[SOLVED] Error: "SIOCSIFADDR: Cannot allocate memory" adding virtual IPv6 Addresses

janneaa · 04-30-2010, 06:39 AM

I get errors trying to virtual IPv6 Addresses i a lab environment. It works fine up to 2033 (?) adrressses, but when I try to add more i get "Cannot allocate memory" error:

# ifconfig eth0 add 2001:1b70:8282:2021:18:0:20:1
SIOCSIFADDR: Cannot allocate memory

This is on Ubuntu Server 9.10 (64-bit).

Any ideas on why we run into this limitation?
Any kernel parameters that can be changed?

/Janne

foodown · 05-02-2010, 12:31 AM

Well . . . If I understand correctly, and you have assigned 2033 addresses to a network interface, and the error message says that it can't allocate any more memory for yet another address, I think that I can guess what could be wrong . . .

After replicating this error, please post the output of . . .

Code:

vmstat -s

janneaa · 05-03-2010, 11:20 AM

This is the output of "vmstat -s".

Code:

# vmstat -s
    49556192 K total memory
     1269392 K used memory
      163412 K active memory
      120756 K inactive memory
    48286800 K free memory
       99844 K buffer memory
      164896 K swap cache
     5849080 K total swap
           0 K used swap
     5849080 K free swap
     2059179 non-nice user cpu ticks
          34 nice user cpu ticks
     1649679 system cpu ticks
  1567045977 idle cpu ticks
        1455 IO-wait cpu ticks
      264423 IRQ cpu ticks
     1566541 softirq cpu ticks
           0 stolen cpu ticks
      203851 pages paged in
      715786 pages paged out
           0 pages swapped in
           0 pages swapped out
   916165536 interrupts
  1089210944 CPU context switches
  1271920924 boot time
     2184939 forks

foodown · 05-03-2010, 05:40 PM

Well . . . plenty of memory there, isn't there?

It may be that you've found a maximum number of virtual interfaces that your kernel will support on one physical NIC. I cannot, however, find any mention of such a limit in any of the kernel documentation or on any kernel newsgroups or discussion boards.

I will continue to research this. I apologize for the delay.

It is possible that there is a line of preprocessor code in one of the kernel headers which defines this limit, or a limit on memory for the kernel routing table. If this is the case, then, in theory, you could change this line and compile a custom kernel with a higher limit, or, possibly, no limit.

Like I said, I'll keep looking.

janneaa · 05-04-2010, 02:05 AM

Yeah, the memory shouldn't be a problem

I agree that we have found some kind of limit, but note that

It's not per NIC, it's the total number of IPv6 addresses that can be added
it's IPv6, when adding IPv4 addresses, we have used up to 260.000 addresses
the limit not exactly 2033, on another machine (same config) we where abe to add 10 more...

Greatful for any idea's!

foodown · 05-06-2010, 10:59 PM

This could be an unresolved bug in a driver.

https://lists.linux-foundation.org/p...ry/011740.html

Are you using any of the following?

3Com Corporation 3c905B 100BaseTX [Cyclone]
Intel Corp. 82557/8/9 [Ethernet Pro 100]

janneaa · 05-07-2010, 01:33 AM

Solved.

The solution was to increase the max size of the IPv6 routing table:
/proc/sys/net/ipv6/route/max_size, which by default is 4096.

I Increased this an where able to create as many addresses as needed...

/Janne

PS. I guess the error message from ifconfig could be improved?

depeje · 06-03-2013, 02:44 PM

Just want to add this to the internet archives, because this still seems to be an issue today with opensuse 12.2.

The same problem has some other symptoms. The first one is that the box doesn't get an address via RA. Forcing a router solicitation also fails:

Code:

krusty:~ # rdisc6 -m eth0                                                                                                                                                               
Soliciting ff02::2 (ff02::2) on eth0...                                                                                                                                                 
Sending ICMPv6 packet: Network is unreachable

On another machine in the same LAN I couldn't receive a solicitation with wireshark. So either the multicast solicitation didn't reach my wireshark PC, or the solicitation is not sent. I believe in the last case, because rdisc6 seemed to give up really fast.

ps. I make the fix permanent by adding this line to /etc/sysctl.conf:

Code:

net.ipv6.route.max_size = 16384

And just to be complete:

Code:

krusty:~ # vmstat -s
      8193140 K total memory
      7982872 K used memory
      1616208 K active memory
      1765260 K inactive memory
       210268 K free memory
        59044 K buffer memory
      4265084 K swap cache
      5242876 K total swap
      1387644 K used swap
      3855232 K free swap
     34468312 non-nice user cpu ticks
    212006556 nice user cpu ticks
    116870346 system cpu ticks
   1641680882 idle cpu ticks
     33816422 IO-wait cpu ticks
         7695 IRQ cpu ticks
      2502007 softirq cpu ticks
            0 stolen cpu ticks
   3746535773 pages paged in
   5399121983 pages paged out
      1332372 pages swapped in
      1301506 pages swapped out
    475418137 interrupts
   1186859131 CPU context switches
   1359917323 boot time
     35437367 forks

I know this post is very old, but I lost a day because I seem to be the first person on the internet who has this problem with my specific symptoms.