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 35 #include <sys/param.h> 36 #include <sys/systm.h> 37 #include <sys/kernel.h> 38 #include <sys/conf.h> 39 #include <sys/disklabel.h> 40 #include <sys/disklabel64.h> 41 #include <sys/diskslice.h> 42 #include <sys/disk.h> 43 #include <sys/kern_syscall.h> 44 #include <sys/buf2.h> 45 46 /* 47 * Alignment against physical start (verses slice start). We use a megabyte 48 * here. Why do we use a megabyte? Because SSDs already use large 128K 49 * blocks internally (for MLC) and who the hell knows in the future. 50 * 51 * This way if the sysop picks sane values for partition sizes everything 52 * will be nicely aligned, particularly swap for e.g. swapcache, and 53 * clustered operations against larger physical sector sizes for newer HDs, 54 * and so forth. 55 */ 56 #define PALIGN_SIZE (1024 * 1024) 57 #define PALIGN_MASK (PALIGN_SIZE - 1) 58 59 /* 60 * Retrieve the partition start and extent, in blocks. Return 0 on success, 61 * EINVAL on error. 62 */ 63 static int 64 l64_getpartbounds(struct diskslices *ssp, disklabel_t lp, u_int32_t part, 65 u_int64_t *start, u_int64_t *blocks) 66 { 67 struct partition64 *pp; 68 69 if (part >= lp.lab64->d_npartitions) 70 return (EINVAL); 71 72 pp = &lp.lab64->d_partitions[part]; 73 74 if ((pp->p_boffset & (ssp->dss_secsize - 1)) || 75 (pp->p_bsize & (ssp->dss_secsize - 1))) { 76 return (EINVAL); 77 } 78 *start = pp->p_boffset / ssp->dss_secsize; 79 *blocks = pp->p_bsize / ssp->dss_secsize; 80 return(0); 81 } 82 83 /* 84 * Get the filesystem type XXX - diskslices code needs to use uuids 85 */ 86 static void 87 l64_loadpartinfo(disklabel_t lp, u_int32_t part, struct partinfo *dpart) 88 { 89 struct partition64 *pp; 90 const size_t uuid_size = sizeof(struct uuid); 91 92 if (part < lp.lab64->d_npartitions) { 93 pp = &lp.lab64->d_partitions[part]; 94 dpart->fstype_uuid = pp->p_type_uuid; 95 dpart->storage_uuid = pp->p_stor_uuid; 96 dpart->fstype = pp->p_fstype; 97 } else { 98 bzero(&dpart->fstype_uuid, uuid_size); 99 bzero(&dpart->storage_uuid, uuid_size); 100 dpart->fstype = 0; 101 } 102 } 103 104 /* 105 * Get the number of partitions 106 */ 107 static u_int32_t 108 l64_getnumparts(disklabel_t lp) 109 { 110 return(lp.lab64->d_npartitions); 111 } 112 113 static void 114 l64_freedisklabel(disklabel_t *lpp) 115 { 116 kfree((*lpp).lab64, M_DEVBUF); 117 (*lpp).lab64 = NULL; 118 } 119 120 /* 121 * Attempt to read a disk label from a device. 64 bit disklabels are 122 * sector-agnostic and begin at offset 0 on the device. 64 bit disklabels 123 * may only be used with GPT partitioning schemes. 124 * 125 * Returns NULL on sucess, and an error string on failure. 126 */ 127 static const char * 128 l64_readdisklabel(cdev_t dev, struct diskslice *sp, disklabel_t *lpp, 129 struct disk_info *info) 130 { 131 struct buf *bp; 132 struct disklabel64 *dlp; 133 const char *msg; 134 uint32_t savecrc; 135 size_t dlpcrcsize; 136 size_t bpsize; 137 int secsize; 138 139 /* 140 * XXX I/O size is subject to device DMA limitations 141 */ 142 secsize = info->d_media_blksize; 143 bpsize = roundup2(sizeof(*dlp), secsize); 144 145 bp = getpbuf_mem(NULL); 146 KKASSERT(bpsize <= bp->b_bufsize); 147 bp->b_bio1.bio_offset = 0; 148 bp->b_bio1.bio_done = biodone_sync; 149 bp->b_bio1.bio_flags |= BIO_SYNC; 150 bp->b_bcount = bpsize; 151 bp->b_flags &= ~B_INVAL; 152 bp->b_flags |= B_FAILONDIS; 153 bp->b_cmd = BUF_CMD_READ; 154 dev_dstrategy(dev, &bp->b_bio1); 155 156 if (biowait(&bp->b_bio1, "labrd")) { 157 msg = "I/O error"; 158 } else { 159 dlp = (struct disklabel64 *)bp->b_data; 160 dlpcrcsize = offsetof(struct disklabel64, 161 d_partitions[dlp->d_npartitions]) - 162 offsetof(struct disklabel64, d_magic); 163 savecrc = dlp->d_crc; 164 dlp->d_crc = 0; 165 if (dlp->d_magic != DISKMAGIC64) { 166 msg = "no disk label"; 167 } else if (dlp->d_npartitions > MAXPARTITIONS64) { 168 msg = "disklabel64 corrupted, too many partitions"; 169 } else if (savecrc != crc32(&dlp->d_magic, dlpcrcsize)) { 170 msg = "disklabel64 corrupted, bad CRC"; 171 } else { 172 dlp->d_crc = savecrc; 173 (*lpp).lab64 = kmalloc(sizeof(*dlp), 174 M_DEVBUF, M_WAITOK|M_ZERO); 175 *(*lpp).lab64 = *dlp; 176 msg = NULL; 177 } 178 } 179 bp->b_flags |= B_INVAL | B_AGE; 180 relpbuf(bp, NULL); 181 182 return (msg); 183 } 184 185 /* 186 * If everything is good, copy olpx to nlpx. Check to see if any 187 * open partitions would change. 188 */ 189 static int 190 l64_setdisklabel(disklabel_t olpx, disklabel_t nlpx, struct diskslices *ssp, 191 struct diskslice *sp, u_int32_t *openmask) 192 { 193 struct disklabel64 *olp, *nlp; 194 struct partition64 *opp, *npp; 195 uint32_t savecrc; 196 uint64_t slicebsize; 197 size_t nlpcrcsize; 198 int i; 199 200 olp = olpx.lab64; 201 nlp = nlpx.lab64; 202 203 slicebsize = (uint64_t)sp->ds_size * ssp->dss_secsize; 204 205 if (nlp->d_magic != DISKMAGIC64) 206 return (EINVAL); 207 if (nlp->d_npartitions > MAXPARTITIONS64) 208 return (EINVAL); 209 savecrc = nlp->d_crc; 210 nlp->d_crc = 0; 211 nlpcrcsize = offsetof(struct disklabel64, 212 d_partitions[nlp->d_npartitions]) - 213 offsetof(struct disklabel64, d_magic); 214 if (crc32(&nlp->d_magic, nlpcrcsize) != savecrc) { 215 nlp->d_crc = savecrc; 216 return (EINVAL); 217 } 218 nlp->d_crc = savecrc; 219 220 /* 221 * Check if open partitions have changed 222 */ 223 i = 0; 224 while (i < MAXPARTITIONS64) { 225 if (openmask[i >> 5] == 0) { 226 i += 32; 227 continue; 228 } 229 if ((openmask[i >> 5] & (1 << (i & 31))) == 0) { 230 ++i; 231 continue; 232 } 233 if (nlp->d_npartitions <= i) 234 return (EBUSY); 235 opp = &olp->d_partitions[i]; 236 npp = &nlp->d_partitions[i]; 237 if (npp->p_boffset != opp->p_boffset || 238 npp->p_bsize < opp->p_bsize) { 239 return (EBUSY); 240 } 241 242 /* 243 * Do not allow p_type_uuid or p_stor_uuid to change if 244 * the partition is currently open. 245 */ 246 if (bcmp(&npp->p_type_uuid, &opp->p_type_uuid, 247 sizeof(npp->p_type_uuid)) != 0) { 248 return (EBUSY); 249 } 250 if (bcmp(&npp->p_stor_uuid, &opp->p_stor_uuid, 251 sizeof(npp->p_stor_uuid)) != 0) { 252 return (EBUSY); 253 } 254 ++i; 255 } 256 257 /* 258 * Make sure the label and partition offsets and sizes are sane. 259 */ 260 if (nlp->d_total_size > slicebsize) 261 return (ENOSPC); 262 if (nlp->d_total_size & (ssp->dss_secsize - 1)) 263 return (EINVAL); 264 if (nlp->d_bbase & (ssp->dss_secsize - 1)) 265 return (EINVAL); 266 if (nlp->d_pbase & (ssp->dss_secsize - 1)) 267 return (EINVAL); 268 if (nlp->d_pstop & (ssp->dss_secsize - 1)) 269 return (EINVAL); 270 if (nlp->d_abase & (ssp->dss_secsize - 1)) 271 return (EINVAL); 272 273 for (i = 0; i < nlp->d_npartitions; ++i) { 274 npp = &nlp->d_partitions[i]; 275 if (npp->p_bsize == 0) { 276 if (npp->p_boffset != 0) 277 return (EINVAL); 278 continue; 279 } 280 if (npp->p_boffset & (ssp->dss_secsize - 1)) 281 return (EINVAL); 282 if (npp->p_bsize & (ssp->dss_secsize - 1)) 283 return (EINVAL); 284 if (npp->p_boffset < nlp->d_pbase) 285 return (ENOSPC); 286 if (npp->p_boffset + npp->p_bsize > nlp->d_total_size) 287 return (ENOSPC); 288 } 289 290 /* 291 * Structurally we may add code to make modifications above in the 292 * future, so regenerate the crc anyway. 293 */ 294 nlp->d_crc = 0; 295 nlp->d_crc = crc32(&nlp->d_magic, nlpcrcsize); 296 *olp = *nlp; 297 298 return (0); 299 } 300 301 /* 302 * Write disk label back to device after modification. 303 */ 304 static int 305 l64_writedisklabel(cdev_t dev, struct diskslices *ssp, 306 struct diskslice *sp, disklabel_t lpx) 307 { 308 struct disklabel64 *lp; 309 struct disklabel64 *dlp; 310 struct buf *bp; 311 int error = 0; 312 size_t bpsize; 313 int secsize; 314 315 lp = lpx.lab64; 316 317 /* 318 * XXX I/O size is subject to device DMA limitations 319 */ 320 secsize = ssp->dss_secsize; 321 bpsize = roundup2(sizeof(*lp), secsize); 322 323 bp = getpbuf_mem(NULL); 324 KKASSERT(bpsize <= bp->b_bufsize); 325 bp->b_bio1.bio_offset = 0; 326 bp->b_bio1.bio_done = biodone_sync; 327 bp->b_bio1.bio_flags |= BIO_SYNC; 328 bp->b_bcount = bpsize; 329 bp->b_flags |= B_FAILONDIS; 330 331 /* 332 * Because our I/O is larger then the label, and because we do not 333 * write the d_reserved0[] area, do a read-modify-write. 334 */ 335 bp->b_flags &= ~B_INVAL; 336 bp->b_cmd = BUF_CMD_READ; 337 KKASSERT(dkpart(dev) == WHOLE_SLICE_PART); 338 dev_dstrategy(dev, &bp->b_bio1); 339 error = biowait(&bp->b_bio1, "labrd"); 340 if (error) 341 goto done; 342 343 dlp = (void *)bp->b_data; 344 bcopy(&lp->d_magic, &dlp->d_magic, 345 sizeof(*lp) - offsetof(struct disklabel64, d_magic)); 346 bp->b_cmd = BUF_CMD_WRITE; 347 bp->b_bio1.bio_done = biodone_sync; 348 bp->b_bio1.bio_flags |= BIO_SYNC; 349 KKASSERT(dkpart(dev) == WHOLE_SLICE_PART); 350 dev_dstrategy(dev, &bp->b_bio1); 351 error = biowait(&bp->b_bio1, "labwr"); 352 done: 353 bp->b_flags |= B_INVAL | B_AGE; 354 relpbuf(bp, NULL); 355 356 return (error); 357 } 358 359 /* 360 * Create a disklabel based on a disk_info structure for the purposes of 361 * DSO_COMPATLABEL - cases where no real label exists on the storage medium. 362 * 363 * If a diskslice is passed, the label is truncated to the slice. 364 * 365 * NOTE! This is not a legal label because d_bbase and d_pbase are both 366 * set to 0. 367 */ 368 static disklabel_t 369 l64_clone_label(struct disk_info *info, struct diskslice *sp) 370 { 371 struct disklabel64 *lp; 372 disklabel_t res; 373 uint32_t blksize = info->d_media_blksize; 374 size_t lpcrcsize; 375 376 lp = kmalloc(sizeof *lp, M_DEVBUF, M_WAITOK | M_ZERO); 377 378 if (sp) 379 lp->d_total_size = (uint64_t)sp->ds_size * blksize; 380 else 381 lp->d_total_size = info->d_media_blocks * blksize; 382 383 lp->d_magic = DISKMAGIC64; 384 lp->d_align = blksize; 385 lp->d_npartitions = MAXPARTITIONS64; 386 lp->d_pstop = lp->d_total_size; 387 388 /* 389 * Create a dummy 'c' part and a dummy 'a' part (if requested). 390 * Note that the 'c' part is really a hack. 64 bit disklabels 391 * do not use 'c' to mean the raw partition. 392 */ 393 394 lp->d_partitions[2].p_boffset = 0; 395 lp->d_partitions[2].p_bsize = lp->d_total_size; 396 /* XXX SET FS TYPE */ 397 398 if (info->d_dsflags & DSO_COMPATPARTA) { 399 lp->d_partitions[0].p_boffset = 0; 400 lp->d_partitions[0].p_bsize = lp->d_total_size; 401 /* XXX SET FS TYPE */ 402 } 403 404 lpcrcsize = offsetof(struct disklabel64, 405 d_partitions[lp->d_npartitions]) - 406 offsetof(struct disklabel64, d_magic); 407 408 lp->d_crc = crc32(&lp->d_magic, lpcrcsize); 409 res.lab64 = lp; 410 return (res); 411 } 412 413 /* 414 * Create a virgin disklabel64 suitable for writing to the media. 415 * 416 * disklabel64 always reserves 32KB for a boot area and leaves room 417 * for up to RESPARTITIONS64 partitions. 418 */ 419 static void 420 l64_makevirginlabel(disklabel_t lpx, struct diskslices *ssp, 421 struct diskslice *sp, struct disk_info *info) 422 { 423 struct disklabel64 *lp = lpx.lab64; 424 struct partition64 *pp; 425 uint32_t blksize; 426 uint32_t ressize; 427 uint64_t blkmask; /* 64 bits so we can ~ */ 428 size_t lpcrcsize; 429 430 /* 431 * Setup the initial label. Use of a block size of at least 4KB 432 * for calculating the initial reserved areas to allow some degree 433 * of portability between media with different sector sizes. 434 * 435 * Note that the modified blksize is stored in d_align as a hint 436 * to the disklabeling program. 437 */ 438 bzero(lp, sizeof(*lp)); 439 if ((blksize = info->d_media_blksize) < 4096) 440 blksize = 4096; 441 blkmask = blksize - 1; 442 443 if (sp) 444 lp->d_total_size = (uint64_t)sp->ds_size * ssp->dss_secsize; 445 else 446 lp->d_total_size = info->d_media_blocks * info->d_media_blksize; 447 448 lp->d_magic = DISKMAGIC64; 449 lp->d_align = blksize; 450 lp->d_npartitions = MAXPARTITIONS64; 451 kern_uuidgen(&lp->d_stor_uuid, 1); 452 453 ressize = offsetof(struct disklabel64, d_partitions[RESPARTITIONS64]); 454 ressize = (ressize + (uint32_t)blkmask) & ~blkmask; 455 456 /* 457 * NOTE: When calculating pbase take into account the slice offset 458 * so the partitions are at least 32K-aligned relative to the 459 * start of the physical disk. This will accomodate efficient 460 * access to 4096 byte physical sector drives. 461 */ 462 lp->d_bbase = ressize; 463 lp->d_pbase = lp->d_bbase + ((32768 + blkmask) & ~blkmask); 464 lp->d_pbase = (lp->d_pbase + PALIGN_MASK) & ~(uint64_t)PALIGN_MASK; 465 466 /* adjust for slice offset so we are physically aligned */ 467 lp->d_pbase += 32768 - (sp->ds_offset * info->d_media_blksize) % 32768; 468 469 lp->d_pstop = (lp->d_total_size - lp->d_bbase) & ~blkmask; 470 lp->d_abase = lp->d_pstop; 471 472 /* 473 * All partitions are left empty unless DSO_COMPATPARTA is set 474 */ 475 476 if (info->d_dsflags & DSO_COMPATPARTA) { 477 pp = &lp->d_partitions[0]; 478 pp->p_boffset = lp->d_pbase; 479 pp->p_bsize = lp->d_pstop - lp->d_pbase; 480 /* XXX SET FS TYPE */ 481 } 482 483 lpcrcsize = offsetof(struct disklabel64, 484 d_partitions[lp->d_npartitions]) - 485 offsetof(struct disklabel64, d_magic); 486 lp->d_crc = crc32(&lp->d_magic, lpcrcsize); 487 } 488 489 /* 490 * Set the number of blocks at the beginning of the slice which have 491 * been reserved for label operations. This area will be write-protected 492 * when accessed via the slice. 493 * 494 * For now just protect the label area proper. Do not protect the 495 * boot area. Note partitions in 64 bit disklabels do not overlap 496 * the disklabel or boot area. 497 */ 498 static void 499 l64_adjust_label_reserved(struct diskslices *ssp, int slice, 500 struct diskslice *sp) 501 { 502 struct disklabel64 *lp = sp->ds_label.lab64; 503 504 sp->ds_reserved = lp->d_bbase / ssp->dss_secsize; 505 } 506 507 struct disklabel_ops disklabel64_ops = { 508 .labelsize = sizeof(struct disklabel64), 509 .op_readdisklabel = l64_readdisklabel, 510 .op_setdisklabel = l64_setdisklabel, 511 .op_writedisklabel = l64_writedisklabel, 512 .op_clone_label = l64_clone_label, 513 .op_adjust_label_reserved = l64_adjust_label_reserved, 514 .op_getpartbounds = l64_getpartbounds, 515 .op_loadpartinfo = l64_loadpartinfo, 516 .op_getnumparts = l64_getnumparts, 517 .op_makevirginlabel = l64_makevirginlabel, 518 .op_freedisklabel = l64_freedisklabel 519 }; 520 521