TCP/IP Illustrated Volume 2 Chapter 19 Routing Requests and Routing Messages

1
TCP/IP Illustrated Volume 2 Chapter 19 Routing
Requests and Routing Messages
2
Table of Contents

• 19.1 Introduction
• 19.2 rtalloc and rtalloc1 Functions
• 19.3 RTFREE Macro and rtfree Function
• 19.4 rtrequest Function
• 19.5 rt_setgate Function
• 19.6 rtinit Function
• 19.7 rtredirect Function
• 19.8 Routing Message Structure
• 19.9 rt_missmsg Function
• 19.10 rt_ifmsg Function
• 19.11 rt_newaddrmsg Function
• 19.12 rt_msg1 Function
• 19.13 rt_msg2 Function
• 19.14 sysctl_rtable Function
• 19.15 sysctl_dumpentry Function
• 19.16 sysctl_iflist Function
• 19.17 Summary

3
19.17 Summary

• routing messages all have the same format
• fixed-length structure (different fixed length
  according to the type) followed by a variable
  number of socket address structures
• length
• version
• type 3 (different fixed length)
• socket address structures
• one messages per read or write between process
  and kernel
• across a routing socket
• a super user process reads and writes access to
  the kernels routing table
• routed, gated (super user processes)
• without a routing socket
• any process can read the kernels routing table
  using the sysctl

4
19.1 Introduction

• rtalloc
• routing table lookup
• rtalloc1
• routing table lookup
• rtrequest
• adds deletes routing tables
• rtinit
• when the interface goes up or down
• sysctl
• to read entire routing table or
• a list of all configured interfaces and interface
  addresses
• routing messages
• route command, routed, gated generate to a
  routing socket, causing the kernel
• to add a new route
• to delete an existing route
• to modify an existing route
• kernel generates to any routing socket interested
• an interface gone down
• a redirect received

5
19.2 rtalloc and rtalloc1 Functions

• rtalloc
• lookup an entry in the routing table
• ro the pointer to a route structure contained in
  an Internet PCB (used by UDP TCP, Chapter 22)
• returns if
• ro points to an rtentry (ro_rt is nonnull)
• rtentry points to an interface
• the route is up
• otherwise call rtalloc1

6
19.2 rtalloc and rtalloc1 Functions

• rtalloc1
• call rn_match
• calls rnh_matchaddr (initialized to rn_match for
  Internet addresses, Figure 18.17)
• search is successful
• A routing table exists for the protocol family
• rn_match returns a nonnull pointer
• the matching radix_node does not have thee
  RNF_ROOT flag set
• 2 leaves end that mark the end of the tree
  RNF_ROOT flag set
• Search fails
• rts_unreach incremented
• if rtalloc1(2nd argument report) is nonzero, a
  routing message (type RTM_MISS) is generated
• matching radix_node is stored in rt and newrt
• 1st of 2 radix_node structures of rtentry
  contains the leaf node

7
19.2 rtalloc and rtalloc1 Functions

• Create clone entries
• nonzero 2nd argument RTF_CLONING lag set,
  then calls rtrequest with RTM_REOVE to create a
  new rtentry (clone)
• used by ARP and multicast addresses
• Clone creation fails
• newrt is set back to the entry returned by
  rn_match
• increment the reference count
• jump to miss for generating RTM_MISS
• Check for external resolution
• If the newly cloned entry has a RTF_XRESOLVE flag
  set, jump to miss, to generate an RTM_RESOLVE
  message (to notify a user process when the route
  is created)
• Increment reference count for normal successful
  search
• search success but RTF_CLONING flag is not set,
  increment the entrys reference count nonnull
  pointer returns (normal flow)
• p604 Figure 19.3 Summary of operation of rtalloc1
• If default router exists, the first two rows
  (entry not found) are impossible
• increment rt_refcnt and call rtrequst with
  RTM_RESOLVE is OKs

8
19.3 RTFREE Macro and rtfree Function

• RTFREE calls rtfree, of the reference count lt 1,
  otherwise decrements
• rtfree releases rtentry when no more references
  (refer to p719 Figure 22.7)
• if RNT_ACTIVE (still part of the routing table)
  or RNF_ROOT (end marker) is set, then internal
  error
• rttrash
• a debugging counter of the number of routing
  entries not in the routing tree
• incremented by rtrequest when it begins deleting
  a route, decremented in rtfree
• normally 0
• Release interface reference
• if the reference count not negative, then IFAFREE
  decrements the reference count for ifaddr and
  releases it (by ifafree) when it reaches 0
• Releases routing memory
• Free (allocated in rt_setgate in a contiguous
  chunk, for releasing them bya single call)
• routing entry key
• its gateway
• Free rtentry itself

9
19.3 RTFREE Macro and rtfree Function

• Routing Table Reference Counts
• rt_refcnt differs from others
• p560 Figure 18.2 yet the routing table entries
  without any references are not deleted
• in rtfree a reference count 0 is not deleted
  unless the entrys RTF_UP flag is not set
  (cleared by rtrequest when a route is deleted
  from the routing tree)
• Most routes are used in the following fashion
• automatic creation of a route typical Ethernet
  interface point-to-point interface
• rtinit calls rtrequest with RTM_ADD, reference
  count to 1, then rtinit decrements to 0
• manual creation fo a route route command,
  routing daemon
• route_output calling rtrequest with RTM_ADD, then
  route_output decrements it to 0
• IP datagram (TCP or UDP)
• ip_output calls rtalloc, which calls rtalloc1,
  rtalloc1 increment the reference count if the
  route is found
• The located route (held route) points to the
  routing table entry in route structure within a
  protocol control block
• rtentry held by someone else (reference count
  nonzero) cannot be deleted
• RTFREE or rtfree releases a held route
• p231 Figure 8.24 (ip_output), Capter22 (PCB)
• call rtalloc1
• to look up a route (a route to the destination
  exists), but the caller doesnt want to hold the
  route
• increments the counter, then the caller
  immediately decrements it
• delete a route by rtrequest

10
19.4 rtrequest Function

• p607 Figure 19.6 Summary of functions that call
  rtrequest
• rtalloc1
• rtredirect
• rtinit
• route_output
• switch (req) command
• RTM_DELETE
• RTM_RESOLVE
• RTM_ADD
• arguments
• dst a socket address (key)
• sa_family from this key selects the routing table
• flag
• a host route, netmask null (ignoring any value
  the caller passed)
• a route to a network

11
19.4 rtrequest Function

• RTM_DELETE
• Deleting from routing table tree
• returns a pointer to rtentry
• rnh_deladdr (rn_delete, p575 Figure 18.17)
• RTF_UP cleared
• Remove reference to gateway routing table entry
• indirect route through a gateway
• RTFREE decrements the rt_refcnt of the gateways
  entry, and deletes it if the count reaches 0
• rt_gwroute is set to null
• rt is set back
• Call interface request function
• if ifa-gtrtrequest (used by ARP) is defined, call
  ifa-gtrtrequest (Chapter 21)
• Return pointer or release reference
• rttrash global incremented
• if rt_nrt is nonnull (caller wants the pointer to
  the rtentry), then that pointer is returned, but
  the entry cannot be released the caller calls
  rtfree
• if rt_nrt is null, then the entry can be
  released if the reference count is less than or
  equal to 0, then increments it and calls rtfree
• RTM_RESOLVE
• called only from rtalloc1, if RTF_CLONING flag
  set (p603 Figure 19.2)
• ret_nrt points to the entry with the RTF_CLONING
  flag set

12
19.4 rtrequest Function

• RTM_ADD
• Locate corresponding interface
• ifa_ifwithroute finds the ifaddr for dst
• Allocate memory for routing table entry
• R_Malloc
• rtenetry zeroes 2 radix_node for the routing
  tree and the other routing information
• rt_flags are set (RTF_UP set)
• Allocate and copy gateway address
• rtsetgate allocates for dst and gateway (copied)
• sets rt_key, rt_gateway, rt_gwroute
• Copy destination address
• dst copied to rn_key (a new key)
• if network mask is supplied
• then rt_maskedcopy logically ANDs dst and netmask
  (key in the table already ANDed with mask)
  forming the new key (ndst) the search key needs
  to be ANDed
• else dst is copied into the rn_key
• ifconfig le0 inet e104.2252.12.63 0xffffff30
  alias
• netstat (first logically AND the address with
  the mask)

13
19.4 rtrequest Function

• Add entry to routing tree
• rnh_addaddr (rn_addroute from Figure p575 18.17)
  adds rtentry (dst, mask) to the routing table
  tree
• if error (already in), EEXIST
• Store interface pointers
• increments ifaddr reference count and the pointer
  to ifaddr
• stores the pointers to its ifaddr ifnet
• Copy metrics for newly cloned route
• RTM_RESOLVE case, copy entire metrics
• RTM_ADD case, caller set the metrics
• Call interface request function
• if an ifa_rtrequest is defined, calls
  ifa_rtrequest (ARP uses this for RTM_ADD and
  RTM_RESOLVE, Section 21.13)
• line 373 SA ( ) ?
• Return pointer and increment reference count
• if the caller wants, returns a copy of the
  pointer to the new structure through ret_nrt
• increments rt_refcnt

14
19.4 rtrequest Function

• Example Cloned Routes with Network Masks
• RTM_RESOLVE in rtrequest creates the cloned route
  using rt_genmask
• if rt_genmask is nonnull, then the socket address
  pointed by rt_genmask becomes the network mask of
  the newly created route
• p560 Figure 18.2 cloned routes for the local
  Ethernet and for multicast addresses
• class B network, 128.1, a point-to-point link,
  subnet mask 0xffffff00 need
• a routing table entry for all possible 254
  subnets
• a gateway value of a router that knows how to
  reach the link to the 128.1.0.0
• Assume
• the gateway router 128.1.0.0 (mask 0xffff0000) is
  not a default router
• rach of 254 subnets have different RTTs, MTUs,
  delays
• If a separate routing table entry for each
  subnet, whenever a connection is closed , TCP
  would update the routing table with statistics
  (RTT, RTT variance, Figure 27.3)
• by hand using route command one per subnet
• automatically using cloning feature
• creates one entry (dst 128.1.0.0, network mask
  0xfff0000)
• sets RTF_CLONING flag
• sets genmask 0xfffff00
• searches for 128.1.2.3 128.1.2 subnet not exist,
  128.1 is the best match
• creates a new entry (dst 128.1.2, genmask
  0xffffff00)

15
19.5 rt_setgate Function

• each leaf in the routing tree
• a key (rt_key) (rn_key member of radix_node
  (r)-gtrt_nodes-gtrn_key) socket addresses
• a gateway (rt_gateway) socket addresses
• (examples) refer to p560 Figure 18.2
• 127.0.0.1
• gateway member Internet socket address
• by route system call
• 140.252.13.33
• gateway member data-link socket address
  (Ethernet address)
• by ARP
• rt_setgate allocates memory for both (Internet
  socket address and Ethernet address)

16
19.5 rt_setgate Function

• Set lengths from socket address structure
• dlen length of the destination socket address
  structure
• glen length of the gateway socket address
  structure
• ROUNDUP macro rounds up the value up to the next
  multiple of 4 bytes
• Allocate memory
• not yet allocated glen gt sa_len, R_Malloc(new)
  rn_key pointto the new memory
• Use memory already allocated for key and gateway
• point to the existing memory
• Copy newgateway
• Bcopy(gate to rt-gtrt_gateway), sets rt-gtgateway
  to the socket address (newdlen)
• Copy key from old memory to new memory
• copy dst to the new piece of memory (new),
  releases the old piece of memory
• Release gateway routing pointer
• if nonnull rt-gtrt_gwroute, then RTFREE(rt)
• Locate and store new gateway routing pointer
• if indirect route, rtalloc1 locates the entry for
  the new gateway (stored in rt_gwroute)

17
19.6 rtinit Function

• 4 calls to rtinit
• in_control calls twice when the destination
  address of a point-to-point interface is set
  (p172 Figure 6.21)
• RTM_DELETE to delete any existing route to the
  destination
• RTM_ADD to add the new route
• in_ifinit calls to add a network route for a
  broadcast network or a host route for a
  point-to-point link (p171 Figure 6.19)
• in-ifscrub calls to delete an existing route for
  an interface
• Command argument (cmd) is always RTM_ADD or
  RTM_DELETE
• Get deadstination address for route
• if a host, the dst is the the other end of the
  point-to-point link
• otherwise, a network route, dst is the unicast
  address of the interface (ifa_netmask)
• Mask network address with network mask
• if RTM_DELETE, if a network route with a network
  mask, then allocate an mbuf, rt_maskedcopy
  (logically ANDing), dst points to the masked copy
  (destination looked up)
• Search for routing table entry
• rtallocc1 searches the routing table for dst
• if found, reference count is decremented (since
  rtalloc1 incremented it)

18
19.6 rtinit Function

• process request
• rtrequest executes the command (RTM_ADD or
  RTM_DELETE),releases the allocated mbuf
• Generate routing message on successful delete
• returns 0 with a pointer (nrt) to the rt entry
  deleted
• generates a routing socket message by
  rt_newaddmsg
• if the reference count lt 0, increments it and
  releases it by rtfree
• Successful add
• returns 0 with a pointer (nrt) to the rt entry
  added
• decrements the reference count (since rtrequest
  incremented it)
• Incorrect interface
• if the pointer to the interfaces ifaddr in the
  new routing table entry does not equals the
  callers argument (ifa, located by
  ia_ifwithroute), then error (ifa_rtrequest with
  RTM_DELETE, releases ifaddr, sets rt_ifa to the
  callers value, recalls ifa_rtrequest with
  RTM_ADD)
• Generate routing message
• generates a routing socket message by
  rt_newaddrmsg for the RTM_ADD

19
19.7 rtredirect Function

• ICMP redirect
• imcp_input calls rtredirect, then calls
  pfctlinput (udp_ctlinput, tcp_ctlinput goes
  through all Internet protocol control blocks
  inpcbs) (p322 Figure 11.27)
• if the PCB is connected to the foreign address
  (redirected) and holds a route to that foreign
  address, then releases the route by rtfree
• calls rtalloc to look up the foreign address in
  the routing table, possibly finds a new
  (redirected) route
• rtredirect
• validates the information in the redirect
• updates the routing table immediately
• generates a routing socket message
• arguments
• dst the destination IP address not the datagram
  that caused the redirect (HD in p224 Figure 8.18)
• gateway IP address of the router to use as the
  new gatewayfield for the desctination (R2 in p224
  Figure 8.18)
• netmask null when called by icmp_input
• flags RTF_GATEWAY, RTF_HOST
• src IP address of the router that sent the
  redirect (R1 in p224 Figure 8.18)
• rtp null when called by icmp_input

20
19.7 rtredirect Function

• New gateway must be connected
• The new gateway must be directly connected
• Locate routing table entry for destination and
  validate redirect
• rtalloc1 searches the routing table for a route
  to the destination
• must be true
• the RTF_DONE not set
• rtalloc locates a routing table entry for dst
• the address of the router that sent the redirect
  (src) the current rt_gateway for the
  destination
• the interface for the new gateway (the ifa
  returned by ifa_ifwithnet) the current
  interface for the destination (rt_ifa) the new
  gateway must be on the same network as the
  current gateway
• the new gateway cannot redirect this host to
  itself, that is, there cannot exist an attached
  interface with a unicast address or a broadcast
  address equal to gateway ifa_ifwithaddr(gateway)
  should be null (?)
• Must create a new route
• if the route not found, if the located routing
  table entry is the default route, then creates a
  new entry
• A host with access to multiple routers can use
  this to learn of the correct router when the
  default is not correct
• The test for default route is whether the routing
  table entry has an associated mask and if the
  length field of the mask lt 2 (the mask for the
  default route is rn_zeroes, p588 Figure 18.35)

21
19.7 rtredirect Function

• Create new host route
• If a network route, a host redirect (not a
  network redirect), then creates a new host route
  for dst, the existing network route is left alone
• Net/3 ICMP considers all received redirects as
  host redirects (RTF_HOST)
• a hostre direct .vs. a network redirect (?)
• Create route
• rtrequest (setting RTF_GATEWAY RTF_DYNAMIC
  (?))
• netmask is null, since the new route is a host
  route (implied mask of all one bits)
• Modify existing host route
• when the current route to the destination is
  already a host route, do not create a new entry,
  but the existing entry is modified (rt_setgate
  with RTF_MODIFY)
• Ignore if destination is directly connected
• If the current route to dst is direct route
  (RTF_GATEWAY not set), a redirect for a dst
  already directly connected, then returns
  EHOSTUNREACH
• Return pointer and Increment statistics
• a located routing table entry, either returned
  (nonnull rtp and no errors) or released by
  rtfree, increments statistics
• Generate routing message
• clears rt_addrinfo, generates a routing socket
  message by rt_missmsg, then sends it by raw_input
  to any process interested in the redirect

22
19.8 Routing Message Structure

• a fixed-length header up to 8 socket address
• 3 types of fixed length headers
• rt_msghdr
• if_meghr
• ifa_msghdr
• p 571 Figure 18.11 Relationships between the
  various routing functions
• p 570 Figure 18.9 Types of messages exchanged
  across a routing socket
• the first three (the message length, version, and
  type) are the same
• a member (a bit mask) that encodes which of the 8
  potential socket address structures follows the
  header
• rtm_addrs
• ifm_addrs
• ifam_addrs
• p622 Figure 19.16 rt_msghr (the most common)
• p622 Figure 19.16 if_msghr (RTM_IFINFO)
• p622 Figure 19.16 ifa_msghr (RTM_NEWADDR,
  RTM_DELADDR)
• p623 Figure 19.19 Constants used to members of
  rti_info array
• bitmask is only for messages etwen a process and
  the kernel
• array indexes are used within the kernel, refers
  to its rt_addrinfo
• socket addresses always occur in order of
  increasing array index

23
19.8 Routing Message Structure

• p623 Figure 19.20 rt_addrinfo
• RTA_GATEWAY bit set in the rti_addrs, then the
  member rti_infoRTAX_GATEWAY is a pointer to a
  gateway socket address (sockaddr_in containing
  the gateways IP)
• rosock.c defines dst, gate, netmask, genmask,
  ifpaddr, afaaddr, brdaddr
• define dst info.rti_infoRTAX_DST
• rt_addrinfo encountered in
• rtalloc1 (p 603 Figure 19.2) p624 Figure 19.21
• rtredirect (p 618 Figure 19.14) p624 Figure
  19.22
• rti_addrs
• within the kernel not initialized with the
  bitmask (a null pointer in rti_info arrays is a
  nonexistent socket address)
• messages between a process and the kernel needs
  bitmask
• p623 Figure 19.23 route_cb
• 4 counters of the number of routing sockets for
  IP, XNS, OSI, any
• By keeping track of routing socket listeners, the
  kernel avoids building a routing message and
  calling raw_input (to send the message when there
  arent any processes waiting for a message)

24
19.9 rt_missmsg Function

• calls rt_msg1 to build a corresponding
  variable-length message for process in an mbuf
  chain , an then calls raw_input to pass the mbuf
  chain to al appropriate routing sockets
• if there arent any routing socket listeners,
  then returns
• Build message in mbuf chain
• p626 Figure 19.25 Mbuf chain built by rt_msg1
  corresponding p624 Figure 19.22
• raw_input calls sbappendaddr to append the mbuf
  chain to a sockets receive buffer
• Finish building message
• sets rtm_flag, rtm_errno
• copies rti_addrs to rtm_addrs
• 0, but rt_msg1 calculates and stores the
  appropriate bitmask, based on which pointers in
  rti_info are nonnull
• Set protocol of message, call raw_input
• raw_input
• struct sockaddr route_dst 2, PF_ROUTE,
  (where?)
• struct sockaddr route_src 2, PF_ROUTE,
  (where?)
• struct sockproto route_proto PF_ROUTE, /
  p626 Figure 19.26 /
• the family never changes, but the protocol is set
  each time raw_input is called
• 3rd argument of socket system call specifies the
  protocol
• The caller of raw_input sets the sp_protocol
  member of the route_proto to the protocol of the
  routing message (sa_family AF_NET)

25
19.10 rt_ifmsg Function

• if_up if_down calls rt_ifmsg to generate a
  routing socket message
• ifconfig can enable (if_up) and disable (if_down)
  by setting or clearing IFF_UP through
  SIOCSIFFLAGS command
• p123 Figure 4.30
• if no socket listener, returns
• Build message in mbuf chain
• rt_addrinfo set to 0
• rt_msg2 build an appropriate message in an mbuf
  chain
• all socket address pointers in rt_addrinfo are
  null, only fixed-length if_msghdr becomes the
  routing message, no addresses
• Finish building message
• copies the interfaces index, flags, if_data in
  the mbuf
• set ifm_addrs bitmask 0
• Set protocol of message, call raw_input
• protocol of the routing message 0, because it
  can apply to all protocol suites
• raw_input delivers the message (a message about
  an interface, not about some specific
  destination) to the appropriate listeners

26
19.11 rt_newaddrmsg Function

• rtinit calls rt_newaddrmsg with RTM_ADD
  orRTM_DELETE which adds or deletes an address to
  an interface
• if no routing socket listeners, returns
• Generate two routing messages
• RTM_ADD case 1st RTM_NEWADDR 2nd RTM_ADD
• RTM_DELETE case 1st RTM_DELETE 2nd RTM_DELADDR
• RTM_RTM_NEWADDR RTM_DELADDR built from an
  ifa_msghdr (for address)
• RTM_RTM_ADD RTM_DELETE built from an rt_msghdr
  (for inteface)
• Generate message with up to four addresses
• 4 socket address
• ifaddr, ifpaddr, netmask, brdaddr reference
  elements in the rti_info array
• rt_msg1 builds thee appropriate message in an
  mbuf chains
• sa points to the ifa_addr, the family of this
  socket address becones the protocol of the
  routing message
• creates an rt_msghdr message with the information
  about the routing entry that was added or deleted
• Build message
• sets rti_info array sets (netmask, dst, gate, in
  rtsock.c) to rt_mask, rt_key, rt_gateway
• sets sa
• rt_msg1 builds a message in an mbuf chain
• fills additional fields including the bitmask set
  by rt_msg1
• Set protocol of message, call raw_input

27
19.12 rt_msg1 Function

• 3 functions (rt_missmsg, rt_ifmsg, rt_newaddrmsg)
  calls rt_msg1 to build a routing message(p624
  Figure 19.21, p 626 Figure 19.25) in mbuf, then
  calls raw_input to append the mbuf chain to
  sockets receive buffer
• from rt_msghdr and rt_addrinfo (p624 Figure
  19.22)
• Get mbuf and determine fixed size of message
• m mbuf
• len the length of the fixed-size message
• p570 Figure 18.9, 2 ifa_msghdr(RTM_DELADDR,
  RTM_NEWADDR), 1 if_msghdr(RTM_IFINFO), 9
  rt_msghdr
• Verify structure fits in mbuf
• mbuf is referenced by pointer
• the largest is if_msghdr 84 bytes
• MHLEN is 100
• Initialize mbuf packet header and zero structure
• initializes and bzeroes rtm
• Copy socket address structure into mbuf chain
• RTAX_MAX8
• generate RTA_xxx bitmask (p623 Figure 19.19),
• the value 1 is left shifted by the RTAX_xxx
  index, then logically ORed into the rti_addrs
  member
• ROUNDUP socket address, then copy socket address
  (all null pointers) into the mbuf by m_copyback
• Store length, version, and type
• stores the first 3 memners

28
19.13 rt_msg2 Function

• build a routing message in a memory buffer
  (different from rt_msg1)
• from route_output to process the RTM_GET command
  (p651 Figure 20.7 final argument is null)
• from sysctl_dumpentry (p641 Figure 19.39) and
  sysctl_iflist (p643 Figure 19.40) to process a
  sysctl system (final argument is nonnull)
• has an argument to a walkarg
• p632 Figure 19.31
• resulting message in memory buffer (pointed by
  cp)
• cp start of the buffer
• The caller sets cp null and claas rt_msg2, then
  rt_msg2 returns only the length ? route_output
  (p651 Figure 20.7) uses this feature and calls
  rt_msg2 twice
• w
• null when called by route_out
• nonnull when called by sysctl
• Determine size of structure
• based on the message type
• Copy socket address structures
• sets rti_addr bitmask
• copies socket address (if cp is nonnull)
• updates the length
• loops twice for handles walkarg
• whenever w is nonnull, cp is nulls

29
19.13 rt_msg2 Function

• Check if data to be stored
• increments w_neede
• if users buffer size 500, w_needed -500
• as long as w_neede is negative, there is room in
  buffer
• if w_where (a pointer to the buffer) null, the
  process doesnt want the result, just want the
  size of the result, no need for rt_msg2 to malloc
  a buffer, so the process can allocate a buffer
  and call sysctl again
• 5 scenarios p635 Figure 19.34
• Allocate buffer first time or if message length
  increases
• w_tmemsize is initialized to 0 by sysctl_rtable,
  if w_rmem allocated, w_tmemsize is determined, if
  the size of the result increases, allocates a new
  larger buffer
• Go around again and store result
• if w_tmem is nonnull, sets cp to point w_tmem,
  sets second_time 1, and jumps to again
• if w_tmem null, sets the pointer to the buffer
  null
• Store length, version, and type
• if cp is nonnull, the first 3 elements are
  stored, returns the length of the message

30
19.14 sysctl_rtable Function

• handles the sysctl on a routing socket called by
  net_sysctl (p571 Figure 18.11)s
• Figure 19.35 Example of sysctl with routing table
  (from arp example)
• mib0, mib1, mib2, mib3 cause the kernel
  to call sysctl_rtable
• 3 mib4 operations
• NET_RT_DUMP
• returns the routing table (family mib3, 0 all)
• 2,3,4 socket address per message
• rt_key, r_gateway, rt_netmask, rt_genmask (might
  be null)
• NET_RT_FLAGS
• mib5 RTF_xxx flag (p580 Figure 18.25) returns
  only the entries with this flag set
• NET_RT_IFLIST
• return information on all the configured
  interfaces on the ifnet
• if mib5 specified, returns the only the
  interfaces with this flag set
• returns information
• 1 RTM_IFINFO message for each interface
• RTM_NEWADDR message for each ifaddr on the
  interfaces if_addrlist
• the amount of information returned can vary,
  depending the number of routing table entries or
  the number of interfaces
• The 1st sysctl call sets the 3rd argument of
  sysctl null, dont return any data, just return
  the length (4th argument)
• Then call malloc

31
19.14 sysctl_rtable Function

• p638 Figure 19.37 Functions that support the
  sysctl system call for routing sockets
• p639 Figure 19.38 sysctl_rtable function process
  sysctl system call requests
• Validate arguments
• new should be null
• namelen must be 3
• name0 mib3 the address family
• name1 mib4 the operation (NET_RT_xxx, p636)
• name2, mib5 the flags
• Initialize walkarg structure
• sets walkarg 0
• initialize the members
• w_where the pointer in the buffer
• w_given the size of the buffer
• w_needed negative
• w_opname1 / mib4 the operation
  (NET_RT_xxx, p636) /
• w_argname2 / mib5 the flags /
• Dump routing table
• rnh_walktree walk through the tree
• 1st argument the pointer to the touring table

32
19.15 sysctl_dumpentry Function

• sysctl_rtable calls rn_walktree which in turn
  calles sysctl_dumpentry
• 1st argument points to a radix_node (pointer to
  a rtentry)
• 2nd argument points to walkarg Iinitialized by
  sysctl_rtable)
• Check flags of routing table entry
• skips if rt_flags is not equal to the flag
  (mib5, i.e. p636 RTF_LLINFO in ARP)
• Form routing message
• copies 4 pointers (dst, gate, netmask, genmask)
  from the routing tanle entry (dst, gate are
  always nonnull)
• rt_msg2 forms an RTM_GET message
• Copy message back to process
• rt_msg2 allocates a buffer
• forms the remainder of the routing message
  pointed to by w_tmem
• copies the message back to the process by copyout

33
19.16 sysctl_iflist Function

• sysctl_rtable directly calls sysctl_iflist
• for loop iteratates through each interface
  (ifnet)
• while loop proceeds through the limked list of
  ifaddr
• Check interface index
• selects if if_index equals to the nonzero flags
  (mib5)
• Build routing message
• rt_msg2 with RTM_IFINFO returns only the socket
  address ifpaddr
• sets ifpaddr to 0 (since the same info is used
  for generating the subsequent RTM_NEWADDR)
• Copy message back to process
• fills the remainder of if_msghdr, copyout it to
  the process, incremets w_where
• Iterate through address structures, check address
  family
• processes each ifaddr for the interface and
  select only the interface address of given family
  (if nonzero address family (mib3)
• Build routing message
• rt_msg2 with RTM_NEWADDR returns upto 3 socket
  addresses (ifaaddr, netmask, brdaddr)
• Copy message back to process
• fills in the remaindr of ifa_msghdr, copyout to
  the process, increments w_where
• set ifaaddr, netmask, brdaddr to 0 (since the
  same array is used for the next interface message)

34
Utilities for Network Administration

• netstat rn
• search /etc/networks
• route
• in /etc/rc2.d/S69inet
• /etc/defaultrouter
• ifconfig a
• ping sv csmail
• snoop csblade csmail
• snoop tr p 520