4 * Uppsala University and
5 * Swedish University of Agricultural Sciences
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:
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
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.
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.
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.
97 * New xmit() return, do_div and misc clean up by Stephen Hemminger
100 * Randy Dunlap fixed u64 printk compiler warning
106 * Removed unused flags F_SET_SRCMAC & F_SET_SRCIP 041230
115 * Fixed src_mac command to set source mac of packet to value specified in
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 "2.75"
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 */
205 #define F_NO_TIMESTAMP (1<<17) /* Don't timestamp packets (default TS) */
207 /* Thread control flag bits */
208 #define T_STOP (1<<0) /* Stop run */
209 #define T_RUN (1<<1) /* Start run */
210 #define T_REMDEVALL (1<<2) /* Remove all devs */
211 #define T_REMDEV (1<<3) /* Remove one dev */
214 #define M_START_XMIT 0 /* Default normal TX */
215 #define M_NETIF_RECEIVE 1 /* Inject packets into stack */
216 #define M_QUEUE_XMIT 2 /* Inject packet into qdisc */
218 /* If lock -- protects updating of if_list */
219 #define if_lock(t) spin_lock(&(t->if_lock));
220 #define if_unlock(t) spin_unlock(&(t->if_lock));
222 /* Used to help with determining the pkts on receive */
223 #define PKTGEN_MAGIC 0xbe9be955
224 #define PG_PROC_DIR "pktgen"
225 #define PGCTRL "pgctrl"
227 #define MAX_CFLOWS 65536
229 #define VLAN_TAG_SIZE(x) ((x)->vlan_id == 0xffff ? 0 : 4)
230 #define SVLAN_TAG_SIZE(x) ((x)->svlan_id == 0xffff ? 0 : 4)
236 struct xfrm_state *x;
242 #define F_INIT (1<<0) /* flow has been initialized */
246 * Try to keep frequent/infrequent used vars. separated.
248 struct proc_dir_entry *entry; /* proc file */
249 struct pktgen_thread *pg_thread;/* the owner */
250 struct list_head list; /* chaining in the thread's run-queue */
251 struct rcu_head rcu; /* freed by RCU */
253 int running; /* if false, the test will stop */
255 /* If min != max, then we will either do a linear iteration, or
256 * we will do a random selection from within the range.
262 int pkt_overhead; /* overhead for MPLS, VLANs, IPSEC etc */
264 int removal_mark; /* non-zero => the device is marked for
265 * removal by worker thread */
268 u64 delay; /* nano-seconds */
270 __u64 count; /* Default No packets to send */
271 __u64 sofar; /* How many pkts we've sent so far */
272 __u64 tx_bytes; /* How many bytes we've transmitted */
273 __u64 errors; /* Errors when trying to transmit, */
275 /* runtime counters relating to clone_skb */
278 int last_ok; /* Was last skb sent?
279 * Or a failed transmit of some sort?
280 * This will keep sequence numbers in order
285 u64 idle_acc; /* nano-seconds */
290 * Use multiple SKBs during packet gen.
291 * If this number is greater than 1, then
292 * that many copies of the same packet will be
293 * sent before a new packet is allocated.
294 * If you want to send 1024 identical packets
295 * before creating a new packet,
296 * set clone_skb to 1024.
299 char dst_min[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
300 char dst_max[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
301 char src_min[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
302 char src_max[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
304 struct in6_addr in6_saddr;
305 struct in6_addr in6_daddr;
306 struct in6_addr cur_in6_daddr;
307 struct in6_addr cur_in6_saddr;
309 struct in6_addr min_in6_daddr;
310 struct in6_addr max_in6_daddr;
311 struct in6_addr min_in6_saddr;
312 struct in6_addr max_in6_saddr;
314 /* If we're doing ranges, random or incremental, then this
315 * defines the min/max for those ranges.
317 __be32 saddr_min; /* inclusive, source IP address */
318 __be32 saddr_max; /* exclusive, source IP address */
319 __be32 daddr_min; /* inclusive, dest IP address */
320 __be32 daddr_max; /* exclusive, dest IP address */
322 __u16 udp_src_min; /* inclusive, source UDP port */
323 __u16 udp_src_max; /* exclusive, source UDP port */
324 __u16 udp_dst_min; /* inclusive, dest UDP port */
325 __u16 udp_dst_max; /* exclusive, dest UDP port */
328 __u8 tos; /* six MSB of (former) IPv4 TOS
329 are for dscp codepoint */
330 __u8 traffic_class; /* ditto for the (former) Traffic Class in IPv6
331 (see RFC 3260, sec. 4) */
334 unsigned int nr_labels; /* Depth of stack, 0 = no MPLS */
335 __be32 labels[MAX_MPLS_LABELS];
337 /* VLAN/SVLAN (802.1Q/Q-in-Q) */
340 __u16 vlan_id; /* 0xffff means no vlan tag */
344 __u16 svlan_id; /* 0xffff means no svlan tag */
346 __u32 src_mac_count; /* How many MACs to iterate through */
347 __u32 dst_mac_count; /* How many MACs to iterate through */
349 unsigned char dst_mac[ETH_ALEN];
350 unsigned char src_mac[ETH_ALEN];
352 __u32 cur_dst_mac_offset;
353 __u32 cur_src_mac_offset;
365 0x00, 0x80, 0xC8, 0x79, 0xB3, 0xCB,
367 We fill in SRC address later
368 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
372 __u16 pad; /* pad out the hh struct to an even 16 bytes */
374 struct sk_buff *skb; /* skb we are to transmit next, used for when we
375 * are transmitting the same one multiple times
377 struct net_device *odev; /* The out-going device.
378 * Note that the device should have it's
379 * pg_info pointer pointing back to this
381 * Set when the user specifies the out-going
382 * device name (not when the inject is
383 * started as it used to do.)
386 struct flow_state *flows;
387 unsigned int cflows; /* Concurrent flows (config) */
388 unsigned int lflow; /* Flow length (config) */
389 unsigned int nflows; /* accumulated flows (stats) */
390 unsigned int curfl; /* current sequenced flow (state)*/
394 __u32 skb_priority; /* skb priority field */
395 unsigned int burst; /* number of duplicated packets to burst */
396 int node; /* Memory node */
399 __u8 ipsmode; /* IPSEC mode (config) */
400 __u8 ipsproto; /* IPSEC type (config) */
402 struct dst_entry dst;
403 struct dst_ops dstops;
416 static int pg_net_id __read_mostly;
420 struct proc_dir_entry *proc_dir;
421 struct list_head pktgen_threads;
425 struct pktgen_thread {
426 spinlock_t if_lock; /* for list of devices */
427 struct list_head if_list; /* All device here */
428 struct list_head th_list;
429 struct task_struct *tsk;
432 /* Field for thread to receive "posted" events terminate,
438 wait_queue_head_t queue;
439 struct completion start_done;
440 struct pktgen_net *net;
446 static const char version[] =
447 "Packet Generator for packet performance testing. "
448 "Version: " VERSION "\n";
450 static int pktgen_remove_device(struct pktgen_thread *t, struct pktgen_dev *i);
451 static int pktgen_add_device(struct pktgen_thread *t, const char *ifname);
452 static struct pktgen_dev *pktgen_find_dev(struct pktgen_thread *t,
453 const char *ifname, bool exact);
454 static int pktgen_device_event(struct notifier_block *, unsigned long, void *);
455 static void pktgen_run_all_threads(struct pktgen_net *pn);
456 static void pktgen_reset_all_threads(struct pktgen_net *pn);
457 static void pktgen_stop_all_threads_ifs(struct pktgen_net *pn);
459 static void pktgen_stop(struct pktgen_thread *t);
460 static void pktgen_clear_counters(struct pktgen_dev *pkt_dev);
462 /* Module parameters, defaults. */
463 static int pg_count_d __read_mostly = 1000;
464 static int pg_delay_d __read_mostly;
465 static int pg_clone_skb_d __read_mostly;
466 static int debug __read_mostly;
468 static DEFINE_MUTEX(pktgen_thread_lock);
470 static struct notifier_block pktgen_notifier_block = {
471 .notifier_call = pktgen_device_event,
475 * /proc handling functions
479 static int pgctrl_show(struct seq_file *seq, void *v)
481 seq_puts(seq, version);
485 static ssize_t pgctrl_write(struct file *file, const char __user *buf,
486 size_t count, loff_t *ppos)
489 struct pktgen_net *pn = net_generic(current->nsproxy->net_ns, pg_net_id);
491 if (!capable(CAP_NET_ADMIN))
497 if (count > sizeof(data))
498 count = sizeof(data);
500 if (copy_from_user(data, buf, count))
503 data[count - 1] = 0; /* Strip trailing '\n' and terminate string */
505 if (!strcmp(data, "stop"))
506 pktgen_stop_all_threads_ifs(pn);
508 else if (!strcmp(data, "start"))
509 pktgen_run_all_threads(pn);
511 else if (!strcmp(data, "reset"))
512 pktgen_reset_all_threads(pn);
520 static int pgctrl_open(struct inode *inode, struct file *file)
522 return single_open(file, pgctrl_show, PDE_DATA(inode));
525 static const struct file_operations pktgen_fops = {
526 .owner = THIS_MODULE,
530 .write = pgctrl_write,
531 .release = single_release,
534 static int pktgen_if_show(struct seq_file *seq, void *v)
536 const struct pktgen_dev *pkt_dev = seq->private;
541 "Params: count %llu min_pkt_size: %u max_pkt_size: %u\n",
542 (unsigned long long)pkt_dev->count, pkt_dev->min_pkt_size,
543 pkt_dev->max_pkt_size);
546 " frags: %d delay: %llu clone_skb: %d ifname: %s\n",
547 pkt_dev->nfrags, (unsigned long long) pkt_dev->delay,
548 pkt_dev->clone_skb, pkt_dev->odevname);
550 seq_printf(seq, " flows: %u flowlen: %u\n", pkt_dev->cflows,
554 " queue_map_min: %u queue_map_max: %u\n",
555 pkt_dev->queue_map_min,
556 pkt_dev->queue_map_max);
558 if (pkt_dev->skb_priority)
559 seq_printf(seq, " skb_priority: %u\n",
560 pkt_dev->skb_priority);
562 if (pkt_dev->flags & F_IPV6) {
564 " saddr: %pI6c min_saddr: %pI6c max_saddr: %pI6c\n"
565 " daddr: %pI6c min_daddr: %pI6c max_daddr: %pI6c\n",
567 &pkt_dev->min_in6_saddr, &pkt_dev->max_in6_saddr,
569 &pkt_dev->min_in6_daddr, &pkt_dev->max_in6_daddr);
572 " dst_min: %s dst_max: %s\n",
573 pkt_dev->dst_min, pkt_dev->dst_max);
575 " src_min: %s src_max: %s\n",
576 pkt_dev->src_min, pkt_dev->src_max);
579 seq_puts(seq, " src_mac: ");
581 seq_printf(seq, "%pM ",
582 is_zero_ether_addr(pkt_dev->src_mac) ?
583 pkt_dev->odev->dev_addr : pkt_dev->src_mac);
585 seq_puts(seq, "dst_mac: ");
586 seq_printf(seq, "%pM\n", pkt_dev->dst_mac);
589 " udp_src_min: %d udp_src_max: %d"
590 " udp_dst_min: %d udp_dst_max: %d\n",
591 pkt_dev->udp_src_min, pkt_dev->udp_src_max,
592 pkt_dev->udp_dst_min, pkt_dev->udp_dst_max);
595 " src_mac_count: %d dst_mac_count: %d\n",
596 pkt_dev->src_mac_count, pkt_dev->dst_mac_count);
598 if (pkt_dev->nr_labels) {
600 seq_puts(seq, " mpls: ");
601 for (i = 0; i < pkt_dev->nr_labels; i++)
602 seq_printf(seq, "%08x%s", ntohl(pkt_dev->labels[i]),
603 i == pkt_dev->nr_labels-1 ? "\n" : ", ");
606 if (pkt_dev->vlan_id != 0xffff)
607 seq_printf(seq, " vlan_id: %u vlan_p: %u vlan_cfi: %u\n",
608 pkt_dev->vlan_id, pkt_dev->vlan_p,
611 if (pkt_dev->svlan_id != 0xffff)
612 seq_printf(seq, " svlan_id: %u vlan_p: %u vlan_cfi: %u\n",
613 pkt_dev->svlan_id, pkt_dev->svlan_p,
617 seq_printf(seq, " tos: 0x%02x\n", pkt_dev->tos);
619 if (pkt_dev->traffic_class)
620 seq_printf(seq, " traffic_class: 0x%02x\n", pkt_dev->traffic_class);
622 if (pkt_dev->burst > 1)
623 seq_printf(seq, " burst: %d\n", pkt_dev->burst);
625 if (pkt_dev->node >= 0)
626 seq_printf(seq, " node: %d\n", pkt_dev->node);
628 if (pkt_dev->xmit_mode == M_NETIF_RECEIVE)
629 seq_puts(seq, " xmit_mode: netif_receive\n");
630 else if (pkt_dev->xmit_mode == M_QUEUE_XMIT)
631 seq_puts(seq, " xmit_mode: xmit_queue\n");
633 seq_puts(seq, " Flags: ");
635 if (pkt_dev->flags & F_IPV6)
636 seq_puts(seq, "IPV6 ");
638 if (pkt_dev->flags & F_IPSRC_RND)
639 seq_puts(seq, "IPSRC_RND ");
641 if (pkt_dev->flags & F_IPDST_RND)
642 seq_puts(seq, "IPDST_RND ");
644 if (pkt_dev->flags & F_TXSIZE_RND)
645 seq_puts(seq, "TXSIZE_RND ");
647 if (pkt_dev->flags & F_UDPSRC_RND)
648 seq_puts(seq, "UDPSRC_RND ");
650 if (pkt_dev->flags & F_UDPDST_RND)
651 seq_puts(seq, "UDPDST_RND ");
653 if (pkt_dev->flags & F_UDPCSUM)
654 seq_puts(seq, "UDPCSUM ");
656 if (pkt_dev->flags & F_NO_TIMESTAMP)
657 seq_puts(seq, "NO_TIMESTAMP ");
659 if (pkt_dev->flags & F_MPLS_RND)
660 seq_puts(seq, "MPLS_RND ");
662 if (pkt_dev->flags & F_QUEUE_MAP_RND)
663 seq_puts(seq, "QUEUE_MAP_RND ");
665 if (pkt_dev->flags & F_QUEUE_MAP_CPU)
666 seq_puts(seq, "QUEUE_MAP_CPU ");
668 if (pkt_dev->cflows) {
669 if (pkt_dev->flags & F_FLOW_SEQ)
670 seq_puts(seq, "FLOW_SEQ "); /*in sequence flows*/
672 seq_puts(seq, "FLOW_RND ");
676 if (pkt_dev->flags & F_IPSEC_ON) {
677 seq_puts(seq, "IPSEC ");
679 seq_printf(seq, "spi:%u", pkt_dev->spi);
683 if (pkt_dev->flags & F_MACSRC_RND)
684 seq_puts(seq, "MACSRC_RND ");
686 if (pkt_dev->flags & F_MACDST_RND)
687 seq_puts(seq, "MACDST_RND ");
689 if (pkt_dev->flags & F_VID_RND)
690 seq_puts(seq, "VID_RND ");
692 if (pkt_dev->flags & F_SVID_RND)
693 seq_puts(seq, "SVID_RND ");
695 if (pkt_dev->flags & F_NODE)
696 seq_puts(seq, "NODE_ALLOC ");
700 /* not really stopped, more like last-running-at */
701 stopped = pkt_dev->running ? ktime_get() : pkt_dev->stopped_at;
702 idle = pkt_dev->idle_acc;
703 do_div(idle, NSEC_PER_USEC);
706 "Current:\n pkts-sofar: %llu errors: %llu\n",
707 (unsigned long long)pkt_dev->sofar,
708 (unsigned long long)pkt_dev->errors);
711 " started: %lluus stopped: %lluus idle: %lluus\n",
712 (unsigned long long) ktime_to_us(pkt_dev->started_at),
713 (unsigned long long) ktime_to_us(stopped),
714 (unsigned long long) idle);
717 " seq_num: %d cur_dst_mac_offset: %d cur_src_mac_offset: %d\n",
718 pkt_dev->seq_num, pkt_dev->cur_dst_mac_offset,
719 pkt_dev->cur_src_mac_offset);
721 if (pkt_dev->flags & F_IPV6) {
722 seq_printf(seq, " cur_saddr: %pI6c cur_daddr: %pI6c\n",
723 &pkt_dev->cur_in6_saddr,
724 &pkt_dev->cur_in6_daddr);
726 seq_printf(seq, " cur_saddr: %pI4 cur_daddr: %pI4\n",
727 &pkt_dev->cur_saddr, &pkt_dev->cur_daddr);
729 seq_printf(seq, " cur_udp_dst: %d cur_udp_src: %d\n",
730 pkt_dev->cur_udp_dst, pkt_dev->cur_udp_src);
732 seq_printf(seq, " cur_queue_map: %u\n", pkt_dev->cur_queue_map);
734 seq_printf(seq, " flows: %u\n", pkt_dev->nflows);
736 if (pkt_dev->result[0])
737 seq_printf(seq, "Result: %s\n", pkt_dev->result);
739 seq_puts(seq, "Result: Idle\n");
745 static int hex32_arg(const char __user *user_buffer, unsigned long maxlen,
751 for (; i < maxlen; i++) {
755 if (get_user(c, &user_buffer[i]))
757 value = hex_to_bin(c);
766 static int count_trail_chars(const char __user * user_buffer,
771 for (i = 0; i < maxlen; i++) {
773 if (get_user(c, &user_buffer[i]))
791 static long num_arg(const char __user *user_buffer, unsigned long maxlen,
797 for (i = 0; i < maxlen; i++) {
799 if (get_user(c, &user_buffer[i]))
801 if ((c >= '0') && (c <= '9')) {
810 static int strn_len(const char __user * user_buffer, unsigned int maxlen)
814 for (i = 0; i < maxlen; i++) {
816 if (get_user(c, &user_buffer[i]))
833 static ssize_t get_labels(const char __user *buffer, struct pktgen_dev *pkt_dev)
840 pkt_dev->nr_labels = 0;
843 len = hex32_arg(&buffer[i], 8, &tmp);
846 pkt_dev->labels[n] = htonl(tmp);
847 if (pkt_dev->labels[n] & MPLS_STACK_BOTTOM)
848 pkt_dev->flags |= F_MPLS_RND;
850 if (get_user(c, &buffer[i]))
854 if (n >= MAX_MPLS_LABELS)
858 pkt_dev->nr_labels = n;
862 static ssize_t pktgen_if_write(struct file *file,
863 const char __user * user_buffer, size_t count,
866 struct seq_file *seq = file->private_data;
867 struct pktgen_dev *pkt_dev = seq->private;
869 char name[16], valstr[32];
870 unsigned long value = 0;
871 char *pg_result = NULL;
875 pg_result = &(pkt_dev->result[0]);
878 pr_warn("wrong command format\n");
883 tmp = count_trail_chars(user_buffer, max);
885 pr_warn("illegal format\n");
890 /* Read variable name */
892 len = strn_len(&user_buffer[i], sizeof(name) - 1);
896 memset(name, 0, sizeof(name));
897 if (copy_from_user(name, &user_buffer[i], len))
902 len = count_trail_chars(&user_buffer[i], max);
909 size_t copy = min_t(size_t, count, 1023);
911 if (copy_from_user(tb, user_buffer, copy))
914 pr_debug("%s,%lu buffer -:%s:-\n",
915 name, (unsigned long)count, tb);
918 if (!strcmp(name, "min_pkt_size")) {
919 len = num_arg(&user_buffer[i], 10, &value);
924 if (value < 14 + 20 + 8)
926 if (value != pkt_dev->min_pkt_size) {
927 pkt_dev->min_pkt_size = value;
928 pkt_dev->cur_pkt_size = value;
930 sprintf(pg_result, "OK: min_pkt_size=%u",
931 pkt_dev->min_pkt_size);
935 if (!strcmp(name, "max_pkt_size")) {
936 len = num_arg(&user_buffer[i], 10, &value);
941 if (value < 14 + 20 + 8)
943 if (value != pkt_dev->max_pkt_size) {
944 pkt_dev->max_pkt_size = value;
945 pkt_dev->cur_pkt_size = value;
947 sprintf(pg_result, "OK: max_pkt_size=%u",
948 pkt_dev->max_pkt_size);
952 /* Shortcut for min = max */
954 if (!strcmp(name, "pkt_size")) {
955 len = num_arg(&user_buffer[i], 10, &value);
960 if (value < 14 + 20 + 8)
962 if (value != pkt_dev->min_pkt_size) {
963 pkt_dev->min_pkt_size = value;
964 pkt_dev->max_pkt_size = value;
965 pkt_dev->cur_pkt_size = value;
967 sprintf(pg_result, "OK: pkt_size=%u", pkt_dev->min_pkt_size);
971 if (!strcmp(name, "debug")) {
972 len = num_arg(&user_buffer[i], 10, &value);
978 sprintf(pg_result, "OK: debug=%u", debug);
982 if (!strcmp(name, "frags")) {
983 len = num_arg(&user_buffer[i], 10, &value);
988 pkt_dev->nfrags = value;
989 sprintf(pg_result, "OK: frags=%u", pkt_dev->nfrags);
992 if (!strcmp(name, "delay")) {
993 len = num_arg(&user_buffer[i], 10, &value);
998 if (value == 0x7FFFFFFF)
999 pkt_dev->delay = ULLONG_MAX;
1001 pkt_dev->delay = (u64)value;
1003 sprintf(pg_result, "OK: delay=%llu",
1004 (unsigned long long) pkt_dev->delay);
1007 if (!strcmp(name, "rate")) {
1008 len = num_arg(&user_buffer[i], 10, &value);
1015 pkt_dev->delay = pkt_dev->min_pkt_size*8*NSEC_PER_USEC/value;
1017 pr_info("Delay set at: %llu ns\n", pkt_dev->delay);
1019 sprintf(pg_result, "OK: rate=%lu", value);
1022 if (!strcmp(name, "ratep")) {
1023 len = num_arg(&user_buffer[i], 10, &value);
1030 pkt_dev->delay = NSEC_PER_SEC/value;
1032 pr_info("Delay set at: %llu ns\n", pkt_dev->delay);
1034 sprintf(pg_result, "OK: rate=%lu", value);
1037 if (!strcmp(name, "udp_src_min")) {
1038 len = num_arg(&user_buffer[i], 10, &value);
1043 if (value != pkt_dev->udp_src_min) {
1044 pkt_dev->udp_src_min = value;
1045 pkt_dev->cur_udp_src = value;
1047 sprintf(pg_result, "OK: udp_src_min=%u", pkt_dev->udp_src_min);
1050 if (!strcmp(name, "udp_dst_min")) {
1051 len = num_arg(&user_buffer[i], 10, &value);
1056 if (value != pkt_dev->udp_dst_min) {
1057 pkt_dev->udp_dst_min = value;
1058 pkt_dev->cur_udp_dst = value;
1060 sprintf(pg_result, "OK: udp_dst_min=%u", pkt_dev->udp_dst_min);
1063 if (!strcmp(name, "udp_src_max")) {
1064 len = num_arg(&user_buffer[i], 10, &value);
1069 if (value != pkt_dev->udp_src_max) {
1070 pkt_dev->udp_src_max = value;
1071 pkt_dev->cur_udp_src = value;
1073 sprintf(pg_result, "OK: udp_src_max=%u", pkt_dev->udp_src_max);
1076 if (!strcmp(name, "udp_dst_max")) {
1077 len = num_arg(&user_buffer[i], 10, &value);
1082 if (value != pkt_dev->udp_dst_max) {
1083 pkt_dev->udp_dst_max = value;
1084 pkt_dev->cur_udp_dst = value;
1086 sprintf(pg_result, "OK: udp_dst_max=%u", pkt_dev->udp_dst_max);
1089 if (!strcmp(name, "clone_skb")) {
1090 len = num_arg(&user_buffer[i], 10, &value);
1094 ((pkt_dev->xmit_mode == M_NETIF_RECEIVE) ||
1095 !(pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING)))
1098 pkt_dev->clone_skb = value;
1100 sprintf(pg_result, "OK: clone_skb=%d", pkt_dev->clone_skb);
1103 if (!strcmp(name, "count")) {
1104 len = num_arg(&user_buffer[i], 10, &value);
1109 pkt_dev->count = value;
1110 sprintf(pg_result, "OK: count=%llu",
1111 (unsigned long long)pkt_dev->count);
1114 if (!strcmp(name, "src_mac_count")) {
1115 len = num_arg(&user_buffer[i], 10, &value);
1120 if (pkt_dev->src_mac_count != value) {
1121 pkt_dev->src_mac_count = value;
1122 pkt_dev->cur_src_mac_offset = 0;
1124 sprintf(pg_result, "OK: src_mac_count=%d",
1125 pkt_dev->src_mac_count);
1128 if (!strcmp(name, "dst_mac_count")) {
1129 len = num_arg(&user_buffer[i], 10, &value);
1134 if (pkt_dev->dst_mac_count != value) {
1135 pkt_dev->dst_mac_count = value;
1136 pkt_dev->cur_dst_mac_offset = 0;
1138 sprintf(pg_result, "OK: dst_mac_count=%d",
1139 pkt_dev->dst_mac_count);
1142 if (!strcmp(name, "burst")) {
1143 len = num_arg(&user_buffer[i], 10, &value);
1149 ((pkt_dev->xmit_mode == M_QUEUE_XMIT) ||
1150 ((pkt_dev->xmit_mode == M_START_XMIT) &&
1151 (!(pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING)))))
1153 pkt_dev->burst = value < 1 ? 1 : value;
1154 sprintf(pg_result, "OK: burst=%d", pkt_dev->burst);
1157 if (!strcmp(name, "node")) {
1158 len = num_arg(&user_buffer[i], 10, &value);
1164 if (node_possible(value)) {
1165 pkt_dev->node = value;
1166 sprintf(pg_result, "OK: node=%d", pkt_dev->node);
1167 if (pkt_dev->page) {
1168 put_page(pkt_dev->page);
1169 pkt_dev->page = NULL;
1173 sprintf(pg_result, "ERROR: node not possible");
1176 if (!strcmp(name, "xmit_mode")) {
1180 len = strn_len(&user_buffer[i], sizeof(f) - 1);
1184 if (copy_from_user(f, &user_buffer[i], len))
1188 if (strcmp(f, "start_xmit") == 0) {
1189 pkt_dev->xmit_mode = M_START_XMIT;
1190 } else if (strcmp(f, "netif_receive") == 0) {
1191 /* clone_skb set earlier, not supported in this mode */
1192 if (pkt_dev->clone_skb > 0)
1195 pkt_dev->xmit_mode = M_NETIF_RECEIVE;
1197 /* make sure new packet is allocated every time
1198 * pktgen_xmit() is called
1200 pkt_dev->last_ok = 1;
1202 /* override clone_skb if user passed default value
1203 * at module loading time
1205 pkt_dev->clone_skb = 0;
1206 } else if (strcmp(f, "queue_xmit") == 0) {
1207 pkt_dev->xmit_mode = M_QUEUE_XMIT;
1208 pkt_dev->last_ok = 1;
1211 "xmit_mode -:%s:- unknown\nAvailable modes: %s",
1212 f, "start_xmit, netif_receive\n");
1215 sprintf(pg_result, "OK: xmit_mode=%s", f);
1218 if (!strcmp(name, "flag")) {
1221 len = strn_len(&user_buffer[i], sizeof(f) - 1);
1225 if (copy_from_user(f, &user_buffer[i], len))
1228 if (strcmp(f, "IPSRC_RND") == 0)
1229 pkt_dev->flags |= F_IPSRC_RND;
1231 else if (strcmp(f, "!IPSRC_RND") == 0)
1232 pkt_dev->flags &= ~F_IPSRC_RND;
1234 else if (strcmp(f, "TXSIZE_RND") == 0)
1235 pkt_dev->flags |= F_TXSIZE_RND;
1237 else if (strcmp(f, "!TXSIZE_RND") == 0)
1238 pkt_dev->flags &= ~F_TXSIZE_RND;
1240 else if (strcmp(f, "IPDST_RND") == 0)
1241 pkt_dev->flags |= F_IPDST_RND;
1243 else if (strcmp(f, "!IPDST_RND") == 0)
1244 pkt_dev->flags &= ~F_IPDST_RND;
1246 else if (strcmp(f, "UDPSRC_RND") == 0)
1247 pkt_dev->flags |= F_UDPSRC_RND;
1249 else if (strcmp(f, "!UDPSRC_RND") == 0)
1250 pkt_dev->flags &= ~F_UDPSRC_RND;
1252 else if (strcmp(f, "UDPDST_RND") == 0)
1253 pkt_dev->flags |= F_UDPDST_RND;
1255 else if (strcmp(f, "!UDPDST_RND") == 0)
1256 pkt_dev->flags &= ~F_UDPDST_RND;
1258 else if (strcmp(f, "MACSRC_RND") == 0)
1259 pkt_dev->flags |= F_MACSRC_RND;
1261 else if (strcmp(f, "!MACSRC_RND") == 0)
1262 pkt_dev->flags &= ~F_MACSRC_RND;
1264 else if (strcmp(f, "MACDST_RND") == 0)
1265 pkt_dev->flags |= F_MACDST_RND;
1267 else if (strcmp(f, "!MACDST_RND") == 0)
1268 pkt_dev->flags &= ~F_MACDST_RND;
1270 else if (strcmp(f, "MPLS_RND") == 0)
1271 pkt_dev->flags |= F_MPLS_RND;
1273 else if (strcmp(f, "!MPLS_RND") == 0)
1274 pkt_dev->flags &= ~F_MPLS_RND;
1276 else if (strcmp(f, "VID_RND") == 0)
1277 pkt_dev->flags |= F_VID_RND;
1279 else if (strcmp(f, "!VID_RND") == 0)
1280 pkt_dev->flags &= ~F_VID_RND;
1282 else if (strcmp(f, "SVID_RND") == 0)
1283 pkt_dev->flags |= F_SVID_RND;
1285 else if (strcmp(f, "!SVID_RND") == 0)
1286 pkt_dev->flags &= ~F_SVID_RND;
1288 else if (strcmp(f, "FLOW_SEQ") == 0)
1289 pkt_dev->flags |= F_FLOW_SEQ;
1291 else if (strcmp(f, "QUEUE_MAP_RND") == 0)
1292 pkt_dev->flags |= F_QUEUE_MAP_RND;
1294 else if (strcmp(f, "!QUEUE_MAP_RND") == 0)
1295 pkt_dev->flags &= ~F_QUEUE_MAP_RND;
1297 else if (strcmp(f, "QUEUE_MAP_CPU") == 0)
1298 pkt_dev->flags |= F_QUEUE_MAP_CPU;
1300 else if (strcmp(f, "!QUEUE_MAP_CPU") == 0)
1301 pkt_dev->flags &= ~F_QUEUE_MAP_CPU;
1303 else if (strcmp(f, "IPSEC") == 0)
1304 pkt_dev->flags |= F_IPSEC_ON;
1307 else if (strcmp(f, "!IPV6") == 0)
1308 pkt_dev->flags &= ~F_IPV6;
1310 else if (strcmp(f, "NODE_ALLOC") == 0)
1311 pkt_dev->flags |= F_NODE;
1313 else if (strcmp(f, "!NODE_ALLOC") == 0)
1314 pkt_dev->flags &= ~F_NODE;
1316 else if (strcmp(f, "UDPCSUM") == 0)
1317 pkt_dev->flags |= F_UDPCSUM;
1319 else if (strcmp(f, "!UDPCSUM") == 0)
1320 pkt_dev->flags &= ~F_UDPCSUM;
1322 else if (strcmp(f, "NO_TIMESTAMP") == 0)
1323 pkt_dev->flags |= F_NO_TIMESTAMP;
1325 else if (strcmp(f, "!NO_TIMESTAMP") == 0)
1326 pkt_dev->flags &= ~F_NO_TIMESTAMP;
1330 "Flag -:%s:- unknown\nAvailable flags, (prepend ! to un-set flag):\n%s",
1332 "IPSRC_RND, IPDST_RND, UDPSRC_RND, UDPDST_RND, "
1333 "MACSRC_RND, MACDST_RND, TXSIZE_RND, IPV6, "
1334 "MPLS_RND, VID_RND, SVID_RND, FLOW_SEQ, "
1335 "QUEUE_MAP_RND, QUEUE_MAP_CPU, UDPCSUM, "
1343 sprintf(pg_result, "OK: flags=0x%x", pkt_dev->flags);
1346 if (!strcmp(name, "dst_min") || !strcmp(name, "dst")) {
1347 len = strn_len(&user_buffer[i], sizeof(pkt_dev->dst_min) - 1);
1351 if (copy_from_user(buf, &user_buffer[i], len))
1354 if (strcmp(buf, pkt_dev->dst_min) != 0) {
1355 memset(pkt_dev->dst_min, 0, sizeof(pkt_dev->dst_min));
1356 strncpy(pkt_dev->dst_min, buf, len);
1357 pkt_dev->daddr_min = in_aton(pkt_dev->dst_min);
1358 pkt_dev->cur_daddr = pkt_dev->daddr_min;
1361 pr_debug("dst_min set to: %s\n", pkt_dev->dst_min);
1363 sprintf(pg_result, "OK: dst_min=%s", pkt_dev->dst_min);
1366 if (!strcmp(name, "dst_max")) {
1367 len = strn_len(&user_buffer[i], sizeof(pkt_dev->dst_max) - 1);
1372 if (copy_from_user(buf, &user_buffer[i], len))
1376 if (strcmp(buf, pkt_dev->dst_max) != 0) {
1377 memset(pkt_dev->dst_max, 0, sizeof(pkt_dev->dst_max));
1378 strncpy(pkt_dev->dst_max, buf, len);
1379 pkt_dev->daddr_max = in_aton(pkt_dev->dst_max);
1380 pkt_dev->cur_daddr = pkt_dev->daddr_max;
1383 pr_debug("dst_max set to: %s\n", pkt_dev->dst_max);
1385 sprintf(pg_result, "OK: dst_max=%s", pkt_dev->dst_max);
1388 if (!strcmp(name, "dst6")) {
1389 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1393 pkt_dev->flags |= F_IPV6;
1395 if (copy_from_user(buf, &user_buffer[i], len))
1399 in6_pton(buf, -1, pkt_dev->in6_daddr.s6_addr, -1, NULL);
1400 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->in6_daddr);
1402 pkt_dev->cur_in6_daddr = pkt_dev->in6_daddr;
1405 pr_debug("dst6 set to: %s\n", buf);
1408 sprintf(pg_result, "OK: dst6=%s", buf);
1411 if (!strcmp(name, "dst6_min")) {
1412 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1416 pkt_dev->flags |= F_IPV6;
1418 if (copy_from_user(buf, &user_buffer[i], len))
1422 in6_pton(buf, -1, pkt_dev->min_in6_daddr.s6_addr, -1, NULL);
1423 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->min_in6_daddr);
1425 pkt_dev->cur_in6_daddr = pkt_dev->min_in6_daddr;
1427 pr_debug("dst6_min set to: %s\n", buf);
1430 sprintf(pg_result, "OK: dst6_min=%s", buf);
1433 if (!strcmp(name, "dst6_max")) {
1434 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1438 pkt_dev->flags |= F_IPV6;
1440 if (copy_from_user(buf, &user_buffer[i], len))
1444 in6_pton(buf, -1, pkt_dev->max_in6_daddr.s6_addr, -1, NULL);
1445 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->max_in6_daddr);
1448 pr_debug("dst6_max set to: %s\n", buf);
1451 sprintf(pg_result, "OK: dst6_max=%s", buf);
1454 if (!strcmp(name, "src6")) {
1455 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1459 pkt_dev->flags |= F_IPV6;
1461 if (copy_from_user(buf, &user_buffer[i], len))
1465 in6_pton(buf, -1, pkt_dev->in6_saddr.s6_addr, -1, NULL);
1466 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->in6_saddr);
1468 pkt_dev->cur_in6_saddr = pkt_dev->in6_saddr;
1471 pr_debug("src6 set to: %s\n", buf);
1474 sprintf(pg_result, "OK: src6=%s", buf);
1477 if (!strcmp(name, "src_min")) {
1478 len = strn_len(&user_buffer[i], sizeof(pkt_dev->src_min) - 1);
1482 if (copy_from_user(buf, &user_buffer[i], len))
1485 if (strcmp(buf, pkt_dev->src_min) != 0) {
1486 memset(pkt_dev->src_min, 0, sizeof(pkt_dev->src_min));
1487 strncpy(pkt_dev->src_min, buf, len);
1488 pkt_dev->saddr_min = in_aton(pkt_dev->src_min);
1489 pkt_dev->cur_saddr = pkt_dev->saddr_min;
1492 pr_debug("src_min set to: %s\n", pkt_dev->src_min);
1494 sprintf(pg_result, "OK: src_min=%s", pkt_dev->src_min);
1497 if (!strcmp(name, "src_max")) {
1498 len = strn_len(&user_buffer[i], sizeof(pkt_dev->src_max) - 1);
1502 if (copy_from_user(buf, &user_buffer[i], len))
1505 if (strcmp(buf, pkt_dev->src_max) != 0) {
1506 memset(pkt_dev->src_max, 0, sizeof(pkt_dev->src_max));
1507 strncpy(pkt_dev->src_max, buf, len);
1508 pkt_dev->saddr_max = in_aton(pkt_dev->src_max);
1509 pkt_dev->cur_saddr = pkt_dev->saddr_max;
1512 pr_debug("src_max set to: %s\n", pkt_dev->src_max);
1514 sprintf(pg_result, "OK: src_max=%s", pkt_dev->src_max);
1517 if (!strcmp(name, "dst_mac")) {
1518 len = strn_len(&user_buffer[i], sizeof(valstr) - 1);
1522 memset(valstr, 0, sizeof(valstr));
1523 if (copy_from_user(valstr, &user_buffer[i], len))
1526 if (!mac_pton(valstr, pkt_dev->dst_mac))
1528 /* Set up Dest MAC */
1529 ether_addr_copy(&pkt_dev->hh[0], pkt_dev->dst_mac);
1531 sprintf(pg_result, "OK: dstmac %pM", pkt_dev->dst_mac);
1534 if (!strcmp(name, "src_mac")) {
1535 len = strn_len(&user_buffer[i], sizeof(valstr) - 1);
1539 memset(valstr, 0, sizeof(valstr));
1540 if (copy_from_user(valstr, &user_buffer[i], len))
1543 if (!mac_pton(valstr, pkt_dev->src_mac))
1545 /* Set up Src MAC */
1546 ether_addr_copy(&pkt_dev->hh[6], pkt_dev->src_mac);
1548 sprintf(pg_result, "OK: srcmac %pM", pkt_dev->src_mac);
1552 if (!strcmp(name, "clear_counters")) {
1553 pktgen_clear_counters(pkt_dev);
1554 sprintf(pg_result, "OK: Clearing counters.\n");
1558 if (!strcmp(name, "flows")) {
1559 len = num_arg(&user_buffer[i], 10, &value);
1564 if (value > MAX_CFLOWS)
1567 pkt_dev->cflows = value;
1568 sprintf(pg_result, "OK: flows=%u", pkt_dev->cflows);
1572 if (!strcmp(name, "spi")) {
1573 len = num_arg(&user_buffer[i], 10, &value);
1578 pkt_dev->spi = value;
1579 sprintf(pg_result, "OK: spi=%u", pkt_dev->spi);
1583 if (!strcmp(name, "flowlen")) {
1584 len = num_arg(&user_buffer[i], 10, &value);
1589 pkt_dev->lflow = value;
1590 sprintf(pg_result, "OK: flowlen=%u", pkt_dev->lflow);
1594 if (!strcmp(name, "queue_map_min")) {
1595 len = num_arg(&user_buffer[i], 5, &value);
1600 pkt_dev->queue_map_min = value;
1601 sprintf(pg_result, "OK: queue_map_min=%u", pkt_dev->queue_map_min);
1605 if (!strcmp(name, "queue_map_max")) {
1606 len = num_arg(&user_buffer[i], 5, &value);
1611 pkt_dev->queue_map_max = value;
1612 sprintf(pg_result, "OK: queue_map_max=%u", pkt_dev->queue_map_max);
1616 if (!strcmp(name, "mpls")) {
1617 unsigned int n, cnt;
1619 len = get_labels(&user_buffer[i], pkt_dev);
1623 cnt = sprintf(pg_result, "OK: mpls=");
1624 for (n = 0; n < pkt_dev->nr_labels; n++)
1625 cnt += sprintf(pg_result + cnt,
1626 "%08x%s", ntohl(pkt_dev->labels[n]),
1627 n == pkt_dev->nr_labels-1 ? "" : ",");
1629 if (pkt_dev->nr_labels && pkt_dev->vlan_id != 0xffff) {
1630 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */
1631 pkt_dev->svlan_id = 0xffff;
1634 pr_debug("VLAN/SVLAN auto turned off\n");
1639 if (!strcmp(name, "vlan_id")) {
1640 len = num_arg(&user_buffer[i], 4, &value);
1645 if (value <= 4095) {
1646 pkt_dev->vlan_id = value; /* turn on VLAN */
1649 pr_debug("VLAN turned on\n");
1651 if (debug && pkt_dev->nr_labels)
1652 pr_debug("MPLS auto turned off\n");
1654 pkt_dev->nr_labels = 0; /* turn off MPLS */
1655 sprintf(pg_result, "OK: vlan_id=%u", pkt_dev->vlan_id);
1657 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */
1658 pkt_dev->svlan_id = 0xffff;
1661 pr_debug("VLAN/SVLAN turned off\n");
1666 if (!strcmp(name, "vlan_p")) {
1667 len = num_arg(&user_buffer[i], 1, &value);
1672 if ((value <= 7) && (pkt_dev->vlan_id != 0xffff)) {
1673 pkt_dev->vlan_p = value;
1674 sprintf(pg_result, "OK: vlan_p=%u", pkt_dev->vlan_p);
1676 sprintf(pg_result, "ERROR: vlan_p must be 0-7");
1681 if (!strcmp(name, "vlan_cfi")) {
1682 len = num_arg(&user_buffer[i], 1, &value);
1687 if ((value <= 1) && (pkt_dev->vlan_id != 0xffff)) {
1688 pkt_dev->vlan_cfi = value;
1689 sprintf(pg_result, "OK: vlan_cfi=%u", pkt_dev->vlan_cfi);
1691 sprintf(pg_result, "ERROR: vlan_cfi must be 0-1");
1696 if (!strcmp(name, "svlan_id")) {
1697 len = num_arg(&user_buffer[i], 4, &value);
1702 if ((value <= 4095) && ((pkt_dev->vlan_id != 0xffff))) {
1703 pkt_dev->svlan_id = value; /* turn on SVLAN */
1706 pr_debug("SVLAN turned on\n");
1708 if (debug && pkt_dev->nr_labels)
1709 pr_debug("MPLS auto turned off\n");
1711 pkt_dev->nr_labels = 0; /* turn off MPLS */
1712 sprintf(pg_result, "OK: svlan_id=%u", pkt_dev->svlan_id);
1714 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */
1715 pkt_dev->svlan_id = 0xffff;
1718 pr_debug("VLAN/SVLAN turned off\n");
1723 if (!strcmp(name, "svlan_p")) {
1724 len = num_arg(&user_buffer[i], 1, &value);
1729 if ((value <= 7) && (pkt_dev->svlan_id != 0xffff)) {
1730 pkt_dev->svlan_p = value;
1731 sprintf(pg_result, "OK: svlan_p=%u", pkt_dev->svlan_p);
1733 sprintf(pg_result, "ERROR: svlan_p must be 0-7");
1738 if (!strcmp(name, "svlan_cfi")) {
1739 len = num_arg(&user_buffer[i], 1, &value);
1744 if ((value <= 1) && (pkt_dev->svlan_id != 0xffff)) {
1745 pkt_dev->svlan_cfi = value;
1746 sprintf(pg_result, "OK: svlan_cfi=%u", pkt_dev->svlan_cfi);
1748 sprintf(pg_result, "ERROR: svlan_cfi must be 0-1");
1753 if (!strcmp(name, "tos")) {
1754 __u32 tmp_value = 0;
1755 len = hex32_arg(&user_buffer[i], 2, &tmp_value);
1761 pkt_dev->tos = tmp_value;
1762 sprintf(pg_result, "OK: tos=0x%02x", pkt_dev->tos);
1764 sprintf(pg_result, "ERROR: tos must be 00-ff");
1769 if (!strcmp(name, "traffic_class")) {
1770 __u32 tmp_value = 0;
1771 len = hex32_arg(&user_buffer[i], 2, &tmp_value);
1777 pkt_dev->traffic_class = tmp_value;
1778 sprintf(pg_result, "OK: traffic_class=0x%02x", pkt_dev->traffic_class);
1780 sprintf(pg_result, "ERROR: traffic_class must be 00-ff");
1785 if (!strcmp(name, "skb_priority")) {
1786 len = num_arg(&user_buffer[i], 9, &value);
1791 pkt_dev->skb_priority = value;
1792 sprintf(pg_result, "OK: skb_priority=%i",
1793 pkt_dev->skb_priority);
1797 sprintf(pkt_dev->result, "No such parameter \"%s\"", name);
1801 static int pktgen_if_open(struct inode *inode, struct file *file)
1803 return single_open(file, pktgen_if_show, PDE_DATA(inode));
1806 static const struct file_operations pktgen_if_fops = {
1807 .owner = THIS_MODULE,
1808 .open = pktgen_if_open,
1810 .llseek = seq_lseek,
1811 .write = pktgen_if_write,
1812 .release = single_release,
1815 static int pktgen_thread_show(struct seq_file *seq, void *v)
1817 struct pktgen_thread *t = seq->private;
1818 const struct pktgen_dev *pkt_dev;
1822 seq_puts(seq, "Running: ");
1825 list_for_each_entry_rcu(pkt_dev, &t->if_list, list)
1826 if (pkt_dev->running)
1827 seq_printf(seq, "%s ", pkt_dev->odevname);
1829 seq_puts(seq, "\nStopped: ");
1831 list_for_each_entry_rcu(pkt_dev, &t->if_list, list)
1832 if (!pkt_dev->running)
1833 seq_printf(seq, "%s ", pkt_dev->odevname);
1836 seq_printf(seq, "\nResult: %s\n", t->result);
1838 seq_puts(seq, "\nResult: NA\n");
1845 static ssize_t pktgen_thread_write(struct file *file,
1846 const char __user * user_buffer,
1847 size_t count, loff_t * offset)
1849 struct seq_file *seq = file->private_data;
1850 struct pktgen_thread *t = seq->private;
1851 int i, max, len, ret;
1856 // sprintf(pg_result, "Wrong command format");
1861 len = count_trail_chars(user_buffer, max);
1867 /* Read variable name */
1869 len = strn_len(&user_buffer[i], sizeof(name) - 1);
1873 memset(name, 0, sizeof(name));
1874 if (copy_from_user(name, &user_buffer[i], len))
1879 len = count_trail_chars(&user_buffer[i], max);
1886 pr_debug("t=%s, count=%lu\n", name, (unsigned long)count);
1889 pr_err("ERROR: No thread\n");
1894 pg_result = &(t->result[0]);
1896 if (!strcmp(name, "add_device")) {
1899 len = strn_len(&user_buffer[i], sizeof(f) - 1);
1904 if (copy_from_user(f, &user_buffer[i], len))
1907 mutex_lock(&pktgen_thread_lock);
1908 ret = pktgen_add_device(t, f);
1909 mutex_unlock(&pktgen_thread_lock);
1912 sprintf(pg_result, "OK: add_device=%s", f);
1914 sprintf(pg_result, "ERROR: can not add device %s", f);
1918 if (!strcmp(name, "rem_device_all")) {
1919 mutex_lock(&pktgen_thread_lock);
1920 t->control |= T_REMDEVALL;
1921 mutex_unlock(&pktgen_thread_lock);
1922 schedule_timeout_interruptible(msecs_to_jiffies(125)); /* Propagate thread->control */
1924 sprintf(pg_result, "OK: rem_device_all");
1928 if (!strcmp(name, "max_before_softirq")) {
1929 sprintf(pg_result, "OK: Note! max_before_softirq is obsoleted -- Do not use");
1939 static int pktgen_thread_open(struct inode *inode, struct file *file)
1941 return single_open(file, pktgen_thread_show, PDE_DATA(inode));
1944 static const struct file_operations pktgen_thread_fops = {
1945 .owner = THIS_MODULE,
1946 .open = pktgen_thread_open,
1948 .llseek = seq_lseek,
1949 .write = pktgen_thread_write,
1950 .release = single_release,
1953 /* Think find or remove for NN */
1954 static struct pktgen_dev *__pktgen_NN_threads(const struct pktgen_net *pn,
1955 const char *ifname, int remove)
1957 struct pktgen_thread *t;
1958 struct pktgen_dev *pkt_dev = NULL;
1959 bool exact = (remove == FIND);
1961 list_for_each_entry(t, &pn->pktgen_threads, th_list) {
1962 pkt_dev = pktgen_find_dev(t, ifname, exact);
1965 pkt_dev->removal_mark = 1;
1966 t->control |= T_REMDEV;
1975 * mark a device for removal
1977 static void pktgen_mark_device(const struct pktgen_net *pn, const char *ifname)
1979 struct pktgen_dev *pkt_dev = NULL;
1980 const int max_tries = 10, msec_per_try = 125;
1983 mutex_lock(&pktgen_thread_lock);
1984 pr_debug("%s: marking %s for removal\n", __func__, ifname);
1988 pkt_dev = __pktgen_NN_threads(pn, ifname, REMOVE);
1989 if (pkt_dev == NULL)
1990 break; /* success */
1992 mutex_unlock(&pktgen_thread_lock);
1993 pr_debug("%s: waiting for %s to disappear....\n",
1995 schedule_timeout_interruptible(msecs_to_jiffies(msec_per_try));
1996 mutex_lock(&pktgen_thread_lock);
1998 if (++i >= max_tries) {
1999 pr_err("%s: timed out after waiting %d msec for device %s to be removed\n",
2000 __func__, msec_per_try * i, ifname);
2006 mutex_unlock(&pktgen_thread_lock);
2009 static void pktgen_change_name(const struct pktgen_net *pn, struct net_device *dev)
2011 struct pktgen_thread *t;
2013 list_for_each_entry(t, &pn->pktgen_threads, th_list) {
2014 struct pktgen_dev *pkt_dev;
2017 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) {
2018 if (pkt_dev->odev != dev)
2021 proc_remove(pkt_dev->entry);
2023 pkt_dev->entry = proc_create_data(dev->name, 0600,
2027 if (!pkt_dev->entry)
2028 pr_err("can't move proc entry for '%s'\n",
2036 static int pktgen_device_event(struct notifier_block *unused,
2037 unsigned long event, void *ptr)
2039 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2040 struct pktgen_net *pn = net_generic(dev_net(dev), pg_net_id);
2042 if (pn->pktgen_exiting)
2045 /* It is OK that we do not hold the group lock right now,
2046 * as we run under the RTNL lock.
2050 case NETDEV_CHANGENAME:
2051 pktgen_change_name(pn, dev);
2054 case NETDEV_UNREGISTER:
2055 pktgen_mark_device(pn, dev->name);
2062 static struct net_device *pktgen_dev_get_by_name(const struct pktgen_net *pn,
2063 struct pktgen_dev *pkt_dev,
2069 for (i = 0; ifname[i] != '@'; i++) {
2077 return dev_get_by_name(pn->net, b);
2081 /* Associate pktgen_dev with a device. */
2083 static int pktgen_setup_dev(const struct pktgen_net *pn,
2084 struct pktgen_dev *pkt_dev, const char *ifname)
2086 struct net_device *odev;
2089 /* Clean old setups */
2090 if (pkt_dev->odev) {
2091 dev_put(pkt_dev->odev);
2092 pkt_dev->odev = NULL;
2095 odev = pktgen_dev_get_by_name(pn, pkt_dev, ifname);
2097 pr_err("no such netdevice: \"%s\"\n", ifname);
2101 if (odev->type != ARPHRD_ETHER) {
2102 pr_err("not an ethernet device: \"%s\"\n", ifname);
2104 } else if (!netif_running(odev)) {
2105 pr_err("device is down: \"%s\"\n", ifname);
2108 pkt_dev->odev = odev;
2116 /* Read pkt_dev from the interface and set up internal pktgen_dev
2117 * structure to have the right information to create/send packets
2119 static void pktgen_setup_inject(struct pktgen_dev *pkt_dev)
2123 if (!pkt_dev->odev) {
2124 pr_err("ERROR: pkt_dev->odev == NULL in setup_inject\n");
2125 sprintf(pkt_dev->result,
2126 "ERROR: pkt_dev->odev == NULL in setup_inject.\n");
2130 /* make sure that we don't pick a non-existing transmit queue */
2131 ntxq = pkt_dev->odev->real_num_tx_queues;
2133 if (ntxq <= pkt_dev->queue_map_min) {
2134 pr_warn("WARNING: Requested queue_map_min (zero-based) (%d) exceeds valid range [0 - %d] for (%d) queues on %s, resetting\n",
2135 pkt_dev->queue_map_min, (ntxq ?: 1) - 1, ntxq,
2137 pkt_dev->queue_map_min = (ntxq ?: 1) - 1;
2139 if (pkt_dev->queue_map_max >= ntxq) {
2140 pr_warn("WARNING: Requested queue_map_max (zero-based) (%d) exceeds valid range [0 - %d] for (%d) queues on %s, resetting\n",
2141 pkt_dev->queue_map_max, (ntxq ?: 1) - 1, ntxq,
2143 pkt_dev->queue_map_max = (ntxq ?: 1) - 1;
2146 /* Default to the interface's mac if not explicitly set. */
2148 if (is_zero_ether_addr(pkt_dev->src_mac))
2149 ether_addr_copy(&(pkt_dev->hh[6]), pkt_dev->odev->dev_addr);
2151 /* Set up Dest MAC */
2152 ether_addr_copy(&(pkt_dev->hh[0]), pkt_dev->dst_mac);
2154 if (pkt_dev->flags & F_IPV6) {
2155 int i, set = 0, err = 1;
2156 struct inet6_dev *idev;
2158 if (pkt_dev->min_pkt_size == 0) {
2159 pkt_dev->min_pkt_size = 14 + sizeof(struct ipv6hdr)
2160 + sizeof(struct udphdr)
2161 + sizeof(struct pktgen_hdr)
2162 + pkt_dev->pkt_overhead;
2165 for (i = 0; i < IN6_ADDR_HSIZE; i++)
2166 if (pkt_dev->cur_in6_saddr.s6_addr[i]) {
2174 * Use linklevel address if unconfigured.
2176 * use ipv6_get_lladdr if/when it's get exported
2180 idev = __in6_dev_get(pkt_dev->odev);
2182 struct inet6_ifaddr *ifp;
2184 read_lock_bh(&idev->lock);
2185 list_for_each_entry(ifp, &idev->addr_list, if_list) {
2186 if ((ifp->scope & IFA_LINK) &&
2187 !(ifp->flags & IFA_F_TENTATIVE)) {
2188 pkt_dev->cur_in6_saddr = ifp->addr;
2193 read_unlock_bh(&idev->lock);
2197 pr_err("ERROR: IPv6 link address not available\n");
2200 if (pkt_dev->min_pkt_size == 0) {
2201 pkt_dev->min_pkt_size = 14 + sizeof(struct iphdr)
2202 + sizeof(struct udphdr)
2203 + sizeof(struct pktgen_hdr)
2204 + pkt_dev->pkt_overhead;
2207 pkt_dev->saddr_min = 0;
2208 pkt_dev->saddr_max = 0;
2209 if (strlen(pkt_dev->src_min) == 0) {
2211 struct in_device *in_dev;
2214 in_dev = __in_dev_get_rcu(pkt_dev->odev);
2216 if (in_dev->ifa_list) {
2217 pkt_dev->saddr_min =
2218 in_dev->ifa_list->ifa_address;
2219 pkt_dev->saddr_max = pkt_dev->saddr_min;
2224 pkt_dev->saddr_min = in_aton(pkt_dev->src_min);
2225 pkt_dev->saddr_max = in_aton(pkt_dev->src_max);
2228 pkt_dev->daddr_min = in_aton(pkt_dev->dst_min);
2229 pkt_dev->daddr_max = in_aton(pkt_dev->dst_max);
2231 /* Initialize current values. */
2232 pkt_dev->cur_pkt_size = pkt_dev->min_pkt_size;
2233 if (pkt_dev->min_pkt_size > pkt_dev->max_pkt_size)
2234 pkt_dev->max_pkt_size = pkt_dev->min_pkt_size;
2236 pkt_dev->cur_dst_mac_offset = 0;
2237 pkt_dev->cur_src_mac_offset = 0;
2238 pkt_dev->cur_saddr = pkt_dev->saddr_min;
2239 pkt_dev->cur_daddr = pkt_dev->daddr_min;
2240 pkt_dev->cur_udp_dst = pkt_dev->udp_dst_min;
2241 pkt_dev->cur_udp_src = pkt_dev->udp_src_min;
2242 pkt_dev->nflows = 0;
2246 static void spin(struct pktgen_dev *pkt_dev, ktime_t spin_until)
2248 ktime_t start_time, end_time;
2250 struct hrtimer_sleeper t;
2252 hrtimer_init_on_stack(&t.timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
2253 hrtimer_set_expires(&t.timer, spin_until);
2255 remaining = ktime_to_ns(hrtimer_expires_remaining(&t.timer));
2259 start_time = ktime_get();
2260 if (remaining < 100000) {
2261 /* for small delays (<100us), just loop until limit is reached */
2263 end_time = ktime_get();
2264 } while (ktime_compare(end_time, spin_until) < 0);
2266 /* see do_nanosleep */
2267 hrtimer_init_sleeper(&t, current);
2269 set_current_state(TASK_INTERRUPTIBLE);
2270 hrtimer_start_expires(&t.timer, HRTIMER_MODE_ABS);
2275 hrtimer_cancel(&t.timer);
2276 } while (t.task && pkt_dev->running && !signal_pending(current));
2277 __set_current_state(TASK_RUNNING);
2278 end_time = ktime_get();
2281 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(end_time, start_time));
2283 pkt_dev->next_tx = ktime_add_ns(spin_until, pkt_dev->delay);
2284 destroy_hrtimer_on_stack(&t.timer);
2287 static inline void set_pkt_overhead(struct pktgen_dev *pkt_dev)
2289 pkt_dev->pkt_overhead = LL_RESERVED_SPACE(pkt_dev->odev);
2290 pkt_dev->pkt_overhead += pkt_dev->nr_labels*sizeof(u32);
2291 pkt_dev->pkt_overhead += VLAN_TAG_SIZE(pkt_dev);
2292 pkt_dev->pkt_overhead += SVLAN_TAG_SIZE(pkt_dev);
2295 static inline int f_seen(const struct pktgen_dev *pkt_dev, int flow)
2297 return !!(pkt_dev->flows[flow].flags & F_INIT);
2300 static inline int f_pick(struct pktgen_dev *pkt_dev)
2302 int flow = pkt_dev->curfl;
2304 if (pkt_dev->flags & F_FLOW_SEQ) {
2305 if (pkt_dev->flows[flow].count >= pkt_dev->lflow) {
2307 pkt_dev->flows[flow].count = 0;
2308 pkt_dev->flows[flow].flags = 0;
2309 pkt_dev->curfl += 1;
2310 if (pkt_dev->curfl >= pkt_dev->cflows)
2311 pkt_dev->curfl = 0; /*reset */
2314 flow = prandom_u32() % pkt_dev->cflows;
2315 pkt_dev->curfl = flow;
2317 if (pkt_dev->flows[flow].count > pkt_dev->lflow) {
2318 pkt_dev->flows[flow].count = 0;
2319 pkt_dev->flows[flow].flags = 0;
2323 return pkt_dev->curfl;
2328 /* If there was already an IPSEC SA, we keep it as is, else
2329 * we go look for it ...
2331 #define DUMMY_MARK 0
2332 static void get_ipsec_sa(struct pktgen_dev *pkt_dev, int flow)
2334 struct xfrm_state *x = pkt_dev->flows[flow].x;
2335 struct pktgen_net *pn = net_generic(dev_net(pkt_dev->odev), pg_net_id);
2339 /* We need as quick as possible to find the right SA
2340 * Searching with minimum criteria to archieve this.
2342 x = xfrm_state_lookup_byspi(pn->net, htonl(pkt_dev->spi), AF_INET);
2344 /* slow path: we dont already have xfrm_state */
2345 x = xfrm_stateonly_find(pn->net, DUMMY_MARK,
2346 (xfrm_address_t *)&pkt_dev->cur_daddr,
2347 (xfrm_address_t *)&pkt_dev->cur_saddr,
2350 pkt_dev->ipsproto, 0);
2353 pkt_dev->flows[flow].x = x;
2354 set_pkt_overhead(pkt_dev);
2355 pkt_dev->pkt_overhead += x->props.header_len;
2361 static void set_cur_queue_map(struct pktgen_dev *pkt_dev)
2364 if (pkt_dev->flags & F_QUEUE_MAP_CPU)
2365 pkt_dev->cur_queue_map = smp_processor_id();
2367 else if (pkt_dev->queue_map_min <= pkt_dev->queue_map_max) {
2369 if (pkt_dev->flags & F_QUEUE_MAP_RND) {
2371 (pkt_dev->queue_map_max -
2372 pkt_dev->queue_map_min + 1)
2373 + pkt_dev->queue_map_min;
2375 t = pkt_dev->cur_queue_map + 1;
2376 if (t > pkt_dev->queue_map_max)
2377 t = pkt_dev->queue_map_min;
2379 pkt_dev->cur_queue_map = t;
2381 pkt_dev->cur_queue_map = pkt_dev->cur_queue_map % pkt_dev->odev->real_num_tx_queues;
2384 /* Increment/randomize headers according to flags and current values
2385 * for IP src/dest, UDP src/dst port, MAC-Addr src/dst
2387 static void mod_cur_headers(struct pktgen_dev *pkt_dev)
2393 if (pkt_dev->cflows)
2394 flow = f_pick(pkt_dev);
2396 /* Deal with source MAC */
2397 if (pkt_dev->src_mac_count > 1) {
2401 if (pkt_dev->flags & F_MACSRC_RND)
2402 mc = prandom_u32() % pkt_dev->src_mac_count;
2404 mc = pkt_dev->cur_src_mac_offset++;
2405 if (pkt_dev->cur_src_mac_offset >=
2406 pkt_dev->src_mac_count)
2407 pkt_dev->cur_src_mac_offset = 0;
2410 tmp = pkt_dev->src_mac[5] + (mc & 0xFF);
2411 pkt_dev->hh[11] = tmp;
2412 tmp = (pkt_dev->src_mac[4] + ((mc >> 8) & 0xFF) + (tmp >> 8));
2413 pkt_dev->hh[10] = tmp;
2414 tmp = (pkt_dev->src_mac[3] + ((mc >> 16) & 0xFF) + (tmp >> 8));
2415 pkt_dev->hh[9] = tmp;
2416 tmp = (pkt_dev->src_mac[2] + ((mc >> 24) & 0xFF) + (tmp >> 8));
2417 pkt_dev->hh[8] = tmp;
2418 tmp = (pkt_dev->src_mac[1] + (tmp >> 8));
2419 pkt_dev->hh[7] = tmp;
2422 /* Deal with Destination MAC */
2423 if (pkt_dev->dst_mac_count > 1) {
2427 if (pkt_dev->flags & F_MACDST_RND)
2428 mc = prandom_u32() % pkt_dev->dst_mac_count;
2431 mc = pkt_dev->cur_dst_mac_offset++;
2432 if (pkt_dev->cur_dst_mac_offset >=
2433 pkt_dev->dst_mac_count) {
2434 pkt_dev->cur_dst_mac_offset = 0;
2438 tmp = pkt_dev->dst_mac[5] + (mc & 0xFF);
2439 pkt_dev->hh[5] = tmp;
2440 tmp = (pkt_dev->dst_mac[4] + ((mc >> 8) & 0xFF) + (tmp >> 8));
2441 pkt_dev->hh[4] = tmp;
2442 tmp = (pkt_dev->dst_mac[3] + ((mc >> 16) & 0xFF) + (tmp >> 8));
2443 pkt_dev->hh[3] = tmp;
2444 tmp = (pkt_dev->dst_mac[2] + ((mc >> 24) & 0xFF) + (tmp >> 8));
2445 pkt_dev->hh[2] = tmp;
2446 tmp = (pkt_dev->dst_mac[1] + (tmp >> 8));
2447 pkt_dev->hh[1] = tmp;
2450 if (pkt_dev->flags & F_MPLS_RND) {
2452 for (i = 0; i < pkt_dev->nr_labels; i++)
2453 if (pkt_dev->labels[i] & MPLS_STACK_BOTTOM)
2454 pkt_dev->labels[i] = MPLS_STACK_BOTTOM |
2455 ((__force __be32)prandom_u32() &
2459 if ((pkt_dev->flags & F_VID_RND) && (pkt_dev->vlan_id != 0xffff)) {
2460 pkt_dev->vlan_id = prandom_u32() & (4096 - 1);
2463 if ((pkt_dev->flags & F_SVID_RND) && (pkt_dev->svlan_id != 0xffff)) {
2464 pkt_dev->svlan_id = prandom_u32() & (4096 - 1);
2467 if (pkt_dev->udp_src_min < pkt_dev->udp_src_max) {
2468 if (pkt_dev->flags & F_UDPSRC_RND)
2469 pkt_dev->cur_udp_src = prandom_u32() %
2470 (pkt_dev->udp_src_max - pkt_dev->udp_src_min)
2471 + pkt_dev->udp_src_min;
2474 pkt_dev->cur_udp_src++;
2475 if (pkt_dev->cur_udp_src >= pkt_dev->udp_src_max)
2476 pkt_dev->cur_udp_src = pkt_dev->udp_src_min;
2480 if (pkt_dev->udp_dst_min < pkt_dev->udp_dst_max) {
2481 if (pkt_dev->flags & F_UDPDST_RND) {
2482 pkt_dev->cur_udp_dst = prandom_u32() %
2483 (pkt_dev->udp_dst_max - pkt_dev->udp_dst_min)
2484 + pkt_dev->udp_dst_min;
2486 pkt_dev->cur_udp_dst++;
2487 if (pkt_dev->cur_udp_dst >= pkt_dev->udp_dst_max)
2488 pkt_dev->cur_udp_dst = pkt_dev->udp_dst_min;
2492 if (!(pkt_dev->flags & F_IPV6)) {
2494 imn = ntohl(pkt_dev->saddr_min);
2495 imx = ntohl(pkt_dev->saddr_max);
2498 if (pkt_dev->flags & F_IPSRC_RND)
2499 t = prandom_u32() % (imx - imn) + imn;
2501 t = ntohl(pkt_dev->cur_saddr);
2507 pkt_dev->cur_saddr = htonl(t);
2510 if (pkt_dev->cflows && f_seen(pkt_dev, flow)) {
2511 pkt_dev->cur_daddr = pkt_dev->flows[flow].cur_daddr;
2513 imn = ntohl(pkt_dev->daddr_min);
2514 imx = ntohl(pkt_dev->daddr_max);
2518 if (pkt_dev->flags & F_IPDST_RND) {
2524 } while (ipv4_is_loopback(s) ||
2525 ipv4_is_multicast(s) ||
2526 ipv4_is_lbcast(s) ||
2527 ipv4_is_zeronet(s) ||
2528 ipv4_is_local_multicast(s));
2529 pkt_dev->cur_daddr = s;
2531 t = ntohl(pkt_dev->cur_daddr);
2536 pkt_dev->cur_daddr = htonl(t);
2539 if (pkt_dev->cflows) {
2540 pkt_dev->flows[flow].flags |= F_INIT;
2541 pkt_dev->flows[flow].cur_daddr =
2544 if (pkt_dev->flags & F_IPSEC_ON)
2545 get_ipsec_sa(pkt_dev, flow);
2550 } else { /* IPV6 * */
2552 if (!ipv6_addr_any(&pkt_dev->min_in6_daddr)) {
2555 /* Only random destinations yet */
2557 for (i = 0; i < 4; i++) {
2558 pkt_dev->cur_in6_daddr.s6_addr32[i] =
2559 (((__force __be32)prandom_u32() |
2560 pkt_dev->min_in6_daddr.s6_addr32[i]) &
2561 pkt_dev->max_in6_daddr.s6_addr32[i]);
2566 if (pkt_dev->min_pkt_size < pkt_dev->max_pkt_size) {
2568 if (pkt_dev->flags & F_TXSIZE_RND) {
2570 (pkt_dev->max_pkt_size - pkt_dev->min_pkt_size)
2571 + pkt_dev->min_pkt_size;
2573 t = pkt_dev->cur_pkt_size + 1;
2574 if (t > pkt_dev->max_pkt_size)
2575 t = pkt_dev->min_pkt_size;
2577 pkt_dev->cur_pkt_size = t;
2580 set_cur_queue_map(pkt_dev);
2582 pkt_dev->flows[flow].count++;
2587 static u32 pktgen_dst_metrics[RTAX_MAX + 1] = {
2589 [RTAX_HOPLIMIT] = 0x5, /* Set a static hoplimit */
2592 static int pktgen_output_ipsec(struct sk_buff *skb, struct pktgen_dev *pkt_dev)
2594 struct xfrm_state *x = pkt_dev->flows[pkt_dev->curfl].x;
2596 struct net *net = dev_net(pkt_dev->odev);
2600 /* XXX: we dont support tunnel mode for now until
2601 * we resolve the dst issue */
2602 if ((x->props.mode != XFRM_MODE_TRANSPORT) && (pkt_dev->spi == 0))
2605 /* But when user specify an valid SPI, transformation
2606 * supports both transport/tunnel mode + ESP/AH type.
2608 if ((x->props.mode == XFRM_MODE_TUNNEL) && (pkt_dev->spi != 0))
2609 skb->_skb_refdst = (unsigned long)&pkt_dev->dst | SKB_DST_NOREF;
2612 err = x->outer_mode->output(x, skb);
2613 rcu_read_unlock_bh();
2615 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEMODEERROR);
2618 err = x->type->output(x, skb);
2620 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEPROTOERROR);
2623 spin_lock_bh(&x->lock);
2624 x->curlft.bytes += skb->len;
2625 x->curlft.packets++;
2626 spin_unlock_bh(&x->lock);
2631 static void free_SAs(struct pktgen_dev *pkt_dev)
2633 if (pkt_dev->cflows) {
2634 /* let go of the SAs if we have them */
2636 for (i = 0; i < pkt_dev->cflows; i++) {
2637 struct xfrm_state *x = pkt_dev->flows[i].x;
2640 pkt_dev->flows[i].x = NULL;
2646 static int process_ipsec(struct pktgen_dev *pkt_dev,
2647 struct sk_buff *skb, __be16 protocol)
2649 if (pkt_dev->flags & F_IPSEC_ON) {
2650 struct xfrm_state *x = pkt_dev->flows[pkt_dev->curfl].x;
2657 nhead = x->props.header_len - skb_headroom(skb);
2659 ret = pskb_expand_head(skb, nhead, 0, GFP_ATOMIC);
2661 pr_err("Error expanding ipsec packet %d\n",
2667 /* ipsec is not expecting ll header */
2668 skb_pull(skb, ETH_HLEN);
2669 ret = pktgen_output_ipsec(skb, pkt_dev);
2671 pr_err("Error creating ipsec packet %d\n", ret);
2675 eth = (struct ethhdr *)skb_push(skb, ETH_HLEN);
2676 memcpy(eth, pkt_dev->hh, 2 * ETH_ALEN);
2677 eth->h_proto = protocol;
2679 /* Update IPv4 header len as well as checksum value */
2681 iph->tot_len = htons(skb->len - ETH_HLEN);
2692 static void mpls_push(__be32 *mpls, struct pktgen_dev *pkt_dev)
2695 for (i = 0; i < pkt_dev->nr_labels; i++)
2696 *mpls++ = pkt_dev->labels[i] & ~MPLS_STACK_BOTTOM;
2699 *mpls |= MPLS_STACK_BOTTOM;
2702 static inline __be16 build_tci(unsigned int id, unsigned int cfi,
2705 return htons(id | (cfi << 12) | (prio << 13));
2708 static void pktgen_finalize_skb(struct pktgen_dev *pkt_dev, struct sk_buff *skb,
2711 struct timeval timestamp;
2712 struct pktgen_hdr *pgh;
2714 pgh = (struct pktgen_hdr *)skb_put(skb, sizeof(*pgh));
2715 datalen -= sizeof(*pgh);
2717 if (pkt_dev->nfrags <= 0) {
2718 memset(skb_put(skb, datalen), 0, datalen);
2720 int frags = pkt_dev->nfrags;
2725 if (frags > MAX_SKB_FRAGS)
2726 frags = MAX_SKB_FRAGS;
2727 len = datalen - frags * PAGE_SIZE;
2729 memset(skb_put(skb, len), 0, len);
2730 datalen = frags * PAGE_SIZE;
2734 frag_len = (datalen/frags) < PAGE_SIZE ?
2735 (datalen/frags) : PAGE_SIZE;
2736 while (datalen > 0) {
2737 if (unlikely(!pkt_dev->page)) {
2738 int node = numa_node_id();
2740 if (pkt_dev->node >= 0 && (pkt_dev->flags & F_NODE))
2741 node = pkt_dev->node;
2742 pkt_dev->page = alloc_pages_node(node, GFP_KERNEL | __GFP_ZERO, 0);
2746 get_page(pkt_dev->page);
2747 skb_frag_set_page(skb, i, pkt_dev->page);
2748 skb_shinfo(skb)->frags[i].page_offset = 0;
2749 /*last fragment, fill rest of data*/
2750 if (i == (frags - 1))
2751 skb_frag_size_set(&skb_shinfo(skb)->frags[i],
2752 (datalen < PAGE_SIZE ? datalen : PAGE_SIZE));
2754 skb_frag_size_set(&skb_shinfo(skb)->frags[i], frag_len);
2755 datalen -= skb_frag_size(&skb_shinfo(skb)->frags[i]);
2756 skb->len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2757 skb->data_len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2759 skb_shinfo(skb)->nr_frags = i;
2763 /* Stamp the time, and sequence number,
2764 * convert them to network byte order
2766 pgh->pgh_magic = htonl(PKTGEN_MAGIC);
2767 pgh->seq_num = htonl(pkt_dev->seq_num);
2769 if (pkt_dev->flags & F_NO_TIMESTAMP) {
2773 do_gettimeofday(×tamp);
2774 pgh->tv_sec = htonl(timestamp.tv_sec);
2775 pgh->tv_usec = htonl(timestamp.tv_usec);
2779 static struct sk_buff *pktgen_alloc_skb(struct net_device *dev,
2780 struct pktgen_dev *pkt_dev,
2781 unsigned int extralen)
2783 struct sk_buff *skb = NULL;
2784 unsigned int size = pkt_dev->cur_pkt_size + 64 + extralen +
2785 pkt_dev->pkt_overhead;
2787 if (pkt_dev->flags & F_NODE) {
2788 int node = pkt_dev->node >= 0 ? pkt_dev->node : numa_node_id();
2790 skb = __alloc_skb(NET_SKB_PAD + size, GFP_NOWAIT, 0, node);
2792 skb_reserve(skb, NET_SKB_PAD);
2796 skb = __netdev_alloc_skb(dev, size, GFP_NOWAIT);
2800 skb_reserve(skb, LL_RESERVED_SPACE(dev));
2805 static struct sk_buff *fill_packet_ipv4(struct net_device *odev,
2806 struct pktgen_dev *pkt_dev)
2808 struct sk_buff *skb = NULL;
2810 struct udphdr *udph;
2813 __be16 protocol = htons(ETH_P_IP);
2815 __be16 *vlan_tci = NULL; /* Encapsulates priority and VLAN ID */
2816 __be16 *vlan_encapsulated_proto = NULL; /* packet type ID field (or len) for VLAN tag */
2817 __be16 *svlan_tci = NULL; /* Encapsulates priority and SVLAN ID */
2818 __be16 *svlan_encapsulated_proto = NULL; /* packet type ID field (or len) for SVLAN tag */
2821 if (pkt_dev->nr_labels)
2822 protocol = htons(ETH_P_MPLS_UC);
2824 if (pkt_dev->vlan_id != 0xffff)
2825 protocol = htons(ETH_P_8021Q);
2827 /* Update any of the values, used when we're incrementing various
2830 mod_cur_headers(pkt_dev);
2831 queue_map = pkt_dev->cur_queue_map;
2833 datalen = (odev->hard_header_len + 16) & ~0xf;
2835 skb = pktgen_alloc_skb(odev, pkt_dev, datalen);
2837 sprintf(pkt_dev->result, "No memory");
2841 prefetchw(skb->data);
2842 skb_reserve(skb, datalen);
2844 /* Reserve for ethernet and IP header */
2845 eth = (__u8 *) skb_push(skb, 14);
2846 mpls = (__be32 *)skb_put(skb, pkt_dev->nr_labels*sizeof(__u32));
2847 if (pkt_dev->nr_labels)
2848 mpls_push(mpls, pkt_dev);
2850 if (pkt_dev->vlan_id != 0xffff) {
2851 if (pkt_dev->svlan_id != 0xffff) {
2852 svlan_tci = (__be16 *)skb_put(skb, sizeof(__be16));
2853 *svlan_tci = build_tci(pkt_dev->svlan_id,
2856 svlan_encapsulated_proto = (__be16 *)skb_put(skb, sizeof(__be16));
2857 *svlan_encapsulated_proto = htons(ETH_P_8021Q);
2859 vlan_tci = (__be16 *)skb_put(skb, sizeof(__be16));
2860 *vlan_tci = build_tci(pkt_dev->vlan_id,
2863 vlan_encapsulated_proto = (__be16 *)skb_put(skb, sizeof(__be16));
2864 *vlan_encapsulated_proto = htons(ETH_P_IP);
2867 skb_reset_mac_header(skb);
2868 skb_set_network_header(skb, skb->len);
2869 iph = (struct iphdr *) skb_put(skb, sizeof(struct iphdr));
2871 skb_set_transport_header(skb, skb->len);
2872 udph = (struct udphdr *) skb_put(skb, sizeof(struct udphdr));
2873 skb_set_queue_mapping(skb, queue_map);
2874 skb->priority = pkt_dev->skb_priority;
2876 memcpy(eth, pkt_dev->hh, 12);
2877 *(__be16 *) & eth[12] = protocol;
2879 /* Eth + IPh + UDPh + mpls */
2880 datalen = pkt_dev->cur_pkt_size - 14 - 20 - 8 -
2881 pkt_dev->pkt_overhead;
2882 if (datalen < 0 || datalen < sizeof(struct pktgen_hdr))
2883 datalen = sizeof(struct pktgen_hdr);
2885 udph->source = htons(pkt_dev->cur_udp_src);
2886 udph->dest = htons(pkt_dev->cur_udp_dst);
2887 udph->len = htons(datalen + 8); /* DATA + udphdr */
2893 iph->tos = pkt_dev->tos;
2894 iph->protocol = IPPROTO_UDP; /* UDP */
2895 iph->saddr = pkt_dev->cur_saddr;
2896 iph->daddr = pkt_dev->cur_daddr;
2897 iph->id = htons(pkt_dev->ip_id);
2900 iplen = 20 + 8 + datalen;
2901 iph->tot_len = htons(iplen);
2903 skb->protocol = protocol;
2905 skb->pkt_type = PACKET_HOST;
2907 pktgen_finalize_skb(pkt_dev, skb, datalen);
2909 if (!(pkt_dev->flags & F_UDPCSUM)) {
2910 skb->ip_summed = CHECKSUM_NONE;
2911 } else if (odev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM)) {
2912 skb->ip_summed = CHECKSUM_PARTIAL;
2914 udp4_hwcsum(skb, iph->saddr, iph->daddr);
2916 __wsum csum = skb_checksum(skb, skb_transport_offset(skb), datalen + 8, 0);
2918 /* add protocol-dependent pseudo-header */
2919 udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
2920 datalen + 8, IPPROTO_UDP, csum);
2922 if (udph->check == 0)
2923 udph->check = CSUM_MANGLED_0;
2927 if (!process_ipsec(pkt_dev, skb, protocol))
2934 static struct sk_buff *fill_packet_ipv6(struct net_device *odev,
2935 struct pktgen_dev *pkt_dev)
2937 struct sk_buff *skb = NULL;
2939 struct udphdr *udph;
2940 int datalen, udplen;
2941 struct ipv6hdr *iph;
2942 __be16 protocol = htons(ETH_P_IPV6);
2944 __be16 *vlan_tci = NULL; /* Encapsulates priority and VLAN ID */
2945 __be16 *vlan_encapsulated_proto = NULL; /* packet type ID field (or len) for VLAN tag */
2946 __be16 *svlan_tci = NULL; /* Encapsulates priority and SVLAN ID */
2947 __be16 *svlan_encapsulated_proto = NULL; /* packet type ID field (or len) for SVLAN tag */
2950 if (pkt_dev->nr_labels)
2951 protocol = htons(ETH_P_MPLS_UC);
2953 if (pkt_dev->vlan_id != 0xffff)
2954 protocol = htons(ETH_P_8021Q);
2956 /* Update any of the values, used when we're incrementing various
2959 mod_cur_headers(pkt_dev);
2960 queue_map = pkt_dev->cur_queue_map;
2962 skb = pktgen_alloc_skb(odev, pkt_dev, 16);
2964 sprintf(pkt_dev->result, "No memory");
2968 prefetchw(skb->data);
2969 skb_reserve(skb, 16);
2971 /* Reserve for ethernet and IP header */
2972 eth = (__u8 *) skb_push(skb, 14);
2973 mpls = (__be32 *)skb_put(skb, pkt_dev->nr_labels*sizeof(__u32));
2974 if (pkt_dev->nr_labels)
2975 mpls_push(mpls, pkt_dev);
2977 if (pkt_dev->vlan_id != 0xffff) {
2978 if (pkt_dev->svlan_id != 0xffff) {
2979 svlan_tci = (__be16 *)skb_put(skb, sizeof(__be16));
2980 *svlan_tci = build_tci(pkt_dev->svlan_id,
2983 svlan_encapsulated_proto = (__be16 *)skb_put(skb, sizeof(__be16));
2984 *svlan_encapsulated_proto = htons(ETH_P_8021Q);
2986 vlan_tci = (__be16 *)skb_put(skb, sizeof(__be16));
2987 *vlan_tci = build_tci(pkt_dev->vlan_id,
2990 vlan_encapsulated_proto = (__be16 *)skb_put(skb, sizeof(__be16));
2991 *vlan_encapsulated_proto = htons(ETH_P_IPV6);
2994 skb_reset_mac_header(skb);
2995 skb_set_network_header(skb, skb->len);
2996 iph = (struct ipv6hdr *) skb_put(skb, sizeof(struct ipv6hdr));
2998 skb_set_transport_header(skb, skb->len);
2999 udph = (struct udphdr *) skb_put(skb, sizeof(struct udphdr));
3000 skb_set_queue_mapping(skb, queue_map);
3001 skb->priority = pkt_dev->skb_priority;
3003 memcpy(eth, pkt_dev->hh, 12);
3004 *(__be16 *) ð[12] = protocol;
3006 /* Eth + IPh + UDPh + mpls */
3007 datalen = pkt_dev->cur_pkt_size - 14 -
3008 sizeof(struct ipv6hdr) - sizeof(struct udphdr) -
3009 pkt_dev->pkt_overhead;
3011 if (datalen < 0 || datalen < sizeof(struct pktgen_hdr)) {
3012 datalen = sizeof(struct pktgen_hdr);
3013 net_info_ratelimited("increased datalen to %d\n", datalen);
3016 udplen = datalen + sizeof(struct udphdr);
3017 udph->source = htons(pkt_dev->cur_udp_src);
3018 udph->dest = htons(pkt_dev->cur_udp_dst);
3019 udph->len = htons(udplen);
3022 *(__be32 *) iph = htonl(0x60000000); /* Version + flow */
3024 if (pkt_dev->traffic_class) {
3025 /* Version + traffic class + flow (0) */
3026 *(__be32 *)iph |= htonl(0x60000000 | (pkt_dev->traffic_class << 20));
3029 iph->hop_limit = 32;
3031 iph->payload_len = htons(udplen);
3032 iph->nexthdr = IPPROTO_UDP;
3034 iph->daddr = pkt_dev->cur_in6_daddr;
3035 iph->saddr = pkt_dev->cur_in6_saddr;
3037 skb->protocol = protocol;
3039 skb->pkt_type = PACKET_HOST;
3041 pktgen_finalize_skb(pkt_dev, skb, datalen);
3043 if (!(pkt_dev->flags & F_UDPCSUM)) {
3044 skb->ip_summed = CHECKSUM_NONE;
3045 } else if (odev->features & (NETIF_F_HW_CSUM | NETIF_F_IPV6_CSUM)) {
3046 skb->ip_summed = CHECKSUM_PARTIAL;
3047 skb->csum_start = skb_transport_header(skb) - skb->head;
3048 skb->csum_offset = offsetof(struct udphdr, check);
3049 udph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, udplen, IPPROTO_UDP, 0);
3051 __wsum csum = skb_checksum(skb, skb_transport_offset(skb), udplen, 0);
3053 /* add protocol-dependent pseudo-header */
3054 udph->check = csum_ipv6_magic(&iph->saddr, &iph->daddr, udplen, IPPROTO_UDP, csum);
3056 if (udph->check == 0)
3057 udph->check = CSUM_MANGLED_0;
3063 static struct sk_buff *fill_packet(struct net_device *odev,
3064 struct pktgen_dev *pkt_dev)
3066 if (pkt_dev->flags & F_IPV6)
3067 return fill_packet_ipv6(odev, pkt_dev);
3069 return fill_packet_ipv4(odev, pkt_dev);
3072 static void pktgen_clear_counters(struct pktgen_dev *pkt_dev)
3074 pkt_dev->seq_num = 1;
3075 pkt_dev->idle_acc = 0;
3077 pkt_dev->tx_bytes = 0;
3078 pkt_dev->errors = 0;
3081 /* Set up structure for sending pkts, clear counters */
3083 static void pktgen_run(struct pktgen_thread *t)
3085 struct pktgen_dev *pkt_dev;
3091 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) {
3094 * setup odev and create initial packet.
3096 pktgen_setup_inject(pkt_dev);
3098 if (pkt_dev->odev) {
3099 pktgen_clear_counters(pkt_dev);
3100 pkt_dev->skb = NULL;
3101 pkt_dev->started_at = pkt_dev->next_tx = ktime_get();
3103 set_pkt_overhead(pkt_dev);
3105 strcpy(pkt_dev->result, "Starting");
3106 pkt_dev->running = 1; /* Cranke yeself! */
3109 strcpy(pkt_dev->result, "Error starting");
3113 t->control &= ~(T_STOP);
3116 static void pktgen_stop_all_threads_ifs(struct pktgen_net *pn)
3118 struct pktgen_thread *t;
3122 mutex_lock(&pktgen_thread_lock);
3124 list_for_each_entry(t, &pn->pktgen_threads, th_list)
3125 t->control |= T_STOP;
3127 mutex_unlock(&pktgen_thread_lock);
3130 static int thread_is_running(const struct pktgen_thread *t)
3132 const struct pktgen_dev *pkt_dev;
3135 list_for_each_entry_rcu(pkt_dev, &t->if_list, list)
3136 if (pkt_dev->running) {
3144 static int pktgen_wait_thread_run(struct pktgen_thread *t)
3146 while (thread_is_running(t)) {
3148 msleep_interruptible(100);
3150 if (signal_pending(current))
3158 static int pktgen_wait_all_threads_run(struct pktgen_net *pn)
3160 struct pktgen_thread *t;
3163 mutex_lock(&pktgen_thread_lock);
3165 list_for_each_entry(t, &pn->pktgen_threads, th_list) {
3166 sig = pktgen_wait_thread_run(t);
3172 list_for_each_entry(t, &pn->pktgen_threads, th_list)
3173 t->control |= (T_STOP);
3175 mutex_unlock(&pktgen_thread_lock);
3179 static void pktgen_run_all_threads(struct pktgen_net *pn)
3181 struct pktgen_thread *t;
3185 mutex_lock(&pktgen_thread_lock);
3187 list_for_each_entry(t, &pn->pktgen_threads, th_list)
3188 t->control |= (T_RUN);
3190 mutex_unlock(&pktgen_thread_lock);
3192 /* Propagate thread->control */
3193 schedule_timeout_interruptible(msecs_to_jiffies(125));
3195 pktgen_wait_all_threads_run(pn);
3198 static void pktgen_reset_all_threads(struct pktgen_net *pn)
3200 struct pktgen_thread *t;
3204 mutex_lock(&pktgen_thread_lock);
3206 list_for_each_entry(t, &pn->pktgen_threads, th_list)
3207 t->control |= (T_REMDEVALL);
3209 mutex_unlock(&pktgen_thread_lock);
3211 /* Propagate thread->control */
3212 schedule_timeout_interruptible(msecs_to_jiffies(125));
3214 pktgen_wait_all_threads_run(pn);
3217 static void show_results(struct pktgen_dev *pkt_dev, int nr_frags)
3219 __u64 bps, mbps, pps;
3220 char *p = pkt_dev->result;
3221 ktime_t elapsed = ktime_sub(pkt_dev->stopped_at,
3222 pkt_dev->started_at);
3223 ktime_t idle = ns_to_ktime(pkt_dev->idle_acc);
3225 p += sprintf(p, "OK: %llu(c%llu+d%llu) usec, %llu (%dbyte,%dfrags)\n",
3226 (unsigned long long)ktime_to_us(elapsed),
3227 (unsigned long long)ktime_to_us(ktime_sub(elapsed, idle)),
3228 (unsigned long long)ktime_to_us(idle),
3229 (unsigned long long)pkt_dev->sofar,
3230 pkt_dev->cur_pkt_size, nr_frags);
3232 pps = div64_u64(pkt_dev->sofar * NSEC_PER_SEC,
3233 ktime_to_ns(elapsed));
3235 bps = pps * 8 * pkt_dev->cur_pkt_size;
3238 do_div(mbps, 1000000);
3239 p += sprintf(p, " %llupps %lluMb/sec (%llubps) errors: %llu",
3240 (unsigned long long)pps,
3241 (unsigned long long)mbps,
3242 (unsigned long long)bps,
3243 (unsigned long long)pkt_dev->errors);
3246 /* Set stopped-at timer, remove from running list, do counters & statistics */
3247 static int pktgen_stop_device(struct pktgen_dev *pkt_dev)
3249 int nr_frags = pkt_dev->skb ? skb_shinfo(pkt_dev->skb)->nr_frags : -1;
3251 if (!pkt_dev->running) {
3252 pr_warn("interface: %s is already stopped\n",
3257 pkt_dev->running = 0;
3258 kfree_skb(pkt_dev->skb);
3259 pkt_dev->skb = NULL;
3260 pkt_dev->stopped_at = ktime_get();
3262 show_results(pkt_dev, nr_frags);
3267 static struct pktgen_dev *next_to_run(struct pktgen_thread *t)
3269 struct pktgen_dev *pkt_dev, *best = NULL;
3272 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) {
3273 if (!pkt_dev->running)
3277 else if (ktime_compare(pkt_dev->next_tx, best->next_tx) < 0)
3285 static void pktgen_stop(struct pktgen_thread *t)
3287 struct pktgen_dev *pkt_dev;
3293 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) {
3294 pktgen_stop_device(pkt_dev);
3301 * one of our devices needs to be removed - find it
3304 static void pktgen_rem_one_if(struct pktgen_thread *t)
3306 struct list_head *q, *n;
3307 struct pktgen_dev *cur;
3311 list_for_each_safe(q, n, &t->if_list) {
3312 cur = list_entry(q, struct pktgen_dev, list);
3314 if (!cur->removal_mark)
3317 kfree_skb(cur->skb);
3320 pktgen_remove_device(t, cur);
3326 static void pktgen_rem_all_ifs(struct pktgen_thread *t)
3328 struct list_head *q, *n;
3329 struct pktgen_dev *cur;
3333 /* Remove all devices, free mem */
3335 list_for_each_safe(q, n, &t->if_list) {
3336 cur = list_entry(q, struct pktgen_dev, list);
3338 kfree_skb(cur->skb);
3341 pktgen_remove_device(t, cur);
3345 static void pktgen_rem_thread(struct pktgen_thread *t)
3347 /* Remove from the thread list */
3348 remove_proc_entry(t->tsk->comm, t->net->proc_dir);
3351 static void pktgen_resched(struct pktgen_dev *pkt_dev)
3353 ktime_t idle_start = ktime_get();
3355 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(ktime_get(), idle_start));
3358 static void pktgen_wait_for_skb(struct pktgen_dev *pkt_dev)
3360 ktime_t idle_start = ktime_get();
3362 while (atomic_read(&(pkt_dev->skb->users)) != 1) {
3363 if (signal_pending(current))
3367 pktgen_resched(pkt_dev);
3371 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(ktime_get(), idle_start));
3374 static void pktgen_xmit(struct pktgen_dev *pkt_dev)
3376 unsigned int burst = ACCESS_ONCE(pkt_dev->burst);
3377 struct net_device *odev = pkt_dev->odev;
3378 struct netdev_queue *txq;
3379 struct sk_buff *skb;
3382 /* If device is offline, then don't send */
3383 if (unlikely(!netif_running(odev) || !netif_carrier_ok(odev))) {
3384 pktgen_stop_device(pkt_dev);
3388 /* This is max DELAY, this has special meaning of
3391 if (unlikely(pkt_dev->delay == ULLONG_MAX)) {
3392 pkt_dev->next_tx = ktime_add_ns(ktime_get(), ULONG_MAX);
3396 /* If no skb or clone count exhausted then get new one */
3397 if (!pkt_dev->skb || (pkt_dev->last_ok &&
3398 ++pkt_dev->clone_count >= pkt_dev->clone_skb)) {
3399 /* build a new pkt */
3400 kfree_skb(pkt_dev->skb);
3402 pkt_dev->skb = fill_packet(odev, pkt_dev);
3403 if (pkt_dev->skb == NULL) {
3404 pr_err("ERROR: couldn't allocate skb in fill_packet\n");
3406 pkt_dev->clone_count--; /* back out increment, OOM */
3409 pkt_dev->last_pkt_size = pkt_dev->skb->len;
3410 pkt_dev->clone_count = 0; /* reset counter */
3413 if (pkt_dev->delay && pkt_dev->last_ok)
3414 spin(pkt_dev, pkt_dev->next_tx);
3416 if (pkt_dev->xmit_mode == M_NETIF_RECEIVE) {
3418 skb->protocol = eth_type_trans(skb, skb->dev);
3419 atomic_add(burst, &skb->users);
3422 ret = netif_receive_skb(skb);
3423 if (ret == NET_RX_DROP)
3427 if (atomic_read(&skb->users) != burst) {
3428 /* skb was queued by rps/rfs or taps,
3429 * so cannot reuse this skb
3431 atomic_sub(burst - 1, &skb->users);
3432 /* get out of the loop and wait
3433 * until skb is consumed
3437 /* skb was 'freed' by stack, so clean few
3440 #ifdef CONFIG_NET_CLS_ACT
3441 skb->tc_verd = 0; /* reset reclass/redir ttl */
3443 } while (--burst > 0);
3444 goto out; /* Skips xmit_mode M_START_XMIT */
3445 } else if (pkt_dev->xmit_mode == M_QUEUE_XMIT) {
3447 atomic_inc(&pkt_dev->skb->users);
3449 ret = dev_queue_xmit(pkt_dev->skb);
3451 case NET_XMIT_SUCCESS:
3454 pkt_dev->tx_bytes += pkt_dev->last_pkt_size;
3458 /* These are all valid return codes for a qdisc but
3459 * indicate packets are being dropped or will likely
3462 case NETDEV_TX_BUSY:
3463 /* qdisc may call dev_hard_start_xmit directly in cases
3464 * where no queues exist e.g. loopback device, virtual
3465 * devices, etc. In this case we need to handle
3470 net_info_ratelimited("%s xmit error: %d\n",
3471 pkt_dev->odevname, ret);
3477 txq = skb_get_tx_queue(odev, pkt_dev->skb);
3481 HARD_TX_LOCK(odev, txq, smp_processor_id());
3483 if (unlikely(netif_xmit_frozen_or_drv_stopped(txq))) {
3484 ret = NETDEV_TX_BUSY;
3485 pkt_dev->last_ok = 0;
3488 atomic_add(burst, &pkt_dev->skb->users);
3491 ret = netdev_start_xmit(pkt_dev->skb, odev, txq, --burst > 0);
3495 pkt_dev->last_ok = 1;
3498 pkt_dev->tx_bytes += pkt_dev->last_pkt_size;
3499 if (burst > 0 && !netif_xmit_frozen_or_drv_stopped(txq))
3504 /* skb has been consumed */
3507 default: /* Drivers are not supposed to return other values! */
3508 net_info_ratelimited("%s xmit error: %d\n",
3509 pkt_dev->odevname, ret);
3512 case NETDEV_TX_BUSY:
3513 /* Retry it next time */
3514 atomic_dec(&(pkt_dev->skb->users));
3515 pkt_dev->last_ok = 0;
3517 if (unlikely(burst))
3518 atomic_sub(burst, &pkt_dev->skb->users);
3520 HARD_TX_UNLOCK(odev, txq);
3525 /* If pkt_dev->count is zero, then run forever */
3526 if ((pkt_dev->count != 0) && (pkt_dev->sofar >= pkt_dev->count)) {
3527 pktgen_wait_for_skb(pkt_dev);
3529 /* Done with this */
3530 pktgen_stop_device(pkt_dev);
3535 * Main loop of the thread goes here
3538 static int pktgen_thread_worker(void *arg)
3541 struct pktgen_thread *t = arg;
3542 struct pktgen_dev *pkt_dev = NULL;
3545 BUG_ON(smp_processor_id() != cpu);
3547 init_waitqueue_head(&t->queue);
3548 complete(&t->start_done);
3550 pr_debug("starting pktgen/%d: pid=%d\n", cpu, task_pid_nr(current));
3554 while (!kthread_should_stop()) {
3555 pkt_dev = next_to_run(t);
3557 if (unlikely(!pkt_dev && t->control == 0)) {
3558 if (t->net->pktgen_exiting)
3560 wait_event_interruptible_timeout(t->queue,
3567 if (likely(pkt_dev)) {
3568 pktgen_xmit(pkt_dev);
3571 pktgen_resched(pkt_dev);
3576 if (t->control & T_STOP) {
3578 t->control &= ~(T_STOP);
3581 if (t->control & T_RUN) {
3583 t->control &= ~(T_RUN);
3586 if (t->control & T_REMDEVALL) {
3587 pktgen_rem_all_ifs(t);
3588 t->control &= ~(T_REMDEVALL);
3591 if (t->control & T_REMDEV) {
3592 pktgen_rem_one_if(t);
3593 t->control &= ~(T_REMDEV);
3599 pr_debug("%s stopping all device\n", t->tsk->comm);
3602 pr_debug("%s removing all device\n", t->tsk->comm);
3603 pktgen_rem_all_ifs(t);
3605 pr_debug("%s removing thread\n", t->tsk->comm);
3606 pktgen_rem_thread(t);
3611 static struct pktgen_dev *pktgen_find_dev(struct pktgen_thread *t,
3612 const char *ifname, bool exact)
3614 struct pktgen_dev *p, *pkt_dev = NULL;
3615 size_t len = strlen(ifname);
3618 list_for_each_entry_rcu(p, &t->if_list, list)
3619 if (strncmp(p->odevname, ifname, len) == 0) {
3620 if (p->odevname[len]) {
3621 if (exact || p->odevname[len] != '@')
3629 pr_debug("find_dev(%s) returning %p\n", ifname, pkt_dev);
3634 * Adds a dev at front of if_list.
3637 static int add_dev_to_thread(struct pktgen_thread *t,
3638 struct pktgen_dev *pkt_dev)
3642 /* This function cannot be called concurrently, as its called
3643 * under pktgen_thread_lock mutex, but it can run from
3644 * userspace on another CPU than the kthread. The if_lock()
3645 * is used here to sync with concurrent instances of
3646 * _rem_dev_from_if_list() invoked via kthread, which is also
3647 * updating the if_list */
3650 if (pkt_dev->pg_thread) {
3651 pr_err("ERROR: already assigned to a thread\n");
3656 pkt_dev->running = 0;
3657 pkt_dev->pg_thread = t;
3658 list_add_rcu(&pkt_dev->list, &t->if_list);
3665 /* Called under thread lock */
3667 static int pktgen_add_device(struct pktgen_thread *t, const char *ifname)
3669 struct pktgen_dev *pkt_dev;
3671 int node = cpu_to_node(t->cpu);
3673 /* We don't allow a device to be on several threads */
3675 pkt_dev = __pktgen_NN_threads(t->net, ifname, FIND);
3677 pr_err("ERROR: interface already used\n");
3681 pkt_dev = kzalloc_node(sizeof(struct pktgen_dev), GFP_KERNEL, node);
3685 strcpy(pkt_dev->odevname, ifname);
3686 pkt_dev->flows = vzalloc_node(MAX_CFLOWS * sizeof(struct flow_state),
3688 if (pkt_dev->flows == NULL) {
3693 pkt_dev->removal_mark = 0;
3694 pkt_dev->nfrags = 0;
3695 pkt_dev->delay = pg_delay_d;
3696 pkt_dev->count = pg_count_d;
3698 pkt_dev->udp_src_min = 9; /* sink port */
3699 pkt_dev->udp_src_max = 9;
3700 pkt_dev->udp_dst_min = 9;
3701 pkt_dev->udp_dst_max = 9;
3702 pkt_dev->vlan_p = 0;
3703 pkt_dev->vlan_cfi = 0;
3704 pkt_dev->vlan_id = 0xffff;
3705 pkt_dev->svlan_p = 0;
3706 pkt_dev->svlan_cfi = 0;
3707 pkt_dev->svlan_id = 0xffff;
3711 err = pktgen_setup_dev(t->net, pkt_dev, ifname);
3714 if (pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING)
3715 pkt_dev->clone_skb = pg_clone_skb_d;
3717 pkt_dev->entry = proc_create_data(ifname, 0600, t->net->proc_dir,
3718 &pktgen_if_fops, pkt_dev);
3719 if (!pkt_dev->entry) {
3720 pr_err("cannot create %s/%s procfs entry\n",
3721 PG_PROC_DIR, ifname);
3726 pkt_dev->ipsmode = XFRM_MODE_TRANSPORT;
3727 pkt_dev->ipsproto = IPPROTO_ESP;
3729 /* xfrm tunnel mode needs additional dst to extract outter
3730 * ip header protocol/ttl/id field, here creat a phony one.
3731 * instead of looking for a valid rt, which definitely hurting
3732 * performance under such circumstance.
3734 pkt_dev->dstops.family = AF_INET;
3735 pkt_dev->dst.dev = pkt_dev->odev;
3736 dst_init_metrics(&pkt_dev->dst, pktgen_dst_metrics, false);
3737 pkt_dev->dst.child = &pkt_dev->dst;
3738 pkt_dev->dst.ops = &pkt_dev->dstops;
3741 return add_dev_to_thread(t, pkt_dev);
3743 dev_put(pkt_dev->odev);
3748 vfree(pkt_dev->flows);
3753 static int __net_init pktgen_create_thread(int cpu, struct pktgen_net *pn)
3755 struct pktgen_thread *t;
3756 struct proc_dir_entry *pe;
3757 struct task_struct *p;
3759 t = kzalloc_node(sizeof(struct pktgen_thread), GFP_KERNEL,
3762 pr_err("ERROR: out of memory, can't create new thread\n");
3766 spin_lock_init(&t->if_lock);
3769 INIT_LIST_HEAD(&t->if_list);
3771 list_add_tail(&t->th_list, &pn->pktgen_threads);
3772 init_completion(&t->start_done);
3774 p = kthread_create_on_node(pktgen_thread_worker,
3777 "kpktgend_%d", cpu);
3779 pr_err("kernel_thread() failed for cpu %d\n", t->cpu);
3780 list_del(&t->th_list);
3784 kthread_bind(p, cpu);
3787 pe = proc_create_data(t->tsk->comm, 0600, pn->proc_dir,
3788 &pktgen_thread_fops, t);
3790 pr_err("cannot create %s/%s procfs entry\n",
3791 PG_PROC_DIR, t->tsk->comm);
3793 list_del(&t->th_list);
3801 wait_for_completion(&t->start_done);
3807 * Removes a device from the thread if_list.
3809 static void _rem_dev_from_if_list(struct pktgen_thread *t,
3810 struct pktgen_dev *pkt_dev)
3812 struct list_head *q, *n;
3813 struct pktgen_dev *p;
3816 list_for_each_safe(q, n, &t->if_list) {
3817 p = list_entry(q, struct pktgen_dev, list);
3819 list_del_rcu(&p->list);
3824 static int pktgen_remove_device(struct pktgen_thread *t,
3825 struct pktgen_dev *pkt_dev)
3827 pr_debug("remove_device pkt_dev=%p\n", pkt_dev);
3829 if (pkt_dev->running) {
3830 pr_warn("WARNING: trying to remove a running interface, stopping it now\n");
3831 pktgen_stop_device(pkt_dev);
3834 /* Dis-associate from the interface */
3836 if (pkt_dev->odev) {
3837 dev_put(pkt_dev->odev);
3838 pkt_dev->odev = NULL;
3841 /* Remove proc before if_list entry, because add_device uses
3842 * list to determine if interface already exist, avoid race
3843 * with proc_create_data() */
3844 proc_remove(pkt_dev->entry);
3846 /* And update the thread if_list */
3847 _rem_dev_from_if_list(t, pkt_dev);
3852 vfree(pkt_dev->flows);
3854 put_page(pkt_dev->page);
3855 kfree_rcu(pkt_dev, rcu);
3859 static int __net_init pg_net_init(struct net *net)
3861 struct pktgen_net *pn = net_generic(net, pg_net_id);
3862 struct proc_dir_entry *pe;
3866 INIT_LIST_HEAD(&pn->pktgen_threads);
3867 pn->pktgen_exiting = false;
3868 pn->proc_dir = proc_mkdir(PG_PROC_DIR, pn->net->proc_net);
3869 if (!pn->proc_dir) {
3870 pr_warn("cannot create /proc/net/%s\n", PG_PROC_DIR);
3873 pe = proc_create(PGCTRL, 0600, pn->proc_dir, &pktgen_fops);
3875 pr_err("cannot create %s procfs entry\n", PGCTRL);
3880 for_each_online_cpu(cpu) {
3883 err = pktgen_create_thread(cpu, pn);
3885 pr_warn("Cannot create thread for cpu %d (%d)\n",
3889 if (list_empty(&pn->pktgen_threads)) {
3890 pr_err("Initialization failed for all threads\n");
3898 remove_proc_entry(PGCTRL, pn->proc_dir);
3900 remove_proc_entry(PG_PROC_DIR, pn->net->proc_net);
3904 static void __net_exit pg_net_exit(struct net *net)
3906 struct pktgen_net *pn = net_generic(net, pg_net_id);
3907 struct pktgen_thread *t;
3908 struct list_head *q, *n;
3911 /* Stop all interfaces & threads */
3912 pn->pktgen_exiting = true;
3914 mutex_lock(&pktgen_thread_lock);
3915 list_splice_init(&pn->pktgen_threads, &list);
3916 mutex_unlock(&pktgen_thread_lock);
3918 list_for_each_safe(q, n, &list) {
3919 t = list_entry(q, struct pktgen_thread, th_list);
3920 list_del(&t->th_list);
3921 kthread_stop(t->tsk);
3922 put_task_struct(t->tsk);
3926 remove_proc_entry(PGCTRL, pn->proc_dir);
3927 remove_proc_entry(PG_PROC_DIR, pn->net->proc_net);
3930 static struct pernet_operations pg_net_ops = {
3931 .init = pg_net_init,
3932 .exit = pg_net_exit,
3934 .size = sizeof(struct pktgen_net),
3937 static int __init pg_init(void)
3941 pr_info("%s", version);
3942 ret = register_pernet_subsys(&pg_net_ops);
3945 ret = register_netdevice_notifier(&pktgen_notifier_block);
3947 unregister_pernet_subsys(&pg_net_ops);
3952 static void __exit pg_cleanup(void)
3954 unregister_netdevice_notifier(&pktgen_notifier_block);
3955 unregister_pernet_subsys(&pg_net_ops);
3956 /* Don't need rcu_barrier() due to use of kfree_rcu() */
3959 module_init(pg_init);
3960 module_exit(pg_cleanup);
3963 MODULE_DESCRIPTION("Packet Generator tool");
3964 MODULE_LICENSE("GPL");
3965 MODULE_VERSION(VERSION);
3966 module_param(pg_count_d, int, 0);
3967 MODULE_PARM_DESC(pg_count_d, "Default number of packets to inject");
3968 module_param(pg_delay_d, int, 0);
3969 MODULE_PARM_DESC(pg_delay_d, "Default delay between packets (nanoseconds)");
3970 module_param(pg_clone_skb_d, int, 0);
3971 MODULE_PARM_DESC(pg_clone_skb_d, "Default number of copies of the same packet");
3972 module_param(debug, int, 0);
3973 MODULE_PARM_DESC(debug, "Enable debugging of pktgen module");