| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <f36b08ee0909290155u177b1983y2eafa8b353a143e0@mail.gmail.com>
Date: Tue, 29 Sep 2009 11:55:18 +0300
From: Yakov Lerner <iler.ml@...il.com>
To: Eric Dumazet <eric.dumazet@...il.com>
Cc: netdev@...r.kernel.org, davem@...emloft.net
Subject: Re: [PATCH] /proc/net/tcp, overhead removed
On Tue, Sep 29, 2009 at 10:56, Eric Dumazet <eric.dumazet@...il.com> wrote:
>
> Yakov Lerner a écrit :
> > Take 2.
> >
> > "Sharp improvement in performance of /proc/net/tcp when number of
> > sockets is large and hashsize is large.
> > O(numsock * hashsize) time becomes O(numsock + hashsize). On slow
> > processors, speed difference can be x100 and more."
> >
> > I must say that I'm not fully satisfied with my choice of "st->sbucket"
> > for the new preserved index. The better name would be "st->snum".
> > Re-using "st->sbucket" saves 4 bytes, and keeps the patch to one sourcefile.
> > But "st->sbucket" has different meaning in OPENREQ and LISTEN states;
> > this can be confusing.
> > Maybe better add "snum" member to struct tcp_iter_state ?
> >
> > Shall I change subject when sending "take N+1", or keep the old subject ?
> >
> > Signed-off-by: Yakov Lerner <iler.ml@...il.com>
> > ---
> > net/ipv4/tcp_ipv4.c | 35 +++++++++++++++++++++++++++++++++--
> > 1 files changed, 33 insertions(+), 2 deletions(-)
> >
> > diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
> > index 7cda24b..e4c4f19 100644
> > --- a/net/ipv4/tcp_ipv4.c
> > +++ b/net/ipv4/tcp_ipv4.c
> > @@ -1994,13 +1994,14 @@ static inline int empty_bucket(struct tcp_iter_state *st)
> > hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].twchain);
> > }
> >
> > -static void *established_get_first(struct seq_file *seq)
> > +static void *established_get_first_after(struct seq_file *seq, int bucket)
> > {
> > struct tcp_iter_state *st = seq->private;
> > struct net *net = seq_file_net(seq);
> > void *rc = NULL;
> >
> > - for (st->bucket = 0; st->bucket < tcp_hashinfo.ehash_size; ++st->bucket) {
> > + for (st->bucket = bucket; st->bucket < tcp_hashinfo.ehash_size;
> > + ++st->bucket) {
> > struct sock *sk;
> > struct hlist_nulls_node *node;
> > struct inet_timewait_sock *tw;
> > @@ -2010,6 +2011,8 @@ static void *established_get_first(struct seq_file *seq)
> > if (empty_bucket(st))
> > continue;
> >
> > + st->sbucket = st->num;
> > +
> > spin_lock_bh(lock);
> > sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
> > if (sk->sk_family != st->family ||
> > @@ -2036,6 +2039,11 @@ out:
> > return rc;
> > }
> >
> > +static void *established_get_first(struct seq_file *seq)
> > +{
> > + return established_get_first_after(seq, 0);
> > +}
> > +
> > static void *established_get_next(struct seq_file *seq, void *cur)
> > {
> > struct sock *sk = cur;
> > @@ -2064,6 +2072,9 @@ get_tw:
> > while (++st->bucket < tcp_hashinfo.ehash_size &&
> > empty_bucket(st))
> > ;
> > +
> > + st->sbucket = st->num;
> > +
> > if (st->bucket >= tcp_hashinfo.ehash_size)
> > return NULL;
> >
> > @@ -2107,6 +2118,7 @@ static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
> >
> > if (!rc) {
> > st->state = TCP_SEQ_STATE_ESTABLISHED;
> > + st->sbucket = 0;
> > rc = established_get_idx(seq, pos);
> > }
> >
> > @@ -2116,6 +2128,25 @@ static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
> > static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
> > {
> > struct tcp_iter_state *st = seq->private;
> > +
> > + if (*pos && *pos >= st->sbucket &&
> > + (st->state == TCP_SEQ_STATE_ESTABLISHED ||
> > + st->state == TCP_SEQ_STATE_TIME_WAIT)) {
> > + void *cur;
> > + int nskip;
> > +
> > + /* for states estab and tw, st->sbucket is index (*pos) */
> > + /* corresponding to the beginning of bucket st->bucket */
> > +
> > + st->num = st->sbucket;
> > + /* jump to st->bucket, then skip (*pos - st->sbucket) items */
> > + st->state = TCP_SEQ_STATE_ESTABLISHED;
> > + cur = established_get_first_after(seq, st->bucket);
> > + for (nskip = *pos - st->num; cur && nskip > 0; --nskip)
> > + cur = established_get_next(seq, cur);
> > + return cur;
> > + }
> > +
> > st->state = TCP_SEQ_STATE_LISTENING;
> > st->num = 0;
> > return *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
>
> Just in case you are working on "take 3" of the patch, there is a fondamental problem.
>
> All the scalability problems come from the fact that tcp_seq_start()
> *has* to rescan all the tables from the begining, because of lseek() capability
> on /proc/net/tcp file
>
> We probably could disable llseek() (on other positions than start of the file),
> and rely only on internal state (listening/established hashtable, hash bucket, position in chain)
>
> I cannot imagine how an application could rely on lseek() on >0 position in this file.
I thought /proc/net/tcp can both be fast and allow lseek;
(1) when no lseek was issued since last read
(we can detect this), /proc/net/tcp can jump to the
last known bucket (common case), vs
(2) switch to slow mode (scan from the beginning of hash)
when lseek was used , no ?
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists