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 (nla_put_u64_64bit(msg, DEVLINK_ATTR_HEALTH_REPORTER_ERR_COUNT,
 291                               reporter->error_count, DEVLINK_ATTR_PAD))
 292                 goto reporter_nest_cancel;
 293         if (nla_put_u64_64bit(msg, DEVLINK_ATTR_HEALTH_REPORTER_RECOVER_COUNT,
 294                               reporter->recovery_count, DEVLINK_ATTR_PAD))
 295                 goto reporter_nest_cancel;
 296         if (reporter->ops->recover &&
 297             nla_put_u64_64bit(msg, DEVLINK_ATTR_HEALTH_REPORTER_GRACEFUL_PERIOD,
 298                               reporter->graceful_period,
 299                               DEVLINK_ATTR_PAD))
 300                 goto reporter_nest_cancel;
 301         if (reporter->ops->recover &&
 302             nla_put_u8(msg, DEVLINK_ATTR_HEALTH_REPORTER_AUTO_RECOVER,
 303                        reporter->auto_recover))
 304                 goto reporter_nest_cancel;
 305         if (reporter->dump_fmsg &&
 306             nla_put_u64_64bit(msg, DEVLINK_ATTR_HEALTH_REPORTER_DUMP_TS,
 307                               jiffies_to_msecs(reporter->dump_ts),
 308                               DEVLINK_ATTR_PAD))
 309                 goto reporter_nest_cancel;
 310         if (reporter->dump_fmsg &&
 311             nla_put_u64_64bit(msg, DEVLINK_ATTR_HEALTH_REPORTER_DUMP_TS_NS,
 312                               reporter->dump_real_ts, DEVLINK_ATTR_PAD))
 313                 goto reporter_nest_cancel;
 314         if (reporter->ops->dump &&
 315             nla_put_u8(msg, DEVLINK_ATTR_HEALTH_REPORTER_AUTO_DUMP,
 316                        reporter->auto_dump))
 317                 goto reporter_nest_cancel;
 318
 319         nla_nest_end(msg, reporter_attr);
 320         genlmsg_end(msg, hdr);
 321         return 0;
 322
 323 reporter_nest_cancel:
 324         nla_nest_cancel(msg, reporter_attr);
 325 genlmsg_cancel:
 326         genlmsg_cancel(msg, hdr);
 327         return -EMSGSIZE;
 328 }
 329
 330 static struct devlink_health_reporter *
 331 devlink_health_reporter_get_from_attrs(struct devlink *devlink,
 332                                        struct nlattr **attrs)
 333 {
 334         struct devlink_port *devlink_port;
 335         char *reporter_name;
 336
 337         if (!attrs[DEVLINK_ATTR_HEALTH_REPORTER_NAME])
 338                 return NULL;
 339
 340         reporter_name = nla_data(attrs[DEVLINK_ATTR_HEALTH_REPORTER_NAME]);
 341         devlink_port = devlink_port_get_from_attrs(devlink, attrs);
 342         if (IS_ERR(devlink_port))
 343                 return devlink_health_reporter_find_by_name(devlink,
 344                                                             reporter_name);
 345         else
 346                 return devlink_port_health_reporter_find_by_name(devlink_port,
 347                                                                  reporter_name);
 348 }
 349
 350 static struct devlink_health_reporter *
 351 devlink_health_reporter_get_from_info(struct devlink *devlink,
 352                                       struct genl_info *info)
 353 {
 354         return devlink_health_reporter_get_from_attrs(devlink, info->attrs);
 355 }
 356
 357 int devlink_nl_health_reporter_get_doit(struct sk_buff *skb,
 358                                         struct genl_info *info)
 359 {
 360         struct devlink *devlink = info->user_ptr[0];
 361         struct devlink_health_reporter *reporter;
 362         struct sk_buff *msg;
 363         int err;
 364
 365         reporter = devlink_health_reporter_get_from_info(devlink, info);
 366         if (!reporter)
 367                 return -EINVAL;
 368
 369         msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
 370         if (!msg)
 371                 return -ENOMEM;
 372
 373         err = devlink_nl_health_reporter_fill(msg, reporter,
 374                                               DEVLINK_CMD_HEALTH_REPORTER_GET,
 375                                               info->snd_portid, info->snd_seq,
 376                                               0);
 377         if (err) {
 378                 nlmsg_free(msg);
 379                 return err;
 380         }
 381
 382         return genlmsg_reply(msg, info);
 383 }
 384
 385 static int devlink_nl_health_reporter_get_dump_one(struct sk_buff *msg,
 386                                                    struct devlink *devlink,
 387                                                    struct netlink_callback *cb,
 388                                                    int flags)
 389 {
 390         struct devlink_nl_dump_state *state = devlink_dump_state(cb);
 391         const struct genl_info *info = genl_info_dump(cb);
 392         struct devlink_health_reporter *reporter;
 393         unsigned long port_index_end = ULONG_MAX;
 394         struct nlattr **attrs = info->attrs;
 395         unsigned long port_index_start = 0;
 396         struct devlink_port *port;
 397         unsigned long port_index;
 398         int idx = 0;
 399         int err;
 400
 401         if (attrs && attrs[DEVLINK_ATTR_PORT_INDEX]) {
 402                 port_index_start = nla_get_u32(attrs[DEVLINK_ATTR_PORT_INDEX]);
 403                 port_index_end = port_index_start;
 404                 flags |= NLM_F_DUMP_FILTERED;
 405                 goto per_port_dump;
 406         }
 407
 408         list_for_each_entry(reporter, &devlink->reporter_list, list) {
 409                 if (idx < state->idx) {
 410                         idx++;
 411                         continue;
 412                 }
 413                 err = devlink_nl_health_reporter_fill(msg, reporter,
 414                                                       DEVLINK_CMD_HEALTH_REPORTER_GET,
 415                                                       NETLINK_CB(cb->skb).portid,
 416                                                       cb->nlh->nlmsg_seq,
 417                                                       flags);
 418                 if (err) {
 419                         state->idx = idx;
 420                         return err;
 421                 }
 422                 idx++;
 423         }
 424 per_port_dump:
 425         xa_for_each_range(&devlink->ports, port_index, port,
 426                           port_index_start, port_index_end) {
 427                 list_for_each_entry(reporter, &port->reporter_list, list) {
 428                         if (idx < state->idx) {
 429                                 idx++;
 430                                 continue;
 431                         }
 432                         err = devlink_nl_health_reporter_fill(msg, reporter,
 433                                                               DEVLINK_CMD_HEALTH_REPORTER_GET,
 434                                                               NETLINK_CB(cb->skb).portid,
 435                                                               cb->nlh->nlmsg_seq,
 436                                                               flags);
 437                         if (err) {
 438                                 state->idx = idx;
 439                                 return err;
 440                         }
 441                         idx++;
 442                 }
 443         }
 444
 445         return 0;
 446 }
 447
 448 int devlink_nl_health_reporter_get_dumpit(struct sk_buff *skb,
 449                                           struct netlink_callback *cb)
 450 {
 451         return devlink_nl_dumpit(skb, cb,
 452                                  devlink_nl_health_reporter_get_dump_one);
 453 }
 454
 455 int devlink_nl_health_reporter_set_doit(struct sk_buff *skb,
 456                                         struct genl_info *info)
 457 {
 458         struct devlink *devlink = info->user_ptr[0];
 459         struct devlink_health_reporter *reporter;
 460
 461         reporter = devlink_health_reporter_get_from_info(devlink, info);
 462         if (!reporter)
 463                 return -EINVAL;
 464
 465         if (!reporter->ops->recover &&
 466             (info->attrs[DEVLINK_ATTR_HEALTH_REPORTER_GRACEFUL_PERIOD] ||
 467              info->attrs[DEVLINK_ATTR_HEALTH_REPORTER_AUTO_RECOVER]))
 468                 return -EOPNOTSUPP;
 469
 470         if (!reporter->ops->dump &&
 471             info->attrs[DEVLINK_ATTR_HEALTH_REPORTER_AUTO_DUMP])
 472                 return -EOPNOTSUPP;
 473
 474         if (info->attrs[DEVLINK_ATTR_HEALTH_REPORTER_GRACEFUL_PERIOD])
 475                 reporter->graceful_period =
 476                         nla_get_u64(info->attrs[DEVLINK_ATTR_HEALTH_REPORTER_GRACEFUL_PERIOD]);
 477
 478         if (info->attrs[DEVLINK_ATTR_HEALTH_REPORTER_AUTO_RECOVER])
 479                 reporter->auto_recover =
 480                         nla_get_u8(info->attrs[DEVLINK_ATTR_HEALTH_REPORTER_AUTO_RECOVER]);
 481
 482         if (info->attrs[DEVLINK_ATTR_HEALTH_REPORTER_AUTO_DUMP])
 483                 reporter->auto_dump =
 484                 nla_get_u8(info->attrs[DEVLINK_ATTR_HEALTH_REPORTER_AUTO_DUMP]);
 485
 486         return 0;
 487 }
 488
 489 static void devlink_recover_notify(struct devlink_health_reporter *reporter,
 490                                    enum devlink_command cmd)
 491 {
 492         struct devlink *devlink = reporter->devlink;
 493         struct sk_buff *msg;
 494         int err;
 495
 496         WARN_ON(cmd != DEVLINK_CMD_HEALTH_REPORTER_RECOVER);
 497         ASSERT_DEVLINK_REGISTERED(devlink);
 498
 499         msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
 500         if (!msg)
 501                 return;
 502
 503         err = devlink_nl_health_reporter_fill(msg, reporter, cmd, 0, 0, 0);
 504         if (err) {
 505                 nlmsg_free(msg);
 506                 return;
 507         }
 508
 509         genlmsg_multicast_netns(&devlink_nl_family, devlink_net(devlink), msg,
 510                                 0, DEVLINK_MCGRP_CONFIG, GFP_KERNEL);
 511 }
 512
 513 void
 514 devlink_health_reporter_recovery_done(struct devlink_health_reporter *reporter)
 515 {
 516         reporter->recovery_count++;
 517         reporter->last_recovery_ts = jiffies;
 518 }
 519 EXPORT_SYMBOL_GPL(devlink_health_reporter_recovery_done);
 520
 521 static int
 522 devlink_health_reporter_recover(struct devlink_health_reporter *reporter,
 523                                 void *priv_ctx, struct netlink_ext_ack *extack)
 524 {
 525         int err;
 526
 527         if (reporter->health_state == DEVLINK_HEALTH_REPORTER_STATE_HEALTHY)
 528                 return 0;
 529
 530         if (!reporter->ops->recover)
 531                 return -EOPNOTSUPP;
 532
 533         err = reporter->ops->recover(reporter, priv_ctx, extack);
 534         if (err)
 535                 return err;
 536
 537         devlink_health_reporter_recovery_done(reporter);
 538         reporter->health_state = DEVLINK_HEALTH_REPORTER_STATE_HEALTHY;
 539         devlink_recover_notify(reporter, DEVLINK_CMD_HEALTH_REPORTER_RECOVER);
 540
 541         return 0;
 542 }
 543
 544 static void
 545 devlink_health_dump_clear(struct devlink_health_reporter *reporter)
 546 {
 547         if (!reporter->dump_fmsg)
 548                 return;
 549         devlink_fmsg_free(reporter->dump_fmsg);
 550         reporter->dump_fmsg = NULL;
 551 }
 552
 553 static int devlink_health_do_dump(struct devlink_health_reporter *reporter,
 554                                   void *priv_ctx,
 555                                   struct netlink_ext_ack *extack)
 556 {
 557         int err;
 558
 559         if (!reporter->ops->dump)
 560                 return 0;
 561
 562         if (reporter->dump_fmsg)
 563                 return 0;
 564
 565         reporter->dump_fmsg = devlink_fmsg_alloc();
 566         if (!reporter->dump_fmsg)
 567                 return -ENOMEM;
 568
 569         devlink_fmsg_obj_nest_start(reporter->dump_fmsg);
 570
 571         err = reporter->ops->dump(reporter, reporter->dump_fmsg,
 572                                   priv_ctx, extack);
 573         if (err)
 574                 goto dump_err;
 575
 576         devlink_fmsg_obj_nest_end(reporter->dump_fmsg);
 577         err = reporter->dump_fmsg->err;
 578         if (err)
 579                 goto dump_err;
 580
 581         reporter->dump_ts = jiffies;
 582         reporter->dump_real_ts = ktime_get_real_ns();
 583
 584         return 0;
 585
 586 dump_err:
 587         devlink_health_dump_clear(reporter);
 588         return err;
 589 }
 590
 591 int devlink_health_report(struct devlink_health_reporter *reporter,
 592                           const char *msg, void *priv_ctx)
 593 {
 594         enum devlink_health_reporter_state prev_health_state;
 595         struct devlink *devlink = reporter->devlink;
 596         unsigned long recover_ts_threshold;
 597         int ret;
 598
 599         /* write a log message of the current error */
 600         WARN_ON(!msg);
 601         trace_devlink_health_report(devlink, reporter->ops->name, msg);
 602         reporter->error_count++;
 603         prev_health_state = reporter->health_state;
 604         reporter->health_state = DEVLINK_HEALTH_REPORTER_STATE_ERROR;
 605         devlink_recover_notify(reporter, DEVLINK_CMD_HEALTH_REPORTER_RECOVER);
 606
 607         /* abort if the previous error wasn't recovered */
 608         recover_ts_threshold = reporter->last_recovery_ts +
 609                                msecs_to_jiffies(reporter->graceful_period);
 610         if (reporter->auto_recover &&
 611             (prev_health_state != DEVLINK_HEALTH_REPORTER_STATE_HEALTHY ||
 612              (reporter->last_recovery_ts && reporter->recovery_count &&
 613               time_is_after_jiffies(recover_ts_threshold)))) {
 614                 trace_devlink_health_recover_aborted(devlink,
 615                                                      reporter->ops->name,
 616                                                      reporter->health_state,
 617                                                      jiffies -
 618                                                      reporter->last_recovery_ts);
 619                 return -ECANCELED;
 620         }
 621
 622         if (reporter->auto_dump) {
 623                 devl_lock(devlink);
 624                 /* store current dump of current error, for later analysis */
 625                 devlink_health_do_dump(reporter, priv_ctx, NULL);
 626                 devl_unlock(devlink);
 627         }
 628
 629         if (!reporter->auto_recover)
 630                 return 0;
 631
 632         devl_lock(devlink);
 633         ret = devlink_health_reporter_recover(reporter, priv_ctx, NULL);
 634         devl_unlock(devlink);
 635
 636         return ret;
 637 }
 638 EXPORT_SYMBOL_GPL(devlink_health_report);
 639
 640 void
 641 devlink_health_reporter_state_update(struct devlink_health_reporter *reporter,
 642                                      enum devlink_health_reporter_state state)
 643 {
 644         if (WARN_ON(state != DEVLINK_HEALTH_REPORTER_STATE_HEALTHY &&
 645                     state != DEVLINK_HEALTH_REPORTER_STATE_ERROR))
 646                 return;
 647
 648         if (reporter->health_state == state)
 649                 return;
 650
 651         reporter->health_state = state;
 652         trace_devlink_health_reporter_state_update(reporter->devlink,
 653                                                    reporter->ops->name, state);
 654         devlink_recover_notify(reporter, DEVLINK_CMD_HEALTH_REPORTER_RECOVER);
 655 }
 656 EXPORT_SYMBOL_GPL(devlink_health_reporter_state_update);
 657
 658 int devlink_nl_health_reporter_recover_doit(struct sk_buff *skb,
 659                                             struct genl_info *info)
 660 {
 661         struct devlink *devlink = info->user_ptr[0];
 662         struct devlink_health_reporter *reporter;
 663
 664         reporter = devlink_health_reporter_get_from_info(devlink, info);
 665         if (!reporter)
 666                 return -EINVAL;
 667
 668         return devlink_health_reporter_recover(reporter, NULL, info->extack);
 669 }
 670
 671 static void devlink_fmsg_err_if_binary(struct devlink_fmsg *fmsg)
 672 {
 673         if (!fmsg->err && fmsg->putting_binary)
 674                 fmsg->err = -EINVAL;
 675 }
 676
 677 static void devlink_fmsg_nest_common(struct devlink_fmsg *fmsg, int attrtype)
 678 {
 679         struct devlink_fmsg_item *item;
 680
 681         if (fmsg->err)
 682                 return;
 683
 684         item = kzalloc(sizeof(*item), GFP_KERNEL);
 685         if (!item) {
 686                 fmsg->err = -ENOMEM;
 687                 return;
 688         }
 689
 690         item->attrtype = attrtype;
 691         list_add_tail(&item->list, &fmsg->item_list);
 692 }
 693
 694 void devlink_fmsg_obj_nest_start(struct devlink_fmsg *fmsg)
 695 {
 696         devlink_fmsg_err_if_binary(fmsg);
 697         devlink_fmsg_nest_common(fmsg, DEVLINK_ATTR_FMSG_OBJ_NEST_START);
 698 }
 699 EXPORT_SYMBOL_GPL(devlink_fmsg_obj_nest_start);
 700
 701 static void devlink_fmsg_nest_end(struct devlink_fmsg *fmsg)
 702 {
 703         devlink_fmsg_err_if_binary(fmsg);
 704         devlink_fmsg_nest_common(fmsg, DEVLINK_ATTR_FMSG_NEST_END);
 705 }
 706
 707 void devlink_fmsg_obj_nest_end(struct devlink_fmsg *fmsg)
 708 {
 709         devlink_fmsg_nest_end(fmsg);
 710 }
 711 EXPORT_SYMBOL_GPL(devlink_fmsg_obj_nest_end);
 712
 713 #define DEVLINK_FMSG_MAX_SIZE (GENLMSG_DEFAULT_SIZE - GENL_HDRLEN - NLA_HDRLEN)
 714
 715 static void devlink_fmsg_put_name(struct devlink_fmsg *fmsg, const char *name)
 716 {
 717         struct devlink_fmsg_item *item;
 718
 719         devlink_fmsg_err_if_binary(fmsg);
 720         if (fmsg->err)
 721                 return;
 722
 723         if (strlen(name) + 1 > DEVLINK_FMSG_MAX_SIZE) {
 724                 fmsg->err = -EMSGSIZE;
 725                 return;
 726         }
 727
 728         item = kzalloc(sizeof(*item) + strlen(name) + 1, GFP_KERNEL);
 729         if (!item) {
 730                 fmsg->err = -ENOMEM;
 731                 return;
 732         }
 733
 734         item->nla_type = NLA_NUL_STRING;
 735         item->len = strlen(name) + 1;
 736         item->attrtype = DEVLINK_ATTR_FMSG_OBJ_NAME;
 737         memcpy(&item->value, name, item->len);
 738         list_add_tail(&item->list, &fmsg->item_list);
 739 }
 740
 741 void devlink_fmsg_pair_nest_start(struct devlink_fmsg *fmsg, const char *name)
 742 {
 743         devlink_fmsg_err_if_binary(fmsg);
 744         devlink_fmsg_nest_common(fmsg, DEVLINK_ATTR_FMSG_PAIR_NEST_START);
 745         devlink_fmsg_put_name(fmsg, name);
 746 }
 747 EXPORT_SYMBOL_GPL(devlink_fmsg_pair_nest_start);
 748
 749 void devlink_fmsg_pair_nest_end(struct devlink_fmsg *fmsg)
 750 {
 751         devlink_fmsg_nest_end(fmsg);
 752 }
 753 EXPORT_SYMBOL_GPL(devlink_fmsg_pair_nest_end);
 754
 755 void devlink_fmsg_arr_pair_nest_start(struct devlink_fmsg *fmsg,
 756                                       const char *name)
 757 {
 758         devlink_fmsg_pair_nest_start(fmsg, name);
 759         devlink_fmsg_nest_common(fmsg, DEVLINK_ATTR_FMSG_ARR_NEST_START);
 760 }
 761 EXPORT_SYMBOL_GPL(devlink_fmsg_arr_pair_nest_start);
 762
 763 void devlink_fmsg_arr_pair_nest_end(struct devlink_fmsg *fmsg)
 764 {
 765         devlink_fmsg_nest_end(fmsg);
 766         devlink_fmsg_nest_end(fmsg);
 767 }
 768 EXPORT_SYMBOL_GPL(devlink_fmsg_arr_pair_nest_end);
 769
 770 void devlink_fmsg_binary_pair_nest_start(struct devlink_fmsg *fmsg,
 771                                          const char *name)
 772 {
 773         devlink_fmsg_arr_pair_nest_start(fmsg, name);
 774         fmsg->putting_binary = true;
 775 }
 776 EXPORT_SYMBOL_GPL(devlink_fmsg_binary_pair_nest_start);
 777
 778 void devlink_fmsg_binary_pair_nest_end(struct devlink_fmsg *fmsg)
 779 {
 780         if (fmsg->err)
 781                 return;
 782
 783         if (!fmsg->putting_binary)
 784                 fmsg->err = -EINVAL;
 785
 786         fmsg->putting_binary = false;
 787         devlink_fmsg_arr_pair_nest_end(fmsg);
 788 }
 789 EXPORT_SYMBOL_GPL(devlink_fmsg_binary_pair_nest_end);
 790
 791 static void devlink_fmsg_put_value(struct devlink_fmsg *fmsg,
 792                                    const void *value, u16 value_len,
 793                                    u8 value_nla_type)
 794 {
 795         struct devlink_fmsg_item *item;
 796
 797         if (fmsg->err)
 798                 return;
 799
 800         if (value_len > DEVLINK_FMSG_MAX_SIZE) {
 801                 fmsg->err = -EMSGSIZE;
 802                 return;
 803         }
 804
 805         item = kzalloc(sizeof(*item) + value_len, GFP_KERNEL);
 806         if (!item) {
 807                 fmsg->err = -ENOMEM;
 808                 return;
 809         }
 810
 811         item->nla_type = value_nla_type;
 812         item->len = value_len;
 813         item->attrtype = DEVLINK_ATTR_FMSG_OBJ_VALUE_DATA;
 814         memcpy(&item->value, value, item->len);
 815         list_add_tail(&item->list, &fmsg->item_list);
 816 }
 817
 818 static void devlink_fmsg_bool_put(struct devlink_fmsg *fmsg, bool value)
 819 {
 820         devlink_fmsg_err_if_binary(fmsg);
 821         devlink_fmsg_put_value(fmsg, &value, sizeof(value), NLA_FLAG);
 822 }
 823
 824 static void devlink_fmsg_u8_put(struct devlink_fmsg *fmsg, u8 value)
 825 {
 826         devlink_fmsg_err_if_binary(fmsg);
 827         devlink_fmsg_put_value(fmsg, &value, sizeof(value), NLA_U8);
 828 }
 829
 830 void devlink_fmsg_u32_put(struct devlink_fmsg *fmsg, u32 value)
 831 {
 832         devlink_fmsg_err_if_binary(fmsg);
 833         devlink_fmsg_put_value(fmsg, &value, sizeof(value), NLA_U32);
 834 }
 835 EXPORT_SYMBOL_GPL(devlink_fmsg_u32_put);
 836
 837 static void devlink_fmsg_u64_put(struct devlink_fmsg *fmsg, u64 value)
 838 {
 839         devlink_fmsg_err_if_binary(fmsg);
 840         devlink_fmsg_put_value(fmsg, &value, sizeof(value), NLA_U64);
 841 }
 842
 843 void devlink_fmsg_string_put(struct devlink_fmsg *fmsg, const char *value)
 844 {
 845         devlink_fmsg_err_if_binary(fmsg);
 846         devlink_fmsg_put_value(fmsg, value, strlen(value) + 1, NLA_NUL_STRING);
 847 }
 848 EXPORT_SYMBOL_GPL(devlink_fmsg_string_put);
 849
 850 void devlink_fmsg_binary_put(struct devlink_fmsg *fmsg, const void *value,
 851                              u16 value_len)
 852 {
 853         if (!fmsg->err && !fmsg->putting_binary)
 854                 fmsg->err = -EINVAL;
 855
 856         devlink_fmsg_put_value(fmsg, value, value_len, NLA_BINARY);
 857 }
 858 EXPORT_SYMBOL_GPL(devlink_fmsg_binary_put);
 859
 860 void devlink_fmsg_bool_pair_put(struct devlink_fmsg *fmsg, const char *name,
 861                                 bool value)
 862 {
 863         devlink_fmsg_pair_nest_start(fmsg, name);
 864         devlink_fmsg_bool_put(fmsg, value);
 865         devlink_fmsg_pair_nest_end(fmsg);
 866 }
 867 EXPORT_SYMBOL_GPL(devlink_fmsg_bool_pair_put);
 868
 869 void devlink_fmsg_u8_pair_put(struct devlink_fmsg *fmsg, const char *name,
 870                               u8 value)
 871 {
 872         devlink_fmsg_pair_nest_start(fmsg, name);
 873         devlink_fmsg_u8_put(fmsg, value);
 874         devlink_fmsg_pair_nest_end(fmsg);
 875 }
 876 EXPORT_SYMBOL_GPL(devlink_fmsg_u8_pair_put);
 877
 878 void devlink_fmsg_u32_pair_put(struct devlink_fmsg *fmsg, const char *name,
 879                                u32 value)
 880 {
 881         devlink_fmsg_pair_nest_start(fmsg, name);
 882         devlink_fmsg_u32_put(fmsg, value);
 883         devlink_fmsg_pair_nest_end(fmsg);
 884 }
 885 EXPORT_SYMBOL_GPL(devlink_fmsg_u32_pair_put);
 886
 887 void devlink_fmsg_u64_pair_put(struct devlink_fmsg *fmsg, const char *name,
 888                                u64 value)
 889 {
 890         devlink_fmsg_pair_nest_start(fmsg, name);
 891         devlink_fmsg_u64_put(fmsg, value);
 892         devlink_fmsg_pair_nest_end(fmsg);
 893 }
 894 EXPORT_SYMBOL_GPL(devlink_fmsg_u64_pair_put);
 895
 896 void devlink_fmsg_string_pair_put(struct devlink_fmsg *fmsg, const char *name,
 897                                   const char *value)
 898 {
 899         devlink_fmsg_pair_nest_start(fmsg, name);
 900         devlink_fmsg_string_put(fmsg, value);
 901         devlink_fmsg_pair_nest_end(fmsg);
 902 }
 903 EXPORT_SYMBOL_GPL(devlink_fmsg_string_pair_put);
 904
 905 void devlink_fmsg_binary_pair_put(struct devlink_fmsg *fmsg, const char *name,
 906                                   const void *value, u32 value_len)
 907 {
 908         u32 data_size;
 909         u32 offset;
 910
 911         devlink_fmsg_binary_pair_nest_start(fmsg, name);
 912
 913         for (offset = 0; offset < value_len; offset += data_size) {
 914                 data_size = value_len - offset;
 915                 if (data_size > DEVLINK_FMSG_MAX_SIZE)
 916                         data_size = DEVLINK_FMSG_MAX_SIZE;
 917
 918                 devlink_fmsg_binary_put(fmsg, value + offset, data_size);
 919         }
 920
 921         devlink_fmsg_binary_pair_nest_end(fmsg);
 922         fmsg->putting_binary = false;
 923 }
 924 EXPORT_SYMBOL_GPL(devlink_fmsg_binary_pair_put);
 925
 926 static int
 927 devlink_fmsg_item_fill_type(struct devlink_fmsg_item *msg, struct sk_buff *skb)
 928 {
 929         switch (msg->nla_type) {
 930         case NLA_FLAG:
 931         case NLA_U8:
 932         case NLA_U32:
 933         case NLA_U64:
 934         case NLA_NUL_STRING:
 935         case NLA_BINARY:
 936                 return nla_put_u8(skb, DEVLINK_ATTR_FMSG_OBJ_VALUE_TYPE,
 937                                   msg->nla_type);
 938         default:
 939                 return -EINVAL;
 940         }
 941 }
 942
 943 static int
 944 devlink_fmsg_item_fill_data(struct devlink_fmsg_item *msg, struct sk_buff *skb)
 945 {
 946         int attrtype = DEVLINK_ATTR_FMSG_OBJ_VALUE_DATA;
 947         u8 tmp;
 948
 949         switch (msg->nla_type) {
 950         case NLA_FLAG:
 951                 /* Always provide flag data, regardless of its value */
 952                 tmp = *(bool *)msg->value;
 953
 954                 return nla_put_u8(skb, attrtype, tmp);
 955         case NLA_U8:
 956                 return nla_put_u8(skb, attrtype, *(u8 *)msg->value);
 957         case NLA_U32:
 958                 return nla_put_u32(skb, attrtype, *(u32 *)msg->value);
 959         case NLA_U64:
 960                 return nla_put_u64_64bit(skb, attrtype, *(u64 *)msg->value,
 961                                          DEVLINK_ATTR_PAD);
 962         case NLA_NUL_STRING:
 963                 return nla_put_string(skb, attrtype, (char *)&msg->value);
 964         case NLA_BINARY:
 965                 return nla_put(skb, attrtype, msg->len, (void *)&msg->value);
 966         default:
 967                 return -EINVAL;
 968         }
 969 }
 970
 971 static int
 972 devlink_fmsg_prepare_skb(struct devlink_fmsg *fmsg, struct sk_buff *skb,
 973                          int *start)
 974 {
 975         struct devlink_fmsg_item *item;
 976         struct nlattr *fmsg_nlattr;
 977         int err = 0;
 978         int i = 0;
 979
 980         fmsg_nlattr = nla_nest_start_noflag(skb, DEVLINK_ATTR_FMSG);
 981         if (!fmsg_nlattr)
 982                 return -EMSGSIZE;
 983
 984         list_for_each_entry(item, &fmsg->item_list, list) {
 985                 if (i < *start) {
 986                         i++;
 987                         continue;
 988                 }
 989
 990                 switch (item->attrtype) {
 991                 case DEVLINK_ATTR_FMSG_OBJ_NEST_START:
 992                 case DEVLINK_ATTR_FMSG_PAIR_NEST_START:
 993                 case DEVLINK_ATTR_FMSG_ARR_NEST_START:
 994                 case DEVLINK_ATTR_FMSG_NEST_END:
 995                         err = nla_put_flag(skb, item->attrtype);
 996                         break;
 997                 case DEVLINK_ATTR_FMSG_OBJ_VALUE_DATA:
 998                         err = devlink_fmsg_item_fill_type(item, skb);
 999                         if (err)
1000                                 break;
1001                         err = devlink_fmsg_item_fill_data(item, skb);
1002                         break;
1003                 case DEVLINK_ATTR_FMSG_OBJ_NAME:
1004                         err = nla_put_string(skb, item->attrtype,
1005                                              (char *)&item->value);
1006                         break;
1007                 default:
1008                         err = -EINVAL;
1009                         break;
1010                 }
1011                 if (!err)
1012                         *start = ++i;
1013                 else
1014                         break;
1015         }
1016
1017         nla_nest_end(skb, fmsg_nlattr);
1018         return err;
1019 }
1020
1021 static int devlink_fmsg_snd(struct devlink_fmsg *fmsg,
1022                             struct genl_info *info,
1023                             enum devlink_command cmd, int flags)
1024 {
1025         struct nlmsghdr *nlh;
1026         struct sk_buff *skb;
1027         bool last = false;
1028         int index = 0;
1029         void *hdr;
1030         int err;
1031
1032         if (fmsg->err)
1033                 return fmsg->err;
1034
1035         while (!last) {
1036                 int tmp_index = index;
1037
1038                 skb = genlmsg_new(GENLMSG_DEFAULT_SIZE, GFP_KERNEL);
1039                 if (!skb)
1040                         return -ENOMEM;
1041
1042                 hdr = genlmsg_put(skb, info->snd_portid, info->snd_seq,
1043                                   &devlink_nl_family, flags | NLM_F_MULTI, cmd);
1044                 if (!hdr) {
1045                         err = -EMSGSIZE;
1046                         goto nla_put_failure;
1047                 }
1048
1049                 err = devlink_fmsg_prepare_skb(fmsg, skb, &index);
1050                 if (!err)
1051                         last = true;
1052                 else if (err != -EMSGSIZE || tmp_index == index)
1053                         goto nla_put_failure;
1054
1055                 genlmsg_end(skb, hdr);
1056                 err = genlmsg_reply(skb, info);
1057                 if (err)
1058                         return err;
1059         }
1060
1061         skb = genlmsg_new(GENLMSG_DEFAULT_SIZE, GFP_KERNEL);
1062         if (!skb)
1063                 return -ENOMEM;
1064         nlh = nlmsg_put(skb, info->snd_portid, info->snd_seq,
1065                         NLMSG_DONE, 0, flags | NLM_F_MULTI);
1066         if (!nlh) {
1067                 err = -EMSGSIZE;
1068                 goto nla_put_failure;
1069         }
1070
1071         return genlmsg_reply(skb, info);
1072
1073 nla_put_failure:
1074         nlmsg_free(skb);
1075         return err;
1076 }
1077
1078 static int devlink_fmsg_dumpit(struct devlink_fmsg *fmsg, struct sk_buff *skb,
1079                                struct netlink_callback *cb,
1080                                enum devlink_command cmd)
1081 {
1082         struct devlink_nl_dump_state *state = devlink_dump_state(cb);
1083         int index = state->idx;
1084         int tmp_index = index;
1085         void *hdr;
1086         int err;
1087
1088         if (fmsg->err)
1089                 return fmsg->err;
1090
1091         hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1092                           &devlink_nl_family, NLM_F_ACK | NLM_F_MULTI, cmd);
1093         if (!hdr) {
1094                 err = -EMSGSIZE;
1095                 goto nla_put_failure;
1096         }
1097
1098         err = devlink_fmsg_prepare_skb(fmsg, skb, &index);
1099         if ((err && err != -EMSGSIZE) || tmp_index == index)
1100                 goto nla_put_failure;
1101
1102         state->idx = index;
1103         genlmsg_end(skb, hdr);
1104         return skb->len;
1105
1106 nla_put_failure:
1107         genlmsg_cancel(skb, hdr);
1108         return err;
1109 }
1110
1111 int devlink_nl_health_reporter_diagnose_doit(struct sk_buff *skb,
1112                                              struct genl_info *info)
1113 {
1114         struct devlink *devlink = info->user_ptr[0];
1115         struct devlink_health_reporter *reporter;
1116         struct devlink_fmsg *fmsg;
1117         int err;
1118
1119         reporter = devlink_health_reporter_get_from_info(devlink, info);
1120         if (!reporter)
1121                 return -EINVAL;
1122
1123         if (!reporter->ops->diagnose)
1124                 return -EOPNOTSUPP;
1125
1126         fmsg = devlink_fmsg_alloc();
1127         if (!fmsg)
1128                 return -ENOMEM;
1129
1130         devlink_fmsg_obj_nest_start(fmsg);
1131
1132         err = reporter->ops->diagnose(reporter, fmsg, info->extack);
1133         if (err)
1134                 goto out;
1135
1136         devlink_fmsg_obj_nest_end(fmsg);
1137
1138         err = devlink_fmsg_snd(fmsg, info,
1139                                DEVLINK_CMD_HEALTH_REPORTER_DIAGNOSE, 0);
1140
1141 out:
1142         devlink_fmsg_free(fmsg);
1143         return err;
1144 }
1145
1146 static struct devlink_health_reporter *
1147 devlink_health_reporter_get_from_cb_lock(struct netlink_callback *cb)
1148 {
1149         const struct genl_info *info = genl_info_dump(cb);
1150         struct devlink_health_reporter *reporter;
1151         struct nlattr **attrs = info->attrs;
1152         struct devlink *devlink;
1153
1154         devlink = devlink_get_from_attrs_lock(sock_net(cb->skb->sk), attrs);
1155         if (IS_ERR(devlink))
1156                 return NULL;
1157
1158         reporter = devlink_health_reporter_get_from_attrs(devlink, attrs);
1159         if (!reporter) {
1160                 devl_unlock(devlink);
1161                 devlink_put(devlink);
1162         }
1163         return reporter;
1164 }
1165
1166 int devlink_nl_health_reporter_dump_get_dumpit(struct sk_buff *skb,
1167                                                struct netlink_callback *cb)
1168 {
1169         struct devlink_nl_dump_state *state = devlink_dump_state(cb);
1170         struct devlink_health_reporter *reporter;
1171         struct devlink *devlink;
1172         int err;
1173
1174         reporter = devlink_health_reporter_get_from_cb_lock(cb);
1175         if (!reporter)
1176                 return -EINVAL;
1177
1178         devlink = reporter->devlink;
1179         if (!reporter->ops->dump) {
1180                 devl_unlock(devlink);
1181                 devlink_put(devlink);
1182                 return -EOPNOTSUPP;
1183         }
1184
1185         if (!state->idx) {
1186                 err = devlink_health_do_dump(reporter, NULL, cb->extack);
1187                 if (err)
1188                         goto unlock;
1189                 state->dump_ts = reporter->dump_ts;
1190         }
1191         if (!reporter->dump_fmsg || state->dump_ts != reporter->dump_ts) {
1192                 NL_SET_ERR_MSG(cb->extack, "Dump trampled, please retry");
1193                 err = -EAGAIN;
1194                 goto unlock;
1195         }
1196
1197         err = devlink_fmsg_dumpit(reporter->dump_fmsg, skb, cb,
1198                                   DEVLINK_CMD_HEALTH_REPORTER_DUMP_GET);
1199 unlock:
1200         devl_unlock(devlink);
1201         devlink_put(devlink);
1202         return err;
1203 }
1204
1205 int devlink_nl_health_reporter_dump_clear_doit(struct sk_buff *skb,
1206                                                struct genl_info *info)
1207 {
1208         struct devlink *devlink = info->user_ptr[0];
1209         struct devlink_health_reporter *reporter;
1210
1211         reporter = devlink_health_reporter_get_from_info(devlink, info);
1212         if (!reporter)
1213                 return -EINVAL;
1214
1215         if (!reporter->ops->dump)
1216                 return -EOPNOTSUPP;
1217
1218         devlink_health_dump_clear(reporter);
1219         return 0;
1220 }
1221
1222 int devlink_nl_health_reporter_test_doit(struct sk_buff *skb,
1223                                          struct genl_info *info)
1224 {
1225         struct devlink *devlink = info->user_ptr[0];
1226         struct devlink_health_reporter *reporter;
1227
1228         reporter = devlink_health_reporter_get_from_info(devlink, info);
1229         if (!reporter)
1230                 return -EINVAL;
1231
1232         if (!reporter->ops->test)
1233                 return -EOPNOTSUPP;
1234
1235         return reporter->ops->test(reporter, info->extack);
1236 }