/*
* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
- */
-/*
- * Copyright (c) 2011 Whamcloud, Inc.
+ *
+ * Copyright (c) 2011, 2012, Whamcloud, Inc.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
lh->mlh_reg_mode = lm;
lh->mlh_type = MDT_PDO_LOCK;
- if (name != NULL) {
+ if (name != NULL && (name[0] != '\0')) {
LASSERT(namelen > 0);
lh->mlh_pdo_hash = full_name_hash(name, namelen);
} else {
(S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
}
+void mdt_client_compatibility(struct mdt_thread_info *info)
+{
+ struct mdt_body *body;
+ struct ptlrpc_request *req = mdt_info_req(info);
+ struct obd_export *exp = req->rq_export;
+ struct md_attr *ma = &info->mti_attr;
+ struct lu_attr *la = &ma->ma_attr;
+ ENTRY;
+
+ if (exp->exp_connect_flags & OBD_CONNECT_LAYOUTLOCK)
+ /* the client can deal with 16-bit lmm_stripe_count */
+ RETURN_EXIT;
+
+ body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
+
+ if (!mdt_body_has_lov(la, body))
+ RETURN_EXIT;
+
+ /* now we have a reply with a lov for a client not compatible with the
+ * layout lock so we have to clean the layout generation number */
+ if (S_ISREG(la->la_mode))
+ ma->ma_lmm->lmm_layout_gen = 0;
+ EXIT;
+}
+
+
static int mdt_getattr_internal(struct mdt_thread_info *info,
struct mdt_object *o, int ma_need)
{
if (rc == 0)
mdt_counter_incr(req->rq_export, LPROC_MDT_GETATTR);
+ mdt_client_compatibility(info);
mdt_shrink_reply(info);
return rc;
}
LASSERT(!info->mti_cross_ref);
/* Only got the fid of this obj by name */
+ fid_zero(child_fid);
rc = mdo_lookup(info->mti_env, next, lname, child_fid,
&info->mti_spec);
#if 0
name = NULL;
CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
"ldlm_rep = %p\n",
- PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
- ldlm_rep);
+ PFID(mdt_object_fid(parent)),
+ PFID(&reqbody->fid2), ldlm_rep);
} else {
lname = mdt_name(info->mti_env, (char *)name, namelen);
CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
}
if (lname) {
- /* step 1: lock parent */
- lhp = &info->mti_lh[MDT_LH_PARENT];
- mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
- rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
- MDT_LOCAL_LOCK);
- if (unlikely(rc != 0))
- RETURN(rc);
+ /* step 1: lock parent only if parent is a directory */
+ if (S_ISDIR(lu_object_attr(&parent->mot_obj.mo_lu))) {
+ lhp = &info->mti_lh[MDT_LH_PARENT];
+ mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
+ rc = mdt_object_lock(info, parent, lhp,
+ MDS_INODELOCK_UPDATE,
+ MDT_LOCAL_LOCK);
+ if (unlikely(rc != 0))
+ RETURN(rc);
+ }
/* step 2: lookup child's fid by name */
+ fid_zero(child_fid);
rc = mdo_lookup(info->mti_env, next, lname, child_fid,
&info->mti_spec);
if (rc != 0) {
if (rc == -ENOENT)
- mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
+ mdt_set_disposition(info, ldlm_rep,
+ DISP_LOOKUP_NEG);
GOTO(out_parent, rc);
} else
mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
relock:
OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
mdt_lock_handle_init(lhc);
- mdt_lock_reg_init(lhc, LCK_PR);
+ if (child_bits == MDS_INODELOCK_LAYOUT)
+ mdt_lock_reg_init(lhc, LCK_CR);
+ else
+ mdt_lock_reg_init(lhc, LCK_PR);
if (mdt_object_exists(child) == 0) {
- LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
+ LU_OBJECT_DEBUG(D_INODE, info->mti_env,
&child->mot_obj.mo_lu,
"Object doesn't exist!\n");
GOTO(out_child, rc = -ENOENT);
if (unlikely(rc != 0))
GOTO(out_child, rc);
+ /* layout lock is used only on regular files */
+ if ((ma->ma_valid & MA_INODE) &&
+ (ma->ma_attr.la_valid & LA_MODE) &&
+ !S_ISREG(ma->ma_attr.la_mode))
+ child_bits &= ~MDS_INODELOCK_LAYOUT;
+
/* If the file has not been changed for some time, we
* return not only a LOOKUP lock, but also an UPDATE
* lock and this might save us RPC on later STAT. For
mdt_exit_ucred(info);
EXIT;
out_shrink:
+ mdt_client_compatibility(info);
mdt_shrink_reply(info);
return rc;
}
out_ucred:
mdt_exit_ucred(info);
out_shrink:
+ mdt_client_compatibility(info);
mdt_shrink_reply(info);
return rc;
}
else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
lh->mlh_pdo_hash != 0 &&
(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
- OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 10);
+ OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
}
RETURN(rc);
MDT_IT_UNLINK,
MDT_IT_TRUNC,
MDT_IT_GETXATTR,
+ MDT_IT_LAYOUT,
MDT_IT_NR
};
.it_fmt = NULL,
.it_flags = 0,
.it_act = NULL
+ },
+ [MDT_IT_LAYOUT] = {
+ .it_fmt = &RQF_LDLM_INTENT_GETATTR,
+ .it_flags = HABEO_REFERO,
+ .it_act = mdt_intent_getattr
}
};
case MDT_IT_GETATTR:
child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
break;
+ case MDT_IT_LAYOUT: {
+ static int printed = 0;
+
+ if (!printed) {
+ CERROR("layout lock not supported by this version\n");
+ printed = 1;
+ }
+ GOTO(out_shrink, rc = -EINVAL);
+ break;
+ }
default:
- CERROR("Unhandled till now");
+ CERROR("Unsupported intent (%d)\n", opcode);
GOTO(out_shrink, rc = -EINVAL);
}
out_ucred:
mdt_exit_ucred(info);
out_shrink:
+ mdt_client_compatibility(info);
mdt_shrink_reply(info);
return rc;
}
}
rep->lock_policy_res2 = clear_serious(rc);
+ if (rep->lock_policy_res2 == -ENOENT &&
+ mdt_get_disposition(rep, DISP_LOOKUP_NEG))
+ rep->lock_policy_res2 = 0;
+
if (rc == -ENOTCONN || rc == -ENODEV ||
rc == -EOVERFLOW) { /**< if VBR failure then return error */
/*
case IT_GETXATTR:
rc = MDT_IT_GETXATTR;
break;
+ case IT_LAYOUT:
+ rc = MDT_IT_LAYOUT;
+ break;
default:
CERROR("Unknown intent opcode: %ld\n", itcode);
rc = -EINVAL;
/*
* setattr service configuration.
+ *
+ * XXX To keep the compatibility with old client(< 2.2), we need to
+ * preserve this portal for a certain time, it should be removed
+ * eventually. LU-617.
*/
conf = (typeof(conf)) {
.psc_nbufs = MDS_NBUFS,
lu_device_get(d);
lu_ref_add(&d->ld_reference, "lu-stack", &lu_site_init);
- RETURN(d);
+ cfs_spin_lock(&d->ld_site->ls_ld_lock);
+ cfs_list_add_tail(&d->ld_linkage, &d->ld_site->ls_ld_linkage);
+ cfs_spin_unlock(&d->ld_site->ls_ld_lock);
+ RETURN(d);
out_alloc:
ldt->ldt_ops->ldto_device_free(env, d);
type->typ_refcnt--;
mdt_obd_llog_cleanup(obd);
obd_exports_barrier(obd);
obd_zombie_barrier();
+
+ mdt_procfs_fini(m);
+
#ifdef HAVE_QUOTA_SUPPORT
next->md_ops->mdo_quota.mqo_cleanup(env, next);
#endif
*/
mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
- lprocfs_free_per_client_stats(obd);
- lprocfs_free_obd_stats(obd);
- mdt_procfs_fini(m);
-
if (ls) {
struct md_site *mite;
return 0;
}
-static void fsoptions_to_mdt_flags(struct mdt_device *m, char *options)
-{
- char *p = options;
-
- m->mdt_opts.mo_mds_capa = 1;
- m->mdt_opts.mo_oss_capa = 1;
-#ifdef CONFIG_FS_POSIX_ACL
- /* ACLs should be enabled by default (b=13829) */
- m->mdt_opts.mo_acl = 1;
- LCONSOLE_INFO("Enabling ACL\n");
-#else
- m->mdt_opts.mo_acl = 0;
- LCONSOLE_INFO("Disabling ACL\n");
-#endif
-
- if (!options)
- return;
-
- while (*options) {
- int len;
-
- while (*p && *p != ',')
- p++;
-
- len = p - options;
- if ((len == sizeof("user_xattr") - 1) &&
- (memcmp(options, "user_xattr", len) == 0)) {
- m->mdt_opts.mo_user_xattr = 1;
- LCONSOLE_INFO("Enabling user_xattr\n");
- } else if ((len == sizeof("nouser_xattr") - 1) &&
- (memcmp(options, "nouser_xattr", len) == 0)) {
- m->mdt_opts.mo_user_xattr = 0;
- LCONSOLE_INFO("Disabling user_xattr\n");
- } else if ((len == sizeof("noacl") - 1) &&
- (memcmp(options, "noacl", len) == 0)) {
- m->mdt_opts.mo_acl = 0;
- LCONSOLE_INFO("Disabling ACL\n");
- }
-
- if (!*p)
- break;
-
- options = ++p;
- }
-}
-
int mdt_postrecov(const struct lu_env *, struct mdt_device *);
static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
struct lu_site *s;
struct md_site *mite;
const char *identity_upcall = "NONE";
-#ifdef HAVE_QUOTA_SUPPORT
struct md_device *next;
-#endif
int rc;
int node_id;
+ mntopt_t mntopts;
ENTRY;
md_device_init(&m->mdt_md_dev, ldt);
m->mdt_max_cookiesize = sizeof(struct llog_cookie);
m->mdt_som_conf = 0;
- m->mdt_opts.mo_user_xattr = 0;
- m->mdt_opts.mo_acl = 0;
m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
lmi = server_get_mount_2(dev);
if (lmi == NULL) {
RETURN(-EFAULT);
} else {
lsi = s2lsi(lmi->lmi_sb);
- fsoptions_to_mdt_flags(m, lsi->lsi_lmd->lmd_opts);
/* CMD is supported only in IAM mode */
ldd = lsi->lsi_ldd;
LASSERT(num);
CERROR("CMD Operation not allowed in IOP mode\n");
GOTO(err_lmi, rc = -EINVAL);
}
- /* Read recovery timeouts */
- if (lsi->lsi_lmd && lsi->lsi_lmd->lmd_recovery_time_soft)
- obd->obd_recovery_timeout =
- lsi->lsi_lmd->lmd_recovery_time_soft;
- if (lsi->lsi_lmd && lsi->lsi_lmd->lmd_recovery_time_hard)
- obd->obd_recovery_time_hard =
- lsi->lsi_lmd->lmd_recovery_time_hard;
+ obd->u.obt.obt_magic = OBT_MAGIC;
}
cfs_rwlock_init(&m->mdt_sptlrpc_lock);
cfs_spin_lock_init(&m->mdt_ioepoch_lock);
m->mdt_opts.mo_compat_resname = 0;
+ m->mdt_opts.mo_mds_capa = 1;
+ m->mdt_opts.mo_oss_capa = 1;
m->mdt_capa_timeout = CAPA_TIMEOUT;
m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
GOTO(err_free_site, rc);
}
- rc = mdt_procfs_init(m, dev);
- if (rc) {
- CERROR("Can't init MDT lprocfs, rc %d\n", rc);
- GOTO(err_fini_proc, rc);
- }
-
/* set server index */
lu_site2md(s)->ms_node_id = node_id;
rc = mdt_stack_init((struct lu_env *)env, m, cfg, lmi);
if (rc) {
CERROR("Can't init device stack, rc %d\n", rc);
- GOTO(err_fini_proc, rc);
+ GOTO(err_lu_site, rc);
}
rc = lut_init(env, &m->mdt_lut, obd, m->mdt_bottom);
/* set obd_namespace for compatibility with old code */
obd->obd_namespace = m->mdt_namespace;
- /* XXX: to support suppgid for ACL, we enable identity_upcall
- * by default, otherwise, maybe got unexpected -EACCESS. */
- if (m->mdt_opts.mo_acl)
- identity_upcall = MDT_IDENTITY_UPCALL_PATH;
-
- m->mdt_identity_cache = upcall_cache_init(obd->obd_name, identity_upcall,
- &mdt_identity_upcall_cache_ops);
- if (IS_ERR(m->mdt_identity_cache)) {
- rc = PTR_ERR(m->mdt_identity_cache);
- m->mdt_identity_cache = NULL;
- GOTO(err_free_ns, rc);
- }
-
cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
rc = mdt_ck_thread_start(m);
mdt_adapt_sptlrpc_conf(obd, 1);
-#ifdef HAVE_QUOTA_SUPPORT
next = m->mdt_child;
+#ifdef HAVE_QUOTA_SUPPORT
rc = next->md_ops->mdo_quota.mqo_setup(env, next, lmi->lmi_mnt);
if (rc)
GOTO(err_llog_cleanup, rc);
server_put_mount_2(dev, lmi->lmi_mnt);
lmi = NULL;
+ rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
+ &mntopts);
+ if (rc)
+ GOTO(err_quota, rc);
+
+ if (mntopts & MNTOPT_USERXATTR)
+ m->mdt_opts.mo_user_xattr = 1;
+ else
+ m->mdt_opts.mo_user_xattr = 0;
+
+ if (mntopts & MNTOPT_ACL)
+ m->mdt_opts.mo_acl = 1;
+ else
+ m->mdt_opts.mo_acl = 0;
+
+ /* XXX: to support suppgid for ACL, we enable identity_upcall
+ * by default, otherwise, maybe got unexpected -EACCESS. */
+ if (m->mdt_opts.mo_acl)
+ identity_upcall = MDT_IDENTITY_UPCALL_PATH;
+
+ m->mdt_identity_cache = upcall_cache_init(obd->obd_name,identity_upcall,
+ &mdt_identity_upcall_cache_ops);
+ if (IS_ERR(m->mdt_identity_cache)) {
+ rc = PTR_ERR(m->mdt_identity_cache);
+ m->mdt_identity_cache = NULL;
+ GOTO(err_quota, rc);
+ }
+
target_recovery_init(&m->mdt_lut, mdt_recovery_handle);
+ rc = mdt_procfs_init(m, dev);
+ if (rc) {
+ CERROR("Can't init MDT lprocfs, rc %d\n", rc);
+ GOTO(err_recovery, rc);
+ }
+
rc = mdt_start_ptlrpc_service(m);
if (rc)
- GOTO(err_recovery, rc);
+ GOTO(err_procfs, rc);
ping_evictor_start();
err_stop_service:
ping_evictor_stop();
mdt_stop_ptlrpc_service(m);
+err_procfs:
+ mdt_procfs_fini(m);
err_recovery:
target_recovery_fini(obd);
+ upcall_cache_cleanup(m->mdt_identity_cache);
+ m->mdt_identity_cache = NULL;
+err_quota:
#ifdef HAVE_QUOTA_SUPPORT
next->md_ops->mdo_quota.mqo_cleanup(env, next);
#endif
cfs_timer_disarm(&m->mdt_ck_timer);
mdt_ck_thread_stop(m);
err_free_ns:
- upcall_cache_cleanup(m->mdt_identity_cache);
- m->mdt_identity_cache = NULL;
ldlm_namespace_free(m->mdt_namespace, NULL, 0);
obd->obd_namespace = m->mdt_namespace = NULL;
err_fini_seq:
lut_fini(env, &m->mdt_lut);
err_fini_stack:
mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
-err_fini_proc:
- mdt_procfs_fini(m);
+err_lu_site:
lu_site_fini(s);
err_free_site:
OBD_FREE_PTR(mite);
/*
* For interoperability between 1.8 and 2.0,
- * skip old "mdt.group_upcall" param.
*/
{
+ /* Skip old "mdt.group_upcall" param. */
char *param = lustre_cfg_string(cfg, 1);
if (param && !strncmp("mdt.group_upcall", param, 16)) {
CWARN("For 1.8 interoperability, skip this"
" mdt.group_upcall. It is obsolete\n");
break;
}
+ /* Rename old "mdt.quota_type" to "mdd.quota_type. */
+ if (param && !strncmp("mdt.quota_type", param, 14)) {
+ CWARN("Found old param mdt.quota_type, changed"
+ " it to mdd.quota_type.\n");
+ param[2] = 'd';
+ }
}
lprocfs_mdt_init_vars(&lvars);
lu_object_init(o, h, d);
lu_object_add_top(h, o);
o->lo_ops = &mdt_obj_ops;
- cfs_sema_init(&mo->mot_ioepoch_sem, 1);
- cfs_sema_init(&mo->mot_lov_sem, 1);
+ cfs_mutex_init(&mo->mot_ioepoch_mutex);
+ cfs_mutex_init(&mo->mot_lov_mutex);
RETURN(o);
} else
RETURN(NULL);
CFS_INIT_LIST_HEAD(&med->med_open_head);
cfs_spin_lock_init(&med->med_open_lock);
- cfs_sema_init(&med->med_idmap_sem, 1);
+ cfs_mutex_init(&med->med_idmap_mutex);
med->med_idmap = NULL;
cfs_spin_lock(&exp->exp_lock);
exp->exp_connecting = 1;
cfs_spin_unlock(&exp->exp_lock);
+
+ /* self-export doesn't need client data and ldlm initialization */
+ if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
+ &exp->exp_client_uuid)))
+ RETURN(0);
+
rc = lut_client_alloc(exp);
if (rc == 0)
rc = ldlm_init_export(exp);
-
if (rc)
- CERROR("Error %d while initializing export\n", rc);
+ CERROR("%s: Error %d while initializing export\n",
+ exp->exp_obd->obd_name, rc);
RETURN(rc);
}
static int mdt_destroy_export(struct obd_export *exp)
{
- struct mdt_export_data *med;
- int rc = 0;
ENTRY;
- med = &exp->exp_mdt_data;
if (exp_connect_rmtclient(exp))
mdt_cleanup_idmap(&exp->exp_mdt_data);
target_destroy_export(exp);
+ /* destroy can be called from failed obd_setup, so
+ * checking uuid is safer than obd_self_export */
+ if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
+ &exp->exp_client_uuid)))
+ RETURN(0);
+
ldlm_destroy_export(exp);
lut_client_free(exp);
LASSERT(cfs_list_empty(&exp->exp_outstanding_replies));
LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head));
- if (obd_uuid_equals(&exp->exp_client_uuid, &exp->exp_obd->obd_uuid))
- RETURN(0);
- RETURN(rc);
+ RETURN(0);
}
static void mdt_allow_cli(struct mdt_device *m, unsigned int flag)
RETURN(rc);
}
-static int mdt_obd_notify(struct obd_device *host,
+static int mdt_obd_notify(struct obd_device *obd,
struct obd_device *watched,
enum obd_notify_event ev, void *data)
{
- struct mdt_device *mdt = mdt_dev(host->obd_lu_dev);
+ struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
#ifdef HAVE_QUOTA_SUPPORT
struct md_device *next = mdt->mdt_child;
#endif
obj = mdt_object_find(env, mdt, &fp->gf_fid);
if (obj == NULL || IS_ERR(obj)) {
- CDEBUG(D_IOCTL, "no object "DFID": %ld\n",PFID(&fp->gf_fid),
+ CDEBUG(D_IOCTL, "no object "DFID": %ld\n", PFID(&fp->gf_fid),
PTR_ERR(obj));
RETURN(-EINVAL);
}
struct mdt_lock_handle *lh;
int rc;
ENTRY;
+
CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
if (!fid_is_sane(fid))
RETURN(-EINVAL);
* fid, this is error to find remote object here
*/
CERROR("nonlocal object "DFID"\n", PFID(fid));
+ } else if (rc == 0) {
+ *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
+ rc = -ENOENT;
} else {
- version = mo_version_get(mti->mti_env, mdt_object_child(obj));
+ version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
*(__u64 *)data->ioc_inlbuf2 = version;
rc = 0;
}
rc = mdt_device_sync(&env, mdt);
break;
case OBD_IOC_SET_READONLY:
- dt->dd_ops->dt_ro(&env, dt);
+ rc = dt->dd_ops->dt_ro(&env, dt);
break;
case OBD_IOC_ABORT_RECOVERY:
CERROR("Aborting recovery for device %s\n", obd->obd_name);
/* context key: mdt_thread_key */
LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
-/* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
-LU_KEY_INIT_FINI(mdt_txn, struct mdt_txn_info);
-
-struct lu_context_key mdt_txn_key = {
- .lct_tags = LCT_TX_HANDLE,
- .lct_init = mdt_txn_key_init,
- .lct_fini = mdt_txn_key_fini
-};
-
struct md_ucred *mdt_ucred(const struct mdt_thread_info *info)
{
return md_ucred(info->mti_env);
}
/* type constructor/destructor: mdt_type_init, mdt_type_fini */
-LU_TYPE_INIT_FINI(mdt, &mdt_thread_key, &mdt_txn_key);
+LU_TYPE_INIT_FINI(mdt, &mdt_thread_key);
static struct lu_device_type_operations mdt_device_type_ops = {
.ldto_init = mdt_type_init,