]> git.ipfire.org Git - thirdparty/mdadm.git/blobdiff - super1.c
super1: set RESHAPE_NO_BACKUP based on new_offset.
[thirdparty/mdadm.git] / super1.c
index 3236a7e28b301f3ef481b50c6180b8473734133e..a9ca2169bf486f7f68b6670557c071eebb621cd9 100644 (file)
--- a/super1.c
+++ b/super1.c
@@ -269,6 +269,7 @@ static void examine_super1(struct supertype *st, char *homehost)
        int l = homehost ? strlen(homehost) : 0;
        int layout;
        unsigned long long sb_offset;
+       struct mdinfo info;
 
        printf("          Magic : %08x\n", __le32_to_cpu(sb->magic));
        printf("        Version : 1");
@@ -327,7 +328,8 @@ static void examine_super1(struct supertype *st, char *homehost)
        if (sb->data_offset)
                printf("    Data Offset : %llu sectors\n",
                       (unsigned long long)__le64_to_cpu(sb->data_offset));
-       if (sb->new_offset) {
+       if (sb->new_offset &&
+           (__le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET)) {
                unsigned long long offset = __le64_to_cpu(sb->data_offset);
                offset += (signed)(int32_t)__le32_to_cpu(sb->new_offset);
                printf("     New Offset : %llu sectors\n", offset);
@@ -336,6 +338,13 @@ static void examine_super1(struct supertype *st, char *homehost)
               (unsigned long long)__le64_to_cpu(sb->super_offset));
        if (__le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET)
                printf("Recovery Offset : %llu sectors\n", (unsigned long long)__le64_to_cpu(sb->recovery_offset));
+
+       st->ss->getinfo_super(st, &info, NULL);
+       if (info.space_after != 1 &&
+           !(__le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
+               printf("   Unused Space : before=%llu sectors, after=%llu sectors\n",
+                      info.space_before, info.space_after);
+
        printf("          State : %s\n", (__le64_to_cpu(sb->resync_offset)+1)? "active":"clean");
        printf("    Device UUID : ");
        for (i=0; i<16; i++) {
@@ -395,7 +404,7 @@ static void examine_super1(struct supertype *st, char *homehost)
        if (sb->bblog_size && sb->bblog_offset) {
                printf("  Bad Block Log : %d entries available at offset %ld sectors",
                       __le16_to_cpu(sb->bblog_size)*512/8,
-                      (long)__le32_to_cpu(sb->bblog_offset));
+                      (long)(int32_t)__le32_to_cpu(sb->bblog_offset));
                if (sb->feature_map &
                    __cpu_to_le32(MD_FEATURE_BAD_BLOCKS))
                        printf(" - bad blocks present.");
@@ -597,6 +606,143 @@ static void export_examine_super1(struct supertype *st)
               (unsigned long long)__le64_to_cpu(sb->events));
 }
 
+static int copy_metadata1(struct supertype *st, int from, int to)
+{
+       /* Read superblock.  If it looks good, write it out.
+        * Then if a bitmap is present, copy that.
+        * And if a bad-block-list is present, copy that too.
+        */
+       void *buf;
+       unsigned long long dsize, sb_offset;
+       const int bufsize = 4*1024;
+       struct mdp_superblock_1 super, *sb;
+
+       if (posix_memalign(&buf, 4096, bufsize) != 0)
+               return 1;
+
+       if (!get_dev_size(from, NULL, &dsize))
+               goto err;
+
+       dsize >>= 9;
+       if (dsize < 24)
+               goto err;
+       switch(st->minor_version) {
+       case 0:
+               sb_offset = dsize;
+               sb_offset -= 8*2;
+               sb_offset &= ~(4*2-1);
+               break;
+       case 1:
+               sb_offset = 0;
+               break;
+       case 2:
+               sb_offset = 4*2;
+               break;
+       default:
+               goto err;
+       }
+
+       if (lseek64(from, sb_offset << 9, 0) < 0LL)
+               goto err;
+       if (read(from, buf, bufsize) != bufsize)
+               goto err;
+
+       sb = buf;
+       super = *sb; // save most of sb for when we reuse buf
+
+       if (__le32_to_cpu(super.magic) != MD_SB_MAGIC ||
+           __le32_to_cpu(super.major_version) != 1 ||
+           __le64_to_cpu(super.super_offset) != sb_offset ||
+           calc_sb_1_csum(sb) != super.sb_csum)
+               goto err;
+
+       if (lseek64(to, sb_offset << 9, 0) < 0LL)
+               goto err;
+       if (write(to, buf, bufsize) != bufsize)
+               goto err;
+
+       if (super.feature_map & __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET)) {
+               unsigned long long bitmap_offset = sb_offset;
+               int bytes = 4096; // just an estimate.
+               int written = 0;
+               struct align_fd afrom, ato;
+
+               init_afd(&afrom, from);
+               init_afd(&ato, to);
+
+               bitmap_offset += (int32_t)__le32_to_cpu(super.bitmap_offset);
+
+               if (lseek64(from, bitmap_offset<<9, 0) < 0)
+                       goto err;
+               if (lseek64(to, bitmap_offset<<9, 0) < 0)
+                       goto err;
+
+               for (written = 0; written < bytes ; ) {
+                       int n = bytes - written;
+                       if (n > 4096)
+                               n = 4096;
+                       if (aread(&afrom, buf, n) != n)
+                               goto err;
+                       if (written == 0) {
+                               /* have the header, can calculate
+                                * correct bitmap bytes */
+                               bitmap_super_t *bms;
+                               int bits;
+                               bms = (void*)buf;
+                               bits = __le64_to_cpu(bms->sync_size) / (__le32_to_cpu(bms->chunksize)>>9);
+                               bytes = (bits+7) >> 3;
+                               bytes += sizeof(bitmap_super_t);
+                               bytes = ROUND_UP(bytes, 512);
+                               if (n > bytes)
+                                       n =  bytes;
+                       }
+                       if (awrite(&ato, buf, n) != n)
+                               goto err;
+                       written += n;
+               }
+       }
+
+       if (super.bblog_size != 0 &&
+           __le32_to_cpu(super.bblog_size) <= 100 &&
+           super.bblog_offset != 0 &&
+           (super.feature_map & __le32_to_cpu(MD_FEATURE_BAD_BLOCKS))) {
+               /* There is a bad block log */
+               unsigned long long bb_offset = sb_offset;
+               int bytes = __le32_to_cpu(super.bblog_size) * 512;
+               int written = 0;
+               struct align_fd afrom, ato;
+
+               init_afd(&afrom, from);
+               init_afd(&ato, to);
+
+               bb_offset += (int32_t)__le32_to_cpu(super.bblog_offset);
+
+               if (lseek64(from, bb_offset<<9, 0) < 0)
+                       goto err;
+               if (lseek64(to, bb_offset<<9, 0) < 0)
+                       goto err;
+
+               for (written = 0; written < bytes ; ) {
+                       int n = bytes - written;
+                       if (n > 4096)
+                               n = 4096;
+                       if (aread(&afrom, buf, n) != n)
+                               goto err;
+
+                       if (awrite(&ato, buf, n) != n)
+                               goto err;
+                       written += n;
+               }
+       }
+
+       free(buf);
+       return 0;
+
+err:
+       free(buf);
+       return 1;
+}
+
 static void detail_super1(struct supertype *st, char *homehost)
 {
        struct mdp_superblock_1 *sb = st->sb;
@@ -668,7 +814,10 @@ static int examine_badblocks_super1(struct supertype *st, int fd, char *devname)
        }
 
        size = __le32_to_cpu(sb->bblog_size)* 512;
-       posix_memalign((void**)&bbl, 4096, size);
+       if (posix_memalign((void**)&bbl, 4096, size) != 0) {
+               pr_err("%s could not allocate badblocks list\n", __func__);
+               return 0;
+       }
        offset = __le64_to_cpu(sb->super_offset) +
                (int)__le32_to_cpu(sb->bblog_offset);
        offset <<= 9;
@@ -769,15 +918,24 @@ static void getinfo_super1(struct supertype *st, struct mdinfo *info, char *map)
                unsigned long long end;
                info->space_before = info->data_offset;
                end = super_offset;
-               if (info->bitmap_offset < 0)
-                       end += info->bitmap_offset;
+
+               if (sb->bblog_offset && sb->bblog_size) {
+                       unsigned long long bboffset = super_offset;
+                       bboffset += (int32_t)__le32_to_cpu(sb->bblog_offset);
+                       if (bboffset < end)
+                               end = bboffset;
+               }
+
+               if (super_offset + info->bitmap_offset < end)
+                       end = super_offset + info->bitmap_offset;
+
                if (info->data_offset + data_size < end)
                        info->space_after = end - data_size - info->data_offset;
                else
                        info->space_after = 0;
        } else {
-               info->space_before = (info->data_offset -
-                                     super_offset);
+               unsigned long long earliest;
+               earliest = super_offset + (32+4)*2; /* match kernel */
                if (info->bitmap_offset > 0) {
                        unsigned long long bmend = info->bitmap_offset;
                        unsigned long long size = __le64_to_cpu(bsb->sync_size);
@@ -786,15 +944,31 @@ static void getinfo_super1(struct supertype *st, struct mdinfo *info, char *map)
                        size += sizeof(bitmap_super_t);
                        size = ROUND_UP(size, 4096);
                        size /= 512;
-                       size += bmend;
-                       if (size < info->space_before)
-                               info->space_before -= size;
-                       else
-                               info->space_before = 0;
-               } else
-                       info->space_before -= 8; /* superblock */
+                       bmend += size;
+                       if (bmend > earliest)
+                               bmend = earliest;
+               }
+               if (sb->bblog_offset && sb->bblog_size) {
+                       unsigned long long bbend = super_offset;
+                       bbend += (int32_t)__le32_to_cpu(sb->bblog_offset);
+                       bbend += __le32_to_cpu(sb->bblog_size);
+                       if (bbend > earliest)
+                               earliest = bbend;
+               }
+               if (earliest < info->data_offset)
+                       info->space_before = info->data_offset - earliest;
+               else
+                       info->space_before = 0;
                info->space_after = misc->device_size - data_size - info->data_offset;
        }
+       if (info->space_before == 0 && info->space_after == 0) {
+               /* It will look like we don't support data_offset changes,
+                * be we do - it's just that there is no room.
+                * A change that reduced the number of devices should
+                * still be allowed, so set the otherwise useless value of '1'
+                */
+               info->space_after = 1;
+       }
 
        info->disk.raid_disk = -1;
        switch(role) {
@@ -832,7 +1006,8 @@ static void getinfo_super1(struct supertype *st, struct mdinfo *info, char *map)
 
        if (sb->feature_map & __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE)) {
                info->reshape_active = 1;
-               if (info->array.level == 10)
+               if ((sb->feature_map & __le32_to_cpu(MD_FEATURE_NEW_OFFSET)) &&
+                   sb->new_offset != 0)
                        info->reshape_active |= RESHAPE_NO_BACKUP;
                info->reshape_progress = __le64_to_cpu(sb->reshape_position);
                info->new_level = __le32_to_cpu(sb->new_level);
@@ -885,6 +1060,21 @@ static int update_super1(struct supertype *st, struct mdinfo *info,
        int rv = 0;
        struct mdp_superblock_1 *sb = st->sb;
 
+       if (strcmp(update, "homehost") == 0 &&
+           homehost) {
+               /* Note that 'homehost' is special as it is really
+                * a "name" update.
+                */
+               char *c;
+               update = "name";
+               c = strchr(sb->set_name, ':');
+               if (c)
+                       strncpy(info->name, c+1, 31 - (c-sb->set_name));
+               else
+                       strncpy(info->name, sb->set_name, 32);
+               info->name[32] = 0;
+       }
+
        if (strcmp(update, "force-one")==0) {
                /* Not enough devices for a working array,
                 * so bring this one up-to-date
@@ -993,27 +1183,29 @@ static int update_super1(struct supertype *st, struct mdinfo *info,
                 */
                unsigned long long sb_offset = __le64_to_cpu(sb->super_offset);
                unsigned long long data_offset = __le64_to_cpu(sb->data_offset);
-               long bitmap_offset = (long)__le64_to_cpu(sb->bitmap_offset);
+               long bitmap_offset = (long)(int32_t)__le32_to_cpu(sb->bitmap_offset);
                long bm_sectors = 0;
                long space;
 
+#ifndef MDASSEMBLE
                if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
                        struct bitmap_super_s *bsb;
                        bsb = (struct bitmap_super_s *)(((char*)sb)+MAX_SB_SIZE);
                        bm_sectors = bitmap_sectors(bsb);
                }
-
+#endif
                if (sb_offset < data_offset) {
-                       /* 1.1 or 1.2.  Put bbl just before data
+                       /* 1.1 or 1.2.  Put bbl after bitmap leaving at least 32K
                         */
                        long bb_offset;
-                       space = data_offset - sb_offset;
-                       bb_offset = space - 8;
+                       bb_offset = sb_offset + 8;
                        if (bm_sectors && bitmap_offset > 0)
-                               space -= (bitmap_offset + bm_sectors);
-                       else
-                               space -= 8; /* The superblock */
-                       if (space >= 8) {
+                               bb_offset = bitmap_offset + bm_sectors;
+                       while (bb_offset < (long)sb_offset + 8 + 32*2
+                              && bb_offset + 8+8 <= (long)data_offset)
+                               /* too close to bitmap, and room to grow */
+                               bb_offset += 8;
+                       if (bb_offset + 8 <= (long)data_offset) {
                                sb->bblog_size = __cpu_to_le16(8);
                                sb->bblog_offset = __cpu_to_le32(bb_offset);
                        }
@@ -1037,16 +1229,6 @@ static int update_super1(struct supertype *st, struct mdinfo *info,
                        sb->bblog_shift = 0;
                        sb->bblog_offset = 0;
                }
-       } else if (strcmp(update, "homehost") == 0 &&
-                  homehost) {
-               char *c;
-               update = "name";
-               c = strchr(sb->set_name, ':');
-               if (c)
-                       strncpy(info->name, c+1, 31 - (c-sb->set_name));
-               else
-                       strncpy(info->name, sb->set_name, 32);
-               info->name[32] = 0;
        } else if (strcmp(update, "name") == 0) {
                if (info->name[0] == 0)
                        sprintf(info->name, "%d", info->array.md_minor);
@@ -1071,6 +1253,35 @@ static int update_super1(struct supertype *st, struct mdinfo *info,
                        misc->device_size - __le64_to_cpu(sb->data_offset));
                printf("Size is %llu\n", (unsigned long long)
                       __le64_to_cpu(sb->data_size));
+       } else if (strcmp(update, "revert-reshape") == 0) {
+               rv = -2;
+               if (!(sb->feature_map & __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE)))
+                       pr_err("No active reshape to revert on %s\n",
+                              devname);
+               else {
+                       rv = 0;
+                       __u32 temp;
+                       sb->raid_disks = __cpu_to_le32(__le32_to_cpu(sb->raid_disks) +
+                                                      __le32_to_cpu(sb->delta_disks));
+                       if (sb->delta_disks == 0)
+                               sb->feature_map ^= __cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
+                       else
+                               sb->delta_disks = __cpu_to_le32(-__le32_to_cpu(sb->delta_disks));
+
+                       temp = sb->new_layout;
+                       sb->new_layout = sb->layout;
+                       sb->layout = temp;
+
+                       temp = sb->new_chunk;
+                       sb->new_chunk = sb->chunksize;
+                       sb->chunksize = temp;
+
+                       if (sb->feature_map & __cpu_to_le32(MD_FEATURE_NEW_OFFSET)) {
+                               sb->data_offset = __cpu_to_le64(__le64_to_cpu(sb->data_offset) +
+                                                               (long)(int32_t)__le32_to_cpu(sb->new_offset));
+                               sb->new_offset = __cpu_to_le32(-(int32_t)__le32_to_cpu(sb->new_offset));
+                       }
+               }
        } else if (strcmp(update, "_reshape_progress")==0)
                sb->reshape_position = __cpu_to_le64(info->reshape_progress);
        else if (strcmp(update, "writemostly")==0)
@@ -1444,7 +1655,10 @@ static int write_init_super1(struct supertype *st)
 
                        sb->data_offset = __cpu_to_le64(reserved);
                        sb->data_size = __cpu_to_le64(dsize - reserved);
-                       if (reserved >= 16) {
+                       if (reserved >= 8 + 32*2 + 8) {
+                               sb->bblog_size = __cpu_to_le16(8);
+                               sb->bblog_offset = __cpu_to_le32(8 + 32*2);
+                       } else if (reserved >= 16) {
                                sb->bblog_size = __cpu_to_le16(8);
                                sb->bblog_offset = __cpu_to_le32(reserved-8);
                        }
@@ -1476,7 +1690,10 @@ static int write_init_super1(struct supertype *st)
 
                        sb->data_offset = __cpu_to_le64(reserved);
                        sb->data_size = __cpu_to_le64(dsize - reserved);
-                       if (reserved >= 16+16) {
+                       if (reserved >= 16 + 32*2 + 8) {
+                               sb->bblog_size = __cpu_to_le16(8);
+                               sb->bblog_offset = __cpu_to_le32(8 + 32*2);
+                       } else if (reserved >= 16+16) {
                                sb->bblog_size = __cpu_to_le16(8);
                                /* '8' sectors for the bblog, and another '8'
                                 * because we want offset from superblock, not
@@ -1633,9 +1850,6 @@ static int load_super1(struct supertype *st, int fd, char *devname)
                return -EINVAL;
        }
 
-       ioctl(fd, BLKFLSBUF, 0); /* make sure we read current data */
-
-
        if (lseek64(fd, sb_offset << 9, 0)< 0LL) {
                if (devname)
                        pr_err("Cannot seek to superblock on %s: %s\n",
@@ -1714,7 +1928,7 @@ static struct supertype *match_metadata_desc1(char *arg)
 {
        struct supertype *st = xcalloc(1, sizeof(*st));
 
-       st->container_dev = NoMdDev;
+       st->container_devnm[0] = 0;
        st->ss = &super1;
        st->max_devs = MAX_DEVS;
        st->sb = NULL;
@@ -2091,6 +2305,60 @@ static int validate_geometry1(struct supertype *st, int level,
 }
 #endif /* MDASSEMBLE */
 
+void *super1_make_v0(struct supertype *st, struct mdinfo *info, mdp_super_t *sb0)
+{
+       /* Create a v1.0 superblock based on 'info'*/
+       void *ret;
+       struct mdp_superblock_1 *sb;
+       int i;
+       int rfd;
+       unsigned long long offset;
+
+       if (posix_memalign(&ret, 4096, 1024) != 0)
+               return NULL;
+       sb = ret;
+       memset(ret, 0, 1024);
+       sb->magic = __cpu_to_le32(MD_SB_MAGIC);
+       sb->major_version = __cpu_to_le32(1);
+
+       copy_uuid(sb->set_uuid, info->uuid, super1.swapuuid);
+       sprintf(sb->set_name, "%d", sb0->md_minor);
+       sb->ctime = __cpu_to_le32(info->array.ctime+1);
+       sb->level = __cpu_to_le32(info->array.level);
+       sb->layout = __cpu_to_le32(info->array.layout);
+       sb->size = __cpu_to_le64(info->component_size);
+       sb->chunksize = __cpu_to_le32(info->array.chunk_size/512);
+       sb->raid_disks = __cpu_to_le32(info->array.raid_disks);
+       sb->data_size = sb->size;
+       sb->resync_offset = MaxSector;
+       sb->max_dev = __cpu_to_le32(MD_SB_DISKS);
+       sb->dev_number = __cpu_to_le32(info->disk.number);
+       sb->utime = __cpu_to_le64(info->array.utime);
+
+       offset = st->devsize/512 - 8*2;
+       offset &= ~(4*2-1);
+       sb->super_offset = __cpu_to_le64(offset);
+       //*(__u64*)(st->other + 128 + 8 + 8) = __cpu_to_le64(offset);
+
+       if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
+           read(rfd, sb->device_uuid, 16) != 16) {
+               __u32 r[4] = {random(), random(), random(), random()};
+               memcpy(sb->device_uuid, r, 16);
+       }
+       if (rfd >= 0)
+               close(rfd);
+
+       for (i = 0; i < MD_SB_DISKS; i++) {
+               int state = sb0->disks[i].state;
+               sb->dev_roles[i] = 0xFFFF;
+               if ((state & (1<<MD_DISK_SYNC)) &&
+                   !(state & (1<<MD_DISK_FAULTY)))
+                       sb->dev_roles[i] = __cpu_to_le16(sb0->disks[i].raid_disk);
+       }
+       sb->sb_csum = calc_sb_1_csum(sb);
+       return ret;
+}
+
 struct superswitch super1 = {
 #ifndef MDASSEMBLE
        .examine_super = examine_super1,
@@ -2103,6 +2371,7 @@ struct superswitch super1 = {
        .validate_geometry = validate_geometry1,
        .add_to_super = add_to_super1,
        .examine_badblocks = examine_badblocks_super1,
+       .copy_metadata = copy_metadata1,
 #endif
        .match_home = match_home1,
        .uuid_from_super = uuid_from_super1,