sched: remove NET_XMIT_POLICED
[linux/fpc-iii.git] / net / core / pktgen.c
blobf74ab9c3b38fe10610b3a4e8610ed78bc0d8437a
1 /*
2 * Authors:
3 * Copyright 2001, 2002 by Robert Olsson <robert.olsson@its.uu.se>
4 * Uppsala University and
5 * Swedish University of Agricultural Sciences
7 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
8 * Ben Greear <greearb@candelatech.com>
9 * Jens Låås <jens.laas@data.slu.se>
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License
13 * as published by the Free Software Foundation; either version
14 * 2 of the License, or (at your option) any later version.
17 * A tool for loading the network with preconfigurated packets.
18 * The tool is implemented as a linux module. Parameters are output
19 * device, delay (to hard_xmit), number of packets, and whether
20 * to use multiple SKBs or just the same one.
21 * pktgen uses the installed interface's output routine.
23 * Additional hacking by:
25 * Jens.Laas@data.slu.se
26 * Improved by ANK. 010120.
27 * Improved by ANK even more. 010212.
28 * MAC address typo fixed. 010417 --ro
29 * Integrated. 020301 --DaveM
30 * Added multiskb option 020301 --DaveM
31 * Scaling of results. 020417--sigurdur@linpro.no
32 * Significant re-work of the module:
33 * * Convert to threaded model to more efficiently be able to transmit
34 * and receive on multiple interfaces at once.
35 * * Converted many counters to __u64 to allow longer runs.
36 * * Allow configuration of ranges, like min/max IP address, MACs,
37 * and UDP-ports, for both source and destination, and can
38 * set to use a random distribution or sequentially walk the range.
39 * * Can now change most values after starting.
40 * * Place 12-byte packet in UDP payload with magic number,
41 * sequence number, and timestamp.
42 * * Add receiver code that detects dropped pkts, re-ordered pkts, and
43 * latencies (with micro-second) precision.
44 * * Add IOCTL interface to easily get counters & configuration.
45 * --Ben Greear <greearb@candelatech.com>
47 * Renamed multiskb to clone_skb and cleaned up sending core for two distinct
48 * skb modes. A clone_skb=0 mode for Ben "ranges" work and a clone_skb != 0
49 * as a "fastpath" with a configurable number of clones after alloc's.
50 * clone_skb=0 means all packets are allocated this also means ranges time
51 * stamps etc can be used. clone_skb=100 means 1 malloc is followed by 100
52 * clones.
54 * Also moved to /proc/net/pktgen/
55 * --ro
57 * Sept 10: Fixed threading/locking. Lots of bone-headed and more clever
58 * mistakes. Also merged in DaveM's patch in the -pre6 patch.
59 * --Ben Greear <greearb@candelatech.com>
61 * Integrated to 2.5.x 021029 --Lucio Maciel (luciomaciel@zipmail.com.br)
64 * 021124 Finished major redesign and rewrite for new functionality.
65 * See Documentation/networking/pktgen.txt for how to use this.
67 * The new operation:
68 * For each CPU one thread/process is created at start. This process checks
69 * for running devices in the if_list and sends packets until count is 0 it
70 * also the thread checks the thread->control which is used for inter-process
71 * communication. controlling process "posts" operations to the threads this
72 * way.
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.
83 * --ro
85 * Fix refcount off by one if first packet fails, potential null deref,
86 * memleak 030710- KJP
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>
109 * 050103
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>
164 #include <net/udp.h>
165 #include <net/ip6_checksum.h>
166 #include <net/addrconf.h>
167 #ifdef CONFIG_XFRM
168 #include <net/xfrm.h>
169 #endif
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>
177 #include <asm/dma.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 */
213 /* Xmit modes */
214 #define M_START_XMIT 0 /* Default normal TX */
215 #define M_NETIF_RECEIVE 1 /* Inject packets into stack */
217 /* If lock -- protects updating of if_list */
218 #define if_lock(t) spin_lock(&(t->if_lock));
219 #define if_unlock(t) spin_unlock(&(t->if_lock));
221 /* Used to help with determining the pkts on receive */
222 #define PKTGEN_MAGIC 0xbe9be955
223 #define PG_PROC_DIR "pktgen"
224 #define PGCTRL "pgctrl"
226 #define MAX_CFLOWS 65536
228 #define VLAN_TAG_SIZE(x) ((x)->vlan_id == 0xffff ? 0 : 4)
229 #define SVLAN_TAG_SIZE(x) ((x)->svlan_id == 0xffff ? 0 : 4)
231 struct flow_state {
232 __be32 cur_daddr;
233 int count;
234 #ifdef CONFIG_XFRM
235 struct xfrm_state *x;
236 #endif
237 __u32 flags;
240 /* flow flag bits */
241 #define F_INIT (1<<0) /* flow has been initialized */
243 struct pktgen_dev {
245 * Try to keep frequent/infrequent used vars. separated.
247 struct proc_dir_entry *entry; /* proc file */
248 struct pktgen_thread *pg_thread;/* the owner */
249 struct list_head list; /* chaining in the thread's run-queue */
250 struct rcu_head rcu; /* freed by RCU */
252 int running; /* if false, the test will stop */
254 /* If min != max, then we will either do a linear iteration, or
255 * we will do a random selection from within the range.
257 __u32 flags;
258 int xmit_mode;
259 int min_pkt_size;
260 int max_pkt_size;
261 int pkt_overhead; /* overhead for MPLS, VLANs, IPSEC etc */
262 int nfrags;
263 int removal_mark; /* non-zero => the device is marked for
264 * removal by worker thread */
266 struct page *page;
267 u64 delay; /* nano-seconds */
269 __u64 count; /* Default No packets to send */
270 __u64 sofar; /* How many pkts we've sent so far */
271 __u64 tx_bytes; /* How many bytes we've transmitted */
272 __u64 errors; /* Errors when trying to transmit, */
274 /* runtime counters relating to clone_skb */
276 __u32 clone_count;
277 int last_ok; /* Was last skb sent?
278 * Or a failed transmit of some sort?
279 * This will keep sequence numbers in order
281 ktime_t next_tx;
282 ktime_t started_at;
283 ktime_t stopped_at;
284 u64 idle_acc; /* nano-seconds */
286 __u32 seq_num;
288 int clone_skb; /*
289 * Use multiple SKBs during packet gen.
290 * If this number is greater than 1, then
291 * that many copies of the same packet will be
292 * sent before a new packet is allocated.
293 * If you want to send 1024 identical packets
294 * before creating a new packet,
295 * set clone_skb to 1024.
298 char dst_min[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
299 char dst_max[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
300 char src_min[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
301 char src_max[IP_NAME_SZ]; /* IP, ie 1.2.3.4 */
303 struct in6_addr in6_saddr;
304 struct in6_addr in6_daddr;
305 struct in6_addr cur_in6_daddr;
306 struct in6_addr cur_in6_saddr;
307 /* For ranges */
308 struct in6_addr min_in6_daddr;
309 struct in6_addr max_in6_daddr;
310 struct in6_addr min_in6_saddr;
311 struct in6_addr max_in6_saddr;
313 /* If we're doing ranges, random or incremental, then this
314 * defines the min/max for those ranges.
316 __be32 saddr_min; /* inclusive, source IP address */
317 __be32 saddr_max; /* exclusive, source IP address */
318 __be32 daddr_min; /* inclusive, dest IP address */
319 __be32 daddr_max; /* exclusive, dest IP address */
321 __u16 udp_src_min; /* inclusive, source UDP port */
322 __u16 udp_src_max; /* exclusive, source UDP port */
323 __u16 udp_dst_min; /* inclusive, dest UDP port */
324 __u16 udp_dst_max; /* exclusive, dest UDP port */
326 /* DSCP + ECN */
327 __u8 tos; /* six MSB of (former) IPv4 TOS
328 are for dscp codepoint */
329 __u8 traffic_class; /* ditto for the (former) Traffic Class in IPv6
330 (see RFC 3260, sec. 4) */
332 /* MPLS */
333 unsigned int nr_labels; /* Depth of stack, 0 = no MPLS */
334 __be32 labels[MAX_MPLS_LABELS];
336 /* VLAN/SVLAN (802.1Q/Q-in-Q) */
337 __u8 vlan_p;
338 __u8 vlan_cfi;
339 __u16 vlan_id; /* 0xffff means no vlan tag */
341 __u8 svlan_p;
342 __u8 svlan_cfi;
343 __u16 svlan_id; /* 0xffff means no svlan tag */
345 __u32 src_mac_count; /* How many MACs to iterate through */
346 __u32 dst_mac_count; /* How many MACs to iterate through */
348 unsigned char dst_mac[ETH_ALEN];
349 unsigned char src_mac[ETH_ALEN];
351 __u32 cur_dst_mac_offset;
352 __u32 cur_src_mac_offset;
353 __be32 cur_saddr;
354 __be32 cur_daddr;
355 __u16 ip_id;
356 __u16 cur_udp_dst;
357 __u16 cur_udp_src;
358 __u16 cur_queue_map;
359 __u32 cur_pkt_size;
360 __u32 last_pkt_size;
362 __u8 hh[14];
363 /* = {
364 0x00, 0x80, 0xC8, 0x79, 0xB3, 0xCB,
366 We fill in SRC address later
367 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
368 0x08, 0x00
371 __u16 pad; /* pad out the hh struct to an even 16 bytes */
373 struct sk_buff *skb; /* skb we are to transmit next, used for when we
374 * are transmitting the same one multiple times
376 struct net_device *odev; /* The out-going device.
377 * Note that the device should have it's
378 * pg_info pointer pointing back to this
379 * device.
380 * Set when the user specifies the out-going
381 * device name (not when the inject is
382 * started as it used to do.)
384 char odevname[32];
385 struct flow_state *flows;
386 unsigned int cflows; /* Concurrent flows (config) */
387 unsigned int lflow; /* Flow length (config) */
388 unsigned int nflows; /* accumulated flows (stats) */
389 unsigned int curfl; /* current sequenced flow (state)*/
391 u16 queue_map_min;
392 u16 queue_map_max;
393 __u32 skb_priority; /* skb priority field */
394 unsigned int burst; /* number of duplicated packets to burst */
395 int node; /* Memory node */
397 #ifdef CONFIG_XFRM
398 __u8 ipsmode; /* IPSEC mode (config) */
399 __u8 ipsproto; /* IPSEC type (config) */
400 __u32 spi;
401 struct dst_entry dst;
402 struct dst_ops dstops;
403 #endif
404 char result[512];
407 struct pktgen_hdr {
408 __be32 pgh_magic;
409 __be32 seq_num;
410 __be32 tv_sec;
411 __be32 tv_usec;
415 static int pg_net_id __read_mostly;
417 struct pktgen_net {
418 struct net *net;
419 struct proc_dir_entry *proc_dir;
420 struct list_head pktgen_threads;
421 bool pktgen_exiting;
424 struct pktgen_thread {
425 spinlock_t if_lock; /* for list of devices */
426 struct list_head if_list; /* All device here */
427 struct list_head th_list;
428 struct task_struct *tsk;
429 char result[512];
431 /* Field for thread to receive "posted" events terminate,
432 stop ifs etc. */
434 u32 control;
435 int cpu;
437 wait_queue_head_t queue;
438 struct completion start_done;
439 struct pktgen_net *net;
442 #define REMOVE 1
443 #define FIND 0
445 static const char version[] =
446 "Packet Generator for packet performance testing. "
447 "Version: " VERSION "\n";
449 static int pktgen_remove_device(struct pktgen_thread *t, struct pktgen_dev *i);
450 static int pktgen_add_device(struct pktgen_thread *t, const char *ifname);
451 static struct pktgen_dev *pktgen_find_dev(struct pktgen_thread *t,
452 const char *ifname, bool exact);
453 static int pktgen_device_event(struct notifier_block *, unsigned long, void *);
454 static void pktgen_run_all_threads(struct pktgen_net *pn);
455 static void pktgen_reset_all_threads(struct pktgen_net *pn);
456 static void pktgen_stop_all_threads_ifs(struct pktgen_net *pn);
458 static void pktgen_stop(struct pktgen_thread *t);
459 static void pktgen_clear_counters(struct pktgen_dev *pkt_dev);
461 /* Module parameters, defaults. */
462 static int pg_count_d __read_mostly = 1000;
463 static int pg_delay_d __read_mostly;
464 static int pg_clone_skb_d __read_mostly;
465 static int debug __read_mostly;
467 static DEFINE_MUTEX(pktgen_thread_lock);
469 static struct notifier_block pktgen_notifier_block = {
470 .notifier_call = pktgen_device_event,
474 * /proc handling functions
478 static int pgctrl_show(struct seq_file *seq, void *v)
480 seq_puts(seq, version);
481 return 0;
484 static ssize_t pgctrl_write(struct file *file, const char __user *buf,
485 size_t count, loff_t *ppos)
487 char data[128];
488 struct pktgen_net *pn = net_generic(current->nsproxy->net_ns, pg_net_id);
490 if (!capable(CAP_NET_ADMIN))
491 return -EPERM;
493 if (count == 0)
494 return -EINVAL;
496 if (count > sizeof(data))
497 count = sizeof(data);
499 if (copy_from_user(data, buf, count))
500 return -EFAULT;
502 data[count - 1] = 0; /* Strip trailing '\n' and terminate string */
504 if (!strcmp(data, "stop"))
505 pktgen_stop_all_threads_ifs(pn);
507 else if (!strcmp(data, "start"))
508 pktgen_run_all_threads(pn);
510 else if (!strcmp(data, "reset"))
511 pktgen_reset_all_threads(pn);
513 else
514 return -EINVAL;
516 return count;
519 static int pgctrl_open(struct inode *inode, struct file *file)
521 return single_open(file, pgctrl_show, PDE_DATA(inode));
524 static const struct file_operations pktgen_fops = {
525 .owner = THIS_MODULE,
526 .open = pgctrl_open,
527 .read = seq_read,
528 .llseek = seq_lseek,
529 .write = pgctrl_write,
530 .release = single_release,
533 static int pktgen_if_show(struct seq_file *seq, void *v)
535 const struct pktgen_dev *pkt_dev = seq->private;
536 ktime_t stopped;
537 u64 idle;
539 seq_printf(seq,
540 "Params: count %llu min_pkt_size: %u max_pkt_size: %u\n",
541 (unsigned long long)pkt_dev->count, pkt_dev->min_pkt_size,
542 pkt_dev->max_pkt_size);
544 seq_printf(seq,
545 " frags: %d delay: %llu clone_skb: %d ifname: %s\n",
546 pkt_dev->nfrags, (unsigned long long) pkt_dev->delay,
547 pkt_dev->clone_skb, pkt_dev->odevname);
549 seq_printf(seq, " flows: %u flowlen: %u\n", pkt_dev->cflows,
550 pkt_dev->lflow);
552 seq_printf(seq,
553 " queue_map_min: %u queue_map_max: %u\n",
554 pkt_dev->queue_map_min,
555 pkt_dev->queue_map_max);
557 if (pkt_dev->skb_priority)
558 seq_printf(seq, " skb_priority: %u\n",
559 pkt_dev->skb_priority);
561 if (pkt_dev->flags & F_IPV6) {
562 seq_printf(seq,
563 " saddr: %pI6c min_saddr: %pI6c max_saddr: %pI6c\n"
564 " daddr: %pI6c min_daddr: %pI6c max_daddr: %pI6c\n",
565 &pkt_dev->in6_saddr,
566 &pkt_dev->min_in6_saddr, &pkt_dev->max_in6_saddr,
567 &pkt_dev->in6_daddr,
568 &pkt_dev->min_in6_daddr, &pkt_dev->max_in6_daddr);
569 } else {
570 seq_printf(seq,
571 " dst_min: %s dst_max: %s\n",
572 pkt_dev->dst_min, pkt_dev->dst_max);
573 seq_printf(seq,
574 " src_min: %s src_max: %s\n",
575 pkt_dev->src_min, pkt_dev->src_max);
578 seq_puts(seq, " src_mac: ");
580 seq_printf(seq, "%pM ",
581 is_zero_ether_addr(pkt_dev->src_mac) ?
582 pkt_dev->odev->dev_addr : pkt_dev->src_mac);
584 seq_puts(seq, "dst_mac: ");
585 seq_printf(seq, "%pM\n", pkt_dev->dst_mac);
587 seq_printf(seq,
588 " udp_src_min: %d udp_src_max: %d"
589 " udp_dst_min: %d udp_dst_max: %d\n",
590 pkt_dev->udp_src_min, pkt_dev->udp_src_max,
591 pkt_dev->udp_dst_min, pkt_dev->udp_dst_max);
593 seq_printf(seq,
594 " src_mac_count: %d dst_mac_count: %d\n",
595 pkt_dev->src_mac_count, pkt_dev->dst_mac_count);
597 if (pkt_dev->nr_labels) {
598 unsigned int i;
599 seq_puts(seq, " mpls: ");
600 for (i = 0; i < pkt_dev->nr_labels; i++)
601 seq_printf(seq, "%08x%s", ntohl(pkt_dev->labels[i]),
602 i == pkt_dev->nr_labels-1 ? "\n" : ", ");
605 if (pkt_dev->vlan_id != 0xffff)
606 seq_printf(seq, " vlan_id: %u vlan_p: %u vlan_cfi: %u\n",
607 pkt_dev->vlan_id, pkt_dev->vlan_p,
608 pkt_dev->vlan_cfi);
610 if (pkt_dev->svlan_id != 0xffff)
611 seq_printf(seq, " svlan_id: %u vlan_p: %u vlan_cfi: %u\n",
612 pkt_dev->svlan_id, pkt_dev->svlan_p,
613 pkt_dev->svlan_cfi);
615 if (pkt_dev->tos)
616 seq_printf(seq, " tos: 0x%02x\n", pkt_dev->tos);
618 if (pkt_dev->traffic_class)
619 seq_printf(seq, " traffic_class: 0x%02x\n", pkt_dev->traffic_class);
621 if (pkt_dev->burst > 1)
622 seq_printf(seq, " burst: %d\n", pkt_dev->burst);
624 if (pkt_dev->node >= 0)
625 seq_printf(seq, " node: %d\n", pkt_dev->node);
627 if (pkt_dev->xmit_mode == M_NETIF_RECEIVE)
628 seq_puts(seq, " xmit_mode: netif_receive\n");
630 seq_puts(seq, " Flags: ");
632 if (pkt_dev->flags & F_IPV6)
633 seq_puts(seq, "IPV6 ");
635 if (pkt_dev->flags & F_IPSRC_RND)
636 seq_puts(seq, "IPSRC_RND ");
638 if (pkt_dev->flags & F_IPDST_RND)
639 seq_puts(seq, "IPDST_RND ");
641 if (pkt_dev->flags & F_TXSIZE_RND)
642 seq_puts(seq, "TXSIZE_RND ");
644 if (pkt_dev->flags & F_UDPSRC_RND)
645 seq_puts(seq, "UDPSRC_RND ");
647 if (pkt_dev->flags & F_UDPDST_RND)
648 seq_puts(seq, "UDPDST_RND ");
650 if (pkt_dev->flags & F_UDPCSUM)
651 seq_puts(seq, "UDPCSUM ");
653 if (pkt_dev->flags & F_NO_TIMESTAMP)
654 seq_puts(seq, "NO_TIMESTAMP ");
656 if (pkt_dev->flags & F_MPLS_RND)
657 seq_puts(seq, "MPLS_RND ");
659 if (pkt_dev->flags & F_QUEUE_MAP_RND)
660 seq_puts(seq, "QUEUE_MAP_RND ");
662 if (pkt_dev->flags & F_QUEUE_MAP_CPU)
663 seq_puts(seq, "QUEUE_MAP_CPU ");
665 if (pkt_dev->cflows) {
666 if (pkt_dev->flags & F_FLOW_SEQ)
667 seq_puts(seq, "FLOW_SEQ "); /*in sequence flows*/
668 else
669 seq_puts(seq, "FLOW_RND ");
672 #ifdef CONFIG_XFRM
673 if (pkt_dev->flags & F_IPSEC_ON) {
674 seq_puts(seq, "IPSEC ");
675 if (pkt_dev->spi)
676 seq_printf(seq, "spi:%u", pkt_dev->spi);
678 #endif
680 if (pkt_dev->flags & F_MACSRC_RND)
681 seq_puts(seq, "MACSRC_RND ");
683 if (pkt_dev->flags & F_MACDST_RND)
684 seq_puts(seq, "MACDST_RND ");
686 if (pkt_dev->flags & F_VID_RND)
687 seq_puts(seq, "VID_RND ");
689 if (pkt_dev->flags & F_SVID_RND)
690 seq_puts(seq, "SVID_RND ");
692 if (pkt_dev->flags & F_NODE)
693 seq_puts(seq, "NODE_ALLOC ");
695 seq_puts(seq, "\n");
697 /* not really stopped, more like last-running-at */
698 stopped = pkt_dev->running ? ktime_get() : pkt_dev->stopped_at;
699 idle = pkt_dev->idle_acc;
700 do_div(idle, NSEC_PER_USEC);
702 seq_printf(seq,
703 "Current:\n pkts-sofar: %llu errors: %llu\n",
704 (unsigned long long)pkt_dev->sofar,
705 (unsigned long long)pkt_dev->errors);
707 seq_printf(seq,
708 " started: %lluus stopped: %lluus idle: %lluus\n",
709 (unsigned long long) ktime_to_us(pkt_dev->started_at),
710 (unsigned long long) ktime_to_us(stopped),
711 (unsigned long long) idle);
713 seq_printf(seq,
714 " seq_num: %d cur_dst_mac_offset: %d cur_src_mac_offset: %d\n",
715 pkt_dev->seq_num, pkt_dev->cur_dst_mac_offset,
716 pkt_dev->cur_src_mac_offset);
718 if (pkt_dev->flags & F_IPV6) {
719 seq_printf(seq, " cur_saddr: %pI6c cur_daddr: %pI6c\n",
720 &pkt_dev->cur_in6_saddr,
721 &pkt_dev->cur_in6_daddr);
722 } else
723 seq_printf(seq, " cur_saddr: %pI4 cur_daddr: %pI4\n",
724 &pkt_dev->cur_saddr, &pkt_dev->cur_daddr);
726 seq_printf(seq, " cur_udp_dst: %d cur_udp_src: %d\n",
727 pkt_dev->cur_udp_dst, pkt_dev->cur_udp_src);
729 seq_printf(seq, " cur_queue_map: %u\n", pkt_dev->cur_queue_map);
731 seq_printf(seq, " flows: %u\n", pkt_dev->nflows);
733 if (pkt_dev->result[0])
734 seq_printf(seq, "Result: %s\n", pkt_dev->result);
735 else
736 seq_puts(seq, "Result: Idle\n");
738 return 0;
742 static int hex32_arg(const char __user *user_buffer, unsigned long maxlen,
743 __u32 *num)
745 int i = 0;
746 *num = 0;
748 for (; i < maxlen; i++) {
749 int value;
750 char c;
751 *num <<= 4;
752 if (get_user(c, &user_buffer[i]))
753 return -EFAULT;
754 value = hex_to_bin(c);
755 if (value >= 0)
756 *num |= value;
757 else
758 break;
760 return i;
763 static int count_trail_chars(const char __user * user_buffer,
764 unsigned int maxlen)
766 int i;
768 for (i = 0; i < maxlen; i++) {
769 char c;
770 if (get_user(c, &user_buffer[i]))
771 return -EFAULT;
772 switch (c) {
773 case '\"':
774 case '\n':
775 case '\r':
776 case '\t':
777 case ' ':
778 case '=':
779 break;
780 default:
781 goto done;
784 done:
785 return i;
788 static long num_arg(const char __user *user_buffer, unsigned long maxlen,
789 unsigned long *num)
791 int i;
792 *num = 0;
794 for (i = 0; i < maxlen; i++) {
795 char c;
796 if (get_user(c, &user_buffer[i]))
797 return -EFAULT;
798 if ((c >= '0') && (c <= '9')) {
799 *num *= 10;
800 *num += c - '0';
801 } else
802 break;
804 return i;
807 static int strn_len(const char __user * user_buffer, unsigned int maxlen)
809 int i;
811 for (i = 0; i < maxlen; i++) {
812 char c;
813 if (get_user(c, &user_buffer[i]))
814 return -EFAULT;
815 switch (c) {
816 case '\"':
817 case '\n':
818 case '\r':
819 case '\t':
820 case ' ':
821 goto done_str;
822 default:
823 break;
826 done_str:
827 return i;
830 static ssize_t get_labels(const char __user *buffer, struct pktgen_dev *pkt_dev)
832 unsigned int n = 0;
833 char c;
834 ssize_t i = 0;
835 int len;
837 pkt_dev->nr_labels = 0;
838 do {
839 __u32 tmp;
840 len = hex32_arg(&buffer[i], 8, &tmp);
841 if (len <= 0)
842 return len;
843 pkt_dev->labels[n] = htonl(tmp);
844 if (pkt_dev->labels[n] & MPLS_STACK_BOTTOM)
845 pkt_dev->flags |= F_MPLS_RND;
846 i += len;
847 if (get_user(c, &buffer[i]))
848 return -EFAULT;
849 i++;
850 n++;
851 if (n >= MAX_MPLS_LABELS)
852 return -E2BIG;
853 } while (c == ',');
855 pkt_dev->nr_labels = n;
856 return i;
859 static ssize_t pktgen_if_write(struct file *file,
860 const char __user * user_buffer, size_t count,
861 loff_t * offset)
863 struct seq_file *seq = file->private_data;
864 struct pktgen_dev *pkt_dev = seq->private;
865 int i, max, len;
866 char name[16], valstr[32];
867 unsigned long value = 0;
868 char *pg_result = NULL;
869 int tmp = 0;
870 char buf[128];
872 pg_result = &(pkt_dev->result[0]);
874 if (count < 1) {
875 pr_warn("wrong command format\n");
876 return -EINVAL;
879 max = count;
880 tmp = count_trail_chars(user_buffer, max);
881 if (tmp < 0) {
882 pr_warn("illegal format\n");
883 return tmp;
885 i = tmp;
887 /* Read variable name */
889 len = strn_len(&user_buffer[i], sizeof(name) - 1);
890 if (len < 0)
891 return len;
893 memset(name, 0, sizeof(name));
894 if (copy_from_user(name, &user_buffer[i], len))
895 return -EFAULT;
896 i += len;
898 max = count - i;
899 len = count_trail_chars(&user_buffer[i], max);
900 if (len < 0)
901 return len;
903 i += len;
905 if (debug) {
906 size_t copy = min_t(size_t, count, 1023);
907 char tb[copy + 1];
908 if (copy_from_user(tb, user_buffer, copy))
909 return -EFAULT;
910 tb[copy] = 0;
911 pr_debug("%s,%lu buffer -:%s:-\n",
912 name, (unsigned long)count, tb);
915 if (!strcmp(name, "min_pkt_size")) {
916 len = num_arg(&user_buffer[i], 10, &value);
917 if (len < 0)
918 return len;
920 i += len;
921 if (value < 14 + 20 + 8)
922 value = 14 + 20 + 8;
923 if (value != pkt_dev->min_pkt_size) {
924 pkt_dev->min_pkt_size = value;
925 pkt_dev->cur_pkt_size = value;
927 sprintf(pg_result, "OK: min_pkt_size=%u",
928 pkt_dev->min_pkt_size);
929 return count;
932 if (!strcmp(name, "max_pkt_size")) {
933 len = num_arg(&user_buffer[i], 10, &value);
934 if (len < 0)
935 return len;
937 i += len;
938 if (value < 14 + 20 + 8)
939 value = 14 + 20 + 8;
940 if (value != pkt_dev->max_pkt_size) {
941 pkt_dev->max_pkt_size = value;
942 pkt_dev->cur_pkt_size = value;
944 sprintf(pg_result, "OK: max_pkt_size=%u",
945 pkt_dev->max_pkt_size);
946 return count;
949 /* Shortcut for min = max */
951 if (!strcmp(name, "pkt_size")) {
952 len = num_arg(&user_buffer[i], 10, &value);
953 if (len < 0)
954 return len;
956 i += len;
957 if (value < 14 + 20 + 8)
958 value = 14 + 20 + 8;
959 if (value != pkt_dev->min_pkt_size) {
960 pkt_dev->min_pkt_size = value;
961 pkt_dev->max_pkt_size = value;
962 pkt_dev->cur_pkt_size = value;
964 sprintf(pg_result, "OK: pkt_size=%u", pkt_dev->min_pkt_size);
965 return count;
968 if (!strcmp(name, "debug")) {
969 len = num_arg(&user_buffer[i], 10, &value);
970 if (len < 0)
971 return len;
973 i += len;
974 debug = value;
975 sprintf(pg_result, "OK: debug=%u", debug);
976 return count;
979 if (!strcmp(name, "frags")) {
980 len = num_arg(&user_buffer[i], 10, &value);
981 if (len < 0)
982 return len;
984 i += len;
985 pkt_dev->nfrags = value;
986 sprintf(pg_result, "OK: frags=%u", pkt_dev->nfrags);
987 return count;
989 if (!strcmp(name, "delay")) {
990 len = num_arg(&user_buffer[i], 10, &value);
991 if (len < 0)
992 return len;
994 i += len;
995 if (value == 0x7FFFFFFF)
996 pkt_dev->delay = ULLONG_MAX;
997 else
998 pkt_dev->delay = (u64)value;
1000 sprintf(pg_result, "OK: delay=%llu",
1001 (unsigned long long) pkt_dev->delay);
1002 return count;
1004 if (!strcmp(name, "rate")) {
1005 len = num_arg(&user_buffer[i], 10, &value);
1006 if (len < 0)
1007 return len;
1009 i += len;
1010 if (!value)
1011 return len;
1012 pkt_dev->delay = pkt_dev->min_pkt_size*8*NSEC_PER_USEC/value;
1013 if (debug)
1014 pr_info("Delay set at: %llu ns\n", pkt_dev->delay);
1016 sprintf(pg_result, "OK: rate=%lu", value);
1017 return count;
1019 if (!strcmp(name, "ratep")) {
1020 len = num_arg(&user_buffer[i], 10, &value);
1021 if (len < 0)
1022 return len;
1024 i += len;
1025 if (!value)
1026 return len;
1027 pkt_dev->delay = NSEC_PER_SEC/value;
1028 if (debug)
1029 pr_info("Delay set at: %llu ns\n", pkt_dev->delay);
1031 sprintf(pg_result, "OK: rate=%lu", value);
1032 return count;
1034 if (!strcmp(name, "udp_src_min")) {
1035 len = num_arg(&user_buffer[i], 10, &value);
1036 if (len < 0)
1037 return len;
1039 i += len;
1040 if (value != pkt_dev->udp_src_min) {
1041 pkt_dev->udp_src_min = value;
1042 pkt_dev->cur_udp_src = value;
1044 sprintf(pg_result, "OK: udp_src_min=%u", pkt_dev->udp_src_min);
1045 return count;
1047 if (!strcmp(name, "udp_dst_min")) {
1048 len = num_arg(&user_buffer[i], 10, &value);
1049 if (len < 0)
1050 return len;
1052 i += len;
1053 if (value != pkt_dev->udp_dst_min) {
1054 pkt_dev->udp_dst_min = value;
1055 pkt_dev->cur_udp_dst = value;
1057 sprintf(pg_result, "OK: udp_dst_min=%u", pkt_dev->udp_dst_min);
1058 return count;
1060 if (!strcmp(name, "udp_src_max")) {
1061 len = num_arg(&user_buffer[i], 10, &value);
1062 if (len < 0)
1063 return len;
1065 i += len;
1066 if (value != pkt_dev->udp_src_max) {
1067 pkt_dev->udp_src_max = value;
1068 pkt_dev->cur_udp_src = value;
1070 sprintf(pg_result, "OK: udp_src_max=%u", pkt_dev->udp_src_max);
1071 return count;
1073 if (!strcmp(name, "udp_dst_max")) {
1074 len = num_arg(&user_buffer[i], 10, &value);
1075 if (len < 0)
1076 return len;
1078 i += len;
1079 if (value != pkt_dev->udp_dst_max) {
1080 pkt_dev->udp_dst_max = value;
1081 pkt_dev->cur_udp_dst = value;
1083 sprintf(pg_result, "OK: udp_dst_max=%u", pkt_dev->udp_dst_max);
1084 return count;
1086 if (!strcmp(name, "clone_skb")) {
1087 len = num_arg(&user_buffer[i], 10, &value);
1088 if (len < 0)
1089 return len;
1090 if ((value > 0) &&
1091 ((pkt_dev->xmit_mode == M_NETIF_RECEIVE) ||
1092 !(pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING)))
1093 return -ENOTSUPP;
1094 i += len;
1095 pkt_dev->clone_skb = value;
1097 sprintf(pg_result, "OK: clone_skb=%d", pkt_dev->clone_skb);
1098 return count;
1100 if (!strcmp(name, "count")) {
1101 len = num_arg(&user_buffer[i], 10, &value);
1102 if (len < 0)
1103 return len;
1105 i += len;
1106 pkt_dev->count = value;
1107 sprintf(pg_result, "OK: count=%llu",
1108 (unsigned long long)pkt_dev->count);
1109 return count;
1111 if (!strcmp(name, "src_mac_count")) {
1112 len = num_arg(&user_buffer[i], 10, &value);
1113 if (len < 0)
1114 return len;
1116 i += len;
1117 if (pkt_dev->src_mac_count != value) {
1118 pkt_dev->src_mac_count = value;
1119 pkt_dev->cur_src_mac_offset = 0;
1121 sprintf(pg_result, "OK: src_mac_count=%d",
1122 pkt_dev->src_mac_count);
1123 return count;
1125 if (!strcmp(name, "dst_mac_count")) {
1126 len = num_arg(&user_buffer[i], 10, &value);
1127 if (len < 0)
1128 return len;
1130 i += len;
1131 if (pkt_dev->dst_mac_count != value) {
1132 pkt_dev->dst_mac_count = value;
1133 pkt_dev->cur_dst_mac_offset = 0;
1135 sprintf(pg_result, "OK: dst_mac_count=%d",
1136 pkt_dev->dst_mac_count);
1137 return count;
1139 if (!strcmp(name, "burst")) {
1140 len = num_arg(&user_buffer[i], 10, &value);
1141 if (len < 0)
1142 return len;
1144 i += len;
1145 if ((value > 1) && (pkt_dev->xmit_mode == M_START_XMIT) &&
1146 (!(pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING)))
1147 return -ENOTSUPP;
1148 pkt_dev->burst = value < 1 ? 1 : value;
1149 sprintf(pg_result, "OK: burst=%d", pkt_dev->burst);
1150 return count;
1152 if (!strcmp(name, "node")) {
1153 len = num_arg(&user_buffer[i], 10, &value);
1154 if (len < 0)
1155 return len;
1157 i += len;
1159 if (node_possible(value)) {
1160 pkt_dev->node = value;
1161 sprintf(pg_result, "OK: node=%d", pkt_dev->node);
1162 if (pkt_dev->page) {
1163 put_page(pkt_dev->page);
1164 pkt_dev->page = NULL;
1167 else
1168 sprintf(pg_result, "ERROR: node not possible");
1169 return count;
1171 if (!strcmp(name, "xmit_mode")) {
1172 char f[32];
1174 memset(f, 0, 32);
1175 len = strn_len(&user_buffer[i], sizeof(f) - 1);
1176 if (len < 0)
1177 return len;
1179 if (copy_from_user(f, &user_buffer[i], len))
1180 return -EFAULT;
1181 i += len;
1183 if (strcmp(f, "start_xmit") == 0) {
1184 pkt_dev->xmit_mode = M_START_XMIT;
1185 } else if (strcmp(f, "netif_receive") == 0) {
1186 /* clone_skb set earlier, not supported in this mode */
1187 if (pkt_dev->clone_skb > 0)
1188 return -ENOTSUPP;
1190 pkt_dev->xmit_mode = M_NETIF_RECEIVE;
1192 /* make sure new packet is allocated every time
1193 * pktgen_xmit() is called
1195 pkt_dev->last_ok = 1;
1197 /* override clone_skb if user passed default value
1198 * at module loading time
1200 pkt_dev->clone_skb = 0;
1201 } else {
1202 sprintf(pg_result,
1203 "xmit_mode -:%s:- unknown\nAvailable modes: %s",
1204 f, "start_xmit, netif_receive\n");
1205 return count;
1207 sprintf(pg_result, "OK: xmit_mode=%s", f);
1208 return count;
1210 if (!strcmp(name, "flag")) {
1211 char f[32];
1212 memset(f, 0, 32);
1213 len = strn_len(&user_buffer[i], sizeof(f) - 1);
1214 if (len < 0)
1215 return len;
1217 if (copy_from_user(f, &user_buffer[i], len))
1218 return -EFAULT;
1219 i += len;
1220 if (strcmp(f, "IPSRC_RND") == 0)
1221 pkt_dev->flags |= F_IPSRC_RND;
1223 else if (strcmp(f, "!IPSRC_RND") == 0)
1224 pkt_dev->flags &= ~F_IPSRC_RND;
1226 else if (strcmp(f, "TXSIZE_RND") == 0)
1227 pkt_dev->flags |= F_TXSIZE_RND;
1229 else if (strcmp(f, "!TXSIZE_RND") == 0)
1230 pkt_dev->flags &= ~F_TXSIZE_RND;
1232 else if (strcmp(f, "IPDST_RND") == 0)
1233 pkt_dev->flags |= F_IPDST_RND;
1235 else if (strcmp(f, "!IPDST_RND") == 0)
1236 pkt_dev->flags &= ~F_IPDST_RND;
1238 else if (strcmp(f, "UDPSRC_RND") == 0)
1239 pkt_dev->flags |= F_UDPSRC_RND;
1241 else if (strcmp(f, "!UDPSRC_RND") == 0)
1242 pkt_dev->flags &= ~F_UDPSRC_RND;
1244 else if (strcmp(f, "UDPDST_RND") == 0)
1245 pkt_dev->flags |= F_UDPDST_RND;
1247 else if (strcmp(f, "!UDPDST_RND") == 0)
1248 pkt_dev->flags &= ~F_UDPDST_RND;
1250 else if (strcmp(f, "MACSRC_RND") == 0)
1251 pkt_dev->flags |= F_MACSRC_RND;
1253 else if (strcmp(f, "!MACSRC_RND") == 0)
1254 pkt_dev->flags &= ~F_MACSRC_RND;
1256 else if (strcmp(f, "MACDST_RND") == 0)
1257 pkt_dev->flags |= F_MACDST_RND;
1259 else if (strcmp(f, "!MACDST_RND") == 0)
1260 pkt_dev->flags &= ~F_MACDST_RND;
1262 else if (strcmp(f, "MPLS_RND") == 0)
1263 pkt_dev->flags |= F_MPLS_RND;
1265 else if (strcmp(f, "!MPLS_RND") == 0)
1266 pkt_dev->flags &= ~F_MPLS_RND;
1268 else if (strcmp(f, "VID_RND") == 0)
1269 pkt_dev->flags |= F_VID_RND;
1271 else if (strcmp(f, "!VID_RND") == 0)
1272 pkt_dev->flags &= ~F_VID_RND;
1274 else if (strcmp(f, "SVID_RND") == 0)
1275 pkt_dev->flags |= F_SVID_RND;
1277 else if (strcmp(f, "!SVID_RND") == 0)
1278 pkt_dev->flags &= ~F_SVID_RND;
1280 else if (strcmp(f, "FLOW_SEQ") == 0)
1281 pkt_dev->flags |= F_FLOW_SEQ;
1283 else if (strcmp(f, "QUEUE_MAP_RND") == 0)
1284 pkt_dev->flags |= F_QUEUE_MAP_RND;
1286 else if (strcmp(f, "!QUEUE_MAP_RND") == 0)
1287 pkt_dev->flags &= ~F_QUEUE_MAP_RND;
1289 else if (strcmp(f, "QUEUE_MAP_CPU") == 0)
1290 pkt_dev->flags |= F_QUEUE_MAP_CPU;
1292 else if (strcmp(f, "!QUEUE_MAP_CPU") == 0)
1293 pkt_dev->flags &= ~F_QUEUE_MAP_CPU;
1294 #ifdef CONFIG_XFRM
1295 else if (strcmp(f, "IPSEC") == 0)
1296 pkt_dev->flags |= F_IPSEC_ON;
1297 #endif
1299 else if (strcmp(f, "!IPV6") == 0)
1300 pkt_dev->flags &= ~F_IPV6;
1302 else if (strcmp(f, "NODE_ALLOC") == 0)
1303 pkt_dev->flags |= F_NODE;
1305 else if (strcmp(f, "!NODE_ALLOC") == 0)
1306 pkt_dev->flags &= ~F_NODE;
1308 else if (strcmp(f, "UDPCSUM") == 0)
1309 pkt_dev->flags |= F_UDPCSUM;
1311 else if (strcmp(f, "!UDPCSUM") == 0)
1312 pkt_dev->flags &= ~F_UDPCSUM;
1314 else if (strcmp(f, "NO_TIMESTAMP") == 0)
1315 pkt_dev->flags |= F_NO_TIMESTAMP;
1317 else if (strcmp(f, "!NO_TIMESTAMP") == 0)
1318 pkt_dev->flags &= ~F_NO_TIMESTAMP;
1320 else {
1321 sprintf(pg_result,
1322 "Flag -:%s:- unknown\nAvailable flags, (prepend ! to un-set flag):\n%s",
1324 "IPSRC_RND, IPDST_RND, UDPSRC_RND, UDPDST_RND, "
1325 "MACSRC_RND, MACDST_RND, TXSIZE_RND, IPV6, "
1326 "MPLS_RND, VID_RND, SVID_RND, FLOW_SEQ, "
1327 "QUEUE_MAP_RND, QUEUE_MAP_CPU, UDPCSUM, "
1328 "NO_TIMESTAMP, "
1329 #ifdef CONFIG_XFRM
1330 "IPSEC, "
1331 #endif
1332 "NODE_ALLOC\n");
1333 return count;
1335 sprintf(pg_result, "OK: flags=0x%x", pkt_dev->flags);
1336 return count;
1338 if (!strcmp(name, "dst_min") || !strcmp(name, "dst")) {
1339 len = strn_len(&user_buffer[i], sizeof(pkt_dev->dst_min) - 1);
1340 if (len < 0)
1341 return len;
1343 if (copy_from_user(buf, &user_buffer[i], len))
1344 return -EFAULT;
1345 buf[len] = 0;
1346 if (strcmp(buf, pkt_dev->dst_min) != 0) {
1347 memset(pkt_dev->dst_min, 0, sizeof(pkt_dev->dst_min));
1348 strncpy(pkt_dev->dst_min, buf, len);
1349 pkt_dev->daddr_min = in_aton(pkt_dev->dst_min);
1350 pkt_dev->cur_daddr = pkt_dev->daddr_min;
1352 if (debug)
1353 pr_debug("dst_min set to: %s\n", pkt_dev->dst_min);
1354 i += len;
1355 sprintf(pg_result, "OK: dst_min=%s", pkt_dev->dst_min);
1356 return count;
1358 if (!strcmp(name, "dst_max")) {
1359 len = strn_len(&user_buffer[i], sizeof(pkt_dev->dst_max) - 1);
1360 if (len < 0)
1361 return len;
1364 if (copy_from_user(buf, &user_buffer[i], len))
1365 return -EFAULT;
1367 buf[len] = 0;
1368 if (strcmp(buf, pkt_dev->dst_max) != 0) {
1369 memset(pkt_dev->dst_max, 0, sizeof(pkt_dev->dst_max));
1370 strncpy(pkt_dev->dst_max, buf, len);
1371 pkt_dev->daddr_max = in_aton(pkt_dev->dst_max);
1372 pkt_dev->cur_daddr = pkt_dev->daddr_max;
1374 if (debug)
1375 pr_debug("dst_max set to: %s\n", pkt_dev->dst_max);
1376 i += len;
1377 sprintf(pg_result, "OK: dst_max=%s", pkt_dev->dst_max);
1378 return count;
1380 if (!strcmp(name, "dst6")) {
1381 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1382 if (len < 0)
1383 return len;
1385 pkt_dev->flags |= F_IPV6;
1387 if (copy_from_user(buf, &user_buffer[i], len))
1388 return -EFAULT;
1389 buf[len] = 0;
1391 in6_pton(buf, -1, pkt_dev->in6_daddr.s6_addr, -1, NULL);
1392 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->in6_daddr);
1394 pkt_dev->cur_in6_daddr = pkt_dev->in6_daddr;
1396 if (debug)
1397 pr_debug("dst6 set to: %s\n", buf);
1399 i += len;
1400 sprintf(pg_result, "OK: dst6=%s", buf);
1401 return count;
1403 if (!strcmp(name, "dst6_min")) {
1404 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1405 if (len < 0)
1406 return len;
1408 pkt_dev->flags |= F_IPV6;
1410 if (copy_from_user(buf, &user_buffer[i], len))
1411 return -EFAULT;
1412 buf[len] = 0;
1414 in6_pton(buf, -1, pkt_dev->min_in6_daddr.s6_addr, -1, NULL);
1415 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->min_in6_daddr);
1417 pkt_dev->cur_in6_daddr = pkt_dev->min_in6_daddr;
1418 if (debug)
1419 pr_debug("dst6_min set to: %s\n", buf);
1421 i += len;
1422 sprintf(pg_result, "OK: dst6_min=%s", buf);
1423 return count;
1425 if (!strcmp(name, "dst6_max")) {
1426 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1427 if (len < 0)
1428 return len;
1430 pkt_dev->flags |= F_IPV6;
1432 if (copy_from_user(buf, &user_buffer[i], len))
1433 return -EFAULT;
1434 buf[len] = 0;
1436 in6_pton(buf, -1, pkt_dev->max_in6_daddr.s6_addr, -1, NULL);
1437 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->max_in6_daddr);
1439 if (debug)
1440 pr_debug("dst6_max set to: %s\n", buf);
1442 i += len;
1443 sprintf(pg_result, "OK: dst6_max=%s", buf);
1444 return count;
1446 if (!strcmp(name, "src6")) {
1447 len = strn_len(&user_buffer[i], sizeof(buf) - 1);
1448 if (len < 0)
1449 return len;
1451 pkt_dev->flags |= F_IPV6;
1453 if (copy_from_user(buf, &user_buffer[i], len))
1454 return -EFAULT;
1455 buf[len] = 0;
1457 in6_pton(buf, -1, pkt_dev->in6_saddr.s6_addr, -1, NULL);
1458 snprintf(buf, sizeof(buf), "%pI6c", &pkt_dev->in6_saddr);
1460 pkt_dev->cur_in6_saddr = pkt_dev->in6_saddr;
1462 if (debug)
1463 pr_debug("src6 set to: %s\n", buf);
1465 i += len;
1466 sprintf(pg_result, "OK: src6=%s", buf);
1467 return count;
1469 if (!strcmp(name, "src_min")) {
1470 len = strn_len(&user_buffer[i], sizeof(pkt_dev->src_min) - 1);
1471 if (len < 0)
1472 return len;
1474 if (copy_from_user(buf, &user_buffer[i], len))
1475 return -EFAULT;
1476 buf[len] = 0;
1477 if (strcmp(buf, pkt_dev->src_min) != 0) {
1478 memset(pkt_dev->src_min, 0, sizeof(pkt_dev->src_min));
1479 strncpy(pkt_dev->src_min, buf, len);
1480 pkt_dev->saddr_min = in_aton(pkt_dev->src_min);
1481 pkt_dev->cur_saddr = pkt_dev->saddr_min;
1483 if (debug)
1484 pr_debug("src_min set to: %s\n", pkt_dev->src_min);
1485 i += len;
1486 sprintf(pg_result, "OK: src_min=%s", pkt_dev->src_min);
1487 return count;
1489 if (!strcmp(name, "src_max")) {
1490 len = strn_len(&user_buffer[i], sizeof(pkt_dev->src_max) - 1);
1491 if (len < 0)
1492 return len;
1494 if (copy_from_user(buf, &user_buffer[i], len))
1495 return -EFAULT;
1496 buf[len] = 0;
1497 if (strcmp(buf, pkt_dev->src_max) != 0) {
1498 memset(pkt_dev->src_max, 0, sizeof(pkt_dev->src_max));
1499 strncpy(pkt_dev->src_max, buf, len);
1500 pkt_dev->saddr_max = in_aton(pkt_dev->src_max);
1501 pkt_dev->cur_saddr = pkt_dev->saddr_max;
1503 if (debug)
1504 pr_debug("src_max set to: %s\n", pkt_dev->src_max);
1505 i += len;
1506 sprintf(pg_result, "OK: src_max=%s", pkt_dev->src_max);
1507 return count;
1509 if (!strcmp(name, "dst_mac")) {
1510 len = strn_len(&user_buffer[i], sizeof(valstr) - 1);
1511 if (len < 0)
1512 return len;
1514 memset(valstr, 0, sizeof(valstr));
1515 if (copy_from_user(valstr, &user_buffer[i], len))
1516 return -EFAULT;
1518 if (!mac_pton(valstr, pkt_dev->dst_mac))
1519 return -EINVAL;
1520 /* Set up Dest MAC */
1521 ether_addr_copy(&pkt_dev->hh[0], pkt_dev->dst_mac);
1523 sprintf(pg_result, "OK: dstmac %pM", pkt_dev->dst_mac);
1524 return count;
1526 if (!strcmp(name, "src_mac")) {
1527 len = strn_len(&user_buffer[i], sizeof(valstr) - 1);
1528 if (len < 0)
1529 return len;
1531 memset(valstr, 0, sizeof(valstr));
1532 if (copy_from_user(valstr, &user_buffer[i], len))
1533 return -EFAULT;
1535 if (!mac_pton(valstr, pkt_dev->src_mac))
1536 return -EINVAL;
1537 /* Set up Src MAC */
1538 ether_addr_copy(&pkt_dev->hh[6], pkt_dev->src_mac);
1540 sprintf(pg_result, "OK: srcmac %pM", pkt_dev->src_mac);
1541 return count;
1544 if (!strcmp(name, "clear_counters")) {
1545 pktgen_clear_counters(pkt_dev);
1546 sprintf(pg_result, "OK: Clearing counters.\n");
1547 return count;
1550 if (!strcmp(name, "flows")) {
1551 len = num_arg(&user_buffer[i], 10, &value);
1552 if (len < 0)
1553 return len;
1555 i += len;
1556 if (value > MAX_CFLOWS)
1557 value = MAX_CFLOWS;
1559 pkt_dev->cflows = value;
1560 sprintf(pg_result, "OK: flows=%u", pkt_dev->cflows);
1561 return count;
1563 #ifdef CONFIG_XFRM
1564 if (!strcmp(name, "spi")) {
1565 len = num_arg(&user_buffer[i], 10, &value);
1566 if (len < 0)
1567 return len;
1569 i += len;
1570 pkt_dev->spi = value;
1571 sprintf(pg_result, "OK: spi=%u", pkt_dev->spi);
1572 return count;
1574 #endif
1575 if (!strcmp(name, "flowlen")) {
1576 len = num_arg(&user_buffer[i], 10, &value);
1577 if (len < 0)
1578 return len;
1580 i += len;
1581 pkt_dev->lflow = value;
1582 sprintf(pg_result, "OK: flowlen=%u", pkt_dev->lflow);
1583 return count;
1586 if (!strcmp(name, "queue_map_min")) {
1587 len = num_arg(&user_buffer[i], 5, &value);
1588 if (len < 0)
1589 return len;
1591 i += len;
1592 pkt_dev->queue_map_min = value;
1593 sprintf(pg_result, "OK: queue_map_min=%u", pkt_dev->queue_map_min);
1594 return count;
1597 if (!strcmp(name, "queue_map_max")) {
1598 len = num_arg(&user_buffer[i], 5, &value);
1599 if (len < 0)
1600 return len;
1602 i += len;
1603 pkt_dev->queue_map_max = value;
1604 sprintf(pg_result, "OK: queue_map_max=%u", pkt_dev->queue_map_max);
1605 return count;
1608 if (!strcmp(name, "mpls")) {
1609 unsigned int n, cnt;
1611 len = get_labels(&user_buffer[i], pkt_dev);
1612 if (len < 0)
1613 return len;
1614 i += len;
1615 cnt = sprintf(pg_result, "OK: mpls=");
1616 for (n = 0; n < pkt_dev->nr_labels; n++)
1617 cnt += sprintf(pg_result + cnt,
1618 "%08x%s", ntohl(pkt_dev->labels[n]),
1619 n == pkt_dev->nr_labels-1 ? "" : ",");
1621 if (pkt_dev->nr_labels && pkt_dev->vlan_id != 0xffff) {
1622 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */
1623 pkt_dev->svlan_id = 0xffff;
1625 if (debug)
1626 pr_debug("VLAN/SVLAN auto turned off\n");
1628 return count;
1631 if (!strcmp(name, "vlan_id")) {
1632 len = num_arg(&user_buffer[i], 4, &value);
1633 if (len < 0)
1634 return len;
1636 i += len;
1637 if (value <= 4095) {
1638 pkt_dev->vlan_id = value; /* turn on VLAN */
1640 if (debug)
1641 pr_debug("VLAN turned on\n");
1643 if (debug && pkt_dev->nr_labels)
1644 pr_debug("MPLS auto turned off\n");
1646 pkt_dev->nr_labels = 0; /* turn off MPLS */
1647 sprintf(pg_result, "OK: vlan_id=%u", pkt_dev->vlan_id);
1648 } else {
1649 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */
1650 pkt_dev->svlan_id = 0xffff;
1652 if (debug)
1653 pr_debug("VLAN/SVLAN turned off\n");
1655 return count;
1658 if (!strcmp(name, "vlan_p")) {
1659 len = num_arg(&user_buffer[i], 1, &value);
1660 if (len < 0)
1661 return len;
1663 i += len;
1664 if ((value <= 7) && (pkt_dev->vlan_id != 0xffff)) {
1665 pkt_dev->vlan_p = value;
1666 sprintf(pg_result, "OK: vlan_p=%u", pkt_dev->vlan_p);
1667 } else {
1668 sprintf(pg_result, "ERROR: vlan_p must be 0-7");
1670 return count;
1673 if (!strcmp(name, "vlan_cfi")) {
1674 len = num_arg(&user_buffer[i], 1, &value);
1675 if (len < 0)
1676 return len;
1678 i += len;
1679 if ((value <= 1) && (pkt_dev->vlan_id != 0xffff)) {
1680 pkt_dev->vlan_cfi = value;
1681 sprintf(pg_result, "OK: vlan_cfi=%u", pkt_dev->vlan_cfi);
1682 } else {
1683 sprintf(pg_result, "ERROR: vlan_cfi must be 0-1");
1685 return count;
1688 if (!strcmp(name, "svlan_id")) {
1689 len = num_arg(&user_buffer[i], 4, &value);
1690 if (len < 0)
1691 return len;
1693 i += len;
1694 if ((value <= 4095) && ((pkt_dev->vlan_id != 0xffff))) {
1695 pkt_dev->svlan_id = value; /* turn on SVLAN */
1697 if (debug)
1698 pr_debug("SVLAN turned on\n");
1700 if (debug && pkt_dev->nr_labels)
1701 pr_debug("MPLS auto turned off\n");
1703 pkt_dev->nr_labels = 0; /* turn off MPLS */
1704 sprintf(pg_result, "OK: svlan_id=%u", pkt_dev->svlan_id);
1705 } else {
1706 pkt_dev->vlan_id = 0xffff; /* turn off VLAN/SVLAN */
1707 pkt_dev->svlan_id = 0xffff;
1709 if (debug)
1710 pr_debug("VLAN/SVLAN turned off\n");
1712 return count;
1715 if (!strcmp(name, "svlan_p")) {
1716 len = num_arg(&user_buffer[i], 1, &value);
1717 if (len < 0)
1718 return len;
1720 i += len;
1721 if ((value <= 7) && (pkt_dev->svlan_id != 0xffff)) {
1722 pkt_dev->svlan_p = value;
1723 sprintf(pg_result, "OK: svlan_p=%u", pkt_dev->svlan_p);
1724 } else {
1725 sprintf(pg_result, "ERROR: svlan_p must be 0-7");
1727 return count;
1730 if (!strcmp(name, "svlan_cfi")) {
1731 len = num_arg(&user_buffer[i], 1, &value);
1732 if (len < 0)
1733 return len;
1735 i += len;
1736 if ((value <= 1) && (pkt_dev->svlan_id != 0xffff)) {
1737 pkt_dev->svlan_cfi = value;
1738 sprintf(pg_result, "OK: svlan_cfi=%u", pkt_dev->svlan_cfi);
1739 } else {
1740 sprintf(pg_result, "ERROR: svlan_cfi must be 0-1");
1742 return count;
1745 if (!strcmp(name, "tos")) {
1746 __u32 tmp_value = 0;
1747 len = hex32_arg(&user_buffer[i], 2, &tmp_value);
1748 if (len < 0)
1749 return len;
1751 i += len;
1752 if (len == 2) {
1753 pkt_dev->tos = tmp_value;
1754 sprintf(pg_result, "OK: tos=0x%02x", pkt_dev->tos);
1755 } else {
1756 sprintf(pg_result, "ERROR: tos must be 00-ff");
1758 return count;
1761 if (!strcmp(name, "traffic_class")) {
1762 __u32 tmp_value = 0;
1763 len = hex32_arg(&user_buffer[i], 2, &tmp_value);
1764 if (len < 0)
1765 return len;
1767 i += len;
1768 if (len == 2) {
1769 pkt_dev->traffic_class = tmp_value;
1770 sprintf(pg_result, "OK: traffic_class=0x%02x", pkt_dev->traffic_class);
1771 } else {
1772 sprintf(pg_result, "ERROR: traffic_class must be 00-ff");
1774 return count;
1777 if (!strcmp(name, "skb_priority")) {
1778 len = num_arg(&user_buffer[i], 9, &value);
1779 if (len < 0)
1780 return len;
1782 i += len;
1783 pkt_dev->skb_priority = value;
1784 sprintf(pg_result, "OK: skb_priority=%i",
1785 pkt_dev->skb_priority);
1786 return count;
1789 sprintf(pkt_dev->result, "No such parameter \"%s\"", name);
1790 return -EINVAL;
1793 static int pktgen_if_open(struct inode *inode, struct file *file)
1795 return single_open(file, pktgen_if_show, PDE_DATA(inode));
1798 static const struct file_operations pktgen_if_fops = {
1799 .owner = THIS_MODULE,
1800 .open = pktgen_if_open,
1801 .read = seq_read,
1802 .llseek = seq_lseek,
1803 .write = pktgen_if_write,
1804 .release = single_release,
1807 static int pktgen_thread_show(struct seq_file *seq, void *v)
1809 struct pktgen_thread *t = seq->private;
1810 const struct pktgen_dev *pkt_dev;
1812 BUG_ON(!t);
1814 seq_puts(seq, "Running: ");
1816 rcu_read_lock();
1817 list_for_each_entry_rcu(pkt_dev, &t->if_list, list)
1818 if (pkt_dev->running)
1819 seq_printf(seq, "%s ", pkt_dev->odevname);
1821 seq_puts(seq, "\nStopped: ");
1823 list_for_each_entry_rcu(pkt_dev, &t->if_list, list)
1824 if (!pkt_dev->running)
1825 seq_printf(seq, "%s ", pkt_dev->odevname);
1827 if (t->result[0])
1828 seq_printf(seq, "\nResult: %s\n", t->result);
1829 else
1830 seq_puts(seq, "\nResult: NA\n");
1832 rcu_read_unlock();
1834 return 0;
1837 static ssize_t pktgen_thread_write(struct file *file,
1838 const char __user * user_buffer,
1839 size_t count, loff_t * offset)
1841 struct seq_file *seq = file->private_data;
1842 struct pktgen_thread *t = seq->private;
1843 int i, max, len, ret;
1844 char name[40];
1845 char *pg_result;
1847 if (count < 1) {
1848 // sprintf(pg_result, "Wrong command format");
1849 return -EINVAL;
1852 max = count;
1853 len = count_trail_chars(user_buffer, max);
1854 if (len < 0)
1855 return len;
1857 i = len;
1859 /* Read variable name */
1861 len = strn_len(&user_buffer[i], sizeof(name) - 1);
1862 if (len < 0)
1863 return len;
1865 memset(name, 0, sizeof(name));
1866 if (copy_from_user(name, &user_buffer[i], len))
1867 return -EFAULT;
1868 i += len;
1870 max = count - i;
1871 len = count_trail_chars(&user_buffer[i], max);
1872 if (len < 0)
1873 return len;
1875 i += len;
1877 if (debug)
1878 pr_debug("t=%s, count=%lu\n", name, (unsigned long)count);
1880 if (!t) {
1881 pr_err("ERROR: No thread\n");
1882 ret = -EINVAL;
1883 goto out;
1886 pg_result = &(t->result[0]);
1888 if (!strcmp(name, "add_device")) {
1889 char f[32];
1890 memset(f, 0, 32);
1891 len = strn_len(&user_buffer[i], sizeof(f) - 1);
1892 if (len < 0) {
1893 ret = len;
1894 goto out;
1896 if (copy_from_user(f, &user_buffer[i], len))
1897 return -EFAULT;
1898 i += len;
1899 mutex_lock(&pktgen_thread_lock);
1900 ret = pktgen_add_device(t, f);
1901 mutex_unlock(&pktgen_thread_lock);
1902 if (!ret) {
1903 ret = count;
1904 sprintf(pg_result, "OK: add_device=%s", f);
1905 } else
1906 sprintf(pg_result, "ERROR: can not add device %s", f);
1907 goto out;
1910 if (!strcmp(name, "rem_device_all")) {
1911 mutex_lock(&pktgen_thread_lock);
1912 t->control |= T_REMDEVALL;
1913 mutex_unlock(&pktgen_thread_lock);
1914 schedule_timeout_interruptible(msecs_to_jiffies(125)); /* Propagate thread->control */
1915 ret = count;
1916 sprintf(pg_result, "OK: rem_device_all");
1917 goto out;
1920 if (!strcmp(name, "max_before_softirq")) {
1921 sprintf(pg_result, "OK: Note! max_before_softirq is obsoleted -- Do not use");
1922 ret = count;
1923 goto out;
1926 ret = -EINVAL;
1927 out:
1928 return ret;
1931 static int pktgen_thread_open(struct inode *inode, struct file *file)
1933 return single_open(file, pktgen_thread_show, PDE_DATA(inode));
1936 static const struct file_operations pktgen_thread_fops = {
1937 .owner = THIS_MODULE,
1938 .open = pktgen_thread_open,
1939 .read = seq_read,
1940 .llseek = seq_lseek,
1941 .write = pktgen_thread_write,
1942 .release = single_release,
1945 /* Think find or remove for NN */
1946 static struct pktgen_dev *__pktgen_NN_threads(const struct pktgen_net *pn,
1947 const char *ifname, int remove)
1949 struct pktgen_thread *t;
1950 struct pktgen_dev *pkt_dev = NULL;
1951 bool exact = (remove == FIND);
1953 list_for_each_entry(t, &pn->pktgen_threads, th_list) {
1954 pkt_dev = pktgen_find_dev(t, ifname, exact);
1955 if (pkt_dev) {
1956 if (remove) {
1957 pkt_dev->removal_mark = 1;
1958 t->control |= T_REMDEV;
1960 break;
1963 return pkt_dev;
1967 * mark a device for removal
1969 static void pktgen_mark_device(const struct pktgen_net *pn, const char *ifname)
1971 struct pktgen_dev *pkt_dev = NULL;
1972 const int max_tries = 10, msec_per_try = 125;
1973 int i = 0;
1975 mutex_lock(&pktgen_thread_lock);
1976 pr_debug("%s: marking %s for removal\n", __func__, ifname);
1978 while (1) {
1980 pkt_dev = __pktgen_NN_threads(pn, ifname, REMOVE);
1981 if (pkt_dev == NULL)
1982 break; /* success */
1984 mutex_unlock(&pktgen_thread_lock);
1985 pr_debug("%s: waiting for %s to disappear....\n",
1986 __func__, ifname);
1987 schedule_timeout_interruptible(msecs_to_jiffies(msec_per_try));
1988 mutex_lock(&pktgen_thread_lock);
1990 if (++i >= max_tries) {
1991 pr_err("%s: timed out after waiting %d msec for device %s to be removed\n",
1992 __func__, msec_per_try * i, ifname);
1993 break;
1998 mutex_unlock(&pktgen_thread_lock);
2001 static void pktgen_change_name(const struct pktgen_net *pn, struct net_device *dev)
2003 struct pktgen_thread *t;
2005 list_for_each_entry(t, &pn->pktgen_threads, th_list) {
2006 struct pktgen_dev *pkt_dev;
2008 rcu_read_lock();
2009 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) {
2010 if (pkt_dev->odev != dev)
2011 continue;
2013 proc_remove(pkt_dev->entry);
2015 pkt_dev->entry = proc_create_data(dev->name, 0600,
2016 pn->proc_dir,
2017 &pktgen_if_fops,
2018 pkt_dev);
2019 if (!pkt_dev->entry)
2020 pr_err("can't move proc entry for '%s'\n",
2021 dev->name);
2022 break;
2024 rcu_read_unlock();
2028 static int pktgen_device_event(struct notifier_block *unused,
2029 unsigned long event, void *ptr)
2031 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2032 struct pktgen_net *pn = net_generic(dev_net(dev), pg_net_id);
2034 if (pn->pktgen_exiting)
2035 return NOTIFY_DONE;
2037 /* It is OK that we do not hold the group lock right now,
2038 * as we run under the RTNL lock.
2041 switch (event) {
2042 case NETDEV_CHANGENAME:
2043 pktgen_change_name(pn, dev);
2044 break;
2046 case NETDEV_UNREGISTER:
2047 pktgen_mark_device(pn, dev->name);
2048 break;
2051 return NOTIFY_DONE;
2054 static struct net_device *pktgen_dev_get_by_name(const struct pktgen_net *pn,
2055 struct pktgen_dev *pkt_dev,
2056 const char *ifname)
2058 char b[IFNAMSIZ+5];
2059 int i;
2061 for (i = 0; ifname[i] != '@'; i++) {
2062 if (i == IFNAMSIZ)
2063 break;
2065 b[i] = ifname[i];
2067 b[i] = 0;
2069 return dev_get_by_name(pn->net, b);
2073 /* Associate pktgen_dev with a device. */
2075 static int pktgen_setup_dev(const struct pktgen_net *pn,
2076 struct pktgen_dev *pkt_dev, const char *ifname)
2078 struct net_device *odev;
2079 int err;
2081 /* Clean old setups */
2082 if (pkt_dev->odev) {
2083 dev_put(pkt_dev->odev);
2084 pkt_dev->odev = NULL;
2087 odev = pktgen_dev_get_by_name(pn, pkt_dev, ifname);
2088 if (!odev) {
2089 pr_err("no such netdevice: \"%s\"\n", ifname);
2090 return -ENODEV;
2093 if (odev->type != ARPHRD_ETHER) {
2094 pr_err("not an ethernet device: \"%s\"\n", ifname);
2095 err = -EINVAL;
2096 } else if (!netif_running(odev)) {
2097 pr_err("device is down: \"%s\"\n", ifname);
2098 err = -ENETDOWN;
2099 } else {
2100 pkt_dev->odev = odev;
2101 return 0;
2104 dev_put(odev);
2105 return err;
2108 /* Read pkt_dev from the interface and set up internal pktgen_dev
2109 * structure to have the right information to create/send packets
2111 static void pktgen_setup_inject(struct pktgen_dev *pkt_dev)
2113 int ntxq;
2115 if (!pkt_dev->odev) {
2116 pr_err("ERROR: pkt_dev->odev == NULL in setup_inject\n");
2117 sprintf(pkt_dev->result,
2118 "ERROR: pkt_dev->odev == NULL in setup_inject.\n");
2119 return;
2122 /* make sure that we don't pick a non-existing transmit queue */
2123 ntxq = pkt_dev->odev->real_num_tx_queues;
2125 if (ntxq <= pkt_dev->queue_map_min) {
2126 pr_warn("WARNING: Requested queue_map_min (zero-based) (%d) exceeds valid range [0 - %d] for (%d) queues on %s, resetting\n",
2127 pkt_dev->queue_map_min, (ntxq ?: 1) - 1, ntxq,
2128 pkt_dev->odevname);
2129 pkt_dev->queue_map_min = (ntxq ?: 1) - 1;
2131 if (pkt_dev->queue_map_max >= ntxq) {
2132 pr_warn("WARNING: Requested queue_map_max (zero-based) (%d) exceeds valid range [0 - %d] for (%d) queues on %s, resetting\n",
2133 pkt_dev->queue_map_max, (ntxq ?: 1) - 1, ntxq,
2134 pkt_dev->odevname);
2135 pkt_dev->queue_map_max = (ntxq ?: 1) - 1;
2138 /* Default to the interface's mac if not explicitly set. */
2140 if (is_zero_ether_addr(pkt_dev->src_mac))
2141 ether_addr_copy(&(pkt_dev->hh[6]), pkt_dev->odev->dev_addr);
2143 /* Set up Dest MAC */
2144 ether_addr_copy(&(pkt_dev->hh[0]), pkt_dev->dst_mac);
2146 if (pkt_dev->flags & F_IPV6) {
2147 int i, set = 0, err = 1;
2148 struct inet6_dev *idev;
2150 if (pkt_dev->min_pkt_size == 0) {
2151 pkt_dev->min_pkt_size = 14 + sizeof(struct ipv6hdr)
2152 + sizeof(struct udphdr)
2153 + sizeof(struct pktgen_hdr)
2154 + pkt_dev->pkt_overhead;
2157 for (i = 0; i < IN6_ADDR_HSIZE; i++)
2158 if (pkt_dev->cur_in6_saddr.s6_addr[i]) {
2159 set = 1;
2160 break;
2163 if (!set) {
2166 * Use linklevel address if unconfigured.
2168 * use ipv6_get_lladdr if/when it's get exported
2171 rcu_read_lock();
2172 idev = __in6_dev_get(pkt_dev->odev);
2173 if (idev) {
2174 struct inet6_ifaddr *ifp;
2176 read_lock_bh(&idev->lock);
2177 list_for_each_entry(ifp, &idev->addr_list, if_list) {
2178 if ((ifp->scope & IFA_LINK) &&
2179 !(ifp->flags & IFA_F_TENTATIVE)) {
2180 pkt_dev->cur_in6_saddr = ifp->addr;
2181 err = 0;
2182 break;
2185 read_unlock_bh(&idev->lock);
2187 rcu_read_unlock();
2188 if (err)
2189 pr_err("ERROR: IPv6 link address not available\n");
2191 } else {
2192 if (pkt_dev->min_pkt_size == 0) {
2193 pkt_dev->min_pkt_size = 14 + sizeof(struct iphdr)
2194 + sizeof(struct udphdr)
2195 + sizeof(struct pktgen_hdr)
2196 + pkt_dev->pkt_overhead;
2199 pkt_dev->saddr_min = 0;
2200 pkt_dev->saddr_max = 0;
2201 if (strlen(pkt_dev->src_min) == 0) {
2203 struct in_device *in_dev;
2205 rcu_read_lock();
2206 in_dev = __in_dev_get_rcu(pkt_dev->odev);
2207 if (in_dev) {
2208 if (in_dev->ifa_list) {
2209 pkt_dev->saddr_min =
2210 in_dev->ifa_list->ifa_address;
2211 pkt_dev->saddr_max = pkt_dev->saddr_min;
2214 rcu_read_unlock();
2215 } else {
2216 pkt_dev->saddr_min = in_aton(pkt_dev->src_min);
2217 pkt_dev->saddr_max = in_aton(pkt_dev->src_max);
2220 pkt_dev->daddr_min = in_aton(pkt_dev->dst_min);
2221 pkt_dev->daddr_max = in_aton(pkt_dev->dst_max);
2223 /* Initialize current values. */
2224 pkt_dev->cur_pkt_size = pkt_dev->min_pkt_size;
2225 if (pkt_dev->min_pkt_size > pkt_dev->max_pkt_size)
2226 pkt_dev->max_pkt_size = pkt_dev->min_pkt_size;
2228 pkt_dev->cur_dst_mac_offset = 0;
2229 pkt_dev->cur_src_mac_offset = 0;
2230 pkt_dev->cur_saddr = pkt_dev->saddr_min;
2231 pkt_dev->cur_daddr = pkt_dev->daddr_min;
2232 pkt_dev->cur_udp_dst = pkt_dev->udp_dst_min;
2233 pkt_dev->cur_udp_src = pkt_dev->udp_src_min;
2234 pkt_dev->nflows = 0;
2238 static void spin(struct pktgen_dev *pkt_dev, ktime_t spin_until)
2240 ktime_t start_time, end_time;
2241 s64 remaining;
2242 struct hrtimer_sleeper t;
2244 hrtimer_init_on_stack(&t.timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
2245 hrtimer_set_expires(&t.timer, spin_until);
2247 remaining = ktime_to_ns(hrtimer_expires_remaining(&t.timer));
2248 if (remaining <= 0)
2249 goto out;
2251 start_time = ktime_get();
2252 if (remaining < 100000) {
2253 /* for small delays (<100us), just loop until limit is reached */
2254 do {
2255 end_time = ktime_get();
2256 } while (ktime_compare(end_time, spin_until) < 0);
2257 } else {
2258 /* see do_nanosleep */
2259 hrtimer_init_sleeper(&t, current);
2260 do {
2261 set_current_state(TASK_INTERRUPTIBLE);
2262 hrtimer_start_expires(&t.timer, HRTIMER_MODE_ABS);
2264 if (likely(t.task))
2265 schedule();
2267 hrtimer_cancel(&t.timer);
2268 } while (t.task && pkt_dev->running && !signal_pending(current));
2269 __set_current_state(TASK_RUNNING);
2270 end_time = ktime_get();
2273 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(end_time, start_time));
2274 out:
2275 pkt_dev->next_tx = ktime_add_ns(spin_until, pkt_dev->delay);
2276 destroy_hrtimer_on_stack(&t.timer);
2279 static inline void set_pkt_overhead(struct pktgen_dev *pkt_dev)
2281 pkt_dev->pkt_overhead = LL_RESERVED_SPACE(pkt_dev->odev);
2282 pkt_dev->pkt_overhead += pkt_dev->nr_labels*sizeof(u32);
2283 pkt_dev->pkt_overhead += VLAN_TAG_SIZE(pkt_dev);
2284 pkt_dev->pkt_overhead += SVLAN_TAG_SIZE(pkt_dev);
2287 static inline int f_seen(const struct pktgen_dev *pkt_dev, int flow)
2289 return !!(pkt_dev->flows[flow].flags & F_INIT);
2292 static inline int f_pick(struct pktgen_dev *pkt_dev)
2294 int flow = pkt_dev->curfl;
2296 if (pkt_dev->flags & F_FLOW_SEQ) {
2297 if (pkt_dev->flows[flow].count >= pkt_dev->lflow) {
2298 /* reset time */
2299 pkt_dev->flows[flow].count = 0;
2300 pkt_dev->flows[flow].flags = 0;
2301 pkt_dev->curfl += 1;
2302 if (pkt_dev->curfl >= pkt_dev->cflows)
2303 pkt_dev->curfl = 0; /*reset */
2305 } else {
2306 flow = prandom_u32() % pkt_dev->cflows;
2307 pkt_dev->curfl = flow;
2309 if (pkt_dev->flows[flow].count > pkt_dev->lflow) {
2310 pkt_dev->flows[flow].count = 0;
2311 pkt_dev->flows[flow].flags = 0;
2315 return pkt_dev->curfl;
2319 #ifdef CONFIG_XFRM
2320 /* If there was already an IPSEC SA, we keep it as is, else
2321 * we go look for it ...
2323 #define DUMMY_MARK 0
2324 static void get_ipsec_sa(struct pktgen_dev *pkt_dev, int flow)
2326 struct xfrm_state *x = pkt_dev->flows[flow].x;
2327 struct pktgen_net *pn = net_generic(dev_net(pkt_dev->odev), pg_net_id);
2328 if (!x) {
2330 if (pkt_dev->spi) {
2331 /* We need as quick as possible to find the right SA
2332 * Searching with minimum criteria to archieve this.
2334 x = xfrm_state_lookup_byspi(pn->net, htonl(pkt_dev->spi), AF_INET);
2335 } else {
2336 /* slow path: we dont already have xfrm_state */
2337 x = xfrm_stateonly_find(pn->net, DUMMY_MARK,
2338 (xfrm_address_t *)&pkt_dev->cur_daddr,
2339 (xfrm_address_t *)&pkt_dev->cur_saddr,
2340 AF_INET,
2341 pkt_dev->ipsmode,
2342 pkt_dev->ipsproto, 0);
2344 if (x) {
2345 pkt_dev->flows[flow].x = x;
2346 set_pkt_overhead(pkt_dev);
2347 pkt_dev->pkt_overhead += x->props.header_len;
2352 #endif
2353 static void set_cur_queue_map(struct pktgen_dev *pkt_dev)
2356 if (pkt_dev->flags & F_QUEUE_MAP_CPU)
2357 pkt_dev->cur_queue_map = smp_processor_id();
2359 else if (pkt_dev->queue_map_min <= pkt_dev->queue_map_max) {
2360 __u16 t;
2361 if (pkt_dev->flags & F_QUEUE_MAP_RND) {
2362 t = prandom_u32() %
2363 (pkt_dev->queue_map_max -
2364 pkt_dev->queue_map_min + 1)
2365 + pkt_dev->queue_map_min;
2366 } else {
2367 t = pkt_dev->cur_queue_map + 1;
2368 if (t > pkt_dev->queue_map_max)
2369 t = pkt_dev->queue_map_min;
2371 pkt_dev->cur_queue_map = t;
2373 pkt_dev->cur_queue_map = pkt_dev->cur_queue_map % pkt_dev->odev->real_num_tx_queues;
2376 /* Increment/randomize headers according to flags and current values
2377 * for IP src/dest, UDP src/dst port, MAC-Addr src/dst
2379 static void mod_cur_headers(struct pktgen_dev *pkt_dev)
2381 __u32 imn;
2382 __u32 imx;
2383 int flow = 0;
2385 if (pkt_dev->cflows)
2386 flow = f_pick(pkt_dev);
2388 /* Deal with source MAC */
2389 if (pkt_dev->src_mac_count > 1) {
2390 __u32 mc;
2391 __u32 tmp;
2393 if (pkt_dev->flags & F_MACSRC_RND)
2394 mc = prandom_u32() % pkt_dev->src_mac_count;
2395 else {
2396 mc = pkt_dev->cur_src_mac_offset++;
2397 if (pkt_dev->cur_src_mac_offset >=
2398 pkt_dev->src_mac_count)
2399 pkt_dev->cur_src_mac_offset = 0;
2402 tmp = pkt_dev->src_mac[5] + (mc & 0xFF);
2403 pkt_dev->hh[11] = tmp;
2404 tmp = (pkt_dev->src_mac[4] + ((mc >> 8) & 0xFF) + (tmp >> 8));
2405 pkt_dev->hh[10] = tmp;
2406 tmp = (pkt_dev->src_mac[3] + ((mc >> 16) & 0xFF) + (tmp >> 8));
2407 pkt_dev->hh[9] = tmp;
2408 tmp = (pkt_dev->src_mac[2] + ((mc >> 24) & 0xFF) + (tmp >> 8));
2409 pkt_dev->hh[8] = tmp;
2410 tmp = (pkt_dev->src_mac[1] + (tmp >> 8));
2411 pkt_dev->hh[7] = tmp;
2414 /* Deal with Destination MAC */
2415 if (pkt_dev->dst_mac_count > 1) {
2416 __u32 mc;
2417 __u32 tmp;
2419 if (pkt_dev->flags & F_MACDST_RND)
2420 mc = prandom_u32() % pkt_dev->dst_mac_count;
2422 else {
2423 mc = pkt_dev->cur_dst_mac_offset++;
2424 if (pkt_dev->cur_dst_mac_offset >=
2425 pkt_dev->dst_mac_count) {
2426 pkt_dev->cur_dst_mac_offset = 0;
2430 tmp = pkt_dev->dst_mac[5] + (mc & 0xFF);
2431 pkt_dev->hh[5] = tmp;
2432 tmp = (pkt_dev->dst_mac[4] + ((mc >> 8) & 0xFF) + (tmp >> 8));
2433 pkt_dev->hh[4] = tmp;
2434 tmp = (pkt_dev->dst_mac[3] + ((mc >> 16) & 0xFF) + (tmp >> 8));
2435 pkt_dev->hh[3] = tmp;
2436 tmp = (pkt_dev->dst_mac[2] + ((mc >> 24) & 0xFF) + (tmp >> 8));
2437 pkt_dev->hh[2] = tmp;
2438 tmp = (pkt_dev->dst_mac[1] + (tmp >> 8));
2439 pkt_dev->hh[1] = tmp;
2442 if (pkt_dev->flags & F_MPLS_RND) {
2443 unsigned int i;
2444 for (i = 0; i < pkt_dev->nr_labels; i++)
2445 if (pkt_dev->labels[i] & MPLS_STACK_BOTTOM)
2446 pkt_dev->labels[i] = MPLS_STACK_BOTTOM |
2447 ((__force __be32)prandom_u32() &
2448 htonl(0x000fffff));
2451 if ((pkt_dev->flags & F_VID_RND) && (pkt_dev->vlan_id != 0xffff)) {
2452 pkt_dev->vlan_id = prandom_u32() & (4096 - 1);
2455 if ((pkt_dev->flags & F_SVID_RND) && (pkt_dev->svlan_id != 0xffff)) {
2456 pkt_dev->svlan_id = prandom_u32() & (4096 - 1);
2459 if (pkt_dev->udp_src_min < pkt_dev->udp_src_max) {
2460 if (pkt_dev->flags & F_UDPSRC_RND)
2461 pkt_dev->cur_udp_src = prandom_u32() %
2462 (pkt_dev->udp_src_max - pkt_dev->udp_src_min)
2463 + pkt_dev->udp_src_min;
2465 else {
2466 pkt_dev->cur_udp_src++;
2467 if (pkt_dev->cur_udp_src >= pkt_dev->udp_src_max)
2468 pkt_dev->cur_udp_src = pkt_dev->udp_src_min;
2472 if (pkt_dev->udp_dst_min < pkt_dev->udp_dst_max) {
2473 if (pkt_dev->flags & F_UDPDST_RND) {
2474 pkt_dev->cur_udp_dst = prandom_u32() %
2475 (pkt_dev->udp_dst_max - pkt_dev->udp_dst_min)
2476 + pkt_dev->udp_dst_min;
2477 } else {
2478 pkt_dev->cur_udp_dst++;
2479 if (pkt_dev->cur_udp_dst >= pkt_dev->udp_dst_max)
2480 pkt_dev->cur_udp_dst = pkt_dev->udp_dst_min;
2484 if (!(pkt_dev->flags & F_IPV6)) {
2486 imn = ntohl(pkt_dev->saddr_min);
2487 imx = ntohl(pkt_dev->saddr_max);
2488 if (imn < imx) {
2489 __u32 t;
2490 if (pkt_dev->flags & F_IPSRC_RND)
2491 t = prandom_u32() % (imx - imn) + imn;
2492 else {
2493 t = ntohl(pkt_dev->cur_saddr);
2494 t++;
2495 if (t > imx)
2496 t = imn;
2499 pkt_dev->cur_saddr = htonl(t);
2502 if (pkt_dev->cflows && f_seen(pkt_dev, flow)) {
2503 pkt_dev->cur_daddr = pkt_dev->flows[flow].cur_daddr;
2504 } else {
2505 imn = ntohl(pkt_dev->daddr_min);
2506 imx = ntohl(pkt_dev->daddr_max);
2507 if (imn < imx) {
2508 __u32 t;
2509 __be32 s;
2510 if (pkt_dev->flags & F_IPDST_RND) {
2512 do {
2513 t = prandom_u32() %
2514 (imx - imn) + imn;
2515 s = htonl(t);
2516 } while (ipv4_is_loopback(s) ||
2517 ipv4_is_multicast(s) ||
2518 ipv4_is_lbcast(s) ||
2519 ipv4_is_zeronet(s) ||
2520 ipv4_is_local_multicast(s));
2521 pkt_dev->cur_daddr = s;
2522 } else {
2523 t = ntohl(pkt_dev->cur_daddr);
2524 t++;
2525 if (t > imx) {
2526 t = imn;
2528 pkt_dev->cur_daddr = htonl(t);
2531 if (pkt_dev->cflows) {
2532 pkt_dev->flows[flow].flags |= F_INIT;
2533 pkt_dev->flows[flow].cur_daddr =
2534 pkt_dev->cur_daddr;
2535 #ifdef CONFIG_XFRM
2536 if (pkt_dev->flags & F_IPSEC_ON)
2537 get_ipsec_sa(pkt_dev, flow);
2538 #endif
2539 pkt_dev->nflows++;
2542 } else { /* IPV6 * */
2544 if (!ipv6_addr_any(&pkt_dev->min_in6_daddr)) {
2545 int i;
2547 /* Only random destinations yet */
2549 for (i = 0; i < 4; i++) {
2550 pkt_dev->cur_in6_daddr.s6_addr32[i] =
2551 (((__force __be32)prandom_u32() |
2552 pkt_dev->min_in6_daddr.s6_addr32[i]) &
2553 pkt_dev->max_in6_daddr.s6_addr32[i]);
2558 if (pkt_dev->min_pkt_size < pkt_dev->max_pkt_size) {
2559 __u32 t;
2560 if (pkt_dev->flags & F_TXSIZE_RND) {
2561 t = prandom_u32() %
2562 (pkt_dev->max_pkt_size - pkt_dev->min_pkt_size)
2563 + pkt_dev->min_pkt_size;
2564 } else {
2565 t = pkt_dev->cur_pkt_size + 1;
2566 if (t > pkt_dev->max_pkt_size)
2567 t = pkt_dev->min_pkt_size;
2569 pkt_dev->cur_pkt_size = t;
2572 set_cur_queue_map(pkt_dev);
2574 pkt_dev->flows[flow].count++;
2578 #ifdef CONFIG_XFRM
2579 static u32 pktgen_dst_metrics[RTAX_MAX + 1] = {
2581 [RTAX_HOPLIMIT] = 0x5, /* Set a static hoplimit */
2584 static int pktgen_output_ipsec(struct sk_buff *skb, struct pktgen_dev *pkt_dev)
2586 struct xfrm_state *x = pkt_dev->flows[pkt_dev->curfl].x;
2587 int err = 0;
2588 struct net *net = dev_net(pkt_dev->odev);
2590 if (!x)
2591 return 0;
2592 /* XXX: we dont support tunnel mode for now until
2593 * we resolve the dst issue */
2594 if ((x->props.mode != XFRM_MODE_TRANSPORT) && (pkt_dev->spi == 0))
2595 return 0;
2597 /* But when user specify an valid SPI, transformation
2598 * supports both transport/tunnel mode + ESP/AH type.
2600 if ((x->props.mode == XFRM_MODE_TUNNEL) && (pkt_dev->spi != 0))
2601 skb->_skb_refdst = (unsigned long)&pkt_dev->dst | SKB_DST_NOREF;
2603 rcu_read_lock_bh();
2604 err = x->outer_mode->output(x, skb);
2605 rcu_read_unlock_bh();
2606 if (err) {
2607 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEMODEERROR);
2608 goto error;
2610 err = x->type->output(x, skb);
2611 if (err) {
2612 XFRM_INC_STATS(net, LINUX_MIB_XFRMOUTSTATEPROTOERROR);
2613 goto error;
2615 spin_lock_bh(&x->lock);
2616 x->curlft.bytes += skb->len;
2617 x->curlft.packets++;
2618 spin_unlock_bh(&x->lock);
2619 error:
2620 return err;
2623 static void free_SAs(struct pktgen_dev *pkt_dev)
2625 if (pkt_dev->cflows) {
2626 /* let go of the SAs if we have them */
2627 int i;
2628 for (i = 0; i < pkt_dev->cflows; i++) {
2629 struct xfrm_state *x = pkt_dev->flows[i].x;
2630 if (x) {
2631 xfrm_state_put(x);
2632 pkt_dev->flows[i].x = NULL;
2638 static int process_ipsec(struct pktgen_dev *pkt_dev,
2639 struct sk_buff *skb, __be16 protocol)
2641 if (pkt_dev->flags & F_IPSEC_ON) {
2642 struct xfrm_state *x = pkt_dev->flows[pkt_dev->curfl].x;
2643 int nhead = 0;
2644 if (x) {
2645 struct ethhdr *eth;
2646 struct iphdr *iph;
2647 int ret;
2649 nhead = x->props.header_len - skb_headroom(skb);
2650 if (nhead > 0) {
2651 ret = pskb_expand_head(skb, nhead, 0, GFP_ATOMIC);
2652 if (ret < 0) {
2653 pr_err("Error expanding ipsec packet %d\n",
2654 ret);
2655 goto err;
2659 /* ipsec is not expecting ll header */
2660 skb_pull(skb, ETH_HLEN);
2661 ret = pktgen_output_ipsec(skb, pkt_dev);
2662 if (ret) {
2663 pr_err("Error creating ipsec packet %d\n", ret);
2664 goto err;
2666 /* restore ll */
2667 eth = (struct ethhdr *)skb_push(skb, ETH_HLEN);
2668 memcpy(eth, pkt_dev->hh, 2 * ETH_ALEN);
2669 eth->h_proto = protocol;
2671 /* Update IPv4 header len as well as checksum value */
2672 iph = ip_hdr(skb);
2673 iph->tot_len = htons(skb->len - ETH_HLEN);
2674 ip_send_check(iph);
2677 return 1;
2678 err:
2679 kfree_skb(skb);
2680 return 0;
2682 #endif
2684 static void mpls_push(__be32 *mpls, struct pktgen_dev *pkt_dev)
2686 unsigned int i;
2687 for (i = 0; i < pkt_dev->nr_labels; i++)
2688 *mpls++ = pkt_dev->labels[i] & ~MPLS_STACK_BOTTOM;
2690 mpls--;
2691 *mpls |= MPLS_STACK_BOTTOM;
2694 static inline __be16 build_tci(unsigned int id, unsigned int cfi,
2695 unsigned int prio)
2697 return htons(id | (cfi << 12) | (prio << 13));
2700 static void pktgen_finalize_skb(struct pktgen_dev *pkt_dev, struct sk_buff *skb,
2701 int datalen)
2703 struct timeval timestamp;
2704 struct pktgen_hdr *pgh;
2706 pgh = (struct pktgen_hdr *)skb_put(skb, sizeof(*pgh));
2707 datalen -= sizeof(*pgh);
2709 if (pkt_dev->nfrags <= 0) {
2710 memset(skb_put(skb, datalen), 0, datalen);
2711 } else {
2712 int frags = pkt_dev->nfrags;
2713 int i, len;
2714 int frag_len;
2717 if (frags > MAX_SKB_FRAGS)
2718 frags = MAX_SKB_FRAGS;
2719 len = datalen - frags * PAGE_SIZE;
2720 if (len > 0) {
2721 memset(skb_put(skb, len), 0, len);
2722 datalen = frags * PAGE_SIZE;
2725 i = 0;
2726 frag_len = (datalen/frags) < PAGE_SIZE ?
2727 (datalen/frags) : PAGE_SIZE;
2728 while (datalen > 0) {
2729 if (unlikely(!pkt_dev->page)) {
2730 int node = numa_node_id();
2732 if (pkt_dev->node >= 0 && (pkt_dev->flags & F_NODE))
2733 node = pkt_dev->node;
2734 pkt_dev->page = alloc_pages_node(node, GFP_KERNEL | __GFP_ZERO, 0);
2735 if (!pkt_dev->page)
2736 break;
2738 get_page(pkt_dev->page);
2739 skb_frag_set_page(skb, i, pkt_dev->page);
2740 skb_shinfo(skb)->frags[i].page_offset = 0;
2741 /*last fragment, fill rest of data*/
2742 if (i == (frags - 1))
2743 skb_frag_size_set(&skb_shinfo(skb)->frags[i],
2744 (datalen < PAGE_SIZE ? datalen : PAGE_SIZE));
2745 else
2746 skb_frag_size_set(&skb_shinfo(skb)->frags[i], frag_len);
2747 datalen -= skb_frag_size(&skb_shinfo(skb)->frags[i]);
2748 skb->len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2749 skb->data_len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2750 i++;
2751 skb_shinfo(skb)->nr_frags = i;
2755 /* Stamp the time, and sequence number,
2756 * convert them to network byte order
2758 pgh->pgh_magic = htonl(PKTGEN_MAGIC);
2759 pgh->seq_num = htonl(pkt_dev->seq_num);
2761 if (pkt_dev->flags & F_NO_TIMESTAMP) {
2762 pgh->tv_sec = 0;
2763 pgh->tv_usec = 0;
2764 } else {
2765 do_gettimeofday(&timestamp);
2766 pgh->tv_sec = htonl(timestamp.tv_sec);
2767 pgh->tv_usec = htonl(timestamp.tv_usec);
2771 static struct sk_buff *pktgen_alloc_skb(struct net_device *dev,
2772 struct pktgen_dev *pkt_dev,
2773 unsigned int extralen)
2775 struct sk_buff *skb = NULL;
2776 unsigned int size = pkt_dev->cur_pkt_size + 64 + extralen +
2777 pkt_dev->pkt_overhead;
2779 if (pkt_dev->flags & F_NODE) {
2780 int node = pkt_dev->node >= 0 ? pkt_dev->node : numa_node_id();
2782 skb = __alloc_skb(NET_SKB_PAD + size, GFP_NOWAIT, 0, node);
2783 if (likely(skb)) {
2784 skb_reserve(skb, NET_SKB_PAD);
2785 skb->dev = dev;
2787 } else {
2788 skb = __netdev_alloc_skb(dev, size, GFP_NOWAIT);
2791 if (likely(skb))
2792 skb_reserve(skb, LL_RESERVED_SPACE(dev));
2794 return skb;
2797 static struct sk_buff *fill_packet_ipv4(struct net_device *odev,
2798 struct pktgen_dev *pkt_dev)
2800 struct sk_buff *skb = NULL;
2801 __u8 *eth;
2802 struct udphdr *udph;
2803 int datalen, iplen;
2804 struct iphdr *iph;
2805 __be16 protocol = htons(ETH_P_IP);
2806 __be32 *mpls;
2807 __be16 *vlan_tci = NULL; /* Encapsulates priority and VLAN ID */
2808 __be16 *vlan_encapsulated_proto = NULL; /* packet type ID field (or len) for VLAN tag */
2809 __be16 *svlan_tci = NULL; /* Encapsulates priority and SVLAN ID */
2810 __be16 *svlan_encapsulated_proto = NULL; /* packet type ID field (or len) for SVLAN tag */
2811 u16 queue_map;
2813 if (pkt_dev->nr_labels)
2814 protocol = htons(ETH_P_MPLS_UC);
2816 if (pkt_dev->vlan_id != 0xffff)
2817 protocol = htons(ETH_P_8021Q);
2819 /* Update any of the values, used when we're incrementing various
2820 * fields.
2822 mod_cur_headers(pkt_dev);
2823 queue_map = pkt_dev->cur_queue_map;
2825 datalen = (odev->hard_header_len + 16) & ~0xf;
2827 skb = pktgen_alloc_skb(odev, pkt_dev, datalen);
2828 if (!skb) {
2829 sprintf(pkt_dev->result, "No memory");
2830 return NULL;
2833 prefetchw(skb->data);
2834 skb_reserve(skb, datalen);
2836 /* Reserve for ethernet and IP header */
2837 eth = (__u8 *) skb_push(skb, 14);
2838 mpls = (__be32 *)skb_put(skb, pkt_dev->nr_labels*sizeof(__u32));
2839 if (pkt_dev->nr_labels)
2840 mpls_push(mpls, pkt_dev);
2842 if (pkt_dev->vlan_id != 0xffff) {
2843 if (pkt_dev->svlan_id != 0xffff) {
2844 svlan_tci = (__be16 *)skb_put(skb, sizeof(__be16));
2845 *svlan_tci = build_tci(pkt_dev->svlan_id,
2846 pkt_dev->svlan_cfi,
2847 pkt_dev->svlan_p);
2848 svlan_encapsulated_proto = (__be16 *)skb_put(skb, sizeof(__be16));
2849 *svlan_encapsulated_proto = htons(ETH_P_8021Q);
2851 vlan_tci = (__be16 *)skb_put(skb, sizeof(__be16));
2852 *vlan_tci = build_tci(pkt_dev->vlan_id,
2853 pkt_dev->vlan_cfi,
2854 pkt_dev->vlan_p);
2855 vlan_encapsulated_proto = (__be16 *)skb_put(skb, sizeof(__be16));
2856 *vlan_encapsulated_proto = htons(ETH_P_IP);
2859 skb_reset_mac_header(skb);
2860 skb_set_network_header(skb, skb->len);
2861 iph = (struct iphdr *) skb_put(skb, sizeof(struct iphdr));
2863 skb_set_transport_header(skb, skb->len);
2864 udph = (struct udphdr *) skb_put(skb, sizeof(struct udphdr));
2865 skb_set_queue_mapping(skb, queue_map);
2866 skb->priority = pkt_dev->skb_priority;
2868 memcpy(eth, pkt_dev->hh, 12);
2869 *(__be16 *) & eth[12] = protocol;
2871 /* Eth + IPh + UDPh + mpls */
2872 datalen = pkt_dev->cur_pkt_size - 14 - 20 - 8 -
2873 pkt_dev->pkt_overhead;
2874 if (datalen < 0 || datalen < sizeof(struct pktgen_hdr))
2875 datalen = sizeof(struct pktgen_hdr);
2877 udph->source = htons(pkt_dev->cur_udp_src);
2878 udph->dest = htons(pkt_dev->cur_udp_dst);
2879 udph->len = htons(datalen + 8); /* DATA + udphdr */
2880 udph->check = 0;
2882 iph->ihl = 5;
2883 iph->version = 4;
2884 iph->ttl = 32;
2885 iph->tos = pkt_dev->tos;
2886 iph->protocol = IPPROTO_UDP; /* UDP */
2887 iph->saddr = pkt_dev->cur_saddr;
2888 iph->daddr = pkt_dev->cur_daddr;
2889 iph->id = htons(pkt_dev->ip_id);
2890 pkt_dev->ip_id++;
2891 iph->frag_off = 0;
2892 iplen = 20 + 8 + datalen;
2893 iph->tot_len = htons(iplen);
2894 ip_send_check(iph);
2895 skb->protocol = protocol;
2896 skb->dev = odev;
2897 skb->pkt_type = PACKET_HOST;
2899 pktgen_finalize_skb(pkt_dev, skb, datalen);
2901 if (!(pkt_dev->flags & F_UDPCSUM)) {
2902 skb->ip_summed = CHECKSUM_NONE;
2903 } else if (odev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM)) {
2904 skb->ip_summed = CHECKSUM_PARTIAL;
2905 skb->csum = 0;
2906 udp4_hwcsum(skb, iph->saddr, iph->daddr);
2907 } else {
2908 __wsum csum = skb_checksum(skb, skb_transport_offset(skb), datalen + 8, 0);
2910 /* add protocol-dependent pseudo-header */
2911 udph->check = csum_tcpudp_magic(iph->saddr, iph->daddr,
2912 datalen + 8, IPPROTO_UDP, csum);
2914 if (udph->check == 0)
2915 udph->check = CSUM_MANGLED_0;
2918 #ifdef CONFIG_XFRM
2919 if (!process_ipsec(pkt_dev, skb, protocol))
2920 return NULL;
2921 #endif
2923 return skb;
2926 static struct sk_buff *fill_packet_ipv6(struct net_device *odev,
2927 struct pktgen_dev *pkt_dev)
2929 struct sk_buff *skb = NULL;
2930 __u8 *eth;
2931 struct udphdr *udph;
2932 int datalen, udplen;
2933 struct ipv6hdr *iph;
2934 __be16 protocol = htons(ETH_P_IPV6);
2935 __be32 *mpls;
2936 __be16 *vlan_tci = NULL; /* Encapsulates priority and VLAN ID */
2937 __be16 *vlan_encapsulated_proto = NULL; /* packet type ID field (or len) for VLAN tag */
2938 __be16 *svlan_tci = NULL; /* Encapsulates priority and SVLAN ID */
2939 __be16 *svlan_encapsulated_proto = NULL; /* packet type ID field (or len) for SVLAN tag */
2940 u16 queue_map;
2942 if (pkt_dev->nr_labels)
2943 protocol = htons(ETH_P_MPLS_UC);
2945 if (pkt_dev->vlan_id != 0xffff)
2946 protocol = htons(ETH_P_8021Q);
2948 /* Update any of the values, used when we're incrementing various
2949 * fields.
2951 mod_cur_headers(pkt_dev);
2952 queue_map = pkt_dev->cur_queue_map;
2954 skb = pktgen_alloc_skb(odev, pkt_dev, 16);
2955 if (!skb) {
2956 sprintf(pkt_dev->result, "No memory");
2957 return NULL;
2960 prefetchw(skb->data);
2961 skb_reserve(skb, 16);
2963 /* Reserve for ethernet and IP header */
2964 eth = (__u8 *) skb_push(skb, 14);
2965 mpls = (__be32 *)skb_put(skb, pkt_dev->nr_labels*sizeof(__u32));
2966 if (pkt_dev->nr_labels)
2967 mpls_push(mpls, pkt_dev);
2969 if (pkt_dev->vlan_id != 0xffff) {
2970 if (pkt_dev->svlan_id != 0xffff) {
2971 svlan_tci = (__be16 *)skb_put(skb, sizeof(__be16));
2972 *svlan_tci = build_tci(pkt_dev->svlan_id,
2973 pkt_dev->svlan_cfi,
2974 pkt_dev->svlan_p);
2975 svlan_encapsulated_proto = (__be16 *)skb_put(skb, sizeof(__be16));
2976 *svlan_encapsulated_proto = htons(ETH_P_8021Q);
2978 vlan_tci = (__be16 *)skb_put(skb, sizeof(__be16));
2979 *vlan_tci = build_tci(pkt_dev->vlan_id,
2980 pkt_dev->vlan_cfi,
2981 pkt_dev->vlan_p);
2982 vlan_encapsulated_proto = (__be16 *)skb_put(skb, sizeof(__be16));
2983 *vlan_encapsulated_proto = htons(ETH_P_IPV6);
2986 skb_reset_mac_header(skb);
2987 skb_set_network_header(skb, skb->len);
2988 iph = (struct ipv6hdr *) skb_put(skb, sizeof(struct ipv6hdr));
2990 skb_set_transport_header(skb, skb->len);
2991 udph = (struct udphdr *) skb_put(skb, sizeof(struct udphdr));
2992 skb_set_queue_mapping(skb, queue_map);
2993 skb->priority = pkt_dev->skb_priority;
2995 memcpy(eth, pkt_dev->hh, 12);
2996 *(__be16 *) &eth[12] = protocol;
2998 /* Eth + IPh + UDPh + mpls */
2999 datalen = pkt_dev->cur_pkt_size - 14 -
3000 sizeof(struct ipv6hdr) - sizeof(struct udphdr) -
3001 pkt_dev->pkt_overhead;
3003 if (datalen < 0 || datalen < sizeof(struct pktgen_hdr)) {
3004 datalen = sizeof(struct pktgen_hdr);
3005 net_info_ratelimited("increased datalen to %d\n", datalen);
3008 udplen = datalen + sizeof(struct udphdr);
3009 udph->source = htons(pkt_dev->cur_udp_src);
3010 udph->dest = htons(pkt_dev->cur_udp_dst);
3011 udph->len = htons(udplen);
3012 udph->check = 0;
3014 *(__be32 *) iph = htonl(0x60000000); /* Version + flow */
3016 if (pkt_dev->traffic_class) {
3017 /* Version + traffic class + flow (0) */
3018 *(__be32 *)iph |= htonl(0x60000000 | (pkt_dev->traffic_class << 20));
3021 iph->hop_limit = 32;
3023 iph->payload_len = htons(udplen);
3024 iph->nexthdr = IPPROTO_UDP;
3026 iph->daddr = pkt_dev->cur_in6_daddr;
3027 iph->saddr = pkt_dev->cur_in6_saddr;
3029 skb->protocol = protocol;
3030 skb->dev = odev;
3031 skb->pkt_type = PACKET_HOST;
3033 pktgen_finalize_skb(pkt_dev, skb, datalen);
3035 if (!(pkt_dev->flags & F_UDPCSUM)) {
3036 skb->ip_summed = CHECKSUM_NONE;
3037 } else if (odev->features & (NETIF_F_HW_CSUM | NETIF_F_IPV6_CSUM)) {
3038 skb->ip_summed = CHECKSUM_PARTIAL;
3039 skb->csum_start = skb_transport_header(skb) - skb->head;
3040 skb->csum_offset = offsetof(struct udphdr, check);
3041 udph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, udplen, IPPROTO_UDP, 0);
3042 } else {
3043 __wsum csum = skb_checksum(skb, skb_transport_offset(skb), udplen, 0);
3045 /* add protocol-dependent pseudo-header */
3046 udph->check = csum_ipv6_magic(&iph->saddr, &iph->daddr, udplen, IPPROTO_UDP, csum);
3048 if (udph->check == 0)
3049 udph->check = CSUM_MANGLED_0;
3052 return skb;
3055 static struct sk_buff *fill_packet(struct net_device *odev,
3056 struct pktgen_dev *pkt_dev)
3058 if (pkt_dev->flags & F_IPV6)
3059 return fill_packet_ipv6(odev, pkt_dev);
3060 else
3061 return fill_packet_ipv4(odev, pkt_dev);
3064 static void pktgen_clear_counters(struct pktgen_dev *pkt_dev)
3066 pkt_dev->seq_num = 1;
3067 pkt_dev->idle_acc = 0;
3068 pkt_dev->sofar = 0;
3069 pkt_dev->tx_bytes = 0;
3070 pkt_dev->errors = 0;
3073 /* Set up structure for sending pkts, clear counters */
3075 static void pktgen_run(struct pktgen_thread *t)
3077 struct pktgen_dev *pkt_dev;
3078 int started = 0;
3080 func_enter();
3082 rcu_read_lock();
3083 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) {
3086 * setup odev and create initial packet.
3088 pktgen_setup_inject(pkt_dev);
3090 if (pkt_dev->odev) {
3091 pktgen_clear_counters(pkt_dev);
3092 pkt_dev->skb = NULL;
3093 pkt_dev->started_at = pkt_dev->next_tx = ktime_get();
3095 set_pkt_overhead(pkt_dev);
3097 strcpy(pkt_dev->result, "Starting");
3098 pkt_dev->running = 1; /* Cranke yeself! */
3099 started++;
3100 } else
3101 strcpy(pkt_dev->result, "Error starting");
3103 rcu_read_unlock();
3104 if (started)
3105 t->control &= ~(T_STOP);
3108 static void pktgen_stop_all_threads_ifs(struct pktgen_net *pn)
3110 struct pktgen_thread *t;
3112 func_enter();
3114 mutex_lock(&pktgen_thread_lock);
3116 list_for_each_entry(t, &pn->pktgen_threads, th_list)
3117 t->control |= T_STOP;
3119 mutex_unlock(&pktgen_thread_lock);
3122 static int thread_is_running(const struct pktgen_thread *t)
3124 const struct pktgen_dev *pkt_dev;
3126 rcu_read_lock();
3127 list_for_each_entry_rcu(pkt_dev, &t->if_list, list)
3128 if (pkt_dev->running) {
3129 rcu_read_unlock();
3130 return 1;
3132 rcu_read_unlock();
3133 return 0;
3136 static int pktgen_wait_thread_run(struct pktgen_thread *t)
3138 while (thread_is_running(t)) {
3140 msleep_interruptible(100);
3142 if (signal_pending(current))
3143 goto signal;
3145 return 1;
3146 signal:
3147 return 0;
3150 static int pktgen_wait_all_threads_run(struct pktgen_net *pn)
3152 struct pktgen_thread *t;
3153 int sig = 1;
3155 mutex_lock(&pktgen_thread_lock);
3157 list_for_each_entry(t, &pn->pktgen_threads, th_list) {
3158 sig = pktgen_wait_thread_run(t);
3159 if (sig == 0)
3160 break;
3163 if (sig == 0)
3164 list_for_each_entry(t, &pn->pktgen_threads, th_list)
3165 t->control |= (T_STOP);
3167 mutex_unlock(&pktgen_thread_lock);
3168 return sig;
3171 static void pktgen_run_all_threads(struct pktgen_net *pn)
3173 struct pktgen_thread *t;
3175 func_enter();
3177 mutex_lock(&pktgen_thread_lock);
3179 list_for_each_entry(t, &pn->pktgen_threads, th_list)
3180 t->control |= (T_RUN);
3182 mutex_unlock(&pktgen_thread_lock);
3184 /* Propagate thread->control */
3185 schedule_timeout_interruptible(msecs_to_jiffies(125));
3187 pktgen_wait_all_threads_run(pn);
3190 static void pktgen_reset_all_threads(struct pktgen_net *pn)
3192 struct pktgen_thread *t;
3194 func_enter();
3196 mutex_lock(&pktgen_thread_lock);
3198 list_for_each_entry(t, &pn->pktgen_threads, th_list)
3199 t->control |= (T_REMDEVALL);
3201 mutex_unlock(&pktgen_thread_lock);
3203 /* Propagate thread->control */
3204 schedule_timeout_interruptible(msecs_to_jiffies(125));
3206 pktgen_wait_all_threads_run(pn);
3209 static void show_results(struct pktgen_dev *pkt_dev, int nr_frags)
3211 __u64 bps, mbps, pps;
3212 char *p = pkt_dev->result;
3213 ktime_t elapsed = ktime_sub(pkt_dev->stopped_at,
3214 pkt_dev->started_at);
3215 ktime_t idle = ns_to_ktime(pkt_dev->idle_acc);
3217 p += sprintf(p, "OK: %llu(c%llu+d%llu) usec, %llu (%dbyte,%dfrags)\n",
3218 (unsigned long long)ktime_to_us(elapsed),
3219 (unsigned long long)ktime_to_us(ktime_sub(elapsed, idle)),
3220 (unsigned long long)ktime_to_us(idle),
3221 (unsigned long long)pkt_dev->sofar,
3222 pkt_dev->cur_pkt_size, nr_frags);
3224 pps = div64_u64(pkt_dev->sofar * NSEC_PER_SEC,
3225 ktime_to_ns(elapsed));
3227 bps = pps * 8 * pkt_dev->cur_pkt_size;
3229 mbps = bps;
3230 do_div(mbps, 1000000);
3231 p += sprintf(p, " %llupps %lluMb/sec (%llubps) errors: %llu",
3232 (unsigned long long)pps,
3233 (unsigned long long)mbps,
3234 (unsigned long long)bps,
3235 (unsigned long long)pkt_dev->errors);
3238 /* Set stopped-at timer, remove from running list, do counters & statistics */
3239 static int pktgen_stop_device(struct pktgen_dev *pkt_dev)
3241 int nr_frags = pkt_dev->skb ? skb_shinfo(pkt_dev->skb)->nr_frags : -1;
3243 if (!pkt_dev->running) {
3244 pr_warn("interface: %s is already stopped\n",
3245 pkt_dev->odevname);
3246 return -EINVAL;
3249 pkt_dev->running = 0;
3250 kfree_skb(pkt_dev->skb);
3251 pkt_dev->skb = NULL;
3252 pkt_dev->stopped_at = ktime_get();
3254 show_results(pkt_dev, nr_frags);
3256 return 0;
3259 static struct pktgen_dev *next_to_run(struct pktgen_thread *t)
3261 struct pktgen_dev *pkt_dev, *best = NULL;
3263 rcu_read_lock();
3264 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) {
3265 if (!pkt_dev->running)
3266 continue;
3267 if (best == NULL)
3268 best = pkt_dev;
3269 else if (ktime_compare(pkt_dev->next_tx, best->next_tx) < 0)
3270 best = pkt_dev;
3272 rcu_read_unlock();
3274 return best;
3277 static void pktgen_stop(struct pktgen_thread *t)
3279 struct pktgen_dev *pkt_dev;
3281 func_enter();
3283 rcu_read_lock();
3285 list_for_each_entry_rcu(pkt_dev, &t->if_list, list) {
3286 pktgen_stop_device(pkt_dev);
3289 rcu_read_unlock();
3293 * one of our devices needs to be removed - find it
3294 * and remove it
3296 static void pktgen_rem_one_if(struct pktgen_thread *t)
3298 struct list_head *q, *n;
3299 struct pktgen_dev *cur;
3301 func_enter();
3303 list_for_each_safe(q, n, &t->if_list) {
3304 cur = list_entry(q, struct pktgen_dev, list);
3306 if (!cur->removal_mark)
3307 continue;
3309 kfree_skb(cur->skb);
3310 cur->skb = NULL;
3312 pktgen_remove_device(t, cur);
3314 break;
3318 static void pktgen_rem_all_ifs(struct pktgen_thread *t)
3320 struct list_head *q, *n;
3321 struct pktgen_dev *cur;
3323 func_enter();
3325 /* Remove all devices, free mem */
3327 list_for_each_safe(q, n, &t->if_list) {
3328 cur = list_entry(q, struct pktgen_dev, list);
3330 kfree_skb(cur->skb);
3331 cur->skb = NULL;
3333 pktgen_remove_device(t, cur);
3337 static void pktgen_rem_thread(struct pktgen_thread *t)
3339 /* Remove from the thread list */
3340 remove_proc_entry(t->tsk->comm, t->net->proc_dir);
3343 static void pktgen_resched(struct pktgen_dev *pkt_dev)
3345 ktime_t idle_start = ktime_get();
3346 schedule();
3347 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(ktime_get(), idle_start));
3350 static void pktgen_wait_for_skb(struct pktgen_dev *pkt_dev)
3352 ktime_t idle_start = ktime_get();
3354 while (atomic_read(&(pkt_dev->skb->users)) != 1) {
3355 if (signal_pending(current))
3356 break;
3358 if (need_resched())
3359 pktgen_resched(pkt_dev);
3360 else
3361 cpu_relax();
3363 pkt_dev->idle_acc += ktime_to_ns(ktime_sub(ktime_get(), idle_start));
3366 static void pktgen_xmit(struct pktgen_dev *pkt_dev)
3368 unsigned int burst = ACCESS_ONCE(pkt_dev->burst);
3369 struct net_device *odev = pkt_dev->odev;
3370 struct netdev_queue *txq;
3371 struct sk_buff *skb;
3372 int ret;
3374 /* If device is offline, then don't send */
3375 if (unlikely(!netif_running(odev) || !netif_carrier_ok(odev))) {
3376 pktgen_stop_device(pkt_dev);
3377 return;
3380 /* This is max DELAY, this has special meaning of
3381 * "never transmit"
3383 if (unlikely(pkt_dev->delay == ULLONG_MAX)) {
3384 pkt_dev->next_tx = ktime_add_ns(ktime_get(), ULONG_MAX);
3385 return;
3388 /* If no skb or clone count exhausted then get new one */
3389 if (!pkt_dev->skb || (pkt_dev->last_ok &&
3390 ++pkt_dev->clone_count >= pkt_dev->clone_skb)) {
3391 /* build a new pkt */
3392 kfree_skb(pkt_dev->skb);
3394 pkt_dev->skb = fill_packet(odev, pkt_dev);
3395 if (pkt_dev->skb == NULL) {
3396 pr_err("ERROR: couldn't allocate skb in fill_packet\n");
3397 schedule();
3398 pkt_dev->clone_count--; /* back out increment, OOM */
3399 return;
3401 pkt_dev->last_pkt_size = pkt_dev->skb->len;
3402 pkt_dev->clone_count = 0; /* reset counter */
3405 if (pkt_dev->delay && pkt_dev->last_ok)
3406 spin(pkt_dev, pkt_dev->next_tx);
3408 if (pkt_dev->xmit_mode == M_NETIF_RECEIVE) {
3409 skb = pkt_dev->skb;
3410 skb->protocol = eth_type_trans(skb, skb->dev);
3411 atomic_add(burst, &skb->users);
3412 local_bh_disable();
3413 do {
3414 ret = netif_receive_skb(skb);
3415 if (ret == NET_RX_DROP)
3416 pkt_dev->errors++;
3417 pkt_dev->sofar++;
3418 pkt_dev->seq_num++;
3419 if (atomic_read(&skb->users) != burst) {
3420 /* skb was queued by rps/rfs or taps,
3421 * so cannot reuse this skb
3423 atomic_sub(burst - 1, &skb->users);
3424 /* get out of the loop and wait
3425 * until skb is consumed
3427 break;
3429 /* skb was 'freed' by stack, so clean few
3430 * bits and reuse it
3432 #ifdef CONFIG_NET_CLS_ACT
3433 skb->tc_verd = 0; /* reset reclass/redir ttl */
3434 #endif
3435 } while (--burst > 0);
3436 goto out; /* Skips xmit_mode M_START_XMIT */
3439 txq = skb_get_tx_queue(odev, pkt_dev->skb);
3441 local_bh_disable();
3443 HARD_TX_LOCK(odev, txq, smp_processor_id());
3445 if (unlikely(netif_xmit_frozen_or_drv_stopped(txq))) {
3446 ret = NETDEV_TX_BUSY;
3447 pkt_dev->last_ok = 0;
3448 goto unlock;
3450 atomic_add(burst, &pkt_dev->skb->users);
3452 xmit_more:
3453 ret = netdev_start_xmit(pkt_dev->skb, odev, txq, --burst > 0);
3455 switch (ret) {
3456 case NETDEV_TX_OK:
3457 pkt_dev->last_ok = 1;
3458 pkt_dev->sofar++;
3459 pkt_dev->seq_num++;
3460 pkt_dev->tx_bytes += pkt_dev->last_pkt_size;
3461 if (burst > 0 && !netif_xmit_frozen_or_drv_stopped(txq))
3462 goto xmit_more;
3463 break;
3464 case NET_XMIT_DROP:
3465 case NET_XMIT_CN:
3466 /* skb has been consumed */
3467 pkt_dev->errors++;
3468 break;
3469 default: /* Drivers are not supposed to return other values! */
3470 net_info_ratelimited("%s xmit error: %d\n",
3471 pkt_dev->odevname, ret);
3472 pkt_dev->errors++;
3473 /* fallthru */
3474 case NETDEV_TX_BUSY:
3475 /* Retry it next time */
3476 atomic_dec(&(pkt_dev->skb->users));
3477 pkt_dev->last_ok = 0;
3479 if (unlikely(burst))
3480 atomic_sub(burst, &pkt_dev->skb->users);
3481 unlock:
3482 HARD_TX_UNLOCK(odev, txq);
3484 out:
3485 local_bh_enable();
3487 /* If pkt_dev->count is zero, then run forever */
3488 if ((pkt_dev->count != 0) && (pkt_dev->sofar >= pkt_dev->count)) {
3489 pktgen_wait_for_skb(pkt_dev);
3491 /* Done with this */
3492 pktgen_stop_device(pkt_dev);
3497 * Main loop of the thread goes here
3500 static int pktgen_thread_worker(void *arg)
3502 DEFINE_WAIT(wait);
3503 struct pktgen_thread *t = arg;
3504 struct pktgen_dev *pkt_dev = NULL;
3505 int cpu = t->cpu;
3507 BUG_ON(smp_processor_id() != cpu);
3509 init_waitqueue_head(&t->queue);
3510 complete(&t->start_done);
3512 pr_debug("starting pktgen/%d: pid=%d\n", cpu, task_pid_nr(current));
3514 set_freezable();
3516 while (!kthread_should_stop()) {
3517 pkt_dev = next_to_run(t);
3519 if (unlikely(!pkt_dev && t->control == 0)) {
3520 if (t->net->pktgen_exiting)
3521 break;
3522 wait_event_interruptible_timeout(t->queue,
3523 t->control != 0,
3524 HZ/10);
3525 try_to_freeze();
3526 continue;
3529 if (likely(pkt_dev)) {
3530 pktgen_xmit(pkt_dev);
3532 if (need_resched())
3533 pktgen_resched(pkt_dev);
3534 else
3535 cpu_relax();
3538 if (t->control & T_STOP) {
3539 pktgen_stop(t);
3540 t->control &= ~(T_STOP);
3543 if (t->control & T_RUN) {
3544 pktgen_run(t);
3545 t->control &= ~(T_RUN);
3548 if (t->control & T_REMDEVALL) {
3549 pktgen_rem_all_ifs(t);
3550 t->control &= ~(T_REMDEVALL);
3553 if (t->control & T_REMDEV) {
3554 pktgen_rem_one_if(t);
3555 t->control &= ~(T_REMDEV);
3558 try_to_freeze();
3561 pr_debug("%s stopping all device\n", t->tsk->comm);
3562 pktgen_stop(t);
3564 pr_debug("%s removing all device\n", t->tsk->comm);
3565 pktgen_rem_all_ifs(t);
3567 pr_debug("%s removing thread\n", t->tsk->comm);
3568 pktgen_rem_thread(t);
3570 return 0;
3573 static struct pktgen_dev *pktgen_find_dev(struct pktgen_thread *t,
3574 const char *ifname, bool exact)
3576 struct pktgen_dev *p, *pkt_dev = NULL;
3577 size_t len = strlen(ifname);
3579 rcu_read_lock();
3580 list_for_each_entry_rcu(p, &t->if_list, list)
3581 if (strncmp(p->odevname, ifname, len) == 0) {
3582 if (p->odevname[len]) {
3583 if (exact || p->odevname[len] != '@')
3584 continue;
3586 pkt_dev = p;
3587 break;
3590 rcu_read_unlock();
3591 pr_debug("find_dev(%s) returning %p\n", ifname, pkt_dev);
3592 return pkt_dev;
3596 * Adds a dev at front of if_list.
3599 static int add_dev_to_thread(struct pktgen_thread *t,
3600 struct pktgen_dev *pkt_dev)
3602 int rv = 0;
3604 /* This function cannot be called concurrently, as its called
3605 * under pktgen_thread_lock mutex, but it can run from
3606 * userspace on another CPU than the kthread. The if_lock()
3607 * is used here to sync with concurrent instances of
3608 * _rem_dev_from_if_list() invoked via kthread, which is also
3609 * updating the if_list */
3610 if_lock(t);
3612 if (pkt_dev->pg_thread) {
3613 pr_err("ERROR: already assigned to a thread\n");
3614 rv = -EBUSY;
3615 goto out;
3618 pkt_dev->running = 0;
3619 pkt_dev->pg_thread = t;
3620 list_add_rcu(&pkt_dev->list, &t->if_list);
3622 out:
3623 if_unlock(t);
3624 return rv;
3627 /* Called under thread lock */
3629 static int pktgen_add_device(struct pktgen_thread *t, const char *ifname)
3631 struct pktgen_dev *pkt_dev;
3632 int err;
3633 int node = cpu_to_node(t->cpu);
3635 /* We don't allow a device to be on several threads */
3637 pkt_dev = __pktgen_NN_threads(t->net, ifname, FIND);
3638 if (pkt_dev) {
3639 pr_err("ERROR: interface already used\n");
3640 return -EBUSY;
3643 pkt_dev = kzalloc_node(sizeof(struct pktgen_dev), GFP_KERNEL, node);
3644 if (!pkt_dev)
3645 return -ENOMEM;
3647 strcpy(pkt_dev->odevname, ifname);
3648 pkt_dev->flows = vzalloc_node(MAX_CFLOWS * sizeof(struct flow_state),
3649 node);
3650 if (pkt_dev->flows == NULL) {
3651 kfree(pkt_dev);
3652 return -ENOMEM;
3655 pkt_dev->removal_mark = 0;
3656 pkt_dev->nfrags = 0;
3657 pkt_dev->delay = pg_delay_d;
3658 pkt_dev->count = pg_count_d;
3659 pkt_dev->sofar = 0;
3660 pkt_dev->udp_src_min = 9; /* sink port */
3661 pkt_dev->udp_src_max = 9;
3662 pkt_dev->udp_dst_min = 9;
3663 pkt_dev->udp_dst_max = 9;
3664 pkt_dev->vlan_p = 0;
3665 pkt_dev->vlan_cfi = 0;
3666 pkt_dev->vlan_id = 0xffff;
3667 pkt_dev->svlan_p = 0;
3668 pkt_dev->svlan_cfi = 0;
3669 pkt_dev->svlan_id = 0xffff;
3670 pkt_dev->burst = 1;
3671 pkt_dev->node = -1;
3673 err = pktgen_setup_dev(t->net, pkt_dev, ifname);
3674 if (err)
3675 goto out1;
3676 if (pkt_dev->odev->priv_flags & IFF_TX_SKB_SHARING)
3677 pkt_dev->clone_skb = pg_clone_skb_d;
3679 pkt_dev->entry = proc_create_data(ifname, 0600, t->net->proc_dir,
3680 &pktgen_if_fops, pkt_dev);
3681 if (!pkt_dev->entry) {
3682 pr_err("cannot create %s/%s procfs entry\n",
3683 PG_PROC_DIR, ifname);
3684 err = -EINVAL;
3685 goto out2;
3687 #ifdef CONFIG_XFRM
3688 pkt_dev->ipsmode = XFRM_MODE_TRANSPORT;
3689 pkt_dev->ipsproto = IPPROTO_ESP;
3691 /* xfrm tunnel mode needs additional dst to extract outter
3692 * ip header protocol/ttl/id field, here creat a phony one.
3693 * instead of looking for a valid rt, which definitely hurting
3694 * performance under such circumstance.
3696 pkt_dev->dstops.family = AF_INET;
3697 pkt_dev->dst.dev = pkt_dev->odev;
3698 dst_init_metrics(&pkt_dev->dst, pktgen_dst_metrics, false);
3699 pkt_dev->dst.child = &pkt_dev->dst;
3700 pkt_dev->dst.ops = &pkt_dev->dstops;
3701 #endif
3703 return add_dev_to_thread(t, pkt_dev);
3704 out2:
3705 dev_put(pkt_dev->odev);
3706 out1:
3707 #ifdef CONFIG_XFRM
3708 free_SAs(pkt_dev);
3709 #endif
3710 vfree(pkt_dev->flows);
3711 kfree(pkt_dev);
3712 return err;
3715 static int __net_init pktgen_create_thread(int cpu, struct pktgen_net *pn)
3717 struct pktgen_thread *t;
3718 struct proc_dir_entry *pe;
3719 struct task_struct *p;
3721 t = kzalloc_node(sizeof(struct pktgen_thread), GFP_KERNEL,
3722 cpu_to_node(cpu));
3723 if (!t) {
3724 pr_err("ERROR: out of memory, can't create new thread\n");
3725 return -ENOMEM;
3728 spin_lock_init(&t->if_lock);
3729 t->cpu = cpu;
3731 INIT_LIST_HEAD(&t->if_list);
3733 list_add_tail(&t->th_list, &pn->pktgen_threads);
3734 init_completion(&t->start_done);
3736 p = kthread_create_on_node(pktgen_thread_worker,
3738 cpu_to_node(cpu),
3739 "kpktgend_%d", cpu);
3740 if (IS_ERR(p)) {
3741 pr_err("kernel_thread() failed for cpu %d\n", t->cpu);
3742 list_del(&t->th_list);
3743 kfree(t);
3744 return PTR_ERR(p);
3746 kthread_bind(p, cpu);
3747 t->tsk = p;
3749 pe = proc_create_data(t->tsk->comm, 0600, pn->proc_dir,
3750 &pktgen_thread_fops, t);
3751 if (!pe) {
3752 pr_err("cannot create %s/%s procfs entry\n",
3753 PG_PROC_DIR, t->tsk->comm);
3754 kthread_stop(p);
3755 list_del(&t->th_list);
3756 kfree(t);
3757 return -EINVAL;
3760 t->net = pn;
3761 get_task_struct(p);
3762 wake_up_process(p);
3763 wait_for_completion(&t->start_done);
3765 return 0;
3769 * Removes a device from the thread if_list.
3771 static void _rem_dev_from_if_list(struct pktgen_thread *t,
3772 struct pktgen_dev *pkt_dev)
3774 struct list_head *q, *n;
3775 struct pktgen_dev *p;
3777 if_lock(t);
3778 list_for_each_safe(q, n, &t->if_list) {
3779 p = list_entry(q, struct pktgen_dev, list);
3780 if (p == pkt_dev)
3781 list_del_rcu(&p->list);
3783 if_unlock(t);
3786 static int pktgen_remove_device(struct pktgen_thread *t,
3787 struct pktgen_dev *pkt_dev)
3789 pr_debug("remove_device pkt_dev=%p\n", pkt_dev);
3791 if (pkt_dev->running) {
3792 pr_warn("WARNING: trying to remove a running interface, stopping it now\n");
3793 pktgen_stop_device(pkt_dev);
3796 /* Dis-associate from the interface */
3798 if (pkt_dev->odev) {
3799 dev_put(pkt_dev->odev);
3800 pkt_dev->odev = NULL;
3803 /* Remove proc before if_list entry, because add_device uses
3804 * list to determine if interface already exist, avoid race
3805 * with proc_create_data() */
3806 proc_remove(pkt_dev->entry);
3808 /* And update the thread if_list */
3809 _rem_dev_from_if_list(t, pkt_dev);
3811 #ifdef CONFIG_XFRM
3812 free_SAs(pkt_dev);
3813 #endif
3814 vfree(pkt_dev->flows);
3815 if (pkt_dev->page)
3816 put_page(pkt_dev->page);
3817 kfree_rcu(pkt_dev, rcu);
3818 return 0;
3821 static int __net_init pg_net_init(struct net *net)
3823 struct pktgen_net *pn = net_generic(net, pg_net_id);
3824 struct proc_dir_entry *pe;
3825 int cpu, ret = 0;
3827 pn->net = net;
3828 INIT_LIST_HEAD(&pn->pktgen_threads);
3829 pn->pktgen_exiting = false;
3830 pn->proc_dir = proc_mkdir(PG_PROC_DIR, pn->net->proc_net);
3831 if (!pn->proc_dir) {
3832 pr_warn("cannot create /proc/net/%s\n", PG_PROC_DIR);
3833 return -ENODEV;
3835 pe = proc_create(PGCTRL, 0600, pn->proc_dir, &pktgen_fops);
3836 if (pe == NULL) {
3837 pr_err("cannot create %s procfs entry\n", PGCTRL);
3838 ret = -EINVAL;
3839 goto remove;
3842 for_each_online_cpu(cpu) {
3843 int err;
3845 err = pktgen_create_thread(cpu, pn);
3846 if (err)
3847 pr_warn("Cannot create thread for cpu %d (%d)\n",
3848 cpu, err);
3851 if (list_empty(&pn->pktgen_threads)) {
3852 pr_err("Initialization failed for all threads\n");
3853 ret = -ENODEV;
3854 goto remove_entry;
3857 return 0;
3859 remove_entry:
3860 remove_proc_entry(PGCTRL, pn->proc_dir);
3861 remove:
3862 remove_proc_entry(PG_PROC_DIR, pn->net->proc_net);
3863 return ret;
3866 static void __net_exit pg_net_exit(struct net *net)
3868 struct pktgen_net *pn = net_generic(net, pg_net_id);
3869 struct pktgen_thread *t;
3870 struct list_head *q, *n;
3871 LIST_HEAD(list);
3873 /* Stop all interfaces & threads */
3874 pn->pktgen_exiting = true;
3876 mutex_lock(&pktgen_thread_lock);
3877 list_splice_init(&pn->pktgen_threads, &list);
3878 mutex_unlock(&pktgen_thread_lock);
3880 list_for_each_safe(q, n, &list) {
3881 t = list_entry(q, struct pktgen_thread, th_list);
3882 list_del(&t->th_list);
3883 kthread_stop(t->tsk);
3884 put_task_struct(t->tsk);
3885 kfree(t);
3888 remove_proc_entry(PGCTRL, pn->proc_dir);
3889 remove_proc_entry(PG_PROC_DIR, pn->net->proc_net);
3892 static struct pernet_operations pg_net_ops = {
3893 .init = pg_net_init,
3894 .exit = pg_net_exit,
3895 .id = &pg_net_id,
3896 .size = sizeof(struct pktgen_net),
3899 static int __init pg_init(void)
3901 int ret = 0;
3903 pr_info("%s", version);
3904 ret = register_pernet_subsys(&pg_net_ops);
3905 if (ret)
3906 return ret;
3907 ret = register_netdevice_notifier(&pktgen_notifier_block);
3908 if (ret)
3909 unregister_pernet_subsys(&pg_net_ops);
3911 return ret;
3914 static void __exit pg_cleanup(void)
3916 unregister_netdevice_notifier(&pktgen_notifier_block);
3917 unregister_pernet_subsys(&pg_net_ops);
3918 /* Don't need rcu_barrier() due to use of kfree_rcu() */
3921 module_init(pg_init);
3922 module_exit(pg_cleanup);
3924 MODULE_AUTHOR("Robert Olsson <robert.olsson@its.uu.se>");
3925 MODULE_DESCRIPTION("Packet Generator tool");
3926 MODULE_LICENSE("GPL");
3927 MODULE_VERSION(VERSION);
3928 module_param(pg_count_d, int, 0);
3929 MODULE_PARM_DESC(pg_count_d, "Default number of packets to inject");
3930 module_param(pg_delay_d, int, 0);
3931 MODULE_PARM_DESC(pg_delay_d, "Default delay between packets (nanoseconds)");
3932 module_param(pg_clone_skb_d, int, 0);
3933 MODULE_PARM_DESC(pg_clone_skb_d, "Default number of copies of the same packet");
3934 module_param(debug, int, 0);
3935 MODULE_PARM_DESC(debug, "Enable debugging of pktgen module");