gnu has --reference=FILE .. maybe it'd be worth using that (for that case)

maybe --reinherit, though that might be worth more thought e.g. should it recurse _up_ then apply encountered inherited ACLs in order (to the target files, not everything inbetween) on the way down

well nto might probably should

(more thought)

I think you'd want to have some way to constrain how far up it looked.

nomad: so after a few too many laaaaate night sessions, the ice driver is (finally) sending and receiving traffic

though a lot of the offloads are still disabled at this point and still missing some other things that need to be implemented)

in particular, the RX checksum bits look to be a bit of a tedious pain (as surprising as that might sound)

though i might import (w/ attribution of course) the rather large static table from the fbsd driver that actually let's you basically lookup when the IP or TCP/UDP checksom has been validated

(building that from hand is the tedious part)

on a somewhat related subject, I was reminded of something I saw from the hyper-v.. I think it's more of a general NDIS thing (though trying to find the actual relevant docs has not proven fruitful, so there's a fair amount of guesswork/assumption)

where they have what they call a 'chimney'

I _think_ the idea is that it allows for selected packets to jump to the front of the line when transmitting

which seems like it might be useful for certain things (imagine getting handed a long chain of packets (mblk_t->b_next) to transmit.. that could tie the TX ring up for a bit and being able to cut in line for things like certain TCP notifications seems like it could maybe be useful?

though I haven't looked to see how often mac sends more than one packet at a time down to the driver

(I'm guessing maybe more likely when the driver applies backpressure and returns mblk_ts in it's mc_tx(9E) function

jbk: I'm trying to work out by what analogy that could be "chimney" and am failig

not that that's not what it means, but I'd love to read the docs

I guess... hotter gasses rise faster?

and handwave handwave

no idea, and since I wasn't able to find good docs, I'm not 100% sure on it

other than it seemed to be a separate 'channel' for sending to the device

I get way too interesetd in how people come up with words

why I thought maybe it was a way to cut in line

at least it seems like something like that could be useful

a multi-flue chimney, perhaps?

a google search for [ndis chimney] turned up: learn.microsoft.com/en-us/windows-h…re/drivers/network/full-tcp-offload

ooooh

ok

so not that :)

(err not what I thought)

"This architecture is called a 'chimney offload' architecture because it provides a direct connection, called a 'chimney,' between applications and an offload-capable NIC. The chimney enables the NIC to perform TCP processing for offloaded connections, including maintaining the protocol state. "

so goes from the basement to above the roof

bypassing all the protocol layers on the floors in between

well I guess I should say Herbert Lom's version of it

Complete with eye twitch.

exactly

jbk: do you need testers for ICE or are you not quite there yet?

not quite there yet -- I need to fix some TX bits, at least add some more stats

and probably do a bit more load just to make sure it's not mostly stable

k

the RX problem ended up being silly, but given the size and complexity, probably not too unsurprising given how many interruptions and times i've had to set it aside

though even having copilot analyize the code (_no_ code generation -- just review) couldn't find it either :P

i'm fixing some TX bugs (or trying to today.. i keep getting interrupted by other work things)

ooh nice.. new hard drive arrived...

which means I can use the old one (it's not dead, but has a lot of bad sectors.. more than any reserve) as a guinea pig for doing FW updates :P

can anyone kick redmine?