]> git.ipfire.org Git - thirdparty/mdadm.git/blobdiff - Assemble.c
Can now mostly assemble DDF arrays
[thirdparty/mdadm.git] / Assemble.c
index 1dc0ffc15bf171d313ee4a9e791e2575886aa791..f10491b63f48b2c7849aaebf7c94c13ddea3f32c 100644 (file)
  */
 
 #include       "mdadm.h"
+#include       <ctype.h>
+
+static int name_matches(char *found, char *required, char *homehost)
+{
+       /* See if the name found matches the required name, possibly
+        * prefixed with 'homehost'
+        */
+       char fnd[33];
+
+       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)
+                       return 1;
+       }
+       return 0;
+}
 
 int Assemble(struct supertype *st, char *mddev, int mdfd,
-            mddev_ident_t ident, char *conffile,
+            mddev_ident_t ident,
             mddev_dev_t devlist, char *backup_file,
             int readonly, int runstop,
             char *update, char *homehost,
@@ -90,18 +111,16 @@ int Assemble(struct supertype *st, char *mddev, int mdfd,
         *    START_ARRAY
         *
         */
+       int clean = 0;
+       int must_close = 0;
        int old_linux = 0;
-       int vers;
-       void *first_super = NULL, *super = NULL;
+       int vers = 0; /* Keep gcc quite - it really is initialised */
        struct {
                char *devname;
-               unsigned int major, minor;
-               unsigned int oldmajor, oldminor;
-               long long events;
-               int uptodate;
-               int state;
-               int raid_disk;
-               int disk_nr;
+               int uptodate; /* set once we decide that this device is as
+                              * recent as everything else in the array.
+                              */
+               struct mdinfo i;
        } *devices;
        int *best = NULL; /* indexed by raid_disk */
        unsigned int bestcnt = 0;
@@ -113,33 +132,36 @@ int Assemble(struct supertype *st, char *mddev, int mdfd,
        int chosen_drive;
        int change = 0;
        int inargv = 0;
-       int start_partial_ok = (runstop >= 0) && (force || devlist==NULL);
+       int bitmap_done;
+       int start_partial_ok = (runstop >= 0) && (force || devlist==NULL || mdfd < 0);
        unsigned int num_devs;
        mddev_dev_t tmpdev;
        struct mdinfo info;
        char *avail;
        int nextspare = 0;
-       
-       vers = md_get_version(mdfd);
-       if (vers <= 0) {
-               fprintf(stderr, Name ": %s appears not to be an md device.\n", mddev);
-               return 1;
-       }
-       if (vers < 9000) {
-               fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
-                       "    Upgrade your kernel or try --build\n");
-               return 1;
-       }
+
        if (get_linux_version() < 2004000)
                old_linux = 1;
 
-       if (ioctl(mdfd, GET_ARRAY_INFO, &info.array)>=0) {
-               fprintf(stderr, Name ": device %s already active - cannot assemble it\n",
-                       mddev);
-               return 1;
-       }
-       ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
+       if (mdfd >= 0) {
+               vers = md_get_version(mdfd);
+               if (vers <= 0) {
+                       fprintf(stderr, Name ": %s appears not to be an md device.\n", mddev);
+                       return 1;
+               }
+               if (vers < 9000) {
+                       fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
+                               "    Upgrade your kernel or try --build\n");
+                       return 1;
+               }
 
+               if (ioctl(mdfd, GET_ARRAY_INFO, &info.array)>=0) {
+                       fprintf(stderr, Name ": device %s already active - cannot assemble it\n",
+                               mddev);
+                       return 1;
+               }
+               ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
+       }
        /*
         * If any subdevs are listed, then any that don't
         * match ident are discarded.  Remainder must all match and
@@ -153,16 +175,22 @@ int Assemble(struct supertype *st, char *mddev, int mdfd,
            ident->super_minor < 0 &&
            ident->devices == NULL) {
                fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
-                       mddev);
+                       mddev ? mddev : "further assembly");
                return 1;
        }
        if (devlist == NULL)
-               devlist = conf_get_devs(conffile);
-       else inargv = 1;
+               devlist = conf_get_devs();
+       else if (mdfd >= 0)
+               inargv = 1;
+
+ try_again:
 
        tmpdev = devlist; num_devs = 0;
        while (tmpdev) {
-               num_devs++;
+               if (tmpdev->used)
+                       tmpdev->used = 2;
+               else
+                       num_devs++;
                tmpdev = tmpdev->next;
        }
        devices = malloc(num_devs * sizeof(*devices));
@@ -171,16 +199,21 @@ int Assemble(struct supertype *st, char *mddev, int mdfd,
 
        if (verbose>0)
            fprintf(stderr, Name ": looking for devices for %s\n",
-                   mddev);
+                   mddev ? mddev : "further assembly");
 
-       while ( devlist) {
-               char *devname;
+       /* first walk the list of devices to find a consistent set
+        * that match the criterea, if that is possible.
+        * We flag the one we like with 'used'.
+        */
+       for (tmpdev = devlist;
+            tmpdev;
+            tmpdev = tmpdev->next) {
+               char *devname = tmpdev->devname;
                int dfd;
                struct stat stb;
-               struct supertype *tst = st;
+               struct supertype *tst = dup_super(st);
 
-               devname = devlist->devname;
-               devlist = devlist->next;
+               if (tmpdev->used > 1) continue;
 
                if (ident->devices &&
                    !match_oneof(ident->devices, devname)) {
@@ -189,99 +222,210 @@ int Assemble(struct supertype *st, char *mddev, int mdfd,
                        continue;
                }
 
-               if (super) {
-                       free(super);
-                       super = NULL;
-               }
-               
                dfd = dev_open(devname, O_RDONLY|O_EXCL);
                if (dfd < 0) {
                        if ((inargv && verbose >= 0) || verbose > 0)
                                fprintf(stderr, Name ": cannot open device %s: %s\n",
                                        devname, strerror(errno));
+                       tmpdev->used = 2;
                } else if (fstat(dfd, &stb)< 0) {
                        /* Impossible! */
                        fprintf(stderr, Name ": fstat failed for %s: %s\n",
                                devname, strerror(errno));
+                       tmpdev->used = 2;
                } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
                        fprintf(stderr, Name ": %s is not a block device.\n",
                                devname);
+                       tmpdev->used = 2;
                } else if (!tst && (tst = guess_super(dfd)) == NULL) {
                        if ((inargv && verbose >= 0) || verbose > 0)
-                               fprintf(stderr, Name ": no recogniseable superblock\n");
-               } else if (tst->ss->load_super(tst,dfd, &super, NULL)) {
+                               fprintf(stderr, Name ": no recogniseable superblock on %s\n",
+                                       devname);
+                       tmpdev->used = 2;
+               } else if (tst->ss->load_super(tst,dfd, NULL)) {
                        if ((inargv && verbose >= 0) || verbose > 0)
                                fprintf( stderr, Name ": no RAID superblock on %s\n",
                                         devname);
                } else {
-                       tst->ss->getinfo_super(&info, super);
+                       tst->ss->getinfo_super(tst, &info);
                }
                if (dfd >= 0) close(dfd);
 
                if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
-                   (!super || same_uuid(info.uuid, ident->uuid, tst->ss->swapuuid)==0)) {
+                   (!tst || !tst->sb ||
+                    same_uuid(info.uuid, ident->uuid, tst->ss->swapuuid)==0)) {
                        if ((inargv && verbose >= 0) || verbose > 0)
                                fprintf(stderr, Name ": %s has wrong uuid.\n",
                                        devname);
-                       continue;
+                       goto loop;
                }
                if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
-                   (!super || strncmp(info.name, ident->name, 32)!=0)) {
+                   (!tst || !tst->sb ||
+                    name_matches(info.name, ident->name, homehost)==0)) {
                        if ((inargv && verbose >= 0) || verbose > 0)
                                fprintf(stderr, Name ": %s has wrong name.\n",
                                        devname);
-                       continue;
+                       goto loop;
                }
                if (ident->super_minor != UnSet &&
-                   (!super || ident->super_minor != info.array.md_minor)) {
+                   (!tst || !tst->sb ||
+                    ident->super_minor != info.array.md_minor)) {
                        if ((inargv && verbose >= 0) || verbose > 0)
                                fprintf(stderr, Name ": %s has wrong super-minor.\n",
                                        devname);
-                       continue;
+                       goto loop;
                }
                if (ident->level != UnSet &&
-                   (!super|| ident->level != info.array.level)) {
+                   (!tst || !tst->sb ||
+                    ident->level != info.array.level)) {
                        if ((inargv && verbose >= 0) || verbose > 0)
                                fprintf(stderr, Name ": %s has wrong raid level.\n",
                                        devname);
-                       continue;
+                       goto loop;
                }
                if (ident->raid_disks != UnSet &&
-                   (!super || ident->raid_disks!= info.array.raid_disks)) {
+                   (!tst || !tst->sb ||
+                    ident->raid_disks!= info.array.raid_disks)) {
                        if ((inargv && verbose >= 0) || verbose > 0)
                                fprintf(stderr, Name ": %s requires wrong number of drives.\n",
                                        devname);
-                       continue;
+                       goto loop;
                }
-
-               /* If we are this far, then we are commited to this device.
+               if (mdfd < 0) {
+                       if (tst == NULL || tst->sb == NULL)
+                               continue;
+                       if (update == NULL &&
+                           tst->ss->match_home(tst, homehost)==0) {
+                               if ((inargv && verbose >= 0) || verbose > 0)
+                                       fprintf(stderr, Name ": %s is not built for host %s.\n",
+                                               devname, homehost);
+                               /* Auto-assemble, and this is not a usable host */
+                               /* if update != NULL, we are updating the host
+                                * name... */
+                               goto loop;
+                       }
+               }
+               /* If we are this far, then we are nearly commited to this device.
                 * If the super_block doesn't exist, or doesn't match others,
-                * then we cannot continue
+                * then we probably cannot continue
+                * However if one of the arrays is for the homehost, and
+                * the other isn't that can disambiguate.
                 */
 
-               if (!super) {
+               if (!tst || !tst->sb) {
                        fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
                                devname);
-                       free(first_super);
+                       if (st)
+                               st->ss->free_super(st);
                        return 1;
                }
 
-
-               st = tst; /* commit to this format, if haven't already */
-               if (st->ss->compare_super(&first_super, super)) {
+               if (st == NULL)
+                       st = dup_super(tst);
+               if (st->minor_version == -1)
+                       st->minor_version = tst->minor_version;
+               if (st->ss != tst->ss ||
+                   st->minor_version != tst->minor_version ||
+                   st->ss->compare_super(st, tst) != 0) {
+                       /* Some mismatch. If exactly one array matches this host,
+                        * we can resolve on that one.
+                        * Or, if we are auto assembling, we just ignore the second
+                        * for now.
+                        */
+                       if (mdfd < 0)
+                               goto loop;
+                       if (homehost) {
+                               int first = st->ss->match_home(st, homehost);
+                               int last = tst->ss->match_home(tst, homehost);
+                               if (first+last == 1) {
+                                       /* We can do something */
+                                       if (first) {/* just ignore this one */
+                                               if ((inargv && verbose >= 0) || verbose > 0)
+                                                       fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
+                                                               devname);
+                                               goto loop;
+                                       } else { /* reject all those sofar */
+                                               mddev_dev_t td;
+                                               if ((inargv && verbose >= 0) || verbose > 0)
+                                                       fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
+                                                               devname);
+                                               for (td=devlist; td != tmpdev; td=td->next)
+                                                       if (td->used == 1)
+                                                               td->used = 0;
+                                               tmpdev->used = 1;
+                                               goto loop;
+                                       }
+                               }
+                       }
                        fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
                                devname);
-                       free(super);
-                       free(first_super);
+                       tst->ss->free_super(tst);
+                       st->ss->free_super(st);
                        return 1;
                }
 
+               tmpdev->used = 1;
+
+       loop:
+               tst->ss->free_super(tst);
+       }
+
+       if (mdfd < 0) {
+               /* So... it is up to me to open the device.
+                * We create a name '/dev/md/XXX' based on the info in the
+                * superblock, and call open_mddev on that
+                */
+               mdu_array_info_t inf;
+               char *c;
+               if (!st->sb) {
+                       return 2;
+               }
+               st->ss->getinfo_super(st, &info);
+               c = strchr(info.name, ':');
+               if (c) c++; else c= info.name;
+               if (isdigit(*c) && ((ident->autof & 7)==4 || (ident->autof&7)==6))
+                       /* /dev/md/d0 style for partitionable */
+                       asprintf(&mddev, "/dev/md/d%s", c);
+               else
+                       asprintf(&mddev, "/dev/md/%s", c);
+               mdfd = open_mddev(mddev, ident->autof);
+               if (mdfd < 0) {
+                       st->ss->free_super(st);
+                       free(devices);
+                       goto try_again;
+               }
+               vers = md_get_version(mdfd);
+               if (ioctl(mdfd, GET_ARRAY_INFO, &inf)==0) {
+                       for (tmpdev = devlist ;
+                            tmpdev && tmpdev->used != 1;
+                            tmpdev = tmpdev->next)
+                               ;
+                       fprintf(stderr, Name ": %s already active, cannot restart it!\n", mddev);
+                       if (tmpdev)
+                               fprintf(stderr, Name ":   %s needed for %s...\n",
+                                       mddev, tmpdev->devname);
+                       close(mdfd);
+                       mdfd = -1;
+                       st->ss->free_super(st);
+                       free(devices);
+                       goto try_again;
+               }
+               must_close = 1;
+       }
+
+       /* Ok, no bad inconsistancy, we can try updating etc */
+       bitmap_done = 0;
+       for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
+               char *devname = tmpdev->devname;
+               struct stat stb;
                /* looks like a good enough match to update the super block if needed */
+#ifndef MDASSEMBLE
                if (update) {
+                       int dfd;
                        /* prepare useful information in info structures */
                        struct stat stb2;
+                       struct supertype *tst;
                        fstat(mdfd, &stb2);
-                       info.array.md_minor = minor(stb2.st_rdev);
 
                        if (strcmp(update, "uuid")==0 &&
                            !ident->uuid_set) {
@@ -295,57 +439,88 @@ int Assemble(struct supertype *st, char *mddev, int mdfd,
                                }
                                if (rfd >= 0) close(rfd);
                        }
+                       dfd = dev_open(devname, O_RDWR|O_EXCL);
+
+                       remove_partitions(dfd);
+
+                       tst = dup_super(st);
+                       tst->ss->load_super(tst, dfd, NULL);
+                       tst->ss->getinfo_super(tst, &info);
+
                        memcpy(info.uuid, ident->uuid, 16);
                        strcpy(info.name, ident->name);
-                       st->ss->update_super(&info, super, update, devname, verbose,
-                                            ident->uuid_set, homehost);
+                       info.array.md_minor = minor(stb2.st_rdev);
+
+                       tst->ss->update_super(tst, &info, update,
+                                             devname, verbose,
+                                             ident->uuid_set, homehost);
                        if (strcmp(update, "uuid")==0 &&
                            !ident->uuid_set) {
                                ident->uuid_set = 1;
                                memcpy(ident->uuid, info.uuid, 16);
                        }
-                       dfd = dev_open(devname, O_RDWR|O_EXCL);
                        if (dfd < 0)
                                fprintf(stderr, Name ": Cannot open %s for superblock update\n",
                                        devname);
-                       else if (st->ss->store_super(st, dfd, super))
+                       else if (tst->ss->store_super(tst, dfd))
                                fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
                                        devname);
                        if (dfd >= 0)
                                close(dfd);
 
                        if (strcmp(update, "uuid")==0 &&
-                           ident->bitmap_fd)
-                               bitmap_update_uuid(ident->bitmap_fd, info.uuid);
+                           ident->bitmap_fd >= 0 && !bitmap_done) {
+                               if (bitmap_update_uuid(ident->bitmap_fd,
+                                                      info.uuid,
+                                                      tst->ss->swapuuid) != 0)
+                                       fprintf(stderr, Name ": Could not update uuid on external bitmap.\n");
+                               else
+                                       bitmap_done = 1;
+                       }
+                       tst->ss->free_super(tst);
+               } else
+#endif
+               {
+                       struct supertype *tst = dup_super(st);;
+                       int dfd;
+                       dfd = dev_open(devname, O_RDWR|O_EXCL);
+
+                       remove_partitions(dfd);
+
+                       tst->ss->load_super(tst, dfd, NULL);
+                       tst->ss->getinfo_super(tst, &info);
+                       tst->ss->free_super(tst);
+                       close(dfd);
                }
 
+               stat(devname, &stb);
+
                if (verbose > 0)
                        fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
                                devname, mddev, info.disk.raid_disk);
                devices[devcnt].devname = devname;
-               devices[devcnt].major = major(stb.st_rdev);
-               devices[devcnt].minor = minor(stb.st_rdev);
-               devices[devcnt].oldmajor = info.disk.major;
-               devices[devcnt].oldminor = info.disk.minor;
-               devices[devcnt].events = info.events;
-               devices[devcnt].raid_disk = info.disk.raid_disk;
-               devices[devcnt].disk_nr = info.disk.number;
                devices[devcnt].uptodate = 0;
-               devices[devcnt].state = info.disk.state;
+               devices[devcnt].i = info;
+               devices[devcnt].i.disk.major = major(stb.st_rdev);
+               devices[devcnt].i.disk.minor = minor(stb.st_rdev);
                if (most_recent < devcnt) {
-                       if (devices[devcnt].events
-                           > devices[most_recent].events)
+                       if (devices[devcnt].i.events
+                           > devices[most_recent].i.events)
                                most_recent = devcnt;
                }
                if (info.array.level == -4)
                        /* with multipath, the raid_disk from the superblock is meaningless */
                        i = devcnt;
                else
-                       i = devices[devcnt].raid_disk;
+                       i = devices[devcnt].i.disk.raid_disk;
                if (i+1 == 0) {
                        if (nextspare < info.array.raid_disks)
                                nextspare = info.array.raid_disks;
                        i = nextspare++;
+               } else {
+                       if (i >= info.array.raid_disks &&
+                           i >= nextspare)
+                               nextspare = i+1;
                }
                if (i < 10000) {
                        if (i >= bestcnt) {
@@ -361,28 +536,51 @@ int Assemble(struct supertype *st, char *mddev, int mdfd,
                                best = newbest;
                                bestcnt = newbestcnt;
                        }
+                       if (best[i] >=0 &&
+                           devices[best[i]].i.events
+                           == devices[devcnt].i.events
+                           && (devices[best[i]].i.disk.minor
+                               != devices[devcnt].i.disk.minor)
+                           && st->ss->major == 0
+                           && info.array.level != -4) {
+                               /* two different devices with identical superblock.
+                                * Could be a mis-detection caused by overlapping
+                                * partitions.  fail-safe.
+                                */
+                               fprintf(stderr, Name ": 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"
+                                       );
+                               if (must_close) close(mdfd);
+                               return 1;
+                       }
                        if (best[i] == -1
-                           || devices[best[i]].events < devices[devcnt].events)
+                           || (devices[best[i]].i.events
+                               < devices[devcnt].i.events))
                                best[i] = devcnt;
                }
                devcnt++;
        }
 
-       if (super)
-               free(super);
-       super = NULL;
-
        if (update && strcmp(update, "byteorder")==0)
                st->minor_version = 90;
 
        if (devcnt == 0) {
                fprintf(stderr, Name ": no devices found for %s\n",
                        mddev);
-               free(first_super);
+               st->ss->free_super(st);
+               if (must_close) close(mdfd);
                return 1;
        }
 
-       st->ss->getinfo_super(&info, first_super);
+       st->ss->getinfo_super(st, &info);
+       clean = info.array.state & 1;
 
        /* now we have some devices that might be suitable.
         * I wonder how many
@@ -401,13 +599,14 @@ int Assemble(struct supertype *st, char *mddev, int mdfd,
                 * as they don't make sense
                 */
                if (info.array.level != -4)
-                       if (!(devices[j].state & (1<<MD_DISK_SYNC))) {
-                               if (!(devices[j].state & (1<<MD_DISK_FAULTY)))
+                       if (!(devices[j].i.disk.state & (1<<MD_DISK_SYNC))) {
+                               if (!(devices[j].i.disk.state
+                                     & (1<<MD_DISK_FAULTY)))
                                        sparecnt++;
                                continue;
                        }
-               if (devices[j].events+event_margin >=
-                   devices[most_recent].events) {
+               if (devices[j].i.events+event_margin >=
+                   devices[most_recent].i.events) {
                        devices[j].uptodate = 1;
                        if (i < info.array.raid_disks) {
                                okcnt++;
@@ -417,61 +616,84 @@ int Assemble(struct supertype *st, char *mddev, int mdfd,
                }
        }
        while (force && !enough(info.array.level, info.array.raid_disks,
-                               info.array.layout,
+                               info.array.layout, 1,
                                avail, okcnt)) {
                /* Choose the newest best drive which is
                 * not up-to-date, update the superblock
                 * and add it.
                 */
                int fd;
+               struct supertype *tst;
+               long long current_events;
                chosen_drive = -1;
                for (i=0; i<info.array.raid_disks && i < bestcnt; i++) {
                        int j = best[i];
                        if (j>=0 &&
                            !devices[j].uptodate &&
-                           devices[j].events > 0 &&
+                           devices[j].i.events > 0 &&
                            (chosen_drive < 0 ||
-                            devices[j].events > devices[chosen_drive].events))
+                            devices[j].i.events
+                            > devices[chosen_drive].i.events))
                                chosen_drive = j;
                }
                if (chosen_drive < 0)
                        break;
+               current_events = devices[chosen_drive].i.events;
+       add_another:
                if (verbose >= 0)
                        fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
-                               devices[chosen_drive].devname, devices[chosen_drive].raid_disk,
-                               (int)(devices[chosen_drive].events),
-                               (int)(devices[most_recent].events));
+                               devices[chosen_drive].devname,
+                               devices[chosen_drive].i.disk.raid_disk,
+                               (int)(devices[chosen_drive].i.events),
+                               (int)(devices[most_recent].i.events));
                fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
                if (fd < 0) {
                        fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
                                devices[chosen_drive].devname);
-                       devices[chosen_drive].events = 0;
+                       devices[chosen_drive].i.events = 0;
                        continue;
                }
-               if (st->ss->load_super(st,fd, &super, NULL)) {
+               tst = dup_super(st);
+               if (tst->ss->load_super(st,fd, NULL)) {
                        close(fd);
                        fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
                                devices[chosen_drive].devname);
-                       devices[chosen_drive].events = 0;
+                       devices[chosen_drive].i.events = 0;
                        continue;
                }
-               info.events = devices[most_recent].events;
-               st->ss->update_super(&info, super, "force", devices[chosen_drive].devname, verbose, 0, NULL);
+               info.events = devices[most_recent].i.events;
+               tst->ss->update_super(tst, &info, "force-one",
+                                    devices[chosen_drive].devname, verbose,
+                                    0, NULL);
 
-               if (st->ss->store_super(st, fd, super)) {
+               if (tst->ss->store_super(tst, fd)) {
                        close(fd);
                        fprintf(stderr, Name ": Could not re-write superblock on %s\n",
                                devices[chosen_drive].devname);
-                       devices[chosen_drive].events = 0;
-                       free(super);
+                       devices[chosen_drive].i.events = 0;
+                       tst->ss->free_super(tst);
                        continue;
                }
                close(fd);
-               devices[chosen_drive].events = devices[most_recent].events;
+               devices[chosen_drive].i.events = devices[most_recent].i.events;
                devices[chosen_drive].uptodate = 1;
                avail[chosen_drive] = 1;
                okcnt++;
-               free(super);
+               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
+                */
+               for (i=0; i<info.array.raid_disks && i < bestcnt ; i++) {
+                       int j = best[i];
+                       if (j >= 0 &&
+                           !devices[j].uptodate &&
+                           devices[j].i.events > 0 &&
+                           devices[j].i.events == current_events) {
+                               chosen_drive = j;
+                               goto add_another;
+                       }
+               }
        }
 
        /* Now we want to look at the superblock which the kernel will base things on
@@ -481,7 +703,7 @@ int Assemble(struct supertype *st, char *mddev, int mdfd,
         * superblock.
         */
        chosen_drive = -1;
-       super = NULL;
+       st->ss->free_super(st);
        for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
                int j = best[i];
                int fd;
@@ -494,21 +716,24 @@ int Assemble(struct supertype *st, char *mddev, int mdfd,
                if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
                        fprintf(stderr, Name ": Cannot open %s: %s\n",
                                devices[j].devname, strerror(errno));
+                       if (must_close) close(mdfd);
                        return 1;
                }
-               if (st->ss->load_super(st,fd, &super, NULL)) {
+               if (st->ss->load_super(st,fd, NULL)) {
                        close(fd);
                        fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
                                devices[j].devname);
+                       if (must_close) close(mdfd);
                        return 1;
                }
                close(fd);
        }
-       if (super == NULL) {
+       if (st->sb == NULL) {
                fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
+               if (must_close) close(mdfd);
                return 1;
        }
-       st->ss->getinfo_super(&info, super);
+       st->ss->getinfo_super(st, &info);
        for (i=0; i<bestcnt; i++) {
                int j = best[i];
                unsigned int desired_state;
@@ -522,12 +747,11 @@ int Assemble(struct supertype *st, char *mddev, int mdfd,
                        continue;
                if (!devices[j].uptodate)
                        continue;
-               info.disk.number = devices[j].disk_nr;
-               info.disk.raid_disk = i;
-               info.disk.state = desired_state;
 
-               if (devices[j].uptodate &&
-                   st->ss->update_super(&info, super, "assemble", NULL, verbose, 0, NULL)) {
+               devices[j].i.disk.state = desired_state;
+
+               if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
+                                        verbose, 0, NULL)) {
                        if (force) {
                                if (verbose >= 0)
                                        fprintf(stderr, Name ": "
@@ -542,17 +766,20 @@ int Assemble(struct supertype *st, char *mddev, int mdfd,
                        }
                }
 #if 0
-               if (!devices[j].uptodate &&
-                   !(super.disks[i].state & (1 << MD_DISK_FAULTY))) {
+               if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
                        fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
                                i, mddev);
                }
 #endif
        }
-       if (force && okcnt == info.array.raid_disks-1) {
-               /* FIXME check event count */
-               change += st->ss->update_super(&info, super, "force",
-                                       devices[chosen_drive].devname, verbose, 0, NULL);
+       if (force && !clean &&
+           !enough(info.array.level, info.array.raid_disks,
+                   info.array.layout, clean,
+                   avail, okcnt)) {
+               change += st->ss->update_super(st, &info, "force-array",
+                                       devices[chosen_drive].devname, verbose,
+                                              0, NULL);
+               clean = 1;
        }
 
        if (change) {
@@ -561,12 +788,14 @@ int Assemble(struct supertype *st, char *mddev, int mdfd,
                if (fd < 0) {
                        fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
                                devices[chosen_drive].devname);
+                       if (must_close) close(mdfd);
                        return 1;
                }
-               if (st->ss->store_super(st, fd, super)) {
+               if (st->ss->store_super(st, fd)) {
                        close(fd);
                        fprintf(stderr, Name ": Could not re-write superblock on %s\n",
                                devices[chosen_drive].devname);
+                       if (must_close) close(mdfd);
                        return 1;
                }
                close(fd);
@@ -601,6 +830,7 @@ int Assemble(struct supertype *st, char *mddev, int mdfd,
                }
                if (err) {
                        fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
+                       if (must_close) close(mdfd);
                        return err;
                }
        }
@@ -613,6 +843,24 @@ int Assemble(struct supertype *st, char *mddev, int mdfd,
        /* Almost ready to actually *do* something */
        if (!old_linux) {
                int rv;
+
+#ifndef MDASSEMBLE
+               struct mdinfo *sra;
+               if (st->ss->external) {
+                       char ver[100];
+                       strcat(strcpy(ver, "external:"), st->ss->text_version);
+                       sra = sysfs_read(mdfd, 0, 0);
+                       if ((vers % 100) < 2 ||
+                           sra == NULL ||
+                           sysfs_set_str(sra, NULL, "metadata_version",
+                                         ver) < 0) {
+                               fprintf(stderr, Name ": This kernel does not "
+                                       "support external metadata.\n");
+                               return 1;
+                       }
+                       rv = sysfs_set_array(sra, &info);
+               } else
+#endif
                if ((vers % 100) >= 1) { /* can use different versions */
                        mdu_array_info_t inf;
                        memset(&inf, 0, sizeof(inf));
@@ -625,11 +873,13 @@ int Assemble(struct supertype *st, char *mddev, int mdfd,
                if (rv) {
                        fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
                                mddev, strerror(errno));
+                       if (must_close) close(mdfd);
                        return 1;
                }
                if (ident->bitmap_fd >= 0) {
                        if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
                                fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
+                               if (must_close) close(mdfd);
                                return 1;
                        }
                } else if (ident->bitmap_file) {
@@ -638,11 +888,13 @@ int Assemble(struct supertype *st, char *mddev, int mdfd,
                        if (bmfd < 0) {
                                fprintf(stderr, Name ": Could not open bitmap file %s\n",
                                        ident->bitmap_file);
+                               if (must_close) close(mdfd);
                                return 1;
                        }
                        if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
                                fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
                                close(bmfd);
+                               if (must_close) close(mdfd);
                                return 1;
                        }
                        close(bmfd);
@@ -659,30 +911,66 @@ int Assemble(struct supertype *st, char *mddev, int mdfd,
                                j = chosen_drive;
 
                        if (j >= 0 /* && devices[j].uptodate */) {
-                               mdu_disk_info_t disk;
-                               memset(&disk, 0, sizeof(disk));
-                               disk.major = devices[j].major;
-                               disk.minor = devices[j].minor;
-                               if (ioctl(mdfd, ADD_NEW_DISK, &disk)!=0) {
-                                       fprintf(stderr, Name ": failed to add %s to %s: %s\n",
+#ifndef MDASSEMBLE
+                               if (st->ss->external) {
+                                       int fd = dev_open(devices[j].devname,
+                                                         O_RDONLY);
+                                       if (fd < 0)
+                                               rv = 1;
+                                       else {
+                                               devices[j].i.disk.number =
+                                                       devices[j].i.disk.raid_disk;
+                                               st->ss->getinfo_super_n(st,
+                                                              &devices[j].i);
+                                               rv = sysfs_add_disk(sra, fd,
+                                                             &devices[j].i);
+                                               close(fd);
+                                       }
+                               } else
+#endif
+                                       rv = ioctl(mdfd, ADD_NEW_DISK,
+                                         &devices[j].i.disk);
+                               if (rv) {
+                                       fprintf(stderr, Name ": failed to add "
+                                                       "%s to %s: %s\n",
                                                devices[j].devname,
                                                mddev,
                                                strerror(errno));
-                                       if (i < info.array.raid_disks || i == bestcnt)
+                                       if (i < info.array.raid_disks
+                                           || i == bestcnt)
                                                okcnt--;
                                        else
                                                sparecnt--;
                                } else if (verbose > 0)
-                                       fprintf(stderr, Name ": added %s to %s as %d\n",
-                                               devices[j].devname, mddev, devices[j].raid_disk);
+                                       fprintf(stderr, Name ": added %s "
+                                                       "to %s as %d\n",
+                                               devices[j].devname, mddev,
+                                               devices[j].i.disk.raid_disk);
                        } else if (verbose > 0 && i < info.array.raid_disks)
-                               fprintf(stderr, Name ": no uptodate device for slot %d of %s\n",
+                               fprintf(stderr, Name ": no uptodate device for "
+                                               "slot %d of %s\n",
                                        i, mddev);
                }
-               
+
+               if (info.array.level == LEVEL_CONTAINER) {
+                       if (verbose >= 0) {
+                               fprintf(stderr, Name ": Container %s has been "
+                                       "assembled with %d drive%s",
+                                       mddev, okcnt, okcnt==1?"":"s");
+                               if (okcnt < info.array.raid_disks)
+                                       fprintf(stderr, " (out of %d)",
+                                               info.array.raid_disks);
+                               fprintf(stderr, "\n");
+                       }
+                       if (must_close)
+                               close(mdfd);
+                       return 0;
+               }
+
                if (runstop == 1 ||
                    (runstop <= 0 &&
-                    ( enough(info.array.level, info.array.raid_disks, info.array.layout, avail, okcnt) &&
+                    ( enough(info.array.level, info.array.raid_disks,
+                             info.array.layout, clean, avail, okcnt) &&
                       (okcnt >= req_cnt || start_partial_ok)
                             ))) {
                        if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
@@ -695,10 +983,51 @@ int Assemble(struct supertype *st, char *mddev, int mdfd,
                                                fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
                                        fprintf(stderr, ".\n");
                                }
+                               if (must_close) {
+                                       int usecs = 1;
+                                       close(mdfd);
+                                       /* There is a nasty race with 'mdadm --monitor'.
+                                        * If it opens this device before we close it,
+                                        * it gets an incomplete open on which IO
+                                        * doesn't work and the capacity is
+                                        * wrong.
+                                        * If we reopen (to check for layered devices)
+                                        * before --monitor closes, we loose.
+                                        *
+                                        * So: wait upto 1 second for there to be
+                                        * a non-zero capacity.
+                                        */
+                                       while (usecs < 1000) {
+                                               mdfd = open(mddev, O_RDONLY);
+                                               if (mdfd >= 0) {
+                                                       unsigned long long size;
+                                                       if (get_dev_size(mdfd, NULL, &size) &&
+                                                           size > 0)
+                                                               break;
+                                                       close(mdfd);
+                                               }
+                                               usleep(usecs);
+                                               usecs <<= 1;
+                                       }
+                               }
                                return 0;
                        }
                        fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
                                mddev, strerror(errno));
+
+                       if (!enough(info.array.level, info.array.raid_disks,
+                                   info.array.layout, 1, avail, okcnt))
+                               fprintf(stderr, Name ": Not enough devices to "
+                                       "start the array.\n");
+                       else if (!enough(info.array.level,
+                                        info.array.raid_disks,
+                                        info.array.layout, clean,
+                                        avail, okcnt))
+                               fprintf(stderr, Name ": Not enough devices to "
+                                       "start the array while not clean "
+                                       "- consider --force.\n");
+
+                       if (must_close) close(mdfd);
                        return 1;
                }
                if (runstop == -1) {
@@ -707,14 +1036,23 @@ int Assemble(struct supertype *st, char *mddev, int mdfd,
                        if (okcnt != info.array.raid_disks)
                                fprintf(stderr, " (out of %d)", info.array.raid_disks);
                        fprintf(stderr, ", but not started.\n");
+                       if (must_close) close(mdfd);
                        return 0;
                }
-               if (verbose >= 0) {
+               if (verbose >= -1) {
                        fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
                        if (sparecnt)
                                fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
-                       if (!enough(info.array.level, info.array.raid_disks, info.array.layout, avail, okcnt))
+                       if (!enough(info.array.level, info.array.raid_disks,
+                                   info.array.layout, 1, avail, okcnt))
                                fprintf(stderr, " - not enough to start the array.\n");
+                       else if (!enough(info.array.level,
+                                        info.array.raid_disks,
+                                        info.array.layout, clean,
+                                        avail, okcnt))
+                               fprintf(stderr, " - not enough to start the "
+                                       "array while not clean - consider "
+                                       "--force.\n");
                        else {
                                if (req_cnt == info.array.raid_disks)
                                        fprintf(stderr, " - need all %d to start it", req_cnt);
@@ -723,6 +1061,7 @@ int Assemble(struct supertype *st, char *mddev, int mdfd,
                                fprintf(stderr, " (use --run to insist).\n");
                        }
                }
+               if (must_close) close(mdfd);
                return 1;
        } else {
                /* The "chosen_drive" is a good choice, and if necessary, the superblock has
@@ -730,13 +1069,14 @@ int Assemble(struct supertype *st, char *mddev, int mdfd,
                 * so we can just start the array
                 */
                unsigned long dev;
-               dev = makedev(devices[chosen_drive].major,
-                           devices[chosen_drive].minor);
+               dev = makedev(devices[chosen_drive].i.disk.major,
+                           devices[chosen_drive].i.disk.minor);
                if (ioctl(mdfd, START_ARRAY, dev)) {
                    fprintf(stderr, Name ": Cannot start array: %s\n",
                            strerror(errno));
                }
-               
+
        }
+       if (must_close) close(mdfd);
        return 0;
 }