reconfig.c

/*
 * Copyright (C) 2006 IBM, all rights reserved.
 * Written by Darrick Wong <djwong@us.ibm.com>,
 * James Simshaw <simshawj@us.ibm.com>, and
 * Adam DiCarlo <bikko@us.ibm.com>
 *
 * Copyright (C) 2006-2008 Heinz Mauelshagen, Red Hat GmbH
 *                     All rights reserved
 *
 * Copyright (C) 2007   Intel Corporation. All rights reserved.
 * November, 2007 - additions for Create, Delete, Rebuild & Raid 10.
 *
 * See file LICENSE at the top of this source tree for license information.
 */
#include "internal.h"
int dso = 0;

#define     add_to_log(entry, log)  \
      list_add_tail(&(entry)->changes, &(log));

static inline int
alloc_entry(struct change **entry, struct lib_context *lc)
{
      return (*entry = dbg_malloc(sizeof(*entry))) ? 0 : -ENOMEM;
}

static int
nuke_spare(struct lib_context *lc, struct raid_dev *rd)
{
      printf("Nuking Spare\n");
      list_del_init(&rd->devs);
      return 0;
}

int
dso_end_rebuild(struct lib_context *lc, int arg)
{
      struct raid_set *sub_rs;
      const char *vol_name = lc->options[LC_REBUILD_SET].arg.str;

      sub_rs = find_set(lc, NULL, vol_name, FIND_ALL);
      if (sub_rs) {
            struct raid_set *rs = NULL;

            rs = find_group(lc, sub_rs);
            if ((rs) && S_OK(sub_rs->status)) {
                  struct raid_dev *check_rd = RD_RS(sub_rs);
                  enum status state = s_ok;

                  if (check_rd->fmt->metadata_handler)
                        state = check_rd->fmt->metadata_handler(lc,
                                                      GET_REBUILD_STATE,
                                                      NULL,
                                                      (void *)
                                                      sub_rs);

                  if (state != s_nosync) {
                        /* cannot rebuild */
                        log_print(lc,
                                "Volume \"%s\" is not in rebuild state (current: %u)\n",
                                sub_rs->name, state);
                        return 1;
                  }

                  if (check_rd->fmt->metadata_handler)
                        check_rd->fmt->metadata_handler(lc,
                                                UPDATE_REBUILD_STATE,
                                                NULL,
                                                (void *) rs);
            } else {
                  log_print(lc,
                          "Volume \"%s\" is not in rebuild state \n",
                          vol_name);
                  return 1;
            }
      } else
            log_print(lc, "raid volume \"%s\" not found\n", vol_name);

      return 0;
}

int rebuild_config_raidset(struct lib_context *lc, struct raid_set *rs);

void
show_raid_stack(struct lib_context *lc)
{
      struct raid_set *_rs;
      log_dbg(lc, "RM: Discovered raid sets:");
      list_for_each_entry(_rs, LC_RS(lc), list) {
            struct raid_dev *_rd;
            struct raid_set *_rss;
            log_dbg(lc, "RM: GROUP name: \"%s\"", _rs->name);
            list_for_each_entry(_rd, &_rs->devs, devs) {
                  log_dbg(lc, "RM: GROUP_DISK name: \"%s\"",
                        (_rd->di) ? _rd->di->path : "UNKNOWN");
            }
            list_for_each_entry(_rss, &_rs->sets, list) {
                  struct raid_dev *_rsd;
                  struct raid_set *_rsss;

                  log_dbg(lc, "RM:   SUPERSET name: \"%s\"", _rss->name);

                  list_for_each_entry(_rsd, &_rss->devs, devs) {
                        log_dbg(lc, "RM:   SUPERSET_DISK name: \"%s\"",
                              (_rsd->di) ? _rsd->
                              di->path : "UNKNOWN");
                  }

                  list_for_each_entry(_rsss, &_rss->sets, list) {
                        struct raid_dev *_rssd;
                        log_dbg(lc, "RM:     SUBSET name: \"%s\"",
                              _rsss->name);
                        list_for_each_entry(_rssd, &_rsss->devs, devs) {
                              log_dbg(lc,
                                    "RM:     SUBSET_DISK name: \"%s\"",
                                    (_rssd->di) ? _rssd->
                                    di->path : "UNKNOWN");
                        }
                  }
            }
      }
}

/* Add a device to a RAID1 set and start the resync */
static int
add_dev_to_raid(struct lib_context *lc, struct raid_set *rs,
            struct raid_dev *rd)
{
      int ret = 0;
      const char *vol_name = lc->options[LC_REBUILD_SET].arg.str;

      struct raid_set *sub_rs, *crs;
      struct raid_dev *check_rd;


      LIST_HEAD(log);         /* playback log */
      sub_rs = find_set(lc, NULL, vol_name, FIND_ALL);
      check_rd = list_entry(rs->devs.next, typeof(*rd), devs);

      if (rd) {
            if (check_rd->fmt->create) {
                  struct raid_dev *tmp;
                  if ((ret = check_rd->fmt->create(lc, rs))) {

                        list_for_each_entry(tmp, &rs->devs, devs) {
                              write_dev(lc, tmp, 0);
                        }
                  } else {
                        log_print(lc, "metadata fmt update failed\n");
                        goto err;
                  }
            } else {
                  log_print(lc, "create failed fmt handler missing\n");
                  goto err;
            }

            struct handler_info info;

            if (lc->options[LC_REBUILD_SET].opt) {
                  if (check_rd->fmt->metadata_handler) {
                        if (!check_rd->
                            fmt->metadata_handler(lc,
                                            GET_REBUILD_DRIVE_NO,
                                            &info,
                                            (void *) sub_rs)) {
                              LOG_ERR(lc, 0,
                                    "can't get rebuild drive !");
                        }
                  }
            }

            if (info.data.i32 != -1) {
                  struct raid_dev *before_rd, *tmp_rd;
                  int idx = 0;
                  list_for_each_entry_safe(before_rd, tmp_rd,
                                     &sub_rs->devs, devs) {
                        if ((idx == info.data.i32)
                            && (&rd->devs != &before_rd->devs)) {
                              list_del(&rd->devs)
                                    list_add_tail(&rd->devs,
                                                &before_rd->devs);
                              break;
                        }
                        idx++;
                  }
            }
            show_raid_stack(lc);
            log_dbg(lc, "RM: REBUILD drivie #: \"%d\"", info.data);
            show_raid_stack(lc);
      }

      /* Reconfigure device mapper */
      // FIXME: is nosync enough? rs->status |= s_inconsistent;
      rs->status = s_ok;
      if ((sub_rs = find_set(lc, NULL, vol_name, FIND_ALL))) {
            sub_rs->status = s_ok;

            list_for_each_entry(crs, &sub_rs->sets, list)
                  crs->status = s_ok;
      }

      change_set(lc, A_ACTIVATE, rs);
      rs->status |= s_nosync;

      if ((sub_rs = find_set(lc, NULL, vol_name, FIND_ALL))) {
            sub_rs->status |= s_nosync;
            list_for_each_entry(crs, &sub_rs->sets, list) {
                  crs->status |= s_nosync;
            }
      }

      ret = change_set(lc, A_RELOAD, rs);
      // FIXME: might need this later: change_set(lc, A_DEACTIVATE,rs);
      if (!ret)
            goto err;

      if (!dso) {
#ifdef DMRAID_LED
            struct raid_dev *_rd;
            list_for_each_entry(_rd, &sub_rs->devs, devs)
                  led(strrchr(_rd->di->path, '/') + 1, LED_REBUILD);
#endif

            if (check_rd->fmt->metadata_handler)
                  check_rd->fmt->metadata_handler(lc,
                                          UPDATE_REBUILD_STATE,
                                          NULL, (void *) rs);
      }

      /* End transaction */
      end_log(lc, &log);
      return 0;

      err:
      revert_log(lc, &log);
      return ret;
}

struct dev_info *find_disk(struct lib_context *lc, char *dp);

/* check if disk is under a raid set */
int
check_busy_disk(struct lib_context *lc, struct raid_dev *check_rd)
{
      struct raid_dev *rd;

      if (!check_rd)
            return 0;

      if (!check_rd->di)
            return 0;

      list_for_each_entry(rd, LC_RD(lc), list) {
            if (rd->di == check_rd->di)
                  return 0;
      }

      return 1;
}

int
add_dev_to_array(struct lib_context *lc, struct raid_set *rs,
             uint build_metadata, struct raid_dev *hot_spare_rd)
{
      const char *set_name = lc->options[LC_REBUILD_SET].arg.str;
      struct raid_dev *rd = NULL;
      struct raid_set *sub_rs;

      if ((hot_spare_rd) || (build_metadata)) {
            struct raid_dev tmp_rd;
            struct raid_dev *new_rd = hot_spare_rd;
            enum type type = t_raid1;

            sub_rs = find_set(lc, NULL, set_name, FIND_ALL);
            if (sub_rs == NULL) {
                  log_print(lc, "Volume \"%s\" not found\n", set_name);
                  return 1;
            }


            type = sub_rs->type;

            if (new_rd == NULL) {
                  const char *disk_name =
                        lc->options[LC_REBUILD_DISK].arg.str;

                  new_rd = &tmp_rd;
                  /* for non-hot-spare */
                  if (!(new_rd->di = find_disk(lc, (char *) disk_name)))
                        LOG_ERR(lc, 0, "failed to find disk %s",
                              disk_name);

                  /* check if disk is a member of another raid set */
                  if (!(check_busy_disk(lc, new_rd)))
                        LOG_ERR(lc, 0,
                              "disk %s cannot be used to rebuilding",
                              disk_name);

                  new_rd->fmt = get_format(rs);
            }

            /* add a rd to group raid set */
            if ((rd = alloc_raid_dev(lc, "rebuild")) == NULL)
                  LOG_ERR(lc, 1,
                        "failed to allocate space for a raid_dev");

            memset(rd, 0, sizeof(*rd));
            rd->name = NULL;

            rd->di = new_rd->di;
            rd->fmt = new_rd->fmt;

            rd->status = s_init;
            rd->type = type;
            rd->offset = 0;
            rd->sectors = 0;

            list_add_tail(&rd->list, LC_RD(lc));
            list_add_tail(&rd->devs, &rs->devs);

            /* add a spare to raid set */
            sub_rs = find_set(lc, NULL, set_name, FIND_ALL);

            if (!(rd = alloc_raid_dev(lc, __func__)))
                  LOG_ERR(lc, 1,
                        "failed to allocate space for a raid_dev");

            rd->name = NULL;
            rd->di = new_rd->di;
            rd->fmt = new_rd->fmt;
            rd->status = s_init;
            rd->type = type;
            rd->offset = 0;
            rd->sectors = 0;
            list_add_tail(&rd->devs, &sub_rs->devs);
            sub_rs->total_devs++;
      }

      add_dev_to_raid(lc, rs, rd);
      return 0;
}

/* Remove a disk from a raid1 */
static int
del_dev_in_raid1(struct lib_context *lc, struct raid_set *rs,
             struct raid_dev *rd)
{
      int ret;
      struct raid_dev *tmp;
      struct change *entry;
      LIST_HEAD(log);         /* Playback log */

      /* Remove device from the raid set */
      ret = alloc_entry(&entry, lc);
      if (ret)
            goto err;

      entry->type = DELETE_FROM_SET;
      entry->rs = rs;
      entry->rd = rd;
      add_to_log(entry, log);
      list_del_init(&rd->devs);
      rd->type = t_spare;

      /* Check that this is a sane configuration */
      list_for_each_entry(tmp, &rs->devs, devs) {
            ret = tmp->fmt->check(lc, rs);
            if (ret)
                  goto err;
      }

      /* Write the metadata of the drive we're removing _first_ */
      ret = alloc_entry(&entry, lc);
      if (ret)
            goto err;

      entry->type = WRITE_METADATA;
      entry->rd = rd;
      add_to_log(entry, log);
      ret = write_dev(lc, rd, 0);
      if (!ret)
            goto err;

      /* Write metadatas of every device in the set */
      list_for_each_entry(tmp, &rs->devs, devs) {
            if (tmp == rd)
                  continue;

            ret = alloc_entry(&entry, lc);
            if (ret)
                  goto err;

            entry->type = WRITE_METADATA;
            entry->rd = tmp;
            add_to_log(entry, log);
            ret = write_dev(lc, tmp, 0);
            if (!ret)
                  goto err;
      }

      /* Reconfigure device mapper */
      rs->status |= s_inconsistent;
      rs->status |= s_nosync;
      ret = change_set(lc, A_RELOAD, rs);
      if (!ret)
            goto err;

      /* End transaction */
      end_log(lc, &log);
      return 0;

      err:
      revert_log(lc, &log);
      return ret;
}

/* Corelate type and function to handle addition/removel of RAID device */
struct handler {
      enum change_type type;
      int (*func) (struct lib_context * lc, struct raid_set * rs,
                 struct raid_dev * rd);
};

/* Call the function to handle addition/removal of a RAID device */
static int
handle_dev(struct lib_context *lc, struct handler *h,
         struct raid_set *rs, struct raid_dev *rd)
{
      do {
            if (h->type == rs->type)
                  return h->func(lc, rs, rd);
      } while ((++h)->type != t_undef);

      LOG_ERR(lc, -ENOENT, "%s: no handler for %x", __func__, rs->type);
}

/* Add a disk to an array. */
int
add_dev_to_set(struct lib_context *lc, struct raid_set *rs, struct raid_dev *rd)
{
      struct handler handlers[] = {
            {t_raid1, add_dev_to_raid},
            {t_undef, NULL},
      };

      if (T_SPARE(rd))
            nuke_spare(lc, rd);
      else if (!list_empty(&rd->devs))
            LOG_ERR(lc, -EBUSY, "%s: disk already in another set!",
                  __func__);

      if (T_GROUP(rd))
            LOG_ERR(lc, -EISDIR,
                  "%s: can't add a group raid_dev to a raid_set.",
                  __func__);

      return handle_dev(lc, handlers, rs, rd);
}

/* Remove a disk from an array */
int
del_dev_in_set(struct lib_context *lc, struct raid_set *rs, struct raid_dev *rd)
{
      struct handler handlers[] = {
            {t_raid1, del_dev_in_raid1},
            {t_undef, NULL},
      };

      if (list_empty(&rd->devs))
            LOG_ERR(lc, -EBUSY, "%s: disk is not in a set!", __func__);

      /* FIXME: Not sure if this is true. */
      if (T_GROUP(rd))
            LOG_ERR(lc, -EISDIR,
                  "%s: can't remove a group raid_dev from a raid_set.",
                  __func__);

      return handle_dev(lc, handlers, rs, rd);
}

/*
 * Find group of raid_set to which sub_rs belongs 
 * Serves one level stacked raids.
 */
struct raid_set *
find_group(struct lib_context *lc, struct raid_set *sub_rs)
{
      struct raid_set *tmp = NULL, *r = NULL, *r2 = NULL;

      list_for_each_entry(tmp, LC_RS(lc), list) {
            if (T_GROUP(tmp)) {
                  list_for_each_entry(r, &tmp->sets, list) {
                        if (r == sub_rs)
                              return tmp;
                        else if (SETS(r)) {
                              list_for_each_entry(r2, &r->sets, list) {
                                    if (r2 == sub_rs)
                                          return tmp;
                              }
                        }
                  }
            }
      }

      /* Group not found. */
      return NULL;
}

struct raid_dev *find_spare(struct lib_context *lc, struct raid_set *sub_rs,
                      struct raid_set **spare_set);

int
_rebuild_raidset(struct lib_context *lc, struct raid_set *sub_rs,
             char *set_name)
{
      struct raid_set *spare_set = NULL, *rs = NULL;
      struct raid_dev *rd = NULL;
      int driveRebuild = 1;

      rs = find_group(lc, sub_rs);

      /* raid 0 cannot be rebuild - exit */
      if (T_RAID0(sub_rs) && (!SETS(sub_rs))) {
            log_print(lc, "Rebuild: raid0 cannot be rebuild\n");
            return 1;
      }
      /* FIXME - work-aroud for status reporting */
      if (S_BROKEN(sub_rs->status) ||
          S_INCONSISTENT(sub_rs->status)) {
            if (lc->options[LC_REBUILD_DISK].opt == 0) {
                  /* find the spare drive */
                  if ((rd = find_spare(lc, sub_rs, &spare_set)) == NULL) {
                        log_print(lc,
                                "Rebuild: a hot-spare drive not found for a volume: \"%s\"."
                                " Need a drive to rebuild a volume.\n",
                                sub_rs->name);
                        return 1;
                  }

            }
      } else if (S_OK(sub_rs->status)) {
            struct raid_dev *check_rd = RD_RS(sub_rs);
            enum status state = s_ok;

            if (check_rd && (check_rd->fmt->metadata_handler))
                  state = check_rd->fmt->metadata_handler(lc,
                                                GET_REBUILD_STATE,
                                                NULL, (void *)
                                                sub_rs);

            if (state != s_nosync) {
                  /* cannot rebuild */
                  log_print(lc,
                          "Volume \"%s\" is not in rebuild state (current: %u)",
                          sub_rs->name, state);
                  log_print(lc,
                          "Rebuild: cannot rebuild from current state!\n");
                  return 1;
            }
            driveRebuild = 0;
      } else if (!(S_NOSYNC(sub_rs->status))) {
            /* cannot rebuild */
            log_print(lc, "Rebuild: cannot rebuild from current state!\n");
            return 1;
      }


      sub_rs->status = s_nosync;
      rs->status = s_nosync;

      /* set the name for rebuild set (function down the path are using this variable to
       * retrive the raid set that is rebuild 
       */


      dbg_free((char *) lc->options[LC_REBUILD_SET].arg.str);
      lc->options[LC_REBUILD_SET].arg.str =
            (const char *) dbg_malloc(strlen(sub_rs->name) + 1);
      strcpy((char *) lc->options[LC_REBUILD_SET].arg.str, sub_rs->name);

      if (!(add_dev_to_array(lc, rs,
                         (driveRebuild
                        && lc->options[LC_REBUILD_DISK].opt)
                         || rd, rd))) {
            log_dbg(lc, "rebuild: raid \"%s\" rebuild finished\n",
                  set_name);
      } else {
            /* error - raid failed to be rebuilded */
            log_print(lc, "Rebuild: raid \"%s\" rebuild failed\n",
                    set_name);
            return 1;
      }
      return 0;
}


int
rebuild_raidset(struct lib_context *lc, char *set_name)
{
      struct raid_set *sub_rs = NULL;
      int ret = 0;

      sub_rs = find_set(lc, NULL, set_name, FIND_ALL);
      if (sub_rs) {
            /* for stacked subsets go throu the stack to retrive the subsets that:
               - do not contain subsets
               - are eligible for rebuild 
             */
            if (SETS(sub_rs)) {
                  enum status curr_state = s_ok;
                  struct raid_set *r = NULL;

                  /* check sub-set that are in Ok state */
                  curr_state = s_ok;

                  /* check for all subsets that have state equal curr_state */
                  list_for_each_entry(r, &sub_rs->sets, list) {
                        if (r->status & curr_state) {
                              ret |= _rebuild_raidset(lc, r,
                                                set_name);
                        }
                  }

                  /* TBD change to s_inconsisten when the states are reported in right way */
                  curr_state = s_broken | s_inconsistent;
                  list_for_each_entry(r, &sub_rs->sets, list) {
                        if (r->status & curr_state) {
                              ret |= _rebuild_raidset(lc, r,
                                                set_name);
                        }
                  }
            } else
                  ret |= _rebuild_raidset(lc, sub_rs, set_name);

      } else
            log_print(lc, "raid volume \"%s\" not found\n", set_name);

      return ret;
}


int
write_set_spare(struct lib_context *lc, void *v)
{
      int ret = 1;
      struct raid_set *r, *rs = v;
      struct raid_dev *rd;

      /* Decend hierarchy */
      list_for_each_entry(r, &rs->sets, list) {
            /*
             * FIXME: does it make sense to try the rest of the subset
             *        in case we fail writing one ?
             */
            if (!write_set_spare(lc, (void *) r))
                  log_err(lc, "writing RAID subset \"%s\", continuing",
                        r->name);
      }

      /* Write metadata to the RAID devices of a set */
      list_for_each_entry(rd, &rs->devs, devs) {
            /*
             * FIXME: does it make sense to try the rest of the
             *        devices in case we fail writing one ?
             */
            if (!(T_GROUP(rs))) {
                  if (!write_dev(lc, rd, 0)) {
                        log_err(lc,
                              "writing RAID device \"%s\", continuing",
                              rd->di->path);
                        ret = 0;
                  }
            }
      }

      return ret;
}



static int
add_spare_dev_to_raid(struct lib_context *lc, struct raid_set *rs)
{
      int ret = 0;
      struct dmraid_format *fmt_hand;


      /* find format handler */
      fmt_hand = get_format(rs);

      if (!(fmt_hand->create)) {
            LOG_ERR(lc, 0,
                  "metadata creation is not supported in \"%s\" format",
                  fmt_hand->name);
      } else {
            if ((ret = fmt_hand->create(lc, rs)))
                  ret = write_set_spare(lc, rs);

            if (!ret)
                  log_print(lc, "metadata fmt update failed\n");
      }

      return ret;
}

int
add_spare_dev_to_array(struct lib_context *lc, struct raid_set *rs)
{
      struct dev_info *di;
      struct dmraid_format *fmt_hand;
      struct raid_dev *rd;
      struct raid_set *rs_sub;

      const char *disk_name = lc->options[LC_REBUILD_DISK].arg.str;


      /* find format handler */
      fmt_hand = get_format(rs);

      /* add a spare rs to raid set */
      if (!(rs_sub = alloc_raid_set(lc, "rebuild")))
            return 0;

      rs_sub->name = NULL;
      rs_sub->size = 0;
      rs_sub->stride = 0;
      rs_sub->type = t_spare;
      rs_sub->flags = 0;
      rs_sub->status = s_init;
      list_add_tail(&rs_sub->list, &rs->sets);

      /* Find disk by name. */
      if (!(di = find_disk(lc, (char *) disk_name)))
            LOG_ERR(lc, 0, "failed to find disk %s", disk_name);

      /* Add a rd to group raid set. */
      if ((rd = alloc_raid_dev(lc, "rebuild")) == NULL)
            LOG_ERR(lc, 0, "failed to allocate space for a raid_dev");

      rd->name = NULL;
      rd->di = di;
      rd->fmt = fmt_hand;
      rd->status = s_init;
      rd->type = t_spare;
      rd->offset = 0;
      rd->sectors = 0;

      /* add dev to lc list and to group rs */
      list_add_tail(&rd->list, LC_RD(lc));
      list_add_tail(&rd->devs, &rs->devs);

      if (!(rd = alloc_raid_dev(lc, "rebuild")))
            LOG_ERR(lc, 0, "failed to allocate space for a raid_dev");

      rd->name = NULL;
      rd->di = di;
      rd->fmt = fmt_hand;
      rd->status = s_init;
      rd->type = t_spare;
      rd->offset = 0;
      rd->sectors = 0;
      list_add_tail(&rd->devs, &rs_sub->devs);
      return add_spare_dev_to_raid(lc, rs);
}


/* Add a disk to raid set as spare disk. */
int
hot_spare_add(struct lib_context *lc, struct raid_set *rs)
{
      const char *vol_name = lc->options[LC_HOT_SPARE_SET].arg.str;
      int ret_func;
      struct dmraid_format *fmt_hand;


      if (!(!OPT_FORMAT(lc) &&
            OPT_REBUILD_DISK(lc) &&
            OPT_HOT_SPARE_SET(lc)))
            return 0;

      if (!(fmt_hand = get_format(rs)))
            LOG_ERR(lc, 0, "unknown metadata format");

      if (!(fmt_hand->metadata_handler))
            LOG_ERR(lc, 0,
                  "metadata_handler() is not supported in \"%s\" format",
                  fmt_hand->name);

      ret_func = fmt_hand->metadata_handler(lc, CHECK_HOT_SPARE, NULL,
                                    (void *) rs);

      if (!ret_func)
            LOG_ERR(lc, 0,
                  "hot-spare cannot be added to existing raid "
                  "set \"%s\" in \"%s\" format",
                  vol_name, fmt_hand->name);

      return add_spare_dev_to_array(lc, rs);
}