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 1.3 2007/06/19 06:07:57 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         secsize = info->d_media_blksize;
 122         bpsize = (sizeof(*dlp) + secsize - 1) & ~(secsize - 1);
 123
 124         bp = geteblk(bpsize);
 125         bp->b_bio1.bio_offset = 0;
 126         bp->b_bcount = bpsize;
 127         bp->b_flags &= ~B_INVAL;
 128         bp->b_cmd = BUF_CMD_READ;
 129         dev_dstrategy(dev, &bp->b_bio1);
 130
 131         if (biowait(bp)) {
 132                 msg = "I/O error";
 133         } else {
 134                 dlp = (struct disklabel64 *)bp->b_data;
 135                 dlpcrcsize = offsetof(struct disklabel64,
 136                                       d_partitions[dlp->d_npartitions]) -
 137                              offsetof(struct disklabel64, d_magic);
 138                 savecrc = dlp->d_crc;
 139                 dlp->d_crc = 0;
 140                 if (dlp->d_magic != DISKMAGIC64) {
 141                         msg = "no disk label";
 142                 } else if (dlp->d_npartitions > MAXPARTITIONS64) {
 143                         msg = "disklabel64 corrupted, too many partitions";
 144                 } else if (savecrc != crc32(&dlp->d_magic, dlpcrcsize)) {
 145                         msg = "disklabel64 corrupted, bad CRC";
 146                 } else {
 147                         dlp->d_crc = savecrc;
 148                         (*lpp).lab64 = kmalloc(sizeof(*dlp),
 149                                                M_DEVBUF, M_WAITOK|M_ZERO);
 150                         *(*lpp).lab64 = *dlp;
 151                         msg = NULL;
 152                 }
 153         }
 154         bp->b_flags |= B_INVAL | B_AGE;
 155         brelse(bp);
 156         return (msg);
 157 }
 158
 159 /*
 160  * If everything is good, copy olpx to nlpx.  Check to see if any
 161  * open partitions would change.
 162  */
 163 static int
 164 l64_setdisklabel(disklabel_t olpx, disklabel_t nlpx, struct diskslices *ssp,
 165                  struct diskslice *sp, u_int32_t *openmask)
 166 {
 167         struct disklabel64 *olp, *nlp;
 168         struct partition64 *opp, *npp;
 169         uint32_t savecrc;
 170         uint64_t slicebsize;
 171         size_t nlpcrcsize;
 172         int part;
 173         int i;
 174
 175         olp = olpx.lab64;
 176         nlp = nlpx.lab64;
 177
 178         slicebsize = (uint64_t)sp->ds_size * ssp->dss_secsize;
 179
 180         if (nlp->d_magic != DISKMAGIC64)
 181                 return (EINVAL);
 182         if (nlp->d_npartitions > MAXPARTITIONS64)
 183                 return (EINVAL);
 184         savecrc = nlp->d_crc;
 185         nlp->d_crc = 0;
 186         nlpcrcsize = offsetof(struct disklabel64,
 187                               d_partitions[nlp->d_npartitions]) -
 188                      offsetof(struct disklabel64, d_magic);
 189         if (crc32(&nlp->d_magic, nlpcrcsize) != savecrc) {
 190                 nlp->d_crc = savecrc;
 191                 return (EINVAL);
 192         }
 193         nlp->d_crc = savecrc;
 194
 195         /*
 196          * Check if open partitions have changed
 197          */
 198         i = 0;
 199         while (i < 128) {
 200                 if (openmask[i >> 5] == 0) {
 201                         i += 32;
 202                         continue;
 203                 }
 204                 if ((openmask[i >> 5] & (1 << (i & 31))) == 0) {
 205                         ++i;
 206                         continue;
 207                 }
 208                 if (nlp->d_npartitions <= i)
 209                         return (EBUSY);
 210                 opp = &olp->d_partitions[i];
 211                 npp = &nlp->d_partitions[i];
 212                 if (npp->p_boffset != opp->p_boffset ||
 213                     npp->p_bsize < opp->p_bsize) {
 214                         return (EBUSY);
 215                 }
 216
 217                 /*
 218                  * Do not allow p_type_uuid or p_stor_uuid to change if
 219                  * the partition is currently open.
 220                  */
 221                 if (bcmp(&npp->p_type_uuid, &opp->p_type_uuid,
 222                      sizeof(npp->p_type_uuid)) != 0) {
 223                         return (EBUSY);
 224                 }
 225                 if (bcmp(&npp->p_stor_uuid, &opp->p_stor_uuid,
 226                      sizeof(npp->p_stor_uuid)) != 0) {
 227                         return (EBUSY);
 228                 }
 229                 ++i;
 230         }
 231
 232         /*
 233          * Make sure the label and partition offsets and sizes are sane.
 234          */
 235         if (nlp->d_total_size > slicebsize)
 236                 return (ENOSPC);
 237         if (nlp->d_total_size & (ssp->dss_secsize - 1))
 238                 return (EINVAL);
 239         if (nlp->d_bbase & (ssp->dss_secsize - 1))
 240                 return (EINVAL);
 241         if (nlp->d_pbase & (ssp->dss_secsize - 1))
 242                 return (EINVAL);
 243         if (nlp->d_pstop & (ssp->dss_secsize - 1))
 244                 return (EINVAL);
 245         if (nlp->d_abase & (ssp->dss_secsize - 1))
 246                 return (EINVAL);
 247
 248         for (part = 0; part < nlp->d_npartitions; ++part) {
 249                 npp = &nlp->d_partitions[i];
 250                 if (npp->p_bsize == 0) {
 251                         if (npp->p_boffset != 0)
 252                                 return (EINVAL);
 253                         continue;
 254                 }
 255                 if (npp->p_boffset & (ssp->dss_secsize - 1))
 256                         return (EINVAL);
 257                 if (npp->p_bsize & (ssp->dss_secsize - 1))
 258                         return (EINVAL);
 259                 if (npp->p_boffset < nlp->d_pbase)
 260                         return (ENOSPC);
 261                 if (npp->p_boffset + npp->p_bsize > nlp->d_total_size)
 262                         return (ENOSPC);
 263         }
 264
 265         /*
 266          * Structurally we may add code to make modifications above in the
 267          * future, so regenerate the crc anyway.
 268          */
 269         nlp->d_crc = 0;
 270         nlp->d_crc = crc32(&nlp->d_magic, nlpcrcsize);
 271         *olp = *nlp;
 272
 273         return (0);
 274 }
 275
 276 /*
 277  * Write disk label back to device after modification.
 278  */
 279 static int
 280 l64_writedisklabel(cdev_t dev, struct diskslices *ssp,
 281                    struct diskslice *sp, disklabel_t lpx)
 282 {
 283         struct disklabel64 *lp;
 284         struct disklabel64 *dlp;
 285         struct buf *bp;
 286         int error = 0;
 287         size_t bpsize;
 288         int secsize;
 289
 290         lp = lpx.lab64;
 291
 292         secsize = ssp->dss_secsize;
 293         bpsize = (sizeof(*lp) + secsize - 1) & ~(secsize - 1);
 294
 295         bp = geteblk(bpsize);
 296         bp->b_bio1.bio_offset = 0;
 297         bp->b_bcount = bpsize;
 298
 299         /*
 300          * Because our I/O is larger then the label, and because we do not
 301          * write the d_reserved0[] area, do a read-modify-write.
 302          */
 303         bp->b_flags &= ~B_INVAL;
 304         bp->b_cmd = BUF_CMD_READ;
 305         dev_dstrategy(dkmodpart(dev, WHOLE_SLICE_PART), &bp->b_bio1);
 306         error = biowait(bp);
 307         if (error)
 308                 goto done;
 309
 310         dlp = (void *)bp->b_data;
 311         bcopy(&lp->d_magic, &dlp->d_magic,
 312               sizeof(*lp) - offsetof(struct disklabel64, d_magic));
 313         bp->b_cmd = BUF_CMD_WRITE;
 314         dev_dstrategy(dkmodpart(dev, WHOLE_SLICE_PART), &bp->b_bio1);
 315         error = biowait(bp);
 316 done:
 317         bp->b_flags |= B_INVAL | B_AGE;
 318         brelse(bp);
 319         return (error);
 320 }
 321
 322 /*
 323  * Create a disklabel based on a disk_info structure for the purposes of
 324  * DSO_COMPATLABEL - cases where no real label exists on the storage medium.
 325  *
 326  * If a diskslice is passed, the label is truncated to the slice.
 327  *
 328  * NOTE!  This is not a legal label because d_bbase and d_pbase are both
 329  * set to 0.
 330  */
 331 static disklabel_t
 332 l64_clone_label(struct disk_info *info, struct diskslice *sp)
 333 {
 334         struct disklabel64 *lp;
 335         disklabel_t res;
 336         uint32_t blksize = info->d_media_blksize;
 337         size_t lpcrcsize;
 338
 339         lp = kmalloc(sizeof *lp, M_DEVBUF, M_WAITOK | M_ZERO);
 340
 341         if (sp)
 342                 lp->d_total_size = (uint64_t)sp->ds_size * blksize;
 343         else
 344                 lp->d_total_size = info->d_media_blocks * blksize;
 345
 346         lp->d_magic = DISKMAGIC64;
 347         lp->d_align = blksize;
 348         lp->d_npartitions = MAXPARTITIONS64;
 349         lp->d_pstop = lp->d_total_size;
 350
 351         /*
 352          * Create a dummy 'c' part and a dummy 'a' part (if requested).
 353          * Note that the 'c' part is really a hack.  64 bit disklabels
 354          * do not use 'c' to mean the raw partition.
 355          */
 356
 357         lp->d_partitions[2].p_boffset = 0;
 358         lp->d_partitions[2].p_bsize = lp->d_total_size;
 359         /* XXX SET FS TYPE */
 360
 361         if (info->d_dsflags & DSO_COMPATPARTA) {
 362                 lp->d_partitions[0].p_boffset = 0;
 363                 lp->d_partitions[0].p_bsize = lp->d_total_size;
 364                 /* XXX SET FS TYPE */
 365         }
 366
 367         lpcrcsize = offsetof(struct disklabel64,
 368                              d_partitions[lp->d_npartitions]) -
 369                     offsetof(struct disklabel64, d_magic);
 370
 371         lp->d_crc = crc32(&lp->d_magic, lpcrcsize);
 372         res.lab64 = lp;
 373         return (res);
 374 }
 375
 376 /*
 377  * Create a virgin disklabel64 suitable for writing to the media.
 378  *
 379  * disklabel64 always reserves 32KB for a boot area and leaves room
 380  * for up to RESPARTITIONS64 partitions.
 381  */
 382 static void
 383 l64_makevirginlabel(disklabel_t lpx, struct diskslices *ssp,
 384                     struct diskslice *sp, struct disk_info *info)
 385 {
 386         struct disklabel64 *lp = lpx.lab64;
 387         struct partition64 *pp;
 388         uint32_t blksize;
 389         uint32_t ressize;
 390         uint64_t blkmask;       /* 64 bits so we can ~ */
 391         size_t lpcrcsize;
 392
 393         /*
 394          * Setup the initial label.  Use of a block size of at least 4KB
 395          * for calculating the initial reserved areas to allow some degree
 396          * of portability between media with different sector sizes.
 397          *
 398          * Note that the modified blksize is stored in d_align as a hint
 399          * to the disklabeling program.
 400          */
 401         bzero(lp, sizeof(*lp));
 402         if ((blksize = info->d_media_blksize) < 4096)
 403                 blksize = 4096;
 404         blkmask = blksize - 1;
 405
 406         if (sp)
 407                 lp->d_total_size = (uint64_t)sp->ds_size * ssp->dss_secsize;
 408         else
 409                 lp->d_total_size = info->d_media_blocks * info->d_media_blksize;
 410
 411         lp->d_magic = DISKMAGIC64;
 412         lp->d_align = blksize;
 413         lp->d_npartitions = MAXPARTITIONS64;
 414         kern_uuidgen(&lp->d_obj_uuid, 1);
 415
 416         ressize = offsetof(struct disklabel64, d_partitions[RESPARTITIONS64]);
 417         ressize = (ressize + (uint32_t)blkmask) & ~blkmask;
 418
 419         lp->d_bbase = ressize;
 420         lp->d_pbase = lp->d_bbase + ((32768 + blkmask) & ~blkmask);
 421         lp->d_pstop = (lp->d_total_size - lp->d_bbase) & ~blkmask;
 422         lp->d_abase = lp->d_pstop;
 423
 424         /*
 425          * All partitions are left empty unless DSO_COMPATPARTA is set
 426          */
 427
 428         if (info->d_dsflags & DSO_COMPATPARTA) {
 429                 pp = &lp->d_partitions[0];
 430                 pp->p_boffset = lp->d_pbase;
 431                 pp->p_bsize = lp->d_pstop - lp->d_pbase;
 432                 /* XXX SET FS TYPE */
 433         }
 434
 435         lpcrcsize = offsetof(struct disklabel64,
 436                              d_partitions[lp->d_npartitions]) -
 437                     offsetof(struct disklabel64, d_magic);
 438         lp->d_crc = crc32(&lp->d_magic, lpcrcsize);
 439 }
 440
 441 /*
 442  * Set the number of blocks at the beginning of the slice which have
 443  * been reserved for label operations.  This area will be write-protected
 444  * when accessed via the slice.
 445  *
 446  * For now just protect the label area proper.  Do not protect the
 447  * boot area.  Note partitions in 64 bit disklabels do not overlap
 448  * the disklabel or boot area.
 449  */
 450 static void
 451 l64_adjust_label_reserved(struct diskslices *ssp, int slice,
 452                           struct diskslice *sp)
 453 {
 454         struct disklabel64 *lp = sp->ds_label.lab64;
 455
 456         sp->ds_reserved = lp->d_bbase / ssp->dss_secsize;
 457 }
 458
 459 struct disklabel_ops disklabel64_ops = {
 460         .labelsize = sizeof(struct disklabel64),
 461         .op_readdisklabel = l64_readdisklabel,
 462         .op_setdisklabel = l64_setdisklabel,
 463         .op_writedisklabel = l64_writedisklabel,
 464         .op_clone_label = l64_clone_label,
 465         .op_adjust_label_reserved = l64_adjust_label_reserved,
 466         .op_getpartbounds = l64_getpartbounds,
 467         .op_loadpartinfo = l64_loadpartinfo,
 468         .op_getnumparts = l64_getnumparts,
 469         .op_makevirginlabel = l64_makevirginlabel
 470 };
 471