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 = geteblk(bpsize); 146 bp->b_bio1.bio_offset = 0; 147 bp->b_bio1.bio_done = biodone_sync; 148 bp->b_bio1.bio_flags |= BIO_SYNC; 149 bp->b_bcount = bpsize; 150 bp->b_flags &= ~B_INVAL; 151 bp->b_flags |= B_FAILONDIS; 152 bp->b_cmd = BUF_CMD_READ; 153 dev_dstrategy(dev, &bp->b_bio1); 154 155 if (biowait(&bp->b_bio1, "labrd")) { 156 msg = "I/O error"; 157 } else { 158 dlp = (struct disklabel64 *)bp->b_data; 159 dlpcrcsize = offsetof(struct disklabel64, 160 d_partitions[dlp->d_npartitions]) - 161 offsetof(struct disklabel64, d_magic); 162 savecrc = dlp->d_crc; 163 dlp->d_crc = 0; 164 if (dlp->d_magic != DISKMAGIC64) { 165 msg = "no disk label"; 166 } else if (dlp->d_npartitions > MAXPARTITIONS64) { 167 msg = "disklabel64 corrupted, too many partitions"; 168 } else if (savecrc != crc32(&dlp->d_magic, dlpcrcsize)) { 169 msg = "disklabel64 corrupted, bad CRC"; 170 } else { 171 dlp->d_crc = savecrc; 172 (*lpp).lab64 = kmalloc(sizeof(*dlp), 173 M_DEVBUF, M_WAITOK|M_ZERO); 174 *(*lpp).lab64 = *dlp; 175 msg = NULL; 176 } 177 } 178 bp->b_flags |= B_INVAL | B_AGE; 179 brelse(bp); 180 return (msg); 181 } 182 183 /* 184 * If everything is good, copy olpx to nlpx. Check to see if any 185 * open partitions would change. 186 */ 187 static int 188 l64_setdisklabel(disklabel_t olpx, disklabel_t nlpx, struct diskslices *ssp, 189 struct diskslice *sp, u_int32_t *openmask) 190 { 191 struct disklabel64 *olp, *nlp; 192 struct partition64 *opp, *npp; 193 uint32_t savecrc; 194 uint64_t slicebsize; 195 size_t nlpcrcsize; 196 int i; 197 198 olp = olpx.lab64; 199 nlp = nlpx.lab64; 200 201 slicebsize = (uint64_t)sp->ds_size * ssp->dss_secsize; 202 203 if (nlp->d_magic != DISKMAGIC64) 204 return (EINVAL); 205 if (nlp->d_npartitions > MAXPARTITIONS64) 206 return (EINVAL); 207 savecrc = nlp->d_crc; 208 nlp->d_crc = 0; 209 nlpcrcsize = offsetof(struct disklabel64, 210 d_partitions[nlp->d_npartitions]) - 211 offsetof(struct disklabel64, d_magic); 212 if (crc32(&nlp->d_magic, nlpcrcsize) != savecrc) { 213 nlp->d_crc = savecrc; 214 return (EINVAL); 215 } 216 nlp->d_crc = savecrc; 217 218 /* 219 * Check if open partitions have changed 220 */ 221 i = 0; 222 while (i < MAXPARTITIONS64) { 223 if (openmask[i >> 5] == 0) { 224 i += 32; 225 continue; 226 } 227 if ((openmask[i >> 5] & (1 << (i & 31))) == 0) { 228 ++i; 229 continue; 230 } 231 if (nlp->d_npartitions <= i) 232 return (EBUSY); 233 opp = &olp->d_partitions[i]; 234 npp = &nlp->d_partitions[i]; 235 if (npp->p_boffset != opp->p_boffset || 236 npp->p_bsize < opp->p_bsize) { 237 return (EBUSY); 238 } 239 240 /* 241 * Do not allow p_type_uuid or p_stor_uuid to change if 242 * the partition is currently open. 243 */ 244 if (bcmp(&npp->p_type_uuid, &opp->p_type_uuid, 245 sizeof(npp->p_type_uuid)) != 0) { 246 return (EBUSY); 247 } 248 if (bcmp(&npp->p_stor_uuid, &opp->p_stor_uuid, 249 sizeof(npp->p_stor_uuid)) != 0) { 250 return (EBUSY); 251 } 252 ++i; 253 } 254 255 /* 256 * Make sure the label and partition offsets and sizes are sane. 257 */ 258 if (nlp->d_total_size > slicebsize) 259 return (ENOSPC); 260 if (nlp->d_total_size & (ssp->dss_secsize - 1)) 261 return (EINVAL); 262 if (nlp->d_bbase & (ssp->dss_secsize - 1)) 263 return (EINVAL); 264 if (nlp->d_pbase & (ssp->dss_secsize - 1)) 265 return (EINVAL); 266 if (nlp->d_pstop & (ssp->dss_secsize - 1)) 267 return (EINVAL); 268 if (nlp->d_abase & (ssp->dss_secsize - 1)) 269 return (EINVAL); 270 271 for (i = 0; i < nlp->d_npartitions; ++i) { 272 npp = &nlp->d_partitions[i]; 273 if (npp->p_bsize == 0) { 274 if (npp->p_boffset != 0) 275 return (EINVAL); 276 continue; 277 } 278 if (npp->p_boffset & (ssp->dss_secsize - 1)) 279 return (EINVAL); 280 if (npp->p_bsize & (ssp->dss_secsize - 1)) 281 return (EINVAL); 282 if (npp->p_boffset < nlp->d_pbase) 283 return (ENOSPC); 284 if (npp->p_boffset + npp->p_bsize > nlp->d_total_size) 285 return (ENOSPC); 286 } 287 288 /* 289 * Structurally we may add code to make modifications above in the 290 * future, so regenerate the crc anyway. 291 */ 292 nlp->d_crc = 0; 293 nlp->d_crc = crc32(&nlp->d_magic, nlpcrcsize); 294 *olp = *nlp; 295 296 return (0); 297 } 298 299 /* 300 * Write disk label back to device after modification. 301 */ 302 static int 303 l64_writedisklabel(cdev_t dev, struct diskslices *ssp, 304 struct diskslice *sp, disklabel_t lpx) 305 { 306 struct disklabel64 *lp; 307 struct disklabel64 *dlp; 308 struct buf *bp; 309 int error = 0; 310 size_t bpsize; 311 int secsize; 312 313 lp = lpx.lab64; 314 315 /* 316 * XXX I/O size is subject to device DMA limitations 317 */ 318 secsize = ssp->dss_secsize; 319 bpsize = roundup2(sizeof(*lp), secsize); 320 321 bp = geteblk(bpsize); 322 bp->b_bio1.bio_offset = 0; 323 bp->b_bio1.bio_done = biodone_sync; 324 bp->b_bio1.bio_flags |= BIO_SYNC; 325 bp->b_bcount = bpsize; 326 bp->b_flags |= B_FAILONDIS; 327 328 /* 329 * Because our I/O is larger then the label, and because we do not 330 * write the d_reserved0[] area, do a read-modify-write. 331 */ 332 bp->b_flags &= ~B_INVAL; 333 bp->b_cmd = BUF_CMD_READ; 334 KKASSERT(dkpart(dev) == WHOLE_SLICE_PART); 335 dev_dstrategy(dev, &bp->b_bio1); 336 error = biowait(&bp->b_bio1, "labrd"); 337 if (error) 338 goto done; 339 340 dlp = (void *)bp->b_data; 341 bcopy(&lp->d_magic, &dlp->d_magic, 342 sizeof(*lp) - offsetof(struct disklabel64, d_magic)); 343 bp->b_cmd = BUF_CMD_WRITE; 344 bp->b_bio1.bio_done = biodone_sync; 345 bp->b_bio1.bio_flags |= BIO_SYNC; 346 KKASSERT(dkpart(dev) == WHOLE_SLICE_PART); 347 dev_dstrategy(dev, &bp->b_bio1); 348 error = biowait(&bp->b_bio1, "labwr"); 349 done: 350 bp->b_flags |= B_INVAL | B_AGE; 351 brelse(bp); 352 return (error); 353 } 354 355 /* 356 * Create a disklabel based on a disk_info structure for the purposes of 357 * DSO_COMPATLABEL - cases where no real label exists on the storage medium. 358 * 359 * If a diskslice is passed, the label is truncated to the slice. 360 * 361 * NOTE! This is not a legal label because d_bbase and d_pbase are both 362 * set to 0. 363 */ 364 static disklabel_t 365 l64_clone_label(struct disk_info *info, struct diskslice *sp) 366 { 367 struct disklabel64 *lp; 368 disklabel_t res; 369 uint32_t blksize = info->d_media_blksize; 370 size_t lpcrcsize; 371 372 lp = kmalloc(sizeof *lp, M_DEVBUF, M_WAITOK | M_ZERO); 373 374 if (sp) 375 lp->d_total_size = (uint64_t)sp->ds_size * blksize; 376 else 377 lp->d_total_size = info->d_media_blocks * blksize; 378 379 lp->d_magic = DISKMAGIC64; 380 lp->d_align = blksize; 381 lp->d_npartitions = MAXPARTITIONS64; 382 lp->d_pstop = lp->d_total_size; 383 384 /* 385 * Create a dummy 'c' part and a dummy 'a' part (if requested). 386 * Note that the 'c' part is really a hack. 64 bit disklabels 387 * do not use 'c' to mean the raw partition. 388 */ 389 390 lp->d_partitions[2].p_boffset = 0; 391 lp->d_partitions[2].p_bsize = lp->d_total_size; 392 /* XXX SET FS TYPE */ 393 394 if (info->d_dsflags & DSO_COMPATPARTA) { 395 lp->d_partitions[0].p_boffset = 0; 396 lp->d_partitions[0].p_bsize = lp->d_total_size; 397 /* XXX SET FS TYPE */ 398 } 399 400 lpcrcsize = offsetof(struct disklabel64, 401 d_partitions[lp->d_npartitions]) - 402 offsetof(struct disklabel64, d_magic); 403 404 lp->d_crc = crc32(&lp->d_magic, lpcrcsize); 405 res.lab64 = lp; 406 return (res); 407 } 408 409 /* 410 * Create a virgin disklabel64 suitable for writing to the media. 411 * 412 * disklabel64 always reserves 32KB for a boot area and leaves room 413 * for up to RESPARTITIONS64 partitions. 414 */ 415 static void 416 l64_makevirginlabel(disklabel_t lpx, struct diskslices *ssp, 417 struct diskslice *sp, struct disk_info *info) 418 { 419 struct disklabel64 *lp = lpx.lab64; 420 struct partition64 *pp; 421 uint32_t blksize; 422 uint32_t ressize; 423 uint64_t blkmask; /* 64 bits so we can ~ */ 424 size_t lpcrcsize; 425 426 /* 427 * Setup the initial label. Use of a block size of at least 4KB 428 * for calculating the initial reserved areas to allow some degree 429 * of portability between media with different sector sizes. 430 * 431 * Note that the modified blksize is stored in d_align as a hint 432 * to the disklabeling program. 433 */ 434 bzero(lp, sizeof(*lp)); 435 if ((blksize = info->d_media_blksize) < 4096) 436 blksize = 4096; 437 blkmask = blksize - 1; 438 439 if (sp) 440 lp->d_total_size = (uint64_t)sp->ds_size * ssp->dss_secsize; 441 else 442 lp->d_total_size = info->d_media_blocks * info->d_media_blksize; 443 444 lp->d_magic = DISKMAGIC64; 445 lp->d_align = blksize; 446 lp->d_npartitions = MAXPARTITIONS64; 447 kern_uuidgen(&lp->d_stor_uuid, 1); 448 449 ressize = offsetof(struct disklabel64, d_partitions[RESPARTITIONS64]); 450 ressize = (ressize + (uint32_t)blkmask) & ~blkmask; 451 452 /* 453 * NOTE: When calculating pbase take into account the slice offset 454 * so the partitions are at least 32K-aligned relative to the 455 * start of the physical disk. This will accomodate efficient 456 * access to 4096 byte physical sector drives. 457 */ 458 lp->d_bbase = ressize; 459 lp->d_pbase = lp->d_bbase + ((32768 + blkmask) & ~blkmask); 460 lp->d_pbase = (lp->d_pbase + PALIGN_MASK) & ~(uint64_t)PALIGN_MASK; 461 462 /* adjust for slice offset so we are physically aligned */ 463 lp->d_pbase += 32768 - (sp->ds_offset * info->d_media_blksize) % 32768; 464 465 lp->d_pstop = (lp->d_total_size - lp->d_bbase) & ~blkmask; 466 lp->d_abase = lp->d_pstop; 467 468 /* 469 * All partitions are left empty unless DSO_COMPATPARTA is set 470 */ 471 472 if (info->d_dsflags & DSO_COMPATPARTA) { 473 pp = &lp->d_partitions[0]; 474 pp->p_boffset = lp->d_pbase; 475 pp->p_bsize = lp->d_pstop - lp->d_pbase; 476 /* XXX SET FS TYPE */ 477 } 478 479 lpcrcsize = offsetof(struct disklabel64, 480 d_partitions[lp->d_npartitions]) - 481 offsetof(struct disklabel64, d_magic); 482 lp->d_crc = crc32(&lp->d_magic, lpcrcsize); 483 } 484 485 /* 486 * Set the number of blocks at the beginning of the slice which have 487 * been reserved for label operations. This area will be write-protected 488 * when accessed via the slice. 489 * 490 * For now just protect the label area proper. Do not protect the 491 * boot area. Note partitions in 64 bit disklabels do not overlap 492 * the disklabel or boot area. 493 */ 494 static void 495 l64_adjust_label_reserved(struct diskslices *ssp, int slice, 496 struct diskslice *sp) 497 { 498 struct disklabel64 *lp = sp->ds_label.lab64; 499 500 sp->ds_reserved = lp->d_bbase / ssp->dss_secsize; 501 } 502 503 struct disklabel_ops disklabel64_ops = { 504 .labelsize = sizeof(struct disklabel64), 505 .op_readdisklabel = l64_readdisklabel, 506 .op_setdisklabel = l64_setdisklabel, 507 .op_writedisklabel = l64_writedisklabel, 508 .op_clone_label = l64_clone_label, 509 .op_adjust_label_reserved = l64_adjust_label_reserved, 510 .op_getpartbounds = l64_getpartbounds, 511 .op_loadpartinfo = l64_loadpartinfo, 512 .op_getnumparts = l64_getnumparts, 513 .op_makevirginlabel = l64_makevirginlabel, 514 .op_freedisklabel = l64_freedisklabel 515 }; 516 517