/**
* send a write success reply
That API fill up the common header with the SP_READ_RSP opcode
insert the transaction_id associated with the inittial request transaction id
insert a status OK
insert the length of the data payload
In case of a success it is up to the called function to release the xmit buffer
@param p : pointer to the root transaction context used for the read
@retval none
*/
void rozofs_storcli_write_reply_success(rozofs_storcli_ctx_t *p)
{
int ret;
uint8_t *pbuf; /* pointer to the part that follows the header length */
uint32_t *header_len_p; /* pointer to the array that contains the length of the rpc message*/
XDR xdrs;
int len;
storcli_status_ret_t status;
storcli_write_arg_no_data_t *storcli_write_rq_p = NULL;
status.status = STORCLI_SUCCESS;
status.storcli_status_ret_t_u.error = 0; /* NS */
storcli_write_rq_p = (storcli_write_arg_no_data_t*)&p->storcli_write_arg;
STORCLI_STOP_NORTH_PROF(p,write,storcli_write_rq_p->len);
/*
** create xdr structure on top of the buffer that will be used for sending the response
*/
header_len_p = (uint32_t*)ruc_buf_getPayload(p->xmitBuf);
pbuf = (uint8_t*) (header_len_p+1);
len = (int)ruc_buf_getMaxPayloadLen(p->xmitBuf);
len -= sizeof(uint32_t);
xdrmem_create(&xdrs,(char*)pbuf,len,XDR_ENCODE);
if (rozofs_encode_rpc_reply(&xdrs,(xdrproc_t)xdr_sp_status_ret_t,(caddr_t)&status,p->src_transaction_id) != TRUE)
{
severe("rpc reply encoding error");
goto error;
}
/*
** compute the total length of the message for the rpc header and add 4 bytes more bytes for
** the ruc buffer to take care of the header length of the rpc message.
*/
int total_len = xdr_getpos(&xdrs) ;
*header_len_p = htonl(0x80000000 | total_len);
total_len +=sizeof(uint32_t);
ruc_buf_setPayloadLen(p->xmitBuf,total_len);
/*
** Clear the reference of the seqnum to prevent any late response to be processed
** by setting seqnum to 0 any late response is ignored and the associated ressources
** will released (buffer associated with the response). This typically permits to
** avoid sending again the response while this has already been done
*/
p->read_seqnum = 0;
/*
** Get the callback for sending back the response:
** A callback is needed since the request for read might be local or remote
*/
ret = (*p->response_cbk)(p->xmitBuf,p->socketRef,p->user_param);
if (ret == 0)
{
/**
* success so remove the reference of the xmit buffer since it is up to the called
* function to release it
*/
p->xmitBuf = NULL;
}
error:
return;
}
/*
** That function is called when all the projection are ready to be sent
@param working_ctx_p: pointer to the root context associated with the top level write request
*/
void rozofs_storcli_truncate_req_processing(rozofs_storcli_ctx_t *working_ctx_p)
{
storcli_truncate_arg_t *storcli_truncate_rq_p = (storcli_truncate_arg_t*)&working_ctx_p->storcli_truncate_arg;
int ret;
int errcode;
/*
** Check whether the truncate operates on a ROZOFS block size bondary. If it is not
** the case, we must read the block from the disk to then remove the extra data at
** the end of the block.
*/
if (storcli_truncate_rq_p->last_seg != 0) {
working_ctx_p->write_ctx_lock = 1; /* Avoid direct response on internal read error */
ret = rozofs_storcli_internal_read_before_truncate_req(working_ctx_p);
working_ctx_p->write_ctx_lock = 0;
if (ret < 0)
{
errcode = errno;
severe("fatal error on internal read");
goto fail;
}
/* Wait for the internal response */
return;
}
rozofs_storcli_truncate_req_processing_exec(working_ctx_p,NULL);
return;
fail:
/*
** we fall in that case when we run out of resource-> that case is a BUG !!
*/
rozofs_storcli_write_reply_error(working_ctx_p,errcode);
/*
** release the root transaction context
*/
STORCLI_STOP_NORTH_PROF(working_ctx_p,truncate,0);
rozofs_storcli_release_context(working_ctx_p);
errno = errcode;
return;
}
/*
** That function is called when all the projection are ready to be sent
@param working_ctx_p: pointer to the root context associated with the top level write request
*/
void rozofs_storcli_write_repair_req_processing(rozofs_storcli_ctx_t *working_ctx_p)
{
storcli_read_arg_t *storcli_read_rq_p = (storcli_read_arg_t*)&working_ctx_p->storcli_read_arg;
uint8_t layout = storcli_read_rq_p->layout;
uint8_t rozofs_forward;
uint8_t projection_id;
int error=0;
int ret;
rozofs_storcli_projection_ctx_t *prj_cxt_p = working_ctx_p->prj_ctx;
uint8_t bsize = storcli_read_rq_p->bsize;
int prj_size_in_msg = rozofs_get_max_psize_in_msg(layout,bsize);
sp_write_repair_arg_no_bins_t *request;
sp_write_repair_arg_no_bins_t repair_prj_args;
sp_write_repair2_arg_no_bins_t *request2;
sp_write_repair2_arg_no_bins_t repair2_prj_args;
rozofs_forward = rozofs_get_rozofs_forward(layout);
/*
** check if the buffer is still valid: we might face the situation where the rozofsmount
** time-out and re-allocate the write buffer located in shared memory for another
** transaction (either read or write:
** the control must take place only where here is the presence of a shared memory for the write
*/
error = 0;
if (working_ctx_p->shared_mem_p!= NULL)
{
uint32_t *xid_p = (uint32_t*)working_ctx_p->shared_mem_p;
if (*xid_p != working_ctx_p->src_transaction_id)
{
/*
** the source has aborted the request
*/
error = EPROTO;
}
}
/*
** send back the response of the read request towards rozofsmount
*/
rozofs_storcli_read_reply_success(working_ctx_p);
/*
** allocate a sequence number for the working context:
** This is mandatory to avoid any confusion with a late response of the previous read request
*/
working_ctx_p->read_seqnum = rozofs_storcli_allocate_read_seqnum();
/*
** check if it make sense to send the repaired blocks
*/
if (error)
{
/*
** the requester has released the buffer and it could be possible that the
** rozofsmount uses it for another purpose, so the data that have been repaired
** might be wrong, so don't take the right to write wrong data for which we can can
** a good crc !!
*/
goto fail;
}
/*
** We have enough storage, so initiate the transaction towards the storage for each
** projection
*/
for (projection_id = 0; projection_id < rozofs_forward; projection_id++)
{
void *xmit_buf;
int ret;
/*
** skip the projections for which no error has been detected
*/
if (storcli_storage_supports_repair2) {
if (ROZOFS_BITMAP64_TEST_ALL0(working_ctx_p->prj_ctx[projection_id].crc_err_bitmap)) continue;
}
else {
if (working_ctx_p->prj_ctx[projection_id].crc_err_bitmap[0] == 0) continue;
}
xmit_buf = prj_cxt_p[projection_id].prj_buf;
if (xmit_buf == NULL)
{
/*
** fatal error since the ressource control already took place
*/
error = EIO;
goto fail;
}
/*
** fill partially the common header
*/
if (storcli_storage_supports_repair2) {
request2 = &repair2_prj_args;
request2->cid = storcli_read_rq_p->cid;
request2->sid = (uint8_t) rozofs_storcli_lbg_prj_get_sid(working_ctx_p->lbg_assoc_tb,prj_cxt_p[projection_id].stor_idx);
request2->layout = storcli_read_rq_p->layout;
request2->bsize = storcli_read_rq_p->bsize;
/*
** the case of spare 1 must not occur because repair is done for th eoptimal distribution only
*/
if (prj_cxt_p[projection_id].stor_idx >= rozofs_forward) request2->spare = 1;
else request2->spare = 0;
memcpy(request2->dist_set, storcli_read_rq_p->dist_set, ROZOFS_SAFE_MAX_STORCLI*sizeof (uint8_t));
memcpy(request2->fid, storcli_read_rq_p->fid, sizeof (sp_uuid_t));
//CRCrequest->proj_id = projection_id;
request2->proj_id = rozofs_storcli_get_mojette_proj_id(storcli_read_rq_p->dist_set,request2->sid,rozofs_forward);
request2->bid = storcli_read_rq_p->bid;
request2->bitmap[0] = working_ctx_p->prj_ctx[projection_id].crc_err_bitmap[0];
request2->bitmap[1] = working_ctx_p->prj_ctx[projection_id].crc_err_bitmap[1];
request2->bitmap[2] = working_ctx_p->prj_ctx[projection_id].crc_err_bitmap[2];
int nb_blocks = ROZOFS_BITMAP64_NB_BIT1(request2->bitmap);
request2->nb_proj = nb_blocks;
/*
** set the length of the bins part: need to compute the number of blocks
*/
int bins_len = (prj_size_in_msg * nb_blocks);
request2->len = bins_len; /**< bins length MUST be in bytes !!! */
uint32_t lbg_id = rozofs_storcli_lbg_prj_get_lbg(working_ctx_p->lbg_assoc_tb,prj_cxt_p[projection_id].stor_idx);
STORCLI_START_NORTH_PROF((&working_ctx_p->prj_ctx[projection_id]),repair_prj,bins_len);
/*
** caution we might have a direct reply if there is a direct error at load balancing group while
** ateempting to send the RPC message-> typically a disconnection of the TCP connection
** As a consequence the response fct 'rozofs_storcli_write_repair_req_processing_cbk) can be called
** prior returning from rozofs_sorcli_send_rq_common')
** anticipate the status of the xmit state of the projection and lock the section to
** avoid a reply error before returning from rozofs_sorcli_send_rq_common()
** --> need to take care because the write context is released after the reply error sent to rozofsmount
*/
working_ctx_p->write_ctx_lock = 1;
prj_cxt_p[projection_id].prj_state = ROZOFS_PRJ_WR_IN_PRG;
ret = rozofs_sorcli_send_rq_common(lbg_id,ROZOFS_TMR_GET(TMR_STORAGE_PROGRAM),STORAGE_PROGRAM,STORAGE_VERSION,SP_WRITE_REPAIR2,
(xdrproc_t) xdr_sp_write_repair2_arg_no_bins_t, (caddr_t) request2,
xmit_buf,
working_ctx_p->read_seqnum,
(uint32_t) projection_id,
bins_len,
rozofs_storcli_write_repair_req_processing_cbk,
(void*)working_ctx_p);
}
else {
request = &repair_prj_args;
request->cid = storcli_read_rq_p->cid;
request->sid = (uint8_t) rozofs_storcli_lbg_prj_get_sid(working_ctx_p->lbg_assoc_tb,prj_cxt_p[projection_id].stor_idx);
request->layout = storcli_read_rq_p->layout;
request->bsize = storcli_read_rq_p->bsize;
/*
** the case of spare 1 must not occur because repair is done for th eoptimal distribution only
*/
if (prj_cxt_p[projection_id].stor_idx >= rozofs_forward) request->spare = 1;
else request->spare = 0;
memcpy(request->dist_set, storcli_read_rq_p->dist_set, ROZOFS_SAFE_MAX_STORCLI*sizeof (uint8_t));
memcpy(request->fid, storcli_read_rq_p->fid, sizeof (sp_uuid_t));
//CRCrequest->proj_id = projection_id;
request->proj_id = rozofs_storcli_get_mojette_proj_id(storcli_read_rq_p->dist_set,request->sid,rozofs_forward);
request->bid = storcli_read_rq_p->bid;
request->bitmap = working_ctx_p->prj_ctx[projection_id].crc_err_bitmap[0];
int nb_blocks = ROZOFS_BITMAP64_NB_BIT1_FUNC((uint8_t*)&request->bitmap,8);
request->nb_proj = nb_blocks;
/*
** set the length of the bins part: need to compute the number of blocks
*/
int bins_len = (prj_size_in_msg * nb_blocks);
request->len = bins_len; /**< bins length MUST be in bytes !!! */
uint32_t lbg_id = rozofs_storcli_lbg_prj_get_lbg(working_ctx_p->lbg_assoc_tb,prj_cxt_p[projection_id].stor_idx);
STORCLI_START_NORTH_PROF((&working_ctx_p->prj_ctx[projection_id]),repair_prj,bins_len);
/*
** caution we might have a direct reply if there is a direct error at load balancing group while
** ateempting to send the RPC message-> typically a disconnection of the TCP connection
** As a consequence the response fct 'rozofs_storcli_write_repair_req_processing_cbk) can be called
** prior returning from rozofs_sorcli_send_rq_common')
** anticipate the status of the xmit state of the projection and lock the section to
** avoid a reply error before returning from rozofs_sorcli_send_rq_common()
** --> need to take care because the write context is released after the reply error sent to rozofsmount
*/
working_ctx_p->write_ctx_lock = 1;
prj_cxt_p[projection_id].prj_state = ROZOFS_PRJ_WR_IN_PRG;
ret = rozofs_sorcli_send_rq_common(lbg_id,ROZOFS_TMR_GET(TMR_STORAGE_PROGRAM),STORAGE_PROGRAM,STORAGE_VERSION,SP_WRITE_REPAIR,
(xdrproc_t) xdr_sp_write_repair_arg_no_bins_t, (caddr_t) request,
xmit_buf,
working_ctx_p->read_seqnum,
(uint32_t) projection_id,
bins_len,
rozofs_storcli_write_repair_req_processing_cbk,
(void*)working_ctx_p);
}
working_ctx_p->write_ctx_lock = 0;
if (ret < 0)
{
/*
** there is no retry, just keep on with a potential other projection to repair
*/
STORCLI_ERR_PROF(repair_prj_err);
STORCLI_STOP_NORTH_PROF((&working_ctx_p->prj_ctx[projection_id]),repair_prj,0);
prj_cxt_p[projection_id].prj_state = ROZOFS_PRJ_WR_ERROR;
continue;
}
else
{
/*
** check if the state has not been changed: -> it might be possible to get a direct error
*/
if (prj_cxt_p[projection_id].prj_state == ROZOFS_PRJ_WR_ERROR)
{
/*
** it looks like that we cannot repair that preojection, check if there is some other
*/
STORCLI_STOP_NORTH_PROF((&working_ctx_p->prj_ctx[projection_id]),repair_prj,0);
}
}
}
/*
** check if there some write repair request pending, in such a case we wait for the end of the repair
** (answer from the storage node
*/
ret = rozofs_storcli_all_prj_write_repair_check(storcli_read_rq_p->layout,
working_ctx_p->prj_ctx);
if (ret == 0)
{
/*
** there is some pending write
*/
return;
}
fail:
/*
** release the root transaction context
*/
STORCLI_STOP_NORTH_PROF(working_ctx_p,repair,0);
rozofs_storcli_release_context(working_ctx_p);
return;
}
/**
Initial write repair request
Here it is assumed that storclo is working with the context that has been allocated
@param working_ctx_p: pointer to the working context of a read transaction
@retval : TRUE-> xmit ready event expected
@retval : FALSE-> xmit ready event not expected
*/
void rozofs_storcli_repair_req_init(rozofs_storcli_ctx_t *working_ctx_p)
{
int i;
storcli_read_arg_t *storcli_read_rq_p = (storcli_read_arg_t*)&working_ctx_p->storcli_read_arg;
STORCLI_START_NORTH_PROF(working_ctx_p,repair,0);
/*
** set the pointer to to first available data (decoded data)
*/
working_ctx_p->data_write_p = working_ctx_p->data_read_p;
/*
** set now the working variable specific for handling the write
** We need one large buffer per projection that will be written on storage
** we keep the buffer that have been allocated for the read.
*/
uint8_t forward_projection = rozofs_get_rozofs_forward(storcli_read_rq_p->layout);
for (i = 0; i < forward_projection; i++)
{
working_ctx_p->prj_ctx[i].prj_state = ROZOFS_PRJ_WR_IDLE;
if (working_ctx_p->prj_ctx[i].prj_buf == NULL)
{
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!!
*/
severe("out of large buffer");
goto failure;
}
}
/*
** set the pointer to the bins
*/
int position;
// For compatibility between new clients and old storages
if (storcli_storage_supports_repair2) {
position = rozofs_storcli_repair2_get_position_of_first_byte2write();
}
else {
position = rozofs_storcli_repair_get_position_of_first_byte2write();
}
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);
}
/*
** now regenerate the projections that were in error
*/
rozofs_storcli_transform_forward_repair(working_ctx_p,
storcli_read_rq_p->layout,
storcli_read_rq_p->nb_proj,
(char *)working_ctx_p->data_write_p);
/*
** starts the sending of the repaired projections
*/
rozofs_storcli_write_repair_req_processing(working_ctx_p);
return;
failure:
/*
** send back the response of the read request towards rozofsmount
*/
rozofs_storcli_read_reply_success(working_ctx_p);
/*
** release the root transaction context
*/
STORCLI_STOP_NORTH_PROF(working_ctx_p,repair,0);
rozofs_storcli_release_context(working_ctx_p);
}
void rozofs_storcli_truncate_projection_retry(rozofs_storcli_ctx_t *working_ctx_p,uint8_t projection_id,int same_storage_retry_acceptable)
{
uint8_t rozofs_safe;
uint8_t rozofs_forward;
uint8_t layout;
storcli_truncate_arg_t *storcli_truncate_rq_p = (storcli_truncate_arg_t*)&working_ctx_p->storcli_truncate_arg;
int error=0;
int storage_idx;
rozofs_storcli_projection_ctx_t *prj_cxt_p = working_ctx_p->prj_ctx;
rozofs_storcli_lbg_prj_assoc_t *lbg_assoc_p = working_ctx_p->lbg_assoc_tb;
layout = storcli_truncate_rq_p->layout;
rozofs_safe = rozofs_get_rozofs_safe(layout);
rozofs_forward = rozofs_get_rozofs_forward(layout);
/*
** Now update the state of each load balancing group since it might be possible
** that some experience a state change
*/
for (storage_idx = 0; storage_idx < rozofs_safe; storage_idx++)
{
/*
** Check the state of the load Balancing group
*/
rozofs_storcli_lbg_prj_insert_lbg_state(lbg_assoc_p,
storage_idx,
NORTH_LBG_GET_STATE(lbg_assoc_p[storage_idx].lbg_id));
}
/**
* attempt to select a new storage
*/
if (rozofs_storcli_select_storage_idx_for_write (working_ctx_p,rozofs_forward,rozofs_safe,projection_id) < 0)
{
/*
** Cannot select a new storage: OK so now double check if the retry on the same storage is
** acceptable.When it is the case, check if the max retry has not been yet reached
** Otherwise, we are in deep shit-> reject the read request
*/
if (same_storage_retry_acceptable == 0)
{
error = EIO;
prj_cxt_p[projection_id].errcode = error;
goto reject;
}
if (++prj_cxt_p[projection_id].retry_cpt >= ROZOFS_STORCLI_MAX_RETRY)
{
error = EIO;
prj_cxt_p[projection_id].errcode = error;
goto reject;
}
}
/*
** we are lucky since either a get a new storage or the retry counter is not exhausted
*/
sp_truncate_arg_no_bins_t *request;
sp_truncate_arg_no_bins_t truncate_prj_args;
void *xmit_buf;
int ret;
xmit_buf = prj_cxt_p[projection_id].prj_buf;
if (xmit_buf == NULL)
{
/*
** fatal error since the ressource control already took place
*/
error = EFAULT;
prj_cxt_p[projection_id].errcode = error;
goto fatal;
}
/*
** fill partially the common header
*/
retry:
request = &truncate_prj_args;
request->cid = storcli_truncate_rq_p->cid;
request->sid = (uint8_t) rozofs_storcli_lbg_prj_get_sid(working_ctx_p->lbg_assoc_tb,prj_cxt_p[projection_id].stor_idx);
request->layout = layout;
if (prj_cxt_p[projection_id].stor_idx >= rozofs_forward) request->spare = 1;
else request->spare = 0;
memcpy(request->dist_set, storcli_truncate_rq_p->dist_set, ROZOFS_SAFE_MAX*sizeof (uint8_t));
memcpy(request->fid, storcli_truncate_rq_p->fid, sizeof (sp_uuid_t));
request->proj_id = projection_id;
request->bid = storcli_truncate_rq_p->bid;
request->last_seg = storcli_truncate_rq_p->last_seg;
request->last_timestamp = working_ctx_p->timestamp;
/*
** Bins len has been saved in the working context
*/
request->len = working_ctx_p->truncate_bins_len;
uint32_t lbg_id = rozofs_storcli_lbg_prj_get_lbg(working_ctx_p->lbg_assoc_tb,prj_cxt_p[projection_id].stor_idx);
/*
** increment the lock since it might be possible that this procedure is called after a synchronous transaction failu failure
** while the system is still in the initial procedure that triggers the writing of the projection. So it might be possible that
** the lock is already asserted
** as for the initial case, we need to anticipate the xmit state of the projection since the ERROR status might be set
** on a synchronous transaction failure. If that state is set after a positive submission towards the lbg, we might
** overwrite the ERROR state with the IN_PRG state.
*/
working_ctx_p->write_ctx_lock++;
prj_cxt_p[projection_id].prj_state = ROZOFS_PRJ_WR_IN_PRG;
STORCLI_START_NORTH_PROF((&working_ctx_p->prj_ctx[projection_id]),truncate_prj,0);
ret = rozofs_sorcli_send_rq_common(lbg_id,ROZOFS_TMR_GET(TMR_STORAGE_PROGRAM),STORAGE_PROGRAM,STORAGE_VERSION,SP_TRUNCATE,
(xdrproc_t) xdr_sp_truncate_arg_no_bins_t, (caddr_t) request,
xmit_buf,
working_ctx_p->read_seqnum,
(uint32_t) projection_id,
working_ctx_p->truncate_bins_len,
rozofs_storcli_truncate_req_processing_cbk,
(void*)working_ctx_p);
working_ctx_p->write_ctx_lock--;
if (ret < 0)
{
/*
** the communication with the storage seems to be wrong (more than TCP connection temporary down
** attempt to select a new storage
**
*/
STORCLI_STOP_NORTH_PROF((&working_ctx_p->prj_ctx[projection_id]),truncate_prj,0);
if (rozofs_storcli_select_storage_idx_for_write (working_ctx_p,rozofs_forward,rozofs_safe,projection_id) < 0)
{
/*
** Out of storage !!-> too many storages are down
*/
goto fatal;
}
/*
** retry for that projection with a new storage index: WARNING: we assume that xmit buffer has not been released !!!
*/
goto retry;
}
/*
** OK, the buffer has been accepted by the load balancing group, check if there was a direct failure for
** that transaction
*/
if ( prj_cxt_p[projection_id].prj_state == ROZOFS_PRJ_WR_ERROR)
{
error = prj_cxt_p[projection_id].errcode;
goto fatal;
}
return;
/*
**_____________________________________________
** Exception cases
**_____________________________________________
*/
reject:
if (working_ctx_p->write_ctx_lock != 0) return;
/*
** we fall in that case when we run out of storage
*/
rozofs_storcli_write_reply_error(working_ctx_p,error);
/*
** release the root transaction context
*/
STORCLI_STOP_NORTH_PROF(working_ctx_p,truncate,0);
rozofs_storcli_release_context(working_ctx_p);
return;
fatal:
/*
** caution -> reply error is only generated if the ctx_lock is 0
*/
if (working_ctx_p->write_ctx_lock != 0) return;
/*
** we fall in that case when we run out of resource-> that case is a BUG !!
*/
rozofs_storcli_write_reply_error(working_ctx_p,error);
/*
** release the root transaction context
*/
STORCLI_STOP_NORTH_PROF(working_ctx_p,truncate,0);
rozofs_storcli_release_context(working_ctx_p);
return;
}
/*
** That function is called when all the projection are ready to be sent
@param working_ctx_p: pointer to the root context associated with the top level write request
@param data : pointer to the data of the last block to truncate
*/
void rozofs_storcli_truncate_req_processing_exec(rozofs_storcli_ctx_t *working_ctx_p, char * data)
{
storcli_truncate_arg_t *storcli_truncate_rq_p = (storcli_truncate_arg_t*)&working_ctx_p->storcli_truncate_arg;
uint8_t layout = storcli_truncate_rq_p->layout;
uint8_t rozofs_forward;
uint8_t rozofs_safe;
uint8_t projection_id;
int storage_idx;
int error=0;
rozofs_storcli_lbg_prj_assoc_t *lbg_assoc_p = working_ctx_p->lbg_assoc_tb;
rozofs_storcli_projection_ctx_t *prj_cxt_p = working_ctx_p->prj_ctx;
rozofs_forward = rozofs_get_rozofs_forward(layout);
rozofs_safe = rozofs_get_rozofs_safe(layout);
/*
** set the current state of each load balancing group belonging to the rozofs_safe group
*/
for (storage_idx = 0; storage_idx < rozofs_safe; storage_idx++)
{
/*
** Check the state of the load Balancing group
*/
rozofs_storcli_lbg_prj_insert_lbg_state(lbg_assoc_p,
storage_idx,
NORTH_LBG_GET_STATE(lbg_assoc_p[storage_idx].lbg_id));
}
/*
** Now find out a selectable lbg_id for each projection
*/
for (projection_id = 0; projection_id < rozofs_forward; projection_id++)
{
if (rozofs_storcli_select_storage_idx_for_write ( working_ctx_p,rozofs_forward, rozofs_safe,projection_id) < 0)
{
/*
** there is no enough valid storage !!
*/
error = EIO;
goto fail;
}
}
/*
** Let's transform the data to write
*/
working_ctx_p->truncate_bins_len = 0;
if (data != NULL) {
STORCLI_START_KPI(storcli_kpi_transform_forward);
rozofs_storcli_transform_forward(working_ctx_p->prj_ctx,
layout,
0,
1,
working_ctx_p->timestamp,
storcli_truncate_rq_p->last_seg,
data);
STORCLI_STOP_KPI(storcli_kpi_transform_forward,0);
working_ctx_p->truncate_bins_len = rozofs_get_max_psize(layout)*sizeof(bin_t) + sizeof(rozofs_stor_bins_hdr_t);
}
/*
** We have enough storage, so initiate the transaction towards the storage for each
** projection
*/
for (projection_id = 0; projection_id < rozofs_forward; projection_id++)
{
sp_truncate_arg_no_bins_t *request;
sp_truncate_arg_no_bins_t truncate_prj_args;
void *xmit_buf;
int ret;
xmit_buf = prj_cxt_p[projection_id].prj_buf;
if (xmit_buf == NULL)
{
/*
** fatal error since the ressource control already took place
*/
error = EIO;
goto fatal;
}
/*
** fill partially the common header
*/
retry:
request = &truncate_prj_args;
request->cid = storcli_truncate_rq_p->cid;
request->sid = (uint8_t) rozofs_storcli_lbg_prj_get_sid(working_ctx_p->lbg_assoc_tb,prj_cxt_p[projection_id].stor_idx);
request->layout = layout;
if (prj_cxt_p[projection_id].stor_idx >= rozofs_forward) request->spare = 1;
else request->spare = 0;
memcpy(request->dist_set, storcli_truncate_rq_p->dist_set, ROZOFS_SAFE_MAX*sizeof (uint8_t));
memcpy(request->fid, storcli_truncate_rq_p->fid, sizeof (sp_uuid_t));
request->proj_id = projection_id;
request->bid = storcli_truncate_rq_p->bid;
request->last_seg = storcli_truncate_rq_p->last_seg;
request->last_timestamp = working_ctx_p->timestamp;
request->len = working_ctx_p->truncate_bins_len;
uint32_t lbg_id = rozofs_storcli_lbg_prj_get_lbg(working_ctx_p->lbg_assoc_tb,prj_cxt_p[projection_id].stor_idx);
STORCLI_START_NORTH_PROF((&working_ctx_p->prj_ctx[projection_id]),truncate_prj,0);
/*
** caution we might have a direct reply if there is a direct error at load balancing group while
** ateempting to send the RPC message-> typically a disconnection of the TCP connection
** As a consequence the response fct 'rozofs_storcli_truncate_req_processing_cbk) can be called
** prior returning from rozofs_sorcli_send_rq_common')
** anticipate the status of the xmit state of the projection and lock the section to
** avoid a reply error before returning from rozofs_sorcli_send_rq_common()
** --> need to take care because the write context is released after the reply error sent to rozofsmount
*/
working_ctx_p->write_ctx_lock = 1;
prj_cxt_p[projection_id].prj_state = ROZOFS_PRJ_WR_IN_PRG;
ret = rozofs_sorcli_send_rq_common(lbg_id,ROZOFS_TMR_GET(TMR_STORAGE_PROGRAM),STORAGE_PROGRAM,STORAGE_VERSION,SP_TRUNCATE,
(xdrproc_t) xdr_sp_truncate_arg_no_bins_t, (caddr_t) request,
xmit_buf,
working_ctx_p->read_seqnum,
(uint32_t) projection_id,
working_ctx_p->truncate_bins_len,
rozofs_storcli_truncate_req_processing_cbk,
(void*)working_ctx_p);
working_ctx_p->write_ctx_lock = 0;
if (ret < 0)
{
/*
** the communication with the storage seems to be wrong (more than TCP connection temporary down
** attempt to select a new storage
**
*/
if (rozofs_storcli_select_storage_idx_for_write (working_ctx_p,rozofs_forward,rozofs_safe,projection_id) < 0)
{
/*
** Out of storage !!-> too many storages are down
*/
goto fatal;
}
/*
** retry for that projection with a new storage index: WARNING: we assume that xmit buffer has not been released !!!
*/
//#warning: it is assumed that xmit buffer has not been release, need to double check!!
goto retry;
}
else
{
/*
** check if the state has not been changed: -> it might be possible to get a direct error
*/
if (prj_cxt_p[projection_id].prj_state == ROZOFS_PRJ_WR_ERROR)
{
error = prj_cxt_p[projection_id].errcode;
goto fatal;
}
}
}
return;
fail:
/*
** we fall in that case when we run out of resource-> that case is a BUG !!
*/
rozofs_storcli_write_reply_error(working_ctx_p,error);
/*
** release the root transaction context
*/
STORCLI_STOP_NORTH_PROF(working_ctx_p,truncate,0);
rozofs_storcli_release_context(working_ctx_p);
return;
fatal:
/*
** we fall in that case when we run out of resource-> that case is a BUG !!
*/
rozofs_storcli_write_reply_error(working_ctx_p,error);
/*
** release the root transaction context
*/
STORCLI_STOP_NORTH_PROF(working_ctx_p,truncate,0);
rozofs_storcli_release_context(working_ctx_p);
return;
}
/**
* callback for the internal read request triggered by a truncate
potential failure case:
- socket_ref is out of range
- connection is down
@param buffer : pointer to the ruc_buffer that cointains the response
@param socket_ref : non significant
@param user_param_p : pointer to the root context
@retval 0 : successfully submitted to the transport layer
@retval < 0 error, the caller is intended to release the buffer
*/
int rozofs_storcli_internal_read_before_truncate_rsp_cbk(void *buffer,uint32_t socket_ref,void *user_param)
{
int errcode = 0;
rozofs_storcli_ctx_t *working_ctx_p = (rozofs_storcli_ctx_t*)user_param;
storcli_truncate_arg_t * storcli_truncate_rq_p = (storcli_truncate_arg_t*)&working_ctx_p->storcli_truncate_arg;
XDR xdrs;
uint8_t *payload;
char *data;
int position;
int bufsize;
struct rpc_msg rpc_reply;
storcli_status_ret_t rozofs_status;
int data_len;
int error;
rpc_reply.acpted_rply.ar_results.proc = NULL;
/*
** decode the read internal read reply
*/
payload = (uint8_t*) ruc_buf_getPayload(buffer);
payload += sizeof(uint32_t); /* skip length*/
/*
** OK now decode the received message
*/
bufsize = ruc_buf_getPayloadLen(buffer);
bufsize -= sizeof(uint32_t); /* skip length*/
xdrmem_create(&xdrs,(char*)payload,bufsize,XDR_DECODE);
error = 0;
while (1)
{
/*
** decode the rpc part
*/
if (rozofs_xdr_replymsg(&xdrs,&rpc_reply) != TRUE)
{
errno = EPROTO;
error = 1;
break;
}
/*
** decode the status of the operation
*/
if (xdr_storcli_status_ret_t(&xdrs,&rozofs_status)!= TRUE)
{
errno = EPROTO;
error = 1;
break;
}
/*
** check th estatus of the operation
*/
if ( rozofs_status.status != STORCLI_SUCCESS )
{
error = 0;
break;
}
{
int alignment;
/*
** skip the alignment
*/
if (xdr_int(&xdrs, &alignment) != TRUE)
{
errno = EPROTO;
STORCLI_ERR_PROF(read_prj_err);
error = 1;
break;
}
}
/*
** Now get the length of the part that has been read
*/
if (xdr_int(&xdrs, &data_len) != TRUE)
{
errno = EPROTO;
error = 1;
break;
}
break;
}
if (error)
{
severe("error while decoding rpc reply");
goto failure;
}
position = XDR_GETPOS(&xdrs);
data = (char*)(payload+position);
/*
** check the status of the read operation
*/
if (rozofs_status.status != STORCLI_SUCCESS)
{
data = NULL;
}
else {
/*,
** No data returned
*/
if (data_len == 0) {
data = NULL;
}
else if (storcli_truncate_rq_p->last_seg <= data_len) {
memset(data+storcli_truncate_rq_p->last_seg, 0, ROZOFS_BSIZE-storcli_truncate_rq_p->last_seg);
}
else {
memset(data+data_len, 0, ROZOFS_BSIZE-data_len);
}
}
rozofs_storcli_truncate_req_processing_exec(working_ctx_p, data);
ruc_buf_freeBuffer(buffer);
return 0 ;
failure:
ruc_buf_freeBuffer(buffer);
/*
** check if the lock is asserted to prevent direct call to callback
*/
if (working_ctx_p->write_ctx_lock == 1) return 0;
/*
** write failure
*/
rozofs_storcli_write_reply_error(working_ctx_p,errcode);
/*
** release the transaction root context
*/
working_ctx_p->xmitBuf = NULL;
STORCLI_STOP_NORTH_PROF(working_ctx_p,truncate,0);
rozofs_storcli_release_context(working_ctx_p);
return 0 ;
}
/**
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;
}
void rozofs_storcli_read_reply_success(rozofs_storcli_ctx_t *p)
{
int ret;
uint8_t *pbuf; /* pointer to the part that follows the header length */
uint32_t *header_len_p; /* pointer to the array that contains the length of the rpc message*/
XDR xdrs;
int len;
storcli_status_t status = STORCLI_SUCCESS;
int data_len;
uint8_t eof_flag = 0;
/*
** create xdr structure on top of the buffer that will be used for sending the response
*/
header_len_p = (uint32_t*)ruc_buf_getPayload(p->xmitBuf);
pbuf = (uint8_t*) (header_len_p+1);
len = (int)ruc_buf_getMaxPayloadLen(p->xmitBuf);
len -= sizeof(uint32_t);
xdrmem_create(&xdrs,(char*)pbuf,len,XDR_ENCODE);
if (rozofs_encode_rpc_reply(&xdrs,(xdrproc_t)xdr_storcli_status_t,(caddr_t)&status,p->src_transaction_id) != TRUE)
{
severe("rpc reply encoding error");
goto error;
}
/*
** Encode the length of returned data
*/
data_len = rozofs_storcli_transform_get_read_len_in_bytes(p->block_ctx_table,
p->effective_number_of_blocks,
&eof_flag);
STORCLI_STOP_NORTH_PROF(p,read,data_len);
/*
** skip the alignment
*/
int position;
position = xdr_getpos(&xdrs);
position += sizeof(uint32_t);
xdr_setpos(&xdrs,position);
XDR_PUTINT32(&xdrs, (int32_t *)&data_len);
/*
** round up data_len to 4 bytes alignment
*/
if ((data_len%4)!= 0) data_len = (data_len &(~0x3))+4;
/*
** compute the total length of the message for the rpc header and add 4 bytes more bytes for
** the ruc buffer to take care of the header length of the rpc message.
*/
int total_len = xdr_getpos(&xdrs)+data_len ;
*header_len_p = htonl(0x80000000 | total_len);
total_len +=sizeof(uint32_t);
ruc_buf_setPayloadLen(p->xmitBuf,total_len);
/*
** Clear the reference of the seqnum to prevent any late response to be processed
** by setting seqnum to 0 any late response is ignored and the associated ressources
** will released (buffer associated with the response). This typically permits to
** avoid sending again the response while this has already been done
*/
p->read_seqnum = 0;
/*
** Get the callback for sending back the response:
** A callback is needed since the request for read might be local or remote
*/
ret = (*p->response_cbk)(p->xmitBuf,p->socketRef,p->user_param);
if (ret == 0)
{
/**
* success so remove the reference of the xmit buffer since it is up to the called
* function to release it
*/
p->xmitBuf = NULL;
}
error:
// #warning need to consider the case of a local read triggers by a write request. Without a guard time the write working can be lost!!
return;
}