CS 498 Lecture 6 Network Devices

1
CS 498 Lecture 6Network Devices

• Jennifer Hou
• Department of Computer Science
• University of Illinois at Urbana-Champaign
• Reading Chapters 5, The Linux Networking
  Architecture Design and Implementation of
  Network Protocols in the Linux Kernel

2
APIs Needed in Project 3

• Chapter 6 Flow of Time, Linux Device Driver

3
timer_list structure

• (Defined in ltlinux/timer.hgt)
• struct timer_list
• struct list_head list
• unsigned long expires
• unsigned long data
• void (function)(unsigned long)
• jiffies the number of ticks since the last boot
  defined in ltlinux/sched.hgt
• HZ the number of ticks per second defined in
  ltlinux/param.hgt machine dependent.
• timer?function will run when jiffies is equal to
  or greater than timer?expires.

4
Functions Used for Timers

• void init_timer(struct timer_list timer)
• Initializes the timer structure.
• void add_timer(struct timer_list timer)
• Inserts a timer into the global list of active
  timers.
• mod_timer(struct timer_list timer, unsigned long
  expires)
• Changes the time at which a timer expires.
• int del_timer(struct timer_list timer)
• Removes a timer from the list before it expires.
• int del_timer_sync(struct timer_list timer)
• Same as del_timer, and guarantees that when the
  function returns, the timer function is not
  running on any CPU.

5
Sleeping and Awakening

• A wake queue is a queue of processes that are
  waiting for an event
• wait_queue_head_t my_queue
• init_waitqueue_head(my_queue)
• Once the wait queue is declared and initialized,
  a process can be put into sleep on this queue.

6
Sleeping and Awakening

• sleep_on(wait_queue_head_t queue)
• Puts the process to sleep on this queue not
  interruptible.
• interruptible_sleep_on(wait_queue_head_t queue)
• sleep_on_timeout(wait_queue_head_t queue, long
  timeout)
• Puts the process into sleep on this queue, but
  the sleep will not be longer than timeout.
• interruptible_sleep_on_timeout(wait_queue_head_t
  queue, long timeout)

7
Sleeping and Awakening

• wake_up(wait_queue_head_t queue)
• Wakes up all the processes that are waiting on
  this event queue
• wake_up_interruptible(wait_queue_head_t queue)
• Wakes up on the processes that are in
  interruptible sleeps.
• wake_up_sync(wait_queue_head_t queue)
• wake_up_interruptible_sync(wait_queue_head_t
  queue)

8
Network Devices

• An interface between software-based protocols and
  network adapters.
• Two major functions
• Abstract from the technical properties of
  network adapters (that implement different
  layer-1 and layer-2 protocols and are
  manufactured by different vendors).
• Provide a uniform interface for access by
  protocol instances.

9
Network Device Interface
Higher Protocol Instances
Network devices (adapter-independent)
dev.c
netif_rx
dev_queue_xmit
dev_open
dev_close
Network devices interface
net_device
dev-gthard_start_xmit
dev-gtopen
dev-gtstop
Abstraction from Adapter specifics
driver.c
net_start_xmit
net_tx
net_rx
Network driver (adapter-specific)
net_open
net_stop
skb
skb
net_interrupt
skb
10
net_device structure
11
net_device structure

• Represents a general interface between higher
  protocol instances and the hardware used.
• Contains a number of function pointers that
  points to hardware-specific methods of the
  driver.
• Are grouped into three categories
• General fields
• Hardware-specific fields
• Network layer fields
• Device driver methods

12
General Fields in net_device

• name name of the network device, e.g.,
  eth0-eth4, lo (loopback device)
• next points to the next net_device in the list
  of network devices (that starts with dev_base).
• owner a pointer to the module structure of the
  module that creates net_device.
• ifindex a second identifier to a network device.
• When a new network device is created,
  dev_get_index() assigns a new unused index to the
  device.

13
General Fields in net_device

• state state of the device
• LINK_STATE_START the network adapter has been
  opened with dev?open()
• netif_running(dev) can test this flag.
• LINK_STATE_XOFF the transmit buffers of the
  network adapter are all busy.
• netif_queue_stopped(dev) can test this flag.
• netif_stop_queue(dev) sets the LINK_STATE_XOFF
  flag, and no packet can be passed onto the
  adapter.
• netif_start_queue(dev) deletes the
  LINK_STATE_XOFF flag, and resumes accepting
  packets from the kernel.
• Netif_wake_queue(dev) resumes passing packets,
  and trigger the NET_TX software interrupt.

14
General Fields in net_device

• trans_start the time (in jiffies) when the
  transmission of a packet started.
• last_rx the time (in jiffies) when the last
  packet arrived.
• priv a pointer to the private data of a network
  device.
• qdisk a structure of they type Qdisk, and
  governs the serving discipline of the network
  device.
• refcnt the number of references to the network
  device.
• xmit_lock, xmit_lock_owner are used for mutual
  exclusion on the transmit queue.

15
Hardware-Specific Fields

• rmem_end, rmem_start, mem_end, mem_start specify
  the common memory space that the network adapter
  and the kernel share.
• (mem_start-mem_end) designates the buffers for
  packets to be sent (rmem_start-rmem_end)
  designates the location for received packets.
• base_addr I/O basic address set in the drivers
  probing routine during the search for a device.
• Irq the of the interrupt of a network adapter
  set in the drivers probing routine.
• dma the number of the DMA channel.

16
Hardware-Specific Fields

• hard_header_length specifies the length of the
  layer-2 packet header.
• mtu maximum transfer unit in a layer-2 frame.
• tx_queue_len specifies the maximum length of the
  output queue of the network device.
• type specifies the hardware type of the network
  adapter, e.g., ARPHRD_ETHER, ARPHRD_IEEE802,
  ARPHRD_LOOPBACK

17
Hardware-Specific Fields

• addr_len, dev_addrMAX_ADDR_LEN,
  broadcastMAX_ADDR_LEN
• dev_mc_list points to a linear list that
  contains multicast layer-2 addresses.
• set_multicast_list() is used to pass the list to
  the network adapter.
• mc_count the number of addresses in dev_mc_list.

18
Hardware-Specific Fields

• watchdog_timer is initialized when a network
  device starts.
• The handling routine dev_watchdog() is called
  after watchdog_timeo time units, and checks
  whether or not watchdog_timeo time units have
  passed since the last transmission of a packet.
• Yes, tx_timeout() of the driver is called.
• No, the watchdog timer is restarted.

19
Network Layer Fields

• ip_ptr, ip6_ptr point to the information of
  layer 3 protocols that use the network device,
  e.g., ip_ptr points to a structure of type
  in_device (that keeps track of a list of IP
  addresses of the network device, a list of IP
  mluticast groups, and the parameters for the ARP
  protocol).
• family designates the address family of the
  network device, e.g., AF_INET
• pa_alen the length of the addresses, e.g., 4
  bytes.
• pa_addr, pa_braddr, pa_mask the address, the
  broadcast address, and the network mask.
• pa_dstaddr specifies the address of the other
  partner in a point-to-point connection.

20
Network Layer Fields

• flags properties of the network device
• IFF_UP the device is on.
• IFF_ARP the device supports ARP
• IFF_BROADCAST the device is broadcast-enabled.
• IFF_LOOKBACK the device is a loopback network
  device.
• IFF_POINTTOPOINT this is a point-to-point
  connection.
• IFF_PROMISC this flag switches the promiscuous
  mode on.
• IFF_MULTICAST activates the receipt of multicast
  packets.

21
Device-Driver Methods

• init() is used to initialize network devices.
• uninit() is called when the network device is
  unregistered (unregister_netdevice()).
• destructor() is called when dev?refcnt0.
• open() opens a named network device. ifconfig
  calls dev_open() which in turn calls dev?open()
• stop() terminates the activity of a network
  adapter.
• hard_start_xmit() transmit a socket buffer over
  the network device.

22
Device Driver Methods

• get_stats() gets statistics and information
  about the network device.
• get_wireless_stats() returns additional
  information for wireless network adapters.
• set_multicast_list() passes the list of
  multicast MAC addresses to the network adapter.
• tx_timeout() is called when no acknowledgement
  for the packet is received after watchdog_timeo
  time units.
• set_config() is used to change the configuration
  of a network adapter.
• do_ioctl() is used by higher protocols to pass
  adapter-specific ioctl() commands to the network
  adapter.

23
Device Driver Methods

• hard_header() creates a layer-2 packet header.
• rebuild_header() is called only when the hard
  header cache contains wrong information.
• hard_header_cache() fills a layer-2 packet
  header in the hard header cache with passed data.
  Subsequent transmission processes can access a
  layer-2 packet header.
• header_cache_update() changes the layer-2
  destination address in a stored layer-2 packet
  header in the hard header cache.

24
Device Driver Methods

• hard_header_parse() reads the layer-2 sender
  address from the packet data space of a socket
  buffer and copies it to the passed address,
  haddr.
• set_mac_address() Is used to change the layer-2
  address of a network adapter.
• change_mtu() changes the MTU of a network
  adapter.

25
Managing Network Device
26
dev_base List

• All the network devices in the Linux kernel are
  connected in a linear list.
• /proc/net/dev or ifconfig a gives the list of
  currently registered devices.
• When the system is booted or when the driver
  module is loaded, an appropriate adapter is
  found. Then a net_device structure is created,
  initialized and added to the list with
  register_netdevice().

27
List of Network Devices
name eth1
dev_base
state
net_device
next
priv
net_device
...
name eth0
Hardware
state
local2private driver structure
next
MAC layer
priv
...
Network layer
Hardware
open
MAC layer
adapter2_open()
stop
adapter2_stop()
hard_start_xmit
Network layer
...
adapter_start_xmit()
open
adapter_open()
stop
adapter_stop()
hard_start_xmit
adapter_start_xmit()
...
local1private driver structure
28
Functions for Managing Network Devices

• init_netdev(dev,sizeof_priv,mask,setup)
• If dev is null, call net_alloc_dev() to create a
  new net_device structure.
• If dev?name is empty, call dev_alloc_name() to
  allocate a name.
• mask can be used to specify a prefix, e.g.,
  testd produces test0, test1, etc.
• The list dev_boot_setup in the method
  netdev_boot_setup_check() is searched for an
  entry with a matched name. If there is such an
  entry, then the hardware specific fields (e.g.,
  mem_start, mem_end, base_addr) are filled out.

29
Functions for Managing Network Devices

• init_netdev(dev,sizeof_priv,mask,setup)
• The function which setup points to is used to
  initialize layer-2 fields. For example, setup may
  points to ether_setup(dev).
• The net_device is added to dev_base by
  register_netdevice().
• init_netdev() returns with a pointer to the
  net_device structure.

30
init_etherdev()

• struct net_device init_etherdev(struct
  net_device dev, int sizeof_priv)
• return init_netdev(dev, sizeof_priv, "ethd",
  ether_setup)

31
ether_setup

• void ether_setup(struct net_device dev)
• / Fill in the fields of the device structure
  with ethernet-generic values. /
• dev-gtchange_mtu eth_change_mtu
• dev-gthard_header eth_header
• dev-gtrebuild_header eth_rebuild_header
• dev-gtset_mac_address eth_mac_addr
• dev-gthard_header_cache eth_header_cache
• dev-gtheader_cache_update eth_header_cache_upd
  ate
• dev-gthard_header_parse eth_header_parse
• dev-gttype ARPHRD_ETHER
• dev-gthard_header_len ETH_HLEN
• dev-gtmtu 1500 / eth_mtu /
• dev-gtaddr_len ETH_ALEN
• dev-gttx_queue_len 1000 / Ethernet wants
  good queues /
• memset(dev-gtbroadcast,0xFF, ETH_ALEN)
• dev-gtflags IFF_BROADCASTIFF_MULTICAST

32
register_netdevice()

• If dev?init() exists, the network adapter is
  first searched and initialized with the function
  init().
• If a net_device with the same name is already in
  the list dev_base, return an error message
  otherwise, add the dev_device, dev, to the list.
• Sets dev?state to LINK_STATE_PRESENT

33
register_netdevice()

• dev_init_scheduler() (i) sets the scheduling
  discipline to FIFO (ii) initializes the timer
  dev?watchdog_timer with dev_watchdog_init(). The
  timer is not started until dev_open() is invoked.
• The network device is assigned a new index
  (dev?ifindex).
• notifier_call_chain(netdev_chain,
  NETDEV_REGISTER, dev) calls all the registered
  methods in netdev_chain

34
unregister_netdevice(dev)

• Remove dev from the dev_base list.
• If dev?flagIFF_UP, dev?close() is called to
  close the device.
• The queueing discipline is released with
  dev_shutdown().
• All registered functions in the notification
  chain netdev_chain are informed about the
  NETDEV_UNREGISTER event.

35
dev_open()

• If (dev?flags IFF_UP) (!netif_device_present(
  dev)), returns with an error message.
• Invokes the open() function of the network
  driver.
• After the initialization,
• dev? flags IFF_UP
• dev? state LINK_STATE_START
• The multicast state is activated with
  dev_mc_upload()
• The queueing and scheduling disciplines are
  activated with dev_activate(dev).
• dev(watchdog_up(dev) starts the timer. The timer
  calls dev_watchdog() every dev?watchdog_timeo
  ticks to check for transmission problems.
• The notification chain netdev_chain is informed
  of the event NETDEV_UP.

36
dev_close(dev)

• dev_deactivate(dev) removes the packet scheduler
  dev?qdisc.
• The bit LINK_STATE_START in dev?state is deleted.
• dev_watchdog_down() stops the timer.
• dev?stop() deactivates the network adapter.
• notifier_call_chain(netdev_chain, NETDEV_DOWN,
  dev).
• dev?refcnt is decremented by one.

37
dev_get_()

• dev_get_by_name(name) and dev_get(name) search
  for the network device specified by name.
• dev_get_by_index(ifindex)
• dev_getbyhwaddr(type,ha) searches for the network
  device by the MAC address ha and the type.

38
Notification Chain

• notifier_call_chain(netdev_chain, EVENT, dev)
• notifier_chain_register(netdev_chain,nb)
• notifier_chain_unregister(nb)