net/devlink/health.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * Copyright (c) 2016 Mellanox Technologies. All rights reserved.
   4  * Copyright (c) 2016 Jiri Pirko <jiri@mellanox.com>
   5  */
   6
   7 #include <net/genetlink.h>
   8 #include <net/sock.h>
   9 #include <trace/events/devlink.h>
  10 #include "devl_internal.h"
  11
  12 struct devlink_fmsg_item {
  13         struct list_head list;
  14         int attrtype;
  15         u8 nla_type;
  16         u16 len;
  17         int value[];
  18 };
  19
  20 struct devlink_fmsg {
  21         struct list_head item_list;
  22         int err; /* first error encountered on some devlink_fmsg_XXX() call */
  23         bool putting_binary; /* This flag forces enclosing of binary data
  24                               * in an array brackets. It forces using
  25                               * of designated API:
  26                               * devlink_fmsg_binary_pair_nest_start()
  27                               * devlink_fmsg_binary_pair_nest_end()
  28                               */
  29 };
  30
  31 static struct devlink_fmsg *devlink_fmsg_alloc(void)
  32 {
  33         struct devlink_fmsg *fmsg;
  34
  35         fmsg = kzalloc(sizeof(*fmsg), GFP_KERNEL);
  36         if (!fmsg)
  37                 return NULL;
  38
  39         INIT_LIST_HEAD(&fmsg->item_list);
  40
  41         return fmsg;
  42 }
  43
  44 static void devlink_fmsg_free(struct devlink_fmsg *fmsg)
  45 {
  46         struct devlink_fmsg_item *item, *tmp;
  47
  48         list_for_each_entry_safe(item, tmp, &fmsg->item_list, list) {
  49                 list_del(&item->list);
  50                 kfree(item);
  51         }
  52         kfree(fmsg);
  53 }
  54
  55 struct devlink_health_reporter {
  56         struct list_head list;
  57         void *priv;
  58         const struct devlink_health_reporter_ops *ops;
  59         struct devlink *devlink;
  60         struct devlink_port *devlink_port;
  61         struct devlink_fmsg *dump_fmsg;
  62         u64 graceful_period;
  63         bool auto_recover;
  64         bool auto_dump;
  65         u8 health_state;
  66         u64 dump_ts;
  67         u64 dump_real_ts;
  68         u64 error_count;
  69         u64 recovery_count;
  70         u64 last_recovery_ts;
  71 };
  72
  73 void *
  74 devlink_health_reporter_priv(struct devlink_health_reporter *reporter)
  75 {
  76         return reporter->priv;
  77 }
  78 EXPORT_SYMBOL_GPL(devlink_health_reporter_priv);
  79
  80 static struct devlink_health_reporter *
  81 __devlink_health_reporter_find_by_name(struct list_head *reporter_list,
  82                                        const char *reporter_name)
  83 {
  84         struct devlink_health_reporter *reporter;
  85
  86         list_for_each_entry(reporter, reporter_list, list)
  87                 if (!strcmp(reporter->ops->name, reporter_name))
  88                         return reporter;
  89         return NULL;
  90 }
  91
  92 static struct devlink_health_reporter *
  93 devlink_health_reporter_find_by_name(struct devlink *devlink,
  94                                      const char *reporter_name)
  95 {
  96         return __devlink_health_reporter_find_by_name(&devlink->reporter_list,
  97                                                       reporter_name);
  98 }
  99
 100 static struct devlink_health_reporter *
 101 devlink_port_health_reporter_find_by_name(struct devlink_port *devlink_port,
 102                                           const char *reporter_name)
 103 {
 104         return __devlink_health_reporter_find_by_name(&devlink_port->reporter_list,
 105                                                       reporter_name);
 106 }
 107
 108 static struct devlink_health_reporter *
 109 __devlink_health_reporter_create(struct devlink *devlink,
 110                                  const struct devlink_health_reporter_ops *ops,
 111                                  u64 graceful_period, void *priv)
 112 {
 113         struct devlink_health_reporter *reporter;
 114
 115         if (WARN_ON(graceful_period && !ops->recover))
 116                 return ERR_PTR(-EINVAL);
 117
 118         reporter = kzalloc(sizeof(*reporter), GFP_KERNEL);
 119         if (!reporter)
 120                 return ERR_PTR(-ENOMEM);
 121
 122         reporter->priv = priv;
 123         reporter->ops = ops;
 124         reporter->devlink = devlink;
 125         reporter->graceful_period = graceful_period;
 126         reporter->auto_recover = !!ops->recover;
 127         reporter->auto_dump = !!ops->dump;
 128         return reporter;
 129 }
 130
 131 /**
 132  * devl_port_health_reporter_create() - create devlink health reporter for
 133  *                                      specified port instance
 134  *
 135  * @port: devlink_port to which health reports will relate
 136  * @ops: devlink health reporter ops
 137  * @graceful_period: min time (in msec) between recovery attempts
 138  * @priv: driver priv pointer
 139  */
 140 struct devlink_health_reporter *
 141 devl_port_health_reporter_create(struct devlink_port *port,
 142                                  const struct devlink_health_reporter_ops *ops,
 143                                  u64 graceful_period, void *priv)
 144 {
 145         struct devlink_health_reporter *reporter;
 146
 147         devl_assert_locked(port->devlink);
 148
 149         if (__devlink_health_reporter_find_by_name(&port->reporter_list,
 150                                                    ops->name))
 151                 return ERR_PTR(-EEXIST);
 152
 153         reporter = __devlink_health_reporter_create(port->devlink, ops,
 154                                                     graceful_period, priv);
 155         if (IS_ERR(reporter))
 156                 return reporter;
 157
 158         reporter->devlink_port = port;
 159         list_add_tail(&reporter->list, &port->reporter_list);
 160         return reporter;
 161 }
 162 EXPORT_SYMBOL_GPL(devl_port_health_reporter_create);
 163
 164 struct devlink_health_reporter *
 165 devlink_port_health_reporter_create(struct devlink_port *port,
 166                                     const struct devlink_health_reporter_ops *ops,
 167                                     u64 graceful_period, void *priv)
 168 {
 169         struct devlink_health_reporter *reporter;
 170         struct devlink *devlink = port->devlink;
 171
 172         devl_lock(devlink);
 173         reporter = devl_port_health_reporter_create(port, ops,
 174                                                     graceful_period, priv);
 175         devl_unlock(devlink);
 176         return reporter;
 177 }
 178 EXPORT_SYMBOL_GPL(devlink_port_health_reporter_create);
 179
 180 /**
 181  * devl_health_reporter_create - create devlink health reporter
 182  *
 183  * @devlink: devlink instance which the health reports will relate
 184  * @ops: devlink health reporter ops
 185  * @graceful_period: min time (in msec) between recovery attempts
 186  * @priv: driver priv pointer
 187  */
 188 struct devlink_health_reporter *
 189 devl_health_reporter_create(struct devlink *devlink,
 190                             const struct devlink_health_reporter_ops *ops,
 191                             u64 graceful_period, void *priv)
 192 {
 193         struct devlink_health_reporter *reporter;
 194
 195         devl_assert_locked(devlink);
 196
 197         if (devlink_health_reporter_find_by_name(devlink, ops->name))
 198                 return ERR_PTR(-EEXIST);
 199
 200         reporter = __devlink_health_reporter_create(devlink, ops,
 201                                                     graceful_period, priv);
 202         if (IS_ERR(reporter))
 203                 return reporter;
 204
 205         list_add_tail(&reporter->list, &devlink->reporter_list);
 206         return reporter;
 207 }
 208 EXPORT_SYMBOL_GPL(devl_health_reporter_create);
 209
 210 struct devlink_health_reporter *
 211 devlink_health_reporter_create(struct devlink *devlink,
 212                                const struct devlink_health_reporter_ops *ops,
 213                                u64 graceful_period, void *priv)
 214 {
 215         struct devlink_health_reporter *reporter;
 216
 217         devl_lock(devlink);
 218         reporter = devl_health_reporter_create(devlink, ops,
 219                                                graceful_period, priv);
 220         devl_unlock(devlink);
 221         return reporter;
 222 }
 223 EXPORT_SYMBOL_GPL(devlink_health_reporter_create);
 224
 225 static void
 226 devlink_health_reporter_free(struct devlink_health_reporter *reporter)
 227 {
 228         if (reporter->dump_fmsg)
 229                 devlink_fmsg_free(reporter->dump_fmsg);
 230         kfree(reporter);
 231 }
 232
 233 /**
 234  * devl_health_reporter_destroy() - destroy devlink health reporter
 235  *
 236  * @reporter: devlink health reporter to destroy
 237  */
 238 void
 239 devl_health_reporter_destroy(struct devlink_health_reporter *reporter)
 240 {
 241         devl_assert_locked(reporter->devlink);
 242
 243         list_del(&reporter->list);
 244         devlink_health_reporter_free(reporter);
 245 }
 246 EXPORT_SYMBOL_GPL(devl_health_reporter_destroy);
 247
 248 void
 249 devlink_health_reporter_destroy(struct devlink_health_reporter *reporter)
 250 {
 251         struct devlink *devlink = reporter->devlink;
 252
 253         devl_lock(devlink);
 254         devl_health_reporter_destroy(reporter);
 255         devl_unlock(devlink);
 256 }
 257 EXPORT_SYMBOL_GPL(devlink_health_reporter_destroy);
 258
 259 static int
 260 devlink_nl_health_reporter_fill(struct sk_buff *msg,
 261                                 struct devlink_health_reporter *reporter,
 262                                 enum devlink_command cmd, u32 portid,
 263                                 u32 seq, int flags)
 264 {
 265         struct devlink *devlink = reporter->devlink;
 266         struct nlattr *reporter_attr;
 267         void *hdr;
 268
 269         hdr = genlmsg_put(msg, portid, seq, &devlink_nl_family, flags, cmd);
 270         if (!hdr)
 271                 return -EMSGSIZE;
 272
 273         if (devlink_nl_put_handle(msg, devlink))
 274                 goto genlmsg_cancel;
 275
 276         if (reporter->devlink_port) {
 277                 if (nla_put_u32(msg, DEVLINK_ATTR_PORT_INDEX, reporter->devlink_port->index))
 278                         goto genlmsg_cancel;
 279         }
 280         reporter_attr = nla_nest_start_noflag(msg,
 281                                               DEVLINK_ATTR_HEALTH_REPORTER);
 282         if (!reporter_attr)
 283                 goto genlmsg_cancel;
 284         if (nla_put_string(msg, DEVLINK_ATTR_HEALTH_REPORTER_NAME,
 285                            reporter->ops->name))
 286                 goto reporter_nest_cancel;
 287         if (nla_put_u8(msg, DEVLINK_ATTR_HEALTH_REPORTER_STATE,
 288                        reporter->health_state))
 289                 goto reporter_nest_cancel;
 290         if (devlink_nl_put_u64(msg, DEVLINK_ATTR_HEALTH_REPORTER_ERR_COUNT,
 291                                reporter->error_count))
 292                 goto reporter_nest_cancel;
 293         if (devlink_nl_put_u64(msg, DEVLINK_ATTR_HEALTH_REPORTER_RECOVER_COUNT,
 294                                reporter->recovery_count))
 295                 goto reporter_nest_cancel;
 296         if (reporter->ops->recover &&
 297             devlink_nl_put_u64(msg, DEVLINK_ATTR_HEALTH_REPORTER_GRACEFUL_PERIOD,
 298                                reporter->graceful_period))
 299                 goto reporter_nest_cancel;
 300         if (reporter->ops->recover &&
 301             nla_put_u8(msg, DEVLINK_ATTR_HEALTH_REPORTER_AUTO_RECOVER,
 302                        reporter->auto_recover))
 303                 goto reporter_nest_cancel;
 304         if (reporter->dump_fmsg &&
 305             devlink_nl_put_u64(msg, DEVLINK_ATTR_HEALTH_REPORTER_DUMP_TS,
 306                                jiffies_to_msecs(reporter->dump_ts)))
 307                 goto reporter_nest_cancel;
 308         if (reporter->dump_fmsg &&
 309             devlink_nl_put_u64(msg, DEVLINK_ATTR_HEALTH_REPORTER_DUMP_TS_NS,
 310                                reporter->dump_real_ts))
 311                 goto reporter_nest_cancel;
 312         if (reporter->ops->dump &&
 313             nla_put_u8(msg, DEVLINK_ATTR_HEALTH_REPORTER_AUTO_DUMP,
 314                        reporter->auto_dump))
 315                 goto reporter_nest_cancel;
 316
 317         nla_nest_end(msg, reporter_attr);
 318         genlmsg_end(msg, hdr);
 319         return 0;
 320
 321 reporter_nest_cancel:
 322         nla_nest_cancel(msg, reporter_attr);
 323 genlmsg_cancel:
 324         genlmsg_cancel(msg, hdr);
 325         return -EMSGSIZE;
 326 }
 327
 328 static struct devlink_health_reporter *
 329 devlink_health_reporter_get_from_attrs(struct devlink *devlink,
 330                                        struct nlattr **attrs)
 331 {
 332         struct devlink_port *devlink_port;
 333         char *reporter_name;
 334
 335         if (!attrs[DEVLINK_ATTR_HEALTH_REPORTER_NAME])
 336                 return NULL;
 337
 338         reporter_name = nla_data(attrs[DEVLINK_ATTR_HEALTH_REPORTER_NAME]);
 339         devlink_port = devlink_port_get_from_attrs(devlink, attrs);
 340         if (IS_ERR(devlink_port))
 341                 return devlink_health_reporter_find_by_name(devlink,
 342                                                             reporter_name);
 343         else
 344                 return devlink_port_health_reporter_find_by_name(devlink_port,
 345                                                                  reporter_name);
 346 }
 347
 348 static struct devlink_health_reporter *
 349 devlink_health_reporter_get_from_info(struct devlink *devlink,
 350                                       struct genl_info *info)
 351 {
 352         return devlink_health_reporter_get_from_attrs(devlink, info->attrs);
 353 }
 354
 355 int devlink_nl_health_reporter_get_doit(struct sk_buff *skb,
 356                                         struct genl_info *info)
 357 {
 358         struct devlink *devlink = info->user_ptr[0];
 359         struct devlink_health_reporter *reporter;
 360         struct sk_buff *msg;
 361         int err;
 362
 363         reporter = devlink_health_reporter_get_from_info(devlink, info);
 364         if (!reporter)
 365                 return -EINVAL;
 366
 367         msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
 368         if (!msg)
 369                 return -ENOMEM;
 370
 371         err = devlink_nl_health_reporter_fill(msg, reporter,
 372                                               DEVLINK_CMD_HEALTH_REPORTER_GET,
 373                                               info->snd_portid, info->snd_seq,
 374                                               0);
 375         if (err) {
 376                 nlmsg_free(msg);
 377                 return err;
 378         }
 379
 380         return genlmsg_reply(msg, info);
 381 }
 382
 383 static int devlink_nl_health_reporter_get_dump_one(struct sk_buff *msg,
 384                                                    struct devlink *devlink,
 385                                                    struct netlink_callback *cb,
 386                                                    int flags)
 387 {
 388         struct devlink_nl_dump_state *state = devlink_dump_state(cb);
 389         const struct genl_info *info = genl_info_dump(cb);
 390         struct devlink_health_reporter *reporter;
 391         unsigned long port_index_end = ULONG_MAX;
 392         struct nlattr **attrs = info->attrs;
 393         unsigned long port_index_start = 0;
 394         struct devlink_port *port;
 395         unsigned long port_index;
 396         int idx = 0;
 397         int err;
 398
 399         if (attrs && attrs[DEVLINK_ATTR_PORT_INDEX]) {
 400                 port_index_start = nla_get_u32(attrs[DEVLINK_ATTR_PORT_INDEX]);
 401                 port_index_end = port_index_start;
 402                 flags |= NLM_F_DUMP_FILTERED;
 403                 goto per_port_dump;
 404         }
 405
 406         list_for_each_entry(reporter, &devlink->reporter_list, list) {
 407                 if (idx < state->idx) {
 408                         idx++;
 409                         continue;
 410                 }
 411                 err = devlink_nl_health_reporter_fill(msg, reporter,
 412                                                       DEVLINK_CMD_HEALTH_REPORTER_GET,
 413                                                       NETLINK_CB(cb->skb).portid,
 414                                                       cb->nlh->nlmsg_seq,
 415                                                       flags);
 416                 if (err) {
 417                         state->idx = idx;
 418                         return err;
 419                 }
 420                 idx++;
 421         }
 422 per_port_dump:
 423         xa_for_each_range(&devlink->ports, port_index, port,
 424                           port_index_start, port_index_end) {
 425                 list_for_each_entry(reporter, &port->reporter_list, list) {
 426                         if (idx < state->idx) {
 427                                 idx++;
 428                                 continue;
 429                         }
 430                         err = devlink_nl_health_reporter_fill(msg, reporter,
 431                                                               DEVLINK_CMD_HEALTH_REPORTER_GET,
 432                                                               NETLINK_CB(cb->skb).portid,
 433                                                               cb->nlh->nlmsg_seq,
 434                                                               flags);
 435                         if (err) {
 436                                 state->idx = idx;
 437                                 return err;
 438                         }
 439                         idx++;
 440                 }
 441         }
 442
 443         return 0;
 444 }
 445
 446 int devlink_nl_health_reporter_get_dumpit(struct sk_buff *skb,
 447                                           struct netlink_callback *cb)
 448 {
 449         return devlink_nl_dumpit(skb, cb,
 450                                  devlink_nl_health_reporter_get_dump_one);
 451 }
 452
 453 int devlink_nl_health_reporter_set_doit(struct sk_buff *skb,
 454                                         struct genl_info *info)
 455 {
 456         struct devlink *devlink = info->user_ptr[0];
 457         struct devlink_health_reporter *reporter;
 458
 459         reporter = devlink_health_reporter_get_from_info(devlink, info);
 460         if (!reporter)
 461                 return -EINVAL;
 462
 463         if (!reporter->ops->recover &&
 464             (info->attrs[DEVLINK_ATTR_HEALTH_REPORTER_GRACEFUL_PERIOD] ||
 465              info->attrs[DEVLINK_ATTR_HEALTH_REPORTER_AUTO_RECOVER]))
 466                 return -EOPNOTSUPP;
 467
 468         if (!reporter->ops->dump &&
 469             info->attrs[DEVLINK_ATTR_HEALTH_REPORTER_AUTO_DUMP])
 470                 return -EOPNOTSUPP;
 471
 472         if (info->attrs[DEVLINK_ATTR_HEALTH_REPORTER_GRACEFUL_PERIOD])
 473                 reporter->graceful_period =
 474                         nla_get_u64(info->attrs[DEVLINK_ATTR_HEALTH_REPORTER_GRACEFUL_PERIOD]);
 475
 476         if (info->attrs[DEVLINK_ATTR_HEALTH_REPORTER_AUTO_RECOVER])
 477                 reporter->auto_recover =
 478                         nla_get_u8(info->attrs[DEVLINK_ATTR_HEALTH_REPORTER_AUTO_RECOVER]);
 479
 480         if (info->attrs[DEVLINK_ATTR_HEALTH_REPORTER_AUTO_DUMP])
 481                 reporter->auto_dump =
 482                 nla_get_u8(info->attrs[DEVLINK_ATTR_HEALTH_REPORTER_AUTO_DUMP]);
 483
 484         return 0;
 485 }
 486
 487 static void devlink_recover_notify(struct devlink_health_reporter *reporter,
 488                                    enum devlink_command cmd)
 489 {
 490         struct devlink *devlink = reporter->devlink;
 491         struct devlink_obj_desc desc;
 492         struct sk_buff *msg;
 493         int err;
 494
 495         WARN_ON(cmd != DEVLINK_CMD_HEALTH_REPORTER_RECOVER);
 496         ASSERT_DEVLINK_REGISTERED(devlink);
 497
 498         if (!devlink_nl_notify_need(devlink))
 499                 return;
 500
 501         msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
 502         if (!msg)
 503                 return;
 504
 505         err = devlink_nl_health_reporter_fill(msg, reporter, cmd, 0, 0, 0);
 506         if (err) {
 507                 nlmsg_free(msg);
 508                 return;
 509         }
 510
 511         devlink_nl_obj_desc_init(&desc, devlink);
 512         if (reporter->devlink_port)
 513                 devlink_nl_obj_desc_port_set(&desc, reporter->devlink_port);
 514         devlink_nl_notify_send_desc(devlink, msg, &desc);
 515 }
 516
 517 void
 518 devlink_health_reporter_recovery_done(struct devlink_health_reporter *reporter)
 519 {
 520         reporter->recovery_count++;
 521         reporter->last_recovery_ts = jiffies;
 522 }
 523 EXPORT_SYMBOL_GPL(devlink_health_reporter_recovery_done);
 524
 525 static int
 526 devlink_health_reporter_recover(struct devlink_health_reporter *reporter,
 527                                 void *priv_ctx, struct netlink_ext_ack *extack)
 528 {
 529         int err;
 530
 531         if (reporter->health_state == DEVLINK_HEALTH_REPORTER_STATE_HEALTHY)
 532                 return 0;
 533
 534         if (!reporter->ops->recover)
 535                 return -EOPNOTSUPP;
 536
 537         err = reporter->ops->recover(reporter, priv_ctx, extack);
 538         if (err)
 539                 return err;
 540
 541         devlink_health_reporter_recovery_done(reporter);
 542         reporter->health_state = DEVLINK_HEALTH_REPORTER_STATE_HEALTHY;
 543         devlink_recover_notify(reporter, DEVLINK_CMD_HEALTH_REPORTER_RECOVER);
 544
 545         return 0;
 546 }
 547
 548 static void
 549 devlink_health_dump_clear(struct devlink_health_reporter *reporter)
 550 {
 551         if (!reporter->dump_fmsg)
 552                 return;
 553         devlink_fmsg_free(reporter->dump_fmsg);
 554         reporter->dump_fmsg = NULL;
 555 }
 556
 557 static int devlink_health_do_dump(struct devlink_health_reporter *reporter,
 558                                   void *priv_ctx,
 559                                   struct netlink_ext_ack *extack)
 560 {
 561         int err;
 562
 563         if (!reporter->ops->dump)
 564                 return 0;
 565
 566         if (reporter->dump_fmsg)
 567                 return 0;
 568
 569         reporter->dump_fmsg = devlink_fmsg_alloc();
 570         if (!reporter->dump_fmsg)
 571                 return -ENOMEM;
 572
 573         devlink_fmsg_obj_nest_start(reporter->dump_fmsg);
 574
 575         err = reporter->ops->dump(reporter, reporter->dump_fmsg,
 576                                   priv_ctx, extack);
 577         if (err)
 578                 goto dump_err;
 579
 580         devlink_fmsg_obj_nest_end(reporter->dump_fmsg);
 581         err = reporter->dump_fmsg->err;
 582         if (err)
 583                 goto dump_err;
 584
 585         reporter->dump_ts = jiffies;
 586         reporter->dump_real_ts = ktime_get_real_ns();
 587
 588         return 0;
 589
 590 dump_err:
 591         devlink_health_dump_clear(reporter);
 592         return err;
 593 }
 594
 595 int devlink_health_report(struct devlink_health_reporter *reporter,
 596                           const char *msg, void *priv_ctx)
 597 {
 598         enum devlink_health_reporter_state prev_health_state;
 599         struct devlink *devlink = reporter->devlink;
 600         unsigned long recover_ts_threshold;
 601         int ret;
 602
 603         /* write a log message of the current error */
 604         WARN_ON(!msg);
 605         trace_devlink_health_report(devlink, reporter->ops->name, msg);
 606         reporter->error_count++;
 607         prev_health_state = reporter->health_state;
 608         reporter->health_state = DEVLINK_HEALTH_REPORTER_STATE_ERROR;
 609         devlink_recover_notify(reporter, DEVLINK_CMD_HEALTH_REPORTER_RECOVER);
 610
 611         /* abort if the previous error wasn't recovered */
 612         recover_ts_threshold = reporter->last_recovery_ts +
 613                                msecs_to_jiffies(reporter->graceful_period);
 614         if (reporter->auto_recover &&
 615             (prev_health_state != DEVLINK_HEALTH_REPORTER_STATE_HEALTHY ||
 616              (reporter->last_recovery_ts && reporter->recovery_count &&
 617               time_is_after_jiffies(recover_ts_threshold)))) {
 618                 trace_devlink_health_recover_aborted(devlink,
 619                                                      reporter->ops->name,
 620                                                      reporter->health_state,
 621                                                      jiffies -
 622                                                      reporter->last_recovery_ts);
 623                 return -ECANCELED;
 624         }
 625
 626         if (reporter->auto_dump) {
 627                 devl_lock(devlink);
 628                 /* store current dump of current error, for later analysis */
 629                 devlink_health_do_dump(reporter, priv_ctx, NULL);
 630                 devl_unlock(devlink);
 631         }
 632
 633         if (!reporter->auto_recover)
 634                 return 0;
 635
 636         devl_lock(devlink);
 637         ret = devlink_health_reporter_recover(reporter, priv_ctx, NULL);
 638         devl_unlock(devlink);
 639
 640         return ret;
 641 }
 642 EXPORT_SYMBOL_GPL(devlink_health_report);
 643
 644 void
 645 devlink_health_reporter_state_update(struct devlink_health_reporter *reporter,
 646                                      enum devlink_health_reporter_state state)
 647 {
 648         if (WARN_ON(state != DEVLINK_HEALTH_REPORTER_STATE_HEALTHY &&
 649                     state != DEVLINK_HEALTH_REPORTER_STATE_ERROR))
 650                 return;
 651
 652         if (reporter->health_state == state)
 653                 return;
 654
 655         reporter->health_state = state;
 656         trace_devlink_health_reporter_state_update(reporter->devlink,
 657                                                    reporter->ops->name, state);
 658         devlink_recover_notify(reporter, DEVLINK_CMD_HEALTH_REPORTER_RECOVER);
 659 }
 660 EXPORT_SYMBOL_GPL(devlink_health_reporter_state_update);
 661
 662 int devlink_nl_health_reporter_recover_doit(struct sk_buff *skb,
 663                                             struct genl_info *info)
 664 {
 665         struct devlink *devlink = info->user_ptr[0];
 666         struct devlink_health_reporter *reporter;
 667
 668         reporter = devlink_health_reporter_get_from_info(devlink, info);
 669         if (!reporter)
 670                 return -EINVAL;
 671
 672         return devlink_health_reporter_recover(reporter, NULL, info->extack);
 673 }
 674
 675 static void devlink_fmsg_err_if_binary(struct devlink_fmsg *fmsg)
 676 {
 677         if (!fmsg->err && fmsg->putting_binary)
 678                 fmsg->err = -EINVAL;
 679 }
 680
 681 static void devlink_fmsg_nest_common(struct devlink_fmsg *fmsg, int attrtype)
 682 {
 683         struct devlink_fmsg_item *item;
 684
 685         if (fmsg->err)
 686                 return;
 687
 688         item = kzalloc(sizeof(*item), GFP_KERNEL);
 689         if (!item) {
 690                 fmsg->err = -ENOMEM;
 691                 return;
 692         }
 693
 694         item->attrtype = attrtype;
 695         list_add_tail(&item->list, &fmsg->item_list);
 696 }
 697
 698 void devlink_fmsg_obj_nest_start(struct devlink_fmsg *fmsg)
 699 {
 700         devlink_fmsg_err_if_binary(fmsg);
 701         devlink_fmsg_nest_common(fmsg, DEVLINK_ATTR_FMSG_OBJ_NEST_START);
 702 }
 703 EXPORT_SYMBOL_GPL(devlink_fmsg_obj_nest_start);
 704
 705 static void devlink_fmsg_nest_end(struct devlink_fmsg *fmsg)
 706 {
 707         devlink_fmsg_err_if_binary(fmsg);
 708         devlink_fmsg_nest_common(fmsg, DEVLINK_ATTR_FMSG_NEST_END);
 709 }
 710
 711 void devlink_fmsg_obj_nest_end(struct devlink_fmsg *fmsg)
 712 {
 713         devlink_fmsg_nest_end(fmsg);
 714 }
 715 EXPORT_SYMBOL_GPL(devlink_fmsg_obj_nest_end);
 716
 717 #define DEVLINK_FMSG_MAX_SIZE (GENLMSG_DEFAULT_SIZE - GENL_HDRLEN - NLA_HDRLEN)
 718
 719 static void devlink_fmsg_put_name(struct devlink_fmsg *fmsg, const char *name)
 720 {
 721         struct devlink_fmsg_item *item;
 722
 723         devlink_fmsg_err_if_binary(fmsg);
 724         if (fmsg->err)
 725                 return;
 726
 727         if (strlen(name) + 1 > DEVLINK_FMSG_MAX_SIZE) {
 728                 fmsg->err = -EMSGSIZE;
 729                 return;
 730         }
 731
 732         item = kzalloc(sizeof(*item) + strlen(name) + 1, GFP_KERNEL);
 733         if (!item) {
 734                 fmsg->err = -ENOMEM;
 735                 return;
 736         }
 737
 738         item->nla_type = NLA_NUL_STRING;
 739         item->len = strlen(name) + 1;
 740         item->attrtype = DEVLINK_ATTR_FMSG_OBJ_NAME;
 741         memcpy(&item->value, name, item->len);
 742         list_add_tail(&item->list, &fmsg->item_list);
 743 }
 744
 745 void devlink_fmsg_pair_nest_start(struct devlink_fmsg *fmsg, const char *name)
 746 {
 747         devlink_fmsg_err_if_binary(fmsg);
 748         devlink_fmsg_nest_common(fmsg, DEVLINK_ATTR_FMSG_PAIR_NEST_START);
 749         devlink_fmsg_put_name(fmsg, name);
 750 }
 751 EXPORT_SYMBOL_GPL(devlink_fmsg_pair_nest_start);
 752
 753 void devlink_fmsg_pair_nest_end(struct devlink_fmsg *fmsg)
 754 {
 755         devlink_fmsg_nest_end(fmsg);
 756 }
 757 EXPORT_SYMBOL_GPL(devlink_fmsg_pair_nest_end);
 758
 759 void devlink_fmsg_arr_pair_nest_start(struct devlink_fmsg *fmsg,
 760                                       const char *name)
 761 {
 762         devlink_fmsg_pair_nest_start(fmsg, name);
 763         devlink_fmsg_nest_common(fmsg, DEVLINK_ATTR_FMSG_ARR_NEST_START);
 764 }
 765 EXPORT_SYMBOL_GPL(devlink_fmsg_arr_pair_nest_start);
 766
 767 void devlink_fmsg_arr_pair_nest_end(struct devlink_fmsg *fmsg)
 768 {
 769         devlink_fmsg_nest_end(fmsg);
 770         devlink_fmsg_nest_end(fmsg);
 771 }
 772 EXPORT_SYMBOL_GPL(devlink_fmsg_arr_pair_nest_end);
 773
 774 void devlink_fmsg_binary_pair_nest_start(struct devlink_fmsg *fmsg,
 775                                          const char *name)
 776 {
 777         devlink_fmsg_arr_pair_nest_start(fmsg, name);
 778         fmsg->putting_binary = true;
 779 }
 780 EXPORT_SYMBOL_GPL(devlink_fmsg_binary_pair_nest_start);
 781
 782 void devlink_fmsg_binary_pair_nest_end(struct devlink_fmsg *fmsg)
 783 {
 784         if (fmsg->err)
 785                 return;
 786
 787         if (!fmsg->putting_binary)
 788                 fmsg->err = -EINVAL;
 789
 790         fmsg->putting_binary = false;
 791         devlink_fmsg_arr_pair_nest_end(fmsg);
 792 }
 793 EXPORT_SYMBOL_GPL(devlink_fmsg_binary_pair_nest_end);
 794
 795 static void devlink_fmsg_put_value(struct devlink_fmsg *fmsg,
 796                                    const void *value, u16 value_len,
 797                                    u8 value_nla_type)
 798 {
 799         struct devlink_fmsg_item *item;
 800
 801         if (fmsg->err)
 802                 return;
 803
 804         if (value_len > DEVLINK_FMSG_MAX_SIZE) {
 805                 fmsg->err = -EMSGSIZE;
 806                 return;
 807         }
 808
 809         item = kzalloc(sizeof(*item) + value_len, GFP_KERNEL);
 810         if (!item) {
 811                 fmsg->err = -ENOMEM;
 812                 return;
 813         }
 814
 815         item->nla_type = value_nla_type;
 816         item->len = value_len;
 817         item->attrtype = DEVLINK_ATTR_FMSG_OBJ_VALUE_DATA;
 818         memcpy(&item->value, value, item->len);
 819         list_add_tail(&item->list, &fmsg->item_list);
 820 }
 821
 822 static void devlink_fmsg_bool_put(struct devlink_fmsg *fmsg, bool value)
 823 {
 824         devlink_fmsg_err_if_binary(fmsg);
 825         devlink_fmsg_put_value(fmsg, &value, sizeof(value), NLA_FLAG);
 826 }
 827
 828 static void devlink_fmsg_u8_put(struct devlink_fmsg *fmsg, u8 value)
 829 {
 830         devlink_fmsg_err_if_binary(fmsg);
 831         devlink_fmsg_put_value(fmsg, &value, sizeof(value), NLA_U8);
 832 }
 833
 834 void devlink_fmsg_u32_put(struct devlink_fmsg *fmsg, u32 value)
 835 {
 836         devlink_fmsg_err_if_binary(fmsg);
 837         devlink_fmsg_put_value(fmsg, &value, sizeof(value), NLA_U32);
 838 }
 839 EXPORT_SYMBOL_GPL(devlink_fmsg_u32_put);
 840
 841 static void devlink_fmsg_u64_put(struct devlink_fmsg *fmsg, u64 value)
 842 {
 843         devlink_fmsg_err_if_binary(fmsg);
 844         devlink_fmsg_put_value(fmsg, &value, sizeof(value), NLA_U64);
 845 }
 846
 847 void devlink_fmsg_string_put(struct devlink_fmsg *fmsg, const char *value)
 848 {
 849         devlink_fmsg_err_if_binary(fmsg);
 850         devlink_fmsg_put_value(fmsg, value, strlen(value) + 1, NLA_NUL_STRING);
 851 }
 852 EXPORT_SYMBOL_GPL(devlink_fmsg_string_put);
 853
 854 void devlink_fmsg_binary_put(struct devlink_fmsg *fmsg, const void *value,
 855                              u16 value_len)
 856 {
 857         if (!fmsg->err && !fmsg->putting_binary)
 858                 fmsg->err = -EINVAL;
 859
 860         devlink_fmsg_put_value(fmsg, value, value_len, NLA_BINARY);
 861 }
 862 EXPORT_SYMBOL_GPL(devlink_fmsg_binary_put);
 863
 864 void devlink_fmsg_bool_pair_put(struct devlink_fmsg *fmsg, const char *name,
 865                                 bool value)
 866 {
 867         devlink_fmsg_pair_nest_start(fmsg, name);
 868         devlink_fmsg_bool_put(fmsg, value);
 869         devlink_fmsg_pair_nest_end(fmsg);
 870 }
 871 EXPORT_SYMBOL_GPL(devlink_fmsg_bool_pair_put);
 872
 873 void devlink_fmsg_u8_pair_put(struct devlink_fmsg *fmsg, const char *name,
 874                               u8 value)
 875 {
 876         devlink_fmsg_pair_nest_start(fmsg, name);
 877         devlink_fmsg_u8_put(fmsg, value);
 878         devlink_fmsg_pair_nest_end(fmsg);
 879 }
 880 EXPORT_SYMBOL_GPL(devlink_fmsg_u8_pair_put);
 881
 882 void devlink_fmsg_u32_pair_put(struct devlink_fmsg *fmsg, const char *name,
 883                                u32 value)
 884 {
 885         devlink_fmsg_pair_nest_start(fmsg, name);
 886         devlink_fmsg_u32_put(fmsg, value);
 887         devlink_fmsg_pair_nest_end(fmsg);
 888 }
 889 EXPORT_SYMBOL_GPL(devlink_fmsg_u32_pair_put);
 890
 891 void devlink_fmsg_u64_pair_put(struct devlink_fmsg *fmsg, const char *name,
 892                                u64 value)
 893 {
 894         devlink_fmsg_pair_nest_start(fmsg, name);
 895         devlink_fmsg_u64_put(fmsg, value);
 896         devlink_fmsg_pair_nest_end(fmsg);
 897 }
 898 EXPORT_SYMBOL_GPL(devlink_fmsg_u64_pair_put);
 899
 900 void devlink_fmsg_string_pair_put(struct devlink_fmsg *fmsg, const char *name,
 901                                   const char *value)
 902 {
 903         devlink_fmsg_pair_nest_start(fmsg, name);
 904         devlink_fmsg_string_put(fmsg, value);
 905         devlink_fmsg_pair_nest_end(fmsg);
 906 }
 907 EXPORT_SYMBOL_GPL(devlink_fmsg_string_pair_put);
 908
 909 void devlink_fmsg_binary_pair_put(struct devlink_fmsg *fmsg, const char *name,
 910                                   const void *value, u32 value_len)
 911 {
 912         u32 data_size;
 913         u32 offset;
 914
 915         devlink_fmsg_binary_pair_nest_start(fmsg, name);
 916
 917         for (offset = 0; offset < value_len; offset += data_size) {
 918                 data_size = value_len - offset;
 919                 if (data_size > DEVLINK_FMSG_MAX_SIZE)
 920                         data_size = DEVLINK_FMSG_MAX_SIZE;
 921
 922                 devlink_fmsg_binary_put(fmsg, value + offset, data_size);
 923         }
 924
 925         devlink_fmsg_binary_pair_nest_end(fmsg);
 926         fmsg->putting_binary = false;
 927 }
 928 EXPORT_SYMBOL_GPL(devlink_fmsg_binary_pair_put);
 929
 930 static int
 931 devlink_fmsg_item_fill_type(struct devlink_fmsg_item *msg, struct sk_buff *skb)
 932 {
 933         switch (msg->nla_type) {
 934         case NLA_FLAG:
 935         case NLA_U8:
 936         case NLA_U32:
 937         case NLA_U64:
 938         case NLA_NUL_STRING:
 939         case NLA_BINARY:
 940                 return nla_put_u8(skb, DEVLINK_ATTR_FMSG_OBJ_VALUE_TYPE,
 941                                   msg->nla_type);
 942         default:
 943                 return -EINVAL;
 944         }
 945 }
 946
 947 static int
 948 devlink_fmsg_item_fill_data(struct devlink_fmsg_item *msg, struct sk_buff *skb)
 949 {
 950         int attrtype = DEVLINK_ATTR_FMSG_OBJ_VALUE_DATA;
 951         u8 tmp;
 952
 953         switch (msg->nla_type) {
 954         case NLA_FLAG:
 955                 /* Always provide flag data, regardless of its value */
 956                 tmp = *(bool *)msg->value;
 957
 958                 return nla_put_u8(skb, attrtype, tmp);
 959         case NLA_U8:
 960                 return nla_put_u8(skb, attrtype, *(u8 *)msg->value);
 961         case NLA_U32:
 962                 return nla_put_u32(skb, attrtype, *(u32 *)msg->value);
 963         case NLA_U64:
 964                 return devlink_nl_put_u64(skb, attrtype, *(u64 *)msg->value);
 965         case NLA_NUL_STRING:
 966                 return nla_put_string(skb, attrtype, (char *)&msg->value);
 967         case NLA_BINARY:
 968                 return nla_put(skb, attrtype, msg->len, (void *)&msg->value);
 969         default:
 970                 return -EINVAL;
 971         }
 972 }
 973
 974 static int
 975 devlink_fmsg_prepare_skb(struct devlink_fmsg *fmsg, struct sk_buff *skb,
 976                          int *start)
 977 {
 978         struct devlink_fmsg_item *item;
 979         struct nlattr *fmsg_nlattr;
 980         int err = 0;
 981         int i = 0;
 982
 983         fmsg_nlattr = nla_nest_start_noflag(skb, DEVLINK_ATTR_FMSG);
 984         if (!fmsg_nlattr)
 985                 return -EMSGSIZE;
 986
 987         list_for_each_entry(item, &fmsg->item_list, list) {
 988                 if (i < *start) {
 989                         i++;
 990                         continue;
 991                 }
 992
 993                 switch (item->attrtype) {
 994                 case DEVLINK_ATTR_FMSG_OBJ_NEST_START:
 995                 case DEVLINK_ATTR_FMSG_PAIR_NEST_START:
 996                 case DEVLINK_ATTR_FMSG_ARR_NEST_START:
 997                 case DEVLINK_ATTR_FMSG_NEST_END:
 998                         err = nla_put_flag(skb, item->attrtype);
 999                         break;
1000                 case DEVLINK_ATTR_FMSG_OBJ_VALUE_DATA:
1001                         err = devlink_fmsg_item_fill_type(item, skb);
1002                         if (err)
1003                                 break;
1004                         err = devlink_fmsg_item_fill_data(item, skb);
1005                         break;
1006                 case DEVLINK_ATTR_FMSG_OBJ_NAME:
1007                         err = nla_put_string(skb, item->attrtype,
1008                                              (char *)&item->value);
1009                         break;
1010                 default:
1011                         err = -EINVAL;
1012                         break;
1013                 }
1014                 if (!err)
1015                         *start = ++i;
1016                 else
1017                         break;
1018         }
1019
1020         nla_nest_end(skb, fmsg_nlattr);
1021         return err;
1022 }
1023
1024 static int devlink_fmsg_snd(struct devlink_fmsg *fmsg,
1025                             struct genl_info *info,
1026                             enum devlink_command cmd, int flags)
1027 {
1028         struct nlmsghdr *nlh;
1029         struct sk_buff *skb;
1030         bool last = false;
1031         int index = 0;
1032         void *hdr;
1033         int err;
1034
1035         if (fmsg->err)
1036                 return fmsg->err;
1037
1038         while (!last) {
1039                 int tmp_index = index;
1040
1041                 skb = genlmsg_new(GENLMSG_DEFAULT_SIZE, GFP_KERNEL);
1042                 if (!skb)
1043                         return -ENOMEM;
1044
1045                 hdr = genlmsg_put(skb, info->snd_portid, info->snd_seq,
1046                                   &devlink_nl_family, flags | NLM_F_MULTI, cmd);
1047                 if (!hdr) {
1048                         err = -EMSGSIZE;
1049                         goto nla_put_failure;
1050                 }
1051
1052                 err = devlink_fmsg_prepare_skb(fmsg, skb, &index);
1053                 if (!err)
1054                         last = true;
1055                 else if (err != -EMSGSIZE || tmp_index == index)
1056                         goto nla_put_failure;
1057
1058                 genlmsg_end(skb, hdr);
1059                 err = genlmsg_reply(skb, info);
1060                 if (err)
1061                         return err;
1062         }
1063
1064         skb = genlmsg_new(GENLMSG_DEFAULT_SIZE, GFP_KERNEL);
1065         if (!skb)
1066                 return -ENOMEM;
1067         nlh = nlmsg_put(skb, info->snd_portid, info->snd_seq,
1068                         NLMSG_DONE, 0, flags | NLM_F_MULTI);
1069         if (!nlh) {
1070                 err = -EMSGSIZE;
1071                 goto nla_put_failure;
1072         }
1073
1074         return genlmsg_reply(skb, info);
1075
1076 nla_put_failure:
1077         nlmsg_free(skb);
1078         return err;
1079 }
1080
1081 static int devlink_fmsg_dumpit(struct devlink_fmsg *fmsg, struct sk_buff *skb,
1082                                struct netlink_callback *cb,
1083                                enum devlink_command cmd)
1084 {
1085         struct devlink_nl_dump_state *state = devlink_dump_state(cb);
1086         int index = state->idx;
1087         int tmp_index = index;
1088         void *hdr;
1089         int err;
1090
1091         if (fmsg->err)
1092                 return fmsg->err;
1093
1094         hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1095                           &devlink_nl_family, NLM_F_ACK | NLM_F_MULTI, cmd);
1096         if (!hdr) {
1097                 err = -EMSGSIZE;
1098                 goto nla_put_failure;
1099         }
1100
1101         err = devlink_fmsg_prepare_skb(fmsg, skb, &index);
1102         if ((err && err != -EMSGSIZE) || tmp_index == index)
1103                 goto nla_put_failure;
1104
1105         state->idx = index;
1106         genlmsg_end(skb, hdr);
1107         return skb->len;
1108
1109 nla_put_failure:
1110         genlmsg_cancel(skb, hdr);
1111         return err;
1112 }
1113
1114 int devlink_nl_health_reporter_diagnose_doit(struct sk_buff *skb,
1115                                              struct genl_info *info)
1116 {
1117         struct devlink *devlink = info->user_ptr[0];
1118         struct devlink_health_reporter *reporter;
1119         struct devlink_fmsg *fmsg;
1120         int err;
1121
1122         reporter = devlink_health_reporter_get_from_info(devlink, info);
1123         if (!reporter)
1124                 return -EINVAL;
1125
1126         if (!reporter->ops->diagnose)
1127                 return -EOPNOTSUPP;
1128
1129         fmsg = devlink_fmsg_alloc();
1130         if (!fmsg)
1131                 return -ENOMEM;
1132
1133         devlink_fmsg_obj_nest_start(fmsg);
1134
1135         err = reporter->ops->diagnose(reporter, fmsg, info->extack);
1136         if (err)
1137                 goto out;
1138
1139         devlink_fmsg_obj_nest_end(fmsg);
1140
1141         err = devlink_fmsg_snd(fmsg, info,
1142                                DEVLINK_CMD_HEALTH_REPORTER_DIAGNOSE, 0);
1143
1144 out:
1145         devlink_fmsg_free(fmsg);
1146         return err;
1147 }
1148
1149 static struct devlink_health_reporter *
1150 devlink_health_reporter_get_from_cb_lock(struct netlink_callback *cb)
1151 {
1152         const struct genl_info *info = genl_info_dump(cb);
1153         struct devlink_health_reporter *reporter;
1154         struct nlattr **attrs = info->attrs;
1155         struct devlink *devlink;
1156
1157         devlink = devlink_get_from_attrs_lock(sock_net(cb->skb->sk), attrs,
1158                                               false);
1159         if (IS_ERR(devlink))
1160                 return NULL;
1161
1162         reporter = devlink_health_reporter_get_from_attrs(devlink, attrs);
1163         if (!reporter) {
1164                 devl_unlock(devlink);
1165                 devlink_put(devlink);
1166         }
1167         return reporter;
1168 }
1169
1170 int devlink_nl_health_reporter_dump_get_dumpit(struct sk_buff *skb,
1171                                                struct netlink_callback *cb)
1172 {
1173         struct devlink_nl_dump_state *state = devlink_dump_state(cb);
1174         struct devlink_health_reporter *reporter;
1175         struct devlink *devlink;
1176         int err;
1177
1178         reporter = devlink_health_reporter_get_from_cb_lock(cb);
1179         if (!reporter)
1180                 return -EINVAL;
1181
1182         devlink = reporter->devlink;
1183         if (!reporter->ops->dump) {
1184                 devl_unlock(devlink);
1185                 devlink_put(devlink);
1186                 return -EOPNOTSUPP;
1187         }
1188
1189         if (!state->idx) {
1190                 err = devlink_health_do_dump(reporter, NULL, cb->extack);
1191                 if (err)
1192                         goto unlock;
1193                 state->dump_ts = reporter->dump_ts;
1194         }
1195         if (!reporter->dump_fmsg || state->dump_ts != reporter->dump_ts) {
1196                 NL_SET_ERR_MSG(cb->extack, "Dump trampled, please retry");
1197                 err = -EAGAIN;
1198                 goto unlock;
1199         }
1200
1201         err = devlink_fmsg_dumpit(reporter->dump_fmsg, skb, cb,
1202                                   DEVLINK_CMD_HEALTH_REPORTER_DUMP_GET);
1203 unlock:
1204         devl_unlock(devlink);
1205         devlink_put(devlink);
1206         return err;
1207 }
1208
1209 int devlink_nl_health_reporter_dump_clear_doit(struct sk_buff *skb,
1210                                                struct genl_info *info)
1211 {
1212         struct devlink *devlink = info->user_ptr[0];
1213         struct devlink_health_reporter *reporter;
1214
1215         reporter = devlink_health_reporter_get_from_info(devlink, info);
1216         if (!reporter)
1217                 return -EINVAL;
1218
1219         if (!reporter->ops->dump)
1220                 return -EOPNOTSUPP;
1221
1222         devlink_health_dump_clear(reporter);
1223         return 0;
1224 }
1225
1226 int devlink_nl_health_reporter_test_doit(struct sk_buff *skb,
1227                                          struct genl_info *info)
1228 {
1229         struct devlink *devlink = info->user_ptr[0];
1230         struct devlink_health_reporter *reporter;
1231
1232         reporter = devlink_health_reporter_get_from_info(devlink, info);
1233         if (!reporter)
1234                 return -EINVAL;
1235
1236         if (!reporter->ops->test)
1237                 return -EOPNOTSUPP;
1238
1239         return reporter->ops->test(reporter, info->extack);
1240 }