It seems to me the code in iscsit_map_iovec doesn't quite work in certain cases if the sg from the transport layer has offsets in their pages. In the simplest case where there is a problem, assume the first scatterlist entry has an offset. Now also assume that this is a write and we will receive immediate data and then data out PDU's because the command is large. Lets say immediate data is 64K and the total command size is 256 K. Now the first receive will work fine since it contains data_offset = 0 and we use sg->offset in the loop for every sg entry. Now let us say the first page has offset 2K. Then on the next data out PDU, we will get data_offset = 0x10000 (data_offset seems to come from hdr->offset). Now we don't take into account the offsets of the pages we have already received (in this case) in the immediate data as part of the first PDU. This causes us to have an offset into sg entry 16 with page_offset 0, which is incorrect since we should be in sg entry 16 and page_offset 2K. The problem is clearly much worse if each entry in the sg has its own offset and not just the first one. I know this is not a manifested bug in existing code because the way transport code allocates the sg is transport_generic_get_mem does not cause a problem since it is always offset 0. But since the iscsi code uses sg->offset, we probably should do it correctly if the sg was ever setup with page offsets in it. -- To unsubscribe from this list: send the line "unsubscribe target-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
