]> git.ipfire.org Git - thirdparty/mdadm.git/blobdiff - super-intel.c
imsm: provide a detail_platform method
[thirdparty/mdadm.git] / super-intel.c
index 433fa464ce9e1fbf1fb99f23a6e12f57ae52b7fb..8e00d742748a74b95a8e21edfc9eb6b4839eda63 100644 (file)
 #include "mdadm.h"
 #include "mdmon.h"
 #include "sha1.h"
+#include "platform-intel.h"
 #include <values.h>
 #include <scsi/sg.h>
 #include <ctype.h>
+#include <dirent.h>
 
 /* MPB == Metadata Parameter Block */
 #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. "
@@ -213,6 +215,7 @@ struct intel_super {
        int updates_pending; /* count of pending updates for mdmon */
        int creating_imsm; /* flag to indicate container creation */
        int current_vol; /* index of raid device undergoing creation */
+       __u32 create_offset; /* common start for 'current_vol' */
        #define IMSM_MAX_RAID_DEVS 2
        struct imsm_dev *dev_tbl[IMSM_MAX_RAID_DEVS];
        struct dl {
@@ -223,10 +226,14 @@ struct intel_super {
                char *devname;
                struct imsm_disk disk;
                int fd;
+               int extent_cnt;
+               struct extent *e; /* for determining freespace @ create */
        } *disks;
        struct dl *add; /* list of disks to add while mdmon active */
        struct dl *missing; /* disks removed while we weren't looking */
        struct bbm_log *bbm_log;
+       const char *hba; /* device path of the raid controller for this metadata */
+       const struct imsm_orom *orom; /* platform firmware support */
 };
 
 struct extent {
@@ -248,6 +255,10 @@ struct imsm_update_activate_spare {
        struct imsm_update_activate_spare *next;
 };
 
+struct disk_info {
+       __u8 serial[MAX_RAID_SERIAL_LEN];
+};
+
 struct imsm_update_create_array {
        enum imsm_update_type type;
        int dev_idx;
@@ -361,6 +372,20 @@ static size_t sizeof_imsm_dev(struct imsm_dev *dev, int migr_state)
        return size;
 }
 
+#ifndef MDASSEMBLE
+/* retrieve disk serial number list from a metadata update */
+static struct disk_info *get_disk_info(struct imsm_update_create_array *update)
+{
+       void *u = update;
+       struct disk_info *inf;
+
+       inf = u + sizeof(*update) - sizeof(struct imsm_dev) +
+             sizeof_imsm_dev(&update->dev, 0);
+
+       return inf;
+}
+#endif
+
 static struct imsm_dev *__get_imsm_dev(struct imsm_super *mpb, __u8 index)
 {
        int offset;
@@ -438,13 +463,10 @@ static int cmp_extent(const void *av, const void *bv)
        return 0;
 }
 
-static struct extent *get_extents(struct intel_super *super, struct dl *dl)
+static int count_memberships(struct dl *dl, struct intel_super *super)
 {
-       /* find a list of used extents on the given physical device */
-       struct extent *rv, *e;
-       int i, j;
        int memberships = 0;
-       __u32 reservation = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
+       int i, j;
 
        for (i = 0; i < super->anchor->num_raid_devs; i++) {
                struct imsm_dev *dev = get_imsm_dev(super, i);
@@ -457,6 +479,18 @@ static struct extent *get_extents(struct intel_super *super, struct dl *dl)
                                memberships++;
                }
        }
+
+       return memberships;
+}
+
+static struct extent *get_extents(struct intel_super *super, struct dl *dl)
+{
+       /* find a list of used extents on the given physical device */
+       struct extent *rv, *e;
+       int i, j;
+       int memberships = count_memberships(dl, super);
+       __u32 reservation = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
+
        rv = malloc(sizeof(struct extent) * (memberships + 1));
        if (!rv)
                return NULL;
@@ -489,6 +523,13 @@ static struct extent *get_extents(struct intel_super *super, struct dl *dl)
 
                remainder = __le32_to_cpu(dl->disk.total_blocks) - 
                            (last->start + last->size);
+               /* round down to 1k block to satisfy precision of the kernel
+                * 'size' interface
+                */
+               remainder &= ~1UL;
+               /* make sure remainder is still sane */
+               if (remainder < ROUND_UP(super->len, 512) >> 9)
+                       remainder = ROUND_UP(super->len, 512) >> 9;
                if (reservation > remainder)
                        reservation = remainder;
        }
@@ -704,6 +745,252 @@ static void brief_detail_super_imsm(struct supertype *st)
        fname_from_uuid(st, &info, nbuf,'-');
        printf(" UUID=%s", nbuf + 5);
 }
+
+static int imsm_read_serial(int fd, char *devname, __u8 *serial);
+static void fd2devname(int fd, char *name);
+
+static int imsm_enumerate_ports(const char *hba_path, int port_count, int host_base, int verbose)
+{
+       /* dump an unsorted list of devices attached to ahci, as well as
+        * non-connected ports
+        */
+       int hba_len = strlen(hba_path) + 1;
+       struct dirent *ent;
+       DIR *dir;
+       char *path = NULL;
+       int err = 0;
+       unsigned long port_mask = (1 << port_count) - 1;
+
+       if (port_count > sizeof(port_mask) * 8) {
+               if (verbose)
+                       fprintf(stderr, Name ": port_count %d out of range\n", port_count);
+               return 2;
+       }
+
+       /* scroll through /sys/dev/block looking for devices attached to
+        * this hba
+        */
+       dir = opendir("/sys/dev/block");
+       for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) {
+               int fd;
+               char model[64];
+               char vendor[64];
+               char buf[1024];
+               int major, minor;
+               char *device;
+               char *c;
+               int port;
+               int type;
+
+               if (sscanf(ent->d_name, "%d:%d", &major, &minor) != 2)
+                       continue;
+               path = devt_to_devpath(makedev(major, minor));
+               if (!path)
+                       continue;
+               if (!path_attached_to_hba(path, hba_path)) {
+                       free(path);
+                       path = NULL;
+                       continue;
+               }
+
+               /* retrieve the scsi device type */
+               if (asprintf(&device, "/sys/dev/block/%d:%d/device/xxxxxxx", major, minor) < 0) {
+                       if (verbose)
+                               fprintf(stderr, Name ": failed to allocate 'device'\n");
+                       err = 2;
+                       break;
+               }
+               sprintf(device, "/sys/dev/block/%d:%d/device/type", major, minor);
+               if (load_sys(device, buf) != 0) {
+                       if (verbose)
+                               fprintf(stderr, Name ": failed to read device type for %s\n",
+                                       path);
+                       err = 2;
+                       free(device);
+                       break;
+               }
+               type = strtoul(buf, NULL, 10);
+
+               /* if it's not a disk print the vendor and model */
+               if (!(type == 0 || type == 7 || type == 14)) {
+                       vendor[0] = '\0';
+                       model[0] = '\0';
+                       sprintf(device, "/sys/dev/block/%d:%d/device/vendor", major, minor);
+                       if (load_sys(device, buf) == 0) {
+                               strncpy(vendor, buf, sizeof(vendor));
+                               vendor[sizeof(vendor) - 1] = '\0';
+                               c = (char *) &vendor[sizeof(vendor) - 1];
+                               while (isspace(*c) || *c == '\0')
+                                       *c-- = '\0';
+
+                       }
+                       sprintf(device, "/sys/dev/block/%d:%d/device/model", major, minor);
+                       if (load_sys(device, buf) == 0) {
+                               strncpy(model, buf, sizeof(model));
+                               model[sizeof(model) - 1] = '\0';
+                               c = (char *) &model[sizeof(model) - 1];
+                               while (isspace(*c) || *c == '\0')
+                                       *c-- = '\0';
+                       }
+
+                       if (vendor[0] && model[0])
+                               sprintf(buf, "%.64s %.64s", vendor, model);
+                       else
+                               switch (type) { /* numbers from hald/linux/device.c */
+                               case 1: sprintf(buf, "tape"); break;
+                               case 2: sprintf(buf, "printer"); break;
+                               case 3: sprintf(buf, "processor"); break;
+                               case 4:
+                               case 5: sprintf(buf, "cdrom"); break;
+                               case 6: sprintf(buf, "scanner"); break;
+                               case 8: sprintf(buf, "media_changer"); break;
+                               case 9: sprintf(buf, "comm"); break;
+                               case 12: sprintf(buf, "raid"); break;
+                               default: sprintf(buf, "unknown");
+                               }
+               } else
+                       buf[0] = '\0';
+               free(device);
+
+               /* chop device path to 'host%d' and calculate the port number */
+               c = strchr(&path[hba_len], '/');
+               *c = '\0';
+               if (sscanf(&path[hba_len], "host%d", &port) == 1)
+                       port -= host_base;
+               else {
+                       if (verbose) {
+                               *c = '/'; /* repair the full string */
+                               fprintf(stderr, Name ": failed to determine port number for %s\n",
+                                       path);
+                       }
+                       err = 2;
+                       break;
+               }
+
+               /* mark this port as used */
+               port_mask &= ~(1 << port);
+
+               /* print out the device information */
+               if (buf[0]) {
+                       printf("          Port%d : - non-disk device (%s) -\n", port, buf);
+                       continue;
+               }
+
+               fd = dev_open(ent->d_name, O_RDONLY);
+               if (fd < 0)
+                       printf("          Port%d : - disk info unavailable -\n", port);
+               else {
+                       fd2devname(fd, buf);
+                       printf("          Port%d : %s", port, buf);
+                       if (imsm_read_serial(fd, NULL, (__u8 *) buf) == 0)
+                               printf(" (%s)\n", buf);
+                       else
+                               printf("()\n");
+               }
+               close(fd);
+               free(path);
+               path = NULL;
+       }
+       if (path)
+               free(path);
+       if (dir)
+               closedir(dir);
+       if (err == 0) {
+               int i;
+
+               for (i = 0; i < port_count; i++)
+                       if (port_mask & (1 << i))
+                               printf("          Port%d : - no device attached -\n", i);
+       }
+
+       return err;
+}
+
+static int detail_platform_imsm(int verbose)
+{
+       /* There are two components to imsm platform support, the ahci SATA
+        * controller and the option-rom.  To find the SATA controller we
+        * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
+        * controller with the Intel vendor id is present.  This approach
+        * allows mdadm to leverage the kernel's ahci detection logic, with the
+        * caveat that if ahci.ko is not loaded mdadm will not be able to
+        * detect platform raid capabilities.  The option-rom resides in a
+        * platform "Adapter ROM".  We scan for its signature to retrieve the
+        * platform capabilities.  If raid support is disabled in the BIOS the
+        * option-rom capability structure will not be available.
+        */
+       const struct imsm_orom *orom;
+       struct sys_dev *list, *hba;
+       DIR *dir;
+       struct dirent *ent;
+       const char *hba_path;
+       int host_base = 0;
+       int port_count = 0;
+
+       list = find_driver_devices("pci", "ahci");
+       for (hba = list; hba; hba = hba->next)
+               if (devpath_to_vendor(hba->path) == 0x8086)
+                       break;
+
+       if (!hba) {
+               if (verbose)
+                       fprintf(stderr, Name ": unable to find active ahci controller\n");
+               free_sys_dev(&list);
+               return 2;
+       } else if (verbose)
+               fprintf(stderr, Name ": found Intel SATA AHCI Controller\n");
+       hba_path = hba->path;
+       hba->path = NULL;
+       free_sys_dev(&list);
+
+       orom = find_imsm_orom();
+       if (!orom) {
+               if (verbose)
+                       fprintf(stderr, Name ": imsm option-rom not found\n");
+               return 2;
+       }
+
+       printf("       Platform : Intel(R) Matrix Storage Manager\n");
+       printf("        Version : %d.%d.%d.%d\n", orom->major_ver, orom->minor_ver,
+              orom->hotfix_ver, orom->build);
+       printf("    RAID Levels :%s%s%s%s%s\n",
+              imsm_orom_has_raid0(orom) ? " raid0" : "",
+              imsm_orom_has_raid1(orom) ? " raid1" : "",
+              imsm_orom_has_raid1e(orom) ? " raid1e" : "",
+              imsm_orom_has_raid10(orom) ? " raid10" : "",
+              imsm_orom_has_raid5(orom) ? " raid5" : "");
+       printf("      Max Disks : %d\n", orom->tds);
+       printf("    Max Volumes : %d\n", orom->vpa);
+       printf(" I/O Controller : %s\n", hba_path);
+
+       /* find the smallest scsi host number to determine a port number base */
+       dir = opendir(hba_path);
+       for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) {
+               int host;
+
+               if (sscanf(ent->d_name, "host%d", &host) != 1)
+                       continue;
+               if (port_count == 0)
+                       host_base = host;
+               else if (host < host_base)
+                       host_base = host;
+
+               if (host + 1 > port_count + host_base)
+                       port_count = host + 1 - host_base;
+
+       }
+       if (dir)
+               closedir(dir);
+
+       if (!port_count || imsm_enumerate_ports(hba_path, port_count,
+                                               host_base, verbose) != 0) {
+               if (verbose)
+                       fprintf(stderr, Name ": failed to enumerate ports\n");
+               return 2;
+       }
+
+       return 0;
+}
 #endif
 
 static int match_home_imsm(struct supertype *st, char *homehost)
@@ -1122,6 +1409,17 @@ static void serialcpy(__u8 *dest, __u8 *src)
        strncpy((char *) dest, (char *) src, MAX_RAID_SERIAL_LEN);
 }
 
+static struct dl *serial_to_dl(__u8 *serial, struct intel_super *super)
+{
+       struct dl *dl;
+
+       for (dl = super->disks; dl; dl = dl->next)
+               if (serialcmp(dl->serial, serial) == 0)
+                       break;
+
+       return dl;
+}
+
 static int
 load_imsm_disk(int fd, struct intel_super *super, char *devname, int keep_fd)
 {
@@ -1141,10 +1439,7 @@ load_imsm_disk(int fd, struct intel_super *super, char *devname, int keep_fd)
         * super->disks while the current anchor believes it is a raid member,
         * check if we need to update dl->index
         */
-       for (dl = super->disks; dl; dl = dl->next)
-               if (serialcmp(dl->serial, serial) == 0)
-                       break;
-
+       dl = serial_to_dl(serial, super);
        if (!dl)
                dl = malloc(sizeof(*dl));
        else
@@ -1167,6 +1462,7 @@ load_imsm_disk(int fd, struct intel_super *super, char *devname, int keep_fd)
                dl->devname = devname ? strdup(devname) : NULL;
                serialcpy(dl->serial, serial);
                dl->index = -2;
+               dl->e = NULL;
        } else if (keep_fd) {
                close(dl->fd);
                dl->fd = fd;
@@ -1421,6 +1717,8 @@ static void __free_imsm_disk(struct dl *d)
                close(d->fd);
        if (d->devname)
                free(d->devname);
+       if (d->e)
+               free(d->e);
        free(d);
 
 }
@@ -1457,6 +1755,10 @@ static void __free_imsm(struct intel_super *super, int free_disks)
                        free(super->dev_tbl[i]);
                        super->dev_tbl[i] = NULL;
                }
+       if (super->hba) {
+               free((void *) super->hba);
+               super->hba = NULL;
+       }
 }
 
 static void free_imsm(struct intel_super *super)
@@ -1484,6 +1786,23 @@ static struct intel_super *alloc_super(int creating_imsm)
                memset(super, 0, sizeof(*super));
                super->creating_imsm = creating_imsm;
                super->current_vol = -1;
+               super->create_offset = ~((__u32 ) 0);
+               if (!check_env("IMSM_NO_PLATFORM"))
+                       super->orom = find_imsm_orom();
+               if (super->orom) {
+                       struct sys_dev *list, *ent;
+
+                       /* find the first intel ahci controller */
+                       list = find_driver_devices("pci", "ahci");
+                       for (ent = list; ent; ent = ent->next)
+                               if (devpath_to_vendor(ent->path) == 0x8086)
+                                       break;
+                       if (ent) {
+                               super->hba = ent->path;
+                               ent->path = NULL;
+                       }
+                       free_sys_dev(&list);
+               }
        }
 
        return super;
@@ -1503,9 +1822,7 @@ static int find_missing(struct intel_super *super)
 
        for (i = 0; i < mpb->num_disks; i++) {
                disk = __get_imsm_disk(mpb, i);
-               for (dl = super->disks; dl; dl = dl->next)
-                       if (serialcmp(dl->disk.serial, disk->serial) == 0)
-                               break;
+               dl = serial_to_dl(disk->serial, super);
                if (dl)
                        continue;
                /* ok we have a 'disk' without a live entry in
@@ -1768,12 +2085,11 @@ static int init_super_imsm_volume(struct supertype *st, mdu_array_info_t *info,
        int idx = mpb->num_raid_devs;
        int i;
        unsigned long long array_blocks;
-       __u32 offset = 0;
        size_t size_old, size_new;
 
-       if (mpb->num_raid_devs >= 2) {
+       if (super->orom && mpb->num_raid_devs >= super->orom->vpa) {
                fprintf(stderr, Name": This imsm-container already has the "
-                       "maximum of 2 volumes\n");
+                       "maximum of %d volumes\n", super->orom->vpa);
                return 0;
        }
 
@@ -1809,9 +2125,12 @@ static int init_super_imsm_volume(struct supertype *st, mdu_array_info_t *info,
                return 0;
        }
        strncpy((char *) dev->volume, name, MAX_RAID_SERIAL_LEN);
-       array_blocks = calc_array_size(info->level, info->raid_disks,
-                                      info->layout, info->chunk_size,
-                                      info->size*2);
+       if (info->level == 1)
+               array_blocks = info_to_blocks_per_member(info);
+       else
+               array_blocks = calc_array_size(info->level, info->raid_disks,
+                                              info->layout, info->chunk_size,
+                                              info->size*2);
        dev->size_low = __cpu_to_le32((__u32) array_blocks);
        dev->size_high = __cpu_to_le32((__u32) (array_blocks >> 32));
        dev->status = __cpu_to_le32(0);
@@ -1821,15 +2140,8 @@ static int init_super_imsm_volume(struct supertype *st, mdu_array_info_t *info,
        vol->migr_type = MIGR_INIT;
        vol->dirty = 0;
        vol->curr_migr_unit = 0;
-       for (i = 0; i < idx; i++) {
-               struct imsm_dev *prev = get_imsm_dev(super, i);
-               struct imsm_map *pmap = get_imsm_map(prev, 0);
-
-               offset += __le32_to_cpu(pmap->blocks_per_member);
-               offset += IMSM_RESERVED_SECTORS;
-       }
        map = get_imsm_map(dev, 0);
-       map->pba_of_lba0 = __cpu_to_le32(offset);
+       map->pba_of_lba0 = __cpu_to_le32(super->create_offset);
        map->blocks_per_member = __cpu_to_le32(info_to_blocks_per_member(info));
        map->blocks_per_strip = __cpu_to_le16(info_to_blocks_per_strip(info));
        map->num_data_stripes = __cpu_to_le32(info_to_num_data_stripes(info));
@@ -1911,7 +2223,7 @@ static int init_super_imsm(struct supertype *st, mdu_array_info_t *info,
 }
 
 #ifndef MDASSEMBLE
-static void add_to_super_imsm_volume(struct supertype *st, mdu_disk_info_t *dk,
+static int add_to_super_imsm_volume(struct supertype *st, mdu_disk_info_t *dk,
                                     int fd, char *devname)
 {
        struct intel_super *super = st->sb;
@@ -1923,13 +2235,21 @@ static void add_to_super_imsm_volume(struct supertype *st, mdu_disk_info_t *dk,
        dev = get_imsm_dev(super, super->current_vol);
        map = get_imsm_map(dev, 0);
 
+       if (! (dk->state & (1<<MD_DISK_SYNC))) {
+               fprintf(stderr, Name ": %s: Cannot add spare devices to IMSM volume\n",
+                       devname);
+               return 1;
+       }
+
        for (dl = super->disks; dl ; dl = dl->next)
                if (dl->major == dk->major &&
                    dl->minor == dk->minor)
                        break;
 
-       if (!dl || ! (dk->state & (1<<MD_DISK_SYNC)))
-               return;
+       if (!dl) {
+               fprintf(stderr, Name ": %s is not a member of the same container\n", devname);
+               return 1;
+       }
 
        /* add a pristine spare to the metadata */
        if (dl->index < 0) {
@@ -1950,9 +2270,11 @@ static void add_to_super_imsm_volume(struct supertype *st, mdu_disk_info_t *dk,
                sum = __gen_imsm_checksum(mpb);
                mpb->family_num = __cpu_to_le32(sum);
        }
+
+       return 0;
 }
 
-static void add_to_super_imsm(struct supertype *st, mdu_disk_info_t *dk,
+static int add_to_super_imsm(struct supertype *st, mdu_disk_info_t *dk,
                              int fd, char *devname)
 {
        struct intel_super *super = st->sb;
@@ -1962,17 +2284,25 @@ static void add_to_super_imsm(struct supertype *st, mdu_disk_info_t *dk,
        int rv;
        struct stat stb;
 
-       if (super->current_vol >= 0) {
-               add_to_super_imsm_volume(st, dk, fd, devname);
-               return;
+       /* if we are on an RAID enabled platform check that the disk is
+        * attached to the raid controller
+        */
+       if (super->hba && !disk_attached_to_hba(fd, super->hba)) {
+               fprintf(stderr,
+                       Name ": %s is not attached to the raid controller: %s\n",
+                       devname ? : "disk", super->hba);
+               return 1;
        }
 
+       if (super->current_vol >= 0)
+               return add_to_super_imsm_volume(st, dk, fd, devname);
+
        fstat(fd, &stb);
        dd = malloc(sizeof(*dd));
        if (!dd) {
                fprintf(stderr,
                        Name ": malloc failed %s:%d.\n", __func__, __LINE__);
-               abort();
+               return 1;
        }
        memset(dd, 0, sizeof(*dd));
        dd->major = major(stb.st_rdev);
@@ -2005,6 +2335,8 @@ static void add_to_super_imsm(struct supertype *st, mdu_disk_info_t *dk,
                dd->next = super->disks;
                super->disks = dd;
        }
+
+       return 0;
 }
 
 static int store_imsm_mpb(int fd, struct intel_super *super);
@@ -2115,8 +2447,14 @@ static int create_array(struct supertype *st)
        struct imsm_update_create_array *u;
        struct intel_super *super = st->sb;
        struct imsm_dev *dev = get_imsm_dev(super, super->current_vol);
+       struct imsm_map *map = get_imsm_map(dev, 0);
+       struct disk_info *inf;
+       struct imsm_disk *disk;
+       int i;
+       int idx;
 
-       len = sizeof(*u) - sizeof(*dev) + sizeof_imsm_dev(dev, 0);
+       len = sizeof(*u) - sizeof(*dev) + sizeof_imsm_dev(dev, 0) +
+             sizeof(*inf) * map->num_members;
        u = malloc(len);
        if (!u) {
                fprintf(stderr, "%s: failed to allocate update buffer\n",
@@ -2127,6 +2465,12 @@ static int create_array(struct supertype *st)
        u->type = update_create_array;
        u->dev_idx = super->current_vol;
        imsm_copy_dev(&u->dev, dev);
+       inf = get_disk_info(u);
+       for (i = 0; i < map->num_members; i++) {
+               idx = get_imsm_disk_idx(dev, i);
+               disk = get_imsm_disk(super, idx);
+               serialcpy(inf[i].serial, disk->serial);
+       }
        append_metadata_update(st, u, len);
 
        return 0;
@@ -2218,12 +2562,25 @@ static int validate_geometry_imsm_container(struct supertype *st, int level,
 {
        int fd;
        unsigned long long ldsize;
+       const struct imsm_orom *orom;
 
        if (level != LEVEL_CONTAINER)
                return 0;
        if (!dev)
                return 1;
 
+       if (check_env("IMSM_NO_PLATFORM"))
+               orom = NULL;
+       else
+               orom = find_imsm_orom();
+       if (orom && raiddisks > orom->tds) {
+               if (verbose)
+                       fprintf(stderr, Name ": %d exceeds maximum number of"
+                               " platform supported disks: %d\n",
+                               raiddisks, orom->tds);
+               return 0;
+       }
+
        fd = open(dev, O_RDONLY|O_EXCL, 0);
        if (fd < 0) {
                if (verbose)
@@ -2242,6 +2599,132 @@ static int validate_geometry_imsm_container(struct supertype *st, int level,
        return 1;
 }
 
+static unsigned long long find_size(struct extent *e, int *idx, int num_extents)
+{
+       const unsigned long long base_start = e[*idx].start;
+       unsigned long long end = base_start + e[*idx].size;
+       int i;
+
+       if (base_start == end)
+               return 0;
+
+       *idx = *idx + 1;
+       for (i = *idx; i < num_extents; i++) {
+               /* extend overlapping extents */
+               if (e[i].start >= base_start &&
+                   e[i].start <= end) {
+                       if (e[i].size == 0)
+                               return 0;
+                       if (e[i].start + e[i].size > end)
+                               end = e[i].start + e[i].size;
+               } else if (e[i].start > end) {
+                       *idx = i;
+                       break;
+               }
+       }
+
+       return end - base_start;
+}
+
+static unsigned long long merge_extents(struct intel_super *super, int sum_extents)
+{
+       /* build a composite disk with all known extents and generate a new
+        * 'maxsize' given the "all disks in an array must share a common start
+        * offset" constraint
+        */
+       struct extent *e = calloc(sum_extents, sizeof(*e));
+       struct dl *dl;
+       int i, j;
+       int start_extent;
+       unsigned long long pos;
+       unsigned long long start;
+       unsigned long long maxsize;
+       unsigned long reserve;
+
+       if (!e)
+               return ~0ULL; /* error */
+
+       /* coalesce and sort all extents. also, check to see if we need to
+        * reserve space between member arrays
+        */
+       j = 0;
+       for (dl = super->disks; dl; dl = dl->next) {
+               if (!dl->e)
+                       continue;
+               for (i = 0; i < dl->extent_cnt; i++)
+                       e[j++] = dl->e[i];
+       }
+       qsort(e, sum_extents, sizeof(*e), cmp_extent);
+
+       /* merge extents */
+       i = 0;
+       j = 0;
+       while (i < sum_extents) {
+               e[j].start = e[i].start;
+               e[j].size = find_size(e, &i, sum_extents);
+               j++;
+               if (e[j-1].size == 0)
+                       break;
+       }
+
+       pos = 0;
+       maxsize = 0;
+       start_extent = 0;
+       i = 0;
+       do {
+               unsigned long long esize;
+
+               esize = e[i].start - pos;
+               if (esize >= maxsize) {
+                       maxsize = esize;
+                       start = pos;
+                       start_extent = i;
+               }
+               pos = e[i].start + e[i].size;
+               i++;
+       } while (e[i-1].size);
+       free(e);
+
+       if (start_extent > 0)
+               reserve = IMSM_RESERVED_SECTORS; /* gap between raid regions */
+       else
+               reserve = 0;
+
+       if (maxsize < reserve)
+               return ~0ULL;
+
+       super->create_offset = ~((__u32) 0);
+       if (start + reserve > super->create_offset)
+               return ~0ULL; /* start overflows create_offset */
+       super->create_offset = start + reserve;
+
+       return maxsize - reserve;
+}
+
+static int is_raid_level_supported(const struct imsm_orom *orom, int level, int raiddisks)
+{
+       if (level < 0 || level == 6 || level == 4)
+               return 0;
+
+       /* if we have an orom prevent invalid raid levels */
+       if (orom)
+               switch (level) {
+               case 0: return imsm_orom_has_raid0(orom);
+               case 1:
+                       if (raiddisks > 2)
+                               return imsm_orom_has_raid1e(orom);
+                       else
+                               return imsm_orom_has_raid1(orom);
+               case 10: return imsm_orom_has_raid10(orom);
+               case 5: return imsm_orom_has_raid5(orom);
+               }
+       else
+               return 1; /* not on an Intel RAID platform so anything goes */
+
+       return 0;
+}
+
+#define vprintf(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd 
  * FIX ME add ahci details
  */
@@ -2259,19 +2742,28 @@ static int validate_geometry_imsm_volume(struct supertype *st, int level,
        struct extent *e;
        int i;
 
-       if (level == LEVEL_CONTAINER)
+       /* We must have the container info already read in. */
+       if (!super)
                return 0;
 
-       if (level == 1 && raiddisks > 2) {
-               if (verbose)
-                       fprintf(stderr, Name ": imsm does not support more "
-                               "than 2 in a raid1 configuration\n");
+       if (!is_raid_level_supported(super->orom, level, raiddisks)) {
+               vprintf(": platform does not support raid level: %d\n", level);
                return 0;
        }
-
-       /* We must have the container info already read in. */
-       if (!super)
+       if (super->orom && !imsm_orom_has_chunk(super->orom, chunk)) {
+               vprintf(": platform does not support a chunk size of: %d\n", chunk);
+               return 0;
+       }
+       if (layout != imsm_level_to_layout(level)) {
+               if (level == 5)
+                       vprintf(": imsm raid 5 only supports the left-asymmetric layout\n");
+               else if (level == 10)
+                       vprintf(": imsm raid 10 only supports the n2 layout\n");
+               else
+                       vprintf(": imsm unknown layout %#x for this raid level %d\n",
+                               layout, level);
                return 0;
+       }
 
        if (!dev) {
                /* General test:  make sure there is space for
@@ -2319,6 +2811,7 @@ static int validate_geometry_imsm_volume(struct supertype *st, int level,
                }
                return 1;
        }
+
        /* This device must be a member of the set */
        if (stat(dev, &stb) < 0)
                return 0;
@@ -2335,17 +2828,54 @@ static int validate_geometry_imsm_volume(struct supertype *st, int level,
                                "same imsm set\n", dev);
                return 0;
        }
+
+       /* retrieve the largest free space block */
        e = get_extents(super, dl);
        maxsize = 0;
        i = 0;
-       if (e) do {
-               unsigned long long esize;
-               esize = e[i].start - pos;
-               if (esize >= maxsize)
-                       maxsize = esize;
-               pos = e[i].start + e[i].size;
-               i++;
-       } while (e[i-1].size);
+       if (e) {
+               do {
+                       unsigned long long esize;
+
+                       esize = e[i].start - pos;
+                       if (esize >= maxsize)
+                               maxsize = esize;
+                       pos = e[i].start + e[i].size;
+                       i++;
+               } while (e[i-1].size);
+               dl->e = e;
+               dl->extent_cnt = i;
+       } else {
+               if (verbose)
+                       fprintf(stderr, Name ": unable to determine free space for: %s\n",
+                               dev);
+               return 0;
+       }
+       if (maxsize < size) {
+               if (verbose)
+                       fprintf(stderr, Name ": %s not enough space (%llu < %llu)\n",
+                               dev, maxsize, size);
+               return 0;
+       }
+
+       /* count total number of extents for merge */
+       i = 0;
+       for (dl = super->disks; dl; dl = dl->next)
+               if (dl->e)
+                       i += dl->extent_cnt;
+
+       maxsize = merge_extents(super, i);
+       if (maxsize < size) {
+               if (verbose)
+                       fprintf(stderr, Name ": not enough space after merge (%llu < %llu)\n",
+                               maxsize, size);
+               return 0;
+       } else if (maxsize == ~0ULL) {
+               if (verbose)
+                       fprintf(stderr, Name ": failed to merge %d extents\n", i);
+               return 0;
+       }
+
        *freesize = maxsize;
 
        return 1;
@@ -3103,18 +3633,20 @@ static struct mdinfo *imsm_activate_spare(struct active_array *a,
        return rv;
 }
 
-static int disks_overlap(struct imsm_dev *d1, struct imsm_dev *d2)
+static int disks_overlap(struct intel_super *super, int idx, struct imsm_update_create_array *u)
 {
-       struct imsm_map *m1 = get_imsm_map(d1, 0);
-       struct imsm_map *m2 = get_imsm_map(d2, 0);
+       struct imsm_dev *dev = get_imsm_dev(super, idx);
+       struct imsm_map *map = get_imsm_map(dev, 0);
+       struct imsm_map *new_map = get_imsm_map(&u->dev, 0);
+       struct disk_info *inf = get_disk_info(u);
+       struct imsm_disk *disk;
        int i;
        int j;
-       int idx;
 
-       for (i = 0; i < m1->num_members; i++) {
-               idx = get_imsm_disk_idx(d1, i);
-               for (j = 0; j < m2->num_members; j++)
-                       if (idx == get_imsm_disk_idx(d2, j))
+       for (i = 0; i < map->num_members; i++) {
+               disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i));
+               for (j = 0; j < new_map->num_members; j++)
+                       if (serialcmp(disk->serial, inf[j].serial) == 0)
                                return 1;
        }
 
@@ -3260,7 +3792,8 @@ static void imsm_process_update(struct supertype *st,
                unsigned long long start, end;
                unsigned long long new_start, new_end;
                int i;
-               int overlap = 0;
+               struct disk_info *inf;
+               struct dl *dl;
 
                /* handle racing creates: first come first serve */
                if (u->dev_idx < mpb->num_raid_devs) {
@@ -3279,6 +3812,7 @@ static void imsm_process_update(struct supertype *st,
                new_map = get_imsm_map(&u->dev, 0);
                new_start = __le32_to_cpu(new_map->pba_of_lba0);
                new_end = new_start + __le32_to_cpu(new_map->blocks_per_member);
+               inf = get_disk_info(u);
 
                /* handle activate_spare versus create race:
                 * check to make sure that overlapping arrays do not include
@@ -3291,17 +3825,15 @@ static void imsm_process_update(struct supertype *st,
                        end = start + __le32_to_cpu(map->blocks_per_member);
                        if ((new_start >= start && new_start <= end) ||
                            (start >= new_start && start <= new_end))
-                               overlap = 1;
-                       if (overlap && disks_overlap(dev, &u->dev)) {
+                               /* overlap */;
+                       else
+                               continue;
+
+                       if (disks_overlap(super, i, u)) {
                                dprintf("%s: arrays overlap\n", __func__);
                                return;
                        }
                }
-               /* check num_members sanity */
-               if (new_map->num_members > mpb->num_disks) {
-                       dprintf("%s: num_disks out of range\n", __func__);
-                       return;
-               }
 
                /* check that prepare update was successful */
                if (!update->space) {
@@ -3309,26 +3841,40 @@ static void imsm_process_update(struct supertype *st,
                        return;
                }
 
+               /* check that all disks are still active before committing
+                * changes.  FIXME: could we instead handle this by creating a
+                * degraded array?  That's probably not what the user expects,
+                * so better to drop this update on the floor.
+                */
+               for (i = 0; i < new_map->num_members; i++) {
+                       dl = serial_to_dl(inf[i].serial, super);
+                       if (!dl) {
+                               dprintf("%s: disk disappeared\n", __func__);
+                               return;
+                       }
+               }
+
                super->updates_pending++;
+
+               /* convert spares to members and fixup ord_tbl */
+               for (i = 0; i < new_map->num_members; i++) {
+                       dl = serial_to_dl(inf[i].serial, super);
+                       if (dl->index == -1) {
+                               dl->index = mpb->num_disks;
+                               mpb->num_disks++;
+                               dl->disk.status |= CONFIGURED_DISK;
+                               dl->disk.status &= ~SPARE_DISK;
+                       }
+                       set_imsm_ord_tbl_ent(new_map, i, dl->index);
+               }
+
                dev = update->space;
-               map = get_imsm_map(dev, 0);
                update->space = NULL;
                imsm_copy_dev(dev, &u->dev);
-               map = get_imsm_map(dev, 0);
                super->dev_tbl[u->dev_idx] = dev;
                mpb->num_raid_devs++;
 
-               /* fix up flags */
-               for (i = 0; i < map->num_members; i++) {
-                       struct imsm_disk *disk;
-
-                       disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i));
-                       disk->status |= CONFIGURED_DISK;
-                       disk->status &= ~SPARE_DISK;
-               }
-
                imsm_update_version_info(super);
-
                break;
        }
        case update_add_disk:
@@ -3377,9 +3923,31 @@ static void imsm_prepare_update(struct supertype *st,
        switch (type) {
        case update_create_array: {
                struct imsm_update_create_array *u = (void *) update->buf;
+               struct imsm_dev *dev = &u->dev;
+               struct imsm_map *map = get_imsm_map(dev, 0);
+               struct dl *dl;
+               struct disk_info *inf;
+               int i;
+               int activate = 0;
 
-               len = sizeof_imsm_dev(&u->dev, 1);
+               inf = get_disk_info(u);
+               len = sizeof_imsm_dev(dev, 1);
+               /* allocate a new super->dev_tbl entry */
                update->space = malloc(len);
+
+               /* count how many spares will be converted to members */
+               for (i = 0; i < map->num_members; i++) {
+                       dl = serial_to_dl(inf[i].serial, super);
+                       if (!dl) {
+                               /* hmm maybe it failed?, nothing we can do about
+                                * it here
+                                */
+                               continue;
+                       }
+                       if (count_memberships(dl, super) == 0)
+                               activate++;
+               }
+               len += activate * sizeof(struct imsm_disk);
                break;
        default:
                break;
@@ -3469,6 +4037,7 @@ struct superswitch super_imsm = {
        .write_init_super = write_init_super_imsm,
        .validate_geometry = validate_geometry_imsm,
        .add_to_super   = add_to_super_imsm,
+       .detail_platform = detail_platform_imsm,
 #endif
        .match_home     = match_home_imsm,
        .uuid_from_super= uuid_from_super_imsm,
@@ -3487,6 +4056,7 @@ struct superswitch super_imsm = {
        .container_content = container_content_imsm,
 
        .external       = 1,
+       .name = "imsm",
 
 #ifndef MDASSEMBLE
 /* for mdmon */