Errata for Understanding Linux Network Internals

"I could work on my image processing projects at home on an i286/486 computer without having to compete with other students for access to the few Sun stations available at the university labs." Linux doesn't run on a i286 -- it starts at a i386.

To say "When no optional block is required, placeholder is just a pointer to the end of the structure" is not entirely accurate. It is certainly true for a variety of structures, however it is only true when previous data causes your placeholder be aligned to the same alignment as the structure. Case in point: struct abc { int age; char whatever; char placeholder[0]; }; in this case, placeholder[0] certainly does not point to the end of the structure. This is because the compiler may align struct abc to keep age on an architecture specific alignment (perhaps a multiple of 4 for example). Here you could actually think of placeholder taking up 3 bytes if the structure alignment were 4. I've seen others make similar mistakes. It is not correct to think that the offsetof the placeholder field is equal to the sizeof the structure, UNLESS you guarantee that placeholder is aligned properly. You can do that with an align attribute (gcc). For a more compiler independent solution, using a more strongly aligned datatype for placeholder might suffice (perhaps double placeholder[0]).

Maybe a mistake, maybe not. I could be wrong here, but I suspect that the real reason for the skb_reserve(skb,2) is to keep the IP header aligned on a 4 byte boundary rather than a 16 byte boundary. 4 byte alignment makes sense as the IP, TCP and UDP header fields are all aligned on 32 bit boundaries to optimize access by 32 bit processors. A 16 byte alignment would be harder to explain unless the cache line size of the processor happened to be 16 bytes, in which case the code to do this would be over in the arch directory someplace.

In the sentence "skb_pull removes a block of data from the head of the buffer by moving the head pointer forward", the "head pointer" should be "data pointer"

In "between skb->start and skb->end", skb->start should be skb->head

ctl_data should be ctl_table

Network E is not shown in the figure, but the text (last sentence on page 75 and continues on page 76) talks about Network E in figure 4-1.

This paragraph says that dev_alloc_name is called by alloc_netdev. However, this function is called by register_netdev. The problem with this is that the name member of the allocated netdev is still a format string after the call to alloc_netdev and cannot be used to properly identify the device.

"low-latency" should be "low-priority". "high-latency" should be "high-priority"

The AppleTalk Address Resolution Protocol and IPX have the same 24-bit OUI. 00:00:00 belongs to Xerox.

"IE, one instead of eye" should be "1F" in 8 places.

Delete "hosts located in".

"states" should be "nodes".

"Port ID" should be "Port number" in two places.

Add a footnote that MAC addresses can be cloned, so they are not necessarily unique.

Figure 15-9 refers to DT, but does not define it.

"Bridge ID BR3 < Bridge ID BR4" should be "Bridge ID BR2 < Bridge ID BR3"

BR4 should be BR3.

There are two BR4s; the right middle one should be BR3. In the BPDU from BR4 to BR3, the Cost should be 110.

"than BR2's" should be "than BR3's".

BR2 should be BR1.

eth2 should be eth0.

br_uninit should be br_deinit.

"bridge device" should be "STP".

in_dev should be in_device.

"L4 header and checksum" should be "L4 header and payload".

fraghdrlen should be maxfraglen.

dst_input should be dst_output.

128 should be 228 in 3 places. 384 should be 484.

ip_options is defined in net/inet_sock.h

Should be replaced with something like "declaring an array that will be defined by an immediately succeeding declaration."

"head" should be "previous element".

"the packet has been conpletely received" should be "there are no gaps in the received fragments".

"hash" should be "protocol".

Delete "non-" in "non-zero"

NUD_IN_TIMER should include NUD_REACHABLE too.

"NUD_STALE" should be "NUD_DELAY or NUD_PROBE".

Figure 26-13 does not show a transition from NUD_REACHABLE to NUD_DELAY.

"one half" should be "twice".

Delete "a little faster than usual".

Swap eth souce addr and eth dest addr in Figure 28-1. The same error occurs in Figure 27-1 and 27-10 too.

arp_announce should be arp_ignore in 4 places.

"broadcast" should be "loopback".

8 should be 16

The class B network: last address is shown as "192.255.255.255.". It should be 191.255.255.255.

"brd 192.168.1.255" should be "brd 192.168.1.127"

The sentence reads "The hosts of subnet 10.0.0.0/24 and 10.0.0.2/24 ...". The second subnet should be 10.0.2.0/24, not 10.0.0.2/24.

Regarding the explanation of the macros "change_nexthops, for_nexthops, endfor_nexthops", the description "Note that for each macro there are two versions: one used when there is Policy Routing support in the kernel and one when there is no Policy routing. The second one is optimized by ....". In Linux routing, Policy routing refers to multiple routing tables. I believe that the author refers to Multipath support. Also I checked the Linux source code, the macros are defined differently like "#ifdef CONFIG_IP_ROUTE_MULTIPATH" ... I believe that the text incorrectly referred to Policy Routing. It should be Multipath support instead.

" rt_hash_mask rt_hash_log The size (number of buckets) of the hash table rt_hash_table, and the base two logarithm of that number.... " The description of rt_hash_log is accurate. However, rt_hash_mask is not described accurately. It is the size (number of buckets) of the hash table - 1. And the size (number of buckets) is guaranteed to be the power of 2. That's why rt_hash_mask is all 1s and named "mask".

"ip_local_driver" was referenced twice in the figure. It should be changed to "ip_local_deliver".