*
* You should have received a copy of the GNU General Public License
* version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
*
* GPL HEADER END
*/
* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, 2013, Intel Corporation.
+ * Copyright (c) 2011, 2015, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#define DEBUG_SUBSYSTEM S_FLD
-#ifdef __KERNEL__
-# include <libcfs/libcfs.h>
-# include <linux/module.h>
-# include <linux/jbd.h>
-# include <asm/div64.h>
-#else /* __KERNEL__ */
-# include <liblustre.h>
-# include <libcfs/list.h>
-#endif
+#include <libcfs/libcfs.h>
+#include <linux/module.h>
#include <obd.h>
-#include <obd_class.h>
-#include <lustre_ver.h>
#include <obd_support.h>
-#include <lprocfs_status.h>
-
-#include <md_object.h>
#include <lustre_fid.h>
+#include <lustre_fld.h>
#include <lustre_req_layout.h>
+#include <lprocfs_status.h>
#include "fld_internal.h"
-#include <lustre_fid.h>
-
-#ifdef __KERNEL__
/* context key constructor/destructor: fld_key_init, fld_key_fini */
LU_KEY_INIT_FINI(fld, struct fld_thread_info);
/* context key: fld_thread_key */
-LU_CONTEXT_KEY_DEFINE(fld, LCT_MD_THREAD|LCT_DT_THREAD);
-
-cfs_proc_dir_entry_t *fld_type_proc_dir = NULL;
+/* MGS thread may create llog file causing FLD lookup */
+LU_CONTEXT_KEY_DEFINE(fld, LCT_MD_THREAD | LCT_DT_THREAD | LCT_MG_THREAD);
-static int __init fld_mod_init(void)
+int fld_server_mod_init(void)
{
- fld_type_proc_dir = lprocfs_register(LUSTRE_FLD_NAME,
- proc_lustre_root,
- NULL, NULL);
- if (IS_ERR(fld_type_proc_dir))
- return PTR_ERR(fld_type_proc_dir);
-
- LU_CONTEXT_KEY_INIT(&fld_thread_key);
- lu_context_key_register(&fld_thread_key);
- return 0;
+ LU_CONTEXT_KEY_INIT(&fld_thread_key);
+ return lu_context_key_register(&fld_thread_key);
}
-static void __exit fld_mod_exit(void)
+void fld_server_mod_exit(void)
{
- lu_context_key_degister(&fld_thread_key);
- if (fld_type_proc_dir != NULL && !IS_ERR(fld_type_proc_dir)) {
- lprocfs_remove(&fld_type_proc_dir);
- fld_type_proc_dir = NULL;
- }
+ lu_context_key_degister(&fld_thread_key);
}
int fld_declare_server_create(const struct lu_env *env,
struct lu_server_fld *fld,
- struct lu_seq_range *range,
+ const struct lu_seq_range *range,
struct thandle *th)
{
int rc;
* is granted to a server.
*/
int fld_server_create(const struct lu_env *env, struct lu_server_fld *fld,
- struct lu_seq_range *range, struct thandle *th)
+ const struct lu_seq_range *range, struct thandle *th)
{
int rc;
EXPORT_SYMBOL(fld_server_create);
/**
- * Lookup mds by seq, returns a range for given seq.
+ * 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)
+ 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)
+ 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);
+
+ mutex_lock(&fld->lsf_lock);
+ rc1 = fld_insert_entry(env, fld, &lsra->lsra_lsr[i]);
+ mutex_unlock(&fld->lsf_lock);
+
+ 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)
+ 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,
+ u64 seq, struct lu_seq_range *range)
+{
+ struct lu_seq_range *erange;
+ struct fld_thread_info *info;
+ int rc;
+
+ ENTRY;
+
+ info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
+ LASSERT(info != NULL);
+ erange = &info->fti_lrange;
+
+ /* Lookup it in the cache. */
+ rc = fld_cache_lookup(fld->lsf_cache, seq, erange);
+ if (rc == 0) {
+ if (unlikely(fld_range_type(erange) != fld_range_type(range) &&
+ !fld_range_is_any(range))) {
+ CERROR("%s: FLD cache range "DRANGE" does not match requested flag %x: rc = %d\n",
+ fld->lsf_name, PRANGE(erange), range->lsr_flags,
+ -EIO);
+ RETURN(-EIO);
+ }
+ *range = *erange;
+ RETURN(0);
+ }
+ RETURN(rc);
+}
+EXPORT_SYMBOL(fld_local_lookup);
+
+/**
+ * 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)
+ u64 seq, struct lu_seq_range *range)
{
- struct lu_seq_range *erange;
- struct fld_thread_info *info;
- int rc;
- ENTRY;
-
- info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
- erange = &info->fti_lrange;
-
- /* Lookup it in the cache. */
- rc = fld_cache_lookup(fld->lsf_cache, seq, erange);
- if (rc == 0) {
- if (unlikely(erange->lsr_flags != range->lsr_flags) &&
- range->lsr_flags != -1) {
- CERROR("FLD cache found a range "DRANGE" doesn't "
- "match the requested flag %x\n",
- PRANGE(erange), range->lsr_flags);
- RETURN(-EIO);
- }
- *range = *erange;
- RETURN(0);
- }
-
- if (fld->lsf_obj) {
- /* On server side, all entries should be in cache.
- * If we can not find it in cache, just return error */
- CERROR("%s: Can not found the seq "LPX64"\n",
- fld->lsf_name, seq);
- RETURN(-EIO);
+ 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);
+
+ if (index == 0 && rc == LDD_F_SV_TYPE_MDT) {
+ /*
+ * 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 %#llx: rc = %d\n",
+ fld->lsf_name, seq, -ENOENT);
+ RETURN(-ENOENT);
} else {
- LASSERT(fld->lsf_control_exp);
- /* send request to mdt0 i.e. super seq. controller.
+ if (!fld->lsf_control_exp) {
+ CERROR("%s: lookup %#llx, 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)
+ 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)
+ 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;
-
- 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.
- * Old 2.0 liblustre client cannot talk with new 2.1 server. */
- if (!(exp->exp_connect_flags & OBD_CONNECT_64BITHASH) &&
- !(exp->exp_connect_flags & OBD_CONNECT_MDS_MDS) &&
- !(exp->exp_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
- !exp->exp_libclient)
- out->lsr_flags = LU_SEQ_RANGE_MDT;
-
- rc = fld_server_handle(lu_site2seq(site)->ss_server_fld,
- req->rq_svc_thread->t_env,
- *opc, out, info);
- } else
- rc = err_serious(-EPROTO);
-
- RETURN(rc);
-}
+ fld = lu_site2seq(site)->ss_server_fld;
-static void fld_thread_info_init(struct ptlrpc_request *req,
- struct fld_thread_info *info)
-{
- 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);
-}
+ rc = fld_server_lookup(tsi->tsi_env, fld, in->lsr_start, out);
-static void fld_thread_info_fini(struct fld_thread_info *info)
-{
- req_capsule_fini(info->fti_pill);
+ CDEBUG(D_INFO, "%s: FLD req handle: error %d (range: "DRANGE")\n",
+ fld->lsf_name, rc, PRANGE(out));
+
+ RETURN(rc);
}
-static int fld_handle(struct ptlrpc_request *req)
+static int fld_handle_read(struct tgt_session_info *tsi)
{
- struct fld_thread_info *info;
- const struct lu_env *env;
- int rc;
+ 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;
- env = req->rq_svc_thread->t_env;
- LASSERT(env != NULL);
+ ENTRY;
- info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
- LASSERT(info != NULL);
+ req_capsule_set(tsi->tsi_pill, &RQF_FLD_READ);
- fld_thread_info_init(req, info);
- rc = fld_req_handle(req, info);
- fld_thread_info_fini(info);
+ in = req_capsule_client_get(tsi->tsi_pill, &RMF_FLD_MDFLD);
+ if (!in)
+ RETURN(err_serious(-EPROTO));
- return rc;
+ req_capsule_set_size(tsi->tsi_pill, &RMF_GENERIC_DATA, RCL_SERVER,
+ PAGE_SIZE);
+
+ rc = req_capsule_server_pack(tsi->tsi_pill);
+ if (unlikely(rc != 0))
+ RETURN(err_serious(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_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.
* fid_is_local() is supposed to be used in assertion checks only.
*/
int fid_is_local(const struct lu_env *env,
- struct lu_site *site, const struct lu_fid *fid)
+ struct lu_site *site, const struct lu_fid *fid)
{
int result;
struct seq_server_site *ss_site;
struct lu_seq_range *range;
struct fld_thread_info *info;
+
ENTRY;
info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
result = 1; /* conservatively assume fid is local */
ss_site = lu_site2seq(site);
- if (ss_site->ss_client_fld != NULL) {
+ if (ss_site->ss_client_fld) {
int rc;
rc = fld_cache_lookup(ss_site->ss_client_fld->lcf_cache,
}
EXPORT_SYMBOL(fid_is_local);
-static void fld_server_proc_fini(struct lu_server_fld *fld);
-
-#ifdef LPROCFS
-static int fld_server_proc_init(struct lu_server_fld *fld)
+static void fld_server_debugfs_fini(struct lu_server_fld *fld)
{
- 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);
- }
-
- rc = lprocfs_seq_create(fld->lsf_proc_dir, "fldb", 0444,
- &fld_proc_seq_fops, fld);
- if (rc) {
- lprocfs_remove(&fld->lsf_proc_dir);
- fld->lsf_proc_dir = NULL;
- }
-
- RETURN(rc);
+ debugfs_remove_recursive(fld->lsf_debugfs_entry);
}
-static void fld_server_proc_fini(struct lu_server_fld *fld)
+static int fld_server_debugfs_init(struct lu_server_fld *fld)
{
- ENTRY;
- if (fld->lsf_proc_dir != NULL) {
- if (!IS_ERR(fld->lsf_proc_dir))
- lprocfs_remove(&fld->lsf_proc_dir);
- fld->lsf_proc_dir = NULL;
- }
- EXIT;
-}
-#else
-static int fld_server_proc_init(struct lu_server_fld *fld)
-{
- return 0;
-}
+ int rc = 0;
-static void fld_server_proc_fini(struct lu_server_fld *fld)
-{
- return;
+ ENTRY;
+ fld->lsf_debugfs_entry = ldebugfs_register(fld->lsf_name,
+ fld_debugfs_dir,
+ NULL, NULL);
+ if (IS_ERR_OR_NULL(fld->lsf_debugfs_entry)) {
+ rc = fld->lsf_debugfs_entry ? PTR_ERR(fld->lsf_debugfs_entry)
+ : -ENOMEM;
+ fld->lsf_debugfs_entry = NULL;
+ RETURN(rc);
+ }
+
+ rc = ldebugfs_seq_create(fld->lsf_debugfs_entry, "fldb", 0444,
+ &fld_debugfs_seq_fops, fld);
+ if (rc)
+ debugfs_remove_recursive(fld->lsf_debugfs_entry);
+
+ RETURN(rc);
}
-#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,
- __u32 lsr_flags)
+ struct dt_device *dt, const char *prefix, int type)
{
- int cache_size, cache_threshold;
- struct lu_seq_range range;
- int rc;
- ENTRY;
+ 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 && lsr_flags == LU_SEQ_RANGE_MDT) {
- rc = fld_index_init(env, fld, dt);
- if (rc)
- GOTO(out, rc);
- } else
- fld->lsf_obj = NULL;
-
- rc = fld_server_proc_init(fld);
- if (rc)
- GOTO(out, rc);
-
- fld->lsf_control_exp = NULL;
-
- if (lsr_flags == LU_SEQ_RANGE_MDT) {
- /* Insert reserved sequence of "ROOT" and ".lustre"
- * into fld cache. */
- range.lsr_start = FID_SEQ_LOCAL_FILE;
- range.lsr_end = FID_SEQ_DOT_LUSTRE + 1;
- range.lsr_index = 0;
- range.lsr_flags = lsr_flags;
- fld_cache_insert(fld->lsf_cache, &range);
+ 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);
}
- EXIT;
-out:
+
+ rc = fld_index_init(env, fld, dt, type);
if (rc)
- fld_server_fini(env, fld);
+ GOTO(out_cache, rc);
+
+ rc = fld_server_debugfs_init(fld);
+ if (rc)
+ GOTO(out_index, rc);
+
+ fld->lsf_control_exp = NULL;
+ fld->lsf_seq_lookup = fld_server_lookup;
+
+ fld->lsf_seq_lookup = fld_server_lookup;
+ RETURN(0);
+out_index:
+ fld_index_fini(env, fld);
+out_cache:
+ fld_cache_fini(fld->lsf_cache);
return rc;
}
EXPORT_SYMBOL(fld_server_init);
{
ENTRY;
- fld_server_proc_fini(fld);
+ fld_server_debugfs_fini(fld);
fld_index_fini(env, fld);
- if (fld->lsf_cache != NULL) {
+ if (fld->lsf_cache) {
if (!IS_ERR(fld->lsf_cache))
fld_cache_fini(fld->lsf_cache);
fld->lsf_cache = NULL;
}
EXPORT_SYMBOL(fld_server_fini);
-MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
-MODULE_DESCRIPTION("Lustre FLD");
-MODULE_LICENSE("GPL");
-
-cfs_module(mdd, "0.1.0", fld_mod_init, fld_mod_exit);
-#endif
+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);