zoukankan html css js c++ java

rust-vmm 学习

V0.1.0 feature

base knowledge:

Architecture of the Kernel-based Virtual Machine (KVM)

用rust-vmm打造未来的虚拟化架构

KVM内核文档阅读笔记

<Mastering KVM Virtualization>：第二章 KVM内部原理

Using the KVM API (org)

Example for a simple vmm (github, search pub fn test_vm())

kvm-ioctls (github)

run test:　　

cd vmm
FILE_LINE=`git grep -n "pub fn test_vm"` && echo ${FILE_LINE%:*}
cargo test test_vm

FILE_LINE=`git grep -n "pub fn test_vm"` && echo ${FILE_LINE%:*}
src/vm.rs:854
cargo test test_vm |grep "Running target/debug/deps/"
    Finished dev [unoptimized + debuginfo] target(s) in 0.16s
     Running target/debug/deps/vmm-5f7f458cb725298f

rust-gdb target/debug/deps/vmm-5f7f458cb725298f
(gdb) info functions test_vm
All functions matching regular expression "test_vm":

(gdb) help info

(gdb) info sources

(gdb) b src/vm.rs:854
No source file named src/vm.rs.
Make breakpoint pending on future shared library load? (y or [n]) n
(gdb) run --test test_vm

# note: Run with `RUST_BACKTRACE=1` environment variable to display a backtrace.

RUST_BACKTRACE=1 cargo run

vim ./cloud-hypervisor/src/main.rs
:ConqueGdbExe rust-gdb :ConqueGdb target/debug/cloud-hypervisor

(gdb) info functions main
All functions matching regular expression "main":

File cloud-hypervisor/src/main.rs:

static void cloud_hypervisor::main::h9d806d3576285a29(void);

(gdb) b cloud_hypervisor::main::h9d806d3576285a29

(gdb) run --kernel ./hypervisor-fw 
    --disk ./clear-29160-kvm.img 
    --cpus 4 
    --memory 512 
    --net "tap=,mac=,ip=,mask=" 
    --rng

(gdb) run target/debug/cloud-hypervisor 
    --kernel ./linux-cloud-hypervisor/arch/x86/boot/compressed/vmlinux.bin 
    --disk ./clear-29160-kvm.img 
    --cmdline "console=ttyS0 reboot=k panic=1 nomodules i8042.noaux i8042.nomux i8042.nopnp i8042.dumbkbd root=/dev/vda3" 
    --cpus 4 
    --memory 512 
    --net "tap=,mac=,ip=,mask=" 
    --rng

Example: debug a Virtio devices(Rng)

sudo ~/.cargo/bin/rust-gdb cloud-hypervisor/target/debug/cloud-hypervisor
(gdb) i functions Rng::new
File vm-virtio/src/rng.rs:
struct Result<vm_virtio::rng::Rng, std::io::error::Error> vm_virtio::rng::Rng::new::h79792d122294607e(struct &str);
(gdb) b vm_virtio::rng::Rng::new::h79792d122294607e
Breakpoint 1 at 0x6330e4: file vm-virtio/src/rng.rs, line 156.
(gdb) r  --kernel ./linux-cloud-hypervisor/arch/x86/boot/compressed/vmlinux.bin 
    --disk ./clear-29160-kvm.img 
    --cmdline "console=ttyS0 reboot=k panic=1 nomodules i8042.noaux i8042.nomux i8042.nopnp i8042.dumbkbd root=/dev/vda3" 
    --cpus 4 
    --memory 512 
    --net "tap=,mac=,ip=,mask=" 
    --rng
(gdb) bt
#0  vm_virtio::rng::Rng::new::h79792d122294607e (path="/dev/urandom") at vm-virtio/src/rng.rs:156
#1  0x0000555555ac81fa in vmm::vm::DeviceManager::new::hb95c517461758006 (memory=GuestMemoryMmap = {...}, allocator=0x7fffffff65e0, vm_fd=0x7fffffff59f0, 
    vm_cfg=0x7fffffff7a18, msi_capable=true, userspace_ioapic=true) at vmm/src/vm.rs:572
#2  0x0000555555acd235 in vmm::vm::Vm::new::h319ad49fd3f83723 (kvm=0x7fffffff731c, config=VmConfig = {...}) at vmm/src/vm.rs:905
#3  0x0000555555aafed1 in vmm::boot_kernel::hc06685bd1a4087c9 (config=VmConfig = {...}) at vmm/src/lib.rs:59
#4  0x00005555556d377b in cloud_hypervisor::main::hc6d3862453504aa4 () at src/main.rs:114
#5  0x00005555556d0110 in std::rt::lang_start::_$u7b$$u7b$closure$u7d$$u7d$::hb71d053f091b9e72 () at /rustc/91856ed52c58aa5ba66a015354d1cc69e9779bdf/src/libstd/rt.rs:64
#6  0x0000555555c36943 in {{closure}} () at src/libstd/rt.rs:49
#7  do_call<closure,i32> () at src/libstd/panicking.rs:297
#8  0x0000555555c3a0ca in __rust_maybe_catch_panic () at src/libpanic_unwind/lib.rs:87
#9  0x0000555555c3744d in try<i32,closure> () at src/libstd/panicking.rs:276
#10 catch_unwind<closure,i32> () at src/libstd/panic.rs:388
#11 lang_start_internal () at src/libstd/rt.rs:48
#12 0x00005555556d00e9 in std::rt::lang_start::hd3791e31e26e1c55 (main=0x5555556d2720 <cloud_hypervisor::main::hc6d3862453504aa4>, argc=14, argv=0x7fffffffe1a8)
    at /rustc/91856ed52c58aa5ba66a015354d1cc69e9779bdf/src/libstd/rt.rs:64
#13 0x00005555556d3bca in main ()
(gdb) set print pretty on

(gdb) info functions Rng.*activate
(gdb) info functions Rng.*VirtioDevice.*activate
All functions matching regular expression "Rng.*VirtioDevice.*activate":

File vm-virtio/src/rng.rs:
struct Result<(), vm_virtio::ActivateError> _$LT$vm_virtio..rng..Rng$u20$as$u20$vm_virtio..device..VirtioDevice$GT$::activate::hdedcc2490e8a5155(struct Rng *, 
    struct GuestMemoryMmap, struct Arc<alloc::boxed::Box<Fn<(&vm_virtio::queue::Queue)>>>, struct Arc<core::sync::atomic::AtomicUsize>, 
    struct Vec<vm_virtio::queue::Queue>, struct Vec<vmm_sys_util::eventfd::EventFd>);
(gdb) b _$LT$vm_virtio..rng..Rng$u20$as$u20$vm_virtio..device..VirtioDevice$GT$::activate::hdedcc2490e8a5155

(gdb) b vm-virtio/src/rng.rs:236
(gdb) c
Thread 2 "cloud-hyperviso" hit Breakpoint 2, _$LT$vm_virtio..rng..Rng$u20$as$u20$vm_virtio..device..VirtioDevice$GT$::activate::hdedcc2490e8a5155 (self=0x555555f71f10, 
    mem=GuestMemoryMmap = {...}, interrupt_cb=Arc<alloc::boxed::Box<Fn<(&vm_virtio::queue::Queue)>>> = {...}, status=Arc<core::sync::atomic::AtomicUsize> = {...}, 
    queues=Vec<vm_virtio::queue::Queue>(len: 1, cap: 1) = {...}, queue_evts=Vec<vmm_sys_util::eventfd::EventFd>(len: 1, cap: 1) = {...}) at vm-virtio/src/rng.rs:236
236	        if queues.len() != NUM_QUEUES || queue_evts.len() != NUM_QUEUES {
(gdb) bt 
#0  _$LT$vm_virtio..rng..Rng$u20$as$u20$vm_virtio..device..VirtioDevice$GT$::activate::hdedcc2490e8a5155 (self=0x555555f71f10, mem=GuestMemoryMmap = {...}, 
    interrupt_cb=Arc<alloc::boxed::Box<Fn<(&vm_virtio::queue::Queue)>>> = {...}, status=Arc<core::sync::atomic::AtomicUsize> = {...}, 
    queues=Vec<vm_virtio::queue::Queue>(len: 1, cap: 1) = {...}, queue_evts=Vec<vmm_sys_util::eventfd::EventFd>(len: 1, cap: 1) = {...}) at vm-virtio/src/rng.rs:236
#1  0x0000555555b84dc5 in _$LT$vm_virtio..transport..pci_device..VirtioPciDevice$u20$as$u20$pci..device..PciDevice$GT$::write_bar::h7ff0b55cec4a3130 (
    self=0x555555f72170, offset=20, data=&[u8](len: 1) = {...}) at vm-virtio/src/transport/pci_device.rs:549
#2  0x0000555555b85692 in _$LT$vm_virtio..transport..pci_device..VirtioPciDevice$u20$as$u20$devices..bus..BusDevice$GT$::write::h32bdbeb4a3c402c0 (self=0x555555f72170, 
    offset=20, data=&[u8](len: 1) = {...}) at vm-virtio/src/transport/pci_device.rs:590
#3  0x0000555555be16ab in devices::bus::Bus::write::h23c30f9081f573b2 (self=0x7fffd6ec2788, addr=68717891604, data=&[u8](len: 1) = {...}) at devices/src/bus.rs:158
#4  0x0000555555ac561b in vmm::vm::Vcpu::run::h523fa30e0aaa7af6 (self=0x7fffd6ec2760) at vmm/src/vm.rs:363
#5  0x0000555555ae8eb3 in vmm::vm::Vm::start::_$u7b$$u7b$closure$u7d$$u7d$::h643431f00d7471fd () at vmm/src/vm.rs:1110
#6  0x0000555555ad9f12 in std::sys_common::backtrace::__rust_begin_short_backtrace::hd688e526ad581234 (f=...)
    at /rustc/91856ed52c58aa5ba66a015354d1cc69e9779bdf/src/libstd/sys_common/backtrace.rs:135
#7  0x0000555555ad7931 in std::thread::Builder::spawn_unchecked::_$u7b$$u7b$closure$u7d$$u7d$::_$u7b$$u7b$closure$u7d$$u7d$::h8d7f8a62b8d1257c ()
    at /rustc/91856ed52c58aa5ba66a015354d1cc69e9779bdf/src/libstd/thread/mod.rs:469
#8  0x0000555555af14a1 in _$LT$std..panic..AssertUnwindSafe$LT$F$GT$$u20$as$u20$core..ops..function..FnOnce$LT$$LP$$RP$$GT$$GT$::call_once::h1f5925e04de15538 (
    self=..., _args=0) at /rustc/91856ed52c58aa5ba66a015354d1cc69e9779bdf/src/libstd/panic.rs:309
#9  0x0000555555b00eda in std::panicking::try::do_call::hebfb4f2c896f9302 (data=0x7fffd6ec2a40 "20)367UUU")
    at /rustc/91856ed52c58aa5ba66a015354d1cc69e9779bdf/src/libstd/panicking.rs:297
#10 0x0000555555c3a0ca in __rust_maybe_catch_panic () at src/libpanic_unwind/lib.rs:87
#11 0x0000555555b00ca0 in std::panicking::try::h6da33e6171d79492 (f=...) at /rustc/91856ed52c58aa5ba66a015354d1cc69e9779bdf/src/libstd/panicking.rs:276
#12 0x0000555555af15d3 in std::panic::catch_unwind::h1660cc8af66e7984 (f=...) at /rustc/91856ed52c58aa5ba66a015354d1cc69e9779bdf/src/libstd/panic.rs:388
#13 0x0000555555ad7336 in std::thread::Builder::spawn_unchecked::_$u7b$$u7b$closure$u7d$$u7d$::h85cc8e716556c7d2 ()
    at /rustc/91856ed52c58aa5ba66a015354d1cc69e9779bdf/src/libstd/thread/mod.rs:468
#14 0x0000555555ad7c7b in _$LT$F$u20$as$u20$alloc..boxed..FnBox$LT$A$GT$$GT$::call_box::h5eac52d3a365b1f5 (self=0x555555f72d70, args=0)
    at /rustc/91856ed52c58aa5ba66a015354d1cc69e9779bdf/src/liballoc/boxed.rs:749
#15 0x0000555555c3985e in call_once<(),()> () at /rustc/91856ed52c58aa5ba66a015354d1cc69e9779bdf/src/liballoc/boxed.rs:759
#16 start_thread () at src/libstd/sys_common/thread.rs:14
#17 thread_start () at src/libstd/sys/unix/thread.rs:81
#18 0x00007ffff77b56ba in start_thread (arg=0x7fffd6ec4700) at pthread_create.c:333
#19 0x00007ffff72d541d in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:109

GDB commands by function

Print settings

set print pretty on

virtio (Github)

virtqueue的每个部分的内存对齐和大小要求如下表，其中Queue Size对应virtqueue中的最大buffer数，始终为2的n次幂，以特定于总线的方式指定：

Driver将sk_buffer填充进scatterlist table中(只是设置地址没有数据搬移)，然后通过计算得到GPA并将GPA写入Descriptor Table中，同时将Desc chain的head记录到Available Ring中，然后通过PIO的方式通知Device，Device发包并更新Used Ring。

virtio 系列文章

Virtio原理简介

6.1.Virtio 原理与Guest OS driver

Virtio概述和基本原理（KVM半虚拟化驱动）

VIRTIO VRING工作机制分析

VirtIO 代码分析（rust-vmm) cloud-hypervisor Latest commit f9036a1

VM创建过程:

1. src/main.rs 的main函数，解析参数，然后boot_kernel (vmm/src/lib.rs).

2. 创建kvm (kvm_ioctls crate, 打开/dev/kvm得到fd),

3. vmm/src/vm.rs 创建Vm,

(kvm.create_vm() 得到vm的fd),

初始化guest的memory,

CPU的Vec，

DeviceManager::new 来创建设备, 各种设备，包括ＰＣＩ设备在此创建。

4. boot_kernel中，先load_kernel，最后start (创建期望的个数的Vcpu，返回fd, 放在Vec<thread::JoinHandle<()>>中).

创建Ｖcpu时，会传入DeviceManager的各种信息：

io_bus = self.devices.io_bus.clone()

mmio_bus = self.devices.mmio_bus.clone()

ioapic = if let Some(ioapic) = &self.devices.ioapic {Some(ioapic.clone())}

io_bus和mmio_bus都是devices/src/bus.rs::Bus类型　｛devices: BTreeMap<BusRange, Arc<Mutex<BusDevice>>>｝

然后在每个thread中, vcpu.run (ioctls/vcpu.rs pub fn run).

5. 最后，control_loop 通过epoll从stdin读入数据，发给ＶＭ的serial.

PCI设备创建过程（Rng为例）:

1. vmm/src/vm.rs: DeviceManager会调用Rng::new 生成virtio_rng_device

2. DeviceManager::add_virtio_pci_device将virtio_rng_device 生成一个VirtioPciDevice，并加入到pci_root(PciRoot::new)中

PciConfigIo::new(pci_root)

还会调用PciConfigIo::register_mapping 设置bus, 会将 (address, size), VirtioPciDevice插入到bus 的map中。

3. VirtioPciDevice::new 会设置

PCI configuration registers

virtio PCI common configuration

MSI-X config

PCI interrupts

virtio queues,　会Rng转化为queue，并设置queue_evts（EventFd::new), vm_fd.register_ioevent

Guest memory

Setting PCI BAR

PCI设备访问过程（Rng为例mmio为例）:

１. 每个VCPU线程vmm::vm::Vcpu::run会判断VCPU的退出情况。以MmioWrite为例

2. devices/src/bus.rs::write 会遍历找到对应的device 进行　write_bar

3. 调用device.activate，　即Rng::activate, 会在一个线程中调用RngEpollHandler::run

4. process_queue

virtIO相关结构　　

1. struct Descriptor

/// A virtio descriptor constraints with C representive
struct Descriptor {
addr: u64,
len: u32,
flags: u16,
next: u16,
}

2. pub struct DescriptorChain

/// A virtio descriptor chain. 包含了Descriptor的４个字段

此外还包括申请的mmap,

desc_table的地址

desc table的索引

 1 /// A virtio descriptor chain.
 2 pub struct DescriptorChain<'a> {
 3     mem: &'a GuestMemoryMmap,
 4     desc_table: GuestAddress,
 5     queue_size: u16,
 6     ttl: u16, // used to prevent infinite chain cycles
 7 
 8     /// Index into the descriptor table
 9     pub index: u16,
10 
11     /// Guest physical address of device specific data
12     pub addr: GuestAddress,
13 
14     /// Length of device specific data
15     pub len: u32,
16 
17     /// Includes next, write, and indirect bits
18     pub flags: u16,
19 
20     /// Index into the descriptor table of the next descriptor if flags has
21     /// the next bit set
22     pub next: u16,
23 }

3. AvailIter

/// Consuming iterator over all available descriptor chain heads in the queue.

 1 /// Consuming iterator over all available descriptor chain heads in the queue.
 2 pub struct AvailIter<'a, 'b> {
 3     mem: &'a GuestMemoryMmap,
 4     desc_table: GuestAddress,
 5     avail_ring: GuestAddress,
 6     next_index: Wrapping<u16>,
 7     last_index: Wrapping<u16>,
 8     queue_size: u16,
 9     next_avail: &'b mut Wrapping<u16>,
10 }

View Code

4. Queue

/// A virtio queue's parameters. 包含了desc_table, avail_ring, used_ring　和中断向量的索引。

 1 /// A virtio queue's parameters.
 2 pub struct Queue {
 3     /// The maximal size in elements offered by the device
 4     max_size: u16,
 5 
 6     /// The queue size in elements the driver selected
 7     pub size: u16,
 8 
 9     /// Inidcates if the queue is finished with configuration
10     pub ready: bool,
11 
12     /// Interrupt vector index of the queue
13     pub vector: u16,
14 
15     /// Guest physical address of the descriptor table
16     pub desc_table: GuestAddress,
17 
18     /// Guest physical address of the available ring
19     pub avail_ring: GuestAddress,
20 
21     /// Guest physical address of the used ring
22     pub used_ring: GuestAddress,
23 
24     next_avail: Wrapping<u16>,
25     next_used: Wrapping<u16>,
26 }

View Code

5. Rng

Rng 是不包括Queue字段的

VirtioPciDevice 的new时候，生成Queue, 还会生成bar的地址fn allocate_bars。

Queue传入的地址就是为VM申请的guest_memory

VirtioPciDevice::read_bar 和　VirtioPciDevice::write_bar　（VirtioPciDevice被插入了bus中）　会调用common_config: VirtioPciCommonConfig的write 和read Queue的各个field.

PCI bar地址

1. vm new的时候，会初始化一个allocator = SystemAllocator::new.

2. 这个allocator会传给device_manager = DeviceManager::new .　　

3. DeviceManager new的时候，在add_virtio_pci_device的时候，会分配bars = virtio_pci_device.allocate_bars

4. 具体由vm-allocator/src/address.rs::AddressAllocator 的　AddressAllocator::allocate分配bar地址

5. SystemAllocator::allocate_mmio_addresses 会　Reserves a section of `size` bytes of MMIO address space.

PCI passtrough代码分析(从5372554开始支持VFIO)

REF:

VFIO ——将 DMA 映射暴露给用户态（推荐）

vfio-mdev逻辑空间分析 (mdev设备也是基于VFIO)

地址空间的故事（介绍了CCIX）

VFIO driver (VFIO的ORG文档，内核及用户空间源码，内核 3.6)

VFIO简介转自developerworks（很多不错的文章）

FPGA通过vfio实现pci-passthrough给kvm虚拟机

VFIO透传具体操作

Linux iommu和vfio概念空间解构

Cloud Phypervisor的实现

1. 5372554 实现了C语言的vfio-bindings

The default bindings are generated from the 5.0.0 Linux userspace API.

2. 2cec3aa 实现了VFIO API (用户态的调用这些API，操作VFIO, VFIO ——将 DMA 映射暴露给用户态 )

实现了 setup_dma_map， enable_msi, enable_msix, region_read, region_write

VFIO平台无关的接口层

向用户态提供访问硬件设备的接口
向用户态提供配置IOMMU的接

3. db5b476 实现了VFIO PCI device 的各种操作。VFIO实现层（PCI设备实现层）

allocate bar

write/read config register 会read/write region （write_config_register）

write/read bars 也会read/write region （write_bar）

allocate bar 时会 setup_dma_map

4. 20f0116 实现了MSI和MSI-X的设置

vfio: pci: Track MSI and MSI-X capabilities

5. c93d536 设置中断（需要研究一下原理）

vfio: pci: Build the KVM routes

VFIO实现层（PCI设备实现层）

6. b746dd7

vfio: Map MMIO regions into the guest

7. 4d16ca8 实现了PCI passthroug (Doc)

vmm: Support direct device assignment

5ae3144 testcase

8. 927861c .. fa41ddd VFIO的bug fix.

REF:

https://wiki.qemu.org/Documentation/vhost-user-ovs-dpdk

各种ＩＯ虚拟化实现及其优缺点

CCIX简介　（CCIX协议对于一些高性能应用详解）

PCI BAR设置过程

linux下共享内存mmap和DMA(直接访问内存)的使用 (mmap 共享内存例子)

Linux 下三种共享内存方式

详解vhost-user协议及其在OVS DPDK、QEMU和virtio-net驱动中的实现

vhost-user 分析1

OVS与DPDK vHost User端口

Passing File Descriptors (中文)

Vhost-user详解 (详细介绍了地址转换)

进程间传递文件描述符--sendmsg,recvmsg

client.c

#include <stdio.h>
#include <unistd.h>
#include <stdlib.h>
#include <stddef.h>
#include <string.h>
#include <sys/types.h>
#include <errno.h>
#include <sys/un.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <sys/epoll.h>
#include <fcntl.h>
 
#define UNIXSTR_PATH "foo.socket"
#define OPEN_FILE  "test"
 
int main(int argc, char *argv[])
{
	int clifd;
	struct sockaddr_un servaddr;  //IPC
	int ret;
	struct msghdr msg;
	struct iovec iov[1];
	char buf[100];
	union {  //保证cmsghdr和msg_control对齐
		struct cmsghdr cm;
		char control[CMSG_SPACE(sizeof(int))];
	} control_un;
	struct cmsghdr *pcmsg;
	int fd;
 
	clifd  = socket(AF_UNIX, SOCK_STREAM, 0) ;
	if   ( clifd  <  0 ) {
		printf ( "socket failed.
" ) ;
		return  - 1 ;
	}
 
	fd  =  open(OPEN_FILE ,O_CREAT | O_RDWR, 0777);
	if( fd  <  0 ) {
		printf("open test failed.
");
		return -1;
	}
 
	bzero (&servaddr, sizeof(servaddr));
	servaddr.sun_family = AF_UNIX;
	strcpy ( servaddr.sun_path, UNIXSTR_PATH);
 
	ret = connect(clifd, (struct sockaddr*)&servaddr, sizeof(servaddr));
	if(ret < 0) {
		printf ( "connect failed.
" ) ;
		return 0;
	}
	//udp需要,tcp无视
	msg.msg_name = NULL;
	msg.msg_namelen = 0;
	iov[0].iov_base = buf;
	iov[0].iov_len = 100;
	msg.msg_iov = iov;
	msg.msg_iovlen = 1;
	//设置缓冲区和长度
	msg.msg_control = control_un.control;
	msg.msg_controllen = sizeof(control_un.control);
	//直接通过CMSG_FIRSTHDR取得附属数据
	pcmsg = CMSG_FIRSTHDR(&msg);
	pcmsg->cmsg_len = CMSG_LEN(sizeof(int));
	pcmsg->cmsg_level = SOL_SOCKET;
	pcmsg->cmsg_type = SCM_RIGHTS;  //指明发送的是描述符
	*((int*)CMSG_DATA(pcmsg)) == fd;  //把描述符写入辅助数据
 
	ret = sendmsg(clifd, &msg, 0);  //send filedescriptor
	printf ("ret = %d, filedescriptor = %d
", ret, fd);
	return 0 ;
}

gcc -o client client.c

server.c

#include <stdio.h>
#include <unistd.h>
#include <stdlib.h>
#include <stddef.h>
#include <string.h>
#include <sys/types.h>
#include <errno.h>
#include <sys/un.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <sys/epoll.h>
#include <fcntl.h>
 
#define UNIXSTR_PATH "foo.socket"
 
int main(int argc, char *argv[])
{
	int clifd, listenfd;
	struct sockaddr_un servaddr, cliaddr;
	int ret;
	socklen_t clilen;
	struct msghdr msg;
	struct iovec iov[1];
	char buf [100];
	char  *testmsg = "test msg.
";
 
	union {  //对齐
		struct cmsghdr cm;
		char control[CMSG_SPACE(sizeof(int))];
	}  control_un;
	struct cmsghdr  * pcmsg;
	int  recvfd;
 
	listenfd  =  socket ( AF_UNIX ,  SOCK_STREAM ,  0 ) ;
	if(listenfd < 0) {
		printf ( "socket failed.
" ) ;
		return  -1;
	}
 
	unlink(UNIXSTR_PATH) ;
 
	bzero (&servaddr, sizeof(servaddr));
	servaddr.sun_family = AF_UNIX;
	strcpy ( servaddr.sun_path ,  UNIXSTR_PATH ) ;
 
	ret  =  bind ( listenfd, (struct sockaddr*)&servaddr, sizeof(servaddr));
	if(ret < 0) {
		printf ( "bind failed. errno = %d.
" ,  errno ) ;
		close(listenfd);
		return  - 1 ;
	}
 
	listen(listenfd, 5);
 
	while(1) {
		clilen = sizeof( cliaddr );
		clifd = accept( listenfd, (struct sockaddr*)&cliaddr , &clilen);
		if ( clifd < 0 ) {
			printf ( "accept failed.
" ) ;
			continue ;
		}
 
		msg.msg_name  = NULL;
		msg.msg_namelen  = 0;
		//设置数据缓冲区
		iov[0].iov_base = buf;
		iov[0].iov_len = 100;
		msg.msg_iov = iov;
		msg.msg_iovlen = 1;
		//设置辅助数据缓冲区和长度
		msg.msg_control = control_un.control;
		msg.msg_controllen  =  sizeof(control_un.control) ;
		//接收
		ret = recvmsg(clifd , &msg, 0);
		if( ret <= 0 ) {
			return ret;
		}
		//检查是否收到了辅助数据，以及长度
		if((pcmsg = CMSG_FIRSTHDR(&msg) ) != NULL && ( pcmsg->cmsg_len == CMSG_LEN(sizeof(int)))) {
			if ( pcmsg->cmsg_level  !=  SOL_SOCKET ) {
				printf("cmsg_leval is not SOL_SOCKET
");
				continue;
			}
 
			if ( pcmsg->cmsg_type != SCM_RIGHTS ) {
				printf ( "cmsg_type is not SCM_RIGHTS" );
				continue;
			}
			//这就是我们接收的描述符
			recvfd = *((int*)CMSG_DATA(pcmsg));
			printf ( "recv fd = %d
", recvfd );
 
			write ( recvfd, testmsg, strlen(testmsg) + 1);
		}
	}
 
	return 0 ;
}

gcc -o server server.c

redhad有关虚拟化，kvm， vDPA, virtio/net, ceph存储的blog：

What is virtualization?

What is KVM?

What is a hypervisor?

Containers vs VMs

Operate Red Hat Virtualization and KVM securely with sVirt

Red Hat Enterprise Virtualization Hypervisor: KVM now & in the future - 2015 Red Hat Summit

An Icehouse Sneak Peek – OpenStack Networking (Neutron)

Best practices for Red Hat Virtualization 4 technology detail

virtio步骤图示：

Virtio networking: A case study of I/O paravirtualization pdf

虚拟化系列文章

深入浅出KVM（一）丨简介&安装，

KVM 介绍（二）：CPU 和内存虚拟化，

深入浅出KVM（三）丨 I/O 全虚拟化和准虚拟化，

Linux云计算底层技术之 I/O 虚拟化,

什么是NFV（网络功能虚拟化）,

KVM 虚拟化详解,

VMX(1) -- 简介，VMX(2) -- VMCS， VMX(3) -- VMXON Region，

SMMU和IOMMU技术 [一] ，

虚拟化技术 - I/O虚拟化 [一]，

虚拟化技术 - I/O虚拟化 [二]，

Intel VT-d（1）- 简介文章系列

丙部同学的 Virtio 基本概念和设备操作

Virtio Spec Overview

总结：

1，前端填充好desc（addr/len），并更新vring->avail（ring[0]）
2，后端读取avail ring索引，找到desc(if ring[0]=2,then desctable[2] 记录的就是一个逻辑buffer的首个物理块的信息)，填充buffer数据；将buffer索引存在desc，将desc索引存放在used ring中
3，前端读取used ring索引，找到desc，获取buffer数据

查看全文

相关阅读:
python3.0与python2.0有哪些不同
 python常用内置模块,执行系统命令的模块
 06python 之基本数据类型
 python语言简介、解释器、字符编码介绍
 http协议&接口规范&接口测试入门
 基于APPIUM测试微信公众号的UI自动化测试框架（结合Allure2测试报告框架）
SQL注入工具sqlmap的注入过程记录
 unittest框架
 测试转型之路--学习ing
Tomcat分析-启动过程

原文地址：https://www.cnblogs.com/shaohef/p/11025893.html

rust-vmm 学习

V0.1.0 feature

base knowledge:

Using the KVM API (org)

kvm-ioctls (github)

virtio (Github)

VirtIO 代码分析（rust-vmm) cloud-hypervisor Latest commit f9036a1

Test_vm

VM创建过程:

PCI设备创建过程（Rng为例）:

PCI设备访问过程（Rng为例mmio为例）:

virtIO相关结构

PCI bar地址

PCI passtrough代码分析(从5372554开始支持VFIO)

REF:

Cloud Phypervisor的实现

REF:

virtIO相关结构　　