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 warning
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 "2.75"
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__);
188 pf(IPV6) /* Interface in IPV6 Mode */ \
189 pf(IPSRC_RND) /* IP-Src Random */ \
190 pf(IPDST_RND) /* IP-Dst Random */ \
191 pf(TXSIZE_RND) /* Transmit size is random */ \
192 pf(UDPSRC_RND) /* UDP-Src Random */ \
193 pf(UDPDST_RND) /* UDP-Dst Random */ \
194 pf(UDPCSUM) /* Include UDP checksum */ \
195 pf(NO_TIMESTAMP) /* Don't timestamp packets (default TS) */ \
196 pf(MPLS_RND) /* Random MPLS labels */ \
197 pf(QUEUE_MAP_RND) /* queue map Random */ \
198 pf(QUEUE_MAP_CPU) /* queue map mirrors smp_processor_id() */ \
199 pf(FLOW_SEQ) /* Sequential flows */ \
200 pf(IPSEC) /* ipsec on for flows */ \
201 pf(MACSRC_RND) /* MAC-Src Random */ \
202 pf(MACDST_RND) /* MAC-Dst Random */ \
203 pf(VID_RND) /* Random VLAN ID */ \
204 pf(SVID_RND) /* Random SVLAN ID */ \
205 pf(NODE) /* Node memory alloc*/ \
207 #define pf(flag) flag##_SHIFT,
213 /* Device flag bits */
214 #define pf(flag) static const __u32 F_##flag = (1<<flag##_SHIFT);
218 #define pf(flag) __stringify(flag),
219 static char *pkt_flag_names[] = {
224 #define NR_PKT_FLAGS ARRAY_SIZE(pkt_flag_names)
226 /* Thread control flag bits */
227 #define T_STOP (1<<0) /* Stop run */
228 #define T_RUN (1<<1) /* Start run */
229 #define T_REMDEVALL (1<<2) /* Remove all devs */
230 #define T_REMDEV (1<<3) /* Remove one dev */
233 #define M_START_XMIT 0 /* Default normal TX */
234 #define M_NETIF_RECEIVE 1 /* Inject packets into stack */
235 #define M_QUEUE_XMIT 2 /* Inject packet into qdisc */
237 /* If lock -- protects updating of if_list */
238 #define if_lock(t) mutex_lock(&(t->if_lock));
239 #define if_unlock(t) mutex_unlock(&(t->if_lock));
241 /* Used to help with determining the pkts on receive */
242 #define PKTGEN_MAGIC 0xbe9be955
243 #define PG_PROC_DIR "pktgen"
244 #define PGCTRL "pgctrl"
246 #define MAX_CFLOWS 65536
248 #define VLAN_TAG_SIZE(x) ((x)->vlan_id == 0xffff ? 0 : 4)
249 #define SVLAN_TAG_SIZE(x) ((x)->svlan_id == 0xffff ? 0 : 4)
255 struct xfrm_state *x;
261 #define F_INIT (1<<0) /* flow has been initialized */
265 * Try to keep frequent/infrequent used vars. separated.
267 struct proc_dir_entry *entry; /* proc file */
268 struct pktgen_thread *pg_thread;/* the owner */
269 struct list_head list; /* chaining in the thread's run-queue */
270 struct rcu_head rcu; /* freed by RCU */
272 int running; /* if false, the test will stop */
274 /* If min != max, then we will either do a linear iteration, or
275 * we will do a random selection from within the range.
281 int pkt_overhead; /* overhead for MPLS, VLANs, IPSEC etc */
283 int removal_mark; /* non-zero => the device is marked for
284 * removal by worker thread */
287 u64 delay; /* nano-seconds */
289 __u64 count; /* Default No packets to send */
290 __u64 sofar; /* How many pkts we've sent so far */
291 __u64 tx_bytes; /* How many bytes we've transmitted */
292 __u64 errors; /* Errors when trying to transmit, */
294 /* runtime counters relating to clone_skb */
297 int last_ok; /* Was last skb sent?
298 * Or a failed transmit of some sort?
299 * This will keep sequence numbers in order
304 u64 idle_acc; /* nano-seconds */
309 * Use multiple SKBs during packet gen.
310 * If this number is greater than 1, then
311 * that many copies of the same packet will be
312 * sent before a new packet is allocated.
313 * If you want to send 1024 identical packets
314 * before creating a new packet,
315 * set clone_skb to 1024.
318 char dst_min[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
319 char dst_max[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
320 char src_min[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
321 char src_max[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
323 struct in6_addr in6_saddr;
324 struct in6_addr in6_daddr;
325 struct in6_addr cur_in6_daddr;
326 struct in6_addr cur_in6_saddr;
328 struct in6_addr min_in6_daddr;
329 struct in6_addr max_in6_daddr;
330 struct in6_addr min_in6_saddr;
331 struct in6_addr max_in6_saddr;
333 /* If we're doing ranges, random or incremental, then this
334 * defines the min/max for those ranges.
336 __be32 saddr_min; /* inclusive, source IP address */
337 __be32 saddr_max; /* exclusive, source IP address */
338 __be32 daddr_min; /* inclusive, dest IP address */
339 __be32 daddr_max; /* exclusive, dest IP address */
341 __u16 udp_src_min; /* inclusive, source UDP port */
342 __u16 udp_src_max; /* exclusive, source UDP port */
343 __u16 udp_dst_min; /* inclusive, dest UDP port */
344 __u16 udp_dst_max; /* exclusive, dest UDP port */
347 __u8 tos; /* six MSB of (former) IPv4 TOS
348 are for dscp codepoint */
349 __u8 traffic_class; /* ditto for the (former) Traffic Class in IPv6
350 (see RFC 3260, sec. 4) */
353 unsigned int nr_labels; /* Depth of stack, 0 = no MPLS */
354 __be32 labels[MAX_MPLS_LABELS];
356 /* VLAN/SVLAN (802.1Q/Q-in-Q) */
359 __u16 vlan_id; /* 0xffff means no vlan tag */
363 __u16 svlan_id; /* 0xffff means no svlan tag */
365 __u32 src_mac_count; /* How many MACs to iterate through */
366 __u32 dst_mac_count; /* How many MACs to iterate through */
368 unsigned char dst_mac[ETH_ALEN];
369 unsigned char src_mac[ETH_ALEN];
371 __u32 cur_dst_mac_offset;
372 __u32 cur_src_mac_offset;
384 0x00, 0x80, 0xC8, 0x79, 0xB3, 0xCB,
386 We fill in SRC address later
387 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
391 __u16 pad; /* pad out the hh struct to an even 16 bytes */
393 struct sk_buff *skb; /* skb we are to transmit next, used for when we
394 * are transmitting the same one multiple times
396 struct net_device *odev; /* The out-going device.
397 * Note that the device should have it's
398 * pg_info pointer pointing back to this
400 * Set when the user specifies the out-going
401 * device name (not when the inject is
402 * started as it used to do.)
405 struct flow_state *flows;
406 unsigned int cflows; /* Concurrent flows (config) */
407 unsigned int lflow; /* Flow length (config) */
408 unsigned int nflows; /* accumulated flows (stats) */
409 unsigned int curfl; /* current sequenced flow (state)*/
413 __u32 skb_priority; /* skb priority field */
414 unsigned int burst; /* number of duplicated packets to burst */
415 int node; /* Memory node */
418 __u8 ipsmode; /* IPSEC mode (config) */
419 __u8 ipsproto; /* IPSEC type (config) */
421 struct xfrm_dst xdst;
422 struct dst_ops dstops;
435 static unsigned int pg_net_id __read_mostly;
439 struct proc_dir_entry *proc_dir;
440 struct list_head pktgen_threads;
444 struct pktgen_thread {
445 struct mutex if_lock; /* for list of devices */
446 struct list_head if_list; /* All device here */
447 struct list_head th_list;
448 struct task_struct *tsk;
451 /* Field for thread to receive "posted" events terminate,
457 wait_queue_head_t queue;
458 struct completion start_done;
459 struct pktgen_net *net;
465 static const char version[] =
466 "Packet Generator for packet performance testing. "
467 "Version: " VERSION "\n";
469 static int pktgen_remove_device(struct pktgen_thread *t, struct pktgen_dev *i);
470 static int pktgen_add_device(struct pktgen_thread *t, const char *ifname);
471 static struct pktgen_dev *pktgen_find_dev(struct pktgen_thread *t,
472 const char *ifname, bool exact);
473 static int pktgen_device_event(struct notifier_block *, unsigned long, void *);
474 static void pktgen_run_all_threads(struct pktgen_net *pn);
475 static void pktgen_reset_all_threads(struct pktgen_net *pn);
476 static void pktgen_stop_all_threads_ifs(struct pktgen_net *pn);
478 static void pktgen_stop(struct pktgen_thread *t);
479 static void pktgen_clear_counters(struct pktgen_dev *pkt_dev);
481 /* Module parameters, defaults. */
482 static int pg_count_d __read_mostly = 1000;
483 static int pg_delay_d __read_mostly;
484 static int pg_clone_skb_d __read_mostly;
485 static int debug __read_mostly;
487 static DEFINE_MUTEX(pktgen_thread_lock);
489 static struct notifier_block pktgen_notifier_block = {
490 .notifier_call = pktgen_device_event,
494 * /proc handling functions
498 static int pgctrl_show(struct seq_file *seq, void *v)
500 seq_puts(seq, version);
504 static ssize_t pgctrl_write(struct file *file, const char __user *buf,
505 size_t count, loff_t *ppos)
508 struct pktgen_net *pn = net_generic(current->nsproxy->net_ns, pg_net_id);
510 if (!capable(CAP_NET_ADMIN))
516 if (count > sizeof(data))
517 count = sizeof(data);
519 if (copy_from_user(data, buf, count))
522 data[count - 1] = 0; /* Strip trailing '\n' and terminate string */
524 if (!strcmp(data, "stop"))
525 pktgen_stop_all_threads_ifs(pn);
527 else if (!strcmp(data, "start"))
528 pktgen_run_all_threads(pn);
530 else if (!strcmp(data, "reset"))
531 pktgen_reset_all_threads(pn);
539 static int pgctrl_open(struct inode *inode, struct file *file)
541 return single_open(file, pgctrl_show, PDE_DATA(inode));
544 static const struct file_operations pktgen_fops = {
548 .write = pgctrl_write,
549 .release = single_release,
552 static int pktgen_if_show(struct seq_file *seq, void *v)
554 const struct pktgen_dev *pkt_dev = seq->private;
560 "Params: count %llu min_pkt_size: %u max_pkt_size: %u\n",
561 (unsigned long long)pkt_dev->count, pkt_dev->min_pkt_size,
562 pkt_dev->max_pkt_size);
565 " frags: %d delay: %llu clone_skb: %d ifname: %s\n",
566 pkt_dev->nfrags, (unsigned long long) pkt_dev->delay,
567 pkt_dev->clone_skb, pkt_dev->odevname);
569 seq_printf(seq, " flows: %u flowlen: %u\n", pkt_dev->cflows,
573 " queue_map_min: %u queue_map_max: %u\n",
574 pkt_dev->queue_map_min,
575 pkt_dev->queue_map_max);
577 if (pkt_dev->skb_priority)
578 seq_printf(seq, " skb_priority: %u\n",
579 pkt_dev->skb_priority);
581 if (pkt_dev->flags & F_IPV6) {
583 " saddr: %pI6c min_saddr: %pI6c max_saddr: %pI6c\n"
584 " daddr: %pI6c min_daddr: %pI6c max_daddr: %pI6c\n",
586 &pkt_dev->min_in6_saddr, &pkt_dev->max_in6_saddr,
588 &pkt_dev->min_in6_daddr, &pkt_dev->max_in6_daddr);
591 " dst_min: %s dst_max: %s\n",
592 pkt_dev->dst_min, pkt_dev->dst_max);
594 " src_min: %s src_max: %s\n",
595 pkt_dev->src_min, pkt_dev->src_max);
598 seq_puts(seq, " src_mac: ");
600 seq_printf(seq, "%pM ",
601 is_zero_ether_addr(pkt_dev->src_mac) ?
602 pkt_dev->odev->dev_addr : pkt_dev->src_mac);
604 seq_puts(seq, "dst_mac: ");
605 seq_printf(seq, "%pM\n", pkt_dev->dst_mac);
608 " udp_src_min: %d udp_src_max: %d"
609 " udp_dst_min: %d udp_dst_max: %d\n",
610 pkt_dev->udp_src_min, pkt_dev->udp_src_max,
611 pkt_dev->udp_dst_min, pkt_dev->udp_dst_max);
614 " src_mac_count: %d dst_mac_count: %d\n",
615 pkt_dev->src_mac_count, pkt_dev->dst_mac_count);
617 if (pkt_dev->nr_labels) {
618 seq_puts(seq, " mpls: ");
619 for (i = 0; i < pkt_dev->nr_labels; i++)
620 seq_printf(seq, "%08x%s", ntohl(pkt_dev->labels[i]),
621 i == pkt_dev->nr_labels-1 ? "\n" : ", ");
624 if (pkt_dev->vlan_id != 0xffff)
625 seq_printf(seq, " vlan_id: %u vlan_p: %u vlan_cfi: %u\n",
626 pkt_dev->vlan_id, pkt_dev->vlan_p,
629 if (pkt_dev->svlan_id != 0xffff)
630 seq_printf(seq, " svlan_id: %u vlan_p: %u vlan_cfi: %u\n",
631 pkt_dev->svlan_id, pkt_dev->svlan_p,
635 seq_printf(seq, " tos: 0x%02x\n", pkt_dev->tos);
637 if (pkt_dev->traffic_class)
638 seq_printf(seq, " traffic_class: 0x%02x\n", pkt_dev->traffic_class);
640 if (pkt_dev->burst > 1)
641 seq_printf(seq, " burst: %d\n", pkt_dev->burst);
643 if (pkt_dev->node >= 0)
644 seq_printf(seq, " node: %d\n", pkt_dev->node);
646 if (pkt_dev->xmit_mode == M_NETIF_RECEIVE)
647 seq_puts(seq, " xmit_mode: netif_receive\n");
648 else if (pkt_dev->xmit_mode == M_QUEUE_XMIT)
649 seq_puts(seq, " xmit_mode: xmit_queue\n");
651 seq_puts(seq, " Flags: ");
653 for (i = 0; i < NR_PKT_FLAGS; i++) {
655 if (!pkt_dev->cflows)
658 if (pkt_dev->flags & (1 << i))
659 seq_printf(seq, "%s ", pkt_flag_names[i]);
660 else if (i == F_FLOW_SEQ)
661 seq_puts(seq, "FLOW_RND ");
664 if (i == F_IPSEC && pkt_dev->spi)
665 seq_printf(seq, "spi:%u", pkt_dev->spi);
671 /* not really stopped, more like last-running-at */
672 stopped = pkt_dev->running ? ktime_get() : pkt_dev->stopped_at;
673 idle = pkt_dev->idle_acc;
674 do_div(idle, NSEC_PER_USEC);
677 "Current:\n pkts-sofar: %llu errors: %llu\n",
678 (unsigned long long)pkt_dev->sofar,
679 (unsigned long long)pkt_dev->errors);
682 " started: %lluus stopped: %lluus idle: %lluus\n",
683 (unsigned long long) ktime_to_us(pkt_dev->started_at),
684 (unsigned long long) ktime_to_us(stopped),
685 (unsigned long long) idle);
688 " seq_num: %d cur_dst_mac_offset: %d cur_src_mac_offset: %d\n",
689 pkt_dev->seq_num, pkt_dev->cur_dst_mac_offset,
690 pkt_dev->cur_src_mac_offset);
692 if (pkt_dev->flags & F_IPV6) {
693 seq_printf(seq, " cur_saddr: %pI6c cur_daddr: %pI6c\n",
694 &pkt_dev->cur_in6_saddr,
695 &pkt_dev->cur_in6_daddr);
697 seq_printf(seq, " cur_saddr: %pI4 cur_daddr: %pI4\n",
698 &pkt_dev->cur_saddr, &pkt_dev->cur_daddr);
700 seq_printf(seq, " cur_udp_dst: %d cur_udp_src: %d\n",
701 pkt_dev->cur_udp_dst, pkt_dev->cur_udp_src);
703 seq_printf(seq, " cur_queue_map: %u\n", pkt_dev->cur_queue_map);
705 seq_printf(seq, " flows: %u\n", pkt_dev->nflows);
707 if (pkt_dev->result[0])
708 seq_printf(seq, "Result: %s\n", pkt_dev->result);
710 seq_puts(seq, "Result: Idle\n");
716 static int hex32_arg(const char __user *user_buffer, unsigned long maxlen,
722 for (; i < maxlen; i++) {
726 if (get_user(c, &user_buffer[i]))
728 value = hex_to_bin(c);
737 static int count_trail_chars(const char __user * user_buffer,
742 for (i = 0; i < maxlen; i++) {
744 if (get_user(c, &user_buffer[i]))
762 static long num_arg(const char __user *user_buffer, unsigned long maxlen,
768 for (i = 0; i < maxlen; i++) {
770 if (get_user(c, &user_buffer[i]))
772 if ((c >= '0') && (c <= '9')) {
781 static int strn_len(const char __user * user_buffer, unsigned int maxlen)
785 for (i = 0; i < maxlen; i++) {
787 if (get_user(c, &user_buffer[i]))
804 static ssize_t get_labels(const char __user *buffer, struct pktgen_dev *pkt_dev)
811 pkt_dev->nr_labels = 0;
814 len = hex32_arg(&buffer[i], 8, &tmp);
817 pkt_dev->labels[n] = htonl(tmp);
818 if (pkt_dev->labels[n] & MPLS_STACK_BOTTOM)
819 pkt_dev->flags |= F_MPLS_RND;
821 if (get_user(c, &buffer[i]))
825 if (n >= MAX_MPLS_LABELS)
829 pkt_dev->nr_labels = n;
833 static __u32 pktgen_read_flag(const char *f, bool *disable)
842 for (i = 0; i < NR_PKT_FLAGS; i++) {
843 if (!IS_ENABLED(CONFIG_XFRM) && i == IPSEC_SHIFT)
846 /* allow only disabling ipv6 flag */
847 if (!*disable && i == IPV6_SHIFT)
850 if (strcmp(f, pkt_flag_names[i]) == 0)
854 if (strcmp(f, "FLOW_RND") == 0) {
855 *disable = !*disable;
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 + 1, 1024);
910 char *tp = strndup_user(user_buffer, copy);
915 pr_debug("%s,%zu buffer -:%s:-\n", name, count, tp);
919 if (!strcmp(name, "min_pkt_size")) {
920 len = num_arg(&user_buffer[i], 10, &value);
925 if (value < 14 + 20 + 8)
927 if (value != pkt_dev->min_pkt_size) {
928 pkt_dev->min_pkt_size = value;
929 pkt_dev->cur_pkt_size = value;
931 sprintf(pg_result, "OK: min_pkt_size=%u",
932 pkt_dev->min_pkt_size);
936 if (!strcmp(name, "max_pkt_size")) {
937 len = num_arg(&user_buffer[i], 10, &value);
942 if (value < 14 + 20 + 8)
944 if (value != pkt_dev->max_pkt_size) {
945 pkt_dev->max_pkt_size = value;
946 pkt_dev->cur_pkt_size = value;
948 sprintf(pg_result, "OK: max_pkt_size=%u",
949 pkt_dev->max_pkt_size);
953 /* Shortcut for min = max */
955 if (!strcmp(name, "pkt_size")) {
956 len = num_arg(&user_buffer[i], 10, &value);
961 if (value < 14 + 20 + 8)
963 if (value != pkt_dev->min_pkt_size) {
964 pkt_dev->min_pkt_size = value;
965 pkt_dev->max_pkt_size = value;
966 pkt_dev->cur_pkt_size = value;
968 sprintf(pg_result, "OK: pkt_size=%u", pkt_dev->min_pkt_size);
972 if (!strcmp(name, "debug")) {
973 len = num_arg(&user_buffer[i], 10, &value);
979 sprintf(pg_result, "OK: debug=%u", debug);
983 if (!strcmp(name, "frags")) {
984 len = num_arg(&user_buffer[i], 10, &value);
989 pkt_dev->nfrags = value;
990 sprintf(pg_result, "OK: frags=%u", pkt_dev->nfrags);
993 if (!strcmp(name, "delay")) {
994 len = num_arg(&user_buffer[i], 10, &value);
999 if (value == 0x7FFFFFFF)
1000 pkt_dev->delay = ULLONG_MAX;
1002 pkt_dev->delay = (u64)value;
1004 sprintf(pg_result, "OK: delay=%llu",
1005 (unsigned long long) pkt_dev->delay);
1008 if (!strcmp(name, "rate")) {
1009 len = num_arg(&user_buffer[i], 10, &value);
1016 pkt_dev->delay = pkt_dev->min_pkt_size*8*NSEC_PER_USEC/value;
1018 pr_info("Delay set at: %llu ns\n", pkt_dev->delay);
1020 sprintf(pg_result, "OK: rate=%lu", value);
1023 if (!strcmp(name, "ratep")) {
1024 len = num_arg(&user_buffer[i], 10, &value);
1031 pkt_dev->delay = NSEC_PER_SEC/value;
1033 pr_info("Delay set at: %llu ns\n", pkt_dev->delay);
1035 sprintf(pg_result, "OK: rate=%lu", value);
1038 if (!strcmp(name, "udp_src_min")) {
1039 len = num_arg(&user_buffer[i], 10, &value);
1044 if (value != pkt_dev->udp_src_min) {
1045 pkt_dev->udp_src_min = value;
1046 pkt_dev->cur_udp_src = value;
1048 sprintf(pg_result, "OK: udp_src_min=%u", pkt_dev->udp_src_min);
1051 if (!strcmp(name, "udp_dst_min")) {
1052 len = num_arg(&user_buffer[i], 10, &value);
1057 if (value != pkt_dev->udp_dst_min) {
1058 pkt_dev->udp_dst_min = value;
1059 pkt_dev->cur_udp_dst = value;
1061 sprintf(pg_result, "OK: udp_dst_min=%u", pkt_dev->udp_dst_min);
1064 if (!strcmp(name, "udp_src_max")) {
1065 len = num_arg(&user_buffer[i], 10, &value);
1070 if (value != pkt_dev->udp_src_max) {
1071 pkt_dev->udp_src_max = value;
1072 pkt_dev->cur_udp_src = value;
1074 sprintf(pg_result, "OK: udp_src_max=%u", pkt_dev->udp_src_max);
1077 if (!strcmp(name, "udp_dst_max")) {
1078 len = num_arg(&user_buffer[i], 10, &value);
1083 if (value != pkt_dev->udp_dst_max) {
1084 pkt_dev->udp_dst_max = value;
1085 pkt_dev->cur_udp_dst = value;
1087 sprintf(pg_result, "OK: udp_dst_max=%u", pkt_dev->udp_dst_max);
1090 if (!strcmp(name, "clone_skb")) {
1091 len = num_arg(&user_buffer[i], 10, &value);
1095 ((pkt_dev->xmit_mode == M_NETIF_RECEIVE) ||
1096 !(pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING)))
1099 pkt_dev->clone_skb = value;
1101 sprintf(pg_result, "OK: clone_skb=%d", pkt_dev->clone_skb);
1104 if (!strcmp(name, "count")) {
1105 len = num_arg(&user_buffer[i], 10, &value);
1110 pkt_dev->count = value;
1111 sprintf(pg_result, "OK: count=%llu",
1112 (unsigned long long)pkt_dev->count);
1115 if (!strcmp(name, "src_mac_count")) {
1116 len = num_arg(&user_buffer[i], 10, &value);
1121 if (pkt_dev->src_mac_count != value) {
1122 pkt_dev->src_mac_count = value;
1123 pkt_dev->cur_src_mac_offset = 0;
1125 sprintf(pg_result, "OK: src_mac_count=%d",
1126 pkt_dev->src_mac_count);
1129 if (!strcmp(name, "dst_mac_count")) {
1130 len = num_arg(&user_buffer[i], 10, &value);
1135 if (pkt_dev->dst_mac_count != value) {
1136 pkt_dev->dst_mac_count = value;
1137 pkt_dev->cur_dst_mac_offset = 0;
1139 sprintf(pg_result, "OK: dst_mac_count=%d",
1140 pkt_dev->dst_mac_count);
1143 if (!strcmp(name, "burst")) {
1144 len = num_arg(&user_buffer[i], 10, &value);
1150 ((pkt_dev->xmit_mode == M_QUEUE_XMIT) ||
1151 ((pkt_dev->xmit_mode == M_START_XMIT) &&
1152 (!(pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING)))))
1154 pkt_dev->burst = value < 1 ? 1 : value;
1155 sprintf(pg_result, "OK: burst=%d", pkt_dev->burst);
1158 if (!strcmp(name, "node")) {
1159 len = num_arg(&user_buffer[i], 10, &value);
1165 if (node_possible(value)) {
1166 pkt_dev->node = value;
1167 sprintf(pg_result, "OK: node=%d", pkt_dev->node);
1168 if (pkt_dev->page) {
1169 put_page(pkt_dev->page);
1170 pkt_dev->page = NULL;
1174 sprintf(pg_result, "ERROR: node not possible");
1177 if (!strcmp(name, "xmit_mode")) {
1181 len = strn_len(&user_buffer[i], sizeof(f) - 1);
1185 if (copy_from_user(f, &user_buffer[i], len))
1189 if (strcmp(f, "start_xmit") == 0) {
1190 pkt_dev->xmit_mode = M_START_XMIT;
1191 } else if (strcmp(f, "netif_receive") == 0) {
1192 /* clone_skb set earlier, not supported in this mode */
1193 if (pkt_dev->clone_skb > 0)
1196 pkt_dev->xmit_mode = M_NETIF_RECEIVE;
1198 /* make sure new packet is allocated every time
1199 * pktgen_xmit() is called
1201 pkt_dev->last_ok = 1;
1203 /* override clone_skb if user passed default value
1204 * at module loading time
1206 pkt_dev->clone_skb = 0;
1207 } else if (strcmp(f, "queue_xmit") == 0) {
1208 pkt_dev->xmit_mode = M_QUEUE_XMIT;
1209 pkt_dev->last_ok = 1;
1212 "xmit_mode -:%s:- unknown\nAvailable modes: %s",
1213 f, "start_xmit, netif_receive\n");
1216 sprintf(pg_result, "OK: xmit_mode=%s", f);
1219 if (!strcmp(name, "flag")) {
1222 bool disable = false;
1225 len = strn_len(&user_buffer[i], sizeof(f) - 1);
1229 if (copy_from_user(f, &user_buffer[i], len))
1233 flag = pktgen_read_flag(f, &disable);
1237 pkt_dev->flags &= ~flag;
1239 pkt_dev->flags |= flag;
1242 "Flag -:%s:- unknown\nAvailable flags, (prepend ! to un-set flag):\n%s",
1244 "IPSRC_RND, IPDST_RND, UDPSRC_RND, UDPDST_RND, "
1245 "MACSRC_RND, MACDST_RND, TXSIZE_RND, IPV6, "
1246 "MPLS_RND, VID_RND, SVID_RND, FLOW_SEQ, "
1247 "QUEUE_MAP_RND, QUEUE_MAP_CPU, UDPCSUM, "
1255 sprintf(pg_result, "OK: flags=0x%x", pkt_dev->flags);
1258 if (!strcmp(name, "dst_min") || !strcmp(name, "dst")) {
1259 len = strn_len(&user_buffer[i], sizeof(pkt_dev->dst_min) - 1);
1263 if (copy_from_user(buf, &user_buffer[i], len))
1266 if (strcmp(buf, pkt_dev->dst_min) != 0) {
1267 memset(pkt_dev->dst_min, 0, sizeof(pkt_dev->dst_min));
1268 strcpy(pkt_dev->dst_min, buf);
1269 pkt_dev->daddr_min = in_aton(pkt_dev->dst_min);
1270 pkt_dev->cur_daddr = pkt_dev->daddr_min;
1273 pr_debug("dst_min set to: %s\n", pkt_dev->dst_min);
1275 sprintf(pg_result, "OK: dst_min=%s", pkt_dev->dst_min);
1278 if (!strcmp(name, "dst_max")) {
1279 len = strn_len(&user_buffer[i], sizeof(pkt_dev->dst_max) - 1);
1283 if (copy_from_user(buf, &user_buffer[i], len))
1286 if (strcmp(buf, pkt_dev->dst_max) != 0) {
1287 memset(pkt_dev->dst_max, 0, sizeof(pkt_dev->dst_max));
1288 strcpy(pkt_dev->dst_max, buf);
1289 pkt_dev->daddr_max = in_aton(pkt_dev->dst_max);
1290 pkt_dev->cur_daddr = pkt_dev->daddr_max;
1293 pr_debug("dst_max set to: %s\n", pkt_dev->dst_max);
1295 sprintf(pg_result, "OK: dst_max=%s", pkt_dev->dst_max);
1298 if (!strcmp(name, "dst6")) {
1299 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1303 pkt_dev->flags |= F_IPV6;
1305 if (copy_from_user(buf, &user_buffer[i], len))
1309 in6_pton(buf, -1, pkt_dev->in6_daddr.s6_addr, -1, NULL);
1310 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->in6_daddr);
1312 pkt_dev->cur_in6_daddr = pkt_dev->in6_daddr;
1315 pr_debug("dst6 set to: %s\n", buf);
1318 sprintf(pg_result, "OK: dst6=%s", buf);
1321 if (!strcmp(name, "dst6_min")) {
1322 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1326 pkt_dev->flags |= F_IPV6;
1328 if (copy_from_user(buf, &user_buffer[i], len))
1332 in6_pton(buf, -1, pkt_dev->min_in6_daddr.s6_addr, -1, NULL);
1333 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->min_in6_daddr);
1335 pkt_dev->cur_in6_daddr = pkt_dev->min_in6_daddr;
1337 pr_debug("dst6_min set to: %s\n", buf);
1340 sprintf(pg_result, "OK: dst6_min=%s", buf);
1343 if (!strcmp(name, "dst6_max")) {
1344 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1348 pkt_dev->flags |= F_IPV6;
1350 if (copy_from_user(buf, &user_buffer[i], len))
1354 in6_pton(buf, -1, pkt_dev->max_in6_daddr.s6_addr, -1, NULL);
1355 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->max_in6_daddr);
1358 pr_debug("dst6_max set to: %s\n", buf);
1361 sprintf(pg_result, "OK: dst6_max=%s", buf);
1364 if (!strcmp(name, "src6")) {
1365 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1369 pkt_dev->flags |= F_IPV6;
1371 if (copy_from_user(buf, &user_buffer[i], len))
1375 in6_pton(buf, -1, pkt_dev->in6_saddr.s6_addr, -1, NULL);
1376 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->in6_saddr);
1378 pkt_dev->cur_in6_saddr = pkt_dev->in6_saddr;
1381 pr_debug("src6 set to: %s\n", buf);
1384 sprintf(pg_result, "OK: src6=%s", buf);
1387 if (!strcmp(name, "src_min")) {
1388 len = strn_len(&user_buffer[i], sizeof(pkt_dev->src_min) - 1);
1392 if (copy_from_user(buf, &user_buffer[i], len))
1395 if (strcmp(buf, pkt_dev->src_min) != 0) {
1396 memset(pkt_dev->src_min, 0, sizeof(pkt_dev->src_min));
1397 strcpy(pkt_dev->src_min, buf);
1398 pkt_dev->saddr_min = in_aton(pkt_dev->src_min);
1399 pkt_dev->cur_saddr = pkt_dev->saddr_min;
1402 pr_debug("src_min set to: %s\n", pkt_dev->src_min);
1404 sprintf(pg_result, "OK: src_min=%s", pkt_dev->src_min);
1407 if (!strcmp(name, "src_max")) {
1408 len = strn_len(&user_buffer[i], sizeof(pkt_dev->src_max) - 1);
1412 if (copy_from_user(buf, &user_buffer[i], len))
1415 if (strcmp(buf, pkt_dev->src_max) != 0) {
1416 memset(pkt_dev->src_max, 0, sizeof(pkt_dev->src_max));
1417 strcpy(pkt_dev->src_max, buf);
1418 pkt_dev->saddr_max = in_aton(pkt_dev->src_max);
1419 pkt_dev->cur_saddr = pkt_dev->saddr_max;
1422 pr_debug("src_max set to: %s\n", pkt_dev->src_max);
1424 sprintf(pg_result, "OK: src_max=%s", pkt_dev->src_max);
1427 if (!strcmp(name, "dst_mac")) {
1428 len = strn_len(&user_buffer[i], sizeof(valstr) - 1);
1432 memset(valstr, 0, sizeof(valstr));
1433 if (copy_from_user(valstr, &user_buffer[i], len))
1436 if (!mac_pton(valstr, pkt_dev->dst_mac))
1438 /* Set up Dest MAC */
1439 ether_addr_copy(&pkt_dev->hh[0], pkt_dev->dst_mac);
1441 sprintf(pg_result, "OK: dstmac %pM", pkt_dev->dst_mac);
1444 if (!strcmp(name, "src_mac")) {
1445 len = strn_len(&user_buffer[i], sizeof(valstr) - 1);
1449 memset(valstr, 0, sizeof(valstr));
1450 if (copy_from_user(valstr, &user_buffer[i], len))
1453 if (!mac_pton(valstr, pkt_dev->src_mac))
1455 /* Set up Src MAC */
1456 ether_addr_copy(&pkt_dev->hh[6], pkt_dev->src_mac);
1458 sprintf(pg_result, "OK: srcmac %pM", pkt_dev->src_mac);
1462 if (!strcmp(name, "clear_counters")) {
1463 pktgen_clear_counters(pkt_dev);
1464 sprintf(pg_result, "OK: Clearing counters.\n");
1468 if (!strcmp(name, "flows")) {
1469 len = num_arg(&user_buffer[i], 10, &value);
1474 if (value > MAX_CFLOWS)
1477 pkt_dev->cflows = value;
1478 sprintf(pg_result, "OK: flows=%u", pkt_dev->cflows);
1482 if (!strcmp(name, "spi")) {
1483 len = num_arg(&user_buffer[i], 10, &value);
1488 pkt_dev->spi = value;
1489 sprintf(pg_result, "OK: spi=%u", pkt_dev->spi);
1493 if (!strcmp(name, "flowlen")) {
1494 len = num_arg(&user_buffer[i], 10, &value);
1499 pkt_dev->lflow = value;
1500 sprintf(pg_result, "OK: flowlen=%u", pkt_dev->lflow);
1504 if (!strcmp(name, "queue_map_min")) {
1505 len = num_arg(&user_buffer[i], 5, &value);
1510 pkt_dev->queue_map_min = value;
1511 sprintf(pg_result, "OK: queue_map_min=%u", pkt_dev->queue_map_min);
1515 if (!strcmp(name, "queue_map_max")) {
1516 len = num_arg(&user_buffer[i], 5, &value);
1521 pkt_dev->queue_map_max = value;
1522 sprintf(pg_result, "OK: queue_map_max=%u", pkt_dev->queue_map_max);
1526 if (!strcmp(name, "mpls")) {
1527 unsigned int n, cnt;
1529 len = get_labels(&user_buffer[i], pkt_dev);
1533 cnt = sprintf(pg_result, "OK: mpls=");
1534 for (n = 0; n < pkt_dev->nr_labels; n++)
1535 cnt += sprintf(pg_result + cnt,
1536 "%08x%s", ntohl(pkt_dev->labels[n]),
1537 n == pkt_dev->nr_labels-1 ? "" : ",");
1539 if (pkt_dev->nr_labels && pkt_dev->vlan_id != 0xffff) {
1540 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */
1541 pkt_dev->svlan_id = 0xffff;
1544 pr_debug("VLAN/SVLAN auto turned off\n");
1549 if (!strcmp(name, "vlan_id")) {
1550 len = num_arg(&user_buffer[i], 4, &value);
1555 if (value <= 4095) {
1556 pkt_dev->vlan_id = value; /* turn on VLAN */
1559 pr_debug("VLAN turned on\n");
1561 if (debug && pkt_dev->nr_labels)
1562 pr_debug("MPLS auto turned off\n");
1564 pkt_dev->nr_labels = 0; /* turn off MPLS */
1565 sprintf(pg_result, "OK: vlan_id=%u", pkt_dev->vlan_id);
1567 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */
1568 pkt_dev->svlan_id = 0xffff;
1571 pr_debug("VLAN/SVLAN turned off\n");
1576 if (!strcmp(name, "vlan_p")) {
1577 len = num_arg(&user_buffer[i], 1, &value);
1582 if ((value <= 7) && (pkt_dev->vlan_id != 0xffff)) {
1583 pkt_dev->vlan_p = value;
1584 sprintf(pg_result, "OK: vlan_p=%u", pkt_dev->vlan_p);
1586 sprintf(pg_result, "ERROR: vlan_p must be 0-7");
1591 if (!strcmp(name, "vlan_cfi")) {
1592 len = num_arg(&user_buffer[i], 1, &value);
1597 if ((value <= 1) && (pkt_dev->vlan_id != 0xffff)) {
1598 pkt_dev->vlan_cfi = value;
1599 sprintf(pg_result, "OK: vlan_cfi=%u", pkt_dev->vlan_cfi);
1601 sprintf(pg_result, "ERROR: vlan_cfi must be 0-1");
1606 if (!strcmp(name, "svlan_id")) {
1607 len = num_arg(&user_buffer[i], 4, &value);
1612 if ((value <= 4095) && ((pkt_dev->vlan_id != 0xffff))) {
1613 pkt_dev->svlan_id = value; /* turn on SVLAN */
1616 pr_debug("SVLAN turned on\n");
1618 if (debug && pkt_dev->nr_labels)
1619 pr_debug("MPLS auto turned off\n");
1621 pkt_dev->nr_labels = 0; /* turn off MPLS */
1622 sprintf(pg_result, "OK: svlan_id=%u", pkt_dev->svlan_id);
1624 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */
1625 pkt_dev->svlan_id = 0xffff;
1628 pr_debug("VLAN/SVLAN turned off\n");
1633 if (!strcmp(name, "svlan_p")) {
1634 len = num_arg(&user_buffer[i], 1, &value);
1639 if ((value <= 7) && (pkt_dev->svlan_id != 0xffff)) {
1640 pkt_dev->svlan_p = value;
1641 sprintf(pg_result, "OK: svlan_p=%u", pkt_dev->svlan_p);
1643 sprintf(pg_result, "ERROR: svlan_p must be 0-7");
1648 if (!strcmp(name, "svlan_cfi")) {
1649 len = num_arg(&user_buffer[i], 1, &value);
1654 if ((value <= 1) && (pkt_dev->svlan_id != 0xffff)) {
1655 pkt_dev->svlan_cfi = value;
1656 sprintf(pg_result, "OK: svlan_cfi=%u", pkt_dev->svlan_cfi);
1658 sprintf(pg_result, "ERROR: svlan_cfi must be 0-1");
1663 if (!strcmp(name, "tos")) {
1664 __u32 tmp_value = 0;
1665 len = hex32_arg(&user_buffer[i], 2, &tmp_value);
1671 pkt_dev->tos = tmp_value;
1672 sprintf(pg_result, "OK: tos=0x%02x", pkt_dev->tos);
1674 sprintf(pg_result, "ERROR: tos must be 00-ff");
1679 if (!strcmp(name, "traffic_class")) {
1680 __u32 tmp_value = 0;
1681 len = hex32_arg(&user_buffer[i], 2, &tmp_value);
1687 pkt_dev->traffic_class = tmp_value;
1688 sprintf(pg_result, "OK: traffic_class=0x%02x", pkt_dev->traffic_class);
1690 sprintf(pg_result, "ERROR: traffic_class must be 00-ff");
1695 if (!strcmp(name, "skb_priority")) {
1696 len = num_arg(&user_buffer[i], 9, &value);
1701 pkt_dev->skb_priority = value;
1702 sprintf(pg_result, "OK: skb_priority=%i",
1703 pkt_dev->skb_priority);
1707 sprintf(pkt_dev->result, "No such parameter \"%s\"", name);
1711 static int pktgen_if_open(struct inode *inode, struct file *file)
1713 return single_open(file, pktgen_if_show, PDE_DATA(inode));
1716 static const struct file_operations pktgen_if_fops = {
1717 .open = pktgen_if_open,
1719 .llseek = seq_lseek,
1720 .write = pktgen_if_write,
1721 .release = single_release,
1724 static int pktgen_thread_show(struct seq_file *seq, void *v)
1726 struct pktgen_thread *t = seq->private;
1727 const struct pktgen_dev *pkt_dev;
1731 seq_puts(seq, "Running: ");
1734 list_for_each_entry_rcu(pkt_dev, &t->if_list, list)
1735 if (pkt_dev->running)
1736 seq_printf(seq, "%s ", pkt_dev->odevname);
1738 seq_puts(seq, "\nStopped: ");
1740 list_for_each_entry_rcu(pkt_dev, &t->if_list, list)
1741 if (!pkt_dev->running)
1742 seq_printf(seq, "%s ", pkt_dev->odevname);
1745 seq_printf(seq, "\nResult: %s\n", t->result);
1747 seq_puts(seq, "\nResult: NA\n");
1754 static ssize_t pktgen_thread_write(struct file *file,
1755 const char __user * user_buffer,
1756 size_t count, loff_t * offset)
1758 struct seq_file *seq = file->private_data;
1759 struct pktgen_thread *t = seq->private;
1760 int i, max, len, ret;
1765 // sprintf(pg_result, "Wrong command format");
1770 len = count_trail_chars(user_buffer, max);
1776 /* Read variable name */
1778 len = strn_len(&user_buffer[i], sizeof(name) - 1);
1782 memset(name, 0, sizeof(name));
1783 if (copy_from_user(name, &user_buffer[i], len))
1788 len = count_trail_chars(&user_buffer[i], max);
1795 pr_debug("t=%s, count=%lu\n", name, (unsigned long)count);
1798 pr_err("ERROR: No thread\n");
1803 pg_result = &(t->result[0]);
1805 if (!strcmp(name, "add_device")) {
1808 len = strn_len(&user_buffer[i], sizeof(f) - 1);
1813 if (copy_from_user(f, &user_buffer[i], len))
1816 mutex_lock(&pktgen_thread_lock);
1817 ret = pktgen_add_device(t, f);
1818 mutex_unlock(&pktgen_thread_lock);
1821 sprintf(pg_result, "OK: add_device=%s", f);
1823 sprintf(pg_result, "ERROR: can not add device %s", f);
1827 if (!strcmp(name, "rem_device_all")) {
1828 mutex_lock(&pktgen_thread_lock);
1829 t->control |= T_REMDEVALL;
1830 mutex_unlock(&pktgen_thread_lock);
1831 schedule_timeout_interruptible(msecs_to_jiffies(125)); /* Propagate thread->control */
1833 sprintf(pg_result, "OK: rem_device_all");
1837 if (!strcmp(name, "max_before_softirq")) {
1838 sprintf(pg_result, "OK: Note! max_before_softirq is obsoleted -- Do not use");
1848 static int pktgen_thread_open(struct inode *inode, struct file *file)
1850 return single_open(file, pktgen_thread_show, PDE_DATA(inode));
1853 static const struct file_operations pktgen_thread_fops = {
1854 .open = pktgen_thread_open,
1856 .llseek = seq_lseek,
1857 .write = pktgen_thread_write,
1858 .release = single_release,
1861 /* Think find or remove for NN */
1862 static struct pktgen_dev *__pktgen_NN_threads(const struct pktgen_net *pn,
1863 const char *ifname, int remove)
1865 struct pktgen_thread *t;
1866 struct pktgen_dev *pkt_dev = NULL;
1867 bool exact = (remove == FIND);
1869 list_for_each_entry(t, &pn->pktgen_threads, th_list) {
1870 pkt_dev = pktgen_find_dev(t, ifname, exact);
1873 pkt_dev->removal_mark = 1;
1874 t->control |= T_REMDEV;
1883 * mark a device for removal
1885 static void pktgen_mark_device(const struct pktgen_net *pn, const char *ifname)
1887 struct pktgen_dev *pkt_dev = NULL;
1888 const int max_tries = 10, msec_per_try = 125;
1891 mutex_lock(&pktgen_thread_lock);
1892 pr_debug("%s: marking %s for removal\n", __func__, ifname);
1896 pkt_dev = __pktgen_NN_threads(pn, ifname, REMOVE);
1897 if (pkt_dev == NULL)
1898 break; /* success */
1900 mutex_unlock(&pktgen_thread_lock);
1901 pr_debug("%s: waiting for %s to disappear....\n",
1903 schedule_timeout_interruptible(msecs_to_jiffies(msec_per_try));
1904 mutex_lock(&pktgen_thread_lock);
1906 if (++i >= max_tries) {
1907 pr_err("%s: timed out after waiting %d msec for device %s to be removed\n",
1908 __func__, msec_per_try * i, ifname);
1914 mutex_unlock(&pktgen_thread_lock);
1917 static void pktgen_change_name(const struct pktgen_net *pn, struct net_device *dev)
1919 struct pktgen_thread *t;
1921 mutex_lock(&pktgen_thread_lock);
1923 list_for_each_entry(t, &pn->pktgen_threads, th_list) {
1924 struct pktgen_dev *pkt_dev;
1927 list_for_each_entry(pkt_dev, &t->if_list, list) {
1928 if (pkt_dev->odev != dev)
1931 proc_remove(pkt_dev->entry);
1933 pkt_dev->entry = proc_create_data(dev->name, 0600,
1937 if (!pkt_dev->entry)
1938 pr_err("can't move proc entry for '%s'\n",
1944 mutex_unlock(&pktgen_thread_lock);
1947 static int pktgen_device_event(struct notifier_block *unused,
1948 unsigned long event, void *ptr)
1950 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1951 struct pktgen_net *pn = net_generic(dev_net(dev), pg_net_id);
1953 if (pn->pktgen_exiting)
1956 /* It is OK that we do not hold the group lock right now,
1957 * as we run under the RTNL lock.
1961 case NETDEV_CHANGENAME:
1962 pktgen_change_name(pn, dev);
1965 case NETDEV_UNREGISTER:
1966 pktgen_mark_device(pn, dev->name);
1973 static struct net_device *pktgen_dev_get_by_name(const struct pktgen_net *pn,
1974 struct pktgen_dev *pkt_dev,
1980 for (i = 0; ifname[i] != '@'; i++) {
1988 return dev_get_by_name(pn->net, b);
1992 /* Associate pktgen_dev with a device. */
1994 static int pktgen_setup_dev(const struct pktgen_net *pn,
1995 struct pktgen_dev *pkt_dev, const char *ifname)
1997 struct net_device *odev;
2000 /* Clean old setups */
2001 if (pkt_dev->odev) {
2002 dev_put(pkt_dev->odev);
2003 pkt_dev->odev = NULL;
2006 odev = pktgen_dev_get_by_name(pn, pkt_dev, ifname);
2008 pr_err("no such netdevice: \"%s\"\n", ifname);
2012 if (odev->type != ARPHRD_ETHER) {
2013 pr_err("not an ethernet device: \"%s\"\n", ifname);
2015 } else if (!netif_running(odev)) {
2016 pr_err("device is down: \"%s\"\n", ifname);
2019 pkt_dev->odev = odev;
2027 /* Read pkt_dev from the interface and set up internal pktgen_dev
2028 * structure to have the right information to create/send packets
2030 static void pktgen_setup_inject(struct pktgen_dev *pkt_dev)
2034 if (!pkt_dev->odev) {
2035 pr_err("ERROR: pkt_dev->odev == NULL in setup_inject\n");
2036 sprintf(pkt_dev->result,
2037 "ERROR: pkt_dev->odev == NULL in setup_inject.\n");
2041 /* make sure that we don't pick a non-existing transmit queue */
2042 ntxq = pkt_dev->odev->real_num_tx_queues;
2044 if (ntxq <= pkt_dev->queue_map_min) {
2045 pr_warn("WARNING: Requested queue_map_min (zero-based) (%d) exceeds valid range [0 - %d] for (%d) queues on %s, resetting\n",
2046 pkt_dev->queue_map_min, (ntxq ?: 1) - 1, ntxq,
2048 pkt_dev->queue_map_min = (ntxq ?: 1) - 1;
2050 if (pkt_dev->queue_map_max >= ntxq) {
2051 pr_warn("WARNING: Requested queue_map_max (zero-based) (%d) exceeds valid range [0 - %d] for (%d) queues on %s, resetting\n",
2052 pkt_dev->queue_map_max, (ntxq ?: 1) - 1, ntxq,
2054 pkt_dev->queue_map_max = (ntxq ?: 1) - 1;
2057 /* Default to the interface's mac if not explicitly set. */
2059 if (is_zero_ether_addr(pkt_dev->src_mac))
2060 ether_addr_copy(&(pkt_dev->hh[6]), pkt_dev->odev->dev_addr);
2062 /* Set up Dest MAC */
2063 ether_addr_copy(&(pkt_dev->hh[0]), pkt_dev->dst_mac);
2065 if (pkt_dev->flags & F_IPV6) {
2066 int i, set = 0, err = 1;
2067 struct inet6_dev *idev;
2069 if (pkt_dev->min_pkt_size == 0) {
2070 pkt_dev->min_pkt_size = 14 + sizeof(struct ipv6hdr)
2071 + sizeof(struct udphdr)
2072 + sizeof(struct pktgen_hdr)
2073 + pkt_dev->pkt_overhead;
2076 for (i = 0; i < sizeof(struct in6_addr); i++)
2077 if (pkt_dev->cur_in6_saddr.s6_addr[i]) {
2085 * Use linklevel address if unconfigured.
2087 * use ipv6_get_lladdr if/when it's get exported
2091 idev = __in6_dev_get(pkt_dev->odev);
2093 struct inet6_ifaddr *ifp;
2095 read_lock_bh(&idev->lock);
2096 list_for_each_entry(ifp, &idev->addr_list, if_list) {
2097 if ((ifp->scope & IFA_LINK) &&
2098 !(ifp->flags & IFA_F_TENTATIVE)) {
2099 pkt_dev->cur_in6_saddr = ifp->addr;
2104 read_unlock_bh(&idev->lock);
2108 pr_err("ERROR: IPv6 link address not available\n");
2111 if (pkt_dev->min_pkt_size == 0) {
2112 pkt_dev->min_pkt_size = 14 + sizeof(struct iphdr)
2113 + sizeof(struct udphdr)
2114 + sizeof(struct pktgen_hdr)
2115 + pkt_dev->pkt_overhead;
2118 pkt_dev->saddr_min = 0;
2119 pkt_dev->saddr_max = 0;
2120 if (strlen(pkt_dev->src_min) == 0) {
2122 struct in_device *in_dev;
2125 in_dev = __in_dev_get_rcu(pkt_dev->odev);
2127 if (in_dev->ifa_list) {
2128 pkt_dev->saddr_min =
2129 in_dev->ifa_list->ifa_address;
2130 pkt_dev->saddr_max = pkt_dev->saddr_min;
2135 pkt_dev->saddr_min = in_aton(pkt_dev->src_min);
2136 pkt_dev->saddr_max = in_aton(pkt_dev->src_max);
2139 pkt_dev->daddr_min = in_aton(pkt_dev->dst_min);
2140 pkt_dev->daddr_max = in_aton(pkt_dev->dst_max);
2142 /* Initialize current values. */
2143 pkt_dev->cur_pkt_size = pkt_dev->min_pkt_size;
2144 if (pkt_dev->min_pkt_size > pkt_dev->max_pkt_size)
2145 pkt_dev->max_pkt_size = pkt_dev->min_pkt_size;
2147 pkt_dev->cur_dst_mac_offset = 0;
2148 pkt_dev->cur_src_mac_offset = 0;
2149 pkt_dev->cur_saddr = pkt_dev->saddr_min;
2150 pkt_dev->cur_daddr = pkt_dev->daddr_min;
2151 pkt_dev->cur_udp_dst = pkt_dev->udp_dst_min;
2152 pkt_dev->cur_udp_src = pkt_dev->udp_src_min;
2153 pkt_dev->nflows = 0;
2157 static void spin(struct pktgen_dev *pkt_dev, ktime_t spin_until)
2159 ktime_t start_time, end_time;
2161 struct hrtimer_sleeper t;
2163 hrtimer_init_on_stack(&t.timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
2164 hrtimer_set_expires(&t.timer, spin_until);
2166 remaining = ktime_to_ns(hrtimer_expires_remaining(&t.timer));
2170 start_time = ktime_get();
2171 if (remaining < 100000) {
2172 /* for small delays (<100us), just loop until limit is reached */
2174 end_time = ktime_get();
2175 } while (ktime_compare(end_time, spin_until) < 0);
2177 /* see do_nanosleep */
2178 hrtimer_init_sleeper(&t, current);
2180 set_current_state(TASK_INTERRUPTIBLE);
2181 hrtimer_start_expires(&t.timer, HRTIMER_MODE_ABS);
2186 hrtimer_cancel(&t.timer);
2187 } while (t.task && pkt_dev->running && !signal_pending(current));
2188 __set_current_state(TASK_RUNNING);
2189 end_time = ktime_get();
2192 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(end_time, start_time));
2194 pkt_dev->next_tx = ktime_add_ns(spin_until, pkt_dev->delay);
2195 destroy_hrtimer_on_stack(&t.timer);
2198 static inline void set_pkt_overhead(struct pktgen_dev *pkt_dev)
2200 pkt_dev->pkt_overhead = 0;
2201 pkt_dev->pkt_overhead += pkt_dev->nr_labels*sizeof(u32);
2202 pkt_dev->pkt_overhead += VLAN_TAG_SIZE(pkt_dev);
2203 pkt_dev->pkt_overhead += SVLAN_TAG_SIZE(pkt_dev);
2206 static inline int f_seen(const struct pktgen_dev *pkt_dev, int flow)
2208 return !!(pkt_dev->flows[flow].flags & F_INIT);
2211 static inline int f_pick(struct pktgen_dev *pkt_dev)
2213 int flow = pkt_dev->curfl;
2215 if (pkt_dev->flags & F_FLOW_SEQ) {
2216 if (pkt_dev->flows[flow].count >= pkt_dev->lflow) {
2218 pkt_dev->flows[flow].count = 0;
2219 pkt_dev->flows[flow].flags = 0;
2220 pkt_dev->curfl += 1;
2221 if (pkt_dev->curfl >= pkt_dev->cflows)
2222 pkt_dev->curfl = 0; /*reset */
2225 flow = prandom_u32() % pkt_dev->cflows;
2226 pkt_dev->curfl = flow;
2228 if (pkt_dev->flows[flow].count > pkt_dev->lflow) {
2229 pkt_dev->flows[flow].count = 0;
2230 pkt_dev->flows[flow].flags = 0;
2234 return pkt_dev->curfl;
2239 /* If there was already an IPSEC SA, we keep it as is, else
2240 * we go look for it ...
2242 #define DUMMY_MARK 0
2243 static void get_ipsec_sa(struct pktgen_dev *pkt_dev, int flow)
2245 struct xfrm_state *x = pkt_dev->flows[flow].x;
2246 struct pktgen_net *pn = net_generic(dev_net(pkt_dev->odev), pg_net_id);
2250 /* We need as quick as possible to find the right SA
2251 * Searching with minimum criteria to archieve this.
2253 x = xfrm_state_lookup_byspi(pn->net, htonl(pkt_dev->spi), AF_INET);
2255 /* slow path: we dont already have xfrm_state */
2256 x = xfrm_stateonly_find(pn->net, DUMMY_MARK, 0,
2257 (xfrm_address_t *)&pkt_dev->cur_daddr,
2258 (xfrm_address_t *)&pkt_dev->cur_saddr,
2261 pkt_dev->ipsproto, 0);
2264 pkt_dev->flows[flow].x = x;
2265 set_pkt_overhead(pkt_dev);
2266 pkt_dev->pkt_overhead += x->props.header_len;
2272 static void set_cur_queue_map(struct pktgen_dev *pkt_dev)
2275 if (pkt_dev->flags & F_QUEUE_MAP_CPU)
2276 pkt_dev->cur_queue_map = smp_processor_id();
2278 else if (pkt_dev->queue_map_min <= pkt_dev->queue_map_max) {
2280 if (pkt_dev->flags & F_QUEUE_MAP_RND) {
2282 (pkt_dev->queue_map_max -
2283 pkt_dev->queue_map_min + 1)
2284 + pkt_dev->queue_map_min;
2286 t = pkt_dev->cur_queue_map + 1;
2287 if (t > pkt_dev->queue_map_max)
2288 t = pkt_dev->queue_map_min;
2290 pkt_dev->cur_queue_map = t;
2292 pkt_dev->cur_queue_map = pkt_dev->cur_queue_map % pkt_dev->odev->real_num_tx_queues;
2295 /* Increment/randomize headers according to flags and current values
2296 * for IP src/dest, UDP src/dst port, MAC-Addr src/dst
2298 static void mod_cur_headers(struct pktgen_dev *pkt_dev)
2304 if (pkt_dev->cflows)
2305 flow = f_pick(pkt_dev);
2307 /* Deal with source MAC */
2308 if (pkt_dev->src_mac_count > 1) {
2312 if (pkt_dev->flags & F_MACSRC_RND)
2313 mc = prandom_u32() % pkt_dev->src_mac_count;
2315 mc = pkt_dev->cur_src_mac_offset++;
2316 if (pkt_dev->cur_src_mac_offset >=
2317 pkt_dev->src_mac_count)
2318 pkt_dev->cur_src_mac_offset = 0;
2321 tmp = pkt_dev->src_mac[5] + (mc & 0xFF);
2322 pkt_dev->hh[11] = tmp;
2323 tmp = (pkt_dev->src_mac[4] + ((mc >> 8) & 0xFF) + (tmp >> 8));
2324 pkt_dev->hh[10] = tmp;
2325 tmp = (pkt_dev->src_mac[3] + ((mc >> 16) & 0xFF) + (tmp >> 8));
2326 pkt_dev->hh[9] = tmp;
2327 tmp = (pkt_dev->src_mac[2] + ((mc >> 24) & 0xFF) + (tmp >> 8));
2328 pkt_dev->hh[8] = tmp;
2329 tmp = (pkt_dev->src_mac[1] + (tmp >> 8));
2330 pkt_dev->hh[7] = tmp;
2333 /* Deal with Destination MAC */
2334 if (pkt_dev->dst_mac_count > 1) {
2338 if (pkt_dev->flags & F_MACDST_RND)
2339 mc = prandom_u32() % pkt_dev->dst_mac_count;
2342 mc = pkt_dev->cur_dst_mac_offset++;
2343 if (pkt_dev->cur_dst_mac_offset >=
2344 pkt_dev->dst_mac_count) {
2345 pkt_dev->cur_dst_mac_offset = 0;
2349 tmp = pkt_dev->dst_mac[5] + (mc & 0xFF);
2350 pkt_dev->hh[5] = tmp;
2351 tmp = (pkt_dev->dst_mac[4] + ((mc >> 8) & 0xFF) + (tmp >> 8));
2352 pkt_dev->hh[4] = tmp;
2353 tmp = (pkt_dev->dst_mac[3] + ((mc >> 16) & 0xFF) + (tmp >> 8));
2354 pkt_dev->hh[3] = tmp;
2355 tmp = (pkt_dev->dst_mac[2] + ((mc >> 24) & 0xFF) + (tmp >> 8));
2356 pkt_dev->hh[2] = tmp;
2357 tmp = (pkt_dev->dst_mac[1] + (tmp >> 8));
2358 pkt_dev->hh[1] = tmp;
2361 if (pkt_dev->flags & F_MPLS_RND) {
2363 for (i = 0; i < pkt_dev->nr_labels; i++)
2364 if (pkt_dev->labels[i] & MPLS_STACK_BOTTOM)
2365 pkt_dev->labels[i] = MPLS_STACK_BOTTOM |
2366 ((__force __be32)prandom_u32() &
2370 if ((pkt_dev->flags & F_VID_RND) && (pkt_dev->vlan_id != 0xffff)) {
2371 pkt_dev->vlan_id = prandom_u32() & (4096 - 1);
2374 if ((pkt_dev->flags & F_SVID_RND) && (pkt_dev->svlan_id != 0xffff)) {
2375 pkt_dev->svlan_id = prandom_u32() & (4096 - 1);
2378 if (pkt_dev->udp_src_min < pkt_dev->udp_src_max) {
2379 if (pkt_dev->flags & F_UDPSRC_RND)
2380 pkt_dev->cur_udp_src = prandom_u32() %
2381 (pkt_dev->udp_src_max - pkt_dev->udp_src_min)
2382 + pkt_dev->udp_src_min;
2385 pkt_dev->cur_udp_src++;
2386 if (pkt_dev->cur_udp_src >= pkt_dev->udp_src_max)
2387 pkt_dev->cur_udp_src = pkt_dev->udp_src_min;
2391 if (pkt_dev->udp_dst_min < pkt_dev->udp_dst_max) {
2392 if (pkt_dev->flags & F_UDPDST_RND) {
2393 pkt_dev->cur_udp_dst = prandom_u32() %
2394 (pkt_dev->udp_dst_max - pkt_dev->udp_dst_min)
2395 + pkt_dev->udp_dst_min;
2397 pkt_dev->cur_udp_dst++;
2398 if (pkt_dev->cur_udp_dst >= pkt_dev->udp_dst_max)
2399 pkt_dev->cur_udp_dst = pkt_dev->udp_dst_min;
2403 if (!(pkt_dev->flags & F_IPV6)) {
2405 imn = ntohl(pkt_dev->saddr_min);
2406 imx = ntohl(pkt_dev->saddr_max);
2409 if (pkt_dev->flags & F_IPSRC_RND)
2410 t = prandom_u32() % (imx - imn) + imn;
2412 t = ntohl(pkt_dev->cur_saddr);
2418 pkt_dev->cur_saddr = htonl(t);
2421 if (pkt_dev->cflows && f_seen(pkt_dev, flow)) {
2422 pkt_dev->cur_daddr = pkt_dev->flows[flow].cur_daddr;
2424 imn = ntohl(pkt_dev->daddr_min);
2425 imx = ntohl(pkt_dev->daddr_max);
2429 if (pkt_dev->flags & F_IPDST_RND) {
2435 } while (ipv4_is_loopback(s) ||
2436 ipv4_is_multicast(s) ||
2437 ipv4_is_lbcast(s) ||
2438 ipv4_is_zeronet(s) ||
2439 ipv4_is_local_multicast(s));
2440 pkt_dev->cur_daddr = s;
2442 t = ntohl(pkt_dev->cur_daddr);
2447 pkt_dev->cur_daddr = htonl(t);
2450 if (pkt_dev->cflows) {
2451 pkt_dev->flows[flow].flags |= F_INIT;
2452 pkt_dev->flows[flow].cur_daddr =
2455 if (pkt_dev->flags & F_IPSEC)
2456 get_ipsec_sa(pkt_dev, flow);
2461 } else { /* IPV6 * */
2463 if (!ipv6_addr_any(&pkt_dev->min_in6_daddr)) {
2466 /* Only random destinations yet */
2468 for (i = 0; i < 4; i++) {
2469 pkt_dev->cur_in6_daddr.s6_addr32[i] =
2470 (((__force __be32)prandom_u32() |
2471 pkt_dev->min_in6_daddr.s6_addr32[i]) &
2472 pkt_dev->max_in6_daddr.s6_addr32[i]);
2477 if (pkt_dev->min_pkt_size < pkt_dev->max_pkt_size) {
2479 if (pkt_dev->flags & F_TXSIZE_RND) {
2481 (pkt_dev->max_pkt_size - pkt_dev->min_pkt_size)
2482 + pkt_dev->min_pkt_size;
2484 t = pkt_dev->cur_pkt_size + 1;
2485 if (t > pkt_dev->max_pkt_size)
2486 t = pkt_dev->min_pkt_size;
2488 pkt_dev->cur_pkt_size = t;
2491 set_cur_queue_map(pkt_dev);
2493 pkt_dev->flows[flow].count++;
2498 static u32 pktgen_dst_metrics[RTAX_MAX + 1] = {
2500 [RTAX_HOPLIMIT] = 0x5, /* Set a static hoplimit */
2503 static int pktgen_output_ipsec(struct sk_buff *skb, struct pktgen_dev *pkt_dev)
2505 struct xfrm_state *x = pkt_dev->flows[pkt_dev->curfl].x;
2507 struct net *net = dev_net(pkt_dev->odev);
2511 /* XXX: we dont support tunnel mode for now until
2512 * we resolve the dst issue */
2513 if ((x->props.mode != XFRM_MODE_TRANSPORT) && (pkt_dev->spi == 0))
2516 /* But when user specify an valid SPI, transformation
2517 * supports both transport/tunnel mode + ESP/AH type.
2519 if ((x->props.mode == XFRM_MODE_TUNNEL) && (pkt_dev->spi != 0))
2520 skb->_skb_refdst = (unsigned long)&pkt_dev->xdst.u.dst | SKB_DST_NOREF;
2523 err = x->outer_mode->output(x, skb);
2524 rcu_read_unlock_bh();
2526 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEMODEERROR);
2529 err = x->type->output(x, skb);
2531 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEPROTOERROR);
2534 spin_lock_bh(&x->lock);
2535 x->curlft.bytes += skb->len;
2536 x->curlft.packets++;
2537 spin_unlock_bh(&x->lock);
2542 static void free_SAs(struct pktgen_dev *pkt_dev)
2544 if (pkt_dev->cflows) {
2545 /* let go of the SAs if we have them */
2547 for (i = 0; i < pkt_dev->cflows; i++) {
2548 struct xfrm_state *x = pkt_dev->flows[i].x;
2551 pkt_dev->flows[i].x = NULL;
2557 static int process_ipsec(struct pktgen_dev *pkt_dev,
2558 struct sk_buff *skb, __be16 protocol)
2560 if (pkt_dev->flags & F_IPSEC) {
2561 struct xfrm_state *x = pkt_dev->flows[pkt_dev->curfl].x;
2568 nhead = x->props.header_len - skb_headroom(skb);
2570 ret = pskb_expand_head(skb, nhead, 0, GFP_ATOMIC);
2572 pr_err("Error expanding ipsec packet %d\n",
2578 /* ipsec is not expecting ll header */
2579 skb_pull(skb, ETH_HLEN);
2580 ret = pktgen_output_ipsec(skb, pkt_dev);
2582 pr_err("Error creating ipsec packet %d\n", ret);
2586 eth = skb_push(skb, ETH_HLEN);
2587 memcpy(eth, pkt_dev->hh, 2 * ETH_ALEN);
2588 eth->h_proto = protocol;
2590 /* Update IPv4 header len as well as checksum value */
2592 iph->tot_len = htons(skb->len - ETH_HLEN);
2603 static void mpls_push(__be32 *mpls, struct pktgen_dev *pkt_dev)
2606 for (i = 0; i < pkt_dev->nr_labels; i++)
2607 *mpls++ = pkt_dev->labels[i] & ~MPLS_STACK_BOTTOM;
2610 *mpls |= MPLS_STACK_BOTTOM;
2613 static inline __be16 build_tci(unsigned int id, unsigned int cfi,
2616 return htons(id | (cfi << 12) | (prio << 13));
2619 static void pktgen_finalize_skb(struct pktgen_dev *pkt_dev, struct sk_buff *skb,
2622 struct timespec64 timestamp;
2623 struct pktgen_hdr *pgh;
2625 pgh = skb_put(skb, sizeof(*pgh));
2626 datalen -= sizeof(*pgh);
2628 if (pkt_dev->nfrags <= 0) {
2629 skb_put_zero(skb, datalen);
2631 int frags = pkt_dev->nfrags;
2636 if (frags > MAX_SKB_FRAGS)
2637 frags = MAX_SKB_FRAGS;
2638 len = datalen - frags * PAGE_SIZE;
2640 skb_put_zero(skb, len);
2641 datalen = frags * PAGE_SIZE;
2645 frag_len = (datalen/frags) < PAGE_SIZE ?
2646 (datalen/frags) : PAGE_SIZE;
2647 while (datalen > 0) {
2648 if (unlikely(!pkt_dev->page)) {
2649 int node = numa_node_id();
2651 if (pkt_dev->node >= 0 && (pkt_dev->flags & F_NODE))
2652 node = pkt_dev->node;
2653 pkt_dev->page = alloc_pages_node(node, GFP_KERNEL | __GFP_ZERO, 0);
2657 get_page(pkt_dev->page);
2658 skb_frag_set_page(skb, i, pkt_dev->page);
2659 skb_shinfo(skb)->frags[i].page_offset = 0;
2660 /*last fragment, fill rest of data*/
2661 if (i == (frags - 1))
2662 skb_frag_size_set(&skb_shinfo(skb)->frags[i],
2663 (datalen < PAGE_SIZE ? datalen : PAGE_SIZE));
2665 skb_frag_size_set(&skb_shinfo(skb)->frags[i], frag_len);
2666 datalen -= skb_frag_size(&skb_shinfo(skb)->frags[i]);
2667 skb->len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2668 skb->data_len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2670 skb_shinfo(skb)->nr_frags = i;
2674 /* Stamp the time, and sequence number,
2675 * convert them to network byte order
2677 pgh->pgh_magic = htonl(PKTGEN_MAGIC);
2678 pgh->seq_num = htonl(pkt_dev->seq_num);
2680 if (pkt_dev->flags & F_NO_TIMESTAMP) {
2685 * pgh->tv_sec wraps in y2106 when interpreted as unsigned
2686 * as done by wireshark, or y2038 when interpreted as signed.
2687 * This is probably harmless, but if anyone wants to improve
2688 * it, we could introduce a variant that puts 64-bit nanoseconds
2689 * into the respective header bytes.
2690 * This would also be slightly faster to read.
2692 ktime_get_real_ts64(×tamp);
2693 pgh->tv_sec = htonl(timestamp.tv_sec);
2694 pgh->tv_usec = htonl(timestamp.tv_nsec / NSEC_PER_USEC);
2698 static struct sk_buff *pktgen_alloc_skb(struct net_device *dev,
2699 struct pktgen_dev *pkt_dev)
2701 unsigned int extralen = LL_RESERVED_SPACE(dev);
2702 struct sk_buff *skb = NULL;
2705 size = pkt_dev->cur_pkt_size + 64 + extralen + pkt_dev->pkt_overhead;
2706 if (pkt_dev->flags & F_NODE) {
2707 int node = pkt_dev->node >= 0 ? pkt_dev->node : numa_node_id();
2709 skb = __alloc_skb(NET_SKB_PAD + size, GFP_NOWAIT, 0, node);
2711 skb_reserve(skb, NET_SKB_PAD);
2715 skb = __netdev_alloc_skb(dev, size, GFP_NOWAIT);
2718 /* the caller pre-fetches from skb->data and reserves for the mac hdr */
2720 skb_reserve(skb, extralen - 16);
2725 static struct sk_buff *fill_packet_ipv4(struct net_device *odev,
2726 struct pktgen_dev *pkt_dev)
2728 struct sk_buff *skb = NULL;
2730 struct udphdr *udph;
2733 __be16 protocol = htons(ETH_P_IP);
2735 __be16 *vlan_tci = NULL; /* Encapsulates priority and VLAN ID */
2736 __be16 *vlan_encapsulated_proto = NULL; /* packet type ID field (or len) for VLAN tag */
2737 __be16 *svlan_tci = NULL; /* Encapsulates priority and SVLAN ID */
2738 __be16 *svlan_encapsulated_proto = NULL; /* packet type ID field (or len) for SVLAN tag */
2741 if (pkt_dev->nr_labels)
2742 protocol = htons(ETH_P_MPLS_UC);
2744 if (pkt_dev->vlan_id != 0xffff)
2745 protocol = htons(ETH_P_8021Q);
2747 /* Update any of the values, used when we're incrementing various
2750 mod_cur_headers(pkt_dev);
2751 queue_map = pkt_dev->cur_queue_map;
2753 skb = pktgen_alloc_skb(odev, pkt_dev);
2755 sprintf(pkt_dev->result, "No memory");
2759 prefetchw(skb->data);
2760 skb_reserve(skb, 16);
2762 /* Reserve for ethernet and IP header */
2763 eth = skb_push(skb, 14);
2764 mpls = skb_put(skb, pkt_dev->nr_labels * sizeof(__u32));
2765 if (pkt_dev->nr_labels)
2766 mpls_push(mpls, pkt_dev);
2768 if (pkt_dev->vlan_id != 0xffff) {
2769 if (pkt_dev->svlan_id != 0xffff) {
2770 svlan_tci = skb_put(skb, sizeof(__be16));
2771 *svlan_tci = build_tci(pkt_dev->svlan_id,
2774 svlan_encapsulated_proto = skb_put(skb,
2776 *svlan_encapsulated_proto = htons(ETH_P_8021Q);
2778 vlan_tci = skb_put(skb, sizeof(__be16));
2779 *vlan_tci = build_tci(pkt_dev->vlan_id,
2782 vlan_encapsulated_proto = skb_put(skb, sizeof(__be16));
2783 *vlan_encapsulated_proto = htons(ETH_P_IP);
2786 skb_reset_mac_header(skb);
2787 skb_set_network_header(skb, skb->len);
2788 iph = skb_put(skb, sizeof(struct iphdr));
2790 skb_set_transport_header(skb, skb->len);
2791 udph = skb_put(skb, sizeof(struct udphdr));
2792 skb_set_queue_mapping(skb, queue_map);
2793 skb->priority = pkt_dev->skb_priority;
2795 memcpy(eth, pkt_dev->hh, 12);
2796 *(__be16 *) & eth[12] = protocol;
2798 /* Eth + IPh + UDPh + mpls */
2799 datalen = pkt_dev->cur_pkt_size - 14 - 20 - 8 -
2800 pkt_dev->pkt_overhead;
2801 if (datalen < 0 || datalen < sizeof(struct pktgen_hdr))
2802 datalen = sizeof(struct pktgen_hdr);
2804 udph->source = htons(pkt_dev->cur_udp_src);
2805 udph->dest = htons(pkt_dev->cur_udp_dst);
2806 udph->len = htons(datalen + 8); /* DATA + udphdr */
2812 iph->tos = pkt_dev->tos;
2813 iph->protocol = IPPROTO_UDP; /* UDP */
2814 iph->saddr = pkt_dev->cur_saddr;
2815 iph->daddr = pkt_dev->cur_daddr;
2816 iph->id = htons(pkt_dev->ip_id);
2819 iplen = 20 + 8 + datalen;
2820 iph->tot_len = htons(iplen);
2822 skb->protocol = protocol;
2824 skb->pkt_type = PACKET_HOST;
2826 pktgen_finalize_skb(pkt_dev, skb, datalen);
2828 if (!(pkt_dev->flags & F_UDPCSUM)) {
2829 skb->ip_summed = CHECKSUM_NONE;
2830 } else if (odev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM)) {
2831 skb->ip_summed = CHECKSUM_PARTIAL;
2833 udp4_hwcsum(skb, iph->saddr, iph->daddr);
2835 __wsum csum = skb_checksum(skb, skb_transport_offset(skb), datalen + 8, 0);
2837 /* add protocol-dependent pseudo-header */
2838 udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
2839 datalen + 8, IPPROTO_UDP, csum);
2841 if (udph->check == 0)
2842 udph->check = CSUM_MANGLED_0;
2846 if (!process_ipsec(pkt_dev, skb, protocol))
2853 static struct sk_buff *fill_packet_ipv6(struct net_device *odev,
2854 struct pktgen_dev *pkt_dev)
2856 struct sk_buff *skb = NULL;
2858 struct udphdr *udph;
2859 int datalen, udplen;
2860 struct ipv6hdr *iph;
2861 __be16 protocol = htons(ETH_P_IPV6);
2863 __be16 *vlan_tci = NULL; /* Encapsulates priority and VLAN ID */
2864 __be16 *vlan_encapsulated_proto = NULL; /* packet type ID field (or len) for VLAN tag */
2865 __be16 *svlan_tci = NULL; /* Encapsulates priority and SVLAN ID */
2866 __be16 *svlan_encapsulated_proto = NULL; /* packet type ID field (or len) for SVLAN tag */
2869 if (pkt_dev->nr_labels)
2870 protocol = htons(ETH_P_MPLS_UC);
2872 if (pkt_dev->vlan_id != 0xffff)
2873 protocol = htons(ETH_P_8021Q);
2875 /* Update any of the values, used when we're incrementing various
2878 mod_cur_headers(pkt_dev);
2879 queue_map = pkt_dev->cur_queue_map;
2881 skb = pktgen_alloc_skb(odev, pkt_dev);
2883 sprintf(pkt_dev->result, "No memory");
2887 prefetchw(skb->data);
2888 skb_reserve(skb, 16);
2890 /* Reserve for ethernet and IP header */
2891 eth = skb_push(skb, 14);
2892 mpls = skb_put(skb, pkt_dev->nr_labels * sizeof(__u32));
2893 if (pkt_dev->nr_labels)
2894 mpls_push(mpls, pkt_dev);
2896 if (pkt_dev->vlan_id != 0xffff) {
2897 if (pkt_dev->svlan_id != 0xffff) {
2898 svlan_tci = skb_put(skb, sizeof(__be16));
2899 *svlan_tci = build_tci(pkt_dev->svlan_id,
2902 svlan_encapsulated_proto = skb_put(skb,
2904 *svlan_encapsulated_proto = htons(ETH_P_8021Q);
2906 vlan_tci = skb_put(skb, sizeof(__be16));
2907 *vlan_tci = build_tci(pkt_dev->vlan_id,
2910 vlan_encapsulated_proto = skb_put(skb, sizeof(__be16));
2911 *vlan_encapsulated_proto = htons(ETH_P_IPV6);
2914 skb_reset_mac_header(skb);
2915 skb_set_network_header(skb, skb->len);
2916 iph = skb_put(skb, sizeof(struct ipv6hdr));
2918 skb_set_transport_header(skb, skb->len);
2919 udph = skb_put(skb, sizeof(struct udphdr));
2920 skb_set_queue_mapping(skb, queue_map);
2921 skb->priority = pkt_dev->skb_priority;
2923 memcpy(eth, pkt_dev->hh, 12);
2924 *(__be16 *) ð[12] = protocol;
2926 /* Eth + IPh + UDPh + mpls */
2927 datalen = pkt_dev->cur_pkt_size - 14 -
2928 sizeof(struct ipv6hdr) - sizeof(struct udphdr) -
2929 pkt_dev->pkt_overhead;
2931 if (datalen < 0 || datalen < sizeof(struct pktgen_hdr)) {
2932 datalen = sizeof(struct pktgen_hdr);
2933 net_info_ratelimited("increased datalen to %d\n", datalen);
2936 udplen = datalen + sizeof(struct udphdr);
2937 udph->source = htons(pkt_dev->cur_udp_src);
2938 udph->dest = htons(pkt_dev->cur_udp_dst);
2939 udph->len = htons(udplen);
2942 *(__be32 *) iph = htonl(0x60000000); /* Version + flow */
2944 if (pkt_dev->traffic_class) {
2945 /* Version + traffic class + flow (0) */
2946 *(__be32 *)iph |= htonl(0x60000000 | (pkt_dev->traffic_class << 20));
2949 iph->hop_limit = 32;
2951 iph->payload_len = htons(udplen);
2952 iph->nexthdr = IPPROTO_UDP;
2954 iph->daddr = pkt_dev->cur_in6_daddr;
2955 iph->saddr = pkt_dev->cur_in6_saddr;
2957 skb->protocol = protocol;
2959 skb->pkt_type = PACKET_HOST;
2961 pktgen_finalize_skb(pkt_dev, skb, datalen);
2963 if (!(pkt_dev->flags & F_UDPCSUM)) {
2964 skb->ip_summed = CHECKSUM_NONE;
2965 } else if (odev->features & (NETIF_F_HW_CSUM | NETIF_F_IPV6_CSUM)) {
2966 skb->ip_summed = CHECKSUM_PARTIAL;
2967 skb->csum_start = skb_transport_header(skb) - skb->head;
2968 skb->csum_offset = offsetof(struct udphdr, check);
2969 udph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, udplen, IPPROTO_UDP, 0);
2971 __wsum csum = skb_checksum(skb, skb_transport_offset(skb), udplen, 0);
2973 /* add protocol-dependent pseudo-header */
2974 udph->check = csum_ipv6_magic(&iph->saddr, &iph->daddr, udplen, IPPROTO_UDP, csum);
2976 if (udph->check == 0)
2977 udph->check = CSUM_MANGLED_0;
2983 static struct sk_buff *fill_packet(struct net_device *odev,
2984 struct pktgen_dev *pkt_dev)
2986 if (pkt_dev->flags & F_IPV6)
2987 return fill_packet_ipv6(odev, pkt_dev);
2989 return fill_packet_ipv4(odev, pkt_dev);
2992 static void pktgen_clear_counters(struct pktgen_dev *pkt_dev)
2994 pkt_dev->seq_num = 1;
2995 pkt_dev->idle_acc = 0;
2997 pkt_dev->tx_bytes = 0;
2998 pkt_dev->errors = 0;
3001 /* Set up structure for sending pkts, clear counters */
3003 static void pktgen_run(struct pktgen_thread *t)
3005 struct pktgen_dev *pkt_dev;
3011 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) {
3014 * setup odev and create initial packet.
3016 pktgen_setup_inject(pkt_dev);
3018 if (pkt_dev->odev) {
3019 pktgen_clear_counters(pkt_dev);
3020 pkt_dev->skb = NULL;
3021 pkt_dev->started_at = pkt_dev->next_tx = ktime_get();
3023 set_pkt_overhead(pkt_dev);
3025 strcpy(pkt_dev->result, "Starting");
3026 pkt_dev->running = 1; /* Cranke yeself! */
3029 strcpy(pkt_dev->result, "Error starting");
3033 t->control &= ~(T_STOP);
3036 static void pktgen_stop_all_threads_ifs(struct pktgen_net *pn)
3038 struct pktgen_thread *t;
3042 mutex_lock(&pktgen_thread_lock);
3044 list_for_each_entry(t, &pn->pktgen_threads, th_list)
3045 t->control |= T_STOP;
3047 mutex_unlock(&pktgen_thread_lock);
3050 static int thread_is_running(const struct pktgen_thread *t)
3052 const struct pktgen_dev *pkt_dev;
3055 list_for_each_entry_rcu(pkt_dev, &t->if_list, list)
3056 if (pkt_dev->running) {
3064 static int pktgen_wait_thread_run(struct pktgen_thread *t)
3066 while (thread_is_running(t)) {
3068 msleep_interruptible(100);
3070 if (signal_pending(current))
3078 static int pktgen_wait_all_threads_run(struct pktgen_net *pn)
3080 struct pktgen_thread *t;
3083 mutex_lock(&pktgen_thread_lock);
3085 list_for_each_entry(t, &pn->pktgen_threads, th_list) {
3086 sig = pktgen_wait_thread_run(t);
3092 list_for_each_entry(t, &pn->pktgen_threads, th_list)
3093 t->control |= (T_STOP);
3095 mutex_unlock(&pktgen_thread_lock);
3099 static void pktgen_run_all_threads(struct pktgen_net *pn)
3101 struct pktgen_thread *t;
3105 mutex_lock(&pktgen_thread_lock);
3107 list_for_each_entry(t, &pn->pktgen_threads, th_list)
3108 t->control |= (T_RUN);
3110 mutex_unlock(&pktgen_thread_lock);
3112 /* Propagate thread->control */
3113 schedule_timeout_interruptible(msecs_to_jiffies(125));
3115 pktgen_wait_all_threads_run(pn);
3118 static void pktgen_reset_all_threads(struct pktgen_net *pn)
3120 struct pktgen_thread *t;
3124 mutex_lock(&pktgen_thread_lock);
3126 list_for_each_entry(t, &pn->pktgen_threads, th_list)
3127 t->control |= (T_REMDEVALL);
3129 mutex_unlock(&pktgen_thread_lock);
3131 /* Propagate thread->control */
3132 schedule_timeout_interruptible(msecs_to_jiffies(125));
3134 pktgen_wait_all_threads_run(pn);
3137 static void show_results(struct pktgen_dev *pkt_dev, int nr_frags)
3139 __u64 bps, mbps, pps;
3140 char *p = pkt_dev->result;
3141 ktime_t elapsed = ktime_sub(pkt_dev->stopped_at,
3142 pkt_dev->started_at);
3143 ktime_t idle = ns_to_ktime(pkt_dev->idle_acc);
3145 p += sprintf(p, "OK: %llu(c%llu+d%llu) usec, %llu (%dbyte,%dfrags)\n",
3146 (unsigned long long)ktime_to_us(elapsed),
3147 (unsigned long long)ktime_to_us(ktime_sub(elapsed, idle)),
3148 (unsigned long long)ktime_to_us(idle),
3149 (unsigned long long)pkt_dev->sofar,
3150 pkt_dev->cur_pkt_size, nr_frags);
3152 pps = div64_u64(pkt_dev->sofar * NSEC_PER_SEC,
3153 ktime_to_ns(elapsed));
3155 bps = pps * 8 * pkt_dev->cur_pkt_size;
3158 do_div(mbps, 1000000);
3159 p += sprintf(p, " %llupps %lluMb/sec (%llubps) errors: %llu",
3160 (unsigned long long)pps,
3161 (unsigned long long)mbps,
3162 (unsigned long long)bps,
3163 (unsigned long long)pkt_dev->errors);
3166 /* Set stopped-at timer, remove from running list, do counters & statistics */
3167 static int pktgen_stop_device(struct pktgen_dev *pkt_dev)
3169 int nr_frags = pkt_dev->skb ? skb_shinfo(pkt_dev->skb)->nr_frags : -1;
3171 if (!pkt_dev->running) {
3172 pr_warn("interface: %s is already stopped\n",
3177 pkt_dev->running = 0;
3178 kfree_skb(pkt_dev->skb);
3179 pkt_dev->skb = NULL;
3180 pkt_dev->stopped_at = ktime_get();
3182 show_results(pkt_dev, nr_frags);
3187 static struct pktgen_dev *next_to_run(struct pktgen_thread *t)
3189 struct pktgen_dev *pkt_dev, *best = NULL;
3192 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) {
3193 if (!pkt_dev->running)
3197 else if (ktime_compare(pkt_dev->next_tx, best->next_tx) < 0)
3205 static void pktgen_stop(struct pktgen_thread *t)
3207 struct pktgen_dev *pkt_dev;
3213 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) {
3214 pktgen_stop_device(pkt_dev);
3221 * one of our devices needs to be removed - find it
3224 static void pktgen_rem_one_if(struct pktgen_thread *t)
3226 struct list_head *q, *n;
3227 struct pktgen_dev *cur;
3231 list_for_each_safe(q, n, &t->if_list) {
3232 cur = list_entry(q, struct pktgen_dev, list);
3234 if (!cur->removal_mark)
3237 kfree_skb(cur->skb);
3240 pktgen_remove_device(t, cur);
3246 static void pktgen_rem_all_ifs(struct pktgen_thread *t)
3248 struct list_head *q, *n;
3249 struct pktgen_dev *cur;
3253 /* Remove all devices, free mem */
3255 list_for_each_safe(q, n, &t->if_list) {
3256 cur = list_entry(q, struct pktgen_dev, list);
3258 kfree_skb(cur->skb);
3261 pktgen_remove_device(t, cur);
3265 static void pktgen_rem_thread(struct pktgen_thread *t)
3267 /* Remove from the thread list */
3268 remove_proc_entry(t->tsk->comm, t->net->proc_dir);
3271 static void pktgen_resched(struct pktgen_dev *pkt_dev)
3273 ktime_t idle_start = ktime_get();
3275 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(ktime_get(), idle_start));
3278 static void pktgen_wait_for_skb(struct pktgen_dev *pkt_dev)
3280 ktime_t idle_start = ktime_get();
3282 while (refcount_read(&(pkt_dev->skb->users)) != 1) {
3283 if (signal_pending(current))
3287 pktgen_resched(pkt_dev);
3291 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(ktime_get(), idle_start));
3294 static void pktgen_xmit(struct pktgen_dev *pkt_dev)
3296 unsigned int burst = READ_ONCE(pkt_dev->burst);
3297 struct net_device *odev = pkt_dev->odev;
3298 struct netdev_queue *txq;
3299 struct sk_buff *skb;
3302 /* If device is offline, then don't send */
3303 if (unlikely(!netif_running(odev) || !netif_carrier_ok(odev))) {
3304 pktgen_stop_device(pkt_dev);
3308 /* This is max DELAY, this has special meaning of
3311 if (unlikely(pkt_dev->delay == ULLONG_MAX)) {
3312 pkt_dev->next_tx = ktime_add_ns(ktime_get(), ULONG_MAX);
3316 /* If no skb or clone count exhausted then get new one */
3317 if (!pkt_dev->skb || (pkt_dev->last_ok &&
3318 ++pkt_dev->clone_count >= pkt_dev->clone_skb)) {
3319 /* build a new pkt */
3320 kfree_skb(pkt_dev->skb);
3322 pkt_dev->skb = fill_packet(odev, pkt_dev);
3323 if (pkt_dev->skb == NULL) {
3324 pr_err("ERROR: couldn't allocate skb in fill_packet\n");
3326 pkt_dev->clone_count--; /* back out increment, OOM */
3329 pkt_dev->last_pkt_size = pkt_dev->skb->len;
3330 pkt_dev->clone_count = 0; /* reset counter */
3333 if (pkt_dev->delay && pkt_dev->last_ok)
3334 spin(pkt_dev, pkt_dev->next_tx);
3336 if (pkt_dev->xmit_mode == M_NETIF_RECEIVE) {
3338 skb->protocol = eth_type_trans(skb, skb->dev);
3339 refcount_add(burst, &skb->users);
3342 ret = netif_receive_skb(skb);
3343 if (ret == NET_RX_DROP)
3347 if (refcount_read(&skb->users) != burst) {
3348 /* skb was queued by rps/rfs or taps,
3349 * so cannot reuse this skb
3351 WARN_ON(refcount_sub_and_test(burst - 1, &skb->users));
3352 /* get out of the loop and wait
3353 * until skb is consumed
3357 /* skb was 'freed' by stack, so clean few
3361 } while (--burst > 0);
3362 goto out; /* Skips xmit_mode M_START_XMIT */
3363 } else if (pkt_dev->xmit_mode == M_QUEUE_XMIT) {
3365 refcount_inc(&pkt_dev->skb->users);
3367 ret = dev_queue_xmit(pkt_dev->skb);
3369 case NET_XMIT_SUCCESS:
3372 pkt_dev->tx_bytes += pkt_dev->last_pkt_size;
3376 /* These are all valid return codes for a qdisc but
3377 * indicate packets are being dropped or will likely
3380 case NETDEV_TX_BUSY:
3381 /* qdisc may call dev_hard_start_xmit directly in cases
3382 * where no queues exist e.g. loopback device, virtual
3383 * devices, etc. In this case we need to handle
3388 net_info_ratelimited("%s xmit error: %d\n",
3389 pkt_dev->odevname, ret);
3395 txq = skb_get_tx_queue(odev, pkt_dev->skb);
3399 HARD_TX_LOCK(odev, txq, smp_processor_id());
3401 if (unlikely(netif_xmit_frozen_or_drv_stopped(txq))) {
3402 ret = NETDEV_TX_BUSY;
3403 pkt_dev->last_ok = 0;
3406 refcount_add(burst, &pkt_dev->skb->users);
3409 ret = netdev_start_xmit(pkt_dev->skb, odev, txq, --burst > 0);
3413 pkt_dev->last_ok = 1;
3416 pkt_dev->tx_bytes += pkt_dev->last_pkt_size;
3417 if (burst > 0 && !netif_xmit_frozen_or_drv_stopped(txq))
3422 /* skb has been consumed */
3425 default: /* Drivers are not supposed to return other values! */
3426 net_info_ratelimited("%s xmit error: %d\n",
3427 pkt_dev->odevname, ret);
3430 case NETDEV_TX_BUSY:
3431 /* Retry it next time */
3432 refcount_dec(&(pkt_dev->skb->users));
3433 pkt_dev->last_ok = 0;
3435 if (unlikely(burst))
3436 WARN_ON(refcount_sub_and_test(burst, &pkt_dev->skb->users));
3438 HARD_TX_UNLOCK(odev, txq);
3443 /* If pkt_dev->count is zero, then run forever */
3444 if ((pkt_dev->count != 0) && (pkt_dev->sofar >= pkt_dev->count)) {
3445 pktgen_wait_for_skb(pkt_dev);
3447 /* Done with this */
3448 pktgen_stop_device(pkt_dev);
3453 * Main loop of the thread goes here
3456 static int pktgen_thread_worker(void *arg)
3459 struct pktgen_thread *t = arg;
3460 struct pktgen_dev *pkt_dev = NULL;
3463 BUG_ON(smp_processor_id() != cpu);
3465 init_waitqueue_head(&t->queue);
3466 complete(&t->start_done);
3468 pr_debug("starting pktgen/%d: pid=%d\n", cpu, task_pid_nr(current));
3472 while (!kthread_should_stop()) {
3473 pkt_dev = next_to_run(t);
3475 if (unlikely(!pkt_dev && t->control == 0)) {
3476 if (t->net->pktgen_exiting)
3478 wait_event_interruptible_timeout(t->queue,
3485 if (likely(pkt_dev)) {
3486 pktgen_xmit(pkt_dev);
3489 pktgen_resched(pkt_dev);
3494 if (t->control & T_STOP) {
3496 t->control &= ~(T_STOP);
3499 if (t->control & T_RUN) {
3501 t->control &= ~(T_RUN);
3504 if (t->control & T_REMDEVALL) {
3505 pktgen_rem_all_ifs(t);
3506 t->control &= ~(T_REMDEVALL);
3509 if (t->control & T_REMDEV) {
3510 pktgen_rem_one_if(t);
3511 t->control &= ~(T_REMDEV);
3517 pr_debug("%s stopping all device\n", t->tsk->comm);
3520 pr_debug("%s removing all device\n", t->tsk->comm);
3521 pktgen_rem_all_ifs(t);
3523 pr_debug("%s removing thread\n", t->tsk->comm);
3524 pktgen_rem_thread(t);
3529 static struct pktgen_dev *pktgen_find_dev(struct pktgen_thread *t,
3530 const char *ifname, bool exact)
3532 struct pktgen_dev *p, *pkt_dev = NULL;
3533 size_t len = strlen(ifname);
3536 list_for_each_entry_rcu(p, &t->if_list, list)
3537 if (strncmp(p->odevname, ifname, len) == 0) {
3538 if (p->odevname[len]) {
3539 if (exact || p->odevname[len] != '@')
3547 pr_debug("find_dev(%s) returning %p\n", ifname, pkt_dev);
3552 * Adds a dev at front of if_list.
3555 static int add_dev_to_thread(struct pktgen_thread *t,
3556 struct pktgen_dev *pkt_dev)
3560 /* This function cannot be called concurrently, as its called
3561 * under pktgen_thread_lock mutex, but it can run from
3562 * userspace on another CPU than the kthread. The if_lock()
3563 * is used here to sync with concurrent instances of
3564 * _rem_dev_from_if_list() invoked via kthread, which is also
3565 * updating the if_list */
3568 if (pkt_dev->pg_thread) {
3569 pr_err("ERROR: already assigned to a thread\n");
3574 pkt_dev->running = 0;
3575 pkt_dev->pg_thread = t;
3576 list_add_rcu(&pkt_dev->list, &t->if_list);
3583 /* Called under thread lock */
3585 static int pktgen_add_device(struct pktgen_thread *t, const char *ifname)
3587 struct pktgen_dev *pkt_dev;
3589 int node = cpu_to_node(t->cpu);
3591 /* We don't allow a device to be on several threads */
3593 pkt_dev = __pktgen_NN_threads(t->net, ifname, FIND);
3595 pr_err("ERROR: interface already used\n");
3599 pkt_dev = kzalloc_node(sizeof(struct pktgen_dev), GFP_KERNEL, node);
3603 strcpy(pkt_dev->odevname, ifname);
3604 pkt_dev->flows = vzalloc_node(array_size(MAX_CFLOWS,
3605 sizeof(struct flow_state)),
3607 if (pkt_dev->flows == NULL) {
3612 pkt_dev->removal_mark = 0;
3613 pkt_dev->nfrags = 0;
3614 pkt_dev->delay = pg_delay_d;
3615 pkt_dev->count = pg_count_d;
3617 pkt_dev->udp_src_min = 9; /* sink port */
3618 pkt_dev->udp_src_max = 9;
3619 pkt_dev->udp_dst_min = 9;
3620 pkt_dev->udp_dst_max = 9;
3621 pkt_dev->vlan_p = 0;
3622 pkt_dev->vlan_cfi = 0;
3623 pkt_dev->vlan_id = 0xffff;
3624 pkt_dev->svlan_p = 0;
3625 pkt_dev->svlan_cfi = 0;
3626 pkt_dev->svlan_id = 0xffff;
3630 err = pktgen_setup_dev(t->net, pkt_dev, ifname);
3633 if (pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING)
3634 pkt_dev->clone_skb = pg_clone_skb_d;
3636 pkt_dev->entry = proc_create_data(ifname, 0600, t->net->proc_dir,
3637 &pktgen_if_fops, pkt_dev);
3638 if (!pkt_dev->entry) {
3639 pr_err("cannot create %s/%s procfs entry\n",
3640 PG_PROC_DIR, ifname);
3645 pkt_dev->ipsmode = XFRM_MODE_TRANSPORT;
3646 pkt_dev->ipsproto = IPPROTO_ESP;
3648 /* xfrm tunnel mode needs additional dst to extract outter
3649 * ip header protocol/ttl/id field, here creat a phony one.
3650 * instead of looking for a valid rt, which definitely hurting
3651 * performance under such circumstance.
3653 pkt_dev->dstops.family = AF_INET;
3654 pkt_dev->xdst.u.dst.dev = pkt_dev->odev;
3655 dst_init_metrics(&pkt_dev->xdst.u.dst, pktgen_dst_metrics, false);
3656 pkt_dev->xdst.child = &pkt_dev->xdst.u.dst;
3657 pkt_dev->xdst.u.dst.ops = &pkt_dev->dstops;
3660 return add_dev_to_thread(t, pkt_dev);
3662 dev_put(pkt_dev->odev);
3667 vfree(pkt_dev->flows);
3672 static int __net_init pktgen_create_thread(int cpu, struct pktgen_net *pn)
3674 struct pktgen_thread *t;
3675 struct proc_dir_entry *pe;
3676 struct task_struct *p;
3678 t = kzalloc_node(sizeof(struct pktgen_thread), GFP_KERNEL,
3681 pr_err("ERROR: out of memory, can't create new thread\n");
3685 mutex_init(&t->if_lock);
3688 INIT_LIST_HEAD(&t->if_list);
3690 list_add_tail(&t->th_list, &pn->pktgen_threads);
3691 init_completion(&t->start_done);
3693 p = kthread_create_on_node(pktgen_thread_worker,
3696 "kpktgend_%d", cpu);
3698 pr_err("kernel_thread() failed for cpu %d\n", t->cpu);
3699 list_del(&t->th_list);
3703 kthread_bind(p, cpu);
3706 pe = proc_create_data(t->tsk->comm, 0600, pn->proc_dir,
3707 &pktgen_thread_fops, t);
3709 pr_err("cannot create %s/%s procfs entry\n",
3710 PG_PROC_DIR, t->tsk->comm);
3712 list_del(&t->th_list);
3720 wait_for_completion(&t->start_done);
3726 * Removes a device from the thread if_list.
3728 static void _rem_dev_from_if_list(struct pktgen_thread *t,
3729 struct pktgen_dev *pkt_dev)
3731 struct list_head *q, *n;
3732 struct pktgen_dev *p;
3735 list_for_each_safe(q, n, &t->if_list) {
3736 p = list_entry(q, struct pktgen_dev, list);
3738 list_del_rcu(&p->list);
3743 static int pktgen_remove_device(struct pktgen_thread *t,
3744 struct pktgen_dev *pkt_dev)
3746 pr_debug("remove_device pkt_dev=%p\n", pkt_dev);
3748 if (pkt_dev->running) {
3749 pr_warn("WARNING: trying to remove a running interface, stopping it now\n");
3750 pktgen_stop_device(pkt_dev);
3753 /* Dis-associate from the interface */
3755 if (pkt_dev->odev) {
3756 dev_put(pkt_dev->odev);
3757 pkt_dev->odev = NULL;
3760 /* Remove proc before if_list entry, because add_device uses
3761 * list to determine if interface already exist, avoid race
3762 * with proc_create_data() */
3763 proc_remove(pkt_dev->entry);
3765 /* And update the thread if_list */
3766 _rem_dev_from_if_list(t, pkt_dev);
3771 vfree(pkt_dev->flows);
3773 put_page(pkt_dev->page);
3774 kfree_rcu(pkt_dev, rcu);
3778 static int __net_init pg_net_init(struct net *net)
3780 struct pktgen_net *pn = net_generic(net, pg_net_id);
3781 struct proc_dir_entry *pe;
3785 INIT_LIST_HEAD(&pn->pktgen_threads);
3786 pn->pktgen_exiting = false;
3787 pn->proc_dir = proc_mkdir(PG_PROC_DIR, pn->net->proc_net);
3788 if (!pn->proc_dir) {
3789 pr_warn("cannot create /proc/net/%s\n", PG_PROC_DIR);
3792 pe = proc_create(PGCTRL, 0600, pn->proc_dir, &pktgen_fops);
3794 pr_err("cannot create %s procfs entry\n", PGCTRL);
3799 for_each_online_cpu(cpu) {
3802 err = pktgen_create_thread(cpu, pn);
3804 pr_warn("Cannot create thread for cpu %d (%d)\n",
3808 if (list_empty(&pn->pktgen_threads)) {
3809 pr_err("Initialization failed for all threads\n");
3817 remove_proc_entry(PGCTRL, pn->proc_dir);
3819 remove_proc_entry(PG_PROC_DIR, pn->net->proc_net);
3823 static void __net_exit pg_net_exit(struct net *net)
3825 struct pktgen_net *pn = net_generic(net, pg_net_id);
3826 struct pktgen_thread *t;
3827 struct list_head *q, *n;
3830 /* Stop all interfaces & threads */
3831 pn->pktgen_exiting = true;
3833 mutex_lock(&pktgen_thread_lock);
3834 list_splice_init(&pn->pktgen_threads, &list);
3835 mutex_unlock(&pktgen_thread_lock);
3837 list_for_each_safe(q, n, &list) {
3838 t = list_entry(q, struct pktgen_thread, th_list);
3839 list_del(&t->th_list);
3840 kthread_stop(t->tsk);
3841 put_task_struct(t->tsk);
3845 remove_proc_entry(PGCTRL, pn->proc_dir);
3846 remove_proc_entry(PG_PROC_DIR, pn->net->proc_net);
3849 static struct pernet_operations pg_net_ops = {
3850 .init = pg_net_init,
3851 .exit = pg_net_exit,
3853 .size = sizeof(struct pktgen_net),
3856 static int __init pg_init(void)
3860 pr_info("%s", version);
3861 ret = register_pernet_subsys(&pg_net_ops);
3864 ret = register_netdevice_notifier(&pktgen_notifier_block);
3866 unregister_pernet_subsys(&pg_net_ops);
3871 static void __exit pg_cleanup(void)
3873 unregister_netdevice_notifier(&pktgen_notifier_block);
3874 unregister_pernet_subsys(&pg_net_ops);
3875 /* Don't need rcu_barrier() due to use of kfree_rcu() */
3878 module_init(pg_init);
3879 module_exit(pg_cleanup);
3881 MODULE_AUTHOR("Robert Olsson <robert.olsson@its.uu.se>");
3882 MODULE_DESCRIPTION("Packet Generator tool");
3883 MODULE_LICENSE("GPL");
3884 MODULE_VERSION(VERSION);
3885 module_param(pg_count_d, int, 0);
3886 MODULE_PARM_DESC(pg_count_d, "Default number of packets to inject");
3887 module_param(pg_delay_d, int, 0);
3888 MODULE_PARM_DESC(pg_delay_d, "Default delay between packets (nanoseconds)");
3889 module_param(pg_clone_skb_d, int, 0);
3890 MODULE_PARM_DESC(pg_clone_skb_d, "Default number of copies of the same packet");
3891 module_param(debug, int, 0);
3892 MODULE_PARM_DESC(debug, "Enable debugging of pktgen module");