* Copyright 2009 Sun Microsystems, Inc. All rights reserved
* Use is subject to license terms.
*
- * Copyright (c) 2011, 2012, Intel, Inc.
+ * Copyright (c) 2012, Intel Corporation.
*
*/
/*
#define DEBUG_SUBSYSTEM S_MDS
#include <obd_class.h>
+#include <lustre_fid.h>
#include <lustre_param.h>
+#include <lustre_update.h>
#include "lod_internal.h"
+/**
+ * Lookup MDT/OST index \a tgt by FID \a fid.
+ *
+ * \param lod LOD to be lookup at.
+ * \param fid FID of object to find MDT/OST.
+ * \param tgt MDT/OST index to return.
+ * \param flags indidcate the FID is on MDS or OST.
+ **/
+int lod_fld_lookup(const struct lu_env *env, struct lod_device *lod,
+ const struct lu_fid *fid, __u32 *tgt, int flags)
+{
+ struct lu_seq_range range;
+ struct lu_server_fld *server_fld;
+ int rc = 0;
+ ENTRY;
+
+ LASSERTF(fid_is_sane(fid), "Invalid FID "DFID"\n", PFID(fid));
+ if (fid_is_idif(fid)) {
+ *tgt = fid_idif_ost_idx(fid);
+ RETURN(rc);
+ }
+
+ if (!lod->lod_initialized || (!fid_seq_in_fldb(fid_seq(fid)))) {
+ LASSERT(lu_site2seq(lod2lu_dev(lod)->ld_site) != NULL);
+ *tgt = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
+ RETURN(rc);
+ }
+
+ server_fld = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_server_fld;
+ range.lsr_flags = flags;
+ rc = fld_server_lookup(env, server_fld, fid_seq(fid), &range);
+ if (rc) {
+ CERROR("%s: Can't find tgt by seq "LPX64", rc %d\n",
+ lod2obd(lod)->obd_name, fid_seq(fid), rc);
+ RETURN(rc);
+ }
+
+ *tgt = range.lsr_index;
+
+ CDEBUG(D_INFO, "LOD: got tgt %x for sequence: "
+ LPX64"\n", *tgt, fid_seq(fid));
+
+ RETURN(rc);
+}
+
extern struct lu_object_operations lod_lu_obj_ops;
+extern struct lu_object_operations lod_lu_robj_ops;
extern struct dt_object_operations lod_obj_ops;
+extern struct dt_lock_operations lod_lock_ops;
/* Slab for OSD object allocation */
cfs_mem_cache_t *lod_object_kmem;
const struct lu_object_header *hdr,
struct lu_device *dev)
{
- struct lu_object *lu_obj;
- struct lod_object *lo;
+ struct lod_object *lod_obj;
+ struct lu_object *lu_obj;
+ const struct lu_fid *fid = &hdr->loh_fid;
+ mdsno_t mds;
+ int rc = 0;
+ ENTRY;
+
+ OBD_SLAB_ALLOC_PTR_GFP(lod_obj, lod_object_kmem, CFS_ALLOC_IO);
+ if (lod_obj == NULL)
+ RETURN(ERR_PTR(-ENOMEM));
- OBD_SLAB_ALLOC_PTR_GFP(lo, lod_object_kmem, CFS_ALLOC_IO);
- if (lo == NULL)
- return NULL;
+ rc = lod_fld_lookup(env, lu2lod_dev(dev), fid, &mds, LU_SEQ_RANGE_MDT);
+ if (rc) {
+ OBD_SLAB_FREE_PTR(lod_obj, lod_object_kmem);
+ RETURN(ERR_PTR(rc));
+ }
- lu_obj = lod2lu_obj(lo);
- dt_object_init(&lo->ldo_obj, NULL, dev);
- lo->ldo_obj.do_ops = &lod_obj_ops;
- lu_obj->lo_ops = &lod_lu_obj_ops;
+ lod_obj->ldo_mds_num = mds;
+ lu_obj = lod2lu_obj(lod_obj);
+ dt_object_init(&lod_obj->ldo_obj, NULL, dev);
+ lod_obj->ldo_obj.do_ops = &lod_obj_ops;
+ lod_obj->ldo_obj.do_lock_ops = &lod_lock_ops;
+ if (likely(mds == lu_site2seq(dev->ld_site)->ss_node_id))
+ lu_obj->lo_ops = &lod_lu_obj_ops;
+ else
+ lu_obj->lo_ops = &lod_lu_robj_ops;
+ RETURN(lu_obj);
+}
- return lu_obj;
+static int lod_cleanup_desc_tgts(const struct lu_env *env,
+ struct lod_device *lod,
+ struct lod_tgt_descs *ltd,
+ struct lustre_cfg *lcfg)
+{
+ struct lu_device *next;
+ int rc = 0;
+ int i;
+
+ lod_getref(ltd);
+ if (ltd->ltd_tgts_size <= 0) {
+ lod_putref(lod, ltd);
+ return 0;
+ }
+ cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
+ struct lod_tgt_desc *tgt;
+ int rc1;
+
+ tgt = LTD_TGT(ltd, i);
+ LASSERT(tgt && tgt->ltd_tgt);
+ next = &tgt->ltd_tgt->dd_lu_dev;
+ rc1 = next->ld_ops->ldo_process_config(env, next, lcfg);
+ if (rc1) {
+ CERROR("%s: error cleaning up LOD index %u: cmd %#x"
+ ": rc = %d\n", lod2obd(lod)->obd_name, i,
+ lcfg->lcfg_command, rc1);
+ rc = rc1;
+ }
+ }
+ lod_putref(lod, ltd);
+ return rc;
}
+static int lodname2mdt_index(char *lodname, int *index)
+{
+ char *ptr, *tmp;
+
+ /* The lodname suppose to be fsname-MDTxxxx-mdtlov */
+ ptr = strrchr(lodname, '-');
+ if (ptr == NULL) {
+ CERROR("invalid MDT index in '%s'\n", lodname);
+ return -EINVAL;
+ }
+
+ if (strncmp(ptr, "-mdtlov", 7) != 0) {
+ CERROR("invalid MDT index in '%s'\n", lodname);
+ return -EINVAL;
+ }
+
+ if ((unsigned long)ptr - (unsigned long)lodname <= 8) {
+ CERROR("invalid MDT index in '%s'\n", lodname);
+ return -EINVAL;
+ }
+
+ if (strncmp(ptr - 8, "-MDT", 4) != 0) {
+ CERROR("invalid MDT index in '%s'\n", lodname);
+ return -EINVAL;
+ }
+
+ *index = simple_strtol(ptr - 4, &tmp, 16);
+ if (*tmp != '-' || *index > INT_MAX || *index < 0) {
+ CERROR("invalid MDT index in '%s'\n", lodname);
+ return -EINVAL;
+ }
+ return 0;
+}
+
+/*
+ * Init client sequence manager which is used by local MDS to talk to sequence
+ * controller on remote node.
+ */
+static int lod_seq_init_cli(const struct lu_env *env,
+ struct lod_device *lod,
+ char *tgtuuid, int index)
+{
+ struct seq_server_site *ss;
+ struct obd_device *osp;
+ int rc;
+ char *prefix;
+ struct obd_uuid obd_uuid;
+ ENTRY;
+
+ ss = lu_site2seq(lod2lu_dev(lod)->ld_site);
+ LASSERT(ss != NULL);
+
+ /* check if this is adding the first MDC and controller is not yet
+ * initialized. */
+ if (index != 0 || ss->ss_client_seq)
+ RETURN(0);
+
+ obd_str2uuid(&obd_uuid, tgtuuid);
+ osp = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
+ &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
+ if (osp == NULL) {
+ CERROR("%s: can't find %s device\n",
+ lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_name,
+ tgtuuid);
+ RETURN(-EINVAL);
+ }
+
+ if (!osp->obd_set_up) {
+ CERROR("target %s not set up\n", osp->obd_name);
+ rc = -EINVAL;
+ }
+
+ LASSERT(ss->ss_control_exp);
+ OBD_ALLOC_PTR(ss->ss_client_seq);
+ if (ss->ss_client_seq == NULL)
+ RETURN(-ENOMEM);
+
+ OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
+ if (!prefix) {
+ OBD_FREE_PTR(ss->ss_client_seq);
+ ss->ss_client_seq = NULL;
+ RETURN(-ENOMEM);
+ }
+
+ snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", osp->obd_name);
+ rc = seq_client_init(ss->ss_client_seq, ss->ss_control_exp,
+ LUSTRE_SEQ_METADATA, prefix, NULL);
+ OBD_FREE(prefix, MAX_OBD_NAME + 5);
+ if (rc) {
+ OBD_FREE_PTR(ss->ss_client_seq);
+ ss->ss_client_seq = NULL;
+ RETURN(rc);
+ }
+
+ LASSERT(ss->ss_server_seq != NULL);
+ rc = seq_server_set_cli(ss->ss_server_seq, ss->ss_client_seq,
+ env);
+
+ RETURN(rc);
+}
+
+static void lod_seq_fini_cli(struct lod_device *lod)
+{
+ struct seq_server_site *ss;
+
+ ENTRY;
+
+ ss = lu_site2seq(lod2lu_dev(lod)->ld_site);
+ if (ss == NULL) {
+ EXIT;
+ return;
+ }
+
+ if (ss->ss_server_seq)
+ seq_server_set_cli(ss->ss_server_seq,
+ NULL, NULL);
+
+ if (ss->ss_control_exp) {
+ class_export_put(ss->ss_control_exp);
+ ss->ss_control_exp = NULL;
+ }
+
+ EXIT;
+ return;
+}
+
+/**
+ * Procss config log on LOD
+ * \param env environment info
+ * \param dev lod device
+ * \param lcfg config log
+ *
+ * Add osc config log,
+ * marker 20 (flags=0x01, v2.2.49.56) lustre-OST0001 'add osc'
+ * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nidxxx
+ * attach 0:lustre-OST0001-osc-MDT0001 1:osc 2:lustre-MDT0001-mdtlov_UUID
+ * setup 0:lustre-OST0001-osc-MDT0001 1:lustre-OST0001_UUID 2:nid
+ * lov_modify_tgts add 0:lustre-MDT0001-mdtlov 1:lustre-OST0001_UUID 2:1 3:1
+ * marker 20 (flags=0x02, v2.2.49.56) lustre-OST0001 'add osc'
+ *
+ * Add mdc config log
+ * marker 10 (flags=0x01, v2.2.49.56) lustre-MDT0000 'add osp'
+ * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nid
+ * attach 0:lustre-MDT0000-osp-MDT0001 1:osp 2:lustre-MDT0001-mdtlov_UUID
+ * setup 0:lustre-MDT0000-osp-MDT0001 1:lustre-MDT0000_UUID 2:nid
+ * modify_mdc_tgts add 0:lustre-MDT0001 1:lustre-MDT0000_UUID 2:0 3:1
+ * marker 10 (flags=0x02, v2.2.49.56) lustre-MDT0000_UUID 'add osp'
+ **/
static int lod_process_config(const struct lu_env *env,
struct lu_device *dev,
struct lustre_cfg *lcfg)
struct lod_device *lod = lu2lod_dev(dev);
struct lu_device *next = &lod->lod_child->dd_lu_dev;
char *arg1;
- int rc, i;
+ int rc = 0;
ENTRY;
switch(lcfg->lcfg_command) {
-
case LCFG_LOV_DEL_OBD:
case LCFG_LOV_ADD_INA:
- case LCFG_LOV_ADD_OBD: {
+ case LCFG_LOV_ADD_OBD:
+ case LCFG_ADD_MDC: {
__u32 index;
+ __u32 mdt_index;
int gen;
- /* lov_modify_tgts add 0:lov_mdsA 1:osp 2:0 3:1 */
+ /* lov_modify_tgts add 0:lov_mdsA 1:osp 2:0 3:1
+ * modify_mdc_tgts add 0:lustre-MDT0001
+ * 1:lustre-MDT0001-mdc0002
+ * 2:2 3:1*/
arg1 = lustre_cfg_string(lcfg, 1);
if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
GOTO(out, rc = -EINVAL);
- if (lcfg->lcfg_command == LCFG_LOV_ADD_OBD)
- rc = lod_add_device(env, lod, arg1, index, gen, 1);
- else if (lcfg->lcfg_command == LCFG_LOV_ADD_INA)
- rc = lod_add_device(env, lod, arg1, index, gen, 0);
- else
- rc = lod_del_device(env, lod, arg1, index, gen);
+ if (lcfg->lcfg_command == LCFG_LOV_ADD_OBD) {
+ char *mdt;
+ mdt = strstr(lustre_cfg_string(lcfg, 0), "-MDT");
+ /* 1.8 configs don't have "-MDT0000" at the end */
+ if (mdt == NULL) {
+ mdt_index = 0;
+ } else {
+ rc = lodname2mdt_index(
+ lustre_cfg_string(lcfg, 0), &mdt_index);
+ if (rc != 0)
+ GOTO(out, rc);
+ }
+ rc = lod_add_device(env, lod, arg1, index, gen,
+ mdt_index, LUSTRE_OSC_NAME, 1);
+ } else if (lcfg->lcfg_command == LCFG_ADD_MDC) {
+ mdt_index = index;
+ rc = lod_add_device(env, lod, arg1, index, gen,
+ mdt_index, LUSTRE_MDC_NAME, 1);
+ if (rc == 0)
+ rc = lod_seq_init_cli(env, lod, arg1,
+ mdt_index);
+ } else if (lcfg->lcfg_command == LCFG_LOV_ADD_INA) {
+ /*FIXME: Add mdt_index for LCFG_LOV_ADD_INA*/
+ mdt_index = 0;
+ rc = lod_add_device(env, lod, arg1, index, gen,
+ mdt_index, LUSTRE_OSC_NAME, 0);
+ } else {
+ rc = lod_del_device(env, lod,
+ &lod->lod_ost_descs,
+ arg1, index, gen);
+ }
break;
}
if (rc > 0)
rc = 0;
GOTO(out, rc);
- }
-
+ }
case LCFG_CLEANUP:
+ case LCFG_PRE_CLEANUP: {
lu_dev_del_linkage(dev->ld_site, dev);
- lod_getref(lod);
- lod_foreach_ost(lod, i) {
- struct lod_ost_desc *ost;
- ost = OST_TGT(lod, i);
- LASSERT(ost && ost->ltd_ost);
- next = &ost->ltd_ost->dd_lu_dev;
- rc = next->ld_ops->ldo_process_config(env, next, lcfg);
- if (rc)
- CERROR("%s: can't process %u: %d\n",
- lod2obd(lod)->obd_name,
- lcfg->lcfg_command, rc);
- }
- lod_putref(lod);
+ lod_cleanup_desc_tgts(env, lod, &lod->lod_mdt_descs, lcfg);
+ lod_cleanup_desc_tgts(env, lod, &lod->lod_ost_descs, lcfg);
+
+ lod_seq_fini_cli(lod);
+ if (lcfg->lcfg_command == LCFG_PRE_CLEANUP)
+ break;
/*
* do cleanup on underlying storage only when
* all OSPs are cleaned up, as they use that OSD as well
if (rc)
CERROR("error in disconnect from storage: %d\n", rc);
break;
-
+ }
default:
CERROR("%s: unknown command %u\n", lod2obd(lod)->obd_name,
lcfg->lcfg_command);
{
struct lod_device *lod = lu2lod_dev(dev);
struct lu_device *next = &lod->lod_child->dd_lu_dev;
- struct lod_ost_desc *ost;
int i, rc;
ENTRY;
rc = next->ld_ops->ldo_recovery_complete(env, next);
- lod_getref(lod);
- lod_foreach_ost(lod, i) {
- ost = OST_TGT(lod, i);
- LASSERT(ost && ost->ltd_ost);
- next = &ost->ltd_ost->dd_lu_dev;
- rc = next->ld_ops->ldo_recovery_complete(env, next);
- if (rc)
- CERROR("%s: can't complete recovery on #%d: %d\n",
- lod2obd(lod)->obd_name, i, rc);
+ lod_getref(&lod->lod_ost_descs);
+ if (lod->lod_osts_size > 0) {
+ cfs_foreach_bit(lod->lod_ost_bitmap, i) {
+ struct lod_tgt_desc *tgt;
+ tgt = OST_TGT(lod, i);
+ LASSERT(tgt && tgt->ltd_tgt);
+ next = &tgt->ltd_ost->dd_lu_dev;
+ rc = next->ld_ops->ldo_recovery_complete(env, next);
+ if (rc)
+ CERROR("%s: can't complete recovery on #%d:"
+ "%d\n", lod2obd(lod)->obd_name, i, rc);
+ }
}
- lod_putref(lod);
-
+ lod_putref(lod, &lod->lod_ost_descs);
RETURN(rc);
}
ENTRY;
rc = next->ld_ops->ldo_prepare(env, pdev, next);
+ if (rc != 0) {
+ CERROR("%s: prepare bottom error: rc = %d\n",
+ lod2obd(lod)->obd_name, rc);
+ RETURN(rc);
+ }
+
+ lod->lod_initialized = 1;
RETURN(rc);
}
static struct thandle *lod_trans_create(const struct lu_env *env,
struct dt_device *dev)
{
- return dt_trans_create(env, dt2lod_dev(dev)->lod_child);
+ struct thandle *th;
+
+ th = dt_trans_create(env, dt2lod_dev(dev)->lod_child);
+ if (IS_ERR(th))
+ return th;
+
+ CFS_INIT_LIST_HEAD(&th->th_remote_update_list);
+ return th;
+}
+
+static int lod_remote_sync(const struct lu_env *env, struct dt_device *dev,
+ struct thandle *th)
+{
+ struct update_request *update;
+ int rc = 0;
+ ENTRY;
+
+ if (cfs_list_empty(&th->th_remote_update_list))
+ RETURN(0);
+
+ cfs_list_for_each_entry(update, &th->th_remote_update_list,
+ ur_list) {
+ /* In DNE phase I, there should be only one OSP
+ * here, so we will do send/receive one by one,
+ * instead of sending them parallel, will fix this
+ * in Phase II */
+ th->th_current_request = update;
+ rc = dt_trans_start(env, update->ur_dt, th);
+ if (rc != 0) {
+ /* FIXME how to revert the partial results
+ * once error happened? Resolved by 2 Phase commit */
+ update->ur_rc = rc;
+ break;
+ }
+ }
+
+ RETURN(rc);
}
static int lod_trans_start(const struct lu_env *env, struct dt_device *dev,
struct thandle *th)
{
- return dt_trans_start(env, dt2lod_dev(dev)->lod_child, th);
+ struct lod_device *lod = dt2lod_dev((struct dt_device *) dev);
+ int rc;
+
+ rc = lod_remote_sync(env, dev, th);
+ if (rc)
+ return rc;
+
+ return dt_trans_start(env, lod->lod_child, th);
}
static int lod_trans_stop(const struct lu_env *env, struct thandle *th)
{
- /* XXX: we don't know next device, will be fixed with DNE */
- return dt_trans_stop(env, th->th_dev, th);
+ struct update_request *update;
+ struct update_request *tmp;
+ int rc = 0;
+ int rc2 = 0;
+
+ cfs_list_for_each_entry_safe(update, tmp,
+ &th->th_remote_update_list,
+ ur_list) {
+ th->th_current_request = update;
+ rc2 = dt_trans_stop(env, update->ur_dt, th);
+ if (unlikely(rc2 != 0 && rc == 0))
+ rc = rc2;
+ }
+
+ rc2 = dt_trans_stop(env, th->th_dev, th);
+
+ return rc2 != 0 ? rc2 : rc;
}
static void lod_conf_get(const struct lu_env *env,
int rc = 0, i;
ENTRY;
- lod_getref(lod);
+ lod_getref(&lod->lod_ost_descs);
lod_foreach_ost(lod, i) {
ost = OST_TGT(lod, i);
LASSERT(ost && ost->ltd_ost);
break;
}
}
- lod_putref(lod);
+ lod_putref(lod, &lod->lod_ost_descs);
if (rc == 0)
rc = dt_sync(env, lod->lod_child);
RETURN(rc);
}
+static int lod_tgt_desc_init(struct lod_tgt_descs *ltd)
+{
+ mutex_init(<d->ltd_mutex);
+ init_rwsem(<d->ltd_rw_sem);
+
+ /* the OST array and bitmap are allocated/grown dynamically as OSTs are
+ * added to the LOD, see lod_add_device() */
+ ltd->ltd_tgt_bitmap = CFS_ALLOCATE_BITMAP(32);
+ if (ltd->ltd_tgt_bitmap == NULL)
+ RETURN(-ENOMEM);
+
+ ltd->ltd_tgts_size = 32;
+ ltd->ltd_tgtnr = 0;
+
+ ltd->ltd_death_row = 0;
+ ltd->ltd_refcount = 0;
+ return 0;
+}
+
static int lod_init0(const struct lu_env *env, struct lod_device *lod,
struct lu_device_type *ldt, struct lustre_cfg *cfg)
{
if (rc)
GOTO(out_pools, rc);
- mutex_init(&lod->lod_mutex);
- init_rwsem(&lod->lod_rw_sem);
spin_lock_init(&lod->lod_desc_lock);
+ spin_lock_init(&lod->lod_connects_lock);
+ lod_tgt_desc_init(&lod->lod_mdt_descs);
+ lod_tgt_desc_init(&lod->lod_ost_descs);
RETURN(0);
struct lu_device *d)
{
struct lod_device *lod = lu2lod_dev(d);
+ int rc;
ENTRY;
lod_pools_fini(lod);
lod_procfs_fini(lod);
+ rc = lod_fini_tgt(lod, &lod->lod_ost_descs);
+ if (rc)
+ CERROR("%s:can not fini ost descs %d\n",
+ lod2obd(lod)->obd_name, rc);
+
+ rc = lod_fini_tgt(lod, &lod->lod_mdt_descs);
+ if (rc)
+ CERROR("%s:can not fini mdt descs %d\n",
+ lod2obd(lod)->obd_name, rc);
+
RETURN(NULL);
}
*exp = class_conn2export(&conn);
- mutex_lock(&lod->lod_mutex);
+ spin_lock(&lod->lod_connects_lock);
lod->lod_connects++;
/* at the moment we expect the only user */
LASSERT(lod->lod_connects == 1);
- mutex_unlock(&lod->lod_mutex);
+ spin_unlock(&lod->lod_connects_lock);
RETURN(0);
}
ENTRY;
/* Only disconnect the underlying layers on the final disconnect. */
- mutex_lock(&lod->lod_mutex);
+ spin_lock(&lod->lod_connects_lock);
lod->lod_connects--;
if (lod->lod_connects != 0) {
/* why should there be more than 1 connect? */
- mutex_unlock(&lod->lod_mutex);
+ spin_unlock(&lod->lod_connects_lock);
CERROR("%s: disconnect #%d\n", exp->exp_obd->obd_name,
lod->lod_connects);
goto out;
}
- mutex_unlock(&lod->lod_mutex);
+ spin_unlock(&lod->lod_connects_lock);
/* the last user of lod has gone, let's release the device */
release = 1;
ENTRY;
LASSERT(d);
- lod_getref(d);
+ lod_getref(&d->lod_ost_descs);
lod_foreach_ost(d, i) {
ost = OST_TGT(d, i);
LASSERT(ost && ost->ltd_ost);
if (rc == 0)
break;
}
- lod_putref(d);
+ lod_putref(d, &d->lod_ost_descs);
RETURN(rc);
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff