#include <obd_support.h>
#include <lustre_fid.h>
#include <lustre_fld.h>
-#include <lustre_mdt.h> /* err_serious() */
+#include <md_object.h>
#include <lustre_req_layout.h>
#include <lprocfs_status.h>
#include "fld_internal.h"
LU_KEY_INIT_FINI(fld, struct fld_thread_info);
/* context key: fld_thread_key */
+/* MGS thread may create llog file causing FLD lookup */
LU_CONTEXT_KEY_DEFINE(fld, LCT_MD_THREAD | LCT_DT_THREAD | LCT_MG_THREAD);
int fld_server_mod_init(void)
EXPORT_SYMBOL(fld_server_create);
/**
- * Lookup mds by seq, returns a range for given seq.
- *
- * If that entry is not cached in fld cache, request is sent to super
- * sequence controller node (MDT0). All other MDT[1...N] and client
- * cache fld entries, but this cache is not persistent.
- */
-int fld_server_lookup(const struct lu_env *env, struct lu_server_fld *fld,
- seqno_t seq, struct lu_seq_range *range)
+ * Extract index information from fld name like srv-fsname-MDT0000
+ **/
+int fld_name_to_index(const char *name, __u32 *index)
+{
+ char *dash;
+ int rc;
+ ENTRY;
+
+ CDEBUG(D_INFO, "get index from %s\n", name);
+ dash = strrchr(name, '-');
+ if (dash == NULL)
+ RETURN(-EINVAL);
+ dash++;
+ rc = target_name2index(dash, index, NULL);
+ RETURN(rc);
+}
+
+/**
+ * Retrieve fldb entry from MDT0 and add to local FLDB and cache.
+ **/
+int fld_update_from_controller(const struct lu_env *env,
+ struct lu_server_fld *fld)
+{
+ struct fld_thread_info *info;
+ struct lu_seq_range *range;
+ struct lu_seq_range_array *lsra;
+ __u32 index;
+ struct ptlrpc_request *req;
+ int rc;
+ int i;
+ ENTRY;
+
+ /* Update only happens during initalization, i.e. local FLDB
+ * does not exist yet */
+ if (!fld->lsf_new)
+ RETURN(0);
+
+ rc = fld_name_to_index(fld->lsf_name, &index);
+ if (rc < 0)
+ RETURN(rc);
+
+ /* No need update fldb for MDT0 */
+ if (index == 0)
+ RETURN(0);
+
+ info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
+ LASSERT(info != NULL);
+ range = &info->fti_lrange;
+ memset(range, 0, sizeof(*range));
+ range->lsr_index = index;
+ fld_range_set_mdt(range);
+
+ do {
+ rc = fld_client_rpc(fld->lsf_control_exp, range, FLD_READ,
+ &req);
+ if (rc != 0 && rc != -EAGAIN)
+ GOTO(out, rc);
+
+ LASSERT(req != NULL);
+ lsra = (struct lu_seq_range_array *)req_capsule_server_get(
+ &req->rq_pill, &RMF_GENERIC_DATA);
+ if (lsra == NULL)
+ GOTO(out, rc = -EPROTO);
+
+ range_array_le_to_cpu(lsra, lsra);
+ for (i = 0; i < lsra->lsra_count; i++) {
+ int rc1;
+
+ if (lsra->lsra_lsr[i].lsr_flags != LU_SEQ_RANGE_MDT)
+ GOTO(out, rc = -EINVAL);
+
+ if (lsra->lsra_lsr[i].lsr_index != index)
+ GOTO(out, rc = -EINVAL);
+
+ rc1 = fld_insert_entry(env, fld, &lsra->lsra_lsr[i]);
+ if (rc1 != 0)
+ GOTO(out, rc = rc1);
+ }
+ if (rc == -EAGAIN)
+ *range = lsra->lsra_lsr[lsra->lsra_count - 1];
+ } while (rc == -EAGAIN);
+
+ fld->lsf_new = 1;
+out:
+ if (req != NULL)
+ ptlrpc_req_finished(req);
+
+ RETURN(rc);
+}
+EXPORT_SYMBOL(fld_update_from_controller);
+
+/**
+ * Lookup sequece in local cache/fldb.
+ **/
+int fld_local_lookup(const struct lu_env *env, struct lu_server_fld *fld,
+ seqno_t seq, struct lu_seq_range *range)
{
struct lu_seq_range *erange;
struct fld_thread_info *info;
*range = *erange;
RETURN(0);
}
+ RETURN(rc);
+}
+EXPORT_SYMBOL(fld_local_lookup);
- if (fld->lsf_obj) {
+/**
+ * Lookup MDT/OST by seq, returns a range for given seq.
+ *
+ * If that entry is not cached in fld cache, request is sent to super
+ * sequence controller node (MDT0). All other MDT[1...N] and client
+ * cache fld entries, but this cache is not persistent.
+ */
+int fld_server_lookup(const struct lu_env *env, struct lu_server_fld *fld,
+ seqno_t seq, struct lu_seq_range *range)
+{
+ __u32 index;
+ int rc;
+ ENTRY;
+
+ rc = fld_local_lookup(env, fld, seq, range);
+ if (likely(rc == 0))
+ RETURN(rc);
+
+ rc = fld_name_to_index(fld->lsf_name, &index);
+ if (rc < 0)
+ RETURN(rc);
+ else
+ rc = 0;
+
+ if (index == 0) {
/* On server side, all entries should be in cache.
* If we can not find it in cache, just return error */
CERROR("%s: Cannot find sequence "LPX64": rc = %d\n",
fld->lsf_name, seq, -EIO);
RETURN(-EIO);
} else {
- LASSERT(fld->lsf_control_exp);
+ if (fld->lsf_control_exp == NULL) {
+ CERROR("%s: lookup "LPX64", but not connects to MDT0"
+ "yet: rc = %d.\n", fld->lsf_name, seq, -EIO);
+ RETURN(-EIO);
+ }
/* send request to mdt0 i.e. super seq. controller.
* This is temporary solution, long term solution is fld
* replication on all mdt servers.
*/
range->lsr_start = seq;
rc = fld_client_rpc(fld->lsf_control_exp,
- range, FLD_LOOKUP);
+ range, FLD_QUERY, NULL);
if (rc == 0)
fld_cache_insert(fld->lsf_cache, range);
}
* All MDT server handle fld lookup operation. But only MDT0 has fld index.
* if entry is not found in cache we need to forward lookup request to MDT0
*/
-
-static int fld_server_handle(struct lu_server_fld *fld,
- const struct lu_env *env,
- __u32 opc, struct lu_seq_range *range,
- struct fld_thread_info *info)
+static int fld_handle_lookup(struct tgt_session_info *tsi)
{
- int rc;
- ENTRY;
+ struct obd_export *exp = tsi->tsi_exp;
+ struct lu_site *site = exp->exp_obd->obd_lu_dev->ld_site;
+ struct lu_server_fld *fld;
+ struct lu_seq_range *in;
+ struct lu_seq_range *out;
+ int rc;
- switch (opc) {
- case FLD_LOOKUP:
- rc = fld_server_lookup(env, fld, range->lsr_start, range);
- break;
- default:
- rc = -EINVAL;
- break;
- }
+ ENTRY;
- CDEBUG(D_INFO, "%s: FLD req handle: error %d (opc: %d, range: "
- DRANGE"\n", fld->lsf_name, rc, opc, PRANGE(range));
+ in = req_capsule_client_get(tsi->tsi_pill, &RMF_FLD_MDFLD);
+ if (in == NULL)
+ RETURN(err_serious(-EPROTO));
- RETURN(rc);
+ rc = req_capsule_server_pack(tsi->tsi_pill);
+ if (unlikely(rc != 0))
+ RETURN(err_serious(rc));
-}
+ out = req_capsule_server_get(tsi->tsi_pill, &RMF_FLD_MDFLD);
+ if (out == NULL)
+ RETURN(err_serious(-EPROTO));
+ *out = *in;
-static int fld_req_handle(struct ptlrpc_request *req,
- struct fld_thread_info *info)
-{
- struct obd_export *exp = req->rq_export;
- struct lu_site *site = exp->exp_obd->obd_lu_dev->ld_site;
- struct lu_seq_range *in;
- struct lu_seq_range *out;
- int rc;
- __u32 *opc;
- ENTRY;
+ fld = lu_site2seq(site)->ss_server_fld;
- rc = req_capsule_server_pack(info->fti_pill);
- if (rc)
- RETURN(err_serious(rc));
-
- opc = req_capsule_client_get(info->fti_pill, &RMF_FLD_OPC);
- if (opc != NULL) {
- in = req_capsule_client_get(info->fti_pill, &RMF_FLD_MDFLD);
- if (in == NULL)
- RETURN(err_serious(-EPROTO));
- out = req_capsule_server_get(info->fti_pill, &RMF_FLD_MDFLD);
- if (out == NULL)
- RETURN(err_serious(-EPROTO));
- *out = *in;
-
- /* For old 2.0 client, the 'lsr_flags' is uninitialized.
- * Set it as 'LU_SEQ_RANGE_MDT' by default. */
- if (!(exp_connect_flags(exp) & OBD_CONNECT_64BITHASH) &&
- !(exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
- !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT) &&
- !exp->exp_libclient)
- fld_range_set_mdt(out);
-
- rc = fld_server_handle(lu_site2seq(site)->ss_server_fld,
- req->rq_svc_thread->t_env,
- *opc, out, info);
- } else {
- rc = err_serious(-EPROTO);
- }
+ rc = fld_server_lookup(tsi->tsi_env, fld, in->lsr_start, out);
+
+ CDEBUG(D_INFO, "%s: FLD req handle: error %d (range: "DRANGE")\n",
+ fld->lsf_name, rc, PRANGE(out));
RETURN(rc);
}
-static void fld_thread_info_init(struct ptlrpc_request *req,
- struct fld_thread_info *info)
+static int fld_handle_read(struct tgt_session_info *tsi)
{
- info->fti_pill = &req->rq_pill;
- /* Init request capsule. */
- req_capsule_init(info->fti_pill, req, RCL_SERVER);
- req_capsule_set(info->fti_pill, &RQF_FLD_QUERY);
-}
+ struct obd_export *exp = tsi->tsi_exp;
+ struct lu_site *site = exp->exp_obd->obd_lu_dev->ld_site;
+ struct lu_seq_range *in;
+ void *data;
+ int rc;
-static void fld_thread_info_fini(struct fld_thread_info *info)
-{
- req_capsule_fini(info->fti_pill);
-}
+ ENTRY;
-static int fld_handle(struct ptlrpc_request *req)
-{
- struct fld_thread_info *info;
- const struct lu_env *env;
- int rc;
+ req_capsule_set(tsi->tsi_pill, &RQF_FLD_READ);
- env = req->rq_svc_thread->t_env;
- LASSERT(env != NULL);
+ in = req_capsule_client_get(tsi->tsi_pill, &RMF_FLD_MDFLD);
+ if (in == NULL)
+ RETURN(err_serious(-EPROTO));
- info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
- LASSERT(info != NULL);
+ req_capsule_set_size(tsi->tsi_pill, &RMF_GENERIC_DATA, RCL_SERVER,
+ PAGE_CACHE_SIZE);
- fld_thread_info_init(req, info);
- rc = fld_req_handle(req, info);
- fld_thread_info_fini(info);
+ rc = req_capsule_server_pack(tsi->tsi_pill);
+ if (unlikely(rc != 0))
+ RETURN(err_serious(rc));
- return rc;
+ data = req_capsule_server_get(tsi->tsi_pill, &RMF_GENERIC_DATA);
+
+ rc = fld_server_read(tsi->tsi_env, lu_site2seq(site)->ss_server_fld,
+ in, data, PAGE_CACHE_SIZE);
+ RETURN(rc);
}
-/*
- * Entry point for handling FLD RPCs called from MDT.
- */
-int fld_query(struct com_thread_info *info)
+static int fld_handle_query(struct tgt_session_info *tsi)
{
- return fld_handle(info->cti_pill->rc_req);
+ int rc;
+
+ ENTRY;
+
+ req_capsule_set(tsi->tsi_pill, &RQF_FLD_QUERY);
+
+ rc = fld_handle_lookup(tsi);
+
+ RETURN(rc);
}
-EXPORT_SYMBOL(fld_query);
/*
* Returns true, if fid is local to this server node.
int rc = 0;
ENTRY;
- fld->lsf_proc_dir = lprocfs_register(fld->lsf_name,
- fld_type_proc_dir,
- fld_server_proc_list, fld);
- if (IS_ERR(fld->lsf_proc_dir)) {
- rc = PTR_ERR(fld->lsf_proc_dir);
- RETURN(rc);
- }
+ fld->lsf_proc_dir = lprocfs_seq_register(fld->lsf_name,
+ fld_type_proc_dir,
+ fld_server_proc_list, fld);
+ if (IS_ERR(fld->lsf_proc_dir)) {
+ rc = PTR_ERR(fld->lsf_proc_dir);
+ RETURN(rc);
+ }
rc = lprocfs_seq_create(fld->lsf_proc_dir, "fldb", 0444,
&fld_proc_seq_fops, fld);
#endif
int fld_server_init(const struct lu_env *env, struct lu_server_fld *fld,
- struct dt_device *dt, const char *prefix, int mds_node_id,
- int type)
+ struct dt_device *dt, const char *prefix, int type)
{
int cache_size, cache_threshold;
int rc;
+
ENTRY;
- snprintf(fld->lsf_name, sizeof(fld->lsf_name),
- "srv-%s", prefix);
+ snprintf(fld->lsf_name, sizeof(fld->lsf_name), "srv-%s", prefix);
- cache_size = FLD_SERVER_CACHE_SIZE /
- sizeof(struct fld_cache_entry);
+ cache_size = FLD_SERVER_CACHE_SIZE / sizeof(struct fld_cache_entry);
- cache_threshold = cache_size *
- FLD_SERVER_CACHE_THRESHOLD / 100;
+ cache_threshold = cache_size * FLD_SERVER_CACHE_THRESHOLD / 100;
mutex_init(&fld->lsf_lock);
- fld->lsf_cache = fld_cache_init(fld->lsf_name,
- cache_size, cache_threshold);
- if (IS_ERR(fld->lsf_cache)) {
- rc = PTR_ERR(fld->lsf_cache);
- fld->lsf_cache = NULL;
- GOTO(out, rc);
- }
-
- if (!mds_node_id && type == LU_SEQ_RANGE_MDT) {
- rc = fld_index_init(env, fld, dt);
- if (rc)
- GOTO(out, rc);
- } else {
- fld->lsf_obj = NULL;
+ fld->lsf_cache = fld_cache_init(fld->lsf_name, cache_size,
+ cache_threshold);
+ if (IS_ERR(fld->lsf_cache)) {
+ rc = PTR_ERR(fld->lsf_cache);
+ fld->lsf_cache = NULL;
+ RETURN(rc);
}
- rc = fld_server_proc_init(fld);
- if (rc)
- GOTO(out, rc);
+ rc = fld_index_init(env, fld, dt);
+ if (rc)
+ GOTO(out_cache, rc);
- fld->lsf_control_exp = NULL;
+ rc = fld_server_proc_init(fld);
+ if (rc)
+ GOTO(out_index, rc);
- GOTO(out, rc);
+ fld->lsf_control_exp = NULL;
-out:
- if (rc)
- fld_server_fini(env, fld);
+ RETURN(0);
+out_index:
+ fld_index_fini(env, fld);
+out_cache:
+ fld_cache_fini(fld->lsf_cache);
return rc;
}
EXPORT_SYMBOL(fld_server_init);
EXIT;
}
EXPORT_SYMBOL(fld_server_fini);
+
+struct tgt_handler fld_handlers[] = {
+TGT_FLD_HDL_VAR(0, FLD_QUERY, fld_handle_query),
+TGT_FLD_HDL_VAR(0, FLD_READ, fld_handle_read),
+};
+EXPORT_SYMBOL(fld_handlers);