X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosp%2Fosp_md_object.c;h=37ce566ec51c251830547d3804f47e79165cc70e;hp=fa86222bb3232dfe79c277f6e8e9ed9a50091e47;hb=27659339a94d5eb1789e08c796c953f6effebc11;hpb=de8572645d287d17c409b99dabdf176822d91486 diff --git a/lustre/osp/osp_md_object.c b/lustre/osp/osp_md_object.c index fa86222..37ce566 100644 --- a/lustre/osp/osp_md_object.c +++ b/lustre/osp/osp_md_object.c @@ -20,7 +20,7 @@ * GPL HEADER END */ /* - * Copyright (c) 2013, 2014, Intel Corporation. + * Copyright (c) 2013, 2017, Intel Corporation. */ /* * lustre/osp/osp_md_object.c @@ -53,11 +53,12 @@ #define DEBUG_SUBSYSTEM S_MDS +#include #include #include "osp_internal.h" -static const char dot[] = "."; -static const char dotdot[] = ".."; +#define OUT_UPDATE_BUFFER_SIZE_ADD 4096 +#define OUT_UPDATE_BUFFER_SIZE_MAX (256 * 4096) /* 1M update size now */ /** * Interpreter call for object creation @@ -76,23 +77,30 @@ static const char dotdot[] = ".."; * * \retval only return 0 for now */ -static int osp_object_create_interpreter(const struct lu_env *env, - struct object_update_reply *reply, - struct ptlrpc_request *req, - struct osp_object *obj, - void *data, int index, int rc) +static int osp_create_interpreter(const struct lu_env *env, + struct object_update_reply *reply, + struct ptlrpc_request *req, + struct osp_object *obj, + void *data, int index, int rc) { - if (rc != 0) { + if (rc != 0 && rc != -EEXIST) { obj->opo_obj.do_lu.lo_header->loh_attr &= ~LOHA_EXISTS; obj->opo_non_exist = 1; } + + /* + * invalidate opo cache for the object after the object is created, so + * attr_get will try to get attr from remote object. + */ + osp_obj_invalidate_cache(obj); + return 0; } /** * Implementation of dt_object_operations::do_declare_create * - * Create the dt_update_request to track the update for this OSP + * Create the osp_update_request to track the update for this OSP * in the transaction. * * \param[in] env execution environment @@ -105,16 +113,30 @@ static int osp_object_create_interpreter(const struct lu_env *env, * \retval 0 if preparation succeeds. * \retval negative errno if preparation fails. */ -int osp_md_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) +int osp_md_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) { return osp_trans_update_request_create(th); } +struct object_update * +update_buffer_get_update(struct object_update_request *request, + unsigned int index) +{ + void *ptr; + int i; + + if (index > request->ourq_count) + return NULL; + + ptr = &request->ourq_updates[0]; + for (i = 0; i < index; i++) + ptr += object_update_size(ptr); + + return ptr; +} + /** * Implementation of dt_object_operations::do_create * @@ -131,30 +153,34 @@ int osp_md_declare_object_create(const struct lu_env *env, * \retval 0 if packing creation succeeds. * \retval negative errno if packing creation fails. */ -int osp_md_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) +int osp_md_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 dt_update_request *update; - int rc; + struct osp_update_request *update; + struct osp_object *obj = dt2osp_obj(dt); + int rc; - update = thandle_to_dt_update_request(th); + update = thandle_to_osp_update_request(th); LASSERT(update != NULL); - rc = out_create_pack(env, &update->dur_buf, - lu_object_fid(&dt->do_lu), attr, hint, dof, - update->dur_batchid); + LASSERT(attr->la_valid & LA_TYPE); + rc = OSP_UPDATE_RPC_PACK(env, out_create_pack, update, + lu_object_fid(&dt->do_lu), attr, hint, dof); if (rc != 0) GOTO(out, rc); - rc = osp_insert_update_callback(env, update, dt2osp_obj(dt), attr, - osp_object_create_interpreter); + rc = osp_insert_update_callback(env, update, dt2osp_obj(dt), NULL, + osp_create_interpreter); if (rc < 0) GOTO(out, rc); dt->do_lu.lo_header->loh_attr |= LOHA_EXISTS | (attr->la_mode & S_IFMT); dt2osp_obj(dt)->opo_non_exist = 0; + obj->opo_stale = 0; + + obj->opo_attr = *attr; out: return rc; } @@ -162,7 +188,7 @@ out: /** * Implementation of dt_object_operations::do_declare_ref_del * - * Create the dt_update_request to track the update for this OSP + * Create the osp_update_request to track the update for this OSP * in the transaction. * * \param[in] env execution environment @@ -194,22 +220,21 @@ static int osp_md_declare_ref_del(const struct lu_env *env, static int osp_md_ref_del(const struct lu_env *env, struct dt_object *dt, struct thandle *th) { - struct dt_update_request *update; - int rc; + struct osp_update_request *update; + int rc; - update = thandle_to_dt_update_request(th); + update = thandle_to_osp_update_request(th); LASSERT(update != NULL); - rc = out_ref_del_pack(env, &update->dur_buf, - lu_object_fid(&dt->do_lu), - update->dur_batchid); + rc = OSP_UPDATE_RPC_PACK(env, out_ref_del_pack, update, + lu_object_fid(&dt->do_lu)); return rc; } /** * Implementation of dt_object_operations::do_declare_ref_del * - * Create the dt_update_request to track the update for this OSP + * Create the osp_update_request to track the update for this OSP * in the transaction. * * \param[in] env execution environment @@ -241,15 +266,14 @@ static int osp_md_declare_ref_add(const struct lu_env *env, static int osp_md_ref_add(const struct lu_env *env, struct dt_object *dt, struct thandle *th) { - struct dt_update_request *update; - int rc; + struct osp_update_request *update; + int rc; - update = thandle_to_dt_update_request(th); + update = thandle_to_osp_update_request(th); LASSERT(update != NULL); - rc = out_ref_add_pack(env, &update->dur_buf, - lu_object_fid(&dt->do_lu), - update->dur_batchid); + rc = OSP_UPDATE_RPC_PACK(env, out_ref_add_pack, update, + lu_object_fid(&dt->do_lu)); return rc; } @@ -282,7 +306,7 @@ static void osp_md_ah_init(const struct lu_env *env, /** * Implementation of dt_object_operations::do_declare_attr_get * - * Create the dt_update_request to track the update for this OSP + * Create the osp_update_request to track the update for this OSP * in the transaction. * * \param[in] env execution environment @@ -318,33 +342,31 @@ int osp_md_declare_attr_set(const struct lu_env *env, struct dt_object *dt, int osp_md_attr_set(const struct lu_env *env, struct dt_object *dt, const struct lu_attr *attr, struct thandle *th) { - struct dt_update_request *update; - int rc; + struct osp_update_request *update; + int rc; - update = thandle_to_dt_update_request(th); + update = thandle_to_osp_update_request(th); LASSERT(update != NULL); - rc = out_attr_set_pack(env, &update->dur_buf, - lu_object_fid(&dt->do_lu), attr, - update->dur_batchid); - + rc = OSP_UPDATE_RPC_PACK(env, out_attr_set_pack, update, + lu_object_fid(&dt->do_lu), attr); return rc; } /** * Implementation of dt_object_operations::do_read_lock * - * osp_md_object_{read,write}_lock() will only lock the remote object in the + * osp_md_{read,write}_lock() will only lock the remote object in the * local cache, which uses the semaphore (opo_sem) inside the osp_object to * lock the object. Note: it will not lock the object in the whole cluster, * which relies on the LDLM lock. * * \param[in] env execution environment * \param[in] dt object to be locked - * \param[in] role lock role from MDD layer, see mdd_object_role(). + * \param[in] role lock role from MDD layer, see dt_object_role(). */ -static void osp_md_object_read_lock(const struct lu_env *env, - struct dt_object *dt, unsigned role) +static void osp_md_read_lock(const struct lu_env *env, struct dt_object *dt, + unsigned role) { struct osp_object *obj = dt2osp_obj(dt); @@ -361,10 +383,10 @@ static void osp_md_object_read_lock(const struct lu_env *env, * * \param[in] env execution environment * \param[in] dt object to be locked - * \param[in] role lock role from MDD layer, see mdd_object_role(). + * \param[in] role lock role from MDD layer, see dt_object_role(). */ -static void osp_md_object_write_lock(const struct lu_env *env, - struct dt_object *dt, unsigned role) +static void osp_md_write_lock(const struct lu_env *env, struct dt_object *dt, + unsigned role) { struct osp_object *obj = dt2osp_obj(dt); @@ -382,8 +404,7 @@ static void osp_md_object_write_lock(const struct lu_env *env, * \param[in] env execution environment * \param[in] dt object to be unlocked */ -static void osp_md_object_read_unlock(const struct lu_env *env, - struct dt_object *dt) +static void osp_md_read_unlock(const struct lu_env *env, struct dt_object *dt) { struct osp_object *obj = dt2osp_obj(dt); @@ -398,8 +419,7 @@ static void osp_md_object_read_unlock(const struct lu_env *env, * \param[in] env execution environment * \param[in] dt object to be unlocked */ -static void osp_md_object_write_unlock(const struct lu_env *env, - struct dt_object *dt) +static void osp_md_write_unlock(const struct lu_env *env, struct dt_object *dt) { struct osp_object *obj = dt2osp_obj(dt); @@ -416,8 +436,7 @@ static void osp_md_object_write_unlock(const struct lu_env *env, * \param[in] env execution environment * \param[in] dt object to be tested */ -static int osp_md_object_write_locked(const struct lu_env *env, - struct dt_object *dt) +static int osp_md_write_locked(const struct lu_env *env, struct dt_object *dt) { struct osp_object *obj = dt2osp_obj(dt); @@ -444,7 +463,7 @@ static int osp_md_index_lookup(const struct lu_env *env, struct dt_object *dt, struct lu_buf *lbuf = &osp_env_info(env)->osi_lb2; struct osp_device *osp = lu2osp_dev(dt->do_lu.lo_dev); struct dt_device *dt_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; struct lu_fid *fid; @@ -455,12 +474,12 @@ static int osp_md_index_lookup(const struct lu_env *env, struct dt_object *dt, * just create an update buffer, instead of attaching the * update_remote list of the thandle. */ - update = dt_update_request_create(dt_dev); + update = osp_update_request_create(dt_dev); if (IS_ERR(update)) RETURN(PTR_ERR(update)); - rc = out_index_lookup_pack(env, &update->dur_buf, - lu_object_fid(&dt->do_lu), rec, key); + rc = OSP_UPDATE_RPC_PACK(env, out_index_lookup_pack, update, + lu_object_fid(&dt->do_lu), rec, key); if (rc != 0) { CERROR("%s: Insert update error: rc = %d\n", dt_dev->dd_lu_dev.ld_obd->obd_name, rc); @@ -511,7 +530,7 @@ out: if (req != NULL) ptlrpc_req_finished(req); - dt_update_request_destroy(update); + osp_update_request_destroy(env, update); return rc; } @@ -519,7 +538,7 @@ out: /** * Implementation of dt_index_operations::dio_declare_insert * - * Create the dt_update_request to track the update for this OSP + * Create the osp_update_request to track the update for this OSP * in the transaction. * * \param[in] env execution environment @@ -551,34 +570,29 @@ static int osp_md_declare_index_insert(const struct lu_env *env, * \param[in] rec record of the index to be inserted * \param[in] key key of the index to be inserted * \param[in] th the transaction handle - * \param[in] ignore_quota quota enforcement for insert * * \retval 0 if packing index insert succeeds. * \retval negative errno if packing fails. */ -static int osp_md_index_insert(const struct lu_env *env, - struct dt_object *dt, +static int osp_md_index_insert(const struct lu_env *env, struct dt_object *dt, const struct dt_rec *rec, - const struct dt_key *key, - struct thandle *th, - int ignore_quota) + const struct dt_key *key, struct thandle *th) { - struct osp_thandle *oth = thandle_to_osp_thandle(th); - struct dt_update_request *update = oth->ot_dur; - int rc; - + struct osp_update_request *update; + int rc; - rc = out_index_insert_pack(env, &update->dur_buf, - lu_object_fid(&dt->do_lu), rec, key, - update->dur_batchid); + update = thandle_to_osp_update_request(th); + LASSERT(update != NULL); + rc = OSP_UPDATE_RPC_PACK(env, out_index_insert_pack, update, + lu_object_fid(&dt->do_lu), rec, key); return rc; } /** * Implementation of dt_index_operations::dio_declare_delete * - * Create the dt_update_request to track the update for this OSP + * Create the osp_update_request to track the update for this OSP * in the transaction. * * \param[in] env execution environment @@ -616,15 +630,15 @@ static int osp_md_index_delete(const struct lu_env *env, const struct dt_key *key, struct thandle *th) { - struct dt_update_request *update; + struct osp_update_request *update; int rc; - update = thandle_to_dt_update_request(th); + update = thandle_to_osp_update_request(th); LASSERT(update != NULL); - rc = out_index_delete_pack(env, &update->dur_buf, - lu_object_fid(&dt->do_lu), key, - update->dur_batchid); + rc = OSP_UPDATE_RPC_PACK(env, out_index_delete_pack, update, + lu_object_fid(&dt->do_lu), key); + return rc; } @@ -638,7 +652,7 @@ static int osp_md_index_delete(const struct lu_env *env, * \param[in] env execution environment * \param[in] di iterator of this iteration * - * \retval 0 if the pointer is advanced successfuly. + * \retval 0 if the pointer is advanced successfully. * \retval 1 if it reaches to the end of the index object. * \retval negative errno if the pointer cannot be advanced. */ @@ -801,6 +815,95 @@ const struct dt_index_operations osp_md_index_ops = { }; /** + * Implement OSP layer dt_object_operations::do_xattr_list() interface. + * + * List extended attribute from the specified MDT/OST object, result is not + * cached because this is called by directory migration only. + * + * \param[in] env pointer to the thread context + * \param[in] dt pointer to the OSP layer dt_object + * \param[out] buf pointer to the lu_buf to hold the extended attribute + * + * \retval positive bytes used/required in the buffer + * \retval negative error number on failure + */ +static int osp_md_xattr_list(const struct lu_env *env, struct dt_object *dt, + const struct lu_buf *buf) +{ + 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 osp_update_request *update = NULL; + struct ptlrpc_request *req = NULL; + struct object_update_reply *reply; + const char *dname = dt->do_lu.lo_dev->ld_obd->obd_name; + int rc = 0; + + ENTRY; + + LASSERT(buf); + + if (unlikely(obj->opo_non_exist)) + RETURN(-ENOENT); + + update = osp_update_request_create(dev); + if (IS_ERR(update)) + RETURN(PTR_ERR(update)); + + rc = OSP_UPDATE_RPC_PACK(env, out_xattr_list_pack, update, + lu_object_fid(&dt->do_lu), buf->lb_len); + if (rc) { + CERROR("%s: Insert update error "DFID": rc = %d\n", + dname, PFID(lu_object_fid(&dt->do_lu)), rc); + GOTO(out, rc); + } + + rc = osp_remote_sync(env, osp, update, &req); + if (rc < 0) { + if (rc == -ENOENT) { + dt->do_lu.lo_header->loh_attr &= ~LOHA_EXISTS; + obj->opo_non_exist = 1; + } + 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) { + DEBUG_REQ(D_ERROR, req, + "%s: Wrong version %x expected %x "DFID": rc = %d", + 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); + if (rc < 0) + GOTO(out, rc); + + if (!buf->lb_buf) + GOTO(out, rc); + + if (unlikely(buf->lb_len < rbuf->lb_len)) + GOTO(out, rc = -ERANGE); + + memcpy(buf->lb_buf, rbuf->lb_buf, rbuf->lb_len); + EXIT; + +out: + if (req) + ptlrpc_req_finished(req); + + if (update && !IS_ERR(update)) + osp_update_request_destroy(env, update); + + return rc; +} + +/** * Implementation of dt_object_operations::do_index_try * * Try to initialize the index API pointer for the given object. This @@ -826,7 +929,9 @@ static int osp_md_index_try(const struct lu_env *env, * Implementation of dt_object_operations::do_object_lock * * Enqueue a lock (by ldlm_cli_enqueue()) of remote object on the remote MDT, - * which will lock the object in the global namespace. + * which will lock the object in the global namespace. And because the + * cross-MDT locks are relatively rare compared with normal local MDT operation, + * let's release it right away, instead of putting it into the LRU list. * * \param[in] env execution environment * \param[in] dt object to be locked @@ -841,37 +946,33 @@ static int osp_md_object_lock(const struct lu_env *env, struct dt_object *dt, struct lustre_handle *lh, struct ldlm_enqueue_info *einfo, - ldlm_policy_data_t *policy) + union ldlm_policy_data *policy) { struct ldlm_res_id *res_id; - struct dt_device *dt_dev = lu2dt_dev(dt->do_lu.lo_dev); - struct osp_device *osp = dt2osp_dev(dt_dev); + struct osp_device *osp = dt2osp_dev(lu2dt_dev(dt->do_lu.lo_dev)); struct ptlrpc_request *req; int rc = 0; - __u64 flags = 0; - ldlm_mode_t mode; + __u64 flags = LDLM_FL_NO_LRU; + ENTRY; res_id = einfo->ei_res_id; LASSERT(res_id != NULL); - mode = ldlm_lock_match(osp->opd_obd->obd_namespace, - LDLM_FL_BLOCK_GRANTED, res_id, - einfo->ei_type, policy, - einfo->ei_mode, lh, 0); - if (mode > 0) - return ELDLM_OK; + if (einfo->ei_mode & (LCK_EX | LCK_PW)) + flags |= LDLM_FL_COS_INCOMPAT; req = ldlm_enqueue_pack(osp->opd_exp, 0); if (IS_ERR(req)) RETURN(PTR_ERR(req)); + osp_set_req_replay(osp, req); rc = ldlm_cli_enqueue(osp->opd_exp, &req, einfo, res_id, - (const ldlm_policy_data_t *)policy, + (const union ldlm_policy_data *)policy, &flags, NULL, 0, LVB_T_NONE, lh, 0); ptlrpc_req_finished(req); - return rc == ELDLM_OK ? 0 : -EIO; + RETURN(rc == ELDLM_OK ? 0 : -EIO); } /** @@ -889,7 +990,7 @@ static int osp_md_object_lock(const struct lu_env *env, static int osp_md_object_unlock(const struct lu_env *env, struct dt_object *dt, struct ldlm_enqueue_info *einfo, - ldlm_policy_data_t *policy) + union ldlm_policy_data *policy) { struct lustre_handle *lockh = einfo->ei_cbdata; @@ -912,8 +1013,8 @@ static int osp_md_object_unlock(const struct lu_env *env, * \retval 0 for success * \retval negative error number on failure */ -int osp_md_declare_object_destroy(const struct lu_env *env, - struct dt_object *dt, struct thandle *th) +int osp_md_declare_destroy(const struct lu_env *env, struct dt_object *dt, + struct thandle *th) { return osp_trans_update_request_create(th); } @@ -933,51 +1034,53 @@ int osp_md_declare_object_destroy(const struct lu_env *env, * \retval 0 for success * \retval negative error number on failure */ -int osp_md_object_destroy(const struct lu_env *env, struct dt_object *dt, - struct thandle *th) +int osp_md_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); - struct dt_update_request *update; - int rc = 0; + struct osp_object *o = dt2osp_obj(dt); + struct osp_device *osp = lu2osp_dev(dt->do_lu.lo_dev); + struct osp_update_request *update; + int rc = 0; ENTRY; o->opo_non_exist = 1; LASSERT(osp->opd_connect_mdt); - update = thandle_to_dt_update_request(th); + update = thandle_to_osp_update_request(th); LASSERT(update != NULL); - rc = out_object_destroy_pack(env, &update->dur_buf, - lu_object_fid(&dt->do_lu), update->dur_batchid); + rc = OSP_UPDATE_RPC_PACK(env, out_destroy_pack, update, + lu_object_fid(&dt->do_lu)); if (rc != 0) RETURN(rc); - /* not needed in cache any more */ set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags); + rc = osp_insert_update_callback(env, update, dt2osp_obj(dt), NULL, + NULL); RETURN(rc); } struct dt_object_operations osp_md_obj_ops = { - .do_read_lock = osp_md_object_read_lock, - .do_write_lock = osp_md_object_write_lock, - .do_read_unlock = osp_md_object_read_unlock, - .do_write_unlock = osp_md_object_write_unlock, - .do_write_locked = osp_md_object_write_locked, - .do_declare_create = osp_md_declare_object_create, - .do_create = osp_md_object_create, + .do_read_lock = osp_md_read_lock, + .do_write_lock = osp_md_write_lock, + .do_read_unlock = osp_md_read_unlock, + .do_write_unlock = osp_md_write_unlock, + .do_write_locked = osp_md_write_locked, + .do_declare_create = osp_md_declare_create, + .do_create = osp_md_create, .do_declare_ref_add = osp_md_declare_ref_add, .do_ref_add = osp_md_ref_add, .do_declare_ref_del = osp_md_declare_ref_del, .do_ref_del = osp_md_ref_del, - .do_declare_destroy = osp_md_declare_object_destroy, - .do_destroy = osp_md_object_destroy, + .do_declare_destroy = osp_md_declare_destroy, + .do_destroy = osp_md_destroy, .do_ah_init = osp_md_ah_init, .do_attr_get = osp_attr_get, .do_declare_attr_set = osp_md_declare_attr_set, .do_attr_set = osp_md_attr_set, .do_xattr_get = osp_xattr_get, + .do_xattr_list = osp_md_xattr_list, .do_declare_xattr_set = osp_declare_xattr_set, .do_xattr_set = osp_xattr_set, .do_declare_xattr_del = osp_declare_xattr_del, @@ -985,12 +1088,13 @@ struct dt_object_operations osp_md_obj_ops = { .do_index_try = osp_md_index_try, .do_object_lock = osp_md_object_lock, .do_object_unlock = osp_md_object_unlock, + .do_invalidate = osp_invalidate, }; /** * Implementation of dt_body_operations::dbo_declare_write * - * Create the dt_update_request to track the update for this OSP + * Create the osp_update_request to track the update for this OSP * in the transaction. * * \param[in] env execution environment @@ -1007,7 +1111,20 @@ static ssize_t osp_md_declare_write(const struct lu_env *env, const struct lu_buf *buf, loff_t pos, struct thandle *th) { - return osp_trans_update_request_create(th); + struct osp_device *osp = dt2osp_dev(th->th_dev); + int rc; + + rc = osp_trans_update_request_create(th); + if (rc != 0) + return rc; + + if (osp->opd_update == NULL) + return 0; + + if (dt2osp_obj(dt)->opo_stale) + return -ESTALE; + + return 0; } /** @@ -1021,33 +1138,162 @@ static ssize_t osp_md_declare_write(const struct lu_env *env, * \param[in] buf buffer to write which includes an embedded size field * \param[in] pos offet in the object to start writing at * \param[in] th transaction handle - * \param[in] ignore_quota quota enforcement for this write * * \retval the buffer size in bytes if packing succeeds. * \retval negative errno if packing fails. */ static ssize_t osp_md_write(const struct lu_env *env, struct dt_object *dt, const struct lu_buf *buf, loff_t *pos, - struct thandle *th, int ignore_quota) + struct thandle *th) { - struct dt_update_request *update; + struct osp_object *obj = dt2osp_obj(dt); + struct osp_update_request *update; + struct osp_thandle *oth = thandle_to_osp_thandle(th); ssize_t rc; + ENTRY; - update = thandle_to_dt_update_request(th); + update = thandle_to_osp_update_request(th); LASSERT(update != NULL); - rc = out_write_pack(env, &update->dur_buf, lu_object_fid(&dt->do_lu), - buf, *pos, update->dur_batchid); + CDEBUG(D_INFO, "write "DFID" offset = %llu length = %zu\n", + PFID(lu_object_fid(&dt->do_lu)), *pos, buf->lb_len); + + rc = OSP_UPDATE_RPC_PACK(env, out_write_pack, update, + lu_object_fid(&dt->do_lu), buf, *pos); if (rc < 0) - return rc; + RETURN(rc); + + rc = osp_check_and_set_rpc_version(oth, obj); + if (rc < 0) + RETURN(rc); /* XXX: how about the write error happened later? */ *pos += buf->lb_len; - return buf->lb_len; + + if (obj->opo_attr.la_valid & LA_SIZE && obj->opo_attr.la_size < *pos) + obj->opo_attr.la_size = *pos; + + spin_lock(&obj->opo_lock); + if (list_empty(&obj->opo_invalidate_cb_list)) { + lu_object_get(&obj->opo_obj.do_lu); + + list_add_tail(&obj->opo_invalidate_cb_list, + &update->our_invalidate_cb_list); + } + spin_unlock(&obj->opo_lock); + + RETURN(buf->lb_len); +} + +static inline void orr_le_to_cpu(struct out_read_reply *orr_dst, + const struct out_read_reply *orr_src) +{ + orr_dst->orr_size = le32_to_cpu(orr_src->orr_size); + orr_dst->orr_padding = le32_to_cpu(orr_src->orr_padding); + orr_dst->orr_offset = le64_to_cpu(orr_dst->orr_offset); +} + + + +static ssize_t osp_md_read(const struct lu_env *env, struct dt_object *dt, + struct lu_buf *rbuf, loff_t *pos) +{ + struct osp_device *osp = lu2osp_dev(dt->do_lu.lo_dev); + struct dt_device *dt_dev = &osp->opd_dt_dev; + struct lu_buf *lbuf = &osp_env_info(env)->osi_lb2; + char *ptr = rbuf->lb_buf; + struct osp_update_request *update; + struct ptlrpc_request *req = NULL; + struct out_read_reply *orr; + struct ptlrpc_bulk_desc *desc; + struct object_update_reply *reply; + int pages; + int rc; + ENTRY; + + /* Because it needs send the update buffer right away, + * just create an update buffer, instead of attaching the + * update_remote list of the thandle. */ + update = osp_update_request_create(dt_dev); + if (IS_ERR(update)) + RETURN(PTR_ERR(update)); + + rc = OSP_UPDATE_RPC_PACK(env, out_read_pack, update, + lu_object_fid(&dt->do_lu), + rbuf->lb_len, *pos); + if (rc != 0) { + CERROR("%s: cannot insert update: rc = %d\n", + dt_dev->dd_lu_dev.ld_obd->obd_name, rc); + GOTO(out_update, rc); + } + + CDEBUG(D_INFO, "%s "DFID" read offset %llu size %zu\n", + dt_dev->dd_lu_dev.ld_obd->obd_name, + PFID(lu_object_fid(&dt->do_lu)), *pos, rbuf->lb_len); + rc = osp_prep_update_req(env, osp->opd_obd->u.cli.cl_import, update, + &req); + if (rc != 0) + GOTO(out_update, rc); + + /* First *and* last might be partial pages, hence +1 */ + pages = DIV_ROUND_UP(rbuf->lb_len, PAGE_SIZE) + 1; + + /* allocate bulk descriptor */ + desc = ptlrpc_prep_bulk_imp(req, pages, 1, + PTLRPC_BULK_PUT_SINK | PTLRPC_BULK_BUF_KIOV, + MDS_BULK_PORTAL, + &ptlrpc_bulk_kiov_nopin_ops); + if (desc == NULL) + GOTO(out, rc = -ENOMEM); + + desc->bd_frag_ops->add_iov_frag(desc, ptr, rbuf->lb_len); + + osp_set_req_replay(osp, req); + req->rq_bulk_read = 1; + /* send request to master and wait for RPC to complete */ + rc = ptlrpc_queue_wait(req); + if (rc != 0) + GOTO(out, rc); + + rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk, + req->rq_bulk->bd_nob_transferred); + if (rc < 0) + 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: invalid update reply magic %x expected %x:" + " rc = %d\n", dt_dev->dd_lu_dev.ld_obd->obd_name, + reply->ourp_magic, UPDATE_REPLY_MAGIC, -EPROTO); + GOTO(out, rc = -EPROTO); + } + + rc = object_update_result_data_get(reply, lbuf, 0); + if (rc < 0) + GOTO(out, rc); + + if (lbuf->lb_len < sizeof(*orr)) + GOTO(out, rc = -EPROTO); + + orr = lbuf->lb_buf; + orr_le_to_cpu(orr, orr); + rc = orr->orr_size; + *pos = orr->orr_offset; +out: + ptlrpc_req_finished(req); + +out_update: + osp_update_request_destroy(env, update); + + RETURN(rc); } /* These body operation will be used to write symlinks during migration etc */ struct dt_body_operations osp_md_body_ops = { .dbo_declare_write = osp_md_declare_write, .dbo_write = osp_md_write, + .dbo_read = osp_md_read, };