* GPL HEADER END
*/
/*
- * Copyright 2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*/
/*
void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
const struct lu_attr *attr, const struct lu_fid *fid)
{
- struct md_attr *ma = &info->mti_attr;
+ struct md_attr *ma = &info->mti_attr;
LASSERT(ma->ma_valid & MA_INODE);
+ b->atime = attr->la_atime;
+ b->mtime = attr->la_mtime;
+ b->ctime = attr->la_ctime;
+ b->mode = attr->la_mode;
+ b->size = attr->la_size;
+ b->blocks = attr->la_blocks;
+ b->uid = attr->la_uid;
+ b->gid = attr->la_gid;
+ b->flags = attr->la_flags;
+ b->nlink = attr->la_nlink;
+ b->rdev = attr->la_rdev;
+
/*XXX should pack the reply body according to lu_valid*/
b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID |
OBD_MD_FLGID | OBD_MD_FLTYPE |
} else if (ma->ma_need & MA_LOV && ma->ma_lmm_size == 0) {
/* means no objects are allocated on osts. */
LASSERT(!(ma->ma_valid & MA_LOV));
- LASSERT(attr->la_blocks == 0);
+ /* just ignore blocks occupied by extend attributes on MDS */
+ b->blocks = 0;
/* if no object is allocated on osts, the size on mds is valid. b=22272 */
b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
}
- b->atime = attr->la_atime;
- b->mtime = attr->la_mtime;
- b->ctime = attr->la_ctime;
- b->mode = attr->la_mode;
- b->size = attr->la_size;
- b->blocks = attr->la_blocks;
- b->uid = attr->la_uid;
- b->gid = attr->la_gid;
- b->flags = attr->la_flags;
- b->nlink = attr->la_nlink;
- b->rdev = attr->la_rdev;
-
if (fid) {
b->fid1 = *fid;
b->valid |= OBD_MD_FLID;
if (info)
mdt_body_reverse_idmap(info, b);
+
+ if (b->valid & OBD_MD_FLSIZE)
+ CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
+ PFID(fid), (unsigned long long)b->size);
}
static inline int mdt_body_has_lov(const struct lu_attr *la,
repbody->eadatasize = ma->ma_lmv_size;
repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
}
- if (!(ma->ma_valid & MA_LOV) && !(ma->ma_valid & MA_LMV)) {
- repbody->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
- }
} else if (S_ISLNK(la->la_mode) &&
reqbody->valid & OBD_MD_LINKNAME) {
buffer->lb_buf = ma->ma_lmm;
struct lu_name *lname = NULL;
const char *name = NULL;
int namelen = 0;
- struct mdt_lock_handle *lhp;
+ struct mdt_lock_handle *lhp = NULL;
struct ldlm_lock *lock;
struct ldlm_res_id *res_id;
int is_resent;
if (namelen == 0) {
reqbody = req_capsule_client_get(info->mti_pill,
&RMF_MDT_BODY);
- LASSERT(fid_is_sane(&reqbody->fid2));
- name = NULL;
+ if (unlikely(reqbody == NULL))
+ RETURN(err_serious(-EFAULT));
+ if (unlikely(!fid_is_sane(&reqbody->fid2)))
+ RETURN(err_serious(-EINVAL));
+
+ name = NULL;
CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
"ldlm_rep = %p\n",
PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
RETURN(rc);
}
- /* 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);
-
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 2: lookup child's fid by name */
rc = mdo_lookup(info->mti_env, next, lname, child_fid,
&info->mti_spec);
LDLM_LOCK_PUT(lock);
rc = 0;
} else {
- struct md_attr *ma;
relock:
- ma = &info->mti_attr;
-
OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
mdt_lock_handle_init(lhc);
mdt_lock_reg_init(lhc, LCK_PR);
LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
&child->mot_obj.mo_lu,
"Object doesn't exist!\n");
- GOTO(out_child, rc = -ESTALE);
+ GOTO(out_child, rc = -ENOENT);
}
- ma->ma_valid = 0;
- ma->ma_need = MA_INODE;
- rc = mo_attr_get(info->mti_env, next, ma);
- if (unlikely(rc != 0))
- GOTO(out_child, rc);
+ if (!(child_bits & MDS_INODELOCK_UPDATE)) {
+ struct md_attr *ma = &info->mti_attr;
- /* 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 directories, it also
- * let negative dentry starts working for this dir. */
- if (ma->ma_valid & MA_INODE &&
- ma->ma_attr.la_valid & LA_CTIME &&
- info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
- ma->ma_attr.la_ctime < cfs_time_current_sec())
- child_bits |= MDS_INODELOCK_UPDATE;
+ ma->ma_valid = 0;
+ ma->ma_need = MA_INODE;
+ rc = mo_attr_get(info->mti_env,
+ mdt_object_child(child), ma);
+ if (unlikely(rc != 0))
+ GOTO(out_child, rc);
+
+ /* 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
+ * directories, it also let negative dentry starts
+ * working for this dir. */
+ if (ma->ma_valid & MA_INODE &&
+ ma->ma_attr.la_valid & LA_CTIME &&
+ info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
+ ma->ma_attr.la_ctime < cfs_time_current_sec())
+ child_bits |= MDS_INODELOCK_UPDATE;
+ }
rc = mdt_object_lock(info, child, lhc, child_bits,
MDT_CROSS_LOCK);
out_child:
mdt_object_put(info->mti_env, child);
out_parent:
- mdt_object_unlock(info, parent, lhp, 1);
+ if (lhp)
+ mdt_object_unlock(info, parent, lhp, 1);
return rc;
}
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
GOTO(abort_bulk, rc = 0);
- timeout = (int) req->rq_deadline - cfs_time_current_sec();
- if (timeout < 0)
- CERROR("Req deadline already passed %lu (now: %lu)\n",
- req->rq_deadline, cfs_time_current_sec());
- *lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(max(timeout, 1)),
- cfs_time_seconds(1), NULL, NULL);
- rc = l_wait_event(desc->bd_waitq, !ptlrpc_server_bulk_active(desc) ||
- exp->exp_failed || exp->exp_abort_active_req, lwi);
- LASSERT (rc == 0 || rc == -ETIMEDOUT);
+ do {
+ timeout = (int) req->rq_deadline - cfs_time_current_sec();
+ if (timeout < 0)
+ CERROR("Req deadline already passed %lu (now: %lu)\n",
+ req->rq_deadline, cfs_time_current_sec());
+ *lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(max(timeout, 1)),
+ cfs_time_seconds(1), NULL, NULL);
+ rc = l_wait_event(desc->bd_waitq,
+ !ptlrpc_server_bulk_active(desc) ||
+ exp->exp_failed ||
+ exp->exp_abort_active_req, lwi);
+ LASSERT (rc == 0 || rc == -ETIMEDOUT);
+ } while ((rc == -ETIMEDOUT) &&
+ (req->rq_deadline > cfs_time_current_sec()));
if (rc == 0) {
if (desc->bd_success &&
static int mdt_enqueue(struct mdt_thread_info *info)
{
struct ptlrpc_request *req;
- __u64 req_bits;
int rc;
/*
LASSERT(info->mti_dlm_req != NULL);
req = mdt_info_req(info);
-
- /*
- * Lock without inodebits makes no sense and will oops later in
- * ldlm. Let's check it now to see if we have wrong lock from client or
- * bits get corrupted somewhere in mdt_intent_policy().
- */
- req_bits = info->mti_dlm_req->lock_desc.l_policy_data.l_inodebits.bits;
- /* This is disabled because we need to support liblustre flock.
- * LASSERT(req_bits != 0);
- */
-
rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
req, info->mti_dlm_req, &cbs);
info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
void *data, int flag)
{
- struct obd_device *obd = lock->l_resource->lr_namespace->ns_obd;
+ struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
int rc;
ENTRY;
RETURN(-ESTALE);
} else {
/* Non-dir object shouldn't have PDO lock */
- LASSERT(S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)));
+ if (!S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)))
+ RETURN(-ENOTDIR);
}
}
dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
if (dlm_req != NULL) {
- if (info->mti_mdt->mdt_opts.mo_compat_resname)
- rc = mdt_lock_resname_compat(info->mti_mdt,
- dlm_req);
- info->mti_dlm_req = dlm_req;
+ if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
+ LDLM_IBITS &&
+ dlm_req->lock_desc.l_policy_data.\
+ l_inodebits.bits == 0)) {
+ /*
+ * Lock without inodebits makes no sense and
+ * will oops later in ldlm. If client miss to
+ * set such bits, do not trigger ASSERTION.
+ *
+ * For liblustre flock case, it maybe zero.
+ */
+ rc = -EPROTO;
+ } else {
+ if (info->mti_mdt->mdt_opts.mo_compat_resname)
+ rc = mdt_lock_resname_compat(
+ info->mti_mdt,
+ dlm_req);
+ info->mti_dlm_req = dlm_req;
+ }
} else {
rc = -EFAULT;
}
* checked here.
*/
if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
+ LASSERTF(rc == 0, "Error occurred but lock handle "
+ "is still in use\n");
rep->lock_policy_res2 = 0;
rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
RETURN(rc);
req_capsule_extend(pill, &RQF_LDLM_INTENT);
it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
if (it != NULL) {
- const struct ldlm_request *dlmreq;
- __u64 req_bits;
-
rc = mdt_intent_opc(it->opc, info, lockp, flags);
if (rc == 0)
rc = ELDLM_OK;
- /*
- * Lock without inodebits makes no sense and will oops
+ /* Lock without inodebits makes no sense and will oops
* later in ldlm. Let's check it now to see if we have
- * wrong lock from client or bits get corrupted
- * somewhere in mdt_intent_opc().
- */
- dlmreq = info->mti_dlm_req;
- req_bits = dlmreq->lock_desc.l_policy_data.l_inodebits.bits;
- LASSERT(req_bits != 0);
-
+ * ibits corrupted somewhere in mdt_intent_opc().
+ * The case for client miss to set ibits has been
+ * processed by others. */
+ LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
+ lr_type == LDLM_IBITS,
+ info->mti_dlm_req->lock_desc.\
+ l_policy_data.l_inodebits.bits != 0));
} else
rc = err_serious(-EFAULT);
} else {
if (m->mdt_regular_service == NULL)
RETURN(-ENOMEM);
- rc = ptlrpc_start_threads(NULL, m->mdt_regular_service);
+ rc = ptlrpc_start_threads(m->mdt_regular_service);
if (rc)
GOTO(err_mdt_svc, rc);
GOTO(err_mdt_svc, rc = -ENOMEM);
}
- rc = ptlrpc_start_threads(NULL, m->mdt_readpage_service);
+ rc = ptlrpc_start_threads(m->mdt_readpage_service);
/*
* setattr service configuration.
GOTO(err_mdt_svc, rc = -ENOMEM);
}
- rc = ptlrpc_start_threads(NULL, m->mdt_setattr_service);
+ rc = ptlrpc_start_threads(m->mdt_setattr_service);
if (rc)
GOTO(err_mdt_svc, rc);
GOTO(err_mdt_svc, rc = -ENOMEM);
}
- rc = ptlrpc_start_threads(NULL, m->mdt_mdsc_service);
+ rc = ptlrpc_start_threads(m->mdt_mdsc_service);
if (rc)
GOTO(err_mdt_svc, rc);
GOTO(err_mdt_svc, rc = -ENOMEM);
}
- rc = ptlrpc_start_threads(NULL, m->mdt_mdss_service);
+ rc = ptlrpc_start_threads(m->mdt_mdss_service);
if (rc)
GOTO(err_mdt_svc, rc);
GOTO(err_mdt_svc, rc = -ENOMEM);
}
- rc = ptlrpc_start_threads(NULL, m->mdt_dtss_service);
+ rc = ptlrpc_start_threads(m->mdt_dtss_service);
if (rc)
GOTO(err_mdt_svc, rc);
GOTO(err_mdt_svc, rc = -ENOMEM);
}
- rc = ptlrpc_start_threads(NULL, m->mdt_fld_service);
+ rc = ptlrpc_start_threads(m->mdt_fld_service);
if (rc)
GOTO(err_mdt_svc, rc);
GOTO(err_mdt_svc, rc = -ENOMEM);
}
- rc = ptlrpc_start_threads(NULL, m->mdt_xmds_service);
+ rc = ptlrpc_start_threads(m->mdt_xmds_service);
if (rc)
GOTO(err_mdt_svc, rc);
LCONSOLE_INFO("Disabling ACL\n");
}
+ if (!*p)
+ break;
+
options = ++p;
}
}
LUSTRE_MDT_NAME"-%p", m);
m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
LDLM_NAMESPACE_SERVER,
- LDLM_NAMESPACE_GREEDY);
+ LDLM_NAMESPACE_GREEDY,
+ LDLM_NS_TYPE_MDT);
if (m->mdt_namespace == NULL)
GOTO(err_fini_seq, rc = -ENOMEM);
void *localdata)
{
struct mdt_thread_info *info;
- struct lsd_client_data *lcd;
struct obd_export *lexp;
struct lustre_handle conn = { 0 };
struct mdt_device *mdt;
rc = mdt_connect_internal(lexp, mdt, data);
if (rc == 0) {
- OBD_ALLOC_PTR(lcd);
- if (lcd != NULL) {
- struct mdt_thread_info *mti;
- mti = lu_context_key_get(&env->le_ctx,
- &mdt_thread_key);
- LASSERT(mti != NULL);
- mti->mti_exp = lexp;
- memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
- lexp->exp_target_data.ted_lcd = lcd;
- rc = mdt_client_new(env, mdt);
- if (rc == 0)
- mdt_export_stats_init(obd, lexp, localdata);
- } else {
- rc = -ENOMEM;
- }
+ struct mdt_thread_info *mti;
+ struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
+ LASSERT(lcd);
+ mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
+ LASSERT(mti != NULL);
+ mti->mti_exp = lexp;
+ memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
+ rc = mdt_client_new(env, mdt);
+ if (rc == 0)
+ mdt_export_stats_init(obd, lexp, localdata);
}
out:
- if (rc != 0)
+ if (rc != 0) {
class_disconnect(lexp);
- else
+ *exp = NULL;
+ } else {
*exp = lexp;
+ }
RETURN(rc);
}
int cookie_size;
lmm_size = mdt->mdt_max_mdsize;
- OBD_ALLOC(ma->ma_lmm, lmm_size);
+ OBD_ALLOC_LARGE(ma->ma_lmm, lmm_size);
if (ma->ma_lmm == NULL)
GOTO(out_lmm, rc = -ENOMEM);
cookie_size = mdt->mdt_max_cookiesize;
- OBD_ALLOC(ma->ma_cookie, cookie_size);
+ OBD_ALLOC_LARGE(ma->ma_cookie, cookie_size);
if (ma->ma_cookie == NULL)
GOTO(out_cookie, rc = -ENOMEM);
ma->ma_need = 0;
/* It is not for setattr, just tell MDD to send
* DESTROY RPC to OSS if needed */
- ma->ma_attr_flags = MDS_CLOSE_CLEANUP;
ma->ma_valid = MA_FLAGS;
+ ma->ma_attr_flags = MDS_CLOSE_CLEANUP;
+ /* Don't unlink orphan on failover umount, LU-184 */
+ if (exp->exp_flags & OBD_OPT_FAILOVER)
+ ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
mdt_mfd_close(info, mfd);
}
- OBD_FREE(ma->ma_cookie, cookie_size);
+ OBD_FREE_LARGE(ma->ma_cookie, cookie_size);
ma->ma_cookie = NULL;
out_cookie:
- OBD_FREE(ma->ma_lmm, lmm_size);
+ OBD_FREE_LARGE(ma->ma_lmm, lmm_size);
ma->ma_lmm = NULL;
}
out_lmm:
info->mti_mdt = NULL;
/* cleanup client slot early */
/* Do not erase record for recoverable client. */
- if (!obd->obd_fail || exp->exp_failed)
+ if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
mdt_client_del(&env, mdt);
lu_env_fini(&env);
cfs_spin_lock(&exp->exp_lock);
exp->exp_connecting = 1;
cfs_spin_unlock(&exp->exp_lock);
- rc = ldlm_init_export(exp);
+ rc = lut_client_alloc(exp);
+ if (rc == 0)
+ rc = ldlm_init_export(exp);
+
if (rc)
CERROR("Error %d while initializing export\n", rc);
RETURN(rc);
target_destroy_export(exp);
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);
- lut_client_free(exp);
RETURN(rc);
}
if (cfs_test_bit(MDT_FL_CFGLOG, &m->mdt_state) &&
cfs_test_bit(MDT_FL_SYNCED, &m->mdt_state)) {
struct obd_device *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
-
+
/* Open for clients */
if (obd->obd_no_conn) {
- cfs_spin_lock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_lock(&obd->obd_dev_lock);
obd->obd_no_conn = 0;
- cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
+ cfs_spin_unlock(&obd->obd_dev_lock);
}
}
}
m->mdt_max_mdsize, m->mdt_max_cookiesize);
mdt_allow_cli(m, CONFIG_SYNC);
if (data)
- (*(__u64 *)data) = m->mdt_lut.lut_mount_count;
+ (*(__u64 *)data) =
+ m->mdt_lut.lut_obd->u.obt.obt_mount_count;
break;
case MD_NO_TRANS:
mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
{
struct lu_device *ld = md2lu_dev(mdt->mdt_child);
- struct obd_device *obd = mdt2obd_dev(mdt);
#ifdef HAVE_QUOTA_SUPPORT
+ struct obd_device *obd = mdt2obd_dev(mdt);
struct md_device *next = mdt->mdt_child;
#endif
int rc;
} else {
mdt_max_threads = MDT_MAX_THREADS;
mdt_min_threads = MDT_MIN_THREADS;
- if (mdt_min_threads < MDT_NUM_THREADS)
- mdt_min_threads = MDT_NUM_THREADS;
}
lprocfs_mdt_init_vars(&lvars);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff