Skip to contentSkip to conversation

neighbour.c, pneigh_get_next() skips published entry

PATCH In review Andrew Morton 4 participants 5 messages Sep 25, 2006 – Sep 27, 2006

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 neighbour.c, pneigh_get_next() skips published entry

I've been sitting on this patch because afaik the problem which it purports to fix remains unfixed.

Should I drop it??

Thanks.

From: "Jari Takkala" <[email protected]>

Fix a problem where output from /proc/net/arp skips a record when the full output does not fit into the users read() buffer.

To reproduce: publish a large number of ARP entries (more than 10 required on my system). Run 'dd if=/proc/net/arp of=arp-1024.out bs=1024'. View the output, one entry will be missing.

[akpm: submitted before, discussion ended inconclusively, iirc]

---

 net/core/neighbour.c |    6 ++++++
 1 file changed, 6 insertions(+)
net/core/neighbour.c renamed from net/core/neighbour.c~neighbourc-pneigh_get_next-skips-published-entry +6 -0
diff -puN net/core/neighbour.c~neighbourc-pneigh_get_next-skips-published-entry net/core/neighbour.c
@@ -2209,6 +2209,12 @@ static struct pneigh_entry *pneigh_get_n
 	struct neigh_seq_state *state = seq->private;
 	struct neigh_table *tbl = state->tbl;
 
+	if (pos != NULL && *pos == 1 &&
+			(pn->next || tbl->phash_buckets[state->bucket])) {
+		--(*pos);
+		return pn;
+	}
+
 	pn = pn->next;
 	while (!pn) {
 		if (++state->bucket > PNEIGH_HASHMASK)

_

maintainer
#1 lore raw

From: Andrew Morton <[email protected]>
Date: Mon, 25 Sep 2006 16:45:35 -0700

Quoted 6 lines

I've been sitting on this patch because afaik the problem which it purports to fix remains unfixed.

Should I drop it??

Thanks.

Please drop it, the patch submitted didn't give us the feedback and test results we asked for which is necessary to pinpoint the true issue here.

#2 lore raw
Quoted 6 lines · On Mon, 25 Sep 2006 16:47:31 -0700 (PDT) David Miller <[email protected]>…

From: Andrew Morton <[email protected]>
Date: Mon, 25 Sep 2006 16:45:35 -0700

Quoted 3 lines

I've been sitting on this patch because afaik the problem which it purports to fix remains unfixed.

Quoted 7 lines
Quoted 3 lines

Should I drop it??

Thanks.

Please drop it, the patch submitted didn't give us the feedback and test results we asked for which is necessary to pinpoint the true issue here.

Well that's why I hang onto such patches: so I can bug people about it every few months. Consider it the world's dumbest bug-tracking system.

But I have a feeling I'll get shouted at if I try it again with this one (looks at the 18-month-old tulip-fix-for-64-bit-mips.patch), so yeah, I'll drop it.

#3 lore raw
In reply to David Miller
Quoted 10 lines · David Miller
Quoted 6 lines

I've been sitting on this patch because afaik the problem which it purports to fix remains unfixed.

Should I drop it??

Thanks.

Please drop it, the patch submitted didn't give us the feedback and test results we asked for which is necessary to pinpoint the true issue here.

It is faster to reproduce with a smaller block size.
I haven't looked in detail to find the cause, but I did notice that neigh_seq_start() does a pos_minus_one adjustment, and neigh_seq_next() does not.

First, with only one entry, both block sizes work:

# dd if=/proc/net/arp bs=1024
IP address       HW type     Flags       HW address            Mask     Device
10.46.1.1        0x1         0x2         00:04:23:C8:8D:E9     *        eth0
0+1 records in
0+1 records out
156 bytes (156 B) copied, 0.000161 seconds, 969 kB/s
# dd if=/proc/net/arp bs=128
IP address       HW type     Flags       HW address            Mask     Device
10.46.1.1        0x1         0x2         00:04:23:C8:8D:E9     *        eth0
1+1 records in
1+1 records out
156 bytes (156 B) copied, 0.000193 seconds, 808 kB/s

But add another entry, and it is lost with bs=128:

# arp -Ds 1.1.1.1 eth1 pub
# dd if=/proc/net/arp bs=1024
IP address       HW type     Flags       HW address            Mask     Device
10.46.1.1        0x1         0x2         00:04:23:C8:8D:E9     *        eth0
1.1.1.1          0x1         0xc         00:00:00:00:00:00     *        eth1
0+1 records in
0+1 records out
233 bytes (233 B) copied, 2.6e-05 seconds, 9.0 MB/s
# dd if=/proc/net/arp bs=128
IP address       HW type     Flags       HW address            Mask     Device
10.46.1.1        0x1         0x2         00:04:23:C8:8D:E9     *        eth0
1+1 records in
1+1 records out
156 bytes (156 B) copied, 7.9e-05 seconds, 2.0 MB/s

Add more entries, but still only one is lost:

# arp -Ds 1.1.1.2 eth1 pub
# arp -Ds 1.1.1.3 eth1 pub
# arp -Ds 1.1.1.4 eth1 pub
# arp -Ds 1.1.1.5 eth1 pub
# dd if=/proc/net/arp bs=128
IP address       HW type     Flags       HW address            Mask     Device
10.46.1.1        0x1         0x2         00:04:23:C8:8D:E9     *        eth0
1.1.1.3          0x1         0xc         00:00:00:00:00:00     *        eth1
1.1.1.2          0x1         0xc         00:00:00:00:00:00     *        eth1
1.1.1.5          0x1         0xc         00:00:00:00:00:00     *        eth1
1.1.1.4          0x1         0xc         00:00:00:00:00:00     *        eth1
3+1 records in
3+1 records out
464 bytes (464 B) copied, 3.6e-05 seconds, 12.9 MB/s
#4 lore raw
Quoted 5 lines · Philip Craig

It is faster to reproduce with a smaller block size.
I haven't looked in detail to find the cause, but I did notice that neigh_seq_start() does a pos_minus_one adjustment, and neigh_seq_next() does not.

Yeah I can reproduce this now. I think you actually sent me the instructions before but I didn't get around to testing it. Sorry about that.

I'll look into this.

Cheers,

Signature · 4 lines
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt