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_bio1.bio_done = biodone_sync;
 130         bp->b_bio1.bio_flags |= BIO_SYNC;
 131         bp->b_bcount = bpsize;
 132         bp->b_flags &= ~B_INVAL;
 133         bp->b_cmd = BUF_CMD_READ;
 134         dev_dstrategy(dev, &bp->b_bio1);
 135
 136         if (biowait(&bp->b_bio1, "labrd")) {
 137                 msg = "I/O error";
 138         } else {
 139                 dlp = (struct disklabel64 *)bp->b_data;
 140                 dlpcrcsize = offsetof(struct disklabel64,
 141                                       d_partitions[dlp->d_npartitions]) -
 142                              offsetof(struct disklabel64, d_magic);
 143                 savecrc = dlp->d_crc;
 144                 dlp->d_crc = 0;
 145                 if (dlp->d_magic != DISKMAGIC64) {
 146                         msg = "no disk label";
 147                 } else if (dlp->d_npartitions > MAXPARTITIONS64) {
 148                         msg = "disklabel64 corrupted, too many partitions";
 149                 } else if (savecrc != crc32(&dlp->d_magic, dlpcrcsize)) {
 150                         msg = "disklabel64 corrupted, bad CRC";
 151                 } else {
 152                         dlp->d_crc = savecrc;
 153                         (*lpp).lab64 = kmalloc(sizeof(*dlp),
 154                                                M_DEVBUF, M_WAITOK|M_ZERO);
 155                         *(*lpp).lab64 = *dlp;
 156                         msg = NULL;
 157                 }
 158         }
 159         bp->b_flags |= B_INVAL | B_AGE;
 160         brelse(bp);
 161         return (msg);
 162 }
 163
 164 /*
 165  * If everything is good, copy olpx to nlpx.  Check to see if any
 166  * open partitions would change.
 167  */
 168 static int
 169 l64_setdisklabel(disklabel_t olpx, disklabel_t nlpx, struct diskslices *ssp,
 170                  struct diskslice *sp, u_int32_t *openmask)
 171 {
 172         struct disklabel64 *olp, *nlp;
 173         struct partition64 *opp, *npp;
 174         uint32_t savecrc;
 175         uint64_t slicebsize;
 176         size_t nlpcrcsize;
 177         int i;
 178
 179         olp = olpx.lab64;
 180         nlp = nlpx.lab64;
 181
 182         slicebsize = (uint64_t)sp->ds_size * ssp->dss_secsize;
 183
 184         if (nlp->d_magic != DISKMAGIC64)
 185                 return (EINVAL);
 186         if (nlp->d_npartitions > MAXPARTITIONS64)
 187                 return (EINVAL);
 188         savecrc = nlp->d_crc;
 189         nlp->d_crc = 0;
 190         nlpcrcsize = offsetof(struct disklabel64,
 191                               d_partitions[nlp->d_npartitions]) -
 192                      offsetof(struct disklabel64, d_magic);
 193         if (crc32(&nlp->d_magic, nlpcrcsize) != savecrc) {
 194                 nlp->d_crc = savecrc;
 195                 return (EINVAL);
 196         }
 197         nlp->d_crc = savecrc;
 198
 199         /*
 200          * Check if open partitions have changed
 201          */
 202         i = 0;
 203         while (i < MAXPARTITIONS64) {
 204                 if (openmask[i >> 5] == 0) {
 205                         i += 32;
 206                         continue;
 207                 }
 208                 if ((openmask[i >> 5] & (1 << (i & 31))) == 0) {
 209                         ++i;
 210                         continue;
 211                 }
 212                 if (nlp->d_npartitions <= i)
 213                         return (EBUSY);
 214                 opp = &olp->d_partitions[i];
 215                 npp = &nlp->d_partitions[i];
 216                 if (npp->p_boffset != opp->p_boffset ||
 217                     npp->p_bsize < opp->p_bsize) {
 218                         return (EBUSY);
 219                 }
 220
 221                 /*
 222                  * Do not allow p_type_uuid or p_stor_uuid to change if
 223                  * the partition is currently open.
 224                  */
 225                 if (bcmp(&npp->p_type_uuid, &opp->p_type_uuid,
 226                      sizeof(npp->p_type_uuid)) != 0) {
 227                         return (EBUSY);
 228                 }
 229                 if (bcmp(&npp->p_stor_uuid, &opp->p_stor_uuid,
 230                      sizeof(npp->p_stor_uuid)) != 0) {
 231                         return (EBUSY);
 232                 }
 233                 ++i;
 234         }
 235
 236         /*
 237          * Make sure the label and partition offsets and sizes are sane.
 238          */
 239         if (nlp->d_total_size > slicebsize)
 240                 return (ENOSPC);
 241         if (nlp->d_total_size & (ssp->dss_secsize - 1))
 242                 return (EINVAL);
 243         if (nlp->d_bbase & (ssp->dss_secsize - 1))
 244                 return (EINVAL);
 245         if (nlp->d_pbase & (ssp->dss_secsize - 1))
 246                 return (EINVAL);
 247         if (nlp->d_pstop & (ssp->dss_secsize - 1))
 248                 return (EINVAL);
 249         if (nlp->d_abase & (ssp->dss_secsize - 1))
 250                 return (EINVAL);
 251
 252         for (i = 0; i < nlp->d_npartitions; ++i) {
 253                 npp = &nlp->d_partitions[i];
 254                 if (npp->p_bsize == 0) {
 255                         if (npp->p_boffset != 0)
 256                                 return (EINVAL);
 257                         continue;
 258                 }
 259                 if (npp->p_boffset & (ssp->dss_secsize - 1))
 260                         return (EINVAL);
 261                 if (npp->p_bsize & (ssp->dss_secsize - 1))
 262                         return (EINVAL);
 263                 if (npp->p_boffset < nlp->d_pbase)
 264                         return (ENOSPC);
 265                 if (npp->p_boffset + npp->p_bsize > nlp->d_total_size)
 266                         return (ENOSPC);
 267         }
 268
 269         /*
 270          * Structurally we may add code to make modifications above in the
 271          * future, so regenerate the crc anyway.
 272          */
 273         nlp->d_crc = 0;
 274         nlp->d_crc = crc32(&nlp->d_magic, nlpcrcsize);
 275         *olp = *nlp;
 276
 277         return (0);
 278 }
 279
 280 /*
 281  * Write disk label back to device after modification.
 282  */
 283 static int
 284 l64_writedisklabel(cdev_t dev, struct diskslices *ssp,
 285                    struct diskslice *sp, disklabel_t lpx)
 286 {
 287         struct disklabel64 *lp;
 288         struct disklabel64 *dlp;
 289         struct buf *bp;
 290         int error = 0;
 291         size_t bpsize;
 292         int secsize;
 293
 294         lp = lpx.lab64;
 295
 296         /*
 297          * XXX I/O size is subject to device DMA limitations
 298          */
 299         secsize = ssp->dss_secsize;
 300         bpsize = (sizeof(*lp) + secsize - 1) & ~(secsize - 1);
 301
 302         bp = geteblk(bpsize);
 303         bp->b_bio1.bio_offset = 0;
 304         bp->b_bio1.bio_done = biodone_sync;
 305         bp->b_bio1.bio_flags |= BIO_SYNC;
 306         bp->b_bcount = bpsize;
 307
 308         /*
 309          * Because our I/O is larger then the label, and because we do not
 310          * write the d_reserved0[] area, do a read-modify-write.
 311          */
 312         bp->b_flags &= ~B_INVAL;
 313         bp->b_cmd = BUF_CMD_READ;
 314         KKASSERT(dkpart(dev) == WHOLE_SLICE_PART);
 315         dev_dstrategy(dev, &bp->b_bio1);
 316         error = biowait(&bp->b_bio1, "labrd");
 317         if (error)
 318                 goto done;
 319
 320         dlp = (void *)bp->b_data;
 321         bcopy(&lp->d_magic, &dlp->d_magic,
 322               sizeof(*lp) - offsetof(struct disklabel64, d_magic));
 323         bp->b_cmd = BUF_CMD_WRITE;
 324         bp->b_bio1.bio_done = biodone_sync;
 325         bp->b_bio1.bio_flags |= BIO_SYNC;
 326         KKASSERT(dkpart(dev) == WHOLE_SLICE_PART);
 327         dev_dstrategy(dev, &bp->b_bio1);
 328         error = biowait(&bp->b_bio1, "labwr");
 329 done:
 330         bp->b_flags |= B_INVAL | B_AGE;
 331         brelse(bp);
 332         return (error);
 333 }
 334
 335 /*
 336  * Create a disklabel based on a disk_info structure for the purposes of
 337  * DSO_COMPATLABEL - cases where no real label exists on the storage medium.
 338  *
 339  * If a diskslice is passed, the label is truncated to the slice.
 340  *
 341  * NOTE!  This is not a legal label because d_bbase and d_pbase are both
 342  * set to 0.
 343  */
 344 static disklabel_t
 345 l64_clone_label(struct disk_info *info, struct diskslice *sp)
 346 {
 347         struct disklabel64 *lp;
 348         disklabel_t res;
 349         uint32_t blksize = info->d_media_blksize;
 350         size_t lpcrcsize;
 351
 352         lp = kmalloc(sizeof *lp, M_DEVBUF, M_WAITOK | M_ZERO);
 353
 354         if (sp)
 355                 lp->d_total_size = (uint64_t)sp->ds_size * blksize;
 356         else
 357                 lp->d_total_size = info->d_media_blocks * blksize;
 358
 359         lp->d_magic = DISKMAGIC64;
 360         lp->d_align = blksize;
 361         lp->d_npartitions = MAXPARTITIONS64;
 362         lp->d_pstop = lp->d_total_size;
 363
 364         /*
 365          * Create a dummy 'c' part and a dummy 'a' part (if requested).
 366          * Note that the 'c' part is really a hack.  64 bit disklabels
 367          * do not use 'c' to mean the raw partition.
 368          */
 369
 370         lp->d_partitions[2].p_boffset = 0;
 371         lp->d_partitions[2].p_bsize = lp->d_total_size;
 372         /* XXX SET FS TYPE */
 373
 374         if (info->d_dsflags & DSO_COMPATPARTA) {
 375                 lp->d_partitions[0].p_boffset = 0;
 376                 lp->d_partitions[0].p_bsize = lp->d_total_size;
 377                 /* XXX SET FS TYPE */
 378         }
 379
 380         lpcrcsize = offsetof(struct disklabel64,
 381                              d_partitions[lp->d_npartitions]) -
 382                     offsetof(struct disklabel64, d_magic);
 383
 384         lp->d_crc = crc32(&lp->d_magic, lpcrcsize);
 385         res.lab64 = lp;
 386         return (res);
 387 }
 388
 389 /*
 390  * Create a virgin disklabel64 suitable for writing to the media.
 391  *
 392  * disklabel64 always reserves 32KB for a boot area and leaves room
 393  * for up to RESPARTITIONS64 partitions.
 394  */
 395 static void
 396 l64_makevirginlabel(disklabel_t lpx, struct diskslices *ssp,
 397                     struct diskslice *sp, struct disk_info *info)
 398 {
 399         struct disklabel64 *lp = lpx.lab64;
 400         struct partition64 *pp;
 401         uint32_t blksize;
 402         uint32_t ressize;
 403         uint64_t blkmask;       /* 64 bits so we can ~ */
 404         size_t lpcrcsize;
 405
 406         /*
 407          * Setup the initial label.  Use of a block size of at least 4KB
 408          * for calculating the initial reserved areas to allow some degree
 409          * of portability between media with different sector sizes.
 410          *
 411          * Note that the modified blksize is stored in d_align as a hint
 412          * to the disklabeling program.
 413          */
 414         bzero(lp, sizeof(*lp));
 415         if ((blksize = info->d_media_blksize) < 4096)
 416                 blksize = 4096;
 417         blkmask = blksize - 1;
 418
 419         if (sp)
 420                 lp->d_total_size = (uint64_t)sp->ds_size * ssp->dss_secsize;
 421         else
 422                 lp->d_total_size = info->d_media_blocks * info->d_media_blksize;
 423
 424         lp->d_magic = DISKMAGIC64;
 425         lp->d_align = blksize;
 426         lp->d_npartitions = MAXPARTITIONS64;
 427         kern_uuidgen(&lp->d_stor_uuid, 1);
 428
 429         ressize = offsetof(struct disklabel64, d_partitions[RESPARTITIONS64]);
 430         ressize = (ressize + (uint32_t)blkmask) & ~blkmask;
 431
 432         lp->d_bbase = ressize;
 433         lp->d_pbase = lp->d_bbase + ((32768 + blkmask) & ~blkmask);
 434         lp->d_pstop = (lp->d_total_size - lp->d_bbase) & ~blkmask;
 435         lp->d_abase = lp->d_pstop;
 436
 437         /*
 438          * All partitions are left empty unless DSO_COMPATPARTA is set
 439          */
 440
 441         if (info->d_dsflags & DSO_COMPATPARTA) {
 442                 pp = &lp->d_partitions[0];
 443                 pp->p_boffset = lp->d_pbase;
 444                 pp->p_bsize = lp->d_pstop - lp->d_pbase;
 445                 /* XXX SET FS TYPE */
 446         }
 447
 448         lpcrcsize = offsetof(struct disklabel64,
 449                              d_partitions[lp->d_npartitions]) -
 450                     offsetof(struct disklabel64, d_magic);
 451         lp->d_crc = crc32(&lp->d_magic, lpcrcsize);
 452 }
 453
 454 /*
 455  * Set the number of blocks at the beginning of the slice which have
 456  * been reserved for label operations.  This area will be write-protected
 457  * when accessed via the slice.
 458  *
 459  * For now just protect the label area proper.  Do not protect the
 460  * boot area.  Note partitions in 64 bit disklabels do not overlap
 461  * the disklabel or boot area.
 462  */
 463 static void
 464 l64_adjust_label_reserved(struct diskslices *ssp, int slice,
 465                           struct diskslice *sp)
 466 {
 467         struct disklabel64 *lp = sp->ds_label.lab64;
 468
 469         sp->ds_reserved = lp->d_bbase / ssp->dss_secsize;
 470 }
 471
 472 struct disklabel_ops disklabel64_ops = {
 473         .labelsize = sizeof(struct disklabel64),
 474         .op_readdisklabel = l64_readdisklabel,
 475         .op_setdisklabel = l64_setdisklabel,
 476         .op_writedisklabel = l64_writedisklabel,
 477         .op_clone_label = l64_clone_label,
 478         .op_adjust_label_reserved = l64_adjust_label_reserved,
 479         .op_getpartbounds = l64_getpartbounds,
 480         .op_loadpartinfo = l64_loadpartinfo,
 481         .op_getnumparts = l64_getnumparts,
 482         .op_makevirginlabel = l64_makevirginlabel
 483 };
 484