[dpdk] 熟悉SDK与初步使用 (三)(IP Fragmentation源码分析)
对例子IP Fragmentation的熟悉,使用,以及源码分析。
功能:
该例子的功能有二:
一: 将IP分片?
二: 根据路由表,做包转发。 路由表如下:
IP_FRAG: Socket 1: adding route 100.10.0.0/16 (port 0) IP_FRAG: Socket 1: adding route 100.20.0.0/16 (port 1) IP_FRAG: Socket 1: adding route 100.30.0.0/16 (port 2) IP_FRAG: Socket 1: adding route 100.40.0.0/16 (port 3) IP_FRAG: Socket 1: adding route 100.50.0.0/16 (port 4) IP_FRAG: Socket 1: adding route 100.60.0.0/16 (port 5) IP_FRAG: Socket 1: adding route 100.70.0.0/16 (port 6) IP_FRAG: Socket 1: adding route 100.80.0.0/16 (port 7) IP_FRAG: Socket 1: adding route 0101:0101:0101:0101:0101:0101:0101:0101/48 (port 0) IP_FRAG: Socket 1: adding route 0201:0101:0101:0101:0101:0101:0101:0101/48 (port 1) IP_FRAG: Socket 1: adding route 0301:0101:0101:0101:0101:0101:0101:0101/48 (port 2) IP_FRAG: Socket 1: adding route 0401:0101:0101:0101:0101:0101:0101:0101/48 (port 3) IP_FRAG: Socket 1: adding route 0501:0101:0101:0101:0101:0101:0101:0101/48 (port 4) IP_FRAG: Socket 1: adding route 0601:0101:0101:0101:0101:0101:0101:0101/48 (port 5) IP_FRAG: Socket 1: adding route 0701:0101:0101:0101:0101:0101:0101:0101/48 (port 6) IP_FRAG: Socket 1: adding route 0801:0101:0101:0101:0101:0101:0101:0101/48 (port 7)
问题一:
main()函数大概是这样的:标红的三行将与下面叙述的事情相关
int main(int argc, char **argv) { ... ... /* init EAL */ ret = rte_eal_init(argc, argv); if (ret < 0) rte_exit(EXIT_FAILURE, "rte_eal_init failed"); ... ... /* launch per-lcore init on every lcore */ rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER); RTE_LCORE_FOREACH_SLAVE(lcore_id) { if (rte_eal_wait_lcore(lcore_id) < 0) return -1; } return 0; }
其中,函数 rte_eal_wait_lcore 的实现如下:
/* * Wait until a lcore finished its job. */ int rte_eal_wait_lcore(unsigned slave_id) { if (lcore_config[slave_id].state == WAIT) return 0; while (lcore_config[slave_id].state != WAIT && lcore_config[slave_id].state != FINISHED); rte_rmb(); /* we are in finished state, go to wait state */ lcore_config[slave_id].state = WAIT; return lcore_config[slave_id].ret; }
阅读红色部分,可以很明显的发现,这是一个死循环啊!!! 从字面意义上来看,main函数在完成了remote_launch之后,主进程会在这个函数里等等子进程结束。
这样的话,用一个死循环来等,难道不会有问题吗??? 所以我要的debug它一下看看怎么回事。 于是,为了达到这个目的,我分别经历了下文中的问题二三四。终于debug成功了。解答如下:
解答起来其实也很简单,只需要看下 rte_eal_mp_remote_launch() 函数的代码,就明白了。它的代码如下:
66 /* 67 * Check that every SLAVE lcores are in WAIT state, then call 68 * rte_eal_remote_launch() for all of them. If call_master is true 69 * (set to CALL_MASTER), also call the function on the master lcore. 70 */ 71 int 72 rte_eal_mp_remote_launch(int (*f)(void *), void *arg, 73 enum rte_rmt_call_master_t call_master) 74 { 75 int lcore_id; 76 int master = rte_get_master_lcore(); 77 78 /* check state of lcores */ 79 RTE_LCORE_FOREACH_SLAVE(lcore_id) { 80 if (lcore_config[lcore_id].state != WAIT) 81 return -EBUSY; 82 } 83 84 /* send messages to cores */ 85 RTE_LCORE_FOREACH_SLAVE(lcore_id) { 86 rte_eal_remote_launch(f, arg, lcore_id); 87 } 88 89 if (call_master == CALL_MASTER) { 90 lcore_config[master].ret = f(arg); 91 lcore_config[master].state = FINISHED; 92 } 93 94 return 0; 95 }
从第90行可以看出。主进程在这里进入了业务逻辑,所以直到程序退出之前。它都没有机会执行前边的那个死循环。也就是说,主进程当进入死循环的时候,也说明其他进程即将结束。并不会存在长期空跑CPU的情况。 不过,如果业务逻辑写错了呢? 子进程并没有如逾期退出的话,是否会进入循环? 这里暂时先留下这个疑问。
另一个需要纪录下来的东西是。所有有需要的函数,实际上在rte_eal_init() 函数中便都创建完成了。remote_launch()函数实际上只是为其他进程传递一个启动运行的消息。
具体消息内容,目前我没有深入分析。
问题二:
运行不起来,启用DEBUG,gdb跟踪一下。
这个makefile也是很那难用的。摸索了一下,有几个命令,比较有用的如下:
[root@dpdk dpdk]# make help
[root@dpdk dpdk]# make V=yes D=yes
以上命令并没有用,到各模块的MAKEFILE里,将-O3手工改成-g,重新编译,才奏效。
问题三:
通过gdb发现,启动不了跟网卡特效有关系。
a。初始化函数中默认的参数是启用 硬checksum 等 offload 特性的。由于我模拟的网卡不支持,只能关掉。
static const struct rte_eth_conf port_conf = { .rxmode = { .max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE, .split_hdr_size = 0, .header_split = 0, /**< Header Split disabled */ .hw_ip_checksum = 0, /**< IP checksum offload enabled */ .hw_vlan_filter = 0, /**< VLAN filtering disabled */ .jumbo_frame = 1, /**< Jumbo Frame Support enabled */ .hw_strip_crc = 0, /**< CRC stripped by hardware */ }, .txmode = { .mq_mode = ETH_MQ_TX_NONE, }, };
b. 另一处修改
/* init one TX queue per couple (lcore,port) */ queueid = 0; for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) { if (rte_lcore_is_enabled(lcore_id) == 0) continue; socket = (int) rte_lcore_to_socket_id(lcore_id); printf("txq=%u,%d ", lcore_id, queueid); fflush(stdout); rte_eth_dev_info_get(portid, &dev_info); txconf = &dev_info.default_txconf; txconf->txq_flags = 0 | ETH_TXQ_FLAGS_NOXSUMS; ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd, socket, txconf); if (ret < 0) { printf("\n"); rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: " "err=%d, port=%d\n", ret, portid); } qconf = &lcore_queue_conf[lcore_id]; qconf->tx_queue_id[portid] = queueid; queueid++; }
c. 我之前模拟的网卡不支持多队列,经过学习研究,让 qemu/kvm 支持了多队列。另写了一篇,如下:
[Virtualization][qemu][kvm][virtio] 使用 QEMU/KVM 模拟网卡多队列
启动成功:
[root@dpdk build]# ./ip_fragmentation -l 6,7 -- -p 3 EAL: Detected 8 lcore(s) EAL: Probing VFIO support... EAL: WARNING: cpu flags constant_tsc=yes nonstop_tsc=no -> using unreliable clock cycles ! PMD: bnxt_rte_pmd_init() called for (null) EAL: PCI device 0000:00:03.0 on NUMA socket -1 EAL: probe driver: 1af4:1000 rte_virtio_pmd EAL: PCI device 0000:00:04.0 on NUMA socket -1 EAL: probe driver: 1af4:1000 rte_virtio_pmd EAL: PCI device 0000:00:05.0 on NUMA socket -1 EAL: probe driver: 1af4:1000 rte_virtio_pmd IP_FRAG: Creating direct mempool on socket 1 IP_FRAG: Creating indirect mempool on socket 1 IP_FRAG: Creating LPM table on socket 1 IP_FRAG: Creating LPM6 table on socket 1 Initializing port 0 on lcore 6... Address:00:00:00:01:00:01 txq=6,0 txq=7,1 Initializing port 1 on lcore 7... Address:00:00:00:01:00:02 txq=6,0 txq=7,1 IP_FRAG: Socket 1: adding route 100.10.0.0/16 (port 0) IP_FRAG: Socket 1: adding route 100.20.0.0/16 (port 1) IP_FRAG: Socket 1: adding route 100.30.0.0/16 (port 2) IP_FRAG: Socket 1: adding route 100.40.0.0/16 (port 3) IP_FRAG: Socket 1: adding route 100.50.0.0/16 (port 4) IP_FRAG: Socket 1: adding route 100.60.0.0/16 (port 5) IP_FRAG: Socket 1: adding route 100.70.0.0/16 (port 6) IP_FRAG: Socket 1: adding route 100.80.0.0/16 (port 7) IP_FRAG: Socket 1: adding route 0101:0101:0101:0101:0101:0101:0101:0101/48 (port 0) IP_FRAG: Socket 1: adding route 0201:0101:0101:0101:0101:0101:0101:0101/48 (port 1) IP_FRAG: Socket 1: adding route 0301:0101:0101:0101:0101:0101:0101:0101/48 (port 2) IP_FRAG: Socket 1: adding route 0401:0101:0101:0101:0101:0101:0101:0101/48 (port 3) IP_FRAG: Socket 1: adding route 0501:0101:0101:0101:0101:0101:0101:0101/48 (port 4) IP_FRAG: Socket 1: adding route 0601:0101:0101:0101:0101:0101:0101:0101/48 (port 5) IP_FRAG: Socket 1: adding route 0701:0101:0101:0101:0101:0101:0101:0101/48 (port 6) IP_FRAG: Socket 1: adding route 0801:0101:0101:0101:0101:0101:0101:0101/48 (port 7) Checking link status done Port 0 Link Up - speed 10000 Mbps - full-duplex Port 1 Link Up - speed 10000 Mbps - full-duplex IP_FRAG: entering main loop on lcore 7 IP_FRAG: -- lcoreid=7 portid=1 IP_FRAG: entering main loop on lcore 6 IP_FRAG: -- lcoreid=6 portid=0
问题四:
如何查看编译选项,使用的静态库。修改编译选项,启动debug等? 唯一的办法是makefile。结构还是很清晰的。但是,依然需要花很长的时间读。
打印编译命令的方法如下:
修改文件 /sdk/@dpdk/dpdk-stable-16.07.1 mk/internal/rte.compile-pre.mk 中的 C_TO_O_DO 变量: 第101行,为新增内容。
99 C_TO_O_DO = @set -e; \ 100 echo $(C_TO_O_DISP); \ 101 echo $(C_TO_O); \ 102 $(C_TO_O) && \ 103 $(PMDINFO_TO_O) && \ 104 echo $(C_TO_O_CMD) > $(call obj2cmd,$(@)) && \ 105 sed 's,'$@':,dep_'$@' =,' $(call obj2dep,$(@)).tmp > $(call obj2dep,$(@)) && \ 106 rm -f $(call obj2dep,$(@)).tmp 107
打印链接命令的方法如下:
修改文件 /sdk/@dpdk/dpdk-stable-16.07.1 mk/rte.app.mk 中的 O_TO_EXE_DO 变量: 第209行,为新增内容。
207 O_TO_EXE_DO = @set -e; \ 208 echo $(O_TO_EXE_DISP); \ 209 echo $(O_TO_EXE); \ 210 $(O_TO_EXE) && \ 211 echo $(O_TO_EXE_CMD) > $(call exe2cmd,$(@)) 212
实现效果如下:
[root@dpdk ip_fragmentation]# make echo "xxxxccccxxxx" xxxxccccxxxx CC main.o gcc -Wp,-MD,./.main.o.d.tmp -m64 -pthread -march=native -DRTE_MACHINE_CPUFLAG_SSE -DRTE_MACHINE_CPUFLAG_SSE2 -DRTE_MACHINE_CPUFLAG_SSE3 -DRTE_MACHINE_CPUFLAG_SSSE3 -DRTE_MACHINE_CPUFLAG_SSE4_1 -DRTE_MACHINE_CPUFLAG_SSE4_2 -I/root/src/sdk/@dpdk/dpdk-stable-16.07.1/examples/ip_fragmentation/build/include -I/root/dpdk//x86_64-native-linuxapp-gcc/include -include /root/dpdk//x86_64-native-linuxapp-gcc/include/rte_config.h -g -W -Wall -Wstrict-prototypes -Wmissing-prototypes -Wmissing-declarations -Wold-style-definition -Wpointer-arith -Wcast-align -Wnested-externs -Wcast-qual -Wformat-nonliteral -Wformat-security -Wundef -Wwrite-strings -Wno-return-type -o main.o -c /root/src/sdk/@dpdk/dpdk-stable-16.07.1/examples/ip_fragmentation/main.c LD ip_fragmentation gcc -o ip_fragmentation -m64 -pthread -march=native -DRTE_MACHINE_CPUFLAG_SSE -DRTE_MACHINE_CPUFLAG_SSE2 -DRTE_MACHINE_CPUFLAG_SSE3 -DRTE_MACHINE_CPUFLAG_SSSE3 -DRTE_MACHINE_CPUFLAG_SSE4_1 -DRTE_MACHINE_CPUFLAG_SSE4_2 -I/root/src/sdk/@dpdk/dpdk-stable-16.07.1/examples/ip_fragmentation/build/include -I/root/dpdk//x86_64-native-linuxapp-gcc/include -include /root/dpdk//x86_64-native-linuxapp-gcc/include/rte_config.h -g -W -Wall -Wstrict-prototypes -Wmissing-prototypes -Wmissing-declarations -Wold-style-definition -Wpointer-arith -Wcast-align -Wnested-externs -Wcast-qual -Wformat-nonliteral -Wformat-security -Wundef -Wwrite-strings main.o -L/root/dpdk//x86_64-native-linuxapp-gcc/lib -Wl,-lrte_kni -Wl,-lrte_pipeline -Wl,-lrte_table -Wl,-lrte_port -Wl,-lrte_pdump -Wl,-lrte_distributor -Wl,-lrte_reorder -Wl,-lrte_ip_frag -Wl,-lrte_meter -Wl,-lrte_sched -Wl,-lrte_lpm -Wl,--whole-archive -Wl,-lrte_acl -Wl,--no-whole-archive -Wl,-lrte_jobstats -Wl,-lrte_power -Wl,--whole-archive -Wl,-lrte_timer -Wl,-lrte_hash -Wl,-lrte_vhost -Wl,-lrte_kvargs -Wl,-lrte_mbuf -Wl,-lethdev -Wl,-lrte_cryptodev -Wl,-lrte_mempool -Wl,-lrte_ring -Wl,-lrte_eal -Wl,-lrte_cmdline -Wl,-lrte_cfgfile -Wl,-lrte_pmd_bond -Wl,-lrte_pmd_af_packet -Wl,-lrte_pmd_bnxt -Wl,-lrte_pmd_cxgbe -Wl,-lrte_pmd_e1000 -Wl,-lrte_pmd_ena -Wl,-lrte_pmd_enic -Wl,-lrte_pmd_fm10k -Wl,-lrte_pmd_i40e -Wl,-lrte_pmd_ixgbe -Wl,-lrte_pmd_null -Wl,-lrte_pmd_ring -Wl,-lrte_pmd_virtio -Wl,-lrte_pmd_vhost -Wl,-lrte_pmd_vmxnet3_uio -Wl,-lrte_pmd_null_crypto -Wl,--no-whole-archive -Wl,-lrt -Wl,-lm -Wl,-ldl -Wl,-export-dynamic -Wl,-export-dynamic -L/root/src/sdk/@dpdk/dpdk-stable-16.07.1/examples/ip_fragmentation/build/lib -L/root/dpdk//x86_64-native-linuxapp-gcc/lib -Wl,--as-needed -Wl,-Map=ip_fragmentation.map -Wl,--cref INSTALL-APP ip_fragmentation INSTALL-MAP ip_fragmentation.map [root@dpdk ip_fragmentation]#
问题五
启动例子程序之后,做发包测试。发现所有包都被源端口转发回来。通过代码可以看到,默认的路由规则就是源端口转发回来。
a。怎么方便的发包呢? 除了已知的 tcpreplay 可以在端口上将包原样转发以外。还可以使用 tcpreplay-edit 对包内容做一下修改后在发送。我就是通过这种方法来测试例子中的路由表的。
/home/tong/Data [tong@T7] [11:24] > sudo tcpreplay-edit -i tap-dpdk-1 -D 0.0.0.0/0:100.20.0.0/16 --enet-dmac=00:00:00:01:00:01 -L1 oicq-bak.pcap Actual: 2 packets (162 bytes) sent in 30.01 seconds. Rated: 5.3 Bps, 0.000 Mbps, 0.06 pps Flows: 1 flows, 0.03 fps, 2 flow packets, 0 non-flow Statistics for network device: tap-dpdk-1 Attempted packets: 2 Successful packets: 2 Failed packets: 0 Truncated packets: 0 Retried packets (ENOBUFS): 0 Retried packets (EAGAIN): 0
b。通过 debug 发现收包之后的结构如下:
其中,红色部分的结构代表了报文类型。很显然,可以看出来,程序在这个时间点,并没有识别到,该报文是IPv4 or IPv6。在后续做转发的代码逻辑里,会对IP类型进行判断,在此结构体数值下,该代码判断为既不是v4,也不是v6。故进入了默认路由,从源端口发了出来。
所以,到目前为止,并不知道为什么报类型没有被识别。可以有三种情况,1,代码依赖了硬件来处理,而我的模拟网卡不能处理。2. pmd处理,又要我是虚拟机,故没有处理。3. 例子代码有误。应该在代码某处,调用一个识别的函数进行处理。
总之,在还未完全高清楚一个常规包处理流程和逻辑时。该问题暂搁置。
(gdb) p *m $26 = {cacheline0 = 0x7fffd4dd4300, buf_addr = 0x7fffd4dd4380, buf_physaddr = 2059223936, buf_len = 2176, rearm_data = 0x7fffd4dd4312 "\216", data_off = 142, { refcnt_atomic = {cnt = 1}, refcnt = 1}, nb_segs = 1 '\001', port = 0 '\000', ol_flags = 0, rx_descriptor_fields1 = 0x7fffd4dd4320, {packet_type = 0, { l2_type = 0, l3_type = 0, l4_type = 0, tun_type = 0, inner_l2_type = 0, inner_l3_type = 0, inner_l4_type = 0}}, pkt_len = 67, data_len = 67, vlan_tci = 0, hash = {rss = 0, fdir = {{{hash = 0, id = 0}, lo = 0}, hi = 0}, sched = {lo = 0, hi = 0}, usr = 0}, seqn = 0, vlan_tci_outer = 0, cacheline1 = 0x7fffd4dd4340, { userdata = 0x0, udata64 = 0}, pool = 0x7fffd60436c0, next = 0x0, {tx_offload = 0, {l2_len = 0, l3_len = 0, l4_len = 0, tso_segsz = 0, outer_l3_len = 0, outer_l2_len = 0}}, priv_size = 0, timesync = 0} (gdb)
c. 改了qemu,模拟了硬件checksum,仍不行。通过debug分析代码,好像是virtio pmd driver就不支持硬件checksum。新的qemu命令如下(注意红色部分)。并没有找到方法怎么在guest里通过查看的方式确认网卡是否支持 checksum offload.
/home/tong/VM/dpdk [tong@T7] [19:06] > cat start-multiqueue.sh sudo qemu-system-x86_64 -nographic -vnc 127.0.0.1:1 -enable-kvm \ -m 2G -cpu Nehalem -smp cores=2,threads=2,sockets=2 \ -numa node,mem=1G,cpus=0-3,nodeid=0 \ -numa node,mem=1G,cpus=4-7,nodeid=1 \ -drive file=disk.img,if=virtio \ -net nic,vlan=0,model=virtio,macaddr='00:00:00:01:00:00' \ -device virtio-net-pci,netdev=dev1,mac='00:00:00:01:00:01',vectors=34,mq=on,csum=on,guest_csum=on \ -device virtio-net-pci,netdev=dev2,mac='00:00:00:01:00:02',vectors=34,mq=on,csum=on,guest_csum=on \ -device virtio-net-pci,netdev=dev3,mac='00:00:00:01:00:03',vectors=34,mq=on,csum=on,guest_csum=on \ -net tap,vlan=0,ifname=tap-dpdk-ctrl \ -netdev tap,ifname=tap-dpdk-1,script=no,downscript=no,vhost=on,queues=16,id=dev1 \ -netdev tap,ifname=tap-dpdk-2,script=no,downscript=no,vhost=on,queues=16,id=dev2 \ -netdev tap,ifname=tap-dpdk-3,script=no,downscript=no,vhost=on,queues=16,id=dev3 & # -device vfio-pci,host='0000:00:19.0' \ #ne2k_pci,i82551,i82557b,i82559er,rtl8139,e1000,pcnet,virtio /home/tong/VM/dpdk [tong@T7] [19:06] >