1/* 2 * Authors: 3 * Copyright 2001, 2002 by Robert Olsson <robert.olsson@its.uu.se> 4 * Uppsala University and 5 * Swedish University of Agricultural Sciences 6 * 7 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> 8 * Ben Greear <greearb@candelatech.com> 9 * Jens Låås <jens.laas@data.slu.se> 10 * 11 * This program is free software; you can redistribute it and/or 12 * modify it under the terms of the GNU General Public License 13 * as published by the Free Software Foundation; either version 14 * 2 of the License, or (at your option) any later version. 15 * 16 * 17 * A tool for loading the network with preconfigurated packets. 18 * The tool is implemented as a linux module. Parameters are output 19 * device, delay (to hard_xmit), number of packets, and whether 20 * to use multiple SKBs or just the same one. 21 * pktgen uses the installed interface's output routine. 22 * 23 * Additional hacking by: 24 * 25 * Jens.Laas@data.slu.se 26 * Improved by ANK. 010120. 27 * Improved by ANK even more. 010212. 28 * MAC address typo fixed. 010417 --ro 29 * Integrated. 020301 --DaveM 30 * Added multiskb option 020301 --DaveM 31 * Scaling of results. 020417--sigurdur@linpro.no 32 * Significant re-work of the module: 33 * * Convert to threaded model to more efficiently be able to transmit 34 * and receive on multiple interfaces at once. 35 * * Converted many counters to __u64 to allow longer runs. 36 * * Allow configuration of ranges, like min/max IP address, MACs, 37 * and UDP-ports, for both source and destination, and can 38 * set to use a random distribution or sequentially walk the range. 39 * * Can now change most values after starting. 40 * * Place 12-byte packet in UDP payload with magic number, 41 * sequence number, and timestamp. 42 * * Add receiver code that detects dropped pkts, re-ordered pkts, and 43 * latencies (with micro-second) precision. 44 * * Add IOCTL interface to easily get counters & configuration. 45 * --Ben Greear <greearb@candelatech.com> 46 * 47 * Renamed multiskb to clone_skb and cleaned up sending core for two distinct 48 * skb modes. A clone_skb=0 mode for Ben "ranges" work and a clone_skb != 0 49 * as a "fastpath" with a configurable number of clones after alloc's. 50 * clone_skb=0 means all packets are allocated this also means ranges time 51 * stamps etc can be used. clone_skb=100 means 1 malloc is followed by 100 52 * clones. 53 * 54 * Also moved to /proc/net/pktgen/ 55 * --ro 56 * 57 * Sept 10: Fixed threading/locking. Lots of bone-headed and more clever 58 * mistakes. Also merged in DaveM's patch in the -pre6 patch. 59 * --Ben Greear <greearb@candelatech.com> 60 * 61 * Integrated to 2.5.x 021029 --Lucio Maciel (luciomaciel@zipmail.com.br) 62 * 63 * 64 * 021124 Finished major redesign and rewrite for new functionality. 65 * See Documentation/networking/pktgen.txt for how to use this. 66 * 67 * The new operation: 68 * For each CPU one thread/process is created at start. This process checks 69 * for running devices in the if_list and sends packets until count is 0 it 70 * also the thread checks the thread->control which is used for inter-process 71 * communication. controlling process "posts" operations to the threads this 72 * way. 73 * The if_list is RCU protected, and the if_lock remains to protect updating 74 * of if_list, from "add_device" as it invoked from userspace (via proc write). 75 * 76 * By design there should only be *one* "controlling" process. In practice 77 * multiple write accesses gives unpredictable result. Understood by "write" 78 * to /proc gives result code thats should be read be the "writer". 79 * For practical use this should be no problem. 80 * 81 * Note when adding devices to a specific CPU there good idea to also assign 82 * /proc/irq/XX/smp_affinity so TX-interrupts gets bound to the same CPU. 83 * --ro 84 * 85 * Fix refcount off by one if first packet fails, potential null deref, 86 * memleak 030710- KJP 87 * 88 * First "ranges" functionality for ipv6 030726 --ro 89 * 90 * Included flow support. 030802 ANK. 91 * 92 * Fixed unaligned access on IA-64 Grant Grundler <grundler@parisc-linux.org> 93 * 94 * Remove if fix from added Harald Welte <laforge@netfilter.org> 040419 95 * ia64 compilation fix from Aron Griffis <aron@hp.com> 040604 96 * 97 * New xmit() return, do_div and misc clean up by Stephen Hemminger 98 * <shemminger@osdl.org> 040923 99 * 100 * Randy Dunlap fixed u64 printk compiler waring 101 * 102 * Remove FCS from BW calculation. Lennert Buytenhek <buytenh@wantstofly.org> 103 * New time handling. Lennert Buytenhek <buytenh@wantstofly.org> 041213 104 * 105 * Corrections from Nikolai Malykh (nmalykh@bilim.com) 106 * Removed unused flags F_SET_SRCMAC & F_SET_SRCIP 041230 107 * 108 * interruptible_sleep_on_timeout() replaced Nishanth Aravamudan <nacc@us.ibm.com> 109 * 050103 110 * 111 * MPLS support by Steven Whitehouse <steve@chygwyn.com> 112 * 113 * 802.1Q/Q-in-Q support by Francesco Fondelli (FF) <francesco.fondelli@gmail.com> 114 * 115 * Fixed src_mac command to set source mac of packet to value specified in 116 * command by Adit Ranadive <adit.262@gmail.com> 117 * 118 */ 119 120#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 121 122#include <linux/sys.h> 123#include <linux/types.h> 124#include <linux/module.h> 125#include <linux/moduleparam.h> 126#include <linux/kernel.h> 127#include <linux/mutex.h> 128#include <linux/sched.h> 129#include <linux/slab.h> 130#include <linux/vmalloc.h> 131#include <linux/unistd.h> 132#include <linux/string.h> 133#include <linux/ptrace.h> 134#include <linux/errno.h> 135#include <linux/ioport.h> 136#include <linux/interrupt.h> 137#include <linux/capability.h> 138#include <linux/hrtimer.h> 139#include <linux/freezer.h> 140#include <linux/delay.h> 141#include <linux/timer.h> 142#include <linux/list.h> 143#include <linux/init.h> 144#include <linux/skbuff.h> 145#include <linux/netdevice.h> 146#include <linux/inet.h> 147#include <linux/inetdevice.h> 148#include <linux/rtnetlink.h> 149#include <linux/if_arp.h> 150#include <linux/if_vlan.h> 151#include <linux/in.h> 152#include <linux/ip.h> 153#include <linux/ipv6.h> 154#include <linux/udp.h> 155#include <linux/proc_fs.h> 156#include <linux/seq_file.h> 157#include <linux/wait.h> 158#include <linux/etherdevice.h> 159#include <linux/kthread.h> 160#include <linux/prefetch.h> 161#include <net/net_namespace.h> 162#include <net/checksum.h> 163#include <net/ipv6.h> 164#include <net/udp.h> 165#include <net/ip6_checksum.h> 166#include <net/addrconf.h> 167#ifdef CONFIG_XFRM 168#include <net/xfrm.h> 169#endif 170#include <net/netns/generic.h> 171#include <asm/byteorder.h> 172#include <linux/rcupdate.h> 173#include <linux/bitops.h> 174#include <linux/io.h> 175#include <linux/timex.h> 176#include <linux/uaccess.h> 177#include <asm/dma.h> 178#include <asm/div64.h> /* do_div */ 179 180#define VERSION "2.74" 181#define IP_NAME_SZ 32 182#define MAX_MPLS_LABELS 16 /* This is the max label stack depth */ 183#define MPLS_STACK_BOTTOM htonl(0x00000100) 184 185#define func_enter() pr_debug("entering %s\n", __func__); 186 187/* Device flag bits */ 188#define F_IPSRC_RND (1<<0) /* IP-Src Random */ 189#define F_IPDST_RND (1<<1) /* IP-Dst Random */ 190#define F_UDPSRC_RND (1<<2) /* UDP-Src Random */ 191#define F_UDPDST_RND (1<<3) /* UDP-Dst Random */ 192#define F_MACSRC_RND (1<<4) /* MAC-Src Random */ 193#define F_MACDST_RND (1<<5) /* MAC-Dst Random */ 194#define F_TXSIZE_RND (1<<6) /* Transmit size is random */ 195#define F_IPV6 (1<<7) /* Interface in IPV6 Mode */ 196#define F_MPLS_RND (1<<8) /* Random MPLS labels */ 197#define F_VID_RND (1<<9) /* Random VLAN ID */ 198#define F_SVID_RND (1<<10) /* Random SVLAN ID */ 199#define F_FLOW_SEQ (1<<11) /* Sequential flows */ 200#define F_IPSEC_ON (1<<12) /* ipsec on for flows */ 201#define F_QUEUE_MAP_RND (1<<13) /* queue map Random */ 202#define F_QUEUE_MAP_CPU (1<<14) /* queue map mirrors smp_processor_id() */ 203#define F_NODE (1<<15) /* Node memory alloc*/ 204#define F_UDPCSUM (1<<16) /* Include UDP checksum */ 205#define F_NO_TIMESTAMP (1<<17) /* Don't timestamp packets (default TS) */ 206 207/* Thread control flag bits */ 208#define T_STOP (1<<0) /* Stop run */ 209#define T_RUN (1<<1) /* Start run */ 210#define T_REMDEVALL (1<<2) /* Remove all devs */ 211#define T_REMDEV (1<<3) /* Remove one dev */ 212 213/* If lock -- protects updating of if_list */ 214#define if_lock(t) spin_lock(&(t->if_lock)); 215#define if_unlock(t) spin_unlock(&(t->if_lock)); 216 217/* Used to help with determining the pkts on receive */ 218#define PKTGEN_MAGIC 0xbe9be955 219#define PG_PROC_DIR "pktgen" 220#define PGCTRL "pgctrl" 221 222#define MAX_CFLOWS 65536 223 224#define VLAN_TAG_SIZE(x) ((x)->vlan_id == 0xffff ? 0 : 4) 225#define SVLAN_TAG_SIZE(x) ((x)->svlan_id == 0xffff ? 0 : 4) 226 227struct flow_state { 228 __be32 cur_daddr; 229 int count; 230#ifdef CONFIG_XFRM 231 struct xfrm_state *x; 232#endif 233 __u32 flags; 234}; 235 236/* flow flag bits */ 237#define F_INIT (1<<0) /* flow has been initialized */ 238 239struct pktgen_dev { 240 /* 241 * Try to keep frequent/infrequent used vars. separated. 242 */ 243 struct proc_dir_entry *entry; /* proc file */ 244 struct pktgen_thread *pg_thread;/* the owner */ 245 struct list_head list; /* chaining in the thread's run-queue */ 246 struct rcu_head rcu; /* freed by RCU */ 247 248 int running; /* if false, the test will stop */ 249 250 /* If min != max, then we will either do a linear iteration, or 251 * we will do a random selection from within the range. 252 */ 253 __u32 flags; 254 int removal_mark; /* non-zero => the device is marked for 255 * removal by worker thread */ 256 257 int min_pkt_size; 258 int max_pkt_size; 259 int pkt_overhead; /* overhead for MPLS, VLANs, IPSEC etc */ 260 int nfrags; 261 struct page *page; 262 u64 delay; /* nano-seconds */ 263 264 __u64 count; /* Default No packets to send */ 265 __u64 sofar; /* How many pkts we've sent so far */ 266 __u64 tx_bytes; /* How many bytes we've transmitted */ 267 __u64 errors; /* Errors when trying to transmit, */ 268 269 /* runtime counters relating to clone_skb */ 270 271 __u64 allocated_skbs; 272 __u32 clone_count; 273 int last_ok; /* Was last skb sent? 274 * Or a failed transmit of some sort? 275 * This will keep sequence numbers in order 276 */ 277 ktime_t next_tx; 278 ktime_t started_at; 279 ktime_t stopped_at; 280 u64 idle_acc; /* nano-seconds */ 281 282 __u32 seq_num; 283 284 int clone_skb; /* 285 * Use multiple SKBs during packet gen. 286 * If this number is greater than 1, then 287 * that many copies of the same packet will be 288 * sent before a new packet is allocated. 289 * If you want to send 1024 identical packets 290 * before creating a new packet, 291 * set clone_skb to 1024. 292 */ 293 294 char dst_min[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */ 295 char dst_max[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */ 296 char src_min[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */ 297 char src_max[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */ 298 299 struct in6_addr in6_saddr; 300 struct in6_addr in6_daddr; 301 struct in6_addr cur_in6_daddr; 302 struct in6_addr cur_in6_saddr; 303 /* For ranges */ 304 struct in6_addr min_in6_daddr; 305 struct in6_addr max_in6_daddr; 306 struct in6_addr min_in6_saddr; 307 struct in6_addr max_in6_saddr; 308 309 /* If we're doing ranges, random or incremental, then this 310 * defines the min/max for those ranges. 311 */ 312 __be32 saddr_min; /* inclusive, source IP address */ 313 __be32 saddr_max; /* exclusive, source IP address */ 314 __be32 daddr_min; /* inclusive, dest IP address */ 315 __be32 daddr_max; /* exclusive, dest IP address */ 316 317 __u16 udp_src_min; /* inclusive, source UDP port */ 318 __u16 udp_src_max; /* exclusive, source UDP port */ 319 __u16 udp_dst_min; /* inclusive, dest UDP port */ 320 __u16 udp_dst_max; /* exclusive, dest UDP port */ 321 322 /* DSCP + ECN */ 323 __u8 tos; /* six MSB of (former) IPv4 TOS 324 are for dscp codepoint */ 325 __u8 traffic_class; /* ditto for the (former) Traffic Class in IPv6 326 (see RFC 3260, sec. 4) */ 327 328 /* MPLS */ 329 unsigned int nr_labels; /* Depth of stack, 0 = no MPLS */ 330 __be32 labels[MAX_MPLS_LABELS]; 331 332 /* VLAN/SVLAN (802.1Q/Q-in-Q) */ 333 __u8 vlan_p; 334 __u8 vlan_cfi; 335 __u16 vlan_id; /* 0xffff means no vlan tag */ 336 337 __u8 svlan_p; 338 __u8 svlan_cfi; 339 __u16 svlan_id; /* 0xffff means no svlan tag */ 340 341 __u32 src_mac_count; /* How many MACs to iterate through */ 342 __u32 dst_mac_count; /* How many MACs to iterate through */ 343 344 unsigned char dst_mac[ETH_ALEN]; 345 unsigned char src_mac[ETH_ALEN]; 346 347 __u32 cur_dst_mac_offset; 348 __u32 cur_src_mac_offset; 349 __be32 cur_saddr; 350 __be32 cur_daddr; 351 __u16 ip_id; 352 __u16 cur_udp_dst; 353 __u16 cur_udp_src; 354 __u16 cur_queue_map; 355 __u32 cur_pkt_size; 356 __u32 last_pkt_size; 357 358 __u8 hh[14]; 359 /* = { 360 0x00, 0x80, 0xC8, 0x79, 0xB3, 0xCB, 361 362 We fill in SRC address later 363 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 364 0x08, 0x00 365 }; 366 */ 367 __u16 pad; /* pad out the hh struct to an even 16 bytes */ 368 369 struct sk_buff *skb; /* skb we are to transmit next, used for when we 370 * are transmitting the same one multiple times 371 */ 372 struct net_device *odev; /* The out-going device. 373 * Note that the device should have it's 374 * pg_info pointer pointing back to this 375 * device. 376 * Set when the user specifies the out-going 377 * device name (not when the inject is 378 * started as it used to do.) 379 */ 380 char odevname[32]; 381 struct flow_state *flows; 382 unsigned int cflows; /* Concurrent flows (config) */ 383 unsigned int lflow; /* Flow length (config) */ 384 unsigned int nflows; /* accumulated flows (stats) */ 385 unsigned int curfl; /* current sequenced flow (state)*/ 386 387 u16 queue_map_min; 388 u16 queue_map_max; 389 __u32 skb_priority; /* skb priority field */ 390 unsigned int burst; /* number of duplicated packets to burst */ 391 int node; /* Memory node */ 392 393#ifdef CONFIG_XFRM 394 __u8 ipsmode; /* IPSEC mode (config) */ 395 __u8 ipsproto; /* IPSEC type (config) */ 396 __u32 spi; 397 struct dst_entry dst; 398 struct dst_ops dstops; 399#endif 400 char result[512]; 401}; 402 403struct pktgen_hdr { 404 __be32 pgh_magic; 405 __be32 seq_num; 406 __be32 tv_sec; 407 __be32 tv_usec; 408}; 409 410 411static int pg_net_id __read_mostly; 412 413struct pktgen_net { 414 struct net *net; 415 struct proc_dir_entry *proc_dir; 416 struct list_head pktgen_threads; 417 bool pktgen_exiting; 418}; 419 420struct pktgen_thread { 421 spinlock_t if_lock; /* for list of devices */ 422 struct list_head if_list; /* All device here */ 423 struct list_head th_list; 424 struct task_struct *tsk; 425 char result[512]; 426 427 /* Field for thread to receive "posted" events terminate, 428 stop ifs etc. */ 429 430 u32 control; 431 int cpu; 432 433 wait_queue_head_t queue; 434 struct completion start_done; 435 struct pktgen_net *net; 436}; 437 438#define REMOVE 1 439#define FIND 0 440 441static const char version[] = 442 "Packet Generator for packet performance testing. " 443 "Version: " VERSION "\n"; 444 445static int pktgen_remove_device(struct pktgen_thread *t, struct pktgen_dev *i); 446static int pktgen_add_device(struct pktgen_thread *t, const char *ifname); 447static struct pktgen_dev *pktgen_find_dev(struct pktgen_thread *t, 448 const char *ifname, bool exact); 449static int pktgen_device_event(struct notifier_block *, unsigned long, void *); 450static void pktgen_run_all_threads(struct pktgen_net *pn); 451static void pktgen_reset_all_threads(struct pktgen_net *pn); 452static void pktgen_stop_all_threads_ifs(struct pktgen_net *pn); 453 454static void pktgen_stop(struct pktgen_thread *t); 455static void pktgen_clear_counters(struct pktgen_dev *pkt_dev); 456 457/* Module parameters, defaults. */ 458static int pg_count_d __read_mostly = 1000; 459static int pg_delay_d __read_mostly; 460static int pg_clone_skb_d __read_mostly; 461static int debug __read_mostly; 462 463static DEFINE_MUTEX(pktgen_thread_lock); 464 465static struct notifier_block pktgen_notifier_block = { 466 .notifier_call = pktgen_device_event, 467}; 468 469/* 470 * /proc handling functions 471 * 472 */ 473 474static int pgctrl_show(struct seq_file *seq, void *v) 475{ 476 seq_puts(seq, version); 477 return 0; 478} 479 480static ssize_t pgctrl_write(struct file *file, const char __user *buf, 481 size_t count, loff_t *ppos) 482{ 483 char data[128]; 484 struct pktgen_net *pn = net_generic(current->nsproxy->net_ns, pg_net_id); 485 486 if (!capable(CAP_NET_ADMIN)) 487 return -EPERM; 488 489 if (count == 0) 490 return -EINVAL; 491 492 if (count > sizeof(data)) 493 count = sizeof(data); 494 495 if (copy_from_user(data, buf, count)) 496 return -EFAULT; 497 498 data[count - 1] = 0; /* Strip trailing '\n' and terminate string */ 499 500 if (!strcmp(data, "stop")) 501 pktgen_stop_all_threads_ifs(pn); 502 503 else if (!strcmp(data, "start")) 504 pktgen_run_all_threads(pn); 505 506 else if (!strcmp(data, "reset")) 507 pktgen_reset_all_threads(pn); 508 509 else 510 pr_warn("Unknown command: %s\n", data); 511 512 return count; 513} 514 515static int pgctrl_open(struct inode *inode, struct file *file) 516{ 517 return single_open(file, pgctrl_show, PDE_DATA(inode)); 518} 519 520static const struct file_operations pktgen_fops = { 521 .owner = THIS_MODULE, 522 .open = pgctrl_open, 523 .read = seq_read, 524 .llseek = seq_lseek, 525 .write = pgctrl_write, 526 .release = single_release, 527}; 528 529static int pktgen_if_show(struct seq_file *seq, void *v) 530{ 531 const struct pktgen_dev *pkt_dev = seq->private; 532 ktime_t stopped; 533 u64 idle; 534 535 seq_printf(seq, 536 "Params: count %llu min_pkt_size: %u max_pkt_size: %u\n", 537 (unsigned long long)pkt_dev->count, pkt_dev->min_pkt_size, 538 pkt_dev->max_pkt_size); 539 540 seq_printf(seq, 541 " frags: %d delay: %llu clone_skb: %d ifname: %s\n", 542 pkt_dev->nfrags, (unsigned long long) pkt_dev->delay, 543 pkt_dev->clone_skb, pkt_dev->odevname); 544 545 seq_printf(seq, " flows: %u flowlen: %u\n", pkt_dev->cflows, 546 pkt_dev->lflow); 547 548 seq_printf(seq, 549 " queue_map_min: %u queue_map_max: %u\n", 550 pkt_dev->queue_map_min, 551 pkt_dev->queue_map_max); 552 553 if (pkt_dev->skb_priority) 554 seq_printf(seq, " skb_priority: %u\n", 555 pkt_dev->skb_priority); 556 557 if (pkt_dev->flags & F_IPV6) { 558 seq_printf(seq, 559 " saddr: %pI6c min_saddr: %pI6c max_saddr: %pI6c\n" 560 " daddr: %pI6c min_daddr: %pI6c max_daddr: %pI6c\n", 561 &pkt_dev->in6_saddr, 562 &pkt_dev->min_in6_saddr, &pkt_dev->max_in6_saddr, 563 &pkt_dev->in6_daddr, 564 &pkt_dev->min_in6_daddr, &pkt_dev->max_in6_daddr); 565 } else { 566 seq_printf(seq, 567 " dst_min: %s dst_max: %s\n", 568 pkt_dev->dst_min, pkt_dev->dst_max); 569 seq_printf(seq, 570 " src_min: %s src_max: %s\n", 571 pkt_dev->src_min, pkt_dev->src_max); 572 } 573 574 seq_puts(seq, " src_mac: "); 575 576 seq_printf(seq, "%pM ", 577 is_zero_ether_addr(pkt_dev->src_mac) ? 578 pkt_dev->odev->dev_addr : pkt_dev->src_mac); 579 580 seq_puts(seq, "dst_mac: "); 581 seq_printf(seq, "%pM\n", pkt_dev->dst_mac); 582 583 seq_printf(seq, 584 " udp_src_min: %d udp_src_max: %d" 585 " udp_dst_min: %d udp_dst_max: %d\n", 586 pkt_dev->udp_src_min, pkt_dev->udp_src_max, 587 pkt_dev->udp_dst_min, pkt_dev->udp_dst_max); 588 589 seq_printf(seq, 590 " src_mac_count: %d dst_mac_count: %d\n", 591 pkt_dev->src_mac_count, pkt_dev->dst_mac_count); 592 593 if (pkt_dev->nr_labels) { 594 unsigned int i; 595 seq_puts(seq, " mpls: "); 596 for (i = 0; i < pkt_dev->nr_labels; i++) 597 seq_printf(seq, "%08x%s", ntohl(pkt_dev->labels[i]), 598 i == pkt_dev->nr_labels-1 ? "\n" : ", "); 599 } 600 601 if (pkt_dev->vlan_id != 0xffff) 602 seq_printf(seq, " vlan_id: %u vlan_p: %u vlan_cfi: %u\n", 603 pkt_dev->vlan_id, pkt_dev->vlan_p, 604 pkt_dev->vlan_cfi); 605 606 if (pkt_dev->svlan_id != 0xffff) 607 seq_printf(seq, " svlan_id: %u vlan_p: %u vlan_cfi: %u\n", 608 pkt_dev->svlan_id, pkt_dev->svlan_p, 609 pkt_dev->svlan_cfi); 610 611 if (pkt_dev->tos) 612 seq_printf(seq, " tos: 0x%02x\n", pkt_dev->tos); 613 614 if (pkt_dev->traffic_class) 615 seq_printf(seq, " traffic_class: 0x%02x\n", pkt_dev->traffic_class); 616 617 if (pkt_dev->burst > 1) 618 seq_printf(seq, " burst: %d\n", pkt_dev->burst); 619 620 if (pkt_dev->node >= 0) 621 seq_printf(seq, " node: %d\n", pkt_dev->node); 622 623 seq_puts(seq, " Flags: "); 624 625 if (pkt_dev->flags & F_IPV6) 626 seq_puts(seq, "IPV6 "); 627 628 if (pkt_dev->flags & F_IPSRC_RND) 629 seq_puts(seq, "IPSRC_RND "); 630 631 if (pkt_dev->flags & F_IPDST_RND) 632 seq_puts(seq, "IPDST_RND "); 633 634 if (pkt_dev->flags & F_TXSIZE_RND) 635 seq_puts(seq, "TXSIZE_RND "); 636 637 if (pkt_dev->flags & F_UDPSRC_RND) 638 seq_puts(seq, "UDPSRC_RND "); 639 640 if (pkt_dev->flags & F_UDPDST_RND) 641 seq_puts(seq, "UDPDST_RND "); 642 643 if (pkt_dev->flags & F_UDPCSUM) 644 seq_puts(seq, "UDPCSUM "); 645 646 if (pkt_dev->flags & F_NO_TIMESTAMP) 647 seq_puts(seq, "NO_TIMESTAMP "); 648 649 if (pkt_dev->flags & F_MPLS_RND) 650 seq_puts(seq, "MPLS_RND "); 651 652 if (pkt_dev->flags & F_QUEUE_MAP_RND) 653 seq_puts(seq, "QUEUE_MAP_RND "); 654 655 if (pkt_dev->flags & F_QUEUE_MAP_CPU) 656 seq_puts(seq, "QUEUE_MAP_CPU "); 657 658 if (pkt_dev->cflows) { 659 if (pkt_dev->flags & F_FLOW_SEQ) 660 seq_puts(seq, "FLOW_SEQ "); /*in sequence flows*/ 661 else 662 seq_puts(seq, "FLOW_RND "); 663 } 664 665#ifdef CONFIG_XFRM 666 if (pkt_dev->flags & F_IPSEC_ON) { 667 seq_puts(seq, "IPSEC "); 668 if (pkt_dev->spi) 669 seq_printf(seq, "spi:%u", pkt_dev->spi); 670 } 671#endif 672 673 if (pkt_dev->flags & F_MACSRC_RND) 674 seq_puts(seq, "MACSRC_RND "); 675 676 if (pkt_dev->flags & F_MACDST_RND) 677 seq_puts(seq, "MACDST_RND "); 678 679 if (pkt_dev->flags & F_VID_RND) 680 seq_puts(seq, "VID_RND "); 681 682 if (pkt_dev->flags & F_SVID_RND) 683 seq_puts(seq, "SVID_RND "); 684 685 if (pkt_dev->flags & F_NODE) 686 seq_puts(seq, "NODE_ALLOC "); 687 688 seq_puts(seq, "\n"); 689 690 /* not really stopped, more like last-running-at */ 691 stopped = pkt_dev->running ? ktime_get() : pkt_dev->stopped_at; 692 idle = pkt_dev->idle_acc; 693 do_div(idle, NSEC_PER_USEC); 694 695 seq_printf(seq, 696 "Current:\n pkts-sofar: %llu errors: %llu\n", 697 (unsigned long long)pkt_dev->sofar, 698 (unsigned long long)pkt_dev->errors); 699 700 seq_printf(seq, 701 " started: %lluus stopped: %lluus idle: %lluus\n", 702 (unsigned long long) ktime_to_us(pkt_dev->started_at), 703 (unsigned long long) ktime_to_us(stopped), 704 (unsigned long long) idle); 705 706 seq_printf(seq, 707 " seq_num: %d cur_dst_mac_offset: %d cur_src_mac_offset: %d\n", 708 pkt_dev->seq_num, pkt_dev->cur_dst_mac_offset, 709 pkt_dev->cur_src_mac_offset); 710 711 if (pkt_dev->flags & F_IPV6) { 712 seq_printf(seq, " cur_saddr: %pI6c cur_daddr: %pI6c\n", 713 &pkt_dev->cur_in6_saddr, 714 &pkt_dev->cur_in6_daddr); 715 } else 716 seq_printf(seq, " cur_saddr: %pI4 cur_daddr: %pI4\n", 717 &pkt_dev->cur_saddr, &pkt_dev->cur_daddr); 718 719 seq_printf(seq, " cur_udp_dst: %d cur_udp_src: %d\n", 720 pkt_dev->cur_udp_dst, pkt_dev->cur_udp_src); 721 722 seq_printf(seq, " cur_queue_map: %u\n", pkt_dev->cur_queue_map); 723 724 seq_printf(seq, " flows: %u\n", pkt_dev->nflows); 725 726 if (pkt_dev->result[0]) 727 seq_printf(seq, "Result: %s\n", pkt_dev->result); 728 else 729 seq_puts(seq, "Result: Idle\n"); 730 731 return 0; 732} 733 734 735static int hex32_arg(const char __user *user_buffer, unsigned long maxlen, 736 __u32 *num) 737{ 738 int i = 0; 739 *num = 0; 740 741 for (; i < maxlen; i++) { 742 int value; 743 char c; 744 *num <<= 4; 745 if (get_user(c, &user_buffer[i])) 746 return -EFAULT; 747 value = hex_to_bin(c); 748 if (value >= 0) 749 *num |= value; 750 else 751 break; 752 } 753 return i; 754} 755 756static int count_trail_chars(const char __user * user_buffer, 757 unsigned int maxlen) 758{ 759 int i; 760 761 for (i = 0; i < maxlen; i++) { 762 char c; 763 if (get_user(c, &user_buffer[i])) 764 return -EFAULT; 765 switch (c) { 766 case '\"': 767 case '\n': 768 case '\r': 769 case '\t': 770 case ' ': 771 case '=': 772 break; 773 default: 774 goto done; 775 } 776 } 777done: 778 return i; 779} 780 781static long num_arg(const char __user *user_buffer, unsigned long maxlen, 782 unsigned long *num) 783{ 784 int i; 785 *num = 0; 786 787 for (i = 0; i < maxlen; i++) { 788 char c; 789 if (get_user(c, &user_buffer[i])) 790 return -EFAULT; 791 if ((c >= '0') && (c <= '9')) { 792 *num *= 10; 793 *num += c - '0'; 794 } else 795 break; 796 } 797 return i; 798} 799 800static int strn_len(const char __user * user_buffer, unsigned int maxlen) 801{ 802 int i; 803 804 for (i = 0; i < maxlen; i++) { 805 char c; 806 if (get_user(c, &user_buffer[i])) 807 return -EFAULT; 808 switch (c) { 809 case '\"': 810 case '\n': 811 case '\r': 812 case '\t': 813 case ' ': 814 goto done_str; 815 default: 816 break; 817 } 818 } 819done_str: 820 return i; 821} 822 823static ssize_t get_labels(const char __user *buffer, struct pktgen_dev *pkt_dev) 824{ 825 unsigned int n = 0; 826 char c; 827 ssize_t i = 0; 828 int len; 829 830 pkt_dev->nr_labels = 0; 831 do { 832 __u32 tmp; 833 len = hex32_arg(&buffer[i], 8, &tmp); 834 if (len <= 0) 835 return len; 836 pkt_dev->labels[n] = htonl(tmp); 837 if (pkt_dev->labels[n] & MPLS_STACK_BOTTOM) 838 pkt_dev->flags |= F_MPLS_RND; 839 i += len; 840 if (get_user(c, &buffer[i])) 841 return -EFAULT; 842 i++; 843 n++; 844 if (n >= MAX_MPLS_LABELS) 845 return -E2BIG; 846 } while (c == ','); 847 848 pkt_dev->nr_labels = n; 849 return i; 850} 851 852static ssize_t pktgen_if_write(struct file *file, 853 const char __user * user_buffer, size_t count, 854 loff_t * offset) 855{ 856 struct seq_file *seq = file->private_data; 857 struct pktgen_dev *pkt_dev = seq->private; 858 int i, max, len; 859 char name[16], valstr[32]; 860 unsigned long value = 0; 861 char *pg_result = NULL; 862 int tmp = 0; 863 char buf[128]; 864 865 pg_result = &(pkt_dev->result[0]); 866 867 if (count < 1) { 868 pr_warn("wrong command format\n"); 869 return -EINVAL; 870 } 871 872 max = count; 873 tmp = count_trail_chars(user_buffer, max); 874 if (tmp < 0) { 875 pr_warn("illegal format\n"); 876 return tmp; 877 } 878 i = tmp; 879 880 /* Read variable name */ 881 882 len = strn_len(&user_buffer[i], sizeof(name) - 1); 883 if (len < 0) 884 return len; 885 886 memset(name, 0, sizeof(name)); 887 if (copy_from_user(name, &user_buffer[i], len)) 888 return -EFAULT; 889 i += len; 890 891 max = count - i; 892 len = count_trail_chars(&user_buffer[i], max); 893 if (len < 0) 894 return len; 895 896 i += len; 897 898 if (debug) { 899 size_t copy = min_t(size_t, count, 1023); 900 char tb[copy + 1]; 901 if (copy_from_user(tb, user_buffer, copy)) 902 return -EFAULT; 903 tb[copy] = 0; 904 pr_debug("%s,%lu buffer -:%s:-\n", 905 name, (unsigned long)count, tb); 906 } 907 908 if (!strcmp(name, "min_pkt_size")) { 909 len = num_arg(&user_buffer[i], 10, &value); 910 if (len < 0) 911 return len; 912 913 i += len; 914 if (value < 14 + 20 + 8) 915 value = 14 + 20 + 8; 916 if (value != pkt_dev->min_pkt_size) { 917 pkt_dev->min_pkt_size = value; 918 pkt_dev->cur_pkt_size = value; 919 } 920 sprintf(pg_result, "OK: min_pkt_size=%u", 921 pkt_dev->min_pkt_size); 922 return count; 923 } 924 925 if (!strcmp(name, "max_pkt_size")) { 926 len = num_arg(&user_buffer[i], 10, &value); 927 if (len < 0) 928 return len; 929 930 i += len; 931 if (value < 14 + 20 + 8) 932 value = 14 + 20 + 8; 933 if (value != pkt_dev->max_pkt_size) { 934 pkt_dev->max_pkt_size = value; 935 pkt_dev->cur_pkt_size = value; 936 } 937 sprintf(pg_result, "OK: max_pkt_size=%u", 938 pkt_dev->max_pkt_size); 939 return count; 940 } 941 942 /* Shortcut for min = max */ 943 944 if (!strcmp(name, "pkt_size")) { 945 len = num_arg(&user_buffer[i], 10, &value); 946 if (len < 0) 947 return len; 948 949 i += len; 950 if (value < 14 + 20 + 8) 951 value = 14 + 20 + 8; 952 if (value != pkt_dev->min_pkt_size) { 953 pkt_dev->min_pkt_size = value; 954 pkt_dev->max_pkt_size = value; 955 pkt_dev->cur_pkt_size = value; 956 } 957 sprintf(pg_result, "OK: pkt_size=%u", pkt_dev->min_pkt_size); 958 return count; 959 } 960 961 if (!strcmp(name, "debug")) { 962 len = num_arg(&user_buffer[i], 10, &value); 963 if (len < 0) 964 return len; 965 966 i += len; 967 debug = value; 968 sprintf(pg_result, "OK: debug=%u", debug); 969 return count; 970 } 971 972 if (!strcmp(name, "frags")) { 973 len = num_arg(&user_buffer[i], 10, &value); 974 if (len < 0) 975 return len; 976 977 i += len; 978 pkt_dev->nfrags = value; 979 sprintf(pg_result, "OK: frags=%u", pkt_dev->nfrags); 980 return count; 981 } 982 if (!strcmp(name, "delay")) { 983 len = num_arg(&user_buffer[i], 10, &value); 984 if (len < 0) 985 return len; 986 987 i += len; 988 if (value == 0x7FFFFFFF) 989 pkt_dev->delay = ULLONG_MAX; 990 else 991 pkt_dev->delay = (u64)value; 992 993 sprintf(pg_result, "OK: delay=%llu", 994 (unsigned long long) pkt_dev->delay); 995 return count; 996 } 997 if (!strcmp(name, "rate")) { 998 len = num_arg(&user_buffer[i], 10, &value); 999 if (len < 0) 1000 return len; 1001 1002 i += len; 1003 if (!value) 1004 return len; 1005 pkt_dev->delay = pkt_dev->min_pkt_size*8*NSEC_PER_USEC/value; 1006 if (debug) 1007 pr_info("Delay set at: %llu ns\n", pkt_dev->delay); 1008 1009 sprintf(pg_result, "OK: rate=%lu", value); 1010 return count; 1011 } 1012 if (!strcmp(name, "ratep")) { 1013 len = num_arg(&user_buffer[i], 10, &value); 1014 if (len < 0) 1015 return len; 1016 1017 i += len; 1018 if (!value) 1019 return len; 1020 pkt_dev->delay = NSEC_PER_SEC/value; 1021 if (debug) 1022 pr_info("Delay set at: %llu ns\n", pkt_dev->delay); 1023 1024 sprintf(pg_result, "OK: rate=%lu", value); 1025 return count; 1026 } 1027 if (!strcmp(name, "udp_src_min")) { 1028 len = num_arg(&user_buffer[i], 10, &value); 1029 if (len < 0) 1030 return len; 1031 1032 i += len; 1033 if (value != pkt_dev->udp_src_min) { 1034 pkt_dev->udp_src_min = value; 1035 pkt_dev->cur_udp_src = value; 1036 } 1037 sprintf(pg_result, "OK: udp_src_min=%u", pkt_dev->udp_src_min); 1038 return count; 1039 } 1040 if (!strcmp(name, "udp_dst_min")) { 1041 len = num_arg(&user_buffer[i], 10, &value); 1042 if (len < 0) 1043 return len; 1044 1045 i += len; 1046 if (value != pkt_dev->udp_dst_min) { 1047 pkt_dev->udp_dst_min = value; 1048 pkt_dev->cur_udp_dst = value; 1049 } 1050 sprintf(pg_result, "OK: udp_dst_min=%u", pkt_dev->udp_dst_min); 1051 return count; 1052 } 1053 if (!strcmp(name, "udp_src_max")) { 1054 len = num_arg(&user_buffer[i], 10, &value); 1055 if (len < 0) 1056 return len; 1057 1058 i += len; 1059 if (value != pkt_dev->udp_src_max) { 1060 pkt_dev->udp_src_max = value; 1061 pkt_dev->cur_udp_src = value; 1062 } 1063 sprintf(pg_result, "OK: udp_src_max=%u", pkt_dev->udp_src_max); 1064 return count; 1065 } 1066 if (!strcmp(name, "udp_dst_max")) { 1067 len = num_arg(&user_buffer[i], 10, &value); 1068 if (len < 0) 1069 return len; 1070 1071 i += len; 1072 if (value != pkt_dev->udp_dst_max) { 1073 pkt_dev->udp_dst_max = value; 1074 pkt_dev->cur_udp_dst = value; 1075 } 1076 sprintf(pg_result, "OK: udp_dst_max=%u", pkt_dev->udp_dst_max); 1077 return count; 1078 } 1079 if (!strcmp(name, "clone_skb")) { 1080 len = num_arg(&user_buffer[i], 10, &value); 1081 if (len < 0) 1082 return len; 1083 if ((value > 0) && 1084 (!(pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING))) 1085 return -ENOTSUPP; 1086 i += len; 1087 pkt_dev->clone_skb = value; 1088 1089 sprintf(pg_result, "OK: clone_skb=%d", pkt_dev->clone_skb); 1090 return count; 1091 } 1092 if (!strcmp(name, "count")) { 1093 len = num_arg(&user_buffer[i], 10, &value); 1094 if (len < 0) 1095 return len; 1096 1097 i += len; 1098 pkt_dev->count = value; 1099 sprintf(pg_result, "OK: count=%llu", 1100 (unsigned long long)pkt_dev->count); 1101 return count; 1102 } 1103 if (!strcmp(name, "src_mac_count")) { 1104 len = num_arg(&user_buffer[i], 10, &value); 1105 if (len < 0) 1106 return len; 1107 1108 i += len; 1109 if (pkt_dev->src_mac_count != value) { 1110 pkt_dev->src_mac_count = value; 1111 pkt_dev->cur_src_mac_offset = 0; 1112 } 1113 sprintf(pg_result, "OK: src_mac_count=%d", 1114 pkt_dev->src_mac_count); 1115 return count; 1116 } 1117 if (!strcmp(name, "dst_mac_count")) { 1118 len = num_arg(&user_buffer[i], 10, &value); 1119 if (len < 0) 1120 return len; 1121 1122 i += len; 1123 if (pkt_dev->dst_mac_count != value) { 1124 pkt_dev->dst_mac_count = value; 1125 pkt_dev->cur_dst_mac_offset = 0; 1126 } 1127 sprintf(pg_result, "OK: dst_mac_count=%d", 1128 pkt_dev->dst_mac_count); 1129 return count; 1130 } 1131 if (!strcmp(name, "burst")) { 1132 len = num_arg(&user_buffer[i], 10, &value); 1133 if (len < 0) 1134 return len; 1135 1136 i += len; 1137 pkt_dev->burst = value < 1 ? 1 : value; 1138 sprintf(pg_result, "OK: burst=%d", pkt_dev->burst); 1139 return count; 1140 } 1141 if (!strcmp(name, "node")) { 1142 len = num_arg(&user_buffer[i], 10, &value); 1143 if (len < 0) 1144 return len; 1145 1146 i += len; 1147 1148 if (node_possible(value)) { 1149 pkt_dev->node = value; 1150 sprintf(pg_result, "OK: node=%d", pkt_dev->node); 1151 if (pkt_dev->page) { 1152 put_page(pkt_dev->page); 1153 pkt_dev->page = NULL; 1154 } 1155 } 1156 else 1157 sprintf(pg_result, "ERROR: node not possible"); 1158 return count; 1159 } 1160 if (!strcmp(name, "flag")) { 1161 char f[32]; 1162 memset(f, 0, 32); 1163 len = strn_len(&user_buffer[i], sizeof(f) - 1); 1164 if (len < 0) 1165 return len; 1166 1167 if (copy_from_user(f, &user_buffer[i], len)) 1168 return -EFAULT; 1169 i += len; 1170 if (strcmp(f, "IPSRC_RND") == 0) 1171 pkt_dev->flags |= F_IPSRC_RND; 1172 1173 else if (strcmp(f, "!IPSRC_RND") == 0) 1174 pkt_dev->flags &= ~F_IPSRC_RND; 1175 1176 else if (strcmp(f, "TXSIZE_RND") == 0) 1177 pkt_dev->flags |= F_TXSIZE_RND; 1178 1179 else if (strcmp(f, "!TXSIZE_RND") == 0) 1180 pkt_dev->flags &= ~F_TXSIZE_RND; 1181 1182 else if (strcmp(f, "IPDST_RND") == 0) 1183 pkt_dev->flags |= F_IPDST_RND; 1184 1185 else if (strcmp(f, "!IPDST_RND") == 0) 1186 pkt_dev->flags &= ~F_IPDST_RND; 1187 1188 else if (strcmp(f, "UDPSRC_RND") == 0) 1189 pkt_dev->flags |= F_UDPSRC_RND; 1190 1191 else if (strcmp(f, "!UDPSRC_RND") == 0) 1192 pkt_dev->flags &= ~F_UDPSRC_RND; 1193 1194 else if (strcmp(f, "UDPDST_RND") == 0) 1195 pkt_dev->flags |= F_UDPDST_RND; 1196 1197 else if (strcmp(f, "!UDPDST_RND") == 0) 1198 pkt_dev->flags &= ~F_UDPDST_RND; 1199 1200 else if (strcmp(f, "MACSRC_RND") == 0) 1201 pkt_dev->flags |= F_MACSRC_RND; 1202 1203 else if (strcmp(f, "!MACSRC_RND") == 0) 1204 pkt_dev->flags &= ~F_MACSRC_RND; 1205 1206 else if (strcmp(f, "MACDST_RND") == 0) 1207 pkt_dev->flags |= F_MACDST_RND; 1208 1209 else if (strcmp(f, "!MACDST_RND") == 0) 1210 pkt_dev->flags &= ~F_MACDST_RND; 1211 1212 else if (strcmp(f, "MPLS_RND") == 0) 1213 pkt_dev->flags |= F_MPLS_RND; 1214 1215 else if (strcmp(f, "!MPLS_RND") == 0) 1216 pkt_dev->flags &= ~F_MPLS_RND; 1217 1218 else if (strcmp(f, "VID_RND") == 0) 1219 pkt_dev->flags |= F_VID_RND; 1220 1221 else if (strcmp(f, "!VID_RND") == 0) 1222 pkt_dev->flags &= ~F_VID_RND; 1223 1224 else if (strcmp(f, "SVID_RND") == 0) 1225 pkt_dev->flags |= F_SVID_RND; 1226 1227 else if (strcmp(f, "!SVID_RND") == 0) 1228 pkt_dev->flags &= ~F_SVID_RND; 1229 1230 else if (strcmp(f, "FLOW_SEQ") == 0) 1231 pkt_dev->flags |= F_FLOW_SEQ; 1232 1233 else if (strcmp(f, "QUEUE_MAP_RND") == 0) 1234 pkt_dev->flags |= F_QUEUE_MAP_RND; 1235 1236 else if (strcmp(f, "!QUEUE_MAP_RND") == 0) 1237 pkt_dev->flags &= ~F_QUEUE_MAP_RND; 1238 1239 else if (strcmp(f, "QUEUE_MAP_CPU") == 0) 1240 pkt_dev->flags |= F_QUEUE_MAP_CPU; 1241 1242 else if (strcmp(f, "!QUEUE_MAP_CPU") == 0) 1243 pkt_dev->flags &= ~F_QUEUE_MAP_CPU; 1244#ifdef CONFIG_XFRM 1245 else if (strcmp(f, "IPSEC") == 0) 1246 pkt_dev->flags |= F_IPSEC_ON; 1247#endif 1248 1249 else if (strcmp(f, "!IPV6") == 0) 1250 pkt_dev->flags &= ~F_IPV6; 1251 1252 else if (strcmp(f, "NODE_ALLOC") == 0) 1253 pkt_dev->flags |= F_NODE; 1254 1255 else if (strcmp(f, "!NODE_ALLOC") == 0) 1256 pkt_dev->flags &= ~F_NODE; 1257 1258 else if (strcmp(f, "UDPCSUM") == 0) 1259 pkt_dev->flags |= F_UDPCSUM; 1260 1261 else if (strcmp(f, "!UDPCSUM") == 0) 1262 pkt_dev->flags &= ~F_UDPCSUM; 1263 1264 else if (strcmp(f, "NO_TIMESTAMP") == 0) 1265 pkt_dev->flags |= F_NO_TIMESTAMP; 1266 1267 else { 1268 sprintf(pg_result, 1269 "Flag -:%s:- unknown\nAvailable flags, (prepend ! to un-set flag):\n%s", 1270 f, 1271 "IPSRC_RND, IPDST_RND, UDPSRC_RND, UDPDST_RND, " 1272 "MACSRC_RND, MACDST_RND, TXSIZE_RND, IPV6, " 1273 "MPLS_RND, VID_RND, SVID_RND, FLOW_SEQ, " 1274 "QUEUE_MAP_RND, QUEUE_MAP_CPU, UDPCSUM, " 1275 "NO_TIMESTAMP, " 1276#ifdef CONFIG_XFRM 1277 "IPSEC, " 1278#endif 1279 "NODE_ALLOC\n"); 1280 return count; 1281 } 1282 sprintf(pg_result, "OK: flags=0x%x", pkt_dev->flags); 1283 return count; 1284 } 1285 if (!strcmp(name, "dst_min") || !strcmp(name, "dst")) { 1286 len = strn_len(&user_buffer[i], sizeof(pkt_dev->dst_min) - 1); 1287 if (len < 0) 1288 return len; 1289 1290 if (copy_from_user(buf, &user_buffer[i], len)) 1291 return -EFAULT; 1292 buf[len] = 0; 1293 if (strcmp(buf, pkt_dev->dst_min) != 0) { 1294 memset(pkt_dev->dst_min, 0, sizeof(pkt_dev->dst_min)); 1295 strncpy(pkt_dev->dst_min, buf, len); 1296 pkt_dev->daddr_min = in_aton(pkt_dev->dst_min); 1297 pkt_dev->cur_daddr = pkt_dev->daddr_min; 1298 } 1299 if (debug) 1300 pr_debug("dst_min set to: %s\n", pkt_dev->dst_min); 1301 i += len; 1302 sprintf(pg_result, "OK: dst_min=%s", pkt_dev->dst_min); 1303 return count; 1304 } 1305 if (!strcmp(name, "dst_max")) { 1306 len = strn_len(&user_buffer[i], sizeof(pkt_dev->dst_max) - 1); 1307 if (len < 0) 1308 return len; 1309 1310 1311 if (copy_from_user(buf, &user_buffer[i], len)) 1312 return -EFAULT; 1313 1314 buf[len] = 0; 1315 if (strcmp(buf, pkt_dev->dst_max) != 0) { 1316 memset(pkt_dev->dst_max, 0, sizeof(pkt_dev->dst_max)); 1317 strncpy(pkt_dev->dst_max, buf, len); 1318 pkt_dev->daddr_max = in_aton(pkt_dev->dst_max); 1319 pkt_dev->cur_daddr = pkt_dev->daddr_max; 1320 } 1321 if (debug) 1322 pr_debug("dst_max set to: %s\n", pkt_dev->dst_max); 1323 i += len; 1324 sprintf(pg_result, "OK: dst_max=%s", pkt_dev->dst_max); 1325 return count; 1326 } 1327 if (!strcmp(name, "dst6")) { 1328 len = strn_len(&user_buffer[i], sizeof(buf) - 1); 1329 if (len < 0) 1330 return len; 1331 1332 pkt_dev->flags |= F_IPV6; 1333 1334 if (copy_from_user(buf, &user_buffer[i], len)) 1335 return -EFAULT; 1336 buf[len] = 0; 1337 1338 in6_pton(buf, -1, pkt_dev->in6_daddr.s6_addr, -1, NULL); 1339 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->in6_daddr); 1340 1341 pkt_dev->cur_in6_daddr = pkt_dev->in6_daddr; 1342 1343 if (debug) 1344 pr_debug("dst6 set to: %s\n", buf); 1345 1346 i += len; 1347 sprintf(pg_result, "OK: dst6=%s", buf); 1348 return count; 1349 } 1350 if (!strcmp(name, "dst6_min")) { 1351 len = strn_len(&user_buffer[i], sizeof(buf) - 1); 1352 if (len < 0) 1353 return len; 1354 1355 pkt_dev->flags |= F_IPV6; 1356 1357 if (copy_from_user(buf, &user_buffer[i], len)) 1358 return -EFAULT; 1359 buf[len] = 0; 1360 1361 in6_pton(buf, -1, pkt_dev->min_in6_daddr.s6_addr, -1, NULL); 1362 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->min_in6_daddr); 1363 1364 pkt_dev->cur_in6_daddr = pkt_dev->min_in6_daddr; 1365 if (debug) 1366 pr_debug("dst6_min set to: %s\n", buf); 1367 1368 i += len; 1369 sprintf(pg_result, "OK: dst6_min=%s", buf); 1370 return count; 1371 } 1372 if (!strcmp(name, "dst6_max")) { 1373 len = strn_len(&user_buffer[i], sizeof(buf) - 1); 1374 if (len < 0) 1375 return len; 1376 1377 pkt_dev->flags |= F_IPV6; 1378 1379 if (copy_from_user(buf, &user_buffer[i], len)) 1380 return -EFAULT; 1381 buf[len] = 0; 1382 1383 in6_pton(buf, -1, pkt_dev->max_in6_daddr.s6_addr, -1, NULL); 1384 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->max_in6_daddr); 1385 1386 if (debug) 1387 pr_debug("dst6_max set to: %s\n", buf); 1388 1389 i += len; 1390 sprintf(pg_result, "OK: dst6_max=%s", buf); 1391 return count; 1392 } 1393 if (!strcmp(name, "src6")) { 1394 len = strn_len(&user_buffer[i], sizeof(buf) - 1); 1395 if (len < 0) 1396 return len; 1397 1398 pkt_dev->flags |= F_IPV6; 1399 1400 if (copy_from_user(buf, &user_buffer[i], len)) 1401 return -EFAULT; 1402 buf[len] = 0; 1403 1404 in6_pton(buf, -1, pkt_dev->in6_saddr.s6_addr, -1, NULL); 1405 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->in6_saddr); 1406 1407 pkt_dev->cur_in6_saddr = pkt_dev->in6_saddr; 1408 1409 if (debug) 1410 pr_debug("src6 set to: %s\n", buf); 1411 1412 i += len; 1413 sprintf(pg_result, "OK: src6=%s", buf); 1414 return count; 1415 } 1416 if (!strcmp(name, "src_min")) { 1417 len = strn_len(&user_buffer[i], sizeof(pkt_dev->src_min) - 1); 1418 if (len < 0) 1419 return len; 1420 1421 if (copy_from_user(buf, &user_buffer[i], len)) 1422 return -EFAULT; 1423 buf[len] = 0; 1424 if (strcmp(buf, pkt_dev->src_min) != 0) { 1425 memset(pkt_dev->src_min, 0, sizeof(pkt_dev->src_min)); 1426 strncpy(pkt_dev->src_min, buf, len); 1427 pkt_dev->saddr_min = in_aton(pkt_dev->src_min); 1428 pkt_dev->cur_saddr = pkt_dev->saddr_min; 1429 } 1430 if (debug) 1431 pr_debug("src_min set to: %s\n", pkt_dev->src_min); 1432 i += len; 1433 sprintf(pg_result, "OK: src_min=%s", pkt_dev->src_min); 1434 return count; 1435 } 1436 if (!strcmp(name, "src_max")) { 1437 len = strn_len(&user_buffer[i], sizeof(pkt_dev->src_max) - 1); 1438 if (len < 0) 1439 return len; 1440 1441 if (copy_from_user(buf, &user_buffer[i], len)) 1442 return -EFAULT; 1443 buf[len] = 0; 1444 if (strcmp(buf, pkt_dev->src_max) != 0) { 1445 memset(pkt_dev->src_max, 0, sizeof(pkt_dev->src_max)); 1446 strncpy(pkt_dev->src_max, buf, len); 1447 pkt_dev->saddr_max = in_aton(pkt_dev->src_max); 1448 pkt_dev->cur_saddr = pkt_dev->saddr_max; 1449 } 1450 if (debug) 1451 pr_debug("src_max set to: %s\n", pkt_dev->src_max); 1452 i += len; 1453 sprintf(pg_result, "OK: src_max=%s", pkt_dev->src_max); 1454 return count; 1455 } 1456 if (!strcmp(name, "dst_mac")) { 1457 len = strn_len(&user_buffer[i], sizeof(valstr) - 1); 1458 if (len < 0) 1459 return len; 1460 1461 memset(valstr, 0, sizeof(valstr)); 1462 if (copy_from_user(valstr, &user_buffer[i], len)) 1463 return -EFAULT; 1464 1465 if (!mac_pton(valstr, pkt_dev->dst_mac)) 1466 return -EINVAL; 1467 /* Set up Dest MAC */ 1468 ether_addr_copy(&pkt_dev->hh[0], pkt_dev->dst_mac); 1469 1470 sprintf(pg_result, "OK: dstmac %pM", pkt_dev->dst_mac); 1471 return count; 1472 } 1473 if (!strcmp(name, "src_mac")) { 1474 len = strn_len(&user_buffer[i], sizeof(valstr) - 1); 1475 if (len < 0) 1476 return len; 1477 1478 memset(valstr, 0, sizeof(valstr)); 1479 if (copy_from_user(valstr, &user_buffer[i], len)) 1480 return -EFAULT; 1481 1482 if (!mac_pton(valstr, pkt_dev->src_mac)) 1483 return -EINVAL; 1484 /* Set up Src MAC */ 1485 ether_addr_copy(&pkt_dev->hh[6], pkt_dev->src_mac); 1486 1487 sprintf(pg_result, "OK: srcmac %pM", pkt_dev->src_mac); 1488 return count; 1489 } 1490 1491 if (!strcmp(name, "clear_counters")) { 1492 pktgen_clear_counters(pkt_dev); 1493 sprintf(pg_result, "OK: Clearing counters.\n"); 1494 return count; 1495 } 1496 1497 if (!strcmp(name, "flows")) { 1498 len = num_arg(&user_buffer[i], 10, &value); 1499 if (len < 0) 1500 return len; 1501 1502 i += len; 1503 if (value > MAX_CFLOWS) 1504 value = MAX_CFLOWS; 1505 1506 pkt_dev->cflows = value; 1507 sprintf(pg_result, "OK: flows=%u", pkt_dev->cflows); 1508 return count; 1509 } 1510#ifdef CONFIG_XFRM 1511 if (!strcmp(name, "spi")) { 1512 len = num_arg(&user_buffer[i], 10, &value); 1513 if (len < 0) 1514 return len; 1515 1516 i += len; 1517 pkt_dev->spi = value; 1518 sprintf(pg_result, "OK: spi=%u", pkt_dev->spi); 1519 return count; 1520 } 1521#endif 1522 if (!strcmp(name, "flowlen")) { 1523 len = num_arg(&user_buffer[i], 10, &value); 1524 if (len < 0) 1525 return len; 1526 1527 i += len; 1528 pkt_dev->lflow = value; 1529 sprintf(pg_result, "OK: flowlen=%u", pkt_dev->lflow); 1530 return count; 1531 } 1532 1533 if (!strcmp(name, "queue_map_min")) { 1534 len = num_arg(&user_buffer[i], 5, &value); 1535 if (len < 0) 1536 return len; 1537 1538 i += len; 1539 pkt_dev->queue_map_min = value; 1540 sprintf(pg_result, "OK: queue_map_min=%u", pkt_dev->queue_map_min); 1541 return count; 1542 } 1543 1544 if (!strcmp(name, "queue_map_max")) { 1545 len = num_arg(&user_buffer[i], 5, &value); 1546 if (len < 0) 1547 return len; 1548 1549 i += len; 1550 pkt_dev->queue_map_max = value; 1551 sprintf(pg_result, "OK: queue_map_max=%u", pkt_dev->queue_map_max); 1552 return count; 1553 } 1554 1555 if (!strcmp(name, "mpls")) { 1556 unsigned int n, cnt; 1557 1558 len = get_labels(&user_buffer[i], pkt_dev); 1559 if (len < 0) 1560 return len; 1561 i += len; 1562 cnt = sprintf(pg_result, "OK: mpls="); 1563 for (n = 0; n < pkt_dev->nr_labels; n++) 1564 cnt += sprintf(pg_result + cnt, 1565 "%08x%s", ntohl(pkt_dev->labels[n]), 1566 n == pkt_dev->nr_labels-1 ? "" : ","); 1567 1568 if (pkt_dev->nr_labels && pkt_dev->vlan_id != 0xffff) { 1569 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */ 1570 pkt_dev->svlan_id = 0xffff; 1571 1572 if (debug) 1573 pr_debug("VLAN/SVLAN auto turned off\n"); 1574 } 1575 return count; 1576 } 1577 1578 if (!strcmp(name, "vlan_id")) { 1579 len = num_arg(&user_buffer[i], 4, &value); 1580 if (len < 0) 1581 return len; 1582 1583 i += len; 1584 if (value <= 4095) { 1585 pkt_dev->vlan_id = value; /* turn on VLAN */ 1586 1587 if (debug) 1588 pr_debug("VLAN turned on\n"); 1589 1590 if (debug && pkt_dev->nr_labels) 1591 pr_debug("MPLS auto turned off\n"); 1592 1593 pkt_dev->nr_labels = 0; /* turn off MPLS */ 1594 sprintf(pg_result, "OK: vlan_id=%u", pkt_dev->vlan_id); 1595 } else { 1596 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */ 1597 pkt_dev->svlan_id = 0xffff; 1598 1599 if (debug) 1600 pr_debug("VLAN/SVLAN turned off\n"); 1601 } 1602 return count; 1603 } 1604 1605 if (!strcmp(name, "vlan_p")) { 1606 len = num_arg(&user_buffer[i], 1, &value); 1607 if (len < 0) 1608 return len; 1609 1610 i += len; 1611 if ((value <= 7) && (pkt_dev->vlan_id != 0xffff)) { 1612 pkt_dev->vlan_p = value; 1613 sprintf(pg_result, "OK: vlan_p=%u", pkt_dev->vlan_p); 1614 } else { 1615 sprintf(pg_result, "ERROR: vlan_p must be 0-7"); 1616 } 1617 return count; 1618 } 1619 1620 if (!strcmp(name, "vlan_cfi")) { 1621 len = num_arg(&user_buffer[i], 1, &value); 1622 if (len < 0) 1623 return len; 1624 1625 i += len; 1626 if ((value <= 1) && (pkt_dev->vlan_id != 0xffff)) { 1627 pkt_dev->vlan_cfi = value; 1628 sprintf(pg_result, "OK: vlan_cfi=%u", pkt_dev->vlan_cfi); 1629 } else { 1630 sprintf(pg_result, "ERROR: vlan_cfi must be 0-1"); 1631 } 1632 return count; 1633 } 1634 1635 if (!strcmp(name, "svlan_id")) { 1636 len = num_arg(&user_buffer[i], 4, &value); 1637 if (len < 0) 1638 return len; 1639 1640 i += len; 1641 if ((value <= 4095) && ((pkt_dev->vlan_id != 0xffff))) { 1642 pkt_dev->svlan_id = value; /* turn on SVLAN */ 1643 1644 if (debug) 1645 pr_debug("SVLAN turned on\n"); 1646 1647 if (debug && pkt_dev->nr_labels) 1648 pr_debug("MPLS auto turned off\n"); 1649 1650 pkt_dev->nr_labels = 0; /* turn off MPLS */ 1651 sprintf(pg_result, "OK: svlan_id=%u", pkt_dev->svlan_id); 1652 } else { 1653 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */ 1654 pkt_dev->svlan_id = 0xffff; 1655 1656 if (debug) 1657 pr_debug("VLAN/SVLAN turned off\n"); 1658 } 1659 return count; 1660 } 1661 1662 if (!strcmp(name, "svlan_p")) { 1663 len = num_arg(&user_buffer[i], 1, &value); 1664 if (len < 0) 1665 return len; 1666 1667 i += len; 1668 if ((value <= 7) && (pkt_dev->svlan_id != 0xffff)) { 1669 pkt_dev->svlan_p = value; 1670 sprintf(pg_result, "OK: svlan_p=%u", pkt_dev->svlan_p); 1671 } else { 1672 sprintf(pg_result, "ERROR: svlan_p must be 0-7"); 1673 } 1674 return count; 1675 } 1676 1677 if (!strcmp(name, "svlan_cfi")) { 1678 len = num_arg(&user_buffer[i], 1, &value); 1679 if (len < 0) 1680 return len; 1681 1682 i += len; 1683 if ((value <= 1) && (pkt_dev->svlan_id != 0xffff)) { 1684 pkt_dev->svlan_cfi = value; 1685 sprintf(pg_result, "OK: svlan_cfi=%u", pkt_dev->svlan_cfi); 1686 } else { 1687 sprintf(pg_result, "ERROR: svlan_cfi must be 0-1"); 1688 } 1689 return count; 1690 } 1691 1692 if (!strcmp(name, "tos")) { 1693 __u32 tmp_value = 0; 1694 len = hex32_arg(&user_buffer[i], 2, &tmp_value); 1695 if (len < 0) 1696 return len; 1697 1698 i += len; 1699 if (len == 2) { 1700 pkt_dev->tos = tmp_value; 1701 sprintf(pg_result, "OK: tos=0x%02x", pkt_dev->tos); 1702 } else { 1703 sprintf(pg_result, "ERROR: tos must be 00-ff"); 1704 } 1705 return count; 1706 } 1707 1708 if (!strcmp(name, "traffic_class")) { 1709 __u32 tmp_value = 0; 1710 len = hex32_arg(&user_buffer[i], 2, &tmp_value); 1711 if (len < 0) 1712 return len; 1713 1714 i += len; 1715 if (len == 2) { 1716 pkt_dev->traffic_class = tmp_value; 1717 sprintf(pg_result, "OK: traffic_class=0x%02x", pkt_dev->traffic_class); 1718 } else { 1719 sprintf(pg_result, "ERROR: traffic_class must be 00-ff"); 1720 } 1721 return count; 1722 } 1723 1724 if (!strcmp(name, "skb_priority")) { 1725 len = num_arg(&user_buffer[i], 9, &value); 1726 if (len < 0) 1727 return len; 1728 1729 i += len; 1730 pkt_dev->skb_priority = value; 1731 sprintf(pg_result, "OK: skb_priority=%i", 1732 pkt_dev->skb_priority); 1733 return count; 1734 } 1735 1736 sprintf(pkt_dev->result, "No such parameter \"%s\"", name); 1737 return -EINVAL; 1738} 1739 1740static int pktgen_if_open(struct inode *inode, struct file *file) 1741{ 1742 return single_open(file, pktgen_if_show, PDE_DATA(inode)); 1743} 1744 1745static const struct file_operations pktgen_if_fops = { 1746 .owner = THIS_MODULE, 1747 .open = pktgen_if_open, 1748 .read = seq_read, 1749 .llseek = seq_lseek, 1750 .write = pktgen_if_write, 1751 .release = single_release, 1752}; 1753 1754static int pktgen_thread_show(struct seq_file *seq, void *v) 1755{ 1756 struct pktgen_thread *t = seq->private; 1757 const struct pktgen_dev *pkt_dev; 1758 1759 BUG_ON(!t); 1760 1761 seq_puts(seq, "Running: "); 1762 1763 rcu_read_lock(); 1764 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) 1765 if (pkt_dev->running) 1766 seq_printf(seq, "%s ", pkt_dev->odevname); 1767 1768 seq_puts(seq, "\nStopped: "); 1769 1770 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) 1771 if (!pkt_dev->running) 1772 seq_printf(seq, "%s ", pkt_dev->odevname); 1773 1774 if (t->result[0]) 1775 seq_printf(seq, "\nResult: %s\n", t->result); 1776 else 1777 seq_puts(seq, "\nResult: NA\n"); 1778 1779 rcu_read_unlock(); 1780 1781 return 0; 1782} 1783 1784static ssize_t pktgen_thread_write(struct file *file, 1785 const char __user * user_buffer, 1786 size_t count, loff_t * offset) 1787{ 1788 struct seq_file *seq = file->private_data; 1789 struct pktgen_thread *t = seq->private; 1790 int i, max, len, ret; 1791 char name[40]; 1792 char *pg_result; 1793 1794 if (count < 1) { 1795 // sprintf(pg_result, "Wrong command format"); 1796 return -EINVAL; 1797 } 1798 1799 max = count; 1800 len = count_trail_chars(user_buffer, max); 1801 if (len < 0) 1802 return len; 1803 1804 i = len; 1805 1806 /* Read variable name */ 1807 1808 len = strn_len(&user_buffer[i], sizeof(name) - 1); 1809 if (len < 0) 1810 return len; 1811 1812 memset(name, 0, sizeof(name)); 1813 if (copy_from_user(name, &user_buffer[i], len)) 1814 return -EFAULT; 1815 i += len; 1816 1817 max = count - i; 1818 len = count_trail_chars(&user_buffer[i], max); 1819 if (len < 0) 1820 return len; 1821 1822 i += len; 1823 1824 if (debug) 1825 pr_debug("t=%s, count=%lu\n", name, (unsigned long)count); 1826 1827 if (!t) { 1828 pr_err("ERROR: No thread\n"); 1829 ret = -EINVAL; 1830 goto out; 1831 } 1832 1833 pg_result = &(t->result[0]); 1834 1835 if (!strcmp(name, "add_device")) { 1836 char f[32]; 1837 memset(f, 0, 32); 1838 len = strn_len(&user_buffer[i], sizeof(f) - 1); 1839 if (len < 0) { 1840 ret = len; 1841 goto out; 1842 } 1843 if (copy_from_user(f, &user_buffer[i], len)) 1844 return -EFAULT; 1845 i += len; 1846 mutex_lock(&pktgen_thread_lock); 1847 ret = pktgen_add_device(t, f); 1848 mutex_unlock(&pktgen_thread_lock); 1849 if (!ret) { 1850 ret = count; 1851 sprintf(pg_result, "OK: add_device=%s", f); 1852 } else 1853 sprintf(pg_result, "ERROR: can not add device %s", f); 1854 goto out; 1855 } 1856 1857 if (!strcmp(name, "rem_device_all")) { 1858 mutex_lock(&pktgen_thread_lock); 1859 t->control |= T_REMDEVALL; 1860 mutex_unlock(&pktgen_thread_lock); 1861 schedule_timeout_interruptible(msecs_to_jiffies(125)); /* Propagate thread->control */ 1862 ret = count; 1863 sprintf(pg_result, "OK: rem_device_all"); 1864 goto out; 1865 } 1866 1867 if (!strcmp(name, "max_before_softirq")) { 1868 sprintf(pg_result, "OK: Note! max_before_softirq is obsoleted -- Do not use"); 1869 ret = count; 1870 goto out; 1871 } 1872 1873 ret = -EINVAL; 1874out: 1875 return ret; 1876} 1877 1878static int pktgen_thread_open(struct inode *inode, struct file *file) 1879{ 1880 return single_open(file, pktgen_thread_show, PDE_DATA(inode)); 1881} 1882 1883static const struct file_operations pktgen_thread_fops = { 1884 .owner = THIS_MODULE, 1885 .open = pktgen_thread_open, 1886 .read = seq_read, 1887 .llseek = seq_lseek, 1888 .write = pktgen_thread_write, 1889 .release = single_release, 1890}; 1891 1892/* Think find or remove for NN */ 1893static struct pktgen_dev *__pktgen_NN_threads(const struct pktgen_net *pn, 1894 const char *ifname, int remove) 1895{ 1896 struct pktgen_thread *t; 1897 struct pktgen_dev *pkt_dev = NULL; 1898 bool exact = (remove == FIND); 1899 1900 list_for_each_entry(t, &pn->pktgen_threads, th_list) { 1901 pkt_dev = pktgen_find_dev(t, ifname, exact); 1902 if (pkt_dev) { 1903 if (remove) { 1904 pkt_dev->removal_mark = 1; 1905 t->control |= T_REMDEV; 1906 } 1907 break; 1908 } 1909 } 1910 return pkt_dev; 1911} 1912 1913/* 1914 * mark a device for removal 1915 */ 1916static void pktgen_mark_device(const struct pktgen_net *pn, const char *ifname) 1917{ 1918 struct pktgen_dev *pkt_dev = NULL; 1919 const int max_tries = 10, msec_per_try = 125; 1920 int i = 0; 1921 1922 mutex_lock(&pktgen_thread_lock); 1923 pr_debug("%s: marking %s for removal\n", __func__, ifname); 1924 1925 while (1) { 1926 1927 pkt_dev = __pktgen_NN_threads(pn, ifname, REMOVE); 1928 if (pkt_dev == NULL) 1929 break; /* success */ 1930 1931 mutex_unlock(&pktgen_thread_lock); 1932 pr_debug("%s: waiting for %s to disappear....\n", 1933 __func__, ifname); 1934 schedule_timeout_interruptible(msecs_to_jiffies(msec_per_try)); 1935 mutex_lock(&pktgen_thread_lock); 1936 1937 if (++i >= max_tries) { 1938 pr_err("%s: timed out after waiting %d msec for device %s to be removed\n", 1939 __func__, msec_per_try * i, ifname); 1940 break; 1941 } 1942 1943 } 1944 1945 mutex_unlock(&pktgen_thread_lock); 1946} 1947 1948static void pktgen_change_name(const struct pktgen_net *pn, struct net_device *dev) 1949{ 1950 struct pktgen_thread *t; 1951 1952 list_for_each_entry(t, &pn->pktgen_threads, th_list) { 1953 struct pktgen_dev *pkt_dev; 1954 1955 rcu_read_lock(); 1956 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) { 1957 if (pkt_dev->odev != dev) 1958 continue; 1959 1960 proc_remove(pkt_dev->entry); 1961 1962 pkt_dev->entry = proc_create_data(dev->name, 0600, 1963 pn->proc_dir, 1964 &pktgen_if_fops, 1965 pkt_dev); 1966 if (!pkt_dev->entry) 1967 pr_err("can't move proc entry for '%s'\n", 1968 dev->name); 1969 break; 1970 } 1971 rcu_read_unlock(); 1972 } 1973} 1974 1975static int pktgen_device_event(struct notifier_block *unused, 1976 unsigned long event, void *ptr) 1977{ 1978 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1979 struct pktgen_net *pn = net_generic(dev_net(dev), pg_net_id); 1980 1981 if (pn->pktgen_exiting) 1982 return NOTIFY_DONE; 1983 1984 /* It is OK that we do not hold the group lock right now, 1985 * as we run under the RTNL lock. 1986 */ 1987 1988 switch (event) { 1989 case NETDEV_CHANGENAME: 1990 pktgen_change_name(pn, dev); 1991 break; 1992 1993 case NETDEV_UNREGISTER: 1994 pktgen_mark_device(pn, dev->name); 1995 break; 1996 } 1997 1998 return NOTIFY_DONE; 1999} 2000 2001static struct net_device *pktgen_dev_get_by_name(const struct pktgen_net *pn, 2002 struct pktgen_dev *pkt_dev, 2003 const char *ifname) 2004{ 2005 char b[IFNAMSIZ+5]; 2006 int i; 2007 2008 for (i = 0; ifname[i] != '@'; i++) { 2009 if (i == IFNAMSIZ) 2010 break; 2011 2012 b[i] = ifname[i]; 2013 } 2014 b[i] = 0; 2015 2016 return dev_get_by_name(pn->net, b); 2017} 2018 2019 2020/* Associate pktgen_dev with a device. */ 2021 2022static int pktgen_setup_dev(const struct pktgen_net *pn, 2023 struct pktgen_dev *pkt_dev, const char *ifname) 2024{ 2025 struct net_device *odev; 2026 int err; 2027 2028 /* Clean old setups */ 2029 if (pkt_dev->odev) { 2030 dev_put(pkt_dev->odev); 2031 pkt_dev->odev = NULL; 2032 } 2033 2034 odev = pktgen_dev_get_by_name(pn, pkt_dev, ifname); 2035 if (!odev) { 2036 pr_err("no such netdevice: \"%s\"\n", ifname); 2037 return -ENODEV; 2038 } 2039 2040 if (odev->type != ARPHRD_ETHER) { 2041 pr_err("not an ethernet device: \"%s\"\n", ifname); 2042 err = -EINVAL; 2043 } else if (!netif_running(odev)) { 2044 pr_err("device is down: \"%s\"\n", ifname); 2045 err = -ENETDOWN; 2046 } else { 2047 pkt_dev->odev = odev; 2048 return 0; 2049 } 2050 2051 dev_put(odev); 2052 return err; 2053} 2054 2055/* Read pkt_dev from the interface and set up internal pktgen_dev 2056 * structure to have the right information to create/send packets 2057 */ 2058static void pktgen_setup_inject(struct pktgen_dev *pkt_dev) 2059{ 2060 int ntxq; 2061 2062 if (!pkt_dev->odev) { 2063 pr_err("ERROR: pkt_dev->odev == NULL in setup_inject\n"); 2064 sprintf(pkt_dev->result, 2065 "ERROR: pkt_dev->odev == NULL in setup_inject.\n"); 2066 return; 2067 } 2068 2069 /* make sure that we don't pick a non-existing transmit queue */ 2070 ntxq = pkt_dev->odev->real_num_tx_queues; 2071 2072 if (ntxq <= pkt_dev->queue_map_min) { 2073 pr_warn("WARNING: Requested queue_map_min (zero-based) (%d) exceeds valid range [0 - %d] for (%d) queues on %s, resetting\n", 2074 pkt_dev->queue_map_min, (ntxq ?: 1) - 1, ntxq, 2075 pkt_dev->odevname); 2076 pkt_dev->queue_map_min = (ntxq ?: 1) - 1; 2077 } 2078 if (pkt_dev->queue_map_max >= ntxq) { 2079 pr_warn("WARNING: Requested queue_map_max (zero-based) (%d) exceeds valid range [0 - %d] for (%d) queues on %s, resetting\n", 2080 pkt_dev->queue_map_max, (ntxq ?: 1) - 1, ntxq, 2081 pkt_dev->odevname); 2082 pkt_dev->queue_map_max = (ntxq ?: 1) - 1; 2083 } 2084 2085 /* Default to the interface's mac if not explicitly set. */ 2086 2087 if (is_zero_ether_addr(pkt_dev->src_mac)) 2088 ether_addr_copy(&(pkt_dev->hh[6]), pkt_dev->odev->dev_addr); 2089 2090 /* Set up Dest MAC */ 2091 ether_addr_copy(&(pkt_dev->hh[0]), pkt_dev->dst_mac); 2092 2093 if (pkt_dev->flags & F_IPV6) { 2094 int i, set = 0, err = 1; 2095 struct inet6_dev *idev; 2096 2097 if (pkt_dev->min_pkt_size == 0) { 2098 pkt_dev->min_pkt_size = 14 + sizeof(struct ipv6hdr) 2099 + sizeof(struct udphdr) 2100 + sizeof(struct pktgen_hdr) 2101 + pkt_dev->pkt_overhead; 2102 } 2103 2104 for (i = 0; i < IN6_ADDR_HSIZE; i++) 2105 if (pkt_dev->cur_in6_saddr.s6_addr[i]) { 2106 set = 1; 2107 break; 2108 } 2109 2110 if (!set) { 2111 2112 /* 2113 * Use linklevel address if unconfigured. 2114 * 2115 * use ipv6_get_lladdr if/when it's get exported 2116 */ 2117 2118 rcu_read_lock(); 2119 idev = __in6_dev_get(pkt_dev->odev); 2120 if (idev) { 2121 struct inet6_ifaddr *ifp; 2122 2123 read_lock_bh(&idev->lock); 2124 list_for_each_entry(ifp, &idev->addr_list, if_list) { 2125 if ((ifp->scope & IFA_LINK) && 2126 !(ifp->flags & IFA_F_TENTATIVE)) { 2127 pkt_dev->cur_in6_saddr = ifp->addr; 2128 err = 0; 2129 break; 2130 } 2131 } 2132 read_unlock_bh(&idev->lock); 2133 } 2134 rcu_read_unlock(); 2135 if (err) 2136 pr_err("ERROR: IPv6 link address not available\n"); 2137 } 2138 } else { 2139 if (pkt_dev->min_pkt_size == 0) { 2140 pkt_dev->min_pkt_size = 14 + sizeof(struct iphdr) 2141 + sizeof(struct udphdr) 2142 + sizeof(struct pktgen_hdr) 2143 + pkt_dev->pkt_overhead; 2144 } 2145 2146 pkt_dev->saddr_min = 0; 2147 pkt_dev->saddr_max = 0; 2148 if (strlen(pkt_dev->src_min) == 0) { 2149 2150 struct in_device *in_dev; 2151 2152 rcu_read_lock(); 2153 in_dev = __in_dev_get_rcu(pkt_dev->odev); 2154 if (in_dev) { 2155 if (in_dev->ifa_list) { 2156 pkt_dev->saddr_min = 2157 in_dev->ifa_list->ifa_address; 2158 pkt_dev->saddr_max = pkt_dev->saddr_min; 2159 } 2160 } 2161 rcu_read_unlock(); 2162 } else { 2163 pkt_dev->saddr_min = in_aton(pkt_dev->src_min); 2164 pkt_dev->saddr_max = in_aton(pkt_dev->src_max); 2165 } 2166 2167 pkt_dev->daddr_min = in_aton(pkt_dev->dst_min); 2168 pkt_dev->daddr_max = in_aton(pkt_dev->dst_max); 2169 } 2170 /* Initialize current values. */ 2171 pkt_dev->cur_pkt_size = pkt_dev->min_pkt_size; 2172 if (pkt_dev->min_pkt_size > pkt_dev->max_pkt_size) 2173 pkt_dev->max_pkt_size = pkt_dev->min_pkt_size; 2174 2175 pkt_dev->cur_dst_mac_offset = 0; 2176 pkt_dev->cur_src_mac_offset = 0; 2177 pkt_dev->cur_saddr = pkt_dev->saddr_min; 2178 pkt_dev->cur_daddr = pkt_dev->daddr_min; 2179 pkt_dev->cur_udp_dst = pkt_dev->udp_dst_min; 2180 pkt_dev->cur_udp_src = pkt_dev->udp_src_min; 2181 pkt_dev->nflows = 0; 2182} 2183 2184 2185static void spin(struct pktgen_dev *pkt_dev, ktime_t spin_until) 2186{ 2187 ktime_t start_time, end_time; 2188 s64 remaining; 2189 struct hrtimer_sleeper t; 2190 2191 hrtimer_init_on_stack(&t.timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); 2192 hrtimer_set_expires(&t.timer, spin_until); 2193 2194 remaining = ktime_to_ns(hrtimer_expires_remaining(&t.timer)); 2195 if (remaining <= 0) { 2196 pkt_dev->next_tx = ktime_add_ns(spin_until, pkt_dev->delay); 2197 return; 2198 } 2199 2200 start_time = ktime_get(); 2201 if (remaining < 100000) { 2202 /* for small delays (<100us), just loop until limit is reached */ 2203 do { 2204 end_time = ktime_get(); 2205 } while (ktime_compare(end_time, spin_until) < 0); 2206 } else { 2207 /* see do_nanosleep */ 2208 hrtimer_init_sleeper(&t, current); 2209 do { 2210 set_current_state(TASK_INTERRUPTIBLE); 2211 hrtimer_start_expires(&t.timer, HRTIMER_MODE_ABS); 2212 if (!hrtimer_active(&t.timer)) 2213 t.task = NULL; 2214 2215 if (likely(t.task)) 2216 schedule(); 2217 2218 hrtimer_cancel(&t.timer); 2219 } while (t.task && pkt_dev->running && !signal_pending(current)); 2220 __set_current_state(TASK_RUNNING); 2221 end_time = ktime_get(); 2222 } 2223 2224 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(end_time, start_time)); 2225 pkt_dev->next_tx = ktime_add_ns(spin_until, pkt_dev->delay); 2226} 2227 2228static inline void set_pkt_overhead(struct pktgen_dev *pkt_dev) 2229{ 2230 pkt_dev->pkt_overhead = 0; 2231 pkt_dev->pkt_overhead += pkt_dev->nr_labels*sizeof(u32); 2232 pkt_dev->pkt_overhead += VLAN_TAG_SIZE(pkt_dev); 2233 pkt_dev->pkt_overhead += SVLAN_TAG_SIZE(pkt_dev); 2234} 2235 2236static inline int f_seen(const struct pktgen_dev *pkt_dev, int flow) 2237{ 2238 return !!(pkt_dev->flows[flow].flags & F_INIT); 2239} 2240 2241static inline int f_pick(struct pktgen_dev *pkt_dev) 2242{ 2243 int flow = pkt_dev->curfl; 2244 2245 if (pkt_dev->flags & F_FLOW_SEQ) { 2246 if (pkt_dev->flows[flow].count >= pkt_dev->lflow) { 2247 /* reset time */ 2248 pkt_dev->flows[flow].count = 0; 2249 pkt_dev->flows[flow].flags = 0; 2250 pkt_dev->curfl += 1; 2251 if (pkt_dev->curfl >= pkt_dev->cflows) 2252 pkt_dev->curfl = 0; /*reset */ 2253 } 2254 } else { 2255 flow = prandom_u32() % pkt_dev->cflows; 2256 pkt_dev->curfl = flow; 2257 2258 if (pkt_dev->flows[flow].count > pkt_dev->lflow) { 2259 pkt_dev->flows[flow].count = 0; 2260 pkt_dev->flows[flow].flags = 0; 2261 } 2262 } 2263 2264 return pkt_dev->curfl; 2265} 2266 2267 2268#ifdef CONFIG_XFRM 2269/* If there was already an IPSEC SA, we keep it as is, else 2270 * we go look for it ... 2271*/ 2272#define DUMMY_MARK 0 2273static void get_ipsec_sa(struct pktgen_dev *pkt_dev, int flow) 2274{ 2275 struct xfrm_state *x = pkt_dev->flows[flow].x; 2276 struct pktgen_net *pn = net_generic(dev_net(pkt_dev->odev), pg_net_id); 2277 if (!x) { 2278 2279 if (pkt_dev->spi) { 2280 /* We need as quick as possible to find the right SA 2281 * Searching with minimum criteria to archieve this. 2282 */ 2283 x = xfrm_state_lookup_byspi(pn->net, htonl(pkt_dev->spi), AF_INET); 2284 } else { 2285 /* slow path: we dont already have xfrm_state */ 2286 x = xfrm_stateonly_find(pn->net, DUMMY_MARK, 2287 (xfrm_address_t *)&pkt_dev->cur_daddr, 2288 (xfrm_address_t *)&pkt_dev->cur_saddr, 2289 AF_INET, 2290 pkt_dev->ipsmode, 2291 pkt_dev->ipsproto, 0); 2292 } 2293 if (x) { 2294 pkt_dev->flows[flow].x = x; 2295 set_pkt_overhead(pkt_dev); 2296 pkt_dev->pkt_overhead += x->props.header_len; 2297 } 2298 2299 } 2300} 2301#endif 2302static void set_cur_queue_map(struct pktgen_dev *pkt_dev) 2303{ 2304 2305 if (pkt_dev->flags & F_QUEUE_MAP_CPU) 2306 pkt_dev->cur_queue_map = smp_processor_id(); 2307 2308 else if (pkt_dev->queue_map_min <= pkt_dev->queue_map_max) { 2309 __u16 t; 2310 if (pkt_dev->flags & F_QUEUE_MAP_RND) { 2311 t = prandom_u32() % 2312 (pkt_dev->queue_map_max - 2313 pkt_dev->queue_map_min + 1) 2314 + pkt_dev->queue_map_min; 2315 } else { 2316 t = pkt_dev->cur_queue_map + 1; 2317 if (t > pkt_dev->queue_map_max) 2318 t = pkt_dev->queue_map_min; 2319 } 2320 pkt_dev->cur_queue_map = t; 2321 } 2322 pkt_dev->cur_queue_map = pkt_dev->cur_queue_map % pkt_dev->odev->real_num_tx_queues; 2323} 2324 2325/* Increment/randomize headers according to flags and current values 2326 * for IP src/dest, UDP src/dst port, MAC-Addr src/dst 2327 */ 2328static void mod_cur_headers(struct pktgen_dev *pkt_dev) 2329{ 2330 __u32 imn; 2331 __u32 imx; 2332 int flow = 0; 2333 2334 if (pkt_dev->cflows) 2335 flow = f_pick(pkt_dev); 2336 2337 /* Deal with source MAC */ 2338 if (pkt_dev->src_mac_count > 1) { 2339 __u32 mc; 2340 __u32 tmp; 2341 2342 if (pkt_dev->flags & F_MACSRC_RND) 2343 mc = prandom_u32() % pkt_dev->src_mac_count; 2344 else { 2345 mc = pkt_dev->cur_src_mac_offset++; 2346 if (pkt_dev->cur_src_mac_offset >= 2347 pkt_dev->src_mac_count) 2348 pkt_dev->cur_src_mac_offset = 0; 2349 } 2350 2351 tmp = pkt_dev->src_mac[5] + (mc & 0xFF); 2352 pkt_dev->hh[11] = tmp; 2353 tmp = (pkt_dev->src_mac[4] + ((mc >> 8) & 0xFF) + (tmp >> 8)); 2354 pkt_dev->hh[10] = tmp; 2355 tmp = (pkt_dev->src_mac[3] + ((mc >> 16) & 0xFF) + (tmp >> 8)); 2356 pkt_dev->hh[9] = tmp; 2357 tmp = (pkt_dev->src_mac[2] + ((mc >> 24) & 0xFF) + (tmp >> 8)); 2358 pkt_dev->hh[8] = tmp; 2359 tmp = (pkt_dev->src_mac[1] + (tmp >> 8)); 2360 pkt_dev->hh[7] = tmp; 2361 } 2362 2363 /* Deal with Destination MAC */ 2364 if (pkt_dev->dst_mac_count > 1) { 2365 __u32 mc; 2366 __u32 tmp; 2367 2368 if (pkt_dev->flags & F_MACDST_RND) 2369 mc = prandom_u32() % pkt_dev->dst_mac_count; 2370 2371 else { 2372 mc = pkt_dev->cur_dst_mac_offset++; 2373 if (pkt_dev->cur_dst_mac_offset >= 2374 pkt_dev->dst_mac_count) { 2375 pkt_dev->cur_dst_mac_offset = 0; 2376 } 2377 } 2378 2379 tmp = pkt_dev->dst_mac[5] + (mc & 0xFF); 2380 pkt_dev->hh[5] = tmp; 2381 tmp = (pkt_dev->dst_mac[4] + ((mc >> 8) & 0xFF) + (tmp >> 8)); 2382 pkt_dev->hh[4] = tmp; 2383 tmp = (pkt_dev->dst_mac[3] + ((mc >> 16) & 0xFF) + (tmp >> 8)); 2384 pkt_dev->hh[3] = tmp; 2385 tmp = (pkt_dev->dst_mac[2] + ((mc >> 24) & 0xFF) + (tmp >> 8)); 2386 pkt_dev->hh[2] = tmp; 2387 tmp = (pkt_dev->dst_mac[1] + (tmp >> 8)); 2388 pkt_dev->hh[1] = tmp; 2389 } 2390 2391 if (pkt_dev->flags & F_MPLS_RND) { 2392 unsigned int i; 2393 for (i = 0; i < pkt_dev->nr_labels; i++) 2394 if (pkt_dev->labels[i] & MPLS_STACK_BOTTOM) 2395 pkt_dev->labels[i] = MPLS_STACK_BOTTOM | 2396 ((__force __be32)prandom_u32() & 2397 htonl(0x000fffff)); 2398 } 2399 2400 if ((pkt_dev->flags & F_VID_RND) && (pkt_dev->vlan_id != 0xffff)) { 2401 pkt_dev->vlan_id = prandom_u32() & (4096 - 1); 2402 } 2403 2404 if ((pkt_dev->flags & F_SVID_RND) && (pkt_dev->svlan_id != 0xffff)) { 2405 pkt_dev->svlan_id = prandom_u32() & (4096 - 1); 2406 } 2407 2408 if (pkt_dev->udp_src_min < pkt_dev->udp_src_max) { 2409 if (pkt_dev->flags & F_UDPSRC_RND) 2410 pkt_dev->cur_udp_src = prandom_u32() % 2411 (pkt_dev->udp_src_max - pkt_dev->udp_src_min) 2412 + pkt_dev->udp_src_min; 2413 2414 else { 2415 pkt_dev->cur_udp_src++; 2416 if (pkt_dev->cur_udp_src >= pkt_dev->udp_src_max) 2417 pkt_dev->cur_udp_src = pkt_dev->udp_src_min; 2418 } 2419 } 2420 2421 if (pkt_dev->udp_dst_min < pkt_dev->udp_dst_max) { 2422 if (pkt_dev->flags & F_UDPDST_RND) { 2423 pkt_dev->cur_udp_dst = prandom_u32() % 2424 (pkt_dev->udp_dst_max - pkt_dev->udp_dst_min) 2425 + pkt_dev->udp_dst_min; 2426 } else { 2427 pkt_dev->cur_udp_dst++; 2428 if (pkt_dev->cur_udp_dst >= pkt_dev->udp_dst_max) 2429 pkt_dev->cur_udp_dst = pkt_dev->udp_dst_min; 2430 } 2431 } 2432 2433 if (!(pkt_dev->flags & F_IPV6)) { 2434 2435 imn = ntohl(pkt_dev->saddr_min); 2436 imx = ntohl(pkt_dev->saddr_max); 2437 if (imn < imx) { 2438 __u32 t; 2439 if (pkt_dev->flags & F_IPSRC_RND) 2440 t = prandom_u32() % (imx - imn) + imn; 2441 else { 2442 t = ntohl(pkt_dev->cur_saddr); 2443 t++; 2444 if (t > imx) 2445 t = imn; 2446 2447 } 2448 pkt_dev->cur_saddr = htonl(t); 2449 } 2450 2451 if (pkt_dev->cflows && f_seen(pkt_dev, flow)) { 2452 pkt_dev->cur_daddr = pkt_dev->flows[flow].cur_daddr; 2453 } else { 2454 imn = ntohl(pkt_dev->daddr_min); 2455 imx = ntohl(pkt_dev->daddr_max); 2456 if (imn < imx) { 2457 __u32 t; 2458 __be32 s; 2459 if (pkt_dev->flags & F_IPDST_RND) { 2460 2461 do { 2462 t = prandom_u32() % 2463 (imx - imn) + imn; 2464 s = htonl(t); 2465 } while (ipv4_is_loopback(s) || 2466 ipv4_is_multicast(s) || 2467 ipv4_is_lbcast(s) || 2468 ipv4_is_zeronet(s) || 2469 ipv4_is_local_multicast(s)); 2470 pkt_dev->cur_daddr = s; 2471 } else { 2472 t = ntohl(pkt_dev->cur_daddr); 2473 t++; 2474 if (t > imx) { 2475 t = imn; 2476 } 2477 pkt_dev->cur_daddr = htonl(t); 2478 } 2479 } 2480 if (pkt_dev->cflows) { 2481 pkt_dev->flows[flow].flags |= F_INIT; 2482 pkt_dev->flows[flow].cur_daddr = 2483 pkt_dev->cur_daddr; 2484#ifdef CONFIG_XFRM 2485 if (pkt_dev->flags & F_IPSEC_ON) 2486 get_ipsec_sa(pkt_dev, flow); 2487#endif 2488 pkt_dev->nflows++; 2489 } 2490 } 2491 } else { /* IPV6 * */ 2492 2493 if (!ipv6_addr_any(&pkt_dev->min_in6_daddr)) { 2494 int i; 2495 2496 /* Only random destinations yet */ 2497 2498 for (i = 0; i < 4; i++) { 2499 pkt_dev->cur_in6_daddr.s6_addr32[i] = 2500 (((__force __be32)prandom_u32() | 2501 pkt_dev->min_in6_daddr.s6_addr32[i]) & 2502 pkt_dev->max_in6_daddr.s6_addr32[i]); 2503 } 2504 } 2505 } 2506 2507 if (pkt_dev->min_pkt_size < pkt_dev->max_pkt_size) { 2508 __u32 t; 2509 if (pkt_dev->flags & F_TXSIZE_RND) { 2510 t = prandom_u32() % 2511 (pkt_dev->max_pkt_size - pkt_dev->min_pkt_size) 2512 + pkt_dev->min_pkt_size; 2513 } else { 2514 t = pkt_dev->cur_pkt_size + 1; 2515 if (t > pkt_dev->max_pkt_size) 2516 t = pkt_dev->min_pkt_size; 2517 } 2518 pkt_dev->cur_pkt_size = t; 2519 } 2520 2521 set_cur_queue_map(pkt_dev); 2522 2523 pkt_dev->flows[flow].count++; 2524} 2525 2526 2527#ifdef CONFIG_XFRM 2528static u32 pktgen_dst_metrics[RTAX_MAX + 1] = { 2529 2530 [RTAX_HOPLIMIT] = 0x5, /* Set a static hoplimit */ 2531}; 2532 2533static int pktgen_output_ipsec(struct sk_buff *skb, struct pktgen_dev *pkt_dev) 2534{ 2535 struct xfrm_state *x = pkt_dev->flows[pkt_dev->curfl].x; 2536 int err = 0; 2537 struct net *net = dev_net(pkt_dev->odev); 2538 2539 if (!x) 2540 return 0; 2541 /* XXX: we dont support tunnel mode for now until 2542 * we resolve the dst issue */ 2543 if ((x->props.mode != XFRM_MODE_TRANSPORT) && (pkt_dev->spi == 0)) 2544 return 0; 2545 2546 /* But when user specify an valid SPI, transformation 2547 * supports both transport/tunnel mode + ESP/AH type. 2548 */ 2549 if ((x->props.mode == XFRM_MODE_TUNNEL) && (pkt_dev->spi != 0)) 2550 skb->_skb_refdst = (unsigned long)&pkt_dev->dst | SKB_DST_NOREF; 2551 2552 rcu_read_lock_bh(); 2553 err = x->outer_mode->output(x, skb); 2554 rcu_read_unlock_bh(); 2555 if (err) { 2556 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEMODEERROR); 2557 goto error; 2558 } 2559 err = x->type->output(x, skb); 2560 if (err) { 2561 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEPROTOERROR); 2562 goto error; 2563 } 2564 spin_lock_bh(&x->lock); 2565 x->curlft.bytes += skb->len; 2566 x->curlft.packets++; 2567 spin_unlock_bh(&x->lock); 2568error: 2569 return err; 2570} 2571 2572static void free_SAs(struct pktgen_dev *pkt_dev) 2573{ 2574 if (pkt_dev->cflows) { 2575 /* let go of the SAs if we have them */ 2576 int i; 2577 for (i = 0; i < pkt_dev->cflows; i++) { 2578 struct xfrm_state *x = pkt_dev->flows[i].x; 2579 if (x) { 2580 xfrm_state_put(x); 2581 pkt_dev->flows[i].x = NULL; 2582 } 2583 } 2584 } 2585} 2586 2587static int process_ipsec(struct pktgen_dev *pkt_dev, 2588 struct sk_buff *skb, __be16 protocol) 2589{ 2590 if (pkt_dev->flags & F_IPSEC_ON) { 2591 struct xfrm_state *x = pkt_dev->flows[pkt_dev->curfl].x; 2592 int nhead = 0; 2593 if (x) { 2594 int ret; 2595 __u8 *eth; 2596 struct iphdr *iph; 2597 2598 nhead = x->props.header_len - skb_headroom(skb); 2599 if (nhead > 0) { 2600 ret = pskb_expand_head(skb, nhead, 0, GFP_ATOMIC); 2601 if (ret < 0) { 2602 pr_err("Error expanding ipsec packet %d\n", 2603 ret); 2604 goto err; 2605 } 2606 } 2607 2608 /* ipsec is not expecting ll header */ 2609 skb_pull(skb, ETH_HLEN); 2610 ret = pktgen_output_ipsec(skb, pkt_dev); 2611 if (ret) { 2612 pr_err("Error creating ipsec packet %d\n", ret); 2613 goto err; 2614 } 2615 /* restore ll */ 2616 eth = (__u8 *) skb_push(skb, ETH_HLEN); 2617 memcpy(eth, pkt_dev->hh, 12); 2618 *(u16 *) ð[12] = protocol; 2619 2620 /* Update IPv4 header len as well as checksum value */ 2621 iph = ip_hdr(skb); 2622 iph->tot_len = htons(skb->len - ETH_HLEN); 2623 ip_send_check(iph); 2624 } 2625 } 2626 return 1; 2627err: 2628 kfree_skb(skb); 2629 return 0; 2630} 2631#endif 2632 2633static void mpls_push(__be32 *mpls, struct pktgen_dev *pkt_dev) 2634{ 2635 unsigned int i; 2636 for (i = 0; i < pkt_dev->nr_labels; i++) 2637 *mpls++ = pkt_dev->labels[i] & ~MPLS_STACK_BOTTOM; 2638 2639 mpls--; 2640 *mpls |= MPLS_STACK_BOTTOM; 2641} 2642 2643static inline __be16 build_tci(unsigned int id, unsigned int cfi, 2644 unsigned int prio) 2645{ 2646 return htons(id | (cfi << 12) | (prio << 13)); 2647} 2648 2649static void pktgen_finalize_skb(struct pktgen_dev *pkt_dev, struct sk_buff *skb, 2650 int datalen) 2651{ 2652 struct timeval timestamp; 2653 struct pktgen_hdr *pgh; 2654 2655 pgh = (struct pktgen_hdr *)skb_put(skb, sizeof(*pgh)); 2656 datalen -= sizeof(*pgh); 2657 2658 if (pkt_dev->nfrags <= 0) { 2659 memset(skb_put(skb, datalen), 0, datalen); 2660 } else { 2661 int frags = pkt_dev->nfrags; 2662 int i, len; 2663 int frag_len; 2664 2665 2666 if (frags > MAX_SKB_FRAGS) 2667 frags = MAX_SKB_FRAGS; 2668 len = datalen - frags * PAGE_SIZE; 2669 if (len > 0) { 2670 memset(skb_put(skb, len), 0, len); 2671 datalen = frags * PAGE_SIZE; 2672 } 2673 2674 i = 0; 2675 frag_len = (datalen/frags) < PAGE_SIZE ? 2676 (datalen/frags) : PAGE_SIZE; 2677 while (datalen > 0) { 2678 if (unlikely(!pkt_dev->page)) { 2679 int node = numa_node_id(); 2680 2681 if (pkt_dev->node >= 0 && (pkt_dev->flags & F_NODE)) 2682 node = pkt_dev->node; 2683 pkt_dev->page = alloc_pages_node(node, GFP_KERNEL | __GFP_ZERO, 0); 2684 if (!pkt_dev->page) 2685 break; 2686 } 2687 get_page(pkt_dev->page); 2688 skb_frag_set_page(skb, i, pkt_dev->page); 2689 skb_shinfo(skb)->frags[i].page_offset = 0; 2690 /*last fragment, fill rest of data*/ 2691 if (i == (frags - 1)) 2692 skb_frag_size_set(&skb_shinfo(skb)->frags[i], 2693 (datalen < PAGE_SIZE ? datalen : PAGE_SIZE)); 2694 else 2695 skb_frag_size_set(&skb_shinfo(skb)->frags[i], frag_len); 2696 datalen -= skb_frag_size(&skb_shinfo(skb)->frags[i]); 2697 skb->len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 2698 skb->data_len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 2699 i++; 2700 skb_shinfo(skb)->nr_frags = i; 2701 } 2702 } 2703 2704 /* Stamp the time, and sequence number, 2705 * convert them to network byte order 2706 */ 2707 pgh->pgh_magic = htonl(PKTGEN_MAGIC); 2708 pgh->seq_num = htonl(pkt_dev->seq_num); 2709 2710 if (pkt_dev->flags & F_NO_TIMESTAMP) { 2711 pgh->tv_sec = 0; 2712 pgh->tv_usec = 0; 2713 } else { 2714 do_gettimeofday(×tamp); 2715 pgh->tv_sec = htonl(timestamp.tv_sec); 2716 pgh->tv_usec = htonl(timestamp.tv_usec); 2717 } 2718} 2719 2720static struct sk_buff *pktgen_alloc_skb(struct net_device *dev, 2721 struct pktgen_dev *pkt_dev, 2722 unsigned int extralen) 2723{ 2724 struct sk_buff *skb = NULL; 2725 unsigned int size = pkt_dev->cur_pkt_size + 64 + extralen + 2726 pkt_dev->pkt_overhead; 2727 2728 if (pkt_dev->flags & F_NODE) { 2729 int node = pkt_dev->node >= 0 ? pkt_dev->node : numa_node_id(); 2730 2731 skb = __alloc_skb(NET_SKB_PAD + size, GFP_NOWAIT, 0, node); 2732 if (likely(skb)) { 2733 skb_reserve(skb, NET_SKB_PAD); 2734 skb->dev = dev; 2735 } 2736 } else { 2737 skb = __netdev_alloc_skb(dev, size, GFP_NOWAIT); 2738 } 2739 2740 return skb; 2741} 2742 2743static struct sk_buff *fill_packet_ipv4(struct net_device *odev, 2744 struct pktgen_dev *pkt_dev) 2745{ 2746 struct sk_buff *skb = NULL; 2747 __u8 *eth; 2748 struct udphdr *udph; 2749 int datalen, iplen; 2750 struct iphdr *iph; 2751 __be16 protocol = htons(ETH_P_IP); 2752 __be32 *mpls; 2753 __be16 *vlan_tci = NULL; /* Encapsulates priority and VLAN ID */ 2754 __be16 *vlan_encapsulated_proto = NULL; /* packet type ID field (or len) for VLAN tag */ 2755 __be16 *svlan_tci = NULL; /* Encapsulates priority and SVLAN ID */ 2756 __be16 *svlan_encapsulated_proto = NULL; /* packet type ID field (or len) for SVLAN tag */ 2757 u16 queue_map; 2758 2759 if (pkt_dev->nr_labels) 2760 protocol = htons(ETH_P_MPLS_UC); 2761 2762 if (pkt_dev->vlan_id != 0xffff) 2763 protocol = htons(ETH_P_8021Q); 2764 2765 /* Update any of the values, used when we're incrementing various 2766 * fields. 2767 */ 2768 mod_cur_headers(pkt_dev); 2769 queue_map = pkt_dev->cur_queue_map; 2770 2771 datalen = (odev->hard_header_len + 16) & ~0xf; 2772 2773 skb = pktgen_alloc_skb(odev, pkt_dev, datalen); 2774 if (!skb) { 2775 sprintf(pkt_dev->result, "No memory"); 2776 return NULL; 2777 } 2778 2779 prefetchw(skb->data); 2780 skb_reserve(skb, datalen); 2781 2782 /* Reserve for ethernet and IP header */ 2783 eth = (__u8 *) skb_push(skb, 14); 2784 mpls = (__be32 *)skb_put(skb, pkt_dev->nr_labels*sizeof(__u32)); 2785 if (pkt_dev->nr_labels) 2786 mpls_push(mpls, pkt_dev); 2787 2788 if (pkt_dev->vlan_id != 0xffff) { 2789 if (pkt_dev->svlan_id != 0xffff) { 2790 svlan_tci = (__be16 *)skb_put(skb, sizeof(__be16)); 2791 *svlan_tci = build_tci(pkt_dev->svlan_id, 2792 pkt_dev->svlan_cfi, 2793 pkt_dev->svlan_p); 2794 svlan_encapsulated_proto = (__be16 *)skb_put(skb, sizeof(__be16)); 2795 *svlan_encapsulated_proto = htons(ETH_P_8021Q); 2796 } 2797 vlan_tci = (__be16 *)skb_put(skb, sizeof(__be16)); 2798 *vlan_tci = build_tci(pkt_dev->vlan_id, 2799 pkt_dev->vlan_cfi, 2800 pkt_dev->vlan_p); 2801 vlan_encapsulated_proto = (__be16 *)skb_put(skb, sizeof(__be16)); 2802 *vlan_encapsulated_proto = htons(ETH_P_IP); 2803 } 2804 2805 skb_set_mac_header(skb, 0); 2806 skb_set_network_header(skb, skb->len); 2807 iph = (struct iphdr *) skb_put(skb, sizeof(struct iphdr)); 2808 2809 skb_set_transport_header(skb, skb->len); 2810 udph = (struct udphdr *) skb_put(skb, sizeof(struct udphdr)); 2811 skb_set_queue_mapping(skb, queue_map); 2812 skb->priority = pkt_dev->skb_priority; 2813 2814 memcpy(eth, pkt_dev->hh, 12); 2815 *(__be16 *) & eth[12] = protocol; 2816 2817 /* Eth + IPh + UDPh + mpls */ 2818 datalen = pkt_dev->cur_pkt_size - 14 - 20 - 8 - 2819 pkt_dev->pkt_overhead; 2820 if (datalen < 0 || datalen < sizeof(struct pktgen_hdr)) 2821 datalen = sizeof(struct pktgen_hdr); 2822 2823 udph->source = htons(pkt_dev->cur_udp_src); 2824 udph->dest = htons(pkt_dev->cur_udp_dst); 2825 udph->len = htons(datalen + 8); /* DATA + udphdr */ 2826 udph->check = 0; 2827 2828 iph->ihl = 5; 2829 iph->version = 4; 2830 iph->ttl = 32; 2831 iph->tos = pkt_dev->tos; 2832 iph->protocol = IPPROTO_UDP; /* UDP */ 2833 iph->saddr = pkt_dev->cur_saddr; 2834 iph->daddr = pkt_dev->cur_daddr; 2835 iph->id = htons(pkt_dev->ip_id); 2836 pkt_dev->ip_id++; 2837 iph->frag_off = 0; 2838 iplen = 20 + 8 + datalen; 2839 iph->tot_len = htons(iplen); 2840 ip_send_check(iph); 2841 skb->protocol = protocol; 2842 skb->dev = odev; 2843 skb->pkt_type = PACKET_HOST; 2844 2845 if (!(pkt_dev->flags & F_UDPCSUM)) { 2846 skb->ip_summed = CHECKSUM_NONE; 2847 } else if (odev->features & NETIF_F_V4_CSUM) { 2848 skb->ip_summed = CHECKSUM_PARTIAL; 2849 skb->csum = 0; 2850 udp4_hwcsum(skb, udph->source, udph->dest); 2851 } else { 2852 __wsum csum = udp_csum(skb); 2853 2854 /* add protocol-dependent pseudo-header */ 2855 udph->check = csum_tcpudp_magic(udph->source, udph->dest, 2856 datalen + 8, IPPROTO_UDP, csum); 2857 2858 if (udph->check == 0) 2859 udph->check = CSUM_MANGLED_0; 2860 } 2861 2862 pktgen_finalize_skb(pkt_dev, skb, datalen); 2863 2864#ifdef CONFIG_XFRM 2865 if (!process_ipsec(pkt_dev, skb, protocol)) 2866 return NULL; 2867#endif 2868 2869 return skb; 2870} 2871 2872static struct sk_buff *fill_packet_ipv6(struct net_device *odev, 2873 struct pktgen_dev *pkt_dev) 2874{ 2875 struct sk_buff *skb = NULL; 2876 __u8 *eth; 2877 struct udphdr *udph; 2878 int datalen, udplen; 2879 struct ipv6hdr *iph; 2880 __be16 protocol = htons(ETH_P_IPV6); 2881 __be32 *mpls; 2882 __be16 *vlan_tci = NULL; /* Encapsulates priority and VLAN ID */ 2883 __be16 *vlan_encapsulated_proto = NULL; /* packet type ID field (or len) for VLAN tag */ 2884 __be16 *svlan_tci = NULL; /* Encapsulates priority and SVLAN ID */ 2885 __be16 *svlan_encapsulated_proto = NULL; /* packet type ID field (or len) for SVLAN tag */ 2886 u16 queue_map; 2887 2888 if (pkt_dev->nr_labels) 2889 protocol = htons(ETH_P_MPLS_UC); 2890 2891 if (pkt_dev->vlan_id != 0xffff) 2892 protocol = htons(ETH_P_8021Q); 2893 2894 /* Update any of the values, used when we're incrementing various 2895 * fields. 2896 */ 2897 mod_cur_headers(pkt_dev); 2898 queue_map = pkt_dev->cur_queue_map; 2899 2900 skb = pktgen_alloc_skb(odev, pkt_dev, 16); 2901 if (!skb) { 2902 sprintf(pkt_dev->result, "No memory"); 2903 return NULL; 2904 } 2905 2906 prefetchw(skb->data); 2907 skb_reserve(skb, 16); 2908 2909 /* Reserve for ethernet and IP header */ 2910 eth = (__u8 *) skb_push(skb, 14); 2911 mpls = (__be32 *)skb_put(skb, pkt_dev->nr_labels*sizeof(__u32)); 2912 if (pkt_dev->nr_labels) 2913 mpls_push(mpls, pkt_dev); 2914 2915 if (pkt_dev->vlan_id != 0xffff) { 2916 if (pkt_dev->svlan_id != 0xffff) { 2917 svlan_tci = (__be16 *)skb_put(skb, sizeof(__be16)); 2918 *svlan_tci = build_tci(pkt_dev->svlan_id, 2919 pkt_dev->svlan_cfi, 2920 pkt_dev->svlan_p); 2921 svlan_encapsulated_proto = (__be16 *)skb_put(skb, sizeof(__be16)); 2922 *svlan_encapsulated_proto = htons(ETH_P_8021Q); 2923 } 2924 vlan_tci = (__be16 *)skb_put(skb, sizeof(__be16)); 2925 *vlan_tci = build_tci(pkt_dev->vlan_id, 2926 pkt_dev->vlan_cfi, 2927 pkt_dev->vlan_p); 2928 vlan_encapsulated_proto = (__be16 *)skb_put(skb, sizeof(__be16)); 2929 *vlan_encapsulated_proto = htons(ETH_P_IPV6); 2930 } 2931 2932 skb_set_mac_header(skb, 0); 2933 skb_set_network_header(skb, skb->len); 2934 iph = (struct ipv6hdr *) skb_put(skb, sizeof(struct ipv6hdr)); 2935 2936 skb_set_transport_header(skb, skb->len); 2937 udph = (struct udphdr *) skb_put(skb, sizeof(struct udphdr)); 2938 skb_set_queue_mapping(skb, queue_map); 2939 skb->priority = pkt_dev->skb_priority; 2940 2941 memcpy(eth, pkt_dev->hh, 12); 2942 *(__be16 *) ð[12] = protocol; 2943 2944 /* Eth + IPh + UDPh + mpls */ 2945 datalen = pkt_dev->cur_pkt_size - 14 - 2946 sizeof(struct ipv6hdr) - sizeof(struct udphdr) - 2947 pkt_dev->pkt_overhead; 2948 2949 if (datalen < 0 || datalen < sizeof(struct pktgen_hdr)) { 2950 datalen = sizeof(struct pktgen_hdr); 2951 net_info_ratelimited("increased datalen to %d\n", datalen); 2952 } 2953 2954 udplen = datalen + sizeof(struct udphdr); 2955 udph->source = htons(pkt_dev->cur_udp_src); 2956 udph->dest = htons(pkt_dev->cur_udp_dst); 2957 udph->len = htons(udplen); 2958 udph->check = 0; 2959 2960 *(__be32 *) iph = htonl(0x60000000); /* Version + flow */ 2961 2962 if (pkt_dev->traffic_class) { 2963 /* Version + traffic class + flow (0) */ 2964 *(__be32 *)iph |= htonl(0x60000000 | (pkt_dev->traffic_class << 20)); 2965 } 2966 2967 iph->hop_limit = 32; 2968 2969 iph->payload_len = htons(udplen); 2970 iph->nexthdr = IPPROTO_UDP; 2971 2972 iph->daddr = pkt_dev->cur_in6_daddr; 2973 iph->saddr = pkt_dev->cur_in6_saddr; 2974 2975 skb->protocol = protocol; 2976 skb->dev = odev; 2977 skb->pkt_type = PACKET_HOST; 2978 2979 if (!(pkt_dev->flags & F_UDPCSUM)) { 2980 skb->ip_summed = CHECKSUM_NONE; 2981 } else if (odev->features & NETIF_F_V6_CSUM) { 2982 skb->ip_summed = CHECKSUM_PARTIAL; 2983 skb->csum_start = skb_transport_header(skb) - skb->head; 2984 skb->csum_offset = offsetof(struct udphdr, check); 2985 udph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, udplen, IPPROTO_UDP, 0); 2986 } else { 2987 __wsum csum = udp_csum(skb); 2988 2989 /* add protocol-dependent pseudo-header */ 2990 udph->check = csum_ipv6_magic(&iph->saddr, &iph->daddr, udplen, IPPROTO_UDP, csum); 2991 2992 if (udph->check == 0) 2993 udph->check = CSUM_MANGLED_0; 2994 } 2995 2996 pktgen_finalize_skb(pkt_dev, skb, datalen); 2997 2998 return skb; 2999} 3000 3001static struct sk_buff *fill_packet(struct net_device *odev, 3002 struct pktgen_dev *pkt_dev) 3003{ 3004 if (pkt_dev->flags & F_IPV6) 3005 return fill_packet_ipv6(odev, pkt_dev); 3006 else 3007 return fill_packet_ipv4(odev, pkt_dev); 3008} 3009 3010static void pktgen_clear_counters(struct pktgen_dev *pkt_dev) 3011{ 3012 pkt_dev->seq_num = 1; 3013 pkt_dev->idle_acc = 0; 3014 pkt_dev->sofar = 0; 3015 pkt_dev->tx_bytes = 0; 3016 pkt_dev->errors = 0; 3017} 3018 3019/* Set up structure for sending pkts, clear counters */ 3020 3021static void pktgen_run(struct pktgen_thread *t) 3022{ 3023 struct pktgen_dev *pkt_dev; 3024 int started = 0; 3025 3026 func_enter(); 3027 3028 rcu_read_lock(); 3029 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) { 3030 3031 /* 3032 * setup odev and create initial packet. 3033 */ 3034 pktgen_setup_inject(pkt_dev); 3035 3036 if (pkt_dev->odev) { 3037 pktgen_clear_counters(pkt_dev); 3038 pkt_dev->skb = NULL; 3039 pkt_dev->started_at = pkt_dev->next_tx = ktime_get(); 3040 3041 set_pkt_overhead(pkt_dev); 3042 3043 strcpy(pkt_dev->result, "Starting"); 3044 pkt_dev->running = 1; /* Cranke yeself! */ 3045 started++; 3046 } else 3047 strcpy(pkt_dev->result, "Error starting"); 3048 } 3049 rcu_read_unlock(); 3050 if (started) 3051 t->control &= ~(T_STOP); 3052} 3053 3054static void pktgen_stop_all_threads_ifs(struct pktgen_net *pn) 3055{ 3056 struct pktgen_thread *t; 3057 3058 func_enter(); 3059 3060 mutex_lock(&pktgen_thread_lock); 3061 3062 list_for_each_entry(t, &pn->pktgen_threads, th_list) 3063 t->control |= T_STOP; 3064 3065 mutex_unlock(&pktgen_thread_lock); 3066} 3067 3068static int thread_is_running(const struct pktgen_thread *t) 3069{ 3070 const struct pktgen_dev *pkt_dev; 3071 3072 rcu_read_lock(); 3073 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) 3074 if (pkt_dev->running) { 3075 rcu_read_unlock(); 3076 return 1; 3077 } 3078 rcu_read_unlock(); 3079 return 0; 3080} 3081 3082static int pktgen_wait_thread_run(struct pktgen_thread *t) 3083{ 3084 while (thread_is_running(t)) { 3085 3086 msleep_interruptible(100); 3087 3088 if (signal_pending(current)) 3089 goto signal; 3090 } 3091 return 1; 3092signal: 3093 return 0; 3094} 3095 3096static int pktgen_wait_all_threads_run(struct pktgen_net *pn) 3097{ 3098 struct pktgen_thread *t; 3099 int sig = 1; 3100 3101 mutex_lock(&pktgen_thread_lock); 3102 3103 list_for_each_entry(t, &pn->pktgen_threads, th_list) { 3104 sig = pktgen_wait_thread_run(t); 3105 if (sig == 0) 3106 break; 3107 } 3108 3109 if (sig == 0) 3110 list_for_each_entry(t, &pn->pktgen_threads, th_list) 3111 t->control |= (T_STOP); 3112 3113 mutex_unlock(&pktgen_thread_lock); 3114 return sig; 3115} 3116 3117static void pktgen_run_all_threads(struct pktgen_net *pn) 3118{ 3119 struct pktgen_thread *t; 3120 3121 func_enter(); 3122 3123 mutex_lock(&pktgen_thread_lock); 3124 3125 list_for_each_entry(t, &pn->pktgen_threads, th_list) 3126 t->control |= (T_RUN); 3127 3128 mutex_unlock(&pktgen_thread_lock); 3129 3130 /* Propagate thread->control */ 3131 schedule_timeout_interruptible(msecs_to_jiffies(125)); 3132 3133 pktgen_wait_all_threads_run(pn); 3134} 3135 3136static void pktgen_reset_all_threads(struct pktgen_net *pn) 3137{ 3138 struct pktgen_thread *t; 3139 3140 func_enter(); 3141 3142 mutex_lock(&pktgen_thread_lock); 3143 3144 list_for_each_entry(t, &pn->pktgen_threads, th_list) 3145 t->control |= (T_REMDEVALL); 3146 3147 mutex_unlock(&pktgen_thread_lock); 3148 3149 /* Propagate thread->control */ 3150 schedule_timeout_interruptible(msecs_to_jiffies(125)); 3151 3152 pktgen_wait_all_threads_run(pn); 3153} 3154 3155static void show_results(struct pktgen_dev *pkt_dev, int nr_frags) 3156{ 3157 __u64 bps, mbps, pps; 3158 char *p = pkt_dev->result; 3159 ktime_t elapsed = ktime_sub(pkt_dev->stopped_at, 3160 pkt_dev->started_at); 3161 ktime_t idle = ns_to_ktime(pkt_dev->idle_acc); 3162 3163 p += sprintf(p, "OK: %llu(c%llu+d%llu) usec, %llu (%dbyte,%dfrags)\n", 3164 (unsigned long long)ktime_to_us(elapsed), 3165 (unsigned long long)ktime_to_us(ktime_sub(elapsed, idle)), 3166 (unsigned long long)ktime_to_us(idle), 3167 (unsigned long long)pkt_dev->sofar, 3168 pkt_dev->cur_pkt_size, nr_frags); 3169 3170 pps = div64_u64(pkt_dev->sofar * NSEC_PER_SEC, 3171 ktime_to_ns(elapsed)); 3172 3173 bps = pps * 8 * pkt_dev->cur_pkt_size; 3174 3175 mbps = bps; 3176 do_div(mbps, 1000000); 3177 p += sprintf(p, " %llupps %lluMb/sec (%llubps) errors: %llu", 3178 (unsigned long long)pps, 3179 (unsigned long long)mbps, 3180 (unsigned long long)bps, 3181 (unsigned long long)pkt_dev->errors); 3182} 3183 3184/* Set stopped-at timer, remove from running list, do counters & statistics */ 3185static int pktgen_stop_device(struct pktgen_dev *pkt_dev) 3186{ 3187 int nr_frags = pkt_dev->skb ? skb_shinfo(pkt_dev->skb)->nr_frags : -1; 3188 3189 if (!pkt_dev->running) { 3190 pr_warn("interface: %s is already stopped\n", 3191 pkt_dev->odevname); 3192 return -EINVAL; 3193 } 3194 3195 pkt_dev->running = 0; 3196 kfree_skb(pkt_dev->skb); 3197 pkt_dev->skb = NULL; 3198 pkt_dev->stopped_at = ktime_get(); 3199 3200 show_results(pkt_dev, nr_frags); 3201 3202 return 0; 3203} 3204 3205static struct pktgen_dev *next_to_run(struct pktgen_thread *t) 3206{ 3207 struct pktgen_dev *pkt_dev, *best = NULL; 3208 3209 rcu_read_lock(); 3210 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) { 3211 if (!pkt_dev->running) 3212 continue; 3213 if (best == NULL) 3214 best = pkt_dev; 3215 else if (ktime_compare(pkt_dev->next_tx, best->next_tx) < 0) 3216 best = pkt_dev; 3217 } 3218 rcu_read_unlock(); 3219 3220 return best; 3221} 3222 3223static void pktgen_stop(struct pktgen_thread *t) 3224{ 3225 struct pktgen_dev *pkt_dev; 3226 3227 func_enter(); 3228 3229 rcu_read_lock(); 3230 3231 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) { 3232 pktgen_stop_device(pkt_dev); 3233 } 3234 3235 rcu_read_unlock(); 3236} 3237 3238/* 3239 * one of our devices needs to be removed - find it 3240 * and remove it 3241 */ 3242static void pktgen_rem_one_if(struct pktgen_thread *t) 3243{ 3244 struct list_head *q, *n; 3245 struct pktgen_dev *cur; 3246 3247 func_enter(); 3248 3249 list_for_each_safe(q, n, &t->if_list) { 3250 cur = list_entry(q, struct pktgen_dev, list); 3251 3252 if (!cur->removal_mark) 3253 continue; 3254 3255 kfree_skb(cur->skb); 3256 cur->skb = NULL; 3257 3258 pktgen_remove_device(t, cur); 3259 3260 break; 3261 } 3262} 3263 3264static void pktgen_rem_all_ifs(struct pktgen_thread *t) 3265{ 3266 struct list_head *q, *n; 3267 struct pktgen_dev *cur; 3268 3269 func_enter(); 3270 3271 /* Remove all devices, free mem */ 3272 3273 list_for_each_safe(q, n, &t->if_list) { 3274 cur = list_entry(q, struct pktgen_dev, list); 3275 3276 kfree_skb(cur->skb); 3277 cur->skb = NULL; 3278 3279 pktgen_remove_device(t, cur); 3280 } 3281} 3282 3283static void pktgen_rem_thread(struct pktgen_thread *t) 3284{ 3285 /* Remove from the thread list */ 3286 remove_proc_entry(t->tsk->comm, t->net->proc_dir); 3287} 3288 3289static void pktgen_resched(struct pktgen_dev *pkt_dev) 3290{ 3291 ktime_t idle_start = ktime_get(); 3292 schedule(); 3293 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(ktime_get(), idle_start)); 3294} 3295 3296static void pktgen_wait_for_skb(struct pktgen_dev *pkt_dev) 3297{ 3298 ktime_t idle_start = ktime_get(); 3299 3300 while (atomic_read(&(pkt_dev->skb->users)) != 1) { 3301 if (signal_pending(current)) 3302 break; 3303 3304 if (need_resched()) 3305 pktgen_resched(pkt_dev); 3306 else 3307 cpu_relax(); 3308 } 3309 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(ktime_get(), idle_start)); 3310} 3311 3312static void pktgen_xmit(struct pktgen_dev *pkt_dev) 3313{ 3314 unsigned int burst = ACCESS_ONCE(pkt_dev->burst); 3315 struct net_device *odev = pkt_dev->odev; 3316 struct netdev_queue *txq; 3317 int ret; 3318 3319 /* If device is offline, then don't send */ 3320 if (unlikely(!netif_running(odev) || !netif_carrier_ok(odev))) { 3321 pktgen_stop_device(pkt_dev); 3322 return; 3323 } 3324 3325 /* This is max DELAY, this has special meaning of 3326 * "never transmit" 3327 */ 3328 if (unlikely(pkt_dev->delay == ULLONG_MAX)) { 3329 pkt_dev->next_tx = ktime_add_ns(ktime_get(), ULONG_MAX); 3330 return; 3331 } 3332 3333 /* If no skb or clone count exhausted then get new one */ 3334 if (!pkt_dev->skb || (pkt_dev->last_ok && 3335 ++pkt_dev->clone_count >= pkt_dev->clone_skb)) { 3336 /* build a new pkt */ 3337 kfree_skb(pkt_dev->skb); 3338 3339 pkt_dev->skb = fill_packet(odev, pkt_dev); 3340 if (pkt_dev->skb == NULL) { 3341 pr_err("ERROR: couldn't allocate skb in fill_packet\n"); 3342 schedule(); 3343 pkt_dev->clone_count--; /* back out increment, OOM */ 3344 return; 3345 } 3346 pkt_dev->last_pkt_size = pkt_dev->skb->len; 3347 pkt_dev->allocated_skbs++; 3348 pkt_dev->clone_count = 0; /* reset counter */ 3349 } 3350 3351 if (pkt_dev->delay && pkt_dev->last_ok) 3352 spin(pkt_dev, pkt_dev->next_tx); 3353 3354 txq = skb_get_tx_queue(odev, pkt_dev->skb); 3355 3356 local_bh_disable(); 3357 3358 HARD_TX_LOCK(odev, txq, smp_processor_id()); 3359 3360 if (unlikely(netif_xmit_frozen_or_drv_stopped(txq))) { 3361 ret = NETDEV_TX_BUSY; 3362 pkt_dev->last_ok = 0; 3363 goto unlock; 3364 } 3365 atomic_add(burst, &pkt_dev->skb->users); 3366 3367xmit_more: 3368 ret = netdev_start_xmit(pkt_dev->skb, odev, txq, --burst > 0); 3369 3370 switch (ret) { 3371 case NETDEV_TX_OK: 3372 pkt_dev->last_ok = 1; 3373 pkt_dev->sofar++; 3374 pkt_dev->seq_num++; 3375 pkt_dev->tx_bytes += pkt_dev->last_pkt_size; 3376 if (burst > 0 && !netif_xmit_frozen_or_drv_stopped(txq)) 3377 goto xmit_more; 3378 break; 3379 case NET_XMIT_DROP: 3380 case NET_XMIT_CN: 3381 case NET_XMIT_POLICED: 3382 /* skb has been consumed */ 3383 pkt_dev->errors++; 3384 break; 3385 default: /* Drivers are not supposed to return other values! */ 3386 net_info_ratelimited("%s xmit error: %d\n", 3387 pkt_dev->odevname, ret); 3388 pkt_dev->errors++; 3389 /* fallthru */ 3390 case NETDEV_TX_LOCKED: 3391 case NETDEV_TX_BUSY: 3392 /* Retry it next time */ 3393 atomic_dec(&(pkt_dev->skb->users)); 3394 pkt_dev->last_ok = 0; 3395 } 3396 if (unlikely(burst)) 3397 atomic_sub(burst, &pkt_dev->skb->users); 3398unlock: 3399 HARD_TX_UNLOCK(odev, txq); 3400 3401 local_bh_enable(); 3402 3403 /* If pkt_dev->count is zero, then run forever */ 3404 if ((pkt_dev->count != 0) && (pkt_dev->sofar >= pkt_dev->count)) { 3405 pktgen_wait_for_skb(pkt_dev); 3406 3407 /* Done with this */ 3408 pktgen_stop_device(pkt_dev); 3409 } 3410} 3411 3412/* 3413 * Main loop of the thread goes here 3414 */ 3415 3416static int pktgen_thread_worker(void *arg) 3417{ 3418 DEFINE_WAIT(wait); 3419 struct pktgen_thread *t = arg; 3420 struct pktgen_dev *pkt_dev = NULL; 3421 int cpu = t->cpu; 3422 3423 BUG_ON(smp_processor_id() != cpu); 3424 3425 init_waitqueue_head(&t->queue); 3426 complete(&t->start_done); 3427 3428 pr_debug("starting pktgen/%d: pid=%d\n", cpu, task_pid_nr(current)); 3429 3430 set_freezable(); 3431 3432 __set_current_state(TASK_RUNNING); 3433 3434 while (!kthread_should_stop()) { 3435 pkt_dev = next_to_run(t); 3436 3437 if (unlikely(!pkt_dev && t->control == 0)) { 3438 if (t->net->pktgen_exiting) 3439 break; 3440 wait_event_interruptible_timeout(t->queue, 3441 t->control != 0, 3442 HZ/10); 3443 try_to_freeze(); 3444 continue; 3445 } 3446 3447 if (likely(pkt_dev)) { 3448 pktgen_xmit(pkt_dev); 3449 3450 if (need_resched()) 3451 pktgen_resched(pkt_dev); 3452 else 3453 cpu_relax(); 3454 } 3455 3456 if (t->control & T_STOP) { 3457 pktgen_stop(t); 3458 t->control &= ~(T_STOP); 3459 } 3460 3461 if (t->control & T_RUN) { 3462 pktgen_run(t); 3463 t->control &= ~(T_RUN); 3464 } 3465 3466 if (t->control & T_REMDEVALL) { 3467 pktgen_rem_all_ifs(t); 3468 t->control &= ~(T_REMDEVALL); 3469 } 3470 3471 if (t->control & T_REMDEV) { 3472 pktgen_rem_one_if(t); 3473 t->control &= ~(T_REMDEV); 3474 } 3475 3476 try_to_freeze(); 3477 } 3478 set_current_state(TASK_INTERRUPTIBLE); 3479 3480 pr_debug("%s stopping all device\n", t->tsk->comm); 3481 pktgen_stop(t); 3482 3483 pr_debug("%s removing all device\n", t->tsk->comm); 3484 pktgen_rem_all_ifs(t); 3485 3486 pr_debug("%s removing thread\n", t->tsk->comm); 3487 pktgen_rem_thread(t); 3488 3489 /* Wait for kthread_stop */ 3490 while (!kthread_should_stop()) { 3491 set_current_state(TASK_INTERRUPTIBLE); 3492 schedule(); 3493 } 3494 __set_current_state(TASK_RUNNING); 3495 3496 return 0; 3497} 3498 3499static struct pktgen_dev *pktgen_find_dev(struct pktgen_thread *t, 3500 const char *ifname, bool exact) 3501{ 3502 struct pktgen_dev *p, *pkt_dev = NULL; 3503 size_t len = strlen(ifname); 3504 3505 rcu_read_lock(); 3506 list_for_each_entry_rcu(p, &t->if_list, list) 3507 if (strncmp(p->odevname, ifname, len) == 0) { 3508 if (p->odevname[len]) { 3509 if (exact || p->odevname[len] != '@') 3510 continue; 3511 } 3512 pkt_dev = p; 3513 break; 3514 } 3515 3516 rcu_read_unlock(); 3517 pr_debug("find_dev(%s) returning %p\n", ifname, pkt_dev); 3518 return pkt_dev; 3519} 3520 3521/* 3522 * Adds a dev at front of if_list. 3523 */ 3524 3525static int add_dev_to_thread(struct pktgen_thread *t, 3526 struct pktgen_dev *pkt_dev) 3527{ 3528 int rv = 0; 3529 3530 /* This function cannot be called concurrently, as its called 3531 * under pktgen_thread_lock mutex, but it can run from 3532 * userspace on another CPU than the kthread. The if_lock() 3533 * is used here to sync with concurrent instances of 3534 * _rem_dev_from_if_list() invoked via kthread, which is also 3535 * updating the if_list */ 3536 if_lock(t); 3537 3538 if (pkt_dev->pg_thread) { 3539 pr_err("ERROR: already assigned to a thread\n"); 3540 rv = -EBUSY; 3541 goto out; 3542 } 3543 3544 pkt_dev->running = 0; 3545 pkt_dev->pg_thread = t; 3546 list_add_rcu(&pkt_dev->list, &t->if_list); 3547 3548out: 3549 if_unlock(t); 3550 return rv; 3551} 3552 3553/* Called under thread lock */ 3554 3555static int pktgen_add_device(struct pktgen_thread *t, const char *ifname) 3556{ 3557 struct pktgen_dev *pkt_dev; 3558 int err; 3559 int node = cpu_to_node(t->cpu); 3560 3561 /* We don't allow a device to be on several threads */ 3562 3563 pkt_dev = __pktgen_NN_threads(t->net, ifname, FIND); 3564 if (pkt_dev) { 3565 pr_err("ERROR: interface already used\n"); 3566 return -EBUSY; 3567 } 3568 3569 pkt_dev = kzalloc_node(sizeof(struct pktgen_dev), GFP_KERNEL, node); 3570 if (!pkt_dev) 3571 return -ENOMEM; 3572 3573 strcpy(pkt_dev->odevname, ifname); 3574 pkt_dev->flows = vzalloc_node(MAX_CFLOWS * sizeof(struct flow_state), 3575 node); 3576 if (pkt_dev->flows == NULL) { 3577 kfree(pkt_dev); 3578 return -ENOMEM; 3579 } 3580 3581 pkt_dev->removal_mark = 0; 3582 pkt_dev->nfrags = 0; 3583 pkt_dev->delay = pg_delay_d; 3584 pkt_dev->count = pg_count_d; 3585 pkt_dev->sofar = 0; 3586 pkt_dev->udp_src_min = 9; /* sink port */ 3587 pkt_dev->udp_src_max = 9; 3588 pkt_dev->udp_dst_min = 9; 3589 pkt_dev->udp_dst_max = 9; 3590 pkt_dev->vlan_p = 0; 3591 pkt_dev->vlan_cfi = 0; 3592 pkt_dev->vlan_id = 0xffff; 3593 pkt_dev->svlan_p = 0; 3594 pkt_dev->svlan_cfi = 0; 3595 pkt_dev->svlan_id = 0xffff; 3596 pkt_dev->burst = 1; 3597 pkt_dev->node = -1; 3598 3599 err = pktgen_setup_dev(t->net, pkt_dev, ifname); 3600 if (err) 3601 goto out1; 3602 if (pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING) 3603 pkt_dev->clone_skb = pg_clone_skb_d; 3604 3605 pkt_dev->entry = proc_create_data(ifname, 0600, t->net->proc_dir, 3606 &pktgen_if_fops, pkt_dev); 3607 if (!pkt_dev->entry) { 3608 pr_err("cannot create %s/%s procfs entry\n", 3609 PG_PROC_DIR, ifname); 3610 err = -EINVAL; 3611 goto out2; 3612 } 3613#ifdef CONFIG_XFRM 3614 pkt_dev->ipsmode = XFRM_MODE_TRANSPORT; 3615 pkt_dev->ipsproto = IPPROTO_ESP; 3616 3617 /* xfrm tunnel mode needs additional dst to extract outter 3618 * ip header protocol/ttl/id field, here creat a phony one. 3619 * instead of looking for a valid rt, which definitely hurting 3620 * performance under such circumstance. 3621 */ 3622 pkt_dev->dstops.family = AF_INET; 3623 pkt_dev->dst.dev = pkt_dev->odev; 3624 dst_init_metrics(&pkt_dev->dst, pktgen_dst_metrics, false); 3625 pkt_dev->dst.child = &pkt_dev->dst; 3626 pkt_dev->dst.ops = &pkt_dev->dstops; 3627#endif 3628 3629 return add_dev_to_thread(t, pkt_dev); 3630out2: 3631 dev_put(pkt_dev->odev); 3632out1: 3633#ifdef CONFIG_XFRM 3634 free_SAs(pkt_dev); 3635#endif 3636 vfree(pkt_dev->flows); 3637 kfree(pkt_dev); 3638 return err; 3639} 3640 3641static int __net_init pktgen_create_thread(int cpu, struct pktgen_net *pn) 3642{ 3643 struct pktgen_thread *t; 3644 struct proc_dir_entry *pe; 3645 struct task_struct *p; 3646 3647 t = kzalloc_node(sizeof(struct pktgen_thread), GFP_KERNEL, 3648 cpu_to_node(cpu)); 3649 if (!t) { 3650 pr_err("ERROR: out of memory, can't create new thread\n"); 3651 return -ENOMEM; 3652 } 3653 3654 spin_lock_init(&t->if_lock); 3655 t->cpu = cpu; 3656 3657 INIT_LIST_HEAD(&t->if_list); 3658 3659 list_add_tail(&t->th_list, &pn->pktgen_threads); 3660 init_completion(&t->start_done); 3661 3662 p = kthread_create_on_node(pktgen_thread_worker, 3663 t, 3664 cpu_to_node(cpu), 3665 "kpktgend_%d", cpu); 3666 if (IS_ERR(p)) { 3667 pr_err("kernel_thread() failed for cpu %d\n", t->cpu); 3668 list_del(&t->th_list); 3669 kfree(t); 3670 return PTR_ERR(p); 3671 } 3672 kthread_bind(p, cpu); 3673 t->tsk = p; 3674 3675 pe = proc_create_data(t->tsk->comm, 0600, pn->proc_dir, 3676 &pktgen_thread_fops, t); 3677 if (!pe) { 3678 pr_err("cannot create %s/%s procfs entry\n", 3679 PG_PROC_DIR, t->tsk->comm); 3680 kthread_stop(p); 3681 list_del(&t->th_list); 3682 kfree(t); 3683 return -EINVAL; 3684 } 3685 3686 t->net = pn; 3687 wake_up_process(p); 3688 wait_for_completion(&t->start_done); 3689 3690 return 0; 3691} 3692 3693/* 3694 * Removes a device from the thread if_list. 3695 */ 3696static void _rem_dev_from_if_list(struct pktgen_thread *t, 3697 struct pktgen_dev *pkt_dev) 3698{ 3699 struct list_head *q, *n; 3700 struct pktgen_dev *p; 3701 3702 if_lock(t); 3703 list_for_each_safe(q, n, &t->if_list) { 3704 p = list_entry(q, struct pktgen_dev, list); 3705 if (p == pkt_dev) 3706 list_del_rcu(&p->list); 3707 } 3708 if_unlock(t); 3709} 3710 3711static int pktgen_remove_device(struct pktgen_thread *t, 3712 struct pktgen_dev *pkt_dev) 3713{ 3714 pr_debug("remove_device pkt_dev=%p\n", pkt_dev); 3715 3716 if (pkt_dev->running) { 3717 pr_warn("WARNING: trying to remove a running interface, stopping it now\n"); 3718 pktgen_stop_device(pkt_dev); 3719 } 3720 3721 /* Dis-associate from the interface */ 3722 3723 if (pkt_dev->odev) { 3724 dev_put(pkt_dev->odev); 3725 pkt_dev->odev = NULL; 3726 } 3727 3728 /* Remove proc before if_list entry, because add_device uses 3729 * list to determine if interface already exist, avoid race 3730 * with proc_create_data() */ 3731 if (pkt_dev->entry) 3732 proc_remove(pkt_dev->entry); 3733 3734 /* And update the thread if_list */ 3735 _rem_dev_from_if_list(t, pkt_dev); 3736 3737#ifdef CONFIG_XFRM 3738 free_SAs(pkt_dev); 3739#endif 3740 vfree(pkt_dev->flows); 3741 if (pkt_dev->page) 3742 put_page(pkt_dev->page); 3743 kfree_rcu(pkt_dev, rcu); 3744 return 0; 3745} 3746 3747static int __net_init pg_net_init(struct net *net) 3748{ 3749 struct pktgen_net *pn = net_generic(net, pg_net_id); 3750 struct proc_dir_entry *pe; 3751 int cpu, ret = 0; 3752 3753 pn->net = net; 3754 INIT_LIST_HEAD(&pn->pktgen_threads); 3755 pn->pktgen_exiting = false; 3756 pn->proc_dir = proc_mkdir(PG_PROC_DIR, pn->net->proc_net); 3757 if (!pn->proc_dir) { 3758 pr_warn("cannot create /proc/net/%s\n", PG_PROC_DIR); 3759 return -ENODEV; 3760 } 3761 pe = proc_create(PGCTRL, 0600, pn->proc_dir, &pktgen_fops); 3762 if (pe == NULL) { 3763 pr_err("cannot create %s procfs entry\n", PGCTRL); 3764 ret = -EINVAL; 3765 goto remove; 3766 } 3767 3768 for_each_online_cpu(cpu) { 3769 int err; 3770 3771 err = pktgen_create_thread(cpu, pn); 3772 if (err) 3773 pr_warn("Cannot create thread for cpu %d (%d)\n", 3774 cpu, err); 3775 } 3776 3777 if (list_empty(&pn->pktgen_threads)) { 3778 pr_err("Initialization failed for all threads\n"); 3779 ret = -ENODEV; 3780 goto remove_entry; 3781 } 3782 3783 return 0; 3784 3785remove_entry: 3786 remove_proc_entry(PGCTRL, pn->proc_dir); 3787remove: 3788 remove_proc_entry(PG_PROC_DIR, pn->net->proc_net); 3789 return ret; 3790} 3791 3792static void __net_exit pg_net_exit(struct net *net) 3793{ 3794 struct pktgen_net *pn = net_generic(net, pg_net_id); 3795 struct pktgen_thread *t; 3796 struct list_head *q, *n; 3797 LIST_HEAD(list); 3798 3799 /* Stop all interfaces & threads */ 3800 pn->pktgen_exiting = true; 3801 3802 mutex_lock(&pktgen_thread_lock); 3803 list_splice_init(&pn->pktgen_threads, &list); 3804 mutex_unlock(&pktgen_thread_lock); 3805 3806 list_for_each_safe(q, n, &list) { 3807 t = list_entry(q, struct pktgen_thread, th_list); 3808 list_del(&t->th_list); 3809 kthread_stop(t->tsk); 3810 kfree(t); 3811 } 3812 3813 remove_proc_entry(PGCTRL, pn->proc_dir); 3814 remove_proc_entry(PG_PROC_DIR, pn->net->proc_net); 3815} 3816 3817static struct pernet_operations pg_net_ops = { 3818 .init = pg_net_init, 3819 .exit = pg_net_exit, 3820 .id = &pg_net_id, 3821 .size = sizeof(struct pktgen_net), 3822}; 3823 3824static int __init pg_init(void) 3825{ 3826 int ret = 0; 3827 3828 pr_info("%s", version); 3829 ret = register_pernet_subsys(&pg_net_ops); 3830 if (ret) 3831 return ret; 3832 ret = register_netdevice_notifier(&pktgen_notifier_block); 3833 if (ret) 3834 unregister_pernet_subsys(&pg_net_ops); 3835 3836 return ret; 3837} 3838 3839static void __exit pg_cleanup(void) 3840{ 3841 unregister_netdevice_notifier(&pktgen_notifier_block); 3842 unregister_pernet_subsys(&pg_net_ops); 3843 /* Don't need rcu_barrier() due to use of kfree_rcu() */ 3844} 3845 3846module_init(pg_init); 3847module_exit(pg_cleanup); 3848 3849MODULE_AUTHOR("Robert Olsson <robert.olsson@its.uu.se>"); 3850MODULE_DESCRIPTION("Packet Generator tool"); 3851MODULE_LICENSE("GPL"); 3852MODULE_VERSION(VERSION); 3853module_param(pg_count_d, int, 0); 3854MODULE_PARM_DESC(pg_count_d, "Default number of packets to inject"); 3855module_param(pg_delay_d, int, 0); 3856MODULE_PARM_DESC(pg_delay_d, "Default delay between packets (nanoseconds)"); 3857module_param(pg_clone_skb_d, int, 0); 3858MODULE_PARM_DESC(pg_clone_skb_d, "Default number of copies of the same packet"); 3859module_param(debug, int, 0); 3860MODULE_PARM_DESC(debug, "Enable debugging of pktgen module"); 3861