Print this page
NEX-5736 implement autoreplace matching based on FRU slot number
NEX-6200 hot spares are not reactivated after reinserting into enclosure
NEX-9403 need to update FRU for spare and l2cache devices
NEX-9404 remove lofi autoreplace support from syseventd
NEX-9409 hotsparing doesn't work for vdevs without FRU
NEX-9424 zfs`vdev_online() needs better notification about state changes
Portions contributed by: Alek Pinchuk <alek@nexenta.com>
Portions contributed by: Josef 'Jeff' Sipek <josef.sipek@nexenta.com>
Reviewed by: Roman Strashkin <roman.strashkin@nexenta.com>
Reviewed by: Steve Peng <steve.peng@nexenta.com>
Reviewed by: Sanjay Nadkarni <sanjay.nadkarni@nexenta.com>
NEX-2846 Enable Automatic/Intelligent Hot Sparing capability
Reviewed by: Jeffry Molanus <jeffry.molanus@nexenta.com>
Reviewed by: Roman Strashkin <roman.strashkin@nexenta.com>
Reviewed by: Saso Kiselkov <saso.kiselkov@nexenta.com>

Split Close
Expand all
Collapse all
          --- old/usr/src/lib/libzfs/common/libzfs_fru.c
          +++ new/usr/src/lib/libzfs/common/libzfs_fru.c
↓ open down ↓ 14 lines elided ↑ open up ↑
  15   15   * If applicable, add the following below this CDDL HEADER, with the
  16   16   * fields enclosed by brackets "[]" replaced with your own identifying
  17   17   * information: Portions Copyright [yyyy] [name of copyright owner]
  18   18   *
  19   19   * CDDL HEADER END
  20   20   */
  21   21  
  22   22  /*
  23   23   * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  24   24   * Use is subject to license terms.
       25 + * Copyright 2017 Nexenta Systems, Inc.
  25   26   */
  26   27  
  27   28  #include <dlfcn.h>
  28   29  #include <errno.h>
  29   30  #include <libintl.h>
  30   31  #include <link.h>
  31   32  #include <pthread.h>
  32   33  #include <strings.h>
  33   34  #include <unistd.h>
  34   35  
  35   36  #include <libzfs.h>
  36   37  
  37   38  #include <fm/libtopo.h>
       39 +#include <fm/topo_hc.h>
  38   40  #include <sys/fm/protocol.h>
  39   41  #include <sys/systeminfo.h>
  40   42  
  41   43  #include "libzfs_impl.h"
  42   44  
  43   45  /*
  44   46   * This file is responsible for determining the relationship between I/O
  45   47   * devices paths and physical locations.  In the world of MPxIO and external
  46   48   * enclosures, the device path is not synonymous with the physical location.
  47   49   * If you remove a drive and insert it into a different slot, it will end up
↓ open down ↓ 22 lines elided ↑ open up ↑
  70   72  static topo_walk_t *(*_topo_walk_init)(topo_hdl_t *, const char *,
  71   73      topo_walk_cb_t, void *, int *);
  72   74  static int (*_topo_walk_step)(topo_walk_t *, int);
  73   75  static void (*_topo_walk_fini)(topo_walk_t *);
  74   76  static void (*_topo_hdl_strfree)(topo_hdl_t *, char *);
  75   77  static char *(*_topo_node_name)(tnode_t *);
  76   78  static int (*_topo_prop_get_string)(tnode_t *, const char *, const char *,
  77   79      char **, int *);
  78   80  static int (*_topo_node_fru)(tnode_t *, nvlist_t **, nvlist_t *, int *);
  79   81  static int (*_topo_fmri_nvl2str)(topo_hdl_t *, nvlist_t *, char **, int *);
       82 +static int (*_topo_fmri_str2nvl)(topo_hdl_t *, const char *, nvlist_t **,
       83 +    int *);
  80   84  static int (*_topo_fmri_strcmp_noauth)(topo_hdl_t *, const char *,
  81   85      const char *);
  82   86  
  83   87  #define ZFS_FRU_HASH_SIZE       257
  84   88  
  85   89  static size_t
  86   90  fru_strhash(const char *key)
  87   91  {
  88   92          ulong_t g, h = 0;
  89   93          const char *p;
↓ open down ↓ 89 lines elided ↑ open up ↑
 179  183          idx = fru_strhash(frup->zf_device);
 180  184          frup->zf_chain = hdl->libzfs_fru_hash[idx];
 181  185          hdl->libzfs_fru_hash[idx] = frup;
 182  186          frup->zf_next = hdl->libzfs_fru_list;
 183  187          hdl->libzfs_fru_list = frup;
 184  188  
 185  189          return (TOPO_WALK_NEXT);
 186  190  }
 187  191  
 188  192  /*
      193 + * Given a disk FRU, check that FRU contains a slot number and remove FRU
      194 + * details that aren't needed when comparing FRUs by slot number.
      195 + */
      196 +static char *
      197 +diskfru_to_slot(libzfs_handle_t *hdl, const char *diskfru)
      198 +{
      199 +        nvlist_t *nvl, **hc;
      200 +        char *hc_name, *tmp = NULL;
      201 +        int ret, i;
      202 +        uint_t hc_cnt;
      203 +
      204 +        /* string -> nvlist */
      205 +        if (_topo_fmri_str2nvl(hdl->libzfs_topo_hdl, diskfru, &nvl, &ret) != 0)
      206 +                return (NULL);
      207 +
      208 +        /* Need slot (bay) number in the FRU */
      209 +        if (nvlist_lookup_nvlist_array(nvl, FM_FMRI_HC_LIST, &hc,
      210 +            &hc_cnt) != 0)
      211 +                goto out;
      212 +
      213 +        for (i = 0; i < hc_cnt; i++) {
      214 +                if (nvlist_lookup_string(hc[i], FM_FMRI_HC_NAME,
      215 +                    &hc_name) == 0 && strcmp(hc_name, BAY) == 0)
      216 +                        break;
      217 +        }
      218 +        if (i == hc_cnt)
      219 +                goto out;
      220 +
      221 +        /* Drop the unwanted components */
      222 +        (void) nvlist_remove_all(nvl, FM_FMRI_HC_SERIAL_ID);
      223 +        (void) nvlist_remove_all(nvl, FM_FMRI_HC_PART);
      224 +        (void) nvlist_remove_all(nvl, FM_FMRI_HC_REVISION);
      225 +
      226 +        /* nvlist -> string */
      227 +        if (_topo_fmri_nvl2str(hdl->libzfs_topo_hdl, nvl, &tmp, &ret) != 0)
      228 +                tmp = NULL;
      229 +out:
      230 +        nvlist_free(nvl);
      231 +        return (tmp);
      232 +}
      233 +
      234 +/*
      235 + * Check if given FRUs match by slot number to skip comparing disk specific
      236 + * fields of the FRU.
      237 + */
      238 +/* ARGSUSED */
      239 +int
      240 +libzfs_fru_cmp_slot(libzfs_handle_t *hdl, const char *a, const char *b,
      241 +    size_t len)
      242 +{
      243 +        char *slota, *slotb;
      244 +        int ret = -1;
      245 +
      246 +        if (a == NULL || b == NULL)
      247 +                return (-1);
      248 +
      249 +        slota = diskfru_to_slot(hdl, a);
      250 +        slotb = diskfru_to_slot(hdl, b);
      251 +
      252 +        if (slota != NULL && slotb != NULL)
      253 +                ret = strcmp(slota, slotb);
      254 +
      255 +        _topo_hdl_strfree(hdl->libzfs_topo_hdl, slota);
      256 +        _topo_hdl_strfree(hdl->libzfs_topo_hdl, slotb);
      257 +
      258 +        return (ret);
      259 +}
      260 +
      261 +/*
 189  262   * Called during initialization to setup the dynamic libtopo connection.
 190  263   */
 191  264  #pragma init(libzfs_init_fru)
 192  265  static void
 193  266  libzfs_init_fru(void)
 194  267  {
 195  268          char path[MAXPATHLEN];
 196  269          char isa[257];
 197  270  
 198  271  #if defined(_LP64)
↓ open down ↓ 25 lines elided ↑ open up ↑
 224  297          _topo_hdl_strfree = (void (*)())
 225  298              dlsym(_topo_dlhandle, "topo_hdl_strfree");
 226  299          _topo_node_name = (char *(*)())
 227  300              dlsym(_topo_dlhandle, "topo_node_name");
 228  301          _topo_prop_get_string = (int (*)())
 229  302              dlsym(_topo_dlhandle, "topo_prop_get_string");
 230  303          _topo_node_fru = (int (*)())
 231  304              dlsym(_topo_dlhandle, "topo_node_fru");
 232  305          _topo_fmri_nvl2str = (int (*)())
 233  306              dlsym(_topo_dlhandle, "topo_fmri_nvl2str");
      307 +        _topo_fmri_str2nvl = (int (*)())
      308 +            dlsym(_topo_dlhandle, "topo_fmri_str2nvl");
 234  309          _topo_fmri_strcmp_noauth = (int (*)())
 235  310              dlsym(_topo_dlhandle, "topo_fmri_strcmp_noauth");
 236  311  
 237  312          if (_topo_open == NULL || _topo_close == NULL ||
 238  313              _topo_snap_hold == NULL || _topo_snap_release == NULL ||
 239  314              _topo_walk_init == NULL || _topo_walk_step == NULL ||
 240  315              _topo_walk_fini == NULL || _topo_hdl_strfree == NULL ||
 241  316              _topo_node_name == NULL || _topo_prop_get_string == NULL ||
 242  317              _topo_node_fru == NULL || _topo_fmri_nvl2str == NULL ||
 243      -            _topo_fmri_strcmp_noauth == NULL) {
      318 +            _topo_fmri_str2nvl == NULL || _topo_fmri_strcmp_noauth == NULL) {
 244  319                  (void) dlclose(_topo_dlhandle);
 245  320                  _topo_dlhandle = NULL;
 246  321          }
 247  322  }
 248  323  
 249  324  /*
 250  325   * Refresh the mappings from device path -> FMRI.  We do this by walking the
 251  326   * hc topology looking for disk nodes, and recording the io/devfs-path and FRU.
 252  327   * Note that we strip out the disk-specific authority information (serial,
 253  328   * part, revision, etc) so that we are left with only the identifying
↓ open down ↓ 31 lines elided ↑ open up ↑
 285  360              calloc(ZFS_FRU_HASH_SIZE, sizeof (void *))) == NULL)
 286  361                  return;
 287  362  
 288  363          /*
 289  364           * We now have a topo snapshot, so iterate over the hc topology looking
 290  365           * for disks to add to the hash.
 291  366           */
 292  367          twp = _topo_walk_init(thp, FM_FMRI_SCHEME_HC,
 293  368              libzfs_fru_gather, hdl, &err);
 294  369          if (twp != NULL) {
 295      -                (void) _topo_walk_step(twp, TOPO_WALK_CHILD);
      370 +                int status;
      371 +
      372 +                status = _topo_walk_step(twp, TOPO_WALK_CHILD);
      373 +                assert(status != TOPO_WALK_NEXT);
 296  374                  _topo_walk_fini(twp);
 297  375          }
 298  376  }
 299  377  
 300  378  /*
 301  379   * Given a devfs path, return the FRU for the device, if known.  This will
 302  380   * automatically call libzfs_fru_refresh() if it hasn't already been called by
 303  381   * the consumer.  The string returned is valid until the next call to
 304  382   * libzfs_fru_refresh().
 305  383   */
↓ open down ↓ 107 lines elided ↑ open up ↑
 413  491  
 414  492          chassisid += 12;
 415  493          len = strlen(hdl->libzfs_chassis_id);
 416  494          if (strncmp(chassisid, hdl->libzfs_chassis_id, len) == 0 &&
 417  495              (chassisid[len] == '/' || chassisid[len] == ':'))
 418  496                  return (B_FALSE);
 419  497  
 420  498          return (B_TRUE);
 421  499  }
 422  500  
      501 +/*
      502 + * Check if both FRUs belong to the same enclosure.
      503 + */
      504 +boolean_t
      505 +libzfs_fru_cmp_enclosure(const char *fru_a, const char *fru_b)
      506 +{
      507 +        int a, b;
      508 +        char *encl_a, *encl_b;
      509 +        const char *encl_str = "/ses-enclosure=";
      510 +        size_t encl_str_len = strlen(encl_str);
      511 +
      512 +        encl_a = strstr(fru_a, encl_str);
      513 +        encl_b = strstr(fru_b, encl_str);
      514 +        /* If both FRUs don't contain enclosure field, consider it a match */
      515 +        if (encl_a == NULL && encl_b == NULL)
      516 +                return (B_TRUE);
      517 +        /* If one FRU has the enclosure field, but the other one doesn't */
      518 +        if (encl_a == NULL || encl_b == NULL)
      519 +                return (B_FALSE);
      520 +
      521 +        encl_a += encl_str_len;
      522 +        encl_b += encl_str_len;
      523 +        if (sscanf(encl_a, "%d", &a) != 1 || sscanf(encl_b, "%d", &b) != 1)
      524 +                return (B_FALSE);
      525 +
      526 +        return (a == b);
      527 +}
      528 +
 423  529  /*
 424  530   * Clear memory associated with the FRU hash.
 425  531   */
 426  532  void
 427  533  libzfs_fru_clear(libzfs_handle_t *hdl, boolean_t final)
 428  534  {
 429  535          libzfs_fru_t *frup;
 430  536  
 431  537          while ((frup = hdl->libzfs_fru_list) != NULL) {
 432  538                  hdl->libzfs_fru_list = frup->zf_next;
↓ open down ↓ 20 lines elided ↑ open up ↑
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX