On 23 August 2016 at 09:18, Shaun Tancheff <shaun.tancheff@xxxxxxxxxxx> wrote: > On Tue, Aug 23, 2016 at 3:37 AM, Tom Yan <tom.ty89@xxxxxxxxx> wrote: >> On 23 August 2016 at 07:30, Shaun Tancheff <shaun@xxxxxxxxxxxx> wrote: > >> If we really want/need to avoid hitting some real buffer limit (e.g. >> maximum length of scatter/gather list?), then we should in some way >> check n_block against that. If it is too large we then return >> used_bytes = 0 (optionally with some follow-up to add a response to >> such return value or so). > > Yes there is a real buffer limit, I can think of these two options: > 1- Assume the setups from sd_setup_discard_cmnd() and/ > or sd_setup_write_same_cmnd() are providing an sglist of > sdp->sector_size via scsi_init_io() That sounds completely wrong. The scatter/gather list we are talking about here has nothing to do with the SCSI or block layer anymore. The SATL has _already_ parsed the SCSI Write Same (16) command and is packing ranges/payload according to that in this stage. If there is any limit it would probably the max_segment allowed by the host driver (e.g. ahci). It doesn't seem to make sense to me either that we would need to prevent sglist overflow in such level. Doesn't that mean we would need to do the same checking (specifically, as in hard coding checks in all kinds of procedures) in every use of scatter/gather list? That doesn't sound right at all. > > 2- Find (or write) a suitable sg_get_size(sgl, nents) to walk the > sglist and calculate the available buffer size. > > #2 sounds like more fun but I'm not sure it's what people would prefer to see. No idea if such thing exists / makes sense at all. > > -- > Shaun -- To unsubscribe from this list: send the line "unsubscribe linux-scsi" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
