]> git.ipfire.org Git - thirdparty/mdadm.git/blobdiff - Assemble.c
Introduce random_uuid() helper function
[thirdparty/mdadm.git] / Assemble.c
index fdb38e4e9064d5ae2abb7273711e141e67da093a..3da090330019928021e5a5d999b5575ae541a1fe 100644 (file)
@@ -1,7 +1,7 @@
 /*
  * mdadm - manage Linux "md" devices aka RAID arrays.
  *
- * Copyright (C) 2001-2013 Neil Brown <neilb@suse.de>
+ * Copyright (C) 2001-2016 Neil Brown <neilb@suse.com>
  *
  *
  *    This program is free software; you can redistribute it and/or modify
 #include       "mdadm.h"
 #include       <ctype.h>
 
-static int name_matches(char *found, char *required, char *homehost)
+static int name_matches(char *found, char *required, char *homehost, int require_homehost)
 {
        /* See if the name found matches the required name, possibly
         * prefixed with 'homehost'
         */
-       char fnd[33];
+       char *sep;
+       unsigned int l;
 
-       strncpy(fnd, found, 32);
-       fnd[32] = 0;
        if (strcmp(found, required)==0)
                return 1;
-       if (homehost) {
-               int l = strlen(homehost);
-               if (l < 32 && fnd[l] == ':' &&
-                   strcmp(fnd+l+1, required)==0)
+       sep = strchr(found, ':');
+       if (!sep)
+               return 0;
+       l = sep - found;
+       if (strncmp(found, "any:", 4) == 0 ||
+           (homehost && strcmp(homehost, "any") == 0) ||
+           !require_homehost ||
+           (homehost && strlen(homehost) == l &&
+            strncmp(found, homehost, l) == 0)) {
+               /* matching homehost */
+               if (strcmp(sep+1, required) == 0)
                        return 1;
        }
        return 0;
@@ -73,7 +79,7 @@ static int is_member_busy(char *metadata_version)
 static int ident_matches(struct mddev_ident *ident,
                         struct mdinfo *content,
                         struct supertype *tst,
-                        char *homehost,
+                        char *homehost, int require_homehost,
                         char *update, char *devname)
 {
 
@@ -85,7 +91,7 @@ static int ident_matches(struct mddev_ident *ident,
                return 0;
        }
        if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
-           name_matches(content->name, ident->name, homehost)==0) {
+           name_matches(content->name, ident->name, homehost, require_homehost)==0) {
                if (devname)
                        pr_err("%s has wrong name.\n", devname);
                return 0;
@@ -233,8 +239,7 @@ static int select_devices(struct mddev_dev *devlist,
                                   !conf_test_metadata(tst->ss->name, (pol = devid_policy(stb.st_rdev)),
                                                       tst->ss->match_home(tst, c->homehost) == 1)) {
                                if (report_mismatch)
-                                       pr_err("%s has metadata type %s for which "
-                                              "auto-assembly is disabled\n",
+                                       pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
                                               devname, tst->ss->name);
                                tmpdev->used = 2;
                        } else
@@ -245,7 +250,9 @@ static int select_devices(struct mddev_dev *devlist,
                                        pr_err("no recogniseable superblock on %s\n",
                                               devname);
                                tmpdev->used = 2;
-                       } else if (tst->ss->load_super(tst,dfd, NULL)) {
+                       } else if ((tst->ignore_hw_compat = 0),
+                                  tst->ss->load_super(tst, dfd,
+                                                      report_mismatch ? devname : NULL)) {
                                if (report_mismatch)
                                        pr_err("no RAID superblock on %s\n",
                                               devname);
@@ -259,8 +266,7 @@ static int select_devices(struct mddev_dev *devlist,
                                   !conf_test_metadata(tst->ss->name, (pol = devid_policy(stb.st_rdev)),
                                                       tst->ss->match_home(tst, c->homehost) == 1)) {
                                if (report_mismatch)
-                                       pr_err("%s has metadata type %s for which "
-                                              "auto-assembly is disabled\n",
+                                       pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
                                               devname, tst->ss->name);
                                tmpdev->used = 2;
                        }
@@ -327,7 +333,8 @@ static int select_devices(struct mddev_dev *devlist,
                             content = content->next) {
 
                                if (!ident_matches(ident, content, tst,
-                                                  c->homehost, c->update,
+                                                  c->homehost, c->require_homehost,
+                                                  c->update,
                                                   report_mismatch ? devname : NULL))
                                        /* message already printed */;
                                else if (is_member_busy(content->text_version)) {
@@ -350,8 +357,7 @@ static int select_devices(struct mddev_dev *devlist,
 
                        st = tst; tst = NULL;
                        if (!auto_assem && inargv && tmpdev->next != NULL) {
-                               pr_err("%s is a container, but is not "
-                                      "only device given: confused and aborting\n",
+                               pr_err("%s is a container, but is not only device given: confused and aborting\n",
                                       devname);
                                st->ss->free_super(st);
                                dev_policy_free(pol);
@@ -366,36 +372,43 @@ static int select_devices(struct mddev_dev *devlist,
                        tmpdev = NULL;
                        goto loop;
                } else {
-                       int rv = 0;
-                       struct mddev_ident *match;
-
                        content = *contentp;
                        tst->ss->getinfo_super(tst, content, NULL);
 
                        if (!ident_matches(ident, content, tst,
-                                          c->homehost, c->update,
+                                          c->homehost, c->require_homehost,
+                                          c->update,
                                           report_mismatch ? devname : NULL))
                                goto loop;
 
-                       match = conf_match(tst, content, devname,
-                                          report_mismatch ? c->verbose : -1,
-                                          &rv);
-                       if (!match && rv == 2)
-                               goto loop;
-                       if (match && match->devname &&
-                           strcasecmp(match->devname, "<ignore>") == 0) {
-                               if (report_mismatch)
-                                       pr_err("%s is a member of an explicitly ignored array\n",
-                                              devname);
-                               goto loop;
-                       }
-                       if (match && !ident_matches(match, content, tst,
-                                                   c->homehost, c->update,
-                                                   report_mismatch ? devname : NULL))
-                               /* Array exists  in mdadm.conf but some
-                                * details don't match, so reject it
+                       if (auto_assem) {
+                               /* Never auto-assemble things that conflict
+                                * with mdadm.conf in some way
                                 */
-                               goto loop;
+                               struct mddev_ident *match;
+                               int rv = 0;
+
+                               match = conf_match(tst, content, devname,
+                                                  report_mismatch ? c->verbose : -1,
+                                                  &rv);
+                               if (!match && rv == 2)
+                                       goto loop;
+                               if (match && match->devname &&
+                                   strcasecmp(match->devname, "<ignore>") == 0) {
+                                       if (report_mismatch)
+                                               pr_err("%s is a member of an explicitly ignored array\n",
+                                                      devname);
+                                       goto loop;
+                               }
+                               if (match && !ident_matches(match, content, tst,
+                                                           c->homehost, c->require_homehost,
+                                                           c->update,
+                                                           report_mismatch ? devname : NULL))
+                                       /* Array exists  in mdadm.conf but some
+                                        * details don't match, so reject it
+                                        */
+                                       goto loop;
+                       }
 
                        /* should be safe to try an exclusive open now, we
                         * have rejected anything that some other mdadm might
@@ -551,7 +564,7 @@ struct devs {
 };
 
 static int load_devices(struct devs *devices, char *devmap,
-                       struct mddev_ident *ident, struct supertype *st,
+                       struct mddev_ident *ident, struct supertype **stp,
                        struct mddev_dev *devlist, struct context *c,
                        struct mdinfo *content,
                        int mdfd, char *mddev,
@@ -567,36 +580,28 @@ static int load_devices(struct devs *devices, char *devmap,
        int most_recent = -1;
        int bestcnt = 0;
        int *best = *bestp;
+       struct supertype *st = *stp;
 
        for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) {
                char *devname = tmpdev->devname;
                struct stat stb;
+               struct supertype *tst;
                int i;
+               int dfd;
 
                if (tmpdev->used != 1)
                        continue;
                /* looks like a good enough match to update the super block if needed */
 #ifndef MDASSEMBLE
                if (c->update) {
-                       int dfd;
                        /* prepare useful information in info structures */
                        struct stat stb2;
-                       struct supertype *tst;
                        int err;
                        fstat(mdfd, &stb2);
 
-                       if (strcmp(c->update, "uuid")==0 &&
-                           !ident->uuid_set) {
-                               int rfd;
-                               if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
-                                   read(rfd, ident->uuid, 16) != 16) {
-                                       *(__u32*)(ident->uuid) = random();
-                                       *(__u32*)(ident->uuid+1) = random();
-                                       *(__u32*)(ident->uuid+2) = random();
-                                       *(__u32*)(ident->uuid+3) = random();
-                               }
-                               if (rfd >= 0) close(rfd);
-                       }
+                       if (strcmp(c->update, "uuid") == 0 && !ident->uuid_set)
+                               random_uuid((__u8 *)ident->uuid);
+
                        dfd = dev_open(devname,
                                       tmpdev->disposition == 'I'
                                       ? O_RDWR : (O_RDWR|O_EXCL));
@@ -610,6 +615,9 @@ static int load_devices(struct devs *devices, char *devmap,
                                close(mdfd);
                                free(devices);
                                free(devmap);
+                               tst->ss->free_super(tst);
+                               free(tst);
+                               *stp = st;
                                return -1;
                        }
                        tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
@@ -620,6 +628,19 @@ static int load_devices(struct devs *devices, char *devmap,
 
                        if (strcmp(c->update, "byteorder") == 0)
                                err = 0;
+                       else if (strcmp(c->update, "home-cluster") == 0) {
+                               tst->cluster_name = c->homecluster;
+                               err = tst->ss->write_bitmap(tst, dfd, NameUpdate);
+                       } else if (strcmp(c->update, "nodes") == 0) {
+                               tst->nodes = c->nodes;
+                               err = tst->ss->write_bitmap(tst, dfd, NodeNumUpdate);
+                       } else if (strcmp(c->update, "revert-reshape") == 0 &&
+                                  c->invalid_backup)
+                               err = tst->ss->update_super(tst, content,
+                                                           "revert-reshape-nobackup",
+                                                           devname, c->verbose,
+                                                           ident->uuid_set,
+                                                           c->homehost);
                        else
                                err = tst->ss->update_super(tst, content, c->update,
                                                            devname, c->verbose,
@@ -627,8 +648,7 @@ static int load_devices(struct devs *devices, char *devmap,
                                                            c->homehost);
                        if (err < 0) {
                                if (err == -1)
-                                       pr_err("--update=%s not understood"
-                                              " for %s metadata\n",
+                                       pr_err("--update=%s not understood for %s metadata\n",
                                               c->update, tst->ss->name);
                                tst->ss->free_super(tst);
                                free(tst);
@@ -636,6 +656,7 @@ static int load_devices(struct devs *devices, char *devmap,
                                close(dfd);
                                free(devices);
                                free(devmap);
+                               *stp = st;
                                return -1;
                        }
                        if (strcmp(c->update, "uuid")==0 &&
@@ -646,7 +667,6 @@ static int load_devices(struct devs *devices, char *devmap,
                        if (tst->ss->store_super(tst, dfd))
                                pr_err("Could not re-write superblock on %s.\n",
                                       devname);
-                       close(dfd);
 
                        if (strcmp(c->update, "uuid")==0 &&
                            ident->bitmap_fd >= 0 && !bitmap_done) {
@@ -657,15 +677,13 @@ static int load_devices(struct devs *devices, char *devmap,
                                else
                                        bitmap_done = 1;
                        }
-                       tst->ss->free_super(tst);
                } else
 #endif
                {
-                       struct supertype *tst = dup_super(st);
-                       int dfd;
                        dfd = dev_open(devname,
                                       tmpdev->disposition == 'I'
                                       ? O_RDWR : (O_RDWR|O_EXCL));
+                       tst = dup_super(st);
 
                        if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
                                pr_err("cannot re-read metadata from %s - aborting\n",
@@ -675,14 +693,16 @@ static int load_devices(struct devs *devices, char *devmap,
                                close(mdfd);
                                free(devices);
                                free(devmap);
+                               tst->ss->free_super(tst);
+                               free(tst);
+                               *stp = st;
                                return -1;
                        }
                        tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
-                       tst->ss->free_super(tst);
-                       close(dfd);
                }
 
-               stat(devname, &stb);
+               fstat(dfd, &stb);
+               close(dfd);
 
                if (c->verbose > 0)
                        pr_err("%s is identified as a member of %s, slot %d%s.\n",
@@ -698,16 +718,22 @@ static int load_devices(struct devs *devices, char *devmap,
                if (devices[devcnt].i.disk.state == 6) {
                        if (most_recent < 0 ||
                            devices[devcnt].i.events
-                           > devices[most_recent].i.events)
+                           > devices[most_recent].i.events) {
+                               struct supertype *tmp = tst;
+                               tst = st;
+                               st = tmp;
                                most_recent = devcnt;
+                       }
                }
+               tst->ss->free_super(tst);
+               free(tst);
 
                if (content->array.level == LEVEL_MULTIPATH)
                        /* with multipath, the raid_disk from the superblock is meaningless */
                        i = devcnt;
                else
                        i = devices[devcnt].i.disk.raid_disk;
-               if (i+1 == 0) {
+               if (i+1 == 0 || i == MD_DISK_ROLE_JOURNAL) {
                        if (nextspare < content->array.raid_disks*2)
                                nextspare = content->array.raid_disks*2;
                        i = nextspare++;
@@ -746,12 +772,9 @@ static int load_devices(struct devs *devices, char *devmap,
                                 * Could be a mis-detection caused by overlapping
                                 * partitions.  fail-safe.
                                 */
-                               pr_err("WARNING %s and %s appear"
-                                      " to have very similar superblocks.\n"
-                                      "      If they are really different, "
-                                      "please --zero the superblock on one\n"
-                                      "      If they are the same or overlap,"
-                                      " please remove one from %s.\n",
+                               pr_err("WARNING %s and %s appear to have very similar superblocks.\n"
+                                      "      If they are really different, please --zero the superblock on one\n"
+                                      "      If they are the same or overlap, please remove one from %s.\n",
                                       devices[best[i]].devname, devname,
                                       inargv ? "the list" :
                                       "the\n      DEVICE list in mdadm.conf"
@@ -759,6 +782,7 @@ static int load_devices(struct devs *devices, char *devmap,
                                close(mdfd);
                                free(devices);
                                free(devmap);
+                               *stp = st;
                                return -1;
                        }
                        if (best[i] == -1
@@ -772,6 +796,7 @@ static int load_devices(struct devs *devices, char *devmap,
                *most_recentp = most_recent;
        *bestcntp = bestcnt;
        *bestp = best;
+       *stp = st;
        return devcnt;
 }
 
@@ -807,12 +832,37 @@ static int force_array(struct mdinfo *content,
                     i < content->array.raid_disks * 2 && i < bestcnt;
                     i += 2) {
                        int j = best[i];
-                       if (j>=0 &&
-                           !devices[j].uptodate &&
-                           devices[j].i.recovery_start == MaxSector &&
-                           (chosen_drive < 0 ||
+                       if (j < 0)
+                               continue;
+                       if (devices[j].uptodate)
+                               continue;
+                       if (devices[j].i.recovery_start != MaxSector) {
+                               int delta;
+                               if (!devices[j].i.reshape_active ||
+                                   devices[j].i.delta_disks <= 0)
+                                       continue;
+                               /* When increasing number of devices, an
+                                * added device also appears to be
+                                * recovering.  It is safe to include it
+                                * as long as it won't be a source of
+                                * data.
+                                * For now, just allow for last data
+                                * devices in RAID4 or last devices in RAID4/5/6.
+                                */
+                               delta = devices[j].i.delta_disks;
+                               if (devices[j].i.array.level >= 4 &&
+                                   devices[j].i.array.level <= 6 &&
+                                   i/2 >= content->array.raid_disks - delta)
+                                       /* OK */;
+                               else if (devices[j].i.array.level == 4 &&
+                                        i/2 >= content->array.raid_disks - delta - 1)
+                                       /* OK */;
+                               else
+                                       continue;
+                       }
+                       if (chosen_drive < 0 ||
                             devices[j].i.events
-                            > devices[chosen_drive].i.events))
+                           > devices[chosen_drive].i.events)
                                chosen_drive = j;
                }
                if (chosen_drive < 0)
@@ -861,7 +911,6 @@ static int force_array(struct mdinfo *content,
                avail[chosen_drive] = 1;
                okcnt++;
                tst->ss->free_super(tst);
-
                /* If there are any other drives of the same vintage,
                 * add them in as well.  We can't lose and we might gain
                 */
@@ -892,6 +941,7 @@ static int start_array(int mdfd,
                       unsigned int okcnt,
                       unsigned int sparecnt,
                       unsigned int rebuilding_cnt,
+                      unsigned int journalcnt,
                       struct context *c,
                       int clean, char *avail,
                       int start_partial_ok,
@@ -903,6 +953,15 @@ static int start_array(int mdfd,
        int i;
        unsigned int req_cnt;
 
+       if (content->journal_device_required && (content->journal_clean == 0)) {
+               if (!c->force) {
+                       pr_err("Not safe to assemble with missing or stale journal device, consider --force.\n");
+                       return 1;
+               }
+               pr_err("Journal is missing or stale, starting array read only.\n");
+               c->readonly = 1;
+       }
+
        rv = set_array_info(mdfd, st, content);
        if (rv && !err_ok) {
                pr_err("failed to set array info for %s: %s\n",
@@ -950,8 +1009,7 @@ static int start_array(int mdfd,
                        rv = add_disk(mdfd, st, content, &devices[j].i);
 
                        if (rv) {
-                               pr_err("failed to add "
-                                      "%s to %s: %s\n",
+                               pr_err("failed to add %s to %s: %s\n",
                                       devices[j].devname,
                                       mddev,
                                       strerror(errno));
@@ -975,19 +1033,35 @@ static int start_array(int mdfd,
                } else if (c->verbose > 0 && i < content->array.raid_disks*2
                           && (i&1) == 0)
                        pr_err("no uptodate device for slot %d of %s\n",
-                              i, mddev);
+                              i/2, mddev);
        }
 
        if (content->array.level == LEVEL_CONTAINER) {
                if (c->verbose >= 0) {
-                       pr_err("Container %s has been "
-                              "assembled with %d drive%s",
-                              mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
+                       pr_err("Container %s has been assembled with %d drive%s",
+                              mddev, okcnt+sparecnt+journalcnt,
+                              okcnt+sparecnt+journalcnt==1?"":"s");
                        if (okcnt < (unsigned)content->array.raid_disks)
                                fprintf(stderr, " (out of %d)",
                                        content->array.raid_disks);
                        fprintf(stderr, "\n");
                }
+
+               if (st->ss->validate_container) {
+                       struct mdinfo *devices_list;
+                       struct mdinfo *info_devices = xmalloc(sizeof(struct mdinfo)*(okcnt+sparecnt));
+                       unsigned int count;
+                       devices_list = NULL;
+                       for (count = 0; count < okcnt+sparecnt; count++) {
+                               info_devices[count] = devices[count].i;
+                               info_devices[count].next = devices_list;
+                               devices_list = &info_devices[count];
+                       }
+                       if (st->ss->validate_container(devices_list))
+                               pr_err("Mismatch detected!\n");
+                       free(info_devices);
+               }
+
                st->ss->free_super(st);
                sysfs_uevent(content, "change");
                if (err_ok && okcnt < (unsigned)content->array.raid_disks)
@@ -1031,7 +1105,7 @@ static int start_array(int mdfd,
                                           "array_state", "readonly");
                        if (rv == 0)
                                rv = Grow_continue(mdfd, st, content,
-                                                  c->backup_file,
+                                                  c->backup_file, 0,
                                                   c->freeze_reshape);
                } else if (c->readonly &&
                           sysfs_attribute_available(
@@ -1052,6 +1126,8 @@ static int start_array(int mdfd,
                                        fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
                                if (sparecnt)
                                        fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
+                               if (content->journal_clean)
+                                       fprintf(stderr, " and %d journal", journalcnt);
                                fprintf(stderr, ".\n");
                        }
                        if (content->reshape_active &&
@@ -1060,12 +1136,16 @@ static int start_array(int mdfd,
                                /* might need to increase the size
                                 * of the stripe cache - default is 256
                                 */
-                               if (256 < 4 * (content->array.chunk_size/4096)) {
+                               int chunk_size = content->array.chunk_size;
+                               if (content->reshape_active &&
+                                   content->new_chunk > chunk_size)
+                                       chunk_size = content->new_chunk;
+                               if (256 < 4 * ((chunk_size+4065)/4096)) {
                                        struct mdinfo *sra = sysfs_read(mdfd, NULL, 0);
                                        if (sra)
                                                sysfs_set_num(sra, NULL,
                                                              "stripe_cache_size",
-                                                             (4 * content->array.chunk_size / 4096) + 1);
+                                                             (4 * chunk_size / 4096) + 1);
                                        sysfs_free(sra);
                                }
                        }
@@ -1107,15 +1187,12 @@ static int start_array(int mdfd,
 
                if (!enough(content->array.level, content->array.raid_disks,
                            content->array.layout, 1, avail))
-                       pr_err("Not enough devices to "
-                              "start the array.\n");
+                       pr_err("Not enough devices to start the array.\n");
                else if (!enough(content->array.level,
                                 content->array.raid_disks,
                                 content->array.layout, clean,
                                 avail))
-                       pr_err("Not enough devices to "
-                              "start the array while not clean "
-                              "- consider --force.\n");
+                       pr_err("Not enough devices to start the array while not clean - consider --force.\n");
 
                return 1;
        }
@@ -1140,9 +1217,7 @@ static int start_array(int mdfd,
                                 content->array.raid_disks,
                                 content->array.layout, clean,
                                 avail))
-                       fprintf(stderr, " - not enough to start the "
-                               "array while not clean - consider "
-                               "--force.\n");
+                       fprintf(stderr, " - not enough to start the array while not clean - consider --force.\n");
                else {
                        if (req_cnt == (unsigned)content->array.raid_disks)
                                fprintf(stderr, " - need all %d to start it", req_cnt);
@@ -1224,7 +1299,8 @@ int Assemble(struct supertype *st, char *mddev,
        int *best = NULL; /* indexed by raid_disk */
        int bestcnt = 0;
        int devcnt;
-       unsigned int okcnt, sparecnt, rebuilding_cnt, replcnt;
+       unsigned int okcnt, sparecnt, rebuilding_cnt, replcnt, journalcnt;
+       int journal_clean = 0;
        int i;
        int was_forced = 0;
        int most_recent = 0;
@@ -1280,7 +1356,7 @@ try_again:
                       mddev ? mddev : "further assembly");
 
        content = &info;
-       if (st)
+       if (st && c->force)
                st->ignore_hw_compat = 1;
        num_devs = select_devices(devlist, ident, &st, &content, c,
                                  inargv, auto_assem);
@@ -1304,7 +1380,10 @@ try_again:
         */
        if (map_lock(&map))
                pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
-       mp = map_by_uuid(&map, content->uuid);
+       if (c->update && strcmp(c->update,"uuid") == 0)
+               mp = NULL;
+       else
+               mp = map_by_uuid(&map, content->uuid);
        if (mp) {
                struct mdinfo *dv;
                /* array already exists. */
@@ -1432,7 +1511,7 @@ try_again:
        /* Ok, no bad inconsistancy, we can try updating etc */
        devices = xcalloc(num_devs, sizeof(*devices));
        devmap = xcalloc(num_devs, content->array.raid_disks);
-       devcnt = load_devices(devices, devmap, ident, st, devlist,
+       devcnt = load_devices(devices, devmap, ident, &st, devlist,
                              c, content, mdfd, mddev,
                              &most_recent, &bestcnt, &best, inargv);
        if (devcnt < 0)
@@ -1462,6 +1541,7 @@ try_again:
        okcnt = 0;
        replcnt = 0;
        sparecnt=0;
+       journalcnt=0;
        rebuilding_cnt=0;
        for (i=0; i< bestcnt; i++) {
                int j = best[i];
@@ -1472,8 +1552,13 @@ try_again:
                /* note: we ignore error flags in multipath arrays
                 * as they don't make sense
                 */
-               if (content->array.level != LEVEL_MULTIPATH)
-                       if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
+               if (content->array.level != LEVEL_MULTIPATH) {
+                       if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL)) {
+                               if (content->journal_device_required)
+                                       journalcnt++;
+                               else    /* unexpected journal, mark as faulty */
+                                       devices[j].i.disk.state |= (1<<MD_DISK_FAULTY);
+                       } else if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
                                if (!(devices[j].i.disk.state
                                      & (1<<MD_DISK_FAULTY))) {
                                        devices[j].uptodate = 1;
@@ -1481,6 +1566,7 @@ try_again:
                                }
                                continue;
                        }
+               }
                /* If this device thinks that 'most_recent' has failed, then
                 * we must reject this device.
                 */
@@ -1504,12 +1590,12 @@ try_again:
                    devices[most_recent].i.events
                        ) {
                        devices[j].uptodate = 1;
+                       if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL))
+                               journal_clean = 1;
                        if (i < content->array.raid_disks * 2) {
                                if (devices[j].i.recovery_start == MaxSector ||
                                    (content->reshape_active &&
-                                    ((i >= content->array.raid_disks - content->delta_disks) ||
-                                     (i >= content->array.raid_disks - content->delta_disks - 1
-                                      && content->array.level == 4)))) {
+                                    i >= content->array.raid_disks - content->delta_disks)) {
                                        if (!avail[i/2]) {
                                                okcnt++;
                                                avail[i/2]=1;
@@ -1517,7 +1603,7 @@ try_again:
                                                replcnt++;
                                } else
                                        rebuilding_cnt++;
-                       } else
+                       } else if (devices[j].i.disk.raid_disk != MD_DISK_ROLE_JOURNAL)
                                sparecnt++;
                }
        }
@@ -1577,11 +1663,15 @@ try_again:
 #ifndef MDASSEMBLE
        sysfs_init(content, mdfd, NULL);
 #endif
+       /* after reload context, store journal_clean in context */
+       content->journal_clean = journal_clean;
        for (i=0; i<bestcnt; i++) {
                int j = best[i];
                unsigned int desired_state;
 
-               if (i >= content->array.raid_disks * 2)
+               if (devices[j].i.disk.raid_disk == MD_DISK_ROLE_JOURNAL)
+                       desired_state = (1<<MD_DISK_JOURNAL);
+               else if (i >= content->array.raid_disks * 2)
                        desired_state = 0;
                else if (i & 1)
                        desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_REPLACEMENT);
@@ -1664,9 +1754,10 @@ try_again:
                int err = 0;
                int *fdlist = xmalloc(sizeof(int)* bestcnt);
                if (c->verbose > 0)
-                       pr_err(":%s has an active reshape - checking "
-                              "if critical section needs to be restored\n",
+                       pr_err("%s has an active reshape - checking if critical section needs to be restored\n",
                               chosen_name);
+               if (!c->backup_file)
+                       c->backup_file = locate_backup(content->sys_name);
                enable_fds(bestcnt/2);
                for (i = 0; i < bestcnt/2; i++) {
                        int j = best[i*2];
@@ -1691,8 +1782,7 @@ try_again:
                                                   c->backup_file, c->verbose > 0);
                        if (err && c->invalid_backup) {
                                if (c->verbose > 0)
-                                       pr_err("continuing"
-                                              " without restoring backup\n");
+                                       pr_err("continuing without restoring backup\n");
                                err = 0;
                        }
                }
@@ -1728,7 +1818,7 @@ try_again:
        rv = start_array(mdfd, mddev, content,
                         st, ident, best, bestcnt,
                         chosen_drive, devices, okcnt, sparecnt,
-                        rebuilding_cnt,
+                        rebuilding_cnt, journalcnt,
                         c,
                         clean, avail, start_partial_ok,
                         pre_exist != NULL,
@@ -1778,18 +1868,17 @@ int assemble_container_content(struct supertype *st, int mdfd,
                               struct mdinfo *content, struct context *c,
                               char *chosen_name, int *result)
 {
-       struct mdinfo *dev, *sra;
+       struct mdinfo *dev, *sra, *dev2;
        int working = 0, preexist = 0;
        int expansion = 0;
-       struct map_ent *map = NULL;
        int old_raid_disks;
        int start_reshape;
-       char *avail = NULL;
+       char *avail;
        int err;
 
        sysfs_init(content, mdfd, NULL);
 
-       sra = sysfs_read(mdfd, NULL, GET_VERSION);
+       sra = sysfs_read(mdfd, NULL, GET_VERSION|GET_DEVS);
        if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0) {
                if (content->array.major_version == -1 &&
                    content->array.minor_version == -2 &&
@@ -1797,8 +1886,7 @@ int assemble_container_content(struct supertype *st, int mdfd,
                    content->text_version[0] == '/')
                        content->text_version[0] = '-';
                if (sysfs_set_array(content, md_get_version(mdfd)) != 0) {
-                       if (sra)
-                               sysfs_free(sra);
+                       sysfs_free(sra);
                        return 1;
                }
        }
@@ -1816,8 +1904,22 @@ int assemble_container_content(struct supertype *st, int mdfd,
        if (st->ss->external && content->recovery_blocked && start_reshape)
                block_subarray(content);
 
-       if (sra)
-               sysfs_free(sra);
+       for (dev2 = sra->devs; dev2; dev2 = dev2->next) {
+               for (dev = content->devs; dev; dev = dev->next)
+                       if (dev2->disk.major == dev->disk.major &&
+                           dev2->disk.minor == dev->disk.minor)
+                               break;
+               if (dev)
+                       continue;
+               /* Don't want this one any more */
+               if (sysfs_set_str(sra, dev2, "slot", "none") < 0 &&
+                   errno == EBUSY) {
+                       pr_err("Cannot remove old device %s: not updating %s\n", dev2->sys_name, sra->sys_name);
+                       sysfs_free(sra);
+                       return 1;
+               }
+               sysfs_set_str(sra, dev2, "state", "remove");
+       }
        old_raid_disks = content->array.raid_disks - content->delta_disks;
        avail = xcalloc(content->array.raid_disks, 1);
        for (dev = content->devs; dev; dev = dev->next) {
@@ -1832,15 +1934,14 @@ int assemble_container_content(struct supertype *st, int mdfd,
                } else if (errno == EEXIST)
                        preexist++;
        }
+       sysfs_free(sra);
        if (working + expansion == 0 && c->runstop <= 0) {
                free(avail);
                return 1;/* Nothing new, don't try to start */
        }
-       map_update(&map, fd2devnm(mdfd),
-                  content->text_version,
+       map_update(NULL, fd2devnm(mdfd), content->text_version,
                   content->uuid, chosen_name);
 
-
        if (enough(content->array.level, content->array.raid_disks,
                   content->array.layout, content->array.state & 1, avail) == 0) {
                if (c->export && result)
@@ -1879,7 +1980,7 @@ int assemble_container_content(struct supertype *st, int mdfd,
                int spare = content->array.raid_disks + expansion;
                if (restore_backup(st, content,
                                   working,
-                                  spare, c->backup_file, c->verbose) == 1)
+                                  spare, &c->backup_file, c->verbose) == 1)
                        return 1;
 
                err = sysfs_set_str(content, NULL,
@@ -1897,7 +1998,7 @@ int assemble_container_content(struct supertype *st, int mdfd,
                }
 
                err = Grow_continue(mdfd, st, content, c->backup_file,
-                                   c->freeze_reshape);
+                                   0, c->freeze_reshape);
        } else switch(content->array.level) {
                case LEVEL_LINEAR:
                case LEVEL_MULTIPATH: