                   &locked);
-                       }
                }
        } else if (KEY_IS(KEY_LAST_FID)) {
                struct ofd_device       *ofd = ofd_exp(exp);
@@ -1332,6 +1308,16 @@ static int ofd_getattr_hdl(struct tgt_session_info *tsi)
                        repbody->oa.o_valid |= OBD_MD_FLDATAVERSION;
                        repbody->oa.o_data_version = curr_version;
                }
+
+               if (fo->ofo_ff.ff_layout_version > 0) {
+                       repbody->oa.o_valid |= OBD_MD_LAYOUT_VERSION;
+                       repbody->oa.o_layout_version =
+                            fo->ofo_ff.ff_layout_version + fo->ofo_ff.ff_range;
+
+                       CDEBUG(D_INODE, DFID": get layout version: %u\n",
+                              PFID(&tsi->tsi_fid),
+                              repbody->oa.o_layout_version);
+               }
        }
 
        ofd_object_put(tsi->tsi_env, fo);
@@ -1367,7 +1353,6 @@ static int ofd_setattr_hdl(struct tgt_session_info *tsi)
        struct ost_body         *repbody;
        struct ldlm_resource    *res;
        struct ofd_object       *fo;
-       struct filter_fid       *ff = NULL;
        int                      rc = 0;
 
        ENTRY;
@@ -1407,13 +1392,8 @@ static int ofd_setattr_hdl(struct tgt_session_info *tsi)
        la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid);
        fti->fti_attr.la_valid &= ~LA_TYPE;
 
-       if (body->oa.o_valid & OBD_MD_FLFID) {
-               ff = &fti->fti_mds_fid;
-               ofd_prepare_fidea(ff, &body->oa);
-       }
-
        /* setting objects attributes (including owner/group) */
-       rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, ff);
+       rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, &body->oa);
        if (rc != 0)
                GOTO(out_put, rc);
 
@@ -1922,10 +1902,12 @@ static int ofd_statfs_hdl(struct tgt_session_info *tsi)
 
        ENTRY;
 
+       OBD_FAIL_TIMEOUT(OBD_FAIL_OST_STATFS_DELAY, 10);
+
        osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS);
 
        rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs,
-                       cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
+                       ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS, 0);
        if (rc != 0)
                CERROR("%s: statfs failed: rc = %d\n",
                       tgt_name(tsi->tsi_tgt), rc);
@@ -2017,7 +1999,6 @@ static int ofd_punch_hdl(struct tgt_session_info *tsi)
        struct ldlm_namespace   *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
        struct ldlm_resource    *res;
        struct ofd_object       *fo;
-       struct filter_fid       *ff = NULL;
        __u64                    flags = 0;
        struct lustre_handle     lh = { 0, };
        int                      rc;
@@ -2078,13 +2059,8 @@ static int ofd_punch_hdl(struct tgt_session_info *tsi)
        info->fti_attr.la_size = start;
        info->fti_attr.la_valid |= LA_SIZE;
 
-       if (oa->o_valid & OBD_MD_FLFID) {
-               ff = &info->fti_mds_fid;
-               ofd_prepare_fidea(ff, oa);
-       }
-
        rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
-                             ff, (struct obdo *)oa);
+                             (struct obdo *)oa);
        if (rc)
                GOTO(out_put, rc);
 
@@ -2105,7 +2081,12 @@ out:
                res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
                                        LDLM_EXTENT, 0);
                if (!IS_ERR(res)) {
+                       struct ost_lvb *res_lvb;
+
                        ldlm_res_lvbo_update(res, NULL, 0);
+                       res_lvb = res->lr_lvb_data;
+                       repbody->oa.o_valid |= OBD_MD_FLBLOCKS;
+                       repbody->oa.o_blocks = res_lvb->lvb_blocks;
                        ldlm_resource_putref(res);
                }
        }
@@ -2430,10 +2411,14 @@ static void ofd_prolong_extent_locks(struct tgt_session_info *tsi,
                                LASSERT(lock->l_export == data->lpa_export);
                                ldlm_lock_prolong_one(lock, data);
                                LDLM_LOCK_PUT(lock);
-                               RETURN_EXIT;
+                               if (data->lpa_locks_cnt > 0)
+                                       RETURN_EXIT;
+                               /* The lock was destroyed probably lets try
+                                * resource tree. */
+                       } else {
+                               lock->l_last_used = ktime_get();
+                               LDLM_LOCK_PUT(lock);
                        }
-                       lock->l_last_used = ktime_get();
-                       LDLM_LOCK_PUT(lock);
                }
        }
 
@@ -2902,7 +2887,6 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
        struct ofd_thread_info *info = NULL;
        struct obd_device *obd;
        struct tg_grants_data *tgd = &m->ofd_lut.lut_tgd;
-       struct obd_statfs *osfs;
        struct lu_fid fid;
        struct nm_config_file *nodemap_config;
        struct obd_device_target *obt;
@@ -2928,24 +2912,11 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
 
        spin_lock_init(&m->ofd_flags_lock);
        m->ofd_raid_degraded = 0;
+       m->ofd_checksum_t10pi_enforce = 0;
        m->ofd_syncjournal = 0;
        ofd_slc_set(m);
-       tgd->tgd_grant_compat_disable = 0;
        m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT;
 
-       /* statfs data */
-       spin_lock_init(&tgd->tgd_osfs_lock);
-       tgd->tgd_osfs_age = cfs_time_shift_64(-1000);
-       tgd->tgd_osfs_unstable = 0;
-       tgd->tgd_statfs_inflight = 0;
-       tgd->tgd_osfs_inflight = 0;
-
-       /* grant data */
-       spin_lock_init(&tgd->tgd_grant_lock);
-       tgd->tgd_tot_dirty = 0;
-       tgd->tgd_tot_granted = 0;
-       tgd->tgd_tot_pending = 0;
-
        m->ofd_seq_count = 0;
        init_waitqueue_head(&m->ofd_inconsistency_thread.t_ctl_waitq);
        INIT_LIST_HEAD(&m->ofd_inconsistency_list);
@@ -3008,36 +2979,20 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
        ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
                           "filter_ldlm_cb_client", &obd->obd_ldlm_client);
 
-       dt_conf_get(env, m->ofd_osd, &m->ofd_lut.lut_dt_conf);
-
        rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice,
                      OBD_FAIL_OST_ALL_REQUEST_NET,
                      OBD_FAIL_OST_ALL_REPLY_NET);
        if (rc)
                GOTO(err_free_ns, rc);
 
-       /* populate cached statfs data */
-       osfs = &ofd_info(env)->fti_u.osfs;
-       rc = tgt_statfs_internal(env, &m->ofd_lut, osfs, 0, NULL);
-       if (rc != 0) {
-               CERROR("%s: can't get statfs data, rc %d\n", obd->obd_name, rc);
-               GOTO(err_fini_lut, rc);
-       }
-       if (!is_power_of_2(osfs->os_bsize)) {
-               CERROR("%s: blocksize (%d) is not a power of 2\n",
-                       obd->obd_name, osfs->os_bsize);
-               GOTO(err_fini_lut, rc = -EPROTO);
-       }
-       tgd->tgd_blockbits = fls(osfs->os_bsize) - 1;
+       tgd->tgd_reserved_pcnt = 0;
 
-       if (DT_DEF_BRW_SIZE < (1U << tgd->tgd_blockbits))
-               m->ofd_brw_size = 1U << tgd->tgd_blockbits;
-       else
-               m->ofd_brw_size = DT_DEF_BRW_SIZE;
-
-       m->ofd_cksum_types_supported = cksum_types_supported_server();
+       m->ofd_brw_size = m->ofd_lut.lut_dt_conf.ddp_brw_size;
+       m->ofd_cksum_types_supported =
+               obd_cksum_types_supported_server(obd->obd_name);
        m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT;
-       if (osfs->os_bsize * osfs->os_blocks < OFD_PRECREATE_SMALL_FS)
+       if (tgd->tgd_osfs.os_bsize * tgd->tgd_osfs.os_blocks <
+           OFD_PRECREATE_SMALL_FS)
                m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL;
 
        rc = ofd_fs_setup(env, m, obd);
@@ -3108,6 +3063,7 @@ static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
        stop.ls_status = LS_PAUSED;
        stop.ls_flags = 0;
        lfsck_stop(env, m->ofd_osd, &stop);
+       ofd_stack_pre_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
        target_recovery_fini(obd);
        if (m->ofd_namespace != NULL)
                ldlm_namespace_free_prior(m->ofd_namespace, NULL,
@@ -3123,11 +3079,6 @@ static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
        nm_config_file_deregister_tgt(env, obd->u.obt.obt_nodemap_config_file);
        obd->u.obt.obt_nodemap_config_file = NULL;
 
-       if (m->ofd_los != NULL) {
-               local_oid_storage_fini(env, m->ofd_los);
-               m->ofd_los = NULL;
-       }
-
        if (m->ofd_namespace != NULL) {
                ldlm_namespace_free_post(m->ofd_namespace);
                d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
@@ -3260,13 +3211,6 @@ static int __init ofd_init(void)
                return(rc);
        }
 
-       rc = ofd_dlm_init();
-       if (rc) {
-               lu_kmem_fini(ofd_caches);
-               ofd_fmd_exit();
-               return rc;
-       }
-
        rc = class_register_type(&ofd_obd_ops, NULL, true, NULL,
                                 LUSTRE_OST_NAME, &ofd_device_type);
        return rc;
@@ -3281,7 +3225,6 @@ static int __init ofd_init(void)
 static void __exit ofd_exit(void)
 {
        ofd_fmd_exit();
-       ofd_dlm_exit();
        lu_kmem_fini(ofd_caches);
        class_unregister_type(LUSTRE_OST_NAME);
 }