splice: fix problems with sys_tee()
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / core / pktgen.c
blobc23e9c06ee237203e792f421aabf6a3005996009
1 /*
2 * Authors:
3 * Copyright 2001, 2002 by Robert Olsson <robert.olsson@its.uu.se>
4 * Uppsala University and
5 * Swedish University of Agricultural Sciences
7 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
8 * Ben Greear <greearb@candelatech.com>
9 * Jens Låås <jens.laas@data.slu.se>
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License
13 * as published by the Free Software Foundation; either version
14 * 2 of the License, or (at your option) any later version.
17 * A tool for loading the network with preconfigurated packets.
18 * The tool is implemented as a linux module. Parameters are output
19 * device, delay (to hard_xmit), number of packets, and whether
20 * to use multiple SKBs or just the same one.
21 * pktgen uses the installed interface's output routine.
23 * Additional hacking by:
25 * Jens.Laas@data.slu.se
26 * Improved by ANK. 010120.
27 * Improved by ANK even more. 010212.
28 * MAC address typo fixed. 010417 --ro
29 * Integrated. 020301 --DaveM
30 * Added multiskb option 020301 --DaveM
31 * Scaling of results. 020417--sigurdur@linpro.no
32 * Significant re-work of the module:
33 * * Convert to threaded model to more efficiently be able to transmit
34 * and receive on multiple interfaces at once.
35 * * Converted many counters to __u64 to allow longer runs.
36 * * Allow configuration of ranges, like min/max IP address, MACs,
37 * and UDP-ports, for both source and destination, and can
38 * set to use a random distribution or sequentially walk the range.
39 * * Can now change most values after starting.
40 * * Place 12-byte packet in UDP payload with magic number,
41 * sequence number, and timestamp.
42 * * Add receiver code that detects dropped pkts, re-ordered pkts, and
43 * latencies (with micro-second) precision.
44 * * Add IOCTL interface to easily get counters & configuration.
45 * --Ben Greear <greearb@candelatech.com>
47 * Renamed multiskb to clone_skb and cleaned up sending core for two distinct
48 * skb modes. A clone_skb=0 mode for Ben "ranges" work and a clone_skb != 0
49 * as a "fastpath" with a configurable number of clones after alloc's.
50 * clone_skb=0 means all packets are allocated this also means ranges time
51 * stamps etc can be used. clone_skb=100 means 1 malloc is followed by 100
52 * clones.
54 * Also moved to /proc/net/pktgen/
55 * --ro
57 * Sept 10: Fixed threading/locking. Lots of bone-headed and more clever
58 * mistakes. Also merged in DaveM's patch in the -pre6 patch.
59 * --Ben Greear <greearb@candelatech.com>
61 * Integrated to 2.5.x 021029 --Lucio Maciel (luciomaciel@zipmail.com.br)
64 * 021124 Finished major redesign and rewrite for new functionality.
65 * See Documentation/networking/pktgen.txt for how to use this.
67 * The new operation:
68 * For each CPU one thread/process is created at start. This process checks
69 * for running devices in the if_list and sends packets until count is 0 it
70 * also the thread checks the thread->control which is used for inter-process
71 * communication. controlling process "posts" operations to the threads this
72 * way. The if_lock should be possible to remove when add/rem_device is merged
73 * into this too.
75 * By design there should only be *one* "controlling" process. In practice
76 * multiple write accesses gives unpredictable result. Understood by "write"
77 * to /proc gives result code thats should be read be the "writer".
78 * For practical use this should be no problem.
80 * Note when adding devices to a specific CPU there good idea to also assign
81 * /proc/irq/XX/smp_affinity so TX-interrupts gets bound to the same CPU.
82 * --ro
84 * Fix refcount off by one if first packet fails, potential null deref,
85 * memleak 030710- KJP
87 * First "ranges" functionality for ipv6 030726 --ro
89 * Included flow support. 030802 ANK.
91 * Fixed unaligned access on IA-64 Grant Grundler <grundler@parisc-linux.org>
93 * Remove if fix from added Harald Welte <laforge@netfilter.org> 040419
94 * ia64 compilation fix from Aron Griffis <aron@hp.com> 040604
96 * New xmit() return, do_div and misc clean up by Stephen Hemminger
97 * <shemminger@osdl.org> 040923
99 * Randy Dunlap fixed u64 printk compiler waring
101 * Remove FCS from BW calculation. Lennert Buytenhek <buytenh@wantstofly.org>
102 * New time handling. Lennert Buytenhek <buytenh@wantstofly.org> 041213
104 * Corrections from Nikolai Malykh (nmalykh@bilim.com)
105 * Removed unused flags F_SET_SRCMAC & F_SET_SRCIP 041230
107 * interruptible_sleep_on_timeout() replaced Nishanth Aravamudan <nacc@us.ibm.com>
108 * 050103
110 * MPLS support by Steven Whitehouse <steve@chygwyn.com>
113 #include <linux/sys.h>
114 #include <linux/types.h>
115 #include <linux/module.h>
116 #include <linux/moduleparam.h>
117 #include <linux/kernel.h>
118 #include <linux/smp_lock.h>
119 #include <linux/mutex.h>
120 #include <linux/sched.h>
121 #include <linux/slab.h>
122 #include <linux/vmalloc.h>
123 #include <linux/unistd.h>
124 #include <linux/string.h>
125 #include <linux/ptrace.h>
126 #include <linux/errno.h>
127 #include <linux/ioport.h>
128 #include <linux/interrupt.h>
129 #include <linux/capability.h>
130 #include <linux/delay.h>
131 #include <linux/timer.h>
132 #include <linux/list.h>
133 #include <linux/init.h>
134 #include <linux/skbuff.h>
135 #include <linux/netdevice.h>
136 #include <linux/inet.h>
137 #include <linux/inetdevice.h>
138 #include <linux/rtnetlink.h>
139 #include <linux/if_arp.h>
140 #include <linux/in.h>
141 #include <linux/ip.h>
142 #include <linux/ipv6.h>
143 #include <linux/udp.h>
144 #include <linux/proc_fs.h>
145 #include <linux/seq_file.h>
146 #include <linux/wait.h>
147 #include <linux/etherdevice.h>
148 #include <net/checksum.h>
149 #include <net/ipv6.h>
150 #include <net/addrconf.h>
151 #include <asm/byteorder.h>
152 #include <linux/rcupdate.h>
153 #include <asm/bitops.h>
154 #include <asm/io.h>
155 #include <asm/dma.h>
156 #include <asm/uaccess.h>
157 #include <asm/div64.h> /* do_div */
158 #include <asm/timex.h>
160 #define VERSION "pktgen v2.67: Packet Generator for packet performance testing.\n"
162 /* #define PG_DEBUG(a) a */
163 #define PG_DEBUG(a)
165 /* The buckets are exponential in 'width' */
166 #define LAT_BUCKETS_MAX 32
167 #define IP_NAME_SZ 32
168 #define MAX_MPLS_LABELS 16 /* This is the max label stack depth */
169 #define MPLS_STACK_BOTTOM __constant_htonl(0x00000100)
171 /* Device flag bits */
172 #define F_IPSRC_RND (1<<0) /* IP-Src Random */
173 #define F_IPDST_RND (1<<1) /* IP-Dst Random */
174 #define F_UDPSRC_RND (1<<2) /* UDP-Src Random */
175 #define F_UDPDST_RND (1<<3) /* UDP-Dst Random */
176 #define F_MACSRC_RND (1<<4) /* MAC-Src Random */
177 #define F_MACDST_RND (1<<5) /* MAC-Dst Random */
178 #define F_TXSIZE_RND (1<<6) /* Transmit size is random */
179 #define F_IPV6 (1<<7) /* Interface in IPV6 Mode */
180 #define F_MPLS_RND (1<<8) /* Random MPLS labels */
182 /* Thread control flag bits */
183 #define T_TERMINATE (1<<0)
184 #define T_STOP (1<<1) /* Stop run */
185 #define T_RUN (1<<2) /* Start run */
186 #define T_REMDEVALL (1<<3) /* Remove all devs */
187 #define T_REMDEV (1<<4) /* Remove one dev */
189 /* If lock -- can be removed after some work */
190 #define if_lock(t) spin_lock(&(t->if_lock));
191 #define if_unlock(t) spin_unlock(&(t->if_lock));
193 /* Used to help with determining the pkts on receive */
194 #define PKTGEN_MAGIC 0xbe9be955
195 #define PG_PROC_DIR "pktgen"
196 #define PGCTRL "pgctrl"
197 static struct proc_dir_entry *pg_proc_dir = NULL;
199 #define MAX_CFLOWS 65536
201 struct flow_state {
202 __u32 cur_daddr;
203 int count;
206 struct pktgen_dev {
209 * Try to keep frequent/infrequent used vars. separated.
212 char ifname[IFNAMSIZ];
213 char result[512];
215 struct pktgen_thread *pg_thread; /* the owner */
216 struct list_head list; /* Used for chaining in the thread's run-queue */
218 int running; /* if this changes to false, the test will stop */
220 /* If min != max, then we will either do a linear iteration, or
221 * we will do a random selection from within the range.
223 __u32 flags;
224 int removal_mark; /* non-zero => the device is marked for
225 * removal by worker thread */
227 int min_pkt_size; /* = ETH_ZLEN; */
228 int max_pkt_size; /* = ETH_ZLEN; */
229 int nfrags;
230 __u32 delay_us; /* Default delay */
231 __u32 delay_ns;
232 __u64 count; /* Default No packets to send */
233 __u64 sofar; /* How many pkts we've sent so far */
234 __u64 tx_bytes; /* How many bytes we've transmitted */
235 __u64 errors; /* Errors when trying to transmit, pkts will be re-sent */
237 /* runtime counters relating to clone_skb */
238 __u64 next_tx_us; /* timestamp of when to tx next */
239 __u32 next_tx_ns;
241 __u64 allocated_skbs;
242 __u32 clone_count;
243 int last_ok; /* Was last skb sent?
244 * Or a failed transmit of some sort? This will keep
245 * sequence numbers in order, for example.
247 __u64 started_at; /* micro-seconds */
248 __u64 stopped_at; /* micro-seconds */
249 __u64 idle_acc; /* micro-seconds */
250 __u32 seq_num;
252 int clone_skb; /* Use multiple SKBs during packet gen. If this number
253 * is greater than 1, then that many copies of the same
254 * packet will be sent before a new packet is allocated.
255 * For instance, if you want to send 1024 identical packets
256 * before creating a new packet, set clone_skb to 1024.
259 char dst_min[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
260 char dst_max[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
261 char src_min[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
262 char src_max[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
264 struct in6_addr in6_saddr;
265 struct in6_addr in6_daddr;
266 struct in6_addr cur_in6_daddr;
267 struct in6_addr cur_in6_saddr;
268 /* For ranges */
269 struct in6_addr min_in6_daddr;
270 struct in6_addr max_in6_daddr;
271 struct in6_addr min_in6_saddr;
272 struct in6_addr max_in6_saddr;
274 /* If we're doing ranges, random or incremental, then this
275 * defines the min/max for those ranges.
277 __u32 saddr_min; /* inclusive, source IP address */
278 __u32 saddr_max; /* exclusive, source IP address */
279 __u32 daddr_min; /* inclusive, dest IP address */
280 __u32 daddr_max; /* exclusive, dest IP address */
282 __u16 udp_src_min; /* inclusive, source UDP port */
283 __u16 udp_src_max; /* exclusive, source UDP port */
284 __u16 udp_dst_min; /* inclusive, dest UDP port */
285 __u16 udp_dst_max; /* exclusive, dest UDP port */
287 /* MPLS */
288 unsigned nr_labels; /* Depth of stack, 0 = no MPLS */
289 __be32 labels[MAX_MPLS_LABELS];
291 __u32 src_mac_count; /* How many MACs to iterate through */
292 __u32 dst_mac_count; /* How many MACs to iterate through */
294 unsigned char dst_mac[ETH_ALEN];
295 unsigned char src_mac[ETH_ALEN];
297 __u32 cur_dst_mac_offset;
298 __u32 cur_src_mac_offset;
299 __u32 cur_saddr;
300 __u32 cur_daddr;
301 __u16 cur_udp_dst;
302 __u16 cur_udp_src;
303 __u32 cur_pkt_size;
305 __u8 hh[14];
306 /* = {
307 0x00, 0x80, 0xC8, 0x79, 0xB3, 0xCB,
309 We fill in SRC address later
310 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
311 0x08, 0x00
314 __u16 pad; /* pad out the hh struct to an even 16 bytes */
316 struct sk_buff *skb; /* skb we are to transmit next, mainly used for when we
317 * are transmitting the same one multiple times
319 struct net_device *odev; /* The out-going device. Note that the device should
320 * have it's pg_info pointer pointing back to this
321 * device. This will be set when the user specifies
322 * the out-going device name (not when the inject is
323 * started as it used to do.)
325 struct flow_state *flows;
326 unsigned cflows; /* Concurrent flows (config) */
327 unsigned lflow; /* Flow length (config) */
328 unsigned nflows; /* accumulated flows (stats) */
331 struct pktgen_hdr {
332 __u32 pgh_magic;
333 __u32 seq_num;
334 __u32 tv_sec;
335 __u32 tv_usec;
338 struct pktgen_thread {
339 spinlock_t if_lock;
340 struct list_head if_list; /* All device here */
341 struct list_head th_list;
342 int removed;
343 char name[32];
344 char result[512];
345 u32 max_before_softirq; /* We'll call do_softirq to prevent starvation. */
347 /* Field for thread to receive "posted" events terminate, stop ifs etc. */
349 u32 control;
350 int pid;
351 int cpu;
353 wait_queue_head_t queue;
356 #define REMOVE 1
357 #define FIND 0
359 /* This code works around the fact that do_div cannot handle two 64-bit
360 numbers, and regular 64-bit division doesn't work on x86 kernels.
361 --Ben
364 #define PG_DIV 0
366 /* This was emailed to LMKL by: Chris Caputo <ccaputo@alt.net>
367 * Function copied/adapted/optimized from:
369 * nemesis.sourceforge.net/browse/lib/static/intmath/ix86/intmath.c.html
371 * Copyright 1994, University of Cambridge Computer Laboratory
372 * All Rights Reserved.
375 static inline s64 divremdi3(s64 x, s64 y, int type)
377 u64 a = (x < 0) ? -x : x;
378 u64 b = (y < 0) ? -y : y;
379 u64 res = 0, d = 1;
381 if (b > 0) {
382 while (b < a) {
383 b <<= 1;
384 d <<= 1;
388 do {
389 if (a >= b) {
390 a -= b;
391 res += d;
393 b >>= 1;
394 d >>= 1;
396 while (d);
398 if (PG_DIV == type) {
399 return (((x ^ y) & (1ll << 63)) == 0) ? res : -(s64) res;
400 } else {
401 return ((x & (1ll << 63)) == 0) ? a : -(s64) a;
405 /* End of hacks to deal with 64-bit math on x86 */
407 /** Convert to milliseconds */
408 static inline __u64 tv_to_ms(const struct timeval *tv)
410 __u64 ms = tv->tv_usec / 1000;
411 ms += (__u64) tv->tv_sec * (__u64) 1000;
412 return ms;
415 /** Convert to micro-seconds */
416 static inline __u64 tv_to_us(const struct timeval *tv)
418 __u64 us = tv->tv_usec;
419 us += (__u64) tv->tv_sec * (__u64) 1000000;
420 return us;
423 static inline __u64 pg_div(__u64 n, __u32 base)
425 __u64 tmp = n;
426 do_div(tmp, base);
427 /* printk("pktgen: pg_div, n: %llu base: %d rv: %llu\n",
428 n, base, tmp); */
429 return tmp;
432 static inline __u64 pg_div64(__u64 n, __u64 base)
434 __u64 tmp = n;
436 * How do we know if the architecture we are running on
437 * supports division with 64 bit base?
440 #if defined(__sparc_v9__) || defined(__powerpc64__) || defined(__alpha__) || defined(__x86_64__) || defined(__ia64__)
442 do_div(tmp, base);
443 #else
444 tmp = divremdi3(n, base, PG_DIV);
445 #endif
446 return tmp;
449 static inline u32 pktgen_random(void)
451 #if 0
452 __u32 n;
453 get_random_bytes(&n, 4);
454 return n;
455 #else
456 return net_random();
457 #endif
460 static inline __u64 getCurMs(void)
462 struct timeval tv;
463 do_gettimeofday(&tv);
464 return tv_to_ms(&tv);
467 static inline __u64 getCurUs(void)
469 struct timeval tv;
470 do_gettimeofday(&tv);
471 return tv_to_us(&tv);
474 static inline __u64 tv_diff(const struct timeval *a, const struct timeval *b)
476 return tv_to_us(a) - tv_to_us(b);
479 /* old include end */
481 static char version[] __initdata = VERSION;
483 static int pktgen_remove_device(struct pktgen_thread *t, struct pktgen_dev *i);
484 static int pktgen_add_device(struct pktgen_thread *t, const char *ifname);
485 static struct pktgen_dev *pktgen_find_dev(struct pktgen_thread *t,
486 const char *ifname);
487 static int pktgen_device_event(struct notifier_block *, unsigned long, void *);
488 static void pktgen_run_all_threads(void);
489 static void pktgen_stop_all_threads_ifs(void);
490 static int pktgen_stop_device(struct pktgen_dev *pkt_dev);
491 static void pktgen_stop(struct pktgen_thread *t);
492 static void pktgen_clear_counters(struct pktgen_dev *pkt_dev);
493 static int pktgen_mark_device(const char *ifname);
494 static unsigned int scan_ip6(const char *s, char ip[16]);
495 static unsigned int fmt_ip6(char *s, const char ip[16]);
497 /* Module parameters, defaults. */
498 static int pg_count_d = 1000; /* 1000 pkts by default */
499 static int pg_delay_d;
500 static int pg_clone_skb_d;
501 static int debug;
503 static DEFINE_MUTEX(pktgen_thread_lock);
504 static LIST_HEAD(pktgen_threads);
506 static struct notifier_block pktgen_notifier_block = {
507 .notifier_call = pktgen_device_event,
511 * /proc handling functions
515 static int pgctrl_show(struct seq_file *seq, void *v)
517 seq_puts(seq, VERSION);
518 return 0;
521 static ssize_t pgctrl_write(struct file *file, const char __user * buf,
522 size_t count, loff_t * ppos)
524 int err = 0;
525 char data[128];
527 if (!capable(CAP_NET_ADMIN)) {
528 err = -EPERM;
529 goto out;
532 if (count > sizeof(data))
533 count = sizeof(data);
535 if (copy_from_user(data, buf, count)) {
536 err = -EFAULT;
537 goto out;
539 data[count - 1] = 0; /* Make string */
541 if (!strcmp(data, "stop"))
542 pktgen_stop_all_threads_ifs();
544 else if (!strcmp(data, "start"))
545 pktgen_run_all_threads();
547 else
548 printk("pktgen: Unknown command: %s\n", data);
550 err = count;
552 out:
553 return err;
556 static int pgctrl_open(struct inode *inode, struct file *file)
558 return single_open(file, pgctrl_show, PDE(inode)->data);
561 static struct file_operations pktgen_fops = {
562 .owner = THIS_MODULE,
563 .open = pgctrl_open,
564 .read = seq_read,
565 .llseek = seq_lseek,
566 .write = pgctrl_write,
567 .release = single_release,
570 static int pktgen_if_show(struct seq_file *seq, void *v)
572 int i;
573 struct pktgen_dev *pkt_dev = seq->private;
574 __u64 sa;
575 __u64 stopped;
576 __u64 now = getCurUs();
578 seq_printf(seq,
579 "Params: count %llu min_pkt_size: %u max_pkt_size: %u\n",
580 (unsigned long long)pkt_dev->count, pkt_dev->min_pkt_size,
581 pkt_dev->max_pkt_size);
583 seq_printf(seq,
584 " frags: %d delay: %u clone_skb: %d ifname: %s\n",
585 pkt_dev->nfrags,
586 1000 * pkt_dev->delay_us + pkt_dev->delay_ns,
587 pkt_dev->clone_skb, pkt_dev->ifname);
589 seq_printf(seq, " flows: %u flowlen: %u\n", pkt_dev->cflows,
590 pkt_dev->lflow);
592 if (pkt_dev->flags & F_IPV6) {
593 char b1[128], b2[128], b3[128];
594 fmt_ip6(b1, pkt_dev->in6_saddr.s6_addr);
595 fmt_ip6(b2, pkt_dev->min_in6_saddr.s6_addr);
596 fmt_ip6(b3, pkt_dev->max_in6_saddr.s6_addr);
597 seq_printf(seq,
598 " saddr: %s min_saddr: %s max_saddr: %s\n", b1,
599 b2, b3);
601 fmt_ip6(b1, pkt_dev->in6_daddr.s6_addr);
602 fmt_ip6(b2, pkt_dev->min_in6_daddr.s6_addr);
603 fmt_ip6(b3, pkt_dev->max_in6_daddr.s6_addr);
604 seq_printf(seq,
605 " daddr: %s min_daddr: %s max_daddr: %s\n", b1,
606 b2, b3);
608 } else
609 seq_printf(seq,
610 " dst_min: %s dst_max: %s\n src_min: %s src_max: %s\n",
611 pkt_dev->dst_min, pkt_dev->dst_max, pkt_dev->src_min,
612 pkt_dev->src_max);
614 seq_puts(seq, " src_mac: ");
616 if (is_zero_ether_addr(pkt_dev->src_mac))
617 for (i = 0; i < 6; i++)
618 seq_printf(seq, "%02X%s", pkt_dev->odev->dev_addr[i],
619 i == 5 ? " " : ":");
620 else
621 for (i = 0; i < 6; i++)
622 seq_printf(seq, "%02X%s", pkt_dev->src_mac[i],
623 i == 5 ? " " : ":");
625 seq_printf(seq, "dst_mac: ");
626 for (i = 0; i < 6; i++)
627 seq_printf(seq, "%02X%s", pkt_dev->dst_mac[i],
628 i == 5 ? "\n" : ":");
630 seq_printf(seq,
631 " udp_src_min: %d udp_src_max: %d udp_dst_min: %d udp_dst_max: %d\n",
632 pkt_dev->udp_src_min, pkt_dev->udp_src_max,
633 pkt_dev->udp_dst_min, pkt_dev->udp_dst_max);
635 seq_printf(seq,
636 " src_mac_count: %d dst_mac_count: %d\n",
637 pkt_dev->src_mac_count, pkt_dev->dst_mac_count);
639 if (pkt_dev->nr_labels) {
640 unsigned i;
641 seq_printf(seq, " mpls: ");
642 for(i = 0; i < pkt_dev->nr_labels; i++)
643 seq_printf(seq, "%08x%s", ntohl(pkt_dev->labels[i]),
644 i == pkt_dev->nr_labels-1 ? "\n" : ", ");
647 seq_printf(seq, " Flags: ");
649 if (pkt_dev->flags & F_IPV6)
650 seq_printf(seq, "IPV6 ");
652 if (pkt_dev->flags & F_IPSRC_RND)
653 seq_printf(seq, "IPSRC_RND ");
655 if (pkt_dev->flags & F_IPDST_RND)
656 seq_printf(seq, "IPDST_RND ");
658 if (pkt_dev->flags & F_TXSIZE_RND)
659 seq_printf(seq, "TXSIZE_RND ");
661 if (pkt_dev->flags & F_UDPSRC_RND)
662 seq_printf(seq, "UDPSRC_RND ");
664 if (pkt_dev->flags & F_UDPDST_RND)
665 seq_printf(seq, "UDPDST_RND ");
667 if (pkt_dev->flags & F_MPLS_RND)
668 seq_printf(seq, "MPLS_RND ");
670 if (pkt_dev->flags & F_MACSRC_RND)
671 seq_printf(seq, "MACSRC_RND ");
673 if (pkt_dev->flags & F_MACDST_RND)
674 seq_printf(seq, "MACDST_RND ");
676 seq_puts(seq, "\n");
678 sa = pkt_dev->started_at;
679 stopped = pkt_dev->stopped_at;
680 if (pkt_dev->running)
681 stopped = now; /* not really stopped, more like last-running-at */
683 seq_printf(seq,
684 "Current:\n pkts-sofar: %llu errors: %llu\n started: %lluus stopped: %lluus idle: %lluus\n",
685 (unsigned long long)pkt_dev->sofar,
686 (unsigned long long)pkt_dev->errors, (unsigned long long)sa,
687 (unsigned long long)stopped,
688 (unsigned long long)pkt_dev->idle_acc);
690 seq_printf(seq,
691 " seq_num: %d cur_dst_mac_offset: %d cur_src_mac_offset: %d\n",
692 pkt_dev->seq_num, pkt_dev->cur_dst_mac_offset,
693 pkt_dev->cur_src_mac_offset);
695 if (pkt_dev->flags & F_IPV6) {
696 char b1[128], b2[128];
697 fmt_ip6(b1, pkt_dev->cur_in6_daddr.s6_addr);
698 fmt_ip6(b2, pkt_dev->cur_in6_saddr.s6_addr);
699 seq_printf(seq, " cur_saddr: %s cur_daddr: %s\n", b2, b1);
700 } else
701 seq_printf(seq, " cur_saddr: 0x%x cur_daddr: 0x%x\n",
702 pkt_dev->cur_saddr, pkt_dev->cur_daddr);
704 seq_printf(seq, " cur_udp_dst: %d cur_udp_src: %d\n",
705 pkt_dev->cur_udp_dst, pkt_dev->cur_udp_src);
707 seq_printf(seq, " flows: %u\n", pkt_dev->nflows);
709 if (pkt_dev->result[0])
710 seq_printf(seq, "Result: %s\n", pkt_dev->result);
711 else
712 seq_printf(seq, "Result: Idle\n");
714 return 0;
718 static int hex32_arg(const char __user *user_buffer, __u32 *num)
720 int i = 0;
721 *num = 0;
723 for(; i < 8; i++) {
724 char c;
725 *num <<= 4;
726 if (get_user(c, &user_buffer[i]))
727 return -EFAULT;
728 if ((c >= '0') && (c <= '9'))
729 *num |= c - '0';
730 else if ((c >= 'a') && (c <= 'f'))
731 *num |= c - 'a' + 10;
732 else if ((c >= 'A') && (c <= 'F'))
733 *num |= c - 'A' + 10;
734 else
735 break;
737 return i;
740 static int count_trail_chars(const char __user * user_buffer,
741 unsigned int maxlen)
743 int i;
745 for (i = 0; i < maxlen; i++) {
746 char c;
747 if (get_user(c, &user_buffer[i]))
748 return -EFAULT;
749 switch (c) {
750 case '\"':
751 case '\n':
752 case '\r':
753 case '\t':
754 case ' ':
755 case '=':
756 break;
757 default:
758 goto done;
761 done:
762 return i;
765 static unsigned long num_arg(const char __user * user_buffer,
766 unsigned long maxlen, unsigned long *num)
768 int i = 0;
769 *num = 0;
771 for (; i < maxlen; i++) {
772 char c;
773 if (get_user(c, &user_buffer[i]))
774 return -EFAULT;
775 if ((c >= '0') && (c <= '9')) {
776 *num *= 10;
777 *num += c - '0';
778 } else
779 break;
781 return i;
784 static int strn_len(const char __user * user_buffer, unsigned int maxlen)
786 int i = 0;
788 for (; i < maxlen; i++) {
789 char c;
790 if (get_user(c, &user_buffer[i]))
791 return -EFAULT;
792 switch (c) {
793 case '\"':
794 case '\n':
795 case '\r':
796 case '\t':
797 case ' ':
798 goto done_str;
799 break;
800 default:
801 break;
804 done_str:
805 return i;
808 static ssize_t get_labels(const char __user *buffer, struct pktgen_dev *pkt_dev)
810 unsigned n = 0;
811 char c;
812 ssize_t i = 0;
813 int len;
815 pkt_dev->nr_labels = 0;
816 do {
817 __u32 tmp;
818 len = hex32_arg(&buffer[i], &tmp);
819 if (len <= 0)
820 return len;
821 pkt_dev->labels[n] = htonl(tmp);
822 if (pkt_dev->labels[n] & MPLS_STACK_BOTTOM)
823 pkt_dev->flags |= F_MPLS_RND;
824 i += len;
825 if (get_user(c, &buffer[i]))
826 return -EFAULT;
827 i++;
828 n++;
829 if (n >= MAX_MPLS_LABELS)
830 return -E2BIG;
831 } while(c == ',');
833 pkt_dev->nr_labels = n;
834 return i;
837 static ssize_t pktgen_if_write(struct file *file,
838 const char __user * user_buffer, size_t count,
839 loff_t * offset)
841 struct seq_file *seq = (struct seq_file *)file->private_data;
842 struct pktgen_dev *pkt_dev = seq->private;
843 int i = 0, max, len;
844 char name[16], valstr[32];
845 unsigned long value = 0;
846 char *pg_result = NULL;
847 int tmp = 0;
848 char buf[128];
850 pg_result = &(pkt_dev->result[0]);
852 if (count < 1) {
853 printk("pktgen: wrong command format\n");
854 return -EINVAL;
857 max = count - i;
858 tmp = count_trail_chars(&user_buffer[i], max);
859 if (tmp < 0) {
860 printk("pktgen: illegal format\n");
861 return tmp;
863 i += tmp;
865 /* Read variable name */
867 len = strn_len(&user_buffer[i], sizeof(name) - 1);
868 if (len < 0) {
869 return len;
871 memset(name, 0, sizeof(name));
872 if (copy_from_user(name, &user_buffer[i], len))
873 return -EFAULT;
874 i += len;
876 max = count - i;
877 len = count_trail_chars(&user_buffer[i], max);
878 if (len < 0)
879 return len;
881 i += len;
883 if (debug) {
884 char tb[count + 1];
885 if (copy_from_user(tb, user_buffer, count))
886 return -EFAULT;
887 tb[count] = 0;
888 printk("pktgen: %s,%lu buffer -:%s:-\n", name,
889 (unsigned long)count, tb);
892 if (!strcmp(name, "min_pkt_size")) {
893 len = num_arg(&user_buffer[i], 10, &value);
894 if (len < 0) {
895 return len;
897 i += len;
898 if (value < 14 + 20 + 8)
899 value = 14 + 20 + 8;
900 if (value != pkt_dev->min_pkt_size) {
901 pkt_dev->min_pkt_size = value;
902 pkt_dev->cur_pkt_size = value;
904 sprintf(pg_result, "OK: min_pkt_size=%u",
905 pkt_dev->min_pkt_size);
906 return count;
909 if (!strcmp(name, "max_pkt_size")) {
910 len = num_arg(&user_buffer[i], 10, &value);
911 if (len < 0) {
912 return len;
914 i += len;
915 if (value < 14 + 20 + 8)
916 value = 14 + 20 + 8;
917 if (value != pkt_dev->max_pkt_size) {
918 pkt_dev->max_pkt_size = value;
919 pkt_dev->cur_pkt_size = value;
921 sprintf(pg_result, "OK: max_pkt_size=%u",
922 pkt_dev->max_pkt_size);
923 return count;
926 /* Shortcut for min = max */
928 if (!strcmp(name, "pkt_size")) {
929 len = num_arg(&user_buffer[i], 10, &value);
930 if (len < 0) {
931 return len;
933 i += len;
934 if (value < 14 + 20 + 8)
935 value = 14 + 20 + 8;
936 if (value != pkt_dev->min_pkt_size) {
937 pkt_dev->min_pkt_size = value;
938 pkt_dev->max_pkt_size = value;
939 pkt_dev->cur_pkt_size = value;
941 sprintf(pg_result, "OK: pkt_size=%u", pkt_dev->min_pkt_size);
942 return count;
945 if (!strcmp(name, "debug")) {
946 len = num_arg(&user_buffer[i], 10, &value);
947 if (len < 0) {
948 return len;
950 i += len;
951 debug = value;
952 sprintf(pg_result, "OK: debug=%u", debug);
953 return count;
956 if (!strcmp(name, "frags")) {
957 len = num_arg(&user_buffer[i], 10, &value);
958 if (len < 0) {
959 return len;
961 i += len;
962 pkt_dev->nfrags = value;
963 sprintf(pg_result, "OK: frags=%u", pkt_dev->nfrags);
964 return count;
966 if (!strcmp(name, "delay")) {
967 len = num_arg(&user_buffer[i], 10, &value);
968 if (len < 0) {
969 return len;
971 i += len;
972 if (value == 0x7FFFFFFF) {
973 pkt_dev->delay_us = 0x7FFFFFFF;
974 pkt_dev->delay_ns = 0;
975 } else {
976 pkt_dev->delay_us = value / 1000;
977 pkt_dev->delay_ns = value % 1000;
979 sprintf(pg_result, "OK: delay=%u",
980 1000 * pkt_dev->delay_us + pkt_dev->delay_ns);
981 return count;
983 if (!strcmp(name, "udp_src_min")) {
984 len = num_arg(&user_buffer[i], 10, &value);
985 if (len < 0) {
986 return len;
988 i += len;
989 if (value != pkt_dev->udp_src_min) {
990 pkt_dev->udp_src_min = value;
991 pkt_dev->cur_udp_src = value;
993 sprintf(pg_result, "OK: udp_src_min=%u", pkt_dev->udp_src_min);
994 return count;
996 if (!strcmp(name, "udp_dst_min")) {
997 len = num_arg(&user_buffer[i], 10, &value);
998 if (len < 0) {
999 return len;
1001 i += len;
1002 if (value != pkt_dev->udp_dst_min) {
1003 pkt_dev->udp_dst_min = value;
1004 pkt_dev->cur_udp_dst = value;
1006 sprintf(pg_result, "OK: udp_dst_min=%u", pkt_dev->udp_dst_min);
1007 return count;
1009 if (!strcmp(name, "udp_src_max")) {
1010 len = num_arg(&user_buffer[i], 10, &value);
1011 if (len < 0) {
1012 return len;
1014 i += len;
1015 if (value != pkt_dev->udp_src_max) {
1016 pkt_dev->udp_src_max = value;
1017 pkt_dev->cur_udp_src = value;
1019 sprintf(pg_result, "OK: udp_src_max=%u", pkt_dev->udp_src_max);
1020 return count;
1022 if (!strcmp(name, "udp_dst_max")) {
1023 len = num_arg(&user_buffer[i], 10, &value);
1024 if (len < 0) {
1025 return len;
1027 i += len;
1028 if (value != pkt_dev->udp_dst_max) {
1029 pkt_dev->udp_dst_max = value;
1030 pkt_dev->cur_udp_dst = value;
1032 sprintf(pg_result, "OK: udp_dst_max=%u", pkt_dev->udp_dst_max);
1033 return count;
1035 if (!strcmp(name, "clone_skb")) {
1036 len = num_arg(&user_buffer[i], 10, &value);
1037 if (len < 0) {
1038 return len;
1040 i += len;
1041 pkt_dev->clone_skb = value;
1043 sprintf(pg_result, "OK: clone_skb=%d", pkt_dev->clone_skb);
1044 return count;
1046 if (!strcmp(name, "count")) {
1047 len = num_arg(&user_buffer[i], 10, &value);
1048 if (len < 0) {
1049 return len;
1051 i += len;
1052 pkt_dev->count = value;
1053 sprintf(pg_result, "OK: count=%llu",
1054 (unsigned long long)pkt_dev->count);
1055 return count;
1057 if (!strcmp(name, "src_mac_count")) {
1058 len = num_arg(&user_buffer[i], 10, &value);
1059 if (len < 0) {
1060 return len;
1062 i += len;
1063 if (pkt_dev->src_mac_count != value) {
1064 pkt_dev->src_mac_count = value;
1065 pkt_dev->cur_src_mac_offset = 0;
1067 sprintf(pg_result, "OK: src_mac_count=%d",
1068 pkt_dev->src_mac_count);
1069 return count;
1071 if (!strcmp(name, "dst_mac_count")) {
1072 len = num_arg(&user_buffer[i], 10, &value);
1073 if (len < 0) {
1074 return len;
1076 i += len;
1077 if (pkt_dev->dst_mac_count != value) {
1078 pkt_dev->dst_mac_count = value;
1079 pkt_dev->cur_dst_mac_offset = 0;
1081 sprintf(pg_result, "OK: dst_mac_count=%d",
1082 pkt_dev->dst_mac_count);
1083 return count;
1085 if (!strcmp(name, "flag")) {
1086 char f[32];
1087 memset(f, 0, 32);
1088 len = strn_len(&user_buffer[i], sizeof(f) - 1);
1089 if (len < 0) {
1090 return len;
1092 if (copy_from_user(f, &user_buffer[i], len))
1093 return -EFAULT;
1094 i += len;
1095 if (strcmp(f, "IPSRC_RND") == 0)
1096 pkt_dev->flags |= F_IPSRC_RND;
1098 else if (strcmp(f, "!IPSRC_RND") == 0)
1099 pkt_dev->flags &= ~F_IPSRC_RND;
1101 else if (strcmp(f, "TXSIZE_RND") == 0)
1102 pkt_dev->flags |= F_TXSIZE_RND;
1104 else if (strcmp(f, "!TXSIZE_RND") == 0)
1105 pkt_dev->flags &= ~F_TXSIZE_RND;
1107 else if (strcmp(f, "IPDST_RND") == 0)
1108 pkt_dev->flags |= F_IPDST_RND;
1110 else if (strcmp(f, "!IPDST_RND") == 0)
1111 pkt_dev->flags &= ~F_IPDST_RND;
1113 else if (strcmp(f, "UDPSRC_RND") == 0)
1114 pkt_dev->flags |= F_UDPSRC_RND;
1116 else if (strcmp(f, "!UDPSRC_RND") == 0)
1117 pkt_dev->flags &= ~F_UDPSRC_RND;
1119 else if (strcmp(f, "UDPDST_RND") == 0)
1120 pkt_dev->flags |= F_UDPDST_RND;
1122 else if (strcmp(f, "!UDPDST_RND") == 0)
1123 pkt_dev->flags &= ~F_UDPDST_RND;
1125 else if (strcmp(f, "MACSRC_RND") == 0)
1126 pkt_dev->flags |= F_MACSRC_RND;
1128 else if (strcmp(f, "!MACSRC_RND") == 0)
1129 pkt_dev->flags &= ~F_MACSRC_RND;
1131 else if (strcmp(f, "MACDST_RND") == 0)
1132 pkt_dev->flags |= F_MACDST_RND;
1134 else if (strcmp(f, "!MACDST_RND") == 0)
1135 pkt_dev->flags &= ~F_MACDST_RND;
1137 else if (strcmp(f, "MPLS_RND") == 0)
1138 pkt_dev->flags |= F_MPLS_RND;
1140 else if (strcmp(f, "!MPLS_RND") == 0)
1141 pkt_dev->flags &= ~F_MPLS_RND;
1143 else {
1144 sprintf(pg_result,
1145 "Flag -:%s:- unknown\nAvailable flags, (prepend ! to un-set flag):\n%s",
1147 "IPSRC_RND, IPDST_RND, TXSIZE_RND, UDPSRC_RND, UDPDST_RND, MACSRC_RND, MACDST_RND\n");
1148 return count;
1150 sprintf(pg_result, "OK: flags=0x%x", pkt_dev->flags);
1151 return count;
1153 if (!strcmp(name, "dst_min") || !strcmp(name, "dst")) {
1154 len = strn_len(&user_buffer[i], sizeof(pkt_dev->dst_min) - 1);
1155 if (len < 0) {
1156 return len;
1159 if (copy_from_user(buf, &user_buffer[i], len))
1160 return -EFAULT;
1161 buf[len] = 0;
1162 if (strcmp(buf, pkt_dev->dst_min) != 0) {
1163 memset(pkt_dev->dst_min, 0, sizeof(pkt_dev->dst_min));
1164 strncpy(pkt_dev->dst_min, buf, len);
1165 pkt_dev->daddr_min = in_aton(pkt_dev->dst_min);
1166 pkt_dev->cur_daddr = pkt_dev->daddr_min;
1168 if (debug)
1169 printk("pktgen: dst_min set to: %s\n",
1170 pkt_dev->dst_min);
1171 i += len;
1172 sprintf(pg_result, "OK: dst_min=%s", pkt_dev->dst_min);
1173 return count;
1175 if (!strcmp(name, "dst_max")) {
1176 len = strn_len(&user_buffer[i], sizeof(pkt_dev->dst_max) - 1);
1177 if (len < 0) {
1178 return len;
1181 if (copy_from_user(buf, &user_buffer[i], len))
1182 return -EFAULT;
1184 buf[len] = 0;
1185 if (strcmp(buf, pkt_dev->dst_max) != 0) {
1186 memset(pkt_dev->dst_max, 0, sizeof(pkt_dev->dst_max));
1187 strncpy(pkt_dev->dst_max, buf, len);
1188 pkt_dev->daddr_max = in_aton(pkt_dev->dst_max);
1189 pkt_dev->cur_daddr = pkt_dev->daddr_max;
1191 if (debug)
1192 printk("pktgen: dst_max set to: %s\n",
1193 pkt_dev->dst_max);
1194 i += len;
1195 sprintf(pg_result, "OK: dst_max=%s", pkt_dev->dst_max);
1196 return count;
1198 if (!strcmp(name, "dst6")) {
1199 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1200 if (len < 0)
1201 return len;
1203 pkt_dev->flags |= F_IPV6;
1205 if (copy_from_user(buf, &user_buffer[i], len))
1206 return -EFAULT;
1207 buf[len] = 0;
1209 scan_ip6(buf, pkt_dev->in6_daddr.s6_addr);
1210 fmt_ip6(buf, pkt_dev->in6_daddr.s6_addr);
1212 ipv6_addr_copy(&pkt_dev->cur_in6_daddr, &pkt_dev->in6_daddr);
1214 if (debug)
1215 printk("pktgen: dst6 set to: %s\n", buf);
1217 i += len;
1218 sprintf(pg_result, "OK: dst6=%s", buf);
1219 return count;
1221 if (!strcmp(name, "dst6_min")) {
1222 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1223 if (len < 0)
1224 return len;
1226 pkt_dev->flags |= F_IPV6;
1228 if (copy_from_user(buf, &user_buffer[i], len))
1229 return -EFAULT;
1230 buf[len] = 0;
1232 scan_ip6(buf, pkt_dev->min_in6_daddr.s6_addr);
1233 fmt_ip6(buf, pkt_dev->min_in6_daddr.s6_addr);
1235 ipv6_addr_copy(&pkt_dev->cur_in6_daddr,
1236 &pkt_dev->min_in6_daddr);
1237 if (debug)
1238 printk("pktgen: dst6_min set to: %s\n", buf);
1240 i += len;
1241 sprintf(pg_result, "OK: dst6_min=%s", buf);
1242 return count;
1244 if (!strcmp(name, "dst6_max")) {
1245 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1246 if (len < 0)
1247 return len;
1249 pkt_dev->flags |= F_IPV6;
1251 if (copy_from_user(buf, &user_buffer[i], len))
1252 return -EFAULT;
1253 buf[len] = 0;
1255 scan_ip6(buf, pkt_dev->max_in6_daddr.s6_addr);
1256 fmt_ip6(buf, pkt_dev->max_in6_daddr.s6_addr);
1258 if (debug)
1259 printk("pktgen: dst6_max set to: %s\n", buf);
1261 i += len;
1262 sprintf(pg_result, "OK: dst6_max=%s", buf);
1263 return count;
1265 if (!strcmp(name, "src6")) {
1266 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1267 if (len < 0)
1268 return len;
1270 pkt_dev->flags |= F_IPV6;
1272 if (copy_from_user(buf, &user_buffer[i], len))
1273 return -EFAULT;
1274 buf[len] = 0;
1276 scan_ip6(buf, pkt_dev->in6_saddr.s6_addr);
1277 fmt_ip6(buf, pkt_dev->in6_saddr.s6_addr);
1279 ipv6_addr_copy(&pkt_dev->cur_in6_saddr, &pkt_dev->in6_saddr);
1281 if (debug)
1282 printk("pktgen: src6 set to: %s\n", buf);
1284 i += len;
1285 sprintf(pg_result, "OK: src6=%s", buf);
1286 return count;
1288 if (!strcmp(name, "src_min")) {
1289 len = strn_len(&user_buffer[i], sizeof(pkt_dev->src_min) - 1);
1290 if (len < 0) {
1291 return len;
1293 if (copy_from_user(buf, &user_buffer[i], len))
1294 return -EFAULT;
1295 buf[len] = 0;
1296 if (strcmp(buf, pkt_dev->src_min) != 0) {
1297 memset(pkt_dev->src_min, 0, sizeof(pkt_dev->src_min));
1298 strncpy(pkt_dev->src_min, buf, len);
1299 pkt_dev->saddr_min = in_aton(pkt_dev->src_min);
1300 pkt_dev->cur_saddr = pkt_dev->saddr_min;
1302 if (debug)
1303 printk("pktgen: src_min set to: %s\n",
1304 pkt_dev->src_min);
1305 i += len;
1306 sprintf(pg_result, "OK: src_min=%s", pkt_dev->src_min);
1307 return count;
1309 if (!strcmp(name, "src_max")) {
1310 len = strn_len(&user_buffer[i], sizeof(pkt_dev->src_max) - 1);
1311 if (len < 0) {
1312 return len;
1314 if (copy_from_user(buf, &user_buffer[i], len))
1315 return -EFAULT;
1316 buf[len] = 0;
1317 if (strcmp(buf, pkt_dev->src_max) != 0) {
1318 memset(pkt_dev->src_max, 0, sizeof(pkt_dev->src_max));
1319 strncpy(pkt_dev->src_max, buf, len);
1320 pkt_dev->saddr_max = in_aton(pkt_dev->src_max);
1321 pkt_dev->cur_saddr = pkt_dev->saddr_max;
1323 if (debug)
1324 printk("pktgen: src_max set to: %s\n",
1325 pkt_dev->src_max);
1326 i += len;
1327 sprintf(pg_result, "OK: src_max=%s", pkt_dev->src_max);
1328 return count;
1330 if (!strcmp(name, "dst_mac")) {
1331 char *v = valstr;
1332 unsigned char old_dmac[ETH_ALEN];
1333 unsigned char *m = pkt_dev->dst_mac;
1334 memcpy(old_dmac, pkt_dev->dst_mac, ETH_ALEN);
1336 len = strn_len(&user_buffer[i], sizeof(valstr) - 1);
1337 if (len < 0) {
1338 return len;
1340 memset(valstr, 0, sizeof(valstr));
1341 if (copy_from_user(valstr, &user_buffer[i], len))
1342 return -EFAULT;
1343 i += len;
1345 for (*m = 0; *v && m < pkt_dev->dst_mac + 6; v++) {
1346 if (*v >= '0' && *v <= '9') {
1347 *m *= 16;
1348 *m += *v - '0';
1350 if (*v >= 'A' && *v <= 'F') {
1351 *m *= 16;
1352 *m += *v - 'A' + 10;
1354 if (*v >= 'a' && *v <= 'f') {
1355 *m *= 16;
1356 *m += *v - 'a' + 10;
1358 if (*v == ':') {
1359 m++;
1360 *m = 0;
1364 /* Set up Dest MAC */
1365 if (compare_ether_addr(old_dmac, pkt_dev->dst_mac))
1366 memcpy(&(pkt_dev->hh[0]), pkt_dev->dst_mac, ETH_ALEN);
1368 sprintf(pg_result, "OK: dstmac");
1369 return count;
1371 if (!strcmp(name, "src_mac")) {
1372 char *v = valstr;
1373 unsigned char *m = pkt_dev->src_mac;
1375 len = strn_len(&user_buffer[i], sizeof(valstr) - 1);
1376 if (len < 0) {
1377 return len;
1379 memset(valstr, 0, sizeof(valstr));
1380 if (copy_from_user(valstr, &user_buffer[i], len))
1381 return -EFAULT;
1382 i += len;
1384 for (*m = 0; *v && m < pkt_dev->src_mac + 6; v++) {
1385 if (*v >= '0' && *v <= '9') {
1386 *m *= 16;
1387 *m += *v - '0';
1389 if (*v >= 'A' && *v <= 'F') {
1390 *m *= 16;
1391 *m += *v - 'A' + 10;
1393 if (*v >= 'a' && *v <= 'f') {
1394 *m *= 16;
1395 *m += *v - 'a' + 10;
1397 if (*v == ':') {
1398 m++;
1399 *m = 0;
1403 sprintf(pg_result, "OK: srcmac");
1404 return count;
1407 if (!strcmp(name, "clear_counters")) {
1408 pktgen_clear_counters(pkt_dev);
1409 sprintf(pg_result, "OK: Clearing counters.\n");
1410 return count;
1413 if (!strcmp(name, "flows")) {
1414 len = num_arg(&user_buffer[i], 10, &value);
1415 if (len < 0) {
1416 return len;
1418 i += len;
1419 if (value > MAX_CFLOWS)
1420 value = MAX_CFLOWS;
1422 pkt_dev->cflows = value;
1423 sprintf(pg_result, "OK: flows=%u", pkt_dev->cflows);
1424 return count;
1427 if (!strcmp(name, "flowlen")) {
1428 len = num_arg(&user_buffer[i], 10, &value);
1429 if (len < 0) {
1430 return len;
1432 i += len;
1433 pkt_dev->lflow = value;
1434 sprintf(pg_result, "OK: flowlen=%u", pkt_dev->lflow);
1435 return count;
1438 if (!strcmp(name, "mpls")) {
1439 unsigned n, offset;
1440 len = get_labels(&user_buffer[i], pkt_dev);
1441 if (len < 0) { return len; }
1442 i += len;
1443 offset = sprintf(pg_result, "OK: mpls=");
1444 for(n = 0; n < pkt_dev->nr_labels; n++)
1445 offset += sprintf(pg_result + offset,
1446 "%08x%s", ntohl(pkt_dev->labels[n]),
1447 n == pkt_dev->nr_labels-1 ? "" : ",");
1448 return count;
1451 sprintf(pkt_dev->result, "No such parameter \"%s\"", name);
1452 return -EINVAL;
1455 static int pktgen_if_open(struct inode *inode, struct file *file)
1457 return single_open(file, pktgen_if_show, PDE(inode)->data);
1460 static struct file_operations pktgen_if_fops = {
1461 .owner = THIS_MODULE,
1462 .open = pktgen_if_open,
1463 .read = seq_read,
1464 .llseek = seq_lseek,
1465 .write = pktgen_if_write,
1466 .release = single_release,
1469 static int pktgen_thread_show(struct seq_file *seq, void *v)
1471 struct pktgen_thread *t = seq->private;
1472 struct pktgen_dev *pkt_dev;
1474 BUG_ON(!t);
1476 seq_printf(seq, "Name: %s max_before_softirq: %d\n",
1477 t->name, t->max_before_softirq);
1479 seq_printf(seq, "Running: ");
1481 if_lock(t);
1482 list_for_each_entry(pkt_dev, &t->if_list, list)
1483 if (pkt_dev->running)
1484 seq_printf(seq, "%s ", pkt_dev->ifname);
1486 seq_printf(seq, "\nStopped: ");
1488 list_for_each_entry(pkt_dev, &t->if_list, list)
1489 if (!pkt_dev->running)
1490 seq_printf(seq, "%s ", pkt_dev->ifname);
1492 if (t->result[0])
1493 seq_printf(seq, "\nResult: %s\n", t->result);
1494 else
1495 seq_printf(seq, "\nResult: NA\n");
1497 if_unlock(t);
1499 return 0;
1502 static ssize_t pktgen_thread_write(struct file *file,
1503 const char __user * user_buffer,
1504 size_t count, loff_t * offset)
1506 struct seq_file *seq = (struct seq_file *)file->private_data;
1507 struct pktgen_thread *t = seq->private;
1508 int i = 0, max, len, ret;
1509 char name[40];
1510 char *pg_result;
1511 unsigned long value = 0;
1513 if (count < 1) {
1514 // sprintf(pg_result, "Wrong command format");
1515 return -EINVAL;
1518 max = count - i;
1519 len = count_trail_chars(&user_buffer[i], max);
1520 if (len < 0)
1521 return len;
1523 i += len;
1525 /* Read variable name */
1527 len = strn_len(&user_buffer[i], sizeof(name) - 1);
1528 if (len < 0)
1529 return len;
1531 memset(name, 0, sizeof(name));
1532 if (copy_from_user(name, &user_buffer[i], len))
1533 return -EFAULT;
1534 i += len;
1536 max = count - i;
1537 len = count_trail_chars(&user_buffer[i], max);
1538 if (len < 0)
1539 return len;
1541 i += len;
1543 if (debug)
1544 printk("pktgen: t=%s, count=%lu\n", name, (unsigned long)count);
1546 if (!t) {
1547 printk("pktgen: ERROR: No thread\n");
1548 ret = -EINVAL;
1549 goto out;
1552 pg_result = &(t->result[0]);
1554 if (!strcmp(name, "add_device")) {
1555 char f[32];
1556 memset(f, 0, 32);
1557 len = strn_len(&user_buffer[i], sizeof(f) - 1);
1558 if (len < 0) {
1559 ret = len;
1560 goto out;
1562 if (copy_from_user(f, &user_buffer[i], len))
1563 return -EFAULT;
1564 i += len;
1565 mutex_lock(&pktgen_thread_lock);
1566 pktgen_add_device(t, f);
1567 mutex_unlock(&pktgen_thread_lock);
1568 ret = count;
1569 sprintf(pg_result, "OK: add_device=%s", f);
1570 goto out;
1573 if (!strcmp(name, "rem_device_all")) {
1574 mutex_lock(&pktgen_thread_lock);
1575 t->control |= T_REMDEVALL;
1576 mutex_unlock(&pktgen_thread_lock);
1577 schedule_timeout_interruptible(msecs_to_jiffies(125)); /* Propagate thread->control */
1578 ret = count;
1579 sprintf(pg_result, "OK: rem_device_all");
1580 goto out;
1583 if (!strcmp(name, "max_before_softirq")) {
1584 len = num_arg(&user_buffer[i], 10, &value);
1585 mutex_lock(&pktgen_thread_lock);
1586 t->max_before_softirq = value;
1587 mutex_unlock(&pktgen_thread_lock);
1588 ret = count;
1589 sprintf(pg_result, "OK: max_before_softirq=%lu", value);
1590 goto out;
1593 ret = -EINVAL;
1594 out:
1595 return ret;
1598 static int pktgen_thread_open(struct inode *inode, struct file *file)
1600 return single_open(file, pktgen_thread_show, PDE(inode)->data);
1603 static struct file_operations pktgen_thread_fops = {
1604 .owner = THIS_MODULE,
1605 .open = pktgen_thread_open,
1606 .read = seq_read,
1607 .llseek = seq_lseek,
1608 .write = pktgen_thread_write,
1609 .release = single_release,
1612 /* Think find or remove for NN */
1613 static struct pktgen_dev *__pktgen_NN_threads(const char *ifname, int remove)
1615 struct pktgen_thread *t;
1616 struct pktgen_dev *pkt_dev = NULL;
1618 list_for_each_entry(t, &pktgen_threads, th_list) {
1619 pkt_dev = pktgen_find_dev(t, ifname);
1620 if (pkt_dev) {
1621 if (remove) {
1622 if_lock(t);
1623 pkt_dev->removal_mark = 1;
1624 t->control |= T_REMDEV;
1625 if_unlock(t);
1627 break;
1630 return pkt_dev;
1634 * mark a device for removal
1636 static int pktgen_mark_device(const char *ifname)
1638 struct pktgen_dev *pkt_dev = NULL;
1639 const int max_tries = 10, msec_per_try = 125;
1640 int i = 0;
1641 int ret = 0;
1643 mutex_lock(&pktgen_thread_lock);
1644 PG_DEBUG(printk("pktgen: pktgen_mark_device marking %s for removal\n",
1645 ifname));
1647 while (1) {
1649 pkt_dev = __pktgen_NN_threads(ifname, REMOVE);
1650 if (pkt_dev == NULL)
1651 break; /* success */
1653 mutex_unlock(&pktgen_thread_lock);
1654 PG_DEBUG(printk("pktgen: pktgen_mark_device waiting for %s "
1655 "to disappear....\n", ifname));
1656 schedule_timeout_interruptible(msecs_to_jiffies(msec_per_try));
1657 mutex_lock(&pktgen_thread_lock);
1659 if (++i >= max_tries) {
1660 printk("pktgen_mark_device: timed out after waiting "
1661 "%d msec for device %s to be removed\n",
1662 msec_per_try * i, ifname);
1663 ret = 1;
1664 break;
1669 mutex_unlock(&pktgen_thread_lock);
1671 return ret;
1674 static int pktgen_device_event(struct notifier_block *unused,
1675 unsigned long event, void *ptr)
1677 struct net_device *dev = (struct net_device *)(ptr);
1679 /* It is OK that we do not hold the group lock right now,
1680 * as we run under the RTNL lock.
1683 switch (event) {
1684 case NETDEV_CHANGEADDR:
1685 case NETDEV_GOING_DOWN:
1686 case NETDEV_DOWN:
1687 case NETDEV_UP:
1688 /* Ignore for now */
1689 break;
1691 case NETDEV_UNREGISTER:
1692 pktgen_mark_device(dev->name);
1693 break;
1696 return NOTIFY_DONE;
1699 /* Associate pktgen_dev with a device. */
1701 static struct net_device *pktgen_setup_dev(struct pktgen_dev *pkt_dev)
1703 struct net_device *odev;
1705 /* Clean old setups */
1707 if (pkt_dev->odev) {
1708 dev_put(pkt_dev->odev);
1709 pkt_dev->odev = NULL;
1712 odev = dev_get_by_name(pkt_dev->ifname);
1714 if (!odev) {
1715 printk("pktgen: no such netdevice: \"%s\"\n", pkt_dev->ifname);
1716 goto out;
1718 if (odev->type != ARPHRD_ETHER) {
1719 printk("pktgen: not an ethernet device: \"%s\"\n",
1720 pkt_dev->ifname);
1721 goto out_put;
1723 if (!netif_running(odev)) {
1724 printk("pktgen: device is down: \"%s\"\n", pkt_dev->ifname);
1725 goto out_put;
1727 pkt_dev->odev = odev;
1729 return pkt_dev->odev;
1731 out_put:
1732 dev_put(odev);
1733 out:
1734 return NULL;
1738 /* Read pkt_dev from the interface and set up internal pktgen_dev
1739 * structure to have the right information to create/send packets
1741 static void pktgen_setup_inject(struct pktgen_dev *pkt_dev)
1743 /* Try once more, just in case it works now. */
1744 if (!pkt_dev->odev)
1745 pktgen_setup_dev(pkt_dev);
1747 if (!pkt_dev->odev) {
1748 printk("pktgen: ERROR: pkt_dev->odev == NULL in setup_inject.\n");
1749 sprintf(pkt_dev->result,
1750 "ERROR: pkt_dev->odev == NULL in setup_inject.\n");
1751 return;
1754 /* Default to the interface's mac if not explicitly set. */
1756 if (is_zero_ether_addr(pkt_dev->src_mac))
1757 memcpy(&(pkt_dev->hh[6]), pkt_dev->odev->dev_addr, ETH_ALEN);
1759 /* Set up Dest MAC */
1760 memcpy(&(pkt_dev->hh[0]), pkt_dev->dst_mac, ETH_ALEN);
1762 /* Set up pkt size */
1763 pkt_dev->cur_pkt_size = pkt_dev->min_pkt_size;
1765 if (pkt_dev->flags & F_IPV6) {
1767 * Skip this automatic address setting until locks or functions
1768 * gets exported
1771 #ifdef NOTNOW
1772 int i, set = 0, err = 1;
1773 struct inet6_dev *idev;
1775 for (i = 0; i < IN6_ADDR_HSIZE; i++)
1776 if (pkt_dev->cur_in6_saddr.s6_addr[i]) {
1777 set = 1;
1778 break;
1781 if (!set) {
1784 * Use linklevel address if unconfigured.
1786 * use ipv6_get_lladdr if/when it's get exported
1789 read_lock(&addrconf_lock);
1790 if ((idev = __in6_dev_get(pkt_dev->odev)) != NULL) {
1791 struct inet6_ifaddr *ifp;
1793 read_lock_bh(&idev->lock);
1794 for (ifp = idev->addr_list; ifp;
1795 ifp = ifp->if_next) {
1796 if (ifp->scope == IFA_LINK
1797 && !(ifp->
1798 flags & IFA_F_TENTATIVE)) {
1799 ipv6_addr_copy(&pkt_dev->
1800 cur_in6_saddr,
1801 &ifp->addr);
1802 err = 0;
1803 break;
1806 read_unlock_bh(&idev->lock);
1808 read_unlock(&addrconf_lock);
1809 if (err)
1810 printk("pktgen: ERROR: IPv6 link address not availble.\n");
1812 #endif
1813 } else {
1814 pkt_dev->saddr_min = 0;
1815 pkt_dev->saddr_max = 0;
1816 if (strlen(pkt_dev->src_min) == 0) {
1818 struct in_device *in_dev;
1820 rcu_read_lock();
1821 in_dev = __in_dev_get_rcu(pkt_dev->odev);
1822 if (in_dev) {
1823 if (in_dev->ifa_list) {
1824 pkt_dev->saddr_min =
1825 in_dev->ifa_list->ifa_address;
1826 pkt_dev->saddr_max = pkt_dev->saddr_min;
1829 rcu_read_unlock();
1830 } else {
1831 pkt_dev->saddr_min = in_aton(pkt_dev->src_min);
1832 pkt_dev->saddr_max = in_aton(pkt_dev->src_max);
1835 pkt_dev->daddr_min = in_aton(pkt_dev->dst_min);
1836 pkt_dev->daddr_max = in_aton(pkt_dev->dst_max);
1838 /* Initialize current values. */
1839 pkt_dev->cur_dst_mac_offset = 0;
1840 pkt_dev->cur_src_mac_offset = 0;
1841 pkt_dev->cur_saddr = pkt_dev->saddr_min;
1842 pkt_dev->cur_daddr = pkt_dev->daddr_min;
1843 pkt_dev->cur_udp_dst = pkt_dev->udp_dst_min;
1844 pkt_dev->cur_udp_src = pkt_dev->udp_src_min;
1845 pkt_dev->nflows = 0;
1848 static void spin(struct pktgen_dev *pkt_dev, __u64 spin_until_us)
1850 __u64 start;
1851 __u64 now;
1853 start = now = getCurUs();
1854 printk(KERN_INFO "sleeping for %d\n", (int)(spin_until_us - now));
1855 while (now < spin_until_us) {
1856 /* TODO: optimize sleeping behavior */
1857 if (spin_until_us - now > jiffies_to_usecs(1) + 1)
1858 schedule_timeout_interruptible(1);
1859 else if (spin_until_us - now > 100) {
1860 do_softirq();
1861 if (!pkt_dev->running)
1862 return;
1863 if (need_resched())
1864 schedule();
1867 now = getCurUs();
1870 pkt_dev->idle_acc += now - start;
1873 /* Increment/randomize headers according to flags and current values
1874 * for IP src/dest, UDP src/dst port, MAC-Addr src/dst
1876 static void mod_cur_headers(struct pktgen_dev *pkt_dev)
1878 __u32 imn;
1879 __u32 imx;
1880 int flow = 0;
1882 if (pkt_dev->cflows) {
1883 flow = pktgen_random() % pkt_dev->cflows;
1885 if (pkt_dev->flows[flow].count > pkt_dev->lflow)
1886 pkt_dev->flows[flow].count = 0;
1889 /* Deal with source MAC */
1890 if (pkt_dev->src_mac_count > 1) {
1891 __u32 mc;
1892 __u32 tmp;
1894 if (pkt_dev->flags & F_MACSRC_RND)
1895 mc = pktgen_random() % (pkt_dev->src_mac_count);
1896 else {
1897 mc = pkt_dev->cur_src_mac_offset++;
1898 if (pkt_dev->cur_src_mac_offset >
1899 pkt_dev->src_mac_count)
1900 pkt_dev->cur_src_mac_offset = 0;
1903 tmp = pkt_dev->src_mac[5] + (mc & 0xFF);
1904 pkt_dev->hh[11] = tmp;
1905 tmp = (pkt_dev->src_mac[4] + ((mc >> 8) & 0xFF) + (tmp >> 8));
1906 pkt_dev->hh[10] = tmp;
1907 tmp = (pkt_dev->src_mac[3] + ((mc >> 16) & 0xFF) + (tmp >> 8));
1908 pkt_dev->hh[9] = tmp;
1909 tmp = (pkt_dev->src_mac[2] + ((mc >> 24) & 0xFF) + (tmp >> 8));
1910 pkt_dev->hh[8] = tmp;
1911 tmp = (pkt_dev->src_mac[1] + (tmp >> 8));
1912 pkt_dev->hh[7] = tmp;
1915 /* Deal with Destination MAC */
1916 if (pkt_dev->dst_mac_count > 1) {
1917 __u32 mc;
1918 __u32 tmp;
1920 if (pkt_dev->flags & F_MACDST_RND)
1921 mc = pktgen_random() % (pkt_dev->dst_mac_count);
1923 else {
1924 mc = pkt_dev->cur_dst_mac_offset++;
1925 if (pkt_dev->cur_dst_mac_offset >
1926 pkt_dev->dst_mac_count) {
1927 pkt_dev->cur_dst_mac_offset = 0;
1931 tmp = pkt_dev->dst_mac[5] + (mc & 0xFF);
1932 pkt_dev->hh[5] = tmp;
1933 tmp = (pkt_dev->dst_mac[4] + ((mc >> 8) & 0xFF) + (tmp >> 8));
1934 pkt_dev->hh[4] = tmp;
1935 tmp = (pkt_dev->dst_mac[3] + ((mc >> 16) & 0xFF) + (tmp >> 8));
1936 pkt_dev->hh[3] = tmp;
1937 tmp = (pkt_dev->dst_mac[2] + ((mc >> 24) & 0xFF) + (tmp >> 8));
1938 pkt_dev->hh[2] = tmp;
1939 tmp = (pkt_dev->dst_mac[1] + (tmp >> 8));
1940 pkt_dev->hh[1] = tmp;
1943 if (pkt_dev->flags & F_MPLS_RND) {
1944 unsigned i;
1945 for(i = 0; i < pkt_dev->nr_labels; i++)
1946 if (pkt_dev->labels[i] & MPLS_STACK_BOTTOM)
1947 pkt_dev->labels[i] = MPLS_STACK_BOTTOM |
1948 (pktgen_random() &
1949 htonl(0x000fffff));
1952 if (pkt_dev->udp_src_min < pkt_dev->udp_src_max) {
1953 if (pkt_dev->flags & F_UDPSRC_RND)
1954 pkt_dev->cur_udp_src =
1955 ((pktgen_random() %
1956 (pkt_dev->udp_src_max - pkt_dev->udp_src_min)) +
1957 pkt_dev->udp_src_min);
1959 else {
1960 pkt_dev->cur_udp_src++;
1961 if (pkt_dev->cur_udp_src >= pkt_dev->udp_src_max)
1962 pkt_dev->cur_udp_src = pkt_dev->udp_src_min;
1966 if (pkt_dev->udp_dst_min < pkt_dev->udp_dst_max) {
1967 if (pkt_dev->flags & F_UDPDST_RND) {
1968 pkt_dev->cur_udp_dst =
1969 ((pktgen_random() %
1970 (pkt_dev->udp_dst_max - pkt_dev->udp_dst_min)) +
1971 pkt_dev->udp_dst_min);
1972 } else {
1973 pkt_dev->cur_udp_dst++;
1974 if (pkt_dev->cur_udp_dst >= pkt_dev->udp_dst_max)
1975 pkt_dev->cur_udp_dst = pkt_dev->udp_dst_min;
1979 if (!(pkt_dev->flags & F_IPV6)) {
1981 if ((imn = ntohl(pkt_dev->saddr_min)) < (imx =
1982 ntohl(pkt_dev->
1983 saddr_max))) {
1984 __u32 t;
1985 if (pkt_dev->flags & F_IPSRC_RND)
1986 t = ((pktgen_random() % (imx - imn)) + imn);
1987 else {
1988 t = ntohl(pkt_dev->cur_saddr);
1989 t++;
1990 if (t > imx) {
1991 t = imn;
1994 pkt_dev->cur_saddr = htonl(t);
1997 if (pkt_dev->cflows && pkt_dev->flows[flow].count != 0) {
1998 pkt_dev->cur_daddr = pkt_dev->flows[flow].cur_daddr;
1999 } else {
2001 if ((imn = ntohl(pkt_dev->daddr_min)) < (imx =
2002 ntohl(pkt_dev->
2003 daddr_max)))
2005 __u32 t;
2006 if (pkt_dev->flags & F_IPDST_RND) {
2008 t = ((pktgen_random() % (imx - imn)) +
2009 imn);
2010 t = htonl(t);
2012 while (LOOPBACK(t) || MULTICAST(t)
2013 || BADCLASS(t) || ZERONET(t)
2014 || LOCAL_MCAST(t)) {
2015 t = ((pktgen_random() %
2016 (imx - imn)) + imn);
2017 t = htonl(t);
2019 pkt_dev->cur_daddr = t;
2022 else {
2023 t = ntohl(pkt_dev->cur_daddr);
2024 t++;
2025 if (t > imx) {
2026 t = imn;
2028 pkt_dev->cur_daddr = htonl(t);
2031 if (pkt_dev->cflows) {
2032 pkt_dev->flows[flow].cur_daddr =
2033 pkt_dev->cur_daddr;
2034 pkt_dev->nflows++;
2037 } else { /* IPV6 * */
2039 if (pkt_dev->min_in6_daddr.s6_addr32[0] == 0 &&
2040 pkt_dev->min_in6_daddr.s6_addr32[1] == 0 &&
2041 pkt_dev->min_in6_daddr.s6_addr32[2] == 0 &&
2042 pkt_dev->min_in6_daddr.s6_addr32[3] == 0) ;
2043 else {
2044 int i;
2046 /* Only random destinations yet */
2048 for (i = 0; i < 4; i++) {
2049 pkt_dev->cur_in6_daddr.s6_addr32[i] =
2050 ((pktgen_random() |
2051 pkt_dev->min_in6_daddr.s6_addr32[i]) &
2052 pkt_dev->max_in6_daddr.s6_addr32[i]);
2057 if (pkt_dev->min_pkt_size < pkt_dev->max_pkt_size) {
2058 __u32 t;
2059 if (pkt_dev->flags & F_TXSIZE_RND) {
2060 t = ((pktgen_random() %
2061 (pkt_dev->max_pkt_size - pkt_dev->min_pkt_size))
2062 + pkt_dev->min_pkt_size);
2063 } else {
2064 t = pkt_dev->cur_pkt_size + 1;
2065 if (t > pkt_dev->max_pkt_size)
2066 t = pkt_dev->min_pkt_size;
2068 pkt_dev->cur_pkt_size = t;
2071 pkt_dev->flows[flow].count++;
2074 static void mpls_push(__be32 *mpls, struct pktgen_dev *pkt_dev)
2076 unsigned i;
2077 for(i = 0; i < pkt_dev->nr_labels; i++) {
2078 *mpls++ = pkt_dev->labels[i] & ~MPLS_STACK_BOTTOM;
2080 mpls--;
2081 *mpls |= MPLS_STACK_BOTTOM;
2084 static struct sk_buff *fill_packet_ipv4(struct net_device *odev,
2085 struct pktgen_dev *pkt_dev)
2087 struct sk_buff *skb = NULL;
2088 __u8 *eth;
2089 struct udphdr *udph;
2090 int datalen, iplen;
2091 struct iphdr *iph;
2092 struct pktgen_hdr *pgh = NULL;
2093 __be16 protocol = __constant_htons(ETH_P_IP);
2094 __be32 *mpls;
2096 if (pkt_dev->nr_labels)
2097 protocol = __constant_htons(ETH_P_MPLS_UC);
2099 /* Update any of the values, used when we're incrementing various
2100 * fields.
2102 mod_cur_headers(pkt_dev);
2104 datalen = (odev->hard_header_len + 16) & ~0xf;
2105 skb = alloc_skb(pkt_dev->cur_pkt_size + 64 + datalen +
2106 pkt_dev->nr_labels*sizeof(u32), GFP_ATOMIC);
2107 if (!skb) {
2108 sprintf(pkt_dev->result, "No memory");
2109 return NULL;
2112 skb_reserve(skb, datalen);
2114 /* Reserve for ethernet and IP header */
2115 eth = (__u8 *) skb_push(skb, 14);
2116 mpls = (__be32 *)skb_put(skb, pkt_dev->nr_labels*sizeof(__u32));
2117 if (pkt_dev->nr_labels)
2118 mpls_push(mpls, pkt_dev);
2119 iph = (struct iphdr *)skb_put(skb, sizeof(struct iphdr));
2120 udph = (struct udphdr *)skb_put(skb, sizeof(struct udphdr));
2122 memcpy(eth, pkt_dev->hh, 12);
2123 *(u16 *) & eth[12] = protocol;
2125 /* Eth + IPh + UDPh + mpls */
2126 datalen = pkt_dev->cur_pkt_size - 14 - 20 - 8 -
2127 pkt_dev->nr_labels*sizeof(u32);
2128 if (datalen < sizeof(struct pktgen_hdr))
2129 datalen = sizeof(struct pktgen_hdr);
2131 udph->source = htons(pkt_dev->cur_udp_src);
2132 udph->dest = htons(pkt_dev->cur_udp_dst);
2133 udph->len = htons(datalen + 8); /* DATA + udphdr */
2134 udph->check = 0; /* No checksum */
2136 iph->ihl = 5;
2137 iph->version = 4;
2138 iph->ttl = 32;
2139 iph->tos = 0;
2140 iph->protocol = IPPROTO_UDP; /* UDP */
2141 iph->saddr = pkt_dev->cur_saddr;
2142 iph->daddr = pkt_dev->cur_daddr;
2143 iph->frag_off = 0;
2144 iplen = 20 + 8 + datalen;
2145 iph->tot_len = htons(iplen);
2146 iph->check = 0;
2147 iph->check = ip_fast_csum((void *)iph, iph->ihl);
2148 skb->protocol = protocol;
2149 skb->mac.raw = ((u8 *) iph) - 14 - pkt_dev->nr_labels*sizeof(u32);
2150 skb->dev = odev;
2151 skb->pkt_type = PACKET_HOST;
2153 if (pkt_dev->nfrags <= 0)
2154 pgh = (struct pktgen_hdr *)skb_put(skb, datalen);
2155 else {
2156 int frags = pkt_dev->nfrags;
2157 int i;
2159 pgh = (struct pktgen_hdr *)(((char *)(udph)) + 8);
2161 if (frags > MAX_SKB_FRAGS)
2162 frags = MAX_SKB_FRAGS;
2163 if (datalen > frags * PAGE_SIZE) {
2164 skb_put(skb, datalen - frags * PAGE_SIZE);
2165 datalen = frags * PAGE_SIZE;
2168 i = 0;
2169 while (datalen > 0) {
2170 struct page *page = alloc_pages(GFP_KERNEL, 0);
2171 skb_shinfo(skb)->frags[i].page = page;
2172 skb_shinfo(skb)->frags[i].page_offset = 0;
2173 skb_shinfo(skb)->frags[i].size =
2174 (datalen < PAGE_SIZE ? datalen : PAGE_SIZE);
2175 datalen -= skb_shinfo(skb)->frags[i].size;
2176 skb->len += skb_shinfo(skb)->frags[i].size;
2177 skb->data_len += skb_shinfo(skb)->frags[i].size;
2178 i++;
2179 skb_shinfo(skb)->nr_frags = i;
2182 while (i < frags) {
2183 int rem;
2185 if (i == 0)
2186 break;
2188 rem = skb_shinfo(skb)->frags[i - 1].size / 2;
2189 if (rem == 0)
2190 break;
2192 skb_shinfo(skb)->frags[i - 1].size -= rem;
2194 skb_shinfo(skb)->frags[i] =
2195 skb_shinfo(skb)->frags[i - 1];
2196 get_page(skb_shinfo(skb)->frags[i].page);
2197 skb_shinfo(skb)->frags[i].page =
2198 skb_shinfo(skb)->frags[i - 1].page;
2199 skb_shinfo(skb)->frags[i].page_offset +=
2200 skb_shinfo(skb)->frags[i - 1].size;
2201 skb_shinfo(skb)->frags[i].size = rem;
2202 i++;
2203 skb_shinfo(skb)->nr_frags = i;
2207 /* Stamp the time, and sequence number, convert them to network byte order */
2209 if (pgh) {
2210 struct timeval timestamp;
2212 pgh->pgh_magic = htonl(PKTGEN_MAGIC);
2213 pgh->seq_num = htonl(pkt_dev->seq_num);
2215 do_gettimeofday(&timestamp);
2216 pgh->tv_sec = htonl(timestamp.tv_sec);
2217 pgh->tv_usec = htonl(timestamp.tv_usec);
2219 pkt_dev->seq_num++;
2221 return skb;
2225 * scan_ip6, fmt_ip taken from dietlibc-0.21
2226 * Author Felix von Leitner <felix-dietlibc@fefe.de>
2228 * Slightly modified for kernel.
2229 * Should be candidate for net/ipv4/utils.c
2230 * --ro
2233 static unsigned int scan_ip6(const char *s, char ip[16])
2235 unsigned int i;
2236 unsigned int len = 0;
2237 unsigned long u;
2238 char suffix[16];
2239 unsigned int prefixlen = 0;
2240 unsigned int suffixlen = 0;
2241 __u32 tmp;
2243 for (i = 0; i < 16; i++)
2244 ip[i] = 0;
2246 for (;;) {
2247 if (*s == ':') {
2248 len++;
2249 if (s[1] == ':') { /* Found "::", skip to part 2 */
2250 s += 2;
2251 len++;
2252 break;
2254 s++;
2257 char *tmp;
2258 u = simple_strtoul(s, &tmp, 16);
2259 i = tmp - s;
2262 if (!i)
2263 return 0;
2264 if (prefixlen == 12 && s[i] == '.') {
2266 /* the last 4 bytes may be written as IPv4 address */
2268 tmp = in_aton(s);
2269 memcpy((struct in_addr *)(ip + 12), &tmp, sizeof(tmp));
2270 return i + len;
2272 ip[prefixlen++] = (u >> 8);
2273 ip[prefixlen++] = (u & 255);
2274 s += i;
2275 len += i;
2276 if (prefixlen == 16)
2277 return len;
2280 /* part 2, after "::" */
2281 for (;;) {
2282 if (*s == ':') {
2283 if (suffixlen == 0)
2284 break;
2285 s++;
2286 len++;
2287 } else if (suffixlen != 0)
2288 break;
2290 char *tmp;
2291 u = simple_strtol(s, &tmp, 16);
2292 i = tmp - s;
2294 if (!i) {
2295 if (*s)
2296 len--;
2297 break;
2299 if (suffixlen + prefixlen <= 12 && s[i] == '.') {
2300 tmp = in_aton(s);
2301 memcpy((struct in_addr *)(suffix + suffixlen), &tmp,
2302 sizeof(tmp));
2303 suffixlen += 4;
2304 len += strlen(s);
2305 break;
2307 suffix[suffixlen++] = (u >> 8);
2308 suffix[suffixlen++] = (u & 255);
2309 s += i;
2310 len += i;
2311 if (prefixlen + suffixlen == 16)
2312 break;
2314 for (i = 0; i < suffixlen; i++)
2315 ip[16 - suffixlen + i] = suffix[i];
2316 return len;
2319 static char tohex(char hexdigit)
2321 return hexdigit > 9 ? hexdigit + 'a' - 10 : hexdigit + '0';
2324 static int fmt_xlong(char *s, unsigned int i)
2326 char *bak = s;
2327 *s = tohex((i >> 12) & 0xf);
2328 if (s != bak || *s != '0')
2329 ++s;
2330 *s = tohex((i >> 8) & 0xf);
2331 if (s != bak || *s != '0')
2332 ++s;
2333 *s = tohex((i >> 4) & 0xf);
2334 if (s != bak || *s != '0')
2335 ++s;
2336 *s = tohex(i & 0xf);
2337 return s - bak + 1;
2340 static unsigned int fmt_ip6(char *s, const char ip[16])
2342 unsigned int len;
2343 unsigned int i;
2344 unsigned int temp;
2345 unsigned int compressing;
2346 int j;
2348 len = 0;
2349 compressing = 0;
2350 for (j = 0; j < 16; j += 2) {
2352 #ifdef V4MAPPEDPREFIX
2353 if (j == 12 && !memcmp(ip, V4mappedprefix, 12)) {
2354 inet_ntoa_r(*(struct in_addr *)(ip + 12), s);
2355 temp = strlen(s);
2356 return len + temp;
2358 #endif
2359 temp = ((unsigned long)(unsigned char)ip[j] << 8) +
2360 (unsigned long)(unsigned char)ip[j + 1];
2361 if (temp == 0) {
2362 if (!compressing) {
2363 compressing = 1;
2364 if (j == 0) {
2365 *s++ = ':';
2366 ++len;
2369 } else {
2370 if (compressing) {
2371 compressing = 0;
2372 *s++ = ':';
2373 ++len;
2375 i = fmt_xlong(s, temp);
2376 len += i;
2377 s += i;
2378 if (j < 14) {
2379 *s++ = ':';
2380 ++len;
2384 if (compressing) {
2385 *s++ = ':';
2386 ++len;
2388 *s = 0;
2389 return len;
2392 static struct sk_buff *fill_packet_ipv6(struct net_device *odev,
2393 struct pktgen_dev *pkt_dev)
2395 struct sk_buff *skb = NULL;
2396 __u8 *eth;
2397 struct udphdr *udph;
2398 int datalen;
2399 struct ipv6hdr *iph;
2400 struct pktgen_hdr *pgh = NULL;
2401 __be16 protocol = __constant_htons(ETH_P_IPV6);
2402 __be32 *mpls;
2404 if (pkt_dev->nr_labels)
2405 protocol = __constant_htons(ETH_P_MPLS_UC);
2407 /* Update any of the values, used when we're incrementing various
2408 * fields.
2410 mod_cur_headers(pkt_dev);
2412 skb = alloc_skb(pkt_dev->cur_pkt_size + 64 + 16 +
2413 pkt_dev->nr_labels*sizeof(u32), GFP_ATOMIC);
2414 if (!skb) {
2415 sprintf(pkt_dev->result, "No memory");
2416 return NULL;
2419 skb_reserve(skb, 16);
2421 /* Reserve for ethernet and IP header */
2422 eth = (__u8 *) skb_push(skb, 14);
2423 mpls = (__be32 *)skb_put(skb, pkt_dev->nr_labels*sizeof(__u32));
2424 if (pkt_dev->nr_labels)
2425 mpls_push(mpls, pkt_dev);
2426 iph = (struct ipv6hdr *)skb_put(skb, sizeof(struct ipv6hdr));
2427 udph = (struct udphdr *)skb_put(skb, sizeof(struct udphdr));
2429 memcpy(eth, pkt_dev->hh, 12);
2430 *(u16 *) & eth[12] = __constant_htons(ETH_P_IPV6);
2432 /* Eth + IPh + UDPh + mpls */
2433 datalen = pkt_dev->cur_pkt_size - 14 -
2434 sizeof(struct ipv6hdr) - sizeof(struct udphdr) -
2435 pkt_dev->nr_labels*sizeof(u32);
2437 if (datalen < sizeof(struct pktgen_hdr)) {
2438 datalen = sizeof(struct pktgen_hdr);
2439 if (net_ratelimit())
2440 printk(KERN_INFO "pktgen: increased datalen to %d\n",
2441 datalen);
2444 udph->source = htons(pkt_dev->cur_udp_src);
2445 udph->dest = htons(pkt_dev->cur_udp_dst);
2446 udph->len = htons(datalen + sizeof(struct udphdr));
2447 udph->check = 0; /* No checksum */
2449 *(u32 *) iph = __constant_htonl(0x60000000); /* Version + flow */
2451 iph->hop_limit = 32;
2453 iph->payload_len = htons(sizeof(struct udphdr) + datalen);
2454 iph->nexthdr = IPPROTO_UDP;
2456 ipv6_addr_copy(&iph->daddr, &pkt_dev->cur_in6_daddr);
2457 ipv6_addr_copy(&iph->saddr, &pkt_dev->cur_in6_saddr);
2459 skb->mac.raw = ((u8 *) iph) - 14 - pkt_dev->nr_labels*sizeof(u32);
2460 skb->protocol = protocol;
2461 skb->dev = odev;
2462 skb->pkt_type = PACKET_HOST;
2464 if (pkt_dev->nfrags <= 0)
2465 pgh = (struct pktgen_hdr *)skb_put(skb, datalen);
2466 else {
2467 int frags = pkt_dev->nfrags;
2468 int i;
2470 pgh = (struct pktgen_hdr *)(((char *)(udph)) + 8);
2472 if (frags > MAX_SKB_FRAGS)
2473 frags = MAX_SKB_FRAGS;
2474 if (datalen > frags * PAGE_SIZE) {
2475 skb_put(skb, datalen - frags * PAGE_SIZE);
2476 datalen = frags * PAGE_SIZE;
2479 i = 0;
2480 while (datalen > 0) {
2481 struct page *page = alloc_pages(GFP_KERNEL, 0);
2482 skb_shinfo(skb)->frags[i].page = page;
2483 skb_shinfo(skb)->frags[i].page_offset = 0;
2484 skb_shinfo(skb)->frags[i].size =
2485 (datalen < PAGE_SIZE ? datalen : PAGE_SIZE);
2486 datalen -= skb_shinfo(skb)->frags[i].size;
2487 skb->len += skb_shinfo(skb)->frags[i].size;
2488 skb->data_len += skb_shinfo(skb)->frags[i].size;
2489 i++;
2490 skb_shinfo(skb)->nr_frags = i;
2493 while (i < frags) {
2494 int rem;
2496 if (i == 0)
2497 break;
2499 rem = skb_shinfo(skb)->frags[i - 1].size / 2;
2500 if (rem == 0)
2501 break;
2503 skb_shinfo(skb)->frags[i - 1].size -= rem;
2505 skb_shinfo(skb)->frags[i] =
2506 skb_shinfo(skb)->frags[i - 1];
2507 get_page(skb_shinfo(skb)->frags[i].page);
2508 skb_shinfo(skb)->frags[i].page =
2509 skb_shinfo(skb)->frags[i - 1].page;
2510 skb_shinfo(skb)->frags[i].page_offset +=
2511 skb_shinfo(skb)->frags[i - 1].size;
2512 skb_shinfo(skb)->frags[i].size = rem;
2513 i++;
2514 skb_shinfo(skb)->nr_frags = i;
2518 /* Stamp the time, and sequence number, convert them to network byte order */
2519 /* should we update cloned packets too ? */
2520 if (pgh) {
2521 struct timeval timestamp;
2523 pgh->pgh_magic = htonl(PKTGEN_MAGIC);
2524 pgh->seq_num = htonl(pkt_dev->seq_num);
2526 do_gettimeofday(&timestamp);
2527 pgh->tv_sec = htonl(timestamp.tv_sec);
2528 pgh->tv_usec = htonl(timestamp.tv_usec);
2530 pkt_dev->seq_num++;
2532 return skb;
2535 static inline struct sk_buff *fill_packet(struct net_device *odev,
2536 struct pktgen_dev *pkt_dev)
2538 if (pkt_dev->flags & F_IPV6)
2539 return fill_packet_ipv6(odev, pkt_dev);
2540 else
2541 return fill_packet_ipv4(odev, pkt_dev);
2544 static void pktgen_clear_counters(struct pktgen_dev *pkt_dev)
2546 pkt_dev->seq_num = 1;
2547 pkt_dev->idle_acc = 0;
2548 pkt_dev->sofar = 0;
2549 pkt_dev->tx_bytes = 0;
2550 pkt_dev->errors = 0;
2553 /* Set up structure for sending pkts, clear counters */
2555 static void pktgen_run(struct pktgen_thread *t)
2557 struct pktgen_dev *pkt_dev;
2558 int started = 0;
2560 PG_DEBUG(printk("pktgen: entering pktgen_run. %p\n", t));
2562 if_lock(t);
2563 list_for_each_entry(pkt_dev, &t->if_list, list) {
2566 * setup odev and create initial packet.
2568 pktgen_setup_inject(pkt_dev);
2570 if (pkt_dev->odev) {
2571 pktgen_clear_counters(pkt_dev);
2572 pkt_dev->running = 1; /* Cranke yeself! */
2573 pkt_dev->skb = NULL;
2574 pkt_dev->started_at = getCurUs();
2575 pkt_dev->next_tx_us = getCurUs(); /* Transmit immediately */
2576 pkt_dev->next_tx_ns = 0;
2578 strcpy(pkt_dev->result, "Starting");
2579 started++;
2580 } else
2581 strcpy(pkt_dev->result, "Error starting");
2583 if_unlock(t);
2584 if (started)
2585 t->control &= ~(T_STOP);
2588 static void pktgen_stop_all_threads_ifs(void)
2590 struct pktgen_thread *t;
2592 PG_DEBUG(printk("pktgen: entering pktgen_stop_all_threads_ifs.\n"));
2594 mutex_lock(&pktgen_thread_lock);
2596 list_for_each_entry(t, &pktgen_threads, th_list)
2597 t->control |= T_STOP;
2599 mutex_unlock(&pktgen_thread_lock);
2602 static int thread_is_running(struct pktgen_thread *t)
2604 struct pktgen_dev *pkt_dev;
2605 int res = 0;
2607 list_for_each_entry(pkt_dev, &t->if_list, list)
2608 if (pkt_dev->running) {
2609 res = 1;
2610 break;
2612 return res;
2615 static int pktgen_wait_thread_run(struct pktgen_thread *t)
2617 if_lock(t);
2619 while (thread_is_running(t)) {
2621 if_unlock(t);
2623 msleep_interruptible(100);
2625 if (signal_pending(current))
2626 goto signal;
2627 if_lock(t);
2629 if_unlock(t);
2630 return 1;
2631 signal:
2632 return 0;
2635 static int pktgen_wait_all_threads_run(void)
2637 struct pktgen_thread *t;
2638 int sig = 1;
2640 mutex_lock(&pktgen_thread_lock);
2642 list_for_each_entry(t, &pktgen_threads, th_list) {
2643 sig = pktgen_wait_thread_run(t);
2644 if (sig == 0)
2645 break;
2648 if (sig == 0)
2649 list_for_each_entry(t, &pktgen_threads, th_list)
2650 t->control |= (T_STOP);
2652 mutex_unlock(&pktgen_thread_lock);
2653 return sig;
2656 static void pktgen_run_all_threads(void)
2658 struct pktgen_thread *t;
2660 PG_DEBUG(printk("pktgen: entering pktgen_run_all_threads.\n"));
2662 mutex_lock(&pktgen_thread_lock);
2664 list_for_each_entry(t, &pktgen_threads, th_list)
2665 t->control |= (T_RUN);
2667 mutex_unlock(&pktgen_thread_lock);
2669 schedule_timeout_interruptible(msecs_to_jiffies(125)); /* Propagate thread->control */
2671 pktgen_wait_all_threads_run();
2674 static void show_results(struct pktgen_dev *pkt_dev, int nr_frags)
2676 __u64 total_us, bps, mbps, pps, idle;
2677 char *p = pkt_dev->result;
2679 total_us = pkt_dev->stopped_at - pkt_dev->started_at;
2681 idle = pkt_dev->idle_acc;
2683 p += sprintf(p, "OK: %llu(c%llu+d%llu) usec, %llu (%dbyte,%dfrags)\n",
2684 (unsigned long long)total_us,
2685 (unsigned long long)(total_us - idle),
2686 (unsigned long long)idle,
2687 (unsigned long long)pkt_dev->sofar,
2688 pkt_dev->cur_pkt_size, nr_frags);
2690 pps = pkt_dev->sofar * USEC_PER_SEC;
2692 while ((total_us >> 32) != 0) {
2693 pps >>= 1;
2694 total_us >>= 1;
2697 do_div(pps, total_us);
2699 bps = pps * 8 * pkt_dev->cur_pkt_size;
2701 mbps = bps;
2702 do_div(mbps, 1000000);
2703 p += sprintf(p, " %llupps %lluMb/sec (%llubps) errors: %llu",
2704 (unsigned long long)pps,
2705 (unsigned long long)mbps,
2706 (unsigned long long)bps,
2707 (unsigned long long)pkt_dev->errors);
2710 /* Set stopped-at timer, remove from running list, do counters & statistics */
2712 static int pktgen_stop_device(struct pktgen_dev *pkt_dev)
2714 int nr_frags = pkt_dev->skb ? skb_shinfo(pkt_dev->skb)->nr_frags : -1;
2716 if (!pkt_dev->running) {
2717 printk("pktgen: interface: %s is already stopped\n",
2718 pkt_dev->ifname);
2719 return -EINVAL;
2722 pkt_dev->stopped_at = getCurUs();
2723 pkt_dev->running = 0;
2725 show_results(pkt_dev, nr_frags);
2727 return 0;
2730 static struct pktgen_dev *next_to_run(struct pktgen_thread *t)
2732 struct pktgen_dev *pkt_dev, *best = NULL;
2734 if_lock(t);
2736 list_for_each_entry(pkt_dev, &t->if_list, list) {
2737 if (!pkt_dev->running)
2738 continue;
2739 if (best == NULL)
2740 best = pkt_dev;
2741 else if (pkt_dev->next_tx_us < best->next_tx_us)
2742 best = pkt_dev;
2744 if_unlock(t);
2745 return best;
2748 static void pktgen_stop(struct pktgen_thread *t)
2750 struct pktgen_dev *pkt_dev;
2752 PG_DEBUG(printk("pktgen: entering pktgen_stop\n"));
2754 if_lock(t);
2756 list_for_each_entry(pkt_dev, &t->if_list, list) {
2757 pktgen_stop_device(pkt_dev);
2758 if (pkt_dev->skb)
2759 kfree_skb(pkt_dev->skb);
2761 pkt_dev->skb = NULL;
2764 if_unlock(t);
2768 * one of our devices needs to be removed - find it
2769 * and remove it
2771 static void pktgen_rem_one_if(struct pktgen_thread *t)
2773 struct list_head *q, *n;
2774 struct pktgen_dev *cur;
2776 PG_DEBUG(printk("pktgen: entering pktgen_rem_one_if\n"));
2778 if_lock(t);
2780 list_for_each_safe(q, n, &t->if_list) {
2781 cur = list_entry(q, struct pktgen_dev, list);
2783 if (!cur->removal_mark)
2784 continue;
2786 if (cur->skb)
2787 kfree_skb(cur->skb);
2788 cur->skb = NULL;
2790 pktgen_remove_device(t, cur);
2792 break;
2795 if_unlock(t);
2798 static void pktgen_rem_all_ifs(struct pktgen_thread *t)
2800 struct list_head *q, *n;
2801 struct pktgen_dev *cur;
2803 /* Remove all devices, free mem */
2805 PG_DEBUG(printk("pktgen: entering pktgen_rem_all_ifs\n"));
2806 if_lock(t);
2808 list_for_each_safe(q, n, &t->if_list) {
2809 cur = list_entry(q, struct pktgen_dev, list);
2811 if (cur->skb)
2812 kfree_skb(cur->skb);
2813 cur->skb = NULL;
2815 pktgen_remove_device(t, cur);
2818 if_unlock(t);
2821 static void pktgen_rem_thread(struct pktgen_thread *t)
2823 /* Remove from the thread list */
2825 remove_proc_entry(t->name, pg_proc_dir);
2827 mutex_lock(&pktgen_thread_lock);
2829 list_del(&t->th_list);
2831 mutex_unlock(&pktgen_thread_lock);
2834 static __inline__ void pktgen_xmit(struct pktgen_dev *pkt_dev)
2836 struct net_device *odev = NULL;
2837 __u64 idle_start = 0;
2838 int ret;
2840 odev = pkt_dev->odev;
2842 if (pkt_dev->delay_us || pkt_dev->delay_ns) {
2843 u64 now;
2845 now = getCurUs();
2846 if (now < pkt_dev->next_tx_us)
2847 spin(pkt_dev, pkt_dev->next_tx_us);
2849 /* This is max DELAY, this has special meaning of
2850 * "never transmit"
2852 if (pkt_dev->delay_us == 0x7FFFFFFF) {
2853 pkt_dev->next_tx_us = getCurUs() + pkt_dev->delay_us;
2854 pkt_dev->next_tx_ns = pkt_dev->delay_ns;
2855 goto out;
2859 if (netif_queue_stopped(odev) || need_resched()) {
2860 idle_start = getCurUs();
2862 if (!netif_running(odev)) {
2863 pktgen_stop_device(pkt_dev);
2864 if (pkt_dev->skb)
2865 kfree_skb(pkt_dev->skb);
2866 pkt_dev->skb = NULL;
2867 goto out;
2869 if (need_resched())
2870 schedule();
2872 pkt_dev->idle_acc += getCurUs() - idle_start;
2874 if (netif_queue_stopped(odev)) {
2875 pkt_dev->next_tx_us = getCurUs(); /* TODO */
2876 pkt_dev->next_tx_ns = 0;
2877 goto out; /* Try the next interface */
2881 if (pkt_dev->last_ok || !pkt_dev->skb) {
2882 if ((++pkt_dev->clone_count >= pkt_dev->clone_skb)
2883 || (!pkt_dev->skb)) {
2884 /* build a new pkt */
2885 if (pkt_dev->skb)
2886 kfree_skb(pkt_dev->skb);
2888 pkt_dev->skb = fill_packet(odev, pkt_dev);
2889 if (pkt_dev->skb == NULL) {
2890 printk("pktgen: ERROR: couldn't allocate skb in fill_packet.\n");
2891 schedule();
2892 pkt_dev->clone_count--; /* back out increment, OOM */
2893 goto out;
2895 pkt_dev->allocated_skbs++;
2896 pkt_dev->clone_count = 0; /* reset counter */
2900 spin_lock_bh(&odev->xmit_lock);
2901 if (!netif_queue_stopped(odev)) {
2903 atomic_inc(&(pkt_dev->skb->users));
2904 retry_now:
2905 ret = odev->hard_start_xmit(pkt_dev->skb, odev);
2906 if (likely(ret == NETDEV_TX_OK)) {
2907 pkt_dev->last_ok = 1;
2908 pkt_dev->sofar++;
2909 pkt_dev->seq_num++;
2910 pkt_dev->tx_bytes += pkt_dev->cur_pkt_size;
2912 } else if (ret == NETDEV_TX_LOCKED
2913 && (odev->features & NETIF_F_LLTX)) {
2914 cpu_relax();
2915 goto retry_now;
2916 } else { /* Retry it next time */
2918 atomic_dec(&(pkt_dev->skb->users));
2920 if (debug && net_ratelimit())
2921 printk(KERN_INFO "pktgen: Hard xmit error\n");
2923 pkt_dev->errors++;
2924 pkt_dev->last_ok = 0;
2927 pkt_dev->next_tx_us = getCurUs();
2928 pkt_dev->next_tx_ns = 0;
2930 pkt_dev->next_tx_us += pkt_dev->delay_us;
2931 pkt_dev->next_tx_ns += pkt_dev->delay_ns;
2933 if (pkt_dev->next_tx_ns > 1000) {
2934 pkt_dev->next_tx_us++;
2935 pkt_dev->next_tx_ns -= 1000;
2939 else { /* Retry it next time */
2940 pkt_dev->last_ok = 0;
2941 pkt_dev->next_tx_us = getCurUs(); /* TODO */
2942 pkt_dev->next_tx_ns = 0;
2945 spin_unlock_bh(&odev->xmit_lock);
2947 /* If pkt_dev->count is zero, then run forever */
2948 if ((pkt_dev->count != 0) && (pkt_dev->sofar >= pkt_dev->count)) {
2949 if (atomic_read(&(pkt_dev->skb->users)) != 1) {
2950 idle_start = getCurUs();
2951 while (atomic_read(&(pkt_dev->skb->users)) != 1) {
2952 if (signal_pending(current)) {
2953 break;
2955 schedule();
2957 pkt_dev->idle_acc += getCurUs() - idle_start;
2960 /* Done with this */
2961 pktgen_stop_device(pkt_dev);
2962 if (pkt_dev->skb)
2963 kfree_skb(pkt_dev->skb);
2964 pkt_dev->skb = NULL;
2966 out:;
2970 * Main loop of the thread goes here
2973 static void pktgen_thread_worker(struct pktgen_thread *t)
2975 DEFINE_WAIT(wait);
2976 struct pktgen_dev *pkt_dev = NULL;
2977 int cpu = t->cpu;
2978 sigset_t tmpsig;
2979 u32 max_before_softirq;
2980 u32 tx_since_softirq = 0;
2982 daemonize("pktgen/%d", cpu);
2984 /* Block all signals except SIGKILL, SIGSTOP and SIGTERM */
2986 spin_lock_irq(&current->sighand->siglock);
2987 tmpsig = current->blocked;
2988 siginitsetinv(&current->blocked,
2989 sigmask(SIGKILL) | sigmask(SIGSTOP) | sigmask(SIGTERM));
2991 recalc_sigpending();
2992 spin_unlock_irq(&current->sighand->siglock);
2994 /* Migrate to the right CPU */
2995 set_cpus_allowed(current, cpumask_of_cpu(cpu));
2996 if (smp_processor_id() != cpu)
2997 BUG();
2999 init_waitqueue_head(&t->queue);
3001 t->control &= ~(T_TERMINATE);
3002 t->control &= ~(T_RUN);
3003 t->control &= ~(T_STOP);
3004 t->control &= ~(T_REMDEVALL);
3005 t->control &= ~(T_REMDEV);
3007 t->pid = current->pid;
3009 PG_DEBUG(printk("pktgen: starting pktgen/%d: pid=%d\n", cpu, current->pid));
3011 max_before_softirq = t->max_before_softirq;
3013 __set_current_state(TASK_INTERRUPTIBLE);
3014 mb();
3016 while (1) {
3018 __set_current_state(TASK_RUNNING);
3021 * Get next dev to xmit -- if any.
3024 pkt_dev = next_to_run(t);
3026 if (pkt_dev) {
3028 pktgen_xmit(pkt_dev);
3031 * We like to stay RUNNING but must also give
3032 * others fair share.
3035 tx_since_softirq += pkt_dev->last_ok;
3037 if (tx_since_softirq > max_before_softirq) {
3038 if (local_softirq_pending())
3039 do_softirq();
3040 tx_since_softirq = 0;
3042 } else {
3043 prepare_to_wait(&(t->queue), &wait, TASK_INTERRUPTIBLE);
3044 schedule_timeout(HZ / 10);
3045 finish_wait(&(t->queue), &wait);
3049 * Back from sleep, either due to the timeout or signal.
3050 * We check if we have any "posted" work for us.
3053 if (t->control & T_TERMINATE || signal_pending(current))
3054 /* we received a request to terminate ourself */
3055 break;
3057 if (t->control & T_STOP) {
3058 pktgen_stop(t);
3059 t->control &= ~(T_STOP);
3062 if (t->control & T_RUN) {
3063 pktgen_run(t);
3064 t->control &= ~(T_RUN);
3067 if (t->control & T_REMDEVALL) {
3068 pktgen_rem_all_ifs(t);
3069 t->control &= ~(T_REMDEVALL);
3072 if (t->control & T_REMDEV) {
3073 pktgen_rem_one_if(t);
3074 t->control &= ~(T_REMDEV);
3077 if (need_resched())
3078 schedule();
3081 PG_DEBUG(printk("pktgen: %s stopping all device\n", t->name));
3082 pktgen_stop(t);
3084 PG_DEBUG(printk("pktgen: %s removing all device\n", t->name));
3085 pktgen_rem_all_ifs(t);
3087 PG_DEBUG(printk("pktgen: %s removing thread.\n", t->name));
3088 pktgen_rem_thread(t);
3090 t->removed = 1;
3093 static struct pktgen_dev *pktgen_find_dev(struct pktgen_thread *t,
3094 const char *ifname)
3096 struct pktgen_dev *p, *pkt_dev = NULL;
3097 if_lock(t);
3099 list_for_each_entry(p, &t->if_list, list)
3100 if (strncmp(p->ifname, ifname, IFNAMSIZ) == 0) {
3101 pkt_dev = p;
3102 break;
3105 if_unlock(t);
3106 PG_DEBUG(printk("pktgen: find_dev(%s) returning %p\n", ifname, pkt_dev));
3107 return pkt_dev;
3111 * Adds a dev at front of if_list.
3114 static int add_dev_to_thread(struct pktgen_thread *t,
3115 struct pktgen_dev *pkt_dev)
3117 int rv = 0;
3119 if_lock(t);
3121 if (pkt_dev->pg_thread) {
3122 printk("pktgen: ERROR: already assigned to a thread.\n");
3123 rv = -EBUSY;
3124 goto out;
3127 list_add(&pkt_dev->list, &t->if_list);
3128 pkt_dev->pg_thread = t;
3129 pkt_dev->running = 0;
3131 out:
3132 if_unlock(t);
3133 return rv;
3136 /* Called under thread lock */
3138 static int pktgen_add_device(struct pktgen_thread *t, const char *ifname)
3140 struct pktgen_dev *pkt_dev;
3141 struct proc_dir_entry *pe;
3143 /* We don't allow a device to be on several threads */
3145 pkt_dev = __pktgen_NN_threads(ifname, FIND);
3146 if (pkt_dev) {
3147 printk("pktgen: ERROR: interface already used.\n");
3148 return -EBUSY;
3151 pkt_dev = kzalloc(sizeof(struct pktgen_dev), GFP_KERNEL);
3152 if (!pkt_dev)
3153 return -ENOMEM;
3155 pkt_dev->flows = vmalloc(MAX_CFLOWS * sizeof(struct flow_state));
3156 if (pkt_dev->flows == NULL) {
3157 kfree(pkt_dev);
3158 return -ENOMEM;
3160 memset(pkt_dev->flows, 0, MAX_CFLOWS * sizeof(struct flow_state));
3162 pkt_dev->removal_mark = 0;
3163 pkt_dev->min_pkt_size = ETH_ZLEN;
3164 pkt_dev->max_pkt_size = ETH_ZLEN;
3165 pkt_dev->nfrags = 0;
3166 pkt_dev->clone_skb = pg_clone_skb_d;
3167 pkt_dev->delay_us = pg_delay_d / 1000;
3168 pkt_dev->delay_ns = pg_delay_d % 1000;
3169 pkt_dev->count = pg_count_d;
3170 pkt_dev->sofar = 0;
3171 pkt_dev->udp_src_min = 9; /* sink port */
3172 pkt_dev->udp_src_max = 9;
3173 pkt_dev->udp_dst_min = 9;
3174 pkt_dev->udp_dst_max = 9;
3176 strncpy(pkt_dev->ifname, ifname, IFNAMSIZ);
3178 if (!pktgen_setup_dev(pkt_dev)) {
3179 printk("pktgen: ERROR: pktgen_setup_dev failed.\n");
3180 if (pkt_dev->flows)
3181 vfree(pkt_dev->flows);
3182 kfree(pkt_dev);
3183 return -ENODEV;
3186 pe = create_proc_entry(ifname, 0600, pg_proc_dir);
3187 if (!pe) {
3188 printk("pktgen: cannot create %s/%s procfs entry.\n",
3189 PG_PROC_DIR, ifname);
3190 if (pkt_dev->flows)
3191 vfree(pkt_dev->flows);
3192 kfree(pkt_dev);
3193 return -EINVAL;
3195 pe->proc_fops = &pktgen_if_fops;
3196 pe->data = pkt_dev;
3198 return add_dev_to_thread(t, pkt_dev);
3201 static struct pktgen_thread *__init pktgen_find_thread(const char *name)
3203 struct pktgen_thread *t;
3205 mutex_lock(&pktgen_thread_lock);
3207 list_for_each_entry(t, &pktgen_threads, th_list)
3208 if (strcmp(t->name, name) == 0) {
3209 mutex_unlock(&pktgen_thread_lock);
3210 return t;
3213 mutex_unlock(&pktgen_thread_lock);
3214 return NULL;
3217 static int __init pktgen_create_thread(const char *name, int cpu)
3219 int err;
3220 struct pktgen_thread *t = NULL;
3221 struct proc_dir_entry *pe;
3223 if (strlen(name) > 31) {
3224 printk("pktgen: ERROR: Thread name cannot be more than 31 characters.\n");
3225 return -EINVAL;
3228 if (pktgen_find_thread(name)) {
3229 printk("pktgen: ERROR: thread: %s already exists\n", name);
3230 return -EINVAL;
3233 t = kzalloc(sizeof(struct pktgen_thread), GFP_KERNEL);
3234 if (!t) {
3235 printk("pktgen: ERROR: out of memory, can't create new thread.\n");
3236 return -ENOMEM;
3239 strcpy(t->name, name);
3240 spin_lock_init(&t->if_lock);
3241 t->cpu = cpu;
3243 pe = create_proc_entry(t->name, 0600, pg_proc_dir);
3244 if (!pe) {
3245 printk("pktgen: cannot create %s/%s procfs entry.\n",
3246 PG_PROC_DIR, t->name);
3247 kfree(t);
3248 return -EINVAL;
3251 pe->proc_fops = &pktgen_thread_fops;
3252 pe->data = t;
3254 INIT_LIST_HEAD(&t->if_list);
3256 list_add_tail(&t->th_list, &pktgen_threads);
3258 t->removed = 0;
3260 err = kernel_thread((void *)pktgen_thread_worker, (void *)t,
3261 CLONE_FS | CLONE_FILES | CLONE_SIGHAND);
3262 if (err < 0) {
3263 printk("pktgen: kernel_thread() failed for cpu %d\n", t->cpu);
3264 remove_proc_entry(t->name, pg_proc_dir);
3265 list_del(&t->th_list);
3266 kfree(t);
3267 return err;
3270 return 0;
3274 * Removes a device from the thread if_list.
3276 static void _rem_dev_from_if_list(struct pktgen_thread *t,
3277 struct pktgen_dev *pkt_dev)
3279 struct list_head *q, *n;
3280 struct pktgen_dev *p;
3282 list_for_each_safe(q, n, &t->if_list) {
3283 p = list_entry(q, struct pktgen_dev, list);
3284 if (p == pkt_dev)
3285 list_del(&p->list);
3289 static int pktgen_remove_device(struct pktgen_thread *t,
3290 struct pktgen_dev *pkt_dev)
3293 PG_DEBUG(printk("pktgen: remove_device pkt_dev=%p\n", pkt_dev));
3295 if (pkt_dev->running) {
3296 printk("pktgen:WARNING: trying to remove a running interface, stopping it now.\n");
3297 pktgen_stop_device(pkt_dev);
3300 /* Dis-associate from the interface */
3302 if (pkt_dev->odev) {
3303 dev_put(pkt_dev->odev);
3304 pkt_dev->odev = NULL;
3307 /* And update the thread if_list */
3309 _rem_dev_from_if_list(t, pkt_dev);
3311 /* Clean up proc file system */
3313 remove_proc_entry(pkt_dev->ifname, pg_proc_dir);
3315 if (pkt_dev->flows)
3316 vfree(pkt_dev->flows);
3317 kfree(pkt_dev);
3318 return 0;
3321 static int __init pg_init(void)
3323 int cpu;
3324 struct proc_dir_entry *pe;
3326 printk(version);
3328 pg_proc_dir = proc_mkdir(PG_PROC_DIR, proc_net);
3329 if (!pg_proc_dir)
3330 return -ENODEV;
3331 pg_proc_dir->owner = THIS_MODULE;
3333 pe = create_proc_entry(PGCTRL, 0600, pg_proc_dir);
3334 if (pe == NULL) {
3335 printk("pktgen: ERROR: cannot create %s procfs entry.\n",
3336 PGCTRL);
3337 proc_net_remove(PG_PROC_DIR);
3338 return -EINVAL;
3341 pe->proc_fops = &pktgen_fops;
3342 pe->data = NULL;
3344 /* Register us to receive netdevice events */
3345 register_netdevice_notifier(&pktgen_notifier_block);
3347 for_each_online_cpu(cpu) {
3348 int err;
3349 char buf[30];
3351 sprintf(buf, "kpktgend_%i", cpu);
3352 err = pktgen_create_thread(buf, cpu);
3353 if (err)
3354 printk("pktgen: WARNING: Cannot create thread for cpu %d (%d)\n",
3355 cpu, err);
3358 if (list_empty(&pktgen_threads)) {
3359 printk("pktgen: ERROR: Initialization failed for all threads\n");
3360 unregister_netdevice_notifier(&pktgen_notifier_block);
3361 remove_proc_entry(PGCTRL, pg_proc_dir);
3362 proc_net_remove(PG_PROC_DIR);
3363 return -ENODEV;
3366 return 0;
3369 static void __exit pg_cleanup(void)
3371 struct pktgen_thread *t;
3372 struct list_head *q, *n;
3373 wait_queue_head_t queue;
3374 init_waitqueue_head(&queue);
3376 /* Stop all interfaces & threads */
3378 list_for_each_safe(q, n, &pktgen_threads) {
3379 t = list_entry(q, struct pktgen_thread, th_list);
3380 t->control |= (T_TERMINATE);
3382 wait_event_interruptible_timeout(queue, (t->removed == 1), HZ);
3385 /* Un-register us from receiving netdevice events */
3386 unregister_netdevice_notifier(&pktgen_notifier_block);
3388 /* Clean up proc file system */
3389 remove_proc_entry(PGCTRL, pg_proc_dir);
3390 proc_net_remove(PG_PROC_DIR);
3393 module_init(pg_init);
3394 module_exit(pg_cleanup);
3396 MODULE_AUTHOR("Robert Olsson <robert.olsson@its.uu.se");
3397 MODULE_DESCRIPTION("Packet Generator tool");
3398 MODULE_LICENSE("GPL");
3399 module_param(pg_count_d, int, 0);
3400 module_param(pg_delay_d, int, 0);
3401 module_param(pg_clone_skb_d, int, 0);
3402 module_param(debug, int, 0);