block request execution failing on RHEL6.0 (2.6.32-71.el6.x86_64)
Hi All, I have written a func to send a CDB on a scsi device from the kernel. For that I am getting a request and queueing for execution. But blk_execute_rq() func is setting errors field of request to 65536 and sense length is set but all the fields in sense header is 0. I dont know what I am missing, any suggestion would be great help. snippet of the code =========================== rq = blk_get_request(q, rw, __GFP_WAIT); if(!rq){ goto out; } if (buflen && blk_rq_map_kern(q, rq, buf, buflen, __GFP_WAIT)){ goto out; } rq->cmd_len = cmdlen; memcpy(rq->cmd, cmd, cmdlen); memset(sense, 0, sizeof(sense)); rq->sense_len = 0; rq->sense = sense; rq->cmd_type |= REQ_TYPE_BLOCK_PC; rq->cmd_type |= REQ_NOMERGE; rq->timeout = WRITE_SCSI_TIMEOUT; //WRITE_SCSI_TIMEOUT is set to 60 sec blk_execute_rq(q, bd_disk, rq, 0); error = rq->errors; if( error ){ dbg("error in blk_execute_rq error %d",error); } if(rq->sense_len){ process_sense_info(rq->sense); } else { info("no sense available"); } ========================================== thanks, ajit PS:We can not use sg_ioctl interface because code expects user space application only to invoke sg_ioctl that the reason it does a copyin.
Hi All, I have found the issue. in kernel prior to RHEL 6.0 blk_pc_request(rq) is define as: #define blk_pc_request(rq) ((rq)->flags & REQ_BLOCK_PC) while in RHEL 6.0 it is define as: #define blk_pc_request(rq) ((rq)->cmd_type == REQ_TYPE_BLOCK_PC) So, the issue was adding I was REQ_NOMERGE flag in cmd_type field of the request. This was making blk_pc_request(rq) to return false and thats why IOs were not going to scsi layer. thanks, ajit On Wed, Apr 20, 2011 at 6:06 PM, ajit jain <ajit4mail@gmail.com> wrote:
Hi All,
I have written a func to send a CDB on a scsi device from the kernel. For that I am getting a request and queueing for execution. But blk_execute_rq() func is setting errors field of request to 65536 and sense length is set but all the fields in sense header is 0.
I dont know what I am missing, any suggestion would be great help.
snippet of the code =========================== rq = blk_get_request(q, rw, __GFP_WAIT); if(!rq){ goto out; } if (buflen && blk_rq_map_kern(q, rq, buf, buflen, __GFP_WAIT)){ goto out; } rq->cmd_len = cmdlen; memcpy(rq->cmd, cmd, cmdlen); memset(sense, 0, sizeof(sense)); rq->sense_len = 0; rq->sense = sense; rq->cmd_type |= REQ_TYPE_BLOCK_PC; rq->cmd_type |= REQ_NOMERGE; rq->timeout = WRITE_SCSI_TIMEOUT; //WRITE_SCSI_TIMEOUT is set to 60 sec blk_execute_rq(q, bd_disk, rq, 0); error = rq->errors; if( error ){ dbg("error in blk_execute_rq error %d",error); } if(rq->sense_len){ process_sense_info(rq->sense); } else { info("no sense available"); } ==========================================
thanks, ajit
PS:We can not use sg_ioctl interface because code expects user space application only to invoke sg_ioctl that the reason it does a copyin.
On Wed, Apr 27, 2011 at 00:12, ajit jain <ajit4mail@gmail.com> wrote:
Hi All,
I have found the issue.
in kernel prior to RHEL 6.0 blk_pc_request(rq) is define as: #define blk_pc_request(rq) ((rq)->flags & REQ_BLOCK_PC) while in RHEL 6.0 it is define as: #define blk_pc_request(rq) ((rq)->cmd_type == REQ_TYPE_BLOCK_PC)
So, the issue was adding I was REQ_NOMERGE flag in cmd_type field of the request. This was making blk_pc_request(rq) to return false and thats why IOs were not going to scsi layer.
Wow, great research! May I suggest something? I think you need to inform this bug to RHEL kernel team.... quite likely it's a serious regression. -- regards, Mulyadi Santosa Freelance Linux trainer and consultant blog: the-hydra.blogspot.com training: mulyaditraining.blogspot.com
participants (2)
-
ajit jain -
Mulyadi Santosa