kernel - Spread IPIs out to more priority levels

[dragonfly.git] / sbin / newfs_hammer2 / newfs_hammer2.c

/*
 * Copyright (c) 2011-2015 The DragonFly Project.  All rights reserved.
 *
 * This code is derived from software contributed to The DragonFly Project
 * by Matthew Dillon <dillon@dragonflybsd.org>
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 *
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in
 *    the documentation and/or other materials provided with the
 *    distribution.
 * 3. Neither the name of The DragonFly Project nor the names of its
 *    contributors may be used to endorse or promote products derived
 *    from this software without specific, prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 */

#include <sys/types.h>
#include <sys/diskslice.h>
#include <sys/diskmbr.h>
#include <sys/stat.h>
#include <sys/time.h>
#include <sys/sysctl.h>
#include <vfs/hammer2/hammer2_xxhash.h>
#include <vfs/hammer2/hammer2_disk.h>

#include <stdio.h>
#include <stdlib.h>
#include <stdarg.h>
#include <stddef.h>
#include <unistd.h>
#include <string.h>
#include <errno.h>
#include <fcntl.h>
#include <assert.h>
#include <err.h>
#include <uuid.h>

#define MAXLABELS       HAMMER2_SET_COUNT

#define hammer2_icrc32(buf, size)       iscsi_crc32((buf), (size))
#define hammer2_icrc32c(buf, size, crc) iscsi_crc32_ext((buf), (size), (crc))
uint32_t iscsi_crc32(const void *buf, size_t size);
uint32_t iscsi_crc32_ext(const void *buf, size_t size, uint32_t ocrc);

static hammer2_off_t check_volume(const char *path, int *fdp);
static int64_t getsize(const char *str, int64_t minval, int64_t maxval, int pw);
static const char *sizetostr(hammer2_off_t size);
static uint64_t nowtime(void);
static int blkrefary_cmp(const void *b1, const void *b2);
static void usage(void);

static void format_hammer2(int fd, hammer2_off_t total_space,
                                hammer2_off_t free_space);
static void alloc_direct(hammer2_off_t *basep, hammer2_blockref_t *bref,
                                size_t bytes);
static hammer2_key_t dirhash(const unsigned char *name, size_t len);

static int Hammer2Version = -1;
static int ForceOpt = 0;
static uuid_t Hammer2_FSType;   /* static filesystem type id for HAMMER2 */
static uuid_t Hammer2_VolFSID;  /* unique filesystem id in volu header */
static uuid_t Hammer2_SupCLID;  /* PFS cluster id in super-root inode */
static uuid_t Hammer2_SupFSID;  /* PFS unique id in super-root inode */
static uuid_t Hammer2_PfsCLID[MAXLABELS];
static uuid_t Hammer2_PfsFSID[MAXLABELS];
static const char *Label[MAXLABELS];
static hammer2_off_t BootAreaSize;
static hammer2_off_t AuxAreaSize;
static int NLabels;

#define GIG     ((hammer2_off_t)1024*1024*1024)

int
main(int ac, char **av)
{
        uint32_t status;
        hammer2_off_t total_space;
        hammer2_off_t free_space;
        hammer2_off_t reserved_space;
        int ch;
        int fd = -1;
        int i;
        int defaultlabels = 1;
        char *vol_fsid;
        char *sup_clid_name;
        char *sup_fsid_name;
        char *pfs_clid_name;
        char *pfs_fsid_name;

        Label[NLabels++] = "LOCAL";

        /*
         * Sanity check basic filesystem structures.  No cookies for us
         * if it gets broken!
         */
        assert(sizeof(hammer2_volume_data_t) == HAMMER2_VOLUME_BYTES);
        assert(sizeof(hammer2_inode_data_t) == HAMMER2_INODE_BYTES);
        assert(sizeof(hammer2_blockref_t) == HAMMER2_BLOCKREF_BYTES);

        /*
         * Generate a filesystem id and lookup the filesystem type
         */
        srandomdev();
        uuidgen(&Hammer2_VolFSID, 1);
        uuidgen(&Hammer2_SupCLID, 1);
        uuidgen(&Hammer2_SupFSID, 1);
        uuid_from_string(HAMMER2_UUID_STRING, &Hammer2_FSType, &status);
        /*uuid_name_lookup(&Hammer2_FSType, "DragonFly HAMMER2", &status);*/
        if (status != uuid_s_ok) {
                errx(1, "uuids file does not have the DragonFly "
                        "HAMMER2 filesystem type");
        }

        /*
         * Parse arguments
         */
        while ((ch = getopt(ac, av, "fL:b:m:r:V:")) != -1) {
                switch(ch) {
                case 'f':
                        ForceOpt = 1;
                        break;
                case 'L':
                        defaultlabels = 0;
                        if (strcasecmp(optarg, "none") == 0) {
                                break;
                        }
                        if (NLabels >= MAXLABELS) {
                                errx(1,
                                     "Limit of %d local labels",
                                     MAXLABELS - 1);
                        }
                        Label[NLabels++] = optarg;
                        if (strlen(Label[NLabels-1]) > HAMMER2_INODE_MAXNAME) {
                                errx(1, "Root directory label too long "
                                        "(64 chars max)\n");
                        }
                        break;
                case 'b':
                        BootAreaSize = getsize(optarg,
                                         HAMMER2_NEWFS_ALIGN,
                                         HAMMER2_BOOT_MAX_BYTES, 2);
                        break;
                case 'r':
                        AuxAreaSize = getsize(optarg,
                                         HAMMER2_NEWFS_ALIGN,
                                         HAMMER2_REDO_MAX_BYTES, 2);
                        break;
                case 'V':
                        Hammer2Version = strtol(optarg, NULL, 0);
                        if (Hammer2Version < HAMMER2_VOL_VERSION_MIN ||
                            Hammer2Version >= HAMMER2_VOL_VERSION_WIP) {
                                errx(1,
                                     "I don't understand how to format "
                                     "HAMMER2 version %d\n",
                                     Hammer2Version);
                        }
                        break;
                default:
                        usage();
                        break;
                }
        }

        /*
         * Check Hammer2 version
         */
        if (Hammer2Version < 0) {
                size_t olen = sizeof(Hammer2Version);
                Hammer2Version = HAMMER2_VOL_VERSION_DEFAULT;
                if (sysctlbyname("vfs.hammer2.supported_version",
                                 &Hammer2Version, &olen, NULL, 0) == 0) {
                        if (Hammer2Version >= HAMMER2_VOL_VERSION_WIP) {
                                Hammer2Version = HAMMER2_VOL_VERSION_WIP - 1;
                                fprintf(stderr,
                                        "newfs_hammer2: WARNING: HAMMER2 VFS "
                                        "supports higher version than I "
                                        "understand,\n"
                                        "using version %d\n",
                                        Hammer2Version);
                        }
                } else {
                        fprintf(stderr,
                                "newfs_hammer2: WARNING: HAMMER2 VFS not "
                                "loaded, cannot get version info.\n"
                                "Using version %d\n",
                                HAMMER2_VOL_VERSION_DEFAULT);
                }
        }

        ac -= optind;
        av += optind;

        if (ac != 1 || av[0][0] == 0) {
                fprintf(stderr, "Exactly one disk device must be specified\n");
                exit(1);
        }

        /*
         * Adjust Label[] and NLabels.
         */
        if (defaultlabels == 0) {
                NLabels = 1;
        } else {
                char c = av[0][strlen(av[0]) - 1];
                if (c == 'a')
                        Label[NLabels++] = "BOOT";
                else if (c == 'd')
                        Label[NLabels++] = "ROOT";
                else
                        Label[NLabels++] = "DATA";
        }

        /*
         * Collect volume information.
         */
        total_space = check_volume(av[0], &fd);

        /*
         * ~typically 8MB alignment to avoid edge cases for reserved blocks
         * and so raid stripes (if any) operate efficiently.
         */
        total_space &= ~HAMMER2_VOLUME_ALIGNMASK64;

        /*
         * Calculate defaults for the boot area size and round to the
         * volume alignment boundary.
         *
         * NOTE: These areas are currently not used for booting but are
         *       reserved for future filesystem expansion.
         */
        if (BootAreaSize == 0) {
                BootAreaSize = HAMMER2_BOOT_NOM_BYTES;
                while (BootAreaSize > total_space / 20)
                        BootAreaSize >>= 1;
                if (BootAreaSize < HAMMER2_BOOT_MIN_BYTES)
                        BootAreaSize = HAMMER2_BOOT_MIN_BYTES;
        } else if (BootAreaSize < HAMMER2_BOOT_MIN_BYTES) {
                BootAreaSize = HAMMER2_BOOT_MIN_BYTES;
        }
        BootAreaSize = (BootAreaSize + HAMMER2_VOLUME_ALIGNMASK64) &
                       ~HAMMER2_VOLUME_ALIGNMASK64;

        /*
         * Calculate defaults for the redo area size and round to the
         * volume alignment boundary.
         *
         * NOTE: These areas are currently not used for logging but are
         *       reserved for future filesystem expansion.
         */
        if (AuxAreaSize == 0) {
                AuxAreaSize = HAMMER2_REDO_NOM_BYTES;
                while (AuxAreaSize > total_space / 20)
                        AuxAreaSize >>= 1;
                if (AuxAreaSize < HAMMER2_REDO_MIN_BYTES)
                        AuxAreaSize = HAMMER2_REDO_MIN_BYTES;
        } else if (AuxAreaSize < HAMMER2_REDO_MIN_BYTES) {
                AuxAreaSize = HAMMER2_REDO_MIN_BYTES;
        }
        AuxAreaSize = (AuxAreaSize + HAMMER2_VOLUME_ALIGNMASK64) &
                       ~HAMMER2_VOLUME_ALIGNMASK64;

        /*
         * We'll need to stuff this in the volume header soon.
         */
        uuid_to_string(&Hammer2_VolFSID, &vol_fsid, &status);
        uuid_to_string(&Hammer2_SupCLID, &sup_clid_name, &status);
        uuid_to_string(&Hammer2_SupFSID, &sup_fsid_name, &status);

        /*
         * Calculate the amount of reserved space.  HAMMER2_ZONE_SEG (4MB)
         * is reserved at the beginning of every 2GB of storage, rounded up.
         * Thus a 200MB filesystem will still have a 4MB reserve area.
         *
         * We also include the boot and redo areas in the reserve.  The
         * reserve is used to help 'df' calculate the amount of available
         * space.
         */
        reserved_space = ((total_space + HAMMER2_ZONE_MASK64) /
                          HAMMER2_ZONE_BYTES64) * HAMMER2_ZONE_SEG64;

        free_space = total_space - reserved_space -
                     BootAreaSize - AuxAreaSize;

        format_hammer2(fd, total_space, free_space);
        fsync(fd);
        close(fd);

        printf("---------------------------------------------\n");
        printf("version:          %d\n", Hammer2Version);
        printf("total-size:       %s (%jd bytes)\n",
               sizetostr(total_space),
               (intmax_t)total_space);
        printf("boot-area-size:   %s\n", sizetostr(BootAreaSize));
        printf("aux-area-size:    %s\n", sizetostr(AuxAreaSize));
        printf("topo-reserved:    %s\n", sizetostr(reserved_space));
        printf("free-space:       %s\n", sizetostr(free_space));
        printf("vol-fsid:         %s\n", vol_fsid);
        printf("sup-clid:         %s\n", sup_clid_name);
        printf("sup-fsid:         %s\n", sup_fsid_name);
        for (i = 0; i < NLabels; ++i) {
                printf("PFS \"%s\"\n", Label[i]);
                uuid_to_string(&Hammer2_PfsCLID[i], &pfs_clid_name, &status);
                uuid_to_string(&Hammer2_PfsFSID[i], &pfs_fsid_name, &status);
                printf("    clid %s\n", pfs_clid_name);
                printf("    fsid %s\n", pfs_fsid_name);
        }
        printf("\n");

        return(0);
}

static
void
usage(void)
{
        fprintf(stderr,
                "usage: newfs_hammer2 -L label [-f] [-b bootsize] "
                "[-r redosize] [-V version] special ...\n"
        );
        exit(1);
}

/*
 * Convert the size in bytes to a human readable string.
 */
static
const char *
sizetostr(hammer2_off_t size)
{
        static char buf[32];

        if (size < 1024 / 2) {
                snprintf(buf, sizeof(buf), "%6.2f", (double)size);
        } else if (size < 1024 * 1024 / 2) {
                snprintf(buf, sizeof(buf), "%6.2fKB",
                        (double)size / 1024);
        } else if (size < 1024 * 1024 * 1024LL / 2) {
                snprintf(buf, sizeof(buf), "%6.2fMB",
                        (double)size / (1024 * 1024));
        } else if (size < 1024 * 1024 * 1024LL * 1024LL / 2) {
                snprintf(buf, sizeof(buf), "%6.2fGB",
                        (double)size / (1024 * 1024 * 1024LL));
        } else {
                snprintf(buf, sizeof(buf), "%6.2fTB",
                        (double)size / (1024 * 1024 * 1024LL * 1024LL));
        }
        return(buf);
}

/*
 * Convert a string to a 64 bit signed integer with various requirements.
 */
static int64_t
getsize(const char *str, int64_t minval, int64_t maxval, int powerof2)
{
        int64_t val;
        char *ptr;

        val = strtoll(str, &ptr, 0);
        switch(*ptr) {
        case 't':
        case 'T':
                val *= 1024;
                /* fall through */
        case 'g':
        case 'G':
                val *= 1024;
                /* fall through */
        case 'm':
        case 'M':
                val *= 1024;
                /* fall through */
        case 'k':
        case 'K':
                val *= 1024;
                break;
        default:
                errx(1, "Unknown suffix in number '%s'\n", str);
                /* not reached */
        }
        if (ptr[1]) {
                errx(1, "Unknown suffix in number '%s'\n", str);
                /* not reached */
        }
        if (val < minval) {
                errx(1, "Value too small: %s, min is %s\n",
                     str, sizetostr(minval));
                /* not reached */
        }
        if (val > maxval) {
                errx(1, "Value too large: %s, max is %s\n",
                     str, sizetostr(maxval));
                /* not reached */
        }
        if ((powerof2 & 1) && (val ^ (val - 1)) != ((val << 1) - 1)) {
                errx(1, "Value not power of 2: %s\n", str);
                /* not reached */
        }
        if ((powerof2 & 2) && (val & HAMMER2_NEWFS_ALIGNMASK)) {
                errx(1, "Value not an integral multiple of %dK: %s",
                     HAMMER2_NEWFS_ALIGN / 1024, str);
                /* not reached */
        }
        return(val);
}

static uint64_t
nowtime(void)
{
        struct timeval tv;
        uint64_t xtime;

        gettimeofday(&tv, NULL);
        xtime = tv.tv_sec * 1000000LL + tv.tv_usec;
        return(xtime);
}

/*
 * Figure out how big the volume is.
 */
static
hammer2_off_t
check_volume(const char *path, int *fdp)
{
        struct partinfo pinfo;
        struct stat st;
        hammer2_off_t size;

        /*
         * Get basic information about the volume
         */
        *fdp = open(path, O_RDWR);
        if (*fdp < 0)
                err(1, "Unable to open %s R+W", path);
        if (ioctl(*fdp, DIOCGPART, &pinfo) < 0) {
                /*
                 * Allow the formatting of regular files as HAMMER2 volumes
                 */
                if (fstat(*fdp, &st) < 0)
                        err(1, "Unable to stat %s", path);
                size = st.st_size;
        } else {
                /*
                 * When formatting a block device as a HAMMER2 volume the
                 * sector size must be compatible.  HAMMER2 uses 64K
                 * filesystem buffers but logical buffers for direct I/O
                 * can be as small as HAMMER2_LOGSIZE (16KB).
                 */
                if (pinfo.reserved_blocks) {
                        errx(1, "HAMMER2 cannot be placed in a partition "
                                "which overlaps the disklabel or MBR");
                }
                if (pinfo.media_blksize > HAMMER2_PBUFSIZE ||
                    HAMMER2_PBUFSIZE % pinfo.media_blksize) {
                        errx(1, "A media sector size of %d is not supported",
                             pinfo.media_blksize);
                }
                size = pinfo.media_size;
        }
        printf("Volume %-15s size %s\n", path, sizetostr(size));
        return (size);
}

/*
 * Create the volume header, the super-root directory inode, and
 * the writable snapshot subdirectory (named via the label) which
 * is to be the initial mount point, or at least the first mount point.
 *
 * [----reserved_area----][boot_area][aux_area]
 * [[vol_hdr]...         ]                     [sroot][root]...
 *
 * The sroot and root inodes eat 512 bytes each.  newfs labels can only be
 * 64 bytes so the root (snapshot) inode does not need to extend past 512
 * bytes.  We use the correct hash slot correct but note that because
 * directory hashes are chained 16x, any slot in the inode will work.
 *
 * Also format the allocation map.
 *
 * NOTE: The passed total_space is 8MB-aligned to avoid edge cases.
 */
static
void
format_hammer2(int fd, hammer2_off_t total_space, hammer2_off_t free_space)
{
        char *buf = malloc(HAMMER2_PBUFSIZE);
        hammer2_volume_data_t *vol;
        hammer2_inode_data_t *rawip;
        hammer2_blockref_t sroot_blockref;
        hammer2_blockref_t root_blockref[MAXLABELS];
        uint64_t now;
        hammer2_off_t volu_base = 0;
        hammer2_off_t boot_base = HAMMER2_ZONE_SEG;
        hammer2_off_t aux_base = boot_base + BootAreaSize;
        hammer2_off_t alloc_base = aux_base + AuxAreaSize;
        hammer2_off_t tmp_base;
        size_t n;
        int i;

        /*
         * Clear the entire reserve for the first 2G segment and
         * make sure we can write to the last block.
         */
        bzero(buf, HAMMER2_PBUFSIZE);
        tmp_base = volu_base;
        for (i = 0; i < HAMMER2_ZONE_BLOCKS_SEG; ++i) {
                n = pwrite(fd, buf, HAMMER2_PBUFSIZE, tmp_base);
                if (n != HAMMER2_PBUFSIZE) {
                        perror("write");
                        exit(1);
                }
                tmp_base += HAMMER2_PBUFSIZE;
        }

        n = pwrite(fd, buf, HAMMER2_PBUFSIZE,
                   volu_base + total_space - HAMMER2_PBUFSIZE);
        if (n != HAMMER2_PBUFSIZE) {
                perror("write (at-end-of-volume)");
                exit(1);
        }

        /*
         * Make sure alloc_base won't cross the reserved area at the
         * beginning of each 2GB zone.
         *
         * Reserve space for the super-root inode and the root inode.
         * Make sure they are in the same 64K block to simplify our code.
         */
        assert((alloc_base & HAMMER2_PBUFMASK) == 0);
        assert(alloc_base < HAMMER2_ZONE_BYTES64 - HAMMER2_ZONE_SEG);
        now = nowtime();
        bzero(buf, HAMMER2_PBUFSIZE);

        alloc_base &= ~HAMMER2_PBUFMASK64;
        alloc_direct(&alloc_base, &sroot_blockref, HAMMER2_INODE_BYTES);

        for (i = 0; i < NLabels; ++i) {
                uuidgen(&Hammer2_PfsCLID[i], 1);
                uuidgen(&Hammer2_PfsFSID[i], 1);

                alloc_direct(&alloc_base, &root_blockref[i],
                             HAMMER2_INODE_BYTES);
                assert(((sroot_blockref.data_off ^ root_blockref[i].data_off) &
                        HAMMER2_OFF_MASK_HI) == 0);

                /*
                 * Format the root directory inode, which is left empty.
                 */
                rawip = (void *)(buf + (HAMMER2_OFF_MASK_LO &
                                        root_blockref[i].data_off));
                rawip->meta.version = HAMMER2_INODE_VERSION_ONE;
                rawip->meta.ctime = now;
                rawip->meta.mtime = now;
                /* rawip->atime = now; NOT IMPL MUST BE ZERO */
                rawip->meta.btime = now;
                rawip->meta.type = HAMMER2_OBJTYPE_DIRECTORY;
                rawip->meta.mode = 0755;
                rawip->meta.inum = 1;   /* root inode, inumber 1 */
                rawip->meta.nlinks = 1; /* directory link count compat */

                rawip->meta.name_len = strlen(Label[i]);
                bcopy(Label[i], rawip->filename, rawip->meta.name_len);
                rawip->meta.name_key =
                                dirhash(rawip->filename, rawip->meta.name_len);

                /*
                 * Compression mode and supported copyids.
                 *
                 * Do not allow compression when creating any "BOOT" label
                 * (pfs-create also does the same if the pfs is named "BOOT")
                 */
                if (strcasecmp(Label[i], "BOOT") == 0) {
                        rawip->meta.comp_algo = HAMMER2_ENC_ALGO(
                                                    HAMMER2_COMP_AUTOZERO);
                        rawip->meta.check_algo = HAMMER2_ENC_ALGO(
                                                    HAMMER2_CHECK_XXHASH64);
                } else  {
                        rawip->meta.comp_algo = HAMMER2_ENC_ALGO(
                                                    HAMMER2_COMP_NEWFS_DEFAULT);
                        rawip->meta.check_algo = HAMMER2_ENC_ALGO(
                                                    HAMMER2_CHECK_XXHASH64);
                }

                /*
                 * NOTE: We leave nmasters set to 0, which means that we
                 *       don't know how many masters there are.  The quorum
                 *       calculation will effectively be 1 ( 0 / 2 + 1 ).
                 */
                rawip->meta.pfs_clid = Hammer2_PfsCLID[i];
                rawip->meta.pfs_fsid = Hammer2_PfsFSID[i];
                rawip->meta.pfs_type = HAMMER2_PFSTYPE_MASTER;
                rawip->meta.op_flags |= HAMMER2_OPFLAG_PFSROOT;

                /* first allocatable inode number */
                rawip->meta.pfs_inum = 16;

                /* rawip->u.blockset is left empty */

                /*
                 * The root blockref will be stored in the super-root inode as
                 * the only directory entry.  The copyid here is the actual
                 * copyid of the storage ref.
                 *
                 * The key field for a directory entry's blockref is
                 * essentially the name key for the entry.
                 */
                root_blockref[i].key = rawip->meta.name_key;
                root_blockref[i].copyid = HAMMER2_COPYID_LOCAL;
                root_blockref[i].keybits = 0;
                root_blockref[i].check.xxhash64.value =
                                XXH64(rawip, sizeof(*rawip), XXH_HAMMER2_SEED);
                root_blockref[i].type = HAMMER2_BREF_TYPE_INODE;
                root_blockref[i].methods =
                                HAMMER2_ENC_CHECK(HAMMER2_CHECK_XXHASH64) |
                                HAMMER2_ENC_COMP(HAMMER2_COMP_NONE);
                root_blockref[i].mirror_tid = 16;
                root_blockref[i].flags = HAMMER2_BREF_FLAG_PFSROOT;
        }

        /*
         * Format the super-root directory inode, giving it one directory
         * entry (root_blockref) and fixup the icrc method.
         *
         * The superroot contains one directory entry pointing at the root
         * inode (named via the label).  Inodes contain one blockset which
         * is fully associative so we can put the entry anywhere without
         * having to worry about the hash.  Use index 0.
         */
        rawip = (void *)(buf + (HAMMER2_OFF_MASK_LO & sroot_blockref.data_off));
        rawip->meta.version = HAMMER2_INODE_VERSION_ONE;
        rawip->meta.ctime = now;
        rawip->meta.mtime = now;
        /* rawip->meta.atime = now; NOT IMPL MUST BE ZERO */
        rawip->meta.btime = now;
        rawip->meta.type = HAMMER2_OBJTYPE_DIRECTORY;
        rawip->meta.mode = 0700;        /* super-root - root only */
        rawip->meta.inum = 0;           /* super root inode, inumber 0 */
        rawip->meta.nlinks = 2;         /* directory link count compat */

        rawip->meta.name_len = 0;       /* super-root is unnamed */
        rawip->meta.name_key = 0;

        rawip->meta.comp_algo = HAMMER2_ENC_ALGO(HAMMER2_COMP_AUTOZERO);
        rawip->meta.check_algo = HAMMER2_ENC_ALGO(HAMMER2_CHECK_XXHASH64);

        /*
         * The super-root is flagged as a PFS and typically given its own
         * random FSID, making it possible to mirror an entire HAMMER2 disk
         * snapshots and all if desired.  PFS ids are used to match up
         * mirror sources and targets and cluster copy sources and targets.
         *
         * (XXX whole-disk logical mirroring is not really supported in
         *  the first attempt because each PFS is in its own modify/mirror
         *  transaction id domain, so normal mechanics cannot cross a PFS
         *  boundary).
         */
        rawip->meta.pfs_clid = Hammer2_SupCLID;
        rawip->meta.pfs_fsid = Hammer2_SupFSID;
        rawip->meta.pfs_type = HAMMER2_PFSTYPE_SUPROOT;
        snprintf(rawip->filename, sizeof(rawip->filename), "SUPROOT");
        rawip->meta.name_key = 0;
        rawip->meta.name_len = strlen(rawip->filename);

        /* The super-root has an inode number of 0 */
        rawip->meta.pfs_inum = 0;

        /*
         * Currently newfs_hammer2 just throws the PFS inodes into the
         * top-level block table at the volume root and doesn't try to
         * create an indirect block, so we are limited to ~4 at filesystem
         * creation time.  More can be added after mounting.
         */
        qsort(root_blockref, NLabels, sizeof(root_blockref[0]), blkrefary_cmp);
        for (i = 0; i < NLabels; ++i)
                rawip->u.blockset.blockref[i] = root_blockref[i];

        /*
         * The sroot blockref will be stored in the volume header.
         */
        sroot_blockref.copyid = HAMMER2_COPYID_LOCAL;
        sroot_blockref.keybits = 0;
        sroot_blockref.check.xxhash64.value =
                                        XXH64(rawip, sizeof(*rawip), XXH_HAMMER2_SEED);
        sroot_blockref.type = HAMMER2_BREF_TYPE_INODE;
        sroot_blockref.methods = HAMMER2_ENC_CHECK(HAMMER2_CHECK_XXHASH64) |
                                 HAMMER2_ENC_COMP(HAMMER2_COMP_AUTOZERO);
        sroot_blockref.mirror_tid = 16;
        rawip = NULL;

        /*
         * Write out the 64K HAMMER2 block containing the root and sroot.
         */
        n = pwrite(fd, buf, HAMMER2_PBUFSIZE,
                   sroot_blockref.data_off & HAMMER2_OFF_MASK_HI);
        if (n != HAMMER2_PBUFSIZE) {
                perror("write");
                exit(1);
        }

        /*
         * Format the volume header.
         *
         * The volume header points to sroot_blockref.  Also be absolutely
         * sure that allocator_beg is set.
         */
        bzero(buf, HAMMER2_PBUFSIZE);
        vol = (void *)buf;

        vol->magic = HAMMER2_VOLUME_ID_HBO;
        vol->boot_beg = boot_base;
        vol->boot_end = boot_base + BootAreaSize;
        vol->aux_beg = aux_base;
        vol->aux_end = aux_base + AuxAreaSize;
        vol->volu_size = total_space;
        vol->version = Hammer2Version;
        vol->flags = 0;

        vol->fsid = Hammer2_VolFSID;
        vol->fstype = Hammer2_FSType;

        vol->peer_type = DMSG_PEER_HAMMER2;     /* LNK_CONN identification */

        vol->allocator_size = free_space;
        vol->allocator_free = free_space;
        vol->allocator_beg = alloc_base;

        vol->sroot_blockset.blockref[0] = sroot_blockref;
        vol->mirror_tid = 16;   /* all blockref mirror TIDs set to 16 */
        vol->freemap_tid = 16;  /* all blockref mirror TIDs set to 16 */
        vol->icrc_sects[HAMMER2_VOL_ICRC_SECT1] =
                        hammer2_icrc32((char *)vol + HAMMER2_VOLUME_ICRC1_OFF,
                                       HAMMER2_VOLUME_ICRC1_SIZE);

        /*
         * Set ICRC_SECT0 after all remaining elements of sect0 have been
         * populated in the volume header.  Note hat ICRC_SECT* (except for
         * SECT0) are part of sect0.
         */
        vol->icrc_sects[HAMMER2_VOL_ICRC_SECT0] =
                        hammer2_icrc32((char *)vol + HAMMER2_VOLUME_ICRC0_OFF,
                                       HAMMER2_VOLUME_ICRC0_SIZE);
        vol->icrc_volheader =
                        hammer2_icrc32((char *)vol + HAMMER2_VOLUME_ICRCVH_OFF,
                                       HAMMER2_VOLUME_ICRCVH_SIZE);

        /*
         * Write the volume header and all alternates.
         */
        for (i = 0; i < HAMMER2_NUM_VOLHDRS; ++i) {
                if (i * HAMMER2_ZONE_BYTES64 >= total_space)
                        break;
                n = pwrite(fd, buf, HAMMER2_PBUFSIZE,
                           volu_base + i * HAMMER2_ZONE_BYTES64);
                if (n != HAMMER2_PBUFSIZE) {
                        perror("write");
                        exit(1);
                }
        }

        /*
         * Cleanup
         */
        free(buf);
}

static void
alloc_direct(hammer2_off_t *basep, hammer2_blockref_t *bref, size_t bytes)
{
        int radix;

        radix = 0;
        assert(bytes);
        while ((bytes & 1) == 0) {
                bytes >>= 1;
                ++radix;
        }
        assert(bytes == 1);
        if (radix < HAMMER2_RADIX_MIN)
                radix = HAMMER2_RADIX_MIN;

        bzero(bref, sizeof(*bref));
        bref->data_off = *basep | radix;
        bref->vradix = radix;

        *basep += 1U << radix;
}

/*
 * Borrow HAMMER1's directory hash algorithm #1 with a few modifications.
 * The filename is split into fields which are hashed separately and then
 * added together.
 *
 * Differences include: bit 63 must be set to 1 for HAMMER2 (HAMMER1 sets
 * it to 0), this is because bit63=0 is used for hidden hardlinked inodes.
 * (This means we do not need to do a 0-check/or-with-0x100000000 either).
 *
 * Also, the iscsi crc code is used instead of the old crc32 code.
 */
static hammer2_key_t
dirhash(const unsigned char *name, size_t len)
{
        const unsigned char *aname = name;
        uint32_t crcx;
        uint64_t key;
        size_t i;
        size_t j;

        /*
         * Filesystem version 6 or better will create directories
         * using the ALG1 dirhash.  This hash breaks the filename
         * up into domains separated by special characters and
         * hashes each domain independently.
         *
         * We also do a simple sub-sort using the first character
         * of the filename in the top 5-bits.
         */
        key = 0;

        /*
         * m32
         */
        crcx = 0;
        for (i = j = 0; i < len; ++i) {
                if (aname[i] == '.' ||
                    aname[i] == '-' ||
                    aname[i] == '_' ||
                    aname[i] == '~') {
                        if (i != j)
                                crcx += hammer2_icrc32(aname + j, i - j);
                        j = i + 1;
                }
        }
        if (i != j)
                crcx += hammer2_icrc32(aname + j, i - j);

        /*
         * The directory hash utilizes the top 32 bits of the 64-bit key.
         * Bit 63 must be set to 1.
         */
        crcx |= 0x80000000U;
        key |= (uint64_t)crcx << 32;

        /*
         * l16 - crc of entire filename
         *
         * This crc reduces degenerate hash collision conditions
         */
        crcx = hammer2_icrc32(aname, len);
        crcx = crcx ^ (crcx << 16);
        key |= crcx & 0xFFFF0000U;

        /*
         * Set bit 15.  This allows readdir to strip bit 63 so a positive
         * 64-bit cookie/offset can always be returned, and still guarantee
         * that the values 0x0000-0x7FFF are available for artificial entries.
         * ('.' and '..').
         */
        key |= 0x8000U;

        return (key);
}

static int
blkrefary_cmp(const void *b1, const void *b2)
{
        const hammer2_blockref_t *bref1 = b1;
        const hammer2_blockref_t *bref2 = b2;
        if (bref1->key < bref2->key)
                return(-1);
        if (bref1->key > bref2->key)
                return(1);
        return 0;
}