-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
* GPL HEADER START
*
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
* GPL HEADER END
*/
/*
- * Copyright 2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
+ *
+ * Copyright (c) 2011, 2012, Whamcloud, Inc.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
*
* Author: Yury Umanets <umka@clusterfs.com>
* Author: WangDi <wangdi@clusterfs.com>
+ * Author: Pravin Shelar <pravin.shelar@sun.com>
*/
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
#define DEBUG_SUBSYSTEM S_FLD
#ifdef __KERNEL__
#include <lprocfs_status.h>
#include <md_object.h>
+#include <lustre_fid.h>
#include <lustre_req_layout.h>
#include "fld_internal.h"
+#include <lustre_fid.h>
#ifdef __KERNEL__
cfs_proc_dir_entry_t *fld_type_proc_dir = NULL;
+static struct lu_local_obj_desc llod_fld_index = {
+ .llod_name = fld_index_name,
+ .llod_oid = FLD_INDEX_OID,
+ .llod_is_index = 1,
+ .llod_feat = &fld_index_features,
+};
+
static int __init fld_mod_init(void)
{
fld_type_proc_dir = lprocfs_register(LUSTRE_FLD_NAME,
if (IS_ERR(fld_type_proc_dir))
return PTR_ERR(fld_type_proc_dir);
+ llo_local_obj_register(&llod_fld_index);
+
LU_CONTEXT_KEY_INIT(&fld_thread_key);
lu_context_key_register(&fld_thread_key);
return 0;
static void __exit fld_mod_exit(void)
{
+ llo_local_obj_unregister(&llod_fld_index);
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);
}
}
-/* Insert index entry and update cache. */
-int fld_server_create(struct lu_server_fld *fld,
- const struct lu_env *env,
- seqno_t seq, mdsno_t mds)
+int fld_declare_server_create(struct lu_server_fld *fld,
+ const struct lu_env *env,
+ struct thandle *th)
{
+ struct dt_object *dt_obj = fld->lsf_obj;
int rc;
+
ENTRY;
-
- rc = fld_index_create(fld, env, seq, mds);
-
- if (rc == 0) {
- /*
- * Do not return result of calling fld_cache_insert()
- * here. First of all because it may return -EEXISTS. Another
- * reason is that, we do not want to stop proceeding even after
- * cache errors.
- */
- fld_cache_insert(fld->lsf_cache, seq, mds);
- }
+ if (fld->lsf_no_range_lookup) {
+ /* Stub for underlying FS which can't lookup ranges */
+ return 0;
+ }
+
+ /* for ldiskfs OSD it's enough to declare operation with any ops
+ * with DMU we'll probably need to specify exact key/value */
+ rc = dt_obj->do_index_ops->dio_declare_delete(env, dt_obj, NULL, th);
+ if (rc)
+ GOTO(out, rc);
+ rc = dt_obj->do_index_ops->dio_declare_delete(env, dt_obj, NULL, th);
+ if (rc)
+ GOTO(out, rc);
+ rc = dt_obj->do_index_ops->dio_declare_insert(env, dt_obj,
+ NULL, NULL, th);
+out:
RETURN(rc);
}
-EXPORT_SYMBOL(fld_server_create);
+EXPORT_SYMBOL(fld_declare_server_create);
+
+/**
+ * Insert FLD index entry and update FLD cache.
+ *
+ * First it try to merge given range with existing range then update
+ * FLD index and FLD cache accordingly. FLD index consistency is maintained
+ * by this function.
+ * This function is called from the sequence allocator when a super-sequence
+ * is granted to a server.
+ */
-/* Delete index entry. */
-int fld_server_delete(struct lu_server_fld *fld,
+int fld_server_create(struct lu_server_fld *fld,
const struct lu_env *env,
- seqno_t seq)
+ struct lu_seq_range *add_range,
+ struct thandle *th)
{
- int rc;
+ struct lu_seq_range *erange;
+ struct lu_seq_range *new;
+ struct fld_thread_info *info;
+ int rc = 0;
+ int do_merge=0;
+
ENTRY;
- fld_cache_delete(fld->lsf_cache, seq);
- rc = fld_index_delete(fld, env, seq);
-
+ info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
+ cfs_mutex_lock(&fld->lsf_lock);
+
+ erange = &info->fti_lrange;
+ new = &info->fti_irange;
+ *new = *add_range;
+
+ /* STEP 1: try to merge with previous range */
+ rc = fld_index_lookup(fld, env, new->lsr_start, erange);
+ if (rc == 0) {
+ /* in case of range overlap, the location must be same */
+ if (range_compare_loc(new, erange) != 0) {
+ CERROR("the start of given range "DRANGE" conflict to"
+ "an existing range "DRANGE"\n",
+ PRANGE(new), PRANGE(erange));
+ GOTO(out, rc = -EIO);
+ }
+
+ if (new->lsr_end < erange->lsr_end)
+ GOTO(out, rc);
+ do_merge = 1;
+ } else if (rc == -ENOENT) {
+ /* check for merge case: optimizes for single mds lustre.
+ * As entry does not exist, returned entry must be left side
+ * entry compared to start of new range (ref dio_lookup()).
+ * So try to merge from left.
+ */
+ if (new->lsr_start == erange->lsr_end &&
+ range_compare_loc(new, erange) == 0)
+ do_merge = 1;
+ } else {
+ /* no overlap allowed in fld, so failure in lookup is error */
+ GOTO(out, rc);
+ }
+
+ if (do_merge) {
+ /* new range will be merged with the existing one.
+ * delete this range at first. */
+ rc = fld_index_delete(fld, env, erange, th);
+ if (rc != 0)
+ GOTO(out, rc);
+
+ new->lsr_start = min(erange->lsr_start, new->lsr_start);
+ new->lsr_end = max(erange->lsr_end, new->lsr_end);
+ do_merge = 0;
+ }
+
+ /* STEP 2: try to merge with next range */
+ rc = fld_index_lookup(fld, env, new->lsr_end, erange);
+ if (rc == 0) {
+ /* found a matched range, meaning we're either
+ * overlapping or ajacent, must merge with it. */
+ do_merge = 1;
+ } else if (rc == -ENOENT) {
+ /* this range is left of new range end point */
+ LASSERT(erange->lsr_end <= new->lsr_end);
+ /*
+ * the found left range must be either:
+ * 1. withing new range.
+ * 2. left of new range (no overlapping).
+ * because if they're partly overlapping, the STEP 1 must have
+ * been removed this range.
+ */
+ LASSERTF(erange->lsr_start > new->lsr_start ||
+ erange->lsr_end < new->lsr_start ||
+ (erange->lsr_end == new->lsr_start &&
+ range_compare_loc(new, erange) != 0),
+ "left "DRANGE", new "DRANGE"\n",
+ PRANGE(erange), PRANGE(new));
+
+ /* if it's within the new range, merge it */
+ if (erange->lsr_start > new->lsr_start)
+ do_merge = 1;
+ } else {
+ GOTO(out, rc);
+ }
+
+ if (do_merge) {
+ if (range_compare_loc(new, erange) != 0) {
+ CERROR("the end of given range "DRANGE" overlaps "
+ "with an existing range "DRANGE"\n",
+ PRANGE(new), PRANGE(erange));
+ GOTO(out, rc = -EIO);
+ }
+
+ /* merge with next range */
+ rc = fld_index_delete(fld, env, erange, th);
+ if (rc != 0)
+ GOTO(out, rc);
+
+ new->lsr_start = min(erange->lsr_start, new->lsr_start);
+ new->lsr_end = max(erange->lsr_end, new->lsr_end);
+ }
+
+ /* now update fld entry. */
+ rc = fld_index_create(fld, env, new, th);
+
+ LASSERT(rc != -EEXIST);
+out:
+ if (rc == 0)
+ fld_cache_insert(fld->lsf_cache, new);
+
+ cfs_mutex_unlock(&fld->lsf_lock);
+
+ CDEBUG((rc != 0 ? D_ERROR : D_INFO),
+ "%s: FLD create: given range : "DRANGE
+ "after merge "DRANGE" rc = %d \n", fld->lsf_name,
+ PRANGE(add_range), PRANGE(new), rc);
+
RETURN(rc);
}
-EXPORT_SYMBOL(fld_server_delete);
+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.
+ */
-/* Lookup mds by seq. */
int fld_server_lookup(struct lu_server_fld *fld,
const struct lu_env *env,
- seqno_t seq, mdsno_t *mds)
+ seqno_t 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, mds);
- if (rc == 0)
+ rc = fld_cache_lookup(fld->lsf_cache, seq, erange);
+ if (rc == 0) {
+ if (unlikely(erange->lsr_flags != range->lsr_flags)) {
+ 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);
+ }
- rc = fld_index_lookup(fld, env, seq, mds);
- if (rc == 0) {
- /*
- * Do not return error here as well. See previous comment in
- * same situation in function fld_server_create().
+ if (fld->lsf_obj) {
+ rc = fld_index_lookup(fld, env, seq, erange);
+ if (rc == 0) {
+ if (unlikely(erange->lsr_flags != range->lsr_flags)) {
+ CERROR("FLD found a range "DRANGE" doesn't "
+ "match the requested flag %x\n",
+ PRANGE(erange), range->lsr_flags);
+ RETURN(-EIO);
+ }
+ *range = *erange;
+ }
+ } else {
+ LASSERT(fld->lsf_control_exp);
+ /* send request to mdt0 i.e. super seq. controller.
+ * This is temporary solution, long term solution is fld
+ * replication on all mdt servers.
*/
- fld_cache_insert(fld->lsf_cache, seq, *mds);
+ rc = fld_client_rpc(fld->lsf_control_exp,
+ range, FLD_LOOKUP);
}
+
+ if (rc == 0)
+ fld_cache_insert(fld->lsf_cache, range);
+
RETURN(rc);
}
EXPORT_SYMBOL(fld_server_lookup);
+/**
+ * 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 md_fld *mf,
+ __u32 opc, struct lu_seq_range *range,
struct fld_thread_info *info)
{
int rc;
ENTRY;
switch (opc) {
- case FLD_CREATE:
- rc = fld_server_create(fld, env,
- mf->mf_seq, mf->mf_mds);
-
- /* Do not return -EEXIST error for resent case */
- if ((info->fti_flags & MSG_RESENT) && rc == -EEXIST)
- rc = 0;
- break;
- case FLD_DELETE:
- rc = fld_server_delete(fld, env, mf->mf_seq);
-
- /* Do not return -ENOENT error for resent case */
- if ((info->fti_flags & MSG_RESENT) && rc == -ENOENT)
- rc = 0;
- break;
case FLD_LOOKUP:
rc = fld_server_lookup(fld, env,
- mf->mf_seq, &mf->mf_mds);
+ range->lsr_start, range);
break;
default:
rc = -EINVAL;
break;
}
- CDEBUG(D_INFO, "%s: FLD req handle: error %d (opc: %d, seq: "
- LPX64", mds: "LPU64")\n", fld->lsf_name, rc, opc,
- mf->mf_seq, mf->mf_mds);
-
+ CDEBUG(D_INFO, "%s: FLD req handle: error %d (opc: %d, range: "
+ DRANGE"\n", fld->lsf_name, rc, opc, PRANGE(range));
+
RETURN(rc);
}
static int fld_req_handle(struct ptlrpc_request *req,
struct fld_thread_info *info)
{
- struct lu_site *site;
- struct md_fld *in;
- struct md_fld *out;
+ 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;
- site = req->rq_export->exp_obd->obd_lu_dev->ld_site;
-
rc = req_capsule_server_pack(info->fti_pill);
if (rc)
RETURN(err_serious(rc));
RETURN(err_serious(-EPROTO));
*out = *in;
- rc = fld_server_handle(site->ls_server_fld,
+ /* 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_libclient)
+ out->lsr_flags = LU_SEQ_RANGE_MDT;
+
+ rc = fld_server_handle(lu_site2md(site)->ms_server_fld,
req->rq_svc_thread->t_env,
*opc, out, info);
} else
static void fld_thread_info_init(struct ptlrpc_request *req,
struct fld_thread_info *info)
{
- info->fti_flags = lustre_msg_get_flags(req->rq_reqmsg);
-
info->fti_pill = &req->rq_pill;
/* Init request capsule. */
req_capsule_init(info->fti_pill, req, RCL_SERVER);
*
* fid_is_local() is supposed to be used in assertion checks only.
*/
-int fid_is_local(struct lu_site *site, const struct lu_fid *fid)
+int fid_is_local(const struct lu_env *env,
+ struct lu_site *site, const struct lu_fid *fid)
{
int result;
+ struct md_site *msite;
+ struct lu_seq_range *range;
+ struct fld_thread_info *info;
+ ENTRY;
+
+ info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
+ range = &info->fti_lrange;
result = 1; /* conservatively assume fid is local */
- if (site->ls_client_fld != NULL) {
- mdsno_t mds;
+ msite = lu_site2md(site);
+ if (msite->ms_client_fld != NULL) {
int rc;
- rc = fld_cache_lookup(site->ls_client_fld->lcf_cache,
- fid_seq(fid), &mds);
+ rc = fld_cache_lookup(msite->ms_client_fld->lcf_cache,
+ fid_seq(fid), range);
if (rc == 0)
- result = (mds == site->ls_node_id);
+ result = (range->lsr_index == msite->ms_node_id);
}
return result;
}
RETURN(rc);
}
- 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);
}
static void fld_server_proc_fini(struct lu_server_fld *fld)
#endif
int fld_server_init(struct lu_server_fld *fld, struct dt_device *dt,
- const char *prefix, const struct lu_env *env)
+ const char *prefix, const struct lu_env *env,
+ int mds_node_id)
{
int cache_size, cache_threshold;
+ struct lu_seq_range range;
int rc;
ENTRY;
cache_threshold = cache_size *
FLD_SERVER_CACHE_THRESHOLD / 100;
+ cfs_mutex_init(&fld->lsf_lock);
fld->lsf_cache = fld_cache_init(fld->lsf_name,
- FLD_SERVER_HTABLE_SIZE,
cache_size, cache_threshold);
if (IS_ERR(fld->lsf_cache)) {
rc = PTR_ERR(fld->lsf_cache);
GOTO(out, rc);
}
- rc = fld_index_init(fld, env, dt);
- if (rc)
- GOTO(out, rc);
+ if (!mds_node_id) {
+ rc = fld_index_init(fld, env, 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;
+
+ /* Insert reserved sequence number of ".lustre" into fld cache. */
+ range.lsr_start = FID_SEQ_DOT_LUSTRE;
+ range.lsr_end = FID_SEQ_DOT_LUSTRE + 1;
+ range.lsr_index = 0;
+ range.lsr_flags = LU_SEQ_RANGE_MDT;
+ fld_cache_insert(fld->lsf_cache, &range);
+
EXIT;
out:
if (rc)
git://git.whamcloud.com / fs / lustre-release.git / blobdiff