kernel/linux/kni/kni_misc.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright(c) 2010-2014 Intel Corporation.
   4  */
   5
   6 #include <linux/version.h>
   7 #include <linux/module.h>
   8 #include <linux/miscdevice.h>
   9 #include <linux/netdevice.h>
  10 #include <linux/etherdevice.h>
  11 #include <linux/pci.h>
  12 #include <linux/kthread.h>
  13 #include <linux/rwsem.h>
  14 #include <linux/mutex.h>
  15 #include <linux/nsproxy.h>
  16 #include <net/net_namespace.h>
  17 #include <net/netns/generic.h>
  18
  19 #include <rte_kni_common.h>
  20
  21 #include "compat.h"
  22 #include "kni_dev.h"
  23
  24 MODULE_LICENSE("Dual BSD/GPL");
  25 MODULE_AUTHOR("Intel Corporation");
  26 MODULE_DESCRIPTION("Kernel Module for managing kni devices");
  27
  28 #define KNI_RX_LOOP_NUM 1000
  29
  30 #define KNI_MAX_DEVICES 32
  31
  32 /* loopback mode */
  33 static char *lo_mode;
  34
  35 /* Kernel thread mode */
  36 static char *kthread_mode;
  37 static uint32_t multiple_kthread_on;
  38
  39 /* Default carrier state for created KNI network interfaces */
  40 static char *carrier;
  41 uint32_t dflt_carrier;
  42
  43 #define KNI_DEV_IN_USE_BIT_NUM 0 /* Bit number for device in use */
  44
  45 static int kni_net_id;
  46
  47 struct kni_net {
  48         unsigned long device_in_use; /* device in use flag */
  49         struct mutex kni_kthread_lock;
  50         struct task_struct *kni_kthread;
  51         struct rw_semaphore kni_list_lock;
  52         struct list_head kni_list_head;
  53 };
  54
  55 static int __net_init
  56 kni_init_net(struct net *net)
  57 {
  58 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
  59         struct kni_net *knet = net_generic(net, kni_net_id);
  60
  61         memset(knet, 0, sizeof(*knet));
  62 #else
  63         struct kni_net *knet;
  64         int ret;
  65
  66         knet = kzalloc(sizeof(struct kni_net), GFP_KERNEL);
  67         if (!knet) {
  68                 ret = -ENOMEM;
  69                 return ret;
  70         }
  71 #endif
  72
  73         /* Clear the bit of device in use */
  74         clear_bit(KNI_DEV_IN_USE_BIT_NUM, &knet->device_in_use);
  75
  76         mutex_init(&knet->kni_kthread_lock);
  77
  78         init_rwsem(&knet->kni_list_lock);
  79         INIT_LIST_HEAD(&knet->kni_list_head);
  80
  81 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
  82         return 0;
  83 #else
  84         ret = net_assign_generic(net, kni_net_id, knet);
  85         if (ret < 0)
  86                 kfree(knet);
  87
  88         return ret;
  89 #endif
  90 }
  91
  92 static void __net_exit
  93 kni_exit_net(struct net *net)
  94 {
  95         struct kni_net *knet __maybe_unused;
  96
  97         knet = net_generic(net, kni_net_id);
  98         mutex_destroy(&knet->kni_kthread_lock);
  99
 100 #ifndef HAVE_SIMPLIFIED_PERNET_OPERATIONS
 101         kfree(knet);
 102 #endif
 103 }
 104
 105 static struct pernet_operations kni_net_ops = {
 106         .init = kni_init_net,
 107         .exit = kni_exit_net,
 108 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
 109         .id   = &kni_net_id,
 110         .size = sizeof(struct kni_net),
 111 #endif
 112 };
 113
 114 static int
 115 kni_thread_single(void *data)
 116 {
 117         struct kni_net *knet = data;
 118         int j;
 119         struct kni_dev *dev;
 120
 121         while (!kthread_should_stop()) {
 122                 down_read(&knet->kni_list_lock);
 123                 for (j = 0; j < KNI_RX_LOOP_NUM; j++) {
 124                         list_for_each_entry(dev, &knet->kni_list_head, list) {
 125                                 kni_net_rx(dev);
 126                                 kni_net_poll_resp(dev);
 127                         }
 128                 }
 129                 up_read(&knet->kni_list_lock);
 130 #ifdef RTE_KNI_PREEMPT_DEFAULT
 131                 /* reschedule out for a while */
 132                 schedule_timeout_interruptible(
 133                         usecs_to_jiffies(KNI_KTHREAD_RESCHEDULE_INTERVAL));
 134 #endif
 135         }
 136
 137         return 0;
 138 }
 139
 140 static int
 141 kni_thread_multiple(void *param)
 142 {
 143         int j;
 144         struct kni_dev *dev = param;
 145
 146         while (!kthread_should_stop()) {
 147                 for (j = 0; j < KNI_RX_LOOP_NUM; j++) {
 148                         kni_net_rx(dev);
 149                         kni_net_poll_resp(dev);
 150                 }
 151 #ifdef RTE_KNI_PREEMPT_DEFAULT
 152                 schedule_timeout_interruptible(
 153                         usecs_to_jiffies(KNI_KTHREAD_RESCHEDULE_INTERVAL));
 154 #endif
 155         }
 156
 157         return 0;
 158 }
 159
 160 static int
 161 kni_open(struct inode *inode, struct file *file)
 162 {
 163         struct net *net = current->nsproxy->net_ns;
 164         struct kni_net *knet = net_generic(net, kni_net_id);
 165
 166         /* kni device can be opened by one user only per netns */
 167         if (test_and_set_bit(KNI_DEV_IN_USE_BIT_NUM, &knet->device_in_use))
 168                 return -EBUSY;
 169
 170         file->private_data = get_net(net);
 171         pr_debug("/dev/kni opened\n");
 172
 173         return 0;
 174 }
 175
 176 static int
 177 kni_dev_remove(struct kni_dev *dev)
 178 {
 179         if (!dev)
 180                 return -ENODEV;
 181
 182         if (dev->net_dev) {
 183                 unregister_netdev(dev->net_dev);
 184                 free_netdev(dev->net_dev);
 185         }
 186
 187         kni_net_release_fifo_phy(dev);
 188
 189         return 0;
 190 }
 191
 192 static int
 193 kni_release(struct inode *inode, struct file *file)
 194 {
 195         struct net *net = file->private_data;
 196         struct kni_net *knet = net_generic(net, kni_net_id);
 197         struct kni_dev *dev, *n;
 198
 199         /* Stop kernel thread for single mode */
 200         if (multiple_kthread_on == 0) {
 201                 mutex_lock(&knet->kni_kthread_lock);
 202                 /* Stop kernel thread */
 203                 if (knet->kni_kthread != NULL) {
 204                         kthread_stop(knet->kni_kthread);
 205                         knet->kni_kthread = NULL;
 206                 }
 207                 mutex_unlock(&knet->kni_kthread_lock);
 208         }
 209
 210         down_write(&knet->kni_list_lock);
 211         list_for_each_entry_safe(dev, n, &knet->kni_list_head, list) {
 212                 /* Stop kernel thread for multiple mode */
 213                 if (multiple_kthread_on && dev->pthread != NULL) {
 214                         kthread_stop(dev->pthread);
 215                         dev->pthread = NULL;
 216                 }
 217
 218                 kni_dev_remove(dev);
 219                 list_del(&dev->list);
 220         }
 221         up_write(&knet->kni_list_lock);
 222
 223         /* Clear the bit of device in use */
 224         clear_bit(KNI_DEV_IN_USE_BIT_NUM, &knet->device_in_use);
 225
 226         put_net(net);
 227         pr_debug("/dev/kni closed\n");
 228
 229         return 0;
 230 }
 231
 232 static int
 233 kni_check_param(struct kni_dev *kni, struct rte_kni_device_info *dev)
 234 {
 235         if (!kni || !dev)
 236                 return -1;
 237
 238         /* Check if network name has been used */
 239         if (!strncmp(kni->name, dev->name, RTE_KNI_NAMESIZE)) {
 240                 pr_err("KNI name %s duplicated\n", dev->name);
 241                 return -1;
 242         }
 243
 244         return 0;
 245 }
 246
 247 static int
 248 kni_run_thread(struct kni_net *knet, struct kni_dev *kni, uint8_t force_bind)
 249 {
 250         /**
 251          * Create a new kernel thread for multiple mode, set its core affinity,
 252          * and finally wake it up.
 253          */
 254         if (multiple_kthread_on) {
 255                 kni->pthread = kthread_create(kni_thread_multiple,
 256                         (void *)kni, "kni_%s", kni->name);
 257                 if (IS_ERR(kni->pthread)) {
 258                         kni_dev_remove(kni);
 259                         return -ECANCELED;
 260                 }
 261
 262                 if (force_bind)
 263                         kthread_bind(kni->pthread, kni->core_id);
 264                 wake_up_process(kni->pthread);
 265         } else {
 266                 mutex_lock(&knet->kni_kthread_lock);
 267
 268                 if (knet->kni_kthread == NULL) {
 269                         knet->kni_kthread = kthread_create(kni_thread_single,
 270                                 (void *)knet, "kni_single");
 271                         if (IS_ERR(knet->kni_kthread)) {
 272                                 mutex_unlock(&knet->kni_kthread_lock);
 273                                 kni_dev_remove(kni);
 274                                 return -ECANCELED;
 275                         }
 276
 277                         if (force_bind)
 278                                 kthread_bind(knet->kni_kthread, kni->core_id);
 279                         wake_up_process(knet->kni_kthread);
 280                 }
 281
 282                 mutex_unlock(&knet->kni_kthread_lock);
 283         }
 284
 285         return 0;
 286 }
 287
 288 static int
 289 kni_ioctl_create(struct net *net, uint32_t ioctl_num,
 290                 unsigned long ioctl_param)
 291 {
 292         struct kni_net *knet = net_generic(net, kni_net_id);
 293         int ret;
 294         struct rte_kni_device_info dev_info;
 295         struct net_device *net_dev = NULL;
 296         struct kni_dev *kni, *dev, *n;
 297
 298         pr_info("Creating kni...\n");
 299         /* Check the buffer size, to avoid warning */
 300         if (_IOC_SIZE(ioctl_num) > sizeof(dev_info))
 301                 return -EINVAL;
 302
 303         /* Copy kni info from user space */
 304         ret = copy_from_user(&dev_info, (void *)ioctl_param, sizeof(dev_info));
 305         if (ret) {
 306                 pr_err("copy_from_user in kni_ioctl_create");
 307                 return -EIO;
 308         }
 309
 310         /* Check if name is zero-ended */
 311         if (strnlen(dev_info.name, sizeof(dev_info.name)) == sizeof(dev_info.name)) {
 312                 pr_err("kni.name not zero-terminated");
 313                 return -EINVAL;
 314         }
 315
 316         /**
 317          * Check if the cpu core id is valid for binding.
 318          */
 319         if (dev_info.force_bind && !cpu_online(dev_info.core_id)) {
 320                 pr_err("cpu %u is not online\n", dev_info.core_id);
 321                 return -EINVAL;
 322         }
 323
 324         /* Check if it has been created */
 325         down_read(&knet->kni_list_lock);
 326         list_for_each_entry_safe(dev, n, &knet->kni_list_head, list) {
 327                 if (kni_check_param(dev, &dev_info) < 0) {
 328                         up_read(&knet->kni_list_lock);
 329                         return -EINVAL;
 330                 }
 331         }
 332         up_read(&knet->kni_list_lock);
 333
 334         net_dev = alloc_netdev(sizeof(struct kni_dev), dev_info.name,
 335 #ifdef NET_NAME_USER
 336                                                         NET_NAME_USER,
 337 #endif
 338                                                         kni_net_init);
 339         if (net_dev == NULL) {
 340                 pr_err("error allocating device \"%s\"\n", dev_info.name);
 341                 return -EBUSY;
 342         }
 343
 344         dev_net_set(net_dev, net);
 345
 346         kni = netdev_priv(net_dev);
 347
 348         kni->net_dev = net_dev;
 349         kni->group_id = dev_info.group_id;
 350         kni->core_id = dev_info.core_id;
 351         strncpy(kni->name, dev_info.name, RTE_KNI_NAMESIZE);
 352
 353         /* Translate user space info into kernel space info */
 354         kni->tx_q = phys_to_virt(dev_info.tx_phys);
 355         kni->rx_q = phys_to_virt(dev_info.rx_phys);
 356         kni->alloc_q = phys_to_virt(dev_info.alloc_phys);
 357         kni->free_q = phys_to_virt(dev_info.free_phys);
 358
 359         kni->req_q = phys_to_virt(dev_info.req_phys);
 360         kni->resp_q = phys_to_virt(dev_info.resp_phys);
 361         kni->sync_va = dev_info.sync_va;
 362         kni->sync_kva = phys_to_virt(dev_info.sync_phys);
 363
 364         kni->mbuf_size = dev_info.mbuf_size;
 365
 366         pr_debug("tx_phys:      0x%016llx, tx_q addr:      0x%p\n",
 367                 (unsigned long long) dev_info.tx_phys, kni->tx_q);
 368         pr_debug("rx_phys:      0x%016llx, rx_q addr:      0x%p\n",
 369                 (unsigned long long) dev_info.rx_phys, kni->rx_q);
 370         pr_debug("alloc_phys:   0x%016llx, alloc_q addr:   0x%p\n",
 371                 (unsigned long long) dev_info.alloc_phys, kni->alloc_q);
 372         pr_debug("free_phys:    0x%016llx, free_q addr:    0x%p\n",
 373                 (unsigned long long) dev_info.free_phys, kni->free_q);
 374         pr_debug("req_phys:     0x%016llx, req_q addr:     0x%p\n",
 375                 (unsigned long long) dev_info.req_phys, kni->req_q);
 376         pr_debug("resp_phys:    0x%016llx, resp_q addr:    0x%p\n",
 377                 (unsigned long long) dev_info.resp_phys, kni->resp_q);
 378         pr_debug("mbuf_size:    %u\n", kni->mbuf_size);
 379
 380         pr_debug("PCI: %02x:%02x.%02x %04x:%04x\n",
 381                                         dev_info.bus,
 382                                         dev_info.devid,
 383                                         dev_info.function,
 384                                         dev_info.vendor_id,
 385                                         dev_info.device_id);
 386         /* if user has provided a valid mac address */
 387         if (is_valid_ether_addr(dev_info.mac_addr))
 388                 memcpy(net_dev->dev_addr, dev_info.mac_addr, ETH_ALEN);
 389         else
 390                 /*
 391                  * Generate random mac address. eth_random_addr() is the
 392                  * newer version of generating mac address in kernel.
 393                  */
 394                 random_ether_addr(net_dev->dev_addr);
 395
 396         if (dev_info.mtu)
 397                 net_dev->mtu = dev_info.mtu;
 398 #ifdef HAVE_MAX_MTU_PARAM
 399         net_dev->max_mtu = net_dev->mtu;
 400 #endif
 401
 402         ret = register_netdev(net_dev);
 403         if (ret) {
 404                 pr_err("error %i registering device \"%s\"\n",
 405                                         ret, dev_info.name);
 406                 kni->net_dev = NULL;
 407                 kni_dev_remove(kni);
 408                 free_netdev(net_dev);
 409                 return -ENODEV;
 410         }
 411
 412         netif_carrier_off(net_dev);
 413
 414         ret = kni_run_thread(knet, kni, dev_info.force_bind);
 415         if (ret != 0)
 416                 return ret;
 417
 418         down_write(&knet->kni_list_lock);
 419         list_add(&kni->list, &knet->kni_list_head);
 420         up_write(&knet->kni_list_lock);
 421
 422         return 0;
 423 }
 424
 425 static int
 426 kni_ioctl_release(struct net *net, uint32_t ioctl_num,
 427                 unsigned long ioctl_param)
 428 {
 429         struct kni_net *knet = net_generic(net, kni_net_id);
 430         int ret = -EINVAL;
 431         struct kni_dev *dev, *n;
 432         struct rte_kni_device_info dev_info;
 433
 434         if (_IOC_SIZE(ioctl_num) > sizeof(dev_info))
 435                 return -EINVAL;
 436
 437         ret = copy_from_user(&dev_info, (void *)ioctl_param, sizeof(dev_info));
 438         if (ret) {
 439                 pr_err("copy_from_user in kni_ioctl_release");
 440                 return -EIO;
 441         }
 442
 443         /* Release the network device according to its name */
 444         if (strlen(dev_info.name) == 0)
 445                 return ret;
 446
 447         down_write(&knet->kni_list_lock);
 448         list_for_each_entry_safe(dev, n, &knet->kni_list_head, list) {
 449                 if (strncmp(dev->name, dev_info.name, RTE_KNI_NAMESIZE) != 0)
 450                         continue;
 451
 452                 if (multiple_kthread_on && dev->pthread != NULL) {
 453                         kthread_stop(dev->pthread);
 454                         dev->pthread = NULL;
 455                 }
 456
 457                 kni_dev_remove(dev);
 458                 list_del(&dev->list);
 459                 ret = 0;
 460                 break;
 461         }
 462         up_write(&knet->kni_list_lock);
 463         pr_info("%s release kni named %s\n",
 464                 (ret == 0 ? "Successfully" : "Unsuccessfully"), dev_info.name);
 465
 466         return ret;
 467 }
 468
 469 static int
 470 kni_ioctl(struct inode *inode, uint32_t ioctl_num, unsigned long ioctl_param)
 471 {
 472         int ret = -EINVAL;
 473         struct net *net = current->nsproxy->net_ns;
 474
 475         pr_debug("IOCTL num=0x%0x param=0x%0lx\n", ioctl_num, ioctl_param);
 476
 477         /*
 478          * Switch according to the ioctl called
 479          */
 480         switch (_IOC_NR(ioctl_num)) {
 481         case _IOC_NR(RTE_KNI_IOCTL_TEST):
 482                 /* For test only, not used */
 483                 break;
 484         case _IOC_NR(RTE_KNI_IOCTL_CREATE):
 485                 ret = kni_ioctl_create(net, ioctl_num, ioctl_param);
 486                 break;
 487         case _IOC_NR(RTE_KNI_IOCTL_RELEASE):
 488                 ret = kni_ioctl_release(net, ioctl_num, ioctl_param);
 489                 break;
 490         default:
 491                 pr_debug("IOCTL default\n");
 492                 break;
 493         }
 494
 495         return ret;
 496 }
 497
 498 static int
 499 kni_compat_ioctl(struct inode *inode, uint32_t ioctl_num,
 500                 unsigned long ioctl_param)
 501 {
 502         /* 32 bits app on 64 bits OS to be supported later */
 503         pr_debug("Not implemented.\n");
 504
 505         return -EINVAL;
 506 }
 507
 508 static const struct file_operations kni_fops = {
 509         .owner = THIS_MODULE,
 510         .open = kni_open,
 511         .release = kni_release,
 512         .unlocked_ioctl = (void *)kni_ioctl,
 513         .compat_ioctl = (void *)kni_compat_ioctl,
 514 };
 515
 516 static struct miscdevice kni_misc = {
 517         .minor = MISC_DYNAMIC_MINOR,
 518         .name = KNI_DEVICE,
 519         .fops = &kni_fops,
 520 };
 521
 522 static int __init
 523 kni_parse_kthread_mode(void)
 524 {
 525         if (!kthread_mode)
 526                 return 0;
 527
 528         if (strcmp(kthread_mode, "single") == 0)
 529                 return 0;
 530         else if (strcmp(kthread_mode, "multiple") == 0)
 531                 multiple_kthread_on = 1;
 532         else
 533                 return -1;
 534
 535         return 0;
 536 }
 537
 538 static int __init
 539 kni_parse_carrier_state(void)
 540 {
 541         if (!carrier) {
 542                 dflt_carrier = 0;
 543                 return 0;
 544         }
 545
 546         if (strcmp(carrier, "off") == 0)
 547                 dflt_carrier = 0;
 548         else if (strcmp(carrier, "on") == 0)
 549                 dflt_carrier = 1;
 550         else
 551                 return -1;
 552
 553         return 0;
 554 }
 555
 556 static int __init
 557 kni_init(void)
 558 {
 559         int rc;
 560
 561         if (kni_parse_kthread_mode() < 0) {
 562                 pr_err("Invalid parameter for kthread_mode\n");
 563                 return -EINVAL;
 564         }
 565
 566         if (multiple_kthread_on == 0)
 567                 pr_debug("Single kernel thread for all KNI devices\n");
 568         else
 569                 pr_debug("Multiple kernel thread mode enabled\n");
 570
 571         if (kni_parse_carrier_state() < 0) {
 572                 pr_err("Invalid parameter for carrier\n");
 573                 return -EINVAL;
 574         }
 575
 576         if (dflt_carrier == 0)
 577                 pr_debug("Default carrier state set to off.\n");
 578         else
 579                 pr_debug("Default carrier state set to on.\n");
 580
 581 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
 582         rc = register_pernet_subsys(&kni_net_ops);
 583 #else
 584         rc = register_pernet_gen_subsys(&kni_net_id, &kni_net_ops);
 585 #endif
 586         if (rc)
 587                 return -EPERM;
 588
 589         rc = misc_register(&kni_misc);
 590         if (rc != 0) {
 591                 pr_err("Misc registration failed\n");
 592                 goto out;
 593         }
 594
 595         /* Configure the lo mode according to the input parameter */
 596         kni_net_config_lo_mode(lo_mode);
 597
 598         return 0;
 599
 600 out:
 601 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
 602         unregister_pernet_subsys(&kni_net_ops);
 603 #else
 604         unregister_pernet_gen_subsys(kni_net_id, &kni_net_ops);
 605 #endif
 606         return rc;
 607 }
 608
 609 static void __exit
 610 kni_exit(void)
 611 {
 612         misc_deregister(&kni_misc);
 613 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS
 614         unregister_pernet_subsys(&kni_net_ops);
 615 #else
 616         unregister_pernet_gen_subsys(kni_net_id, &kni_net_ops);
 617 #endif
 618 }
 619
 620 module_init(kni_init);
 621 module_exit(kni_exit);
 622
 623 module_param(lo_mode, charp, 0644);
 624 MODULE_PARM_DESC(lo_mode,
 625 "KNI loopback mode (default=lo_mode_none):\n"
 626 "\t\tlo_mode_none        Kernel loopback disabled\n"
 627 "\t\tlo_mode_fifo        Enable kernel loopback with fifo\n"
 628 "\t\tlo_mode_fifo_skb    Enable kernel loopback with fifo and skb buffer\n"
 629 "\t\t"
 630 );
 631
 632 module_param(kthread_mode, charp, 0644);
 633 MODULE_PARM_DESC(kthread_mode,
 634 "Kernel thread mode (default=single):\n"
 635 "\t\tsingle    Single kernel thread mode enabled.\n"
 636 "\t\tmultiple  Multiple kernel thread mode enabled.\n"
 637 "\t\t"
 638 );
 639
 640 module_param(carrier, charp, 0644);
 641 MODULE_PARM_DESC(carrier,
 642 "Default carrier state for KNI interface (default=off):\n"
 643 "\t\toff   Interfaces will be created with carrier state set to off.\n"
 644 "\t\ton    Interfaces will be created with carrier state set to on.\n"
 645 "\t\t"
 646 );