linux/net/core/pktgen.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * Authors:
   4 * Copyright 2001, 2002 by Robert Olsson <robert.olsson@its.uu.se>
   5 *                             Uppsala University and
   6 *                             Swedish University of Agricultural Sciences
   7 *
   8 * Alexey Kuznetsov  <kuznet@ms2.inr.ac.ru>
   9 * Ben Greear <greearb@candelatech.com>
  10 * Jens Låås <jens.laas@data.slu.se>
  11 *
  12 * A tool for loading the network with preconfigurated packets.
  13 * The tool is implemented as a linux module.  Parameters are output
  14 * device, delay (to hard_xmit), number of packets, and whether
  15 * to use multiple SKBs or just the same one.
  16 * pktgen uses the installed interface's output routine.
  17 *
  18 * Additional hacking by:
  19 *
  20 * Jens.Laas@data.slu.se
  21 * Improved by ANK. 010120.
  22 * Improved by ANK even more. 010212.
  23 * MAC address typo fixed. 010417 --ro
  24 * Integrated.  020301 --DaveM
  25 * Added multiskb option 020301 --DaveM
  26 * Scaling of results. 020417--sigurdur@linpro.no
  27 * Significant re-work of the module:
  28 *   *  Convert to threaded model to more efficiently be able to transmit
  29 *       and receive on multiple interfaces at once.
  30 *   *  Converted many counters to __u64 to allow longer runs.
  31 *   *  Allow configuration of ranges, like min/max IP address, MACs,
  32 *       and UDP-ports, for both source and destination, and can
  33 *       set to use a random distribution or sequentially walk the range.
  34 *   *  Can now change most values after starting.
  35 *   *  Place 12-byte packet in UDP payload with magic number,
  36 *       sequence number, and timestamp.
  37 *   *  Add receiver code that detects dropped pkts, re-ordered pkts, and
  38 *       latencies (with micro-second) precision.
  39 *   *  Add IOCTL interface to easily get counters & configuration.
  40 *   --Ben Greear <greearb@candelatech.com>
  41 *
  42 * Renamed multiskb to clone_skb and cleaned up sending core for two distinct
  43 * skb modes. A clone_skb=0 mode for Ben "ranges" work and a clone_skb != 0
  44 * as a "fastpath" with a configurable number of clones after alloc's.
  45 * clone_skb=0 means all packets are allocated this also means ranges time
  46 * stamps etc can be used. clone_skb=100 means 1 malloc is followed by 100
  47 * clones.
  48 *
  49 * Also moved to /proc/net/pktgen/
  50 * --ro
  51 *
  52 * Sept 10:  Fixed threading/locking.  Lots of bone-headed and more clever
  53 *    mistakes.  Also merged in DaveM's patch in the -pre6 patch.
  54 * --Ben Greear <greearb@candelatech.com>
  55 *
  56 * Integrated to 2.5.x 021029 --Lucio Maciel (luciomaciel@zipmail.com.br)
  57 *
  58 * 021124 Finished major redesign and rewrite for new functionality.
  59 * See Documentation/networking/pktgen.rst for how to use this.
  60 *
  61 * The new operation:
  62 * For each CPU one thread/process is created at start. This process checks
  63 * for running devices in the if_list and sends packets until count is 0 it
  64 * also the thread checks the thread->control which is used for inter-process
  65 * communication. controlling process "posts" operations to the threads this
  66 * way.
  67 * The if_list is RCU protected, and the if_lock remains to protect updating
  68 * of if_list, from "add_device" as it invoked from userspace (via proc write).
  69 *
  70 * By design there should only be *one* "controlling" process. In practice
  71 * multiple write accesses gives unpredictable result. Understood by "write"
  72 * to /proc gives result code thats should be read be the "writer".
  73 * For practical use this should be no problem.
  74 *
  75 * Note when adding devices to a specific CPU there good idea to also assign
  76 * /proc/irq/XX/smp_affinity so TX-interrupts gets bound to the same CPU.
  77 * --ro
  78 *
  79 * Fix refcount off by one if first packet fails, potential null deref,
  80 * memleak 030710- KJP
  81 *
  82 * First "ranges" functionality for ipv6 030726 --ro
  83 *
  84 * Included flow support. 030802 ANK.
  85 *
  86 * Fixed unaligned access on IA-64 Grant Grundler <grundler@parisc-linux.org>
  87 *
  88 * Remove if fix from added Harald Welte <laforge@netfilter.org> 040419
  89 * ia64 compilation fix from  Aron Griffis <aron@hp.com> 040604
  90 *
  91 * New xmit() return, do_div and misc clean up by Stephen Hemminger
  92 * <shemminger@osdl.org> 040923
  93 *
  94 * Randy Dunlap fixed u64 printk compiler warning
  95 *
  96 * Remove FCS from BW calculation.  Lennert Buytenhek <buytenh@wantstofly.org>
  97 * New time handling. Lennert Buytenhek <buytenh@wantstofly.org> 041213
  98 *
  99 * Corrections from Nikolai Malykh (nmalykh@bilim.com)
 100 * Removed unused flags F_SET_SRCMAC & F_SET_SRCIP 041230
 101 *
 102 * interruptible_sleep_on_timeout() replaced Nishanth Aravamudan <nacc@us.ibm.com>
 103 * 050103
 104 *
 105 * MPLS support by Steven Whitehouse <steve@chygwyn.com>
 106 *
 107 * 802.1Q/Q-in-Q support by Francesco Fondelli (FF) <francesco.fondelli@gmail.com>
 108 *
 109 * Fixed src_mac command to set source mac of packet to value specified in
 110 * command by Adit Ranadive <adit.262@gmail.com>
 111 */
 112
 113#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 114
 115#include <linux/sys.h>
 116#include <linux/types.h>
 117#include <linux/module.h>
 118#include <linux/moduleparam.h>
 119#include <linux/kernel.h>
 120#include <linux/mutex.h>
 121#include <linux/sched.h>
 122#include <linux/slab.h>
 123#include <linux/vmalloc.h>
 124#include <linux/unistd.h>
 125#include <linux/string.h>
 126#include <linux/ptrace.h>
 127#include <linux/errno.h>
 128#include <linux/ioport.h>
 129#include <linux/interrupt.h>
 130#include <linux/capability.h>
 131#include <linux/hrtimer.h>
 132#include <linux/freezer.h>
 133#include <linux/delay.h>
 134#include <linux/timer.h>
 135#include <linux/list.h>
 136#include <linux/init.h>
 137#include <linux/skbuff.h>
 138#include <linux/netdevice.h>
 139#include <linux/inet.h>
 140#include <linux/inetdevice.h>
 141#include <linux/rtnetlink.h>
 142#include <linux/if_arp.h>
 143#include <linux/if_vlan.h>
 144#include <linux/in.h>
 145#include <linux/ip.h>
 146#include <linux/ipv6.h>
 147#include <linux/udp.h>
 148#include <linux/proc_fs.h>
 149#include <linux/seq_file.h>
 150#include <linux/wait.h>
 151#include <linux/etherdevice.h>
 152#include <linux/kthread.h>
 153#include <linux/prefetch.h>
 154#include <linux/mmzone.h>
 155#include <net/net_namespace.h>
 156#include <net/checksum.h>
 157#include <net/ipv6.h>
 158#include <net/udp.h>
 159#include <net/ip6_checksum.h>
 160#include <net/addrconf.h>
 161#ifdef CONFIG_XFRM
 162#include <net/xfrm.h>
 163#endif
 164#include <net/netns/generic.h>
 165#include <asm/byteorder.h>
 166#include <linux/rcupdate.h>
 167#include <linux/bitops.h>
 168#include <linux/io.h>
 169#include <linux/timex.h>
 170#include <linux/uaccess.h>
 171#include <asm/dma.h>
 172#include <asm/div64.h>          /* do_div */
 173
 174#define VERSION "2.75"
 175#define IP_NAME_SZ 32
 176#define MAX_MPLS_LABELS 16 /* This is the max label stack depth */
 177#define MPLS_STACK_BOTTOM htonl(0x00000100)
 178
 179#define func_enter() pr_debug("entering %s\n", __func__);
 180
 181#define PKT_FLAGS                                                       \
 182        pf(IPV6)                /* Interface in IPV6 Mode */            \
 183        pf(IPSRC_RND)           /* IP-Src Random  */                    \
 184        pf(IPDST_RND)           /* IP-Dst Random  */                    \
 185        pf(TXSIZE_RND)          /* Transmit size is random */           \
 186        pf(UDPSRC_RND)          /* UDP-Src Random */                    \
 187        pf(UDPDST_RND)          /* UDP-Dst Random */                    \
 188        pf(UDPCSUM)             /* Include UDP checksum */              \
 189        pf(NO_TIMESTAMP)        /* Don't timestamp packets (default TS) */ \
 190        pf(MPLS_RND)            /* Random MPLS labels */                \
 191        pf(QUEUE_MAP_RND)       /* queue map Random */                  \
 192        pf(QUEUE_MAP_CPU)       /* queue map mirrors smp_processor_id() */ \
 193        pf(FLOW_SEQ)            /* Sequential flows */                  \
 194        pf(IPSEC)               /* ipsec on for flows */                \
 195        pf(MACSRC_RND)          /* MAC-Src Random */                    \
 196        pf(MACDST_RND)          /* MAC-Dst Random */                    \
 197        pf(VID_RND)             /* Random VLAN ID */                    \
 198        pf(SVID_RND)            /* Random SVLAN ID */                   \
 199        pf(NODE)                /* Node memory alloc*/                  \
 200
 201#define pf(flag)                flag##_SHIFT,
 202enum pkt_flags {
 203        PKT_FLAGS
 204};
 205#undef pf
 206
 207/* Device flag bits */
 208#define pf(flag)                static const __u32 F_##flag = (1<<flag##_SHIFT);
 209PKT_FLAGS
 210#undef pf
 211
 212#define pf(flag)                __stringify(flag),
 213static char *pkt_flag_names[] = {
 214        PKT_FLAGS
 215};
 216#undef pf
 217
 218#define NR_PKT_FLAGS            ARRAY_SIZE(pkt_flag_names)
 219
 220/* Thread control flag bits */
 221#define T_STOP        (1<<0)    /* Stop run */
 222#define T_RUN         (1<<1)    /* Start run */
 223#define T_REMDEVALL   (1<<2)    /* Remove all devs */
 224#define T_REMDEV      (1<<3)    /* Remove one dev */
 225
 226/* Xmit modes */
 227#define M_START_XMIT            0       /* Default normal TX */
 228#define M_NETIF_RECEIVE         1       /* Inject packets into stack */
 229#define M_QUEUE_XMIT            2       /* Inject packet into qdisc */
 230
 231/* If lock -- protects updating of if_list */
 232#define   if_lock(t)           mutex_lock(&(t->if_lock));
 233#define   if_unlock(t)           mutex_unlock(&(t->if_lock));
 234
 235/* Used to help with determining the pkts on receive */
 236#define PKTGEN_MAGIC 0xbe9be955
 237#define PG_PROC_DIR "pktgen"
 238#define PGCTRL      "pgctrl"
 239
 240#define MAX_CFLOWS  65536
 241
 242#define VLAN_TAG_SIZE(x) ((x)->vlan_id == 0xffff ? 0 : 4)
 243#define SVLAN_TAG_SIZE(x) ((x)->svlan_id == 0xffff ? 0 : 4)
 244
 245struct flow_state {
 246        __be32 cur_daddr;
 247        int count;
 248#ifdef CONFIG_XFRM
 249        struct xfrm_state *x;
 250#endif
 251        __u32 flags;
 252};
 253
 254/* flow flag bits */
 255#define F_INIT   (1<<0)         /* flow has been initialized */
 256
 257struct pktgen_dev {
 258        /*
 259         * Try to keep frequent/infrequent used vars. separated.
 260         */
 261        struct proc_dir_entry *entry;   /* proc file */
 262        struct pktgen_thread *pg_thread;/* the owner */
 263        struct list_head list;          /* chaining in the thread's run-queue */
 264        struct rcu_head  rcu;           /* freed by RCU */
 265
 266        int running;            /* if false, the test will stop */
 267
 268        /* If min != max, then we will either do a linear iteration, or
 269         * we will do a random selection from within the range.
 270         */
 271        __u32 flags;
 272        int xmit_mode;
 273        int min_pkt_size;
 274        int max_pkt_size;
 275        int pkt_overhead;       /* overhead for MPLS, VLANs, IPSEC etc */
 276        int nfrags;
 277        int removal_mark;       /* non-zero => the device is marked for
 278                                 * removal by worker thread */
 279
 280        struct page *page;
 281        u64 delay;              /* nano-seconds */
 282
 283        __u64 count;            /* Default No packets to send */
 284        __u64 sofar;            /* How many pkts we've sent so far */
 285        __u64 tx_bytes;         /* How many bytes we've transmitted */
 286        __u64 errors;           /* Errors when trying to transmit, */
 287
 288        /* runtime counters relating to clone_skb */
 289
 290        __u32 clone_count;
 291        int last_ok;            /* Was last skb sent?
 292                                 * Or a failed transmit of some sort?
 293                                 * This will keep sequence numbers in order
 294                                 */
 295        ktime_t next_tx;
 296        ktime_t started_at;
 297        ktime_t stopped_at;
 298        u64     idle_acc;       /* nano-seconds */
 299
 300        __u32 seq_num;
 301
 302        int clone_skb;          /*
 303                                 * Use multiple SKBs during packet gen.
 304                                 * If this number is greater than 1, then
 305                                 * that many copies of the same packet will be
 306                                 * sent before a new packet is allocated.
 307                                 * If you want to send 1024 identical packets
 308                                 * before creating a new packet,
 309                                 * set clone_skb to 1024.
 310                                 */
 311
 312        char dst_min[IP_NAME_SZ];       /* IP, ie 1.2.3.4 */
 313        char dst_max[IP_NAME_SZ];       /* IP, ie 1.2.3.4 */
 314        char src_min[IP_NAME_SZ];       /* IP, ie 1.2.3.4 */
 315        char src_max[IP_NAME_SZ];       /* IP, ie 1.2.3.4 */
 316
 317        struct in6_addr in6_saddr;
 318        struct in6_addr in6_daddr;
 319        struct in6_addr cur_in6_daddr;
 320        struct in6_addr cur_in6_saddr;
 321        /* For ranges */
 322        struct in6_addr min_in6_daddr;
 323        struct in6_addr max_in6_daddr;
 324        struct in6_addr min_in6_saddr;
 325        struct in6_addr max_in6_saddr;
 326
 327        /* If we're doing ranges, random or incremental, then this
 328         * defines the min/max for those ranges.
 329         */
 330        __be32 saddr_min;       /* inclusive, source IP address */
 331        __be32 saddr_max;       /* exclusive, source IP address */
 332        __be32 daddr_min;       /* inclusive, dest IP address */
 333        __be32 daddr_max;       /* exclusive, dest IP address */
 334
 335        __u16 udp_src_min;      /* inclusive, source UDP port */
 336        __u16 udp_src_max;      /* exclusive, source UDP port */
 337        __u16 udp_dst_min;      /* inclusive, dest UDP port */
 338        __u16 udp_dst_max;      /* exclusive, dest UDP port */
 339
 340        /* DSCP + ECN */
 341        __u8 tos;            /* six MSB of (former) IPv4 TOS
 342                                are for dscp codepoint */
 343        __u8 traffic_class;  /* ditto for the (former) Traffic Class in IPv6
 344                                (see RFC 3260, sec. 4) */
 345
 346        /* MPLS */
 347        unsigned int nr_labels; /* Depth of stack, 0 = no MPLS */
 348        __be32 labels[MAX_MPLS_LABELS];
 349
 350        /* VLAN/SVLAN (802.1Q/Q-in-Q) */
 351        __u8  vlan_p;
 352        __u8  vlan_cfi;
 353        __u16 vlan_id;  /* 0xffff means no vlan tag */
 354
 355        __u8  svlan_p;
 356        __u8  svlan_cfi;
 357        __u16 svlan_id; /* 0xffff means no svlan tag */
 358
 359        __u32 src_mac_count;    /* How many MACs to iterate through */
 360        __u32 dst_mac_count;    /* How many MACs to iterate through */
 361
 362        unsigned char dst_mac[ETH_ALEN];
 363        unsigned char src_mac[ETH_ALEN];
 364
 365        __u32 cur_dst_mac_offset;
 366        __u32 cur_src_mac_offset;
 367        __be32 cur_saddr;
 368        __be32 cur_daddr;
 369        __u16 ip_id;
 370        __u16 cur_udp_dst;
 371        __u16 cur_udp_src;
 372        __u16 cur_queue_map;
 373        __u32 cur_pkt_size;
 374        __u32 last_pkt_size;
 375
 376        __u8 hh[14];
 377        /* = {
 378           0x00, 0x80, 0xC8, 0x79, 0xB3, 0xCB,
 379
 380           We fill in SRC address later
 381           0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 382           0x08, 0x00
 383           };
 384         */
 385        __u16 pad;              /* pad out the hh struct to an even 16 bytes */
 386
 387        struct sk_buff *skb;    /* skb we are to transmit next, used for when we
 388                                 * are transmitting the same one multiple times
 389                                 */
 390        struct net_device *odev; /* The out-going device.
 391                                  * Note that the device should have it's
 392                                  * pg_info pointer pointing back to this
 393                                  * device.
 394                                  * Set when the user specifies the out-going
 395                                  * device name (not when the inject is
 396                                  * started as it used to do.)
 397                                  */
 398        char odevname[32];
 399        struct flow_state *flows;
 400        unsigned int cflows;    /* Concurrent flows (config) */
 401        unsigned int lflow;             /* Flow length  (config) */
 402        unsigned int nflows;    /* accumulated flows (stats) */
 403        unsigned int curfl;             /* current sequenced flow (state)*/
 404
 405        u16 queue_map_min;
 406        u16 queue_map_max;
 407        __u32 skb_priority;     /* skb priority field */
 408        unsigned int burst;     /* number of duplicated packets to burst */
 409        int node;               /* Memory node */
 410
 411#ifdef CONFIG_XFRM
 412        __u8    ipsmode;                /* IPSEC mode (config) */
 413        __u8    ipsproto;               /* IPSEC type (config) */
 414        __u32   spi;
 415        struct xfrm_dst xdst;
 416        struct dst_ops dstops;
 417#endif
 418        char result[512];
 419};
 420
 421struct pktgen_hdr {
 422        __be32 pgh_magic;
 423        __be32 seq_num;
 424        __be32 tv_sec;
 425        __be32 tv_usec;
 426};
 427
 428
 429static unsigned int pg_net_id __read_mostly;
 430
 431struct pktgen_net {
 432        struct net              *net;
 433        struct proc_dir_entry   *proc_dir;
 434        struct list_head        pktgen_threads;
 435        bool                    pktgen_exiting;
 436};
 437
 438struct pktgen_thread {
 439        struct mutex if_lock;           /* for list of devices */
 440        struct list_head if_list;       /* All device here */
 441        struct list_head th_list;
 442        struct task_struct *tsk;
 443        char result[512];
 444
 445        /* Field for thread to receive "posted" events terminate,
 446           stop ifs etc. */
 447
 448        u32 control;
 449        int cpu;
 450
 451        wait_queue_head_t queue;
 452        struct completion start_done;
 453        struct pktgen_net *net;
 454};
 455
 456#define REMOVE 1
 457#define FIND   0
 458
 459static const char version[] =
 460        "Packet Generator for packet performance testing. "
 461        "Version: " VERSION "\n";
 462
 463static int pktgen_remove_device(struct pktgen_thread *t, struct pktgen_dev *i);
 464static int pktgen_add_device(struct pktgen_thread *t, const char *ifname);
 465static struct pktgen_dev *pktgen_find_dev(struct pktgen_thread *t,
 466                                          const char *ifname, bool exact);
 467static int pktgen_device_event(struct notifier_block *, unsigned long, void *);
 468static void pktgen_run_all_threads(struct pktgen_net *pn);
 469static void pktgen_reset_all_threads(struct pktgen_net *pn);
 470static void pktgen_stop_all_threads(struct pktgen_net *pn);
 471
 472static void pktgen_stop(struct pktgen_thread *t);
 473static void pktgen_clear_counters(struct pktgen_dev *pkt_dev);
 474
 475/* Module parameters, defaults. */
 476static int pg_count_d __read_mostly = 1000;
 477static int pg_delay_d __read_mostly;
 478static int pg_clone_skb_d  __read_mostly;
 479static int debug  __read_mostly;
 480
 481static DEFINE_MUTEX(pktgen_thread_lock);
 482
 483static struct notifier_block pktgen_notifier_block = {
 484        .notifier_call = pktgen_device_event,
 485};
 486
 487/*
 488 * /proc handling functions
 489 *
 490 */
 491
 492static int pgctrl_show(struct seq_file *seq, void *v)
 493{
 494        seq_puts(seq, version);
 495        return 0;
 496}
 497
 498static ssize_t pgctrl_write(struct file *file, const char __user *buf,
 499                            size_t count, loff_t *ppos)
 500{
 501        char data[128];
 502        struct pktgen_net *pn = net_generic(current->nsproxy->net_ns, pg_net_id);
 503
 504        if (!capable(CAP_NET_ADMIN))
 505                return -EPERM;
 506
 507        if (count == 0)
 508                return -EINVAL;
 509
 510        if (count > sizeof(data))
 511                count = sizeof(data);
 512
 513        if (copy_from_user(data, buf, count))
 514                return -EFAULT;
 515
 516        data[count - 1] = 0;    /* Strip trailing '\n' and terminate string */
 517
 518        if (!strcmp(data, "stop"))
 519                pktgen_stop_all_threads(pn);
 520        else if (!strcmp(data, "start"))
 521                pktgen_run_all_threads(pn);
 522        else if (!strcmp(data, "reset"))
 523                pktgen_reset_all_threads(pn);
 524        else
 525                return -EINVAL;
 526
 527        return count;
 528}
 529
 530static int pgctrl_open(struct inode *inode, struct file *file)
 531{
 532        return single_open(file, pgctrl_show, PDE_DATA(inode));
 533}
 534
 535static const struct proc_ops pktgen_proc_ops = {
 536        .proc_open      = pgctrl_open,
 537        .proc_read      = seq_read,
 538        .proc_lseek     = seq_lseek,
 539        .proc_write     = pgctrl_write,
 540        .proc_release   = single_release,
 541};
 542
 543static int pktgen_if_show(struct seq_file *seq, void *v)
 544{
 545        const struct pktgen_dev *pkt_dev = seq->private;
 546        ktime_t stopped;
 547        unsigned int i;
 548        u64 idle;
 549
 550        seq_printf(seq,
 551                   "Params: count %llu  min_pkt_size: %u  max_pkt_size: %u\n",
 552                   (unsigned long long)pkt_dev->count, pkt_dev->min_pkt_size,
 553                   pkt_dev->max_pkt_size);
 554
 555        seq_printf(seq,
 556                   "     frags: %d  delay: %llu  clone_skb: %d  ifname: %s\n",
 557                   pkt_dev->nfrags, (unsigned long long) pkt_dev->delay,
 558                   pkt_dev->clone_skb, pkt_dev->odevname);
 559
 560        seq_printf(seq, "     flows: %u flowlen: %u\n", pkt_dev->cflows,
 561                   pkt_dev->lflow);
 562
 563        seq_printf(seq,
 564                   "     queue_map_min: %u  queue_map_max: %u\n",
 565                   pkt_dev->queue_map_min,
 566                   pkt_dev->queue_map_max);
 567
 568        if (pkt_dev->skb_priority)
 569                seq_printf(seq, "     skb_priority: %u\n",
 570                           pkt_dev->skb_priority);
 571
 572        if (pkt_dev->flags & F_IPV6) {
 573                seq_printf(seq,
 574                           "     saddr: %pI6c  min_saddr: %pI6c  max_saddr: %pI6c\n"
 575                           "     daddr: %pI6c  min_daddr: %pI6c  max_daddr: %pI6c\n",
 576                           &pkt_dev->in6_saddr,
 577                           &pkt_dev->min_in6_saddr, &pkt_dev->max_in6_saddr,
 578                           &pkt_dev->in6_daddr,
 579                           &pkt_dev->min_in6_daddr, &pkt_dev->max_in6_daddr);
 580        } else {
 581                seq_printf(seq,
 582                           "     dst_min: %s  dst_max: %s\n",
 583                           pkt_dev->dst_min, pkt_dev->dst_max);
 584                seq_printf(seq,
 585                           "     src_min: %s  src_max: %s\n",
 586                           pkt_dev->src_min, pkt_dev->src_max);
 587        }
 588
 589        seq_puts(seq, "     src_mac: ");
 590
 591        seq_printf(seq, "%pM ",
 592                   is_zero_ether_addr(pkt_dev->src_mac) ?
 593                             pkt_dev->odev->dev_addr : pkt_dev->src_mac);
 594
 595        seq_puts(seq, "dst_mac: ");
 596        seq_printf(seq, "%pM\n", pkt_dev->dst_mac);
 597
 598        seq_printf(seq,
 599                   "     udp_src_min: %d  udp_src_max: %d"
 600                   "  udp_dst_min: %d  udp_dst_max: %d\n",
 601                   pkt_dev->udp_src_min, pkt_dev->udp_src_max,
 602                   pkt_dev->udp_dst_min, pkt_dev->udp_dst_max);
 603
 604        seq_printf(seq,
 605                   "     src_mac_count: %d  dst_mac_count: %d\n",
 606                   pkt_dev->src_mac_count, pkt_dev->dst_mac_count);
 607
 608        if (pkt_dev->nr_labels) {
 609                seq_puts(seq, "     mpls: ");
 610                for (i = 0; i < pkt_dev->nr_labels; i++)
 611                        seq_printf(seq, "%08x%s", ntohl(pkt_dev->labels[i]),
 612                                   i == pkt_dev->nr_labels-1 ? "\n" : ", ");
 613        }
 614
 615        if (pkt_dev->vlan_id != 0xffff)
 616                seq_printf(seq, "     vlan_id: %u  vlan_p: %u  vlan_cfi: %u\n",
 617                           pkt_dev->vlan_id, pkt_dev->vlan_p,
 618                           pkt_dev->vlan_cfi);
 619
 620        if (pkt_dev->svlan_id != 0xffff)
 621                seq_printf(seq, "     svlan_id: %u  vlan_p: %u  vlan_cfi: %u\n",
 622                           pkt_dev->svlan_id, pkt_dev->svlan_p,
 623                           pkt_dev->svlan_cfi);
 624
 625        if (pkt_dev->tos)
 626                seq_printf(seq, "     tos: 0x%02x\n", pkt_dev->tos);
 627
 628        if (pkt_dev->traffic_class)
 629                seq_printf(seq, "     traffic_class: 0x%02x\n", pkt_dev->traffic_class);
 630
 631        if (pkt_dev->burst > 1)
 632                seq_printf(seq, "     burst: %d\n", pkt_dev->burst);
 633
 634        if (pkt_dev->node >= 0)
 635                seq_printf(seq, "     node: %d\n", pkt_dev->node);
 636
 637        if (pkt_dev->xmit_mode == M_NETIF_RECEIVE)
 638                seq_puts(seq, "     xmit_mode: netif_receive\n");
 639        else if (pkt_dev->xmit_mode == M_QUEUE_XMIT)
 640                seq_puts(seq, "     xmit_mode: xmit_queue\n");
 641
 642        seq_puts(seq, "     Flags: ");
 643
 644        for (i = 0; i < NR_PKT_FLAGS; i++) {
 645                if (i == F_FLOW_SEQ)
 646                        if (!pkt_dev->cflows)
 647                                continue;
 648
 649                if (pkt_dev->flags & (1 << i))
 650                        seq_printf(seq, "%s  ", pkt_flag_names[i]);
 651                else if (i == F_FLOW_SEQ)
 652                        seq_puts(seq, "FLOW_RND  ");
 653
 654#ifdef CONFIG_XFRM
 655                if (i == F_IPSEC && pkt_dev->spi)
 656                        seq_printf(seq, "spi:%u", pkt_dev->spi);
 657#endif
 658        }
 659
 660        seq_puts(seq, "\n");
 661
 662        /* not really stopped, more like last-running-at */
 663        stopped = pkt_dev->running ? ktime_get() : pkt_dev->stopped_at;
 664        idle = pkt_dev->idle_acc;
 665        do_div(idle, NSEC_PER_USEC);
 666
 667        seq_printf(seq,
 668                   "Current:\n     pkts-sofar: %llu  errors: %llu\n",
 669                   (unsigned long long)pkt_dev->sofar,
 670                   (unsigned long long)pkt_dev->errors);
 671
 672        seq_printf(seq,
 673                   "     started: %lluus  stopped: %lluus idle: %lluus\n",
 674                   (unsigned long long) ktime_to_us(pkt_dev->started_at),
 675                   (unsigned long long) ktime_to_us(stopped),
 676                   (unsigned long long) idle);
 677
 678        seq_printf(seq,
 679                   "     seq_num: %d  cur_dst_mac_offset: %d  cur_src_mac_offset: %d\n",
 680                   pkt_dev->seq_num, pkt_dev->cur_dst_mac_offset,
 681                   pkt_dev->cur_src_mac_offset);
 682
 683        if (pkt_dev->flags & F_IPV6) {
 684                seq_printf(seq, "     cur_saddr: %pI6c  cur_daddr: %pI6c\n",
 685                                &pkt_dev->cur_in6_saddr,
 686                                &pkt_dev->cur_in6_daddr);
 687        } else
 688                seq_printf(seq, "     cur_saddr: %pI4  cur_daddr: %pI4\n",
 689                           &pkt_dev->cur_saddr, &pkt_dev->cur_daddr);
 690
 691        seq_printf(seq, "     cur_udp_dst: %d  cur_udp_src: %d\n",
 692                   pkt_dev->cur_udp_dst, pkt_dev->cur_udp_src);
 693
 694        seq_printf(seq, "     cur_queue_map: %u\n", pkt_dev->cur_queue_map);
 695
 696        seq_printf(seq, "     flows: %u\n", pkt_dev->nflows);
 697
 698        if (pkt_dev->result[0])
 699                seq_printf(seq, "Result: %s\n", pkt_dev->result);
 700        else
 701                seq_puts(seq, "Result: Idle\n");
 702
 703        return 0;
 704}
 705
 706
 707static int hex32_arg(const char __user *user_buffer, unsigned long maxlen,
 708                     __u32 *num)
 709{
 710        int i = 0;
 711        *num = 0;
 712
 713        for (; i < maxlen; i++) {
 714                int value;
 715                char c;
 716                *num <<= 4;
 717                if (get_user(c, &user_buffer[i]))
 718                        return -EFAULT;
 719                value = hex_to_bin(c);
 720                if (value >= 0)
 721                        *num |= value;
 722                else
 723                        break;
 724        }
 725        return i;
 726}
 727
 728static int count_trail_chars(const char __user * user_buffer,
 729                             unsigned int maxlen)
 730{
 731        int i;
 732
 733        for (i = 0; i < maxlen; i++) {
 734                char c;
 735                if (get_user(c, &user_buffer[i]))
 736                        return -EFAULT;
 737                switch (c) {
 738                case '\"':
 739                case '\n':
 740                case '\r':
 741                case '\t':
 742                case ' ':
 743                case '=':
 744                        break;
 745                default:
 746                        goto done;
 747                }
 748        }
 749done:
 750        return i;
 751}
 752
 753static long num_arg(const char __user *user_buffer, unsigned long maxlen,
 754                                unsigned long *num)
 755{
 756        int i;
 757        *num = 0;
 758
 759        for (i = 0; i < maxlen; i++) {
 760                char c;
 761                if (get_user(c, &user_buffer[i]))
 762                        return -EFAULT;
 763                if ((c >= '0') && (c <= '9')) {
 764                        *num *= 10;
 765                        *num += c - '0';
 766                } else
 767                        break;
 768        }
 769        return i;
 770}
 771
 772static int strn_len(const char __user * user_buffer, unsigned int maxlen)
 773{
 774        int i;
 775
 776        for (i = 0; i < maxlen; i++) {
 777                char c;
 778                if (get_user(c, &user_buffer[i]))
 779                        return -EFAULT;
 780                switch (c) {
 781                case '\"':
 782                case '\n':
 783                case '\r':
 784                case '\t':
 785                case ' ':
 786                        goto done_str;
 787                default:
 788                        break;
 789                }
 790        }
 791done_str:
 792        return i;
 793}
 794
 795static ssize_t get_labels(const char __user *buffer, struct pktgen_dev *pkt_dev)
 796{
 797        unsigned int n = 0;
 798        char c;
 799        ssize_t i = 0;
 800        int len;
 801
 802        pkt_dev->nr_labels = 0;
 803        do {
 804                __u32 tmp;
 805                len = hex32_arg(&buffer[i], 8, &tmp);
 806                if (len <= 0)
 807                        return len;
 808                pkt_dev->labels[n] = htonl(tmp);
 809                if (pkt_dev->labels[n] & MPLS_STACK_BOTTOM)
 810                        pkt_dev->flags |= F_MPLS_RND;
 811                i += len;
 812                if (get_user(c, &buffer[i]))
 813                        return -EFAULT;
 814                i++;
 815                n++;
 816                if (n >= MAX_MPLS_LABELS)
 817                        return -E2BIG;
 818        } while (c == ',');
 819
 820        pkt_dev->nr_labels = n;
 821        return i;
 822}
 823
 824static __u32 pktgen_read_flag(const char *f, bool *disable)
 825{
 826        __u32 i;
 827
 828        if (f[0] == '!') {
 829                *disable = true;
 830                f++;
 831        }
 832
 833        for (i = 0; i < NR_PKT_FLAGS; i++) {
 834                if (!IS_ENABLED(CONFIG_XFRM) && i == IPSEC_SHIFT)
 835                        continue;
 836
 837                /* allow only disabling ipv6 flag */
 838                if (!*disable && i == IPV6_SHIFT)
 839                        continue;
 840
 841                if (strcmp(f, pkt_flag_names[i]) == 0)
 842                        return 1 << i;
 843        }
 844
 845        if (strcmp(f, "FLOW_RND") == 0) {
 846                *disable = !*disable;
 847                return F_FLOW_SEQ;
 848        }
 849
 850        return 0;
 851}
 852
 853static ssize_t pktgen_if_write(struct file *file,
 854                               const char __user * user_buffer, size_t count,
 855                               loff_t * offset)
 856{
 857        struct seq_file *seq = file->private_data;
 858        struct pktgen_dev *pkt_dev = seq->private;
 859        int i, max, len;
 860        char name[16], valstr[32];
 861        unsigned long value = 0;
 862        char *pg_result = NULL;
 863        int tmp = 0;
 864        char buf[128];
 865
 866        pg_result = &(pkt_dev->result[0]);
 867
 868        if (count < 1) {
 869                pr_warn("wrong command format\n");
 870                return -EINVAL;
 871        }
 872
 873        max = count;
 874        tmp = count_trail_chars(user_buffer, max);
 875        if (tmp < 0) {
 876                pr_warn("illegal format\n");
 877                return tmp;
 878        }
 879        i = tmp;
 880
 881        /* Read variable name */
 882
 883        len = strn_len(&user_buffer[i], sizeof(name) - 1);
 884        if (len < 0)
 885                return len;
 886
 887        memset(name, 0, sizeof(name));
 888        if (copy_from_user(name, &user_buffer[i], len))
 889                return -EFAULT;
 890        i += len;
 891
 892        max = count - i;
 893        len = count_trail_chars(&user_buffer[i], max);
 894        if (len < 0)
 895                return len;
 896
 897        i += len;
 898
 899        if (debug) {
 900                size_t copy = min_t(size_t, count + 1, 1024);
 901                char *tp = strndup_user(user_buffer, copy);
 902
 903                if (IS_ERR(tp))
 904                        return PTR_ERR(tp);
 905
 906                pr_debug("%s,%zu  buffer -:%s:-\n", name, count, tp);
 907                kfree(tp);
 908        }
 909
 910        if (!strcmp(name, "min_pkt_size")) {
 911                len = num_arg(&user_buffer[i], 10, &value);
 912                if (len < 0)
 913                        return len;
 914
 915                i += len;
 916                if (value < 14 + 20 + 8)
 917                        value = 14 + 20 + 8;
 918                if (value != pkt_dev->min_pkt_size) {
 919                        pkt_dev->min_pkt_size = value;
 920                        pkt_dev->cur_pkt_size = value;
 921                }
 922                sprintf(pg_result, "OK: min_pkt_size=%d",
 923                        pkt_dev->min_pkt_size);
 924                return count;
 925        }
 926
 927        if (!strcmp(name, "max_pkt_size")) {
 928                len = num_arg(&user_buffer[i], 10, &value);
 929                if (len < 0)
 930                        return len;
 931
 932                i += len;
 933                if (value < 14 + 20 + 8)
 934                        value = 14 + 20 + 8;
 935                if (value != pkt_dev->max_pkt_size) {
 936                        pkt_dev->max_pkt_size = value;
 937                        pkt_dev->cur_pkt_size = value;
 938                }
 939                sprintf(pg_result, "OK: max_pkt_size=%d",
 940                        pkt_dev->max_pkt_size);
 941                return count;
 942        }
 943
 944        /* Shortcut for min = max */
 945
 946        if (!strcmp(name, "pkt_size")) {
 947                len = num_arg(&user_buffer[i], 10, &value);
 948                if (len < 0)
 949                        return len;
 950
 951                i += len;
 952                if (value < 14 + 20 + 8)
 953                        value = 14 + 20 + 8;
 954                if (value != pkt_dev->min_pkt_size) {
 955                        pkt_dev->min_pkt_size = value;
 956                        pkt_dev->max_pkt_size = value;
 957                        pkt_dev->cur_pkt_size = value;
 958                }
 959                sprintf(pg_result, "OK: pkt_size=%d", pkt_dev->min_pkt_size);
 960                return count;
 961        }
 962
 963        if (!strcmp(name, "debug")) {
 964                len = num_arg(&user_buffer[i], 10, &value);
 965                if (len < 0)
 966                        return len;
 967
 968                i += len;
 969                debug = value;
 970                sprintf(pg_result, "OK: debug=%u", debug);
 971                return count;
 972        }
 973
 974        if (!strcmp(name, "frags")) {
 975                len = num_arg(&user_buffer[i], 10, &value);
 976                if (len < 0)
 977                        return len;
 978
 979                i += len;
 980                pkt_dev->nfrags = value;
 981                sprintf(pg_result, "OK: frags=%d", pkt_dev->nfrags);
 982                return count;
 983        }
 984        if (!strcmp(name, "delay")) {
 985                len = num_arg(&user_buffer[i], 10, &value);
 986                if (len < 0)
 987                        return len;
 988
 989                i += len;
 990                if (value == 0x7FFFFFFF)
 991                        pkt_dev->delay = ULLONG_MAX;
 992                else
 993                        pkt_dev->delay = (u64)value;
 994
 995                sprintf(pg_result, "OK: delay=%llu",
 996                        (unsigned long long) pkt_dev->delay);
 997                return count;
 998        }
 999        if (!strcmp(name, "rate")) {
1000                len = num_arg(&user_buffer[i], 10, &value);
1001                if (len < 0)
1002                        return len;
1003
1004                i += len;
1005                if (!value)
1006                        return len;
1007                pkt_dev->delay = pkt_dev->min_pkt_size*8*NSEC_PER_USEC/value;
1008                if (debug)
1009                        pr_info("Delay set at: %llu ns\n", pkt_dev->delay);
1010
1011                sprintf(pg_result, "OK: rate=%lu", value);
1012                return count;
1013        }
1014        if (!strcmp(name, "ratep")) {
1015                len = num_arg(&user_buffer[i], 10, &value);
1016                if (len < 0)
1017                        return len;
1018
1019                i += len;
1020                if (!value)
1021                        return len;
1022                pkt_dev->delay = NSEC_PER_SEC/value;
1023                if (debug)
1024                        pr_info("Delay set at: %llu ns\n", pkt_dev->delay);
1025
1026                sprintf(pg_result, "OK: rate=%lu", value);
1027                return count;
1028        }
1029        if (!strcmp(name, "udp_src_min")) {
1030                len = num_arg(&user_buffer[i], 10, &value);
1031                if (len < 0)
1032                        return len;
1033
1034                i += len;
1035                if (value != pkt_dev->udp_src_min) {
1036                        pkt_dev->udp_src_min = value;
1037                        pkt_dev->cur_udp_src = value;
1038                }
1039                sprintf(pg_result, "OK: udp_src_min=%u", pkt_dev->udp_src_min);
1040                return count;
1041        }
1042        if (!strcmp(name, "udp_dst_min")) {
1043                len = num_arg(&user_buffer[i], 10, &value);
1044                if (len < 0)
1045                        return len;
1046
1047                i += len;
1048                if (value != pkt_dev->udp_dst_min) {
1049                        pkt_dev->udp_dst_min = value;
1050                        pkt_dev->cur_udp_dst = value;
1051                }
1052                sprintf(pg_result, "OK: udp_dst_min=%u", pkt_dev->udp_dst_min);
1053                return count;
1054        }
1055        if (!strcmp(name, "udp_src_max")) {
1056                len = num_arg(&user_buffer[i], 10, &value);
1057                if (len < 0)
1058                        return len;
1059
1060                i += len;
1061                if (value != pkt_dev->udp_src_max) {
1062                        pkt_dev->udp_src_max = value;
1063                        pkt_dev->cur_udp_src = value;
1064                }
1065                sprintf(pg_result, "OK: udp_src_max=%u", pkt_dev->udp_src_max);
1066                return count;
1067        }
1068        if (!strcmp(name, "udp_dst_max")) {
1069                len = num_arg(&user_buffer[i], 10, &value);
1070                if (len < 0)
1071                        return len;
1072
1073                i += len;
1074                if (value != pkt_dev->udp_dst_max) {
1075                        pkt_dev->udp_dst_max = value;
1076                        pkt_dev->cur_udp_dst = value;
1077                }
1078                sprintf(pg_result, "OK: udp_dst_max=%u", pkt_dev->udp_dst_max);
1079                return count;
1080        }
1081        if (!strcmp(name, "clone_skb")) {
1082                len = num_arg(&user_buffer[i], 10, &value);
1083                if (len < 0)
1084                        return len;
1085                if ((value > 0) &&
1086                    ((pkt_dev->xmit_mode == M_NETIF_RECEIVE) ||
1087                     !(pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING)))
1088                        return -ENOTSUPP;
1089                i += len;
1090                pkt_dev->clone_skb = value;
1091
1092                sprintf(pg_result, "OK: clone_skb=%d", pkt_dev->clone_skb);
1093                return count;
1094        }
1095        if (!strcmp(name, "count")) {
1096                len = num_arg(&user_buffer[i], 10, &value);
1097                if (len < 0)
1098                        return len;
1099
1100                i += len;
1101                pkt_dev->count = value;
1102                sprintf(pg_result, "OK: count=%llu",
1103                        (unsigned long long)pkt_dev->count);
1104                return count;
1105        }
1106        if (!strcmp(name, "src_mac_count")) {
1107                len = num_arg(&user_buffer[i], 10, &value);
1108                if (len < 0)
1109                        return len;
1110
1111                i += len;
1112                if (pkt_dev->src_mac_count != value) {
1113                        pkt_dev->src_mac_count = value;
1114                        pkt_dev->cur_src_mac_offset = 0;
1115                }
1116                sprintf(pg_result, "OK: src_mac_count=%d",
1117                        pkt_dev->src_mac_count);
1118                return count;
1119        }
1120        if (!strcmp(name, "dst_mac_count")) {
1121                len = num_arg(&user_buffer[i], 10, &value);
1122                if (len < 0)
1123                        return len;
1124
1125                i += len;
1126                if (pkt_dev->dst_mac_count != value) {
1127                        pkt_dev->dst_mac_count = value;
1128                        pkt_dev->cur_dst_mac_offset = 0;
1129                }
1130                sprintf(pg_result, "OK: dst_mac_count=%d",
1131                        pkt_dev->dst_mac_count);
1132                return count;
1133        }
1134        if (!strcmp(name, "burst")) {
1135                len = num_arg(&user_buffer[i], 10, &value);
1136                if (len < 0)
1137                        return len;
1138
1139                i += len;
1140                if ((value > 1) &&
1141                    ((pkt_dev->xmit_mode == M_QUEUE_XMIT) ||
1142                     ((pkt_dev->xmit_mode == M_START_XMIT) &&
1143                     (!(pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING)))))
1144                        return -ENOTSUPP;
1145                pkt_dev->burst = value < 1 ? 1 : value;
1146                sprintf(pg_result, "OK: burst=%u", pkt_dev->burst);
1147                return count;
1148        }
1149        if (!strcmp(name, "node")) {
1150                len = num_arg(&user_buffer[i], 10, &value);
1151                if (len < 0)
1152                        return len;
1153
1154                i += len;
1155
1156                if (node_possible(value)) {
1157                        pkt_dev->node = value;
1158                        sprintf(pg_result, "OK: node=%d", pkt_dev->node);
1159                        if (pkt_dev->page) {
1160                                put_page(pkt_dev->page);
1161                                pkt_dev->page = NULL;
1162                        }
1163                }
1164                else
1165                        sprintf(pg_result, "ERROR: node not possible");
1166                return count;
1167        }
1168        if (!strcmp(name, "xmit_mode")) {
1169                char f[32];
1170
1171                memset(f, 0, 32);
1172                len = strn_len(&user_buffer[i], sizeof(f) - 1);
1173                if (len < 0)
1174                        return len;
1175
1176                if (copy_from_user(f, &user_buffer[i], len))
1177                        return -EFAULT;
1178                i += len;
1179
1180                if (strcmp(f, "start_xmit") == 0) {
1181                        pkt_dev->xmit_mode = M_START_XMIT;
1182                } else if (strcmp(f, "netif_receive") == 0) {
1183                        /* clone_skb set earlier, not supported in this mode */
1184                        if (pkt_dev->clone_skb > 0)
1185                                return -ENOTSUPP;
1186
1187                        pkt_dev->xmit_mode = M_NETIF_RECEIVE;
1188
1189                        /* make sure new packet is allocated every time
1190                         * pktgen_xmit() is called
1191                         */
1192                        pkt_dev->last_ok = 1;
1193
1194                        /* override clone_skb if user passed default value
1195                         * at module loading time
1196                         */
1197                        pkt_dev->clone_skb = 0;
1198                } else if (strcmp(f, "queue_xmit") == 0) {
1199                        pkt_dev->xmit_mode = M_QUEUE_XMIT;
1200                        pkt_dev->last_ok = 1;
1201                } else {
1202                        sprintf(pg_result,
1203                                "xmit_mode -:%s:- unknown\nAvailable modes: %s",
1204                                f, "start_xmit, netif_receive\n");
1205                        return count;
1206                }
1207                sprintf(pg_result, "OK: xmit_mode=%s", f);
1208                return count;
1209        }
1210        if (!strcmp(name, "flag")) {
1211                __u32 flag;
1212                char f[32];
1213                bool disable = false;
1214
1215                memset(f, 0, 32);
1216                len = strn_len(&user_buffer[i], sizeof(f) - 1);
1217                if (len < 0)
1218                        return len;
1219
1220                if (copy_from_user(f, &user_buffer[i], len))
1221                        return -EFAULT;
1222                i += len;
1223
1224                flag = pktgen_read_flag(f, &disable);
1225
1226                if (flag) {
1227                        if (disable)
1228                                pkt_dev->flags &= ~flag;
1229                        else
1230                                pkt_dev->flags |= flag;
1231                } else {
1232                        sprintf(pg_result,
1233                                "Flag -:%s:- unknown\nAvailable flags, (prepend ! to un-set flag):\n%s",
1234                                f,
1235                                "IPSRC_RND, IPDST_RND, UDPSRC_RND, UDPDST_RND, "
1236                                "MACSRC_RND, MACDST_RND, TXSIZE_RND, IPV6, "
1237                                "MPLS_RND, VID_RND, SVID_RND, FLOW_SEQ, "
1238                                "QUEUE_MAP_RND, QUEUE_MAP_CPU, UDPCSUM, "
1239                                "NO_TIMESTAMP, "
1240#ifdef CONFIG_XFRM
1241                                "IPSEC, "
1242#endif
1243                                "NODE_ALLOC\n");
1244                        return count;
1245                }
1246                sprintf(pg_result, "OK: flags=0x%x", pkt_dev->flags);
1247                return count;
1248        }
1249        if (!strcmp(name, "dst_min") || !strcmp(name, "dst")) {
1250                len = strn_len(&user_buffer[i], sizeof(pkt_dev->dst_min) - 1);
1251                if (len < 0)
1252                        return len;
1253
1254                if (copy_from_user(buf, &user_buffer[i], len))
1255                        return -EFAULT;
1256                buf[len] = 0;
1257                if (strcmp(buf, pkt_dev->dst_min) != 0) {
1258                        memset(pkt_dev->dst_min, 0, sizeof(pkt_dev->dst_min));
1259                        strcpy(pkt_dev->dst_min, buf);
1260                        pkt_dev->daddr_min = in_aton(pkt_dev->dst_min);
1261                        pkt_dev->cur_daddr = pkt_dev->daddr_min;
1262                }
1263                if (debug)
1264                        pr_debug("dst_min set to: %s\n", pkt_dev->dst_min);
1265                i += len;
1266                sprintf(pg_result, "OK: dst_min=%s", pkt_dev->dst_min);
1267                return count;
1268        }
1269        if (!strcmp(name, "dst_max")) {
1270                len = strn_len(&user_buffer[i], sizeof(pkt_dev->dst_max) - 1);
1271                if (len < 0)
1272                        return len;
1273
1274                if (copy_from_user(buf, &user_buffer[i], len))
1275                        return -EFAULT;
1276                buf[len] = 0;
1277                if (strcmp(buf, pkt_dev->dst_max) != 0) {
1278                        memset(pkt_dev->dst_max, 0, sizeof(pkt_dev->dst_max));
1279                        strcpy(pkt_dev->dst_max, buf);
1280                        pkt_dev->daddr_max = in_aton(pkt_dev->dst_max);
1281                        pkt_dev->cur_daddr = pkt_dev->daddr_max;
1282                }
1283                if (debug)
1284                        pr_debug("dst_max set to: %s\n", pkt_dev->dst_max);
1285                i += len;
1286                sprintf(pg_result, "OK: dst_max=%s", pkt_dev->dst_max);
1287                return count;
1288        }
1289        if (!strcmp(name, "dst6")) {
1290                len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1291                if (len < 0)
1292                        return len;
1293
1294                pkt_dev->flags |= F_IPV6;
1295
1296                if (copy_from_user(buf, &user_buffer[i], len))
1297                        return -EFAULT;
1298                buf[len] = 0;
1299
1300                in6_pton(buf, -1, pkt_dev->in6_daddr.s6_addr, -1, NULL);
1301                snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->in6_daddr);
1302
1303                pkt_dev->cur_in6_daddr = pkt_dev->in6_daddr;
1304
1305                if (debug)
1306                        pr_debug("dst6 set to: %s\n", buf);
1307
1308                i += len;
1309                sprintf(pg_result, "OK: dst6=%s", buf);
1310                return count;
1311        }
1312        if (!strcmp(name, "dst6_min")) {
1313                len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1314                if (len < 0)
1315                        return len;
1316
1317                pkt_dev->flags |= F_IPV6;
1318
1319                if (copy_from_user(buf, &user_buffer[i], len))
1320                        return -EFAULT;
1321                buf[len] = 0;
1322
1323                in6_pton(buf, -1, pkt_dev->min_in6_daddr.s6_addr, -1, NULL);
1324                snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->min_in6_daddr);
1325
1326                pkt_dev->cur_in6_daddr = pkt_dev->min_in6_daddr;
1327                if (debug)
1328                        pr_debug("dst6_min set to: %s\n", buf);
1329
1330                i += len;
1331                sprintf(pg_result, "OK: dst6_min=%s", buf);
1332                return count;
1333        }
1334        if (!strcmp(name, "dst6_max")) {
1335                len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1336                if (len < 0)
1337                        return len;
1338
1339                pkt_dev->flags |= F_IPV6;
1340
1341                if (copy_from_user(buf, &user_buffer[i], len))
1342                        return -EFAULT;
1343                buf[len] = 0;
1344
1345                in6_pton(buf, -1, pkt_dev->max_in6_daddr.s6_addr, -1, NULL);
1346                snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->max_in6_daddr);
1347
1348                if (debug)
1349                        pr_debug("dst6_max set to: %s\n", buf);
1350
1351                i += len;
1352                sprintf(pg_result, "OK: dst6_max=%s", buf);
1353                return count;
1354        }
1355        if (!strcmp(name, "src6")) {
1356                len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1357                if (len < 0)
1358                        return len;
1359
1360                pkt_dev->flags |= F_IPV6;
1361
1362                if (copy_from_user(buf, &user_buffer[i], len))
1363                        return -EFAULT;
1364                buf[len] = 0;
1365
1366                in6_pton(buf, -1, pkt_dev->in6_saddr.s6_addr, -1, NULL);
1367                snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->in6_saddr);
1368
1369                pkt_dev->cur_in6_saddr = pkt_dev->in6_saddr;
1370
1371                if (debug)
1372                        pr_debug("src6 set to: %s\n", buf);
1373
1374                i += len;
1375                sprintf(pg_result, "OK: src6=%s", buf);
1376                return count;
1377        }
1378        if (!strcmp(name, "src_min")) {
1379                len = strn_len(&user_buffer[i], sizeof(pkt_dev->src_min) - 1);
1380                if (len < 0)
1381                        return len;
1382
1383                if (copy_from_user(buf, &user_buffer[i], len))
1384                        return -EFAULT;
1385                buf[len] = 0;
1386                if (strcmp(buf, pkt_dev->src_min) != 0) {
1387                        memset(pkt_dev->src_min, 0, sizeof(pkt_dev->src_min));
1388                        strcpy(pkt_dev->src_min, buf);
1389                        pkt_dev->saddr_min = in_aton(pkt_dev->src_min);
1390                        pkt_dev->cur_saddr = pkt_dev->saddr_min;
1391                }
1392                if (debug)
1393                        pr_debug("src_min set to: %s\n", pkt_dev->src_min);
1394                i += len;
1395                sprintf(pg_result, "OK: src_min=%s", pkt_dev->src_min);
1396                return count;
1397        }
1398        if (!strcmp(name, "src_max")) {
1399                len = strn_len(&user_buffer[i], sizeof(pkt_dev->src_max) - 1);
1400                if (len < 0)
1401                        return len;
1402
1403                if (copy_from_user(buf, &user_buffer[i], len))
1404                        return -EFAULT;
1405                buf[len] = 0;
1406                if (strcmp(buf, pkt_dev->src_max) != 0) {
1407                        memset(pkt_dev->src_max, 0, sizeof(pkt_dev->src_max));
1408                        strcpy(pkt_dev->src_max, buf);
1409                        pkt_dev->saddr_max = in_aton(pkt_dev->src_max);
1410                        pkt_dev->cur_saddr = pkt_dev->saddr_max;
1411                }
1412                if (debug)
1413                        pr_debug("src_max set to: %s\n", pkt_dev->src_max);
1414                i += len;
1415                sprintf(pg_result, "OK: src_max=%s", pkt_dev->src_max);
1416                return count;
1417        }
1418        if (!strcmp(name, "dst_mac")) {
1419                len = strn_len(&user_buffer[i], sizeof(valstr) - 1);
1420                if (len < 0)
1421                        return len;
1422
1423                memset(valstr, 0, sizeof(valstr));
1424                if (copy_from_user(valstr, &user_buffer[i], len))
1425                        return -EFAULT;
1426
1427                if (!mac_pton(valstr, pkt_dev->dst_mac))
1428                        return -EINVAL;
1429                /* Set up Dest MAC */
1430                ether_addr_copy(&pkt_dev->hh[0], pkt_dev->dst_mac);
1431
1432                sprintf(pg_result, "OK: dstmac %pM", pkt_dev->dst_mac);
1433                return count;
1434        }
1435        if (!strcmp(name, "src_mac")) {
1436                len = strn_len(&user_buffer[i], sizeof(valstr) - 1);
1437                if (len < 0)
1438                        return len;
1439
1440                memset(valstr, 0, sizeof(valstr));
1441                if (copy_from_user(valstr, &user_buffer[i], len))
1442                        return -EFAULT;
1443
1444                if (!mac_pton(valstr, pkt_dev->src_mac))
1445                        return -EINVAL;
1446                /* Set up Src MAC */
1447                ether_addr_copy(&pkt_dev->hh[6], pkt_dev->src_mac);
1448
1449                sprintf(pg_result, "OK: srcmac %pM", pkt_dev->src_mac);
1450                return count;
1451        }
1452
1453        if (!strcmp(name, "clear_counters")) {
1454                pktgen_clear_counters(pkt_dev);
1455                sprintf(pg_result, "OK: Clearing counters.\n");
1456                return count;
1457        }
1458
1459        if (!strcmp(name, "flows")) {
1460                len = num_arg(&user_buffer[i], 10, &value);
1461                if (len < 0)
1462                        return len;
1463
1464                i += len;
1465                if (value > MAX_CFLOWS)
1466                        value = MAX_CFLOWS;
1467
1468                pkt_dev->cflows = value;
1469                sprintf(pg_result, "OK: flows=%u", pkt_dev->cflows);
1470                return count;
1471        }
1472#ifdef CONFIG_XFRM
1473        if (!strcmp(name, "spi")) {
1474                len = num_arg(&user_buffer[i], 10, &value);
1475                if (len < 0)
1476                        return len;
1477
1478                i += len;
1479                pkt_dev->spi = value;
1480                sprintf(pg_result, "OK: spi=%u", pkt_dev->spi);
1481                return count;
1482        }
1483#endif
1484        if (!strcmp(name, "flowlen")) {
1485                len = num_arg(&user_buffer[i], 10, &value);
1486                if (len < 0)
1487                        return len;
1488
1489                i += len;
1490                pkt_dev->lflow = value;
1491                sprintf(pg_result, "OK: flowlen=%u", pkt_dev->lflow);
1492                return count;
1493        }
1494
1495        if (!strcmp(name, "queue_map_min")) {
1496                len = num_arg(&user_buffer[i], 5, &value);
1497                if (len < 0)
1498                        return len;
1499
1500                i += len;
1501                pkt_dev->queue_map_min = value;
1502                sprintf(pg_result, "OK: queue_map_min=%u", pkt_dev->queue_map_min);
1503                return count;
1504        }
1505
1506        if (!strcmp(name, "queue_map_max")) {
1507                len = num_arg(&user_buffer[i], 5, &value);
1508                if (len < 0)
1509                        return len;
1510
1511                i += len;
1512                pkt_dev->queue_map_max = value;
1513                sprintf(pg_result, "OK: queue_map_max=%u", pkt_dev->queue_map_max);
1514                return count;
1515        }
1516
1517        if (!strcmp(name, "mpls")) {
1518                unsigned int n, cnt;
1519
1520                len = get_labels(&user_buffer[i], pkt_dev);
1521                if (len < 0)
1522                        return len;
1523                i += len;
1524                cnt = sprintf(pg_result, "OK: mpls=");
1525                for (n = 0; n < pkt_dev->nr_labels; n++)
1526                        cnt += sprintf(pg_result + cnt,
1527                                       "%08x%s", ntohl(pkt_dev->labels[n]),
1528                                       n == pkt_dev->nr_labels-1 ? "" : ",");
1529
1530                if (pkt_dev->nr_labels && pkt_dev->vlan_id != 0xffff) {
1531                        pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */
1532                        pkt_dev->svlan_id = 0xffff;
1533
1534                        if (debug)
1535                                pr_debug("VLAN/SVLAN auto turned off\n");
1536                }
1537                return count;
1538        }
1539
1540        if (!strcmp(name, "vlan_id")) {
1541                len = num_arg(&user_buffer[i], 4, &value);
1542                if (len < 0)
1543                        return len;
1544
1545                i += len;
1546                if (value <= 4095) {
1547                        pkt_dev->vlan_id = value;  /* turn on VLAN */
1548
1549                        if (debug)
1550                                pr_debug("VLAN turned on\n");
1551
1552                        if (debug && pkt_dev->nr_labels)
1553                                pr_debug("MPLS auto turned off\n");
1554
1555                        pkt_dev->nr_labels = 0;    /* turn off MPLS */
1556                        sprintf(pg_result, "OK: vlan_id=%u", pkt_dev->vlan_id);
1557                } else {
1558                        pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */
1559                        pkt_dev->svlan_id = 0xffff;
1560
1561                        if (debug)
1562                                pr_debug("VLAN/SVLAN turned off\n");
1563                }
1564                return count;
1565        }
1566
1567        if (!strcmp(name, "vlan_p")) {
1568                len = num_arg(&user_buffer[i], 1, &value);
1569                if (len < 0)
1570                        return len;
1571
1572                i += len;
1573                if ((value <= 7) && (pkt_dev->vlan_id != 0xffff)) {
1574                        pkt_dev->vlan_p = value;
1575                        sprintf(pg_result, "OK: vlan_p=%u", pkt_dev->vlan_p);
1576                } else {
1577                        sprintf(pg_result, "ERROR: vlan_p must be 0-7");
1578                }
1579                return count;
1580        }
1581
1582        if (!strcmp(name, "vlan_cfi")) {
1583                len = num_arg(&user_buffer[i], 1, &value);
1584                if (len < 0)
1585                        return len;
1586
1587                i += len;
1588                if ((value <= 1) && (pkt_dev->vlan_id != 0xffff)) {
1589                        pkt_dev->vlan_cfi = value;
1590                        sprintf(pg_result, "OK: vlan_cfi=%u", pkt_dev->vlan_cfi);
1591                } else {
1592                        sprintf(pg_result, "ERROR: vlan_cfi must be 0-1");
1593                }
1594                return count;
1595        }
1596
1597        if (!strcmp(name, "svlan_id")) {
1598                len = num_arg(&user_buffer[i], 4, &value);
1599                if (len < 0)
1600                        return len;
1601
1602                i += len;
1603                if ((value <= 4095) && ((pkt_dev->vlan_id != 0xffff))) {
1604                        pkt_dev->svlan_id = value;  /* turn on SVLAN */
1605
1606                        if (debug)
1607                                pr_debug("SVLAN turned on\n");
1608
1609                        if (debug && pkt_dev->nr_labels)
1610                                pr_debug("MPLS auto turned off\n");
1611
1612                        pkt_dev->nr_labels = 0;    /* turn off MPLS */
1613                        sprintf(pg_result, "OK: svlan_id=%u", pkt_dev->svlan_id);
1614                } else {
1615                        pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */
1616                        pkt_dev->svlan_id = 0xffff;
1617
1618                        if (debug)
1619                                pr_debug("VLAN/SVLAN turned off\n");
1620                }
1621                return count;
1622        }
1623
1624        if (!strcmp(name, "svlan_p")) {
1625                len = num_arg(&user_buffer[i], 1, &value);
1626                if (len < 0)
1627                        return len;
1628
1629                i += len;
1630                if ((value <= 7) && (pkt_dev->svlan_id != 0xffff)) {
1631                        pkt_dev->svlan_p = value;
1632                        sprintf(pg_result, "OK: svlan_p=%u", pkt_dev->svlan_p);
1633                } else {
1634                        sprintf(pg_result, "ERROR: svlan_p must be 0-7");
1635                }
1636                return count;
1637        }
1638
1639        if (!strcmp(name, "svlan_cfi")) {
1640                len = num_arg(&user_buffer[i], 1, &value);
1641                if (len < 0)
1642                        return len;
1643
1644                i += len;
1645                if ((value <= 1) && (pkt_dev->svlan_id != 0xffff)) {
1646                        pkt_dev->svlan_cfi = value;
1647                        sprintf(pg_result, "OK: svlan_cfi=%u", pkt_dev->svlan_cfi);
1648                } else {
1649                        sprintf(pg_result, "ERROR: svlan_cfi must be 0-1");
1650                }
1651                return count;
1652        }
1653
1654        if (!strcmp(name, "tos")) {
1655                __u32 tmp_value = 0;
1656                len = hex32_arg(&user_buffer[i], 2, &tmp_value);
1657                if (len < 0)
1658                        return len;
1659
1660                i += len;
1661                if (len == 2) {
1662                        pkt_dev->tos = tmp_value;
1663                        sprintf(pg_result, "OK: tos=0x%02x", pkt_dev->tos);
1664                } else {
1665                        sprintf(pg_result, "ERROR: tos must be 00-ff");
1666                }
1667                return count;
1668        }
1669
1670        if (!strcmp(name, "traffic_class")) {
1671                __u32 tmp_value = 0;
1672                len = hex32_arg(&user_buffer[i], 2, &tmp_value);
1673                if (len < 0)
1674                        return len;
1675
1676                i += len;
1677                if (len == 2) {
1678                        pkt_dev->traffic_class = tmp_value;
1679                        sprintf(pg_result, "OK: traffic_class=0x%02x", pkt_dev->traffic_class);
1680                } else {
1681                        sprintf(pg_result, "ERROR: traffic_class must be 00-ff");
1682                }
1683                return count;
1684        }
1685
1686        if (!strcmp(name, "skb_priority")) {
1687                len = num_arg(&user_buffer[i], 9, &value);
1688                if (len < 0)
1689                        return len;
1690
1691                i += len;
1692                pkt_dev->skb_priority = value;
1693                sprintf(pg_result, "OK: skb_priority=%i",
1694                        pkt_dev->skb_priority);
1695                return count;
1696        }
1697
1698        sprintf(pkt_dev->result, "No such parameter \"%s\"", name);
1699        return -EINVAL;
1700}
1701
1702static int pktgen_if_open(struct inode *inode, struct file *file)
1703{
1704        return single_open(file, pktgen_if_show, PDE_DATA(inode));
1705}
1706
1707static const struct proc_ops pktgen_if_proc_ops = {
1708        .proc_open      = pktgen_if_open,
1709        .proc_read      = seq_read,
1710        .proc_lseek     = seq_lseek,
1711        .proc_write     = pktgen_if_write,
1712        .proc_release   = single_release,
1713};
1714
1715static int pktgen_thread_show(struct seq_file *seq, void *v)
1716{
1717        struct pktgen_thread *t = seq->private;
1718        const struct pktgen_dev *pkt_dev;
1719
1720        BUG_ON(!t);
1721
1722        seq_puts(seq, "Running: ");
1723
1724        rcu_read_lock();
1725        list_for_each_entry_rcu(pkt_dev, &t->if_list, list)
1726                if (pkt_dev->running)
1727                        seq_printf(seq, "%s ", pkt_dev->odevname);
1728
1729        seq_puts(seq, "\nStopped: ");
1730
1731        list_for_each_entry_rcu(pkt_dev, &t->if_list, list)
1732                if (!pkt_dev->running)
1733                        seq_printf(seq, "%s ", pkt_dev->odevname);
1734
1735        if (t->result[0])
1736                seq_printf(seq, "\nResult: %s\n", t->result);
1737        else
1738                seq_puts(seq, "\nResult: NA\n");
1739
1740        rcu_read_unlock();
1741
1742        return 0;
1743}
1744
1745static ssize_t pktgen_thread_write(struct file *file,
1746                                   const char __user * user_buffer,
1747                                   size_t count, loff_t * offset)
1748{
1749        struct seq_file *seq = file->private_data;
1750        struct pktgen_thread *t = seq->private;
1751        int i, max, len, ret;
1752        char name[40];
1753        char *pg_result;
1754
1755        if (count < 1) {
1756                //      sprintf(pg_result, "Wrong command format");
1757                return -EINVAL;
1758        }
1759
1760        max = count;
1761        len = count_trail_chars(user_buffer, max);
1762        if (len < 0)
1763                return len;
1764
1765        i = len;
1766
1767        /* Read variable name */
1768
1769        len = strn_len(&user_buffer[i], sizeof(name) - 1);
1770        if (len < 0)
1771                return len;
1772
1773        memset(name, 0, sizeof(name));
1774        if (copy_from_user(name, &user_buffer[i], len))
1775                return -EFAULT;
1776        i += len;
1777
1778        max = count - i;
1779        len = count_trail_chars(&user_buffer[i], max);
1780        if (len < 0)
1781                return len;
1782
1783        i += len;
1784
1785        if (debug)
1786                pr_debug("t=%s, count=%lu\n", name, (unsigned long)count);
1787
1788        if (!t) {
1789                pr_err("ERROR: No thread\n");
1790                ret = -EINVAL;
1791                goto out;
1792        }
1793
1794        pg_result = &(t->result[0]);
1795
1796        if (!strcmp(name, "add_device")) {
1797                char f[32];
1798                memset(f, 0, 32);
1799                len = strn_len(&user_buffer[i], sizeof(f) - 1);
1800                if (len < 0) {
1801                        ret = len;
1802                        goto out;
1803                }
1804                if (copy_from_user(f, &user_buffer[i], len))
1805                        return -EFAULT;
1806                i += len;
1807                mutex_lock(&pktgen_thread_lock);
1808                ret = pktgen_add_device(t, f);
1809                mutex_unlock(&pktgen_thread_lock);
1810                if (!ret) {
1811                        ret = count;
1812                        sprintf(pg_result, "OK: add_device=%s", f);
1813                } else
1814                        sprintf(pg_result, "ERROR: can not add device %s", f);
1815                goto out;
1816        }
1817
1818        if (!strcmp(name, "rem_device_all")) {
1819                mutex_lock(&pktgen_thread_lock);
1820                t->control |= T_REMDEVALL;
1821                mutex_unlock(&pktgen_thread_lock);
1822                schedule_timeout_interruptible(msecs_to_jiffies(125));  /* Propagate thread->control  */
1823                ret = count;
1824                sprintf(pg_result, "OK: rem_device_all");
1825                goto out;
1826        }
1827
1828        if (!strcmp(name, "max_before_softirq")) {
1829                sprintf(pg_result, "OK: Note! max_before_softirq is obsoleted -- Do not use");
1830                ret = count;
1831                goto out;
1832        }
1833
1834        ret = -EINVAL;
1835out:
1836        return ret;
1837}
1838
1839static int pktgen_thread_open(struct inode *inode, struct file *file)
1840{
1841        return single_open(file, pktgen_thread_show, PDE_DATA(inode));
1842}
1843
1844static const struct proc_ops pktgen_thread_proc_ops = {
1845        .proc_open      = pktgen_thread_open,
1846        .proc_read      = seq_read,
1847        .proc_lseek     = seq_lseek,
1848        .proc_write     = pktgen_thread_write,
1849        .proc_release   = single_release,
1850};
1851
1852/* Think find or remove for NN */
1853static struct pktgen_dev *__pktgen_NN_threads(const struct pktgen_net *pn,
1854                                              const char *ifname, int remove)
1855{
1856        struct pktgen_thread *t;
1857        struct pktgen_dev *pkt_dev = NULL;
1858        bool exact = (remove == FIND);
1859
1860        list_for_each_entry(t, &pn->pktgen_threads, th_list) {
1861                pkt_dev = pktgen_find_dev(t, ifname, exact);
1862                if (pkt_dev) {
1863                        if (remove) {
1864                                pkt_dev->removal_mark = 1;
1865                                t->control |= T_REMDEV;
1866                        }
1867                        break;
1868                }
1869        }
1870        return pkt_dev;
1871}
1872
1873/*
1874 * mark a device for removal
1875 */
1876static void pktgen_mark_device(const struct pktgen_net *pn, const char *ifname)
1877{
1878        struct pktgen_dev *pkt_dev = NULL;
1879        const int max_tries = 10, msec_per_try = 125;
1880        int i = 0;
1881
1882        mutex_lock(&pktgen_thread_lock);
1883        pr_debug("%s: marking %s for removal\n", __func__, ifname);
1884
1885        while (1) {
1886
1887                pkt_dev = __pktgen_NN_threads(pn, ifname, REMOVE);
1888                if (pkt_dev == NULL)
1889                        break;  /* success */
1890
1891                mutex_unlock(&pktgen_thread_lock);
1892                pr_debug("%s: waiting for %s to disappear....\n",
1893                         __func__, ifname);
1894                schedule_timeout_interruptible(msecs_to_jiffies(msec_per_try));
1895                mutex_lock(&pktgen_thread_lock);
1896
1897                if (++i >= max_tries) {
1898                        pr_err("%s: timed out after waiting %d msec for device %s to be removed\n",
1899                               __func__, msec_per_try * i, ifname);
1900                        break;
1901                }
1902
1903        }
1904
1905        mutex_unlock(&pktgen_thread_lock);
1906}
1907
1908static void pktgen_change_name(const struct pktgen_net *pn, struct net_device *dev)
1909{
1910        struct pktgen_thread *t;
1911
1912        mutex_lock(&pktgen_thread_lock);
1913
1914        list_for_each_entry(t, &pn->pktgen_threads, th_list) {
1915                struct pktgen_dev *pkt_dev;
1916
1917                if_lock(t);
1918                list_for_each_entry(pkt_dev, &t->if_list, list) {
1919                        if (pkt_dev->odev != dev)
1920                                continue;
1921
1922                        proc_remove(pkt_dev->entry);
1923
1924                        pkt_dev->entry = proc_create_data(dev->name, 0600,
1925                                                          pn->proc_dir,
1926                                                          &pktgen_if_proc_ops,
1927                                                          pkt_dev);
1928                        if (!pkt_dev->entry)
1929                                pr_err("can't move proc entry for '%s'\n",
1930                                       dev->name);
1931                        break;
1932                }
1933                if_unlock(t);
1934        }
1935        mutex_unlock(&pktgen_thread_lock);
1936}
1937
1938static int pktgen_device_event(struct notifier_block *unused,
1939                               unsigned long event, void *ptr)
1940{
1941        struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1942        struct pktgen_net *pn = net_generic(dev_net(dev), pg_net_id);
1943
1944        if (pn->pktgen_exiting)
1945                return NOTIFY_DONE;
1946
1947        /* It is OK that we do not hold the group lock right now,
1948         * as we run under the RTNL lock.
1949         */
1950
1951        switch (event) {
1952        case NETDEV_CHANGENAME:
1953                pktgen_change_name(pn, dev);
1954                break;
1955
1956        case NETDEV_UNREGISTER:
1957                pktgen_mark_device(pn, dev->name);
1958                break;
1959        }
1960
1961        return NOTIFY_DONE;
1962}
1963
1964static struct net_device *pktgen_dev_get_by_name(const struct pktgen_net *pn,
1965                                                 struct pktgen_dev *pkt_dev,
1966                                                 const char *ifname)
1967{
1968        char b[IFNAMSIZ+5];
1969        int i;
1970
1971        for (i = 0; ifname[i] != '@'; i++) {
1972                if (i == IFNAMSIZ)
1973                        break;
1974
1975                b[i] = ifname[i];
1976        }
1977        b[i] = 0;
1978
1979        return dev_get_by_name(pn->net, b);
1980}
1981
1982
1983/* Associate pktgen_dev with a device. */
1984
1985static int pktgen_setup_dev(const struct pktgen_net *pn,
1986                            struct pktgen_dev *pkt_dev, const char *ifname)
1987{
1988        struct net_device *odev;
1989        int err;
1990
1991        /* Clean old setups */
1992        if (pkt_dev->odev) {
1993                dev_put(pkt_dev->odev);
1994                pkt_dev->odev = NULL;
1995        }
1996
1997        odev = pktgen_dev_get_by_name(pn, pkt_dev, ifname);
1998        if (!odev) {
1999                pr_err("no such netdevice: \"%s\"\n", ifname);
2000                return -ENODEV;
2001        }
2002
2003        if (odev->type != ARPHRD_ETHER && odev->type != ARPHRD_LOOPBACK) {
2004                pr_err("not an ethernet or loopback device: \"%s\"\n", ifname);
2005                err = -EINVAL;
2006        } else if (!netif_running(odev)) {
2007                pr_err("device is down: \"%s\"\n", ifname);
2008                err = -ENETDOWN;
2009        } else {
2010                pkt_dev->odev = odev;
2011                return 0;
2012        }
2013
2014        dev_put(odev);
2015        return err;
2016}
2017
2018/* Read pkt_dev from the interface and set up internal pktgen_dev
2019 * structure to have the right information to create/send packets
2020 */
2021static void pktgen_setup_inject(struct pktgen_dev *pkt_dev)
2022{
2023        int ntxq;
2024
2025        if (!pkt_dev->odev) {
2026                pr_err("ERROR: pkt_dev->odev == NULL in setup_inject\n");
2027                sprintf(pkt_dev->result,
2028                        "ERROR: pkt_dev->odev == NULL in setup_inject.\n");
2029                return;
2030        }
2031
2032        /* make sure that we don't pick a non-existing transmit queue */
2033        ntxq = pkt_dev->odev->real_num_tx_queues;
2034
2035        if (ntxq <= pkt_dev->queue_map_min) {
2036                pr_warn("WARNING: Requested queue_map_min (zero-based) (%d) exceeds valid range [0 - %d] for (%d) queues on %s, resetting\n",
2037                        pkt_dev->queue_map_min, (ntxq ?: 1) - 1, ntxq,
2038                        pkt_dev->odevname);
2039                pkt_dev->queue_map_min = (ntxq ?: 1) - 1;
2040        }
2041        if (pkt_dev->queue_map_max >= ntxq) {
2042                pr_warn("WARNING: Requested queue_map_max (zero-based) (%d) exceeds valid range [0 - %d] for (%d) queues on %s, resetting\n",
2043                        pkt_dev->queue_map_max, (ntxq ?: 1) - 1, ntxq,
2044                        pkt_dev->odevname);
2045                pkt_dev->queue_map_max = (ntxq ?: 1) - 1;
2046        }
2047
2048        /* Default to the interface's mac if not explicitly set. */
2049
2050        if (is_zero_ether_addr(pkt_dev->src_mac))
2051                ether_addr_copy(&(pkt_dev->hh[6]), pkt_dev->odev->dev_addr);
2052
2053        /* Set up Dest MAC */
2054        ether_addr_copy(&(pkt_dev->hh[0]), pkt_dev->dst_mac);
2055
2056        if (pkt_dev->flags & F_IPV6) {
2057                int i, set = 0, err = 1;
2058                struct inet6_dev *idev;
2059
2060                if (pkt_dev->min_pkt_size == 0) {
2061                        pkt_dev->min_pkt_size = 14 + sizeof(struct ipv6hdr)
2062                                                + sizeof(struct udphdr)
2063                                                + sizeof(struct pktgen_hdr)
2064                                                + pkt_dev->pkt_overhead;
2065                }
2066
2067                for (i = 0; i < sizeof(struct in6_addr); i++)
2068                        if (pkt_dev->cur_in6_saddr.s6_addr[i]) {
2069                                set = 1;
2070                                break;
2071                        }
2072
2073                if (!set) {
2074
2075                        /*
2076                         * Use linklevel address if unconfigured.
2077                         *
2078                         * use ipv6_get_lladdr if/when it's get exported
2079                         */
2080
2081                        rcu_read_lock();
2082                        idev = __in6_dev_get(pkt_dev->odev);
2083                        if (idev) {
2084                                struct inet6_ifaddr *ifp;
2085
2086                                read_lock_bh(&idev->lock);
2087                                list_for_each_entry(ifp, &idev->addr_list, if_list) {
2088                                        if ((ifp->scope & IFA_LINK) &&
2089                                            !(ifp->flags & IFA_F_TENTATIVE)) {
2090                                                pkt_dev->cur_in6_saddr = ifp->addr;
2091                                                err = 0;
2092                                                break;
2093                                        }
2094                                }
2095                                read_unlock_bh(&idev->lock);
2096                        }
2097                        rcu_read_unlock();
2098                        if (err)
2099                                pr_err("ERROR: IPv6 link address not available\n");
2100                }
2101        } else {
2102                if (pkt_dev->min_pkt_size == 0) {
2103                        pkt_dev->min_pkt_size = 14 + sizeof(struct iphdr)
2104                                                + sizeof(struct udphdr)
2105                                                + sizeof(struct pktgen_hdr)
2106                                                + pkt_dev->pkt_overhead;
2107                }
2108
2109                pkt_dev->saddr_min = 0;
2110                pkt_dev->saddr_max = 0;
2111                if (strlen(pkt_dev->src_min) == 0) {
2112
2113                        struct in_device *in_dev;
2114
2115                        rcu_read_lock();
2116                        in_dev = __in_dev_get_rcu(pkt_dev->odev);
2117                        if (in_dev) {
2118                                const struct in_ifaddr *ifa;
2119
2120                                ifa = rcu_dereference(in_dev->ifa_list);
2121                                if (ifa) {
2122                                        pkt_dev->saddr_min = ifa->ifa_address;
2123                                        pkt_dev->saddr_max = pkt_dev->saddr_min;
2124                                }
2125                        }
2126                        rcu_read_unlock();
2127                } else {
2128                        pkt_dev->saddr_min = in_aton(pkt_dev->src_min);
2129                        pkt_dev->saddr_max = in_aton(pkt_dev->src_max);
2130                }
2131
2132                pkt_dev->daddr_min = in_aton(pkt_dev->dst_min);
2133                pkt_dev->daddr_max = in_aton(pkt_dev->dst_max);
2134        }
2135        /* Initialize current values. */
2136        pkt_dev->cur_pkt_size = pkt_dev->min_pkt_size;
2137        if (pkt_dev->min_pkt_size > pkt_dev->max_pkt_size)
2138                pkt_dev->max_pkt_size = pkt_dev->min_pkt_size;
2139
2140        pkt_dev->cur_dst_mac_offset = 0;
2141        pkt_dev->cur_src_mac_offset = 0;
2142        pkt_dev->cur_saddr = pkt_dev->saddr_min;
2143        pkt_dev->cur_daddr = pkt_dev->daddr_min;
2144        pkt_dev->cur_udp_dst = pkt_dev->udp_dst_min;
2145        pkt_dev->cur_udp_src = pkt_dev->udp_src_min;
2146        pkt_dev->nflows = 0;
2147}
2148
2149
2150static void spin(struct pktgen_dev *pkt_dev, ktime_t spin_until)
2151{
2152        ktime_t start_time, end_time;
2153        s64 remaining;
2154        struct hrtimer_sleeper t;
2155
2156        hrtimer_init_sleeper_on_stack(&t, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
2157        hrtimer_set_expires(&t.timer, spin_until);
2158
2159        remaining = ktime_to_ns(hrtimer_expires_remaining(&t.timer));
2160        if (remaining <= 0)
2161                goto out;
2162
2163        start_time = ktime_get();
2164        if (remaining < 100000) {
2165                /* for small delays (<100us), just loop until limit is reached */
2166                do {
2167                        end_time = ktime_get();
2168                } while (ktime_compare(end_time, spin_until) < 0);
2169        } else {
2170                do {
2171                        set_current_state(TASK_INTERRUPTIBLE);
2172                        hrtimer_sleeper_start_expires(&t, HRTIMER_MODE_ABS);
2173
2174                        if (likely(t.task))
2175                                schedule();
2176
2177                        hrtimer_cancel(&t.timer);
2178                } while (t.task && pkt_dev->running && !signal_pending(current));
2179                __set_current_state(TASK_RUNNING);
2180                end_time = ktime_get();
2181        }
2182
2183        pkt_dev->idle_acc += ktime_to_ns(ktime_sub(end_time, start_time));
2184out:
2185        pkt_dev->next_tx = ktime_add_ns(spin_until, pkt_dev->delay);
2186        destroy_hrtimer_on_stack(&t.timer);
2187}
2188
2189static inline void set_pkt_overhead(struct pktgen_dev *pkt_dev)
2190{
2191        pkt_dev->pkt_overhead = 0;
2192        pkt_dev->pkt_overhead += pkt_dev->nr_labels*sizeof(u32);
2193        pkt_dev->pkt_overhead += VLAN_TAG_SIZE(pkt_dev);
2194        pkt_dev->pkt_overhead += SVLAN_TAG_SIZE(pkt_dev);
2195}
2196
2197static inline int f_seen(const struct pktgen_dev *pkt_dev, int flow)
2198{
2199        return !!(pkt_dev->flows[flow].flags & F_INIT);
2200}
2201
2202static inline int f_pick(struct pktgen_dev *pkt_dev)
2203{
2204        int flow = pkt_dev->curfl;
2205
2206        if (pkt_dev->flags & F_FLOW_SEQ) {
2207                if (pkt_dev->flows[flow].count >= pkt_dev->lflow) {
2208                        /* reset time */
2209                        pkt_dev->flows[flow].count = 0;
2210                        pkt_dev->flows[flow].flags = 0;
2211                        pkt_dev->curfl += 1;
2212                        if (pkt_dev->curfl >= pkt_dev->cflows)
2213                                pkt_dev->curfl = 0; /*reset */
2214                }
2215        } else {
2216                flow = prandom_u32() % pkt_dev->cflows;
2217                pkt_dev->curfl = flow;
2218
2219                if (pkt_dev->flows[flow].count > pkt_dev->lflow) {
2220                        pkt_dev->flows[flow].count = 0;
2221                        pkt_dev->flows[flow].flags = 0;
2222                }
2223        }
2224
2225        return pkt_dev->curfl;
2226}
2227
2228
2229#ifdef CONFIG_XFRM
2230/* If there was already an IPSEC SA, we keep it as is, else
2231 * we go look for it ...
2232*/
2233#define DUMMY_MARK 0
2234static void get_ipsec_sa(struct pktgen_dev *pkt_dev, int flow)
2235{
2236        struct xfrm_state *x = pkt_dev->flows[flow].x;
2237        struct pktgen_net *pn = net_generic(dev_net(pkt_dev->odev), pg_net_id);
2238        if (!x) {
2239
2240                if (pkt_dev->spi) {
2241                        /* We need as quick as possible to find the right SA
2242                         * Searching with minimum criteria to archieve this.
2243                         */
2244                        x = xfrm_state_lookup_byspi(pn->net, htonl(pkt_dev->spi), AF_INET);
2245                } else {
2246                        /* slow path: we dont already have xfrm_state */
2247                        x = xfrm_stateonly_find(pn->net, DUMMY_MARK, 0,
2248                                                (xfrm_address_t *)&pkt_dev->cur_daddr,
2249                                                (xfrm_address_t *)&pkt_dev->cur_saddr,
2250                                                AF_INET,
2251                                                pkt_dev->ipsmode,
2252                                                pkt_dev->ipsproto, 0);
2253                }
2254                if (x) {
2255                        pkt_dev->flows[flow].x = x;
2256                        set_pkt_overhead(pkt_dev);
2257                        pkt_dev->pkt_overhead += x->props.header_len;
2258                }
2259
2260        }
2261}
2262#endif
2263static void set_cur_queue_map(struct pktgen_dev *pkt_dev)
2264{
2265
2266        if (pkt_dev->flags & F_QUEUE_MAP_CPU)
2267                pkt_dev->cur_queue_map = smp_processor_id();
2268
2269        else if (pkt_dev->queue_map_min <= pkt_dev->queue_map_max) {
2270                __u16 t;
2271                if (pkt_dev->flags & F_QUEUE_MAP_RND) {
2272                        t = prandom_u32() %
2273                                (pkt_dev->queue_map_max -
2274                                 pkt_dev->queue_map_min + 1)
2275                                + pkt_dev->queue_map_min;
2276                } else {
2277                        t = pkt_dev->cur_queue_map + 1;
2278                        if (t > pkt_dev->queue_map_max)
2279                                t = pkt_dev->queue_map_min;
2280                }
2281                pkt_dev->cur_queue_map = t;
2282        }
2283        pkt_dev->cur_queue_map  = pkt_dev->cur_queue_map % pkt_dev->odev->real_num_tx_queues;
2284}
2285
2286/* Increment/randomize headers according to flags and current values
2287 * for IP src/dest, UDP src/dst port, MAC-Addr src/dst
2288 */
2289static void mod_cur_headers(struct pktgen_dev *pkt_dev)
2290{
2291        __u32 imn;
2292        __u32 imx;
2293        int flow = 0;
2294
2295        if (pkt_dev->cflows)
2296                flow = f_pick(pkt_dev);
2297
2298        /*  Deal with source MAC */
2299        if (pkt_dev->src_mac_count > 1) {
2300                __u32 mc;
2301                __u32 tmp;
2302
2303                if (pkt_dev->flags & F_MACSRC_RND)
2304                        mc = prandom_u32() % pkt_dev->src_mac_count;
2305                else {
2306                        mc = pkt_dev->cur_src_mac_offset++;
2307                        if (pkt_dev->cur_src_mac_offset >=
2308                            pkt_dev->src_mac_count)
2309                                pkt_dev->cur_src_mac_offset = 0;
2310                }
2311
2312                tmp = pkt_dev->src_mac[5] + (mc & 0xFF);
2313                pkt_dev->hh[11] = tmp;
2314                tmp = (pkt_dev->src_mac[4] + ((mc >> 8) & 0xFF) + (tmp >> 8));
2315                pkt_dev->hh[10] = tmp;
2316                tmp = (pkt_dev->src_mac[3] + ((mc >> 16) & 0xFF) + (tmp >> 8));
2317                pkt_dev->hh[9] = tmp;
2318                tmp = (pkt_dev->src_mac[2] + ((mc >> 24) & 0xFF) + (tmp >> 8));
2319                pkt_dev->hh[8] = tmp;
2320                tmp = (pkt_dev->src_mac[1] + (tmp >> 8));
2321                pkt_dev->hh[7] = tmp;
2322        }
2323
2324        /*  Deal with Destination MAC */
2325        if (pkt_dev->dst_mac_count > 1) {
2326                __u32 mc;
2327                __u32 tmp;
2328
2329                if (pkt_dev->flags & F_MACDST_RND)
2330                        mc = prandom_u32() % pkt_dev->dst_mac_count;
2331
2332                else {
2333                        mc = pkt_dev->cur_dst_mac_offset++;
2334                        if (pkt_dev->cur_dst_mac_offset >=
2335                            pkt_dev->dst_mac_count) {
2336                                pkt_dev->cur_dst_mac_offset = 0;
2337                        }
2338                }
2339
2340                tmp = pkt_dev->dst_mac[5] + (mc & 0xFF);
2341                pkt_dev->hh[5] = tmp;
2342                tmp = (pkt_dev->dst_mac[4] + ((mc >> 8) & 0xFF) + (tmp >> 8));
2343                pkt_dev->hh[4] = tmp;
2344                tmp = (pkt_dev->dst_mac[3] + ((mc >> 16) & 0xFF) + (tmp >> 8));
2345                pkt_dev->hh[3] = tmp;
2346                tmp = (pkt_dev->dst_mac[2] + ((mc >> 24) & 0xFF) + (tmp >> 8));
2347                pkt_dev->hh[2] = tmp;
2348                tmp = (pkt_dev->dst_mac[1] + (tmp >> 8));
2349                pkt_dev->hh[1] = tmp;
2350        }
2351
2352        if (pkt_dev->flags & F_MPLS_RND) {
2353                unsigned int i;
2354                for (i = 0; i < pkt_dev->nr_labels; i++)
2355                        if (pkt_dev->labels[i] & MPLS_STACK_BOTTOM)
2356                                pkt_dev->labels[i] = MPLS_STACK_BOTTOM |
2357                                             ((__force __be32)prandom_u32() &
2358                                                      htonl(0x000fffff));
2359        }
2360
2361        if ((pkt_dev->flags & F_VID_RND) && (pkt_dev->vlan_id != 0xffff)) {
2362                pkt_dev->vlan_id = prandom_u32() & (4096 - 1);
2363        }
2364
2365        if ((pkt_dev->flags & F_SVID_RND) && (pkt_dev->svlan_id != 0xffff)) {
2366                pkt_dev->svlan_id = prandom_u32() & (4096 - 1);
2367        }
2368
2369        if (pkt_dev->udp_src_min < pkt_dev->udp_src_max) {
2370                if (pkt_dev->flags & F_UDPSRC_RND)
2371                        pkt_dev->cur_udp_src = prandom_u32() %
2372                                (pkt_dev->udp_src_max - pkt_dev->udp_src_min)
2373                                + pkt_dev->udp_src_min;
2374
2375                else {
2376                        pkt_dev->cur_udp_src++;
2377                        if (pkt_dev->cur_udp_src >= pkt_dev->udp_src_max)
2378                                pkt_dev->cur_udp_src = pkt_dev->udp_src_min;
2379                }
2380        }
2381
2382        if (pkt_dev->udp_dst_min < pkt_dev->udp_dst_max) {
2383                if (pkt_dev->flags & F_UDPDST_RND) {
2384                        pkt_dev->cur_udp_dst = prandom_u32() %
2385                                (pkt_dev->udp_dst_max - pkt_dev->udp_dst_min)
2386                                + pkt_dev->udp_dst_min;
2387                } else {
2388                        pkt_dev->cur_udp_dst++;
2389                        if (pkt_dev->cur_udp_dst >= pkt_dev->udp_dst_max)
2390                                pkt_dev->cur_udp_dst = pkt_dev->udp_dst_min;
2391                }
2392        }
2393
2394        if (!(pkt_dev->flags & F_IPV6)) {
2395
2396                imn = ntohl(pkt_dev->saddr_min);
2397                imx = ntohl(pkt_dev->saddr_max);
2398                if (imn < imx) {
2399                        __u32 t;
2400                        if (pkt_dev->flags & F_IPSRC_RND)
2401                                t = prandom_u32() % (imx - imn) + imn;
2402                        else {
2403                                t = ntohl(pkt_dev->cur_saddr);
2404                                t++;
2405                                if (t > imx)
2406                                        t = imn;
2407
2408                        }
2409                        pkt_dev->cur_saddr = htonl(t);
2410                }
2411
2412                if (pkt_dev->cflows && f_seen(pkt_dev, flow)) {
2413                        pkt_dev->cur_daddr = pkt_dev->flows[flow].cur_daddr;
2414                } else {
2415                        imn = ntohl(pkt_dev->daddr_min);
2416                        imx = ntohl(pkt_dev->daddr_max);
2417                        if (imn < imx) {
2418                                __u32 t;
2419                                __be32 s;
2420                                if (pkt_dev->flags & F_IPDST_RND) {
2421
2422                                        do {
2423                                                t = prandom_u32() %
2424                                                        (imx - imn) + imn;
2425                                                s = htonl(t);
2426                                        } while (ipv4_is_loopback(s) ||
2427                                                ipv4_is_multicast(s) ||
2428                                                ipv4_is_lbcast(s) ||
2429                                                ipv4_is_zeronet(s) ||
2430                                                ipv4_is_local_multicast(s));
2431                                        pkt_dev->cur_daddr = s;
2432                                } else {
2433                                        t = ntohl(pkt_dev->cur_daddr);
2434                                        t++;
2435                                        if (t > imx) {
2436                                                t = imn;
2437                                        }
2438                                        pkt_dev->cur_daddr = htonl(t);
2439                                }
2440                        }
2441                        if (pkt_dev->cflows) {
2442                                pkt_dev->flows[flow].flags |= F_INIT;
2443                                pkt_dev->flows[flow].cur_daddr =
2444                                    pkt_dev->cur_daddr;
2445#ifdef CONFIG_XFRM
2446                                if (pkt_dev->flags & F_IPSEC)
2447                                        get_ipsec_sa(pkt_dev, flow);
2448#endif
2449                                pkt_dev->nflows++;
2450                        }
2451                }
2452        } else {                /* IPV6 * */
2453
2454                if (!ipv6_addr_any(&pkt_dev->min_in6_daddr)) {
2455                        int i;
2456
2457                        /* Only random destinations yet */
2458
2459                        for (i = 0; i < 4; i++) {
2460                                pkt_dev->cur_in6_daddr.s6_addr32[i] =
2461                                    (((__force __be32)prandom_u32() |
2462                                      pkt_dev->min_in6_daddr.s6_addr32[i]) &
2463                                     pkt_dev->max_in6_daddr.s6_addr32[i]);
2464                        }
2465                }
2466        }
2467
2468        if (pkt_dev->min_pkt_size < pkt_dev->max_pkt_size) {
2469                __u32 t;
2470                if (pkt_dev->flags & F_TXSIZE_RND) {
2471                        t = prandom_u32() %
2472                                (pkt_dev->max_pkt_size - pkt_dev->min_pkt_size)
2473                                + pkt_dev->min_pkt_size;
2474                } else {
2475                        t = pkt_dev->cur_pkt_size + 1;
2476                        if (t > pkt_dev->max_pkt_size)
2477                                t = pkt_dev->min_pkt_size;
2478                }
2479                pkt_dev->cur_pkt_size = t;
2480        }
2481
2482        set_cur_queue_map(pkt_dev);
2483
2484        pkt_dev->flows[flow].count++;
2485}
2486
2487
2488#ifdef CONFIG_XFRM
2489static u32 pktgen_dst_metrics[RTAX_MAX + 1] = {
2490
2491        [RTAX_HOPLIMIT] = 0x5, /* Set a static hoplimit */
2492};
2493
2494static int pktgen_output_ipsec(struct sk_buff *skb, struct pktgen_dev *pkt_dev)
2495{
2496        struct xfrm_state *x = pkt_dev->flows[pkt_dev->curfl].x;
2497        int err = 0;
2498        struct net *net = dev_net(pkt_dev->odev);
2499
2500        if (!x)
2501                return 0;
2502        /* XXX: we dont support tunnel mode for now until
2503         * we resolve the dst issue */
2504        if ((x->props.mode != XFRM_MODE_TRANSPORT) && (pkt_dev->spi == 0))
2505                return 0;
2506
2507        /* But when user specify an valid SPI, transformation
2508         * supports both transport/tunnel mode + ESP/AH type.
2509         */
2510        if ((x->props.mode == XFRM_MODE_TUNNEL) && (pkt_dev->spi != 0))
2511                skb->_skb_refdst = (unsigned long)&pkt_dev->xdst.u.dst | SKB_DST_NOREF;
2512
2513        rcu_read_lock_bh();
2514        err = pktgen_xfrm_outer_mode_output(x, skb);
2515        rcu_read_unlock_bh();
2516        if (err) {
2517                XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEMODEERROR);
2518                goto error;
2519        }
2520        err = x->type->output(x, skb);
2521        if (err) {
2522                XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEPROTOERROR);
2523                goto error;
2524        }
2525        spin_lock_bh(&x->lock);
2526        x->curlft.bytes += skb->len;
2527        x->curlft.packets++;
2528        spin_unlock_bh(&x->lock);
2529error:
2530        return err;
2531}
2532
2533static void free_SAs(struct pktgen_dev *pkt_dev)
2534{
2535        if (pkt_dev->cflows) {
2536                /* let go of the SAs if we have them */
2537                int i;
2538                for (i = 0; i < pkt_dev->cflows; i++) {
2539                        struct xfrm_state *x = pkt_dev->flows[i].x;
2540                        if (x) {
2541                                xfrm_state_put(x);
2542                                pkt_dev->flows[i].x = NULL;
2543                        }
2544                }
2545        }
2546}
2547
2548static int process_ipsec(struct pktgen_dev *pkt_dev,
2549                              struct sk_buff *skb, __be16 protocol)
2550{
2551        if (pkt_dev->flags & F_IPSEC) {
2552                struct xfrm_state *x = pkt_dev->flows[pkt_dev->curfl].x;
2553                int nhead = 0;
2554                if (x) {
2555                        struct ethhdr *eth;
2556                        struct iphdr *iph;
2557                        int ret;
2558
2559                        nhead = x->props.header_len - skb_headroom(skb);
2560                        if (nhead > 0) {
2561                                ret = pskb_expand_head(skb, nhead, 0, GFP_ATOMIC);
2562                                if (ret < 0) {
2563                                        pr_err("Error expanding ipsec packet %d\n",
2564                                               ret);
2565                                        goto err;
2566                                }
2567                        }
2568
2569                        /* ipsec is not expecting ll header */
2570                        skb_pull(skb, ETH_HLEN);
2571                        ret = pktgen_output_ipsec(skb, pkt_dev);
2572                        if (ret) {
2573                                pr_err("Error creating ipsec packet %d\n", ret);
2574                                goto err;
2575                        }
2576                        /* restore ll */
2577                        eth = skb_push(skb, ETH_HLEN);
2578                        memcpy(eth, pkt_dev->hh, 2 * ETH_ALEN);
2579                        eth->h_proto = protocol;
2580
2581                        /* Update IPv4 header len as well as checksum value */
2582                        iph = ip_hdr(skb);
2583                        iph->tot_len = htons(skb->len - ETH_HLEN);
2584                        ip_send_check(iph);
2585                }
2586        }
2587        return 1;
2588err:
2589        kfree_skb(skb);
2590        return 0;
2591}
2592#endif
2593
2594static void mpls_push(__be32 *mpls, struct pktgen_dev *pkt_dev)
2595{
2596        unsigned int i;
2597        for (i = 0; i < pkt_dev->nr_labels; i++)
2598                *mpls++ = pkt_dev->labels[i] & ~MPLS_STACK_BOTTOM;
2599
2600        mpls--;
2601        *mpls |= MPLS_STACK_BOTTOM;
2602}
2603
2604static inline __be16 build_tci(unsigned int id, unsigned int cfi,
2605                               unsigned int prio)
2606{
2607        return htons(id | (cfi << 12) | (prio << 13));
2608}
2609
2610static void pktgen_finalize_skb(struct pktgen_dev *pkt_dev, struct sk_buff *skb,
2611                                int datalen)
2612{
2613        struct timespec64 timestamp;
2614        struct pktgen_hdr *pgh;
2615
2616        pgh = skb_put(skb, sizeof(*pgh));
2617        datalen -= sizeof(*pgh);
2618
2619        if (pkt_dev->nfrags <= 0) {
2620                skb_put_zero(skb, datalen);
2621        } else {
2622                int frags = pkt_dev->nfrags;
2623                int i, len;
2624                int frag_len;
2625
2626
2627                if (frags > MAX_SKB_FRAGS)
2628                        frags = MAX_SKB_FRAGS;
2629                len = datalen - frags * PAGE_SIZE;
2630                if (len > 0) {
2631                        skb_put_zero(skb, len);
2632                        datalen = frags * PAGE_SIZE;
2633                }
2634
2635                i = 0;
2636                frag_len = (datalen/frags) < PAGE_SIZE ?
2637                           (datalen/frags) : PAGE_SIZE;
2638                while (datalen > 0) {
2639                        if (unlikely(!pkt_dev->page)) {
2640                                int node = numa_node_id();
2641
2642                                if (pkt_dev->node >= 0 && (pkt_dev->flags & F_NODE))
2643                                        node = pkt_dev->node;
2644                                pkt_dev->page = alloc_pages_node(node, GFP_KERNEL | __GFP_ZERO, 0);
2645                                if (!pkt_dev->page)
2646                                        break;
2647                        }
2648                        get_page(pkt_dev->page);
2649                        skb_frag_set_page(skb, i, pkt_dev->page);
2650                        skb_frag_off_set(&skb_shinfo(skb)->frags[i], 0);
2651                        /*last fragment, fill rest of data*/
2652                        if (i == (frags - 1))
2653                                skb_frag_size_set(&skb_shinfo(skb)->frags[i],
2654                                    (datalen < PAGE_SIZE ? datalen : PAGE_SIZE));
2655                        else
2656                                skb_frag_size_set(&skb_shinfo(skb)->frags[i], frag_len);
2657                        datalen -= skb_frag_size(&skb_shinfo(skb)->frags[i]);
2658                        skb->len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2659                        skb->data_len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2660                        i++;
2661                        skb_shinfo(skb)->nr_frags = i;
2662                }
2663        }
2664
2665        /* Stamp the time, and sequence number,
2666         * convert them to network byte order
2667         */
2668        pgh->pgh_magic = htonl(PKTGEN_MAGIC);
2669        pgh->seq_num = htonl(pkt_dev->seq_num);
2670
2671        if (pkt_dev->flags & F_NO_TIMESTAMP) {
2672                pgh->tv_sec = 0;
2673                pgh->tv_usec = 0;
2674        } else {
2675                /*
2676                 * pgh->tv_sec wraps in y2106 when interpreted as unsigned
2677                 * as done by wireshark, or y2038 when interpreted as signed.
2678                 * This is probably harmless, but if anyone wants to improve
2679                 * it, we could introduce a variant that puts 64-bit nanoseconds
2680                 * into the respective header bytes.
2681                 * This would also be slightly faster to read.
2682                 */
2683                ktime_get_real_ts64(&timestamp);
2684                pgh->tv_sec = htonl(timestamp.tv_sec);
2685                pgh->tv_usec = htonl(timestamp.tv_nsec / NSEC_PER_USEC);
2686        }
2687}
2688
2689static struct sk_buff *pktgen_alloc_skb(struct net_device *dev,
2690                                        struct pktgen_dev *pkt_dev)
2691{
2692        unsigned int extralen = LL_RESERVED_SPACE(dev);
2693        struct sk_buff *skb = NULL;
2694        unsigned int size;
2695
2696        size = pkt_dev->cur_pkt_size + 64 + extralen + pkt_dev->pkt_overhead;
2697        if (pkt_dev->flags & F_NODE) {
2698                int node = pkt_dev->node >= 0 ? pkt_dev->node : numa_node_id();
2699
2700                skb = __alloc_skb(NET_SKB_PAD + size, GFP_NOWAIT, 0, node);
2701                if (likely(skb)) {
2702                        skb_reserve(skb, NET_SKB_PAD);
2703                        skb->dev = dev;
2704                }
2705        } else {
2706                 skb = __netdev_alloc_skb(dev, size, GFP_NOWAIT);
2707        }
2708
2709        /* the caller pre-fetches from skb->data and reserves for the mac hdr */
2710        if (likely(skb))
2711                skb_reserve(skb, extralen - 16);
2712
2713        return skb;
2714}
2715
2716static struct sk_buff *fill_packet_ipv4(struct net_device *odev,
2717                                        struct pktgen_dev *pkt_dev)
2718{
2719        struct sk_buff *skb = NULL;
2720        __u8 *eth;
2721        struct udphdr *udph;
2722        int datalen, iplen;
2723        struct iphdr *iph;
2724        __be16 protocol = htons(ETH_P_IP);
2725        __be32 *mpls;
2726        __be16 *vlan_tci = NULL;                 /* Encapsulates priority and VLAN ID */
2727        __be16 *vlan_encapsulated_proto = NULL;  /* packet type ID field (or len) for VLAN tag */
2728        __be16 *svlan_tci = NULL;                /* Encapsulates priority and SVLAN ID */
2729        __be16 *svlan_encapsulated_proto = NULL; /* packet type ID field (or len) for SVLAN tag */
2730        u16 queue_map;
2731
2732        if (pkt_dev->nr_labels)
2733                protocol = htons(ETH_P_MPLS_UC);
2734
2735        if (pkt_dev->vlan_id != 0xffff)
2736                protocol = htons(ETH_P_8021Q);
2737
2738        /* Update any of the values, used when we're incrementing various
2739         * fields.
2740         */
2741        mod_cur_headers(pkt_dev);
2742        queue_map = pkt_dev->cur_queue_map;
2743
2744        skb = pktgen_alloc_skb(odev, pkt_dev);
2745        if (!skb) {
2746                sprintf(pkt_dev->result, "No memory");
2747                return NULL;
2748        }
2749
2750        prefetchw(skb->data);
2751        skb_reserve(skb, 16);
2752
2753        /*  Reserve for ethernet and IP header  */
2754        eth = skb_push(skb, 14);
2755        mpls = skb_put(skb, pkt_dev->nr_labels * sizeof(__u32));
2756        if (pkt_dev->nr_labels)
2757                mpls_push(mpls, pkt_dev);
2758
2759        if (pkt_dev->vlan_id != 0xffff) {
2760                if (pkt_dev->svlan_id != 0xffff) {
2761                        svlan_tci = skb_put(skb, sizeof(__be16));
2762                        *svlan_tci = build_tci(pkt_dev->svlan_id,
2763                                               pkt_dev->svlan_cfi,
2764                                               pkt_dev->svlan_p);
2765                        svlan_encapsulated_proto = skb_put(skb,
2766                                                           sizeof(__be16));
2767                        *svlan_encapsulated_proto = htons(ETH_P_8021Q);
2768                }
2769                vlan_tci = skb_put(skb, sizeof(__be16));
2770                *vlan_tci = build_tci(pkt_dev->vlan_id,
2771                                      pkt_dev->vlan_cfi,
2772                                      pkt_dev->vlan_p);
2773                vlan_encapsulated_proto = skb_put(skb, sizeof(__be16));
2774                *vlan_encapsulated_proto = htons(ETH_P_IP);
2775        }
2776
2777        skb_reset_mac_header(skb);
2778        skb_set_network_header(skb, skb->len);
2779        iph = skb_put(skb, sizeof(struct iphdr));
2780
2781        skb_set_transport_header(skb, skb->len);
2782        udph = skb_put(skb, sizeof(struct udphdr));
2783        skb_set_queue_mapping(skb, queue_map);
2784        skb->priority = pkt_dev->skb_priority;
2785
2786        memcpy(eth, pkt_dev->hh, 12);
2787        *(__be16 *) & eth[12] = protocol;
2788
2789        /* Eth + IPh + UDPh + mpls */
2790        datalen = pkt_dev->cur_pkt_size - 14 - 20 - 8 -
2791                  pkt_dev->pkt_overhead;
2792        if (datalen < 0 || datalen < sizeof(struct pktgen_hdr))
2793                datalen = sizeof(struct pktgen_hdr);
2794
2795        udph->source = htons(pkt_dev->cur_udp_src);
2796        udph->dest = htons(pkt_dev->cur_udp_dst);
2797        udph->len = htons(datalen + 8); /* DATA + udphdr */
2798        udph->check = 0;
2799
2800        iph->ihl = 5;
2801        iph->version = 4;
2802        iph->ttl = 32;
2803        iph->tos = pkt_dev->tos;
2804        iph->protocol = IPPROTO_UDP;    /* UDP */
2805        iph->saddr = pkt_dev->cur_saddr;
2806        iph->daddr = pkt_dev->cur_daddr;
2807        iph->id = htons(pkt_dev->ip_id);
2808        pkt_dev->ip_id++;
2809        iph->frag_off = 0;
2810        iplen = 20 + 8 + datalen;
2811        iph->tot_len = htons(iplen);
2812        ip_send_check(iph);
2813        skb->protocol = protocol;
2814        skb->dev = odev;
2815        skb->pkt_type = PACKET_HOST;
2816
2817        pktgen_finalize_skb(pkt_dev, skb, datalen);
2818
2819        if (!(pkt_dev->flags & F_UDPCSUM)) {
2820                skb->ip_summed = CHECKSUM_NONE;
2821        } else if (odev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM)) {
2822                skb->ip_summed = CHECKSUM_PARTIAL;
2823                skb->csum = 0;
2824                udp4_hwcsum(skb, iph->saddr, iph->daddr);
2825        } else {
2826                __wsum csum = skb_checksum(skb, skb_transport_offset(skb), datalen + 8, 0);
2827
2828                /* add protocol-dependent pseudo-header */
2829                udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
2830                                                datalen + 8, IPPROTO_UDP, csum);
2831
2832                if (udph->check == 0)
2833                        udph->check = CSUM_MANGLED_0;
2834        }
2835
2836#ifdef CONFIG_XFRM
2837        if (!process_ipsec(pkt_dev, skb, protocol))
2838                return NULL;
2839#endif
2840
2841        return skb;
2842}
2843
2844static struct sk_buff *fill_packet_ipv6(struct net_device *odev,
2845                                        struct pktgen_dev *pkt_dev)
2846{
2847        struct sk_buff *skb = NULL;
2848        __u8 *eth;
2849        struct udphdr *udph;
2850        int datalen, udplen;
2851        struct ipv6hdr *iph;
2852        __be16 protocol = htons(ETH_P_IPV6);
2853        __be32 *mpls;
2854        __be16 *vlan_tci = NULL;                 /* Encapsulates priority and VLAN ID */
2855        __be16 *vlan_encapsulated_proto = NULL;  /* packet type ID field (or len) for VLAN tag */
2856        __be16 *svlan_tci = NULL;                /* Encapsulates priority and SVLAN ID */
2857        __be16 *svlan_encapsulated_proto = NULL; /* packet type ID field (or len) for SVLAN tag */
2858        u16 queue_map;
2859
2860        if (pkt_dev->nr_labels)
2861                protocol = htons(ETH_P_MPLS_UC);
2862
2863        if (pkt_dev->vlan_id != 0xffff)
2864                protocol = htons(ETH_P_8021Q);
2865
2866        /* Update any of the values, used when we're incrementing various
2867         * fields.
2868         */
2869        mod_cur_headers(pkt_dev);
2870        queue_map = pkt_dev->cur_queue_map;
2871
2872        skb = pktgen_alloc_skb(odev, pkt_dev);
2873        if (!skb) {
2874                sprintf(pkt_dev->result, "No memory");
2875                return NULL;
2876        }
2877
2878        prefetchw(skb->data);
2879        skb_reserve(skb, 16);
2880
2881        /*  Reserve for ethernet and IP header  */
2882        eth = skb_push(skb, 14);
2883        mpls = skb_put(skb, pkt_dev->nr_labels * sizeof(__u32));
2884        if (pkt_dev->nr_labels)
2885                mpls_push(mpls, pkt_dev);
2886
2887        if (pkt_dev->vlan_id != 0xffff) {
2888                if (pkt_dev->svlan_id != 0xffff) {
2889                        svlan_tci = skb_put(skb, sizeof(__be16));
2890                        *svlan_tci = build_tci(pkt_dev->svlan_id,
2891                                               pkt_dev->svlan_cfi,
2892                                               pkt_dev->svlan_p);
2893                        svlan_encapsulated_proto = skb_put(skb,
2894                                                           sizeof(__be16));
2895                        *svlan_encapsulated_proto = htons(ETH_P_8021Q);
2896                }
2897                vlan_tci = skb_put(skb, sizeof(__be16));
2898                *vlan_tci = build_tci(pkt_dev->vlan_id,
2899                                      pkt_dev->vlan_cfi,
2900                                      pkt_dev->vlan_p);
2901                vlan_encapsulated_proto = skb_put(skb, sizeof(__be16));
2902                *vlan_encapsulated_proto = htons(ETH_P_IPV6);
2903        }
2904
2905        skb_reset_mac_header(skb);
2906        skb_set_network_header(skb, skb->len);
2907        iph = skb_put(skb, sizeof(struct ipv6hdr));
2908
2909        skb_set_transport_header(skb, skb->len);
2910        udph = skb_put(skb, sizeof(struct udphdr));
2911        skb_set_queue_mapping(skb, queue_map);
2912        skb->priority = pkt_dev->skb_priority;
2913
2914        memcpy(eth, pkt_dev->hh, 12);
2915        *(__be16 *) &eth[12] = protocol;
2916
2917        /* Eth + IPh + UDPh + mpls */
2918        datalen = pkt_dev->cur_pkt_size - 14 -
2919                  sizeof(struct ipv6hdr) - sizeof(struct udphdr) -
2920                  pkt_dev->pkt_overhead;
2921
2922        if (datalen < 0 || datalen < sizeof(struct pktgen_hdr)) {
2923                datalen = sizeof(struct pktgen_hdr);
2924                net_info_ratelimited("increased datalen to %d\n", datalen);
2925        }
2926
2927        udplen = datalen + sizeof(struct udphdr);
2928        udph->source = htons(pkt_dev->cur_udp_src);
2929        udph->dest = htons(pkt_dev->cur_udp_dst);
2930        udph->len = htons(udplen);
2931        udph->check = 0;
2932
2933        *(__be32 *) iph = htonl(0x60000000);    /* Version + flow */
2934
2935        if (pkt_dev->traffic_class) {
2936                /* Version + traffic class + flow (0) */
2937                *(__be32 *)iph |= htonl(0x60000000 | (pkt_dev->traffic_class << 20));
2938        }
2939
2940        iph->hop_limit = 32;
2941
2942        iph->payload_len = htons(udplen);
2943        iph->nexthdr = IPPROTO_UDP;
2944
2945        iph->daddr = pkt_dev->cur_in6_daddr;
2946        iph->saddr = pkt_dev->cur_in6_saddr;
2947
2948        skb->protocol = protocol;
2949        skb->dev = odev;
2950        skb->pkt_type = PACKET_HOST;
2951
2952        pktgen_finalize_skb(pkt_dev, skb, datalen);
2953
2954        if (!(pkt_dev->flags & F_UDPCSUM)) {
2955                skb->ip_summed = CHECKSUM_NONE;
2956        } else if (odev->features & (NETIF_F_HW_CSUM | NETIF_F_IPV6_CSUM)) {
2957                skb->ip_summed = CHECKSUM_PARTIAL;
2958                skb->csum_start = skb_transport_header(skb) - skb->head;
2959                skb->csum_offset = offsetof(struct udphdr, check);
2960                udph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, udplen, IPPROTO_UDP, 0);
2961        } else {
2962                __wsum csum = skb_checksum(skb, skb_transport_offset(skb), udplen, 0);
2963
2964                /* add protocol-dependent pseudo-header */
2965                udph->check = csum_ipv6_magic(&iph->saddr, &iph->daddr, udplen, IPPROTO_UDP, csum);
2966
2967                if (udph->check == 0)
2968                        udph->check = CSUM_MANGLED_0;
2969        }
2970
2971        return skb;
2972}
2973
2974static struct sk_buff *fill_packet(struct net_device *odev,
2975                                   struct pktgen_dev *pkt_dev)
2976{
2977        if (pkt_dev->flags & F_IPV6)
2978                return fill_packet_ipv6(odev, pkt_dev);
2979        else
2980                return fill_packet_ipv4(odev, pkt_dev);
2981}
2982
2983static void pktgen_clear_counters(struct pktgen_dev *pkt_dev)
2984{
2985        pkt_dev->seq_num = 1;
2986        pkt_dev->idle_acc = 0;
2987        pkt_dev->sofar = 0;
2988        pkt_dev->tx_bytes = 0;
2989        pkt_dev->errors = 0;
2990}
2991
2992/* Set up structure for sending pkts, clear counters */
2993
2994static void pktgen_run(struct pktgen_thread *t)
2995{
2996        struct pktgen_dev *pkt_dev;
2997        int started = 0;
2998
2999        func_enter();
3000
3001        rcu_read_lock();
3002        list_for_each_entry_rcu(pkt_dev, &t->if_list, list) {
3003
3004                /*
3005                 * setup odev and create initial packet.
3006                 */
3007                pktgen_setup_inject(pkt_dev);
3008
3009                if (pkt_dev->odev) {
3010                        pktgen_clear_counters(pkt_dev);
3011                        pkt_dev->skb = NULL;
3012                        pkt_dev->started_at = pkt_dev->next_tx = ktime_get();
3013
3014                        set_pkt_overhead(pkt_dev);
3015
3016                        strcpy(pkt_dev->result, "Starting");
3017                        pkt_dev->running = 1;   /* Cranke yeself! */
3018                        started++;
3019                } else
3020                        strcpy(pkt_dev->result, "Error starting");
3021        }
3022        rcu_read_unlock();
3023        if (started)
3024                t->control &= ~(T_STOP);
3025}
3026
3027static void pktgen_handle_all_threads(struct pktgen_net *pn, u32 flags)
3028{
3029        struct pktgen_thread *t;
3030
3031        mutex_lock(&pktgen_thread_lock);
3032
3033        list_for_each_entry(t, &pn->pktgen_threads, th_list)
3034                t->control |= (flags);
3035
3036        mutex_unlock(&pktgen_thread_lock);
3037}
3038
3039static void pktgen_stop_all_threads(struct pktgen_net *pn)
3040{
3041        func_enter();
3042
3043        pktgen_handle_all_threads(pn, T_STOP);
3044}
3045
3046static int thread_is_running(const struct pktgen_thread *t)
3047{
3048        const struct pktgen_dev *pkt_dev;
3049
3050        rcu_read_lock();
3051        list_for_each_entry_rcu(pkt_dev, &t->if_list, list)
3052                if (pkt_dev->running) {
3053                        rcu_read_unlock();
3054                        return 1;
3055                }
3056        rcu_read_unlock();
3057        return 0;
3058}
3059
3060static int pktgen_wait_thread_run(struct pktgen_thread *t)
3061{
3062        while (thread_is_running(t)) {
3063
3064                /* note: 't' will still be around even after the unlock/lock
3065                 * cycle because pktgen_thread threads are only cleared at
3066                 * net exit
3067                 */
3068                mutex_unlock(&pktgen_thread_lock);
3069                msleep_interruptible(100);
3070                mutex_lock(&pktgen_thread_lock);
3071
3072                if (signal_pending(current))
3073                        goto signal;
3074        }
3075        return 1;
3076signal:
3077        return 0;
3078}
3079
3080static int pktgen_wait_all_threads_run(struct pktgen_net *pn)
3081{
3082        struct pktgen_thread *t;
3083        int sig = 1;
3084
3085        /* prevent from racing with rmmod */
3086        if (!try_module_get(THIS_MODULE))
3087                return sig;
3088
3089        mutex_lock(&pktgen_thread_lock);
3090
3091        list_for_each_entry(t, &pn->pktgen_threads, th_list) {
3092                sig = pktgen_wait_thread_run(t);
3093                if (sig == 0)
3094                        break;
3095        }
3096
3097        if (sig == 0)
3098                list_for_each_entry(t, &pn->pktgen_threads, th_list)
3099                        t->control |= (T_STOP);
3100
3101        mutex_unlock(&pktgen_thread_lock);
3102        module_put(THIS_MODULE);
3103        return sig;
3104}
3105
3106static void pktgen_run_all_threads(struct pktgen_net *pn)
3107{
3108        func_enter();
3109
3110        pktgen_handle_all_threads(pn, T_RUN);
3111
3112        /* Propagate thread->control  */
3113        schedule_timeout_interruptible(msecs_to_jiffies(125));
3114
3115        pktgen_wait_all_threads_run(pn);
3116}
3117
3118static void pktgen_reset_all_threads(struct pktgen_net *pn)
3119{
3120        func_enter();
3121
3122        pktgen_handle_all_threads(pn, T_REMDEVALL);
3123
3124        /* Propagate thread->control  */
3125        schedule_timeout_interruptible(msecs_to_jiffies(125));
3126
3127        pktgen_wait_all_threads_run(pn);
3128}
3129
3130static void show_results(struct pktgen_dev *pkt_dev, int nr_frags)
3131{
3132        __u64 bps, mbps, pps;
3133        char *p = pkt_dev->result;
3134        ktime_t elapsed = ktime_sub(pkt_dev->stopped_at,
3135                                    pkt_dev->started_at);
3136        ktime_t idle = ns_to_ktime(pkt_dev->idle_acc);
3137
3138        p += sprintf(p, "OK: %llu(c%llu+d%llu) usec, %llu (%dbyte,%dfrags)\n",
3139                     (unsigned long long)ktime_to_us(elapsed),
3140                     (unsigned long long)ktime_to_us(ktime_sub(elapsed, idle)),
3141                     (unsigned long long)ktime_to_us(idle),
3142                     (unsigned long long)pkt_dev->sofar,
3143                     pkt_dev->cur_pkt_size, nr_frags);
3144
3145        pps = div64_u64(pkt_dev->sofar * NSEC_PER_SEC,
3146                        ktime_to_ns(elapsed));
3147
3148        bps = pps * 8 * pkt_dev->cur_pkt_size;
3149
3150        mbps = bps;
3151        do_div(mbps, 1000000);
3152        p += sprintf(p, "  %llupps %lluMb/sec (%llubps) errors: %llu",
3153                     (unsigned long long)pps,
3154                     (unsigned long long)mbps,
3155                     (unsigned long long)bps,
3156                     (unsigned long long)pkt_dev->errors);
3157}
3158
3159/* Set stopped-at timer, remove from running list, do counters & statistics */
3160static int pktgen_stop_device(struct pktgen_dev *pkt_dev)
3161{
3162        int nr_frags = pkt_dev->skb ? skb_shinfo(pkt_dev->skb)->nr_frags : -1;
3163
3164        if (!pkt_dev->running) {
3165                pr_warn("interface: %s is already stopped\n",
3166                        pkt_dev->odevname);
3167                return -EINVAL;
3168        }
3169
3170        pkt_dev->running = 0;
3171        kfree_skb(pkt_dev->skb);
3172        pkt_dev->skb = NULL;
3173        pkt_dev->stopped_at = ktime_get();
3174
3175        show_results(pkt_dev, nr_frags);
3176
3177        return 0;
3178}
3179
3180static struct pktgen_dev *next_to_run(struct pktgen_thread *t)
3181{
3182        struct pktgen_dev *pkt_dev, *best = NULL;
3183
3184        rcu_read_lock();
3185        list_for_each_entry_rcu(pkt_dev, &t->if_list, list) {
3186                if (!pkt_dev->running)
3187                        continue;
3188                if (best == NULL)
3189                        best = pkt_dev;
3190                else if (ktime_compare(pkt_dev->next_tx, best->next_tx) < 0)
3191                        best = pkt_dev;
3192        }
3193        rcu_read_unlock();
3194
3195        return best;
3196}
3197
3198static void pktgen_stop(struct pktgen_thread *t)
3199{
3200        struct pktgen_dev *pkt_dev;
3201
3202        func_enter();
3203
3204        rcu_read_lock();
3205
3206        list_for_each_entry_rcu(pkt_dev, &t->if_list, list) {
3207                pktgen_stop_device(pkt_dev);
3208        }
3209
3210        rcu_read_unlock();
3211}
3212
3213/*
3214 * one of our devices needs to be removed - find it
3215 * and remove it
3216 */
3217static void pktgen_rem_one_if(struct pktgen_thread *t)
3218{
3219        struct list_head *q, *n;
3220        struct pktgen_dev *cur;
3221
3222        func_enter();
3223
3224        list_for_each_safe(q, n, &t->if_list) {
3225                cur = list_entry(q, struct pktgen_dev, list);
3226
3227                if (!cur->removal_mark)
3228                        continue;
3229
3230                kfree_skb(cur->skb);
3231                cur->skb = NULL;
3232
3233                pktgen_remove_device(t, cur);
3234
3235                break;
3236        }
3237}
3238
3239static void pktgen_rem_all_ifs(struct pktgen_thread *t)
3240{
3241        struct list_head *q, *n;
3242        struct pktgen_dev *cur;
3243
3244        func_enter();
3245
3246        /* Remove all devices, free mem */
3247
3248        list_for_each_safe(q, n, &t->if_list) {
3249                cur = list_entry(q, struct pktgen_dev, list);
3250
3251                kfree_skb(cur->skb);
3252                cur->skb = NULL;
3253
3254                pktgen_remove_device(t, cur);
3255        }
3256}
3257
3258static void pktgen_rem_thread(struct pktgen_thread *t)
3259{
3260        /* Remove from the thread list */
3261        remove_proc_entry(t->tsk->comm, t->net->proc_dir);
3262}
3263
3264static void pktgen_resched(struct pktgen_dev *pkt_dev)
3265{
3266        ktime_t idle_start = ktime_get();
3267        schedule();
3268        pkt_dev->idle_acc += ktime_to_ns(ktime_sub(ktime_get(), idle_start));
3269}
3270
3271static void pktgen_wait_for_skb(struct pktgen_dev *pkt_dev)
3272{
3273        ktime_t idle_start = ktime_get();
3274
3275        while (refcount_read(&(pkt_dev->skb->users)) != 1) {
3276                if (signal_pending(current))
3277                        break;
3278
3279                if (need_resched())
3280                        pktgen_resched(pkt_dev);
3281                else
3282                        cpu_relax();
3283        }
3284        pkt_dev->idle_acc += ktime_to_ns(ktime_sub(ktime_get(), idle_start));
3285}
3286
3287static void pktgen_xmit(struct pktgen_dev *pkt_dev)
3288{
3289        unsigned int burst = READ_ONCE(pkt_dev->burst);
3290        struct net_device *odev = pkt_dev->odev;
3291        struct netdev_queue *txq;
3292        struct sk_buff *skb;
3293        int ret;
3294
3295        /* If device is offline, then don't send */
3296        if (unlikely(!netif_running(odev) || !netif_carrier_ok(odev))) {
3297                pktgen_stop_device(pkt_dev);
3298                return;
3299        }
3300
3301        /* This is max DELAY, this has special meaning of
3302         * "never transmit"
3303         */
3304        if (unlikely(pkt_dev->delay == ULLONG_MAX)) {
3305                pkt_dev->next_tx = ktime_add_ns(ktime_get(), ULONG_MAX);
3306                return;
3307        }
3308
3309        /* If no skb or clone count exhausted then get new one */
3310        if (!pkt_dev->skb || (pkt_dev->last_ok &&
3311                              ++pkt_dev->clone_count >= pkt_dev->clone_skb)) {
3312                /* build a new pkt */
3313                kfree_skb(pkt_dev->skb);
3314
3315                pkt_dev->skb = fill_packet(odev, pkt_dev);
3316                if (pkt_dev->skb == NULL) {
3317                        pr_err("ERROR: couldn't allocate skb in fill_packet\n");
3318                        schedule();
3319                        pkt_dev->clone_count--; /* back out increment, OOM */
3320                        return;
3321                }
3322                pkt_dev->last_pkt_size = pkt_dev->skb->len;
3323                pkt_dev->clone_count = 0;       /* reset counter */
3324        }
3325
3326        if (pkt_dev->delay && pkt_dev->last_ok)
3327                spin(pkt_dev, pkt_dev->next_tx);
3328
3329        if (pkt_dev->xmit_mode == M_NETIF_RECEIVE) {
3330                skb = pkt_dev->skb;
3331                skb->protocol = eth_type_trans(skb, skb->dev);
3332                refcount_add(burst, &skb->users);
3333                local_bh_disable();
3334                do {
3335                        ret = netif_receive_skb(skb);
3336                        if (ret == NET_RX_DROP)
3337                                pkt_dev->errors++;
3338                        pkt_dev->sofar++;
3339                        pkt_dev->seq_num++;
3340                        if (refcount_read(&skb->users) != burst) {
3341                                /* skb was queued by rps/rfs or taps,
3342                                 * so cannot reuse this skb
3343                                 */
3344                                WARN_ON(refcount_sub_and_test(burst - 1, &skb->users));
3345                                /* get out of the loop and wait
3346                                 * until skb is consumed
3347                                 */
3348                                break;
3349                        }
3350                        /* skb was 'freed' by stack, so clean few
3351                         * bits and reuse it
3352                         */
3353                        skb_reset_redirect(skb);
3354                } while (--burst > 0);
3355                goto out; /* Skips xmit_mode M_START_XMIT */
3356        } else if (pkt_dev->xmit_mode == M_QUEUE_XMIT) {
3357                local_bh_disable();
3358                refcount_inc(&pkt_dev->skb->users);
3359
3360                ret = dev_queue_xmit(pkt_dev->skb);
3361                switch (ret) {
3362                case NET_XMIT_SUCCESS:
3363                        pkt_dev->sofar++;
3364                        pkt_dev->seq_num++;
3365                        pkt_dev->tx_bytes += pkt_dev->last_pkt_size;
3366                        break;
3367                case NET_XMIT_DROP:
3368                case NET_XMIT_CN:
3369                /* These are all valid return codes for a qdisc but
3370                 * indicate packets are being dropped or will likely
3371                 * be dropped soon.
3372                 */
3373                case NETDEV_TX_BUSY:
3374                /* qdisc may call dev_hard_start_xmit directly in cases
3375                 * where no queues exist e.g. loopback device, virtual
3376                 * devices, etc. In this case we need to handle
3377                 * NETDEV_TX_ codes.
3378                 */
3379                default:
3380                        pkt_dev->errors++;
3381                        net_info_ratelimited("%s xmit error: %d\n",
3382                                             pkt_dev->odevname, ret);
3383                        break;
3384                }
3385                goto out;
3386        }
3387
3388        txq = skb_get_tx_queue(odev, pkt_dev->skb);
3389
3390        local_bh_disable();
3391
3392        HARD_TX_LOCK(odev, txq, smp_processor_id());
3393
3394        if (unlikely(netif_xmit_frozen_or_drv_stopped(txq))) {
3395                pkt_dev->last_ok = 0;
3396                goto unlock;
3397        }
3398        refcount_add(burst, &pkt_dev->skb->users);
3399
3400xmit_more:
3401        ret = netdev_start_xmit(pkt_dev->skb, odev, txq, --burst > 0);
3402
3403        switch (ret) {
3404        case NETDEV_TX_OK:
3405                pkt_dev->last_ok = 1;
3406                pkt_dev->sofar++;
3407                pkt_dev->seq_num++;
3408                pkt_dev->tx_bytes += pkt_dev->last_pkt_size;
3409                if (burst > 0 && !netif_xmit_frozen_or_drv_stopped(txq))
3410                        goto xmit_more;
3411                break;
3412        case NET_XMIT_DROP:
3413        case NET_XMIT_CN:
3414                /* skb has been consumed */
3415                pkt_dev->errors++;
3416                break;
3417        default: /* Drivers are not supposed to return other values! */
3418                net_info_ratelimited("%s xmit error: %d\n",
3419                                     pkt_dev->odevname, ret);
3420                pkt_dev->errors++;
3421                fallthrough;
3422        case NETDEV_TX_BUSY:
3423                /* Retry it next time */
3424                refcount_dec(&(pkt_dev->skb->users));
3425                pkt_dev->last_ok = 0;
3426        }
3427        if (unlikely(burst))
3428                WARN_ON(refcount_sub_and_test(burst, &pkt_dev->skb->users));
3429unlock:
3430        HARD_TX_UNLOCK(odev, txq);
3431
3432out:
3433        local_bh_enable();
3434
3435        /* If pkt_dev->count is zero, then run forever */
3436        if ((pkt_dev->count != 0) && (pkt_dev->sofar >= pkt_dev->count)) {
3437                pktgen_wait_for_skb(pkt_dev);
3438
3439                /* Done with this */
3440                pktgen_stop_device(pkt_dev);
3441        }
3442}
3443
3444/*
3445 * Main loop of the thread goes here
3446 */
3447
3448static int pktgen_thread_worker(void *arg)
3449{
3450        DEFINE_WAIT(wait);
3451        struct pktgen_thread *t = arg;
3452        struct pktgen_dev *pkt_dev = NULL;
3453        int cpu = t->cpu;
3454
3455        WARN_ON(smp_processor_id() != cpu);
3456
3457        init_waitqueue_head(&t->queue);
3458        complete(&t->start_done);
3459
3460        pr_debug("starting pktgen/%d:  pid=%d\n", cpu, task_pid_nr(current));
3461
3462        set_freezable();
3463
3464        while (!kthread_should_stop()) {
3465                pkt_dev = next_to_run(t);
3466
3467                if (unlikely(!pkt_dev && t->control == 0)) {
3468                        if (t->net->pktgen_exiting)
3469                                break;
3470                        wait_event_interruptible_timeout(t->queue,
3471                                                         t->control != 0,
3472                                                         HZ/10);
3473                        try_to_freeze();
3474                        continue;
3475                }
3476
3477                if (likely(pkt_dev)) {
3478                        pktgen_xmit(pkt_dev);
3479
3480                        if (need_resched())
3481                                pktgen_resched(pkt_dev);
3482                        else
3483                                cpu_relax();
3484                }
3485
3486                if (t->control & T_STOP) {
3487                        pktgen_stop(t);
3488                        t->control &= ~(T_STOP);
3489                }
3490
3491                if (t->control & T_RUN) {
3492                        pktgen_run(t);
3493                        t->control &= ~(T_RUN);
3494                }
3495
3496                if (t->control & T_REMDEVALL) {
3497                        pktgen_rem_all_ifs(t);
3498                        t->control &= ~(T_REMDEVALL);
3499                }
3500
3501                if (t->control & T_REMDEV) {
3502                        pktgen_rem_one_if(t);
3503                        t->control &= ~(T_REMDEV);
3504                }
3505
3506                try_to_freeze();
3507        }
3508
3509        pr_debug("%s stopping all device\n", t->tsk->comm);
3510        pktgen_stop(t);
3511
3512        pr_debug("%s removing all device\n", t->tsk->comm);
3513        pktgen_rem_all_ifs(t);
3514
3515        pr_debug("%s removing thread\n", t->tsk->comm);
3516        pktgen_rem_thread(t);
3517
3518        return 0;
3519}
3520
3521static struct pktgen_dev *pktgen_find_dev(struct pktgen_thread *t,
3522                                          const char *ifname, bool exact)
3523{
3524        struct pktgen_dev *p, *pkt_dev = NULL;
3525        size_t len = strlen(ifname);
3526
3527        rcu_read_lock();
3528        list_for_each_entry_rcu(p, &t->if_list, list)
3529                if (strncmp(p->odevname, ifname, len) == 0) {
3530                        if (p->odevname[len]) {
3531                                if (exact || p->odevname[len] != '@')
3532                                        continue;
3533                        }
3534                        pkt_dev = p;
3535                        break;
3536                }
3537
3538        rcu_read_unlock();
3539        pr_debug("find_dev(%s) returning %p\n", ifname, pkt_dev);
3540        return pkt_dev;
3541}
3542
3543/*
3544 * Adds a dev at front of if_list.
3545 */
3546
3547static int add_dev_to_thread(struct pktgen_thread *t,
3548                             struct pktgen_dev *pkt_dev)
3549{
3550        int rv = 0;
3551
3552        /* This function cannot be called concurrently, as its called
3553         * under pktgen_thread_lock mutex, but it can run from
3554         * userspace on another CPU than the kthread.  The if_lock()
3555         * is used here to sync with concurrent instances of
3556         * _rem_dev_from_if_list() invoked via kthread, which is also
3557         * updating the if_list */
3558        if_lock(t);
3559
3560        if (pkt_dev->pg_thread) {
3561                pr_err("ERROR: already assigned to a thread\n");
3562                rv = -EBUSY;
3563                goto out;
3564        }
3565
3566        pkt_dev->running = 0;
3567        pkt_dev->pg_thread = t;
3568        list_add_rcu(&pkt_dev->list, &t->if_list);
3569
3570out:
3571        if_unlock(t);
3572        return rv;
3573}
3574
3575/* Called under thread lock */
3576
3577static int pktgen_add_device(struct pktgen_thread *t, const char *ifname)
3578{
3579        struct pktgen_dev *pkt_dev;
3580        int err;
3581        int node = cpu_to_node(t->cpu);
3582
3583        /* We don't allow a device to be on several threads */
3584
3585        pkt_dev = __pktgen_NN_threads(t->net, ifname, FIND);
3586        if (pkt_dev) {
3587                pr_err("ERROR: interface already used\n");
3588                return -EBUSY;
3589        }
3590
3591        pkt_dev = kzalloc_node(sizeof(struct pktgen_dev), GFP_KERNEL, node);
3592        if (!pkt_dev)
3593                return -ENOMEM;
3594
3595        strcpy(pkt_dev->odevname, ifname);
3596        pkt_dev->flows = vzalloc_node(array_size(MAX_CFLOWS,
3597                                                 sizeof(struct flow_state)),
3598                                      node);
3599        if (pkt_dev->flows == NULL) {
3600                kfree(pkt_dev);
3601                return -ENOMEM;
3602        }
3603
3604        pkt_dev->removal_mark = 0;
3605        pkt_dev->nfrags = 0;
3606        pkt_dev->delay = pg_delay_d;
3607        pkt_dev->count = pg_count_d;
3608        pkt_dev->sofar = 0;
3609        pkt_dev->udp_src_min = 9;       /* sink port */
3610        pkt_dev->udp_src_max = 9;
3611        pkt_dev->udp_dst_min = 9;
3612        pkt_dev->udp_dst_max = 9;
3613        pkt_dev->vlan_p = 0;
3614        pkt_dev->vlan_cfi = 0;
3615        pkt_dev->vlan_id = 0xffff;
3616        pkt_dev->svlan_p = 0;
3617        pkt_dev->svlan_cfi = 0;
3618        pkt_dev->svlan_id = 0xffff;
3619        pkt_dev->burst = 1;
3620        pkt_dev->node = NUMA_NO_NODE;
3621
3622        err = pktgen_setup_dev(t->net, pkt_dev, ifname);
3623        if (err)
3624                goto out1;
3625        if (pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING)
3626                pkt_dev->clone_skb = pg_clone_skb_d;
3627
3628        pkt_dev->entry = proc_create_data(ifname, 0600, t->net->proc_dir,
3629                                          &pktgen_if_proc_ops, pkt_dev);
3630        if (!pkt_dev->entry) {
3631                pr_err("cannot create %s/%s procfs entry\n",
3632                       PG_PROC_DIR, ifname);
3633                err = -EINVAL;
3634                goto out2;
3635        }
3636#ifdef CONFIG_XFRM
3637        pkt_dev->ipsmode = XFRM_MODE_TRANSPORT;
3638        pkt_dev->ipsproto = IPPROTO_ESP;
3639
3640        /* xfrm tunnel mode needs additional dst to extract outter
3641         * ip header protocol/ttl/id field, here creat a phony one.
3642         * instead of looking for a valid rt, which definitely hurting
3643         * performance under such circumstance.
3644         */
3645        pkt_dev->dstops.family = AF_INET;
3646        pkt_dev->xdst.u.dst.dev = pkt_dev->odev;
3647        dst_init_metrics(&pkt_dev->xdst.u.dst, pktgen_dst_metrics, false);
3648        pkt_dev->xdst.child = &pkt_dev->xdst.u.dst;
3649        pkt_dev->xdst.u.dst.ops = &pkt_dev->dstops;
3650#endif
3651
3652        return add_dev_to_thread(t, pkt_dev);
3653out2:
3654        dev_put(pkt_dev->odev);
3655out1:
3656#ifdef CONFIG_XFRM
3657        free_SAs(pkt_dev);
3658#endif
3659        vfree(pkt_dev->flows);
3660        kfree(pkt_dev);
3661        return err;
3662}
3663
3664static int __net_init pktgen_create_thread(int cpu, struct pktgen_net *pn)
3665{
3666        struct pktgen_thread *t;
3667        struct proc_dir_entry *pe;
3668        struct task_struct *p;
3669
3670        t = kzalloc_node(sizeof(struct pktgen_thread), GFP_KERNEL,
3671                         cpu_to_node(cpu));
3672        if (!t) {
3673                pr_err("ERROR: out of memory, can't create new thread\n");
3674                return -ENOMEM;
3675        }
3676
3677        mutex_init(&t->if_lock);
3678        t->cpu = cpu;
3679
3680        INIT_LIST_HEAD(&t->if_list);
3681
3682        list_add_tail(&t->th_list, &pn->pktgen_threads);
3683        init_completion(&t->start_done);
3684
3685        p = kthread_create_on_node(pktgen_thread_worker,
3686                                   t,
3687                                   cpu_to_node(cpu),
3688                                   "kpktgend_%d", cpu);
3689        if (IS_ERR(p)) {
3690                pr_err("kthread_create_on_node() failed for cpu %d\n", t->cpu);
3691                list_del(&t->th_list);
3692                kfree(t);
3693                return PTR_ERR(p);
3694        }
3695        kthread_bind(p, cpu);
3696        t->tsk = p;
3697
3698        pe = proc_create_data(t->tsk->comm, 0600, pn->proc_dir,
3699                              &pktgen_thread_proc_ops, t);
3700        if (!pe) {
3701                pr_err("cannot create %s/%s procfs entry\n",
3702                       PG_PROC_DIR, t->tsk->comm);
3703                kthread_stop(p);
3704                list_del(&t->th_list);
3705                kfree(t);
3706                return -EINVAL;
3707        }
3708
3709        t->net = pn;
3710        get_task_struct(p);
3711        wake_up_process(p);
3712        wait_for_completion(&t->start_done);
3713
3714        return 0;
3715}
3716
3717/*
3718 * Removes a device from the thread if_list.
3719 */
3720static void _rem_dev_from_if_list(struct pktgen_thread *t,
3721                                  struct pktgen_dev *pkt_dev)
3722{
3723        struct list_head *q, *n;
3724        struct pktgen_dev *p;
3725
3726        if_lock(t);
3727        list_for_each_safe(q, n, &t->if_list) {
3728                p = list_entry(q, struct pktgen_dev, list);
3729                if (p == pkt_dev)
3730                        list_del_rcu(&p->list);
3731        }
3732        if_unlock(t);
3733}
3734
3735static int pktgen_remove_device(struct pktgen_thread *t,
3736                                struct pktgen_dev *pkt_dev)
3737{
3738        pr_debug("remove_device pkt_dev=%p\n", pkt_dev);
3739
3740        if (pkt_dev->running) {
3741                pr_warn("WARNING: trying to remove a running interface, stopping it now\n");
3742                pktgen_stop_device(pkt_dev);
3743        }
3744
3745        /* Dis-associate from the interface */
3746
3747        if (pkt_dev->odev) {
3748                dev_put(pkt_dev->odev);
3749                pkt_dev->odev = NULL;
3750        }
3751
3752        /* Remove proc before if_list entry, because add_device uses
3753         * list to determine if interface already exist, avoid race
3754         * with proc_create_data() */
3755        proc_remove(pkt_dev->entry);
3756
3757        /* And update the thread if_list */
3758        _rem_dev_from_if_list(t, pkt_dev);
3759
3760#ifdef CONFIG_XFRM
3761        free_SAs(pkt_dev);
3762#endif
3763        vfree(pkt_dev->flows);
3764        if (pkt_dev->page)
3765                put_page(pkt_dev->page);
3766        kfree_rcu(pkt_dev, rcu);
3767        return 0;
3768}
3769
3770static int __net_init pg_net_init(struct net *net)
3771{
3772        struct pktgen_net *pn = net_generic(net, pg_net_id);
3773        struct proc_dir_entry *pe;
3774        int cpu, ret = 0;
3775
3776        pn->net = net;
3777        INIT_LIST_HEAD(&pn->pktgen_threads);
3778        pn->pktgen_exiting = false;
3779        pn->proc_dir = proc_mkdir(PG_PROC_DIR, pn->net->proc_net);
3780        if (!pn->proc_dir) {
3781                pr_warn("cannot create /proc/net/%s\n", PG_PROC_DIR);
3782                return -ENODEV;
3783        }
3784        pe = proc_create(PGCTRL, 0600, pn->proc_dir, &pktgen_proc_ops);
3785        if (pe == NULL) {
3786                pr_err("cannot create %s procfs entry\n", PGCTRL);
3787                ret = -EINVAL;
3788                goto remove;
3789        }
3790
3791        for_each_online_cpu(cpu) {
3792                int err;
3793
3794                err = pktgen_create_thread(cpu, pn);
3795                if (err)
3796                        pr_warn("Cannot create thread for cpu %d (%d)\n",
3797                                   cpu, err);
3798        }
3799
3800        if (list_empty(&pn->pktgen_threads)) {
3801                pr_err("Initialization failed for all threads\n");
3802                ret = -ENODEV;
3803                goto remove_entry;
3804        }
3805
3806        return 0;
3807
3808remove_entry:
3809        remove_proc_entry(PGCTRL, pn->proc_dir);
3810remove:
3811        remove_proc_entry(PG_PROC_DIR, pn->net->proc_net);
3812        return ret;
3813}
3814
3815static void __net_exit pg_net_exit(struct net *net)
3816{
3817        struct pktgen_net *pn = net_generic(net, pg_net_id);
3818        struct pktgen_thread *t;
3819        struct list_head *q, *n;
3820        LIST_HEAD(list);
3821
3822        /* Stop all interfaces & threads */
3823        pn->pktgen_exiting = true;
3824
3825        mutex_lock(&pktgen_thread_lock);
3826        list_splice_init(&pn->pktgen_threads, &list);
3827        mutex_unlock(&pktgen_thread_lock);
3828
3829        list_for_each_safe(q, n, &list) {
3830                t = list_entry(q, struct pktgen_thread, th_list);
3831                list_del(&t->th_list);
3832                kthread_stop(t->tsk);
3833                put_task_struct(t->tsk);
3834                kfree(t);
3835        }
3836
3837        remove_proc_entry(PGCTRL, pn->proc_dir);
3838        remove_proc_entry(PG_PROC_DIR, pn->net->proc_net);
3839}
3840
3841static struct pernet_operations pg_net_ops = {
3842        .init = pg_net_init,
3843        .exit = pg_net_exit,
3844        .id   = &pg_net_id,
3845        .size = sizeof(struct pktgen_net),
3846};
3847
3848static int __init pg_init(void)
3849{
3850        int ret = 0;
3851
3852        pr_info("%s", version);
3853        ret = register_pernet_subsys(&pg_net_ops);
3854        if (ret)
3855                return ret;
3856        ret = register_netdevice_notifier(&pktgen_notifier_block);
3857        if (ret)
3858                unregister_pernet_subsys(&pg_net_ops);
3859
3860        return ret;
3861}
3862
3863static void __exit pg_cleanup(void)
3864{
3865        unregister_netdevice_notifier(&pktgen_notifier_block);
3866        unregister_pernet_subsys(&pg_net_ops);
3867        /* Don't need rcu_barrier() due to use of kfree_rcu() */
3868}
3869
3870module_init(pg_init);
3871module_exit(pg_cleanup);
3872
3873MODULE_AUTHOR("Robert Olsson <robert.olsson@its.uu.se>");
3874MODULE_DESCRIPTION("Packet Generator tool");
3875MODULE_LICENSE("GPL");
3876MODULE_VERSION(VERSION);
3877module_param(pg_count_d, int, 0);
3878MODULE_PARM_DESC(pg_count_d, "Default number of packets to inject");
3879module_param(pg_delay_d, int, 0);
3880MODULE_PARM_DESC(pg_delay_d, "Default delay between packets (nanoseconds)");
3881module_param(pg_clone_skb_d, int, 0);
3882MODULE_PARM_DESC(pg_clone_skb_d, "Default number of copies of the same packet");
3883module_param(debug, int, 0);
3884MODULE_PARM_DESC(debug, "Enable debugging of pktgen module");
3885