Figure 1: After alloc_skb
|
Документ взят из кэша поисковой машины. Адрес
оригинального документа
: http://www.arcetri.astro.it/irlab/library/linux/khg/HyperNews/get/net/net-intro.html
Дата изменения: Thu Mar 23 12:38:18 2000 Дата индексирования: Tue Sep 14 23:06:24 2010 Кодировка: |
Reprinted with permission of Linux Journal, from issue 29, September 1996. Some changes have been made to accomodate the web. This article was originally written for the Kernel Korner column. The Kernel Korner series has included many other articles of interest to Linux kernel hackers, as well.
The Linux operating system implements the industry-standard Berkeley socket API, which has its origins in the BSD unix developments (4.2/4.3/4.4 BSD). In this article, we will look at the way the memory management and buffering is implemented for network layers and network device drivers under the existing Linux kernel, as well as explain how and why some things have changed over time.
The networking layer tries to be fairly object-oriented in its design, as indeed is much of the Linux kernel. The core structure of the networking code goes back to the initial networking and socket implementations by Ross Biro and Orest Zborowski respectively. The key objects are:
The primary goal of the sk_buff routines is to provide a consistent and efficient buffer handling method for all of the network layers, and by being consistent to make it possible to provide higher level sk_buff and socket handling facilities to all the protocols.
An sk_buff is a control structure with a block of memory attached. There are two primary sets of functions provided in the sk_buff library. Firstly routines to manipulate doubly linked lists of sk_buffs, secondly functions for controlling the attached memory. The buffers are held on linked lists optimised for the common network operations of append to end and remove from start. As so much of the networking functionality occurs during interrupts these routines are written to be atomic. The small extra overhead this causes is well worth the pain it saves in bug hunting.
We use the list operations to manage groups of packets as they arrive from the network, and as we send them to the physical interfaces. We use the memory manipulation routines for handling the contents of packets in a standardised and efficient manner.
At its most basic level, a list of buffers is managed using functions like this:
void append_frame(char *buf, int len)
{
struct sk_buff *skb=alloc_skb(len, GFP_ATOMIC);
if(skb==NULL)
my_dropped++;
else
{
skb_put(skb,len);
memcpy(skb->data,data,len);
skb_append(&my_list, skb);
}
}
void process_queue(void)
{
struct sk_buff *skb;
while((skb=skb_dequeue(&my_list))!=NULL)
{
process_data(skb);
kfree_skb(skb, FREE_READ);
}
}
These two fairly simplistic pieces of code actually demonstrate
the receive packet mechanism quite accurately. The
append_frame() function is similar to the code called
from an interrupt by a device driver receiving a packet, and
process_frame() is similar to the code called to feed
data into the protocols. If you go and look in net/core/dev.c
at netif_rx() and net_bh(), you will see that
they manage buffers similarly. They are far more complex, as
they have to feed packets to the right protocol and manage flow
control, but the basic operations are the same. This is just as
true if you look at buffers going from the protocol code to a
user application.
The example also shows the use of one of the data control functions, skb_put(). Here it is used to reserve space in the buffer for the data we wish to pass down.
Let's look at append_frame(). The alloc_skb() fucntion obtains a buffer of len bytes (Figure 1), which consists of:
Immediately after a buffer has been allocated, all the available room is at the end. A further function named skb_reserve() (Figure 2) can be called before data is added allows you to specify that some of the room should be at the beginning. Thus, many sending routines start with something like:
skb=alloc_skb(len+headspace, GFP_KERNEL);
skb_reserve(skb, headspace);
skb_put(skb,len);
memcpy_fromfs(skb->data,data,len);
pass_to_m_protocol(skb);
In systems such as BSD unix you don't need to know in advance how much space you will need as it uses chains of small buffers (mbufs) for its network buffers. Linux chooses to use linear buffers and save space in advance (often wasting a few bytes to allow for the worst case) because linear buffers make many other things much faster.
Now to return to the list functions. Linux provides the following operations:
Figure 3: An sk_buff containing data
Figure 4: After skb_put has been called on the buffer
Figure 5: After an skb_push has occured on the previous buffer
Figure 6: Network device data flow
The semantics of allocating and queueing buffers for sockets also involve flow control rules and for sending a whole list of interactions with signals and optional settings such as non blocking. Two routines are designed to make this easy for most protocols.
The sock_queue_rcv_skb() function is used to handle incoming data flow control and is normally used in the form:
sk=my_find_socket(whatever);
if(sock_queue_rcv_skb(sk,skb)==-1)
{
myproto_stats.dropped++;
kfree_skb(skb,FREE_READ);
return;
}
This function uses the socket read queue counters to prevent
vast amounts of data being queued to a socket. After a limit is
hit, data is discarded. It is up to the application to read
fast enough, or as in TCP, for the protocol to do flow control
over the network. TCP actually tells the sending machine to
shut up when it can no longer queue data.
On the sending side, sock_alloc_send_skb() handles signal handling, the non blocking flag, and all the semantics of blocking until there is space in the send queue so you cannot tie up all of memory with data queued for a slow interface. Many protocol send routines have this function doing almost all the work:
skb=sock_alloc_send_skb(sk,....)
if(skb==NULL)
return -err;
skb->sk=sk;
skb_reserve(skb, headroom);
skb_put(skb,len);
memcpy(skb->data, data, len);
protocol_do_something(skb);
Most of this we have met before. The very important line is skb->sk=sk. The sock_alloc_send_skb() has charged the memory for the buffer to the socket. By setting skb->sk we tell the kernel that whoever does a kfree_skb() on the buffer should cause the socket to be credited the memory for the buffer. Thus when a device has sent a buffer and frees it the user will be able to send more.
All Linux network devices follow the same interface although many functions available in that interface will not be needed for all devices. An object oriented mentality is used and each device is an object with a series of methods that are filled into a structure. Each method is called with the device itself as the first argument. This is done to get around the lack of the C++ concept of this within the C language.
The file drivers/net/skeleton.c contains the skeleton of a network device driver. View or print a copy from a recent kernel and follow along throughout the rest of the article.
Each network device deals entirely in the transmission of network buffers from the protocols to the physical media, and in receiving and decoding the responses the hardware generates. Incoming frames are turned into network buffers, identified by protocol and delivered to netif_rx(). This function then passes the frames off to the protocol layer for further processing.
Each device provides a set of additional methods for the handling of stopping, starting, control and physical encapsulation of packets. These and all the other control information are collected together in the device structures that are used to manage each device.
All Linux network devices have a unique name. This is not in any way related to the file system names devices may have, and indeed network devices do not normally have a filesystem representation, although you may create a device which is tied to device drivers. Traditionally the name indicates only the type of a device rather than its maker. Multiple devices of the same type are numbered upwards from 0. Thus ethernet devices are known as ``eth0'', ``eth1'', ``eth2'' etc. The naming scheme is important as it allows users to write programs or system configuration in terms of ``an ethernet card'' rather than worrying about the manufacturer of the board and forcing reconfiguration if a board is changed.
The following names are currently used for generic devices:
If possible, a new device should pick a name that reflects existing practice. When you are adding a whole new physical layer type you should look for other people working on such a project and use a common naming scheme.
Certain physical layers present multiple logical interfaces over one media. Both ATM and Frame Relay have this property, as does multi-drop KISS in the amateur radio environment. Under such circumstances a driver needs to exist for each active channel. The Linux networking code is structured in such a way as to make this managable without excessive additional code, and the name registration scheme allows you to create and remove interfaces almost at will as channels come into and out of existance. The proposed convention for such names is still under some discussion, as the simple scheme of ``sl0a'', ``sl0b'', "sl0c" works for basic devices like multidrop KISS, but does not cope with multiple frame relay connections where a virtual channel may be moved across physical boards.
Each device is created by filling in a struct device object and passing it to the register_netdev(struct device *) call. This links your device structure into the kernel network device tables. As the structure you pass in is used by the kernel, you must not free this until you have unloaded the device with void unregister_netdev(struct device *) calls. These calls are normally done at boot time, or module load and unload.
The kernel will not object if you create multiple devices with the same name, it will break. Therefore, if your driver is a loadable module you should use the struct device *dev_get(const char *name) call to ensure the name is not already in use. If it is in use, you should fail or pick another name. You may not use unregister_netdev() to unregister the other device with the name if you discover a clash!
A typical code sequence for registration is:
int register_my_device(void)
{
int i=0;
for(i=0;i<100;i++)
{
sprintf(mydevice.name,"mydev%d",i);
if(dev_get(mydevice.name)==NULL)
{
if(register_netdev(&mydevice)!=0)
return -EIO;
return 0;
}
}
printk("100 mydevs loaded. Unable to load more.\n");
return -ENFILE;
}
All the generic information and methods for each network device are kept in the device structure. To create a device you need to fill most of these in. This section covers how they should be set up.
The next block of parameters are used to maintain the location of a device within the device address spaces of the architecture. The irq field holds the interrupt (IRQ) the device is using. This is normally set at boot, or by the initialization function. If an interrupt is not used, not currently known, or not assigned, the value zero should be used. The interrupt can be set in a variety of fashions. The auto-irq facilities of the kernel may be used to probe for the device interrupt, or the interrupt may be set when loading the network module. Network drivers normally use a global int called irq for this so that users can load the module with insmod mydevice irq=5 style commands. Finally, the IRQ may be set dynamically from the ifconfig command. This causes a call to your device that will be discussed later on.
The base_addr field is the base I/O space address the device resides at. If the device uses no I/O locations or is running on a system with no I/O space concept this field should be zero. When this is user settable, it is normally set by a global variable called io. The interface I/O address may also be set with ifconfig.
Two hardware shared memory ranges are defined for things like ISA bus shared memory ethernet cards. For current purposes, the rmem_start and rmem_end fields are obsolete and should be loaded with 0. The mem_start and mem_end addresses should be loaded with the start and end of the shared memory block used by this device. If no shared memory block is used, then the value 0 should be stored. Those devices that allow the user to specify this parameter use a global variable called mem to set the memory base, and set the mem_end appropriately themselves.
The dma variable holds the DMA channel in use by the device. Linux allows DMA (like interrupts) to be automatically probed. If no DMA channel is used, or the DMA channel is not yet set, the value 0 is used. This may have to change, since the latest PC boards allow ISA bus DMA channel 0 to be used by hardware boards and do not just tie it to memory refresh. If the user can set the DMA channel the global variable dma is used.
It is important to realise that the physical information is provided for control and user viewing (as well as the driver's internal functions), and does not register these areas to prevent them being reused. Thus the device driver must also allocate and register the I/O, DMA and interrupt lines it wishes to use, using the same kernel functions as any other device driver. [See the recent Kernel Korner articles on writing a character device driver in issues 23, 24, 25, 26, and 28 of Linux Journal.]
The if_port field holds the physical media type for multi-media devices such as combo ethernet boards.
In order for the network protocol layers to perform in a sensible manner, the device has to provide a set of capability flags and variables. These are also maintained in the device structure.
The mtu is the largest payload that can be sent over this interface (that is, the largest packet size not including any bottom layer headers that the device itself will provide). This is used by the protocol layers such as IP to select suitable packet sizes to send. There are minimums imposed by each protocol. A device is not usable for IPX without a 576 byte frame size or higher. IP needs at least 72 bytes, and does not perform sensibly below about 200 bytes. It is up to the protocol layers to decide whether to co-operate with your device.
The family is always set to AF_INET and indicates the protocol family the device is using. Linux allows a device to be using multiple protocol families at once, and maintains this information solely to look more like the standard BSD networking API.
The interface hardware type (type) field is taken from a table
of physical media types. The values used by the ARP protocol
(see RFC1700) are used for those media supporting ARP and
additional values are assigned for other physical layers. New
values are added when neccessary both to the kernel and to
net-tools which is the package containing programs like
ifconfig that need to be able to decode this field. The
fields defined as of Linux pre2.0.5 are:
From RFC1700:
Those interfaces marked unused are defined types but without any current support on the existing net-tools. The Linux kernel provides additional generic support routines for devices using ethernet and token ring.
The pa_addr field is used to hold the IP address when the interface is up. Interfaces should start down with this variable clear. pa_brdaddr is used to hold the configured broadcast address, pa_dstaddr the target of a point to point link and pa_mask the IP netmask of the interface. All of these can be initialised to zero. The pa_alen field holds the length of an address (in our case an IP address), this should be initialised to 4.
The hard_header_len is the number of bytes the device desires at the start of a network buffer it is passed. It does not have to be the number of bytes of physical header that will be added, although this is normal. A device can use this to provide itself a scratchpad at the start of each buffer.
In the 1.2.x series kernels, the skb->data pointer will point to the buffer start and you must avoid sending your scratchpad yourself. This also means for devices with variable length headers you will need to allocate max_size+1 bytes and keep a length byte at the start so you know where the header really begins (the header should be contiguous with the data). Linux 1.3.x makes life much simpler and ensures you will have at least as much room as you asked free at the start of the buffer. It is up to you to use skb_push() appropriately as was discussed in the section on networking buffers.
The physical media addresses (if any) are maintained in dev_addr and broadcast respectively. These are byte arrays and addresses smaller than the size of the array are stored starting from the left. The addr_len field is used to hold the length of a hardware address. With many media there is no hardware address, and this should be set to zero. For some other interfaces the address must be set by a user program. The ifconfig tool permits the setting of an interface hardware address. In this case it need not be set initially, but the open code should take care not to allow a device to start transmitting without an address being set.
A set of flags are used to maintain the interface properties. Some of these are ``compatibility'' items and as such not directly useful. The flags are:
Packets are queued for an interface by the kernel protocol code. Within each device, buffs[] is an array of packet queues for each kernel priority level. These are maintained entirely by the kernel code, but must be initialised by the device itself on boot up. The intialisation code used is:
int ct=0;
while(ct<DEV_NUMBUFFS)
{
skb_queue_head_init(&dev->buffs[ct]);
ct++;
}
All other fields should be initialised to 0.
The device gets to select the queue length it wants by setting the field dev->tx_queue_len to the maximum number of frames the kernel should queue for the device. Typically this is around 100 for ethernet and 10 for serial lines. A device can modify this dynamically, although its effect will lag the change slightly.
Each network device has to provide a set of actual functions (methods) for the basic low level operations. It should also provide a set of support functions that interface the protocol layer to the protocol requirements of the link layer it is providing.
The init method is called when the device is initialised and registered with the system. It should perform any low level verification and checking needed, and return an error code if the device is not present, areas cannot be registered or it is otherwise unable to proceed. If the init method returns an error the register_netdev() call returns the error code and the device is not created.
All devices must provide a transmit function. It is possible for a device to exist that cannot transmit. In this case the device needs a transmit function that simply frees the buffer it is passed. The dummy device has exactly this functionality on transmit.
The dev->hard_start_xmit() function is called and provides the driver with its own device pointer and network buffer (an sk_buff) to transmit. If your device is unable to accept the buffer, it should return 1 and set dev->tbusy to a non-zero value. This will queue the buffer and it may be retried again later, although there is no guarantee that the buffer will be retried. If the protocol layer decides to free the buffer the driver has rejected, then it will not be offered back to the device. If the device knows the buffer cannot be transmitted in the near future, for example due to bad congestion, it can call dev_kfree_skb() to dump the buffer and return 0 indicating the buffer is processed.
If there is room the buffer should be processed. The buffer handed down already contains all the headers, including link layer headers, neccessary and need only be actually loaded into the hardware for transmission. In addition, the buffer is locked. This means that the device driver has absolute ownership of the buffer until it chooses to relinquish it. The contents of an sk_buff remain read-only, except that you are guaranteed that the next/previous pointers are free so you can use the sk_buff list primitives to build internal chains of buffers.
When the buffer h