*
* You should have received a copy of the GNU General Public License
* version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
*
* GPL HEADER END
*/
* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2012, 2015, Intel Corporation.
+ * Copyright (c) 2012, 2017, Intel Corporation.
*/
/*
* lustre/osp/osp_object.c
return osp->opd_storage->dd_lu_dev.ld_site->ld_seq_site->ss_node_id;
}
+static inline const char *osp_dto2name(struct osp_object *obj)
+{
+ return obj->opo_obj.do_lu.lo_dev->ld_obd->obd_name;
+}
+
static inline bool is_ost_obj(struct lu_object *lo)
{
return !lu2osp_dev(lo->lo_dev)->opd_connect_mdt;
}
+static inline void __osp_oac_xattr_assignment(struct osp_object *obj,
+ struct osp_xattr_entry *oxe,
+ const struct lu_buf *buf)
+{
+ if (buf->lb_len > 0)
+ memcpy(oxe->oxe_value, buf->lb_buf, buf->lb_len);
+
+ oxe->oxe_vallen = buf->lb_len;
+ oxe->oxe_exist = 1;
+ oxe->oxe_ready = 1;
+}
+
/**
* Assign FID to the OST object.
*
lu_object_assign_fid(env, &o->opo_obj.do_lu, &osi->osi_fid);
}
+#define OXE_DEFAULT_LEN 16
+
+/**
+ * Release reference from the OSP object extended attribute entry.
+ *
+ * If it is the last reference, then free the entry.
+ *
+ * \param[in] oxe pointer to the OSP object extended attribute entry.
+ */
+static inline void osp_oac_xattr_put(struct osp_xattr_entry *oxe)
+{
+ if (atomic_dec_and_test(&oxe->oxe_ref)) {
+ LASSERT(list_empty(&oxe->oxe_list));
+
+ OBD_FREE(oxe, oxe->oxe_buflen);
+ }
+}
+
/**
* Find the named extended attribute in the OSP object attributes cache.
*
spin_lock(&obj->opo_lock);
oxe = osp_oac_xattr_find_locked(obj, name, strlen(name));
- if (oxe != NULL) {
+ if (oxe) {
if (unlink)
list_del_init(&oxe->oxe_list);
else
osp_oac_xattr_find_or_add(struct osp_object *obj, const char *name, size_t len)
{
struct osp_xattr_entry *oxe;
- struct osp_xattr_entry *tmp = NULL;
- size_t namelen = strlen(name);
- size_t size = sizeof(*oxe) + namelen + 1 + len;
+ struct osp_xattr_entry *tmp = NULL;
+ size_t namelen = strlen(name);
+ size_t size = sizeof(*oxe) + namelen + 1 +
+ (len ? len : OXE_DEFAULT_LEN);
oxe = osp_oac_xattr_find(obj, name, false);
- if (oxe != NULL)
+ if (oxe)
return oxe;
OBD_ALLOC(oxe, size);
- if (unlikely(oxe == NULL))
+ if (unlikely(!oxe))
return NULL;
INIT_LIST_HEAD(&oxe->oxe_list);
spin_lock(&obj->opo_lock);
tmp = osp_oac_xattr_find_locked(obj, name, namelen);
- if (tmp == NULL)
+ if (!tmp)
list_add_tail(&oxe->oxe_list, &obj->opo_xattr_list);
else
atomic_inc(&tmp->oxe_ref);
spin_unlock(&obj->opo_lock);
- if (tmp != NULL) {
+ if (tmp) {
OBD_FREE(oxe, size);
oxe = tmp;
}
}
/**
- * Add the given extended attribute to the OSP object attributes cache.
+ * Assign the cached OST-object's EA with the given value.
*
- * If there is an old extended attributed entry with the same name,
- * remove it from the cache and return it via the parameter \a poxe.
+ * If the current EA entry in cache has not enough space to hold the new
+ * value, remove it, create a new one, then assign with the given value.
*
* \param[in] obj pointer to the OSP object
- * \param[in,out] poxe double pointer to the OSP object extended attribute
- * entry: the new extended attribute entry is transferred
- * via such pointer target, and if old the extended
- * attribute entry exists, then it will be returned back
- * via such pointer target.
- * \param[in] len the length of the (new) extended attribute value
- *
- * \retval pointer to the new extended attribute entry
- * \retval NULL for failure cases.
+ * \param[in] oxe pointer to the cached EA entry to be assigned
+ * \param[in] buf pointer to the buffer with new EA value
+ *
+ * \retval pointer to the new created EA entry in cache if
+ * current entry is not big enough; otherwise, the
+ * input 'oxe' will be returned.
*/
static struct osp_xattr_entry *
-osp_oac_xattr_replace(struct osp_object *obj,
- struct osp_xattr_entry **poxe, size_t len)
+osp_oac_xattr_assignment(struct osp_object *obj, struct osp_xattr_entry *oxe,
+ const struct lu_buf *buf)
{
- struct osp_xattr_entry *oxe;
- size_t namelen = (*poxe)->oxe_namelen;
- size_t size = sizeof(*oxe) + namelen + 1 + len;
-
- OBD_ALLOC(oxe, size);
- if (unlikely(oxe == NULL))
- return NULL;
-
- INIT_LIST_HEAD(&oxe->oxe_list);
- oxe->oxe_buflen = size;
- oxe->oxe_namelen = namelen;
- memcpy(oxe->oxe_buf, (*poxe)->oxe_buf, namelen);
- oxe->oxe_value = oxe->oxe_buf + namelen + 1;
- /* One ref is for the caller, the other is for the entry on the list. */
- atomic_set(&oxe->oxe_ref, 2);
+ struct osp_xattr_entry *new = NULL;
+ struct osp_xattr_entry *old = NULL;
+ int namelen = oxe->oxe_namelen;
+ size_t size = sizeof(*oxe) + namelen + 1 + buf->lb_len;
+ bool unlink_only = false;
+
+ if (oxe->oxe_buflen < size) {
+ OBD_ALLOC(new, size);
+ if (likely(new)) {
+ INIT_LIST_HEAD(&new->oxe_list);
+ new->oxe_buflen = size;
+ new->oxe_namelen = namelen;
+ memcpy(new->oxe_buf, oxe->oxe_buf, namelen);
+ new->oxe_value = new->oxe_buf + namelen + 1;
+ /* One ref is for the caller,
+ * the other is for the entry on the list. */
+ atomic_set(&new->oxe_ref, 2);
+ __osp_oac_xattr_assignment(obj, new, buf);
+ } else {
+ unlink_only = true;
+ CWARN("%s: cannot update cached xattr %.*s of "DFID"\n",
+ osp_dto2name(obj), namelen, oxe->oxe_buf,
+ PFID(lu_object_fid(&obj->opo_obj.do_lu)));
+ }
+ }
spin_lock(&obj->opo_lock);
- *poxe = osp_oac_xattr_find_locked(obj, oxe->oxe_buf, namelen);
- LASSERT(*poxe != NULL);
+ old = osp_oac_xattr_find_locked(obj, oxe->oxe_buf, namelen);
+ if (likely(old)) {
+ if (new) {
+ /* Unlink the 'old'. */
+ list_del_init(&old->oxe_list);
- list_del_init(&(*poxe)->oxe_list);
- list_add_tail(&oxe->oxe_list, &obj->opo_xattr_list);
- spin_unlock(&obj->opo_lock);
+ /* Drop the ref for 'old' on list. */
+ osp_oac_xattr_put(old);
- return oxe;
-}
+ /* Drop the ref for current using. */
+ osp_oac_xattr_put(oxe);
+ oxe = new;
-/**
- * Release reference from the OSP object extended attribute entry.
- *
- * If it is the last reference, then free the entry.
- *
- * \param[in] oxe pointer to the OSP object extended attribute entry.
- */
-static inline void osp_oac_xattr_put(struct osp_xattr_entry *oxe)
-{
- if (atomic_dec_and_test(&oxe->oxe_ref)) {
- LASSERT(list_empty(&oxe->oxe_list));
+ /* Insert 'new' into list. */
+ list_add_tail(&new->oxe_list, &obj->opo_xattr_list);
+ } else if (unlink_only) {
+ /* Unlink the 'old'. */
+ list_del_init(&old->oxe_list);
- OBD_FREE(oxe, oxe->oxe_buflen);
+ /* Drop the ref for 'old' on list. */
+ osp_oac_xattr_put(old);
+ } else {
+ __osp_oac_xattr_assignment(obj, oxe, buf);
+ }
+ } else if (new) {
+ /* Drop the ref for current using. */
+ osp_oac_xattr_put(oxe);
+ oxe = new;
+
+ /* Someone unlinked the 'old' by race,
+ * insert the 'new' one into list. */
+ list_add_tail(&new->oxe_list, &obj->opo_xattr_list);
}
+ spin_unlock(&obj->opo_lock);
+
+ return oxe;
}
/**
lustre_swab_obdo(wobdo);
lustre_get_wire_obdo(NULL, lobdo, wobdo);
- spin_lock(&obj->opo_lock);
- la_from_obdo(&obj->opo_attr, lobdo, lobdo->o_valid);
- if (attr != NULL)
- *attr = obj->opo_attr;
- spin_unlock(&obj->opo_lock);
+ if (obj) {
+ spin_lock(&obj->opo_lock);
+ la_from_obdo(&obj->opo_attr, lobdo, lobdo->o_valid);
+ spin_unlock(&obj->opo_lock);
+ }
+ if (attr)
+ la_from_obdo(attr, lobdo, lobdo->o_valid);
return 0;
}
struct osp_update_request *update;
struct object_update_reply *reply;
struct ptlrpc_request *req = NULL;
- int rc = 0;
+ int invalidated, cache = 0, rc = 0;
ENTRY;
if (is_ost_obj(&dt->do_lu) && obj->opo_non_exist)
if (IS_ERR(update))
RETURN(PTR_ERR(update));
- rc = osp_update_rpc_pack(env, attr_get, update, OUT_ATTR_GET,
+ rc = OSP_UPDATE_RPC_PACK(env, out_attr_get_pack, update,
lu_object_fid(&dt->do_lu));
if (rc != 0) {
CERROR("%s: Insert update error "DFID": rc = %d\n",
dev->dd_lu_dev.ld_obd->obd_name,
PFID(lu_object_fid(&dt->do_lu)), rc);
- GOTO(out, rc);
+ GOTO(out_req, rc);
}
+ invalidated = atomic_read(&obj->opo_invalidate_seq);
+
rc = osp_remote_sync(env, osp, update, &req);
+
+ down_read(&obj->opo_invalidate_sem);
+ if (invalidated == atomic_read(&obj->opo_invalidate_seq)) {
+ /* no invalited has came so far, we can cache the attrs */
+ cache = 1;
+ }
+
if (rc != 0) {
if (rc == -ENOENT) {
osp2lu_obj(obj)->lo_header->loh_attr &= ~LOHA_EXISTS;
- obj->opo_non_exist = 1;
+ if (cache)
+ obj->opo_non_exist = 1;
} else {
CERROR("%s:osp_attr_get update error "DFID": rc = %d\n",
dev->dd_lu_dev.ld_obd->obd_name,
if (reply == NULL || reply->ourp_magic != UPDATE_REPLY_MAGIC)
GOTO(out, rc = -EPROTO);
- rc = osp_get_attr_from_reply(env, reply, req, attr, obj, 0);
+ rc = osp_get_attr_from_reply(env, reply, req, attr,
+ cache ? obj : NULL, 0);
if (rc != 0)
GOTO(out, rc);
spin_lock(&obj->opo_lock);
- obj->opo_stale = 0;
+ if (cache)
+ obj->opo_stale = 0;
spin_unlock(&obj->opo_lock);
GOTO(out, rc);
out:
+ up_read(&obj->opo_invalidate_sem);
+
+out_req:
if (req != NULL)
ptlrpc_req_finished(req);
- osp_update_request_destroy(update);
+ osp_update_request_destroy(env, update);
return rc;
}
RETURN(rc);
}
- if (!(attr->la_valid & (LA_UID | LA_GID)))
+ if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
RETURN(0);
- /* track all UID/GID changes via llog */
+ /* track all UID/GID, projid, and layout version changes via llog */
rc = osp_sync_declare_add(env, o, MDS_SETATTR64_REC, th);
return 0;
int rc = 0;
ENTRY;
- /* we're interested in uid/gid changes only */
- if (!(attr->la_valid & (LA_UID | LA_GID)))
+ /* we're interested in uid/gid/projid/layout version changes only */
+ if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
RETURN(0);
if (!is_only_remote_trans(th)) {
- rc = osp_sync_add(env, o, MDS_SETATTR64_REC, th, attr);
- /* XXX: send new uid/gid to OST ASAP? */
+ if (attr->la_flags & LUSTRE_SET_SYNC_FL) {
+ struct ptlrpc_request *req = NULL;
+ struct osp_update_request *update = NULL;
+ struct osp_device *osp = lu2osp_dev(dt->do_lu.lo_dev);
+
+ update = osp_update_request_create(&osp->opd_dt_dev);
+ if (IS_ERR(update))
+ RETURN(PTR_ERR(update));
+
+ rc = OSP_UPDATE_RPC_PACK(env, out_attr_set_pack, update,
+ lu_object_fid(&dt->do_lu),
+ attr);
+ if (rc != 0) {
+ CERROR("%s: update error "DFID": rc = %d\n",
+ osp->opd_obd->obd_name,
+ PFID(lu_object_fid(&dt->do_lu)), rc);
+
+ osp_update_request_destroy(env, update);
+ RETURN(rc);
+ }
+
+ rc = osp_remote_sync(env, osp, update, &req);
+ if (req != NULL)
+ ptlrpc_req_finished(req);
+
+ osp_update_request_destroy(env, update);
+ } else {
+ rc = osp_sync_add(env, o, MDS_SETATTR64_REC, th, attr);
+ /* XXX: send new uid/gid to OST ASAP? */
+ }
} else {
struct lu_attr *la;
la->la_gid = attr->la_gid;
la->la_valid |= LA_GID;
}
+ if (attr->la_valid & LA_PROJID) {
+ la->la_projid = attr->la_projid;
+ la->la_valid |= LA_PROJID;
+ }
spin_unlock(&o->opo_lock);
}
struct osp_object *obj,
void *data, int index, int rc)
{
- struct osp_xattr_entry *oxe = data;
- struct lu_buf *rbuf = &osp_env_info(env)->osi_lb2;
+ struct osp_xattr_entry *oxe = data;
- if (rc == 0) {
+ spin_lock(&obj->opo_lock);
+ if (rc >= 0) {
+ struct lu_buf *rbuf = &osp_env_info(env)->osi_lb2;
size_t len = sizeof(*oxe) + oxe->oxe_namelen + 1;
rc = object_update_result_data_get(reply, rbuf, index);
- if (rc < 0 || rbuf->lb_len > (oxe->oxe_buflen - len)) {
- spin_lock(&obj->opo_lock);
- oxe->oxe_ready = 0;
- spin_unlock(&obj->opo_lock);
- osp_oac_xattr_put(oxe);
+ if (rc == -ENOENT || rc == -ENODATA || rc == 0) {
+ oxe->oxe_exist = 0;
+ oxe->oxe_ready = 1;
+ goto unlock;
+ }
- return rc < 0 ? rc : -ERANGE;
+ if (unlikely(rc < 0) ||
+ rbuf->lb_len > (oxe->oxe_buflen - len)) {
+ oxe->oxe_ready = 0;
+ goto unlock;
}
- spin_lock(&obj->opo_lock);
- oxe->oxe_vallen = rbuf->lb_len;
- memcpy(oxe->oxe_value, rbuf->lb_buf, rbuf->lb_len);
- oxe->oxe_exist = 1;
- oxe->oxe_ready = 1;
- spin_unlock(&obj->opo_lock);
+ __osp_oac_xattr_assignment(obj, oxe, rbuf);
} else if (rc == -ENOENT || rc == -ENODATA) {
- spin_lock(&obj->opo_lock);
oxe->oxe_exist = 0;
oxe->oxe_ready = 1;
- spin_unlock(&obj->opo_lock);
} else {
- spin_lock(&obj->opo_lock);
oxe->oxe_ready = 0;
- spin_unlock(&obj->opo_lock);
}
+unlock:
+ spin_unlock(&obj->opo_lock);
+
+ /* Put the reference obtained in the osp_declare_xattr_get(). */
osp_oac_xattr_put(oxe);
return 0;
struct osp_object *obj = dt2osp_obj(dt);
struct osp_device *osp = lu2osp_dev(dt->do_lu.lo_dev);
struct osp_xattr_entry *oxe;
- __u16 namelen = strlen(name);
int rc = 0;
+ __u16 len;
LASSERT(buf != NULL);
LASSERT(name != NULL);
- /* If only for xattr size, return directly. */
if (unlikely(buf->lb_len == 0))
- return 0;
+ return -EINVAL;
oxe = osp_oac_xattr_find_or_add(obj, name, buf->lb_len);
if (oxe == NULL)
return -ENOMEM;
+ len = strlen(name) + 1;
mutex_lock(&osp->opd_async_requests_mutex);
rc = osp_insert_async_request(env, OUT_XATTR_GET, obj, 1,
- &namelen, (const void **)&name,
+ &len, (const void **)&name,
oxe, buf->lb_len,
osp_xattr_get_interpterer);
if (rc != 0) {
struct ptlrpc_request *req = NULL;
struct object_update_reply *reply;
struct osp_xattr_entry *oxe = NULL;
- const char *dname = dt->do_lu.lo_dev->ld_obd->obd_name;
- int rc = 0;
+ const char *dname = osp_dto2name(obj);
+ int invalidated, rc = 0;
ENTRY;
LASSERT(buf != NULL);
if (unlikely(obj->opo_non_exist))
RETURN(-ENOENT);
+ invalidated = atomic_read(&obj->opo_invalidate_seq);
+
oxe = osp_oac_xattr_find(obj, name, false);
if (oxe != NULL) {
spin_lock(&obj->opo_lock);
}
update = osp_update_request_create(dev);
if (IS_ERR(update))
- GOTO(out, rc = PTR_ERR(update));
+ GOTO(out_req, rc = PTR_ERR(update));
- rc = osp_update_rpc_pack(env, xattr_get, update, OUT_XATTR_GET,
+ rc = OSP_UPDATE_RPC_PACK(env, out_xattr_get_pack, update,
lu_object_fid(&dt->do_lu), name, buf->lb_len);
if (rc != 0) {
CERROR("%s: Insert update error "DFID": rc = %d\n",
dname, PFID(lu_object_fid(&dt->do_lu)), rc);
- GOTO(out, rc);
+ GOTO(out_req, rc);
}
rc = osp_remote_sync(env, osp, update, &req);
+
+ down_read(&obj->opo_invalidate_sem);
+ if (invalidated != atomic_read(&obj->opo_invalidate_seq)) {
+ /* invalidated has been requested, we can't cache the result */
+ if (rc < 0) {
+ if (rc == -ENOENT)
+ dt->do_lu.lo_header->loh_attr &= ~LOHA_EXISTS;
+ GOTO(out, rc);
+ }
+ reply = req_capsule_server_sized_get(&req->rq_pill,
+ &RMF_OUT_UPDATE_REPLY,
+ OUT_UPDATE_REPLY_SIZE);
+ if (reply->ourp_magic != UPDATE_REPLY_MAGIC) {
+ CERROR("%s: Wrong version %x expected %x "DFID
+ ": rc = %d\n", dname, reply->ourp_magic,
+ UPDATE_REPLY_MAGIC,
+ PFID(lu_object_fid(&dt->do_lu)), -EPROTO);
+ GOTO(out, rc = -EPROTO);
+ }
+ rc = object_update_result_data_get(reply, rbuf, 0);
+ GOTO(out, rc);
+ }
+
if (rc < 0) {
if (rc == -ENOENT) {
dt->do_lu.lo_header->loh_attr &= ~LOHA_EXISTS;
}
rc = object_update_result_data_get(reply, rbuf, 0);
- if (rc < 0)
- GOTO(out, rc);
+ if (rc < 0 || rbuf->lb_len == 0) {
+ if (oxe == NULL && rc == -ENODATA) {
+ oxe = osp_oac_xattr_find_or_add(obj, name, buf->lb_len);
+ if (oxe == NULL) {
+ rc = -ENOMEM;
+ CWARN("%s: Fail to add xattr (%s) to cache for "
+ DFID" (1): rc = %d\n", dname, name,
+ PFID(lu_object_fid(&dt->do_lu)), rc);
+ GOTO(out, rc);
+ }
+ }
- if (buf->lb_buf == NULL)
- GOTO(out, rc);
+ if (oxe) {
+ spin_lock(&obj->opo_lock);
+ if (unlikely(rc == -ENODATA)) {
+ oxe->oxe_exist = 0;
+ oxe->oxe_ready = 1;
+ } else {
+ oxe->oxe_ready = 0;
+ }
+ spin_unlock(&obj->opo_lock);
+ }
- if (unlikely(buf->lb_len < rbuf->lb_len))
- GOTO(out, rc = -ERANGE);
+ GOTO(out, rc);
+ }
- memcpy(buf->lb_buf, rbuf->lb_buf, rbuf->lb_len);
+ /* For detecting EA size. */
+ if (!buf->lb_buf)
+ GOTO(out, rc);
- if (oxe == NULL) {
+ if (!oxe) {
oxe = osp_oac_xattr_find_or_add(obj, name, rbuf->lb_len);
- if (oxe == NULL) {
+ if (!oxe) {
CWARN("%s: Fail to add xattr (%s) to "
"cache for "DFID" (2): rc = %d\n",
dname, name, PFID(lu_object_fid(&dt->do_lu)), rc);
}
}
- if (oxe->oxe_buflen - oxe->oxe_namelen - 1 < rbuf->lb_len) {
- struct osp_xattr_entry *old = oxe;
- struct osp_xattr_entry *tmp;
+ oxe = osp_oac_xattr_assignment(obj, oxe, rbuf);
- tmp = osp_oac_xattr_replace(obj, &old, rbuf->lb_len);
- osp_oac_xattr_put(oxe);
- oxe = tmp;
- if (tmp == NULL) {
- CWARN("%s: Fail to update xattr (%s) to "
- "cache for "DFID": rc = %d\n",
- dname, name, PFID(lu_object_fid(&dt->do_lu)), rc);
- spin_lock(&obj->opo_lock);
- old->oxe_ready = 0;
- spin_unlock(&obj->opo_lock);
+ GOTO(out, rc);
- GOTO(out, rc);
- }
+out:
+ up_read(&obj->opo_invalidate_sem);
- /* Drop the ref for entry on list. */
- osp_oac_xattr_put(old);
+out_req:
+ if (rc > 0 && buf->lb_buf) {
+ if (unlikely(buf->lb_len < rbuf->lb_len))
+ rc = -ERANGE;
+ else
+ memcpy(buf->lb_buf, rbuf->lb_buf, rbuf->lb_len);
}
- spin_lock(&obj->opo_lock);
- oxe->oxe_vallen = rbuf->lb_len;
- memcpy(oxe->oxe_value, rbuf->lb_buf, rbuf->lb_len);
- oxe->oxe_exist = 1;
- oxe->oxe_ready = 1;
- spin_unlock(&obj->opo_lock);
-
- GOTO(out, rc);
-
-out:
- if (req != NULL)
+ if (req)
ptlrpc_req_finished(req);
- if (update != NULL && !IS_ERR(update))
- osp_update_request_destroy(update);
+ if (update && !IS_ERR(update))
+ osp_update_request_destroy(env, update);
- if (oxe != NULL)
+ if (oxe)
osp_oac_xattr_put(oxe);
return rc;
const struct lu_buf *buf, const char *name, int fl,
struct thandle *th)
{
- struct osp_object *o = dt2osp_obj(dt);
+ struct osp_object *o = dt2osp_obj(dt);
struct osp_update_request *update;
- struct osp_xattr_entry *oxe;
- int rc;
+ struct osp_xattr_entry *oxe;
+ int rc;
ENTRY;
update = thandle_to_osp_update_request(th);
CDEBUG(D_INODE, DFID" set xattr '%s' with size %zd\n",
PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
- rc = osp_update_rpc_pack(env, xattr_set, update, OUT_XATTR_SET,
+ rc = OSP_UPDATE_RPC_PACK(env, out_xattr_set_pack, update,
lu_object_fid(&dt->do_lu), buf, name, fl);
if (rc != 0)
RETURN(rc);
+ /* Do not cache linkEA that may be self-adjusted by peers
+ * under EA overflow case. */
+ if (strcmp(name, XATTR_NAME_LINK) == 0) {
+ oxe = osp_oac_xattr_find(o, name, true);
+ if (oxe != NULL)
+ osp_oac_xattr_put(oxe);
+
+ RETURN(0);
+ }
+
oxe = osp_oac_xattr_find_or_add(o, name, buf->lb_len);
if (oxe == NULL) {
CWARN("%s: cannot cache xattr '%s' of "DFID"\n",
- dt->do_lu.lo_dev->ld_obd->obd_name,
- name, PFID(lu_object_fid(&dt->do_lu)));
+ osp_dto2name(o), name, PFID(lu_object_fid(&dt->do_lu)));
RETURN(0);
}
- if (oxe->oxe_buflen - oxe->oxe_namelen - 1 < buf->lb_len) {
- struct osp_xattr_entry *old = oxe;
- struct osp_xattr_entry *tmp;
-
- tmp = osp_oac_xattr_replace(o, &old, buf->lb_len);
+ oxe = osp_oac_xattr_assignment(o, oxe, buf);
+ if (oxe)
osp_oac_xattr_put(oxe);
- oxe = tmp;
- if (tmp == NULL) {
- CWARN("%s: cannot update cached xattr '%s' of "DFID"\n",
- dt->do_lu.lo_dev->ld_obd->obd_name,
- name, PFID(lu_object_fid(&dt->do_lu)));
- spin_lock(&o->opo_lock);
- old->oxe_ready = 0;
- spin_unlock(&o->opo_lock);
-
- RETURN(0);
- }
-
- /* Drop the ref for entry on list. */
- osp_oac_xattr_put(old);
- }
-
- spin_lock(&o->opo_lock);
- oxe->oxe_vallen = buf->lb_len;
- memcpy(oxe->oxe_value, buf->lb_buf, buf->lb_len);
- oxe->oxe_exist = 1;
- oxe->oxe_ready = 1;
- spin_unlock(&o->opo_lock);
- osp_oac_xattr_put(oxe);
RETURN(0);
}
const char *name, struct thandle *th)
{
struct osp_update_request *update;
- const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
- struct osp_object *o = dt2osp_obj(dt);
- struct osp_xattr_entry *oxe;
- int rc;
+ const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
+ struct osp_object *o = dt2osp_obj(dt);
+ struct osp_xattr_entry *oxe;
+ int rc;
update = thandle_to_osp_update_request(th);
LASSERT(update != NULL);
- rc = osp_update_rpc_pack(env, xattr_del, update, OUT_XATTR_DEL,
- fid, name);
+ rc = OSP_UPDATE_RPC_PACK(env, out_xattr_del_pack, update, fid, name);
if (rc != 0)
return rc;
return 0;
}
+void osp_obj_invalidate_cache(struct osp_object *obj)
+{
+ struct osp_xattr_entry *oxe;
+ struct osp_xattr_entry *tmp;
+
+ spin_lock(&obj->opo_lock);
+ list_for_each_entry_safe(oxe, tmp, &obj->opo_xattr_list, oxe_list) {
+ oxe->oxe_ready = 0;
+ list_del_init(&oxe->oxe_list);
+ osp_oac_xattr_put(oxe);
+ }
+ obj->opo_attr.la_valid = 0;
+ spin_unlock(&obj->opo_lock);
+}
+
/**
* Implement OSP layer dt_object_operations::do_invalidate() interface.
*
int osp_invalidate(const struct lu_env *env, struct dt_object *dt)
{
struct osp_object *obj = dt2osp_obj(dt);
- struct osp_xattr_entry *oxe;
- struct osp_xattr_entry *tmp;
ENTRY;
+ CDEBUG(D_HA, "Invalidate osp_object "DFID"\n",
+ PFID(lu_object_fid(&dt->do_lu)));
+
+ /* serialize attr/EA set vs. invalidation */
+ down_write(&obj->opo_invalidate_sem);
+
+ /* this should invalidate all in-flights */
+ atomic_inc(&obj->opo_invalidate_seq);
+
spin_lock(&obj->opo_lock);
- list_for_each_entry_safe(oxe, tmp, &obj->opo_xattr_list, oxe_list) {
- oxe->oxe_ready = 0;
- list_del_init(&oxe->oxe_list);
- osp_oac_xattr_put(oxe);
- }
- obj->opo_attr.la_valid = 0;
- obj->opo_stale = 1;
+ /* do not mark new objects stale */
+ if (obj->opo_attr.la_valid)
+ obj->opo_stale = 1;
+ obj->opo_non_exist = 0;
spin_unlock(&obj->opo_lock);
+ osp_obj_invalidate_cache(obj);
+
+ up_write(&obj->opo_invalidate_sem);
+
RETURN(0);
}
*
* If the transaction is a remote transaction and the FID for the OST-object
* has been assigned already, then handle it as creating (remote) MDT object
- * via osp_md_declare_object_create(). This function is usually used for LFSCK
+ * via osp_md_declare_create(). This function is usually used for LFSCK
* to re-create the lost OST object. Otherwise, if it is not replay case, the
* OSP will reserve pre-created object for the subsequent create operation;
* if the MDT side cached pre-created objects are less than some threshold,
* \retval 0 for success
* \retval negative error number on failure
*/
-static int osp_declare_object_create(const struct lu_env *env,
- struct dt_object *dt,
- struct lu_attr *attr,
- struct dt_allocation_hint *hint,
- struct dt_object_format *dof,
- struct thandle *th)
+static int osp_declare_create(const struct lu_env *env, struct dt_object *dt,
+ struct lu_attr *attr,
+ struct dt_allocation_hint *hint,
+ struct dt_object_format *dof, struct thandle *th)
{
struct osp_thread_info *osi = osp_env_info(env);
struct osp_device *d = lu2osp_dev(dt->do_lu.lo_dev);
if (is_only_remote_trans(th) && !fid_is_zero(fid)) {
LASSERT(fid_is_sane(fid));
- rc = osp_md_declare_object_create(env, dt, attr, hint, dof, th);
+ rc = osp_md_declare_create(env, dt, attr, hint, dof, th);
RETURN(rc);
}
if (unlikely(!fid_is_zero(fid))) {
/* replay case: caller knows fid */
- osi->osi_off = sizeof(osi->osi_id) * d->opd_index;
- osi->osi_lb.lb_len = sizeof(osi->osi_id);
- osi->osi_lb.lb_buf = NULL;
-
+ osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off, NULL,
+ d->opd_index);
rc = dt_declare_record_write(env, d->opd_last_used_oid_file,
&osi->osi_lb, osi->osi_off,
local_th);
o->opo_reserved = 1;
/* common for all OSPs file hystorically */
- osi->osi_off = sizeof(osi->osi_id) * d->opd_index;
- osi->osi_lb.lb_len = sizeof(osi->osi_id);
- osi->osi_lb.lb_buf = NULL;
+ osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off, NULL,
+ d->opd_index);
rc = dt_declare_record_write(env, d->opd_last_used_oid_file,
&osi->osi_lb, osi->osi_off,
local_th);
*
* If the transaction is a remote transaction and the FID for the OST-object
* has been assigned already, then handle it as handling MDT object via the
- * osp_md_object_create(). For other cases, the OSP will assign FID to the
+ * osp_md_create(). For other cases, the OSP will assign FID to the
* object to be created, and update last_used Object ID (OID) file.
*
* \param[in] env pointer to the thread context
* \retval 0 for success
* \retval negative error number on failure
*/
-static int osp_object_create(const struct lu_env *env, struct dt_object *dt,
- struct lu_attr *attr,
- struct dt_allocation_hint *hint,
- struct dt_object_format *dof, struct thandle *th)
+static int osp_create(const struct lu_env *env, struct dt_object *dt,
+ struct lu_attr *attr, struct dt_allocation_hint *hint,
+ struct dt_object_format *dof, struct thandle *th)
{
struct osp_thread_info *osi = osp_env_info(env);
struct osp_device *d = lu2osp_dev(dt->do_lu.lo_dev);
!fid_is_zero(lu_object_fid(&dt->do_lu))) {
LASSERT(fid_is_sane(lu_object_fid(&dt->do_lu)));
- rc = osp_md_object_create(env, dt, attr, hint, dof, th);
+ rc = osp_md_create(env, dt, attr, hint, dof, th);
if (rc == 0)
o->opo_non_exist = 0;
if (d->opd_gap_count > 0) {
int count = d->opd_gap_count;
- ostid_set_id(&osi->osi_oi,
- fid_oid(&d->opd_gap_start_fid));
+ rc = ostid_set_id(&osi->osi_oi,
+ fid_oid(&d->opd_gap_start_fid));
+ if (rc) {
+ spin_unlock(&d->opd_pre_lock);
+ RETURN(rc);
+ }
d->opd_gap_count = 0;
spin_unlock(&d->opd_pre_lock);
/* Only need update last_used oid file, seq file will only be update
* during seq rollover */
osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off,
- &d->opd_last_used_fid.f_oid, d->opd_index);
+ &d->opd_last_id, d->opd_index);
rc = dt_record_write(env, d->opd_last_used_oid_file, &osi->osi_lb,
&osi->osi_off, local_th);
* \retval 0 for success
* \retval negative error number on failure
*/
-int osp_declare_object_destroy(const struct lu_env *env,
- struct dt_object *dt, struct thandle *th)
+int osp_declare_destroy(const struct lu_env *env, struct dt_object *dt,
+ struct thandle *th)
{
struct osp_object *o = dt2osp_obj(dt);
struct osp_device *osp = lu2osp_dev(dt->do_lu.lo_dev);
ENTRY;
LASSERT(!osp->opd_connect_mdt);
- rc = osp_sync_declare_add(env, o, MDS_UNLINK64_REC, th);
+
+ if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ))
+ rc = osp_sync_declare_add(env, o, MDS_UNLINK64_REC, th);
RETURN(rc);
}
* \retval 0 for success
* \retval negative error number on failure
*/
-static int osp_object_destroy(const struct lu_env *env, struct dt_object *dt,
- struct thandle *th)
+static int osp_destroy(const struct lu_env *env, struct dt_object *dt,
+ struct thandle *th)
{
struct osp_object *o = dt2osp_obj(dt);
struct osp_device *osp = lu2osp_dev(dt->do_lu.lo_dev);
o->opo_non_exist = 1;
LASSERT(!osp->opd_connect_mdt);
- /* once transaction is committed put proper command on
- * the queue going to our OST. */
- rc = osp_sync_add(env, o, MDS_UNLINK64_REC, th, NULL);
- if (rc < 0)
- RETURN(rc);
+
+ if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ)) {
+ /* once transaction is committed put proper command on
+ * the queue going to our OST. */
+ rc = osp_sync_add(env, o, MDS_UNLINK64_REC, th, NULL);
+ if (rc < 0)
+ RETURN(rc);
+ }
/* not needed in cache any more */
set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
struct dt_object *dt,
const struct dt_rec *rec,
const struct dt_key *key,
- struct thandle *handle,
- int ignore_quota)
+ struct thandle *handle)
{
return -EOPNOTSUPP;
}
struct lu_device *dev = it->ooi_obj->do_lu.lo_dev;
struct osp_device *osp = lu2osp_dev(dev);
struct page **pages;
- struct lu_device *top_device;
struct ptlrpc_request *req = NULL;
struct ptlrpc_bulk_desc *desc;
struct idx_info *ii;
/* 1MB bulk */
npages = min_t(unsigned int, OFD_MAX_BRW_SIZE, 1 << 20);
- npages /= PAGE_CACHE_SIZE;
+ npages /= PAGE_SIZE;
OBD_ALLOC(pages, npages * sizeof(*pages));
if (pages == NULL)
it->ooi_pages = pages;
it->ooi_total_npages = npages;
for (i = 0; i < npages; i++) {
- pages[i] = alloc_page(GFP_IOFS);
+ pages[i] = alloc_page(GFP_NOFS);
if (pages[i] == NULL)
RETURN(-ENOMEM);
}
RETURN(rc);
}
- /* Let's allow this request during recovery, otherwise
- * if the remote target is also in recovery status,
- * it might cause deadlock */
- top_device = dev->ld_site->ls_top_dev;
- if (top_device->ld_obd->obd_recovering)
- req->rq_allow_replay = 1;
-
+ osp_set_req_replay(osp, req);
req->rq_request_portal = OUT_PORTAL;
ii = req_capsule_client_get(&req->rq_pill, &RMF_IDX_INFO);
memset(ii, 0, sizeof(*ii));
PTLRPC_BULK_PUT_SINK | PTLRPC_BULK_BUF_KIOV,
MDS_BULK_PORTAL,
&ptlrpc_bulk_kiov_pin_ops);
- if (desc == NULL) {
- ptlrpc_request_free(req);
- RETURN(-ENOMEM);
- }
+ if (desc == NULL)
+ GOTO(out, rc = -ENOMEM);
for (i = 0; i < npages; i++)
desc->bd_frag_ops->add_kiov_frag(desc, pages[i], 0,
- PAGE_CACHE_SIZE);
+ PAGE_SIZE);
ptlrpc_request_set_replen(req);
rc = ptlrpc_queue_wait(req);
GOTO(out, rc = -EPROTO);
npages = (ii->ii_count + LU_PAGE_COUNT - 1) >>
- (PAGE_CACHE_SHIFT - LU_PAGE_SHIFT);
+ (PAGE_SHIFT - LU_PAGE_SHIFT);
if (npages > it->ooi_total_npages) {
CERROR("%s: returned more pages than expected, %u > %u\n",
osp->opd_obd->obd_name, npages, it->ooi_total_npages);
GOTO(out, rc = -EINVAL);
}
+ it->ooi_rec_size = ii->ii_recsize;
it->ooi_valid_npages = npages;
if (ptlrpc_rep_need_swab(req))
it->ooi_swab = 1;
it->ooi_pos_ent++;
if (it->ooi_pos_ent < idxpage->lip_nr) {
- it->ooi_ent =
+ if (it->ooi_rec_size ==
+ sizeof(struct lu_orphan_rec_v3)) {
+ it->ooi_ent =
+ (struct lu_orphan_ent_v3 *)idxpage->lip_entries+
+ it->ooi_pos_ent;
+ if (it->ooi_swab)
+ lustre_swab_orphan_ent_v3(it->ooi_ent);
+ } else if (it->ooi_rec_size ==
+ sizeof(struct lu_orphan_rec_v2)) {
+ it->ooi_ent =
+ (struct lu_orphan_ent_v2 *)idxpage->lip_entries+
+ it->ooi_pos_ent;
+ if (it->ooi_swab)
+ lustre_swab_orphan_ent_v2(it->ooi_ent);
+ } else {
+ it->ooi_ent =
(struct lu_orphan_ent *)idxpage->lip_entries +
it->ooi_pos_ent;
- if (it->ooi_swab)
- lustre_swab_orphan_ent(it->ooi_ent);
+ if (it->ooi_swab)
+ lustre_swab_orphan_ent(it->ooi_ent);
+ }
RETURN(0);
}
}
static int osp_orphan_it_rec(const struct lu_env *env, const struct dt_it *di,
struct dt_rec *rec, __u32 attr)
{
- struct osp_it *it = (struct osp_it *)di;
- struct lu_orphan_ent *ent = (struct lu_orphan_ent *)it->ooi_ent;
+ struct osp_it *it = (struct osp_it *)di;
+
+ if (likely(it->ooi_ent)) {
+ if (it->ooi_rec_size == sizeof(struct lu_orphan_rec_v3)) {
+ struct lu_orphan_ent_v3 *ent =
+ (struct lu_orphan_ent_v3 *)it->ooi_ent;
- if (likely(ent != NULL)) {
- *(struct lu_orphan_rec *)rec = ent->loe_rec;
+ *(struct lu_orphan_rec_v3 *)rec = ent->loe_rec;
+ } else if (it->ooi_rec_size ==
+ sizeof(struct lu_orphan_rec_v2)) {
+ struct lu_orphan_ent_v2 *ent =
+ (struct lu_orphan_ent_v2 *)it->ooi_ent;
+
+ *(struct lu_orphan_rec_v2 *)rec = ent->loe_rec;
+ } else {
+ struct lu_orphan_ent *ent =
+ (struct lu_orphan_ent *)it->ooi_ent;
+
+ *(struct lu_orphan_rec *)rec = ent->loe_rec;
+ }
return 0;
}
.do_xattr_get = osp_xattr_get,
.do_declare_xattr_set = osp_declare_xattr_set,
.do_xattr_set = osp_xattr_set,
- .do_declare_create = osp_declare_object_create,
- .do_create = osp_object_create,
- .do_declare_destroy = osp_declare_object_destroy,
- .do_destroy = osp_object_destroy,
+ .do_declare_create = osp_declare_create,
+ .do_create = osp_create,
+ .do_declare_destroy = osp_declare_destroy,
+ .do_destroy = osp_destroy,
.do_index_try = osp_index_try,
};
spin_lock_init(&po->opo_lock);
o->lo_header->loh_attr |= LOHA_REMOTE;
INIT_LIST_HEAD(&po->opo_xattr_list);
+ INIT_LIST_HEAD(&po->opo_invalidate_cb_list);
+ init_rwsem(&po->opo_invalidate_sem);
if (is_ost_obj(o)) {
po->opo_obj.do_ops = &osp_obj_ops;
d->opd_pre_reserved--;
spin_unlock(&d->opd_pre_lock);
+ /*
+ * Check that osp_precreate_cleanup_orphans is not blocked
+ * due to opd_pre_reserved > 0.
+ */
+ if (unlikely(d->opd_pre_reserved == 0 &&
+ (d->opd_pre_recovering || d->opd_pre_status)))
+ wake_up(&d->opd_pre_waitq);
+
/* not needed in cache any more */
set_bit(LU_OBJECT_HEARD_BANSHEE, &o->lo_header->loh_flags);
}