sys/kern/subr_disklabel64.c

   1 /*
   2  * Copyright (c) 2007 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>
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  *
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in
  15  *    the documentation and/or other materials provided with the
  16  *    distribution.
  17  * 3. Neither the name of The DragonFly Project nor the names of its
  18  *    contributors may be used to endorse or promote products derived
  19  *    from this software without specific, prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  *
  34  * $DragonFly: src/sys/kern/subr_disklabel64.c,v 1.5 2007/07/20 17:21:51 dillon Exp $
  35  */
  36
  37 #include <sys/param.h>
  38 #include <sys/systm.h>
  39 #include <sys/kernel.h>
  40 #include <sys/conf.h>
  41 #include <sys/disklabel.h>
  42 #include <sys/disklabel64.h>
  43 #include <sys/diskslice.h>
  44 #include <sys/disk.h>
  45 #include <sys/kern_syscall.h>
  46 #include <sys/buf2.h>
  47
  48 /*
  49  * Retrieve the partition start and extent, in blocks.  Return 0 on success,
  50  * EINVAL on error.
  51  */
  52 static int
  53 l64_getpartbounds(struct diskslices *ssp, disklabel_t lp, u_int32_t part,
  54                   u_int64_t *start, u_int64_t *blocks)
  55 {
  56         struct partition64 *pp;
  57
  58         if (part >= lp.lab64->d_npartitions)
  59                 return (EINVAL);
  60
  61         pp = &lp.lab64->d_partitions[part];
  62
  63         if ((pp->p_boffset & (ssp->dss_secsize - 1)) ||
  64             (pp->p_bsize & (ssp->dss_secsize - 1))) {
  65                 return (EINVAL);
  66         }
  67         *start = pp->p_boffset / ssp->dss_secsize;
  68         *blocks = pp->p_bsize / ssp->dss_secsize;
  69         return(0);
  70 }
  71
  72 /*
  73  * Get the filesystem type XXX - diskslices code needs to use uuids
  74  */
  75 static void
  76 l64_loadpartinfo(disklabel_t lp, u_int32_t part, struct partinfo *dpart)
  77 {
  78         struct partition64 *pp;
  79         const size_t uuid_size = sizeof(struct uuid);
  80
  81         if (part < lp.lab64->d_npartitions) {
  82                 pp = &lp.lab64->d_partitions[part];
  83                 dpart->fstype_uuid = pp->p_type_uuid;
  84                 dpart->storage_uuid = pp->p_stor_uuid;
  85                 dpart->fstype = pp->p_fstype;
  86         } else {
  87                 bzero(&dpart->fstype_uuid, uuid_size);
  88                 bzero(&dpart->storage_uuid, uuid_size);
  89                 dpart->fstype = 0;
  90         }
  91 }
  92
  93 /*
  94  * Get the number of partitions
  95  */
  96 static u_int32_t
  97 l64_getnumparts(disklabel_t lp)
  98 {
  99         return(lp.lab64->d_npartitions);
 100 }
 101
 102 /*
 103  * Attempt to read a disk label from a device.  64 bit disklabels are
 104  * sector-agnostic and begin at offset 0 on the device.  64 bit disklabels
 105  * may only be used with GPT partitioning schemes.
 106  *
 107  * Returns NULL on sucess, and an error string on failure.
 108  */
 109 static const char *
 110 l64_readdisklabel(cdev_t dev, struct diskslice *sp, disklabel_t *lpp,
 111                   struct disk_info *info)
 112 {
 113         struct buf *bp;
 114         struct disklabel64 *dlp;
 115         const char *msg;
 116         uint32_t savecrc;
 117         size_t dlpcrcsize;
 118         size_t bpsize;
 119         int secsize;
 120
 121         /*
 122          * XXX I/O size is subject to device DMA limitations
 123          */
 124         secsize = info->d_media_blksize;
 125         bpsize = (sizeof(*dlp) + secsize - 1) & ~(secsize - 1);
 126
 127         bp = geteblk(bpsize);
 128         bp->b_bio1.bio_offset = 0;
 129         bp->b_bcount = bpsize;
 130         bp->b_flags &= ~B_INVAL;
 131         bp->b_cmd = BUF_CMD_READ;
 132         dev_dstrategy(dev, &bp->b_bio1);
 133
 134         if (biowait(bp)) {
 135                 msg = "I/O error";
 136         } else {
 137                 dlp = (struct disklabel64 *)bp->b_data;
 138                 dlpcrcsize = offsetof(struct disklabel64,
 139                                       d_partitions[dlp->d_npartitions]) -
 140                              offsetof(struct disklabel64, d_magic);
 141                 savecrc = dlp->d_crc;
 142                 dlp->d_crc = 0;
 143                 if (dlp->d_magic != DISKMAGIC64) {
 144                         msg = "no disk label";
 145                 } else if (dlp->d_npartitions > MAXPARTITIONS64) {
 146                         msg = "disklabel64 corrupted, too many partitions";
 147                 } else if (savecrc != crc32(&dlp->d_magic, dlpcrcsize)) {
 148                         msg = "disklabel64 corrupted, bad CRC";
 149                 } else {
 150                         dlp->d_crc = savecrc;
 151                         (*lpp).lab64 = kmalloc(sizeof(*dlp),
 152                                                M_DEVBUF, M_WAITOK|M_ZERO);
 153                         *(*lpp).lab64 = *dlp;
 154                         msg = NULL;
 155                 }
 156         }
 157         bp->b_flags |= B_INVAL | B_AGE;
 158         brelse(bp);
 159         return (msg);
 160 }
 161
 162 /*
 163  * If everything is good, copy olpx to nlpx.  Check to see if any
 164  * open partitions would change.
 165  */
 166 static int
 167 l64_setdisklabel(disklabel_t olpx, disklabel_t nlpx, struct diskslices *ssp,
 168                  struct diskslice *sp, u_int32_t *openmask)
 169 {
 170         struct disklabel64 *olp, *nlp;
 171         struct partition64 *opp, *npp;
 172         uint32_t savecrc;
 173         uint64_t slicebsize;
 174         size_t nlpcrcsize;
 175         int part;
 176         int i;
 177
 178         olp = olpx.lab64;
 179         nlp = nlpx.lab64;
 180
 181         slicebsize = (uint64_t)sp->ds_size * ssp->dss_secsize;
 182
 183         if (nlp->d_magic != DISKMAGIC64)
 184                 return (EINVAL);
 185         if (nlp->d_npartitions > MAXPARTITIONS64)
 186                 return (EINVAL);
 187         savecrc = nlp->d_crc;
 188         nlp->d_crc = 0;
 189         nlpcrcsize = offsetof(struct disklabel64,
 190                               d_partitions[nlp->d_npartitions]) -
 191                      offsetof(struct disklabel64, d_magic);
 192         if (crc32(&nlp->d_magic, nlpcrcsize) != savecrc) {
 193                 nlp->d_crc = savecrc;
 194                 return (EINVAL);
 195         }
 196         nlp->d_crc = savecrc;
 197
 198         /*
 199          * Check if open partitions have changed
 200          */
 201         i = 0;
 202         while (i < 128) {
 203                 if (openmask[i >> 5] == 0) {
 204                         i += 32;
 205                         continue;
 206                 }
 207                 if ((openmask[i >> 5] & (1 << (i & 31))) == 0) {
 208                         ++i;
 209                         continue;
 210                 }
 211                 if (nlp->d_npartitions <= i)
 212                         return (EBUSY);
 213                 opp = &olp->d_partitions[i];
 214                 npp = &nlp->d_partitions[i];
 215                 if (npp->p_boffset != opp->p_boffset ||
 216                     npp->p_bsize < opp->p_bsize) {
 217                         return (EBUSY);
 218                 }
 219
 220                 /*
 221                  * Do not allow p_type_uuid or p_stor_uuid to change if
 222                  * the partition is currently open.
 223                  */
 224                 if (bcmp(&npp->p_type_uuid, &opp->p_type_uuid,
 225                      sizeof(npp->p_type_uuid)) != 0) {
 226                         return (EBUSY);
 227                 }
 228                 if (bcmp(&npp->p_stor_uuid, &opp->p_stor_uuid,
 229                      sizeof(npp->p_stor_uuid)) != 0) {
 230                         return (EBUSY);
 231                 }
 232                 ++i;
 233         }
 234
 235         /*
 236          * Make sure the label and partition offsets and sizes are sane.
 237          */
 238         if (nlp->d_total_size > slicebsize)
 239                 return (ENOSPC);
 240         if (nlp->d_total_size & (ssp->dss_secsize - 1))
 241                 return (EINVAL);
 242         if (nlp->d_bbase & (ssp->dss_secsize - 1))
 243                 return (EINVAL);
 244         if (nlp->d_pbase & (ssp->dss_secsize - 1))
 245                 return (EINVAL);
 246         if (nlp->d_pstop & (ssp->dss_secsize - 1))
 247                 return (EINVAL);
 248         if (nlp->d_abase & (ssp->dss_secsize - 1))
 249                 return (EINVAL);
 250
 251         for (part = 0; part < nlp->d_npartitions; ++part) {
 252                 npp = &nlp->d_partitions[i];
 253                 if (npp->p_bsize == 0) {
 254                         if (npp->p_boffset != 0)
 255                                 return (EINVAL);
 256                         continue;
 257                 }
 258                 if (npp->p_boffset & (ssp->dss_secsize - 1))
 259                         return (EINVAL);
 260                 if (npp->p_bsize & (ssp->dss_secsize - 1))
 261                         return (EINVAL);
 262                 if (npp->p_boffset < nlp->d_pbase)
 263                         return (ENOSPC);
 264                 if (npp->p_boffset + npp->p_bsize > nlp->d_total_size)
 265                         return (ENOSPC);
 266         }
 267
 268         /*
 269          * Structurally we may add code to make modifications above in the
 270          * future, so regenerate the crc anyway.
 271          */
 272         nlp->d_crc = 0;
 273         nlp->d_crc = crc32(&nlp->d_magic, nlpcrcsize);
 274         *olp = *nlp;
 275
 276         return (0);
 277 }
 278
 279 /*
 280  * Write disk label back to device after modification.
 281  */
 282 static int
 283 l64_writedisklabel(cdev_t dev, struct diskslices *ssp,
 284                    struct diskslice *sp, disklabel_t lpx)
 285 {
 286         struct disklabel64 *lp;
 287         struct disklabel64 *dlp;
 288         struct buf *bp;
 289         int error = 0;
 290         size_t bpsize;
 291         int secsize;
 292
 293         lp = lpx.lab64;
 294
 295         /*
 296          * XXX I/O size is subject to device DMA limitations
 297          */
 298         secsize = ssp->dss_secsize;
 299         bpsize = (sizeof(*lp) + secsize - 1) & ~(secsize - 1);
 300
 301         bp = geteblk(bpsize);
 302         bp->b_bio1.bio_offset = 0;
 303         bp->b_bcount = bpsize;
 304
 305         /*
 306          * Because our I/O is larger then the label, and because we do not
 307          * write the d_reserved0[] area, do a read-modify-write.
 308          */
 309         bp->b_flags &= ~B_INVAL;
 310         bp->b_cmd = BUF_CMD_READ;
 311         dev_dstrategy(dkmodpart(dev, WHOLE_SLICE_PART), &bp->b_bio1);
 312         error = biowait(bp);
 313         if (error)
 314                 goto done;
 315
 316         dlp = (void *)bp->b_data;
 317         bcopy(&lp->d_magic, &dlp->d_magic,
 318               sizeof(*lp) - offsetof(struct disklabel64, d_magic));
 319         bp->b_cmd = BUF_CMD_WRITE;
 320         dev_dstrategy(dkmodpart(dev, WHOLE_SLICE_PART), &bp->b_bio1);
 321         error = biowait(bp);
 322 done:
 323         bp->b_flags |= B_INVAL | B_AGE;
 324         brelse(bp);
 325         return (error);
 326 }
 327
 328 /*
 329  * Create a disklabel based on a disk_info structure for the purposes of
 330  * DSO_COMPATLABEL - cases where no real label exists on the storage medium.
 331  *
 332  * If a diskslice is passed, the label is truncated to the slice.
 333  *
 334  * NOTE!  This is not a legal label because d_bbase and d_pbase are both
 335  * set to 0.
 336  */
 337 static disklabel_t
 338 l64_clone_label(struct disk_info *info, struct diskslice *sp)
 339 {
 340         struct disklabel64 *lp;
 341         disklabel_t res;
 342         uint32_t blksize = info->d_media_blksize;
 343         size_t lpcrcsize;
 344
 345         lp = kmalloc(sizeof *lp, M_DEVBUF, M_WAITOK | M_ZERO);
 346
 347         if (sp)
 348                 lp->d_total_size = (uint64_t)sp->ds_size * blksize;
 349         else
 350                 lp->d_total_size = info->d_media_blocks * blksize;
 351
 352         lp->d_magic = DISKMAGIC64;
 353         lp->d_align = blksize;
 354         lp->d_npartitions = MAXPARTITIONS64;
 355         lp->d_pstop = lp->d_total_size;
 356
 357         /*
 358          * Create a dummy 'c' part and a dummy 'a' part (if requested).
 359          * Note that the 'c' part is really a hack.  64 bit disklabels
 360          * do not use 'c' to mean the raw partition.
 361          */
 362
 363         lp->d_partitions[2].p_boffset = 0;
 364         lp->d_partitions[2].p_bsize = lp->d_total_size;
 365         /* XXX SET FS TYPE */
 366
 367         if (info->d_dsflags & DSO_COMPATPARTA) {
 368                 lp->d_partitions[0].p_boffset = 0;
 369                 lp->d_partitions[0].p_bsize = lp->d_total_size;
 370                 /* XXX SET FS TYPE */
 371         }
 372
 373         lpcrcsize = offsetof(struct disklabel64,
 374                              d_partitions[lp->d_npartitions]) -
 375                     offsetof(struct disklabel64, d_magic);
 376
 377         lp->d_crc = crc32(&lp->d_magic, lpcrcsize);
 378         res.lab64 = lp;
 379         return (res);
 380 }
 381
 382 /*
 383  * Create a virgin disklabel64 suitable for writing to the media.
 384  *
 385  * disklabel64 always reserves 32KB for a boot area and leaves room
 386  * for up to RESPARTITIONS64 partitions.
 387  */
 388 static void
 389 l64_makevirginlabel(disklabel_t lpx, struct diskslices *ssp,
 390                     struct diskslice *sp, struct disk_info *info)
 391 {
 392         struct disklabel64 *lp = lpx.lab64;
 393         struct partition64 *pp;
 394         uint32_t blksize;
 395         uint32_t ressize;
 396         uint64_t blkmask;       /* 64 bits so we can ~ */
 397         size_t lpcrcsize;
 398
 399         /*
 400          * Setup the initial label.  Use of a block size of at least 4KB
 401          * for calculating the initial reserved areas to allow some degree
 402          * of portability between media with different sector sizes.
 403          *
 404          * Note that the modified blksize is stored in d_align as a hint
 405          * to the disklabeling program.
 406          */
 407         bzero(lp, sizeof(*lp));
 408         if ((blksize = info->d_media_blksize) < 4096)
 409                 blksize = 4096;
 410         blkmask = blksize - 1;
 411
 412         if (sp)
 413                 lp->d_total_size = (uint64_t)sp->ds_size * ssp->dss_secsize;
 414         else
 415                 lp->d_total_size = info->d_media_blocks * info->d_media_blksize;
 416
 417         lp->d_magic = DISKMAGIC64;
 418         lp->d_align = blksize;
 419         lp->d_npartitions = MAXPARTITIONS64;
 420         kern_uuidgen(&lp->d_stor_uuid, 1);
 421
 422         ressize = offsetof(struct disklabel64, d_partitions[RESPARTITIONS64]);
 423         ressize = (ressize + (uint32_t)blkmask) & ~blkmask;
 424
 425         lp->d_bbase = ressize;
 426         lp->d_pbase = lp->d_bbase + ((32768 + blkmask) & ~blkmask);
 427         lp->d_pstop = (lp->d_total_size - lp->d_bbase) & ~blkmask;
 428         lp->d_abase = lp->d_pstop;
 429
 430         /*
 431          * All partitions are left empty unless DSO_COMPATPARTA is set
 432          */
 433
 434         if (info->d_dsflags & DSO_COMPATPARTA) {
 435                 pp = &lp->d_partitions[0];
 436                 pp->p_boffset = lp->d_pbase;
 437                 pp->p_bsize = lp->d_pstop - lp->d_pbase;
 438                 /* XXX SET FS TYPE */
 439         }
 440
 441         lpcrcsize = offsetof(struct disklabel64,
 442                              d_partitions[lp->d_npartitions]) -
 443                     offsetof(struct disklabel64, d_magic);
 444         lp->d_crc = crc32(&lp->d_magic, lpcrcsize);
 445 }
 446
 447 /*
 448  * Set the number of blocks at the beginning of the slice which have
 449  * been reserved for label operations.  This area will be write-protected
 450  * when accessed via the slice.
 451  *
 452  * For now just protect the label area proper.  Do not protect the
 453  * boot area.  Note partitions in 64 bit disklabels do not overlap
 454  * the disklabel or boot area.
 455  */
 456 static void
 457 l64_adjust_label_reserved(struct diskslices *ssp, int slice,
 458                           struct diskslice *sp)
 459 {
 460         struct disklabel64 *lp = sp->ds_label.lab64;
 461
 462         sp->ds_reserved = lp->d_bbase / ssp->dss_secsize;
 463 }
 464
 465 struct disklabel_ops disklabel64_ops = {
 466         .labelsize = sizeof(struct disklabel64),
 467         .op_readdisklabel = l64_readdisklabel,
 468         .op_setdisklabel = l64_setdisklabel,
 469         .op_writedisklabel = l64_writedisklabel,
 470         .op_clone_label = l64_clone_label,
 471         .op_adjust_label_reserved = l64_adjust_label_reserved,
 472         .op_getpartbounds = l64_getpartbounds,
 473         .op_loadpartinfo = l64_loadpartinfo,
 474         .op_getnumparts = l64_getnumparts,
 475         .op_makevirginlabel = l64_makevirginlabel
 476 };
 477