Re: [PATCH net-next v3 03/15] quic: provide common utilities and data structures

On Tue, Sep 23, 2025 at 7:21 AM Paolo Abeni <[email protected]> wrote:
>
> On 9/19/25 12:34 AM, Xin Long wrote:
> > This patch provides foundational data structures and utilities used
> > throughout the QUIC stack.
> >
> > It introduces packet header types, connection ID support, and address
> > handling. Hash tables are added to manage socket lookup and connection
> > ID mapping.
> >
> > A flexible binary data type is provided, along with helpers for parsing,
> > matching, and memory management. Helpers for encoding and decoding
> > transport parameters and frames are also included.
> >
> > Signed-off-by: Xin Long <[email protected]>
> > ---
> > v3:
> >   - Rework hashtables: split into two types and size them based on
> >     totalram_pages(), similar to SCTP (reported by Paolo).
> >   - struct quic_shash_table: use rwlock instead of spinlock.
>
> Why? rwlock usage should be avoided in networking (as it's unfair, see
> the many refactors replacing rwlock with rcu/plain spinlock)
Interesting, I thought rwlock works better than spinlock in this case.
I will change back to spinlock.

>
> [...]
> > +
> > +static int quic_uhash_table_init(struct quic_uhash_table *ht, u32 max_size, int order)
> > +{
> > +     int i, max_order, size;
> > +
> > +     /* Same sizing logic as in quic_shash_table_init(). */
> > +     max_order = get_order(max_size * sizeof(struct quic_uhash_head));
> > +     order = min(order, max_order);
> > +     do {
> > +             ht->hash = (struct quic_uhash_head *)
> > +                     __get_free_pages(GFP_KERNEL | __GFP_NOWARN, order);
> > +     } while (!ht->hash && --order > 0);
>
> You can avoid a little complexity, and see more consistent behaviour,
> using plain vmalloc() or alloc_large_system_hash() with no fallback.
>
I wanted to use alloc_large_system_hash(), but the memory allocated
by it is usually NOT meant to be freed at runtime. I don't see a free_
function to do it either.

If QUIC works as a kernel module, what should I do with this memory
in module_exit()?

>
> > +/* rfc9000#section-a.3: DecodePacketNumber()
> > + *
> > + * Reconstructs the full packet number from a truncated one.
> > + */
> > +s64 quic_get_num(s64 max_pkt_num, s64 pkt_num, u32 n)
> > +{
> > +     s64 expected = max_pkt_num + 1;
> > +     s64 win = BIT_ULL(n * 8);
> > +     s64 hwin = win / 2;
> > +     s64 mask = win - 1;
> > +     s64 cand;
> > +
> > +     cand = (expected & ~mask) | pkt_num;
> > +     if (cand <= expected - hwin && cand < (1ULL << 62) - win)
> > +             return cand + win;
> > +     if (cand > expected + hwin && cand >= win)
> > +             return cand - win;
> > +     return cand;
>
> The above is a bit obscure to me; replacing magic nubers (62) with macro
> could help. Some more comments also would do.
>
The code is exactly from the commented doc:
/* rfc9000#section-a.3: DecodePacketNumber()

See:
https://datatracker.ietf.org/doc/html/rfc9000#section-a.3

I will bring some comments from there.

Thanks.