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
54 * Also moved to /proc/net/pktgen/
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.
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
73 * The if_list is RCU protected, and the if_lock remains to protect updating
74 * of if_list, from "add_device" as it invoked from userspace (via proc write).
76 * By design there should only be *one* "controlling" process. In practice
77 * multiple write accesses gives unpredictable result. Understood by "write"
78 * to /proc gives result code thats should be read be the "writer".
79 * For practical use this should be no problem.
81 * Note when adding devices to a specific CPU there good idea to also assign
82 * /proc/irq/XX/smp_affinity so TX-interrupts gets bound to the same CPU.
85 * Fix refcount off by one if first packet fails, potential null deref,
88 * First "ranges" functionality for ipv6 030726 --ro
90 * Included flow support. 030802 ANK.
92 * Fixed unaligned access on IA-64 Grant Grundler <grundler@parisc-linux.org>
94 * Remove if fix from added Harald Welte <laforge@netfilter.org> 040419
95 * ia64 compilation fix from Aron Griffis <aron@hp.com> 040604
97 * New xmit() return, do_div and misc clean up by Stephen Hemminger
98 * <shemminger@osdl.org> 040923
100 * Randy Dunlap fixed u64 printk compiler waring
102 * Remove FCS from BW calculation. Lennert Buytenhek <buytenh@wantstofly.org>
103 * New time handling. Lennert Buytenhek <buytenh@wantstofly.org> 041213
105 * Corrections from Nikolai Malykh (nmalykh@bilim.com)
106 * Removed unused flags F_SET_SRCMAC & F_SET_SRCIP 041230
108 * interruptible_sleep_on_timeout() replaced Nishanth Aravamudan <nacc@us.ibm.com>
111 * MPLS support by Steven Whitehouse <steve@chygwyn.com>
113 * 802.1Q/Q-in-Q support by Francesco Fondelli (FF) <francesco.fondelli@gmail.com>
115 * Fixed src_mac command to set source mac of packet to value specified in
116 * command by Adit Ranadive <adit.262@gmail.com>
120 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
122 #include <linux/sys.h>
123 #include <linux/types.h>
124 #include <linux/module.h>
125 #include <linux/moduleparam.h>
126 #include <linux/kernel.h>
127 #include <linux/mutex.h>
128 #include <linux/sched.h>
129 #include <linux/slab.h>
130 #include <linux/vmalloc.h>
131 #include <linux/unistd.h>
132 #include <linux/string.h>
133 #include <linux/ptrace.h>
134 #include <linux/errno.h>
135 #include <linux/ioport.h>
136 #include <linux/interrupt.h>
137 #include <linux/capability.h>
138 #include <linux/hrtimer.h>
139 #include <linux/freezer.h>
140 #include <linux/delay.h>
141 #include <linux/timer.h>
142 #include <linux/list.h>
143 #include <linux/init.h>
144 #include <linux/skbuff.h>
145 #include <linux/netdevice.h>
146 #include <linux/inet.h>
147 #include <linux/inetdevice.h>
148 #include <linux/rtnetlink.h>
149 #include <linux/if_arp.h>
150 #include <linux/if_vlan.h>
151 #include <linux/in.h>
152 #include <linux/ip.h>
153 #include <linux/ipv6.h>
154 #include <linux/udp.h>
155 #include <linux/proc_fs.h>
156 #include <linux/seq_file.h>
157 #include <linux/wait.h>
158 #include <linux/etherdevice.h>
159 #include <linux/kthread.h>
160 #include <linux/prefetch.h>
161 #include <net/net_namespace.h>
162 #include <net/checksum.h>
163 #include <net/ipv6.h>
165 #include <net/ip6_checksum.h>
166 #include <net/addrconf.h>
168 #include <net/xfrm.h>
170 #include <net/netns/generic.h>
171 #include <asm/byteorder.h>
172 #include <linux/rcupdate.h>
173 #include <linux/bitops.h>
174 #include <linux/io.h>
175 #include <linux/timex.h>
176 #include <linux/uaccess.h>
178 #include <asm/div64.h> /* do_div */
180 #define VERSION "2.74"
181 #define IP_NAME_SZ 32
182 #define MAX_MPLS_LABELS 16 /* This is the max label stack depth */
183 #define MPLS_STACK_BOTTOM htonl(0x00000100)
185 #define func_enter() pr_debug("entering %s\n", __func__);
187 /* Device flag bits */
188 #define F_IPSRC_RND (1<<0) /* IP-Src Random */
189 #define F_IPDST_RND (1<<1) /* IP-Dst Random */
190 #define F_UDPSRC_RND (1<<2) /* UDP-Src Random */
191 #define F_UDPDST_RND (1<<3) /* UDP-Dst Random */
192 #define F_MACSRC_RND (1<<4) /* MAC-Src Random */
193 #define F_MACDST_RND (1<<5) /* MAC-Dst Random */
194 #define F_TXSIZE_RND (1<<6) /* Transmit size is random */
195 #define F_IPV6 (1<<7) /* Interface in IPV6 Mode */
196 #define F_MPLS_RND (1<<8) /* Random MPLS labels */
197 #define F_VID_RND (1<<9) /* Random VLAN ID */
198 #define F_SVID_RND (1<<10) /* Random SVLAN ID */
199 #define F_FLOW_SEQ (1<<11) /* Sequential flows */
200 #define F_IPSEC_ON (1<<12) /* ipsec on for flows */
201 #define F_QUEUE_MAP_RND (1<<13) /* queue map Random */
202 #define F_QUEUE_MAP_CPU (1<<14) /* queue map mirrors smp_processor_id() */
203 #define F_NODE (1<<15) /* Node memory alloc*/
204 #define F_UDPCSUM (1<<16) /* Include UDP checksum */
206 /* Thread control flag bits */
207 #define T_STOP (1<<0) /* Stop run */
208 #define T_RUN (1<<1) /* Start run */
209 #define T_REMDEVALL (1<<2) /* Remove all devs */
210 #define T_REMDEV (1<<3) /* Remove one dev */
212 /* If lock -- protects updating of if_list */
213 #define if_lock(t) spin_lock(&(t->if_lock));
214 #define if_unlock(t) spin_unlock(&(t->if_lock));
216 /* Used to help with determining the pkts on receive */
217 #define PKTGEN_MAGIC 0xbe9be955
218 #define PG_PROC_DIR "pktgen"
219 #define PGCTRL "pgctrl"
221 #define MAX_CFLOWS 65536
223 #define VLAN_TAG_SIZE(x) ((x)->vlan_id == 0xffff ? 0 : 4)
224 #define SVLAN_TAG_SIZE(x) ((x)->svlan_id == 0xffff ? 0 : 4)
230 struct xfrm_state *x;
236 #define F_INIT (1<<0) /* flow has been initialized */
240 * Try to keep frequent/infrequent used vars. separated.
242 struct proc_dir_entry *entry; /* proc file */
243 struct pktgen_thread *pg_thread;/* the owner */
244 struct list_head list; /* chaining in the thread's run-queue */
245 struct rcu_head rcu; /* freed by RCU */
247 int running; /* if false, the test will stop */
249 /* If min != max, then we will either do a linear iteration, or
250 * we will do a random selection from within the range.
253 int removal_mark; /* non-zero => the device is marked for
254 * removal by worker thread */
258 int pkt_overhead; /* overhead for MPLS, VLANs, IPSEC etc */
261 u64 delay; /* nano-seconds */
263 __u64 count; /* Default No packets to send */
264 __u64 sofar; /* How many pkts we've sent so far */
265 __u64 tx_bytes; /* How many bytes we've transmitted */
266 __u64 errors; /* Errors when trying to transmit, */
268 /* runtime counters relating to clone_skb */
270 __u64 allocated_skbs;
272 int last_ok; /* Was last skb sent?
273 * Or a failed transmit of some sort?
274 * This will keep sequence numbers in order
279 u64 idle_acc; /* nano-seconds */
284 * Use multiple SKBs during packet gen.
285 * If this number is greater than 1, then
286 * that many copies of the same packet will be
287 * sent before a new packet is allocated.
288 * If you want to send 1024 identical packets
289 * before creating a new packet,
290 * set clone_skb to 1024.
293 char dst_min[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
294 char dst_max[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
295 char src_min[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
296 char src_max[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
298 struct in6_addr in6_saddr;
299 struct in6_addr in6_daddr;
300 struct in6_addr cur_in6_daddr;
301 struct in6_addr cur_in6_saddr;
303 struct in6_addr min_in6_daddr;
304 struct in6_addr max_in6_daddr;
305 struct in6_addr min_in6_saddr;
306 struct in6_addr max_in6_saddr;
308 /* If we're doing ranges, random or incremental, then this
309 * defines the min/max for those ranges.
311 __be32 saddr_min; /* inclusive, source IP address */
312 __be32 saddr_max; /* exclusive, source IP address */
313 __be32 daddr_min; /* inclusive, dest IP address */
314 __be32 daddr_max; /* exclusive, dest IP address */
316 __u16 udp_src_min; /* inclusive, source UDP port */
317 __u16 udp_src_max; /* exclusive, source UDP port */
318 __u16 udp_dst_min; /* inclusive, dest UDP port */
319 __u16 udp_dst_max; /* exclusive, dest UDP port */
322 __u8 tos; /* six MSB of (former) IPv4 TOS
323 are for dscp codepoint */
324 __u8 traffic_class; /* ditto for the (former) Traffic Class in IPv6
325 (see RFC 3260, sec. 4) */
328 unsigned int nr_labels; /* Depth of stack, 0 = no MPLS */
329 __be32 labels[MAX_MPLS_LABELS];
331 /* VLAN/SVLAN (802.1Q/Q-in-Q) */
334 __u16 vlan_id; /* 0xffff means no vlan tag */
338 __u16 svlan_id; /* 0xffff means no svlan tag */
340 __u32 src_mac_count; /* How many MACs to iterate through */
341 __u32 dst_mac_count; /* How many MACs to iterate through */
343 unsigned char dst_mac[ETH_ALEN];
344 unsigned char src_mac[ETH_ALEN];
346 __u32 cur_dst_mac_offset;
347 __u32 cur_src_mac_offset;
359 0x00, 0x80, 0xC8, 0x79, 0xB3, 0xCB,
361 We fill in SRC address later
362 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
366 __u16 pad; /* pad out the hh struct to an even 16 bytes */
368 struct sk_buff *skb; /* skb we are to transmit next, used for when we
369 * are transmitting the same one multiple times
371 struct net_device *odev; /* The out-going device.
372 * Note that the device should have it's
373 * pg_info pointer pointing back to this
375 * Set when the user specifies the out-going
376 * device name (not when the inject is
377 * started as it used to do.)
380 struct flow_state *flows;
381 unsigned int cflows; /* Concurrent flows (config) */
382 unsigned int lflow; /* Flow length (config) */
383 unsigned int nflows; /* accumulated flows (stats) */
384 unsigned int curfl; /* current sequenced flow (state)*/
388 __u32 skb_priority; /* skb priority field */
389 int node; /* Memory node */
392 __u8 ipsmode; /* IPSEC mode (config) */
393 __u8 ipsproto; /* IPSEC type (config) */
395 struct dst_entry dst;
396 struct dst_ops dstops;
409 static int pg_net_id __read_mostly;
413 struct proc_dir_entry *proc_dir;
414 struct list_head pktgen_threads;
418 struct pktgen_thread {
419 spinlock_t if_lock; /* for list of devices */
420 struct list_head if_list; /* All device here */
421 struct list_head th_list;
422 struct task_struct *tsk;
425 /* Field for thread to receive "posted" events terminate,
431 wait_queue_head_t queue;
432 struct completion start_done;
433 struct pktgen_net *net;
439 static const char version[] =
440 "Packet Generator for packet performance testing. "
441 "Version: " VERSION "\n";
443 static int pktgen_remove_device(struct pktgen_thread *t, struct pktgen_dev *i);
444 static int pktgen_add_device(struct pktgen_thread *t, const char *ifname);
445 static struct pktgen_dev *pktgen_find_dev(struct pktgen_thread *t,
446 const char *ifname, bool exact);
447 static int pktgen_device_event(struct notifier_block *, unsigned long, void *);
448 static void pktgen_run_all_threads(struct pktgen_net *pn);
449 static void pktgen_reset_all_threads(struct pktgen_net *pn);
450 static void pktgen_stop_all_threads_ifs(struct pktgen_net *pn);
452 static void pktgen_stop(struct pktgen_thread *t);
453 static void pktgen_clear_counters(struct pktgen_dev *pkt_dev);
455 /* Module parameters, defaults. */
456 static int pg_count_d __read_mostly = 1000;
457 static int pg_delay_d __read_mostly;
458 static int pg_clone_skb_d __read_mostly;
459 static int debug __read_mostly;
461 static DEFINE_MUTEX(pktgen_thread_lock);
463 static struct notifier_block pktgen_notifier_block = {
464 .notifier_call = pktgen_device_event,
468 * /proc handling functions
472 static int pgctrl_show(struct seq_file *seq, void *v)
474 seq_puts(seq, version);
478 static ssize_t pgctrl_write(struct file *file, const char __user *buf,
479 size_t count, loff_t *ppos)
482 struct pktgen_net *pn = net_generic(current->nsproxy->net_ns, pg_net_id);
484 if (!capable(CAP_NET_ADMIN))
490 if (count > sizeof(data))
491 count = sizeof(data);
493 if (copy_from_user(data, buf, count))
496 data[count - 1] = 0; /* Strip trailing '\n' and terminate string */
498 if (!strcmp(data, "stop"))
499 pktgen_stop_all_threads_ifs(pn);
501 else if (!strcmp(data, "start"))
502 pktgen_run_all_threads(pn);
504 else if (!strcmp(data, "reset"))
505 pktgen_reset_all_threads(pn);
508 pr_warning("Unknown command: %s\n", data);
513 static int pgctrl_open(struct inode *inode, struct file *file)
515 return single_open(file, pgctrl_show, PDE_DATA(inode));
518 static const struct file_operations pktgen_fops = {
519 .owner = THIS_MODULE,
523 .write = pgctrl_write,
524 .release = single_release,
527 static int pktgen_if_show(struct seq_file *seq, void *v)
529 const struct pktgen_dev *pkt_dev = seq->private;
534 "Params: count %llu min_pkt_size: %u max_pkt_size: %u\n",
535 (unsigned long long)pkt_dev->count, pkt_dev->min_pkt_size,
536 pkt_dev->max_pkt_size);
539 " frags: %d delay: %llu clone_skb: %d ifname: %s\n",
540 pkt_dev->nfrags, (unsigned long long) pkt_dev->delay,
541 pkt_dev->clone_skb, pkt_dev->odevname);
543 seq_printf(seq, " flows: %u flowlen: %u\n", pkt_dev->cflows,
547 " queue_map_min: %u queue_map_max: %u\n",
548 pkt_dev->queue_map_min,
549 pkt_dev->queue_map_max);
551 if (pkt_dev->skb_priority)
552 seq_printf(seq, " skb_priority: %u\n",
553 pkt_dev->skb_priority);
555 if (pkt_dev->flags & F_IPV6) {
557 " saddr: %pI6c min_saddr: %pI6c max_saddr: %pI6c\n"
558 " daddr: %pI6c min_daddr: %pI6c max_daddr: %pI6c\n",
560 &pkt_dev->min_in6_saddr, &pkt_dev->max_in6_saddr,
562 &pkt_dev->min_in6_daddr, &pkt_dev->max_in6_daddr);
565 " dst_min: %s dst_max: %s\n",
566 pkt_dev->dst_min, pkt_dev->dst_max);
568 " src_min: %s src_max: %s\n",
569 pkt_dev->src_min, pkt_dev->src_max);
572 seq_puts(seq, " src_mac: ");
574 seq_printf(seq, "%pM ",
575 is_zero_ether_addr(pkt_dev->src_mac) ?
576 pkt_dev->odev->dev_addr : pkt_dev->src_mac);
578 seq_puts(seq, "dst_mac: ");
579 seq_printf(seq, "%pM\n", pkt_dev->dst_mac);
582 " udp_src_min: %d udp_src_max: %d"
583 " udp_dst_min: %d udp_dst_max: %d\n",
584 pkt_dev->udp_src_min, pkt_dev->udp_src_max,
585 pkt_dev->udp_dst_min, pkt_dev->udp_dst_max);
588 " src_mac_count: %d dst_mac_count: %d\n",
589 pkt_dev->src_mac_count, pkt_dev->dst_mac_count);
591 if (pkt_dev->nr_labels) {
593 seq_puts(seq, " mpls: ");
594 for (i = 0; i < pkt_dev->nr_labels; i++)
595 seq_printf(seq, "%08x%s", ntohl(pkt_dev->labels[i]),
596 i == pkt_dev->nr_labels-1 ? "\n" : ", ");
599 if (pkt_dev->vlan_id != 0xffff)
600 seq_printf(seq, " vlan_id: %u vlan_p: %u vlan_cfi: %u\n",
601 pkt_dev->vlan_id, pkt_dev->vlan_p,
604 if (pkt_dev->svlan_id != 0xffff)
605 seq_printf(seq, " svlan_id: %u vlan_p: %u vlan_cfi: %u\n",
606 pkt_dev->svlan_id, pkt_dev->svlan_p,
610 seq_printf(seq, " tos: 0x%02x\n", pkt_dev->tos);
612 if (pkt_dev->traffic_class)
613 seq_printf(seq, " traffic_class: 0x%02x\n", pkt_dev->traffic_class);
615 if (pkt_dev->node >= 0)
616 seq_printf(seq, " node: %d\n", pkt_dev->node);
618 seq_puts(seq, " Flags: ");
620 if (pkt_dev->flags & F_IPV6)
621 seq_puts(seq, "IPV6 ");
623 if (pkt_dev->flags & F_IPSRC_RND)
624 seq_puts(seq, "IPSRC_RND ");
626 if (pkt_dev->flags & F_IPDST_RND)
627 seq_puts(seq, "IPDST_RND ");
629 if (pkt_dev->flags & F_TXSIZE_RND)
630 seq_puts(seq, "TXSIZE_RND ");
632 if (pkt_dev->flags & F_UDPSRC_RND)
633 seq_puts(seq, "UDPSRC_RND ");
635 if (pkt_dev->flags & F_UDPDST_RND)
636 seq_puts(seq, "UDPDST_RND ");
638 if (pkt_dev->flags & F_UDPCSUM)
639 seq_puts(seq, "UDPCSUM ");
641 if (pkt_dev->flags & F_MPLS_RND)
642 seq_puts(seq, "MPLS_RND ");
644 if (pkt_dev->flags & F_QUEUE_MAP_RND)
645 seq_puts(seq, "QUEUE_MAP_RND ");
647 if (pkt_dev->flags & F_QUEUE_MAP_CPU)
648 seq_puts(seq, "QUEUE_MAP_CPU ");
650 if (pkt_dev->cflows) {
651 if (pkt_dev->flags & F_FLOW_SEQ)
652 seq_puts(seq, "FLOW_SEQ "); /*in sequence flows*/
654 seq_puts(seq, "FLOW_RND ");
658 if (pkt_dev->flags & F_IPSEC_ON) {
659 seq_puts(seq, "IPSEC ");
661 seq_printf(seq, "spi:%u", pkt_dev->spi);
665 if (pkt_dev->flags & F_MACSRC_RND)
666 seq_puts(seq, "MACSRC_RND ");
668 if (pkt_dev->flags & F_MACDST_RND)
669 seq_puts(seq, "MACDST_RND ");
671 if (pkt_dev->flags & F_VID_RND)
672 seq_puts(seq, "VID_RND ");
674 if (pkt_dev->flags & F_SVID_RND)
675 seq_puts(seq, "SVID_RND ");
677 if (pkt_dev->flags & F_NODE)
678 seq_puts(seq, "NODE_ALLOC ");
682 /* not really stopped, more like last-running-at */
683 stopped = pkt_dev->running ? ktime_get() : pkt_dev->stopped_at;
684 idle = pkt_dev->idle_acc;
685 do_div(idle, NSEC_PER_USEC);
688 "Current:\n pkts-sofar: %llu errors: %llu\n",
689 (unsigned long long)pkt_dev->sofar,
690 (unsigned long long)pkt_dev->errors);
693 " started: %lluus stopped: %lluus idle: %lluus\n",
694 (unsigned long long) ktime_to_us(pkt_dev->started_at),
695 (unsigned long long) ktime_to_us(stopped),
696 (unsigned long long) idle);
699 " seq_num: %d cur_dst_mac_offset: %d cur_src_mac_offset: %d\n",
700 pkt_dev->seq_num, pkt_dev->cur_dst_mac_offset,
701 pkt_dev->cur_src_mac_offset);
703 if (pkt_dev->flags & F_IPV6) {
704 seq_printf(seq, " cur_saddr: %pI6c cur_daddr: %pI6c\n",
705 &pkt_dev->cur_in6_saddr,
706 &pkt_dev->cur_in6_daddr);
708 seq_printf(seq, " cur_saddr: %pI4 cur_daddr: %pI4\n",
709 &pkt_dev->cur_saddr, &pkt_dev->cur_daddr);
711 seq_printf(seq, " cur_udp_dst: %d cur_udp_src: %d\n",
712 pkt_dev->cur_udp_dst, pkt_dev->cur_udp_src);
714 seq_printf(seq, " cur_queue_map: %u\n", pkt_dev->cur_queue_map);
716 seq_printf(seq, " flows: %u\n", pkt_dev->nflows);
718 if (pkt_dev->result[0])
719 seq_printf(seq, "Result: %s\n", pkt_dev->result);
721 seq_puts(seq, "Result: Idle\n");
727 static int hex32_arg(const char __user *user_buffer, unsigned long maxlen,
733 for (; i < maxlen; i++) {
737 if (get_user(c, &user_buffer[i]))
739 value = hex_to_bin(c);
748 static int count_trail_chars(const char __user * user_buffer,
753 for (i = 0; i < maxlen; i++) {
755 if (get_user(c, &user_buffer[i]))
773 static long num_arg(const char __user *user_buffer, unsigned long maxlen,
779 for (i = 0; i < maxlen; i++) {
781 if (get_user(c, &user_buffer[i]))
783 if ((c >= '0') && (c <= '9')) {
792 static int strn_len(const char __user * user_buffer, unsigned int maxlen)
796 for (i = 0; i < maxlen; i++) {
798 if (get_user(c, &user_buffer[i]))
816 static ssize_t get_labels(const char __user *buffer, struct pktgen_dev *pkt_dev)
823 pkt_dev->nr_labels = 0;
826 len = hex32_arg(&buffer[i], 8, &tmp);
829 pkt_dev->labels[n] = htonl(tmp);
830 if (pkt_dev->labels[n] & MPLS_STACK_BOTTOM)
831 pkt_dev->flags |= F_MPLS_RND;
833 if (get_user(c, &buffer[i]))
837 if (n >= MAX_MPLS_LABELS)
841 pkt_dev->nr_labels = n;
845 static ssize_t pktgen_if_write(struct file *file,
846 const char __user * user_buffer, size_t count,
849 struct seq_file *seq = file->private_data;
850 struct pktgen_dev *pkt_dev = seq->private;
852 char name[16], valstr[32];
853 unsigned long value = 0;
854 char *pg_result = NULL;
858 pg_result = &(pkt_dev->result[0]);
861 pr_warning("wrong command format\n");
866 tmp = count_trail_chars(user_buffer, max);
868 pr_warning("illegal format\n");
873 /* Read variable name */
875 len = strn_len(&user_buffer[i], sizeof(name) - 1);
879 memset(name, 0, sizeof(name));
880 if (copy_from_user(name, &user_buffer[i], len))
885 len = count_trail_chars(&user_buffer[i], max);
892 size_t copy = min_t(size_t, count, 1023);
894 if (copy_from_user(tb, user_buffer, copy))
897 pr_debug("%s,%lu buffer -:%s:-\n",
898 name, (unsigned long)count, tb);
901 if (!strcmp(name, "min_pkt_size")) {
902 len = num_arg(&user_buffer[i], 10, &value);
907 if (value < 14 + 20 + 8)
909 if (value != pkt_dev->min_pkt_size) {
910 pkt_dev->min_pkt_size = value;
911 pkt_dev->cur_pkt_size = value;
913 sprintf(pg_result, "OK: min_pkt_size=%u",
914 pkt_dev->min_pkt_size);
918 if (!strcmp(name, "max_pkt_size")) {
919 len = num_arg(&user_buffer[i], 10, &value);
924 if (value < 14 + 20 + 8)
926 if (value != pkt_dev->max_pkt_size) {
927 pkt_dev->max_pkt_size = value;
928 pkt_dev->cur_pkt_size = value;
930 sprintf(pg_result, "OK: max_pkt_size=%u",
931 pkt_dev->max_pkt_size);
935 /* Shortcut for min = max */
937 if (!strcmp(name, "pkt_size")) {
938 len = num_arg(&user_buffer[i], 10, &value);
943 if (value < 14 + 20 + 8)
945 if (value != pkt_dev->min_pkt_size) {
946 pkt_dev->min_pkt_size = value;
947 pkt_dev->max_pkt_size = value;
948 pkt_dev->cur_pkt_size = value;
950 sprintf(pg_result, "OK: pkt_size=%u", pkt_dev->min_pkt_size);
954 if (!strcmp(name, "debug")) {
955 len = num_arg(&user_buffer[i], 10, &value);
961 sprintf(pg_result, "OK: debug=%u", debug);
965 if (!strcmp(name, "frags")) {
966 len = num_arg(&user_buffer[i], 10, &value);
971 pkt_dev->nfrags = value;
972 sprintf(pg_result, "OK: frags=%u", pkt_dev->nfrags);
975 if (!strcmp(name, "delay")) {
976 len = num_arg(&user_buffer[i], 10, &value);
981 if (value == 0x7FFFFFFF)
982 pkt_dev->delay = ULLONG_MAX;
984 pkt_dev->delay = (u64)value;
986 sprintf(pg_result, "OK: delay=%llu",
987 (unsigned long long) pkt_dev->delay);
990 if (!strcmp(name, "rate")) {
991 len = num_arg(&user_buffer[i], 10, &value);
998 pkt_dev->delay = pkt_dev->min_pkt_size*8*NSEC_PER_USEC/value;
1000 pr_info("Delay set at: %llu ns\n", pkt_dev->delay);
1002 sprintf(pg_result, "OK: rate=%lu", value);
1005 if (!strcmp(name, "ratep")) {
1006 len = num_arg(&user_buffer[i], 10, &value);
1013 pkt_dev->delay = NSEC_PER_SEC/value;
1015 pr_info("Delay set at: %llu ns\n", pkt_dev->delay);
1017 sprintf(pg_result, "OK: rate=%lu", value);
1020 if (!strcmp(name, "udp_src_min")) {
1021 len = num_arg(&user_buffer[i], 10, &value);
1026 if (value != pkt_dev->udp_src_min) {
1027 pkt_dev->udp_src_min = value;
1028 pkt_dev->cur_udp_src = value;
1030 sprintf(pg_result, "OK: udp_src_min=%u", pkt_dev->udp_src_min);
1033 if (!strcmp(name, "udp_dst_min")) {
1034 len = num_arg(&user_buffer[i], 10, &value);
1039 if (value != pkt_dev->udp_dst_min) {
1040 pkt_dev->udp_dst_min = value;
1041 pkt_dev->cur_udp_dst = value;
1043 sprintf(pg_result, "OK: udp_dst_min=%u", pkt_dev->udp_dst_min);
1046 if (!strcmp(name, "udp_src_max")) {
1047 len = num_arg(&user_buffer[i], 10, &value);
1052 if (value != pkt_dev->udp_src_max) {
1053 pkt_dev->udp_src_max = value;
1054 pkt_dev->cur_udp_src = value;
1056 sprintf(pg_result, "OK: udp_src_max=%u", pkt_dev->udp_src_max);
1059 if (!strcmp(name, "udp_dst_max")) {
1060 len = num_arg(&user_buffer[i], 10, &value);
1065 if (value != pkt_dev->udp_dst_max) {
1066 pkt_dev->udp_dst_max = value;
1067 pkt_dev->cur_udp_dst = value;
1069 sprintf(pg_result, "OK: udp_dst_max=%u", pkt_dev->udp_dst_max);
1072 if (!strcmp(name, "clone_skb")) {
1073 len = num_arg(&user_buffer[i], 10, &value);
1077 (!(pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING)))
1080 pkt_dev->clone_skb = value;
1082 sprintf(pg_result, "OK: clone_skb=%d", pkt_dev->clone_skb);
1085 if (!strcmp(name, "count")) {
1086 len = num_arg(&user_buffer[i], 10, &value);
1091 pkt_dev->count = value;
1092 sprintf(pg_result, "OK: count=%llu",
1093 (unsigned long long)pkt_dev->count);
1096 if (!strcmp(name, "src_mac_count")) {
1097 len = num_arg(&user_buffer[i], 10, &value);
1102 if (pkt_dev->src_mac_count != value) {
1103 pkt_dev->src_mac_count = value;
1104 pkt_dev->cur_src_mac_offset = 0;
1106 sprintf(pg_result, "OK: src_mac_count=%d",
1107 pkt_dev->src_mac_count);
1110 if (!strcmp(name, "dst_mac_count")) {
1111 len = num_arg(&user_buffer[i], 10, &value);
1116 if (pkt_dev->dst_mac_count != value) {
1117 pkt_dev->dst_mac_count = value;
1118 pkt_dev->cur_dst_mac_offset = 0;
1120 sprintf(pg_result, "OK: dst_mac_count=%d",
1121 pkt_dev->dst_mac_count);
1124 if (!strcmp(name, "node")) {
1125 len = num_arg(&user_buffer[i], 10, &value);
1131 if (node_possible(value)) {
1132 pkt_dev->node = value;
1133 sprintf(pg_result, "OK: node=%d", pkt_dev->node);
1134 if (pkt_dev->page) {
1135 put_page(pkt_dev->page);
1136 pkt_dev->page = NULL;
1140 sprintf(pg_result, "ERROR: node not possible");
1143 if (!strcmp(name, "flag")) {
1146 len = strn_len(&user_buffer[i], sizeof(f) - 1);
1150 if (copy_from_user(f, &user_buffer[i], len))
1153 if (strcmp(f, "IPSRC_RND") == 0)
1154 pkt_dev->flags |= F_IPSRC_RND;
1156 else if (strcmp(f, "!IPSRC_RND") == 0)
1157 pkt_dev->flags &= ~F_IPSRC_RND;
1159 else if (strcmp(f, "TXSIZE_RND") == 0)
1160 pkt_dev->flags |= F_TXSIZE_RND;
1162 else if (strcmp(f, "!TXSIZE_RND") == 0)
1163 pkt_dev->flags &= ~F_TXSIZE_RND;
1165 else if (strcmp(f, "IPDST_RND") == 0)
1166 pkt_dev->flags |= F_IPDST_RND;
1168 else if (strcmp(f, "!IPDST_RND") == 0)
1169 pkt_dev->flags &= ~F_IPDST_RND;
1171 else if (strcmp(f, "UDPSRC_RND") == 0)
1172 pkt_dev->flags |= F_UDPSRC_RND;
1174 else if (strcmp(f, "!UDPSRC_RND") == 0)
1175 pkt_dev->flags &= ~F_UDPSRC_RND;
1177 else if (strcmp(f, "UDPDST_RND") == 0)
1178 pkt_dev->flags |= F_UDPDST_RND;
1180 else if (strcmp(f, "!UDPDST_RND") == 0)
1181 pkt_dev->flags &= ~F_UDPDST_RND;
1183 else if (strcmp(f, "MACSRC_RND") == 0)
1184 pkt_dev->flags |= F_MACSRC_RND;
1186 else if (strcmp(f, "!MACSRC_RND") == 0)
1187 pkt_dev->flags &= ~F_MACSRC_RND;
1189 else if (strcmp(f, "MACDST_RND") == 0)
1190 pkt_dev->flags |= F_MACDST_RND;
1192 else if (strcmp(f, "!MACDST_RND") == 0)
1193 pkt_dev->flags &= ~F_MACDST_RND;
1195 else if (strcmp(f, "MPLS_RND") == 0)
1196 pkt_dev->flags |= F_MPLS_RND;
1198 else if (strcmp(f, "!MPLS_RND") == 0)
1199 pkt_dev->flags &= ~F_MPLS_RND;
1201 else if (strcmp(f, "VID_RND") == 0)
1202 pkt_dev->flags |= F_VID_RND;
1204 else if (strcmp(f, "!VID_RND") == 0)
1205 pkt_dev->flags &= ~F_VID_RND;
1207 else if (strcmp(f, "SVID_RND") == 0)
1208 pkt_dev->flags |= F_SVID_RND;
1210 else if (strcmp(f, "!SVID_RND") == 0)
1211 pkt_dev->flags &= ~F_SVID_RND;
1213 else if (strcmp(f, "FLOW_SEQ") == 0)
1214 pkt_dev->flags |= F_FLOW_SEQ;
1216 else if (strcmp(f, "QUEUE_MAP_RND") == 0)
1217 pkt_dev->flags |= F_QUEUE_MAP_RND;
1219 else if (strcmp(f, "!QUEUE_MAP_RND") == 0)
1220 pkt_dev->flags &= ~F_QUEUE_MAP_RND;
1222 else if (strcmp(f, "QUEUE_MAP_CPU") == 0)
1223 pkt_dev->flags |= F_QUEUE_MAP_CPU;
1225 else if (strcmp(f, "!QUEUE_MAP_CPU") == 0)
1226 pkt_dev->flags &= ~F_QUEUE_MAP_CPU;
1228 else if (strcmp(f, "IPSEC") == 0)
1229 pkt_dev->flags |= F_IPSEC_ON;
1232 else if (strcmp(f, "!IPV6") == 0)
1233 pkt_dev->flags &= ~F_IPV6;
1235 else if (strcmp(f, "NODE_ALLOC") == 0)
1236 pkt_dev->flags |= F_NODE;
1238 else if (strcmp(f, "!NODE_ALLOC") == 0)
1239 pkt_dev->flags &= ~F_NODE;
1241 else if (strcmp(f, "UDPCSUM") == 0)
1242 pkt_dev->flags |= F_UDPCSUM;
1244 else if (strcmp(f, "!UDPCSUM") == 0)
1245 pkt_dev->flags &= ~F_UDPCSUM;
1249 "Flag -:%s:- unknown\nAvailable flags, (prepend ! to un-set flag):\n%s",
1251 "IPSRC_RND, IPDST_RND, UDPSRC_RND, UDPDST_RND, "
1252 "MACSRC_RND, MACDST_RND, TXSIZE_RND, IPV6, "
1253 "MPLS_RND, VID_RND, SVID_RND, FLOW_SEQ, "
1254 "QUEUE_MAP_RND, QUEUE_MAP_CPU, UDPCSUM, "
1261 sprintf(pg_result, "OK: flags=0x%x", pkt_dev->flags);
1264 if (!strcmp(name, "dst_min") || !strcmp(name, "dst")) {
1265 len = strn_len(&user_buffer[i], sizeof(pkt_dev->dst_min) - 1);
1269 if (copy_from_user(buf, &user_buffer[i], len))
1272 if (strcmp(buf, pkt_dev->dst_min) != 0) {
1273 memset(pkt_dev->dst_min, 0, sizeof(pkt_dev->dst_min));
1274 strncpy(pkt_dev->dst_min, buf, len);
1275 pkt_dev->daddr_min = in_aton(pkt_dev->dst_min);
1276 pkt_dev->cur_daddr = pkt_dev->daddr_min;
1279 pr_debug("dst_min set to: %s\n", pkt_dev->dst_min);
1281 sprintf(pg_result, "OK: dst_min=%s", pkt_dev->dst_min);
1284 if (!strcmp(name, "dst_max")) {
1285 len = strn_len(&user_buffer[i], sizeof(pkt_dev->dst_max) - 1);
1290 if (copy_from_user(buf, &user_buffer[i], len))
1294 if (strcmp(buf, pkt_dev->dst_max) != 0) {
1295 memset(pkt_dev->dst_max, 0, sizeof(pkt_dev->dst_max));
1296 strncpy(pkt_dev->dst_max, buf, len);
1297 pkt_dev->daddr_max = in_aton(pkt_dev->dst_max);
1298 pkt_dev->cur_daddr = pkt_dev->daddr_max;
1301 pr_debug("dst_max set to: %s\n", pkt_dev->dst_max);
1303 sprintf(pg_result, "OK: dst_max=%s", pkt_dev->dst_max);
1306 if (!strcmp(name, "dst6")) {
1307 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1311 pkt_dev->flags |= F_IPV6;
1313 if (copy_from_user(buf, &user_buffer[i], len))
1317 in6_pton(buf, -1, pkt_dev->in6_daddr.s6_addr, -1, NULL);
1318 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->in6_daddr);
1320 pkt_dev->cur_in6_daddr = pkt_dev->in6_daddr;
1323 pr_debug("dst6 set to: %s\n", buf);
1326 sprintf(pg_result, "OK: dst6=%s", buf);
1329 if (!strcmp(name, "dst6_min")) {
1330 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1334 pkt_dev->flags |= F_IPV6;
1336 if (copy_from_user(buf, &user_buffer[i], len))
1340 in6_pton(buf, -1, pkt_dev->min_in6_daddr.s6_addr, -1, NULL);
1341 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->min_in6_daddr);
1343 pkt_dev->cur_in6_daddr = pkt_dev->min_in6_daddr;
1345 pr_debug("dst6_min set to: %s\n", buf);
1348 sprintf(pg_result, "OK: dst6_min=%s", buf);
1351 if (!strcmp(name, "dst6_max")) {
1352 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1356 pkt_dev->flags |= F_IPV6;
1358 if (copy_from_user(buf, &user_buffer[i], len))
1362 in6_pton(buf, -1, pkt_dev->max_in6_daddr.s6_addr, -1, NULL);
1363 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->max_in6_daddr);
1366 pr_debug("dst6_max set to: %s\n", buf);
1369 sprintf(pg_result, "OK: dst6_max=%s", buf);
1372 if (!strcmp(name, "src6")) {
1373 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1377 pkt_dev->flags |= F_IPV6;
1379 if (copy_from_user(buf, &user_buffer[i], len))
1383 in6_pton(buf, -1, pkt_dev->in6_saddr.s6_addr, -1, NULL);
1384 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->in6_saddr);
1386 pkt_dev->cur_in6_saddr = pkt_dev->in6_saddr;
1389 pr_debug("src6 set to: %s\n", buf);
1392 sprintf(pg_result, "OK: src6=%s", buf);
1395 if (!strcmp(name, "src_min")) {
1396 len = strn_len(&user_buffer[i], sizeof(pkt_dev->src_min) - 1);
1400 if (copy_from_user(buf, &user_buffer[i], len))
1403 if (strcmp(buf, pkt_dev->src_min) != 0) {
1404 memset(pkt_dev->src_min, 0, sizeof(pkt_dev->src_min));
1405 strncpy(pkt_dev->src_min, buf, len);
1406 pkt_dev->saddr_min = in_aton(pkt_dev->src_min);
1407 pkt_dev->cur_saddr = pkt_dev->saddr_min;
1410 pr_debug("src_min set to: %s\n", pkt_dev->src_min);
1412 sprintf(pg_result, "OK: src_min=%s", pkt_dev->src_min);
1415 if (!strcmp(name, "src_max")) {
1416 len = strn_len(&user_buffer[i], sizeof(pkt_dev->src_max) - 1);
1420 if (copy_from_user(buf, &user_buffer[i], len))
1423 if (strcmp(buf, pkt_dev->src_max) != 0) {
1424 memset(pkt_dev->src_max, 0, sizeof(pkt_dev->src_max));
1425 strncpy(pkt_dev->src_max, buf, len);
1426 pkt_dev->saddr_max = in_aton(pkt_dev->src_max);
1427 pkt_dev->cur_saddr = pkt_dev->saddr_max;
1430 pr_debug("src_max set to: %s\n", pkt_dev->src_max);
1432 sprintf(pg_result, "OK: src_max=%s", pkt_dev->src_max);
1435 if (!strcmp(name, "dst_mac")) {
1436 len = strn_len(&user_buffer[i], sizeof(valstr) - 1);
1440 memset(valstr, 0, sizeof(valstr));
1441 if (copy_from_user(valstr, &user_buffer[i], len))
1444 if (!mac_pton(valstr, pkt_dev->dst_mac))
1446 /* Set up Dest MAC */
1447 ether_addr_copy(&pkt_dev->hh[0], pkt_dev->dst_mac);
1449 sprintf(pg_result, "OK: dstmac %pM", pkt_dev->dst_mac);
1452 if (!strcmp(name, "src_mac")) {
1453 len = strn_len(&user_buffer[i], sizeof(valstr) - 1);
1457 memset(valstr, 0, sizeof(valstr));
1458 if (copy_from_user(valstr, &user_buffer[i], len))
1461 if (!mac_pton(valstr, pkt_dev->src_mac))
1463 /* Set up Src MAC */
1464 ether_addr_copy(&pkt_dev->hh[6], pkt_dev->src_mac);
1466 sprintf(pg_result, "OK: srcmac %pM", pkt_dev->src_mac);
1470 if (!strcmp(name, "clear_counters")) {
1471 pktgen_clear_counters(pkt_dev);
1472 sprintf(pg_result, "OK: Clearing counters.\n");
1476 if (!strcmp(name, "flows")) {
1477 len = num_arg(&user_buffer[i], 10, &value);
1482 if (value > MAX_CFLOWS)
1485 pkt_dev->cflows = value;
1486 sprintf(pg_result, "OK: flows=%u", pkt_dev->cflows);
1490 if (!strcmp(name, "spi")) {
1491 len = num_arg(&user_buffer[i], 10, &value);
1496 pkt_dev->spi = value;
1497 sprintf(pg_result, "OK: spi=%u", pkt_dev->spi);
1501 if (!strcmp(name, "flowlen")) {
1502 len = num_arg(&user_buffer[i], 10, &value);
1507 pkt_dev->lflow = value;
1508 sprintf(pg_result, "OK: flowlen=%u", pkt_dev->lflow);
1512 if (!strcmp(name, "queue_map_min")) {
1513 len = num_arg(&user_buffer[i], 5, &value);
1518 pkt_dev->queue_map_min = value;
1519 sprintf(pg_result, "OK: queue_map_min=%u", pkt_dev->queue_map_min);
1523 if (!strcmp(name, "queue_map_max")) {
1524 len = num_arg(&user_buffer[i], 5, &value);
1529 pkt_dev->queue_map_max = value;
1530 sprintf(pg_result, "OK: queue_map_max=%u", pkt_dev->queue_map_max);
1534 if (!strcmp(name, "mpls")) {
1535 unsigned int n, cnt;
1537 len = get_labels(&user_buffer[i], pkt_dev);
1541 cnt = sprintf(pg_result, "OK: mpls=");
1542 for (n = 0; n < pkt_dev->nr_labels; n++)
1543 cnt += sprintf(pg_result + cnt,
1544 "%08x%s", ntohl(pkt_dev->labels[n]),
1545 n == pkt_dev->nr_labels-1 ? "" : ",");
1547 if (pkt_dev->nr_labels && pkt_dev->vlan_id != 0xffff) {
1548 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */
1549 pkt_dev->svlan_id = 0xffff;
1552 pr_debug("VLAN/SVLAN auto turned off\n");
1557 if (!strcmp(name, "vlan_id")) {
1558 len = num_arg(&user_buffer[i], 4, &value);
1563 if (value <= 4095) {
1564 pkt_dev->vlan_id = value; /* turn on VLAN */
1567 pr_debug("VLAN turned on\n");
1569 if (debug && pkt_dev->nr_labels)
1570 pr_debug("MPLS auto turned off\n");
1572 pkt_dev->nr_labels = 0; /* turn off MPLS */
1573 sprintf(pg_result, "OK: vlan_id=%u", pkt_dev->vlan_id);
1575 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */
1576 pkt_dev->svlan_id = 0xffff;
1579 pr_debug("VLAN/SVLAN turned off\n");
1584 if (!strcmp(name, "vlan_p")) {
1585 len = num_arg(&user_buffer[i], 1, &value);
1590 if ((value <= 7) && (pkt_dev->vlan_id != 0xffff)) {
1591 pkt_dev->vlan_p = value;
1592 sprintf(pg_result, "OK: vlan_p=%u", pkt_dev->vlan_p);
1594 sprintf(pg_result, "ERROR: vlan_p must be 0-7");
1599 if (!strcmp(name, "vlan_cfi")) {
1600 len = num_arg(&user_buffer[i], 1, &value);
1605 if ((value <= 1) && (pkt_dev->vlan_id != 0xffff)) {
1606 pkt_dev->vlan_cfi = value;
1607 sprintf(pg_result, "OK: vlan_cfi=%u", pkt_dev->vlan_cfi);
1609 sprintf(pg_result, "ERROR: vlan_cfi must be 0-1");
1614 if (!strcmp(name, "svlan_id")) {
1615 len = num_arg(&user_buffer[i], 4, &value);
1620 if ((value <= 4095) && ((pkt_dev->vlan_id != 0xffff))) {
1621 pkt_dev->svlan_id = value; /* turn on SVLAN */
1624 pr_debug("SVLAN turned on\n");
1626 if (debug && pkt_dev->nr_labels)
1627 pr_debug("MPLS auto turned off\n");
1629 pkt_dev->nr_labels = 0; /* turn off MPLS */
1630 sprintf(pg_result, "OK: svlan_id=%u", pkt_dev->svlan_id);
1632 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */
1633 pkt_dev->svlan_id = 0xffff;
1636 pr_debug("VLAN/SVLAN turned off\n");
1641 if (!strcmp(name, "svlan_p")) {
1642 len = num_arg(&user_buffer[i], 1, &value);
1647 if ((value <= 7) && (pkt_dev->svlan_id != 0xffff)) {
1648 pkt_dev->svlan_p = value;
1649 sprintf(pg_result, "OK: svlan_p=%u", pkt_dev->svlan_p);
1651 sprintf(pg_result, "ERROR: svlan_p must be 0-7");
1656 if (!strcmp(name, "svlan_cfi")) {
1657 len = num_arg(&user_buffer[i], 1, &value);
1662 if ((value <= 1) && (pkt_dev->svlan_id != 0xffff)) {
1663 pkt_dev->svlan_cfi = value;
1664 sprintf(pg_result, "OK: svlan_cfi=%u", pkt_dev->svlan_cfi);
1666 sprintf(pg_result, "ERROR: svlan_cfi must be 0-1");
1671 if (!strcmp(name, "tos")) {
1672 __u32 tmp_value = 0;
1673 len = hex32_arg(&user_buffer[i], 2, &tmp_value);
1679 pkt_dev->tos = tmp_value;
1680 sprintf(pg_result, "OK: tos=0x%02x", pkt_dev->tos);
1682 sprintf(pg_result, "ERROR: tos must be 00-ff");
1687 if (!strcmp(name, "traffic_class")) {
1688 __u32 tmp_value = 0;
1689 len = hex32_arg(&user_buffer[i], 2, &tmp_value);
1695 pkt_dev->traffic_class = tmp_value;
1696 sprintf(pg_result, "OK: traffic_class=0x%02x", pkt_dev->traffic_class);
1698 sprintf(pg_result, "ERROR: traffic_class must be 00-ff");
1703 if (!strcmp(name, "skb_priority")) {
1704 len = num_arg(&user_buffer[i], 9, &value);
1709 pkt_dev->skb_priority = value;
1710 sprintf(pg_result, "OK: skb_priority=%i",
1711 pkt_dev->skb_priority);
1715 sprintf(pkt_dev->result, "No such parameter \"%s\"", name);
1719 static int pktgen_if_open(struct inode *inode, struct file *file)
1721 return single_open(file, pktgen_if_show, PDE_DATA(inode));
1724 static const struct file_operations pktgen_if_fops = {
1725 .owner = THIS_MODULE,
1726 .open = pktgen_if_open,
1728 .llseek = seq_lseek,
1729 .write = pktgen_if_write,
1730 .release = single_release,
1733 static int pktgen_thread_show(struct seq_file *seq, void *v)
1735 struct pktgen_thread *t = seq->private;
1736 const struct pktgen_dev *pkt_dev;
1740 seq_puts(seq, "Running: ");
1743 list_for_each_entry_rcu(pkt_dev, &t->if_list, list)
1744 if (pkt_dev->running)
1745 seq_printf(seq, "%s ", pkt_dev->odevname);
1747 seq_puts(seq, "\nStopped: ");
1749 list_for_each_entry_rcu(pkt_dev, &t->if_list, list)
1750 if (!pkt_dev->running)
1751 seq_printf(seq, "%s ", pkt_dev->odevname);
1754 seq_printf(seq, "\nResult: %s\n", t->result);
1756 seq_puts(seq, "\nResult: NA\n");
1763 static ssize_t pktgen_thread_write(struct file *file,
1764 const char __user * user_buffer,
1765 size_t count, loff_t * offset)
1767 struct seq_file *seq = file->private_data;
1768 struct pktgen_thread *t = seq->private;
1769 int i, max, len, ret;
1774 // sprintf(pg_result, "Wrong command format");
1779 len = count_trail_chars(user_buffer, max);
1785 /* Read variable name */
1787 len = strn_len(&user_buffer[i], sizeof(name) - 1);
1791 memset(name, 0, sizeof(name));
1792 if (copy_from_user(name, &user_buffer[i], len))
1797 len = count_trail_chars(&user_buffer[i], max);
1804 pr_debug("t=%s, count=%lu\n", name, (unsigned long)count);
1807 pr_err("ERROR: No thread\n");
1812 pg_result = &(t->result[0]);
1814 if (!strcmp(name, "add_device")) {
1817 len = strn_len(&user_buffer[i], sizeof(f) - 1);
1822 if (copy_from_user(f, &user_buffer[i], len))
1825 mutex_lock(&pktgen_thread_lock);
1826 ret = pktgen_add_device(t, f);
1827 mutex_unlock(&pktgen_thread_lock);
1830 sprintf(pg_result, "OK: add_device=%s", f);
1832 sprintf(pg_result, "ERROR: can not add device %s", f);
1836 if (!strcmp(name, "rem_device_all")) {
1837 mutex_lock(&pktgen_thread_lock);
1838 t->control |= T_REMDEVALL;
1839 mutex_unlock(&pktgen_thread_lock);
1840 schedule_timeout_interruptible(msecs_to_jiffies(125)); /* Propagate thread->control */
1842 sprintf(pg_result, "OK: rem_device_all");
1846 if (!strcmp(name, "max_before_softirq")) {
1847 sprintf(pg_result, "OK: Note! max_before_softirq is obsoleted -- Do not use");
1857 static int pktgen_thread_open(struct inode *inode, struct file *file)
1859 return single_open(file, pktgen_thread_show, PDE_DATA(inode));
1862 static const struct file_operations pktgen_thread_fops = {
1863 .owner = THIS_MODULE,
1864 .open = pktgen_thread_open,
1866 .llseek = seq_lseek,
1867 .write = pktgen_thread_write,
1868 .release = single_release,
1871 /* Think find or remove for NN */
1872 static struct pktgen_dev *__pktgen_NN_threads(const struct pktgen_net *pn,
1873 const char *ifname, int remove)
1875 struct pktgen_thread *t;
1876 struct pktgen_dev *pkt_dev = NULL;
1877 bool exact = (remove == FIND);
1879 list_for_each_entry(t, &pn->pktgen_threads, th_list) {
1880 pkt_dev = pktgen_find_dev(t, ifname, exact);
1883 pkt_dev->removal_mark = 1;
1884 t->control |= T_REMDEV;
1893 * mark a device for removal
1895 static void pktgen_mark_device(const struct pktgen_net *pn, const char *ifname)
1897 struct pktgen_dev *pkt_dev = NULL;
1898 const int max_tries = 10, msec_per_try = 125;
1901 mutex_lock(&pktgen_thread_lock);
1902 pr_debug("%s: marking %s for removal\n", __func__, ifname);
1906 pkt_dev = __pktgen_NN_threads(pn, ifname, REMOVE);
1907 if (pkt_dev == NULL)
1908 break; /* success */
1910 mutex_unlock(&pktgen_thread_lock);
1911 pr_debug("%s: waiting for %s to disappear....\n",
1913 schedule_timeout_interruptible(msecs_to_jiffies(msec_per_try));
1914 mutex_lock(&pktgen_thread_lock);
1916 if (++i >= max_tries) {
1917 pr_err("%s: timed out after waiting %d msec for device %s to be removed\n",
1918 __func__, msec_per_try * i, ifname);
1924 mutex_unlock(&pktgen_thread_lock);
1927 static void pktgen_change_name(const struct pktgen_net *pn, struct net_device *dev)
1929 struct pktgen_thread *t;
1931 list_for_each_entry(t, &pn->pktgen_threads, th_list) {
1932 struct pktgen_dev *pkt_dev;
1935 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) {
1936 if (pkt_dev->odev != dev)
1939 proc_remove(pkt_dev->entry);
1941 pkt_dev->entry = proc_create_data(dev->name, 0600,
1945 if (!pkt_dev->entry)
1946 pr_err("can't move proc entry for '%s'\n",
1954 static int pktgen_device_event(struct notifier_block *unused,
1955 unsigned long event, void *ptr)
1957 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1958 struct pktgen_net *pn = net_generic(dev_net(dev), pg_net_id);
1960 if (pn->pktgen_exiting)
1963 /* It is OK that we do not hold the group lock right now,
1964 * as we run under the RTNL lock.
1968 case NETDEV_CHANGENAME:
1969 pktgen_change_name(pn, dev);
1972 case NETDEV_UNREGISTER:
1973 pktgen_mark_device(pn, dev->name);
1980 static struct net_device *pktgen_dev_get_by_name(const struct pktgen_net *pn,
1981 struct pktgen_dev *pkt_dev,
1987 for (i = 0; ifname[i] != '@'; i++) {
1995 return dev_get_by_name(pn->net, b);
1999 /* Associate pktgen_dev with a device. */
2001 static int pktgen_setup_dev(const struct pktgen_net *pn,
2002 struct pktgen_dev *pkt_dev, const char *ifname)
2004 struct net_device *odev;
2007 /* Clean old setups */
2008 if (pkt_dev->odev) {
2009 dev_put(pkt_dev->odev);
2010 pkt_dev->odev = NULL;
2013 odev = pktgen_dev_get_by_name(pn, pkt_dev, ifname);
2015 pr_err("no such netdevice: \"%s\"\n", ifname);
2019 if (odev->type != ARPHRD_ETHER) {
2020 pr_err("not an ethernet device: \"%s\"\n", ifname);
2022 } else if (!netif_running(odev)) {
2023 pr_err("device is down: \"%s\"\n", ifname);
2026 pkt_dev->odev = odev;
2034 /* Read pkt_dev from the interface and set up internal pktgen_dev
2035 * structure to have the right information to create/send packets
2037 static void pktgen_setup_inject(struct pktgen_dev *pkt_dev)
2041 if (!pkt_dev->odev) {
2042 pr_err("ERROR: pkt_dev->odev == NULL in setup_inject\n");
2043 sprintf(pkt_dev->result,
2044 "ERROR: pkt_dev->odev == NULL in setup_inject.\n");
2048 /* make sure that we don't pick a non-existing transmit queue */
2049 ntxq = pkt_dev->odev->real_num_tx_queues;
2051 if (ntxq <= pkt_dev->queue_map_min) {
2052 pr_warning("WARNING: Requested queue_map_min (zero-based) (%d) exceeds valid range [0 - %d] for (%d) queues on %s, resetting\n",
2053 pkt_dev->queue_map_min, (ntxq ?: 1) - 1, ntxq,
2055 pkt_dev->queue_map_min = (ntxq ?: 1) - 1;
2057 if (pkt_dev->queue_map_max >= ntxq) {
2058 pr_warning("WARNING: Requested queue_map_max (zero-based) (%d) exceeds valid range [0 - %d] for (%d) queues on %s, resetting\n",
2059 pkt_dev->queue_map_max, (ntxq ?: 1) - 1, ntxq,
2061 pkt_dev->queue_map_max = (ntxq ?: 1) - 1;
2064 /* Default to the interface's mac if not explicitly set. */
2066 if (is_zero_ether_addr(pkt_dev->src_mac))
2067 ether_addr_copy(&(pkt_dev->hh[6]), pkt_dev->odev->dev_addr);
2069 /* Set up Dest MAC */
2070 ether_addr_copy(&(pkt_dev->hh[0]), pkt_dev->dst_mac);
2072 if (pkt_dev->flags & F_IPV6) {
2073 int i, set = 0, err = 1;
2074 struct inet6_dev *idev;
2076 if (pkt_dev->min_pkt_size == 0) {
2077 pkt_dev->min_pkt_size = 14 + sizeof(struct ipv6hdr)
2078 + sizeof(struct udphdr)
2079 + sizeof(struct pktgen_hdr)
2080 + pkt_dev->pkt_overhead;
2083 for (i = 0; i < IN6_ADDR_HSIZE; i++)
2084 if (pkt_dev->cur_in6_saddr.s6_addr[i]) {
2092 * Use linklevel address if unconfigured.
2094 * use ipv6_get_lladdr if/when it's get exported
2098 idev = __in6_dev_get(pkt_dev->odev);
2100 struct inet6_ifaddr *ifp;
2102 read_lock_bh(&idev->lock);
2103 list_for_each_entry(ifp, &idev->addr_list, if_list) {
2104 if ((ifp->scope & IFA_LINK) &&
2105 !(ifp->flags & IFA_F_TENTATIVE)) {
2106 pkt_dev->cur_in6_saddr = ifp->addr;
2111 read_unlock_bh(&idev->lock);
2115 pr_err("ERROR: IPv6 link address not available\n");
2118 if (pkt_dev->min_pkt_size == 0) {
2119 pkt_dev->min_pkt_size = 14 + sizeof(struct iphdr)
2120 + sizeof(struct udphdr)
2121 + sizeof(struct pktgen_hdr)
2122 + pkt_dev->pkt_overhead;
2125 pkt_dev->saddr_min = 0;
2126 pkt_dev->saddr_max = 0;
2127 if (strlen(pkt_dev->src_min) == 0) {
2129 struct in_device *in_dev;
2132 in_dev = __in_dev_get_rcu(pkt_dev->odev);
2134 if (in_dev->ifa_list) {
2135 pkt_dev->saddr_min =
2136 in_dev->ifa_list->ifa_address;
2137 pkt_dev->saddr_max = pkt_dev->saddr_min;
2142 pkt_dev->saddr_min = in_aton(pkt_dev->src_min);
2143 pkt_dev->saddr_max = in_aton(pkt_dev->src_max);
2146 pkt_dev->daddr_min = in_aton(pkt_dev->dst_min);
2147 pkt_dev->daddr_max = in_aton(pkt_dev->dst_max);
2149 /* Initialize current values. */
2150 pkt_dev->cur_pkt_size = pkt_dev->min_pkt_size;
2151 if (pkt_dev->min_pkt_size > pkt_dev->max_pkt_size)
2152 pkt_dev->max_pkt_size = pkt_dev->min_pkt_size;
2154 pkt_dev->cur_dst_mac_offset = 0;
2155 pkt_dev->cur_src_mac_offset = 0;
2156 pkt_dev->cur_saddr = pkt_dev->saddr_min;
2157 pkt_dev->cur_daddr = pkt_dev->daddr_min;
2158 pkt_dev->cur_udp_dst = pkt_dev->udp_dst_min;
2159 pkt_dev->cur_udp_src = pkt_dev->udp_src_min;
2160 pkt_dev->nflows = 0;
2164 static void spin(struct pktgen_dev *pkt_dev, ktime_t spin_until)
2166 ktime_t start_time, end_time;
2168 struct hrtimer_sleeper t;
2170 hrtimer_init_on_stack(&t.timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
2171 hrtimer_set_expires(&t.timer, spin_until);
2173 remaining = ktime_to_ns(hrtimer_expires_remaining(&t.timer));
2174 if (remaining <= 0) {
2175 pkt_dev->next_tx = ktime_add_ns(spin_until, pkt_dev->delay);
2179 start_time = ktime_get();
2180 if (remaining < 100000) {
2181 /* for small delays (<100us), just loop until limit is reached */
2183 end_time = ktime_get();
2184 } while (ktime_compare(end_time, spin_until) < 0);
2186 /* see do_nanosleep */
2187 hrtimer_init_sleeper(&t, current);
2189 set_current_state(TASK_INTERRUPTIBLE);
2190 hrtimer_start_expires(&t.timer, HRTIMER_MODE_ABS);
2191 if (!hrtimer_active(&t.timer))
2197 hrtimer_cancel(&t.timer);
2198 } while (t.task && pkt_dev->running && !signal_pending(current));
2199 __set_current_state(TASK_RUNNING);
2200 end_time = ktime_get();
2203 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(end_time, start_time));
2204 pkt_dev->next_tx = ktime_add_ns(spin_until, pkt_dev->delay);
2207 static inline void set_pkt_overhead(struct pktgen_dev *pkt_dev)
2209 pkt_dev->pkt_overhead = 0;
2210 pkt_dev->pkt_overhead += pkt_dev->nr_labels*sizeof(u32);
2211 pkt_dev->pkt_overhead += VLAN_TAG_SIZE(pkt_dev);
2212 pkt_dev->pkt_overhead += SVLAN_TAG_SIZE(pkt_dev);
2215 static inline int f_seen(const struct pktgen_dev *pkt_dev, int flow)
2217 return !!(pkt_dev->flows[flow].flags & F_INIT);
2220 static inline int f_pick(struct pktgen_dev *pkt_dev)
2222 int flow = pkt_dev->curfl;
2224 if (pkt_dev->flags & F_FLOW_SEQ) {
2225 if (pkt_dev->flows[flow].count >= pkt_dev->lflow) {
2227 pkt_dev->flows[flow].count = 0;
2228 pkt_dev->flows[flow].flags = 0;
2229 pkt_dev->curfl += 1;
2230 if (pkt_dev->curfl >= pkt_dev->cflows)
2231 pkt_dev->curfl = 0; /*reset */
2234 flow = prandom_u32() % pkt_dev->cflows;
2235 pkt_dev->curfl = flow;
2237 if (pkt_dev->flows[flow].count > pkt_dev->lflow) {
2238 pkt_dev->flows[flow].count = 0;
2239 pkt_dev->flows[flow].flags = 0;
2243 return pkt_dev->curfl;
2248 /* If there was already an IPSEC SA, we keep it as is, else
2249 * we go look for it ...
2251 #define DUMMY_MARK 0
2252 static void get_ipsec_sa(struct pktgen_dev *pkt_dev, int flow)
2254 struct xfrm_state *x = pkt_dev->flows[flow].x;
2255 struct pktgen_net *pn = net_generic(dev_net(pkt_dev->odev), pg_net_id);
2259 /* We need as quick as possible to find the right SA
2260 * Searching with minimum criteria to archieve this.
2262 x = xfrm_state_lookup_byspi(pn->net, htonl(pkt_dev->spi), AF_INET);
2264 /* slow path: we dont already have xfrm_state */
2265 x = xfrm_stateonly_find(pn->net, DUMMY_MARK,
2266 (xfrm_address_t *)&pkt_dev->cur_daddr,
2267 (xfrm_address_t *)&pkt_dev->cur_saddr,
2270 pkt_dev->ipsproto, 0);
2273 pkt_dev->flows[flow].x = x;
2274 set_pkt_overhead(pkt_dev);
2275 pkt_dev->pkt_overhead += x->props.header_len;
2281 static void set_cur_queue_map(struct pktgen_dev *pkt_dev)
2284 if (pkt_dev->flags & F_QUEUE_MAP_CPU)
2285 pkt_dev->cur_queue_map = smp_processor_id();
2287 else if (pkt_dev->queue_map_min <= pkt_dev->queue_map_max) {
2289 if (pkt_dev->flags & F_QUEUE_MAP_RND) {
2291 (pkt_dev->queue_map_max -
2292 pkt_dev->queue_map_min + 1)
2293 + pkt_dev->queue_map_min;
2295 t = pkt_dev->cur_queue_map + 1;
2296 if (t > pkt_dev->queue_map_max)
2297 t = pkt_dev->queue_map_min;
2299 pkt_dev->cur_queue_map = t;
2301 pkt_dev->cur_queue_map = pkt_dev->cur_queue_map % pkt_dev->odev->real_num_tx_queues;
2304 /* Increment/randomize headers according to flags and current values
2305 * for IP src/dest, UDP src/dst port, MAC-Addr src/dst
2307 static void mod_cur_headers(struct pktgen_dev *pkt_dev)
2313 if (pkt_dev->cflows)
2314 flow = f_pick(pkt_dev);
2316 /* Deal with source MAC */
2317 if (pkt_dev->src_mac_count > 1) {
2321 if (pkt_dev->flags & F_MACSRC_RND)
2322 mc = prandom_u32() % pkt_dev->src_mac_count;
2324 mc = pkt_dev->cur_src_mac_offset++;
2325 if (pkt_dev->cur_src_mac_offset >=
2326 pkt_dev->src_mac_count)
2327 pkt_dev->cur_src_mac_offset = 0;
2330 tmp = pkt_dev->src_mac[5] + (mc & 0xFF);
2331 pkt_dev->hh[11] = tmp;
2332 tmp = (pkt_dev->src_mac[4] + ((mc >> 8) & 0xFF) + (tmp >> 8));
2333 pkt_dev->hh[10] = tmp;
2334 tmp = (pkt_dev->src_mac[3] + ((mc >> 16) & 0xFF) + (tmp >> 8));
2335 pkt_dev->hh[9] = tmp;
2336 tmp = (pkt_dev->src_mac[2] + ((mc >> 24) & 0xFF) + (tmp >> 8));
2337 pkt_dev->hh[8] = tmp;
2338 tmp = (pkt_dev->src_mac[1] + (tmp >> 8));
2339 pkt_dev->hh[7] = tmp;
2342 /* Deal with Destination MAC */
2343 if (pkt_dev->dst_mac_count > 1) {
2347 if (pkt_dev->flags & F_MACDST_RND)
2348 mc = prandom_u32() % pkt_dev->dst_mac_count;
2351 mc = pkt_dev->cur_dst_mac_offset++;
2352 if (pkt_dev->cur_dst_mac_offset >=
2353 pkt_dev->dst_mac_count) {
2354 pkt_dev->cur_dst_mac_offset = 0;
2358 tmp = pkt_dev->dst_mac[5] + (mc & 0xFF);
2359 pkt_dev->hh[5] = tmp;
2360 tmp = (pkt_dev->dst_mac[4] + ((mc >> 8) & 0xFF) + (tmp >> 8));
2361 pkt_dev->hh[4] = tmp;
2362 tmp = (pkt_dev->dst_mac[3] + ((mc >> 16) & 0xFF) + (tmp >> 8));
2363 pkt_dev->hh[3] = tmp;
2364 tmp = (pkt_dev->dst_mac[2] + ((mc >> 24) & 0xFF) + (tmp >> 8));
2365 pkt_dev->hh[2] = tmp;
2366 tmp = (pkt_dev->dst_mac[1] + (tmp >> 8));
2367 pkt_dev->hh[1] = tmp;
2370 if (pkt_dev->flags & F_MPLS_RND) {
2372 for (i = 0; i < pkt_dev->nr_labels; i++)
2373 if (pkt_dev->labels[i] & MPLS_STACK_BOTTOM)
2374 pkt_dev->labels[i] = MPLS_STACK_BOTTOM |
2375 ((__force __be32)prandom_u32() &
2379 if ((pkt_dev->flags & F_VID_RND) && (pkt_dev->vlan_id != 0xffff)) {
2380 pkt_dev->vlan_id = prandom_u32() & (4096 - 1);
2383 if ((pkt_dev->flags & F_SVID_RND) && (pkt_dev->svlan_id != 0xffff)) {
2384 pkt_dev->svlan_id = prandom_u32() & (4096 - 1);
2387 if (pkt_dev->udp_src_min < pkt_dev->udp_src_max) {
2388 if (pkt_dev->flags & F_UDPSRC_RND)
2389 pkt_dev->cur_udp_src = prandom_u32() %
2390 (pkt_dev->udp_src_max - pkt_dev->udp_src_min)
2391 + pkt_dev->udp_src_min;
2394 pkt_dev->cur_udp_src++;
2395 if (pkt_dev->cur_udp_src >= pkt_dev->udp_src_max)
2396 pkt_dev->cur_udp_src = pkt_dev->udp_src_min;
2400 if (pkt_dev->udp_dst_min < pkt_dev->udp_dst_max) {
2401 if (pkt_dev->flags & F_UDPDST_RND) {
2402 pkt_dev->cur_udp_dst = prandom_u32() %
2403 (pkt_dev->udp_dst_max - pkt_dev->udp_dst_min)
2404 + pkt_dev->udp_dst_min;
2406 pkt_dev->cur_udp_dst++;
2407 if (pkt_dev->cur_udp_dst >= pkt_dev->udp_dst_max)
2408 pkt_dev->cur_udp_dst = pkt_dev->udp_dst_min;
2412 if (!(pkt_dev->flags & F_IPV6)) {
2414 imn = ntohl(pkt_dev->saddr_min);
2415 imx = ntohl(pkt_dev->saddr_max);
2418 if (pkt_dev->flags & F_IPSRC_RND)
2419 t = prandom_u32() % (imx - imn) + imn;
2421 t = ntohl(pkt_dev->cur_saddr);
2427 pkt_dev->cur_saddr = htonl(t);
2430 if (pkt_dev->cflows && f_seen(pkt_dev, flow)) {
2431 pkt_dev->cur_daddr = pkt_dev->flows[flow].cur_daddr;
2433 imn = ntohl(pkt_dev->daddr_min);
2434 imx = ntohl(pkt_dev->daddr_max);
2438 if (pkt_dev->flags & F_IPDST_RND) {
2444 } while (ipv4_is_loopback(s) ||
2445 ipv4_is_multicast(s) ||
2446 ipv4_is_lbcast(s) ||
2447 ipv4_is_zeronet(s) ||
2448 ipv4_is_local_multicast(s));
2449 pkt_dev->cur_daddr = s;
2451 t = ntohl(pkt_dev->cur_daddr);
2456 pkt_dev->cur_daddr = htonl(t);
2459 if (pkt_dev->cflows) {
2460 pkt_dev->flows[flow].flags |= F_INIT;
2461 pkt_dev->flows[flow].cur_daddr =
2464 if (pkt_dev->flags & F_IPSEC_ON)
2465 get_ipsec_sa(pkt_dev, flow);
2470 } else { /* IPV6 * */
2472 if (!ipv6_addr_any(&pkt_dev->min_in6_daddr)) {
2475 /* Only random destinations yet */
2477 for (i = 0; i < 4; i++) {
2478 pkt_dev->cur_in6_daddr.s6_addr32[i] =
2479 (((__force __be32)prandom_u32() |
2480 pkt_dev->min_in6_daddr.s6_addr32[i]) &
2481 pkt_dev->max_in6_daddr.s6_addr32[i]);
2486 if (pkt_dev->min_pkt_size < pkt_dev->max_pkt_size) {
2488 if (pkt_dev->flags & F_TXSIZE_RND) {
2490 (pkt_dev->max_pkt_size - pkt_dev->min_pkt_size)
2491 + pkt_dev->min_pkt_size;
2493 t = pkt_dev->cur_pkt_size + 1;
2494 if (t > pkt_dev->max_pkt_size)
2495 t = pkt_dev->min_pkt_size;
2497 pkt_dev->cur_pkt_size = t;
2500 set_cur_queue_map(pkt_dev);
2502 pkt_dev->flows[flow].count++;
2507 static u32 pktgen_dst_metrics[RTAX_MAX + 1] = {
2509 [RTAX_HOPLIMIT] = 0x5, /* Set a static hoplimit */
2512 static int pktgen_output_ipsec(struct sk_buff *skb, struct pktgen_dev *pkt_dev)
2514 struct xfrm_state *x = pkt_dev->flows[pkt_dev->curfl].x;
2516 struct net *net = dev_net(pkt_dev->odev);
2520 /* XXX: we dont support tunnel mode for now until
2521 * we resolve the dst issue */
2522 if ((x->props.mode != XFRM_MODE_TRANSPORT) && (pkt_dev->spi == 0))
2525 /* But when user specify an valid SPI, transformation
2526 * supports both transport/tunnel mode + ESP/AH type.
2528 if ((x->props.mode == XFRM_MODE_TUNNEL) && (pkt_dev->spi != 0))
2529 skb->_skb_refdst = (unsigned long)&pkt_dev->dst | SKB_DST_NOREF;
2532 err = x->outer_mode->output(x, skb);
2533 rcu_read_unlock_bh();
2535 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEMODEERROR);
2538 err = x->type->output(x, skb);
2540 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEPROTOERROR);
2543 spin_lock_bh(&x->lock);
2544 x->curlft.bytes += skb->len;
2545 x->curlft.packets++;
2546 spin_unlock_bh(&x->lock);
2551 static void free_SAs(struct pktgen_dev *pkt_dev)
2553 if (pkt_dev->cflows) {
2554 /* let go of the SAs if we have them */
2556 for (i = 0; i < pkt_dev->cflows; i++) {
2557 struct xfrm_state *x = pkt_dev->flows[i].x;
2560 pkt_dev->flows[i].x = NULL;
2566 static int process_ipsec(struct pktgen_dev *pkt_dev,
2567 struct sk_buff *skb, __be16 protocol)
2569 if (pkt_dev->flags & F_IPSEC_ON) {
2570 struct xfrm_state *x = pkt_dev->flows[pkt_dev->curfl].x;
2577 nhead = x->props.header_len - skb_headroom(skb);
2579 ret = pskb_expand_head(skb, nhead, 0, GFP_ATOMIC);
2581 pr_err("Error expanding ipsec packet %d\n",
2587 /* ipsec is not expecting ll header */
2588 skb_pull(skb, ETH_HLEN);
2589 ret = pktgen_output_ipsec(skb, pkt_dev);
2591 pr_err("Error creating ipsec packet %d\n", ret);
2595 eth = (__u8 *) skb_push(skb, ETH_HLEN);
2596 memcpy(eth, pkt_dev->hh, 12);
2597 *(u16 *) ð[12] = protocol;
2599 /* Update IPv4 header len as well as checksum value */
2601 iph->tot_len = htons(skb->len - ETH_HLEN);
2612 static void mpls_push(__be32 *mpls, struct pktgen_dev *pkt_dev)
2615 for (i = 0; i < pkt_dev->nr_labels; i++)
2616 *mpls++ = pkt_dev->labels[i] & ~MPLS_STACK_BOTTOM;
2619 *mpls |= MPLS_STACK_BOTTOM;
2622 static inline __be16 build_tci(unsigned int id, unsigned int cfi,
2625 return htons(id | (cfi << 12) | (prio << 13));
2628 static void pktgen_finalize_skb(struct pktgen_dev *pkt_dev, struct sk_buff *skb,
2631 struct timeval timestamp;
2632 struct pktgen_hdr *pgh;
2634 pgh = (struct pktgen_hdr *)skb_put(skb, sizeof(*pgh));
2635 datalen -= sizeof(*pgh);
2637 if (pkt_dev->nfrags <= 0) {
2638 memset(skb_put(skb, datalen), 0, datalen);
2640 int frags = pkt_dev->nfrags;
2645 if (frags > MAX_SKB_FRAGS)
2646 frags = MAX_SKB_FRAGS;
2647 len = datalen - frags * PAGE_SIZE;
2649 memset(skb_put(skb, len), 0, len);
2650 datalen = frags * PAGE_SIZE;
2654 frag_len = (datalen/frags) < PAGE_SIZE ?
2655 (datalen/frags) : PAGE_SIZE;
2656 while (datalen > 0) {
2657 if (unlikely(!pkt_dev->page)) {
2658 int node = numa_node_id();
2660 if (pkt_dev->node >= 0 && (pkt_dev->flags & F_NODE))
2661 node = pkt_dev->node;
2662 pkt_dev->page = alloc_pages_node(node, GFP_KERNEL | __GFP_ZERO, 0);
2666 get_page(pkt_dev->page);
2667 skb_frag_set_page(skb, i, pkt_dev->page);
2668 skb_shinfo(skb)->frags[i].page_offset = 0;
2669 /*last fragment, fill rest of data*/
2670 if (i == (frags - 1))
2671 skb_frag_size_set(&skb_shinfo(skb)->frags[i],
2672 (datalen < PAGE_SIZE ? datalen : PAGE_SIZE));
2674 skb_frag_size_set(&skb_shinfo(skb)->frags[i], frag_len);
2675 datalen -= skb_frag_size(&skb_shinfo(skb)->frags[i]);
2676 skb->len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2677 skb->data_len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2679 skb_shinfo(skb)->nr_frags = i;
2683 /* Stamp the time, and sequence number,
2684 * convert them to network byte order
2686 pgh->pgh_magic = htonl(PKTGEN_MAGIC);
2687 pgh->seq_num = htonl(pkt_dev->seq_num);
2689 do_gettimeofday(×tamp);
2690 pgh->tv_sec = htonl(timestamp.tv_sec);
2691 pgh->tv_usec = htonl(timestamp.tv_usec);
2694 static struct sk_buff *pktgen_alloc_skb(struct net_device *dev,
2695 struct pktgen_dev *pkt_dev,
2696 unsigned int extralen)
2698 struct sk_buff *skb = NULL;
2699 unsigned int size = pkt_dev->cur_pkt_size + 64 + extralen +
2700 pkt_dev->pkt_overhead;
2702 if (pkt_dev->flags & F_NODE) {
2703 int node = pkt_dev->node >= 0 ? pkt_dev->node : numa_node_id();
2705 skb = __alloc_skb(NET_SKB_PAD + size, GFP_NOWAIT, 0, node);
2707 skb_reserve(skb, NET_SKB_PAD);
2711 skb = __netdev_alloc_skb(dev, size, GFP_NOWAIT);
2717 static struct sk_buff *fill_packet_ipv4(struct net_device *odev,
2718 struct pktgen_dev *pkt_dev)
2720 struct sk_buff *skb = NULL;
2722 struct udphdr *udph;
2725 __be16 protocol = htons(ETH_P_IP);
2727 __be16 *vlan_tci = NULL; /* Encapsulates priority and VLAN ID */
2728 __be16 *vlan_encapsulated_proto = NULL; /* packet type ID field (or len) for VLAN tag */
2729 __be16 *svlan_tci = NULL; /* Encapsulates priority and SVLAN ID */
2730 __be16 *svlan_encapsulated_proto = NULL; /* packet type ID field (or len) for SVLAN tag */
2733 if (pkt_dev->nr_labels)
2734 protocol = htons(ETH_P_MPLS_UC);
2736 if (pkt_dev->vlan_id != 0xffff)
2737 protocol = htons(ETH_P_8021Q);
2739 /* Update any of the values, used when we're incrementing various
2742 mod_cur_headers(pkt_dev);
2743 queue_map = pkt_dev->cur_queue_map;
2745 datalen = (odev->hard_header_len + 16) & ~0xf;
2747 skb = pktgen_alloc_skb(odev, pkt_dev, datalen);
2749 sprintf(pkt_dev->result, "No memory");
2753 prefetchw(skb->data);
2754 skb_reserve(skb, datalen);
2756 /* Reserve for ethernet and IP header */
2757 eth = (__u8 *) skb_push(skb, 14);
2758 mpls = (__be32 *)skb_put(skb, pkt_dev->nr_labels*sizeof(__u32));
2759 if (pkt_dev->nr_labels)
2760 mpls_push(mpls, pkt_dev);
2762 if (pkt_dev->vlan_id != 0xffff) {
2763 if (pkt_dev->svlan_id != 0xffff) {
2764 svlan_tci = (__be16 *)skb_put(skb, sizeof(__be16));
2765 *svlan_tci = build_tci(pkt_dev->svlan_id,
2768 svlan_encapsulated_proto = (__be16 *)skb_put(skb, sizeof(__be16));
2769 *svlan_encapsulated_proto = htons(ETH_P_8021Q);
2771 vlan_tci = (__be16 *)skb_put(skb, sizeof(__be16));
2772 *vlan_tci = build_tci(pkt_dev->vlan_id,
2775 vlan_encapsulated_proto = (__be16 *)skb_put(skb, sizeof(__be16));
2776 *vlan_encapsulated_proto = htons(ETH_P_IP);
2779 skb_set_mac_header(skb, 0);
2780 skb_set_network_header(skb, skb->len);
2781 iph = (struct iphdr *) skb_put(skb, sizeof(struct iphdr));
2783 skb_set_transport_header(skb, skb->len);
2784 udph = (struct udphdr *) skb_put(skb, sizeof(struct udphdr));
2785 skb_set_queue_mapping(skb, queue_map);
2786 skb->priority = pkt_dev->skb_priority;
2788 memcpy(eth, pkt_dev->hh, 12);
2789 *(__be16 *) & eth[12] = protocol;
2791 /* Eth + IPh + UDPh + mpls */
2792 datalen = pkt_dev->cur_pkt_size - 14 - 20 - 8 -
2793 pkt_dev->pkt_overhead;
2794 if (datalen < 0 || datalen < sizeof(struct pktgen_hdr))
2795 datalen = sizeof(struct pktgen_hdr);
2797 udph->source = htons(pkt_dev->cur_udp_src);
2798 udph->dest = htons(pkt_dev->cur_udp_dst);
2799 udph->len = htons(datalen + 8); /* DATA + udphdr */
2805 iph->tos = pkt_dev->tos;
2806 iph->protocol = IPPROTO_UDP; /* UDP */
2807 iph->saddr = pkt_dev->cur_saddr;
2808 iph->daddr = pkt_dev->cur_daddr;
2809 iph->id = htons(pkt_dev->ip_id);
2812 iplen = 20 + 8 + datalen;
2813 iph->tot_len = htons(iplen);
2815 skb->protocol = protocol;
2817 skb->pkt_type = PACKET_HOST;
2819 if (!(pkt_dev->flags & F_UDPCSUM)) {
2820 skb->ip_summed = CHECKSUM_NONE;
2821 } else if (odev->features & NETIF_F_V4_CSUM) {
2822 skb->ip_summed = CHECKSUM_PARTIAL;
2824 udp4_hwcsum(skb, udph->source, udph->dest);
2826 __wsum csum = udp_csum(skb);
2828 /* add protocol-dependent pseudo-header */
2829 udph->check = csum_tcpudp_magic(udph->source, udph->dest,
2830 datalen + 8, IPPROTO_UDP, csum);
2832 if (udph->check == 0)
2833 udph->check = CSUM_MANGLED_0;
2836 pktgen_finalize_skb(pkt_dev, skb, datalen);
2839 if (!process_ipsec(pkt_dev, skb, protocol))
2846 static struct sk_buff *fill_packet_ipv6(struct net_device *odev,
2847 struct pktgen_dev *pkt_dev)
2849 struct sk_buff *skb = NULL;
2851 struct udphdr *udph;
2852 int datalen, udplen;
2853 struct ipv6hdr *iph;
2854 __be16 protocol = htons(ETH_P_IPV6);
2856 __be16 *vlan_tci = NULL; /* Encapsulates priority and VLAN ID */
2857 __be16 *vlan_encapsulated_proto = NULL; /* packet type ID field (or len) for VLAN tag */
2858 __be16 *svlan_tci = NULL; /* Encapsulates priority and SVLAN ID */
2859 __be16 *svlan_encapsulated_proto = NULL; /* packet type ID field (or len) for SVLAN tag */
2862 if (pkt_dev->nr_labels)
2863 protocol = htons(ETH_P_MPLS_UC);
2865 if (pkt_dev->vlan_id != 0xffff)
2866 protocol = htons(ETH_P_8021Q);
2868 /* Update any of the values, used when we're incrementing various
2871 mod_cur_headers(pkt_dev);
2872 queue_map = pkt_dev->cur_queue_map;
2874 skb = pktgen_alloc_skb(odev, pkt_dev, 16);
2876 sprintf(pkt_dev->result, "No memory");
2880 prefetchw(skb->data);
2881 skb_reserve(skb, 16);
2883 /* Reserve for ethernet and IP header */
2884 eth = (__u8 *) skb_push(skb, 14);
2885 mpls = (__be32 *)skb_put(skb, pkt_dev->nr_labels*sizeof(__u32));
2886 if (pkt_dev->nr_labels)
2887 mpls_push(mpls, pkt_dev);
2889 if (pkt_dev->vlan_id != 0xffff) {
2890 if (pkt_dev->svlan_id != 0xffff) {
2891 svlan_tci = (__be16 *)skb_put(skb, sizeof(__be16));
2892 *svlan_tci = build_tci(pkt_dev->svlan_id,
2895 svlan_encapsulated_proto = (__be16 *)skb_put(skb, sizeof(__be16));
2896 *svlan_encapsulated_proto = htons(ETH_P_8021Q);
2898 vlan_tci = (__be16 *)skb_put(skb, sizeof(__be16));
2899 *vlan_tci = build_tci(pkt_dev->vlan_id,
2902 vlan_encapsulated_proto = (__be16 *)skb_put(skb, sizeof(__be16));
2903 *vlan_encapsulated_proto = htons(ETH_P_IPV6);
2906 skb_set_mac_header(skb, 0);
2907 skb_set_network_header(skb, skb->len);
2908 iph = (struct ipv6hdr *) skb_put(skb, sizeof(struct ipv6hdr));
2910 skb_set_transport_header(skb, skb->len);
2911 udph = (struct udphdr *) skb_put(skb, sizeof(struct udphdr));
2912 skb_set_queue_mapping(skb, queue_map);
2913 skb->priority = pkt_dev->skb_priority;
2915 memcpy(eth, pkt_dev->hh, 12);
2916 *(__be16 *) ð[12] = protocol;
2918 /* Eth + IPh + UDPh + mpls */
2919 datalen = pkt_dev->cur_pkt_size - 14 -
2920 sizeof(struct ipv6hdr) - sizeof(struct udphdr) -
2921 pkt_dev->pkt_overhead;
2923 if (datalen < 0 || datalen < sizeof(struct pktgen_hdr)) {
2924 datalen = sizeof(struct pktgen_hdr);
2925 net_info_ratelimited("increased datalen to %d\n", datalen);
2928 udplen = datalen + sizeof(struct udphdr);
2929 udph->source = htons(pkt_dev->cur_udp_src);
2930 udph->dest = htons(pkt_dev->cur_udp_dst);
2931 udph->len = htons(udplen);
2934 *(__be32 *) iph = htonl(0x60000000); /* Version + flow */
2936 if (pkt_dev->traffic_class) {
2937 /* Version + traffic class + flow (0) */
2938 *(__be32 *)iph |= htonl(0x60000000 | (pkt_dev->traffic_class << 20));
2941 iph->hop_limit = 32;
2943 iph->payload_len = htons(udplen);
2944 iph->nexthdr = IPPROTO_UDP;
2946 iph->daddr = pkt_dev->cur_in6_daddr;
2947 iph->saddr = pkt_dev->cur_in6_saddr;
2949 skb->protocol = protocol;
2951 skb->pkt_type = PACKET_HOST;
2953 if (!(pkt_dev->flags & F_UDPCSUM)) {
2954 skb->ip_summed = CHECKSUM_NONE;
2955 } else if (odev->features & NETIF_F_V6_CSUM) {
2956 skb->ip_summed = CHECKSUM_PARTIAL;
2957 skb->csum_start = skb_transport_header(skb) - skb->head;
2958 skb->csum_offset = offsetof(struct udphdr, check);
2959 udph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, udplen, IPPROTO_UDP, 0);
2961 __wsum csum = udp_csum(skb);
2963 /* add protocol-dependent pseudo-header */
2964 udph->check = csum_ipv6_magic(&iph->saddr, &iph->daddr, udplen, IPPROTO_UDP, csum);
2966 if (udph->check == 0)
2967 udph->check = CSUM_MANGLED_0;
2970 pktgen_finalize_skb(pkt_dev, skb, datalen);
2975 static struct sk_buff *fill_packet(struct net_device *odev,
2976 struct pktgen_dev *pkt_dev)
2978 if (pkt_dev->flags & F_IPV6)
2979 return fill_packet_ipv6(odev, pkt_dev);
2981 return fill_packet_ipv4(odev, pkt_dev);
2984 static void pktgen_clear_counters(struct pktgen_dev *pkt_dev)
2986 pkt_dev->seq_num = 1;
2987 pkt_dev->idle_acc = 0;
2989 pkt_dev->tx_bytes = 0;
2990 pkt_dev->errors = 0;
2993 /* Set up structure for sending pkts, clear counters */
2995 static void pktgen_run(struct pktgen_thread *t)
2997 struct pktgen_dev *pkt_dev;
3003 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) {
3006 * setup odev and create initial packet.
3008 pktgen_setup_inject(pkt_dev);
3010 if (pkt_dev->odev) {
3011 pktgen_clear_counters(pkt_dev);
3012 pkt_dev->skb = NULL;
3013 pkt_dev->started_at = pkt_dev->next_tx = ktime_get();
3015 set_pkt_overhead(pkt_dev);
3017 strcpy(pkt_dev->result, "Starting");
3018 pkt_dev->running = 1; /* Cranke yeself! */
3021 strcpy(pkt_dev->result, "Error starting");
3025 t->control &= ~(T_STOP);
3028 static void pktgen_stop_all_threads_ifs(struct pktgen_net *pn)
3030 struct pktgen_thread *t;
3034 mutex_lock(&pktgen_thread_lock);
3036 list_for_each_entry(t, &pn->pktgen_threads, th_list)
3037 t->control |= T_STOP;
3039 mutex_unlock(&pktgen_thread_lock);
3042 static int thread_is_running(const struct pktgen_thread *t)
3044 const struct pktgen_dev *pkt_dev;
3047 list_for_each_entry_rcu(pkt_dev, &t->if_list, list)
3048 if (pkt_dev->running) {
3056 static int pktgen_wait_thread_run(struct pktgen_thread *t)
3058 while (thread_is_running(t)) {
3060 msleep_interruptible(100);
3062 if (signal_pending(current))
3070 static int pktgen_wait_all_threads_run(struct pktgen_net *pn)
3072 struct pktgen_thread *t;
3075 mutex_lock(&pktgen_thread_lock);
3077 list_for_each_entry(t, &pn->pktgen_threads, th_list) {
3078 sig = pktgen_wait_thread_run(t);
3084 list_for_each_entry(t, &pn->pktgen_threads, th_list)
3085 t->control |= (T_STOP);
3087 mutex_unlock(&pktgen_thread_lock);
3091 static void pktgen_run_all_threads(struct pktgen_net *pn)
3093 struct pktgen_thread *t;
3097 mutex_lock(&pktgen_thread_lock);
3099 list_for_each_entry(t, &pn->pktgen_threads, th_list)
3100 t->control |= (T_RUN);
3102 mutex_unlock(&pktgen_thread_lock);
3104 /* Propagate thread->control */
3105 schedule_timeout_interruptible(msecs_to_jiffies(125));
3107 pktgen_wait_all_threads_run(pn);
3110 static void pktgen_reset_all_threads(struct pktgen_net *pn)
3112 struct pktgen_thread *t;
3116 mutex_lock(&pktgen_thread_lock);
3118 list_for_each_entry(t, &pn->pktgen_threads, th_list)
3119 t->control |= (T_REMDEVALL);
3121 mutex_unlock(&pktgen_thread_lock);
3123 /* Propagate thread->control */
3124 schedule_timeout_interruptible(msecs_to_jiffies(125));
3126 pktgen_wait_all_threads_run(pn);
3129 static void show_results(struct pktgen_dev *pkt_dev, int nr_frags)
3131 __u64 bps, mbps, pps;
3132 char *p = pkt_dev->result;
3133 ktime_t elapsed = ktime_sub(pkt_dev->stopped_at,
3134 pkt_dev->started_at);
3135 ktime_t idle = ns_to_ktime(pkt_dev->idle_acc);
3137 p += sprintf(p, "OK: %llu(c%llu+d%llu) usec, %llu (%dbyte,%dfrags)\n",
3138 (unsigned long long)ktime_to_us(elapsed),
3139 (unsigned long long)ktime_to_us(ktime_sub(elapsed, idle)),
3140 (unsigned long long)ktime_to_us(idle),
3141 (unsigned long long)pkt_dev->sofar,
3142 pkt_dev->cur_pkt_size, nr_frags);
3144 pps = div64_u64(pkt_dev->sofar * NSEC_PER_SEC,
3145 ktime_to_ns(elapsed));
3147 bps = pps * 8 * pkt_dev->cur_pkt_size;
3150 do_div(mbps, 1000000);
3151 p += sprintf(p, " %llupps %lluMb/sec (%llubps) errors: %llu",
3152 (unsigned long long)pps,
3153 (unsigned long long)mbps,
3154 (unsigned long long)bps,
3155 (unsigned long long)pkt_dev->errors);
3158 /* Set stopped-at timer, remove from running list, do counters & statistics */
3159 static int pktgen_stop_device(struct pktgen_dev *pkt_dev)
3161 int nr_frags = pkt_dev->skb ? skb_shinfo(pkt_dev->skb)->nr_frags : -1;
3163 if (!pkt_dev->running) {
3164 pr_warning("interface: %s is already stopped\n",
3169 pkt_dev->running = 0;
3170 kfree_skb(pkt_dev->skb);
3171 pkt_dev->skb = NULL;
3172 pkt_dev->stopped_at = ktime_get();
3174 show_results(pkt_dev, nr_frags);
3179 static struct pktgen_dev *next_to_run(struct pktgen_thread *t)
3181 struct pktgen_dev *pkt_dev, *best = NULL;
3184 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) {
3185 if (!pkt_dev->running)
3189 else if (ktime_compare(pkt_dev->next_tx, best->next_tx) < 0)
3197 static void pktgen_stop(struct pktgen_thread *t)
3199 struct pktgen_dev *pkt_dev;
3205 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) {
3206 pktgen_stop_device(pkt_dev);
3213 * one of our devices needs to be removed - find it
3216 static void pktgen_rem_one_if(struct pktgen_thread *t)
3218 struct list_head *q, *n;
3219 struct pktgen_dev *cur;
3223 list_for_each_safe(q, n, &t->if_list) {
3224 cur = list_entry(q, struct pktgen_dev, list);
3226 if (!cur->removal_mark)
3229 kfree_skb(cur->skb);
3232 pktgen_remove_device(t, cur);
3238 static void pktgen_rem_all_ifs(struct pktgen_thread *t)
3240 struct list_head *q, *n;
3241 struct pktgen_dev *cur;
3245 /* Remove all devices, free mem */
3247 list_for_each_safe(q, n, &t->if_list) {
3248 cur = list_entry(q, struct pktgen_dev, list);
3250 kfree_skb(cur->skb);
3253 pktgen_remove_device(t, cur);
3257 static void pktgen_rem_thread(struct pktgen_thread *t)
3259 /* Remove from the thread list */
3260 remove_proc_entry(t->tsk->comm, t->net->proc_dir);
3263 static void pktgen_resched(struct pktgen_dev *pkt_dev)
3265 ktime_t idle_start = ktime_get();
3267 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(ktime_get(), idle_start));
3270 static void pktgen_wait_for_skb(struct pktgen_dev *pkt_dev)
3272 ktime_t idle_start = ktime_get();
3274 while (atomic_read(&(pkt_dev->skb->users)) != 1) {
3275 if (signal_pending(current))
3279 pktgen_resched(pkt_dev);
3283 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(ktime_get(), idle_start));
3286 static void pktgen_xmit(struct pktgen_dev *pkt_dev)
3288 struct net_device *odev = pkt_dev->odev;
3289 netdev_tx_t (*xmit)(struct sk_buff *, struct net_device *)
3290 = odev->netdev_ops->ndo_start_xmit;
3291 struct netdev_queue *txq;
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);
3301 /* This is max DELAY, this has special meaning of
3304 if (unlikely(pkt_dev->delay == ULLONG_MAX)) {
3305 pkt_dev->next_tx = ktime_add_ns(ktime_get(), ULONG_MAX);
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);
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");
3319 pkt_dev->clone_count--; /* back out increment, OOM */
3322 pkt_dev->last_pkt_size = pkt_dev->skb->len;
3323 pkt_dev->allocated_skbs++;
3324 pkt_dev->clone_count = 0; /* reset counter */
3327 if (pkt_dev->delay && pkt_dev->last_ok)
3328 spin(pkt_dev, pkt_dev->next_tx);
3330 queue_map = skb_get_queue_mapping(pkt_dev->skb);
3331 txq = netdev_get_tx_queue(odev, queue_map);
3335 HARD_TX_LOCK(odev, txq, smp_processor_id());
3337 if (unlikely(netif_xmit_frozen_or_drv_stopped(txq))) {
3338 ret = NETDEV_TX_BUSY;
3339 pkt_dev->last_ok = 0;
3342 atomic_inc(&(pkt_dev->skb->users));
3343 ret = (*xmit)(pkt_dev->skb, odev);
3347 txq_trans_update(txq);
3348 pkt_dev->last_ok = 1;
3351 pkt_dev->tx_bytes += pkt_dev->last_pkt_size;
3355 case NET_XMIT_POLICED:
3356 /* skb has been consumed */
3359 default: /* Drivers are not supposed to return other values! */
3360 net_info_ratelimited("%s xmit error: %d\n",
3361 pkt_dev->odevname, ret);
3364 case NETDEV_TX_LOCKED:
3365 case NETDEV_TX_BUSY:
3366 /* Retry it next time */
3367 atomic_dec(&(pkt_dev->skb->users));
3368 pkt_dev->last_ok = 0;
3371 HARD_TX_UNLOCK(odev, txq);
3375 /* If pkt_dev->count is zero, then run forever */
3376 if ((pkt_dev->count != 0) && (pkt_dev->sofar >= pkt_dev->count)) {
3377 pktgen_wait_for_skb(pkt_dev);
3379 /* Done with this */
3380 pktgen_stop_device(pkt_dev);
3385 * Main loop of the thread goes here
3388 static int pktgen_thread_worker(void *arg)
3391 struct pktgen_thread *t = arg;
3392 struct pktgen_dev *pkt_dev = NULL;
3395 BUG_ON(smp_processor_id() != cpu);
3397 init_waitqueue_head(&t->queue);
3398 complete(&t->start_done);
3400 pr_debug("starting pktgen/%d: pid=%d\n", cpu, task_pid_nr(current));
3404 __set_current_state(TASK_RUNNING);
3406 while (!kthread_should_stop()) {
3407 pkt_dev = next_to_run(t);
3409 if (unlikely(!pkt_dev && t->control == 0)) {
3410 if (t->net->pktgen_exiting)
3412 wait_event_interruptible_timeout(t->queue,
3419 if (likely(pkt_dev)) {
3420 pktgen_xmit(pkt_dev);
3423 pktgen_resched(pkt_dev);
3428 if (t->control & T_STOP) {
3430 t->control &= ~(T_STOP);
3433 if (t->control & T_RUN) {
3435 t->control &= ~(T_RUN);
3438 if (t->control & T_REMDEVALL) {
3439 pktgen_rem_all_ifs(t);
3440 t->control &= ~(T_REMDEVALL);
3443 if (t->control & T_REMDEV) {
3444 pktgen_rem_one_if(t);
3445 t->control &= ~(T_REMDEV);
3450 set_current_state(TASK_INTERRUPTIBLE);
3452 pr_debug("%s stopping all device\n", t->tsk->comm);
3455 pr_debug("%s removing all device\n", t->tsk->comm);
3456 pktgen_rem_all_ifs(t);
3458 pr_debug("%s removing thread\n", t->tsk->comm);
3459 pktgen_rem_thread(t);
3461 /* Wait for kthread_stop */
3462 while (!kthread_should_stop()) {
3463 set_current_state(TASK_INTERRUPTIBLE);
3466 __set_current_state(TASK_RUNNING);
3471 static struct pktgen_dev *pktgen_find_dev(struct pktgen_thread *t,
3472 const char *ifname, bool exact)
3474 struct pktgen_dev *p, *pkt_dev = NULL;
3475 size_t len = strlen(ifname);
3478 list_for_each_entry_rcu(p, &t->if_list, list)
3479 if (strncmp(p->odevname, ifname, len) == 0) {
3480 if (p->odevname[len]) {
3481 if (exact || p->odevname[len] != '@')
3489 pr_debug("find_dev(%s) returning %p\n", ifname, pkt_dev);
3494 * Adds a dev at front of if_list.
3497 static int add_dev_to_thread(struct pktgen_thread *t,
3498 struct pktgen_dev *pkt_dev)
3502 /* This function cannot be called concurrently, as its called
3503 * under pktgen_thread_lock mutex, but it can run from
3504 * userspace on another CPU than the kthread. The if_lock()
3505 * is used here to sync with concurrent instances of
3506 * _rem_dev_from_if_list() invoked via kthread, which is also
3507 * updating the if_list */
3510 if (pkt_dev->pg_thread) {
3511 pr_err("ERROR: already assigned to a thread\n");
3516 pkt_dev->running = 0;
3517 pkt_dev->pg_thread = t;
3518 list_add_rcu(&pkt_dev->list, &t->if_list);
3525 /* Called under thread lock */
3527 static int pktgen_add_device(struct pktgen_thread *t, const char *ifname)
3529 struct pktgen_dev *pkt_dev;
3531 int node = cpu_to_node(t->cpu);
3533 /* We don't allow a device to be on several threads */
3535 pkt_dev = __pktgen_NN_threads(t->net, ifname, FIND);
3537 pr_err("ERROR: interface already used\n");
3541 pkt_dev = kzalloc_node(sizeof(struct pktgen_dev), GFP_KERNEL, node);
3545 strcpy(pkt_dev->odevname, ifname);
3546 pkt_dev->flows = vzalloc_node(MAX_CFLOWS * sizeof(struct flow_state),
3548 if (pkt_dev->flows == NULL) {
3553 pkt_dev->removal_mark = 0;
3554 pkt_dev->nfrags = 0;
3555 pkt_dev->delay = pg_delay_d;
3556 pkt_dev->count = pg_count_d;
3558 pkt_dev->udp_src_min = 9; /* sink port */
3559 pkt_dev->udp_src_max = 9;
3560 pkt_dev->udp_dst_min = 9;
3561 pkt_dev->udp_dst_max = 9;
3562 pkt_dev->vlan_p = 0;
3563 pkt_dev->vlan_cfi = 0;
3564 pkt_dev->vlan_id = 0xffff;
3565 pkt_dev->svlan_p = 0;
3566 pkt_dev->svlan_cfi = 0;
3567 pkt_dev->svlan_id = 0xffff;
3570 err = pktgen_setup_dev(t->net, pkt_dev, ifname);
3573 if (pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING)
3574 pkt_dev->clone_skb = pg_clone_skb_d;
3576 pkt_dev->entry = proc_create_data(ifname, 0600, t->net->proc_dir,
3577 &pktgen_if_fops, pkt_dev);
3578 if (!pkt_dev->entry) {
3579 pr_err("cannot create %s/%s procfs entry\n",
3580 PG_PROC_DIR, ifname);
3585 pkt_dev->ipsmode = XFRM_MODE_TRANSPORT;
3586 pkt_dev->ipsproto = IPPROTO_ESP;
3588 /* xfrm tunnel mode needs additional dst to extract outter
3589 * ip header protocol/ttl/id field, here creat a phony one.
3590 * instead of looking for a valid rt, which definitely hurting
3591 * performance under such circumstance.
3593 pkt_dev->dstops.family = AF_INET;
3594 pkt_dev->dst.dev = pkt_dev->odev;
3595 dst_init_metrics(&pkt_dev->dst, pktgen_dst_metrics, false);
3596 pkt_dev->dst.child = &pkt_dev->dst;
3597 pkt_dev->dst.ops = &pkt_dev->dstops;
3600 return add_dev_to_thread(t, pkt_dev);
3602 dev_put(pkt_dev->odev);
3607 vfree(pkt_dev->flows);
3612 static int __net_init pktgen_create_thread(int cpu, struct pktgen_net *pn)
3614 struct pktgen_thread *t;
3615 struct proc_dir_entry *pe;
3616 struct task_struct *p;
3618 t = kzalloc_node(sizeof(struct pktgen_thread), GFP_KERNEL,
3621 pr_err("ERROR: out of memory, can't create new thread\n");
3625 spin_lock_init(&t->if_lock);
3628 INIT_LIST_HEAD(&t->if_list);
3630 list_add_tail(&t->th_list, &pn->pktgen_threads);
3631 init_completion(&t->start_done);
3633 p = kthread_create_on_node(pktgen_thread_worker,
3636 "kpktgend_%d", cpu);
3638 pr_err("kernel_thread() failed for cpu %d\n", t->cpu);
3639 list_del(&t->th_list);
3643 kthread_bind(p, cpu);
3646 pe = proc_create_data(t->tsk->comm, 0600, pn->proc_dir,
3647 &pktgen_thread_fops, t);
3649 pr_err("cannot create %s/%s procfs entry\n",
3650 PG_PROC_DIR, t->tsk->comm);
3652 list_del(&t->th_list);
3659 wait_for_completion(&t->start_done);
3665 * Removes a device from the thread if_list.
3667 static void _rem_dev_from_if_list(struct pktgen_thread *t,
3668 struct pktgen_dev *pkt_dev)
3670 struct list_head *q, *n;
3671 struct pktgen_dev *p;
3674 list_for_each_safe(q, n, &t->if_list) {
3675 p = list_entry(q, struct pktgen_dev, list);
3677 list_del_rcu(&p->list);
3682 static int pktgen_remove_device(struct pktgen_thread *t,
3683 struct pktgen_dev *pkt_dev)
3685 pr_debug("remove_device pkt_dev=%p\n", pkt_dev);
3687 if (pkt_dev->running) {
3688 pr_warning("WARNING: trying to remove a running interface, stopping it now\n");
3689 pktgen_stop_device(pkt_dev);
3692 /* Dis-associate from the interface */
3694 if (pkt_dev->odev) {
3695 dev_put(pkt_dev->odev);
3696 pkt_dev->odev = NULL;
3699 /* Remove proc before if_list entry, because add_device uses
3700 * list to determine if interface already exist, avoid race
3701 * with proc_create_data() */
3703 proc_remove(pkt_dev->entry);
3705 /* And update the thread if_list */
3706 _rem_dev_from_if_list(t, pkt_dev);
3711 vfree(pkt_dev->flows);
3713 put_page(pkt_dev->page);
3714 kfree_rcu(pkt_dev, rcu);
3718 static int __net_init pg_net_init(struct net *net)
3720 struct pktgen_net *pn = net_generic(net, pg_net_id);
3721 struct proc_dir_entry *pe;
3725 INIT_LIST_HEAD(&pn->pktgen_threads);
3726 pn->pktgen_exiting = false;
3727 pn->proc_dir = proc_mkdir(PG_PROC_DIR, pn->net->proc_net);
3728 if (!pn->proc_dir) {
3729 pr_warn("cannot create /proc/net/%s\n", PG_PROC_DIR);
3732 pe = proc_create(PGCTRL, 0600, pn->proc_dir, &pktgen_fops);
3734 pr_err("cannot create %s procfs entry\n", PGCTRL);
3739 for_each_online_cpu(cpu) {
3742 err = pktgen_create_thread(cpu, pn);
3744 pr_warn("Cannot create thread for cpu %d (%d)\n",
3748 if (list_empty(&pn->pktgen_threads)) {
3749 pr_err("Initialization failed for all threads\n");
3757 remove_proc_entry(PGCTRL, pn->proc_dir);
3759 remove_proc_entry(PG_PROC_DIR, pn->net->proc_net);
3763 static void __net_exit pg_net_exit(struct net *net)
3765 struct pktgen_net *pn = net_generic(net, pg_net_id);
3766 struct pktgen_thread *t;
3767 struct list_head *q, *n;
3770 /* Stop all interfaces & threads */
3771 pn->pktgen_exiting = true;
3773 mutex_lock(&pktgen_thread_lock);
3774 list_splice_init(&pn->pktgen_threads, &list);
3775 mutex_unlock(&pktgen_thread_lock);
3777 list_for_each_safe(q, n, &list) {
3778 t = list_entry(q, struct pktgen_thread, th_list);
3779 list_del(&t->th_list);
3780 kthread_stop(t->tsk);
3784 remove_proc_entry(PGCTRL, pn->proc_dir);
3785 remove_proc_entry(PG_PROC_DIR, pn->net->proc_net);
3788 static struct pernet_operations pg_net_ops = {
3789 .init = pg_net_init,
3790 .exit = pg_net_exit,
3792 .size = sizeof(struct pktgen_net),
3795 static int __init pg_init(void)
3799 pr_info("%s", version);
3800 ret = register_pernet_subsys(&pg_net_ops);
3803 ret = register_netdevice_notifier(&pktgen_notifier_block);
3805 unregister_pernet_subsys(&pg_net_ops);
3810 static void __exit pg_cleanup(void)
3812 unregister_netdevice_notifier(&pktgen_notifier_block);
3813 unregister_pernet_subsys(&pg_net_ops);
3814 /* Don't need rcu_barrier() due to use of kfree_rcu() */
3817 module_init(pg_init);
3818 module_exit(pg_cleanup);
3820 MODULE_AUTHOR("Robert Olsson <robert.olsson@its.uu.se>");
3821 MODULE_DESCRIPTION("Packet Generator tool");
3822 MODULE_LICENSE("GPL");
3823 MODULE_VERSION(VERSION);
3824 module_param(pg_count_d, int, 0);
3825 MODULE_PARM_DESC(pg_count_d, "Default number of packets to inject");
3826 module_param(pg_delay_d, int, 0);
3827 MODULE_PARM_DESC(pg_delay_d, "Default delay between packets (nanoseconds)");
3828 module_param(pg_clone_skb_d, int, 0);
3829 MODULE_PARM_DESC(pg_clone_skb_d, "Default number of copies of the same packet");
3830 module_param(debug, int, 0);
3831 MODULE_PARM_DESC(debug, "Enable debugging of pktgen module");