1/*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2008 Andrew Thompson <thompsa@FreeBSD.org> 5 * All rights reserved. 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 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 */ 28 29#include <sys/cdefs.h> 30#include <sys/ctype.h> 31#include <sys/param.h> 32#include <sys/bio.h> 33#include <sys/kernel.h> 34#include <sys/limits.h> 35#include <sys/malloc.h> 36#include <sys/queue.h> 37#include <sys/sysctl.h> 38#include <sys/systm.h> 39 40#include <geom/geom.h> 41#include <geom/geom_dbg.h> 42#include <sys/endian.h> 43 44#include <geom/linux_lvm/g_linux_lvm.h> 45 46FEATURE(geom_linux_lvm, "GEOM Linux LVM partitioning support"); 47 48/* Declare malloc(9) label */ 49static MALLOC_DEFINE(M_GLLVM, "gllvm", "GEOM_LINUX_LVM Data"); 50 51/* GEOM class methods */ 52static g_access_t g_llvm_access; 53static g_init_t g_llvm_init; 54static g_orphan_t g_llvm_orphan; 55static g_orphan_t g_llvm_taste_orphan; 56static g_start_t g_llvm_start; 57static g_taste_t g_llvm_taste; 58static g_ctl_destroy_geom_t g_llvm_destroy_geom; 59 60static void g_llvm_done(struct bio *); 61static void g_llvm_remove_disk(struct g_llvm_vg *, struct g_consumer *); 62static int g_llvm_activate_lv(struct g_llvm_vg *, struct g_llvm_lv *); 63static int g_llvm_add_disk(struct g_llvm_vg *, struct g_provider *, char *); 64static void g_llvm_free_vg(struct g_llvm_vg *); 65static int g_llvm_destroy(struct g_llvm_vg *, int); 66static int g_llvm_read_label(struct g_consumer *, struct g_llvm_label *); 67static int g_llvm_read_md(struct g_consumer *, struct g_llvm_metadata *, 68 struct g_llvm_label *); 69 70static int llvm_label_decode(const u_char *, struct g_llvm_label *, 71 int, u_int); 72static int llvm_md_decode(const u_char *, struct g_llvm_metadata *, 73 struct g_llvm_label *); 74static int llvm_textconf_decode(u_char *, int, 75 struct g_llvm_metadata *); 76static int llvm_textconf_decode_pv(char **, char *, struct g_llvm_vg *); 77static int llvm_textconf_decode_lv(char **, char *, struct g_llvm_vg *); 78static int llvm_textconf_decode_sg(char **, char *, struct g_llvm_lv *); 79 80SYSCTL_DECL(_kern_geom); 81SYSCTL_NODE(_kern_geom, OID_AUTO, linux_lvm, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 82 "GEOM_LINUX_LVM stuff"); 83static u_int g_llvm_debug = 0; 84SYSCTL_UINT(_kern_geom_linux_lvm, OID_AUTO, debug, CTLFLAG_RWTUN, &g_llvm_debug, 0, 85 "Debug level"); 86 87LIST_HEAD(, g_llvm_vg) vg_list; 88 89/* 90 * Called to notify geom when it's been opened, and for what intent 91 */ 92static int 93g_llvm_access(struct g_provider *pp, int dr, int dw, int de) 94{ 95 struct g_consumer *c; 96 struct g_llvm_vg *vg; 97 struct g_geom *gp; 98 int error; 99 100 KASSERT(pp != NULL, ("%s: NULL provider", __func__)); 101 gp = pp->geom; 102 KASSERT(gp != NULL, ("%s: NULL geom", __func__)); 103 vg = gp->softc; 104 105 if (vg == NULL) { 106 /* It seems that .access can be called with negative dr,dw,dx 107 * in this case but I want to check for myself */ 108 G_LLVM_DEBUG(0, "access(%d, %d, %d) for %s", 109 dr, dw, de, pp->name); 110 111 /* This should only happen when geom is withered so 112 * allow only negative requests */ 113 KASSERT(dr <= 0 && dw <= 0 && de <= 0, 114 ("%s: Positive access for %s", __func__, pp->name)); 115 if (pp->acr + dr == 0 && pp->acw + dw == 0 && pp->ace + de == 0) 116 G_LLVM_DEBUG(0, 117 "Device %s definitely destroyed", pp->name); 118 return (0); 119 } 120 121 /* Grab an exclusive bit to propagate on our consumers on first open */ 122 if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0) 123 de++; 124 /* ... drop it on close */ 125 if (pp->acr + dr == 0 && pp->acw + dw == 0 && pp->ace + de == 0) 126 de--; 127 128 error = ENXIO; 129 LIST_FOREACH(c, &gp->consumer, consumer) { 130 KASSERT(c != NULL, ("%s: consumer is NULL", __func__)); 131 error = g_access(c, dr, dw, de); 132 if (error != 0) { 133 struct g_consumer *c2; 134 135 /* Backout earlier changes */ 136 LIST_FOREACH(c2, &gp->consumer, consumer) { 137 if (c2 == c) /* all eariler components fixed */ 138 return (error); 139 g_access(c2, -dr, -dw, -de); 140 } 141 } 142 } 143 144 return (error); 145} 146 147/* 148 * Dismantle bio_queue and destroy its components 149 */ 150static void 151bioq_dismantle(struct bio_queue_head *bq) 152{ 153 struct bio *b; 154 155 for (b = bioq_first(bq); b != NULL; b = bioq_first(bq)) { 156 bioq_remove(bq, b); 157 g_destroy_bio(b); 158 } 159} 160 161/* 162 * GEOM .done handler 163 * Can't use standard handler because one requested IO may 164 * fork into additional data IOs 165 */ 166static void 167g_llvm_done(struct bio *b) 168{ 169 struct bio *parent_b; 170 171 parent_b = b->bio_parent; 172 173 if (b->bio_error != 0) { 174 G_LLVM_DEBUG(0, "Error %d for offset=%ju, length=%ju on %s", 175 b->bio_error, b->bio_offset, b->bio_length, 176 b->bio_to->name); 177 if (parent_b->bio_error == 0) 178 parent_b->bio_error = b->bio_error; 179 } 180 181 parent_b->bio_inbed++; 182 parent_b->bio_completed += b->bio_completed; 183 184 if (parent_b->bio_children == parent_b->bio_inbed) { 185 parent_b->bio_completed = parent_b->bio_length; 186 g_io_deliver(parent_b, parent_b->bio_error); 187 } 188 g_destroy_bio(b); 189} 190 191static void 192g_llvm_start(struct bio *bp) 193{ 194 struct g_provider *pp; 195 struct g_llvm_vg *vg; 196 struct g_llvm_pv *pv; 197 struct g_llvm_lv *lv; 198 struct g_llvm_segment *sg; 199 struct bio *cb; 200 struct bio_queue_head bq; 201 size_t chunk_size; 202 off_t offset, length; 203 char *addr; 204 u_int count; 205 206 pp = bp->bio_to; 207 lv = pp->private; 208 vg = pp->geom->softc; 209 210 switch (bp->bio_cmd) { 211 case BIO_READ: 212 case BIO_WRITE: 213 case BIO_DELETE: 214 /* XXX BIO_GETATTR allowed? */ 215 break; 216 default: 217 /* 218 * BIO_SPEEDUP and BIO_FLUSH should pass through to all sg 219 * elements, but aren't. 220 */ 221 g_io_deliver(bp, EOPNOTSUPP); 222 return; 223 } 224 225 bioq_init(&bq); 226 227 chunk_size = vg->vg_extentsize; 228 addr = bp->bio_data; 229 offset = bp->bio_offset; /* virtual offset and length */ 230 length = bp->bio_length; 231 232 while (length > 0) { 233 size_t chunk_index, in_chunk_offset, in_chunk_length; 234 235 pv = NULL; 236 cb = g_clone_bio(bp); 237 if (cb == NULL) { 238 bioq_dismantle(&bq); 239 if (bp->bio_error == 0) 240 bp->bio_error = ENOMEM; 241 g_io_deliver(bp, bp->bio_error); 242 return; 243 } 244 245 /* get the segment and the pv */ 246 if (lv->lv_sgcount == 1) { 247 /* skip much of the calculations for a single sg */ 248 chunk_index = 0; 249 in_chunk_offset = 0; 250 in_chunk_length = length; 251 sg = lv->lv_firstsg; 252 pv = sg->sg_pv; 253 cb->bio_offset = offset + sg->sg_pvoffset; 254 } else { 255 chunk_index = offset / chunk_size; /* round downwards */ 256 in_chunk_offset = offset % chunk_size; 257 in_chunk_length = 258 min(length, chunk_size - in_chunk_offset); 259 260 /* XXX could be faster */ 261 LIST_FOREACH(sg, &lv->lv_segs, sg_next) { 262 if (chunk_index >= sg->sg_start && 263 chunk_index <= sg->sg_end) { 264 /* adjust chunk index for sg start */ 265 chunk_index -= sg->sg_start; 266 pv = sg->sg_pv; 267 break; 268 } 269 } 270 cb->bio_offset = 271 (off_t)chunk_index * (off_t)chunk_size 272 + in_chunk_offset + sg->sg_pvoffset; 273 } 274 275 KASSERT(pv != NULL, ("Can't find PV for chunk %zu", 276 chunk_index)); 277 278 cb->bio_to = pv->pv_gprov; 279 cb->bio_done = g_llvm_done; 280 cb->bio_length = in_chunk_length; 281 cb->bio_data = addr; 282 cb->bio_caller1 = pv; 283 bioq_disksort(&bq, cb); 284 285 G_LLVM_DEBUG(5, 286 "Mapped %s(%ju, %ju) on %s to %zu(%zu,%zu) @ %s:%ju", 287 bp->bio_cmd == BIO_READ ? "R" : "W", 288 offset, length, lv->lv_name, 289 chunk_index, in_chunk_offset, in_chunk_length, 290 pv->pv_name, cb->bio_offset); 291 292 addr += in_chunk_length; 293 length -= in_chunk_length; 294 offset += in_chunk_length; 295 } 296 297 /* Fire off bio's here */ 298 count = 0; 299 for (cb = bioq_first(&bq); cb != NULL; cb = bioq_first(&bq)) { 300 bioq_remove(&bq, cb); 301 pv = cb->bio_caller1; 302 cb->bio_caller1 = NULL; 303 G_LLVM_DEBUG(6, "firing bio to %s, offset=%ju, length=%ju", 304 cb->bio_to->name, cb->bio_offset, cb->bio_length); 305 g_io_request(cb, pv->pv_gcons); 306 count++; 307 } 308 if (count == 0) { /* We handled everything locally */ 309 bp->bio_completed = bp->bio_length; 310 g_io_deliver(bp, 0); 311 } 312} 313 314static void 315g_llvm_remove_disk(struct g_llvm_vg *vg, struct g_consumer *cp) 316{ 317 struct g_llvm_pv *pv; 318 struct g_llvm_lv *lv; 319 struct g_llvm_segment *sg; 320 int found; 321 322 KASSERT(cp != NULL, ("Non-valid disk in %s.", __func__)); 323 pv = (struct g_llvm_pv *)cp->private; 324 325 G_LLVM_DEBUG(0, "Disk %s removed from %s.", cp->provider->name, 326 pv->pv_name); 327 328 LIST_FOREACH(lv, &vg->vg_lvs, lv_next) { 329 /* Find segments that map to this disk */ 330 found = 0; 331 LIST_FOREACH(sg, &lv->lv_segs, sg_next) { 332 if (sg->sg_pv == pv) { 333 sg->sg_pv = NULL; 334 lv->lv_sgactive--; 335 found = 1; 336 break; 337 } 338 } 339 if (found) { 340 G_LLVM_DEBUG(0, "Device %s removed.", 341 lv->lv_gprov->name); 342 g_wither_provider(lv->lv_gprov, ENXIO); 343 lv->lv_gprov = NULL; 344 } 345 } 346 347 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 348 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 349 g_detach(cp); 350 g_destroy_consumer(cp); 351} 352 353static void 354g_llvm_orphan(struct g_consumer *cp) 355{ 356 struct g_llvm_vg *vg; 357 struct g_geom *gp; 358 359 g_topology_assert(); 360 gp = cp->geom; 361 vg = gp->softc; 362 if (vg == NULL) 363 return; 364 365 g_llvm_remove_disk(vg, cp); 366 g_llvm_destroy(vg, 1); 367} 368 369static int 370g_llvm_activate_lv(struct g_llvm_vg *vg, struct g_llvm_lv *lv) 371{ 372 struct g_geom *gp; 373 struct g_provider *pp; 374 375 g_topology_assert(); 376 377 KASSERT(lv->lv_sgactive == lv->lv_sgcount, ("segment missing")); 378 379 gp = vg->vg_geom; 380 pp = g_new_providerf(gp, "linux_lvm/%s-%s", vg->vg_name, lv->lv_name); 381 pp->mediasize = vg->vg_extentsize * (off_t)lv->lv_extentcount; 382 pp->sectorsize = vg->vg_sectorsize; 383 g_error_provider(pp, 0); 384 lv->lv_gprov = pp; 385 pp->private = lv; 386 387 G_LLVM_DEBUG(1, "Created %s, %juM", pp->name, 388 pp->mediasize / (1024*1024)); 389 390 return (0); 391} 392 393static int 394g_llvm_add_disk(struct g_llvm_vg *vg, struct g_provider *pp, char *uuid) 395{ 396 struct g_geom *gp; 397 struct g_consumer *cp, *fcp; 398 struct g_llvm_pv *pv; 399 struct g_llvm_lv *lv; 400 struct g_llvm_segment *sg; 401 int error; 402 403 g_topology_assert(); 404 405 LIST_FOREACH(pv, &vg->vg_pvs, pv_next) { 406 if (strcmp(pv->pv_uuid, uuid) == 0) 407 break; /* found it */ 408 } 409 if (pv == NULL) { 410 G_LLVM_DEBUG(3, "uuid %s not found in pv list", uuid); 411 return (ENOENT); 412 } 413 if (pv->pv_gprov != NULL) { 414 G_LLVM_DEBUG(0, "disk %s already initialised in %s", 415 pv->pv_name, vg->vg_name); 416 return (EEXIST); 417 } 418 419 pv->pv_start *= vg->vg_sectorsize; 420 gp = vg->vg_geom; 421 fcp = LIST_FIRST(&gp->consumer); 422 423 cp = g_new_consumer(gp); 424 error = g_attach(cp, pp); 425 G_LLVM_DEBUG(1, "Attached %s to %s at offset %ju", 426 pp->name, pv->pv_name, pv->pv_start); 427 428 if (error != 0) { 429 G_LLVM_DEBUG(0, "cannot attach %s to %s", 430 pp->name, vg->vg_name); 431 g_destroy_consumer(cp); 432 return (error); 433 } 434 435 if (fcp != NULL) { 436 if (fcp->provider->sectorsize != pp->sectorsize) { 437 G_LLVM_DEBUG(0, "Provider %s of %s has invalid " 438 "sector size (%d)", pp->name, vg->vg_name, 439 pp->sectorsize); 440 return (EINVAL); 441 } 442 if (fcp->acr > 0 || fcp->acw || fcp->ace > 0) { 443 /* Replicate access permissions from first "live" 444 * consumer to the new one */ 445 error = g_access(cp, fcp->acr, fcp->acw, fcp->ace); 446 if (error != 0) { 447 g_detach(cp); 448 g_destroy_consumer(cp); 449 return (error); 450 } 451 } 452 } 453 454 cp->private = pv; 455 pv->pv_gcons = cp; 456 pv->pv_gprov = pp; 457 458 LIST_FOREACH(lv, &vg->vg_lvs, lv_next) { 459 /* Find segments that map to this disk */ 460 LIST_FOREACH(sg, &lv->lv_segs, sg_next) { 461 if (strcmp(sg->sg_pvname, pv->pv_name) == 0) { 462 /* avtivate the segment */ 463 KASSERT(sg->sg_pv == NULL, 464 ("segment already mapped")); 465 sg->sg_pvoffset = 466 (off_t)sg->sg_pvstart * vg->vg_extentsize 467 + pv->pv_start; 468 sg->sg_pv = pv; 469 lv->lv_sgactive++; 470 471 G_LLVM_DEBUG(2, "%s: %d to %d @ %s:%d" 472 " offset %ju sector %ju", 473 lv->lv_name, sg->sg_start, sg->sg_end, 474 sg->sg_pvname, sg->sg_pvstart, 475 sg->sg_pvoffset, 476 sg->sg_pvoffset / vg->vg_sectorsize); 477 } 478 } 479 /* Activate any lvs waiting on this disk */ 480 if (lv->lv_gprov == NULL && lv->lv_sgactive == lv->lv_sgcount) { 481 error = g_llvm_activate_lv(vg, lv); 482 if (error) 483 break; 484 } 485 } 486 return (error); 487} 488 489static void 490g_llvm_init(struct g_class *mp) 491{ 492 LIST_INIT(&vg_list); 493} 494 495static void 496g_llvm_free_vg(struct g_llvm_vg *vg) 497{ 498 struct g_llvm_pv *pv; 499 struct g_llvm_lv *lv; 500 struct g_llvm_segment *sg; 501 502 /* Free all the structures */ 503 while ((pv = LIST_FIRST(&vg->vg_pvs)) != NULL) { 504 LIST_REMOVE(pv, pv_next); 505 free(pv, M_GLLVM); 506 } 507 while ((lv = LIST_FIRST(&vg->vg_lvs)) != NULL) { 508 while ((sg = LIST_FIRST(&lv->lv_segs)) != NULL) { 509 LIST_REMOVE(sg, sg_next); 510 free(sg, M_GLLVM); 511 } 512 LIST_REMOVE(lv, lv_next); 513 free(lv, M_GLLVM); 514 } 515 free(vg, M_GLLVM); 516} 517 518static void 519g_llvm_taste_orphan(struct g_consumer *cp) 520{ 521 522 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 523 cp->provider->name)); 524} 525 526static struct g_geom * 527g_llvm_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 528{ 529 struct g_consumer *cp; 530 struct g_geom *gp; 531 struct g_llvm_label ll; 532 struct g_llvm_metadata md; 533 struct g_llvm_vg *vg; 534 int error; 535 536 bzero(&md, sizeof(md)); 537 538 g_topology_assert(); 539 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 540 gp = g_new_geomf(mp, "linux_lvm:taste"); 541 /* This orphan function should be never called. */ 542 gp->orphan = g_llvm_taste_orphan; 543 cp = g_new_consumer(gp); 544 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE; 545 error = g_attach(cp, pp); 546 if (error == 0) { 547 error = g_llvm_read_label(cp, &ll); 548 if (error == 0) 549 error = g_llvm_read_md(cp, &md, &ll); 550 g_detach(cp); 551 } 552 g_destroy_consumer(cp); 553 g_destroy_geom(gp); 554 if (error != 0) 555 return (NULL); 556 557 vg = md.md_vg; 558 if (vg->vg_geom == NULL) { 559 /* new volume group */ 560 gp = g_new_geomf(mp, "%s", vg->vg_name); 561 gp->start = g_llvm_start; 562 gp->spoiled = g_llvm_orphan; 563 gp->orphan = g_llvm_orphan; 564 gp->access = g_llvm_access; 565 vg->vg_sectorsize = pp->sectorsize; 566 vg->vg_extentsize *= vg->vg_sectorsize; 567 vg->vg_geom = gp; 568 gp->softc = vg; 569 G_LLVM_DEBUG(1, "Created volume %s, extent size %zuK", 570 vg->vg_name, vg->vg_extentsize / 1024); 571 } 572 573 /* initialise this disk in the volume group */ 574 g_llvm_add_disk(vg, pp, ll.ll_uuid); 575 return (vg->vg_geom); 576} 577 578static int 579g_llvm_destroy(struct g_llvm_vg *vg, int force) 580{ 581 struct g_provider *pp; 582 struct g_geom *gp; 583 584 g_topology_assert(); 585 if (vg == NULL) 586 return (ENXIO); 587 gp = vg->vg_geom; 588 589 LIST_FOREACH(pp, &gp->provider, provider) { 590 if (pp->acr != 0 || pp->acw != 0 || pp->ace != 0) { 591 G_LLVM_DEBUG(1, "Device %s is still open (r%dw%de%d)", 592 pp->name, pp->acr, pp->acw, pp->ace); 593 if (!force) 594 return (EBUSY); 595 } 596 } 597 598 LIST_REMOVE(vg, vg_next); 599 g_llvm_free_vg(vg); 600 gp->softc = NULL; 601 g_wither_geom(gp, ENXIO); 602 return (0); 603} 604 605static int 606g_llvm_destroy_geom(struct gctl_req *req __unused, struct g_class *mp __unused, 607 struct g_geom *gp) 608{ 609 struct g_llvm_vg *vg; 610 611 vg = gp->softc; 612 return (g_llvm_destroy(vg, 0)); 613} 614 615int 616g_llvm_read_label(struct g_consumer *cp, struct g_llvm_label *ll) 617{ 618 struct g_provider *pp; 619 u_char *buf; 620 int i, error = 0; 621 622 g_topology_assert(); 623 624 /* The LVM label is stored on the first four sectors */ 625 error = g_access(cp, 1, 0, 0); 626 if (error != 0) 627 return (error); 628 pp = cp->provider; 629 g_topology_unlock(); 630 buf = g_read_data(cp, 0, pp->sectorsize * 4, &error); 631 g_topology_lock(); 632 g_access(cp, -1, 0, 0); 633 if (buf == NULL) { 634 G_LLVM_DEBUG(1, "Cannot read metadata from %s (error=%d)", 635 pp->name, error); 636 return (error); 637 } 638 639 /* Search the four sectors for the LVM label. */ 640 for (i = 0; i < 4; i++) { 641 error = llvm_label_decode(&buf[i * pp->sectorsize], ll, i, 642 pp->sectorsize); 643 if (error == 0) 644 break; /* found it */ 645 } 646 g_free(buf); 647 return (error); 648} 649 650int 651g_llvm_read_md(struct g_consumer *cp, struct g_llvm_metadata *md, 652 struct g_llvm_label *ll) 653{ 654 struct g_provider *pp; 655 u_char *buf; 656 int error; 657 int size; 658 659 g_topology_assert(); 660 661 error = g_access(cp, 1, 0, 0); 662 if (error != 0) 663 return (error); 664 pp = cp->provider; 665 g_topology_unlock(); 666 buf = g_read_data(cp, ll->ll_md_offset, pp->sectorsize, &error); 667 g_topology_lock(); 668 g_access(cp, -1, 0, 0); 669 if (buf == NULL) { 670 G_LLVM_DEBUG(0, "Cannot read metadata from %s (error=%d)", 671 cp->provider->name, error); 672 return (error); 673 } 674 675 error = llvm_md_decode(buf, md, ll); 676 g_free(buf); 677 if (error != 0) { 678 return (error); 679 } 680 681 G_LLVM_DEBUG(1, "reading LVM2 config @ %s:%ju", pp->name, 682 ll->ll_md_offset + md->md_reloffset); 683 error = g_access(cp, 1, 0, 0); 684 if (error != 0) 685 return (error); 686 pp = cp->provider; 687 g_topology_unlock(); 688 /* round up to the nearest sector */ 689 size = md->md_relsize + 690 (pp->sectorsize - md->md_relsize % pp->sectorsize); 691 buf = g_read_data(cp, ll->ll_md_offset + md->md_reloffset, size, &error); 692 g_topology_lock(); 693 g_access(cp, -1, 0, 0); 694 if (buf == NULL) { 695 G_LLVM_DEBUG(0, "Cannot read LVM2 config from %s (error=%d)", 696 pp->name, error); 697 return (error); 698 } 699 buf[md->md_relsize] = '\0'; 700 G_LLVM_DEBUG(10, "LVM config:\n%s\n", buf); 701 error = llvm_textconf_decode(buf, md->md_relsize, md); 702 g_free(buf); 703 704 return (error); 705} 706 707static int 708llvm_label_decode(const u_char *data, struct g_llvm_label *ll, int sector, 709 u_int sectorsize) 710{ 711 uint64_t off; 712 char *uuid; 713 714 /* Magic string */ 715 if (bcmp("LABELONE", data , 8) != 0) 716 return (EINVAL); 717 718 /* We only support LVM2 text format */ 719 if (bcmp("LVM2 001", data + 24, 8) != 0) { 720 G_LLVM_DEBUG(0, "Unsupported LVM format"); 721 return (EINVAL); 722 } 723 724 ll->ll_sector = le64dec(data + 8); 725 ll->ll_crc = le32dec(data + 16); 726 ll->ll_offset = le32dec(data + 20); 727 728 if (ll->ll_sector != sector) { 729 G_LLVM_DEBUG(0, "Expected sector %ju, found at %d", 730 ll->ll_sector, sector); 731 return (EINVAL); 732 } 733 734 /* XXX The minimal possible size of physical volume header is 88 */ 735 if (ll->ll_offset < 32 || ll->ll_offset > sectorsize - 88) { 736 G_LLVM_DEBUG(0, "Invalid physical volume header offset %u", 737 ll->ll_offset); 738 return (EINVAL); 739 } 740 741 off = ll->ll_offset; 742 /* 743 * convert the binary uuid to string format, the format is 744 * xxxxxx-xxxx-xxxx-xxxx-xxxx-xxxx-xxxxxx (6-4-4-4-4-4-6) 745 */ 746 uuid = ll->ll_uuid; 747 bcopy(data + off, uuid, 6); 748 off += 6; 749 uuid += 6; 750 *uuid++ = '-'; 751 for (int i = 0; i < 5; i++) { 752 bcopy(data + off, uuid, 4); 753 off += 4; 754 uuid += 4; 755 *uuid++ = '-'; 756 } 757 bcopy(data + off, uuid, 6); 758 off += 6; 759 uuid += 6; 760 *uuid++ = '\0'; 761 762 ll->ll_size = le64dec(data + off); 763 off += 8; 764 ll->ll_pestart = le64dec(data + off); 765 off += 16; 766 767 /* Only one data section is supported */ 768 if (le64dec(data + off) != 0) { 769 G_LLVM_DEBUG(0, "Only one data section supported"); 770 return (EINVAL); 771 } 772 773 off += 16; 774 ll->ll_md_offset = le64dec(data + off); 775 off += 8; 776 ll->ll_md_size = le64dec(data + off); 777 off += 8; 778 779 G_LLVM_DEBUG(1, "LVM metadata: offset=%ju, size=%ju", ll->ll_md_offset, 780 ll->ll_md_size); 781 782 /* Only one data section is supported */ 783 if (le64dec(data + off) != 0) { 784 G_LLVM_DEBUG(0, "Only one metadata section supported"); 785 return (EINVAL); 786 } 787 788 G_LLVM_DEBUG(2, "label uuid=%s", ll->ll_uuid); 789 G_LLVM_DEBUG(2, "sector=%ju, crc=%u, offset=%u, size=%ju, pestart=%ju", 790 ll->ll_sector, ll->ll_crc, ll->ll_offset, ll->ll_size, 791 ll->ll_pestart); 792 793 return (0); 794} 795 796static int 797llvm_md_decode(const u_char *data, struct g_llvm_metadata *md, 798 struct g_llvm_label *ll) 799{ 800 uint64_t off; 801 char magic[16]; 802 803 off = 0; 804 md->md_csum = le32dec(data + off); 805 off += 4; 806 bcopy(data + off, magic, 16); 807 off += 16; 808 md->md_version = le32dec(data + off); 809 off += 4; 810 md->md_start = le64dec(data + off); 811 off += 8; 812 md->md_size = le64dec(data + off); 813 off += 8; 814 815 if (bcmp(G_LLVM_MAGIC, magic, 16) != 0) { 816 G_LLVM_DEBUG(0, "Incorrect md magic number"); 817 return (EINVAL); 818 } 819 if (md->md_version != 1) { 820 G_LLVM_DEBUG(0, "Incorrect md version number (%u)", 821 md->md_version); 822 return (EINVAL); 823 } 824 if (md->md_start != ll->ll_md_offset) { 825 G_LLVM_DEBUG(0, "Incorrect md offset (%ju)", md->md_start); 826 return (EINVAL); 827 } 828 829 /* Aparently only one is ever returned */ 830 md->md_reloffset = le64dec(data + off); 831 off += 8; 832 md->md_relsize = le64dec(data + off); 833 off += 16; /* XXX skipped checksum */ 834 835 if (le64dec(data + off) != 0) { 836 G_LLVM_DEBUG(0, "Only one reloc supported"); 837 return (EINVAL); 838 } 839 840 G_LLVM_DEBUG(3, "reloc: offset=%ju, size=%ju", 841 md->md_reloffset, md->md_relsize); 842 G_LLVM_DEBUG(3, "md: version=%u, start=%ju, size=%ju", 843 md->md_version, md->md_start, md->md_size); 844 845 return (0); 846} 847 848#define GRAB_INT(key, tok1, tok2, v) \ 849 if (tok1 && tok2 && strncmp(tok1, key, sizeof(key)) == 0) { \ 850 v = strtol(tok2, &tok1, 10); \ 851 if (tok1 == tok2) \ 852 /* strtol did not eat any of the buffer */ \ 853 goto bad; \ 854 continue; \ 855 } 856 857#define GRAB_STR(key, tok1, tok2, v, len) \ 858 if (tok1 && tok2 && strncmp(tok1, key, sizeof(key)) == 0) { \ 859 strsep(&tok2, "\""); \ 860 if (tok2 == NULL) \ 861 continue; \ 862 tok1 = strsep(&tok2, "\""); \ 863 if (tok2 == NULL) \ 864 continue; \ 865 strncpy(v, tok1, len); \ 866 continue; \ 867 } 868 869#define SPLIT(key, value, str) \ 870 key = strsep(&value, str); \ 871 /* strip trailing whitespace on the key */ \ 872 for (char *t = key; *t != '\0'; t++) \ 873 if (isspace(*t)) { \ 874 *t = '\0'; \ 875 break; \ 876 } 877 878static size_t 879llvm_grab_name(char *name, const char *tok) 880{ 881 size_t len; 882 883 len = 0; 884 if (tok == NULL) 885 return (0); 886 if (tok[0] == '-') 887 return (0); 888 if (strcmp(tok, ".") == 0 || strcmp(tok, "..") == 0) 889 return (0); 890 while (tok[len] && (isalpha(tok[len]) || isdigit(tok[len]) || 891 tok[len] == '.' || tok[len] == '_' || tok[len] == '-' || 892 tok[len] == '+') && len < G_LLVM_NAMELEN - 1) 893 len++; 894 bcopy(tok, name, len); 895 name[len] = '\0'; 896 return (len); 897} 898 899static int 900llvm_textconf_decode(u_char *data, int buflen, struct g_llvm_metadata *md) 901{ 902 struct g_llvm_vg *vg; 903 char *buf = data; 904 char *tok, *v; 905 char name[G_LLVM_NAMELEN]; 906 char uuid[G_LLVM_UUIDLEN]; 907 size_t len; 908 909 if (buf == NULL || *buf == '\0') 910 return (EINVAL); 911 912 tok = strsep(&buf, "\n"); 913 if (tok == NULL) 914 return (EINVAL); 915 len = llvm_grab_name(name, tok); 916 if (len == 0) 917 return (EINVAL); 918 919 /* check too see if the vg has already been loaded off another disk */ 920 LIST_FOREACH(vg, &vg_list, vg_next) { 921 if (strcmp(vg->vg_name, name) == 0) { 922 uuid[0] = '\0'; 923 /* grab the volume group uuid */ 924 while ((tok = strsep(&buf, "\n")) != NULL) { 925 if (strstr(tok, "{")) 926 break; 927 if (strstr(tok, "=")) { 928 SPLIT(v, tok, "="); 929 GRAB_STR("id", v, tok, uuid, 930 sizeof(uuid)); 931 } 932 } 933 if (strcmp(vg->vg_uuid, uuid) == 0) { 934 /* existing vg */ 935 md->md_vg = vg; 936 return (0); 937 } 938 /* XXX different volume group with name clash! */ 939 G_LLVM_DEBUG(0, 940 "%s already exists, volume group not loaded", name); 941 return (EINVAL); 942 } 943 } 944 945 vg = malloc(sizeof(*vg), M_GLLVM, M_NOWAIT|M_ZERO); 946 if (vg == NULL) 947 return (ENOMEM); 948 949 strncpy(vg->vg_name, name, sizeof(vg->vg_name)); 950 LIST_INIT(&vg->vg_pvs); 951 LIST_INIT(&vg->vg_lvs); 952 953#define VOL_FOREACH(func, tok, buf, p) \ 954 while ((tok = strsep(buf, "\n")) != NULL) { \ 955 if (strstr(tok, "{")) { \ 956 func(buf, tok, p); \ 957 continue; \ 958 } \ 959 if (strstr(tok, "}")) \ 960 break; \ 961 } 962 963 while ((tok = strsep(&buf, "\n")) != NULL) { 964 if (strcmp(tok, "physical_volumes {") == 0) { 965 VOL_FOREACH(llvm_textconf_decode_pv, tok, &buf, vg); 966 continue; 967 } 968 if (strcmp(tok, "logical_volumes {") == 0) { 969 VOL_FOREACH(llvm_textconf_decode_lv, tok, &buf, vg); 970 continue; 971 } 972 if (strstr(tok, "{")) { 973 G_LLVM_DEBUG(2, "unknown section %s", tok); 974 continue; 975 } 976 977 /* parse 'key = value' lines */ 978 if (strstr(tok, "=")) { 979 SPLIT(v, tok, "="); 980 GRAB_STR("id", v, tok, vg->vg_uuid, sizeof(vg->vg_uuid)); 981 GRAB_INT("extent_size", v, tok, vg->vg_extentsize); 982 continue; 983 } 984 } 985 /* basic checking */ 986 if (vg->vg_extentsize == 0) 987 goto bad; 988 989 md->md_vg = vg; 990 LIST_INSERT_HEAD(&vg_list, vg, vg_next); 991 G_LLVM_DEBUG(3, "vg: name=%s uuid=%s", vg->vg_name, vg->vg_uuid); 992 return(0); 993 994bad: 995 g_llvm_free_vg(vg); 996 return (-1); 997} 998#undef VOL_FOREACH 999 1000static int 1001llvm_textconf_decode_pv(char **buf, char *tok, struct g_llvm_vg *vg) 1002{ 1003 struct g_llvm_pv *pv; 1004 char *v; 1005 size_t len; 1006 1007 if (*buf == NULL || **buf == '\0') 1008 return (EINVAL); 1009 1010 pv = malloc(sizeof(*pv), M_GLLVM, M_NOWAIT|M_ZERO); 1011 if (pv == NULL) 1012 return (ENOMEM); 1013 1014 pv->pv_vg = vg; 1015 len = 0; 1016 if (tok == NULL) 1017 goto bad; 1018 len = llvm_grab_name(pv->pv_name, tok); 1019 if (len == 0) 1020 goto bad; 1021 1022 while ((tok = strsep(buf, "\n")) != NULL) { 1023 if (strstr(tok, "{")) 1024 goto bad; 1025 1026 if (strstr(tok, "}")) 1027 break; 1028 1029 /* parse 'key = value' lines */ 1030 if (strstr(tok, "=")) { 1031 SPLIT(v, tok, "="); 1032 GRAB_STR("id", v, tok, pv->pv_uuid, sizeof(pv->pv_uuid)); 1033 GRAB_INT("pe_start", v, tok, pv->pv_start); 1034 GRAB_INT("pe_count", v, tok, pv->pv_count); 1035 continue; 1036 } 1037 } 1038 if (tok == NULL) 1039 goto bad; 1040 /* basic checking */ 1041 if (pv->pv_count == 0) 1042 goto bad; 1043 1044 LIST_INSERT_HEAD(&vg->vg_pvs, pv, pv_next); 1045 G_LLVM_DEBUG(3, "pv: name=%s uuid=%s", pv->pv_name, pv->pv_uuid); 1046 1047 return (0); 1048bad: 1049 free(pv, M_GLLVM); 1050 return (-1); 1051} 1052 1053static int 1054llvm_textconf_decode_lv(char **buf, char *tok, struct g_llvm_vg *vg) 1055{ 1056 struct g_llvm_lv *lv; 1057 struct g_llvm_segment *sg; 1058 char *v; 1059 size_t len; 1060 1061 if (*buf == NULL || **buf == '\0') 1062 return (EINVAL); 1063 1064 lv = malloc(sizeof(*lv), M_GLLVM, M_NOWAIT|M_ZERO); 1065 if (lv == NULL) 1066 return (ENOMEM); 1067 1068 lv->lv_vg = vg; 1069 LIST_INIT(&lv->lv_segs); 1070 1071 if (tok == NULL) 1072 goto bad; 1073 len = llvm_grab_name(lv->lv_name, tok); 1074 if (len == 0) 1075 goto bad; 1076 1077 while ((tok = strsep(buf, "\n")) != NULL) { 1078 if (strstr(tok, "{")) { 1079 if (strstr(tok, "segment")) { 1080 llvm_textconf_decode_sg(buf, tok, lv); 1081 continue; 1082 } else 1083 /* unexpected section */ 1084 goto bad; 1085 } 1086 1087 if (strstr(tok, "}")) 1088 break; 1089 1090 /* parse 'key = value' lines */ 1091 if (strstr(tok, "=")) { 1092 SPLIT(v, tok, "="); 1093 GRAB_STR("id", v, tok, lv->lv_uuid, sizeof(lv->lv_uuid)); 1094 GRAB_INT("segment_count", v, tok, lv->lv_sgcount); 1095 continue; 1096 } 1097 } 1098 if (tok == NULL) 1099 goto bad; 1100 if (lv->lv_sgcount == 0 || lv->lv_sgcount != lv->lv_numsegs) 1101 /* zero or incomplete segment list */ 1102 goto bad; 1103 1104 /* Optimize for only one segment on the pv */ 1105 lv->lv_firstsg = LIST_FIRST(&lv->lv_segs); 1106 LIST_INSERT_HEAD(&vg->vg_lvs, lv, lv_next); 1107 G_LLVM_DEBUG(3, "lv: name=%s uuid=%s", lv->lv_name, lv->lv_uuid); 1108 1109 return (0); 1110bad: 1111 while ((sg = LIST_FIRST(&lv->lv_segs)) != NULL) { 1112 LIST_REMOVE(sg, sg_next); 1113 free(sg, M_GLLVM); 1114 } 1115 free(lv, M_GLLVM); 1116 return (-1); 1117} 1118 1119static int 1120llvm_textconf_decode_sg(char **buf, char *tok, struct g_llvm_lv *lv) 1121{ 1122 struct g_llvm_segment *sg; 1123 char *v; 1124 int count = 0; 1125 1126 if (*buf == NULL || **buf == '\0') 1127 return (EINVAL); 1128 1129 sg = malloc(sizeof(*sg), M_GLLVM, M_NOWAIT|M_ZERO); 1130 if (sg == NULL) 1131 return (ENOMEM); 1132 1133 while ((tok = strsep(buf, "\n")) != NULL) { 1134 /* only a single linear stripe is supported */ 1135 if (strstr(tok, "stripe_count")) { 1136 SPLIT(v, tok, "="); 1137 GRAB_INT("stripe_count", v, tok, count); 1138 if (count != 1) 1139 goto bad; 1140 } 1141 1142 if (strstr(tok, "{")) 1143 goto bad; 1144 1145 if (strstr(tok, "}")) 1146 break; 1147 1148 if (strcmp(tok, "stripes = [") == 0) { 1149 tok = strsep(buf, "\n"); 1150 if (tok == NULL) 1151 goto bad; 1152 1153 strsep(&tok, "\""); 1154 if (tok == NULL) 1155 goto bad; /* missing open quotes */ 1156 v = strsep(&tok, "\""); 1157 if (tok == NULL) 1158 goto bad; /* missing close quotes */ 1159 strncpy(sg->sg_pvname, v, sizeof(sg->sg_pvname)); 1160 if (*tok != ',') 1161 goto bad; /* missing comma for stripe */ 1162 tok++; 1163 1164 sg->sg_pvstart = strtol(tok, &v, 10); 1165 if (v == tok) 1166 /* strtol did not eat any of the buffer */ 1167 goto bad; 1168 1169 continue; 1170 } 1171 1172 /* parse 'key = value' lines */ 1173 if (strstr(tok, "=")) { 1174 SPLIT(v, tok, "="); 1175 GRAB_INT("start_extent", v, tok, sg->sg_start); 1176 GRAB_INT("extent_count", v, tok, sg->sg_count); 1177 continue; 1178 } 1179 } 1180 if (tok == NULL) 1181 goto bad; 1182 /* basic checking */ 1183 if (count != 1 || sg->sg_count == 0) 1184 goto bad; 1185 1186 sg->sg_end = sg->sg_start + sg->sg_count - 1; 1187 lv->lv_numsegs++; 1188 lv->lv_extentcount += sg->sg_count; 1189 LIST_INSERT_HEAD(&lv->lv_segs, sg, sg_next); 1190 1191 return (0); 1192bad: 1193 free(sg, M_GLLVM); 1194 return (-1); 1195} 1196#undef GRAB_INT 1197#undef GRAB_STR 1198#undef SPLIT 1199 1200static struct g_class g_llvm_class = { 1201 .name = G_LLVM_CLASS_NAME, 1202 .version = G_VERSION, 1203 .init = g_llvm_init, 1204 .taste = g_llvm_taste, 1205 .destroy_geom = g_llvm_destroy_geom 1206}; 1207 1208DECLARE_GEOM_CLASS(g_llvm_class, g_linux_lvm); 1209MODULE_VERSION(geom_linux_lvm, 0); 1210