* Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
* Author: Mikhail Pershin <mike.pershin@intel.com>
*/
+/**
+ * The Logical Object Device (LOD) layer manages access to striped
+ * objects (both regular files and directories). It implements the DT
+ * device and object APIs and is responsible for creating, storing,
+ * and loading striping information as an extended attribute of the
+ * underlying OSD object. LOD is the server side analog of the LOV and
+ * LMV layers on the client side.
+ *
+ * Metadata LU object stack (layers of the same compound LU object,
+ * all have the same FID):
+ *
+ * MDT
+ * | MD API
+ * MDD
+ * | DT API
+ * LOD
+ * / \ DT API
+ * OSD OSP
+ *
+ * During LOD object initialization the localness or remoteness of the
+ * object FID dictates the choice between OSD and OSP.
+ *
+ * An LOD object (file or directory) with N stripes (each has a
+ * different FID):
+ *
+ * LOD
+ * |
+ * +---+---+---+...+
+ * | | | | |
+ * S0 S1 S2 S3 S(N-1) OS[DP] objects, seen as DT objects by LOD
+ *
+ * When upper layers must access an object's stripes (which are
+ * themselves OST or MDT LU objects) LOD finds these objects by their
+ * FIDs and stores them as an array of DT object pointers on the
+ * object. Declarations and operations on LOD objects are received by
+ * LOD (as DT object operations) and performed on the underlying
+ * OS[DP] object and (as needed) on the stripes. From the perspective
+ * of LOD, a stripe-less file (created by mknod() or open with
+ * O_LOV_DELAY_CREATE) is an object which does not yet have stripes,
+ * while a non-striped directory (created by mkdir()) is an object
+ * which will never have stripes.
+ *
+ * The LOD layer also implements a small subset of the OBD device API
+ * to support MDT stack initialization and finalization (an MDD device
+ * connects and disconnects itself to and from the underlying LOD
+ * device), and pool management. In turn LOD uses the OBD device API
+ * to connect it self to the underlying OSD, and to connect itself to
+ * OSP devices representing the MDTs and OSTs that bear the stripes of
+ * its objects.
+ */
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
#define DEBUG_SUBSYSTEM S_MDS
#include <obd_class.h>
+#include <md_object.h>
#include <lustre_fid.h>
#include <lustre_param.h>
#include <lustre_update.h>
* \param type 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 type)
+ const struct lu_fid *fid, __u32 *tgt, int *type)
{
struct lu_seq_range range = { 0 };
struct lu_server_fld *server_fld;
ENTRY;
LASSERTF(fid_is_sane(fid), "Invalid FID "DFID"\n", PFID(fid));
+
if (fid_is_idif(fid)) {
*tgt = fid_idif_ost_idx(fid);
+ *type = LU_SEQ_RANGE_OST;
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;
+ *type = LU_SEQ_RANGE_MDT;
RETURN(rc);
}
server_fld = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_server_fld;
- fld_range_set_type(&range, type);
+ fld_range_set_type(&range, *type);
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);
+ if (rc)
RETURN(rc);
- }
*tgt = range.lsr_index;
+ *type = range.lsr_flags;
CDEBUG(D_INFO, "LOD: got tgt %x for sequence: "
LPX64"\n", *tgt, fid_seq(fid));
}
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;
/* Slab for OSD object allocation */
-cfs_mem_cache_t *lod_object_kmem;
+struct kmem_cache *lod_object_kmem;
static struct lu_kmem_descr lod_caches[] = {
{
{
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);
+ OBD_SLAB_ALLOC_PTR_GFP(lod_obj, lod_object_kmem, GFP_NOFS);
if (lod_obj == NULL)
RETURN(ERR_PTR(-ENOMEM));
- 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));
- }
-
- 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;
- 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;
+ lu_obj->lo_ops = &lod_lu_obj_ops;
+
RETURN(lu_obj);
}
{
struct lu_device *next;
int rc = 0;
- int i;
+ unsigned int i;
lod_getref(ltd);
if (ltd->ltd_tgts_size <= 0) {
return rc;
}
-static int lodname2mdt_index(char *lodname, int *index)
+static int lodname2mdt_index(char *lodname, long *index)
{
char *ptr, *tmp;
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
if (mdt == NULL) {
mdt_index = 0;
} else {
+ long long_index;
rc = lodname2mdt_index(
- lustre_cfg_string(lcfg, 0), &mdt_index);
+ lustre_cfg_string(lcfg, 0),
+ &long_index);
if (rc != 0)
GOTO(out, rc);
+ mdt_index = long_index;
}
rc = lod_add_device(env, lod, arg1, index, gen,
mdt_index, LUSTRE_OSC_NAME, 1);
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;
} else {
rc = lod_del_device(env, lod,
&lod->lod_ost_descs,
- arg1, index, gen);
+ arg1, index, gen, true);
}
break;
}
case LCFG_PARAM: {
- struct lprocfs_static_vars v = { 0 };
- struct obd_device *obd = lod2obd(lod);
-
- lprocfs_lod_init_vars(&v);
+ struct obd_device *obd = lod2obd(lod);
- rc = class_process_proc_param(PARAM_LOV, v.obd_vars, lcfg, obd);
+ rc = class_process_proc_seq_param(PARAM_LOV, obd->obd_vars,
+ lcfg, obd);
if (rc > 0)
rc = 0;
GOTO(out, rc);
lu_dev_del_linkage(dev->ld_site, dev);
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;
/*
{
struct lod_device *lod = lu2lod_dev(dev);
struct lu_device *next = &lod->lod_child->dd_lu_dev;
- int i, rc;
+ unsigned int i;
+ int rc;
ENTRY;
LASSERT(lod->lod_recovery_completed == 0);
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)
{
struct lod_device *lod = dt2lod_dev((struct dt_device *) dev);
- int rc;
+ int rc = 0;
- rc = lod_remote_sync(env, dev, th);
- if (rc)
- return rc;
+ if (unlikely(th->th_update != NULL)) {
+ struct thandle_update *tu = th->th_update;
+ struct dt_update_request *update;
+ list_for_each_entry(update, &tu->tu_remote_update_list,
+ dur_list) {
+ LASSERT(update->dur_dt != NULL);
+ rc = dt_trans_start(env, update->dur_dt, th);
+ if (rc != 0)
+ return rc;
+ }
+ }
return dt_trans_start(env, lod->lod_child, th);
}
-static int lod_trans_stop(const struct lu_env *env, struct thandle *th)
+static int lod_trans_stop(const struct lu_env *env, struct dt_device *dt,
+ struct thandle *th)
{
- struct update_request *update;
- struct update_request *tmp;
- int rc = 0;
- int rc2 = 0;
+ struct thandle_update *tu = th->th_update;
+ struct dt_update_request *update;
+ struct dt_update_request *tmp;
+ int rc2 = 0;
+ int rc;
+ ENTRY;
+
+ rc = dt_trans_stop(env, th->th_dev, th);
+ if (likely(tu == NULL))
+ RETURN(rc);
- 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);
+ list_for_each_entry_safe(update, tmp,
+ &tu->tu_remote_update_list,
+ dur_list) {
+ /* update will be freed inside dt_trans_stop */
+ rc2 = dt_trans_stop(env, update->dur_dt, th);
if (unlikely(rc2 != 0 && rc == 0))
rc = rc2;
}
- rc2 = dt_trans_stop(env, th->th_dev, th);
-
- return rc2 != 0 ? rc2 : rc;
+ RETURN(rc);
}
static void lod_conf_get(const struct lu_env *env,
{
struct lod_device *lod = dt2lod_dev(dev);
struct lod_ost_desc *ost;
- int rc = 0, i;
+ unsigned int i;
+ int rc = 0;
ENTRY;
lod_getref(&lod->lod_ost_descs);
struct lu_device *next = &lod->lod_child->dd_lu_dev;
ENTRY;
- LASSERT(cfs_atomic_read(&lu->ld_ref) == 0);
+ LASSERT(atomic_read(&lu->ld_ref) == 0);
dt_device_fini(&lod->lod_dt_dev);
OBD_FREE_PTR(lod);
RETURN(next);
lod_procfs_fini(lod);
- rc = lod_fini_tgt(lod, &lod->lod_ost_descs);
+ rc = lod_fini_tgt(env, lod, &lod->lod_ost_descs, true);
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);
+ rc = lod_fini_tgt(env, lod, &lod->lod_mdt_descs, false);
if (rc)
CERROR("%s:can not fini mdt descs %d\n",
lod2obd(lod)->obd_name, rc);
info->lti_ea_store = NULL;
info->lti_ea_store_size = 0;
}
+ lu_buf_free(&info->lti_linkea_buf);
OBD_FREE_PTR(info);
}
.ldt_ctx_tags = LCT_MD_THREAD,
};
-static int lod_obd_health_check(const struct lu_env *env,
- struct obd_device *obd)
+static int lod_obd_get_info(const struct lu_env *env, struct obd_export *exp,
+ __u32 keylen, void *key, __u32 *vallen, void *val,
+ struct lov_stripe_md *lsm)
{
- struct lod_device *d = lu2lod_dev(obd->obd_lu_dev);
- struct lod_ost_desc *ost;
- int i, rc = 1;
- ENTRY;
+ int rc = -EINVAL;
- LASSERT(d);
- lod_getref(&d->lod_ost_descs);
- lod_foreach_ost(d, i) {
- ost = OST_TGT(d, i);
- LASSERT(ost && ost->ltd_ost);
- rc = obd_health_check(env, ost->ltd_exp->exp_obd);
- /* one healthy device is enough */
- if (rc == 0)
- break;
+ if (KEY_IS(KEY_OSP_CONNECTED)) {
+ struct obd_device *obd = exp->exp_obd;
+ struct lod_device *d;
+ struct lod_ost_desc *ost;
+ unsigned int i;
+ int rc = 1;
+
+ if (!obd->obd_set_up || obd->obd_stopping)
+ RETURN(-EAGAIN);
+
+ d = lu2lod_dev(obd->obd_lu_dev);
+ lod_getref(&d->lod_ost_descs);
+ lod_foreach_ost(d, i) {
+ ost = OST_TGT(d, i);
+ LASSERT(ost && ost->ltd_ost);
+
+ rc = obd_get_info(env, ost->ltd_exp, keylen, key,
+ vallen, val, lsm);
+ /* one healthy device is enough */
+ if (rc == 0)
+ break;
+ }
+ lod_putref(d, &d->lod_ost_descs);
+ RETURN(rc);
}
- lod_putref(d, &d->lod_ost_descs);
+
RETURN(rc);
}
.o_owner = THIS_MODULE,
.o_connect = lod_obd_connect,
.o_disconnect = lod_obd_disconnect,
- .o_health_check = lod_obd_health_check,
+ .o_get_info = lod_obd_get_info,
.o_pool_new = lod_pool_new,
.o_pool_rem = lod_pool_remove,
.o_pool_add = lod_pool_add,
static int __init lod_mod_init(void)
{
- struct lprocfs_static_vars lvars = { 0 };
- cfs_proc_dir_entry_t *lov_proc_dir;
- int rc;
+ struct obd_type *type;
+ int rc;
rc = lu_kmem_init(lod_caches);
if (rc)
return rc;
- lprocfs_lod_init_vars(&lvars);
-
- rc = class_register_type(&lod_obd_device_ops, NULL, lvars.module_vars,
+ rc = class_register_type(&lod_obd_device_ops, NULL, true, NULL,
+#ifndef HAVE_ONLY_PROCFS_SEQ
+ NULL,
+#endif
LUSTRE_LOD_NAME, &lod_device_type);
if (rc) {
lu_kmem_fini(lod_caches);
}
/* create "lov" entry in procfs for compatibility purposes */
- lov_proc_dir = lprocfs_srch(proc_lustre_root, "lov");
- if (lov_proc_dir == NULL) {
- lov_proc_dir = lprocfs_register("lov", proc_lustre_root,
- NULL, NULL);
- if (IS_ERR(lov_proc_dir))
- CERROR("lod: can't create compat entry \"lov\": %d\n",
- (int)PTR_ERR(lov_proc_dir));
- }
+ type = class_search_type(LUSTRE_LOV_NAME);
+ if (type != NULL && type->typ_procroot != NULL)
+ return rc;
+ type = class_search_type(LUSTRE_LOD_NAME);
+ type->typ_procsym = lprocfs_seq_register("lov", proc_lustre_root,
+ NULL, NULL);
+ if (IS_ERR(type->typ_procsym)) {
+ CERROR("lod: can't create compat entry \"lov\": %d\n",
+ (int)PTR_ERR(type->typ_procsym));
+ type->typ_procsym = NULL;
+ }
return rc;
}
static void __exit lod_mod_exit(void)
{
-
- lprocfs_try_remove_proc_entry("lov", proc_lustre_root);
-
class_unregister_type(LUSTRE_LOD_NAME);
lu_kmem_fini(lod_caches);
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff