geom_vinum_drive.c revision 183514
1/*- 2 * Copyright (c) 2004, 2005 Lukas Ertl 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27#include <sys/cdefs.h> 28__FBSDID("$FreeBSD: head/sys/geom/vinum/geom_vinum_drive.c 183514 2008-10-01 14:50:36Z lulf $"); 29 30#include <sys/param.h> 31#include <sys/bio.h> 32#include <sys/errno.h> 33#include <sys/endian.h> 34#include <sys/conf.h> 35#include <sys/kernel.h> 36#include <sys/kthread.h> 37#include <sys/libkern.h> 38#include <sys/lock.h> 39#include <sys/malloc.h> 40#include <sys/module.h> 41#include <sys/mutex.h> 42#include <sys/sbuf.h> 43#include <sys/systm.h> 44#include <sys/time.h> 45#include <sys/vimage.h> 46 47#include <geom/geom.h> 48#include <geom/vinum/geom_vinum_var.h> 49#include <geom/vinum/geom_vinum.h> 50#include <geom/vinum/geom_vinum_share.h> 51 52#define GV_LEGACY_I386 0 53#define GV_LEGACY_AMD64 1 54#define GV_LEGACY_SPARC64 2 55#define GV_LEGACY_POWERPC 3 56 57static void gv_drive_dead(void *, int); 58static void gv_drive_worker(void *); 59static int gv_legacy_header_type(uint8_t *, int); 60 61/* 62 * Here are the "offset (size)" for the various struct gv_hdr fields, 63 * for the legacy i386 (or 32-bit powerpc), legacy amd64 (or sparc64), and 64 * current (cpu & endian agnostic) versions of the on-disk format of the vinum 65 * header structure: 66 * 67 * i386 amd64 current field 68 * -------- -------- -------- ----- 69 * 0 ( 8) 0 ( 8) 0 ( 8) magic 70 * 8 ( 4) 8 ( 8) 8 ( 8) config_length 71 * 12 (32) 16 (32) 16 (32) label.sysname 72 * 44 (32) 48 (32) 48 (32) label.name 73 * 76 ( 4) 80 ( 8) 80 ( 8) label.date_of_birth.tv_sec 74 * 80 ( 4) 88 ( 8) 88 ( 8) label.date_of_birth.tv_usec 75 * 84 ( 4) 96 ( 8) 96 ( 8) label.last_update.tv_sec 76 * 88 ( 4) 104 ( 8) 104 ( 8) label.last_update.tv_usec 77 * 92 ( 8) 112 ( 8) 112 ( 8) label.drive_size 78 * ======== ======== ======== 79 * 100 120 120 total size 80 * 81 * NOTE: i386 and amd64 formats are stored as little-endian; the current 82 * format uses big-endian (network order). 83 */ 84 85 86/* Checks for legacy format depending on platform. */ 87static int 88gv_legacy_header_type(uint8_t *hdr, int bigendian) 89{ 90 uint32_t *i32; 91 int arch_32, arch_64, i; 92 93 /* Set arch according to endianess. */ 94 if (bigendian) { 95 arch_32 = GV_LEGACY_POWERPC; 96 arch_64 = GV_LEGACY_SPARC64; 97 } else { 98 arch_32 = GV_LEGACY_I386; 99 arch_64 = GV_LEGACY_AMD64; 100 } 101 102 /* if non-empty hostname overlaps 64-bit config_length */ 103 i32 = (uint32_t *)(hdr + 12); 104 if (*i32 != 0) 105 return (arch_32); 106 /* check for non-empty hostname */ 107 if (hdr[16] != 0) 108 return (arch_64); 109 /* check bytes past 32-bit structure */ 110 for (i = 100; i < 120; i++) 111 if (hdr[i] != 0) 112 return (arch_32); 113 /* check for overlapping timestamp */ 114 i32 = (uint32_t *)(hdr + 84); 115 116 if (*i32 == 0) 117 return (arch_64); 118 return (arch_32); 119} 120 121/* 122 * Read the header while taking magic number into account, and write it to 123 * destination pointer. 124 */ 125int 126gv_read_header(struct g_consumer *cp, struct gv_hdr *m_hdr) 127{ 128 struct g_provider *pp; 129 uint64_t magic_machdep; 130 uint8_t *d_hdr; 131 int be, off; 132 133#define GV_GET32(endian) \ 134 endian##32toh(*((uint32_t *)&d_hdr[off])); \ 135 off += 4 136#define GV_GET64(endian) \ 137 endian##64toh(*((uint64_t *)&d_hdr[off])); \ 138 off += 8 139 140 KASSERT(m_hdr != NULL, ("gv_read_header: null m_hdr")); 141 KASSERT(cp != NULL, ("gv_read_header: null cp")); 142 pp = cp->provider; 143 KASSERT(pp != NULL, ("gv_read_header: null pp")); 144 145 d_hdr = g_read_data(cp, GV_HDR_OFFSET, pp->sectorsize, NULL); 146 if (d_hdr == NULL) 147 return (-1); 148 off = 0; 149 m_hdr->magic = GV_GET64(be); 150 magic_machdep = *((uint64_t *)&d_hdr[0]); 151 /* 152 * The big endian machines will have a reverse of GV_OLD_MAGIC, so we 153 * need to decide if we are running on a big endian machine as well as 154 * checking the magic against the reverse of GV_OLD_MAGIC. 155 */ 156 be = (m_hdr->magic == magic_machdep); 157 if (m_hdr->magic == GV_MAGIC) { 158 m_hdr->config_length = GV_GET64(be); 159 off = 16; 160 bcopy(d_hdr + off, m_hdr->label.sysname, GV_HOSTNAME_LEN); 161 off += GV_HOSTNAME_LEN; 162 bcopy(d_hdr + off, m_hdr->label.name, GV_MAXDRIVENAME); 163 off += GV_MAXDRIVENAME; 164 m_hdr->label.date_of_birth.tv_sec = GV_GET64(be); 165 m_hdr->label.date_of_birth.tv_usec = GV_GET64(be); 166 m_hdr->label.last_update.tv_sec = GV_GET64(be); 167 m_hdr->label.last_update.tv_usec = GV_GET64(be); 168 m_hdr->label.drive_size = GV_GET64(be); 169 } else if (m_hdr->magic != GV_OLD_MAGIC && 170 m_hdr->magic != le64toh(GV_OLD_MAGIC)) { 171 /* Not a gvinum drive. */ 172 g_free(d_hdr); 173 return (-1); 174 } else if (gv_legacy_header_type(d_hdr, be) == GV_LEGACY_SPARC64) { 175 printf("VINUM: detected legacy sparc64 header\n"); 176 m_hdr->magic = GV_MAGIC; 177 /* Legacy sparc64 on-disk header */ 178 m_hdr->config_length = GV_GET64(be); 179 bcopy(d_hdr + 16, m_hdr->label.sysname, GV_HOSTNAME_LEN); 180 off += GV_HOSTNAME_LEN; 181 bcopy(d_hdr + 48, m_hdr->label.name, GV_MAXDRIVENAME); 182 off += GV_MAXDRIVENAME; 183 m_hdr->label.date_of_birth.tv_sec = GV_GET64(be); 184 m_hdr->label.date_of_birth.tv_usec = GV_GET64(be); 185 m_hdr->label.last_update.tv_sec = GV_GET64(be); 186 m_hdr->label.last_update.tv_usec = GV_GET64(be); 187 m_hdr->label.drive_size = GV_GET64(be); 188 } else if (gv_legacy_header_type(d_hdr, be) == GV_LEGACY_POWERPC) { 189 printf("VINUM: detected legacy PowerPC header\n"); 190 m_hdr->magic = GV_MAGIC; 191 /* legacy 32-bit big endian on-disk header */ 192 m_hdr->config_length = GV_GET32(be); 193 bcopy(d_hdr + off, m_hdr->label.sysname, GV_HOSTNAME_LEN); 194 off += GV_HOSTNAME_LEN; 195 bcopy(d_hdr + off, m_hdr->label.name, GV_MAXDRIVENAME); 196 off += GV_MAXDRIVENAME; 197 m_hdr->label.date_of_birth.tv_sec = GV_GET32(be); 198 m_hdr->label.date_of_birth.tv_usec = GV_GET32(be); 199 m_hdr->label.last_update.tv_sec = GV_GET32(be); 200 m_hdr->label.last_update.tv_usec = GV_GET32(be); 201 m_hdr->label.drive_size = GV_GET64(be); 202 } else if (gv_legacy_header_type(d_hdr, be) == GV_LEGACY_I386) { 203 printf("VINUM: detected legacy i386 header\n"); 204 m_hdr->magic = GV_MAGIC; 205 /* legacy i386 on-disk header */ 206 m_hdr->config_length = GV_GET32(le); 207 bcopy(d_hdr + off, m_hdr->label.sysname, GV_HOSTNAME_LEN); 208 off += GV_HOSTNAME_LEN; 209 bcopy(d_hdr + off, m_hdr->label.name, GV_MAXDRIVENAME); 210 off += GV_MAXDRIVENAME; 211 m_hdr->label.date_of_birth.tv_sec = GV_GET32(le); 212 m_hdr->label.date_of_birth.tv_usec = GV_GET32(le); 213 m_hdr->label.last_update.tv_sec = GV_GET32(le); 214 m_hdr->label.last_update.tv_usec = GV_GET32(le); 215 m_hdr->label.drive_size = GV_GET64(le); 216 } else { 217 printf("VINUM: detected legacy amd64 header\n"); 218 m_hdr->magic = GV_MAGIC; 219 /* legacy amd64 on-disk header */ 220 m_hdr->config_length = GV_GET64(le); 221 bcopy(d_hdr + 16, m_hdr->label.sysname, GV_HOSTNAME_LEN); 222 off += GV_HOSTNAME_LEN; 223 bcopy(d_hdr + 48, m_hdr->label.name, GV_MAXDRIVENAME); 224 off += GV_MAXDRIVENAME; 225 m_hdr->label.date_of_birth.tv_sec = GV_GET64(le); 226 m_hdr->label.date_of_birth.tv_usec = GV_GET64(le); 227 m_hdr->label.last_update.tv_sec = GV_GET64(le); 228 m_hdr->label.last_update.tv_usec = GV_GET64(le); 229 m_hdr->label.drive_size = GV_GET64(le); 230 } 231 232 g_free(d_hdr); 233 return (0); 234} 235 236/* Write out the gvinum header. */ 237int 238gv_write_header(struct g_consumer *cp, struct gv_hdr *m_hdr) 239{ 240 uint8_t d_hdr[GV_HDR_LEN]; 241 int off, ret; 242 243#define GV_SET32BE(field) \ 244 do { \ 245 *((uint32_t *)&d_hdr[off]) = htobe32(field); \ 246 off += 4; \ 247 } while (0) 248#define GV_SET64BE(field) \ 249 do { \ 250 *((uint64_t *)&d_hdr[off]) = htobe64(field); \ 251 off += 8; \ 252 } while (0) 253 254 KASSERT(m_hdr != NULL, ("gv_write_header: null m_hdr")); 255 256 off = 0; 257 memset(d_hdr, 0, GV_HDR_LEN); 258 GV_SET64BE(m_hdr->magic); 259 GV_SET64BE(m_hdr->config_length); 260 off = 16; 261 bcopy(m_hdr->label.sysname, d_hdr + off, GV_HOSTNAME_LEN); 262 off += GV_HOSTNAME_LEN; 263 bcopy(m_hdr->label.name, d_hdr + off, GV_MAXDRIVENAME); 264 off += GV_MAXDRIVENAME; 265 GV_SET64BE(m_hdr->label.date_of_birth.tv_sec); 266 GV_SET64BE(m_hdr->label.date_of_birth.tv_usec); 267 GV_SET64BE(m_hdr->label.last_update.tv_sec); 268 GV_SET64BE(m_hdr->label.last_update.tv_usec); 269 GV_SET64BE(m_hdr->label.drive_size); 270 271 ret = g_write_data(cp, GV_HDR_OFFSET, d_hdr, GV_HDR_LEN); 272 return (ret); 273} 274 275void 276gv_config_new_drive(struct gv_drive *d) 277{ 278 struct gv_hdr *vhdr; 279 struct gv_freelist *fl; 280 281 KASSERT(d != NULL, ("config_new_drive: NULL d")); 282 283 vhdr = g_malloc(sizeof(*vhdr), M_WAITOK | M_ZERO); 284 vhdr->magic = GV_MAGIC; 285 vhdr->config_length = GV_CFG_LEN; 286 287 mtx_lock(&hostname_mtx); 288 bcopy(G_hostname, vhdr->label.sysname, GV_HOSTNAME_LEN); 289 mtx_unlock(&hostname_mtx); 290 strncpy(vhdr->label.name, d->name, GV_MAXDRIVENAME); 291 microtime(&vhdr->label.date_of_birth); 292 293 d->hdr = vhdr; 294 295 LIST_INIT(&d->subdisks); 296 LIST_INIT(&d->freelist); 297 298 fl = g_malloc(sizeof(struct gv_freelist), M_WAITOK | M_ZERO); 299 fl->offset = GV_DATA_START; 300 fl->size = d->avail; 301 LIST_INSERT_HEAD(&d->freelist, fl, freelist); 302 d->freelist_entries = 1; 303 304 d->bqueue = g_malloc(sizeof(struct bio_queue_head), M_WAITOK | M_ZERO); 305 bioq_init(d->bqueue); 306 mtx_init(&d->bqueue_mtx, "gv_drive", NULL, MTX_DEF); 307 kproc_create(gv_drive_worker, d, NULL, 0, 0, "gv_d %s", d->name); 308 d->flags |= GV_DRIVE_THREAD_ACTIVE; 309} 310 311void 312gv_save_config_all(struct gv_softc *sc) 313{ 314 struct gv_drive *d; 315 316 g_topology_assert(); 317 318 LIST_FOREACH(d, &sc->drives, drive) { 319 if (d->geom == NULL) 320 continue; 321 gv_save_config(NULL, d, sc); 322 } 323} 324 325/* Save the vinum configuration back to disk. */ 326void 327gv_save_config(struct g_consumer *cp, struct gv_drive *d, struct gv_softc *sc) 328{ 329 struct g_geom *gp; 330 struct g_consumer *cp2; 331 struct gv_hdr *vhdr, *hdr; 332 struct sbuf *sb; 333 int error; 334 335 g_topology_assert(); 336 337 KASSERT(d != NULL, ("gv_save_config: null d")); 338 KASSERT(sc != NULL, ("gv_save_config: null sc")); 339 340 /* 341 * We can't save the config on a drive that isn't up, but drives that 342 * were just created aren't officially up yet, so we check a special 343 * flag. 344 */ 345 if ((d->state != GV_DRIVE_UP) && !(d->flags && GV_DRIVE_NEWBORN)) 346 return; 347 348 if (cp == NULL) { 349 gp = d->geom; 350 KASSERT(gp != NULL, ("gv_save_config: null gp")); 351 cp2 = LIST_FIRST(&gp->consumer); 352 KASSERT(cp2 != NULL, ("gv_save_config: null cp2")); 353 } else 354 cp2 = cp; 355 356 vhdr = g_malloc(GV_HDR_LEN, M_WAITOK | M_ZERO); 357 vhdr->magic = GV_MAGIC; 358 vhdr->config_length = GV_CFG_LEN; 359 360 hdr = d->hdr; 361 if (hdr == NULL) { 362 printf("GEOM_VINUM: drive %s has NULL hdr\n", d->name); 363 g_free(vhdr); 364 return; 365 } 366 microtime(&hdr->label.last_update); 367 bcopy(&hdr->label, &vhdr->label, sizeof(struct gv_label)); 368 369 sb = sbuf_new(NULL, NULL, GV_CFG_LEN, SBUF_FIXEDLEN); 370 gv_format_config(sc, sb, 1, NULL); 371 sbuf_finish(sb); 372 373 error = g_access(cp2, 0, 1, 0); 374 if (error) { 375 printf("GEOM_VINUM: g_access failed on drive %s, errno %d\n", 376 d->name, error); 377 sbuf_delete(sb); 378 g_free(vhdr); 379 return; 380 } 381 g_topology_unlock(); 382 383 do { 384 error = gv_write_header(cp2, vhdr); 385 if (error) { 386 printf("GEOM_VINUM: writing vhdr failed on drive %s, " 387 "errno %d", d->name, error); 388 break; 389 } 390 391 error = g_write_data(cp2, GV_CFG_OFFSET, sbuf_data(sb), 392 GV_CFG_LEN); 393 if (error) { 394 printf("GEOM_VINUM: writing first config copy failed " 395 "on drive %s, errno %d", d->name, error); 396 break; 397 } 398 399 error = g_write_data(cp2, GV_CFG_OFFSET + GV_CFG_LEN, 400 sbuf_data(sb), GV_CFG_LEN); 401 if (error) 402 printf("GEOM_VINUM: writing second config copy failed " 403 "on drive %s, errno %d", d->name, error); 404 } while (0); 405 406 g_topology_lock(); 407 g_access(cp2, 0, -1, 0); 408 sbuf_delete(sb); 409 g_free(vhdr); 410 411 if (d->geom != NULL) 412 gv_drive_modify(d); 413} 414 415/* This resembles g_slice_access(). */ 416static int 417gv_drive_access(struct g_provider *pp, int dr, int dw, int de) 418{ 419 struct g_geom *gp; 420 struct g_consumer *cp; 421 struct g_provider *pp2; 422 struct gv_drive *d; 423 struct gv_sd *s, *s2; 424 int error; 425 426 gp = pp->geom; 427 cp = LIST_FIRST(&gp->consumer); 428 if (cp == NULL) 429 return (0); 430 431 d = gp->softc; 432 if (d == NULL) 433 return (0); 434 435 s = pp->private; 436 KASSERT(s != NULL, ("gv_drive_access: NULL s")); 437 438 LIST_FOREACH(s2, &d->subdisks, from_drive) { 439 if (s == s2) 440 continue; 441 if (s->drive_offset + s->size <= s2->drive_offset) 442 continue; 443 if (s2->drive_offset + s2->size <= s->drive_offset) 444 continue; 445 446 /* Overlap. */ 447 pp2 = s2->provider; 448 KASSERT(s2 != NULL, ("gv_drive_access: NULL s2")); 449 if ((pp->acw + dw) > 0 && pp2->ace > 0) 450 return (EPERM); 451 if ((pp->ace + de) > 0 && pp2->acw > 0) 452 return (EPERM); 453 } 454 455 error = g_access(cp, dr, dw, de); 456 return (error); 457} 458 459static void 460gv_drive_done(struct bio *bp) 461{ 462 struct gv_drive *d; 463 464 /* Put the BIO on the worker queue again. */ 465 d = bp->bio_from->geom->softc; 466 bp->bio_cflags |= GV_BIO_DONE; 467 mtx_lock(&d->bqueue_mtx); 468 bioq_insert_tail(d->bqueue, bp); 469 wakeup(d); 470 mtx_unlock(&d->bqueue_mtx); 471} 472 473 474static void 475gv_drive_start(struct bio *bp) 476{ 477 struct gv_drive *d; 478 struct gv_sd *s; 479 480 switch (bp->bio_cmd) { 481 case BIO_READ: 482 case BIO_WRITE: 483 case BIO_DELETE: 484 break; 485 case BIO_GETATTR: 486 default: 487 g_io_deliver(bp, EOPNOTSUPP); 488 return; 489 } 490 491 s = bp->bio_to->private; 492 if ((s->state == GV_SD_DOWN) || (s->state == GV_SD_STALE)) { 493 g_io_deliver(bp, ENXIO); 494 return; 495 } 496 497 d = bp->bio_to->geom->softc; 498 499 /* 500 * Put the BIO on the worker queue, where the worker thread will pick 501 * it up. 502 */ 503 mtx_lock(&d->bqueue_mtx); 504 bioq_disksort(d->bqueue, bp); 505 wakeup(d); 506 mtx_unlock(&d->bqueue_mtx); 507 508} 509 510static void 511gv_drive_worker(void *arg) 512{ 513 struct bio *bp, *cbp; 514 struct g_geom *gp; 515 struct g_provider *pp; 516 struct gv_drive *d; 517 struct gv_sd *s; 518 int error; 519 520 d = arg; 521 522 mtx_lock(&d->bqueue_mtx); 523 for (;;) { 524 /* We were signaled to exit. */ 525 if (d->flags & GV_DRIVE_THREAD_DIE) 526 break; 527 528 /* Take the first BIO from out queue. */ 529 bp = bioq_takefirst(d->bqueue); 530 if (bp == NULL) { 531 msleep(d, &d->bqueue_mtx, PRIBIO, "-", hz/10); 532 continue; 533 } 534 mtx_unlock(&d->bqueue_mtx); 535 536 pp = bp->bio_to; 537 gp = pp->geom; 538 539 /* Completed request. */ 540 if (bp->bio_cflags & GV_BIO_DONE) { 541 error = bp->bio_error; 542 543 /* Deliver the original request. */ 544 g_std_done(bp); 545 546 /* The request had an error, we need to clean up. */ 547 if (error != 0) { 548 g_topology_lock(); 549 gv_set_drive_state(d, GV_DRIVE_DOWN, 550 GV_SETSTATE_FORCE | GV_SETSTATE_CONFIG); 551 g_topology_unlock(); 552 g_post_event(gv_drive_dead, d, M_WAITOK, d, 553 NULL); 554 } 555 556 /* New request, needs to be sent downwards. */ 557 } else { 558 s = pp->private; 559 560 if ((s->state == GV_SD_DOWN) || 561 (s->state == GV_SD_STALE)) { 562 g_io_deliver(bp, ENXIO); 563 mtx_lock(&d->bqueue_mtx); 564 continue; 565 } 566 if (bp->bio_offset > s->size) { 567 g_io_deliver(bp, EINVAL); 568 mtx_lock(&d->bqueue_mtx); 569 continue; 570 } 571 572 cbp = g_clone_bio(bp); 573 if (cbp == NULL) { 574 g_io_deliver(bp, ENOMEM); 575 mtx_lock(&d->bqueue_mtx); 576 continue; 577 } 578 if (cbp->bio_offset + cbp->bio_length > s->size) 579 cbp->bio_length = s->size - 580 cbp->bio_offset; 581 cbp->bio_done = gv_drive_done; 582 cbp->bio_offset += s->drive_offset; 583 g_io_request(cbp, LIST_FIRST(&gp->consumer)); 584 } 585 586 mtx_lock(&d->bqueue_mtx); 587 } 588 589 while ((bp = bioq_takefirst(d->bqueue)) != NULL) { 590 mtx_unlock(&d->bqueue_mtx); 591 if (bp->bio_cflags & GV_BIO_DONE) 592 g_std_done(bp); 593 else 594 g_io_deliver(bp, ENXIO); 595 mtx_lock(&d->bqueue_mtx); 596 } 597 mtx_unlock(&d->bqueue_mtx); 598 d->flags |= GV_DRIVE_THREAD_DEAD; 599 600 kproc_exit(ENXIO); 601} 602 603 604static void 605gv_drive_orphan(struct g_consumer *cp) 606{ 607 struct g_geom *gp; 608 struct gv_drive *d; 609 610 g_topology_assert(); 611 gp = cp->geom; 612 g_trace(G_T_TOPOLOGY, "gv_drive_orphan(%s)", gp->name); 613 d = gp->softc; 614 if (d != NULL) { 615 gv_set_drive_state(d, GV_DRIVE_DOWN, 616 GV_SETSTATE_FORCE | GV_SETSTATE_CONFIG); 617 g_post_event(gv_drive_dead, d, M_WAITOK, d, NULL); 618 } else 619 g_wither_geom(gp, ENXIO); 620} 621 622static struct g_geom * 623gv_drive_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 624{ 625 struct g_geom *gp, *gp2; 626 struct g_consumer *cp; 627 struct gv_drive *d; 628 struct gv_sd *s; 629 struct gv_softc *sc; 630 struct gv_freelist *fl; 631 struct gv_hdr *vhdr; 632 int error; 633 char *buf, errstr[ERRBUFSIZ]; 634 635 vhdr = NULL; 636 d = NULL; 637 638 g_trace(G_T_TOPOLOGY, "gv_drive_taste(%s, %s)", mp->name, pp->name); 639 g_topology_assert(); 640 641 /* Find the VINUM class and its associated geom. */ 642 gp2 = find_vinum_geom(); 643 if (gp2 == NULL) 644 return (NULL); 645 sc = gp2->softc; 646 647 gp = g_new_geomf(mp, "%s.vinumdrive", pp->name); 648 gp->start = gv_drive_start; 649 gp->orphan = gv_drive_orphan; 650 gp->access = gv_drive_access; 651 gp->start = gv_drive_start; 652 653 cp = g_new_consumer(gp); 654 g_attach(cp, pp); 655 error = g_access(cp, 1, 0, 0); 656 if (error) { 657 g_detach(cp); 658 g_destroy_consumer(cp); 659 g_destroy_geom(gp); 660 return (NULL); 661 } 662 663 g_topology_unlock(); 664 665 /* Now check if the provided slice is a valid vinum drive. */ 666 do { 667 vhdr = g_malloc(GV_HDR_LEN, M_WAITOK | M_ZERO); 668 error = gv_read_header(cp, vhdr); 669 if (error) { 670 g_free(vhdr); 671 break; 672 } 673 674 /* A valid vinum drive, let's parse the on-disk information. */ 675 buf = g_read_data(cp, GV_CFG_OFFSET, GV_CFG_LEN, NULL); 676 if (buf == NULL) { 677 g_free(vhdr); 678 break; 679 } 680 g_topology_lock(); 681 gv_parse_config(sc, buf, 1); 682 g_free(buf); 683 684 /* 685 * Let's see if this drive is already known in the 686 * configuration. 687 */ 688 d = gv_find_drive(sc, vhdr->label.name); 689 690 /* We already know about this drive. */ 691 if (d != NULL) { 692 /* Check if this drive already has a geom. */ 693 if (d->geom != NULL) { 694 g_topology_unlock(); 695 g_free(vhdr); 696 break; 697 } 698 bcopy(vhdr, d->hdr, sizeof(*vhdr)); 699 g_free(vhdr); 700 701 /* This is a new drive. */ 702 } else { 703 d = g_malloc(sizeof(*d), M_WAITOK | M_ZERO); 704 705 /* Initialize all needed variables. */ 706 d->size = pp->mediasize - GV_DATA_START; 707 d->avail = d->size; 708 d->hdr = vhdr; 709 strncpy(d->name, vhdr->label.name, GV_MAXDRIVENAME); 710 LIST_INIT(&d->subdisks); 711 LIST_INIT(&d->freelist); 712 713 /* We also need a freelist entry. */ 714 fl = g_malloc(sizeof(*fl), M_WAITOK | M_ZERO); 715 fl->offset = GV_DATA_START; 716 fl->size = d->avail; 717 LIST_INSERT_HEAD(&d->freelist, fl, freelist); 718 d->freelist_entries = 1; 719 720 /* Save it into the main configuration. */ 721 LIST_INSERT_HEAD(&sc->drives, d, drive); 722 } 723 724 /* 725 * Create bio queue, queue mutex and a worker thread, if 726 * necessary. 727 */ 728 if (d->bqueue == NULL) { 729 d->bqueue = g_malloc(sizeof(struct bio_queue_head), 730 M_WAITOK | M_ZERO); 731 bioq_init(d->bqueue); 732 } 733 if (mtx_initialized(&d->bqueue_mtx) == 0) 734 mtx_init(&d->bqueue_mtx, "gv_drive", NULL, MTX_DEF); 735 736 if (!(d->flags & GV_DRIVE_THREAD_ACTIVE)) { 737 kproc_create(gv_drive_worker, d, NULL, 0, 0, 738 "gv_d %s", d->name); 739 d->flags |= GV_DRIVE_THREAD_ACTIVE; 740 } 741 742 g_access(cp, -1, 0, 0); 743 744 gp->softc = d; 745 d->geom = gp; 746 d->vinumconf = sc; 747 strncpy(d->device, pp->name, GV_MAXDRIVENAME); 748 749 /* 750 * Find out which subdisks belong to this drive and crosslink 751 * them. 752 */ 753 LIST_FOREACH(s, &sc->subdisks, sd) { 754 if (!strncmp(s->drive, d->name, GV_MAXDRIVENAME)) 755 /* XXX: errors ignored */ 756 gv_sd_to_drive(sc, d, s, errstr, 757 sizeof(errstr)); 758 } 759 760 /* This drive is now up for sure. */ 761 gv_set_drive_state(d, GV_DRIVE_UP, 0); 762 763 /* 764 * If there are subdisks on this drive, we need to create 765 * providers for them. 766 */ 767 if (d->sdcount) 768 gv_drive_modify(d); 769 770 return (gp); 771 772 } while (0); 773 774 g_topology_lock(); 775 g_access(cp, -1, 0, 0); 776 777 g_detach(cp); 778 g_destroy_consumer(cp); 779 g_destroy_geom(gp); 780 return (NULL); 781} 782 783/* 784 * Modify the providers for the given drive 'd'. It is assumed that the 785 * subdisk list of 'd' is already correctly set up. 786 */ 787void 788gv_drive_modify(struct gv_drive *d) 789{ 790 struct g_geom *gp; 791 struct g_consumer *cp; 792 struct g_provider *pp, *pp2; 793 struct gv_sd *s; 794 795 KASSERT(d != NULL, ("gv_drive_modify: null d")); 796 gp = d->geom; 797 KASSERT(gp != NULL, ("gv_drive_modify: null gp")); 798 cp = LIST_FIRST(&gp->consumer); 799 KASSERT(cp != NULL, ("gv_drive_modify: null cp")); 800 pp = cp->provider; 801 KASSERT(pp != NULL, ("gv_drive_modify: null pp")); 802 803 g_topology_assert(); 804 805 LIST_FOREACH(s, &d->subdisks, from_drive) { 806 /* This subdisk already has a provider. */ 807 if (s->provider != NULL) 808 continue; 809 pp2 = g_new_providerf(gp, "gvinum/sd/%s", s->name); 810 pp2->mediasize = s->size; 811 pp2->sectorsize = pp->sectorsize; 812 g_error_provider(pp2, 0); 813 s->provider = pp2; 814 pp2->private = s; 815 } 816} 817 818static void 819gv_drive_dead(void *arg, int flag) 820{ 821 struct g_geom *gp; 822 struct g_consumer *cp; 823 struct gv_drive *d; 824 struct gv_sd *s; 825 826 g_topology_assert(); 827 KASSERT(arg != NULL, ("gv_drive_dead: NULL arg")); 828 829 if (flag == EV_CANCEL) 830 return; 831 832 d = arg; 833 if (d->state != GV_DRIVE_DOWN) 834 return; 835 836 g_trace(G_T_TOPOLOGY, "gv_drive_dead(%s)", d->name); 837 838 gp = d->geom; 839 if (gp == NULL) 840 return; 841 842 LIST_FOREACH(cp, &gp->consumer, consumer) { 843 if (cp->nstart != cp->nend) { 844 printf("GEOM_VINUM: dead drive '%s' has still " 845 "active requests, can't detach consumer\n", 846 d->name); 847 g_post_event(gv_drive_dead, d, M_WAITOK, d, 848 NULL); 849 return; 850 } 851 if (cp->acr != 0 || cp->acw != 0 || cp->ace != 0) 852 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 853 } 854 855 printf("GEOM_VINUM: lost drive '%s'\n", d->name); 856 d->geom = NULL; 857 LIST_FOREACH(s, &d->subdisks, from_drive) { 858 s->provider = NULL; 859 s->consumer = NULL; 860 } 861 gv_kill_drive_thread(d); 862 gp->softc = NULL; 863 g_wither_geom(gp, ENXIO); 864} 865 866static int 867gv_drive_destroy_geom(struct gctl_req *req, struct g_class *mp, 868 struct g_geom *gp) 869{ 870 struct gv_drive *d; 871 872 g_trace(G_T_TOPOLOGY, "gv_drive_destroy_geom: %s", gp->name); 873 g_topology_assert(); 874 875 d = gp->softc; 876 gv_kill_drive_thread(d); 877 878 g_wither_geom(gp, ENXIO); 879 return (0); 880} 881 882#define VINUMDRIVE_CLASS_NAME "VINUMDRIVE" 883 884static struct g_class g_vinum_drive_class = { 885 .name = VINUMDRIVE_CLASS_NAME, 886 .version = G_VERSION, 887 .taste = gv_drive_taste, 888 .destroy_geom = gv_drive_destroy_geom 889}; 890 891DECLARE_GEOM_CLASS(g_vinum_drive_class, g_vinum_drive); 892