]> git.ipfire.org Git - thirdparty/xfsprogs-dev.git/commitdiff
mkfs: finobt mkfs support
authorBrian Foster <bfoster@redhat.com>
Mon, 26 May 2014 23:06:17 +0000 (09:06 +1000)
committerDave Chinner <david@fromorbit.com>
Mon, 26 May 2014 23:06:17 +0000 (09:06 +1000)
Add the 'finobt' metadata option to mkfs to format an fs with free
inode btree support. If enabled, initialize the associated AGI
header fields and btree root block.

Also, do the initialization of the superblock version and feature
bits (including the new finobt flag) a bit earlier. These fields
must now be initialized prior to the use of XFS_PREALLOC_BLOCKS(),
as the latter returns a value that depends on whether a finobt root
btree block is reserved.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
mkfs/xfs_mkfs.c

index 37c05a98f37a9be2d97d5376cda5dfdb6589eded..188b6b3bfc012c95e586f384a8ef9132901733f8 100644 (file)
@@ -183,6 +183,8 @@ char        *sopts[] = {
 char   *mopts[] = {
 #define        M_CRC           0
        "crc",
+#define M_FINOBT       1
+       "finobt",
        NULL
 };
 
@@ -962,6 +964,7 @@ main(
        struct fs_topology      ft;
        int                     lazy_sb_counters;
        int                     crcs_enabled;
+       int                     finobt;
 
        progname = basename(argv[0]);
        setlocale(LC_ALL, "");
@@ -995,6 +998,7 @@ main(
        worst_freelist = 0;
        lazy_sb_counters = 1;
        crcs_enabled = 0;
+       finobt = 0;
        memset(&fsx, 0, sizeof(fsx));
 
        memset(&xi, 0, sizeof(xi));
@@ -1486,6 +1490,14 @@ _("cannot specify both crc and ftype\n"));
                                                usage();
                                        }
                                        break;
+                               case M_FINOBT:
+                                       if (!value || *value == '\0')
+                                               reqval('m', mopts, M_CRC);
+                                       c = atoi(value);
+                                       if (c < 0 || c > 1)
+                                               illegal(value, "m finobt");
+                                       finobt = c;
+                                       break;
                                default:
                                        unknown('m', value);
                                }
@@ -1827,6 +1839,16 @@ _("32 bit Project IDs always enabled on CRC enabled filesytems\n"));
                }
        }
 
+       /*
+        * The kernel doesn't currently support crc=0,finobt=1 filesystems.
+        * Catch it here, disable finobt and warn the user.
+        */
+       if (finobt && !crcs_enabled) {
+               fprintf(stderr,
+_("warning: finobt not supported without CRC support, disabled.\n"));
+               finobt = 0;
+       }
+
        if (nsflag || nlflag) {
                if (dirblocksize < blocksize ||
                                        dirblocksize > XFS_MAX_BLOCKSIZE) {
@@ -2413,6 +2435,30 @@ _("size %s specified for log subvolume is too large, maximum is %lld blocks\n"),
        mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
        mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
 
+       /*
+        * sb_versionnum and finobt flags must be set before we use
+        * XFS_PREALLOC_BLOCKS().
+        */
+       sbp->sb_features2 = XFS_SB_VERSION2_MKFS(crcs_enabled, lazy_sb_counters,
+                                       attrversion == 2, !projid16bit, 0,
+                                       (!crcs_enabled && dirftype));
+       sbp->sb_versionnum = XFS_SB_VERSION_MKFS(crcs_enabled, iaflag,
+                                       dsunit != 0,
+                                       logversion == 2, attrversion == 1,
+                                       (sectorsize != BBSIZE ||
+                                                       lsectorsize != BBSIZE),
+                                       nci, sbp->sb_features2 != 0);
+       /*
+        * Due to a structure alignment issue, sb_features2 ended up in one
+        * of two locations, the second "incorrect" location represented by
+        * the sb_bad_features2 field. To avoid older kernels mounting
+        * filesystems they shouldn't, set both field to the same value.
+        */
+       sbp->sb_bad_features2 = sbp->sb_features2;
+
+       if (finobt)
+               sbp->sb_features_ro_compat = XFS_SB_FEAT_RO_COMPAT_FINOBT;
+
        if (loginternal) {
                /*
                 * Readjust the log size to fit within an AG if it was sized
@@ -2475,7 +2521,7 @@ _("size %s specified for log subvolume is too large, maximum is %lld blocks\n"),
                printf(_(
                   "meta-data=%-22s isize=%-6d agcount=%lld, agsize=%lld blks\n"
                   "         =%-22s sectsz=%-5u attr=%u, projid32bit=%u\n"
-                  "         =%-22s crc=%u\n"
+                  "         =%-22s crc=%-8u finobt=%u\n"
                   "data     =%-22s bsize=%-6u blocks=%llu, imaxpct=%u\n"
                   "         =%-22s sunit=%-6u swidth=%u blks\n"
                   "naming   =version %-14u bsize=%-6u ascii-ci=%d ftype=%d\n"
@@ -2484,7 +2530,7 @@ _("size %s specified for log subvolume is too large, maximum is %lld blocks\n"),
                   "realtime =%-22s extsz=%-6d blocks=%lld, rtextents=%lld\n"),
                        dfile, isize, (long long)agcount, (long long)agsize,
                        "", sectorsize, attrversion, !projid16bit,
-                       "", crcs_enabled,
+                       "", crcs_enabled, finobt,
                        "", blocksize, (long long)dblocks, imaxpct,
                        "", dsunit, dswidth,
                        dirversion, dirblocksize, nci, dirftype,
@@ -2553,23 +2599,6 @@ _("size %s specified for log subvolume is too large, maximum is %lld blocks\n"),
                sbp->sb_logsectsize = 0;
        }
 
-       sbp->sb_features2 = XFS_SB_VERSION2_MKFS(crcs_enabled, lazy_sb_counters,
-                                  attrversion == 2, !projid16bit, 0,
-                                  (!crcs_enabled && dirftype));
-       sbp->sb_versionnum = XFS_SB_VERSION_MKFS(crcs_enabled, iaflag,
-                                       dsunit != 0,
-                                       logversion == 2, attrversion == 1,
-                                       (sectorsize != BBSIZE ||
-                                                       lsectorsize != BBSIZE),
-                                       nci, sbp->sb_features2 != 0);
-       /*
-        * Due to a structure alignment issue, sb_features2 ended up in one
-        * of two locations, the second "incorrect" location represented by
-        * the sb_bad_features2 field. To avoid older kernels mounting
-        * filesystems they shouldn't, set both field to the same value.
-        */
-       sbp->sb_bad_features2 = sbp->sb_features2;
-
        if (force_overwrite)
                zero_old_xfs_structures(&xi, sbp);
 
@@ -2726,6 +2755,10 @@ _("size %s specified for log subvolume is too large, maximum is %lld blocks\n"),
                agi->agi_count = 0;
                agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
                agi->agi_level = cpu_to_be32(1);
+               if (finobt) {
+                       agi->agi_free_root = cpu_to_be32(XFS_FIBT_BLOCK(mp));
+                       agi->agi_free_level = cpu_to_be32(1);
+               }
                agi->agi_freecount = 0;
                agi->agi_newino = cpu_to_be32(NULLAGINO);
                agi->agi_dirino = cpu_to_be32(NULLAGINO);
@@ -2851,6 +2884,26 @@ _("size %s specified for log subvolume is too large, maximum is %lld blocks\n"),
                        xfs_btree_init_block(mp, buf, XFS_IBT_MAGIC, 0, 0,
                                                agno, 0);
                libxfs_writebuf(buf, LIBXFS_EXIT_ON_FAILURE);
+
+               /*
+                * Free INO btree root block
+                */
+               if (!finobt)
+                       continue;
+
+               buf = libxfs_getbuf(mp->m_ddev_targp,
+                               XFS_AGB_TO_DADDR(mp, agno, XFS_FIBT_BLOCK(mp)),
+                               bsize);
+               buf->b_ops = &xfs_inobt_buf_ops;
+               block = XFS_BUF_TO_BLOCK(buf);
+               memset(block, 0, blocksize);
+               if (xfs_sb_version_hascrc(&mp->m_sb))
+                       xfs_btree_init_block(mp, buf, XFS_FIBT_CRC_MAGIC, 0, 0,
+                                               agno, XFS_BTREE_CRC_BLOCKS);
+               else
+                       xfs_btree_init_block(mp, buf, XFS_FIBT_MAGIC, 0, 0,
+                                               agno, 0);
+               libxfs_writebuf(buf, LIBXFS_EXIT_ON_FAILURE);
        }
 
        /*
@@ -3087,7 +3140,7 @@ usage( void )
 {
        fprintf(stderr, _("Usage: %s\n\
 /* blocksize */                [-b log=n|size=num]\n\
-/* metadata */         [-m crc=[0|1]\n\
+/* metadata */         [-m crc=0|1,finobt=0|1]\n\
 /* data subvol */      [-d agcount=n,agsize=n,file,name=xxx,size=num,\n\
                            (sunit=value,swidth=value|su=num,sw=num|noalign),\n\
                            sectlog=n|sectsize=num\n\