Whamcloud - gitweb
LU-11546 utils: enable large_dir for ldiskfs
[fs/lustre-release.git] / lustre / utils / libmount_utils_ldiskfs.c
index c1f79ae..56dd609 100644 (file)
 
 extern char *progname;
 
-#define L_BLOCK_SIZE 4096
 /* keep it less than LL_FID_NAMELEN */
 #define DUMMY_FILE_NAME_LEN             25
 #define EXT3_DIRENT_SIZE                DUMMY_FILE_NAME_LEN
 
 static void append_unique(char *buf, char *prefix, char *key, char *val,
                          size_t maxbuflen);
-static int is_e2fsprogs_feature_supp(const char *feature);
+static bool is_e2fsprogs_feature_supp(const char *feature);
 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs);
 
 /*
@@ -220,7 +219,7 @@ int ldiskfs_write_ldd(struct mkfs_opts *mop)
        /* Multiple mount protection enabled if failover node specified */
        if (mop->mo_flags & MO_FAILOVER &&
            !is_feature_enabled("mmp", dev)) {
-               if (is_e2fsprogs_feature_supp("-O mmp") == 0) {
+               if (is_e2fsprogs_feature_supp("-O mmp")) {
                        char *command = filepnm;
 
                        snprintf(command, sizeof(filepnm),
@@ -230,8 +229,9 @@ int ldiskfs_write_ldd(struct mkfs_opts *mop)
                                fprintf(stderr,
                                        "%s: Unable to set 'mmp' on %s: %d\n",
                                        progname, dev, ret);
-               } else
+               } else {
                        disp_old_e2fsprogs_msg("mmp", 1);
+               }
        }
 
        ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0,
@@ -459,14 +459,14 @@ int ldiskfs_is_lustre(char *dev, unsigned *mount_type)
  * Firstly we try to use "debugfs supported_features" command to check if
  * the feature is supported. If this fails we try to set this feature with
  * mke2fs to check for its support. */
-static int is_e2fsprogs_feature_supp(const char *feature)
+static bool is_e2fsprogs_feature_supp(const char *feature)
 {
        static char supp_features[4096] = "";
        FILE *fp;
        char cmd[PATH_MAX];
        char imgname[] = "/tmp/test-img-XXXXXX";
-       int fd = -1;
-       int ret = 1;
+       int fd;
+       int ret;
 
        if (supp_features[0] == '\0') {
                snprintf(cmd, sizeof(cmd), "%s -c -R supported_features 2>&1",
@@ -477,20 +477,22 @@ static int is_e2fsprogs_feature_supp(const char *feature)
                fp = popen(cmd, "r");
                if (!fp) {
                        fprintf(stderr, "%s: %s\n", progname, strerror(errno));
-                       return 0;
+               } else {
+                       ret = fread(supp_features, 1,
+                                   sizeof(supp_features) - 1, fp);
+                       supp_features[ret] = '\0';
+                       pclose(fp);
                }
-               ret = fread(supp_features, 1, sizeof(supp_features) - 1, fp);
-               supp_features[ret] = '\0';
-               pclose(fp);
        }
-       if (ret > 0 && strstr(supp_features,
-                             strncmp(feature, "-O ", 3) ? feature : feature+3))
-               return 0;
+
+       if (strstr(supp_features,
+                  strncmp(feature, "-O ", 3) ? feature : feature + 3))
+               return true;
 
        if ((fd = mkstemp(imgname)) < 0)
-               return -1;
-       else
-               close(fd);
+               return false;
+
+       close(fd);
 
        snprintf(cmd, sizeof(cmd), "%s -F %s %s 100 >/dev/null 2>&1",
                 MKE2FS, feature, imgname);
@@ -499,10 +501,9 @@ static int is_e2fsprogs_feature_supp(const char *feature)
        ret = system(cmd);
        unlink(imgname);
 
-       return ret;
+       return ret == 0;
 }
 
-
 /**
  * append_unique: append @key or @key=@val pair to @buf only if @key does not
  *                exists
@@ -540,14 +541,14 @@ static void append_unique(char *buf, char *prefix, char *key, char *val,
 }
 
 static int enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
-                                       size_t maxbuflen, int user_spec)
+                                       size_t maxbuflen, bool user_spec)
 {
-       int enable_64bit = 0;
+       unsigned long long blocks = mop->mo_device_kb / mop->mo_blocksize_kb;
+       bool enable_64bit = false;
 
        /* Enable large block addresses if the LUN is over 2^32 blocks. */
-       if ((mop->mo_device_kb / (L_BLOCK_SIZE >> 10) > UINT32_MAX) &&
-            is_e2fsprogs_feature_supp("-O 64bit") == 0)
-               enable_64bit = 1;
+       if (blocks > 0xffffffffULL && is_e2fsprogs_feature_supp("-O 64bit"))
+               enable_64bit = true;
 
        if (IS_OST(&mop->mo_ldd)) {
                append_unique(anchor, user_spec ? "," : " -O ",
@@ -568,14 +569,14 @@ static int enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
 
        /* Multiple mount protection enabled only if failover node specified */
        if (mop->mo_flags & MO_FAILOVER) {
-               if (is_e2fsprogs_feature_supp("-O mmp") == 0)
+               if (is_e2fsprogs_feature_supp("-O mmp"))
                        append_unique(anchor, ",", "mmp", NULL, maxbuflen);
                else
                        disp_old_e2fsprogs_msg("mmp", 1);
        }
 
        /* Allow more than 65000 subdirectories */
-       if (is_e2fsprogs_feature_supp("-O dir_nlink") == 0)
+       if (is_e2fsprogs_feature_supp("-O dir_nlink"))
                append_unique(anchor, ",", "dir_nlink", NULL, maxbuflen);
 
        /* The following options are only valid for ext4-based ldiskfs.
@@ -584,7 +585,7 @@ static int enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
                return 0;
 
        /* Enable quota by default */
-       if (is_e2fsprogs_feature_supp("-O quota") == 0) {
+       if (is_e2fsprogs_feature_supp("-O quota")) {
                append_unique(anchor, ",", "quota", NULL, maxbuflen);
        } else {
                fatal();
@@ -594,16 +595,30 @@ static int enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
        }
 
        /* Allow files larger than 2TB.  Also needs LU-16, but not harmful. */
-       if (is_e2fsprogs_feature_supp("-O huge_file") == 0)
+       if (is_e2fsprogs_feature_supp("-O huge_file"))
                append_unique(anchor, ",", "huge_file", NULL, maxbuflen);
 
        if (enable_64bit)
                append_unique(anchor, ",", "64bit", NULL, maxbuflen);
 
+       if (blocks >= 0x1000000000 && is_e2fsprogs_feature_supp("-O meta_bg"))
+               append_unique(anchor, ",", "meta_bg", NULL, maxbuflen);
+
+       if (enable_64bit || strstr(mop->mo_mkfsopts, "meta_bg"))
+               append_unique(anchor, ",", "^resize_inode", NULL, maxbuflen);
+
+       /* Allow xattrs larger than one block, stored in a separate inode */
+       if (IS_MDT(&mop->mo_ldd) && is_e2fsprogs_feature_supp("-O ea_inode"))
+               append_unique(anchor, ",", "ea_inode", NULL, maxbuflen);
+
+       /* Allow more than 10M directory entries */
+       if (IS_MDT(&mop->mo_ldd) && is_e2fsprogs_feature_supp("-O large_dir"))
+               append_unique(anchor, ",", "large_dir", NULL, maxbuflen);
+
        /* Cluster inode/block bitmaps and inode table for more efficient IO.
         * Align the flex groups on a 1MB boundary for better performance. */
        /* This -O feature needs to go last, since it adds the "-G" option. */
-       if (is_e2fsprogs_feature_supp("-O flex_bg") == 0) {
+       if (is_e2fsprogs_feature_supp("-O flex_bg")) {
                char tmp_buf[64];
 
                append_unique(anchor, ",", "flex_bg", NULL, maxbuflen);
@@ -611,7 +626,7 @@ static int enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
                if (IS_OST(&mop->mo_ldd) &&
                    strstr(mop->mo_mkfsopts, "-G") == NULL) {
                        snprintf(tmp_buf, sizeof(tmp_buf), " -G %u",
-                                (1 << 20) / L_BLOCK_SIZE);
+                                1024 / mop->mo_blocksize_kb);
                        strscat(anchor, tmp_buf, maxbuflen);
                }
        }
@@ -666,263 +681,297 @@ static char *moveopts_to_end(char *start)
 /* Build fs according to type */
 int ldiskfs_make_lustre(struct mkfs_opts *mop)
 {
-       __u64 device_kb = mop->mo_device_kb, block_count = 0;
        char mkfs_cmd[PATH_MAX];
        char buf[64];
        char *start;
        char *dev;
        int ret = 0, ext_opts = 0;
+       bool enable_64bit = false;
+       long inode_size = 0;
        size_t maxbuflen;
 
+       mop->mo_blocksize_kb = 4;
+
+       start = strstr(mop->mo_mkfsopts, "-b");
+       if (start) {
+               char *end = NULL;
+               long blocksize;
+
+               blocksize = strtol(start + 2, &end, 0);
+               if (end && (*end == 'k' || *end == 'K'))
+                       blocksize *= 1024;
+               /* EXT4_MIN_BLOCK_SIZE || EXT4_MAX_BLOCK_SIZE */
+               if (blocksize < 1024 || blocksize > 65536) {
+                       fprintf(stderr,
+                               "%s: blocksize %lu not in 1024-65536 bytes, normally 4096 bytes\n",
+                               progname, blocksize);
+                       return EINVAL;
+               }
+
+               if ((blocksize & (blocksize - 1)) != 0) {
+                       fprintf(stderr,
+                               "%s: blocksize %lu not a power-of-two value\n",
+                               progname, blocksize);
+                       return EINVAL;
+               }
+               mop->mo_blocksize_kb = blocksize >> 10;
+       }
+
        if (!(mop->mo_flags & MO_IS_LOOP)) {
-               mop->mo_device_kb = get_device_size(mop->mo_device);
+               __u64 device_kb = get_device_size(mop->mo_device);
 
-               if (mop->mo_device_kb == 0)
+               if (device_kb == 0)
                        return ENODEV;
 
                /* Compare to real size */
-               if (device_kb == 0 || device_kb > mop->mo_device_kb)
-                       device_kb = mop->mo_device_kb;
-               else
+               if (mop->mo_device_kb == 0 || device_kb < mop->mo_device_kb)
                        mop->mo_device_kb = device_kb;
        }
 
        if (mop->mo_device_kb != 0) {
+               __u64 block_count;
+
                if (mop->mo_device_kb < 32384) {
                        fprintf(stderr, "%s: size of filesystem must be larger "
                                "than 32MB, but is set to %lldKB\n",
                                progname, (long long)mop->mo_device_kb);
                        return EINVAL;
                }
-               block_count = mop->mo_device_kb / (L_BLOCK_SIZE >> 10);
-               /* If the LUN size is just over 2^32 blocks, limit the
-                * filesystem size to 2^32-1 blocks to avoid problems with
-                * ldiskfs/mkfs not handling this size.  Bug 22906 */
-               if (block_count > 0xffffffffULL && block_count < 0x100002000ULL)
-                       block_count = 0xffffffffULL;
+               block_count = mop->mo_device_kb / mop->mo_blocksize_kb;
+               if (block_count > 0xffffffffULL) {
+                       /* If the LUN size is just over 2^32 blocks, limit the
+                        * filesystem size to 2^32-1 blocks to avoid problems
+                        * with ldiskfs/mkfs not handling this well. b=22906
+                        */
+                       if (block_count < 0x100002000ULL)
+                               mop->mo_device_kb =
+                                       0xffffffffULL * mop->mo_blocksize_kb;
+                       else
+                               enable_64bit = true;
+               }
        }
 
-       if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
-           (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
-           (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
-               long inode_size = 0;
-
-               /* Journal size in MB */
-               if (strstr(mop->mo_mkfsopts, "-J") == NULL &&
-                   device_kb > 1024 * 1024) {
-                       /* Choose our own default journal size */
-                       long journal_mb = 0, max_mb;
-
-                       /* cap journal size at 4GB for MDT,
-                        * leave it at 400MB for OSTs. */
-                       if (IS_MDT(&mop->mo_ldd))
-                               max_mb = 4096;
-                       else if (IS_OST(&mop->mo_ldd))
-                               max_mb = 400;
-                       else /* Use mke2fs default size for MGS */
-                               max_mb = 0;
-
-                       /* Use at most 4% of device for journal */
-                       journal_mb = device_kb * 4 / (1024 * 100);
-                       if (journal_mb > max_mb)
-                               journal_mb = max_mb;
-
-                       if (journal_mb) {
-                               sprintf(buf, " -J size=%ld", journal_mb);
-                               strscat(mop->mo_mkfsopts, buf,
-                                       sizeof(mop->mo_mkfsopts));
-                       }
-               }
+       if ((mop->mo_ldd.ldd_mount_type != LDD_MT_EXT3) &&
+           (mop->mo_ldd.ldd_mount_type != LDD_MT_LDISKFS) &&
+           (mop->mo_ldd.ldd_mount_type != LDD_MT_LDISKFS2)) {
+               fprintf(stderr, "%s: unsupported fs type: %d (%s)\n",
+                       progname, mop->mo_ldd.ldd_mount_type,
+                       MT_STR(&mop->mo_ldd));
 
-               /*
-                * The inode size is constituted by following elements
-                * (assuming all files are in composite layout and has
-                * 3 components):
-                *
-                *   ldiskfs inode size: 160
-                *   MDT extended attributes size, including:
-                *      ext4_xattr_header: 32
-                *      LOV EA size: 32(lov_comp_md_v1) +
-                *                   3 * 40(lov_comp_md_entry_v1) +
-                *                   3 * 32(lov_mds_md) +
-                *                   stripes * 24(lov_ost_data) +
-                *                   16(xattr_entry) + 4("lov")
-                *      LMA EA size: 24(lustre_mdt_attrs) +
-                *                   16(xattr_entry) + 4("lma")
-                *      SOM EA size: 24(lustre_som_attrs) +
-                *                   16(xattr_entry) + 4("som")
-                *      link EA size: 24(link_ea_header) + 18(link_ea_entry) +
-                *                    16(filename) + 16(xattr_entry) + 4("link")
-                *   and some margin for 4-byte alignment, ACLs and other EAs.
-                *
-                * If we say the average filename length is about 32 bytes,
-                * the calculation looks like:
-                * 160 + 32 + (32+3*(40+32)+24*stripes+20) + (24+20) + (24+20) +
-                *  (24+20) + (~42+16+20) + other <= 512*2^m, {m=0,1,2,3}
-                */
-               if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
-                       if (IS_MDT(&mop->mo_ldd)) {
-                               if (mop->mo_stripe_count > 59)
-                                       inode_size = 512; /* bz 7241 */
-                               /* see also "-i" below for EA blocks */
-                               else if (mop->mo_stripe_count > 16)
-                                       inode_size = 2048;
-                               else
-                                       inode_size = 1024;
-                       } else if (IS_OST(&mop->mo_ldd)) {
-                               /* We store MDS FID and necessary composite
-                                * layout information in the OST object EA:
-                                *   ldiskfs inode size: 160
-                                *   OST extended attributes size, including:
-                                *      ext4_xattr_header: 32
-                                *      LMA EA size: 24(lustre_mdt_attrs) +
-                                *                   16(xattr_entry) + 4("lma")
-                                *      FID EA size: 52(filter_fid) +
-                                *                   16(xattr_entry) + 4("fid")
-                                * 160 + 32 + (24+20) + (52+20) = 308
-                                */
-                               inode_size = 512;
-                       }
+               return EINVAL;
+       }
 
-                       if (inode_size > 0) {
-                               sprintf(buf, " -I %ld", inode_size);
-                               strscat(mop->mo_mkfsopts, buf,
-                                       sizeof(mop->mo_mkfsopts));
-                       }
+       /* Journal size in MB */
+       if (strstr(mop->mo_mkfsopts, "-J") == NULL &&
+           mop->mo_device_kb > 1024 * 1024) {
+               /* Choose our own default journal size */
+               long journal_mb = 0, max_mb;
+
+               /* cap journal size at 4GB for MDT, leave at 1GB for OSTs */
+               if (IS_MDT(&mop->mo_ldd))
+                       max_mb = 4096;
+               else if (IS_OST(&mop->mo_ldd))
+                       max_mb = 1024;
+               else /* Use mke2fs default size for MGS */
+                       max_mb = 0;
+
+               /* Use at most 4% of device for journal */
+               journal_mb = mop->mo_device_kb * 4 / (1024 * 100);
+               if (journal_mb > max_mb)
+                       journal_mb = max_mb;
+
+               if (journal_mb) {
+                       snprintf(buf, sizeof(buf), " -J size=%ld", journal_mb);
+                       strscat(mop->mo_mkfsopts, buf,
+                               sizeof(mop->mo_mkfsopts));
                }
+       }
 
-               /* Bytes_per_inode: disk size / num inodes */
-               if (strstr(mop->mo_mkfsopts, "-i") == NULL &&
-                   strstr(mop->mo_mkfsopts, "-N") == NULL) {
-                       long bytes_per_inode = 0;
-
-                       /* Allocate more inodes on MDT devices.  There is
-                        * no data stored on the MDT, and very little extra
-                        * metadata beyond the inode.  It could go down as
-                        * low as 1024 bytes, but this is conservative.
-                        * Account for external EA blocks for wide striping. */
-                       if (IS_MDT(&mop->mo_ldd)) {
-                               bytes_per_inode = inode_size + 1536;
-
-                               if (mop->mo_stripe_count > 59) {
-                                       int extra = mop->mo_stripe_count * 24;
-                                       extra = ((extra - 1) | 4095) + 1;
-                                       bytes_per_inode += extra;
-                               }
-                       }
-
-                       /* Allocate fewer inodes on large OST devices.  Most
-                        * filesystems can be much more aggressive than even
-                        * this, but it is impossible to know in advance. */
-                       if (IS_OST(&mop->mo_ldd)) {
-                               /* OST > 16TB assume average file size 1MB */
-                               if (device_kb > (16ULL << 30))
-                                       bytes_per_inode = 1024 * 1024;
-                               /* OST > 4TB assume average file size 512kB */
-                               else if (device_kb > (4ULL << 30))
-                                       bytes_per_inode = 512 * 1024;
-                               /* OST > 1TB assume average file size 256kB */
-                               else if (device_kb > (1ULL << 30))
-                                       bytes_per_inode = 256 * 1024;
-                               /* OST > 10GB assume average file size 64kB,
-                                * plus a bit so that inodes will fit into a
-                                * 256x flex_bg without overflowing */
-                               else if (device_kb > (10ULL << 20))
-                                       bytes_per_inode = 69905;
-                       }
-
-                       if (bytes_per_inode > 0) {
-                               sprintf(buf, " -i %ld", bytes_per_inode);
-                               strscat(mop->mo_mkfsopts, buf,
-                                       sizeof(mop->mo_mkfsopts));
-                               mop->mo_inode_size = bytes_per_inode;
-                       }
+       /*
+        * The inode size is constituted by following elements
+        * (assuming all files are in composite layout and has
+        * 3 components):
+        *
+        *   ldiskfs inode size: 160
+        *   MDT extended attributes size, including:
+        *      ext4_xattr_header: 32
+        *      LOV EA size: 32(lov_comp_md_v1) +
+        *                   3 * 40(lov_comp_md_entry_v1) +
+        *                   3 * 32(lov_mds_md) +
+        *                   stripes * 24(lov_ost_data) +
+        *                   16(xattr_entry) + 4("lov")
+        *      LMA EA size: 24(lustre_mdt_attrs) +
+        *                   16(xattr_entry) + 4("lma")
+        *      SOM EA size: 24(lustre_som_attrs) +
+        *                   16(xattr_entry) + 4("som")
+        *      link EA size: 24(link_ea_header) + 18(link_ea_entry) +
+        *                    16(filename) + 16(xattr_entry) + 4("link")
+        *   and some margin for 4-byte alignment, ACLs and other EAs.
+        *
+        * If we say the average filename length is about 32 bytes,
+        * the calculation looks like:
+        * 160 + 32 + (32+3*(40+32)+24*stripes+20) + (24+20) + (24+20) +
+        *  (24+20) + (~42+16+20) + other <= 512*2^m, {m=0,1,2,3}
+        */
+       if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
+               if (IS_MDT(&mop->mo_ldd)) {
+                       if (mop->mo_stripe_count > 59)
+                               inode_size = 512; /* bz 7241 */
+                       /* see also "-i" below for EA blocks */
+                       else if (mop->mo_stripe_count > 16)
+                               inode_size = 2048;
+                       else
+                               inode_size = 1024;
+               } else if (IS_OST(&mop->mo_ldd)) {
+                       /* We store MDS FID and necessary composite
+                        * layout information in the OST object EA:
+                        *   ldiskfs inode size: 160
+                        *   OST extended attributes size, including:
+                        *      ext4_xattr_header: 32
+                        *      LMA EA size: 24(lustre_mdt_attrs) +
+                        *                   16(xattr_entry) + 4("lma")
+                        *      FID EA size: 52(filter_fid) +
+                        *                   16(xattr_entry) + 4("fid")
+                        * 160 + 32 + (24+20) + (52+20) = 308
+                        */
+                       inode_size = 512;
                }
 
-               if (verbose < 2) {
-                       strscat(mop->mo_mkfsopts, " -q",
+               if (inode_size > 0) {
+                       snprintf(buf, sizeof(buf), " -I %ld", inode_size);
+                       strscat(mop->mo_mkfsopts, buf,
                                sizeof(mop->mo_mkfsopts));
                }
+       }
 
-               /* start handle -O mkfs options */
-               if ((start = strstr(mop->mo_mkfsopts, "-O")) != NULL) {
-                       if (strstr(start + 2, "-O") != NULL) {
-                               fprintf(stderr,
-                                       "%s: don't specify multiple -O options\n",
-                                       progname);
-                               return EINVAL;
+       /* Bytes_per_inode: disk size / num inodes */
+       if (strstr(mop->mo_mkfsopts, "-i") == NULL &&
+           strstr(mop->mo_mkfsopts, "-N") == NULL) {
+               long bytes_per_inode = 0;
+
+               /* Allocate more inodes on MDT devices.  There is
+                * no data stored on the MDT, and very little extra
+                * metadata beyond the inode.  It could go down as
+                * low as 1024 bytes, but this is conservative.
+                * Account for external EA blocks for wide striping.
+                */
+               if (IS_MDT(&mop->mo_ldd)) {
+                       bytes_per_inode = inode_size + 1536;
+
+                       if (mop->mo_stripe_count > 59) {
+                               int extra = mop->mo_stripe_count * 24;
+
+                               extra = ((extra - 1) | 4095) + 1;
+                               bytes_per_inode += extra;
                        }
-                       start = moveopts_to_end(start);
-                       maxbuflen = sizeof(mop->mo_mkfsopts) -
-                               (start - mop->mo_mkfsopts) - strlen(start);
-                       ret = enable_default_ext4_features(mop, start, maxbuflen, 1);
-               } else {
-                       start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts),
-                             maxbuflen = sizeof(mop->mo_mkfsopts) -
-                                     strlen(mop->mo_mkfsopts);
-                       ret = enable_default_ext4_features(mop, start, maxbuflen, 0);
                }
-               if (ret)
-                       return ret;
-               /* end handle -O mkfs options */
 
-               /* start handle -E mkfs options */
-               if ((start = strstr(mop->mo_mkfsopts, "-E")) != NULL) {
-                       if (strstr(start + 2, "-E") != NULL) {
-                               fprintf(stderr,
-                                       "%s: don't specify multiple -E options\n",
-                                       progname);
-                               return EINVAL;
-                       }
-                       start = moveopts_to_end(start);
-                       maxbuflen = sizeof(mop->mo_mkfsopts) -
-                               (start - mop->mo_mkfsopts) - strlen(start);
-                       ext_opts = 1;
-               } else {
-                       start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
-                       maxbuflen = sizeof(mop->mo_mkfsopts) -
-                               strlen(mop->mo_mkfsopts);
+               /* Allocate fewer inodes on large OST devices.  Most
+                * filesystems can be much more aggressive than even
+                * this, but it is impossible to know in advance.
+                */
+               if (IS_OST(&mop->mo_ldd)) {
+                       /* OST > 16TB assume average file size 1MB */
+                       if (mop->mo_device_kb > (16ULL << 30))
+                               bytes_per_inode = 1024 * 1024;
+                       /* OST > 4TB assume average file size 512kB */
+                       else if (mop->mo_device_kb > (4ULL << 30))
+                               bytes_per_inode = 512 * 1024;
+                       /* OST > 1TB assume average file size 256kB */
+                       else if (mop->mo_device_kb > (1ULL << 30))
+                               bytes_per_inode = 256 * 1024;
+                       /* OST > 10GB assume average file size 64kB,
+                        * plus a bit so that inodes will fit into a
+                        * 256x flex_bg without overflowing.
+                        */
+                       else if (mop->mo_device_kb > (10ULL << 20))
+                               bytes_per_inode = 69905;
                }
 
-               /* In order to align the filesystem metadata on 1MB boundaries,
-                * give a resize value that will reserve a power-of-two group
-                * descriptor blocks, but leave one block for the superblock.
-                * Only useful for filesystems with < 2^32 blocks due to resize
-                * limitations. */
-               if (strstr(mop->mo_mkfsopts, "meta_bg") == NULL &&
-                   IS_OST(&mop->mo_ldd) && mop->mo_device_kb > 100 * 1024 &&
-                   mop->mo_device_kb * 1024 / L_BLOCK_SIZE <= 0xffffffffULL) {
-                       unsigned group_blocks = L_BLOCK_SIZE * 8;
-                       unsigned desc_per_block = L_BLOCK_SIZE / 32;
-                       unsigned resize_blks;
-
-                       resize_blks = (1ULL<<32) - desc_per_block*group_blocks;
-                       snprintf(buf, sizeof(buf), "%u", resize_blks);
-                       append_unique(start, ext_opts ? "," : " -E ",
-                                     "resize", buf, maxbuflen);
-                       ext_opts = 1;
+               if (bytes_per_inode > 0) {
+                       snprintf(buf, sizeof(buf), " -i %ld", bytes_per_inode);
+                       strscat(mop->mo_mkfsopts, buf,
+                               sizeof(mop->mo_mkfsopts));
+                       mop->mo_inode_size = bytes_per_inode;
                }
+       }
 
-               /* Avoid zeroing out the full journal - speeds up mkfs */
-               if (is_e2fsprogs_feature_supp("-E lazy_journal_init") == 0)
-                       append_unique(start, ext_opts ? "," : " -E ",
-                                     "lazy_journal_init", NULL, maxbuflen);
-               /* end handle -E mkfs options */
+       if (verbose < 2)
+               strscat(mop->mo_mkfsopts, " -q", sizeof(mop->mo_mkfsopts));
 
-               /* Allow reformat of full devices (as opposed to
-                  partitions.)  We already checked for mounted dev. */
-               strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
+       /* start handle -O mkfs options */
+       start = strstr(mop->mo_mkfsopts, "-O");
+       if (start) {
+               if (strstr(start + 2, "-O") != NULL) {
+                       fprintf(stderr,
+                               "%s: don't specify multiple -O options\n",
+                               progname);
+                       return EINVAL;
+               }
+               start = moveopts_to_end(start);
+               maxbuflen = sizeof(mop->mo_mkfsopts) -
+                       (start - mop->mo_mkfsopts) - strlen(start);
+               ret = enable_default_ext4_features(mop, start, maxbuflen, 1);
+       } else {
+               start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
+               maxbuflen = sizeof(mop->mo_mkfsopts) - strlen(mop->mo_mkfsopts);
+               ret = enable_default_ext4_features(mop, start, maxbuflen, 0);
+       }
+       if (ret)
+               return ret;
+       /* end handle -O mkfs options */
 
-               snprintf(mkfs_cmd, sizeof(mkfs_cmd),
-                        "%s -j -b %d -L %s ", MKE2FS, L_BLOCK_SIZE,
-                        mop->mo_ldd.ldd_svname);
+       /* start handle -E mkfs options */
+       start = strstr(mop->mo_mkfsopts, "-E");
+       if (start) {
+               if (strstr(start + 2, "-E") != NULL) {
+                       fprintf(stderr,
+                               "%s: don't specify multiple -E options\n",
+                               progname);
+                       return EINVAL;
+               }
+               start = moveopts_to_end(start);
+               maxbuflen = sizeof(mop->mo_mkfsopts) -
+                       (start - mop->mo_mkfsopts) - strlen(start);
+               ext_opts = 1;
        } else {
-               fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
-                       progname, mop->mo_ldd.ldd_mount_type,
-                       MT_STR(&mop->mo_ldd));
-               return EINVAL;
+               start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
+               maxbuflen = sizeof(mop->mo_mkfsopts) - strlen(mop->mo_mkfsopts);
+       }
+
+       /* In order to align the filesystem metadata on 1MB boundaries,
+        * give a resize value that will reserve a power-of-two group
+        * descriptor blocks, but leave one block for the superblock.
+        * Only useful for filesystems with < 2^32 blocks due to resize
+        * limitations.
+        */
+       if (!enable_64bit && strstr(mop->mo_mkfsopts, "meta_bg") == NULL &&
+           IS_OST(&mop->mo_ldd) && mop->mo_device_kb > 100 * 1024) {
+               unsigned int group_blocks = mop->mo_blocksize_kb * 8192;
+               unsigned int desc_per_block = mop->mo_blocksize_kb * 1024 / 32;
+               unsigned int resize_blks;
+
+               resize_blks = (1ULL<<32) - desc_per_block*group_blocks;
+               snprintf(buf, sizeof(buf), "%u", resize_blks);
+               append_unique(start, ext_opts ? "," : " -E ",
+                             "resize", buf, maxbuflen);
+               ext_opts = 1;
        }
 
+       /* Avoid zeroing out the full journal - speeds up mkfs */
+       if (is_e2fsprogs_feature_supp("-E lazy_journal_init"))
+               append_unique(start, ext_opts ? "," : " -E ",
+                             "lazy_journal_init", NULL, maxbuflen);
+       /* end handle -E mkfs options */
+
+       /* Allow reformat of full devices (as opposed to partitions).
+        * We already checked for mounted dev.
+        */
+       strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
+
+       snprintf(mkfs_cmd, sizeof(mkfs_cmd), "%s -j -b %d -L %s ", MKE2FS,
+                mop->mo_blocksize_kb * 1024, mop->mo_ldd.ldd_svname);
+
        /* For loop device format the dev, not the filename */
        dev = mop->mo_device;
        if (mop->mo_flags & MO_IS_LOOP)
@@ -931,16 +980,16 @@ int ldiskfs_make_lustre(struct mkfs_opts *mop)
        vprint("formatting backing filesystem %s on %s\n",
               MT_STR(&mop->mo_ldd), dev);
        vprint("\ttarget name   %s\n", mop->mo_ldd.ldd_svname);
-       vprint("\t4k blocks     %ju\n", (uintmax_t)block_count);
+       vprint("\tkilobytes     %llu\n", mop->mo_device_kb);
        vprint("\toptions       %s\n", mop->mo_mkfsopts);
 
        /* mkfs_cmd's trailing space is important! */
        strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
        strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
        strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
-       if (block_count != 0) {
-               snprintf(buf, sizeof(buf), " %ju",
-                        (uintmax_t)block_count);
+       if (mop->mo_device_kb != 0) {
+               snprintf(buf, sizeof(buf), " %lluk",
+                        (unsigned long long)mop->mo_device_kb);
                strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
        }
 
@@ -1173,7 +1222,9 @@ static int tune_block_dev_scheduler(const char *sys_path, const char *new_sched)
        *e = '\0';
 
        if (strcmp(old_sched, "noop") == 0 ||
-           strcmp(old_sched, new_sched) == 0)
+           strcmp(old_sched, "deadline") == 0 ||
+           strcmp(old_sched, "mq-deadline") == 0 ||
+           strstr(old_sched, new_sched) == 0)
                return 0;
 
        rc = write_file(path, new_sched);
@@ -1210,13 +1261,13 @@ static int tune_block_dev_slaves(const char *sys_path, struct mount_opts *mop)
        }
 
        while ((d = readdir(slaves_dir)) != NULL) {
-               char path[PATH_MAX];
+               char path[PATH_MAX * 2];
                int rc2;
 
                if (d->d_type != DT_LNK)
                        continue;
 
-               snprintf(path, sizeof(path), "%s/%s", slaves_path, d->d_name);
+               snprintf(path, sizeof(path), "/dev/%s", d->d_name);
                rc2 = tune_block_dev(path, mop);
                if (rc2 != 0)
                        rc = rc2;
@@ -1234,7 +1285,7 @@ static int tune_block_dev(const char *src, struct mount_opts *mop)
 {
        struct stat st;
        char sys_path[PATH_MAX];
-       char partition_path[PATH_MAX];
+       char partition_path[PATH_MAX + sizeof("partition")];
        char *real_sys_path = NULL;
        int rc;
 
@@ -1404,7 +1455,7 @@ int ldiskfs_enable_quota(struct mkfs_opts *mop)
        char cmd[512];
        int cmdsz = sizeof(cmd), ret;
 
-       if (is_e2fsprogs_feature_supp("-O quota") != 0) {
+       if (!is_e2fsprogs_feature_supp("-O quota")) {
                fprintf(stderr, "%s: \"-O quota\" is is not supported by "
                        "current e2fsprogs\n", progname);
                return EINVAL;