/*
** 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;
}
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;
}
/**
Apply the transform to a buffer starting at "data". That buffer MUST be ROZOFS_BSIZE
aligned.
The first_block_idx is the index of a ROZOFS_BSIZE array in the output buffer
The number_of_blocks is the number of ROZOFS_BSIZE that must be transform
Notice that the first_block_idx offset applies to the output transform buffer only
not to the input buffer pointed by "data".
*
* @param *working_ctx_p: storcli working context
* @param number_of_blocks: number of blocks to write
* @param *data: pointer to the source data that must be transformed
*
* @return: the length written on success, -1 otherwise (errno is set)
*/
void rozofs_storcli_transform_forward_repair(rozofs_storcli_ctx_t *working_ctx_p,
uint8_t layout,
uint32_t number_of_blocks,
char *data)
{
projection_t rozofs_fwd_projections[ROZOFS_SAFE_MAX_STORCLI];
projection_t *projections; // Table of projections used to transform data
uint16_t projection_id = 0;
uint32_t i = 0;
uint8_t rozofs_forward = rozofs_get_rozofs_forward(layout);
uint8_t rozofs_safe = rozofs_get_rozofs_forward(layout);
uint8_t rozofs_inverse = rozofs_get_rozofs_inverse(layout);
rozofs_storcli_projection_ctx_t *prj_ctx_p = &working_ctx_p->prj_ctx[0];
int empty_block = 0;
uint8_t sid;
int moj_prj_id;
int block_idx;
int k;
storcli_read_arg_t *storcli_read_rq_p = (storcli_read_arg_t*)&working_ctx_p->storcli_read_arg;
uint8_t bsize = storcli_read_rq_p->bsize;
uint32_t bbytes = ROZOFS_BSIZE_BYTES(bsize);
int prj_size_in_msg = rozofs_get_max_psize_in_msg(layout,bsize);
projections = rozofs_fwd_projections;
// For each projection
for (projection_id = 0; projection_id < rozofs_forward; projection_id++) {
projections[projection_id].angle.p = rozofs_get_angles_p(layout,projection_id);
projections[projection_id].angle.q = rozofs_get_angles_q(layout,projection_id);
projections[projection_id].size = rozofs_get_128bits_psizes(layout,bsize,projection_id);
}
/*
** now go through all projection set to find out if there is something to regenerate
*/
for (k = 0; k < rozofs_safe; k++)
{
block_idx = 0;
if (ROZOFS_BITMAP64_TEST_ALL0(prj_ctx_p[k].crc_err_bitmap)) continue;
/*
** Get the sid associated with the projection context
*/
sid = (uint8_t) rozofs_storcli_lbg_prj_get_sid(working_ctx_p->lbg_assoc_tb,
prj_ctx_p[k].stor_idx);
/*
** Get the reference of the Mojette projection_id
*/
moj_prj_id = rozofs_storcli_get_mojette_proj_id(storcli_read_rq_p->dist_set,sid,rozofs_forward);
if (moj_prj_id < 0)
{
/*
** it is the reference of a spare sid, so go to the next projection context
*/
continue;
}
for (i = 0; i < number_of_blocks; i++)
{
if (ROZOFS_BITMAP64_TEST0(i,prj_ctx_p[k].crc_err_bitmap))
{
/*
** nothing to generate for that block
*/
continue;
}
/*
** check for empty block
*/
empty_block = rozofs_data_block_check_empty(data + (i * bbytes), bbytes);
/**
* regenerate the projection for the block for which a crc error has been detected
*/
//CRC projections[moj_prj_id].bins = prj_ctx_p[moj_prj_id].bins +
projections[moj_prj_id].bins = prj_ctx_p[k].bins +
(prj_size_in_msg/sizeof(bin_t)* (0+block_idx));
rozofs_stor_bins_hdr_t *rozofs_bins_hdr_p = (rozofs_stor_bins_hdr_t*)projections[moj_prj_id].bins;
/*
** check if the user data block is empty: if the data block is empty no need to transform
*/
if (empty_block)
{
rozofs_bins_hdr_p->s.projection_id = 0;
rozofs_bins_hdr_p->s.timestamp = 0;
rozofs_bins_hdr_p->s.effective_length = 0;
rozofs_bins_hdr_p->s.filler = 0;
rozofs_bins_hdr_p->s.version = 0;
block_idx++;
continue;
}
/*
** fill the header of the projection
*/
rozofs_bins_hdr_p->s.projection_id = moj_prj_id;
//CRC rozofs_bins_hdr_p->s.timestamp = working_ctx_p->block_ctx_table[block_idx].timestamp;
rozofs_bins_hdr_p->s.timestamp = working_ctx_p->block_ctx_table[i].timestamp;
//CRC rozofs_bins_hdr_p->s.effective_length = working_ctx_p->block_ctx_table[block_idx].effective_length;
rozofs_bins_hdr_p->s.effective_length = working_ctx_p->block_ctx_table[i].effective_length;
rozofs_bins_hdr_p->s.filler = 0;
rozofs_bins_hdr_p->s.version = 0;
/*
** update the pointer to point out the first bins
*/
projections[moj_prj_id].bins += sizeof(rozofs_stor_bins_hdr_t)/sizeof(bin_t);
/*
** do not apply transform for empty block
*/
if (empty_block == 0)
{
/*
** Apply the erasure code transform for the block i
*/
transform128_forward_one_proj((pxl_t *) (data + (i * bbytes)),
rozofs_inverse,
bbytes / rozofs_inverse / sizeof (pxl_t),
moj_prj_id, projections);
/*
** add the footer at the end of the repaired projection
*/
rozofs_stor_bins_footer_t *rozofs_bins_foot_p;
rozofs_bins_foot_p = (rozofs_stor_bins_footer_t*) (projections[moj_prj_id].bins
+ rozofs_get_psizes(layout,bsize,moj_prj_id));
//CRC rozofs_bins_foot_p->timestamp = working_ctx_p->block_ctx_table[block_idx].timestamp;
rozofs_bins_foot_p->timestamp = rozofs_bins_hdr_p->s.timestamp;
}
block_idx++;
}
}
}
/*
** 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;
}
