-/* -*- 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) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
+ *
+ * Copyright (c) 2011, 2013, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
* Lustre is a trademark of Sun Microsystems, Inc.
*/
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
#define DEBUG_SUBSYSTEM S_LMV
#ifdef __KERNEL__
#include <linux/slab.h>
#include <asm/div64.h>
#include <linux/seq_file.h>
#include <linux/namei.h>
+#include <asm/uaccess.h>
#else
#include <liblustre.h>
#endif
#include <lustre/lustre_idl.h>
-#include <lustre_log.h>
#include <obd_support.h>
#include <lustre_lib.h>
#include <lustre_net.h>
#include <lustre_fid.h>
#include "lmv_internal.h"
-/* object cache. */
-cfs_mem_cache_t *obj_cache;
-atomic_t obj_cache_count = ATOMIC_INIT(0);
-
static void lmv_activate_target(struct lmv_obd *lmv,
struct lmv_tgt_desc *tgt,
int activate)
lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
}
-/* Error codes:
+/**
+ * Error codes:
*
* -EINVAL : UUID can't be found in the LMV's target list
* -ENOTCONN: The UUID is found, but the target connection is bad (!)
static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid,
int activate)
{
- struct lmv_tgt_desc *tgt;
- struct obd_device *obd;
- int i, rc = 0;
+ struct lmv_tgt_desc *tgt;
+ struct obd_device *obd;
+ int i;
+ int rc = 0;
ENTRY;
CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
lmv, uuid->uuid, activate);
- spin_lock(&lmv->lmv_lock);
- for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
- if (tgt->ltd_exp == NULL)
- continue;
+ spin_lock(&lmv->lmv_lock);
+ for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
+ tgt = lmv->tgts[i];
+ if (tgt == NULL || tgt->ltd_exp == NULL)
+ continue;
- CDEBUG(D_INFO, "lmv idx %d is %s conn "LPX64"\n",
- i, tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
+ CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
+ tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
- if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
- break;
- }
+ if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
+ break;
+ }
if (i == lmv->desc.ld_tgt_count)
GOTO(out_lmv_lock, rc = -EINVAL);
GOTO(out_lmv_lock, rc);
}
- CDEBUG(D_INFO, "Marking OBD %p %sactive\n",
- obd, activate ? "" : "in");
-
+ CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
+ activate ? "" : "in");
lmv_activate_target(lmv, tgt, activate);
-
EXIT;
out_lmv_lock:
- spin_unlock(&lmv->lmv_lock);
- return rc;
+ spin_unlock(&lmv->lmv_lock);
+ return rc;
}
-static int lmv_set_mdc_data(struct lmv_obd *lmv, struct obd_uuid *uuid,
- struct obd_connect_data *data)
+struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
{
- struct lmv_tgt_desc *tgt;
- int i;
- ENTRY;
-
- LASSERT(data != NULL);
+ struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
- spin_lock(&lmv->lmv_lock);
- for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
- if (tgt->ltd_exp == NULL)
- continue;
-
- if (obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
- lmv->datas[tgt->ltd_idx] = *data;
- break;
- }
- }
- spin_unlock(&lmv->lmv_lock);
- RETURN(0);
+ return obd_get_uuid(lmv->tgts[0]->ltd_exp);
}
static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
enum obd_notify_event ev, void *data)
{
- struct lmv_obd *lmv = &obd->u.lmv;
- struct obd_uuid *uuid;
- int rc = 0;
+ struct obd_connect_data *conn_data;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct obd_uuid *uuid;
+ int rc = 0;
ENTRY;
if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
uuid->uuid, rc);
RETURN(rc);
}
- } else if (ev == OBD_NOTIFY_OCD) {
- struct obd_connect_data *conn_data =
- &watched->u.cli.cl_import->imp_connect_data;
-
- /* Set connect data to desired target, update
- * exp_connect_flags. */
- rc = lmv_set_mdc_data(lmv, uuid, conn_data);
- if (rc) {
- CERROR("can't set connect data to target %s, rc %d\n",
- uuid->uuid, rc);
- RETURN(rc);
- }
-
- /*
- * XXX: Make sure that ocd_connect_flags from all targets are
- * the same. Otherwise one of MDTs runs wrong version or
- * something like this. --umka
- */
- obd->obd_self_export->exp_connect_flags =
- conn_data->ocd_connect_flags;
- }
+ } else if (ev == OBD_NOTIFY_OCD) {
+ conn_data = &watched->u.cli.cl_import->imp_connect_data;
+ /*
+ * XXX: Make sure that ocd_connect_flags from all targets are
+ * the same. Otherwise one of MDTs runs wrong version or
+ * something like this. --umka
+ */
+ obd->obd_self_export->exp_connect_data = *conn_data;
+ }
#if 0
else if (ev == OBD_NOTIFY_DISCON) {
- /* For disconnect event, flush fld cache for failout MDS case. */
+ /*
+ * For disconnect event, flush fld cache for failout MDS case.
+ */
fld_client_flush(&lmv->lmv_fld);
}
#endif
- /* Pass the notification up the chain. */
+ /*
+ * Pass the notification up the chain.
+ */
if (obd->obd_observer)
rc = obd_notify(obd->obd_observer, watched, ev, data);
RETURN(rc);
}
-/* this is fake connect function. Its purpose is to initialize lmv and say
- * caller that everything is okay. Real connection will be performed later. */
+/**
+ * This is fake connect function. Its purpose is to initialize lmv and say
+ * caller that everything is okay. Real connection will be performed later.
+ */
static int lmv_connect(const struct lu_env *env,
- struct lustre_handle *conn, struct obd_device *obd,
+ struct obd_export **exp, struct obd_device *obd,
struct obd_uuid *cluuid, struct obd_connect_data *data,
void *localdata)
{
#ifdef __KERNEL__
struct proc_dir_entry *lmv_proc_dir;
#endif
- struct lmv_obd *lmv = &obd->u.lmv;
- struct obd_export *exp;
- int rc = 0;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lustre_handle conn = { 0 };
+ int rc = 0;
ENTRY;
- rc = class_connect(conn, obd, cluuid);
+ /*
+ * We don't want to actually do the underlying connections more than
+ * once, so keep track.
+ */
+ lmv->refcount++;
+ if (lmv->refcount > 1) {
+ *exp = NULL;
+ RETURN(0);
+ }
+
+ rc = class_connect(&conn, obd, cluuid);
if (rc) {
CERROR("class_connection() returned %d\n", rc);
RETURN(rc);
}
- exp = class_conn2export(conn);
-
- /* we don't want to actually do the underlying connections more than
- * once, so keep track. */
- lmv->refcount++;
- if (lmv->refcount > 1) {
- class_export_put(exp);
- RETURN(0);
- }
+ *exp = class_conn2export(&conn);
+ class_export_get(*exp);
- lmv->exp = exp;
+ lmv->exp = *exp;
lmv->connected = 0;
lmv->cluuid = *cluuid;
}
#endif
- /* all real clients should perform actual connection right away, because
+ /*
+ * All real clients should perform actual connection right away, because
* it is possible, that LMV will not have opportunity to connect targets
* and MDC stuff will be called directly, for instance while reading
- * ../mdc/../kbytesfree procfs file, etc. */
+ * ../mdc/../kbytesfree procfs file, etc.
+ */
if (data->ocd_connect_flags & OBD_CONNECT_REAL)
rc = lmv_check_connect(obd);
static void lmv_set_timeouts(struct obd_device *obd)
{
- struct lmv_tgt_desc *tgts;
- struct lmv_obd *lmv;
- int i;
+ struct lmv_tgt_desc *tgt;
+ struct lmv_obd *lmv;
+ int i;
lmv = &obd->u.lmv;
if (lmv->server_timeout == 0)
if (lmv->connected == 0)
return;
- for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
- if (tgts->ltd_exp == NULL)
- continue;
+ for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
+ tgt = lmv->tgts[i];
+ if (tgt == NULL || tgt->ltd_exp == NULL || tgt->ltd_active == 0)
+ continue;
- obd_set_info_async(tgts->ltd_exp, sizeof(KEY_INTERMDS),
- KEY_INTERMDS, 0, NULL, NULL);
- }
+ obd_set_info_async(NULL, tgt->ltd_exp, sizeof(KEY_INTERMDS),
+ KEY_INTERMDS, 0, NULL, NULL);
+ }
}
static int lmv_init_ea_size(struct obd_export *exp, int easize,
int def_easize, int cookiesize)
{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- int i, rc = 0, change = 0;
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ int i;
+ int rc = 0;
+ int change = 0;
ENTRY;
if (lmv->max_easize < easize) {
if (lmv->connected == 0)
RETURN(0);
- for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
- if (lmv->tgts[i].ltd_exp == NULL) {
- CWARN("%s: NULL export for %d\n", obd->obd_name, i);
- continue;
- }
-
- rc = md_init_ea_size(lmv->tgts[i].ltd_exp, easize, def_easize,
- cookiesize);
- if (rc) {
- CERROR("obd_init_ea_size() failed on MDT target %d, "
- "error %d.\n", i, rc);
- break;
- }
- }
- RETURN(rc);
+ for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
+ if (lmv->tgts[i] == NULL ||
+ lmv->tgts[i]->ltd_exp == NULL ||
+ lmv->tgts[i]->ltd_active == 0) {
+ CWARN("%s: NULL export for %d\n", obd->obd_name, i);
+ continue;
+ }
+
+ rc = md_init_ea_size(lmv->tgts[i]->ltd_exp, easize, def_easize,
+ cookiesize);
+ if (rc) {
+ CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
+ " rc = %d.\n", obd->obd_name, i, rc);
+ break;
+ }
+ }
+ RETURN(rc);
}
#define MAX_STRING_SIZE 128
int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
{
- struct lmv_obd *lmv = &obd->u.lmv;
- struct obd_uuid *cluuid = &lmv->cluuid;
- struct obd_connect_data *mdc_data = NULL;
- struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
- struct lustre_handle conn = {0, };
- struct obd_device *mdc_obd;
- struct obd_export *mdc_exp;
- struct lu_fld_target target;
- int rc;
#ifdef __KERNEL__
- struct proc_dir_entry *lmv_proc_dir;
+ struct proc_dir_entry *lmv_proc_dir;
#endif
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct obd_uuid *cluuid = &lmv->cluuid;
+ struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
+ struct obd_device *mdc_obd;
+ struct obd_export *mdc_exp;
+ struct lu_fld_target target;
+ int rc;
ENTRY;
- /* for MDS: don't connect to yourself */
- if (obd_uuid_equals(&tgt->ltd_uuid, cluuid)) {
- CDEBUG(D_CONFIG, "don't connect back to %s\n", cluuid->uuid);
- /* XXX - the old code didn't increment active tgt count.
- * should we ? */
- RETURN(0);
- }
-
mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
&obd->obd_uuid);
if (!mdc_obd) {
RETURN(-EINVAL);
}
- rc = obd_connect(NULL, &conn, mdc_obd, &lmv_mdc_uuid,
+ rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
&lmv->conn_data, NULL);
if (rc) {
CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
RETURN(rc);
}
- mdc_exp = class_conn2export(&conn);
-
- /* Init fid sequence client for this mdc. */
- rc = obd_fid_init(mdc_exp);
- if (rc)
- RETURN(rc);
+ /*
+ * Init fid sequence client for this mdc and add new fld target.
+ */
+ rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
+ if (rc)
+ RETURN(rc);
- /* Add new FLD target. */
target.ft_srv = NULL;
target.ft_exp = mdc_exp;
target.ft_idx = tgt->ltd_idx;
fld_client_add_target(&lmv->lmv_fld, &target);
- mdc_data = &class_exp2cliimp(mdc_exp)->imp_connect_data;
-
rc = obd_register_observer(mdc_obd, obd);
if (rc) {
obd_disconnect(mdc_exp);
}
if (obd->obd_observer) {
- /* tell the mds_lmv about the new target */
- rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
- OBD_NOTIFY_ACTIVE, (void *)(tgt - lmv->tgts));
- if (rc) {
- obd_disconnect(mdc_exp);
- RETURN(rc);
- }
+ /*
+ * Tell the observer about the new target.
+ */
+ rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
+ OBD_NOTIFY_ACTIVE,
+ (void *)(tgt - lmv->tgts[0]));
+ if (rc) {
+ obd_disconnect(mdc_exp);
+ RETURN(rc);
+ }
}
tgt->ltd_active = 1;
tgt->ltd_exp = mdc_exp;
lmv->desc.ld_active_tgt_count++;
- /* copy connect data, it may be used later */
- lmv->datas[tgt->ltd_idx] = *mdc_data;
-
md_init_ea_size(tgt->ltd_exp, lmv->max_easize,
lmv->max_def_easize, lmv->max_cookiesize);
- CDEBUG(D_CONFIG, "connected to %s(%s) successfully (%d)\n",
+ CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
- atomic_read(&obd->obd_refcount));
+ cfs_atomic_read(&obd->obd_refcount));
#ifdef __KERNEL__
lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
if (lmv_proc_dir) {
struct proc_dir_entry *mdc_symlink;
- char name[MAX_STRING_SIZE + 1];
LASSERT(mdc_obd->obd_type != NULL);
LASSERT(mdc_obd->obd_type->typ_name != NULL);
- name[MAX_STRING_SIZE] = '\0';
- snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
- mdc_obd->obd_type->typ_name,
- mdc_obd->obd_name);
- mdc_symlink = proc_symlink(mdc_obd->obd_name,
- lmv_proc_dir, name);
+ mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
+ lmv_proc_dir,
+ "../../../%s/%s",
+ mdc_obd->obd_type->typ_name,
+ mdc_obd->obd_name);
if (mdc_symlink == NULL) {
- CERROR("could not register LMV target "
+ CERROR("Could not register LMV target "
"/proc/fs/lustre/%s/%s/target_obds/%s.",
obd->obd_type->typ_name, obd->obd_name,
mdc_obd->obd_name);
RETURN(0);
}
-int lmv_add_target(struct obd_device *obd, struct obd_uuid *tgt_uuid)
+static void lmv_del_target(struct lmv_obd *lmv, int index)
{
- struct lmv_obd *lmv = &obd->u.lmv;
+ if (lmv->tgts[index] == NULL)
+ return;
+
+ OBD_FREE_PTR(lmv->tgts[index]);
+ lmv->tgts[index] = NULL;
+ return;
+}
+
+static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
+ __u32 index, int gen)
+{
+ struct lmv_obd *lmv = &obd->u.lmv;
struct lmv_tgt_desc *tgt;
- int rc = 0;
+ int rc = 0;
ENTRY;
- CDEBUG(D_CONFIG, "tgt_uuid: %s.\n", tgt_uuid->uuid);
+ CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
lmv_init_lock(lmv);
- if (lmv->desc.ld_active_tgt_count >= LMV_MAX_TGT_COUNT) {
- lmv_init_unlock(lmv);
- CERROR("can't add %s, LMV module compiled for %d MDCs. "
- "That many MDCs already configured.\n",
- tgt_uuid->uuid, LMV_MAX_TGT_COUNT);
- RETURN(-EINVAL);
- }
- if (lmv->desc.ld_tgt_count == 0) {
- struct obd_device *mdc_obd;
-
- mdc_obd = class_find_client_obd(tgt_uuid, LUSTRE_MDC_NAME,
- &obd->obd_uuid);
- if (!mdc_obd) {
- lmv_init_unlock(lmv);
- CERROR("Target %s not attached\n", tgt_uuid->uuid);
- RETURN(-EINVAL);
- }
-
- rc = obd_llog_init(obd, &obd->obd_olg, mdc_obd, 0, NULL, tgt_uuid);
- if (rc) {
- lmv_init_unlock(lmv);
- CERROR("lmv failed to setup llogging subsystems\n");
- }
- }
- spin_lock(&lmv->lmv_lock);
- tgt = lmv->tgts + lmv->desc.ld_tgt_count++;
- tgt->ltd_uuid = *tgt_uuid;
- spin_unlock(&lmv->lmv_lock);
-
- if (lmv->connected) {
- rc = lmv_connect_mdc(obd, tgt);
- if (rc) {
- spin_lock(&lmv->lmv_lock);
- lmv->desc.ld_tgt_count--;
- memset(tgt, 0, sizeof(*tgt));
- spin_unlock(&lmv->lmv_lock);
+ if (lmv->desc.ld_tgt_count == 0) {
+ struct obd_device *mdc_obd;
+
+ mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
+ &obd->obd_uuid);
+ if (!mdc_obd) {
+ lmv_init_unlock(lmv);
+ CERROR("%s: Target %s not attached: rc = %d\n",
+ obd->obd_name, uuidp->uuid, -EINVAL);
+ RETURN(-EINVAL);
+ }
+ }
+
+ if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
+ tgt = lmv->tgts[index];
+ CERROR("%s: UUID %s already assigned at LOV target index %d:"
+ " rc = %d\n", obd->obd_name,
+ obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
+ lmv_init_unlock(lmv);
+ RETURN(-EEXIST);
+ }
+
+ if (index >= lmv->tgts_size) {
+ /* We need to reallocate the lmv target array. */
+ struct lmv_tgt_desc **newtgts, **old = NULL;
+ __u32 newsize = 1;
+ __u32 oldsize = 0;
+
+ while (newsize < index + 1)
+ newsize = newsize << 1;
+ OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
+ if (newtgts == NULL) {
+ lmv_init_unlock(lmv);
+ RETURN(-ENOMEM);
+ }
+
+ if (lmv->tgts_size) {
+ memcpy(newtgts, lmv->tgts,
+ sizeof(*newtgts) * lmv->tgts_size);
+ old = lmv->tgts;
+ oldsize = lmv->tgts_size;
+ }
+
+ lmv->tgts = newtgts;
+ lmv->tgts_size = newsize;
+ smp_rmb();
+ if (old)
+ OBD_FREE(old, sizeof(*old) * oldsize);
+
+ CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
+ lmv->tgts_size);
+ }
+
+ OBD_ALLOC_PTR(tgt);
+ if (!tgt) {
+ lmv_init_unlock(lmv);
+ RETURN(-ENOMEM);
+ }
+
+ mutex_init(&tgt->ltd_fid_mutex);
+ tgt->ltd_idx = index;
+ tgt->ltd_uuid = *uuidp;
+ tgt->ltd_active = 0;
+ lmv->tgts[index] = tgt;
+ if (index >= lmv->desc.ld_tgt_count)
+ lmv->desc.ld_tgt_count = index + 1;
+
+ if (lmv->connected) {
+ rc = lmv_connect_mdc(obd, tgt);
+ if (rc) {
+ spin_lock(&lmv->lmv_lock);
+ lmv->desc.ld_tgt_count--;
+ memset(tgt, 0, sizeof(*tgt));
+ spin_unlock(&lmv->lmv_lock);
} else {
int easize = sizeof(struct lmv_stripe_md) +
lmv->desc.ld_tgt_count *
RETURN(rc);
}
-/* performs a check if passed obd is connected. If no - connect it. */
int lmv_check_connect(struct obd_device *obd)
{
- struct lmv_obd *lmv = &obd->u.lmv;
- struct lmv_tgt_desc *tgt;
- int i, rc, easize;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lmv_tgt_desc *tgt;
+ int i;
+ int rc;
+ int easize;
ENTRY;
if (lmv->connected)
}
if (lmv->desc.ld_tgt_count == 0) {
+ lmv_init_unlock(lmv);
CERROR("%s: no targets configured.\n", obd->obd_name);
RETURN(-EINVAL);
}
- CDEBUG(D_CONFIG, "time to connect %s to %s\n",
+ CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
lmv->cluuid.uuid, obd->obd_name);
- LASSERT(lmv->tgts != NULL);
+ LASSERT(lmv->tgts != NULL);
- for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
- rc = lmv_connect_mdc(obd, tgt);
- if (rc)
- GOTO(out_disc, rc);
- }
+ for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
+ tgt = lmv->tgts[i];
+ if (tgt == NULL)
+ continue;
+ rc = lmv_connect_mdc(obd, tgt);
+ if (rc)
+ GOTO(out_disc, rc);
+ }
lmv_set_timeouts(obd);
class_export_put(lmv->exp);
out_disc:
while (i-- > 0) {
int rc2;
- --tgt;
+ tgt = lmv->tgts[i];
+ if (tgt == NULL)
+ continue;
tgt->ltd_active = 0;
if (tgt->ltd_exp) {
--lmv->desc.ld_active_tgt_count;
rc2 = obd_disconnect(tgt->ltd_exp);
if (rc2) {
- CERROR("error: LMV target %s disconnect on "
+ CERROR("LMV target %s disconnect on "
"MDC idx %d: error %d\n",
tgt->ltd_uuid.uuid, i, rc2);
}
static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
{
#ifdef __KERNEL__
- struct proc_dir_entry *lmv_proc_dir;
+ struct proc_dir_entry *lmv_proc_dir;
#endif
- struct lmv_obd *lmv = &obd->u.lmv;
- struct obd_device *mdc_obd;
- int rc;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct obd_device *mdc_obd;
+ int rc;
ENTRY;
LASSERT(tgt != NULL);
mdc_obd = class_exp2obd(tgt->ltd_exp);
- if (mdc_obd)
+ if (mdc_obd) {
+ mdc_obd->obd_force = obd->obd_force;
+ mdc_obd->obd_fail = obd->obd_fail;
mdc_obd->obd_no_recov = obd->obd_no_recov;
+ }
#ifdef __KERNEL__
lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
}
}
#endif
- rc = obd_fid_fini(tgt->ltd_exp);
- if (rc)
- CERROR("Can't finanize fids factory\n");
+ rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
+ if (rc)
+ CERROR("Can't finanize fids factory\n");
- CDEBUG(D_OTHER, "Disconnected from %s(%s) successfully\n",
+ CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
tgt->ltd_exp->exp_obd->obd_name,
tgt->ltd_exp->exp_obd->obd_uuid.uuid);
static int lmv_disconnect(struct obd_export *exp)
{
- struct obd_device *obd = class_exp2obd(exp);
+ struct obd_device *obd = class_exp2obd(exp);
#ifdef __KERNEL__
struct proc_dir_entry *lmv_proc_dir;
#endif
- struct lmv_obd *lmv = &obd->u.lmv;
- int rc, i;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ int rc;
+ int i;
ENTRY;
if (!lmv->tgts)
goto out_local;
- /* Only disconnect the underlying layers on the final disconnect. */
+ /*
+ * Only disconnect the underlying layers on the final disconnect.
+ */
lmv->refcount--;
if (lmv->refcount != 0)
goto out_local;
for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
- if (lmv->tgts[i].ltd_exp == NULL)
+ if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
continue;
- lmv_disconnect_mdc(obd, &lmv->tgts[i]);
+
+ lmv_disconnect_mdc(obd, lmv->tgts[i]);
}
#ifdef __KERNEL__
RETURN(rc);
}
+static int lmv_fid2path(struct obd_export *exp, int len, void *karg, void *uarg)
+{
+ struct obd_device *obddev = class_exp2obd(exp);
+ struct lmv_obd *lmv = &obddev->u.lmv;
+ struct getinfo_fid2path *gf;
+ struct lmv_tgt_desc *tgt;
+ struct getinfo_fid2path *remote_gf = NULL;
+ int remote_gf_size = 0;
+ int rc;
+
+ gf = (struct getinfo_fid2path *)karg;
+ tgt = lmv_find_target(lmv, &gf->gf_fid);
+ if (IS_ERR(tgt))
+ RETURN(PTR_ERR(tgt));
+
+repeat_fid2path:
+ rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
+ if (rc != 0 && rc != -EREMOTE)
+ GOTO(out_fid2path, rc);
+
+ /* If remote_gf != NULL, it means just building the
+ * path on the remote MDT, copy this path segement to gf */
+ if (remote_gf != NULL) {
+ struct getinfo_fid2path *ori_gf;
+ char *ptr;
+
+ ori_gf = (struct getinfo_fid2path *)karg;
+ if (strlen(ori_gf->gf_path) +
+ strlen(gf->gf_path) > ori_gf->gf_pathlen)
+ GOTO(out_fid2path, rc = -EOVERFLOW);
+
+ ptr = ori_gf->gf_path;
+
+ memmove(ptr + strlen(gf->gf_path) + 1, ptr,
+ strlen(ori_gf->gf_path));
+
+ strncpy(ptr, gf->gf_path, strlen(gf->gf_path));
+ ptr += strlen(gf->gf_path);
+ *ptr = '/';
+ }
+
+ CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
+ tgt->ltd_exp->exp_obd->obd_name,
+ gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
+ gf->gf_linkno);
+
+ if (rc == 0)
+ GOTO(out_fid2path, rc);
+
+ /* sigh, has to go to another MDT to do path building further */
+ if (remote_gf == NULL) {
+ remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
+ OBD_ALLOC(remote_gf, remote_gf_size);
+ if (remote_gf == NULL)
+ GOTO(out_fid2path, rc = -ENOMEM);
+ remote_gf->gf_pathlen = PATH_MAX;
+ }
+
+ if (!fid_is_sane(&gf->gf_fid)) {
+ CERROR("%s: invalid FID "DFID": rc = %d\n",
+ tgt->ltd_exp->exp_obd->obd_name,
+ PFID(&gf->gf_fid), -EINVAL);
+ GOTO(out_fid2path, rc = -EINVAL);
+ }
+
+ tgt = lmv_find_target(lmv, &gf->gf_fid);
+ if (IS_ERR(tgt))
+ GOTO(out_fid2path, rc = -EINVAL);
+
+ remote_gf->gf_fid = gf->gf_fid;
+ remote_gf->gf_recno = -1;
+ remote_gf->gf_linkno = -1;
+ memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
+ gf = remote_gf;
+ goto repeat_fid2path;
+
+out_fid2path:
+ if (remote_gf != NULL)
+ OBD_FREE(remote_gf, remote_gf_size);
+ RETURN(rc);
+}
+
+static int lmv_hsm_req_count(struct lmv_obd *lmv,
+ const struct hsm_user_request *hur,
+ const struct lmv_tgt_desc *tgt_mds)
+{
+ int i, nr = 0;
+ struct lmv_tgt_desc *curr_tgt;
+
+ /* count how many requests must be sent to the given target */
+ for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
+ curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
+ if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
+ nr++;
+ }
+ return nr;
+}
+
+static void lmv_hsm_req_build(struct lmv_obd *lmv,
+ struct hsm_user_request *hur_in,
+ const struct lmv_tgt_desc *tgt_mds,
+ struct hsm_user_request *hur_out)
+{
+ int i, nr_out;
+ struct lmv_tgt_desc *curr_tgt;
+
+ /* build the hsm_user_request for the given target */
+ hur_out->hur_request = hur_in->hur_request;
+ nr_out = 0;
+ for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
+ curr_tgt = lmv_find_target(lmv,
+ &hur_in->hur_user_item[i].hui_fid);
+ if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
+ hur_out->hur_user_item[nr_out] =
+ hur_in->hur_user_item[i];
+ nr_out++;
+ }
+ }
+ hur_out->hur_request.hr_itemcount = nr_out;
+ memcpy(hur_data(hur_out), hur_data(hur_in),
+ hur_in->hur_request.hr_data_len);
+}
+
+static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
+ struct lustre_kernelcomm *lk, void *uarg)
+{
+ int i, rc = 0;
+ ENTRY;
+
+ /* unregister request (call from llapi_hsm_copytool_fini) */
+ for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
+ /* best effort: try to clean as much as possible
+ * (continue on error) */
+ obd_iocontrol(cmd, lmv->tgts[i]->ltd_exp, len, lk, uarg);
+ }
+
+ /* Whatever the result, remove copytool from kuc groups.
+ * Unreached coordinators will get EPIPE on next requests
+ * and will unregister automatically.
+ */
+ rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group);
+ RETURN(rc);
+}
+
+static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
+ struct lustre_kernelcomm *lk, void *uarg)
+{
+ struct file *filp;
+ int i, j, err;
+ int rc = 0;
+ bool any_set = false;
+ ENTRY;
+
+ /* All or nothing: try to register to all MDS.
+ * In case of failure, unregister from previous MDS,
+ * except if it because of inactive target. */
+ for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
+ err = obd_iocontrol(cmd, lmv->tgts[i]->ltd_exp,
+ len, lk, uarg);
+ if (err) {
+ if (lmv->tgts[i]->ltd_active) {
+ /* permanent error */
+ CERROR("error: iocontrol MDC %s on MDT"
+ "idx %d cmd %x: err = %d\n",
+ lmv->tgts[i]->ltd_uuid.uuid,
+ i, cmd, err);
+ rc = err;
+ lk->lk_flags |= LK_FLG_STOP;
+ /* unregister from previous MDS */
+ for (j = 0; j < i; j++)
+ obd_iocontrol(cmd,
+ lmv->tgts[j]->ltd_exp,
+ len, lk, uarg);
+ RETURN(rc);
+ }
+ /* else: transient error.
+ * kuc will register to the missing MDT
+ * when it is back */
+ } else {
+ any_set = true;
+ }
+ }
+
+ if (!any_set)
+ /* no registration done: return error */
+ RETURN(-ENOTCONN);
+
+ /* at least one registration done, with no failure */
+ filp = fget(lk->lk_wfd);
+ if (filp == NULL) {
+ RETURN(-EBADF);
+ }
+ rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, lk->lk_data);
+ if (rc != 0 && filp != NULL)
+ fput(filp);
+ RETURN(rc);
+}
+
+
+
+
static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
int len, void *karg, void *uarg)
{
- struct obd_device *obddev = class_exp2obd(exp);
- struct lmv_obd *lmv = &obddev->u.lmv;
- int i, rc = 0, set = 0;
+ struct obd_device *obddev = class_exp2obd(exp);
+ struct lmv_obd *lmv = &obddev->u.lmv;
+ int i = 0;
+ int rc = 0;
+ int set = 0;
+ int count = lmv->desc.ld_tgt_count;
ENTRY;
- if (lmv->desc.ld_tgt_count == 0)
+ if (count == 0)
RETURN(-ENOTTY);
switch (cmd) {
__u32 index;
memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
- LASSERT(data->ioc_plen1 == sizeof(struct obd_statfs));
-
- if ((index >= lmv->desc.ld_tgt_count))
+ if ((index >= count))
RETURN(-ENODEV);
- if (!lmv->tgts[index].ltd_active)
- RETURN(-ENODATA);
+ if (lmv->tgts[index] == NULL ||
+ lmv->tgts[index]->ltd_active == 0)
+ RETURN(-ENODATA);
+
+ mdc_obd = class_exp2obd(lmv->tgts[index]->ltd_exp);
+ if (!mdc_obd)
+ RETURN(-EINVAL);
+
+ /* copy UUID */
+ if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
+ min((int) data->ioc_plen2,
+ (int) sizeof(struct obd_uuid))))
+ RETURN(-EFAULT);
+
+ rc = obd_statfs(NULL, lmv->tgts[index]->ltd_exp, &stat_buf,
+ cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
+ 0);
+ if (rc)
+ RETURN(rc);
+ if (copy_to_user(data->ioc_pbuf1, &stat_buf,
+ min((int) data->ioc_plen1,
+ (int) sizeof(stat_buf))))
+ RETURN(-EFAULT);
+ break;
+ }
+ case OBD_IOC_QUOTACTL: {
+ struct if_quotactl *qctl = karg;
+ struct lmv_tgt_desc *tgt = NULL;
+ struct obd_quotactl *oqctl;
+
+ if (qctl->qc_valid == QC_MDTIDX) {
+ if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
+ RETURN(-EINVAL);
+
+ tgt = lmv->tgts[qctl->qc_idx];
+ if (tgt == NULL || tgt->ltd_exp == NULL)
+ RETURN(-EINVAL);
+ } else if (qctl->qc_valid == QC_UUID) {
+ for (i = 0; i < count; i++) {
+ tgt = lmv->tgts[i];
+ if (tgt == NULL)
+ continue;
+ if (!obd_uuid_equals(&tgt->ltd_uuid,
+ &qctl->obd_uuid))
+ continue;
+
+ if (tgt->ltd_exp == NULL)
+ RETURN(-EINVAL);
- mdc_obd = class_exp2obd(lmv->tgts[index].ltd_exp);
- if (!mdc_obd)
+ break;
+ }
+ } else {
RETURN(-EINVAL);
+ }
- /* got statfs data */
- rc = obd_statfs(mdc_obd, &stat_buf,
- cfs_time_current_64() - HZ, 0);
- if (rc)
- RETURN(rc);
- if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
- RETURN(rc);
- /* copy UUID */
- rc = copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
- data->ioc_plen2);
+ if (i >= count)
+ RETURN(-EAGAIN);
+
+ LASSERT(tgt && tgt->ltd_exp);
+ OBD_ALLOC_PTR(oqctl);
+ if (!oqctl)
+ RETURN(-ENOMEM);
+
+ QCTL_COPY(oqctl, qctl);
+ rc = obd_quotactl(tgt->ltd_exp, oqctl);
+ if (rc == 0) {
+ QCTL_COPY(qctl, oqctl);
+ qctl->qc_valid = QC_MDTIDX;
+ qctl->obd_uuid = tgt->ltd_uuid;
+ }
+ OBD_FREE_PTR(oqctl);
break;
}
- default : {
- for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
- int err;
-
- if (lmv->tgts[i].ltd_exp == NULL)
- continue;
-
- err = obd_iocontrol(cmd, lmv->tgts[i].ltd_exp, len,
- karg, uarg);
- if (err) {
- if (lmv->tgts[i].ltd_active) {
- CERROR("error: iocontrol MDC %s on MDT"
- "idx %d cmd %x: err = %d\n",
- lmv->tgts[i].ltd_uuid.uuid,
- i, cmd, err);
- if (!rc)
- rc = err;
- }
- } else
- set = 1;
+ case OBD_IOC_CHANGELOG_SEND:
+ case OBD_IOC_CHANGELOG_CLEAR: {
+ struct ioc_changelog *icc = karg;
+
+ if (icc->icc_mdtindex >= count)
+ RETURN(-ENODEV);
+
+ if (lmv->tgts[icc->icc_mdtindex] == NULL ||
+ lmv->tgts[icc->icc_mdtindex]->ltd_exp == NULL ||
+ lmv->tgts[icc->icc_mdtindex]->ltd_active == 0)
+ RETURN(-ENODEV);
+ rc = obd_iocontrol(cmd, lmv->tgts[icc->icc_mdtindex]->ltd_exp,
+ sizeof(*icc), icc, NULL);
+ break;
+ }
+ case LL_IOC_GET_CONNECT_FLAGS: {
+ if (lmv->tgts[0] == NULL)
+ RETURN(-ENODATA);
+ rc = obd_iocontrol(cmd, lmv->tgts[0]->ltd_exp, len, karg, uarg);
+ break;
+ }
+ case OBD_IOC_FID2PATH: {
+ rc = lmv_fid2path(exp, len, karg, uarg);
+ break;
+ }
+ case LL_IOC_HSM_STATE_GET:
+ case LL_IOC_HSM_STATE_SET:
+ case LL_IOC_HSM_ACTION: {
+ struct md_op_data *op_data = karg;
+ struct lmv_tgt_desc *tgt;
+
+ tgt = lmv_find_target(lmv, &op_data->op_fid1);
+ if (IS_ERR(tgt))
+ RETURN(PTR_ERR(tgt));
+
+ if (tgt->ltd_exp == NULL)
+ RETURN(-EINVAL);
+
+ rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
+ break;
+ }
+ case LL_IOC_HSM_PROGRESS: {
+ const struct hsm_progress_kernel *hpk = karg;
+ struct lmv_tgt_desc *tgt;
+
+ tgt = lmv_find_target(lmv, &hpk->hpk_fid);
+ if (IS_ERR(tgt))
+ RETURN(PTR_ERR(tgt));
+ rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
+ break;
+ }
+ case LL_IOC_HSM_REQUEST: {
+ struct hsm_user_request *hur = karg;
+ struct lmv_tgt_desc *tgt;
+ unsigned int reqcount = hur->hur_request.hr_itemcount;
+
+ if (reqcount == 0)
+ RETURN(0);
+
+ /* if the request is about a single fid
+ * or if there is a single MDS, no need to split
+ * the request. */
+ if (reqcount == 1 || count == 1) {
+ tgt = lmv_find_target(lmv,
+ &hur->hur_user_item[0].hui_fid);
+ if (IS_ERR(tgt))
+ RETURN(PTR_ERR(tgt));
+ rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
+ } else {
+ /* split fid list to their respective MDS */
+ for (i = 0; i < count; i++) {
+ unsigned int nr, reqlen;
+ int rc1;
+ struct hsm_user_request *req;
+
+ nr = lmv_hsm_req_count(lmv, hur, lmv->tgts[i]);
+ if (nr == 0) /* nothing for this MDS */
+ continue;
+
+ /* build a request with fids for this MDS */
+ reqlen = offsetof(typeof(*hur),
+ hur_user_item[nr])
+ + hur->hur_request.hr_data_len;
+ OBD_ALLOC_LARGE(req, reqlen);
+ if (req == NULL)
+ RETURN(-ENOMEM);
+
+ lmv_hsm_req_build(lmv, hur, lmv->tgts[i], req);
+
+ rc1 = obd_iocontrol(cmd, lmv->tgts[i]->ltd_exp,
+ reqlen, req, uarg);
+ if (rc1 != 0 && rc == 0)
+ rc = rc1;
+ OBD_FREE_LARGE(req, reqlen);
+ }
+ }
+ break;
+ }
+ case LL_IOC_LOV_SWAP_LAYOUTS: {
+ struct md_op_data *op_data = karg;
+ struct lmv_tgt_desc *tgt1, *tgt2;
+
+ tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
+ if (IS_ERR(tgt1))
+ RETURN(PTR_ERR(tgt1));
+
+ tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
+ if (IS_ERR(tgt2))
+ RETURN(PTR_ERR(tgt2));
+
+ if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
+ RETURN(-EINVAL);
+
+ /* only files on same MDT can have their layouts swapped */
+ if (tgt1->ltd_idx != tgt2->ltd_idx)
+ RETURN(-EPERM);
+
+ rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
+ break;
+ }
+ case LL_IOC_HSM_CT_START: {
+ struct lustre_kernelcomm *lk = karg;
+ if (lk->lk_flags & LK_FLG_STOP)
+ rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
+ else
+ rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
+ break;
+ }
+ default:
+ for (i = 0; i < count; i++) {
+ struct obd_device *mdc_obd;
+ int err;
+
+ if (lmv->tgts[i] == NULL ||
+ lmv->tgts[i]->ltd_exp == NULL)
+ continue;
+ /* ll_umount_begin() sets force flag but for lmv, not
+ * mdc. Let's pass it through */
+ mdc_obd = class_exp2obd(lmv->tgts[i]->ltd_exp);
+ mdc_obd->obd_force = obddev->obd_force;
+ err = obd_iocontrol(cmd, lmv->tgts[i]->ltd_exp, len,
+ karg, uarg);
+ if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
+ RETURN(err);
+ } else if (err) {
+ if (lmv->tgts[i]->ltd_active) {
+ CERROR("error: iocontrol MDC %s on MDT"
+ "idx %d cmd %x: err = %d\n",
+ lmv->tgts[i]->ltd_uuid.uuid,
+ i, cmd, err);
+ if (!rc)
+ rc = err;
+ }
+ } else
+ set = 1;
}
if (!set && !rc)
rc = -EIO;
}
- }
RETURN(rc);
}
-enum MDS_POLICY {
- CHAR_TYPE,
- NID_TYPE
-};
-
+#if 0
static int lmv_all_chars_policy(int count, const char *name,
int len)
{
static int lmv_nid_policy(struct lmv_obd *lmv)
{
- struct obd_import *imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
- __u32 id;
+ struct obd_import *imp;
+ __u32 id;
+
/*
- * XXX Hack: to get nid we assume that underlying obd device is mdc.
+ * XXX: To get nid we assume that underlying obd device is mdc.
*/
+ imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
return id % lmv->desc.ld_tgt_count;
}
static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
- int type)
+ placement_policy_t placement)
{
- switch (type) {
- case CHAR_TYPE:
+ switch (placement) {
+ case PLACEMENT_CHAR_POLICY:
return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
op_data->op_name,
op_data->op_namelen);
- case NID_TYPE:
+ case PLACEMENT_NID_POLICY:
return lmv_nid_policy(lmv);
default:
break;
}
- CERROR("unsupport type %d \n", type);
+ CERROR("Unsupported placement policy %x\n", placement);
return -EINVAL;
}
+#endif
-/* This is _inode_ placement policy function (not name). */
+/**
+ * This is _inode_ placement policy function (not name).
+ */
static int lmv_placement_policy(struct obd_device *obd,
struct md_op_data *op_data,
mdsno_t *mds)
{
- struct lmv_obd *lmv = &obd->u.lmv;
- struct lmv_obj *obj;
- int rc;
- ENTRY;
-
- LASSERT(mds != NULL);
-
- /*
- * Allocate new fid on target according to operation type and parent
- * home mds.
- */
- obj = lmv_obj_grab(obd, &op_data->op_fid1);
- if (obj != NULL || op_data->op_name == NULL ||
- op_data->op_opc != LUSTRE_OPC_MKDIR) {
- /*
- * Allocate fid for non-dir or for null name or for case parent
- * dir is split.
- */
- if (obj) {
- lmv_obj_put(obj);
-
- /*
- * If we have this flag turned on, and we see that
- * parent dir is split, this means, that caller did not
- * notice split yet. This is race and we would like to
- * let caller know that.
- */
- if (op_data->op_bias & MDS_CHECK_SPLIT)
- RETURN(-ERESTART);
- }
-
- /*
- * Allocate new fid on same mds where parent fid is located and
- * where operation will be sent. In case of split dir, ->op_fid1
- * and ->op_mds here will contain fid and mds of slave directory
- * object (assigned by caller).
- */
- *mds = op_data->op_mds;
- rc = 0;
-
-#if 0
- /* XXX: This should be removed later wehn we sure it is not
- * needed. */
- rc = lmv_fld_lookup(lmv, &op_data->op_fid1, mds);
- if (rc)
- GOTO(out, rc);
-#endif
- } else {
- /*
- * Parent directory is not split and we want to create a
- * directory in it. Let's calculate where to place it according
- * to name.
- */
- *mds = lmv_choose_mds(lmv, op_data, NID_TYPE);
- rc = 0;
- }
- EXIT;
-#if 0
-out:
-#endif
- if (rc) {
- CERROR("Can't choose MDS, err = %d\n", rc);
- } else {
- LASSERT(*mds < lmv->desc.ld_tgt_count);
- }
-
- return rc;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ ENTRY;
+
+ LASSERT(mds != NULL);
+
+ if (lmv->desc.ld_tgt_count == 1) {
+ *mds = 0;
+ RETURN(0);
+ }
+
+ /**
+ * If stripe_offset is provided during setdirstripe
+ * (setdirstripe -i xx), xx MDS will be choosen.
+ */
+ if (op_data->op_cli_flags & CLI_SET_MEA) {
+ struct lmv_user_md *lum;
+
+ lum = (struct lmv_user_md *)op_data->op_data;
+ if (lum->lum_type == LMV_STRIPE_TYPE &&
+ lum->lum_stripe_offset != -1) {
+ if (lum->lum_stripe_offset >= lmv->desc.ld_tgt_count) {
+ CERROR("%s: Stripe_offset %d > MDT count %d:"
+ " rc = %d\n", obd->obd_name,
+ lum->lum_stripe_offset,
+ lmv->desc.ld_tgt_count, -ERANGE);
+ RETURN(-ERANGE);
+ }
+ *mds = lum->lum_stripe_offset;
+ RETURN(0);
+ }
+ }
+
+ /* Allocate new fid on target according to operation type and parent
+ * home mds. */
+ *mds = op_data->op_mds;
+ RETURN(0);
}
int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid,
- mdsno_t mds)
+ mdsno_t mds)
{
- struct lmv_tgt_desc *tgt = &lmv->tgts[mds];
- int rc;
- ENTRY;
+ struct lmv_tgt_desc *tgt;
+ int rc;
+ ENTRY;
+
+ tgt = lmv_get_target(lmv, mds);
+ if (IS_ERR(tgt))
+ RETURN(PTR_ERR(tgt));
- /* New seq alloc and FLD setup should be atomic. */
- down(&tgt->ltd_fid_sem);
+ /*
+ * New seq alloc and FLD setup should be atomic. Otherwise we may find
+ * on server that seq in new allocated fid is not yet known.
+ */
+ mutex_lock(&tgt->ltd_fid_mutex);
- /* Asking underlaying tgt layer to allocate new fid. */
+ if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
+ GOTO(out, rc = -ENODEV);
+
+ /*
+ * Asking underlaying tgt layer to allocate new fid.
+ */
rc = obd_fid_alloc(tgt->ltd_exp, fid, NULL);
if (rc > 0) {
LASSERT(fid_is_sane(fid));
-
- /* Client switches to new sequence, setup FLD. */
- rc = fld_client_create(&lmv->lmv_fld, fid_seq(fid),
- mds, NULL);
- if (rc) {
- CERROR("Can't create fld entry, rc %d\n", rc);
- /* Delete just allocated fid sequence */
- obd_fid_delete(tgt->ltd_exp, NULL);
- }
+ rc = 0;
}
- up(&tgt->ltd_fid_sem);
- RETURN(rc);
+
+ EXIT;
+out:
+ mutex_unlock(&tgt->ltd_fid_mutex);
+ return rc;
}
int lmv_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
struct md_op_data *op_data)
{
- struct obd_device *obd = class_exp2obd(exp);
- struct lmv_obd *lmv = &obd->u.lmv;
- mdsno_t mds;
- int rc;
+ struct obd_device *obd = class_exp2obd(exp);
+ struct lmv_obd *lmv = &obd->u.lmv;
+ mdsno_t mds = 0;
+ int rc;
ENTRY;
LASSERT(op_data != NULL);
RETURN(rc);
}
-static int lmv_fid_delete(struct obd_export *exp, const struct lu_fid *fid)
-{
- ENTRY;
-
- LASSERT(exp && fid);
- if (lmv_obj_delete(exp, fid)) {
- CDEBUG(D_OTHER, "lmv object "DFID" is destroyed.\n",
- PFID(fid));
- }
- RETURN(0);
-}
-
static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
{
- struct lmv_obd *lmv = &obd->u.lmv;
- struct lprocfs_static_vars lvars;
- struct lmv_desc *desc;
- int rc, i = 0;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lprocfs_static_vars lvars;
+ struct lmv_desc *desc;
+ int rc;
ENTRY;
if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
- CERROR("descriptor size wrong: %d > %d\n",
+ CERROR("Lmv descriptor size wrong: %d > %d\n",
(int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
RETURN(-EINVAL);
}
- lmv->tgts_size = LMV_MAX_TGT_COUNT * sizeof(struct lmv_tgt_desc);
-
- OBD_ALLOC(lmv->tgts, lmv->tgts_size);
- if (lmv->tgts == NULL)
- RETURN(-ENOMEM);
-
- for (i = 0; i < LMV_MAX_TGT_COUNT; i++) {
- sema_init(&lmv->tgts[i].ltd_fid_sem, 1);
- lmv->tgts[i].ltd_idx = i;
- }
-
- lmv->datas_size = LMV_MAX_TGT_COUNT * sizeof(struct obd_connect_data);
-
- OBD_ALLOC(lmv->datas, lmv->datas_size);
- if (lmv->datas == NULL)
- GOTO(out_free_tgts, rc = -ENOMEM);
+ OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * 32);
+ if (lmv->tgts == NULL)
+ RETURN(-ENOMEM);
+ lmv->tgts_size = 32;
- obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
- lmv->desc.ld_tgt_count = 0;
- lmv->desc.ld_active_tgt_count = 0;
- lmv->max_cookiesize = 0;
- lmv->max_def_easize = 0;
- lmv->max_easize = 0;
+ obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
+ lmv->desc.ld_tgt_count = 0;
+ lmv->desc.ld_active_tgt_count = 0;
+ lmv->max_cookiesize = 0;
+ lmv->max_def_easize = 0;
+ lmv->max_easize = 0;
+ lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
- spin_lock_init(&lmv->lmv_lock);
- sema_init(&lmv->init_sem, 1);
+ spin_lock_init(&lmv->lmv_lock);
+ mutex_init(&lmv->init_mutex);
- rc = lmv_obj_setup(obd);
- if (rc) {
- CERROR("Can't setup LMV object manager, "
- "error %d.\n", rc);
- GOTO(out_free_datas, rc);
- }
+ lprocfs_lmv_init_vars(&lvars);
- lprocfs_lmv_init_vars(&lvars);
- lprocfs_obd_setup(obd, lvars.obd_vars);
+ lprocfs_obd_setup(obd, lvars.obd_vars);
#ifdef LPROCFS
{
- rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd_status",
+ rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
0444, &lmv_proc_target_fops, obd);
if (rc)
- CWARN("Error adding the target_obd_status file\n");
+ CWARN("%s: error adding LMV target_obd file: rc = %d\n",
+ obd->obd_name, rc);
}
#endif
- rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
- LUSTRE_CLI_FLD_HASH_DHT);
- if (rc) {
- CERROR("can't init FLD, err %d\n",
- rc);
- GOTO(out_free_datas, rc);
- }
+ rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
+ LUSTRE_CLI_FLD_HASH_DHT);
+ if (rc) {
+ CERROR("Can't init FLD, err %d\n", rc);
+ GOTO(out, rc);
+ }
RETURN(0);
-out_free_datas:
- OBD_FREE(lmv->datas, lmv->datas_size);
- lmv->datas = NULL;
-out_free_tgts:
- OBD_FREE(lmv->tgts, lmv->tgts_size);
- lmv->tgts = NULL;
+out:
return rc;
}
static int lmv_cleanup(struct obd_device *obd)
{
- struct lmv_obd *lmv = &obd->u.lmv;
- ENTRY;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ ENTRY;
- fld_client_fini(&lmv->lmv_fld);
- lprocfs_obd_cleanup(obd);
- lmv_obj_cleanup(obd);
- OBD_FREE(lmv->datas, lmv->datas_size);
- OBD_FREE(lmv->tgts, lmv->tgts_size);
-
- RETURN(0);
+ fld_client_fini(&lmv->lmv_fld);
+ if (lmv->tgts != NULL) {
+ int i;
+ for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
+ if (lmv->tgts[i] == NULL)
+ continue;
+ lmv_del_target(lmv, i);
+ }
+ OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
+ lmv->tgts_size = 0;
+ }
+ RETURN(0);
}
static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
{
- struct lustre_cfg *lcfg = buf;
- struct obd_uuid tgt_uuid;
- int rc;
- ENTRY;
-
- switch(lcfg->lcfg_command) {
- case LCFG_ADD_MDC:
- if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(tgt_uuid.uuid))
- GOTO(out, rc = -EINVAL);
-
- obd_str2uuid(&tgt_uuid, lustre_cfg_string(lcfg, 1));
- rc = lmv_add_target(obd, &tgt_uuid);
- GOTO(out, rc);
- default: {
- CERROR("Unknown command: %d\n", lcfg->lcfg_command);
- GOTO(out, rc = -EINVAL);
- }
- }
+ struct lustre_cfg *lcfg = buf;
+ struct obd_uuid obd_uuid;
+ int gen;
+ __u32 index;
+ int rc;
+ ENTRY;
+
+ switch (lcfg->lcfg_command) {
+ case LCFG_ADD_MDC:
+ /* modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID
+ * 2:0 3:1 4:lustre-MDT0000-mdc_UUID */
+ if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
+ GOTO(out, rc = -EINVAL);
+
+ obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
+
+ if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
+ GOTO(out, rc = -EINVAL);
+ if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
+ GOTO(out, rc = -EINVAL);
+ rc = lmv_add_target(obd, &obd_uuid, index, gen);
+ GOTO(out, rc);
+ default:
+ CERROR("Unknown command: %d\n", lcfg->lcfg_command);
+ GOTO(out, rc = -EINVAL);
+ }
out:
- RETURN(rc);
+ RETURN(rc);
}
-static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs,
- __u64 max_age, __u32 flags)
+static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
+ struct obd_statfs *osfs, __u64 max_age, __u32 flags)
{
- struct lmv_obd *lmv = &obd->u.lmv;
- struct obd_statfs *temp;
- int rc = 0, i;
+ struct obd_device *obd = class_exp2obd(exp);
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct obd_statfs *temp;
+ int rc = 0;
+ int i;
ENTRY;
rc = lmv_check_connect(obd);
RETURN(-ENOMEM);
for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
- if (lmv->tgts[i].ltd_exp == NULL)
- continue;
-
- rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, temp,
- max_age, flags);
- if (rc) {
- CERROR("can't stat MDS #%d (%s), error %d\n", i,
- lmv->tgts[i].ltd_exp->exp_obd->obd_name,
- rc);
- GOTO(out_free_temp, rc);
- }
- if (i == 0) {
- *osfs = *temp;
+ if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
+ continue;
+
+ rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
+ max_age, flags);
+ if (rc) {
+ CERROR("can't stat MDS #%d (%s), error %d\n", i,
+ lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
+ rc);
+ GOTO(out_free_temp, rc);
+ }
+
+ if (i == 0) {
+ *osfs = *temp;
+ /* If the statfs is from mount, it will needs
+ * retrieve necessary information from MDT0.
+ * i.e. mount does not need the merged osfs
+ * from all of MDT.
+ * And also clients can be mounted as long as
+ * MDT0 is in service*/
+ if (flags & OBD_STATFS_FOR_MDT0)
+ GOTO(out_free_temp, rc);
} else {
osfs->os_bavail += temp->os_bavail;
osfs->os_blocks += temp->os_blocks;
struct lu_fid *fid,
struct obd_capa **pc)
{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- int rc;
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ int rc;
ENTRY;
rc = lmv_check_connect(obd);
if (rc)
RETURN(rc);
- rc = md_getstatus(lmv->tgts[0].ltd_exp, fid, pc);
-
- RETURN(rc);
+ rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
+ RETURN(rc);
}
static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
const char *input, int input_size, int output_size,
int flags, struct ptlrpc_request **request)
{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- struct obd_export *tgt_exp;
- int rc;
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lmv_tgt_desc *tgt;
+ int rc;
ENTRY;
rc = lmv_check_connect(obd);
if (rc)
RETURN(rc);
- tgt_exp = lmv_find_export(lmv, fid);
- if (IS_ERR(tgt_exp))
- RETURN(PTR_ERR(tgt_exp));
+ tgt = lmv_find_target(lmv, fid);
+ if (IS_ERR(tgt))
+ RETURN(PTR_ERR(tgt));
- rc = md_getxattr(tgt_exp, fid, oc, valid, name, input, input_size,
- output_size, flags, request);
+ rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
+ input_size, output_size, flags, request);
RETURN(rc);
}
int flags, __u32 suppgid,
struct ptlrpc_request **request)
{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- struct obd_export *tgt_exp;
- int rc;
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lmv_tgt_desc *tgt;
+ int rc;
ENTRY;
rc = lmv_check_connect(obd);
if (rc)
RETURN(rc);
- tgt_exp = lmv_find_export(lmv, fid);
- if (IS_ERR(tgt_exp))
- RETURN(PTR_ERR(tgt_exp));
+ tgt = lmv_find_target(lmv, fid);
+ if (IS_ERR(tgt))
+ RETURN(PTR_ERR(tgt));
- rc = md_setxattr(tgt_exp, fid, oc, valid, name,
- input, input_size, output_size, flags, suppgid,
+ rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
+ input_size, output_size, flags, suppgid,
request);
RETURN(rc);
}
-static int lmv_getattr(struct obd_export *exp, const struct lu_fid *fid,
- struct obd_capa *oc, obd_valid valid, int ea_size,
+static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
struct ptlrpc_request **request)
{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- struct obd_export *tgt_exp;
- struct lmv_obj *obj;
- int rc, i;
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lmv_tgt_desc *tgt;
+ int rc;
ENTRY;
rc = lmv_check_connect(obd);
if (rc)
RETURN(rc);
- tgt_exp = lmv_find_export(lmv, fid);
- if (IS_ERR(tgt_exp))
- RETURN(PTR_ERR(tgt_exp));
-
- rc = md_getattr(tgt_exp, fid, oc, valid, ea_size, request);
- if (rc)
- RETURN(rc);
-
- obj = lmv_obj_grab(obd, fid);
-
- CDEBUG(D_OTHER, "GETATTR for "DFID" %s\n", PFID(fid),
- obj ? "(split)" : "");
-
- /*
- * If object is split, then we loop over all the slaves and gather size
- * attribute. In ideal world we would have to gather also mds field from
- * all slaves, as object is spread over the cluster and this is
- * definitely interesting information and it is not good to loss it,
- * but...
- */
- if (obj) {
- struct mdt_body *body;
-
- if (*request == NULL) {
- lmv_obj_put(obj);
- RETURN(rc);
- }
-
- body = req_capsule_server_get(&(*request)->rq_pill,
- &RMF_MDT_BODY);
- LASSERT(body != NULL);
-
- lmv_obj_lock(obj);
-
- for (i = 0; i < obj->lo_objcount; i++) {
- if (lmv->tgts[i].ltd_exp == NULL) {
- CWARN("%s: NULL export for %d\n",
- obd->obd_name, i);
- continue;
- }
-
- /* skip master obj. */
- if (lu_fid_eq(&obj->lo_fid, &obj->lo_inodes[i].li_fid))
- continue;
+ tgt = lmv_find_target(lmv, &op_data->op_fid1);
+ if (IS_ERR(tgt))
+ RETURN(PTR_ERR(tgt));
- lmv_update_body(body, &obj->lo_inodes[i]);
- }
+ if (op_data->op_flags & MF_GET_MDT_IDX) {
+ op_data->op_mds = tgt->ltd_idx;
+ RETURN(0);
+ }
- lmv_obj_unlock(obj);
- lmv_obj_put(obj);
- }
+ rc = md_getattr(tgt->ltd_exp, op_data, request);
RETURN(rc);
}
-static int lmv_change_cbdata(struct obd_export *exp, const struct lu_fid *fid,
- ldlm_iterator_t it, void *data)
+static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- int i, rc;
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ int i;
+ int rc;
ENTRY;
rc = lmv_check_connect(obd);
if (rc)
RETURN(rc);
- CDEBUG(D_OTHER, "CBDATA for "DFID"\n", PFID(fid));
+ CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
- /*
- * With CMD every object can have two locks in different namespaces:
- * lookup lock in space of mds storing direntry and update/open lock in
- * space of mds storing inode.
- */
- for (i = 0; i < lmv->desc.ld_tgt_count; i++)
- md_change_cbdata(lmv->tgts[i].ltd_exp, fid, it, data);
+ /*
+ * With DNE every object can have two locks in different namespaces:
+ * lookup lock in space of MDT storing direntry and update/open lock in
+ * space of MDT storing inode.
+ */
+ for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
+ if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
+ continue;
+ md_null_inode(lmv->tgts[i]->ltd_exp, fid);
+ }
- RETURN(0);
+ RETURN(0);
}
-static int lmv_close(struct obd_export *exp,
- struct md_op_data *op_data,
- struct md_open_data *mod,
- struct ptlrpc_request **request)
+static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
+ ldlm_iterator_t it, void *data)
{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- struct obd_export *tgt_exp;
- int rc;
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ int i;
+ int rc;
ENTRY;
rc = lmv_check_connect(obd);
if (rc)
RETURN(rc);
- tgt_exp = lmv_find_export(lmv, &op_data->op_fid1);
- if (IS_ERR(tgt_exp))
- RETURN(PTR_ERR(tgt_exp));
+ CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
- CDEBUG(D_OTHER, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
- rc = md_close(tgt_exp, op_data, mod, request);
- RETURN(rc);
+ /*
+ * With DNE every object can have two locks in different namespaces:
+ * lookup lock in space of MDT storing direntry and update/open lock in
+ * space of MDT storing inode.
+ */
+ for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
+ if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
+ continue;
+ rc = md_find_cbdata(lmv->tgts[i]->ltd_exp, fid, it, data);
+ if (rc)
+ RETURN(rc);
+ }
+
+ RETURN(rc);
}
-/*
- * Called in the case MDS returns -ERESTART on create on open, what means that
- * directory is split and its LMV presentation object has to be updated.
- */
-int lmv_handle_split(struct obd_export *exp, const struct lu_fid *fid)
+
+static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
+ struct md_open_data *mod, struct ptlrpc_request **request)
{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- struct ptlrpc_request *req = NULL;
- struct obd_export *tgt_exp;
- struct lmv_obj *obj;
- struct lustre_md md;
- int mealen, rc;
- __u64 valid;
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lmv_tgt_desc *tgt;
+ int rc;
ENTRY;
- md.mea = NULL;
- mealen = lmv_get_easize(lmv);
-
- valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA | OBD_MD_MEA;
-
- tgt_exp = lmv_find_export(lmv, fid);
- if (IS_ERR(tgt_exp))
- RETURN(PTR_ERR(tgt_exp));
+ rc = lmv_check_connect(obd);
+ if (rc)
+ RETURN(rc);
- /* time to update mea of parent fid */
- rc = md_getattr(tgt_exp, fid, NULL, valid, mealen, &req);
- if (rc) {
- CERROR("md_getattr() failed, error %d\n", rc);
- GOTO(cleanup, rc);
- }
+ tgt = lmv_find_target(lmv, &op_data->op_fid1);
+ if (IS_ERR(tgt))
+ RETURN(PTR_ERR(tgt));
- rc = md_get_lustre_md(tgt_exp, req, NULL, exp, &md);
- if (rc) {
- CERROR("mdc_get_lustre_md() failed, error %d\n", rc);
- GOTO(cleanup, rc);
- }
+ CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
+ rc = md_close(tgt->ltd_exp, op_data, mod, request);
+ RETURN(rc);
+}
- if (md.mea == NULL)
- GOTO(cleanup, rc = -ENODATA);
+struct lmv_tgt_desc
+*lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
+ struct lu_fid *fid)
+{
+ struct lmv_tgt_desc *tgt;
- obj = lmv_obj_create(exp, fid, md.mea);
- if (IS_ERR(obj))
- rc = PTR_ERR(obj);
- else
- lmv_obj_put(obj);
+ tgt = lmv_find_target(lmv, fid);
+ if (IS_ERR(tgt))
+ return tgt;
- /* XXX LOV STACKING */
- obd_free_memmd(exp, (void *)&md.mea);
+ op_data->op_mds = tgt->ltd_idx;
- EXIT;
-cleanup:
- if (req)
- ptlrpc_req_finished(req);
- return rc;
+ return tgt;
}
int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
const void *data, int datalen, int mode, __u32 uid,
- __u32 gid, __u32 cap_effective, __u64 rdev,
+ __u32 gid, cfs_cap_t cap_effective, __u64 rdev,
struct ptlrpc_request **request)
{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- struct obd_export *tgt_exp;
- struct lmv_obj *obj;
- int rc, loop = 0;
- ENTRY;
-
- rc = lmv_check_connect(obd);
- if (rc)
- RETURN(rc);
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lmv_tgt_desc *tgt;
+ int rc;
+ ENTRY;
- if (!lmv->desc.ld_active_tgt_count)
- RETURN(-EIO);
-repeat:
- ++loop;
- LASSERT(loop <= 2);
- obj = lmv_obj_grab(obd, &op_data->op_fid1);
- if (obj) {
- int mea_idx;
-
- mea_idx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
- op_data->op_name, op_data->op_namelen);
- op_data->op_fid1 = obj->lo_inodes[mea_idx].li_fid;
- op_data->op_bias &= ~MDS_CHECK_SPLIT;
- op_data->op_mds = obj->lo_inodes[mea_idx].li_mds;
- tgt_exp = lmv_get_export(lmv, op_data->op_mds);
- lmv_obj_put(obj);
- } else {
- struct lmv_tgt_desc *tgt;
+ rc = lmv_check_connect(obd);
+ if (rc)
+ RETURN(rc);
- tgt = lmv_find_target(lmv, &op_data->op_fid1);
- op_data->op_bias |= MDS_CHECK_SPLIT;
- op_data->op_mds = tgt->ltd_idx;
- tgt_exp = tgt->ltd_exp;
- }
+ if (!lmv->desc.ld_active_tgt_count)
+ RETURN(-EIO);
- if (IS_ERR(tgt_exp))
- RETURN(PTR_ERR(tgt_exp));
+ tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
+ if (IS_ERR(tgt))
+ RETURN(PTR_ERR(tgt));
- rc = lmv_fid_alloc(exp, &op_data->op_fid2, op_data);
- if (rc == -ERESTART)
- goto repeat;
- else if (rc)
- RETURN(rc);
+ rc = lmv_fid_alloc(exp, &op_data->op_fid2, op_data);
+ if (rc)
+ RETURN(rc);
- CDEBUG(D_OTHER, "CREATE '%*s' on "DFID"\n", op_data->op_namelen,
- op_data->op_name, PFID(&op_data->op_fid1));
+ CDEBUG(D_INODE, "CREATE '%*s' on "DFID" -> mds #%x\n",
+ op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
+ op_data->op_mds);
- op_data->op_flags |= MF_MDC_CANCEL_FID1;
- rc = md_create(tgt_exp, op_data, data, datalen, mode, uid, gid,
- cap_effective, rdev, request);
- if (rc == 0) {
- if (*request == NULL)
- RETURN(rc);
- CDEBUG(D_OTHER, "created - "DFID"\n", PFID(&op_data->op_fid1));
- } else if (rc == -ERESTART) {
- LASSERT(*request != NULL);
- DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
- "Got -ERESTART during create!\n");
- ptlrpc_req_finished(*request);
- *request = NULL;
+ op_data->op_flags |= MF_MDC_CANCEL_FID1;
+ rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
+ cap_effective, rdev, request);
- /*
- * Directory got split. Time to update local object and repeat
- * the request with proper MDS.
- */
- rc = lmv_handle_split(exp, &op_data->op_fid1);
- if (rc == 0) {
- rc = lmv_alloc_slave_fids(obd, &op_data->op_fid1,
- op_data, &op_data->op_fid2);
- if (rc)
- RETURN(rc);
- goto repeat;
- }
- }
- RETURN(rc);
+ if (rc == 0) {
+ if (*request == NULL)
+ RETURN(rc);
+ CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
+ }
+ RETURN(rc);
}
static int lmv_done_writing(struct obd_export *exp,
struct md_op_data *op_data,
struct md_open_data *mod)
{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- struct obd_export *tgt_exp;
- int rc;
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lmv_tgt_desc *tgt;
+ int rc;
ENTRY;
rc = lmv_check_connect(obd);
if (rc)
RETURN(rc);
- tgt_exp = lmv_find_export(lmv, &op_data->op_fid1);
- if (IS_ERR(tgt_exp))
- RETURN(PTR_ERR(tgt_exp));
+ tgt = lmv_find_target(lmv, &op_data->op_fid1);
+ if (IS_ERR(tgt))
+ RETURN(PTR_ERR(tgt));
- rc = md_done_writing(tgt_exp, op_data, mod);
+ rc = md_done_writing(tgt->ltd_exp, op_data, mod);
RETURN(rc);
}
static int
-lmv_enqueue_slaves(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
- struct lookup_intent *it, struct md_op_data *op_data,
- struct lustre_handle *lockh, void *lmm, int lmmsize)
-{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- struct lmv_stripe_md *mea = op_data->op_mea1;
- struct md_op_data *op_data2;
- struct obd_export *tgt_exp;
- int i, rc = 0;
- ENTRY;
-
- OBD_ALLOC_PTR(op_data2);
- if (op_data2 == NULL)
- RETURN(-ENOMEM);
-
- LASSERT(mea != NULL);
- for (i = 0; i < mea->mea_count; i++) {
- memset(op_data2, 0, sizeof(*op_data2));
- op_data2->op_fid1 = mea->mea_ids[i];
- op_data2->op_bias = 0;
-
- tgt_exp = lmv_find_export(lmv, &op_data2->op_fid1);
- if (IS_ERR(tgt_exp))
- GOTO(cleanup, rc = PTR_ERR(tgt_exp));
-
- if (tgt_exp == NULL)
- continue;
-
- rc = md_enqueue(tgt_exp, einfo, it, op_data2,
- lockh + i, lmm, lmmsize, NULL, 0);
-
- CDEBUG(D_OTHER, "take lock on slave "DFID" -> %d/%d\n",
- PFID(&mea->mea_ids[i]), rc, it->d.lustre.it_status);
-
- if (rc)
- GOTO(cleanup, rc);
-
- if (it->d.lustre.it_data) {
- struct ptlrpc_request *req;
- req = (struct ptlrpc_request *)it->d.lustre.it_data;
- ptlrpc_req_finished(req);
- }
-
- if (it->d.lustre.it_status)
- GOTO(cleanup, rc = it->d.lustre.it_status);
- }
-
- EXIT;
-cleanup:
- OBD_FREE_PTR(op_data2);
-
- if (rc != 0) {
- /* drop all taken locks */
- while (--i >= 0) {
- if (lockh[i].cookie)
- ldlm_lock_decref(lockh + i, einfo->ei_mode);
- lockh[i].cookie = 0;
- }
- }
- return rc;
-}
-
-static int
lmv_enqueue_remote(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
struct lookup_intent *it, struct md_op_data *op_data,
struct lustre_handle *lockh, void *lmm, int lmmsize,
int extra_lock_flags)
{
- struct ptlrpc_request *req = it->d.lustre.it_data;
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- struct lustre_handle plock;
- struct obd_export *tgt_exp;
- struct md_op_data *rdata;
- struct lu_fid fid_copy;
- struct mdt_body *body;
- int rc = 0, pmode;
+ struct ptlrpc_request *req = it->d.lustre.it_data;
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lustre_handle plock;
+ struct lmv_tgt_desc *tgt;
+ struct md_op_data *rdata;
+ struct lu_fid fid1;
+ struct mdt_body *body;
+ int rc = 0;
+ int pmode;
ENTRY;
body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
if (!(body->valid & OBD_MD_MDS))
RETURN(0);
- CDEBUG(D_OTHER, "ENQUEUE '%s' on "DFID" -> "DFID"\n",
+ CDEBUG(D_INODE, "REMOTE_ENQUEUE '%s' on "DFID" -> "DFID"\n",
LL_IT2STR(it), PFID(&op_data->op_fid1), PFID(&body->fid1));
- /* We got LOOKUP lock, but we really need attrs */
+ /*
+ * We got LOOKUP lock, but we really need attrs.
+ */
pmode = it->d.lustre.it_lock_mode;
LASSERT(pmode != 0);
memcpy(&plock, lockh, sizeof(plock));
it->d.lustre.it_lock_mode = 0;
it->d.lustre.it_data = NULL;
- fid_copy = body->fid1;
+ fid1 = body->fid1;
it->d.lustre.it_disposition &= ~DISP_ENQ_COMPLETE;
ptlrpc_req_finished(req);
- tgt_exp = lmv_find_export(lmv, &fid_copy);
- if (IS_ERR(tgt_exp))
- GOTO(out, rc = PTR_ERR(tgt_exp));
+ tgt = lmv_find_target(lmv, &fid1);
+ if (IS_ERR(tgt))
+ GOTO(out, rc = PTR_ERR(tgt));
OBD_ALLOC_PTR(rdata);
if (rdata == NULL)
GOTO(out, rc = -ENOMEM);
- rdata->op_fid1 = fid_copy;
+ rdata->op_fid1 = fid1;
rdata->op_bias = MDS_CROSS_REF;
- rc = md_enqueue(tgt_exp, einfo, it, rdata, lockh,
+ rc = md_enqueue(tgt->ltd_exp, einfo, it, rdata, lockh,
lmm, lmmsize, NULL, extra_lock_flags);
OBD_FREE_PTR(rdata);
EXIT;
lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
struct lookup_intent *it, struct md_op_data *op_data,
struct lustre_handle *lockh, void *lmm, int lmmsize,
- struct ptlrpc_request **req, int extra_lock_flags)
+ struct ptlrpc_request **req, __u64 extra_lock_flags)
{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- struct obd_export *tgt_exp = NULL;
- struct lmv_obj *obj;
- int rc;
- ENTRY;
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lmv_tgt_desc *tgt;
+ int rc;
+ ENTRY;
- rc = lmv_check_connect(obd);
- if (rc)
- RETURN(rc);
-
- if (op_data->op_mea1 && it && it->it_op == IT_UNLINK) {
- rc = lmv_enqueue_slaves(exp, einfo, it, op_data,
- lockh, lmm, lmmsize);
- RETURN(rc);
- }
+ rc = lmv_check_connect(obd);
+ if (rc)
+ RETURN(rc);
- if (op_data->op_namelen) {
- obj = lmv_obj_grab(obd, &op_data->op_fid1);
- if (obj) {
- int mea_idx;
-
- /* directory is split. look for right mds for this
- * name */
- mea_idx = raw_name2idx(obj->lo_hashtype,
- obj->lo_objcount,
- (char *)op_data->op_name,
- op_data->op_namelen);
- op_data->op_fid1 = obj->lo_inodes[mea_idx].li_fid;
- tgt_exp = lmv_get_export(lmv, obj->lo_inodes[mea_idx].li_mds);
- lmv_obj_put(obj);
- }
- }
+ CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
+ LL_IT2STR(it), PFID(&op_data->op_fid1));
- if (tgt_exp == NULL)
- tgt_exp = lmv_find_export(lmv, &op_data->op_fid1);
- if (IS_ERR(tgt_exp))
- RETURN(PTR_ERR(tgt_exp));
+ tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
+ if (IS_ERR(tgt))
+ RETURN(PTR_ERR(tgt));
- CDEBUG(D_OTHER, "ENQUEUE '%s' on "DFID"\n", LL_IT2STR(it),
- PFID(&op_data->op_fid1));
+ CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%d\n",
+ LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
- rc = md_enqueue(tgt_exp, einfo, it, op_data, lockh,
- lmm, lmmsize, req, extra_lock_flags);
+ rc = md_enqueue(tgt->ltd_exp, einfo, it, op_data, lockh,
+ lmm, lmmsize, req, extra_lock_flags);
- if (rc == 0 && it && it->it_op == IT_OPEN)
- rc = lmv_enqueue_remote(exp, einfo, it, op_data, lockh,
- lmm, lmmsize, extra_lock_flags);
- RETURN(rc);
+ if (rc == 0 && it && it->it_op == IT_OPEN) {
+ rc = lmv_enqueue_remote(exp, einfo, it, op_data, lockh,
+ lmm, lmmsize, extra_lock_flags);
+ }
+ RETURN(rc);
}
static int
-lmv_getattr_name(struct obd_export *exp, const struct lu_fid *fid,
- struct obd_capa *oc, const char *filename, int namelen,
- obd_valid valid, int ea_size, __u32 suppgid,
+lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
struct ptlrpc_request **request)
{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- struct lu_fid rid = *fid;
- struct obd_export *tgt_exp;
- struct mdt_body *body;
- struct lmv_obj *obj;
- int rc, loop = 0;
- ENTRY;
-
- rc = lmv_check_connect(obd);
- if (rc)
- RETURN(rc);
-
-repeat:
- ++loop;
- LASSERT(loop <= 2);
- obj = lmv_obj_grab(obd, &rid);
- if (obj) {
- int mea_idx;
-
- /* Directory is split. Look for right mds for this name */
- mea_idx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
- filename, namelen - 1);
- rid = obj->lo_inodes[mea_idx].li_fid;
- tgt_exp = lmv_get_export(lmv, obj->lo_inodes[mea_idx].li_mds);
- lmv_obj_put(obj);
- valid &= ~OBD_MD_FLCKSPLIT;
- } else {
- tgt_exp = lmv_find_export(lmv, &rid);
- valid |= OBD_MD_FLCKSPLIT;
- }
- if (IS_ERR(tgt_exp))
- RETURN(PTR_ERR(tgt_exp));
-
- CDEBUG(D_OTHER, "getattr_name for %*s on "DFID" -> "DFID"\n",
- namelen, filename, PFID(fid), PFID(&rid));
-
- rc = md_getattr_name(tgt_exp, &rid, oc, filename, namelen, valid,
- ea_size, suppgid, request);
- if (rc == 0) {
- body = req_capsule_server_get(&(*request)->rq_pill,
- &RMF_MDT_BODY);
- LASSERT(body != NULL);
-
- if (body->valid & OBD_MD_MDS) {
- struct ptlrpc_request *req = NULL;
-
- rid = body->fid1;
- CDEBUG(D_OTHER, "request attrs for "DFID"\n",
- PFID(&rid));
-
- tgt_exp = lmv_find_export(lmv, &rid);
- if (IS_ERR(tgt_exp)) {
- ptlrpc_req_finished(*request);
- RETURN(PTR_ERR(tgt_exp));
- }
-
- rc = md_getattr_name(tgt_exp, &rid, NULL, NULL, 1,
- valid | OBD_MD_FLCROSSREF,
- ea_size, suppgid, &req);
- ptlrpc_req_finished(*request);
- *request = req;
- }
- } else if (rc == -ERESTART) {
- LASSERT(*request != NULL);
- DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
- "Got -ERESTART during getattr!\n");
- ptlrpc_req_finished(*request);
- *request = NULL;
-
- /*
- * Directory got split. Time to update local object and repeat
- * the request with proper MDS.
- */
- rc = lmv_handle_split(exp, &rid);
- if (rc == 0)
- goto repeat;
- }
- RETURN(rc);
+ struct ptlrpc_request *req = NULL;
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lmv_tgt_desc *tgt;
+ struct mdt_body *body;
+ int rc;
+ ENTRY;
+
+ rc = lmv_check_connect(obd);
+ if (rc)
+ RETURN(rc);
+
+ tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
+ if (IS_ERR(tgt))
+ RETURN(PTR_ERR(tgt));
+
+ CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
+ op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
+ tgt->ltd_idx);
+
+ rc = md_getattr_name(tgt->ltd_exp, op_data, request);
+ if (rc != 0)
+ RETURN(rc);
+
+ body = req_capsule_server_get(&(*request)->rq_pill,
+ &RMF_MDT_BODY);
+ LASSERT(body != NULL);
+
+ if (body->valid & OBD_MD_MDS) {
+ struct lu_fid rid = body->fid1;
+ CDEBUG(D_INODE, "Request attrs for "DFID"\n",
+ PFID(&rid));
+
+ tgt = lmv_find_target(lmv, &rid);
+ if (IS_ERR(tgt)) {
+ ptlrpc_req_finished(*request);
+ RETURN(PTR_ERR(tgt));
+ }
+
+ op_data->op_fid1 = rid;
+ op_data->op_valid |= OBD_MD_FLCROSSREF;
+ op_data->op_namelen = 0;
+ op_data->op_name = NULL;
+ rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
+ ptlrpc_req_finished(*request);
+ *request = req;
+ }
+
+ RETURN(rc);
}
#define md_op_data_fid(op_data, fl) \
fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
NULL)
-/* @tgt_exp is the export the metadata request is sent.
- * @fid_exp is the export the cancel should be sent for the current fid.
- * if @fid_exp is NULL, the export is found for the current fid.
- * @op_data keeps the current fid, which is pointed through @flag.
- * @mode, @bits -- lock match parameters. */
-static int lmv_early_cancel(struct lmv_obd *lmv, struct obd_export *tgt_exp,
- struct obd_export *fid_exp,
- struct md_op_data *op_data,
- ldlm_mode_t mode, int bits, int flag)
+static int lmv_early_cancel(struct obd_export *exp, struct md_op_data *op_data,
+ int op_tgt, ldlm_mode_t mode, int bits, int flag)
{
- struct lu_fid *fid = md_op_data_fid(op_data, flag);
- ldlm_policy_data_t policy = {{0}};
- int rc = 0;
+ struct lu_fid *fid = md_op_data_fid(op_data, flag);
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lmv_tgt_desc *tgt;
+ ldlm_policy_data_t policy = {{0}};
+ int rc = 0;
ENTRY;
if (!fid_is_sane(fid))
RETURN(0);
- if (fid_exp == NULL)
- fid_exp = lmv_find_export(lmv, fid);
-
- if (tgt_exp == fid_exp) {
- /* The export is the same as on the target server, cancel
- * will be sent along with the main metadata operation. */
- op_data->op_flags |= flag;
- RETURN(0);
- }
-
- policy.l_inodebits.bits = bits;
- rc = md_cancel_unused(fid_exp, fid, &policy, mode, LDLM_FL_ASYNC, NULL);
- RETURN(rc);
-}
-
-#ifdef EARLY_CANCEL_FOR_STRIPED_DIR_IS_READY
-/* Check if the fid in @op_data pointed to by flag is of the same export(s)
- * as @tgt_exp. Early cancels will be sent later by mdc code, otherwise, call
- * md_cancel_unused for child export(s). */
-static int lmv_early_cancel_stripes(struct obd_export *exp,
- struct obd_export *tgt_exp,
- struct md_op_data *op_data,
- ldlm_mode_t mode, int bits, int flag)
-{
- struct lu_fid *fid = md_op_data_fid(op_data, flag);
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- struct obd_export *st_exp;
- struct lmv_obj *obj;
- int rc = 0;
- ENTRY;
-
- if (!fid_is_sane(fid))
- RETURN(0);
+ tgt = lmv_find_target(lmv, fid);
+ if (IS_ERR(tgt))
+ RETURN(PTR_ERR(tgt));
- obj = lmv_obj_grab(obd, fid);
- if (obj) {
- ldlm_policy_data_t policy = {{0}};
- struct lu_fid *st_fid;
- int i;
+ if (tgt->ltd_idx != op_tgt) {
+ CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
+ policy.l_inodebits.bits = bits;
+ rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
+ mode, LCF_ASYNC, NULL);
+ } else {
+ CDEBUG(D_INODE,
+ "EARLY_CANCEL skip operation target %d on "DFID"\n",
+ op_tgt, PFID(fid));
+ op_data->op_flags |= flag;
+ rc = 0;
+ }
- policy.l_inodebits.bits = bits;
- for (i = 0; i < obj->lo_objcount; i++) {
- st_exp = lmv_get_export(lmv, obj->lo_inodes[i].li_mds);
- st_fid = &obj->lo_inodes[i].li_fid;
- if (tgt_exp != st_exp) {
- rc = md_cancel_unused(st_exp, st_fid, &policy,
- mode, LDLM_FL_ASYNC,
- NULL);
- if (rc)
- break;
- } else {
- /* Some export matches to @tgt_exp, do cancel
- * for its fid in mdc */
- *fid = *st_fid;
- op_data->op_flags |= flag;
- }
- }
- lmv_obj_put(obj);
- } else {
- rc = lmv_early_cancel(lmv, tgt_exp, NULL, op_data,
- mode, bits, flag);
- }
- RETURN(rc);
+ RETURN(rc);
}
-#endif
/*
* llite passes fid of an target inode in op_data->op_fid1 and id of directory in
static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
struct ptlrpc_request **request)
{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- struct obd_export *tgt_exp;
- struct lmv_obj *obj;
- int rc, loop = 0;
- mdsno_t mds;
- ENTRY;
-
- rc = lmv_check_connect(obd);
- if (rc)
- RETURN(rc);
-
-repeat:
- ++loop;
- LASSERT(loop <= 2);
- if (op_data->op_namelen != 0) {
- int mea_idx;
-
- /* Usual link request */
- obj = lmv_obj_grab(obd, &op_data->op_fid2);
- if (obj) {
- mea_idx = raw_name2idx(obj->lo_hashtype,
- obj->lo_objcount,
- op_data->op_name,
- op_data->op_namelen);
- op_data->op_fid2 = obj->lo_inodes[mea_idx].li_fid;
- mds = obj->lo_inodes[mea_idx].li_mds;
- lmv_obj_put(obj);
- } else {
- rc = lmv_fld_lookup(lmv, &op_data->op_fid2, &mds);
- if (rc)
- RETURN(rc);
- }
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lmv_tgt_desc *tgt;
+ int rc;
+ ENTRY;
- CDEBUG(D_OTHER,"link "DFID":%*s to "DFID"\n",
- PFID(&op_data->op_fid2), op_data->op_namelen,
- op_data->op_name, PFID(&op_data->op_fid1));
- } else {
- rc = lmv_fld_lookup(lmv, &op_data->op_fid1, &mds);
- if (rc)
- RETURN(rc);
+ rc = lmv_check_connect(obd);
+ if (rc)
+ RETURN(rc);
- /* request from MDS to acquire i_links for inode by fid1 */
- CDEBUG(D_OTHER, "inc i_nlinks for "DFID"\n",
- PFID(&op_data->op_fid1));
- }
+ LASSERT(op_data->op_namelen != 0);
- CDEBUG(D_OTHER, "forward to MDS #"LPU64" ("DFID")\n",
- mds, PFID(&op_data->op_fid1));
+ CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
+ PFID(&op_data->op_fid2), op_data->op_namelen,
+ op_data->op_name, PFID(&op_data->op_fid1));
- op_data->op_fsuid = current->fsuid;
- op_data->op_fsgid = current->fsgid;
- op_data->op_cap = current->cap_effective;
+ op_data->op_fsuid = cfs_curproc_fsuid();
+ op_data->op_fsgid = cfs_curproc_fsgid();
+ op_data->op_cap = cfs_curproc_cap_pack();
+ tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
+ if (IS_ERR(tgt))
+ RETURN(PTR_ERR(tgt));
- tgt_exp = lmv->tgts[mds].ltd_exp;
- if (op_data->op_namelen) {
- op_data->op_flags |= MF_MDC_CANCEL_FID2;
- /* Cancel UPDATE lock on child (fid1). */
- rc = lmv_early_cancel(lmv, tgt_exp, NULL, op_data, LCK_EX,
- MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
- }
- if (rc == 0)
- rc = md_link(tgt_exp, op_data, request);
- if (rc == -ERESTART) {
- LASSERT(*request != NULL);
- DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
- "Got -ERESTART during link!\n");
- ptlrpc_req_finished(*request);
- *request = NULL;
+ /*
+ * Cancel UPDATE lock on child (fid1).
+ */
+ op_data->op_flags |= MF_MDC_CANCEL_FID2;
+ rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX,
+ MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
+ if (rc != 0)
+ RETURN(rc);
- /*
- * Directory got split. Time to update local object and repeat
- * the request with proper MDS.
- */
- rc = lmv_handle_split(exp, &op_data->op_fid2);
- if (rc == 0)
- goto repeat;
- }
+ rc = md_link(tgt->ltd_exp, op_data, request);
- RETURN(rc);
+ RETURN(rc);
}
static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
const char *old, int oldlen, const char *new, int newlen,
struct ptlrpc_request **request)
{
- struct obd_export *tgt_exp = NULL, *src_exp;
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- int rc, mea_idx, loop = 0;
- struct lmv_obj *obj;
- mdsno_t mds1, mds2;
- ENTRY;
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lmv_tgt_desc *src_tgt;
+ struct lmv_tgt_desc *tgt_tgt;
+ int rc;
+ ENTRY;
- CDEBUG(D_OTHER, "rename %*s in "DFID" to %*s in "DFID"\n",
+ LASSERT(oldlen != 0);
+
+ CDEBUG(D_INODE, "RENAME %*s in "DFID" to %*s in "DFID"\n",
oldlen, old, PFID(&op_data->op_fid1),
newlen, new, PFID(&op_data->op_fid2));
if (rc)
RETURN(rc);
- if (oldlen == 0) {
- /*
- * MDS with old dir entry is asking another MDS to create name
- * there.
- */
- CDEBUG(D_OTHER,
- "create %*s(%d/%d) in "DFID" pointing "
- "to "DFID"\n", newlen, new, oldlen, newlen,
- PFID(&op_data->op_fid2), PFID(&op_data->op_fid1));
-
- rc = lmv_fld_lookup(lmv, &op_data->op_fid2, &mds1);
- if (rc)
- RETURN(rc);
-
- /*
- * Target directory can be split, sowe should forward request to
- * the right MDS.
- */
- obj = lmv_obj_grab(obd, &op_data->op_fid2);
- if (obj) {
- mea_idx = raw_name2idx(obj->lo_hashtype,
- obj->lo_objcount,
- (char *)new, newlen);
- op_data->op_fid2 = obj->lo_inodes[mea_idx].li_fid;
- CDEBUG(D_OTHER, "Parent obj "DFID"\n",
- PFID(&op_data->op_fid2));
- lmv_obj_put(obj);
- }
- goto request;
- }
-
-repeat:
- ++loop;
- LASSERT(loop <= 2);
- obj = lmv_obj_grab(obd, &op_data->op_fid1);
- if (obj) {
- /*
- * directory is already split, so we have to forward request to
- * the right MDS.
- */
- mea_idx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
- (char *)old, oldlen);
- op_data->op_fid1 = obj->lo_inodes[mea_idx].li_fid;
- mds1 = obj->lo_inodes[mea_idx].li_mds;
- CDEBUG(D_OTHER, "Parent obj "DFID"\n", PFID(&op_data->op_fid1));
- lmv_obj_put(obj);
- } else {
- rc = lmv_fld_lookup(lmv, &op_data->op_fid1, &mds1);
- if (rc)
- RETURN(rc);
- }
-
- obj = lmv_obj_grab(obd, &op_data->op_fid2);
- if (obj) {
- /*
- * Directory is already split, so we have to forward request to
- * the right MDS.
- */
- mea_idx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
- (char *)new, newlen);
-
- mds2 = obj->lo_inodes[mea_idx].li_mds;
- op_data->op_fid2 = obj->lo_inodes[mea_idx].li_fid;
- CDEBUG(D_OTHER, "Parent obj "DFID"\n", PFID(&op_data->op_fid2));
- lmv_obj_put(obj);
- } else {
- rc = lmv_fld_lookup(lmv, &op_data->op_fid2, &mds2);
- if (rc)
- RETURN(rc);
- }
-
-request:
- op_data->op_fsuid = current->fsuid;
- op_data->op_fsgid = current->fsgid;
- op_data->op_cap = current->cap_effective;
-
- src_exp = lmv_get_export(lmv, mds1);
- tgt_exp = lmv_get_export(lmv, mds2);
- if (oldlen) {
- /* LOOKUP lock on src child (fid3) should also be cancelled for
- * src_exp in mdc_rename. */
- op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
-
- /* Cancel UPDATE locks on tgt parent (fid2), tgt_exp is its
- * own export. */
- rc = lmv_early_cancel(lmv, src_exp, tgt_exp, op_data, LCK_EX,
- MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
-
- /* Cancel LOOKUP locks on tgt child (fid4) for parent tgt_exp.*/
- if (rc == 0)
- rc = lmv_early_cancel(lmv, src_exp, tgt_exp, op_data,
- LCK_EX, MDS_INODELOCK_LOOKUP,
- MF_MDC_CANCEL_FID4);
-
- /* XXX: the case when child is a striped dir is not supported.
- * Only the master stripe has all locks cancelled early. */
- /* Cancel all the locks on tgt child (fid4). */
- if (rc == 0)
- rc = lmv_early_cancel(lmv, src_exp, NULL, op_data,
- LCK_EX, MDS_INODELOCK_FULL,
- MF_MDC_CANCEL_FID4);
- }
-
- if (rc == 0)
- rc = md_rename(src_exp, op_data, old, oldlen,
- new, newlen, request);
- if (rc == -ERESTART) {
- LASSERT(*request != NULL);
- DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
- "Got -ERESTART during rename!\n");
- ptlrpc_req_finished(*request);
- *request = NULL;
-
- /*
- * Directory got split. Time to update local object and repeat
- * the request with proper MDS.
- */
- rc = lmv_handle_split(exp, &op_data->op_fid1);
- if (rc == 0)
- goto repeat;
- }
- RETURN(rc);
+ op_data->op_fsuid = cfs_curproc_fsuid();
+ op_data->op_fsgid = cfs_curproc_fsgid();
+ op_data->op_cap = cfs_curproc_cap_pack();
+ src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
+ if (IS_ERR(src_tgt))
+ RETURN(PTR_ERR(src_tgt));
+
+ tgt_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
+ if (IS_ERR(tgt_tgt))
+ RETURN(PTR_ERR(tgt_tgt));
+ /*
+ * LOOKUP lock on src child (fid3) should also be cancelled for
+ * src_tgt in mdc_rename.
+ */
+ op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
+
+ /*
+ * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
+ * own target.
+ */
+ rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
+ LCK_EX, MDS_INODELOCK_UPDATE,
+ MF_MDC_CANCEL_FID2);
+
+ /*
+ * Cancel LOOKUP locks on tgt child (fid4) for parent tgt_tgt.
+ */
+ if (rc == 0) {
+ rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
+ LCK_EX, MDS_INODELOCK_LOOKUP,
+ MF_MDC_CANCEL_FID4);
+ }
+
+ /*
+ * Cancel all the locks on tgt child (fid4).
+ */
+ if (rc == 0)
+ rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
+ LCK_EX, MDS_INODELOCK_FULL,
+ MF_MDC_CANCEL_FID4);
+
+ if (rc == 0)
+ rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen,
+ new, newlen, request);
+ RETURN(rc);
}
static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
struct ptlrpc_request **request,
struct md_open_data **mod)
{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- struct ptlrpc_request *req;
- struct obd_export *tgt_exp;
- struct lmv_obj *obj;
- int rc = 0, i;
- ENTRY;
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lmv_tgt_desc *tgt;
+ int rc = 0;
+ ENTRY;
- rc = lmv_check_connect(obd);
- if (rc)
- RETURN(rc);
+ rc = lmv_check_connect(obd);
+ if (rc)
+ RETURN(rc);
- obj = lmv_obj_grab(obd, &op_data->op_fid1);
+ CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
+ PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
- CDEBUG(D_OTHER, "SETATTR for "DFID", valid 0x%x%s\n",
- PFID(&op_data->op_fid1), op_data->op_attr.ia_valid,
- obj ? ", split" : "");
+ op_data->op_flags |= MF_MDC_CANCEL_FID1;
+ tgt = lmv_find_target(lmv, &op_data->op_fid1);
+ if (IS_ERR(tgt))
+ RETURN(PTR_ERR(tgt));
- op_data->op_flags |= MF_MDC_CANCEL_FID1;
- if (obj) {
- for (i = 0; i < obj->lo_objcount; i++) {
- op_data->op_fid1 = obj->lo_inodes[i].li_fid;
+ rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
+ ea2len, request, mod);
- tgt_exp = lmv_get_export(lmv, obj->lo_inodes[i].li_mds);
- if (IS_ERR(tgt_exp)) {
- rc = PTR_ERR(tgt_exp);
- break;
- }
-
- rc = md_setattr(tgt_exp, op_data, ea, ealen,
- ea2, ea2len, &req, mod);
-
- if (lu_fid_eq(&obj->lo_fid, &obj->lo_inodes[i].li_fid)) {
- /*
- * this is master object and this request should
- * be returned back to llite.
- */
- *request = req;
- } else {
- ptlrpc_req_finished(req);
- }
-
- if (rc)
- break;
- }
- lmv_obj_put(obj);
- } else {
- tgt_exp = lmv_find_export(lmv, &op_data->op_fid1);
- if (IS_ERR(tgt_exp))
- RETURN(PTR_ERR(tgt_exp));
-
- rc = md_setattr(tgt_exp, op_data, ea, ealen, ea2,
- ea2len, request, mod);
- }
- RETURN(rc);
+ RETURN(rc);
}
static int lmv_sync(struct obd_export *exp, const struct lu_fid *fid,
struct obd_capa *oc, struct ptlrpc_request **request)
{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- struct obd_export *tgt_exp;
- int rc;
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lmv_tgt_desc *tgt;
+ int rc;
ENTRY;
rc = lmv_check_connect(obd);
if (rc)
RETURN(rc);
- tgt_exp = lmv_find_export(lmv, fid);
- if (IS_ERR(tgt_exp))
- RETURN(PTR_ERR(tgt_exp));
+ tgt = lmv_find_target(lmv, fid);
+ if (IS_ERR(tgt))
+ RETURN(PTR_ERR(tgt));
- rc = md_sync(tgt_exp, fid, oc, request);
+ rc = md_sync(tgt->ltd_exp, fid, oc, request);
RETURN(rc);
}
-/* main purpose of LMV blocking ast is to remove split directory LMV
- * presentation object (struct lmv_obj) attached to the lock being revoked. */
-int lmv_blocking_ast(struct ldlm_lock *lock,
- struct ldlm_lock_desc *desc,
- void *data, int flag)
-{
- struct lustre_handle lockh;
- struct lmv_obj *obj;
- int rc;
- ENTRY;
-
- switch (flag) {
- case LDLM_CB_BLOCKING:
- ldlm_lock2handle(lock, &lockh);
- rc = ldlm_cli_cancel(&lockh);
- if (rc < 0) {
- CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
- RETURN(rc);
- }
- break;
- case LDLM_CB_CANCELING:
- /* time to drop cached attrs for dirobj */
- obj = lock->l_ast_data;
- if (obj) {
- CDEBUG(D_OTHER, "cancel %s on "LPU64"/"LPU64
- ", master "DFID"\n",
- lock->l_resource->lr_name.name[3] == 1 ?
- "LOOKUP" : "UPDATE",
- lock->l_resource->lr_name.name[0],
- lock->l_resource->lr_name.name[1],
- PFID(&obj->lo_fid));
- lmv_obj_put(obj);
- }
- break;
- default:
- LBUG();
- }
- RETURN(0);
-}
-
-static void lmv_hash_adjust(__u64 *hash, __u64 hash_adj)
-{
- __u64 val;
-
- val = le64_to_cpu(*hash);
- if (val < hash_adj)
- val += MAX_HASH_SIZE;
- if (val != DIR_END_OFF)
- *hash = cpu_to_le64(val - hash_adj);
-}
-
-static __u32 lmv_node_rank(struct obd_export *exp, const struct lu_fid *fid)
-{
- __u64 id;
- struct obd_import *imp;
-
- /*
- * XXX Hack: to get nid we assume that underlying obd device is mdc.
- */
- imp = class_exp2cliimp(exp);
- id = imp->imp_connection->c_self + fid_flatten(fid);
-
- CDEBUG(D_INFO, "node rank: "LPX64" "DFID" "LPX64" "LPX64"\n",
- imp->imp_connection->c_self, PFID(fid), id, id ^ (id >> 32));
-
- return id ^ (id >> 32);
-}
-
-static int lmv_readpage(struct obd_export *exp, const struct lu_fid *fid,
- struct obd_capa *oc, __u64 offset64, struct page *page,
- struct ptlrpc_request **request)
-{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- struct obd_export *tgt_exp;
- struct lu_fid rid = *fid;
- struct lmv_obj *obj;
- __u64 offset;
- __u64 hash_adj = 0;
- __u32 rank = 0;
- __u64 seg_size = 0;
- __u64 tgt_tmp = 0;
- int tgt = 0;
- int tgt0 = 0;
- int rc;
- int nr = 0;
- ENTRY;
-
- offset = offset64;
-
- rc = lmv_check_connect(obd);
- if (rc)
- RETURN(rc);
-
- CDEBUG(D_INFO, "READPAGE at "LPX64" from "DFID"\n", offset, PFID(&rid));
-
- obj = lmv_obj_grab(obd, fid);
- if (obj) {
-
- /*
- * This case handle directory lookup in clustered metadata case (i.e.
- * split directory is located on multiple md servers.)
- * each server keeps directory entries for certain range of hashes.
- * E.g. we have N server and suppose hash range is 0 to MAX_HASH.
- * first server will keep records with hashes [ 0 ... MAX_HASH / N - 1],
- * second one with hashes [MAX_HASH / N ... 2 * MAX_HASH / N] and
- * so on....
- * readdir can simply start reading entries from 0 - N server in
- * order but that will not scale well as all client will request dir in
- * to server in same order.
- * Following algorithm does optimization:
- * Instead of doing readdir in 1, 2, ...., N order, client with a
- * rank R does readdir in R, R + 1, ..., N, 1, ... R - 1 order.
- * (every client has rank R)
- * But ll_readdir() expect offset range [0 to MAX_HASH/N) but
- * since client ask dir from MDS{R} client has pages with offsets
- * [R*MAX_HASH/N ... (R + 1)*MAX_HASH/N] there for we do hash_adj
- * on hash values that we get.
- */
-
- struct lmv_inode *loi;
-
- lmv_obj_lock(obj);
-
- nr = obj->lo_objcount;
- LASSERT(nr > 0);
- seg_size = MAX_HASH_SIZE;
- do_div(seg_size, nr);
- loi = obj->lo_inodes;
- rank = lmv_node_rank(lmv_get_export(lmv, loi[0].li_mds),
- fid) % nr;
- tgt_tmp = offset;
- do_div(tgt_tmp, seg_size);
- tgt0 = do_div(tgt_tmp, nr);
- tgt = (tgt0 + rank) % nr;
-
- if (tgt < tgt0)
- /*
- * Wrap around.
- *
- * Last segment has unusual length due to division
- * rounding.
- */
- hash_adj = MAX_HASH_SIZE - seg_size * nr;
- else
- hash_adj = 0;
-
- hash_adj += rank * seg_size;
-
- CDEBUG(D_INFO, "hash_adj: %x "LPX64" "LPX64"/%x -> "LPX64"/%x\n",
- rank, hash_adj, offset, tgt0, offset + hash_adj, tgt);
-
- offset = (offset + hash_adj) & MAX_HASH_SIZE;
- rid = obj->lo_inodes[tgt].li_fid;
- tgt_exp = lmv_get_export(lmv, loi[tgt].li_mds);
-
- CDEBUG(D_INFO, "forward to "DFID" with offset %lu i %d\n",
- PFID(&rid), (unsigned long)offset, tgt);
- } else
- tgt_exp = lmv_find_export(lmv, &rid);
-
- if (IS_ERR(tgt_exp))
- GOTO(cleanup, rc = PTR_ERR(tgt_exp));
-
- rc = md_readpage(tgt_exp, &rid, oc, offset, page, request);
- if (rc)
- GOTO(cleanup, rc);
- if (obj) {
- struct lu_dirpage *dp;
- struct lu_dirent *ent;
-
- dp = cfs_kmap(page);
-
- lmv_hash_adjust(&dp->ldp_hash_start, hash_adj);
- lmv_hash_adjust(&dp->ldp_hash_end, hash_adj);
- LASSERT(le64_to_cpu(dp->ldp_hash_start) <= offset64);
-
- for (ent = lu_dirent_start(dp); ent != NULL;
- ent = lu_dirent_next(ent))
- lmv_hash_adjust(&ent->lde_hash, hash_adj);
-
- if (tgt0 != nr - 1) {
- __u64 end;
-
- end = le64_to_cpu(dp->ldp_hash_end);
- if (end == DIR_END_OFF) {
- dp->ldp_hash_end = cpu_to_le32(seg_size *
- (tgt0 + 1));
- CDEBUG(D_INFO,
- ""DFID" reset end "LPX64" tgt %d\n",
- PFID(&rid),
- le64_to_cpu(dp->ldp_hash_end), tgt);
- }
- }
- cfs_kunmap(page);
- }
- /*
- * Here we could remove "." and ".." from all pages which at not from
- * master. But MDS has only "." and ".." for master dir.
- */
- EXIT;
-cleanup:
- if (obj) {
- lmv_obj_unlock(obj);
- lmv_obj_put(obj);
- }
- return rc;
+/*
+ * Adjust a set of pages, each page containing an array of lu_dirpages,
+ * so that each page can be used as a single logical lu_dirpage.
+ *
+ * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
+ * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
+ * struct lu_dirent. It has size up to LU_PAGE_SIZE. The ldp_hash_end
+ * value is used as a cookie to request the next lu_dirpage in a
+ * directory listing that spans multiple pages (two in this example):
+ * ________
+ * | |
+ * .|--------v------- -----.
+ * |s|e|f|p|ent|ent| ... |ent|
+ * '--|-------------- -----' Each CFS_PAGE contains a single
+ * '------. lu_dirpage.
+ * .---------v------- -----.
+ * |s|e|f|p|ent| 0 | ... | 0 |
+ * '----------------- -----'
+ *
+ * However, on hosts where the native VM page size (PAGE_CACHE_SIZE) is
+ * larger than LU_PAGE_SIZE, a single host page may contain multiple
+ * lu_dirpages. After reading the lu_dirpages from the MDS, the
+ * ldp_hash_end of the first lu_dirpage refers to the one immediately
+ * after it in the same CFS_PAGE (arrows simplified for brevity, but
+ * in general e0==s1, e1==s2, etc.):
+ *
+ * .-------------------- -----.
+ * |s0|e0|f0|p|ent|ent| ... |ent|
+ * |---v---------------- -----|
+ * |s1|e1|f1|p|ent|ent| ... |ent|
+ * |---v---------------- -----| Here, each CFS_PAGE contains
+ * ... multiple lu_dirpages.
+ * |---v---------------- -----|
+ * |s'|e'|f'|p|ent|ent| ... |ent|
+ * '---|---------------- -----'
+ * v
+ * .----------------------------.
+ * | next CFS_PAGE |
+ *
+ * This structure is transformed into a single logical lu_dirpage as follows:
+ *
+ * - Replace e0 with e' so the request for the next lu_dirpage gets the page
+ * labeled 'next CFS_PAGE'.
+ *
+ * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
+ * a hash collision with the next page exists.
+ *
+ * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
+ * to the first entry of the next lu_dirpage.
+ */
+#if PAGE_CACHE_SIZE > LU_PAGE_SIZE
+static void lmv_adjust_dirpages(struct page **pages, int ncfspgs, int nlupgs)
+{
+ int i;
+
+ for (i = 0; i < ncfspgs; i++) {
+ struct lu_dirpage *dp = kmap(pages[i]);
+ struct lu_dirpage *first = dp;
+ struct lu_dirent *end_dirent = NULL;
+ struct lu_dirent *ent;
+ __u64 hash_end = dp->ldp_hash_end;
+ __u32 flags = dp->ldp_flags;
+
+ for (; nlupgs > 1; nlupgs--) {
+ ent = lu_dirent_start(dp);
+ for (end_dirent = ent; ent != NULL;
+ end_dirent = ent, ent = lu_dirent_next(ent));
+
+ /* Advance dp to next lu_dirpage. */
+ dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
+
+ /* Check if we've reached the end of the CFS_PAGE. */
+ if (!((unsigned long)dp & ~CFS_PAGE_MASK))
+ break;
+
+ /* Save the hash and flags of this lu_dirpage. */
+ hash_end = dp->ldp_hash_end;
+ flags = dp->ldp_flags;
+
+ /* Check if lu_dirpage contains no entries. */
+ if (!end_dirent)
+ break;
+
+ /* Enlarge the end entry lde_reclen from 0 to
+ * first entry of next lu_dirpage. */
+ LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
+ end_dirent->lde_reclen =
+ cpu_to_le16((char *)(dp->ldp_entries) -
+ (char *)end_dirent);
+ }
+
+ first->ldp_hash_end = hash_end;
+ first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
+ first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
+
+ kunmap(pages[i]);
+ }
}
+#else
+#define lmv_adjust_dirpages(pages, ncfspgs, nlupgs) do {} while (0)
+#endif /* PAGE_CACHE_SIZE > LU_PAGE_SIZE */
-static int lmv_unlink_slaves(struct obd_export *exp,
- struct md_op_data *op_data,
- struct ptlrpc_request **req)
+static int lmv_readpage(struct obd_export *exp, struct md_op_data *op_data,
+ struct page **pages, struct ptlrpc_request **request)
{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- struct lmv_stripe_md *mea = op_data->op_mea1;
- struct md_op_data *op_data2;
- struct obd_export *tgt_exp;
- int i, rc = 0;
- ENTRY;
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ __u64 offset = op_data->op_offset;
+ int rc;
+ int ncfspgs; /* pages read in PAGE_CACHE_SIZE */
+ int nlupgs; /* pages read in LU_PAGE_SIZE */
+ struct lmv_tgt_desc *tgt;
+ ENTRY;
- OBD_ALLOC_PTR(op_data2);
- if (op_data2 == NULL)
- RETURN(-ENOMEM);
+ rc = lmv_check_connect(obd);
+ if (rc)
+ RETURN(rc);
- op_data2->op_mode = S_IFDIR;
- op_data2->op_fsuid = current->fsuid;
- op_data2->op_fsgid = current->fsgid;
- op_data2->op_bias = 0;
+ CDEBUG(D_INODE, "READPAGE at "LPX64" from "DFID"\n",
+ offset, PFID(&op_data->op_fid1));
- LASSERT(mea != NULL);
- for (i = 0; i < mea->mea_count; i++) {
- memset(op_data2, 0, sizeof(*op_data2));
- op_data2->op_fid1 = mea->mea_ids[i];
- tgt_exp = lmv_find_export(lmv, &op_data2->op_fid1);
- if (IS_ERR(tgt_exp))
- GOTO(out_free_op_data2, rc = PTR_ERR(tgt_exp));
+ tgt = lmv_find_target(lmv, &op_data->op_fid1);
+ if (IS_ERR(tgt))
+ RETURN(PTR_ERR(tgt));
- if (tgt_exp == NULL)
- continue;
+ rc = md_readpage(tgt->ltd_exp, op_data, pages, request);
+ if (rc != 0)
+ RETURN(rc);
- rc = md_unlink(tgt_exp, op_data2, req);
+ ncfspgs = ((*request)->rq_bulk->bd_nob_transferred +
+ PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
+ nlupgs = (*request)->rq_bulk->bd_nob_transferred >> LU_PAGE_SHIFT;
+ LASSERT(!((*request)->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
+ LASSERT(ncfspgs > 0 && ncfspgs <= op_data->op_npages);
- CDEBUG(D_OTHER, "unlink slave "DFID" -> %d\n",
- PFID(&mea->mea_ids[i]), rc);
+ CDEBUG(D_INODE, "read %d(%d)/%d pages\n", ncfspgs, nlupgs,
+ op_data->op_npages);
- if (*req) {
- ptlrpc_req_finished(*req);
- *req = NULL;
- }
- if (rc)
- GOTO(out_free_op_data2, rc);
- }
+ lmv_adjust_dirpages(pages, ncfspgs, nlupgs);
- EXIT;
-out_free_op_data2:
- OBD_FREE_PTR(op_data2);
- return rc;
+ RETURN(rc);
}
static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
struct ptlrpc_request **request)
{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- struct obd_export *tgt_exp = NULL;
- struct lmv_obj *obj;
- int rc, loop = 0;
- ENTRY;
-
- rc = lmv_check_connect(obd);
- if (rc)
- RETURN(rc);
-
- if (op_data->op_namelen == 0 && op_data->op_mea1 != NULL) {
- /* mds asks to remove slave objects */
- rc = lmv_unlink_slaves(exp, op_data, request);
- RETURN(rc);
- }
-
-repeat:
- ++loop;
- LASSERT(loop <= 2);
- if (op_data->op_namelen != 0) {
- int mea_idx;
-
- obj = lmv_obj_grab(obd, &op_data->op_fid1);
- if (obj) {
- mea_idx = raw_name2idx(obj->lo_hashtype,
- obj->lo_objcount,
- op_data->op_name,
- op_data->op_namelen);
- op_data->op_bias &= ~MDS_CHECK_SPLIT;
- op_data->op_fid1 = obj->lo_inodes[mea_idx].li_fid;
- tgt_exp = lmv_get_export(lmv,
- obj->lo_inodes[mea_idx].li_mds);
- lmv_obj_put(obj);
- CDEBUG(D_OTHER, "unlink '%*s' in "DFID" -> %u\n",
- op_data->op_namelen, op_data->op_name,
- PFID(&op_data->op_fid1), mea_idx);
- }
- } else {
- CDEBUG(D_OTHER, "drop i_nlink on "DFID"\n",
- PFID(&op_data->op_fid1));
- }
- if (tgt_exp == NULL) {
- tgt_exp = lmv_find_export(lmv, &op_data->op_fid1);
- if (IS_ERR(tgt_exp))
- RETURN(PTR_ERR(tgt_exp));
- op_data->op_bias |= MDS_CHECK_SPLIT;
- }
-
- op_data->op_fsuid = current->fsuid;
- op_data->op_fsgid = current->fsgid;
- op_data->op_cap = current->cap_effective;
-
- /* If child's fid is given, cancel unused locks for it if it is from
- * another export than parent. */
- if (op_data->op_namelen) {
- /* LOOKUP lock for child (fid3) should also be cancelled on
- * parent tgt_exp in mdc_unlink(). */
- op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
-
- /* XXX: the case when child is a striped dir is not supported.
- * Only the master stripe has all locks cancelled early. */
- /* Cancel FULL locks on child (fid3). */
- rc = lmv_early_cancel(lmv, tgt_exp, NULL, op_data, LCK_EX,
- MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
- }
- if (rc == 0)
- rc = md_unlink(tgt_exp, op_data, request);
- if (rc == -ERESTART) {
- LASSERT(*request != NULL);
- DEBUG_REQ(D_WARNING|D_RPCTRACE, *request,
- "Got -ERESTART during unlink!\n");
- ptlrpc_req_finished(*request);
- *request = NULL;
-
- /*
- * Directory got split. Time to update local object and repeat
- * the request with proper MDS.
- */
- rc = lmv_handle_split(exp, &op_data->op_fid1);
- if (rc == 0)
- goto repeat;
- }
- RETURN(rc);
-}
-
-static int lmv_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
- struct obd_device *tgt, int count,
- struct llog_catid *logid, struct obd_uuid *uuid)
-{
-#if 0
- struct llog_ctxt *ctxt;
- int rc;
- ENTRY;
-
- LASSERT(group == OBD_LLOG_GROUP);
- rc = llog_setup(obd, &obd->obd_olg, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
- &llog_client_ops);
- if (rc == 0) {
- ctxt = llog_group_get_ctxt(&obd->obd_olg, LLOG_CONFIG_REPL_CTXT);
- llog_initiator_connect(ctxt, tgt);
- llog_ctxt_put(ctxt);
- }
- RETURN(rc);
-#else
- return 0;
-#endif
-}
-
-static int lmv_llog_finish(struct obd_device *obd, int count)
-{
- struct llog_ctxt *ctxt;
- int rc = 0;
- ENTRY;
-
- ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
- if (ctxt)
- rc = llog_cleanup(ctxt);
-
- RETURN(rc);
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lmv_tgt_desc *tgt = NULL;
+ struct mdt_body *body;
+ int rc;
+ ENTRY;
+
+ rc = lmv_check_connect(obd);
+ if (rc)
+ RETURN(rc);
+retry:
+ /* Send unlink requests to the MDT where the child is located */
+ if (likely(!fid_is_zero(&op_data->op_fid2)))
+ tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
+ else
+ tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
+ if (IS_ERR(tgt))
+ RETURN(PTR_ERR(tgt));
+
+ op_data->op_fsuid = cfs_curproc_fsuid();
+ op_data->op_fsgid = cfs_curproc_fsgid();
+ op_data->op_cap = cfs_curproc_cap_pack();
+
+ /*
+ * If child's fid is given, cancel unused locks for it if it is from
+ * another export than parent.
+ *
+ * LOOKUP lock for child (fid3) should also be cancelled on parent
+ * tgt_tgt in mdc_unlink().
+ */
+ op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
+
+ /*
+ * Cancel FULL locks on child (fid3).
+ */
+ rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX,
+ MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
+
+ if (rc != 0)
+ RETURN(rc);
+
+ CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%d\n",
+ PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
+
+ rc = md_unlink(tgt->ltd_exp, op_data, request);
+ if (rc != 0 && rc != -EREMOTE)
+ RETURN(rc);
+
+ body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
+ if (body == NULL)
+ RETURN(-EPROTO);
+
+ /* Not cross-ref case, just get out of here. */
+ if (likely(!(body->valid & OBD_MD_MDS)))
+ RETURN(0);
+
+ CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
+ exp->exp_obd->obd_name, PFID(&body->fid1));
+
+ /* This is a remote object, try remote MDT, Note: it may
+ * try more than 1 time here, Considering following case
+ * /mnt/lustre is root on MDT0, remote1 is on MDT1
+ * 1. Initially A does not know where remote1 is, it send
+ * unlink RPC to MDT0, MDT0 return -EREMOTE, it will
+ * resend unlink RPC to MDT1 (retry 1st time).
+ *
+ * 2. During the unlink RPC in flight,
+ * client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
+ * and create new remote1, but on MDT0
+ *
+ * 3. MDT1 get unlink RPC(from A), then do remote lock on
+ * /mnt/lustre, then lookup get fid of remote1, and find
+ * it is remote dir again, and replay -EREMOTE again.
+ *
+ * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
+ *
+ * In theory, it might try unlimited time here, but it should
+ * be very rare case. */
+ op_data->op_fid2 = body->fid1;
+ ptlrpc_req_finished(*request);
+ *request = NULL;
+
+ goto retry;
}
static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
{
+ struct lmv_obd *lmv = &obd->u.lmv;
int rc = 0;
switch (stage) {
* stack. */
break;
case OBD_CLEANUP_EXPORTS:
- rc = obd_llog_finish(obd, 0);
- if (rc != 0)
- CERROR("failed to cleanup llogging subsystems\n");
+ fld_client_proc_fini(&lmv->lmv_fld);
+ lprocfs_obd_cleanup(obd);
break;
default:
break;
RETURN(rc);
}
-static int lmv_get_info(struct obd_export *exp, __u32 keylen,
- void *key, __u32 *vallen, void *val, struct lov_stripe_md *lsm)
+static int lmv_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 obd_device *obd;
- struct lmv_obd *lmv;
- int rc = 0;
+ struct obd_device *obd;
+ struct lmv_obd *lmv;
+ int rc = 0;
ENTRY;
obd = class_exp2obd(exp);
if (obd == NULL) {
- CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
+ CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
exp->exp_handle.h_cookie);
RETURN(-EINVAL);
}
lmv = &obd->u.lmv;
if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
- struct lmv_tgt_desc *tgts;
+ struct lmv_tgt_desc *tgt;
int i;
rc = lmv_check_connect(obd);
RETURN(rc);
LASSERT(*vallen == sizeof(__u32));
- for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count;
- i++, tgts++) {
-
- /* all tgts should be connected when this get called. */
- if (!tgts || !tgts->ltd_exp) {
- CERROR("target not setup?\n");
- continue;
- }
-
- if (!obd_get_info(tgts->ltd_exp, keylen, key,
- vallen, val, NULL))
- RETURN(0);
+ for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
+ tgt = lmv->tgts[i];
+ /*
+ * All tgts should be connected when this gets called.
+ */
+ if (tgt == NULL || tgt->ltd_exp == NULL)
+ continue;
+
+ if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
+ vallen, val, NULL))
+ RETURN(0);
}
RETURN(-EINVAL);
} else if (KEY_IS(KEY_MAX_EASIZE) || KEY_IS(KEY_CONN_DATA)) {
if (rc)
RETURN(rc);
- /* forwarding this request to first MDS, it should know LOV
- * desc. */
- rc = obd_get_info(lmv->tgts[0].ltd_exp, keylen, key,
- vallen, val, NULL);
- if (!rc && KEY_IS(KEY_CONN_DATA)) {
- exp->exp_connect_flags =
- ((struct obd_connect_data *)val)->ocd_connect_flags;
- }
+ /*
+ * Forwarding this request to first MDS, it should know LOV
+ * desc.
+ */
+ rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
+ vallen, val, NULL);
+ if (!rc && KEY_IS(KEY_CONN_DATA))
+ exp->exp_connect_data = *(struct obd_connect_data *)val;
RETURN(rc);
+ } else if (KEY_IS(KEY_TGT_COUNT)) {
+ *((int *)val) = lmv->desc.ld_tgt_count;
+ RETURN(0);
}
- CDEBUG(D_IOCTL, "invalid key\n");
+ CDEBUG(D_IOCTL, "Invalid key\n");
RETURN(-EINVAL);
}
-int lmv_set_info_async(struct obd_export *exp, obd_count keylen,
- void *key, obd_count vallen, void *val,
- struct ptlrpc_request_set *set)
+int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
+ obd_count keylen, void *key, obd_count vallen,
+ void *val, struct ptlrpc_request_set *set)
{
struct lmv_tgt_desc *tgt;
struct obd_device *obd;
obd = class_exp2obd(exp);
if (obd == NULL) {
- CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
+ CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
exp->exp_handle.h_cookie);
RETURN(-EINVAL);
}
lmv = &obd->u.lmv;
- if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
- KEY_IS(KEY_INIT_RECOV_BACKUP)) {
+ if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) {
int i, err = 0;
- for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
- tgt = &lmv->tgts[i];
+ for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
+ tgt = lmv->tgts[i];
- if (!tgt->ltd_exp)
- continue;
+ if (tgt == NULL || tgt->ltd_exp == NULL)
+ continue;
- err = obd_set_info_async(tgt->ltd_exp,
+ err = obd_set_info_async(env, tgt->ltd_exp,
keylen, key, vallen, val, set);
if (err && rc == 0)
rc = err;
int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
struct lov_stripe_md *lsm)
{
- struct obd_device *obd = class_exp2obd(exp);
- struct lmv_obd *lmv = &obd->u.lmv;
- struct lmv_stripe_md *meap, *lsmp;
- int mea_size, i;
+ struct obd_device *obd = class_exp2obd(exp);
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lmv_stripe_md *meap;
+ struct lmv_stripe_md *lsmp;
+ int mea_size;
+ int i;
ENTRY;
mea_size = lmv_get_easize(lmv);
RETURN(mea_size);
if (*lmmp && !lsm) {
- OBD_FREE(*lmmp, mea_size);
+ OBD_FREE_LARGE(*lmmp, mea_size);
*lmmp = NULL;
RETURN(0);
}
if (*lmmp == NULL) {
- OBD_ALLOC(*lmmp, mea_size);
+ OBD_ALLOC_LARGE(*lmmp, mea_size);
if (*lmmp == NULL)
RETURN(-ENOMEM);
}
meap->mea_count = cpu_to_le32(lsmp->mea_count);
meap->mea_master = cpu_to_le32(lsmp->mea_master);
- for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
- meap->mea_ids[i] = meap->mea_ids[i];
- fid_cpu_to_le(&meap->mea_ids[i], &meap->mea_ids[i]);
- }
+ for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
+ meap->mea_ids[i] = lsmp->mea_ids[i];
+ fid_cpu_to_le(&meap->mea_ids[i], &lsmp->mea_ids[i]);
+ }
RETURN(mea_size);
}
int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
struct lov_mds_md *lmm, int lmm_size)
{
- struct obd_device *obd = class_exp2obd(exp);
- struct lmv_stripe_md **tmea = (struct lmv_stripe_md **)lsmp;
- struct lmv_stripe_md *mea = (struct lmv_stripe_md *)lmm;
- struct lmv_obd *lmv = &obd->u.lmv;
- int mea_size, i;
- __u32 magic;
+ struct obd_device *obd = class_exp2obd(exp);
+ struct lmv_stripe_md **tmea = (struct lmv_stripe_md **)lsmp;
+ struct lmv_stripe_md *mea = (struct lmv_stripe_md *)lmm;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ int mea_size;
+ int i;
+ __u32 magic;
ENTRY;
mea_size = lmv_get_easize(lmv);
return mea_size;
if (*lsmp != NULL && lmm == NULL) {
- OBD_FREE(*tmea, mea_size);
+ OBD_FREE_LARGE(*tmea, mea_size);
*lsmp = NULL;
RETURN(0);
}
LASSERT(mea_size == lmm_size);
- OBD_ALLOC(*tmea, mea_size);
+ OBD_ALLOC_LARGE(*tmea, mea_size);
if (*tmea == NULL)
RETURN(-ENOMEM);
{
magic = le32_to_cpu(mea->mea_magic);
} else {
- /* old mea is not handled here */
+ /*
+ * Old mea is not handled here.
+ */
+ CERROR("Old not supportable EA is found\n");
LBUG();
}
RETURN(mea_size);
}
-static int lmv_cancel_unused(struct obd_export *exp,
- const struct lu_fid *fid,
- ldlm_policy_data_t *policy,
- ldlm_mode_t mode, int flags, void *opaque)
+static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
+ ldlm_policy_data_t *policy, ldlm_mode_t mode,
+ ldlm_cancel_flags_t flags, void *opaque)
{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- int rc = 0, err, i;
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ int rc = 0;
+ int err;
+ int i;
ENTRY;
LASSERT(fid != NULL);
for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
- if (!lmv->tgts[i].ltd_exp || !lmv->tgts[i].ltd_active)
- continue;
+ if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL ||
+ lmv->tgts[i]->ltd_active == 0)
+ continue;
- err = md_cancel_unused(lmv->tgts[i].ltd_exp, fid,
- policy, mode, flags, opaque);
- if (!rc)
- rc = err;
+ err = md_cancel_unused(lmv->tgts[i]->ltd_exp, fid,
+ policy, mode, flags, opaque);
+ if (!rc)
+ rc = err;
}
- RETURN(rc);
+ RETURN(rc);
}
-int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data)
+int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
+ __u64 *bits)
{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
+ struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
+ int rc;
+ ENTRY;
- ENTRY;
- RETURN(md_set_lock_data(lmv->tgts[0].ltd_exp, lockh, data));
+ rc = md_set_lock_data(lmv->tgts[0]->ltd_exp, lockh, data, bits);
+ RETURN(rc);
}
-ldlm_mode_t lmv_lock_match(struct obd_export *exp, int flags,
+ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
const struct lu_fid *fid, ldlm_type_t type,
ldlm_policy_data_t *policy, ldlm_mode_t mode,
struct lustre_handle *lockh)
{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- ldlm_mode_t rc;
- int i;
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ ldlm_mode_t rc;
+ int i;
ENTRY;
- CDEBUG(D_OTHER, "lock match for "DFID"\n", PFID(fid));
+ CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
- /* with CMD every object can have two locks in different namespaces:
+ /*
+ * With CMD every object can have two locks in different namespaces:
* lookup lock in space of mds storing direntry and update/open lock in
* space of mds storing inode. Thus we check all targets, not only that
- * one fid was created in. */
+ * one fid was created in.
+ */
for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
- rc = md_lock_match(lmv->tgts[i].ltd_exp, flags, fid,
+ if (lmv->tgts[i] == NULL ||
+ lmv->tgts[i]->ltd_exp == NULL ||
+ lmv->tgts[i]->ltd_active == 0)
+ continue;
+
+ rc = md_lock_match(lmv->tgts[i]->ltd_exp, flags, fid,
type, policy, mode, lockh);
if (rc)
RETURN(rc);
}
int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
- struct obd_export *dt_exp, struct obd_export *md_exp,
- struct lustre_md *md)
+ struct obd_export *dt_exp, struct obd_export *md_exp,
+ struct lustre_md *md)
{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- int rc;
+ struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
- ENTRY;
- rc = md_get_lustre_md(lmv->tgts[0].ltd_exp, req, dt_exp, md_exp, md);
- RETURN(rc);
+ return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
}
int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
-
- ENTRY;
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ ENTRY;
- /* XXX LOV STACKING */
- if (md->mea)
- obd_free_memmd(exp, (void *)&md->mea);
- RETURN(md_free_lustre_md(lmv->tgts[0].ltd_exp, md));
+ if (md->mea)
+ obd_free_memmd(exp, (void *)&md->mea);
+ RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
}
int lmv_set_open_replay_data(struct obd_export *exp,
struct obd_client_handle *och,
struct ptlrpc_request *open_req)
{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- struct obd_export *tgt_exp;
-
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lmv_tgt_desc *tgt;
ENTRY;
- tgt_exp = lmv_find_export(lmv, &och->och_fid);
- if (IS_ERR(tgt_exp))
- RETURN(PTR_ERR(tgt_exp));
+ tgt = lmv_find_target(lmv, &och->och_fid);
+ if (IS_ERR(tgt))
+ RETURN(PTR_ERR(tgt));
- RETURN(md_set_open_replay_data(tgt_exp, och, open_req));
+ RETURN(md_set_open_replay_data(tgt->ltd_exp, och, open_req));
}
int lmv_clear_open_replay_data(struct obd_export *exp,
struct obd_client_handle *och)
{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- struct obd_export *tgt_exp;
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lmv_tgt_desc *tgt;
ENTRY;
- tgt_exp = lmv_find_export(lmv, &och->och_fid);
- if (IS_ERR(tgt_exp))
- RETURN(PTR_ERR(tgt_exp));
+ tgt = lmv_find_target(lmv, &och->och_fid);
+ if (IS_ERR(tgt))
+ RETURN(PTR_ERR(tgt));
- RETURN(md_clear_open_replay_data(tgt_exp, och));
+ RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
}
static int lmv_get_remote_perm(struct obd_export *exp,
struct obd_capa *oc, __u32 suppgid,
struct ptlrpc_request **request)
{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- struct obd_export *tgt_exp;
- int rc;
-
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lmv_tgt_desc *tgt;
+ int rc;
ENTRY;
rc = lmv_check_connect(obd);
if (rc)
RETURN(rc);
- tgt_exp = lmv_find_export(lmv, fid);
- if (IS_ERR(tgt_exp))
- RETURN(PTR_ERR(tgt_exp));
-
- rc = md_get_remote_perm(tgt_exp, fid, oc, suppgid, request);
+ tgt = lmv_find_target(lmv, fid);
+ if (IS_ERR(tgt))
+ RETURN(PTR_ERR(tgt));
+ rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
RETURN(rc);
}
static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
renew_capa_cb_t cb)
{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- struct obd_export *tgt_exp;
- int rc;
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lmv_tgt_desc *tgt;
+ int rc;
ENTRY;
rc = lmv_check_connect(obd);
if (rc)
RETURN(rc);
- tgt_exp = lmv_find_export(lmv, &oc->c_capa.lc_fid);
- if (IS_ERR(tgt_exp))
- RETURN(PTR_ERR(tgt_exp));
+ tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
+ if (IS_ERR(tgt))
+ RETURN(PTR_ERR(tgt));
- rc = md_renew_capa(tgt_exp, oc, cb);
+ rc = md_renew_capa(tgt->ltd_exp, oc, cb);
RETURN(rc);
}
+int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
+ const struct req_msg_field *field, struct obd_capa **oc)
+{
+ struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
+
+ return md_unpack_capa(lmv->tgts[0]->ltd_exp, req, field, oc);
+}
+
int lmv_intent_getattr_async(struct obd_export *exp,
struct md_enqueue_info *minfo,
struct ldlm_enqueue_info *einfo)
{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- struct obd_export *tgt_exp;
- int rc;
+ struct md_op_data *op_data = &minfo->mi_data;
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lmv_tgt_desc *tgt = NULL;
+ int rc;
+ ENTRY;
+
+ rc = lmv_check_connect(obd);
+ if (rc)
+ RETURN(rc);
+
+ tgt = lmv_find_target(lmv, &op_data->op_fid1);
+ if (IS_ERR(tgt))
+ RETURN(PTR_ERR(tgt));
+
+ rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
+ RETURN(rc);
+}
+
+int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
+ struct lu_fid *fid, __u64 *bits)
+{
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lmv_tgt_desc *tgt;
+ int rc;
ENTRY;
rc = lmv_check_connect(obd);
if (rc)
RETURN(rc);
- if (fid_is_zero(&minfo->mi_data.op_fid2))
- tgt_exp = lmv_find_export(lmv, &minfo->mi_data.op_fid1);
- else
- tgt_exp = lmv_find_export(lmv, &minfo->mi_data.op_fid2);
- if (IS_ERR(tgt_exp))
- RETURN(PTR_ERR(tgt_exp));
+ tgt = lmv_find_target(lmv, fid);
+ if (IS_ERR(tgt))
+ RETURN(PTR_ERR(tgt));
- rc = md_intent_getattr_async(tgt_exp, minfo, einfo);
+ rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
RETURN(rc);
}
-int lmv_revalidate_lock(struct obd_export *exp,
- struct lookup_intent *it,
- struct lu_fid *fid)
-{
- struct obd_device *obd = exp->exp_obd;
- struct lmv_obd *lmv = &obd->u.lmv;
- struct obd_export *tgt_exp;
- int rc;
- ENTRY;
+/**
+ * For lmv, only need to send request to master MDT, and the master MDT will
+ * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
+ * we directly fetch data from the slave MDTs.
+ */
+int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
+ struct obd_quotactl *oqctl)
+{
+ struct obd_device *obd = class_exp2obd(exp);
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lmv_tgt_desc *tgt = lmv->tgts[0];
+ int rc = 0, i;
+ __u64 curspace, curinodes;
+ ENTRY;
+
+ if (!lmv->desc.ld_tgt_count || !tgt->ltd_active) {
+ CERROR("master lmv inactive\n");
+ RETURN(-EIO);
+ }
- rc = lmv_check_connect(obd);
- if (rc)
+ if (oqctl->qc_cmd != Q_GETOQUOTA) {
+ rc = obd_quotactl(tgt->ltd_exp, oqctl);
RETURN(rc);
+ }
- tgt_exp = lmv_find_export(lmv, fid);
- if (IS_ERR(tgt_exp))
- RETURN(PTR_ERR(tgt_exp));
+ curspace = curinodes = 0;
+ for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
+ int err;
+ tgt = lmv->tgts[i];
+
+ if (tgt == NULL || tgt->ltd_exp == NULL || tgt->ltd_active == 0)
+ continue;
+ if (!tgt->ltd_active) {
+ CDEBUG(D_HA, "mdt %d is inactive.\n", i);
+ continue;
+ }
+
+ err = obd_quotactl(tgt->ltd_exp, oqctl);
+ if (err) {
+ CERROR("getquota on mdt %d failed. %d\n", i, err);
+ if (!rc)
+ rc = err;
+ } else {
+ curspace += oqctl->qc_dqblk.dqb_curspace;
+ curinodes += oqctl->qc_dqblk.dqb_curinodes;
+ }
+ }
+ oqctl->qc_dqblk.dqb_curspace = curspace;
+ oqctl->qc_dqblk.dqb_curinodes = curinodes;
- rc = md_revalidate_lock(tgt_exp, it, fid);
RETURN(rc);
}
+int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp,
+ struct obd_quotactl *oqctl)
+{
+ struct obd_device *obd = class_exp2obd(exp);
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lmv_tgt_desc *tgt;
+ int i, rc = 0;
+ ENTRY;
+
+ for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
+ int err;
+ tgt = lmv->tgts[i];
+ if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
+ CERROR("lmv idx %d inactive\n", i);
+ RETURN(-EIO);
+ }
+
+ err = obd_quotacheck(tgt->ltd_exp, oqctl);
+ if (err && !rc)
+ rc = err;
+ }
+
+ RETURN(rc);
+}
struct obd_ops lmv_obd_ops = {
.o_owner = THIS_MODULE,
.o_connect = lmv_connect,
.o_disconnect = lmv_disconnect,
.o_statfs = lmv_statfs,
- .o_llog_init = lmv_llog_init,
- .o_llog_finish = lmv_llog_finish,
.o_get_info = lmv_get_info,
.o_set_info_async = lmv_set_info_async,
.o_packmd = lmv_packmd,
.o_unpackmd = lmv_unpackmd,
.o_notify = lmv_notify,
+ .o_get_uuid = lmv_get_uuid,
.o_iocontrol = lmv_iocontrol,
- .o_fid_delete = lmv_fid_delete
+ .o_quotacheck = lmv_quotacheck,
+ .o_quotactl = lmv_quotactl
};
struct md_ops lmv_md_ops = {
.m_getstatus = lmv_getstatus,
- .m_change_cbdata = lmv_change_cbdata,
+ .m_null_inode = lmv_null_inode,
+ .m_find_cbdata = lmv_find_cbdata,
.m_close = lmv_close,
.m_create = lmv_create,
.m_done_writing = lmv_done_writing,
.m_set_open_replay_data = lmv_set_open_replay_data,
.m_clear_open_replay_data = lmv_clear_open_replay_data,
.m_renew_capa = lmv_renew_capa,
+ .m_unpack_capa = lmv_unpack_capa,
.m_get_remote_perm = lmv_get_remote_perm,
.m_intent_getattr_async = lmv_intent_getattr_async,
.m_revalidate_lock = lmv_revalidate_lock
int __init lmv_init(void)
{
struct lprocfs_static_vars lvars;
- int rc;
-
- obj_cache = cfs_mem_cache_create("lmv_objects",
- sizeof(struct lmv_obj),
- 0, 0);
- if (!obj_cache) {
- CERROR("error allocating lmv objects cache\n");
- return -ENOMEM;
- }
+ int rc;
lprocfs_lmv_init_vars(&lvars);
+
rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
lvars.module_vars, LUSTRE_LMV_NAME, NULL);
- if (rc)
- cfs_mem_cache_destroy(obj_cache);
-
return rc;
}
#ifdef __KERNEL__
static void lmv_exit(void)
{
- int rc;
-
class_unregister_type(LUSTRE_LMV_NAME);
-
- rc = cfs_mem_cache_destroy(obj_cache);
- LASSERTF(rc == 0,
- "can't free lmv objects cache, %d object(s)"
- "still in use\n", atomic_read(&obj_cache_count));
}
MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");