X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosp%2Fosp_object.c;h=1ff4a88bfd23595a72f127ead1bbc799b73e1dee;hp=07aeeec10d11bc0b4ae489f877bb52eb43e2cc97;hb=044fbff00b25f127560338d35caa4d89faa4c207;hpb=c5e7e6f29afdb62915219a2cb2bfa09ff3db8034 diff --git a/lustre/osp/osp_object.c b/lustre/osp/osp_object.c index 07aeeec..1ff4a88 100644 --- a/lustre/osp/osp_object.c +++ b/lustre/osp/osp_object.c @@ -85,6 +85,39 @@ * or osp_declare_xattr_get(). That is usually for LFSCK purpose, * but it also can be shared by others. * + * + * XXX: NOT prepare out RPC for remote transaction. ((please refer to the + * comment of osp_trans_create() for remote transaction) + * + * According to our current transaction/dt_object_lock framework (to make + * the cross-MDTs modification for DNE1 to be workable), the transaction + * sponsor will start the transaction firstly, then try to acquire related + * dt_object_lock if needed. Under such rules, if we want to prepare the + * OUT RPC in the transaction declare phase, then related attr/xattr + * should be known without dt_object_lock. But such condition maybe not + * true for some remote transaction case. For example: + * + * For linkEA repairing (by LFSCK) case, before the LFSCK thread obtained + * the dt_object_lock on the target MDT-object, it cannot know whether + * the MDT-object has linkEA or not, neither invalid or not. + * + * Since the LFSCK thread cannot hold dt_object_lock before the remote + * transaction start (otherwise there will be some potential deadlock), + * it cannot prepare related OUT RPC for repairing during the declare + * phase as other normal transactions do. + * + * To resolve the trouble, we will make OSP to prepare related OUT RPC + * after remote transaction started, and trigger the remote updating + * (send RPC) when trans_stop. Then the up layer users, such as LFSCK, + * can follow the general rule to handle trans_start/dt_object_lock + * for repairing linkEA inconsistency without distinguishing remote + * MDT-object. + * + * In fact, above solution for remote transaction should be the normal + * model without considering DNE1. The trouble brought by DNE1 will be + * resolved in DNE2. At that time, this patch can be removed. + * + * * Author: Alex Zhuravlev * Author: Mikhail Pershin */ @@ -135,7 +168,7 @@ static void osp_object_assign_fid(const struct lu_env *env, * \retval 0 for success * \retval negative error number on failure */ -static int osp_oac_init(struct osp_object *obj) +int osp_oac_init(struct osp_object *obj) { struct osp_object_attr *ooa; @@ -286,7 +319,7 @@ osp_oac_xattr_find_or_add(struct osp_object *obj, const char *name, size_t len) * * \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 transfered + * 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. @@ -460,13 +493,11 @@ static int osp_attr_get_interpterer(const struct lu_env *env, * * \param[in] env pointer to the thread context * \param[in] dt pointer to the OSP layer dt_object - * \param[in] capa the capability for this operation * * \retval 0 for success * \retval negative error number on failure */ -static int osp_declare_attr_get(const struct lu_env *env, struct dt_object *dt, - struct lustre_capa *capa) +static int osp_declare_attr_get(const struct lu_env *env, struct dt_object *dt) { struct osp_object *obj = dt2osp_obj(dt); struct osp_device *osp = lu2osp_dev(dt->do_lu.lo_dev); @@ -500,18 +531,17 @@ static int osp_declare_attr_get(const struct lu_env *env, struct dt_object *dt, * \param[in] env pointer to the thread context * \param[in] dt pointer to the OSP layer dt_object * \param[out] attr pointer to the buffer to hold the output attribute - * \param[in] capa the capability for this operation * * \retval 0 for success * \retval negative error number on failure */ int osp_attr_get(const struct lu_env *env, struct dt_object *dt, - struct lu_attr *attr, struct lustre_capa *capa) + struct lu_attr *attr) { struct osp_device *osp = lu2osp_dev(dt->do_lu.lo_dev); struct osp_object *obj = dt2osp_obj(dt); struct dt_device *dev = &osp->opd_dt_dev; - struct dt_update_request *update; + struct osp_update_request *update; struct object_update_reply *reply; struct ptlrpc_request *req = NULL; int rc = 0; @@ -531,12 +561,12 @@ int osp_attr_get(const struct lu_env *env, struct dt_object *dt, spin_unlock(&obj->opo_lock); } - update = dt_update_request_create(dev); + update = osp_update_request_create(dev); if (IS_ERR(update)) RETURN(PTR_ERR(update)); - rc = out_attr_get_pack(env, &update->dur_buf, - lu_object_fid(&dt->do_lu)); + rc = osp_update_rpc_pack(env, attr_get, update, OUT_ATTR_GET, + 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, @@ -545,7 +575,7 @@ int osp_attr_get(const struct lu_env *env, struct dt_object *dt, GOTO(out, rc); } - rc = out_remote_sync(env, osp->opd_obd->u.cli.cl_import, update, &req); + rc = osp_remote_sync(env, osp, update, &req); if (rc != 0) { if (rc == -ENOENT) { osp2lu_obj(obj)->lo_header->loh_attr &= ~LOHA_EXISTS; @@ -577,20 +607,34 @@ out: if (req != NULL) ptlrpc_req_finished(req); - dt_update_request_destroy(update); + osp_update_request_destroy(update); return rc; } -static int __osp_attr_set(const struct lu_env *env, struct dt_object *dt, - const struct lu_attr *attr, struct thandle *th) +/** + * Implement OSP layer dt_object_operations::do_declare_attr_set() interface. + * + * If the transaction is not remote one, then declare the credits that will + * be used for the subsequent llog record for the object's attributes. + * + * \param[in] env pointer to the thread context + * \param[in] dt pointer to the OSP layer dt_object + * \param[in] attr pointer to the attribute to be set + * \param[in] th pointer to the transaction handler + * + * \retval 0 for success + * \retval negative error number on failure + */ +static int osp_declare_attr_set(const struct lu_env *env, struct dt_object *dt, + const struct lu_attr *attr, struct thandle *th) { struct osp_device *d = lu2osp_dev(dt->do_lu.lo_dev); struct osp_object *o = dt2osp_obj(dt); - struct lu_attr *la; - int rc = 0; - ENTRY; + int rc; + if (is_only_remote_trans(th)) + return osp_md_declare_attr_set(env, dt, attr, th); /* * Usually we don't allow server stack to manipulate size * but there is a special case when striping is created @@ -619,97 +663,17 @@ static int __osp_attr_set(const struct lu_env *env, struct dt_object *dt, LASSERT(!dt_object_exists(dt)); osp_object_assign_fid(env, d, o); rc = osp_object_truncate(env, dt, attr->la_size); - if (rc) + if (rc != 0) RETURN(rc); } if (!(attr->la_valid & (LA_UID | LA_GID))) RETURN(0); - if (!is_only_remote_trans(th)) { - if (o->opo_new) - /* no need in logging for new objects being created */ - RETURN(0); - - /* - * track all UID/GID changes via llog - */ - rc = osp_sync_declare_add(env, o, MDS_SETATTR64_REC, th); - } else { - /* It is for OST-object attr_set directly without updating - * local MDT-object attribute. It is usually used by LFSCK. */ - rc = __osp_md_attr_set(env, dt, attr, th); - } - - if (rc != 0 || o->opo_ooa == NULL) - RETURN(rc); - - /* Update the OSP object attributes cache. */ - la = &o->opo_ooa->ooa_attr; - spin_lock(&o->opo_lock); - if (attr->la_valid & LA_UID) { - la->la_uid = attr->la_uid; - la->la_valid |= LA_UID; - } - - if (attr->la_valid & LA_GID) { - la->la_gid = attr->la_gid; - la->la_valid |= LA_GID; - } - spin_unlock(&o->opo_lock); - - RETURN(0); -} - -/** - * Implement OSP layer dt_object_operations::do_declare_attr_set() interface. - * XXX: NOT prepare set_{attr,xattr} RPC for remote transaction. - * - * According to our current transaction/dt_object_lock framework (to make - * the cross-MDTs modification for DNE1 to be workable), the transaction - * sponsor will start the transaction firstly, then try to acquire related - * dt_object_lock if needed. Under such rules, if we want to prepare the - * set_{attr,xattr} RPC in the RPC declare phase, then related attr/xattr - * should be known without dt_object_lock. But such condition maybe not - * true for some remote transaction case. For example: - * - * For linkEA repairing (by LFSCK) case, before the LFSCK thread obtained - * the dt_object_lock on the target MDT-object, it cannot know whether - * the MDT-object has linkEA or not, neither invalid or not. - * - * Since the LFSCK thread cannot hold dt_object_lock before the (remote) - * transaction start (otherwise there will be some potential deadlock), - * it cannot prepare related RPC for repairing during the declare phase - * as other normal transactions do. - * - * To resolve the trouble, we will make OSP to prepare related RPC - * (set_attr/set_xattr/del_xattr) after remote transaction started, - * and trigger the remote updating (RPC sending) when trans_stop. - * Then the up layer users, such as LFSCK, can follow the general - * rule to handle trans_start/dt_object_lock for repairing linkEA - * inconsistency without distinguishing remote MDT-object. - * - * In fact, above solution for remote transaction should be the normal - * model without considering DNE1. The trouble brought by DNE1 will be - * resolved in DNE2. At that time, this patch can be removed. - * - * \param[in] env pointer to the thread context - * \param[in] dt pointer to the OSP layer dt_object - * \param[in] attr pointer to the attribute to be set - * \param[in] th pointer to the transaction handler - * - * \retval 0 for success - * \retval negative error number on failure - */ -static int osp_declare_attr_set(const struct lu_env *env, struct dt_object *dt, - const struct lu_attr *attr, struct thandle *th) -{ - int rc = 0; - - if (!is_only_remote_trans(th)) - rc = __osp_attr_set(env, dt, attr, th); + /* track all UID/GID changes via llog */ + rc = osp_sync_declare_add(env, o, MDS_SETATTR64_REC, th); - return rc; + return 0; } /** @@ -717,11 +681,10 @@ static int osp_declare_attr_set(const struct lu_env *env, struct dt_object *dt, * * Set attribute to the specified OST object. * - * If the transaction is a remote transaction, then related modification - * sub-request has been added in the declare phase and related OUT RPC - * has been triggered at transaction start. Otherwise it will generate - * a MDS_SETATTR64_REC record in the llog. There is a dedicated thread - * to handle the llog asynchronously. + * If the transaction is a remote one, then add OUT_ATTR_SET sub-request + * in the OUT RPC that will be flushed when the remote transaction stop. + * Otherwise, it will generate a MDS_SETATTR64_REC record in the llog that + * will be handled by a dedicated thread asynchronously. * * If the attribute entry exists in the OSP object attributes cache, * then update the cached attribute according to given attribute. @@ -730,42 +693,50 @@ static int osp_declare_attr_set(const struct lu_env *env, struct dt_object *dt, * \param[in] dt pointer to the OSP layer dt_object * \param[in] attr pointer to the attribute to be set * \param[in] th pointer to the transaction handler - * \param[in] capa the capability for this operation * * \retval 0 for success * \retval negative error number on failure */ static int osp_attr_set(const struct lu_env *env, struct dt_object *dt, - const struct lu_attr *attr, struct thandle *th, - struct lustre_capa *capa) + const struct lu_attr *attr, struct thandle *th) { struct osp_object *o = dt2osp_obj(dt); int rc = 0; ENTRY; - if (is_only_remote_trans(th)) { - rc = __osp_attr_set(env, dt, attr, th); - if (rc == 0 && o->opo_new) - o->opo_new = 0; - - RETURN(rc); - } - /* we're interested in uid/gid changes only */ if (!(attr->la_valid & (LA_UID | LA_GID))) RETURN(0); - /* new object, the very first ->attr_set() - * initializing attributes needs no logging - * all subsequent one are subject to the - * logging and synchronization with OST */ - if (o->opo_new) { - o->opo_new = 0; - 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? */ + } else { + struct lu_attr *la; + + /* It is for OST-object attr_set directly without updating + * local MDT-object attribute. It is usually used by LFSCK. */ + rc = osp_md_attr_set(env, dt, attr, th); + CDEBUG(D_INFO, "(1) set attr "DFID": rc = %d\n", + PFID(&dt->do_lu.lo_header->loh_fid), rc); + + if (rc != 0 || o->opo_ooa == NULL) + RETURN(rc); - rc = osp_sync_add(env, o, MDS_SETATTR64_REC, th, attr); - /* XXX: send new uid/gid to OST ASAP? */ + /* Update the OSP object attributes cache. */ + la = &o->opo_ooa->ooa_attr; + spin_lock(&o->opo_lock); + if (attr->la_valid & LA_UID) { + la->la_uid = attr->la_uid; + la->la_valid |= LA_UID; + } + + if (attr->la_valid & LA_GID) { + la->la_gid = attr->la_gid; + la->la_valid |= LA_GID; + } + spin_unlock(&o->opo_lock); + } RETURN(rc); } @@ -848,14 +819,12 @@ static int osp_xattr_get_interpterer(const struct lu_env *env, * \param[in] dt pointer to the OSP layer dt_object * \param[out] buf pointer to the lu_buf to hold the extended attribute * \param[in] name the name for the expected extended attribute - * \param[in] capa the capability for this operation * * \retval 0 for success * \retval negative error number on failure */ static int osp_declare_xattr_get(const struct lu_env *env, struct dt_object *dt, - struct lu_buf *buf, const char *name, - struct lustre_capa *capa) + struct lu_buf *buf, const char *name) { struct osp_object *obj = dt2osp_obj(dt); struct osp_device *osp = lu2osp_dev(dt->do_lu.lo_dev); @@ -888,19 +857,21 @@ static int osp_declare_xattr_get(const struct lu_env *env, struct dt_object *dt, mutex_unlock(&osp->opd_async_requests_mutex); osp_oac_xattr_put(oxe); } else { - struct dt_update_request *update; + struct osp_update_request *our; + struct osp_update_request_sub *ours; /* XXX: Currently, we trigger the batched async OUT * RPC via dt_declare_xattr_get(). It is not * perfect solution, but works well now. * * We will improve it in the future. */ - update = osp->opd_async_requests; - if (update != NULL && update->dur_buf.ub_req != NULL && - update->dur_buf.ub_req->ourq_count > 0) { + our = osp->opd_async_requests; + ours = osp_current_object_update_request(our); + if (ours != NULL && ours->ours_req != NULL && + ours->ours_req->ourq_count > 0) { osp->opd_async_requests = NULL; mutex_unlock(&osp->opd_async_requests_mutex); - rc = osp_unplug_async_request(env, osp, update); + rc = osp_unplug_async_request(env, osp, our); } else { mutex_unlock(&osp->opd_async_requests_mutex); } @@ -928,20 +899,18 @@ static int osp_declare_xattr_get(const struct lu_env *env, struct dt_object *dt, * \param[in] dt pointer to the OSP layer dt_object * \param[out] buf pointer to the lu_buf to hold the extended attribute * \param[in] name the name for the expected extended attribute - * \param[in] capa the capability for this operation * * \retval 0 for success * \retval negative error number on failure */ int osp_xattr_get(const struct lu_env *env, struct dt_object *dt, - struct lu_buf *buf, const char *name, - struct lustre_capa *capa) + struct lu_buf *buf, const char *name) { struct osp_device *osp = lu2osp_dev(dt->do_lu.lo_dev); struct osp_object *obj = dt2osp_obj(dt); struct dt_device *dev = &osp->opd_dt_dev; struct lu_buf *rbuf = &osp_env_info(env)->osi_lb2; - struct dt_update_request *update = NULL; + struct osp_update_request *update = NULL; struct ptlrpc_request *req = NULL; struct object_update_reply *reply; struct osp_xattr_entry *oxe = NULL; @@ -992,19 +961,19 @@ unlock: spin_unlock(&obj->opo_lock); } - update = dt_update_request_create(dev); + update = osp_update_request_create(dev); if (IS_ERR(update)) GOTO(out, rc = PTR_ERR(update)); - rc = out_xattr_get_pack(env, &update->dur_buf, - lu_object_fid(&dt->do_lu), name); + rc = osp_update_rpc_pack(env, xattr_get, update, OUT_XATTR_GET, + lu_object_fid(&dt->do_lu), name); if (rc != 0) { CERROR("%s: Insert update error "DFID": rc = %d\n", dname, PFID(lu_object_fid(&dt->do_lu)), rc); GOTO(out, rc); } - rc = out_remote_sync(env, osp->opd_obd->u.cli.cl_import, update, &req); + rc = osp_remote_sync(env, osp, update, &req); if (rc != 0) { if (rc == -ENOENT) { dt->do_lu.lo_header->loh_attr &= ~LOHA_EXISTS; @@ -1110,7 +1079,7 @@ out: ptlrpc_req_finished(req); if (update != NULL && !IS_ERR(update)) - dt_update_request_destroy(update); + osp_update_request_destroy(update); if (oxe != NULL) osp_oac_xattr_put(oxe); @@ -1118,85 +1087,17 @@ out: return rc; } -static int __osp_xattr_set(const struct lu_env *env, struct dt_object *dt, - const struct lu_buf *buf, const char *name, - int flag, struct thandle *th) -{ - struct osp_object *o = dt2osp_obj(dt); - struct dt_update_request *update; - struct osp_xattr_entry *oxe; - int rc; - ENTRY; - - LASSERT(buf->lb_len > 0 && buf->lb_buf != NULL); - - update = dt_update_request_find_or_create(th, dt); - if (IS_ERR(update)) { - CERROR("%s: Get OSP update buf failed "DFID": rc = %d\n", - dt->do_lu.lo_dev->ld_obd->obd_name, - PFID(lu_object_fid(&dt->do_lu)), - (int)PTR_ERR(update)); - - RETURN(PTR_ERR(update)); - } - - rc = out_xattr_set_pack(env, &update->dur_buf, - lu_object_fid(&dt->do_lu), - buf, name, flag, update->dur_batchid); - if (rc != 0 || o->opo_ooa == NULL) - RETURN(rc); - - 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))); - - 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); - 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); -} - /** * Implement OSP layer dt_object_operations::do_declare_xattr_set() interface. * * Declare that the caller will set extended attribute to the specified * MDT/OST object. * - * This function will add an OUT_XATTR_SET sub-request to the per - * OSP-transaction based request queue which will be flushed when - * the transaction starts. + * If it is non-remote transaction, it will add an OUT_XATTR_SET sub-request + * to the OUT RPC that will be flushed when the transaction start. And if the + * OSP attributes cache is initialized, then check whether the name extended + * attribute entry exists in the cache or not. If yes, replace it; otherwise, + * add the extended attribute to the cache. * * \param[in] env pointer to the thread context * \param[in] dt pointer to the OSP layer dt_object @@ -1213,14 +1114,7 @@ int osp_declare_xattr_set(const struct lu_env *env, struct dt_object *dt, const struct lu_buf *buf, const char *name, int flag, struct thandle *th) { - int rc = 0; - - /* Please check the comment in osp_attr_set() for handling - * remote transaction. */ - if (!is_only_remote_trans(th)) - rc = __osp_xattr_set(env, dt, buf, name, flag, th); - - return rc; + return osp_trans_update_request_create(th); } /** @@ -1228,12 +1122,10 @@ int osp_declare_xattr_set(const struct lu_env *env, struct dt_object *dt, * * Set extended attribute to the specified MDT/OST object. * - * The real modification sub-request has been added in the declare phase - * and related (OUT) RPC has been triggered when transaction start. - * - * If the OSP attributes cache is initialized, then check whether the name - * extended attribute entry exists in the cache or not. If yes, replace it; - * otherwise, add the extended attribute to the cache. + * Add an OUT_XATTR_SET sub-request into the OUT RPC that will be flushed in + * the transaction stop. And if the OSP attributes cache is initialized, then + * check whether the name extended attribute entry exists in the cache or not. + * If yes, replace it; otherwise, add the extended attribute to the cache. * * \param[in] env pointer to the thread context * \param[in] dt pointer to the OSP layer dt_object @@ -1242,55 +1134,71 @@ int osp_declare_xattr_set(const struct lu_env *env, struct dt_object *dt, * \param[in] fl to indicate the detailed set operation: LU_XATTR_CREATE * or LU_XATTR_REPLACE or others * \param[in] th pointer to the transaction handler - * \param[in] capa the capability for this operation * * \retval 0 for success * \retval negative error number on failure */ int osp_xattr_set(const struct lu_env *env, struct dt_object *dt, const struct lu_buf *buf, const char *name, int fl, - struct thandle *th, struct lustre_capa *capa) + struct thandle *th) { - int rc = 0; + struct osp_object *o = dt2osp_obj(dt); + struct osp_update_request *update; + struct osp_xattr_entry *oxe; + int rc; + ENTRY; - CDEBUG(D_INFO, "xattr %s set object "DFID"\n", name, - PFID(&dt->do_lu.lo_header->loh_fid)); + update = thandle_to_osp_update_request(th); + LASSERT(update != NULL); - /* Please check the comment in osp_attr_set() for handling - * remote transaction. */ - if (is_only_remote_trans(th)) - rc = __osp_xattr_set(env, dt, buf, name, fl, th); + CDEBUG(D_INODE, DFID" set xattr '%s' with size %zd\n", + PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len); - return rc; -} + rc = osp_update_rpc_pack(env, xattr_set, update, OUT_XATTR_SET, + lu_object_fid(&dt->do_lu), buf, name, fl); + if (rc != 0 || o->opo_ooa == NULL) + RETURN(rc); -static int __osp_xattr_del(const struct lu_env *env, struct dt_object *dt, - const char *name, struct thandle *th) -{ - struct dt_update_request *update; - const struct lu_fid *fid; - struct osp_object *o = dt2osp_obj(dt); - struct osp_xattr_entry *oxe; - int rc; + 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))); - update = dt_update_request_find_or_create(th, dt); - if (IS_ERR(update)) - return PTR_ERR(update); + RETURN(0); + } - fid = lu_object_fid(&dt->do_lu); + if (oxe->oxe_buflen - oxe->oxe_namelen - 1 < buf->lb_len) { + struct osp_xattr_entry *old = oxe; + struct osp_xattr_entry *tmp; - rc = out_xattr_del_pack(env, &update->dur_buf, fid, name, - update->dur_batchid); + tmp = osp_oac_xattr_replace(o, &old, buf->lb_len); + 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); - if (rc != 0 || o->opo_ooa == NULL) - return rc; + RETURN(0); + } - oxe = osp_oac_xattr_find(o, name, true); - if (oxe != NULL) /* Drop the ref for entry on list. */ - osp_oac_xattr_put(oxe); + osp_oac_xattr_put(old); + } - return 0; + 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); } /** @@ -1299,9 +1207,10 @@ static int __osp_xattr_del(const struct lu_env *env, struct dt_object *dt, * Declare that the caller will delete extended attribute on the specified * MDT/OST object. * - * This function will add an OUT_XATTR_DEL sub-request to the per - * OSP-transaction based request queue which will be flushed when - * transaction start. + * If it is non-remote transaction, it will add an OUT_XATTR_DEL sub-request + * to the OUT RPC that will be flushed when the transaction start. And if the + * name extended attribute entry exists in the OSP attributes cache, then remove + * it from the cache. * * \param[in] env pointer to the thread context * \param[in] dt pointer to the OSP layer dt_object @@ -1314,14 +1223,7 @@ static int __osp_xattr_del(const struct lu_env *env, struct dt_object *dt, int osp_declare_xattr_del(const struct lu_env *env, struct dt_object *dt, const char *name, struct thandle *th) { - int rc = 0; - - /* Please check the comment in osp_attr_set() for handling - * remote transaction. */ - if (!is_only_remote_trans(th)) - rc = __osp_xattr_del(env, dt, name, th); - - return rc; + return osp_trans_update_request_create(th); } /** @@ -1329,36 +1231,42 @@ int osp_declare_xattr_del(const struct lu_env *env, struct dt_object *dt, * * Delete extended attribute on the specified MDT/OST object. * - * The real modification sub-request has been added in the declare phase - * and related (OUT) RPC has been triggered when transaction start. - * - * If the name extended attribute entry exists in the OSP attributes - * cache, then remove it from the cache. + * If it is remote transaction, it will add an OUT_XATTR_DEL sub-request into + * the OUT RPC that will be flushed when the transaction stop. And if the name + * extended attribute entry exists in the OSP attributes cache, then remove it + * from the cache. * * \param[in] env pointer to the thread context * \param[in] dt pointer to the OSP layer dt_object * \param[in] name the name of the extended attribute to be set * \param[in] th pointer to the transaction handler - * \param[in] capa the capability for this operation * * \retval 0 for success * \retval negative error number on failure */ int osp_xattr_del(const struct lu_env *env, struct dt_object *dt, - const char *name, struct thandle *th, - struct lustre_capa *capa) + const char *name, struct thandle *th) { - int rc = 0; + 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; - CDEBUG(D_INFO, "xattr %s del object "DFID"\n", name, - PFID(&dt->do_lu.lo_header->loh_fid)); + update = thandle_to_osp_update_request(th); + LASSERT(update != NULL); - /* Please check the comment in osp_attr_set() for handling - * remote transaction. */ - if (is_only_remote_trans(th)) - rc = __osp_xattr_del(env, dt, name, th); + rc = osp_update_rpc_pack(env, xattr_del, update, OUT_XATTR_DEL, + fid, name); + if (rc != 0 || o->opo_ooa == NULL) + return rc; - return rc; + oxe = osp_oac_xattr_find(o, name, true); + if (oxe != NULL) + /* Drop the ref for entry on list. */ + osp_oac_xattr_put(oxe); + + return 0; } /** @@ -1366,15 +1274,13 @@ int osp_xattr_del(const struct lu_env *env, struct dt_object *dt, * * Declare that the caller will create the OST object. * - * If the transaction is a remote transaction (please refer to the - * comment of osp_trans_create() for remote transaction), then the FID - * for the OST object has been assigned already, and will be handled - * as create (remote) MDT object via osp_md_declare_object_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, then - * it will wakeup the pre-create thread. + * 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 + * 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, + * then it will wakeup the pre-create thread. * * \param[in] env pointer to the thread context * \param[in] dt pointer to the OSP layer dt_object @@ -1398,11 +1304,12 @@ static int osp_declare_object_create(const struct lu_env *env, struct osp_device *d = lu2osp_dev(dt->do_lu.lo_dev); struct osp_object *o = dt2osp_obj(dt); const struct lu_fid *fid = lu_object_fid(&dt->do_lu); + struct thandle *local_th; int rc = 0; ENTRY; - if (is_only_remote_trans(th)) { + 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); @@ -1426,13 +1333,19 @@ static int osp_declare_object_create(const struct lu_env *env, */ /* rc = osp_sync_declare_add(env, o, MDS_UNLINK64_REC, th); */ + local_th = osp_get_storage_thandle(env, th, d); + if (IS_ERR(local_th)) + RETURN(PTR_ERR(local_th)); + 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; + rc = dt_declare_record_write(env, d->opd_last_used_oid_file, - &osi->osi_lb, osi->osi_off, th); + &osi->osi_lb, osi->osi_off, + local_th); RETURN(rc); } @@ -1457,7 +1370,8 @@ static int osp_declare_object_create(const struct lu_env *env, osi->osi_lb.lb_len = sizeof(osi->osi_id); osi->osi_lb.lb_buf = NULL; rc = dt_declare_record_write(env, d->opd_last_used_oid_file, - &osi->osi_lb, osi->osi_off, th); + &osi->osi_lb, osi->osi_off, + local_th); } else { /* not needed in the cache anymore */ set_bit(LU_OBJECT_HEARD_BANSHEE, @@ -1471,12 +1385,9 @@ static int osp_declare_object_create(const struct lu_env *env, * * Create the OST object. * - * For remote transaction case, the real create sub-request has been - * added in the declare phase and related (OUT) RPC has been triggered - * at transaction start. Here, like creating (remote) MDT object, the - * OSP will mark the object existence via osp_md_object_create(). - * - * For non-remote transaction case, the OSP will assign FID to the + * 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 * object to be created, and update last_used Object ID (OID) file. * * \param[in] env pointer to the thread context @@ -1500,9 +1411,11 @@ static int osp_object_create(const struct lu_env *env, struct dt_object *dt, struct osp_object *o = dt2osp_obj(dt); int rc = 0; struct lu_fid *fid = &osi->osi_fid; + struct thandle *local_th; ENTRY; - if (is_only_remote_trans(th)) { + if (is_only_remote_trans(th) && + !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); @@ -1541,6 +1454,9 @@ static int osp_object_create(const struct lu_env *env, struct dt_object *dt, if (osp_precreate_end_seq(env, d) && osp_is_fid_client(d)) th->th_sync = 1; + local_th = osp_get_storage_thandle(env, th, d); + if (IS_ERR(local_th)) + RETURN(PTR_ERR(local_th)); /* * it's OK if the import is inactive by this moment - id was created * by OST earlier, we just need to maintain it consistently on the disk @@ -1569,17 +1485,13 @@ static int osp_object_create(const struct lu_env *env, struct dt_object *dt, } } - /* new object, the very first ->attr_set() - * initializing attributes needs no logging */ - o->opo_new = 1; - /* 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); rc = dt_record_write(env, d->opd_last_used_oid_file, &osi->osi_lb, - &osi->osi_off, th); + &osi->osi_off, local_th); CDEBUG(D_HA, "%s: Wrote last used FID: "DFID", index %d: %d\n", d->opd_obd->obd_name, PFID(fid), d->opd_index, rc); @@ -1606,13 +1518,12 @@ int osp_declare_object_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); int rc = 0; ENTRY; - /* - * track objects to be destroyed via llog - */ + LASSERT(!osp->opd_connect_mdt); rc = osp_sync_declare_add(env, o, MDS_UNLINK64_REC, th); RETURN(rc); @@ -1635,20 +1546,23 @@ int osp_declare_object_destroy(const struct lu_env *env, * \retval 0 for success * \retval negative error number on failure */ -int osp_object_destroy(const struct lu_env *env, struct dt_object *dt, - struct thandle *th) +static int osp_object_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); int rc = 0; ENTRY; o->opo_non_exist = 1; - /* - * once transaction is committed put proper command on - * the queue going to our OST - */ + + 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); /* not needed in cache any more */ set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags); @@ -1659,8 +1573,7 @@ int osp_object_destroy(const struct lu_env *env, struct dt_object *dt, static int osp_orphan_index_lookup(const struct lu_env *env, struct dt_object *dt, struct dt_rec *rec, - const struct dt_key *key, - struct lustre_capa *capa) + const struct dt_key *key) { return -EOPNOTSUPP; } @@ -1679,7 +1592,6 @@ static int osp_orphan_index_insert(const struct lu_env *env, const struct dt_rec *rec, const struct dt_key *key, struct thandle *handle, - struct lustre_capa *capa, int ignore_quota) { return -EOPNOTSUPP; @@ -1696,8 +1608,7 @@ static int osp_orphan_index_declare_delete(const struct lu_env *env, static int osp_orphan_index_delete(const struct lu_env *env, struct dt_object *dt, const struct dt_key *key, - struct thandle *handle, - struct lustre_capa *capa) + struct thandle *handle) { return -EOPNOTSUPP; } @@ -1708,13 +1619,12 @@ static int osp_orphan_index_delete(const struct lu_env *env, * \param[in] env pointer to the thread context * \param[in] dt pointer to the index object to be iterated * \param[in] attr unused - * \param[in] capa the capability for this operation * * \retval pointer to the iteration structure * \retval negative error number on failure */ struct dt_it *osp_it_init(const struct lu_env *env, struct dt_object *dt, - __u32 attr, struct lustre_capa *capa) + __u32 attr) { struct osp_it *it; @@ -1761,7 +1671,7 @@ void osp_it_fini(const struct lu_env *env, struct dt_it *di) * Get more records for the iteration from peer. * * The new records will be filled in an array of pages. The OSP side - * allows 1MB bulk data to be transfered. + * allows 1MB bulk data to be transferred. * * \param[in] env pointer to the thread context * \param[in] it pointer to the iteration structure @@ -1833,15 +1743,18 @@ static int osp_it_fetch(const struct lu_env *env, struct osp_it *it) ptlrpc_at_set_req_timeout(req); - desc = ptlrpc_prep_bulk_imp(req, npages, 1, BULK_PUT_SINK, - MDS_BULK_PORTAL); + desc = ptlrpc_prep_bulk_imp(req, npages, 1, + 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); } for (i = 0; i < npages; i++) - ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_CACHE_SIZE); + desc->bd_frag_ops->add_kiov_frag(desc, pages[i], 0, + PAGE_CACHE_SIZE); ptlrpc_request_set_replen(req); rc = ptlrpc_queue_wait(req); @@ -1995,7 +1908,7 @@ again0: * \retval 0 for success * \retval negative error number on failure */ -int osp_orphan_it_next(const struct lu_env *env, struct dt_it *di) +static int osp_orphan_it_next(const struct lu_env *env, struct dt_it *di) { struct osp_it *it = (struct osp_it *)di; struct lu_idxpage *idxpage; @@ -2036,8 +1949,8 @@ void osp_it_put(const struct lu_env *env, struct dt_it *di) { } -struct dt_key *osp_orphan_it_key(const struct lu_env *env, - const struct dt_it *di) +static struct dt_key *osp_orphan_it_key(const struct lu_env *env, + const struct dt_it *di) { struct osp_it *it = (struct osp_it *)di; struct lu_orphan_ent *ent = (struct lu_orphan_ent *)it->ooi_ent; @@ -2048,13 +1961,14 @@ struct dt_key *osp_orphan_it_key(const struct lu_env *env, return NULL; } -int osp_orphan_it_key_size(const struct lu_env *env, const struct dt_it *di) +static int osp_orphan_it_key_size(const struct lu_env *env, + const struct dt_it *di) { return sizeof(struct lu_fid); } -int osp_orphan_it_rec(const struct lu_env *env, const struct dt_it *di, - struct dt_rec *rec, __u32 attr) +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; @@ -2158,7 +2072,7 @@ static int osp_index_try(const struct lu_env *env, return 0; } -struct dt_object_operations osp_obj_ops = { +static struct dt_object_operations osp_obj_ops = { .do_declare_attr_get = osp_declare_attr_get, .do_attr_get = osp_attr_get, .do_declare_attr_set = osp_declare_attr_set, @@ -2206,8 +2120,7 @@ static int osp_object_init(const struct lu_env *env, struct lu_object *o, po->opo_obj.do_ops = &osp_md_obj_ops; po->opo_obj.do_body_ops = &osp_md_body_ops; - rc = po->opo_obj.do_ops->do_attr_get(env, lu2dt_obj(o), - la, NULL); + rc = po->opo_obj.do_ops->do_attr_get(env, lu2dt_obj(o), la); if (rc == 0) o->lo_header->loh_attr |= LOHA_EXISTS | (la->la_mode & S_IFMT);