/* **__________________________________________________________________________ */ void af_unix_send_stream_fsm(af_unix_ctx_generic_t *socket_p,com_xmit_template_t *xmit_p) { char *pbuf; int write_len; int ret; int inuse; while(1) { switch (xmit_p->state) { case XMIT_READY: /* ** the transmitter is ready to send however we need to double if there is a ** current buffer to send (because we just exit from congestion or if there ** some buffer in the xmit pending queue */ /* ** Check if there is a current buffer to send */ if (xmit_p->bufRefCurrent != NULL) { write_len = (int)ruc_buf_getPayloadLen(xmit_p->bufRefCurrent); /* ** Get the reference of the destination socket (name) from the ruc_buffer) */ xmit_p->nbWrite = 0; xmit_p->nb2Write = write_len; xmit_p->state = XMIT_IN_PRG; } else { /* ** nothing to send !! */ return; } break; case XMIT_IN_PRG: /* ** Check if there is a current buffer to send */ socket_p->stats.totalXmitAttempts++; pbuf = (char *)ruc_buf_getPayload(xmit_p->bufRefCurrent); ret = af_unix_send_stream_generic(socket_p->socketRef,pbuf+xmit_p->nbWrite,xmit_p->nb2Write - xmit_p->nbWrite, &write_len); switch (ret) { case RUC_OK: /* ** release the buffer that has been sent */ xmit_p->xmit_credit++; inuse = ruc_buf_inuse_decrement(xmit_p->bufRefCurrent); if (inuse < 0) { /* ** inuse MUST never be negative so EXIT !!!!! */ fatal("Inuse is negative %d",inuse); } if (socket_p->userXmitDoneCallBack != NULL) { /* ** caution: in that case it is up to the application that provides the callback to release ** the xmit buffer */ if (ruc_buf_get_opaque_ref(xmit_p->bufRefCurrent) == socket_p) { (socket_p->userXmitDoneCallBack)(socket_p->userRef,socket_p->index,xmit_p->bufRefCurrent); } else { if (inuse == 1) { /* ** need an obj remove since that buffer might still queue somewhere : typically ** in the xmit list of a load balacner entry. */ ruc_objRemove((ruc_obj_desc_t*)xmit_p->bufRefCurrent); ruc_buf_freeBuffer(xmit_p->bufRefCurrent); } } } else { if (inuse == 1) { ruc_objRemove((ruc_obj_desc_t*)xmit_p->bufRefCurrent); ruc_buf_freeBuffer(xmit_p->bufRefCurrent); } } xmit_p->bufRefCurrent = NULL; xmit_p->nbWrite = 0; xmit_p->nb2Write = 0; socket_p->stats.totalXmitSuccess++; socket_p->stats.totalXmitBytes += write_len; xmit_p->state = XMIT_CHECK_XMITQ; break; case RUC_PARTIAL: /* ** need to re-attempt writing */ xmit_p->nbWrite += write_len; socket_p->stats.totalXmitBytes += write_len; break; case RUC_WOULDBLOCK: /* ** the socket is congested-> so exit */ socket_p->stats.totalXmitCongested++; xmit_p->congested_flag = 1; xmit_p->eoc_flag = 0; xmit_p->eoc_threshold = AF_UNIX_CONGESTION_DEFAULT_THRESHOLD; xmit_p->state = XMIT_CONGESTED; return ; case RUC_DISC: /* ** something wrong on sending: if the user has a callback use it: ** the transmitter is no more the owner of the buffer */ inuse = ruc_buf_inuse_decrement(xmit_p->bufRefCurrent); if (inuse < 0) { /* * inuse MUST never be negative so EXIT !!!!! */ fatal("Inuse is negative %d",inuse); } socket_p->stats.totalXmitError++; if (socket_p->userDiscCallBack != NULL) { void *bufref = xmit_p->bufRefCurrent; xmit_p->bufRefCurrent = NULL; if (ruc_buf_get_opaque_ref(bufref) != socket_p) { /* ** the buffer is affected to another socket, however it might possible ** that the real owner of the buffer has finished while the buffer is ** still used by that old connection. So it might be necessary to release ** the buffer. ** However in any case the application must not be inform that there was ** an issue while sendig that buffer since the connection is not considered ** anymore. */ if (inuse == 1) { ruc_objRemove((ruc_obj_desc_t*)bufref); ruc_buf_freeBuffer(bufref); } bufref = NULL; } /* ** it is up to the application to release the buffer if the error is fatal: ** caution the internal disconnection MUST be called before the application since ** the application might attempt to perform a direct re-connection */ xmit_p->state = XMIT_DEAD; af_unix_sock_stream_disconnect_internal(socket_p); (socket_p->userDiscCallBack)(socket_p->userRef,socket_p->index,bufref,errno); return; } else { if (inuse == 1) { ruc_objRemove((ruc_obj_desc_t*)xmit_p->bufRefCurrent); ruc_buf_freeBuffer(xmit_p->bufRefCurrent); } xmit_p->bufRefCurrent = NULL; } /* ** general disconnection->need to clean the socket queue */ xmit_p->state = XMIT_DEAD; af_unix_sock_stream_disconnect_internal(socket_p); return ; break; } break; case XMIT_CHECK_XMITQ: /* ** Check the xmit credit */ if (xmit_p->xmit_credit >= xmit_p->xmit_credit_conf) { xmit_p->xmit_credit = 0; /* ** asser the flag to request a re-activation on the next run of the socket ** controller */ xmit_p->xmit_req_flag = 1; return; } /* ** check if there is a pending buffer (case found if there was a previous congestion */ if (xmit_p->bufRefCurrent != NULL) { /* * lest's go and send it */ xmit_p->state = XMIT_IN_PRG; break; } /* ** read the pending Xmit queue (only priority 0 is considered in the current implementation */ xmit_p->bufRefCurrent = com_xmit_pendingQueue_get(xmit_p,0); if (xmit_p->bufRefCurrent == NULL) { /* ** queue is empty */ xmit_p->xmit_credit = 0; xmit_p->state = XMIT_READY; return; } /* ** OK, go back to send that new bufffer */ ruc_buf_inuse_increment(xmit_p->bufRefCurrent); xmit_p->state = XMIT_READY; break; case XMIT_CONGESTED: /* ** the transmitter is congested: check of the threshold has reached 0 */ xmit_p->eoc_threshold--; if (xmit_p->eoc_threshold == 0) { xmit_p->eoc_flag = 1; xmit_p->congested_flag = 0; xmit_p->state = XMIT_IN_PRG; break; } return; case XMIT_DEAD: /* ** the transmitter is dead */ return; } } }
/** Initial truncate request @param socket_ctx_p: pointer to the af unix socket @param socketId: reference of the socket (not used) @param rozofs_storcli_remote_rsp_cbk: callback for sending out the response @retval : TRUE-> xmit ready event expected @retval : FALSE-> xmit ready event not expected */ void rozofs_storcli_truncate_req_init(uint32_t socket_ctx_idx, void *recv_buf,rozofs_storcli_resp_pf_t rozofs_storcli_remote_rsp_cbk) { rozofs_rpc_call_hdr_with_sz_t *com_hdr_p; rozofs_storcli_ctx_t *working_ctx_p = NULL; int i; uint32_t msg_len; /* length of the rpc messsage including the header length */ storcli_truncate_arg_t *storcli_truncate_rq_p = NULL; rozofs_rpc_call_hdr_t hdr; /* structure that contains the rpc header in host format */ int len; /* effective length of application message */ uint8_t *pmsg; /* pointer to the first available byte in the application message */ uint32_t header_len; XDR xdrs; int errcode = EINVAL; /* ** allocate a context for the duration of the write */ working_ctx_p = rozofs_storcli_alloc_context(); if (working_ctx_p == NULL) { /* ** that situation MUST not occur since there the same number of receive buffer and working context!! */ severe("out of working read/write saved context"); goto failure; } storcli_truncate_rq_p = &working_ctx_p->storcli_truncate_arg; STORCLI_START_NORTH_PROF(working_ctx_p,truncate,0); /* ** Get the full length of the message and adjust it the the length of the applicative part (RPC header+application msg) */ msg_len = ruc_buf_getPayloadLen(recv_buf); msg_len -=sizeof(uint32_t); /* ** save the reference of the received socket since it will be needed for sending back the ** response */ working_ctx_p->socketRef = socket_ctx_idx; working_ctx_p->user_param = NULL; working_ctx_p->recv_buf = recv_buf; working_ctx_p->response_cbk = rozofs_storcli_remote_rsp_cbk; /* ** Get the payload of the receive buffer and set the pointer to the array that describes the write request */ com_hdr_p = (rozofs_rpc_call_hdr_with_sz_t*) ruc_buf_getPayload(recv_buf); memcpy(&hdr,&com_hdr_p->hdr,sizeof(rozofs_rpc_call_hdr_t)); /* ** swap the rpc header */ scv_call_hdr_ntoh(&hdr); pmsg = rozofs_storcli_set_ptr_on_nfs_call_msg((char*)&com_hdr_p->hdr,&header_len); if (pmsg == NULL) { errcode = EFAULT; goto failure; } /* ** map the memory on the first applicative RPC byte available and prepare to decode: ** notice that we will not call XDR_FREE since the application MUST ** provide a pointer for storing the file handle */ len = msg_len - header_len; xdrmem_create(&xdrs,(char*)pmsg,len,XDR_DECODE); /* ** store the source transaction id needed for the reply */ working_ctx_p->src_transaction_id = hdr.hdr.xid; /* ** decode the RPC message of the truncate request */ if (xdr_storcli_truncate_arg_t(&xdrs,storcli_truncate_rq_p) == FALSE) { /* ** decoding error */ errcode = EFAULT; severe("rpc trucnate request decoding error"); goto failure; } /* ** init of the load balancing group/ projection association table: ** That table is ordered: the first corresponds to the storage associated with projection 0, second with 1, etc.. ** When build that table, we MUST consider the value of the base which is associated with the distribution */ uint8_t rozofs_safe = rozofs_get_rozofs_safe(storcli_truncate_rq_p->layout); int lbg_in_distribution = 0; for (i = 0; i <rozofs_safe ; i ++) { /* ** Get the load balancing group associated with the sid */ int lbg_id = rozofs_storcli_get_lbg_for_sid(storcli_truncate_rq_p->cid,storcli_truncate_rq_p->dist_set[i]); if (lbg_id < 0) { /* ** there is no associated between the sid and the lbg. It is typically the case ** when a new cluster has been added to the configuration and the client does not ** know yet the configuration change */ severe("sid is unknown !! %d\n",storcli_truncate_rq_p->dist_set[i]); continue; } rozofs_storcli_lbg_prj_insert_lbg_and_sid(working_ctx_p->lbg_assoc_tb,lbg_in_distribution, lbg_id, storcli_truncate_rq_p->dist_set[i]); rozofs_storcli_lbg_prj_insert_lbg_state(working_ctx_p->lbg_assoc_tb, lbg_in_distribution, NORTH_LBG_GET_STATE(working_ctx_p->lbg_assoc_tb[lbg_in_distribution].lbg_id)); lbg_in_distribution++; if (lbg_in_distribution == rozofs_safe) break; } /* ** allocate a small buffer that will be used for sending the response to the truncate request */ working_ctx_p->xmitBuf = ruc_buf_getBuffer(ROZOFS_STORCLI_NORTH_SMALL_POOL); if (working_ctx_p == NULL) { /* ** that situation MUST not occur since there the same number of receive buffer and working context!! */ errcode = ENOMEM; severe("out of small buffer"); goto failure; } /* ** allocate a sequence number for the working context (same aas for read) */ working_ctx_p->read_seqnum = rozofs_storcli_allocate_read_seqnum(); /* ** set now the working variable specific for handling the truncate ** we re-use the structure used for writing even if nothing is written */ uint8_t forward_projection = rozofs_get_rozofs_forward(storcli_truncate_rq_p->layout); for (i = 0; i < forward_projection; i++) { working_ctx_p->prj_ctx[i].prj_state = ROZOFS_PRJ_READ_IDLE; working_ctx_p->prj_ctx[i].prj_buf = ruc_buf_getBuffer(ROZOFS_STORCLI_SOUTH_LARGE_POOL); if (working_ctx_p->prj_ctx[i].prj_buf == NULL) { /* ** that situation MUST not occur since there the same number of receive buffer and working context!! */ errcode = ENOMEM; severe("out of large buffer"); goto failure; } /* ** increment inuse counter on each buffer since we might need to re-use that packet in case ** of retransmission */ working_ctx_p->prj_ctx[i].inuse_valid = 1; ruc_buf_inuse_increment(working_ctx_p->prj_ctx[i].prj_buf); /* ** set the pointer to the bins */ int position = rozofs_storcli_get_position_of_first_byte2write_in_truncate(); uint8_t *pbuf = (uint8_t*)ruc_buf_getPayload(working_ctx_p->prj_ctx[i].prj_buf); working_ctx_p->prj_ctx[i].bins = (bin_t*)(pbuf+position); } /* ** Prepare for request serialization */ memcpy(working_ctx_p->fid_key, storcli_truncate_rq_p->fid, sizeof (sp_uuid_t)); working_ctx_p->opcode_key = STORCLI_TRUNCATE; { /** * lock all the file for a truncate */ uint64_t nb_blocks = 0; nb_blocks--; int ret; ret = stc_rng_insert((void*)working_ctx_p, STORCLI_READ,working_ctx_p->fid_key, 0,nb_blocks, &working_ctx_p->sched_idx); if (ret == 0) { /* ** there is a current request that is processed with the same fid and there is a collision */ return; } /* ** no request pending with that fid, so we can process it right away */ return rozofs_storcli_truncate_req_processing(working_ctx_p); } /* **_____________________________________________ ** Exception cases **_____________________________________________ */ /* ** there was a failure while attempting to allocate a memory ressource. */ failure: /* ** send back the response with the appropriated error code. ** note: The received buffer (rev_buf) is ** intended to be released by this service in case of error or the TCP transmitter ** once it has been passed to the TCP stack. */ rozofs_storcli_reply_error_with_recv_buf(socket_ctx_idx,recv_buf,NULL,rozofs_storcli_remote_rsp_cbk,errcode); /* ** check if the root context was allocated. Free it if is exist */ if (working_ctx_p != NULL) { /* ** remove the reference to the recvbuf to avoid releasing it twice */ STORCLI_STOP_NORTH_PROF(working_ctx_p,truncate,0); working_ctx_p->recv_buf = NULL; rozofs_storcli_release_context(working_ctx_p); } return; }
int storcli_lbg_cnx_sup_is_selectable(int lbg_id) { uint64_t current_date; storcli_lbg_cnx_supervision_t *p; void *xmit_buf = NULL; int ret; if (lbg_id >=STORCLI_MAX_LBG) return 0; p = &storcli_lbg_cnx_supervision_tab[lbg_id]; if (p->state == STORCLI_LBG_RUNNING) return 1; current_date = timer_get_ticker(); // if (current_date > p->expiration_date) return 1; /* ** check if poll is active */ if (p->poll_state == STORCLI_POLL_IN_PRG) return 0; /* ** check the period */ if (current_date > p->next_poll_date) { /* ** attempt to poll */ p->poll_counter++; xmit_buf = ruc_buf_getBuffer(ROZOFS_STORCLI_SOUTH_LARGE_POOL); if (xmit_buf == NULL) { return 0; } p->poll_state = STORCLI_POLL_IN_PRG; /* ** increment the inuse to avoid a release of the xmit buffer by rozofs_sorcli_send_rq_common() */ ruc_buf_inuse_increment(xmit_buf); ret = rozofs_sorcli_send_rq_common(lbg_id,ROZOFS_TMR_GET(TMR_RPC_NULL_PROC_LBG),STORAGE_PROGRAM,STORAGE_VERSION,SP_NULL, (xdrproc_t) xdr_void, (caddr_t) NULL, xmit_buf, lbg_id, 0, 0, rozofs_storcli_sp_null_processing_cbk, (void*)NULL); ruc_buf_inuse_decrement(xmit_buf); if (ret < 0) { /* ** direct need to free the xmit buffer */ ruc_buf_freeBuffer(xmit_buf); return 0; } /* ** Check if there is direct response from tx module */ if (p->poll_state == STORCLI_POLL_ERR) { /* ** set the next expiration date */ p->next_poll_date = current_date+STORCLI_LBG_SP_NULL_INTERVAL; /* ** release the xmit buffer since there was a direct reply from the lbg while attempting to send the buffer */ ruc_buf_freeBuffer(xmit_buf); return 0; } } return 0; }