tcp_lp_cong_avoid(9)

πŸ“¦ version 7.3.1, in nixpkgs unstable

πŸ’… tcp_lp_cong_avoid(9) β€” a little kernel moment

ok so first things first, bestie: this is a page from the Kernel API Manual (yes, that manual), and today's main character is tcp_lp_cong_avoid. it's giving very niche, very kernel-internals, very iykyk πŸ’–


✨ NAME ✨

tcp_lp_cong_avoid β€”

(and honestly? that's the whole name entry. no description. the man page said "she's mysterious" and left it at that πŸ™ˆ)


✨ SYNOPSIS ✨

void tcp_lp_cong_avoid (struct sock *sk , u32 ack , u32 acked );

like… three args, void return, no notes. she's clean 🧚


πŸŽ€ ARGUMENTS πŸŽ€

  • sk β€” the socket we're trying to keep from getting all congested. the whole point of the function, lowkey ⭐
  • ack β€” the current ack sequence number. the now-ness of the ack.
  • acked β€” how many packets got ACKed. a little count, a little tally πŸ“

πŸ’œ DESCRIPTION πŸ’œ

ok so this is the implementation of cong_avoid. it'll only call newReno CA when it's away from inference β€” like, newReno only gets the aux cord when we're not in inference mode, periodt.

and per the TCP-LP paper, this whole situation gets handled in additive increasement β€” that's the paper's term, not mine, don't come for me 😭


🌈 SEE ALSO 🌈

the whole TCP-LP friend group, honestly. go read the siblings, they all ate 🫢


ngl this one is short but she's important β€” it's a real kernel congestion-avoidance callback, so be nice to her πŸ’€

ok literally that's the page!! sourced from ./net/ipv4, dated October 2026, and it's the tcp_lp_cong_avoid(9) moment. mwah πŸ’‹βœ¨πŸŽ€ slay and read your kernel source, bestie πŸ‘‘

πŸ“¦ where this page comes from