-/* -*- 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.
* in the LICENSE file that accompanied this code).
*
* You should have received a copy of the GNU General Public License
- * version 2 along with this program; If not, see [sun.com URL with a
- * copy of GPLv2].
+ * 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
* 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, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
* Author: Yury Umanets <umka@clusterfs.com>
*/
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
#define DEBUG_SUBSYSTEM S_FID
#ifdef __KERNEL__
#include <lustre_fid.h>
#include "fid_internal.h"
+int client_fid_init(struct obd_device *obd,
+ struct obd_export *exp, enum lu_cli_type type)
+{
+ struct client_obd *cli = &obd->u.cli;
+ char *prefix;
+ int rc;
+ ENTRY;
+
+ OBD_ALLOC_PTR(cli->cl_seq);
+ if (cli->cl_seq == NULL)
+ RETURN(-ENOMEM);
+
+ OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
+ if (prefix == NULL)
+ GOTO(out_free_seq, rc = -ENOMEM);
+
+ snprintf(prefix, MAX_OBD_NAME + 5, "cli-%s", obd->obd_name);
+
+ /* Init client side sequence-manager */
+ rc = seq_client_init(cli->cl_seq, exp, type, prefix, NULL);
+ OBD_FREE(prefix, MAX_OBD_NAME + 5);
+ if (rc)
+ GOTO(out_free_seq, rc);
+
+ RETURN(rc);
+out_free_seq:
+ OBD_FREE_PTR(cli->cl_seq);
+ cli->cl_seq = NULL;
+ return rc;
+}
+EXPORT_SYMBOL(client_fid_init);
+
+int client_fid_fini(struct obd_device *obd)
+{
+ struct client_obd *cli = &obd->u.cli;
+ ENTRY;
+
+ if (cli->cl_seq != NULL) {
+ seq_client_fini(cli->cl_seq);
+ OBD_FREE_PTR(cli->cl_seq);
+ cli->cl_seq = NULL;
+ }
+
+ RETURN(0);
+}
+EXPORT_SYMBOL(client_fid_fini);
+
#ifdef __KERNEL__
+static void seq_server_proc_fini(struct lu_server_seq *seq);
+
/* Assigns client to sequence controller node. */
int seq_server_set_cli(struct lu_server_seq *seq,
struct lu_client_seq *cli,
* Ask client for new range, assign that range to ->seq_space and write
* seq state to backing store should be atomic.
*/
- down(&seq->lss_sem);
+ mutex_lock(&seq->lss_mutex);
if (cli == NULL) {
CDEBUG(D_INFO, "%s: Detached sequence client %s\n",
GOTO(out_up, rc = 0);
}
- if (seq->lss_cli != NULL) {
- CERROR("%s: Sequence controller is already "
- "assigned\n", seq->lss_name);
- GOTO(out_up, rc = -EINVAL);
- }
+ if (seq->lss_cli != NULL) {
+ CDEBUG(D_HA, "%s: Sequence controller is already "
+ "assigned\n", seq->lss_name);
+ GOTO(out_up, rc = -EEXIST);
+ }
CDEBUG(D_INFO, "%s: Attached sequence controller %s\n",
seq->lss_name, cli->lcs_name);
- seq->lss_cli = cli;
- EXIT;
+ seq->lss_cli = cli;
+ cli->lcs_space.lsr_index = seq->lss_site->ss_node_id;
+ EXIT;
out_up:
- up(&seq->lss_sem);
+ mutex_unlock(&seq->lss_mutex);
return rc;
}
EXPORT_SYMBOL(seq_server_set_cli);
-
/*
+ * allocate \a w units of sequence from range \a from.
+ */
+static inline void range_alloc(struct lu_seq_range *to,
+ struct lu_seq_range *from,
+ __u64 width)
+{
+ width = min(range_space(from), width);
+ to->lsr_start = from->lsr_start;
+ to->lsr_end = from->lsr_start + width;
+ from->lsr_start += width;
+}
+
+/**
* On controller node, allocate new super sequence for regular sequence server.
+ * As this super sequence controller, this node suppose to maintain fld
+ * and update index.
+ * \a out range always has currect mds node number of requester.
*/
+
static int __seq_server_alloc_super(struct lu_server_seq *seq,
- struct lu_range *in,
- struct lu_range *out,
+ struct lu_seq_range *out,
const struct lu_env *env)
{
- struct lu_range *space = &seq->lss_space;
- int rc;
- ENTRY;
-
- LASSERT(range_is_sane(space));
-
- if (in != NULL) {
- CDEBUG(D_INFO, "%s: Input seq range: "
- DRANGE"\n", seq->lss_name, PRANGE(in));
+ struct lu_seq_range *space = &seq->lss_space;
+ int rc;
+ ENTRY;
+
+ LASSERT(range_is_sane(space));
+
+ if (range_is_exhausted(space)) {
+ CERROR("%s: Sequences space is exhausted\n",
+ seq->lss_name);
+ RETURN(-ENOSPC);
+ } else {
+ range_alloc(out, space, seq->lss_width);
+ }
- if (in->lr_end > space->lr_start)
- space->lr_start = in->lr_end;
- *out = *in;
+ rc = seq_store_update(env, seq, out, 1 /* sync */);
- CDEBUG(D_INFO, "%s: Recovered space: "DRANGE"\n",
- seq->lss_name, PRANGE(space));
- } else {
- if (range_space(space) < seq->lss_width) {
- CWARN("%s: Sequences space to be exhausted soon. "
- "Only "LPU64" sequences left\n", seq->lss_name,
- range_space(space));
- *out = *space;
- space->lr_start = space->lr_end;
- } else if (range_is_exhausted(space)) {
- CERROR("%s: Sequences space is exhausted\n",
- seq->lss_name);
- RETURN(-ENOSPC);
- } else {
- range_alloc(out, space, seq->lss_width);
- }
- }
-
- rc = seq_store_write(seq, env);
- if (rc) {
- CERROR("%s: Can't write space data, rc %d\n",
- seq->lss_name, rc);
- RETURN(rc);
- }
+ LCONSOLE_INFO("%s: super-sequence allocation rc = %d " DRANGE"\n",
+ seq->lss_name, rc, PRANGE(out));
- CDEBUG(D_INFO, "%s: Allocated super-sequence "
- DRANGE"\n", seq->lss_name, PRANGE(out));
-
- RETURN(rc);
+ RETURN(rc);
}
int seq_server_alloc_super(struct lu_server_seq *seq,
- struct lu_range *in,
- struct lu_range *out,
+ struct lu_seq_range *out,
const struct lu_env *env)
{
int rc;
ENTRY;
- down(&seq->lss_sem);
- rc = __seq_server_alloc_super(seq, in, out, env);
- up(&seq->lss_sem);
+ mutex_lock(&seq->lss_mutex);
+ rc = __seq_server_alloc_super(seq, out, env);
+ mutex_unlock(&seq->lss_mutex);
RETURN(rc);
}
-static int __seq_server_alloc_meta(struct lu_server_seq *seq,
- struct lu_range *in,
- struct lu_range *out,
- const struct lu_env *env)
+static int __seq_set_init(const struct lu_env *env,
+ struct lu_server_seq *seq)
{
- struct lu_range *space = &seq->lss_space;
+ struct lu_seq_range *space = &seq->lss_space;
+ int rc;
+
+ range_alloc(&seq->lss_lowater_set, space, seq->lss_set_width);
+ range_alloc(&seq->lss_hiwater_set, space, seq->lss_set_width);
+
+ rc = seq_store_update(env, seq, NULL, 1);
+
+ return rc;
+}
+
+/*
+ * This function implements new seq allocation algorithm using async
+ * updates to seq file on disk. ref bug 18857 for details.
+ * there are four variable to keep track of this process
+ *
+ * lss_space; - available lss_space
+ * lss_lowater_set; - lu_seq_range for all seqs before barrier, i.e. safe to use
+ * lss_hiwater_set; - lu_seq_range after barrier, i.e. allocated but may be
+ * not yet committed
+ *
+ * when lss_lowater_set reaches the end it is replaced with hiwater one and
+ * a write operation is initiated to allocate new hiwater range.
+ * if last seq write opearion is still not commited, current operation is
+ * flaged as sync write op.
+ */
+static int range_alloc_set(const struct lu_env *env,
+ struct lu_seq_range *out,
+ struct lu_server_seq *seq)
+{
+ struct lu_seq_range *space = &seq->lss_space;
+ struct lu_seq_range *loset = &seq->lss_lowater_set;
+ struct lu_seq_range *hiset = &seq->lss_hiwater_set;
int rc = 0;
- ENTRY;
- LASSERT(range_is_sane(space));
+ if (range_is_zero(loset))
+ __seq_set_init(env, seq);
- /*
- * This is recovery case. Adjust super range if input range looks like
- * it is allocated from new super.
- */
- if (in != NULL) {
- CDEBUG(D_INFO, "%s: Input seq range: "
- DRANGE"\n", seq->lss_name, PRANGE(in));
-
- if (range_is_exhausted(space)) {
- /*
- * Server cannot send empty range to client, this is why
- * we check here that range from client is "newer" than
- * exhausted super.
- */
- LASSERT(in->lr_end > space->lr_start);
-
- /*
- * Start is set to end of last allocated, because it
- * *is* already allocated so we take that into account
- * and do not use for other allocations.
- */
- space->lr_start = in->lr_end;
-
- /*
- * End is set to in->lr_start + super sequence
- * allocation unit. That is because in->lr_start is
- * first seq in new allocated range from controller
- * before failure.
- */
- space->lr_end = in->lr_start + LUSTRE_SEQ_SUPER_WIDTH;
-
- if (!seq->lss_cli) {
- CERROR("%s: No sequence controller "
- "is attached.\n", seq->lss_name);
- RETURN(-ENODEV);
- }
-
- /*
- * Let controller know that this is recovery and last
- * obtained range from it was @space.
- */
- rc = seq_client_replay_super(seq->lss_cli, space, env);
- if (rc) {
- CERROR("%s: Can't replay super-sequence, "
- "rc %d\n", seq->lss_name, rc);
- RETURN(rc);
- }
- } else {
- /*
- * Update super start by end from client's range. Super
- * end should not be changed if range was not exhausted.
- */
- if (in->lr_end > space->lr_start)
- space->lr_start = in->lr_end;
- }
+ if (OBD_FAIL_CHECK(OBD_FAIL_SEQ_ALLOC)) /* exhaust set */
+ loset->lsr_start = loset->lsr_end;
- *out = *in;
+ if (range_is_exhausted(loset)) {
+ /* reached high water mark. */
+ struct lu_device *dev = seq->lss_site->ss_lu->ls_top_dev;
+ int obd_num_clients = dev->ld_obd->obd_num_exports;
+ __u64 set_sz;
- CDEBUG(D_INFO, "%s: Recovered space: "DRANGE"\n",
- seq->lss_name, PRANGE(space));
- } else {
- /*
- * XXX: Avoid cascading RPCs using kind of async preallocation
- * when meta-sequence is close to exhausting.
- */
- if (range_is_exhausted(space)) {
- if (!seq->lss_cli) {
- CERROR("%s: No sequence controller "
- "is attached.\n", seq->lss_name);
- RETURN(-ENODEV);
- }
-
- rc = seq_client_alloc_super(seq->lss_cli, env);
- if (rc) {
- CERROR("%s: Can't allocate super-sequence, "
- "rc %d\n", seq->lss_name, rc);
- RETURN(rc);
- }
-
- /* Saving new range to allocation space. */
- *space = seq->lss_cli->lcs_space;
- LASSERT(range_is_sane(space));
- }
+ /* calculate new seq width based on number of clients */
+ set_sz = max(seq->lss_set_width,
+ obd_num_clients * seq->lss_width);
+ set_sz = min(range_space(space), set_sz);
- range_alloc(out, space, seq->lss_width);
- }
+ /* Switch to hiwater range now */
+ *loset = *hiset;
+ /* allocate new hiwater range */
+ range_alloc(hiset, space, set_sz);
- rc = seq_store_write(seq, env);
- if (rc) {
- CERROR("%s: Can't write space data, rc %d\n",
- seq->lss_name, rc);
+ /* update ondisk seq with new *space */
+ rc = seq_store_update(env, seq, NULL, seq->lss_need_sync);
}
- if (rc == 0) {
- CDEBUG(D_INFO, "%s: Allocated meta-sequence "
- DRANGE"\n", seq->lss_name, PRANGE(out));
- }
+ LASSERTF(!range_is_exhausted(loset) || range_is_sane(loset),
+ DRANGE"\n", PRANGE(loset));
+
+ if (rc == 0)
+ range_alloc(out, loset, seq->lss_width);
RETURN(rc);
}
+static int __seq_server_alloc_meta(struct lu_server_seq *seq,
+ struct lu_seq_range *out,
+ const struct lu_env *env)
+{
+ struct lu_seq_range *space = &seq->lss_space;
+ int rc = 0;
+
+ ENTRY;
+
+ LASSERT(range_is_sane(space));
+
+ /* Check if available space ends and allocate new super seq */
+ if (range_is_exhausted(space)) {
+ if (!seq->lss_cli) {
+ CERROR("%s: No sequence controller is attached.\n",
+ seq->lss_name);
+ RETURN(-ENODEV);
+ }
+
+ rc = seq_client_alloc_super(seq->lss_cli, env);
+ if (rc) {
+ CERROR("%s: Can't allocate super-sequence, rc %d\n",
+ seq->lss_name, rc);
+ RETURN(rc);
+ }
+
+ /* Saving new range to allocation space. */
+ *space = seq->lss_cli->lcs_space;
+ LASSERT(range_is_sane(space));
+ }
+
+ rc = range_alloc_set(env, out, seq);
+ if (rc != 0) {
+ CERROR("%s: Allocated meta-sequence failed: rc = %d\n",
+ seq->lss_name, rc);
+ RETURN(rc);
+ }
+
+ CDEBUG(D_INFO, "%s: Allocated meta-sequence " DRANGE"\n",
+ seq->lss_name, PRANGE(out));
+
+ RETURN(rc);
+}
+
int seq_server_alloc_meta(struct lu_server_seq *seq,
- struct lu_range *in,
- struct lu_range *out,
+ struct lu_seq_range *out,
const struct lu_env *env)
{
int rc;
ENTRY;
- down(&seq->lss_sem);
- rc = __seq_server_alloc_meta(seq, in, out, env);
- up(&seq->lss_sem);
+ mutex_lock(&seq->lss_mutex);
+ rc = __seq_server_alloc_meta(seq, out, env);
+ mutex_unlock(&seq->lss_mutex);
RETURN(rc);
}
static int seq_server_handle(struct lu_site *site,
const struct lu_env *env,
- __u32 opc, struct lu_range *in,
- struct lu_range *out)
+ __u32 opc, struct lu_seq_range *out)
{
- int rc;
- ENTRY;
-
- switch (opc) {
- case SEQ_ALLOC_META:
- if (!site->ls_server_seq) {
- CERROR("Sequence server is not "
- "initialized\n");
- RETURN(-EINVAL);
- }
- rc = seq_server_alloc_meta(site->ls_server_seq,
- in, out, env);
- break;
- case SEQ_ALLOC_SUPER:
- if (!site->ls_control_seq) {
- CERROR("Sequence controller is not "
- "initialized\n");
- RETURN(-EINVAL);
- }
- rc = seq_server_alloc_super(site->ls_control_seq,
- in, out, env);
- break;
- default:
- rc = -EINVAL;
- break;
- }
+ int rc;
+ struct seq_server_site *ss_site;
+ ENTRY;
+
+ ss_site = lu_site2seq(site);
+
+ switch (opc) {
+ case SEQ_ALLOC_META:
+ if (!ss_site->ss_server_seq) {
+ CERROR("Sequence server is not "
+ "initialized\n");
+ RETURN(-EINVAL);
+ }
+ rc = seq_server_alloc_meta(ss_site->ss_server_seq, out, env);
+ break;
+ case SEQ_ALLOC_SUPER:
+ if (!ss_site->ss_control_seq) {
+ CERROR("Sequence controller is not "
+ "initialized\n");
+ RETURN(-EINVAL);
+ }
+ rc = seq_server_alloc_super(ss_site->ss_control_seq, out, env);
+ break;
+ default:
+ rc = -EINVAL;
+ break;
+ }
- RETURN(rc);
+ RETURN(rc);
}
static int seq_req_handle(struct ptlrpc_request *req,
const struct lu_env *env,
struct seq_thread_info *info)
{
- struct lu_range *out, *in = NULL;
+ struct lu_seq_range *out, *tmp;
struct lu_site *site;
int rc = -EPROTO;
__u32 *opc;
ENTRY;
+ LASSERT(!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY));
site = req->rq_export->exp_obd->obd_lu_dev->ld_site;
LASSERT(site != NULL);
-
+
rc = req_capsule_server_pack(info->sti_pill);
if (rc)
RETURN(err_serious(rc));
if (out == NULL)
RETURN(err_serious(-EPROTO));
- if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
- in = req_capsule_client_get(info->sti_pill,
- &RMF_SEQ_RANGE);
+ tmp = req_capsule_client_get(info->sti_pill, &RMF_SEQ_RANGE);
- LASSERT(!range_is_zero(in) && range_is_sane(in));
- }
+ /* seq client passed mdt id, we need to pass that using out
+ * range parameter */
- rc = seq_server_handle(site, env, *opc, in, out);
+ out->lsr_index = tmp->lsr_index;
+ out->lsr_flags = tmp->lsr_flags;
+ rc = seq_server_handle(site, env, *opc, out);
} else
rc = err_serious(-EPROTO);
LU_KEY_INIT_FINI(seq, struct seq_thread_info);
/* context key: seq_thread_key */
-LU_CONTEXT_KEY_DEFINE(seq, LCT_MD_THREAD);
+LU_CONTEXT_KEY_DEFINE(seq, LCT_MD_THREAD | LCT_DT_THREAD);
static void seq_thread_info_init(struct ptlrpc_request *req,
struct seq_thread_info *info)
req_capsule_fini(info->sti_pill);
}
-static int seq_handle(struct ptlrpc_request *req)
+int seq_handle(struct ptlrpc_request *req)
{
const struct lu_env *env;
struct seq_thread_info *info;
seq_thread_info_init(req, info);
rc = seq_req_handle(req, env, info);
+ /* XXX: we don't need replay but MDT assign transno in any case,
+ * remove it manually before reply*/
+ lustre_msg_set_transno(req->rq_repmsg, 0);
seq_thread_info_fini(info);
return rc;
}
+EXPORT_SYMBOL(seq_handle);
/*
* Entry point for handling FLD RPCs called from MDT.
}
EXPORT_SYMBOL(seq_query);
-static void seq_server_proc_fini(struct lu_server_seq *seq);
#ifdef LPROCFS
static int seq_server_proc_init(struct lu_server_seq *seq)
}
#endif
+
int seq_server_init(struct lu_server_seq *seq,
- struct dt_device *dev,
- const char *prefix,
- enum lu_mgr_type type,
- const struct lu_env *env)
+ struct dt_device *dev,
+ const char *prefix,
+ enum lu_mgr_type type,
+ struct seq_server_site *ss,
+ const struct lu_env *env)
{
int rc, is_srv = (type == LUSTRE_SEQ_SERVER);
ENTRY;
- LASSERT(dev != NULL);
+ LASSERT(dev != NULL);
LASSERT(prefix != NULL);
+ LASSERT(ss != NULL);
+ LASSERT(ss->ss_lu != NULL);
+
+ seq->lss_cli = NULL;
+ seq->lss_type = type;
+ seq->lss_site = ss;
+ range_init(&seq->lss_space);
- seq->lss_cli = NULL;
- seq->lss_type = type;
- range_zero(&seq->lss_space);
- sema_init(&seq->lss_sem, 1);
+ range_init(&seq->lss_lowater_set);
+ range_init(&seq->lss_hiwater_set);
+ seq->lss_set_width = LUSTRE_SEQ_BATCH_WIDTH;
+
+ mutex_init(&seq->lss_mutex);
seq->lss_width = is_srv ?
LUSTRE_SEQ_META_WIDTH : LUSTRE_SEQ_SUPER_WIDTH;
rc = seq_store_init(seq, env, dev);
if (rc)
GOTO(out, rc);
-
/* Request backing store for saved sequence info. */
rc = seq_store_read(seq, env);
if (rc == -ENODATA) {
LUSTRE_SEQ_ZERO_RANGE:
LUSTRE_SEQ_SPACE_RANGE;
- CDEBUG(D_INFO, "%s: No data found "
- "on store. Initialize space\n",
- seq->lss_name);
+ LASSERT(ss != NULL);
+ seq->lss_space.lsr_index = ss->ss_node_id;
+ LCONSOLE_INFO("%s: No data found "
+ "on store. Initialize space\n",
+ seq->lss_name);
- /* Save default controller value to store. */
- rc = seq_store_write(seq, env);
+ rc = seq_store_update(env, seq, NULL, 0);
if (rc) {
CERROR("%s: Can't write space data, "
"rc %d\n", seq->lss_name, rc);
}
} else if (rc) {
- CERROR("%s: Can't read space data, rc %d\n",
- seq->lss_name, rc);
- GOTO(out, rc);
- }
+ CERROR("%s: Can't read space data, rc %d\n",
+ seq->lss_name, rc);
+ GOTO(out, rc);
+ }
if (is_srv) {
LASSERT(range_is_sane(&seq->lss_space));
rc = seq_server_proc_init(seq);
if (rc)
- GOTO(out, rc);
+ GOTO(out, rc);
- EXIT;
+ EXIT;
out:
- if (rc)
- seq_server_fini(seq, env);
- return rc;
+ if (rc)
+ seq_server_fini(seq, env);
+ return rc;
}
EXPORT_SYMBOL(seq_server_init);
}
EXPORT_SYMBOL(seq_server_fini);
+int seq_site_fini(const struct lu_env *env, struct seq_server_site *ss)
+{
+ if (ss == NULL)
+ RETURN(0);
+
+ if (ss->ss_server_seq) {
+ seq_server_fini(ss->ss_server_seq, env);
+ OBD_FREE_PTR(ss->ss_server_seq);
+ ss->ss_server_seq = NULL;
+ }
+
+ if (ss->ss_control_seq) {
+ seq_server_fini(ss->ss_control_seq, env);
+ OBD_FREE_PTR(ss->ss_control_seq);
+ ss->ss_control_seq = NULL;
+ }
+
+ if (ss->ss_client_seq) {
+ seq_client_fini(ss->ss_client_seq);
+ OBD_FREE_PTR(ss->ss_client_seq);
+ ss->ss_client_seq = NULL;
+ }
+
+ RETURN(0);
+}
+EXPORT_SYMBOL(seq_site_fini);
+
cfs_proc_dir_entry_t *seq_type_proc_dir = NULL;
static int __init fid_mod_init(void)
git://git.whamcloud.com / fs / lustre-release.git / blobdiff