sys/vfs/hammer/hammer_expand.c

   1 /*
   2  * Copyright (c) 2009 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Matthew Dillon <dillon@backplane.com> and
   6  * Michael Neumann <mneumann@ntecs.de>
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  *
  12  * 1. Redistributions of source code must retain the above copyright
  13  *    notice, this list of conditions and the following disclaimer.
  14  * 2. Redistributions in binary form must reproduce the above copyright
  15  *    notice, this list of conditions and the following disclaimer in
  16  *    the documentation and/or other materials provided with the
  17  *    distribution.
  18  * 3. Neither the name of The DragonFly Project nor the names of its
  19  *    contributors may be used to endorse or promote products derived
  20  *    from this software without specific, prior written permission.
  21  *
  22  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  23  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  24  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  25  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  26  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  27  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  28  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  29  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  30  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  31  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  32  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  33  * SUCH DAMAGE.
  34  *
  35  */
  36
  37 #include "hammer.h"
  38 #include <sys/fcntl.h>
  39 #include <sys/nlookup.h>
  40 #include <sys/buf.h>
  41
  42 static int
  43 hammer_format_volume_header(struct hammer_mount *hmp, const char *dev_path,
  44         const char *vol_name, int vol_no, int vol_count,
  45         int64_t vol_size, int64_t boot_area_size, int64_t mem_area_size,
  46         uint64_t *num_layer1_entries_p, uint64_t *layer1_free_blocks);
  47
  48 int
  49 hammer_ioc_expand(hammer_transaction_t trans, hammer_inode_t ip,
  50                 struct hammer_ioc_expand *expand)
  51 {
  52         struct hammer_mount *hmp = trans->hmp;
  53         struct mount *mp = hmp->mp;
  54         int error;
  55
  56         if (mp->mnt_flag & MNT_RDONLY) {
  57                 kprintf("Cannot expand read-only HAMMER filesystem\n");
  58                 return (EINVAL);
  59         }
  60
  61         if (hmp->nvolumes + 1 >= HAMMER_MAX_VOLUMES) {
  62                 kprintf("Max number of HAMMER volumes exceeded\n");
  63                 return (EINVAL);
  64         }
  65
  66         /*
  67          * Find an unused volume number.
  68          */
  69         int free_vol_no = 0;
  70         while (free_vol_no < HAMMER_MAX_VOLUMES &&
  71                RB_LOOKUP(hammer_vol_rb_tree, &hmp->rb_vols_root, free_vol_no)) {
  72                 ++free_vol_no;
  73         }
  74         if (free_vol_no >= HAMMER_MAX_VOLUMES) {
  75                 kprintf("Max number of HAMMER volumes exceeded\n");
  76                 return (EINVAL);
  77         }
  78
  79         uint64_t num_layer1_entries = 0;
  80         uint64_t *layer1_free_blocks =
  81                 kmalloc(1024 * sizeof(uint64_t), M_TEMP, M_WAITOK|M_ZERO);
  82
  83         error = hammer_format_volume_header(
  84                 hmp,
  85                 expand->device_name,
  86                 hmp->rootvol->ondisk->vol_name,
  87                 free_vol_no,
  88                 hmp->nvolumes+1,
  89                 expand->vol_size,
  90                 expand->boot_area_size,
  91                 expand->mem_area_size,
  92                 &num_layer1_entries /* out param */,
  93                 layer1_free_blocks);
  94         KKASSERT(num_layer1_entries < 1024);
  95         if (error)
  96                 goto end;
  97
  98         error = hammer_install_volume(hmp, expand->device_name, NULL);
  99         if (error)
 100                 goto end;
 101
 102         ++hmp->nvolumes;
 103
 104         hammer_sync_lock_sh(trans);
 105         hammer_lock_ex(&hmp->blkmap_lock);
 106
 107         /*
 108          * Set each volumes new value of the vol_count field.
 109          */
 110         for (int vol_no = 0; vol_no < HAMMER_MAX_VOLUMES; ++vol_no) {
 111                 hammer_volume_t volume;
 112                 volume = hammer_get_volume(hmp, vol_no, &error);
 113                 if (volume == NULL && error == ENOENT) {
 114                         /*
 115                          * Skip unused volume numbers
 116                          */
 117                         error = 0;
 118                         continue;
 119                 }
 120                 KKASSERT(error == 0);
 121                 hammer_modify_volume_field(trans, volume, vol_count);
 122                 volume->ondisk->vol_count = hmp->nvolumes;
 123                 hammer_modify_volume_done(volume);
 124                 hammer_rel_volume(volume, 0);
 125         }
 126
 127         /*
 128          * Assign Layer1 entries
 129          */
 130
 131         hammer_volume_t root_volume = NULL;
 132         hammer_blockmap_t freemap;
 133
 134         freemap = &hmp->blockmap[HAMMER_ZONE_FREEMAP_INDEX];
 135         root_volume = hammer_get_root_volume(hmp, &error);
 136         KKASSERT(root_volume && error == 0);
 137
 138         for (uint64_t i_layer1 = 0; i_layer1 < num_layer1_entries; i_layer1++) {
 139                 hammer_buffer_t buffer1 = NULL;
 140                 struct hammer_blockmap_layer1 *layer1;
 141                 hammer_off_t layer1_offset;
 142
 143                 layer1_offset = freemap->phys_offset +
 144                         (free_vol_no * 1024L) *
 145                         sizeof(struct hammer_blockmap_layer1) + i_layer1;
 146
 147                 layer1 = hammer_bread(hmp, layer1_offset, &error, &buffer1);
 148                 KKASSERT(layer1 != NULL && error == 0);
 149                 KKASSERT(layer1->phys_offset == HAMMER_BLOCKMAP_UNAVAIL);
 150
 151                 hammer_modify_buffer(trans, buffer1, layer1, sizeof(*layer1));
 152                 bzero(layer1, sizeof(*layer1));
 153                 layer1->phys_offset = HAMMER_ENCODE_RAW_BUFFER(free_vol_no,
 154                         i_layer1 * HAMMER_LARGEBLOCK_SIZE);
 155
 156                 layer1->blocks_free = layer1_free_blocks[i_layer1];
 157                 layer1->layer1_crc = crc32(layer1, HAMMER_LAYER1_CRCSIZE);
 158
 159                 hammer_modify_buffer_done(buffer1);
 160                 if (buffer1)
 161                         hammer_rel_buffer(buffer1, 0);
 162
 163                 hammer_modify_volume_field(trans, root_volume,
 164                         vol0_stat_freebigblocks);
 165
 166                 root_volume->ondisk->vol0_stat_freebigblocks +=
 167                         layer1_free_blocks[i_layer1];
 168                 hmp->copy_stat_freebigblocks =
 169                         root_volume->ondisk->vol0_stat_freebigblocks;
 170                 hammer_modify_volume_done(root_volume);
 171         } /* for */
 172
 173         hammer_rel_volume(root_volume, 0);
 174
 175         hammer_unlock(&hmp->blkmap_lock);
 176         hammer_sync_unlock(trans);
 177
 178 end:
 179         if (error) {
 180                 kprintf("An error occured: %d\n", error);
 181         }
 182         if (layer1_free_blocks)
 183                 kfree(layer1_free_blocks, M_TEMP);
 184         return (error);
 185 }
 186
 187 static int
 188 hammer_format_volume_header(struct hammer_mount *hmp, const char *dev_path,
 189         const char *vol_name, int vol_no, int vol_count,
 190         int64_t vol_size, int64_t boot_area_size, int64_t mem_area_size,
 191         uint64_t *num_layer1_entries_p, uint64_t *layer1_free_blocks)
 192 {
 193         struct vnode *devvp = NULL;
 194         struct buf *bp = NULL;
 195         struct nlookupdata nd;
 196         struct hammer_volume_ondisk *ondisk;
 197         int error;
 198
 199         /*
 200          * Get the device vnode
 201          */
 202         error = nlookup_init(&nd, dev_path, UIO_SYSSPACE, NLC_FOLLOW);
 203         if (error == 0)
 204                 error = nlookup(&nd);
 205         if (error == 0)
 206                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &devvp);
 207         nlookup_done(&nd);
 208
 209         if (error == 0) {
 210                 if (vn_isdisk(devvp, &error)) {
 211                         error = vfs_mountedon(devvp);
 212                 }
 213         }
 214         if (error == 0 &&
 215             count_udev(devvp->v_umajor, devvp->v_uminor) > 0) {
 216                 error = EBUSY;
 217         }
 218         if (error == 0) {
 219                 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
 220                 error = vinvalbuf(devvp, V_SAVE, 0, 0);
 221                 if (error == 0) {
 222                         error = VOP_OPEN(devvp, FREAD|FWRITE, FSCRED, NULL);
 223                 }
 224                 vn_unlock(devvp);
 225         }
 226         if (error) {
 227                 if (devvp)
 228                         vrele(devvp);
 229                 return (error);
 230         }
 231
 232         /*
 233          * Extract the volume number from the volume header and do various
 234          * sanity checks.
 235          */
 236         KKASSERT(HAMMER_BUFSIZE >= sizeof(struct hammer_volume_ondisk));
 237         error = bread(devvp, 0LL, HAMMER_BUFSIZE, &bp);
 238         if (error || bp->b_bcount < sizeof(struct hammer_volume_ondisk))
 239                 goto late_failure;
 240
 241         ondisk = (struct hammer_volume_ondisk*) bp->b_data;
 242
 243         /*
 244          * Note that we do NOT allow to use a device that contains
 245          * a valid HAMMER signature. It has to be cleaned up with dd
 246          * before.
 247          */
 248         if (ondisk->vol_signature == HAMMER_FSBUF_VOLUME) {
 249                 kprintf("hammer_expand: Formatting of valid HAMMER volume "
 250                         "%s denied. Erase with dd!\n", vol_name);
 251                 error = EFTYPE;
 252                 goto late_failure;
 253         }
 254
 255         bzero(ondisk, sizeof(struct hammer_volume_ondisk));
 256         ksnprintf(ondisk->vol_name, sizeof(ondisk->vol_name), "%s", vol_name);
 257         ondisk->vol_fstype = hmp->rootvol->ondisk->vol_fstype;
 258         ondisk->vol_signature = HAMMER_FSBUF_VOLUME;
 259         ondisk->vol_fsid = hmp->fsid;
 260         ondisk->vol_rootvol = hmp->rootvol->vol_no;
 261         ondisk->vol_no = vol_no;
 262         ondisk->vol_count = vol_count;
 263         ondisk->vol_version = hmp->version;
 264
 265         /*
 266          * Reserve space for (future) header junk, setup our poor-man's
 267          * bigblock allocator.
 268          */
 269         int64_t vol_alloc = HAMMER_BUFSIZE * 16;
 270
 271         ondisk->vol_bot_beg = vol_alloc;
 272         vol_alloc += boot_area_size;
 273         ondisk->vol_mem_beg = vol_alloc;
 274         vol_alloc += mem_area_size;
 275
 276         /*
 277          * The remaining area is the zone 2 buffer allocation area.  These
 278          * buffers
 279          */
 280         ondisk->vol_buf_beg = vol_alloc;
 281         ondisk->vol_buf_end = vol_size & ~(int64_t)HAMMER_BUFMASK;
 282
 283         if (ondisk->vol_buf_end < ondisk->vol_buf_beg) {
 284                 kprintf("volume %d %s is too small to hold the volume header",
 285                      ondisk->vol_no, ondisk->vol_name);
 286                 error = EFTYPE;
 287                 goto late_failure;
 288         }
 289
 290         ondisk->vol_nblocks = (ondisk->vol_buf_end - ondisk->vol_buf_beg) /
 291                               HAMMER_BUFSIZE;
 292         ondisk->vol_blocksize = HAMMER_BUFSIZE;
 293
 294         /*
 295          * Write volume header to disk
 296          */
 297         error = bwrite(bp);
 298         bp = NULL;
 299
 300         /*
 301          * Initialize layer2 freemap
 302          */
 303
 304         /*
 305          * Determine the number of L1 entries we need to represent the
 306          * space of the whole volume. Each L1 entry covers 4 TB of space
 307          * (8MB * 2**19) and we need one L2 big block for each L1 entry.
 308          * L1 entries are stored in the root volume.
 309          */
 310         hammer_off_t off_end = (ondisk->vol_buf_end - ondisk->vol_buf_beg)
 311                 & ~HAMMER_LARGEBLOCK_MASK64;
 312         uint64_t num_layer1_entries = (off_end / HAMMER_BLOCKMAP_LAYER2) +
 313                 ((off_end & HAMMER_BLOCKMAP_LAYER2_MASK) == 0 ? 0 : 1);
 314         *num_layer1_entries_p = num_layer1_entries;
 315
 316         /*
 317          * We allocate all L2 big blocks sequentially from the start of
 318          * the volume.
 319          */
 320         KKASSERT(off_end / HAMMER_LARGEBLOCK_SIZE >= num_layer1_entries);
 321
 322         hammer_off_t layer2_end = num_layer1_entries * HAMMER_LARGEBLOCK_SIZE;
 323         hammer_off_t off = 0;
 324         while (off < layer2_end) {
 325                 error = bread(devvp, ondisk->vol_buf_beg + off,
 326                               HAMMER_BUFSIZE, &bp);
 327                 if (error || bp->b_bcount != HAMMER_BUFSIZE)
 328                         goto late_failure;
 329                 struct hammer_blockmap_layer2 *layer2 = (void*)bp->b_data;
 330
 331                 for (int i = 0; i < HAMMER_BUFSIZE / sizeof(*layer2); ++i) {
 332
 333                         /* the bigblock described by the layer2 entry */
 334                         hammer_off_t bigblock_off = HAMMER_LARGEBLOCK_SIZE *
 335                                 (off / sizeof(*layer2));
 336
 337                         /*
 338                          * To which layer1 entry does the current layer2
 339                          * big block belong?
 340                          *
 341                          * We need this to calculate the free bigblocks
 342                          * which is required for the layer1.
 343                          */
 344                         uint64_t i_layer1 = HAMMER_BLOCKMAP_LAYER1_OFFSET(off) /
 345                                         sizeof(struct hammer_blockmap_layer1);
 346                         KKASSERT(i_layer1 < 1024);
 347
 348                         bzero(layer2, sizeof(*layer2));
 349
 350                         if ((off & HAMMER_LARGEBLOCK_SIZE) == bigblock_off) {
 351                                 /*
 352                                  * Bigblock is part of the layer2 freemap
 353                                  */
 354                                 layer2->zone = HAMMER_ZONE_FREEMAP_INDEX;
 355                                 layer2->append_off = HAMMER_LARGEBLOCK_SIZE;
 356                                 layer2->bytes_free = 0;
 357                         } else if (bigblock_off < off_end) {
 358                                 layer2->zone = 0;
 359                                 layer2->append_off = 0;
 360                                 layer2->bytes_free = HAMMER_LARGEBLOCK_SIZE;
 361                                 ++layer1_free_blocks[i_layer1];
 362                         } else {
 363                                 layer2->zone = HAMMER_ZONE_UNAVAIL_INDEX;
 364                                 layer2->append_off = HAMMER_LARGEBLOCK_SIZE;
 365                                 layer2->bytes_free = 0;
 366                         }
 367                         layer2->entry_crc = crc32(layer2, HAMMER_LAYER2_CRCSIZE);
 368                         off += sizeof(*layer2);
 369                         ++layer2;
 370                 }
 371
 372                 error = bwrite(bp);
 373                 bp = NULL;
 374                 if (error)
 375                         goto late_failure;
 376         }
 377
 378 late_failure:
 379         if (bp)
 380                 brelse(bp);
 381         VOP_CLOSE(devvp, FREAD|FWRITE);
 382         if (devvp)
 383                 vrele(devvp);
 384         return (error);
 385 }