1 /* $OpenBSD: softraid.c,v 1.343 2014/10/30 19:07:54 tedu Exp $ */ 2 /* 3 * Copyright (c) 2007, 2008, 2009 Marco Peereboom <marco@peereboom.us> 4 * Copyright (c) 2008 Chris Kuethe <ckuethe@openbsd.org> 5 * Copyright (c) 2009 Joel Sing <jsing@openbsd.org> 6 * 7 * Permission to use, copy, modify, and distribute this software for any 8 * purpose with or without fee is hereby granted, provided that the above 9 * copyright notice and this permission notice appear in all copies. 10 * 11 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES 12 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF 13 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR 14 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES 15 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN 16 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF 17 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 18 */ 19 20 #include "bio.h" 21 22 #include <sys/param.h> 23 #include <sys/systm.h> 24 #include <sys/buf.h> 25 #include <sys/device.h> 26 #include <sys/ioctl.h> 27 #include <sys/malloc.h> 28 #include <sys/pool.h> 29 #include <sys/kernel.h> 30 #include <sys/disk.h> 31 #include <sys/rwlock.h> 32 #include <sys/queue.h> 33 #include <sys/fcntl.h> 34 #include <sys/disklabel.h> 35 #include <sys/vnode.h> 36 #include <sys/mount.h> 37 #include <sys/sensors.h> 38 #include <sys/stat.h> 39 #include <sys/conf.h> 40 #include <sys/uio.h> 41 #include <sys/task.h> 42 #include <sys/kthread.h> 43 #include <sys/dkio.h> 44 45 #include <crypto/cryptodev.h> 46 47 #include <scsi/scsi_all.h> 48 #include <scsi/scsiconf.h> 49 #include <scsi/scsi_disk.h> 50 51 #include <dev/softraidvar.h> 52 #include <dev/rndvar.h> 53 54 #ifdef HIBERNATE 55 #include <lib/libsa/aes_xts.h> 56 #include <sys/hibernate.h> 57 #include <scsi/sdvar.h> 58 #endif /* HIBERNATE */ 59 60 /* #define SR_FANCY_STATS */ 61 62 #ifdef SR_DEBUG 63 #define SR_FANCY_STATS 64 uint32_t sr_debug = 0 65 /* | SR_D_CMD */ 66 /* | SR_D_MISC */ 67 /* | SR_D_INTR */ 68 /* | SR_D_IOCTL */ 69 /* | SR_D_CCB */ 70 /* | SR_D_WU */ 71 /* | SR_D_META */ 72 /* | SR_D_DIS */ 73 /* | SR_D_STATE */ 74 /* | SR_D_REBUILD */ 75 ; 76 #endif 77 78 struct sr_softc *softraid0; 79 struct sr_uuid sr_bootuuid; 80 u_int8_t sr_bootkey[SR_CRYPTO_MAXKEYBYTES]; 81 82 int sr_match(struct device *, void *, void *); 83 void sr_attach(struct device *, struct device *, void *); 84 int sr_detach(struct device *, int); 85 void sr_map_root(void); 86 87 struct cfattach softraid_ca = { 88 sizeof(struct sr_softc), sr_match, sr_attach, sr_detach, 89 }; 90 91 struct cfdriver softraid_cd = { 92 NULL, "softraid", DV_DULL 93 }; 94 95 /* scsi & discipline */ 96 void sr_scsi_cmd(struct scsi_xfer *); 97 void sr_minphys(struct buf *, struct scsi_link *); 98 int sr_scsi_probe(struct scsi_link *); 99 void sr_copy_internal_data(struct scsi_xfer *, 100 void *, size_t); 101 int sr_scsi_ioctl(struct scsi_link *, u_long, 102 caddr_t, int); 103 int sr_bio_ioctl(struct device *, u_long, caddr_t); 104 int sr_ioctl_inq(struct sr_softc *, struct bioc_inq *); 105 int sr_ioctl_vol(struct sr_softc *, struct bioc_vol *); 106 int sr_ioctl_disk(struct sr_softc *, struct bioc_disk *); 107 int sr_ioctl_setstate(struct sr_softc *, 108 struct bioc_setstate *); 109 int sr_ioctl_createraid(struct sr_softc *, 110 struct bioc_createraid *, int, void *); 111 int sr_ioctl_deleteraid(struct sr_softc *, 112 struct bioc_deleteraid *); 113 int sr_ioctl_discipline(struct sr_softc *, 114 struct bioc_discipline *); 115 int sr_ioctl_installboot(struct sr_softc *, 116 struct bioc_installboot *); 117 void sr_chunks_unwind(struct sr_softc *, 118 struct sr_chunk_head *); 119 void sr_discipline_free(struct sr_discipline *); 120 void sr_discipline_shutdown(struct sr_discipline *, int); 121 int sr_discipline_init(struct sr_discipline *, int); 122 int sr_alloc_resources(struct sr_discipline *); 123 void sr_free_resources(struct sr_discipline *); 124 void sr_set_chunk_state(struct sr_discipline *, int, int); 125 void sr_set_vol_state(struct sr_discipline *); 126 127 /* utility functions */ 128 void sr_shutdown(void); 129 void sr_uuid_generate(struct sr_uuid *); 130 char *sr_uuid_format(struct sr_uuid *); 131 void sr_uuid_print(struct sr_uuid *, int); 132 void sr_checksum_print(u_int8_t *); 133 int sr_boot_assembly(struct sr_softc *); 134 int sr_already_assembled(struct sr_discipline *); 135 int sr_hotspare(struct sr_softc *, dev_t); 136 void sr_hotspare_rebuild(struct sr_discipline *); 137 int sr_rebuild_init(struct sr_discipline *, dev_t, int); 138 void sr_rebuild_start(void *); 139 void sr_rebuild_thread(void *); 140 void sr_rebuild(struct sr_discipline *); 141 void sr_roam_chunks(struct sr_discipline *); 142 int sr_chunk_in_use(struct sr_softc *, dev_t); 143 int sr_rw(struct sr_softc *, dev_t, char *, size_t, 144 daddr_t, long); 145 void sr_wu_done_callback(void *, void *); 146 147 /* don't include these on RAMDISK */ 148 #ifndef SMALL_KERNEL 149 void sr_sensors_refresh(void *); 150 int sr_sensors_create(struct sr_discipline *); 151 void sr_sensors_delete(struct sr_discipline *); 152 #endif 153 154 /* metadata */ 155 int sr_meta_probe(struct sr_discipline *, dev_t *, int); 156 int sr_meta_attach(struct sr_discipline *, int, int); 157 int sr_meta_rw(struct sr_discipline *, dev_t, void *, 158 size_t, daddr_t, long); 159 int sr_meta_clear(struct sr_discipline *); 160 void sr_meta_init(struct sr_discipline *, int, int); 161 void sr_meta_init_complete(struct sr_discipline *); 162 void sr_meta_opt_handler(struct sr_discipline *, 163 struct sr_meta_opt_hdr *); 164 165 /* hotplug magic */ 166 void sr_disk_attach(struct disk *, int); 167 168 struct sr_hotplug_list { 169 void (*sh_hotplug)(struct sr_discipline *, 170 struct disk *, int); 171 struct sr_discipline *sh_sd; 172 173 SLIST_ENTRY(sr_hotplug_list) shl_link; 174 }; 175 SLIST_HEAD(sr_hotplug_list_head, sr_hotplug_list); 176 177 struct sr_hotplug_list_head sr_hotplug_callbacks; 178 extern void (*softraid_disk_attach)(struct disk *, int); 179 180 /* scsi glue */ 181 struct scsi_adapter sr_switch = { 182 sr_scsi_cmd, sr_minphys, sr_scsi_probe, NULL, sr_scsi_ioctl 183 }; 184 185 /* native metadata format */ 186 int sr_meta_native_bootprobe(struct sr_softc *, dev_t, 187 struct sr_boot_chunk_head *); 188 #define SR_META_NOTCLAIMED (0) 189 #define SR_META_CLAIMED (1) 190 int sr_meta_native_probe(struct sr_softc *, 191 struct sr_chunk *); 192 int sr_meta_native_attach(struct sr_discipline *, int); 193 int sr_meta_native_write(struct sr_discipline *, dev_t, 194 struct sr_metadata *,void *); 195 196 #ifdef SR_DEBUG 197 void sr_meta_print(struct sr_metadata *); 198 #else 199 #define sr_meta_print(m) 200 #endif 201 202 /* the metadata driver should remain stateless */ 203 struct sr_meta_driver { 204 daddr_t smd_offset; /* metadata location */ 205 u_int32_t smd_size; /* size of metadata */ 206 207 int (*smd_probe)(struct sr_softc *, 208 struct sr_chunk *); 209 int (*smd_attach)(struct sr_discipline *, int); 210 int (*smd_detach)(struct sr_discipline *); 211 int (*smd_read)(struct sr_discipline *, dev_t, 212 struct sr_metadata *, void *); 213 int (*smd_write)(struct sr_discipline *, dev_t, 214 struct sr_metadata *, void *); 215 int (*smd_validate)(struct sr_discipline *, 216 struct sr_metadata *, void *); 217 } smd[] = { 218 { SR_META_OFFSET, SR_META_SIZE * 512, 219 sr_meta_native_probe, sr_meta_native_attach, NULL, 220 sr_meta_native_read, sr_meta_native_write, NULL }, 221 { 0, 0, NULL, NULL, NULL, NULL } 222 }; 223 224 int 225 sr_meta_attach(struct sr_discipline *sd, int chunk_no, int force) 226 { 227 struct sr_softc *sc = sd->sd_sc; 228 struct sr_chunk_head *cl; 229 struct sr_chunk *ch_entry, *chunk1, *chunk2; 230 int rv = 1, i = 0; 231 232 DNPRINTF(SR_D_META, "%s: sr_meta_attach(%d)\n", DEVNAME(sc)); 233 234 /* in memory copy of metadata */ 235 sd->sd_meta = malloc(SR_META_SIZE * 512, M_DEVBUF, M_ZERO | M_NOWAIT); 236 if (!sd->sd_meta) { 237 sr_error(sc, "could not allocate memory for metadata"); 238 goto bad; 239 } 240 241 if (sd->sd_meta_type != SR_META_F_NATIVE) { 242 /* in memory copy of foreign metadata */ 243 sd->sd_meta_foreign = malloc(smd[sd->sd_meta_type].smd_size, 244 M_DEVBUF, M_ZERO | M_NOWAIT); 245 if (!sd->sd_meta_foreign) { 246 /* unwind frees sd_meta */ 247 sr_error(sc, "could not allocate memory for foreign " 248 "metadata"); 249 goto bad; 250 } 251 } 252 253 /* we have a valid list now create an array index */ 254 cl = &sd->sd_vol.sv_chunk_list; 255 sd->sd_vol.sv_chunks = mallocarray(chunk_no, sizeof(struct sr_chunk *), 256 M_DEVBUF, M_WAITOK | M_ZERO); 257 258 /* fill out chunk array */ 259 i = 0; 260 SLIST_FOREACH(ch_entry, cl, src_link) 261 sd->sd_vol.sv_chunks[i++] = ch_entry; 262 263 /* attach metadata */ 264 if (smd[sd->sd_meta_type].smd_attach(sd, force)) 265 goto bad; 266 267 /* Force chunks into correct order now that metadata is attached. */ 268 SLIST_FOREACH(ch_entry, cl, src_link) 269 SLIST_REMOVE(cl, ch_entry, sr_chunk, src_link); 270 for (i = 0; i < chunk_no; i++) { 271 ch_entry = sd->sd_vol.sv_chunks[i]; 272 chunk2 = NULL; 273 SLIST_FOREACH(chunk1, cl, src_link) { 274 if (chunk1->src_meta.scmi.scm_chunk_id > 275 ch_entry->src_meta.scmi.scm_chunk_id) 276 break; 277 chunk2 = chunk1; 278 } 279 if (chunk2 == NULL) 280 SLIST_INSERT_HEAD(cl, ch_entry, src_link); 281 else 282 SLIST_INSERT_AFTER(chunk2, ch_entry, src_link); 283 } 284 i = 0; 285 SLIST_FOREACH(ch_entry, cl, src_link) 286 sd->sd_vol.sv_chunks[i++] = ch_entry; 287 288 rv = 0; 289 bad: 290 return (rv); 291 } 292 293 int 294 sr_meta_probe(struct sr_discipline *sd, dev_t *dt, int no_chunk) 295 { 296 struct sr_softc *sc = sd->sd_sc; 297 struct vnode *vn; 298 struct sr_chunk *ch_entry, *ch_prev = NULL; 299 struct sr_chunk_head *cl; 300 char devname[32]; 301 int i, d, type, found, prevf, error; 302 dev_t dev; 303 304 DNPRINTF(SR_D_META, "%s: sr_meta_probe(%d)\n", DEVNAME(sc), no_chunk); 305 306 if (no_chunk == 0) 307 goto unwind; 308 309 cl = &sd->sd_vol.sv_chunk_list; 310 311 for (d = 0, prevf = SR_META_F_INVALID; d < no_chunk; d++) { 312 ch_entry = malloc(sizeof(struct sr_chunk), M_DEVBUF, 313 M_WAITOK | M_ZERO); 314 /* keep disks in user supplied order */ 315 if (ch_prev) 316 SLIST_INSERT_AFTER(ch_prev, ch_entry, src_link); 317 else 318 SLIST_INSERT_HEAD(cl, ch_entry, src_link); 319 ch_prev = ch_entry; 320 dev = dt[d]; 321 ch_entry->src_dev_mm = dev; 322 323 if (dev == NODEV) { 324 ch_entry->src_meta.scm_status = BIOC_SDOFFLINE; 325 continue; 326 } else { 327 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 328 if (bdevvp(dev, &vn)) { 329 sr_error(sc, "sr_meta_probe: cannot allocate " 330 "vnode"); 331 goto unwind; 332 } 333 334 /* 335 * XXX leaving dev open for now; move this to attach 336 * and figure out the open/close dance for unwind. 337 */ 338 error = VOP_OPEN(vn, FREAD | FWRITE, NOCRED, curproc); 339 if (error) { 340 DNPRINTF(SR_D_META,"%s: sr_meta_probe can't " 341 "open %s\n", DEVNAME(sc), devname); 342 vput(vn); 343 goto unwind; 344 } 345 346 strlcpy(ch_entry->src_devname, devname, 347 sizeof(ch_entry->src_devname)); 348 ch_entry->src_vn = vn; 349 } 350 351 /* determine if this is a device we understand */ 352 for (i = 0, found = SR_META_F_INVALID; smd[i].smd_probe; i++) { 353 type = smd[i].smd_probe(sc, ch_entry); 354 if (type == SR_META_F_INVALID) 355 continue; 356 else { 357 found = type; 358 break; 359 } 360 } 361 362 if (found == SR_META_F_INVALID) 363 goto unwind; 364 if (prevf == SR_META_F_INVALID) 365 prevf = found; 366 if (prevf != found) { 367 DNPRINTF(SR_D_META, "%s: prevf != found\n", 368 DEVNAME(sc)); 369 goto unwind; 370 } 371 } 372 373 return (prevf); 374 unwind: 375 return (SR_META_F_INVALID); 376 } 377 378 void 379 sr_meta_getdevname(struct sr_softc *sc, dev_t dev, char *buf, int size) 380 { 381 int maj, unit, part; 382 char *name; 383 384 DNPRINTF(SR_D_META, "%s: sr_meta_getdevname(%p, %d)\n", 385 DEVNAME(sc), buf, size); 386 387 if (!buf) 388 return; 389 390 maj = major(dev); 391 part = DISKPART(dev); 392 unit = DISKUNIT(dev); 393 394 name = findblkname(maj); 395 if (name == NULL) 396 return; 397 398 snprintf(buf, size, "%s%d%c", name, unit, part + 'a'); 399 } 400 401 int 402 sr_rw(struct sr_softc *sc, dev_t dev, char *buf, size_t size, daddr_t offset, 403 long flags) 404 { 405 struct vnode *vp; 406 struct buf b; 407 size_t bufsize, dma_bufsize; 408 int rv = 1; 409 char *dma_buf; 410 411 DNPRINTF(SR_D_MISC, "%s: sr_rw(0x%x, %p, %zu, %lld 0x%x)\n", 412 DEVNAME(sc), dev, buf, size, (long long)offset, flags); 413 414 dma_bufsize = (size > MAXPHYS) ? MAXPHYS : size; 415 dma_buf = dma_alloc(dma_bufsize, PR_WAITOK); 416 417 if (bdevvp(dev, &vp)) { 418 printf("%s: sr_rw: failed to allocate vnode\n", DEVNAME(sc)); 419 goto done; 420 } 421 422 while (size > 0) { 423 DNPRINTF(SR_D_MISC, "%s: dma_buf %p, size %d, offset %llu)\n", 424 DEVNAME(sc), dma_buf, size, offset); 425 426 bufsize = (size > MAXPHYS) ? MAXPHYS : size; 427 if (flags == B_WRITE) 428 bcopy(buf, dma_buf, bufsize); 429 430 bzero(&b, sizeof(b)); 431 b.b_flags = flags | B_PHYS; 432 b.b_proc = curproc; 433 b.b_dev = dev; 434 b.b_iodone = NULL; 435 b.b_error = 0; 436 b.b_blkno = offset; 437 b.b_data = dma_buf; 438 b.b_bcount = bufsize; 439 b.b_bufsize = bufsize; 440 b.b_resid = bufsize; 441 b.b_vp = vp; 442 443 if ((b.b_flags & B_READ) == 0) 444 vp->v_numoutput++; 445 446 LIST_INIT(&b.b_dep); 447 VOP_STRATEGY(&b); 448 biowait(&b); 449 450 if (b.b_flags & B_ERROR) { 451 printf("%s: I/O error %d on dev 0x%x at block %llu\n", 452 DEVNAME(sc), b.b_error, dev, b.b_blkno); 453 goto done; 454 } 455 456 if (flags == B_READ) 457 bcopy(dma_buf, buf, bufsize); 458 459 size -= bufsize; 460 buf += bufsize; 461 offset += howmany(bufsize, DEV_BSIZE); 462 } 463 464 rv = 0; 465 466 done: 467 if (vp) 468 vput(vp); 469 470 dma_free(dma_buf, dma_bufsize); 471 472 return (rv); 473 } 474 475 int 476 sr_meta_rw(struct sr_discipline *sd, dev_t dev, void *md, size_t size, 477 daddr_t offset, long flags) 478 { 479 int rv = 1; 480 481 DNPRINTF(SR_D_META, "%s: sr_meta_rw(0x%x, %p, %zu, %lld 0x%x)\n", 482 DEVNAME(sd->sd_sc), dev, md, size, (long long)offset, flags); 483 484 if (md == NULL) { 485 printf("%s: sr_meta_rw: invalid metadata pointer\n", 486 DEVNAME(sd->sd_sc)); 487 goto done; 488 } 489 490 rv = sr_rw(sd->sd_sc, dev, md, size, offset, flags); 491 492 done: 493 return (rv); 494 } 495 496 int 497 sr_meta_clear(struct sr_discipline *sd) 498 { 499 struct sr_softc *sc = sd->sd_sc; 500 struct sr_chunk_head *cl = &sd->sd_vol.sv_chunk_list; 501 struct sr_chunk *ch_entry; 502 void *m; 503 int rv = 1; 504 505 DNPRINTF(SR_D_META, "%s: sr_meta_clear\n", DEVNAME(sc)); 506 507 if (sd->sd_meta_type != SR_META_F_NATIVE) { 508 sr_error(sc, "cannot clear foreign metadata"); 509 goto done; 510 } 511 512 m = malloc(SR_META_SIZE * 512, M_DEVBUF, M_WAITOK | M_ZERO); 513 SLIST_FOREACH(ch_entry, cl, src_link) { 514 if (sr_meta_native_write(sd, ch_entry->src_dev_mm, m, NULL)) { 515 /* XXX mark disk offline */ 516 DNPRINTF(SR_D_META, "%s: sr_meta_clear failed to " 517 "clear %s\n", ch_entry->src_devname); 518 rv++; 519 continue; 520 } 521 bzero(&ch_entry->src_meta, sizeof(ch_entry->src_meta)); 522 } 523 524 bzero(sd->sd_meta, SR_META_SIZE * 512); 525 526 free(m, M_DEVBUF, SR_META_SIZE * 512); 527 rv = 0; 528 done: 529 return (rv); 530 } 531 532 void 533 sr_meta_init(struct sr_discipline *sd, int level, int no_chunk) 534 { 535 struct sr_softc *sc = sd->sd_sc; 536 struct sr_metadata *sm = sd->sd_meta; 537 struct sr_chunk_head *cl = &sd->sd_vol.sv_chunk_list; 538 struct sr_meta_chunk *scm; 539 struct sr_chunk *chunk; 540 int cid = 0; 541 u_int64_t max_chunk_sz = 0, min_chunk_sz = 0; 542 543 DNPRINTF(SR_D_META, "%s: sr_meta_init\n", DEVNAME(sc)); 544 545 if (!sm) 546 return; 547 548 /* Initialise volume metadata. */ 549 sm->ssdi.ssd_magic = SR_MAGIC; 550 sm->ssdi.ssd_version = SR_META_VERSION; 551 sm->ssdi.ssd_vol_flags = sd->sd_meta_flags; 552 sm->ssdi.ssd_volid = 0; 553 sm->ssdi.ssd_chunk_no = no_chunk; 554 sm->ssdi.ssd_level = level; 555 556 sm->ssd_data_offset = SR_DATA_OFFSET; 557 sm->ssd_ondisk = 0; 558 559 sr_uuid_generate(&sm->ssdi.ssd_uuid); 560 561 /* Initialise chunk metadata and get min/max chunk sizes. */ 562 SLIST_FOREACH(chunk, cl, src_link) { 563 scm = &chunk->src_meta; 564 scm->scmi.scm_size = chunk->src_size; 565 scm->scmi.scm_chunk_id = cid++; 566 scm->scm_status = BIOC_SDONLINE; 567 scm->scmi.scm_volid = 0; 568 strlcpy(scm->scmi.scm_devname, chunk->src_devname, 569 sizeof(scm->scmi.scm_devname)); 570 bcopy(&sm->ssdi.ssd_uuid, &scm->scmi.scm_uuid, 571 sizeof(scm->scmi.scm_uuid)); 572 sr_checksum(sc, scm, &scm->scm_checksum, 573 sizeof(scm->scm_checksum)); 574 575 if (min_chunk_sz == 0) 576 min_chunk_sz = scm->scmi.scm_size; 577 min_chunk_sz = MIN(min_chunk_sz, scm->scmi.scm_size); 578 max_chunk_sz = MAX(max_chunk_sz, scm->scmi.scm_size); 579 } 580 581 /* Equalize chunk sizes. */ 582 SLIST_FOREACH(chunk, cl, src_link) 583 chunk->src_meta.scmi.scm_coerced_size = min_chunk_sz; 584 585 sd->sd_vol.sv_chunk_minsz = min_chunk_sz; 586 sd->sd_vol.sv_chunk_maxsz = max_chunk_sz; 587 } 588 589 void 590 sr_meta_init_complete(struct sr_discipline *sd) 591 { 592 #ifdef SR_DEBUG 593 struct sr_softc *sc = sd->sd_sc; 594 #endif 595 struct sr_metadata *sm = sd->sd_meta; 596 597 DNPRINTF(SR_D_META, "%s: sr_meta_complete\n", DEVNAME(sc)); 598 599 /* Complete initialisation of volume metadata. */ 600 strlcpy(sm->ssdi.ssd_vendor, "OPENBSD", sizeof(sm->ssdi.ssd_vendor)); 601 snprintf(sm->ssdi.ssd_product, sizeof(sm->ssdi.ssd_product), 602 "SR %s", sd->sd_name); 603 snprintf(sm->ssdi.ssd_revision, sizeof(sm->ssdi.ssd_revision), 604 "%03d", sm->ssdi.ssd_version); 605 } 606 607 void 608 sr_meta_opt_handler(struct sr_discipline *sd, struct sr_meta_opt_hdr *om) 609 { 610 if (om->som_type != SR_OPT_BOOT) 611 panic("unknown optional metadata type"); 612 } 613 614 void 615 sr_meta_save_callback(void *arg1, void *arg2) 616 { 617 struct sr_discipline *sd = arg1; 618 int s; 619 620 s = splbio(); 621 622 if (sr_meta_save(arg1, SR_META_DIRTY)) 623 printf("%s: save metadata failed\n", DEVNAME(sd->sd_sc)); 624 625 sd->sd_must_flush = 0; 626 splx(s); 627 } 628 629 int 630 sr_meta_save(struct sr_discipline *sd, u_int32_t flags) 631 { 632 struct sr_softc *sc = sd->sd_sc; 633 struct sr_metadata *sm = sd->sd_meta, *m; 634 struct sr_meta_driver *s; 635 struct sr_chunk *src; 636 struct sr_meta_chunk *cm; 637 struct sr_workunit wu; 638 struct sr_meta_opt_hdr *omh; 639 struct sr_meta_opt_item *omi; 640 int i; 641 642 DNPRINTF(SR_D_META, "%s: sr_meta_save %s\n", 643 DEVNAME(sc), sd->sd_meta->ssd_devname); 644 645 if (!sm) { 646 printf("%s: no in memory copy of metadata\n", DEVNAME(sc)); 647 goto bad; 648 } 649 650 /* meta scratchpad */ 651 s = &smd[sd->sd_meta_type]; 652 m = malloc(SR_META_SIZE * 512, M_DEVBUF, M_ZERO | M_NOWAIT); 653 if (!m) { 654 printf("%s: could not allocate metadata scratch area\n", 655 DEVNAME(sc)); 656 goto bad; 657 } 658 659 /* from here on out metadata is updated */ 660 restart: 661 sm->ssd_ondisk++; 662 sm->ssd_meta_flags = flags; 663 bcopy(sm, m, sizeof(*m)); 664 665 /* Chunk metadata. */ 666 cm = (struct sr_meta_chunk *)(m + 1); 667 for (i = 0; i < sm->ssdi.ssd_chunk_no; i++) { 668 src = sd->sd_vol.sv_chunks[i]; 669 bcopy(&src->src_meta, cm, sizeof(*cm)); 670 cm++; 671 } 672 673 /* Optional metadata. */ 674 omh = (struct sr_meta_opt_hdr *)(cm); 675 SLIST_FOREACH(omi, &sd->sd_meta_opt, omi_link) { 676 DNPRINTF(SR_D_META, "%s: saving optional metadata type %u with " 677 "length %u\n", DEVNAME(sc), omi->omi_som->som_type, 678 omi->omi_som->som_length); 679 bzero(&omi->omi_som->som_checksum, MD5_DIGEST_LENGTH); 680 sr_checksum(sc, omi->omi_som, &omi->omi_som->som_checksum, 681 omi->omi_som->som_length); 682 bcopy(omi->omi_som, omh, omi->omi_som->som_length); 683 omh = (struct sr_meta_opt_hdr *)((u_int8_t *)omh + 684 omi->omi_som->som_length); 685 } 686 687 for (i = 0; i < sm->ssdi.ssd_chunk_no; i++) { 688 src = sd->sd_vol.sv_chunks[i]; 689 690 /* skip disks that are offline */ 691 if (src->src_meta.scm_status == BIOC_SDOFFLINE) 692 continue; 693 694 /* calculate metadata checksum for correct chunk */ 695 m->ssdi.ssd_chunk_id = i; 696 sr_checksum(sc, m, &m->ssd_checksum, 697 sizeof(struct sr_meta_invariant)); 698 699 #ifdef SR_DEBUG 700 DNPRINTF(SR_D_META, "%s: sr_meta_save %s: volid: %d " 701 "chunkid: %d checksum: ", 702 DEVNAME(sc), src->src_meta.scmi.scm_devname, 703 m->ssdi.ssd_volid, m->ssdi.ssd_chunk_id); 704 705 if (sr_debug & SR_D_META) 706 sr_checksum_print((u_int8_t *)&m->ssd_checksum); 707 DNPRINTF(SR_D_META, "\n"); 708 sr_meta_print(m); 709 #endif 710 711 /* translate and write to disk */ 712 if (s->smd_write(sd, src->src_dev_mm, m, NULL /* XXX */)) { 713 printf("%s: could not write metadata to %s\n", 714 DEVNAME(sc), src->src_devname); 715 /* restart the meta write */ 716 src->src_meta.scm_status = BIOC_SDOFFLINE; 717 /* XXX recalculate volume status */ 718 goto restart; 719 } 720 } 721 722 /* not all disciplines have sync */ 723 if (sd->sd_scsi_sync) { 724 bzero(&wu, sizeof(wu)); 725 wu.swu_flags |= SR_WUF_FAKE; 726 wu.swu_dis = sd; 727 sd->sd_scsi_sync(&wu); 728 } 729 free(m, M_DEVBUF, SR_META_SIZE * 512); 730 return (0); 731 bad: 732 return (1); 733 } 734 735 int 736 sr_meta_read(struct sr_discipline *sd) 737 { 738 struct sr_softc *sc = sd->sd_sc; 739 struct sr_chunk_head *cl = &sd->sd_vol.sv_chunk_list; 740 struct sr_metadata *sm; 741 struct sr_chunk *ch_entry; 742 struct sr_meta_chunk *cp; 743 struct sr_meta_driver *s; 744 void *fm = NULL; 745 int no_disk = 0, got_meta = 0; 746 747 DNPRINTF(SR_D_META, "%s: sr_meta_read\n", DEVNAME(sc)); 748 749 sm = malloc(SR_META_SIZE * 512, M_DEVBUF, M_WAITOK | M_ZERO); 750 s = &smd[sd->sd_meta_type]; 751 if (sd->sd_meta_type != SR_META_F_NATIVE) 752 fm = malloc(s->smd_size, M_DEVBUF, M_WAITOK | M_ZERO); 753 754 cp = (struct sr_meta_chunk *)(sm + 1); 755 SLIST_FOREACH(ch_entry, cl, src_link) { 756 /* skip disks that are offline */ 757 if (ch_entry->src_meta.scm_status == BIOC_SDOFFLINE) { 758 DNPRINTF(SR_D_META, 759 "%s: %s chunk marked offline, spoofing status\n", 760 DEVNAME(sc), ch_entry->src_devname); 761 cp++; /* adjust chunk pointer to match failure */ 762 continue; 763 } else if (s->smd_read(sd, ch_entry->src_dev_mm, sm, fm)) { 764 /* read and translate */ 765 /* XXX mark chunk offline, elsewhere!! */ 766 ch_entry->src_meta.scm_status = BIOC_SDOFFLINE; 767 cp++; /* adjust chunk pointer to match failure */ 768 DNPRINTF(SR_D_META, "%s: sr_meta_read failed\n", 769 DEVNAME(sc)); 770 continue; 771 } 772 773 if (sm->ssdi.ssd_magic != SR_MAGIC) { 774 DNPRINTF(SR_D_META, "%s: sr_meta_read !SR_MAGIC\n", 775 DEVNAME(sc)); 776 continue; 777 } 778 779 /* validate metadata */ 780 if (sr_meta_validate(sd, ch_entry->src_dev_mm, sm, fm)) { 781 DNPRINTF(SR_D_META, "%s: invalid metadata\n", 782 DEVNAME(sc)); 783 no_disk = -1; 784 goto done; 785 } 786 787 /* assume first chunk contains metadata */ 788 if (got_meta == 0) { 789 sr_meta_opt_load(sc, sm, &sd->sd_meta_opt); 790 bcopy(sm, sd->sd_meta, sizeof(*sd->sd_meta)); 791 got_meta = 1; 792 } 793 794 bcopy(cp, &ch_entry->src_meta, sizeof(ch_entry->src_meta)); 795 796 no_disk++; 797 cp++; 798 } 799 800 free(sm, M_DEVBUF, SR_META_SIZE * 512); 801 free(fm, M_DEVBUF, s->smd_size); 802 803 done: 804 DNPRINTF(SR_D_META, "%s: sr_meta_read found %d parts\n", DEVNAME(sc), 805 no_disk); 806 return (no_disk); 807 } 808 809 void 810 sr_meta_opt_load(struct sr_softc *sc, struct sr_metadata *sm, 811 struct sr_meta_opt_head *som) 812 { 813 struct sr_meta_opt_hdr *omh; 814 struct sr_meta_opt_item *omi; 815 u_int8_t checksum[MD5_DIGEST_LENGTH]; 816 int i; 817 818 /* Process optional metadata. */ 819 omh = (struct sr_meta_opt_hdr *)((u_int8_t *)(sm + 1) + 820 sizeof(struct sr_meta_chunk) * sm->ssdi.ssd_chunk_no); 821 for (i = 0; i < sm->ssdi.ssd_opt_no; i++) { 822 823 omi = malloc(sizeof(struct sr_meta_opt_item), M_DEVBUF, 824 M_WAITOK | M_ZERO); 825 SLIST_INSERT_HEAD(som, omi, omi_link); 826 827 if (omh->som_length == 0) { 828 829 /* Load old fixed length optional metadata. */ 830 DNPRINTF(SR_D_META, "%s: old optional metadata of type " 831 "%u\n", DEVNAME(sc), omh->som_type); 832 833 /* Validate checksum. */ 834 sr_checksum(sc, (void *)omh, &checksum, 835 SR_OLD_META_OPT_SIZE - MD5_DIGEST_LENGTH); 836 if (bcmp(&checksum, (void *)omh + SR_OLD_META_OPT_MD5, 837 sizeof(checksum))) 838 panic("%s: invalid optional metadata " 839 "checksum", DEVNAME(sc)); 840 841 /* Determine correct length. */ 842 switch (omh->som_type) { 843 case SR_OPT_CRYPTO: 844 omh->som_length = sizeof(struct sr_meta_crypto); 845 break; 846 case SR_OPT_BOOT: 847 omh->som_length = sizeof(struct sr_meta_boot); 848 break; 849 case SR_OPT_KEYDISK: 850 omh->som_length = 851 sizeof(struct sr_meta_keydisk); 852 break; 853 default: 854 panic("unknown old optional metadata " 855 "type %u\n", omh->som_type); 856 } 857 858 omi->omi_som = malloc(omh->som_length, M_DEVBUF, 859 M_WAITOK | M_ZERO); 860 bcopy((u_int8_t *)omh + SR_OLD_META_OPT_OFFSET, 861 (u_int8_t *)omi->omi_som + sizeof(*omi->omi_som), 862 omh->som_length - sizeof(*omi->omi_som)); 863 omi->omi_som->som_type = omh->som_type; 864 omi->omi_som->som_length = omh->som_length; 865 866 omh = (struct sr_meta_opt_hdr *)((void *)omh + 867 SR_OLD_META_OPT_SIZE); 868 } else { 869 870 /* Load variable length optional metadata. */ 871 DNPRINTF(SR_D_META, "%s: optional metadata of type %u, " 872 "length %u\n", DEVNAME(sc), omh->som_type, 873 omh->som_length); 874 omi->omi_som = malloc(omh->som_length, M_DEVBUF, 875 M_WAITOK | M_ZERO); 876 bcopy(omh, omi->omi_som, omh->som_length); 877 878 /* Validate checksum. */ 879 bcopy(&omi->omi_som->som_checksum, &checksum, 880 MD5_DIGEST_LENGTH); 881 bzero(&omi->omi_som->som_checksum, MD5_DIGEST_LENGTH); 882 sr_checksum(sc, omi->omi_som, 883 &omi->omi_som->som_checksum, omh->som_length); 884 if (bcmp(&checksum, &omi->omi_som->som_checksum, 885 sizeof(checksum))) 886 panic("%s: invalid optional metadata checksum", 887 DEVNAME(sc)); 888 889 omh = (struct sr_meta_opt_hdr *)((void *)omh + 890 omh->som_length); 891 } 892 } 893 } 894 895 int 896 sr_meta_validate(struct sr_discipline *sd, dev_t dev, struct sr_metadata *sm, 897 void *fm) 898 { 899 struct sr_softc *sc = sd->sd_sc; 900 struct sr_meta_driver *s; 901 #ifdef SR_DEBUG 902 struct sr_meta_chunk *mc; 903 #endif 904 u_int8_t checksum[MD5_DIGEST_LENGTH]; 905 char devname[32]; 906 int rv = 1; 907 908 DNPRINTF(SR_D_META, "%s: sr_meta_validate(%p)\n", DEVNAME(sc), sm); 909 910 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 911 912 s = &smd[sd->sd_meta_type]; 913 if (sd->sd_meta_type != SR_META_F_NATIVE) 914 if (s->smd_validate(sd, sm, fm)) { 915 sr_error(sc, "invalid foreign metadata"); 916 goto done; 917 } 918 919 /* 920 * at this point all foreign metadata has been translated to the native 921 * format and will be treated just like the native format 922 */ 923 924 if (sm->ssdi.ssd_magic != SR_MAGIC) { 925 sr_error(sc, "not valid softraid metadata"); 926 goto done; 927 } 928 929 /* Verify metadata checksum. */ 930 sr_checksum(sc, sm, &checksum, sizeof(struct sr_meta_invariant)); 931 if (bcmp(&checksum, &sm->ssd_checksum, sizeof(checksum))) { 932 sr_error(sc, "invalid metadata checksum"); 933 goto done; 934 } 935 936 /* Handle changes between versions. */ 937 if (sm->ssdi.ssd_version == 3) { 938 939 /* 940 * Version 3 - update metadata version and fix up data offset 941 * value since this did not exist in version 3. 942 */ 943 if (sm->ssd_data_offset == 0) 944 sm->ssd_data_offset = SR_META_V3_DATA_OFFSET; 945 946 } else if (sm->ssdi.ssd_version == 4) { 947 948 /* 949 * Version 4 - original metadata format did not store 950 * data offset so fix this up if necessary. 951 */ 952 if (sm->ssd_data_offset == 0) 953 sm->ssd_data_offset = SR_DATA_OFFSET; 954 955 } else if (sm->ssdi.ssd_version == SR_META_VERSION) { 956 957 /* 958 * Version 5 - variable length optional metadata. Migration 959 * from earlier fixed length optional metadata is handled 960 * in sr_meta_read(). 961 */ 962 963 } else { 964 965 sr_error(sc, "cannot read metadata version %u on %s, " 966 "expected version %u or earlier", 967 sm->ssdi.ssd_version, devname, SR_META_VERSION); 968 goto done; 969 970 } 971 972 /* Update version number and revision string. */ 973 sm->ssdi.ssd_version = SR_META_VERSION; 974 snprintf(sm->ssdi.ssd_revision, sizeof(sm->ssdi.ssd_revision), 975 "%03d", SR_META_VERSION); 976 977 #ifdef SR_DEBUG 978 /* warn if disk changed order */ 979 mc = (struct sr_meta_chunk *)(sm + 1); 980 if (strncmp(mc[sm->ssdi.ssd_chunk_id].scmi.scm_devname, devname, 981 sizeof(mc[sm->ssdi.ssd_chunk_id].scmi.scm_devname))) 982 DNPRINTF(SR_D_META, "%s: roaming device %s -> %s\n", 983 DEVNAME(sc), mc[sm->ssdi.ssd_chunk_id].scmi.scm_devname, 984 devname); 985 #endif 986 987 /* we have meta data on disk */ 988 DNPRINTF(SR_D_META, "%s: sr_meta_validate valid metadata %s\n", 989 DEVNAME(sc), devname); 990 991 rv = 0; 992 done: 993 return (rv); 994 } 995 996 int 997 sr_meta_native_bootprobe(struct sr_softc *sc, dev_t devno, 998 struct sr_boot_chunk_head *bch) 999 { 1000 struct vnode *vn; 1001 struct disklabel label; 1002 struct sr_metadata *md = NULL; 1003 struct sr_discipline *fake_sd = NULL; 1004 struct sr_boot_chunk *bc; 1005 char devname[32]; 1006 dev_t chrdev, rawdev; 1007 int error, i; 1008 int rv = SR_META_NOTCLAIMED; 1009 1010 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe\n", DEVNAME(sc)); 1011 1012 /* 1013 * Use character raw device to avoid SCSI complaints about missing 1014 * media on removable media devices. 1015 */ 1016 chrdev = blktochr(devno); 1017 rawdev = MAKEDISKDEV(major(chrdev), DISKUNIT(devno), RAW_PART); 1018 if (cdevvp(rawdev, &vn)) { 1019 sr_error(sc, "sr_meta_native_bootprobe: cannot allocate vnode"); 1020 goto done; 1021 } 1022 1023 /* open device */ 1024 error = VOP_OPEN(vn, FREAD, NOCRED, curproc); 1025 if (error) { 1026 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe open " 1027 "failed\n", DEVNAME(sc)); 1028 vput(vn); 1029 goto done; 1030 } 1031 1032 /* get disklabel */ 1033 error = VOP_IOCTL(vn, DIOCGDINFO, (caddr_t)&label, FREAD, NOCRED, 1034 curproc); 1035 if (error) { 1036 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe ioctl " 1037 "failed\n", DEVNAME(sc)); 1038 VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1039 vput(vn); 1040 goto done; 1041 } 1042 1043 /* we are done, close device */ 1044 error = VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1045 if (error) { 1046 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe close " 1047 "failed\n", DEVNAME(sc)); 1048 vput(vn); 1049 goto done; 1050 } 1051 vput(vn); 1052 1053 /* Make sure this is a 512-byte/sector device. */ 1054 if (label.d_secsize != DEV_BSIZE) { 1055 DNPRINTF(SR_D_META, "%s: %s has unsupported sector size (%d)", 1056 DEVNAME(sc), devname, label.d_secsize); 1057 goto done; 1058 } 1059 1060 md = malloc(SR_META_SIZE * 512, M_DEVBUF, M_ZERO | M_NOWAIT); 1061 if (md == NULL) { 1062 sr_error(sc, "not enough memory for metadata buffer"); 1063 goto done; 1064 } 1065 1066 /* create fake sd to use utility functions */ 1067 fake_sd = malloc(sizeof(struct sr_discipline), M_DEVBUF, 1068 M_ZERO | M_NOWAIT); 1069 if (fake_sd == NULL) { 1070 sr_error(sc, "not enough memory for fake discipline"); 1071 goto done; 1072 } 1073 fake_sd->sd_sc = sc; 1074 fake_sd->sd_meta_type = SR_META_F_NATIVE; 1075 1076 for (i = 0; i < MAXPARTITIONS; i++) { 1077 if (label.d_partitions[i].p_fstype != FS_RAID) 1078 continue; 1079 1080 /* open partition */ 1081 rawdev = MAKEDISKDEV(major(devno), DISKUNIT(devno), i); 1082 if (bdevvp(rawdev, &vn)) { 1083 sr_error(sc, "sr_meta_native_bootprobe: cannot " 1084 "allocate vnode for partition"); 1085 goto done; 1086 } 1087 error = VOP_OPEN(vn, FREAD, NOCRED, curproc); 1088 if (error) { 1089 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe " 1090 "open failed, partition %d\n", 1091 DEVNAME(sc), i); 1092 vput(vn); 1093 continue; 1094 } 1095 1096 if (sr_meta_native_read(fake_sd, rawdev, md, NULL)) { 1097 sr_error(sc, "native bootprobe could not read native " 1098 "metadata"); 1099 VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1100 vput(vn); 1101 continue; 1102 } 1103 1104 /* are we a softraid partition? */ 1105 if (md->ssdi.ssd_magic != SR_MAGIC) { 1106 VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1107 vput(vn); 1108 continue; 1109 } 1110 1111 sr_meta_getdevname(sc, rawdev, devname, sizeof(devname)); 1112 if (sr_meta_validate(fake_sd, rawdev, md, NULL) == 0) { 1113 /* XXX fix M_WAITOK, this is boot time */ 1114 bc = malloc(sizeof(struct sr_boot_chunk), 1115 M_DEVBUF, M_WAITOK | M_ZERO); 1116 bc->sbc_metadata = malloc(sizeof(struct sr_metadata), 1117 M_DEVBUF, M_WAITOK | M_ZERO); 1118 bcopy(md, bc->sbc_metadata, sizeof(struct sr_metadata)); 1119 bc->sbc_mm = rawdev; 1120 SLIST_INSERT_HEAD(bch, bc, sbc_link); 1121 rv = SR_META_CLAIMED; 1122 } 1123 1124 /* we are done, close partition */ 1125 VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1126 vput(vn); 1127 } 1128 1129 done: 1130 free(fake_sd, M_DEVBUF, sizeof(struct sr_discipline)); 1131 free(md, M_DEVBUF, SR_META_SIZE * 512); 1132 1133 return (rv); 1134 } 1135 1136 int 1137 sr_boot_assembly(struct sr_softc *sc) 1138 { 1139 struct sr_boot_volume_head bvh; 1140 struct sr_boot_chunk_head bch, kdh; 1141 struct sr_boot_volume *bv, *bv1, *bv2; 1142 struct sr_boot_chunk *bc, *bcnext, *bc1, *bc2; 1143 struct sr_disk_head sdklist; 1144 struct sr_disk *sdk; 1145 struct disk *dk; 1146 struct bioc_createraid bcr; 1147 struct sr_meta_chunk *hm; 1148 struct sr_chunk_head *cl; 1149 struct sr_chunk *hotspare, *chunk, *last; 1150 u_int64_t *ondisk = NULL; 1151 dev_t *devs = NULL; 1152 void *data; 1153 char devname[32]; 1154 int rv = 0, i; 1155 1156 DNPRINTF(SR_D_META, "%s: sr_boot_assembly\n", DEVNAME(sc)); 1157 1158 SLIST_INIT(&sdklist); 1159 SLIST_INIT(&bvh); 1160 SLIST_INIT(&bch); 1161 SLIST_INIT(&kdh); 1162 1163 dk = TAILQ_FIRST(&disklist); 1164 while (dk != NULL) { 1165 1166 /* See if this disk has been checked. */ 1167 SLIST_FOREACH(sdk, &sdklist, sdk_link) 1168 if (sdk->sdk_devno == dk->dk_devno) 1169 break; 1170 1171 if (sdk != NULL || dk->dk_devno == NODEV) { 1172 dk = TAILQ_NEXT(dk, dk_link); 1173 continue; 1174 } 1175 1176 /* Add this disk to the list that we've checked. */ 1177 sdk = malloc(sizeof(struct sr_disk), M_DEVBUF, 1178 M_NOWAIT | M_CANFAIL | M_ZERO); 1179 if (sdk == NULL) 1180 goto unwind; 1181 sdk->sdk_devno = dk->dk_devno; 1182 SLIST_INSERT_HEAD(&sdklist, sdk, sdk_link); 1183 1184 /* Only check sd(4) and wd(4) devices. */ 1185 if (strncmp(dk->dk_name, "sd", 2) && 1186 strncmp(dk->dk_name, "wd", 2)) { 1187 dk = TAILQ_NEXT(dk, dk_link); 1188 continue; 1189 } 1190 1191 /* native softraid uses partitions */ 1192 rw_enter_write(&sc->sc_lock); 1193 bio_status_init(&sc->sc_status, &sc->sc_dev); 1194 sr_meta_native_bootprobe(sc, dk->dk_devno, &bch); 1195 rw_exit_write(&sc->sc_lock); 1196 1197 /* probe non-native disks if native failed. */ 1198 1199 /* Restart scan since we may have slept. */ 1200 dk = TAILQ_FIRST(&disklist); 1201 } 1202 1203 /* 1204 * Create a list of volumes and associate chunks with each volume. 1205 */ 1206 for (bc = SLIST_FIRST(&bch); bc != NULL; bc = bcnext) { 1207 1208 bcnext = SLIST_NEXT(bc, sbc_link); 1209 SLIST_REMOVE(&bch, bc, sr_boot_chunk, sbc_link); 1210 bc->sbc_chunk_id = bc->sbc_metadata->ssdi.ssd_chunk_id; 1211 1212 /* Handle key disks separately. */ 1213 if (bc->sbc_metadata->ssdi.ssd_level == SR_KEYDISK_LEVEL) { 1214 SLIST_INSERT_HEAD(&kdh, bc, sbc_link); 1215 continue; 1216 } 1217 1218 SLIST_FOREACH(bv, &bvh, sbv_link) { 1219 if (bcmp(&bc->sbc_metadata->ssdi.ssd_uuid, 1220 &bv->sbv_uuid, 1221 sizeof(bc->sbc_metadata->ssdi.ssd_uuid)) == 0) 1222 break; 1223 } 1224 1225 if (bv == NULL) { 1226 bv = malloc(sizeof(struct sr_boot_volume), 1227 M_DEVBUF, M_NOWAIT | M_CANFAIL | M_ZERO); 1228 if (bv == NULL) { 1229 printf("%s: failed to allocate boot volume\n", 1230 DEVNAME(sc)); 1231 goto unwind; 1232 } 1233 1234 bv->sbv_level = bc->sbc_metadata->ssdi.ssd_level; 1235 bv->sbv_volid = bc->sbc_metadata->ssdi.ssd_volid; 1236 bv->sbv_chunk_no = bc->sbc_metadata->ssdi.ssd_chunk_no; 1237 bv->sbv_flags = bc->sbc_metadata->ssdi.ssd_vol_flags; 1238 bcopy(&bc->sbc_metadata->ssdi.ssd_uuid, &bv->sbv_uuid, 1239 sizeof(bc->sbc_metadata->ssdi.ssd_uuid)); 1240 SLIST_INIT(&bv->sbv_chunks); 1241 1242 /* Maintain volume order. */ 1243 bv2 = NULL; 1244 SLIST_FOREACH(bv1, &bvh, sbv_link) { 1245 if (bv1->sbv_volid > bv->sbv_volid) 1246 break; 1247 bv2 = bv1; 1248 } 1249 if (bv2 == NULL) { 1250 DNPRINTF(SR_D_META, "%s: insert volume %u " 1251 "at head\n", DEVNAME(sc), bv->sbv_volid); 1252 SLIST_INSERT_HEAD(&bvh, bv, sbv_link); 1253 } else { 1254 DNPRINTF(SR_D_META, "%s: insert volume %u " 1255 "after %u\n", DEVNAME(sc), bv->sbv_volid, 1256 bv2->sbv_volid); 1257 SLIST_INSERT_AFTER(bv2, bv, sbv_link); 1258 } 1259 } 1260 1261 /* Maintain chunk order. */ 1262 bc2 = NULL; 1263 SLIST_FOREACH(bc1, &bv->sbv_chunks, sbc_link) { 1264 if (bc1->sbc_chunk_id > bc->sbc_chunk_id) 1265 break; 1266 bc2 = bc1; 1267 } 1268 if (bc2 == NULL) { 1269 DNPRINTF(SR_D_META, "%s: volume %u insert chunk %u " 1270 "at head\n", DEVNAME(sc), bv->sbv_volid, 1271 bc->sbc_chunk_id); 1272 SLIST_INSERT_HEAD(&bv->sbv_chunks, bc, sbc_link); 1273 } else { 1274 DNPRINTF(SR_D_META, "%s: volume %u insert chunk %u " 1275 "after %u\n", DEVNAME(sc), bv->sbv_volid, 1276 bc->sbc_chunk_id, bc2->sbc_chunk_id); 1277 SLIST_INSERT_AFTER(bc2, bc, sbc_link); 1278 } 1279 1280 bv->sbv_chunks_found++; 1281 } 1282 1283 /* Allocate memory for device and ondisk version arrays. */ 1284 devs = mallocarray(BIOC_CRMAXLEN, sizeof(dev_t), M_DEVBUF, 1285 M_NOWAIT | M_CANFAIL); 1286 if (devs == NULL) { 1287 printf("%s: failed to allocate device array\n", DEVNAME(sc)); 1288 goto unwind; 1289 } 1290 ondisk = mallocarray(BIOC_CRMAXLEN, sizeof(u_int64_t), M_DEVBUF, 1291 M_NOWAIT | M_CANFAIL); 1292 if (ondisk == NULL) { 1293 printf("%s: failed to allocate ondisk array\n", DEVNAME(sc)); 1294 goto unwind; 1295 } 1296 1297 /* 1298 * Assemble hotspare "volumes". 1299 */ 1300 SLIST_FOREACH(bv, &bvh, sbv_link) { 1301 1302 /* Check if this is a hotspare "volume". */ 1303 if (bv->sbv_level != SR_HOTSPARE_LEVEL || 1304 bv->sbv_chunk_no != 1) 1305 continue; 1306 1307 #ifdef SR_DEBUG 1308 DNPRINTF(SR_D_META, "%s: assembling hotspare volume ", 1309 DEVNAME(sc)); 1310 if (sr_debug & SR_D_META) 1311 sr_uuid_print(&bv->sbv_uuid, 0); 1312 DNPRINTF(SR_D_META, " volid %u with %u chunks\n", 1313 bv->sbv_volid, bv->sbv_chunk_no); 1314 #endif 1315 1316 /* Create hotspare chunk metadata. */ 1317 hotspare = malloc(sizeof(struct sr_chunk), M_DEVBUF, 1318 M_NOWAIT | M_CANFAIL | M_ZERO); 1319 if (hotspare == NULL) { 1320 printf("%s: failed to allocate hotspare\n", 1321 DEVNAME(sc)); 1322 goto unwind; 1323 } 1324 1325 bc = SLIST_FIRST(&bv->sbv_chunks); 1326 sr_meta_getdevname(sc, bc->sbc_mm, devname, sizeof(devname)); 1327 hotspare->src_dev_mm = bc->sbc_mm; 1328 strlcpy(hotspare->src_devname, devname, 1329 sizeof(hotspare->src_devname)); 1330 hotspare->src_size = bc->sbc_metadata->ssdi.ssd_size; 1331 1332 hm = &hotspare->src_meta; 1333 hm->scmi.scm_volid = SR_HOTSPARE_VOLID; 1334 hm->scmi.scm_chunk_id = 0; 1335 hm->scmi.scm_size = bc->sbc_metadata->ssdi.ssd_size; 1336 hm->scmi.scm_coerced_size = bc->sbc_metadata->ssdi.ssd_size; 1337 strlcpy(hm->scmi.scm_devname, devname, 1338 sizeof(hm->scmi.scm_devname)); 1339 bcopy(&bc->sbc_metadata->ssdi.ssd_uuid, &hm->scmi.scm_uuid, 1340 sizeof(struct sr_uuid)); 1341 1342 sr_checksum(sc, hm, &hm->scm_checksum, 1343 sizeof(struct sr_meta_chunk_invariant)); 1344 1345 hm->scm_status = BIOC_SDHOTSPARE; 1346 1347 /* Add chunk to hotspare list. */ 1348 rw_enter_write(&sc->sc_hs_lock); 1349 cl = &sc->sc_hotspare_list; 1350 if (SLIST_EMPTY(cl)) 1351 SLIST_INSERT_HEAD(cl, hotspare, src_link); 1352 else { 1353 SLIST_FOREACH(chunk, cl, src_link) 1354 last = chunk; 1355 SLIST_INSERT_AFTER(last, hotspare, src_link); 1356 } 1357 sc->sc_hotspare_no++; 1358 rw_exit_write(&sc->sc_hs_lock); 1359 1360 } 1361 1362 /* 1363 * Assemble RAID volumes. 1364 */ 1365 SLIST_FOREACH(bv, &bvh, sbv_link) { 1366 1367 bzero(&bcr, sizeof(bcr)); 1368 data = NULL; 1369 1370 /* Check if this is a hotspare "volume". */ 1371 if (bv->sbv_level == SR_HOTSPARE_LEVEL && 1372 bv->sbv_chunk_no == 1) 1373 continue; 1374 1375 /* 1376 * Skip volumes that are marked as no auto assemble, unless 1377 * this was the volume which we actually booted from. 1378 */ 1379 if (bcmp(&sr_bootuuid, &bv->sbv_uuid, sizeof(sr_bootuuid)) != 0) 1380 if (bv->sbv_flags & BIOC_SCNOAUTOASSEMBLE) 1381 continue; 1382 1383 #ifdef SR_DEBUG 1384 DNPRINTF(SR_D_META, "%s: assembling volume ", DEVNAME(sc)); 1385 if (sr_debug & SR_D_META) 1386 sr_uuid_print(&bv->sbv_uuid, 0); 1387 DNPRINTF(SR_D_META, " volid %u with %u chunks\n", 1388 bv->sbv_volid, bv->sbv_chunk_no); 1389 #endif 1390 1391 /* 1392 * If this is a crypto volume, try to find a matching 1393 * key disk... 1394 */ 1395 bcr.bc_key_disk = NODEV; 1396 if (bv->sbv_level == 'C') { 1397 SLIST_FOREACH(bc, &kdh, sbc_link) { 1398 if (bcmp(&bc->sbc_metadata->ssdi.ssd_uuid, 1399 &bv->sbv_uuid, 1400 sizeof(bc->sbc_metadata->ssdi.ssd_uuid)) 1401 == 0) 1402 bcr.bc_key_disk = bc->sbc_mm; 1403 } 1404 } 1405 1406 for (i = 0; i < BIOC_CRMAXLEN; i++) { 1407 devs[i] = NODEV; /* mark device as illegal */ 1408 ondisk[i] = 0; 1409 } 1410 1411 SLIST_FOREACH(bc, &bv->sbv_chunks, sbc_link) { 1412 if (devs[bc->sbc_chunk_id] != NODEV) { 1413 bv->sbv_chunks_found--; 1414 sr_meta_getdevname(sc, bc->sbc_mm, devname, 1415 sizeof(devname)); 1416 printf("%s: found duplicate chunk %u for " 1417 "volume %u on device %s\n", DEVNAME(sc), 1418 bc->sbc_chunk_id, bv->sbv_volid, devname); 1419 } 1420 1421 if (devs[bc->sbc_chunk_id] == NODEV || 1422 bc->sbc_metadata->ssd_ondisk > 1423 ondisk[bc->sbc_chunk_id]) { 1424 devs[bc->sbc_chunk_id] = bc->sbc_mm; 1425 ondisk[bc->sbc_chunk_id] = 1426 bc->sbc_metadata->ssd_ondisk; 1427 DNPRINTF(SR_D_META, "%s: using ondisk " 1428 "metadata version %llu for chunk %u\n", 1429 DEVNAME(sc), ondisk[bc->sbc_chunk_id], 1430 bc->sbc_chunk_id); 1431 } 1432 } 1433 1434 if (bv->sbv_chunk_no != bv->sbv_chunks_found) { 1435 printf("%s: not all chunks were provided; " 1436 "attempting to bring volume %d online\n", 1437 DEVNAME(sc), bv->sbv_volid); 1438 } 1439 1440 bcr.bc_level = bv->sbv_level; 1441 bcr.bc_dev_list_len = bv->sbv_chunk_no * sizeof(dev_t); 1442 bcr.bc_dev_list = devs; 1443 bcr.bc_flags = BIOC_SCDEVT | 1444 (bv->sbv_flags & BIOC_SCNOAUTOASSEMBLE); 1445 1446 if (bv->sbv_level == 'C' && 1447 bcmp(&sr_bootuuid, &bv->sbv_uuid, sizeof(sr_bootuuid)) == 0) 1448 data = sr_bootkey; 1449 1450 rw_enter_write(&sc->sc_lock); 1451 bio_status_init(&sc->sc_status, &sc->sc_dev); 1452 sr_ioctl_createraid(sc, &bcr, 0, data); 1453 rw_exit_write(&sc->sc_lock); 1454 1455 rv++; 1456 } 1457 1458 /* done with metadata */ 1459 unwind: 1460 /* Free boot volumes and associated chunks. */ 1461 for (bv1 = SLIST_FIRST(&bvh); bv1 != NULL; bv1 = bv2) { 1462 bv2 = SLIST_NEXT(bv1, sbv_link); 1463 for (bc1 = SLIST_FIRST(&bv1->sbv_chunks); bc1 != NULL; 1464 bc1 = bc2) { 1465 bc2 = SLIST_NEXT(bc1, sbc_link); 1466 if (bc1->sbc_metadata) 1467 free(bc1->sbc_metadata, M_DEVBUF, 0); 1468 free(bc1, M_DEVBUF, 0); 1469 } 1470 free(bv1, M_DEVBUF, 0); 1471 } 1472 /* Free keydisks chunks. */ 1473 for (bc1 = SLIST_FIRST(&kdh); bc1 != NULL; bc1 = bc2) { 1474 bc2 = SLIST_NEXT(bc1, sbc_link); 1475 if (bc1->sbc_metadata) 1476 free(bc1->sbc_metadata, M_DEVBUF, 0); 1477 free(bc1, M_DEVBUF, 0); 1478 } 1479 /* Free unallocated chunks. */ 1480 for (bc1 = SLIST_FIRST(&bch); bc1 != NULL; bc1 = bc2) { 1481 bc2 = SLIST_NEXT(bc1, sbc_link); 1482 if (bc1->sbc_metadata) 1483 free(bc1->sbc_metadata, M_DEVBUF, 0); 1484 free(bc1, M_DEVBUF, 0); 1485 } 1486 1487 while (!SLIST_EMPTY(&sdklist)) { 1488 sdk = SLIST_FIRST(&sdklist); 1489 SLIST_REMOVE_HEAD(&sdklist, sdk_link); 1490 free(sdk, M_DEVBUF, 0); 1491 } 1492 1493 free(devs, M_DEVBUF, BIOC_CRMAXLEN * sizeof(dev_t)); 1494 free(ondisk, M_DEVBUF, BIOC_CRMAXLEN * sizeof(u_int64_t)); 1495 1496 return (rv); 1497 } 1498 1499 void 1500 sr_map_root(void) 1501 { 1502 struct sr_softc *sc = softraid0; 1503 struct sr_discipline *sd; 1504 struct sr_meta_opt_item *omi; 1505 struct sr_meta_boot *sbm; 1506 u_char duid[8]; 1507 int i; 1508 1509 DNPRINTF(SR_D_MISC, "%s: sr_map_root\n", DEVNAME(sc)); 1510 1511 if (sc == NULL) 1512 return; 1513 1514 bzero(duid, sizeof(duid)); 1515 if (bcmp(rootduid, duid, sizeof(duid)) == 0) { 1516 DNPRINTF(SR_D_MISC, "%s: root duid is zero\n", DEVNAME(sc)); 1517 return; 1518 } 1519 1520 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 1521 SLIST_FOREACH(omi, &sd->sd_meta_opt, omi_link) { 1522 if (omi->omi_som->som_type != SR_OPT_BOOT) 1523 continue; 1524 sbm = (struct sr_meta_boot *)omi->omi_som; 1525 for (i = 0; i < SR_MAX_BOOT_DISKS; i++) { 1526 if (bcmp(rootduid, sbm->sbm_boot_duid[i], 1527 sizeof(rootduid)) == 0) { 1528 bcopy(sbm->sbm_root_duid, rootduid, 1529 sizeof(rootduid)); 1530 DNPRINTF(SR_D_MISC, "%s: root duid " 1531 "mapped to %02hx%02hx%02hx%02hx" 1532 "%02hx%02hx%02hx%02hx\n", 1533 DEVNAME(sc), rootduid[0], 1534 rootduid[1], rootduid[2], 1535 rootduid[3], rootduid[4], 1536 rootduid[5], rootduid[6], 1537 rootduid[7]); 1538 return; 1539 } 1540 } 1541 } 1542 } 1543 } 1544 1545 int 1546 sr_meta_native_probe(struct sr_softc *sc, struct sr_chunk *ch_entry) 1547 { 1548 struct disklabel label; 1549 char *devname; 1550 int error, part; 1551 daddr_t size; 1552 1553 DNPRINTF(SR_D_META, "%s: sr_meta_native_probe(%s)\n", 1554 DEVNAME(sc), ch_entry->src_devname); 1555 1556 devname = ch_entry->src_devname; 1557 part = DISKPART(ch_entry->src_dev_mm); 1558 1559 /* get disklabel */ 1560 error = VOP_IOCTL(ch_entry->src_vn, DIOCGDINFO, (caddr_t)&label, FREAD, 1561 NOCRED, curproc); 1562 if (error) { 1563 DNPRINTF(SR_D_META, "%s: %s can't obtain disklabel\n", 1564 DEVNAME(sc), devname); 1565 goto unwind; 1566 } 1567 bcopy(label.d_uid, ch_entry->src_duid, sizeof(ch_entry->src_duid)); 1568 1569 /* Make sure this is a 512-byte/sector device. */ 1570 if (label.d_secsize != DEV_BSIZE) { 1571 sr_error(sc, "%s has unsupported sector size (%u)", 1572 devname, label.d_secsize); 1573 goto unwind; 1574 } 1575 1576 /* make sure the partition is of the right type */ 1577 if (label.d_partitions[part].p_fstype != FS_RAID) { 1578 DNPRINTF(SR_D_META, 1579 "%s: %s partition not of type RAID (%d)\n", DEVNAME(sc), 1580 devname, 1581 label.d_partitions[part].p_fstype); 1582 goto unwind; 1583 } 1584 1585 size = DL_SECTOBLK(&label, DL_GETPSIZE(&label.d_partitions[part])) - 1586 SR_DATA_OFFSET; 1587 if (size <= 0) { 1588 DNPRINTF(SR_D_META, "%s: %s partition too small\n", DEVNAME(sc), 1589 devname); 1590 goto unwind; 1591 } 1592 ch_entry->src_size = size; 1593 1594 DNPRINTF(SR_D_META, "%s: probe found %s size %lld\n", DEVNAME(sc), 1595 devname, (long long)size); 1596 1597 return (SR_META_F_NATIVE); 1598 unwind: 1599 DNPRINTF(SR_D_META, "%s: invalid device: %s\n", DEVNAME(sc), 1600 devname ? devname : "nodev"); 1601 return (SR_META_F_INVALID); 1602 } 1603 1604 int 1605 sr_meta_native_attach(struct sr_discipline *sd, int force) 1606 { 1607 struct sr_softc *sc = sd->sd_sc; 1608 struct sr_chunk_head *cl = &sd->sd_vol.sv_chunk_list; 1609 struct sr_metadata *md = NULL; 1610 struct sr_chunk *ch_entry, *ch_next; 1611 struct sr_uuid uuid; 1612 u_int64_t version = 0; 1613 int sr, not_sr, rv = 1, d, expected = -1, old_meta = 0; 1614 1615 DNPRINTF(SR_D_META, "%s: sr_meta_native_attach\n", DEVNAME(sc)); 1616 1617 md = malloc(SR_META_SIZE * 512, M_DEVBUF, M_ZERO | M_NOWAIT); 1618 if (md == NULL) { 1619 sr_error(sc, "not enough memory for metadata buffer"); 1620 goto bad; 1621 } 1622 1623 bzero(&uuid, sizeof uuid); 1624 1625 sr = not_sr = d = 0; 1626 SLIST_FOREACH(ch_entry, cl, src_link) { 1627 if (ch_entry->src_dev_mm == NODEV) 1628 continue; 1629 1630 if (sr_meta_native_read(sd, ch_entry->src_dev_mm, md, NULL)) { 1631 sr_error(sc, "could not read native metadata"); 1632 goto bad; 1633 } 1634 1635 if (md->ssdi.ssd_magic == SR_MAGIC) { 1636 sr++; 1637 ch_entry->src_meta.scmi.scm_chunk_id = 1638 md->ssdi.ssd_chunk_id; 1639 if (d == 0) { 1640 bcopy(&md->ssdi.ssd_uuid, &uuid, sizeof uuid); 1641 expected = md->ssdi.ssd_chunk_no; 1642 version = md->ssd_ondisk; 1643 d++; 1644 continue; 1645 } else if (bcmp(&md->ssdi.ssd_uuid, &uuid, 1646 sizeof uuid)) { 1647 sr_error(sc, "not part of the same volume"); 1648 goto bad; 1649 } 1650 if (md->ssd_ondisk != version) { 1651 old_meta++; 1652 version = MAX(md->ssd_ondisk, version); 1653 } 1654 } else 1655 not_sr++; 1656 } 1657 1658 if (sr && not_sr) { 1659 sr_error(sc, "not all chunks are of the native metadata " 1660 "format"); 1661 goto bad; 1662 } 1663 1664 /* mixed metadata versions; mark bad disks offline */ 1665 if (old_meta) { 1666 d = 0; 1667 for (ch_entry = SLIST_FIRST(cl); ch_entry != NULL; 1668 ch_entry = ch_next, d++) { 1669 ch_next = SLIST_NEXT(ch_entry, src_link); 1670 1671 /* XXX do we want to read this again? */ 1672 if (ch_entry->src_dev_mm == NODEV) 1673 panic("src_dev_mm == NODEV"); 1674 if (sr_meta_native_read(sd, ch_entry->src_dev_mm, md, 1675 NULL)) 1676 sr_warn(sc, "could not read native metadata"); 1677 if (md->ssd_ondisk != version) 1678 sd->sd_vol.sv_chunks[d]->src_meta.scm_status = 1679 BIOC_SDOFFLINE; 1680 } 1681 } 1682 1683 if (expected != sr && !force && expected != -1) { 1684 DNPRINTF(SR_D_META, "%s: not all chunks were provided, trying " 1685 "anyway\n", DEVNAME(sc)); 1686 } 1687 1688 rv = 0; 1689 bad: 1690 free(md, M_DEVBUF, SR_META_SIZE * 512); 1691 return (rv); 1692 } 1693 1694 int 1695 sr_meta_native_read(struct sr_discipline *sd, dev_t dev, 1696 struct sr_metadata *md, void *fm) 1697 { 1698 #ifdef SR_DEBUG 1699 struct sr_softc *sc = sd->sd_sc; 1700 #endif 1701 DNPRINTF(SR_D_META, "%s: sr_meta_native_read(0x%x, %p)\n", 1702 DEVNAME(sc), dev, md); 1703 1704 return (sr_meta_rw(sd, dev, md, SR_META_SIZE * 512, SR_META_OFFSET, 1705 B_READ)); 1706 } 1707 1708 int 1709 sr_meta_native_write(struct sr_discipline *sd, dev_t dev, 1710 struct sr_metadata *md, void *fm) 1711 { 1712 #ifdef SR_DEBUG 1713 struct sr_softc *sc = sd->sd_sc; 1714 #endif 1715 DNPRINTF(SR_D_META, "%s: sr_meta_native_write(0x%x, %p)\n", 1716 DEVNAME(sc), dev, md); 1717 1718 return (sr_meta_rw(sd, dev, md, SR_META_SIZE * 512, SR_META_OFFSET, 1719 B_WRITE)); 1720 } 1721 1722 void 1723 sr_hotplug_register(struct sr_discipline *sd, void *func) 1724 { 1725 struct sr_hotplug_list *mhe; 1726 1727 DNPRINTF(SR_D_MISC, "%s: sr_hotplug_register: %p\n", 1728 DEVNAME(sd->sd_sc), func); 1729 1730 /* make sure we aren't on the list yet */ 1731 SLIST_FOREACH(mhe, &sr_hotplug_callbacks, shl_link) 1732 if (mhe->sh_hotplug == func) 1733 return; 1734 1735 mhe = malloc(sizeof(struct sr_hotplug_list), M_DEVBUF, 1736 M_WAITOK | M_ZERO); 1737 mhe->sh_hotplug = func; 1738 mhe->sh_sd = sd; 1739 SLIST_INSERT_HEAD(&sr_hotplug_callbacks, mhe, shl_link); 1740 } 1741 1742 void 1743 sr_hotplug_unregister(struct sr_discipline *sd, void *func) 1744 { 1745 struct sr_hotplug_list *mhe; 1746 1747 DNPRINTF(SR_D_MISC, "%s: sr_hotplug_unregister: %s %p\n", 1748 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, func); 1749 1750 /* make sure we are on the list yet */ 1751 SLIST_FOREACH(mhe, &sr_hotplug_callbacks, shl_link) 1752 if (mhe->sh_hotplug == func) { 1753 SLIST_REMOVE(&sr_hotplug_callbacks, mhe, 1754 sr_hotplug_list, shl_link); 1755 free(mhe, M_DEVBUF, 0); 1756 if (SLIST_EMPTY(&sr_hotplug_callbacks)) 1757 SLIST_INIT(&sr_hotplug_callbacks); 1758 return; 1759 } 1760 } 1761 1762 void 1763 sr_disk_attach(struct disk *diskp, int action) 1764 { 1765 struct sr_hotplug_list *mhe; 1766 1767 SLIST_FOREACH(mhe, &sr_hotplug_callbacks, shl_link) 1768 if (mhe->sh_sd->sd_ready) 1769 mhe->sh_hotplug(mhe->sh_sd, diskp, action); 1770 } 1771 1772 int 1773 sr_match(struct device *parent, void *match, void *aux) 1774 { 1775 return (1); 1776 } 1777 1778 void 1779 sr_attach(struct device *parent, struct device *self, void *aux) 1780 { 1781 struct sr_softc *sc = (void *)self; 1782 struct scsibus_attach_args saa; 1783 1784 DNPRINTF(SR_D_MISC, "\n%s: sr_attach", DEVNAME(sc)); 1785 1786 if (softraid0 == NULL) 1787 softraid0 = sc; 1788 1789 rw_init(&sc->sc_lock, "sr_lock"); 1790 rw_init(&sc->sc_hs_lock, "sr_hs_lock"); 1791 1792 SLIST_INIT(&sr_hotplug_callbacks); 1793 TAILQ_INIT(&sc->sc_dis_list); 1794 SLIST_INIT(&sc->sc_hotspare_list); 1795 1796 #if NBIO > 0 1797 if (bio_register(&sc->sc_dev, sr_bio_ioctl) != 0) 1798 printf("%s: controller registration failed", DEVNAME(sc)); 1799 #endif /* NBIO > 0 */ 1800 1801 #ifndef SMALL_KERNEL 1802 strlcpy(sc->sc_sensordev.xname, DEVNAME(sc), 1803 sizeof(sc->sc_sensordev.xname)); 1804 sensordev_install(&sc->sc_sensordev); 1805 #endif /* SMALL_KERNEL */ 1806 1807 printf("\n"); 1808 1809 sc->sc_link.adapter_softc = sc; 1810 sc->sc_link.adapter = &sr_switch; 1811 sc->sc_link.adapter_target = SR_MAX_LD; 1812 sc->sc_link.adapter_buswidth = SR_MAX_LD; 1813 sc->sc_link.luns = 1; 1814 1815 bzero(&saa, sizeof(saa)); 1816 saa.saa_sc_link = &sc->sc_link; 1817 1818 sc->sc_scsibus = (struct scsibus_softc *)config_found(&sc->sc_dev, 1819 &saa, scsiprint); 1820 1821 softraid_disk_attach = sr_disk_attach; 1822 1823 sr_boot_assembly(sc); 1824 1825 explicit_bzero(sr_bootkey, sizeof(sr_bootkey)); 1826 } 1827 1828 int 1829 sr_detach(struct device *self, int flags) 1830 { 1831 struct sr_softc *sc = (void *)self; 1832 int rv; 1833 1834 DNPRINTF(SR_D_MISC, "%s: sr_detach\n", DEVNAME(sc)); 1835 1836 softraid_disk_attach = NULL; 1837 1838 sr_shutdown(); 1839 1840 #ifndef SMALL_KERNEL 1841 if (sc->sc_sensor_task != NULL) 1842 sensor_task_unregister(sc->sc_sensor_task); 1843 sensordev_deinstall(&sc->sc_sensordev); 1844 #endif /* SMALL_KERNEL */ 1845 1846 if (sc->sc_scsibus != NULL) { 1847 rv = config_detach((struct device *)sc->sc_scsibus, flags); 1848 if (rv != 0) 1849 return (rv); 1850 sc->sc_scsibus = NULL; 1851 } 1852 1853 return (0); 1854 } 1855 1856 void 1857 sr_info(struct sr_softc *sc, const char *fmt, ...) 1858 { 1859 va_list ap; 1860 1861 rw_assert_wrlock(&sc->sc_lock); 1862 1863 va_start(ap, fmt); 1864 bio_status(&sc->sc_status, 0, BIO_MSG_INFO, fmt, &ap); 1865 va_end(ap); 1866 } 1867 1868 void 1869 sr_warn(struct sr_softc *sc, const char *fmt, ...) 1870 { 1871 va_list ap; 1872 1873 rw_assert_wrlock(&sc->sc_lock); 1874 1875 va_start(ap, fmt); 1876 bio_status(&sc->sc_status, 1, BIO_MSG_WARN, fmt, &ap); 1877 va_end(ap); 1878 } 1879 1880 void 1881 sr_error(struct sr_softc *sc, const char *fmt, ...) 1882 { 1883 va_list ap; 1884 1885 rw_assert_wrlock(&sc->sc_lock); 1886 1887 va_start(ap, fmt); 1888 bio_status(&sc->sc_status, 1, BIO_MSG_ERROR, fmt, &ap); 1889 va_end(ap); 1890 } 1891 1892 void 1893 sr_minphys(struct buf *bp, struct scsi_link *sl) 1894 { 1895 DNPRINTF(SR_D_MISC, "sr_minphys: %d\n", bp->b_bcount); 1896 1897 /* XXX currently using SR_MAXFER = MAXPHYS */ 1898 if (bp->b_bcount > SR_MAXFER) 1899 bp->b_bcount = SR_MAXFER; 1900 minphys(bp); 1901 } 1902 1903 void 1904 sr_copy_internal_data(struct scsi_xfer *xs, void *v, size_t size) 1905 { 1906 size_t copy_cnt; 1907 1908 DNPRINTF(SR_D_MISC, "sr_copy_internal_data xs: %p size: %zu\n", 1909 xs, size); 1910 1911 if (xs->datalen) { 1912 copy_cnt = MIN(size, xs->datalen); 1913 bcopy(v, xs->data, copy_cnt); 1914 } 1915 } 1916 1917 int 1918 sr_ccb_alloc(struct sr_discipline *sd) 1919 { 1920 struct sr_ccb *ccb; 1921 int i; 1922 1923 if (!sd) 1924 return (1); 1925 1926 DNPRINTF(SR_D_CCB, "%s: sr_ccb_alloc\n", DEVNAME(sd->sd_sc)); 1927 1928 if (sd->sd_ccb) 1929 return (1); 1930 1931 sd->sd_ccb = mallocarray(sd->sd_max_wu, 1932 sd->sd_max_ccb_per_wu * sizeof(struct sr_ccb), 1933 M_DEVBUF, M_WAITOK | M_ZERO); 1934 TAILQ_INIT(&sd->sd_ccb_freeq); 1935 for (i = 0; i < sd->sd_max_wu * sd->sd_max_ccb_per_wu; i++) { 1936 ccb = &sd->sd_ccb[i]; 1937 ccb->ccb_dis = sd; 1938 sr_ccb_put(ccb); 1939 } 1940 1941 DNPRINTF(SR_D_CCB, "%s: sr_ccb_alloc ccb: %d\n", 1942 DEVNAME(sd->sd_sc), sd->sd_max_wu * sd->sd_max_ccb_per_wu); 1943 1944 return (0); 1945 } 1946 1947 void 1948 sr_ccb_free(struct sr_discipline *sd) 1949 { 1950 struct sr_ccb *ccb; 1951 1952 if (!sd) 1953 return; 1954 1955 DNPRINTF(SR_D_CCB, "%s: sr_ccb_free %p\n", DEVNAME(sd->sd_sc), sd); 1956 1957 while ((ccb = TAILQ_FIRST(&sd->sd_ccb_freeq)) != NULL) 1958 TAILQ_REMOVE(&sd->sd_ccb_freeq, ccb, ccb_link); 1959 1960 if (sd->sd_ccb) 1961 free(sd->sd_ccb, M_DEVBUF, 0); 1962 } 1963 1964 struct sr_ccb * 1965 sr_ccb_get(struct sr_discipline *sd) 1966 { 1967 struct sr_ccb *ccb; 1968 int s; 1969 1970 s = splbio(); 1971 1972 ccb = TAILQ_FIRST(&sd->sd_ccb_freeq); 1973 if (ccb) { 1974 TAILQ_REMOVE(&sd->sd_ccb_freeq, ccb, ccb_link); 1975 ccb->ccb_state = SR_CCB_INPROGRESS; 1976 } 1977 1978 splx(s); 1979 1980 DNPRINTF(SR_D_CCB, "%s: sr_ccb_get: %p\n", DEVNAME(sd->sd_sc), 1981 ccb); 1982 1983 return (ccb); 1984 } 1985 1986 void 1987 sr_ccb_put(struct sr_ccb *ccb) 1988 { 1989 struct sr_discipline *sd = ccb->ccb_dis; 1990 int s; 1991 1992 DNPRINTF(SR_D_CCB, "%s: sr_ccb_put: %p\n", DEVNAME(sd->sd_sc), 1993 ccb); 1994 1995 s = splbio(); 1996 1997 ccb->ccb_wu = NULL; 1998 ccb->ccb_state = SR_CCB_FREE; 1999 ccb->ccb_target = -1; 2000 ccb->ccb_opaque = NULL; 2001 2002 TAILQ_INSERT_TAIL(&sd->sd_ccb_freeq, ccb, ccb_link); 2003 2004 splx(s); 2005 } 2006 2007 struct sr_ccb * 2008 sr_ccb_rw(struct sr_discipline *sd, int chunk, daddr_t blkno, 2009 daddr_t len, u_int8_t *data, int xsflags, int ccbflags) 2010 { 2011 struct sr_chunk *sc = sd->sd_vol.sv_chunks[chunk]; 2012 struct sr_ccb *ccb = NULL; 2013 2014 ccb = sr_ccb_get(sd); 2015 if (ccb == NULL) 2016 goto out; 2017 2018 ccb->ccb_flags = ccbflags; 2019 ccb->ccb_target = chunk; 2020 2021 ccb->ccb_buf.b_flags = B_PHYS | B_CALL; 2022 if (ISSET(xsflags, SCSI_DATA_IN)) 2023 ccb->ccb_buf.b_flags |= B_READ; 2024 else 2025 ccb->ccb_buf.b_flags |= B_WRITE; 2026 2027 ccb->ccb_buf.b_blkno = blkno; 2028 ccb->ccb_buf.b_bcount = len; 2029 ccb->ccb_buf.b_bufsize = len; 2030 ccb->ccb_buf.b_resid = len; 2031 ccb->ccb_buf.b_data = data; 2032 ccb->ccb_buf.b_error = 0; 2033 ccb->ccb_buf.b_iodone = sd->sd_scsi_intr; 2034 ccb->ccb_buf.b_proc = curproc; 2035 ccb->ccb_buf.b_dev = sc->src_dev_mm; 2036 ccb->ccb_buf.b_vp = sc->src_vn; 2037 ccb->ccb_buf.b_bq = NULL; 2038 2039 if (!ISSET(ccb->ccb_buf.b_flags, B_READ)) 2040 ccb->ccb_buf.b_vp->v_numoutput++; 2041 2042 LIST_INIT(&ccb->ccb_buf.b_dep); 2043 2044 DNPRINTF(SR_D_DIS, "%s: %s %s ccb " 2045 "b_bcount %d b_blkno %lld b_flags 0x%0x b_data %p\n", 2046 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, sd->sd_name, 2047 ccb->ccb_buf.b_bcount, (long long)ccb->ccb_buf.b_blkno, 2048 ccb->ccb_buf.b_flags, ccb->ccb_buf.b_data); 2049 2050 out: 2051 return ccb; 2052 } 2053 2054 void 2055 sr_ccb_done(struct sr_ccb *ccb) 2056 { 2057 struct sr_workunit *wu = ccb->ccb_wu; 2058 struct sr_discipline *sd = wu->swu_dis; 2059 struct sr_softc *sc = sd->sd_sc; 2060 2061 DNPRINTF(SR_D_INTR, "%s: %s %s ccb done b_bcount %d b_resid %d" 2062 " b_flags 0x%0x block %lld target %d\n", 2063 DEVNAME(sc), sd->sd_meta->ssd_devname, sd->sd_name, 2064 ccb->ccb_buf.b_bcount, ccb->ccb_buf.b_resid, ccb->ccb_buf.b_flags, 2065 (long long)ccb->ccb_buf.b_blkno, ccb->ccb_target); 2066 2067 splassert(IPL_BIO); 2068 2069 if (ccb->ccb_target == -1) 2070 panic("%s: invalid target on wu: %p", DEVNAME(sc), wu); 2071 2072 if (ccb->ccb_buf.b_flags & B_ERROR) { 2073 DNPRINTF(SR_D_INTR, "%s: i/o error on block %lld target %d\n", 2074 DEVNAME(sc), (long long)ccb->ccb_buf.b_blkno, 2075 ccb->ccb_target); 2076 if (ISSET(sd->sd_capabilities, SR_CAP_REDUNDANT)) 2077 sd->sd_set_chunk_state(sd, ccb->ccb_target, 2078 BIOC_SDOFFLINE); 2079 else 2080 printf("%s: i/o error on block %lld target %d " 2081 "b_error %d\n", DEVNAME(sc), 2082 (long long)ccb->ccb_buf.b_blkno, ccb->ccb_target, 2083 ccb->ccb_buf.b_error); 2084 ccb->ccb_state = SR_CCB_FAILED; 2085 wu->swu_ios_failed++; 2086 } else { 2087 ccb->ccb_state = SR_CCB_OK; 2088 wu->swu_ios_succeeded++; 2089 } 2090 2091 wu->swu_ios_complete++; 2092 } 2093 2094 int 2095 sr_wu_alloc(struct sr_discipline *sd, int wu_size) 2096 { 2097 struct sr_workunit *wu; 2098 int i, no_wu; 2099 2100 DNPRINTF(SR_D_WU, "%s: sr_wu_alloc %p %d\n", DEVNAME(sd->sd_sc), 2101 sd, sd->sd_max_wu); 2102 2103 no_wu = sd->sd_max_wu; 2104 sd->sd_wu_pending = no_wu; 2105 2106 mtx_init(&sd->sd_wu_mtx, IPL_BIO); 2107 TAILQ_INIT(&sd->sd_wu); 2108 TAILQ_INIT(&sd->sd_wu_freeq); 2109 TAILQ_INIT(&sd->sd_wu_pendq); 2110 TAILQ_INIT(&sd->sd_wu_defq); 2111 2112 for (i = 0; i < no_wu; i++) { 2113 wu = malloc(wu_size, M_DEVBUF, M_WAITOK | M_ZERO); 2114 TAILQ_INSERT_TAIL(&sd->sd_wu, wu, swu_next); 2115 TAILQ_INIT(&wu->swu_ccb); 2116 task_set(&wu->swu_task, sr_wu_done_callback, sd, wu); 2117 wu->swu_dis = sd; 2118 sr_wu_put(sd, wu); 2119 } 2120 2121 return (0); 2122 } 2123 2124 void 2125 sr_wu_free(struct sr_discipline *sd) 2126 { 2127 struct sr_workunit *wu; 2128 2129 DNPRINTF(SR_D_WU, "%s: sr_wu_free %p\n", DEVNAME(sd->sd_sc), sd); 2130 2131 while ((wu = TAILQ_FIRST(&sd->sd_wu_freeq)) != NULL) 2132 TAILQ_REMOVE(&sd->sd_wu_freeq, wu, swu_link); 2133 while ((wu = TAILQ_FIRST(&sd->sd_wu_pendq)) != NULL) 2134 TAILQ_REMOVE(&sd->sd_wu_pendq, wu, swu_link); 2135 while ((wu = TAILQ_FIRST(&sd->sd_wu_defq)) != NULL) 2136 TAILQ_REMOVE(&sd->sd_wu_defq, wu, swu_link); 2137 2138 while ((wu = TAILQ_FIRST(&sd->sd_wu)) != NULL) { 2139 TAILQ_REMOVE(&sd->sd_wu, wu, swu_next); 2140 free(wu, M_DEVBUF, 0); 2141 } 2142 } 2143 2144 void * 2145 sr_wu_get(void *xsd) 2146 { 2147 struct sr_discipline *sd = (struct sr_discipline *)xsd; 2148 struct sr_workunit *wu; 2149 2150 mtx_enter(&sd->sd_wu_mtx); 2151 wu = TAILQ_FIRST(&sd->sd_wu_freeq); 2152 if (wu) { 2153 TAILQ_REMOVE(&sd->sd_wu_freeq, wu, swu_link); 2154 sd->sd_wu_pending++; 2155 } 2156 mtx_leave(&sd->sd_wu_mtx); 2157 2158 DNPRINTF(SR_D_WU, "%s: sr_wu_get: %p\n", DEVNAME(sd->sd_sc), wu); 2159 2160 return (wu); 2161 } 2162 2163 void 2164 sr_wu_put(void *xsd, void *xwu) 2165 { 2166 struct sr_discipline *sd = (struct sr_discipline *)xsd; 2167 struct sr_workunit *wu = (struct sr_workunit *)xwu; 2168 2169 DNPRINTF(SR_D_WU, "%s: sr_wu_put: %p\n", DEVNAME(sd->sd_sc), wu); 2170 2171 sr_wu_release_ccbs(wu); 2172 sr_wu_init(sd, wu); 2173 2174 mtx_enter(&sd->sd_wu_mtx); 2175 TAILQ_INSERT_TAIL(&sd->sd_wu_freeq, wu, swu_link); 2176 sd->sd_wu_pending--; 2177 mtx_leave(&sd->sd_wu_mtx); 2178 } 2179 2180 void 2181 sr_wu_init(struct sr_discipline *sd, struct sr_workunit *wu) 2182 { 2183 int s; 2184 2185 s = splbio(); 2186 if (wu->swu_cb_active == 1) 2187 panic("%s: sr_wu_init got active wu", DEVNAME(sd->sd_sc)); 2188 splx(s); 2189 2190 wu->swu_xs = NULL; 2191 wu->swu_state = SR_WU_FREE; 2192 wu->swu_flags = 0; 2193 wu->swu_blk_start = 0; 2194 wu->swu_blk_end = 0; 2195 wu->swu_collider = NULL; 2196 } 2197 2198 void 2199 sr_wu_enqueue_ccb(struct sr_workunit *wu, struct sr_ccb *ccb) 2200 { 2201 struct sr_discipline *sd = wu->swu_dis; 2202 int s; 2203 2204 s = splbio(); 2205 if (wu->swu_cb_active == 1) 2206 panic("%s: sr_wu_enqueue_ccb got active wu", 2207 DEVNAME(sd->sd_sc)); 2208 ccb->ccb_wu = wu; 2209 wu->swu_io_count++; 2210 TAILQ_INSERT_TAIL(&wu->swu_ccb, ccb, ccb_link); 2211 splx(s); 2212 } 2213 2214 void 2215 sr_wu_release_ccbs(struct sr_workunit *wu) 2216 { 2217 struct sr_ccb *ccb; 2218 2219 /* Return all ccbs that are associated with this workunit. */ 2220 while ((ccb = TAILQ_FIRST(&wu->swu_ccb)) != NULL) { 2221 TAILQ_REMOVE(&wu->swu_ccb, ccb, ccb_link); 2222 sr_ccb_put(ccb); 2223 } 2224 2225 wu->swu_io_count = 0; 2226 wu->swu_ios_complete = 0; 2227 wu->swu_ios_failed = 0; 2228 wu->swu_ios_succeeded = 0; 2229 } 2230 2231 void 2232 sr_wu_done(struct sr_workunit *wu) 2233 { 2234 struct sr_discipline *sd = wu->swu_dis; 2235 2236 DNPRINTF(SR_D_INTR, "%s: sr_wu_done count %d completed %d failed %d\n", 2237 DEVNAME(sd->sd_sc), wu->swu_io_count, wu->swu_ios_complete, 2238 wu->swu_ios_failed); 2239 2240 if (wu->swu_ios_complete < wu->swu_io_count) 2241 return; 2242 2243 task_add(sd->sd_taskq, &wu->swu_task); 2244 } 2245 2246 void 2247 sr_wu_done_callback(void *arg1, void *arg2) 2248 { 2249 struct sr_discipline *sd = (struct sr_discipline *)arg1; 2250 struct sr_workunit *wu = (struct sr_workunit *)arg2; 2251 struct scsi_xfer *xs = wu->swu_xs; 2252 struct sr_workunit *wup; 2253 int s; 2254 2255 /* 2256 * The SR_WUF_DISCIPLINE or SR_WUF_REBUILD flag must be set if 2257 * the work unit is not associated with a scsi_xfer. 2258 */ 2259 KASSERT(xs != NULL || 2260 (wu->swu_flags & (SR_WUF_DISCIPLINE|SR_WUF_REBUILD))); 2261 2262 s = splbio(); 2263 2264 if (xs != NULL) { 2265 if (wu->swu_ios_failed) 2266 xs->error = XS_DRIVER_STUFFUP; 2267 else 2268 xs->error = XS_NOERROR; 2269 } 2270 2271 if (sd->sd_scsi_wu_done) { 2272 if (sd->sd_scsi_wu_done(wu) == SR_WU_RESTART) 2273 goto done; 2274 } 2275 2276 /* Remove work unit from pending queue. */ 2277 TAILQ_FOREACH(wup, &sd->sd_wu_pendq, swu_link) 2278 if (wup == wu) 2279 break; 2280 if (wup == NULL) 2281 panic("%s: wu %p not on pending queue", 2282 DEVNAME(sd->sd_sc), wu); 2283 TAILQ_REMOVE(&sd->sd_wu_pendq, wu, swu_link); 2284 2285 if (wu->swu_collider) { 2286 if (wu->swu_ios_failed) 2287 sr_raid_recreate_wu(wu->swu_collider); 2288 2289 /* XXX Should the collider be failed if this xs failed? */ 2290 sr_raid_startwu(wu->swu_collider); 2291 } 2292 2293 /* 2294 * If a discipline provides its own sd_scsi_done function, then it 2295 * is responsible for calling sr_scsi_done() once I/O is complete. 2296 */ 2297 if (wu->swu_flags & SR_WUF_REBUILD) 2298 wu->swu_flags |= SR_WUF_REBUILDIOCOMP; 2299 if (wu->swu_flags & SR_WUF_WAKEUP) 2300 wakeup(wu); 2301 if (sd->sd_scsi_done) 2302 sd->sd_scsi_done(wu); 2303 else if (wu->swu_flags & SR_WUF_DISCIPLINE) 2304 sr_scsi_wu_put(sd, wu); 2305 else if (!(wu->swu_flags & SR_WUF_REBUILD)) 2306 sr_scsi_done(sd, xs); 2307 2308 done: 2309 splx(s); 2310 } 2311 2312 struct sr_workunit * 2313 sr_scsi_wu_get(struct sr_discipline *sd, int flags) 2314 { 2315 return scsi_io_get(&sd->sd_iopool, flags); 2316 } 2317 2318 void 2319 sr_scsi_wu_put(struct sr_discipline *sd, struct sr_workunit *wu) 2320 { 2321 scsi_io_put(&sd->sd_iopool, wu); 2322 2323 if (sd->sd_sync && sd->sd_wu_pending == 0) 2324 wakeup(sd); 2325 } 2326 2327 void 2328 sr_scsi_done(struct sr_discipline *sd, struct scsi_xfer *xs) 2329 { 2330 DNPRINTF(SR_D_DIS, "%s: sr_scsi_done: xs %p\n", DEVNAME(sd->sd_sc), xs); 2331 2332 if (xs->error == XS_NOERROR) 2333 xs->resid = 0; 2334 2335 scsi_done(xs); 2336 2337 if (sd->sd_sync && sd->sd_wu_pending == 0) 2338 wakeup(sd); 2339 } 2340 2341 void 2342 sr_scsi_cmd(struct scsi_xfer *xs) 2343 { 2344 struct scsi_link *link = xs->sc_link; 2345 struct sr_softc *sc = link->adapter_softc; 2346 struct sr_workunit *wu = xs->io; 2347 struct sr_discipline *sd; 2348 2349 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd target %d xs %p flags %#x\n", 2350 DEVNAME(sc), link->target, xs, xs->flags); 2351 2352 sd = sc->sc_targets[link->target]; 2353 if (sd == NULL) { 2354 printf("%s: sr_scsi_cmd NULL discipline\n", DEVNAME(sc)); 2355 goto stuffup; 2356 } 2357 2358 if (sd->sd_deleted) { 2359 printf("%s: %s device is being deleted, failing io\n", 2360 DEVNAME(sc), sd->sd_meta->ssd_devname); 2361 goto stuffup; 2362 } 2363 2364 /* scsi layer *can* re-send wu without calling sr_wu_put(). */ 2365 sr_wu_release_ccbs(wu); 2366 sr_wu_init(sd, wu); 2367 wu->swu_state = SR_WU_INPROGRESS; 2368 wu->swu_xs = xs; 2369 2370 switch (xs->cmd->opcode) { 2371 case READ_COMMAND: 2372 case READ_BIG: 2373 case READ_16: 2374 case WRITE_COMMAND: 2375 case WRITE_BIG: 2376 case WRITE_16: 2377 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: READ/WRITE %02x\n", 2378 DEVNAME(sc), xs->cmd->opcode); 2379 if (sd->sd_scsi_rw(wu)) 2380 goto stuffup; 2381 break; 2382 2383 case SYNCHRONIZE_CACHE: 2384 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: SYNCHRONIZE_CACHE\n", 2385 DEVNAME(sc)); 2386 if (sd->sd_scsi_sync(wu)) 2387 goto stuffup; 2388 goto complete; 2389 2390 case TEST_UNIT_READY: 2391 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: TEST_UNIT_READY\n", 2392 DEVNAME(sc)); 2393 if (sd->sd_scsi_tur(wu)) 2394 goto stuffup; 2395 goto complete; 2396 2397 case START_STOP: 2398 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: START_STOP\n", 2399 DEVNAME(sc)); 2400 if (sd->sd_scsi_start_stop(wu)) 2401 goto stuffup; 2402 goto complete; 2403 2404 case INQUIRY: 2405 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: INQUIRY\n", 2406 DEVNAME(sc)); 2407 if (sd->sd_scsi_inquiry(wu)) 2408 goto stuffup; 2409 goto complete; 2410 2411 case READ_CAPACITY: 2412 case READ_CAPACITY_16: 2413 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd READ CAPACITY 0x%02x\n", 2414 DEVNAME(sc), xs->cmd->opcode); 2415 if (sd->sd_scsi_read_cap(wu)) 2416 goto stuffup; 2417 goto complete; 2418 2419 case REQUEST_SENSE: 2420 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd REQUEST SENSE\n", 2421 DEVNAME(sc)); 2422 if (sd->sd_scsi_req_sense(wu)) 2423 goto stuffup; 2424 goto complete; 2425 2426 default: 2427 DNPRINTF(SR_D_CMD, "%s: unsupported scsi command %x\n", 2428 DEVNAME(sc), xs->cmd->opcode); 2429 /* XXX might need to add generic function to handle others */ 2430 goto stuffup; 2431 } 2432 2433 return; 2434 stuffup: 2435 if (sd && sd->sd_scsi_sense.error_code) { 2436 xs->error = XS_SENSE; 2437 bcopy(&sd->sd_scsi_sense, &xs->sense, sizeof(xs->sense)); 2438 bzero(&sd->sd_scsi_sense, sizeof(sd->sd_scsi_sense)); 2439 } else { 2440 xs->error = XS_DRIVER_STUFFUP; 2441 } 2442 complete: 2443 sr_scsi_done(sd, xs); 2444 } 2445 2446 int 2447 sr_scsi_probe(struct scsi_link *link) 2448 { 2449 struct sr_softc *sc = link->adapter_softc; 2450 struct sr_discipline *sd; 2451 2452 KASSERT(link->target < SR_MAX_LD && link->lun == 0); 2453 2454 sd = sc->sc_targets[link->target]; 2455 if (sd == NULL) 2456 return (ENODEV); 2457 2458 link->pool = &sd->sd_iopool; 2459 if (sd->sd_openings) 2460 link->openings = sd->sd_openings(sd); 2461 else 2462 link->openings = sd->sd_max_wu; 2463 2464 return (0); 2465 } 2466 2467 int 2468 sr_scsi_ioctl(struct scsi_link *link, u_long cmd, caddr_t addr, int flag) 2469 { 2470 DNPRINTF(SR_D_IOCTL, "%s: sr_scsi_ioctl cmd: %#x\n", 2471 DEVNAME((struct sr_softc *)link->adapter_softc), cmd); 2472 2473 /* Pass bio ioctls through to bio handler. */ 2474 if (IOCGROUP(cmd) == 'B') 2475 return (sr_bio_ioctl(link->adapter_softc, cmd, addr)); 2476 2477 switch (cmd) { 2478 case DIOCGCACHE: 2479 case DIOCSCACHE: 2480 return (EOPNOTSUPP); 2481 default: 2482 return (ENOTTY); 2483 } 2484 } 2485 2486 int 2487 sr_bio_ioctl(struct device *dev, u_long cmd, caddr_t addr) 2488 { 2489 struct sr_softc *sc = (struct sr_softc *)dev; 2490 struct bio *bio = (struct bio *)addr; 2491 int rv = 0; 2492 2493 DNPRINTF(SR_D_IOCTL, "%s: sr_bio_ioctl ", DEVNAME(sc)); 2494 2495 rw_enter_write(&sc->sc_lock); 2496 2497 bio_status_init(&sc->sc_status, &sc->sc_dev); 2498 2499 switch (cmd) { 2500 case BIOCINQ: 2501 DNPRINTF(SR_D_IOCTL, "inq\n"); 2502 rv = sr_ioctl_inq(sc, (struct bioc_inq *)addr); 2503 break; 2504 2505 case BIOCVOL: 2506 DNPRINTF(SR_D_IOCTL, "vol\n"); 2507 rv = sr_ioctl_vol(sc, (struct bioc_vol *)addr); 2508 break; 2509 2510 case BIOCDISK: 2511 DNPRINTF(SR_D_IOCTL, "disk\n"); 2512 rv = sr_ioctl_disk(sc, (struct bioc_disk *)addr); 2513 break; 2514 2515 case BIOCALARM: 2516 DNPRINTF(SR_D_IOCTL, "alarm\n"); 2517 /*rv = sr_ioctl_alarm(sc, (struct bioc_alarm *)addr); */ 2518 break; 2519 2520 case BIOCBLINK: 2521 DNPRINTF(SR_D_IOCTL, "blink\n"); 2522 /*rv = sr_ioctl_blink(sc, (struct bioc_blink *)addr); */ 2523 break; 2524 2525 case BIOCSETSTATE: 2526 DNPRINTF(SR_D_IOCTL, "setstate\n"); 2527 rv = sr_ioctl_setstate(sc, (struct bioc_setstate *)addr); 2528 break; 2529 2530 case BIOCCREATERAID: 2531 DNPRINTF(SR_D_IOCTL, "createraid\n"); 2532 rv = sr_ioctl_createraid(sc, (struct bioc_createraid *)addr, 2533 1, NULL); 2534 break; 2535 2536 case BIOCDELETERAID: 2537 DNPRINTF(SR_D_IOCTL, "deleteraid\n"); 2538 rv = sr_ioctl_deleteraid(sc, (struct bioc_deleteraid *)addr); 2539 break; 2540 2541 case BIOCDISCIPLINE: 2542 DNPRINTF(SR_D_IOCTL, "discipline\n"); 2543 rv = sr_ioctl_discipline(sc, (struct bioc_discipline *)addr); 2544 break; 2545 2546 case BIOCINSTALLBOOT: 2547 DNPRINTF(SR_D_IOCTL, "installboot\n"); 2548 rv = sr_ioctl_installboot(sc, (struct bioc_installboot *)addr); 2549 break; 2550 2551 default: 2552 DNPRINTF(SR_D_IOCTL, "invalid ioctl\n"); 2553 rv = ENOTTY; 2554 } 2555 2556 sc->sc_status.bs_status = (rv ? BIO_STATUS_ERROR : BIO_STATUS_SUCCESS); 2557 2558 bcopy(&sc->sc_status, &bio->bio_status, sizeof(struct bio_status)); 2559 2560 rw_exit_write(&sc->sc_lock); 2561 2562 return (0); 2563 } 2564 2565 int 2566 sr_ioctl_inq(struct sr_softc *sc, struct bioc_inq *bi) 2567 { 2568 struct sr_discipline *sd; 2569 int vol = 0, disk = 0; 2570 2571 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2572 vol++; 2573 disk += sd->sd_meta->ssdi.ssd_chunk_no; 2574 } 2575 2576 strlcpy(bi->bi_dev, sc->sc_dev.dv_xname, sizeof(bi->bi_dev)); 2577 bi->bi_novol = vol + sc->sc_hotspare_no; 2578 bi->bi_nodisk = disk + sc->sc_hotspare_no; 2579 2580 return (0); 2581 } 2582 2583 int 2584 sr_ioctl_vol(struct sr_softc *sc, struct bioc_vol *bv) 2585 { 2586 int vol = -1, rv = EINVAL; 2587 struct sr_discipline *sd; 2588 struct sr_chunk *hotspare; 2589 daddr_t rb, sz; 2590 2591 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2592 vol++; 2593 if (vol != bv->bv_volid) 2594 continue; 2595 2596 bv->bv_status = sd->sd_vol_status; 2597 bv->bv_size = sd->sd_meta->ssdi.ssd_size << DEV_BSHIFT; 2598 bv->bv_level = sd->sd_meta->ssdi.ssd_level; 2599 bv->bv_nodisk = sd->sd_meta->ssdi.ssd_chunk_no; 2600 2601 #ifdef CRYPTO 2602 if (sd->sd_meta->ssdi.ssd_level == 'C' && 2603 sd->mds.mdd_crypto.key_disk != NULL) 2604 bv->bv_nodisk++; 2605 #endif 2606 2607 if (bv->bv_status == BIOC_SVREBUILD) { 2608 sz = sd->sd_meta->ssdi.ssd_size; 2609 rb = sd->sd_meta->ssd_rebuild; 2610 if (rb > 0) 2611 bv->bv_percent = 100 - 2612 ((sz * 100 - rb * 100) / sz) - 1; 2613 else 2614 bv->bv_percent = 0; 2615 } 2616 strlcpy(bv->bv_dev, sd->sd_meta->ssd_devname, 2617 sizeof(bv->bv_dev)); 2618 strlcpy(bv->bv_vendor, sd->sd_meta->ssdi.ssd_vendor, 2619 sizeof(bv->bv_vendor)); 2620 rv = 0; 2621 goto done; 2622 } 2623 2624 /* Check hotspares list. */ 2625 SLIST_FOREACH(hotspare, &sc->sc_hotspare_list, src_link) { 2626 vol++; 2627 if (vol != bv->bv_volid) 2628 continue; 2629 2630 bv->bv_status = BIOC_SVONLINE; 2631 bv->bv_size = hotspare->src_meta.scmi.scm_size << DEV_BSHIFT; 2632 bv->bv_level = -1; /* Hotspare. */ 2633 bv->bv_nodisk = 1; 2634 strlcpy(bv->bv_dev, hotspare->src_meta.scmi.scm_devname, 2635 sizeof(bv->bv_dev)); 2636 strlcpy(bv->bv_vendor, hotspare->src_meta.scmi.scm_devname, 2637 sizeof(bv->bv_vendor)); 2638 rv = 0; 2639 goto done; 2640 } 2641 2642 done: 2643 return (rv); 2644 } 2645 2646 int 2647 sr_ioctl_disk(struct sr_softc *sc, struct bioc_disk *bd) 2648 { 2649 struct sr_discipline *sd; 2650 struct sr_chunk *src, *hotspare; 2651 int vol = -1, rv = EINVAL; 2652 2653 if (bd->bd_diskid < 0) 2654 goto done; 2655 2656 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2657 vol++; 2658 if (vol != bd->bd_volid) 2659 continue; 2660 2661 if (bd->bd_diskid < sd->sd_meta->ssdi.ssd_chunk_no) 2662 src = sd->sd_vol.sv_chunks[bd->bd_diskid]; 2663 #ifdef CRYPTO 2664 else if (bd->bd_diskid == sd->sd_meta->ssdi.ssd_chunk_no && 2665 sd->sd_meta->ssdi.ssd_level == 'C' && 2666 sd->mds.mdd_crypto.key_disk != NULL) 2667 src = sd->mds.mdd_crypto.key_disk; 2668 #endif 2669 else 2670 break; 2671 2672 bd->bd_status = src->src_meta.scm_status; 2673 bd->bd_size = src->src_meta.scmi.scm_size << DEV_BSHIFT; 2674 bd->bd_channel = vol; 2675 bd->bd_target = bd->bd_diskid; 2676 strlcpy(bd->bd_vendor, src->src_meta.scmi.scm_devname, 2677 sizeof(bd->bd_vendor)); 2678 rv = 0; 2679 goto done; 2680 } 2681 2682 /* Check hotspares list. */ 2683 SLIST_FOREACH(hotspare, &sc->sc_hotspare_list, src_link) { 2684 vol++; 2685 if (vol != bd->bd_volid) 2686 continue; 2687 2688 if (bd->bd_diskid != 0) 2689 break; 2690 2691 bd->bd_status = hotspare->src_meta.scm_status; 2692 bd->bd_size = hotspare->src_meta.scmi.scm_size << DEV_BSHIFT; 2693 bd->bd_channel = vol; 2694 bd->bd_target = bd->bd_diskid; 2695 strlcpy(bd->bd_vendor, hotspare->src_meta.scmi.scm_devname, 2696 sizeof(bd->bd_vendor)); 2697 rv = 0; 2698 goto done; 2699 } 2700 2701 done: 2702 return (rv); 2703 } 2704 2705 int 2706 sr_ioctl_setstate(struct sr_softc *sc, struct bioc_setstate *bs) 2707 { 2708 int rv = EINVAL; 2709 int vol = -1, found, c; 2710 struct sr_discipline *sd; 2711 struct sr_chunk *ch_entry; 2712 struct sr_chunk_head *cl; 2713 2714 if (bs->bs_other_id_type == BIOC_SSOTHER_UNUSED) 2715 goto done; 2716 2717 if (bs->bs_status == BIOC_SSHOTSPARE) { 2718 rv = sr_hotspare(sc, (dev_t)bs->bs_other_id); 2719 goto done; 2720 } 2721 2722 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2723 vol++; 2724 if (vol == bs->bs_volid) 2725 break; 2726 } 2727 if (sd == NULL) 2728 goto done; 2729 2730 switch (bs->bs_status) { 2731 case BIOC_SSOFFLINE: 2732 /* Take chunk offline */ 2733 found = c = 0; 2734 cl = &sd->sd_vol.sv_chunk_list; 2735 SLIST_FOREACH(ch_entry, cl, src_link) { 2736 if (ch_entry->src_dev_mm == bs->bs_other_id) { 2737 found = 1; 2738 break; 2739 } 2740 c++; 2741 } 2742 if (found == 0) { 2743 sr_error(sc, "chunk not part of array"); 2744 goto done; 2745 } 2746 2747 /* XXX: check current state first */ 2748 sd->sd_set_chunk_state(sd, c, BIOC_SDOFFLINE); 2749 2750 if (sr_meta_save(sd, SR_META_DIRTY)) { 2751 sr_error(sc, "could not save metadata for %s", 2752 sd->sd_meta->ssd_devname); 2753 goto done; 2754 } 2755 rv = 0; 2756 break; 2757 2758 case BIOC_SDSCRUB: 2759 break; 2760 2761 case BIOC_SSREBUILD: 2762 rv = sr_rebuild_init(sd, (dev_t)bs->bs_other_id, 0); 2763 break; 2764 2765 default: 2766 sr_error(sc, "unsupported state request %d", bs->bs_status); 2767 } 2768 2769 done: 2770 return (rv); 2771 } 2772 2773 int 2774 sr_chunk_in_use(struct sr_softc *sc, dev_t dev) 2775 { 2776 struct sr_discipline *sd; 2777 struct sr_chunk *chunk; 2778 int i; 2779 2780 DNPRINTF(SR_D_MISC, "%s: sr_chunk_in_use(%d)\n", DEVNAME(sc), dev); 2781 2782 if (dev == NODEV) 2783 return BIOC_SDINVALID; 2784 2785 /* See if chunk is already in use. */ 2786 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2787 for (i = 0; i < sd->sd_meta->ssdi.ssd_chunk_no; i++) { 2788 chunk = sd->sd_vol.sv_chunks[i]; 2789 if (chunk->src_dev_mm == dev) 2790 return chunk->src_meta.scm_status; 2791 } 2792 } 2793 2794 /* Check hotspares list. */ 2795 SLIST_FOREACH(chunk, &sc->sc_hotspare_list, src_link) 2796 if (chunk->src_dev_mm == dev) 2797 return chunk->src_meta.scm_status; 2798 2799 return BIOC_SDINVALID; 2800 } 2801 2802 int 2803 sr_hotspare(struct sr_softc *sc, dev_t dev) 2804 { 2805 struct sr_discipline *sd = NULL; 2806 struct sr_metadata *sm = NULL; 2807 struct sr_meta_chunk *hm; 2808 struct sr_chunk_head *cl; 2809 struct sr_chunk *chunk, *last, *hotspare = NULL; 2810 struct sr_uuid uuid; 2811 struct disklabel label; 2812 struct vnode *vn; 2813 daddr_t size; 2814 char devname[32]; 2815 int rv = EINVAL; 2816 int c, part, open = 0; 2817 2818 /* 2819 * Add device to global hotspares list. 2820 */ 2821 2822 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 2823 2824 /* Make sure chunk is not already in use. */ 2825 c = sr_chunk_in_use(sc, dev); 2826 if (c != BIOC_SDINVALID && c != BIOC_SDOFFLINE) { 2827 if (c == BIOC_SDHOTSPARE) 2828 sr_error(sc, "%s is already a hotspare", devname); 2829 else 2830 sr_error(sc, "%s is already in use", devname); 2831 goto done; 2832 } 2833 2834 /* XXX - See if there is an existing degraded volume... */ 2835 2836 /* Open device. */ 2837 if (bdevvp(dev, &vn)) { 2838 sr_error(sc, "sr_hotspare: cannot allocate vnode"); 2839 goto done; 2840 } 2841 if (VOP_OPEN(vn, FREAD | FWRITE, NOCRED, curproc)) { 2842 DNPRINTF(SR_D_META,"%s: sr_hotspare cannot open %s\n", 2843 DEVNAME(sc), devname); 2844 vput(vn); 2845 goto fail; 2846 } 2847 open = 1; /* close dev on error */ 2848 2849 /* Get partition details. */ 2850 part = DISKPART(dev); 2851 if (VOP_IOCTL(vn, DIOCGDINFO, (caddr_t)&label, FREAD, 2852 NOCRED, curproc)) { 2853 DNPRINTF(SR_D_META, "%s: sr_hotspare ioctl failed\n", 2854 DEVNAME(sc)); 2855 VOP_CLOSE(vn, FREAD | FWRITE, NOCRED, curproc); 2856 vput(vn); 2857 goto fail; 2858 } 2859 if (label.d_secsize != DEV_BSIZE) { 2860 sr_error(sc, "%s has unsupported sector size (%u)", 2861 devname, label.d_secsize); 2862 goto fail; 2863 } 2864 if (label.d_partitions[part].p_fstype != FS_RAID) { 2865 sr_error(sc, "%s partition not of type RAID (%d)", 2866 devname, label.d_partitions[part].p_fstype); 2867 goto fail; 2868 } 2869 2870 /* Calculate partition size. */ 2871 size = DL_SECTOBLK(&label, DL_GETPSIZE(&label.d_partitions[part])) - 2872 SR_DATA_OFFSET; 2873 2874 /* 2875 * Create and populate chunk metadata. 2876 */ 2877 2878 sr_uuid_generate(&uuid); 2879 hotspare = malloc(sizeof(struct sr_chunk), M_DEVBUF, M_WAITOK | M_ZERO); 2880 2881 hotspare->src_dev_mm = dev; 2882 hotspare->src_vn = vn; 2883 strlcpy(hotspare->src_devname, devname, sizeof(hm->scmi.scm_devname)); 2884 hotspare->src_size = size; 2885 2886 hm = &hotspare->src_meta; 2887 hm->scmi.scm_volid = SR_HOTSPARE_VOLID; 2888 hm->scmi.scm_chunk_id = 0; 2889 hm->scmi.scm_size = size; 2890 hm->scmi.scm_coerced_size = size; 2891 strlcpy(hm->scmi.scm_devname, devname, sizeof(hm->scmi.scm_devname)); 2892 bcopy(&uuid, &hm->scmi.scm_uuid, sizeof(struct sr_uuid)); 2893 2894 sr_checksum(sc, hm, &hm->scm_checksum, 2895 sizeof(struct sr_meta_chunk_invariant)); 2896 2897 hm->scm_status = BIOC_SDHOTSPARE; 2898 2899 /* 2900 * Create and populate our own discipline and metadata. 2901 */ 2902 2903 sm = malloc(sizeof(struct sr_metadata), M_DEVBUF, M_WAITOK | M_ZERO); 2904 sm->ssdi.ssd_magic = SR_MAGIC; 2905 sm->ssdi.ssd_version = SR_META_VERSION; 2906 sm->ssd_ondisk = 0; 2907 sm->ssdi.ssd_vol_flags = 0; 2908 bcopy(&uuid, &sm->ssdi.ssd_uuid, sizeof(struct sr_uuid)); 2909 sm->ssdi.ssd_chunk_no = 1; 2910 sm->ssdi.ssd_volid = SR_HOTSPARE_VOLID; 2911 sm->ssdi.ssd_level = SR_HOTSPARE_LEVEL; 2912 sm->ssdi.ssd_size = size; 2913 strlcpy(sm->ssdi.ssd_vendor, "OPENBSD", sizeof(sm->ssdi.ssd_vendor)); 2914 snprintf(sm->ssdi.ssd_product, sizeof(sm->ssdi.ssd_product), 2915 "SR %s", "HOTSPARE"); 2916 snprintf(sm->ssdi.ssd_revision, sizeof(sm->ssdi.ssd_revision), 2917 "%03d", SR_META_VERSION); 2918 2919 sd = malloc(sizeof(struct sr_discipline), M_DEVBUF, M_WAITOK | M_ZERO); 2920 sd->sd_sc = sc; 2921 sd->sd_meta = sm; 2922 sd->sd_meta_type = SR_META_F_NATIVE; 2923 sd->sd_vol_status = BIOC_SVONLINE; 2924 strlcpy(sd->sd_name, "HOTSPARE", sizeof(sd->sd_name)); 2925 SLIST_INIT(&sd->sd_meta_opt); 2926 2927 /* Add chunk to volume. */ 2928 sd->sd_vol.sv_chunks = malloc(sizeof(struct sr_chunk *), M_DEVBUF, 2929 M_WAITOK | M_ZERO); 2930 sd->sd_vol.sv_chunks[0] = hotspare; 2931 SLIST_INIT(&sd->sd_vol.sv_chunk_list); 2932 SLIST_INSERT_HEAD(&sd->sd_vol.sv_chunk_list, hotspare, src_link); 2933 2934 /* Save metadata. */ 2935 if (sr_meta_save(sd, SR_META_DIRTY)) { 2936 sr_error(sc, "could not save metadata to %s", devname); 2937 goto fail; 2938 } 2939 2940 /* 2941 * Add chunk to hotspare list. 2942 */ 2943 rw_enter_write(&sc->sc_hs_lock); 2944 cl = &sc->sc_hotspare_list; 2945 if (SLIST_EMPTY(cl)) 2946 SLIST_INSERT_HEAD(cl, hotspare, src_link); 2947 else { 2948 SLIST_FOREACH(chunk, cl, src_link) 2949 last = chunk; 2950 SLIST_INSERT_AFTER(last, hotspare, src_link); 2951 } 2952 sc->sc_hotspare_no++; 2953 rw_exit_write(&sc->sc_hs_lock); 2954 2955 rv = 0; 2956 goto done; 2957 2958 fail: 2959 if (hotspare) 2960 free(hotspare, M_DEVBUF, 0); 2961 2962 done: 2963 if (sd && sd->sd_vol.sv_chunks) 2964 free(sd->sd_vol.sv_chunks, M_DEVBUF, 0); 2965 if (sd) 2966 free(sd, M_DEVBUF, 0); 2967 if (sm) 2968 free(sm, M_DEVBUF, 0); 2969 if (open) { 2970 VOP_CLOSE(vn, FREAD | FWRITE, NOCRED, curproc); 2971 vput(vn); 2972 } 2973 2974 return (rv); 2975 } 2976 2977 void 2978 sr_hotspare_rebuild_callback(void *arg1, void *arg2) 2979 { 2980 sr_hotspare_rebuild((struct sr_discipline *)arg1); 2981 } 2982 2983 void 2984 sr_hotspare_rebuild(struct sr_discipline *sd) 2985 { 2986 struct sr_softc *sc = sd->sd_sc; 2987 struct sr_chunk_head *cl; 2988 struct sr_chunk *hotspare, *chunk = NULL; 2989 struct sr_workunit *wu; 2990 struct sr_ccb *ccb; 2991 int i, s, chunk_no, busy; 2992 2993 /* 2994 * Attempt to locate a hotspare and initiate rebuild. 2995 */ 2996 2997 for (i = 0; i < sd->sd_meta->ssdi.ssd_chunk_no; i++) { 2998 if (sd->sd_vol.sv_chunks[i]->src_meta.scm_status == 2999 BIOC_SDOFFLINE) { 3000 chunk_no = i; 3001 chunk = sd->sd_vol.sv_chunks[i]; 3002 break; 3003 } 3004 } 3005 3006 if (chunk == NULL) { 3007 printf("%s: no offline chunk found on %s!\n", 3008 DEVNAME(sc), sd->sd_meta->ssd_devname); 3009 return; 3010 } 3011 3012 /* See if we have a suitable hotspare... */ 3013 rw_enter_write(&sc->sc_hs_lock); 3014 cl = &sc->sc_hotspare_list; 3015 SLIST_FOREACH(hotspare, cl, src_link) 3016 if (hotspare->src_size >= chunk->src_size) 3017 break; 3018 3019 if (hotspare != NULL) { 3020 3021 printf("%s: %s volume degraded, will attempt to " 3022 "rebuild on hotspare %s\n", DEVNAME(sc), 3023 sd->sd_meta->ssd_devname, hotspare->src_devname); 3024 3025 /* 3026 * Ensure that all pending I/O completes on the failed chunk 3027 * before trying to initiate a rebuild. 3028 */ 3029 i = 0; 3030 do { 3031 busy = 0; 3032 3033 s = splbio(); 3034 TAILQ_FOREACH(wu, &sd->sd_wu_pendq, swu_link) { 3035 TAILQ_FOREACH(ccb, &wu->swu_ccb, ccb_link) { 3036 if (ccb->ccb_target == chunk_no) 3037 busy = 1; 3038 } 3039 } 3040 TAILQ_FOREACH(wu, &sd->sd_wu_defq, swu_link) { 3041 TAILQ_FOREACH(ccb, &wu->swu_ccb, ccb_link) { 3042 if (ccb->ccb_target == chunk_no) 3043 busy = 1; 3044 } 3045 } 3046 splx(s); 3047 3048 if (busy) { 3049 tsleep(sd, PRIBIO, "sr_hotspare", hz); 3050 i++; 3051 } 3052 3053 } while (busy && i < 120); 3054 3055 DNPRINTF(SR_D_META, "%s: waited %i seconds for I/O to " 3056 "complete on failed chunk %s\n", DEVNAME(sc), 3057 i, chunk->src_devname); 3058 3059 if (busy) { 3060 printf("%s: pending I/O failed to complete on " 3061 "failed chunk %s, hotspare rebuild aborted...\n", 3062 DEVNAME(sc), chunk->src_devname); 3063 goto done; 3064 } 3065 3066 s = splbio(); 3067 rw_enter_write(&sc->sc_lock); 3068 bio_status_init(&sc->sc_status, &sc->sc_dev); 3069 if (sr_rebuild_init(sd, hotspare->src_dev_mm, 1) == 0) { 3070 3071 /* Remove hotspare from available list. */ 3072 sc->sc_hotspare_no--; 3073 SLIST_REMOVE(cl, hotspare, sr_chunk, src_link); 3074 free(hotspare, M_DEVBUF, 0); 3075 3076 } 3077 rw_exit_write(&sc->sc_lock); 3078 splx(s); 3079 } 3080 done: 3081 rw_exit_write(&sc->sc_hs_lock); 3082 } 3083 3084 int 3085 sr_rebuild_init(struct sr_discipline *sd, dev_t dev, int hotspare) 3086 { 3087 struct sr_softc *sc = sd->sd_sc; 3088 struct sr_chunk *chunk = NULL; 3089 struct sr_meta_chunk *meta; 3090 struct disklabel label; 3091 struct vnode *vn; 3092 daddr_t size, csize; 3093 char devname[32]; 3094 int rv = EINVAL, open = 0; 3095 int cid, i, part, status; 3096 3097 /* 3098 * Attempt to initiate a rebuild onto the specified device. 3099 */ 3100 3101 if (!(sd->sd_capabilities & SR_CAP_REBUILD)) { 3102 sr_error(sc, "discipline does not support rebuild"); 3103 goto done; 3104 } 3105 3106 /* make sure volume is in the right state */ 3107 if (sd->sd_vol_status == BIOC_SVREBUILD) { 3108 sr_error(sc, "rebuild already in progress"); 3109 goto done; 3110 } 3111 if (sd->sd_vol_status != BIOC_SVDEGRADED) { 3112 sr_error(sc, "volume not degraded"); 3113 goto done; 3114 } 3115 3116 /* Find first offline chunk. */ 3117 for (cid = 0; cid < sd->sd_meta->ssdi.ssd_chunk_no; cid++) { 3118 if (sd->sd_vol.sv_chunks[cid]->src_meta.scm_status == 3119 BIOC_SDOFFLINE) { 3120 chunk = sd->sd_vol.sv_chunks[cid]; 3121 break; 3122 } 3123 } 3124 if (chunk == NULL) { 3125 sr_error(sc, "no offline chunks available to rebuild"); 3126 goto done; 3127 } 3128 3129 /* Get coerced size from another online chunk. */ 3130 for (i = 0; i < sd->sd_meta->ssdi.ssd_chunk_no; i++) { 3131 if (sd->sd_vol.sv_chunks[i]->src_meta.scm_status == 3132 BIOC_SDONLINE) { 3133 meta = &sd->sd_vol.sv_chunks[i]->src_meta; 3134 csize = meta->scmi.scm_coerced_size; 3135 break; 3136 } 3137 } 3138 3139 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 3140 if (bdevvp(dev, &vn)) { 3141 printf("%s: sr_rebuild_init: can't allocate vnode\n", 3142 DEVNAME(sc)); 3143 goto done; 3144 } 3145 if (VOP_OPEN(vn, FREAD | FWRITE, NOCRED, curproc)) { 3146 DNPRINTF(SR_D_META,"%s: sr_ioctl_setstate can't " 3147 "open %s\n", DEVNAME(sc), devname); 3148 vput(vn); 3149 goto done; 3150 } 3151 open = 1; /* close dev on error */ 3152 3153 /* Get disklabel and check partition. */ 3154 part = DISKPART(dev); 3155 if (VOP_IOCTL(vn, DIOCGDINFO, (caddr_t)&label, FREAD, 3156 NOCRED, curproc)) { 3157 DNPRINTF(SR_D_META, "%s: sr_ioctl_setstate ioctl failed\n", 3158 DEVNAME(sc)); 3159 goto done; 3160 } 3161 if (label.d_secsize != DEV_BSIZE) { 3162 sr_error(sc, "%s has unsupported sector size (%u)", 3163 devname, label.d_secsize); 3164 goto done; 3165 } 3166 if (label.d_partitions[part].p_fstype != FS_RAID) { 3167 sr_error(sc, "%s partition not of type RAID (%d)", 3168 devname, label.d_partitions[part].p_fstype); 3169 goto done; 3170 } 3171 3172 /* Is the partition large enough? */ 3173 size = DL_SECTOBLK(&label, DL_GETPSIZE(&label.d_partitions[part])) - 3174 sd->sd_meta->ssd_data_offset; 3175 if (size < csize) { 3176 sr_error(sc, "%s partition too small, at least %lld bytes " 3177 "required", devname, (long long)(csize << DEV_BSHIFT)); 3178 goto done; 3179 } else if (size > csize) 3180 sr_warn(sc, "%s partition too large, wasting %lld bytes", 3181 devname, (long long)((size - csize) << DEV_BSHIFT)); 3182 3183 /* Ensure that this chunk is not already in use. */ 3184 status = sr_chunk_in_use(sc, dev); 3185 if (status != BIOC_SDINVALID && status != BIOC_SDOFFLINE && 3186 !(hotspare && status == BIOC_SDHOTSPARE)) { 3187 sr_error(sc, "%s is already in use", devname); 3188 goto done; 3189 } 3190 3191 /* Reset rebuild counter since we rebuilding onto a new chunk. */ 3192 sd->sd_meta->ssd_rebuild = 0; 3193 3194 open = 0; /* leave dev open from here on out */ 3195 3196 /* Fix up chunk. */ 3197 bcopy(label.d_uid, chunk->src_duid, sizeof(chunk->src_duid)); 3198 chunk->src_dev_mm = dev; 3199 chunk->src_vn = vn; 3200 3201 /* Reconstruct metadata. */ 3202 meta = &chunk->src_meta; 3203 meta->scmi.scm_volid = sd->sd_meta->ssdi.ssd_volid; 3204 meta->scmi.scm_chunk_id = cid; 3205 strlcpy(meta->scmi.scm_devname, devname, 3206 sizeof(meta->scmi.scm_devname)); 3207 meta->scmi.scm_size = size; 3208 meta->scmi.scm_coerced_size = csize; 3209 bcopy(&sd->sd_meta->ssdi.ssd_uuid, &meta->scmi.scm_uuid, 3210 sizeof(meta->scmi.scm_uuid)); 3211 sr_checksum(sc, meta, &meta->scm_checksum, 3212 sizeof(struct sr_meta_chunk_invariant)); 3213 3214 sd->sd_set_chunk_state(sd, cid, BIOC_SDREBUILD); 3215 3216 if (sr_meta_save(sd, SR_META_DIRTY)) { 3217 sr_error(sc, "could not save metadata to %s", devname); 3218 open = 1; 3219 goto done; 3220 } 3221 3222 sr_warn(sc, "rebuild of %s started on %s", 3223 sd->sd_meta->ssd_devname, devname); 3224 3225 sd->sd_reb_abort = 0; 3226 kthread_create_deferred(sr_rebuild_start, sd); 3227 3228 rv = 0; 3229 done: 3230 if (open) { 3231 VOP_CLOSE(vn, FREAD | FWRITE, NOCRED, curproc); 3232 vput(vn); 3233 } 3234 3235 return (rv); 3236 } 3237 3238 void 3239 sr_roam_chunks(struct sr_discipline *sd) 3240 { 3241 struct sr_softc *sc = sd->sd_sc; 3242 struct sr_chunk *chunk; 3243 struct sr_meta_chunk *meta; 3244 int roamed = 0; 3245 3246 /* Have any chunks roamed? */ 3247 SLIST_FOREACH(chunk, &sd->sd_vol.sv_chunk_list, src_link) { 3248 meta = &chunk->src_meta; 3249 if (strncmp(meta->scmi.scm_devname, chunk->src_devname, 3250 sizeof(meta->scmi.scm_devname))) { 3251 3252 printf("%s: roaming device %s -> %s\n", DEVNAME(sc), 3253 meta->scmi.scm_devname, chunk->src_devname); 3254 3255 strlcpy(meta->scmi.scm_devname, chunk->src_devname, 3256 sizeof(meta->scmi.scm_devname)); 3257 3258 roamed++; 3259 } 3260 } 3261 3262 if (roamed) 3263 sr_meta_save(sd, SR_META_DIRTY); 3264 } 3265 3266 int 3267 sr_ioctl_createraid(struct sr_softc *sc, struct bioc_createraid *bc, 3268 int user, void *data) 3269 { 3270 struct sr_meta_opt_item *omi; 3271 struct sr_chunk_head *cl; 3272 struct sr_discipline *sd = NULL; 3273 struct sr_chunk *ch_entry; 3274 struct scsi_link *link; 3275 struct device *dev; 3276 char *uuid, devname[32]; 3277 dev_t *dt; 3278 int i, no_chunk, rv = EINVAL, target, vol; 3279 int no_meta; 3280 3281 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_createraid(%d)\n", 3282 DEVNAME(sc), user); 3283 3284 /* user input */ 3285 if (bc->bc_dev_list_len > BIOC_CRMAXLEN) 3286 goto unwind; 3287 3288 dt = malloc(bc->bc_dev_list_len, M_DEVBUF, M_WAITOK | M_ZERO); 3289 if (user) { 3290 if (copyin(bc->bc_dev_list, dt, bc->bc_dev_list_len) != 0) 3291 goto unwind; 3292 } else 3293 bcopy(bc->bc_dev_list, dt, bc->bc_dev_list_len); 3294 3295 /* Initialise discipline. */ 3296 sd = malloc(sizeof(struct sr_discipline), M_DEVBUF, M_WAITOK | M_ZERO); 3297 sd->sd_sc = sc; 3298 SLIST_INIT(&sd->sd_meta_opt); 3299 sd->sd_taskq = taskq_create("srdis", 1, IPL_BIO); 3300 if (sd->sd_taskq == NULL) { 3301 sr_error(sc, "could not create discipline taskq"); 3302 goto unwind; 3303 } 3304 if (sr_discipline_init(sd, bc->bc_level)) { 3305 sr_error(sc, "could not initialize discipline"); 3306 goto unwind; 3307 } 3308 3309 no_chunk = bc->bc_dev_list_len / sizeof(dev_t); 3310 cl = &sd->sd_vol.sv_chunk_list; 3311 SLIST_INIT(cl); 3312 3313 /* Ensure that chunks are not already in use. */ 3314 for (i = 0; i < no_chunk; i++) { 3315 if (sr_chunk_in_use(sc, dt[i]) != BIOC_SDINVALID) { 3316 sr_meta_getdevname(sc, dt[i], devname, sizeof(devname)); 3317 sr_error(sc, "chunk %s already in use", devname); 3318 goto unwind; 3319 } 3320 } 3321 3322 sd->sd_meta_type = sr_meta_probe(sd, dt, no_chunk); 3323 if (sd->sd_meta_type == SR_META_F_INVALID) { 3324 sr_error(sc, "invalid metadata format"); 3325 goto unwind; 3326 } 3327 3328 if (sr_meta_attach(sd, no_chunk, bc->bc_flags & BIOC_SCFORCE)) 3329 goto unwind; 3330 3331 /* force the raid volume by clearing metadata region */ 3332 if (bc->bc_flags & BIOC_SCFORCE) { 3333 /* make sure disk isn't up and running */ 3334 if (sr_meta_read(sd)) 3335 if (sr_already_assembled(sd)) { 3336 uuid = sr_uuid_format( 3337 &sd->sd_meta->ssdi.ssd_uuid); 3338 sr_error(sc, "disk %s is currently in use; " 3339 "cannot force create", uuid); 3340 free(uuid, M_DEVBUF, 0); 3341 goto unwind; 3342 } 3343 3344 if (sr_meta_clear(sd)) { 3345 sr_error(sc, "failed to clear metadata"); 3346 goto unwind; 3347 } 3348 } 3349 3350 no_meta = sr_meta_read(sd); 3351 if (no_meta == -1) { 3352 3353 /* Corrupt metadata on one or more chunks. */ 3354 sr_error(sc, "one of the chunks has corrupt metadata; " 3355 "aborting assembly"); 3356 goto unwind; 3357 3358 } else if (no_meta == 0) { 3359 3360 /* Initialise volume and chunk metadata. */ 3361 sr_meta_init(sd, bc->bc_level, no_chunk); 3362 sd->sd_vol_status = BIOC_SVONLINE; 3363 sd->sd_meta_flags = bc->bc_flags & BIOC_SCNOAUTOASSEMBLE; 3364 if (sd->sd_create) { 3365 if ((i = sd->sd_create(sd, bc, no_chunk, 3366 sd->sd_vol.sv_chunk_minsz))) { 3367 rv = i; 3368 goto unwind; 3369 } 3370 } 3371 sr_meta_init_complete(sd); 3372 3373 DNPRINTF(SR_D_IOCTL, 3374 "%s: sr_ioctl_createraid: vol_size: %lld\n", 3375 DEVNAME(sc), sd->sd_meta->ssdi.ssd_size); 3376 3377 /* Warn if we've wasted chunk space due to coercing. */ 3378 if ((sd->sd_capabilities & SR_CAP_NON_COERCED) == 0 && 3379 sd->sd_vol.sv_chunk_minsz != sd->sd_vol.sv_chunk_maxsz) 3380 sr_warn(sc, "chunk sizes are not equal; up to %llu " 3381 "blocks wasted per chunk", 3382 sd->sd_vol.sv_chunk_maxsz - 3383 sd->sd_vol.sv_chunk_minsz); 3384 3385 } else { 3386 3387 /* Ensure metadata level matches requested assembly level. */ 3388 if (sd->sd_meta->ssdi.ssd_level != bc->bc_level) { 3389 sr_error(sc, "volume level does not match metadata " 3390 "level"); 3391 goto unwind; 3392 } 3393 3394 if (sr_already_assembled(sd)) { 3395 uuid = sr_uuid_format(&sd->sd_meta->ssdi.ssd_uuid); 3396 sr_error(sc, "disk %s already assembled", uuid); 3397 free(uuid, M_DEVBUF, 0); 3398 goto unwind; 3399 } 3400 3401 if (user == 0 && sd->sd_meta_flags & BIOC_SCNOAUTOASSEMBLE) { 3402 DNPRINTF(SR_D_META, "%s: disk not auto assembled from " 3403 "metadata\n", DEVNAME(sc)); 3404 goto unwind; 3405 } 3406 3407 if (no_meta != no_chunk) 3408 sr_warn(sc, "trying to bring up %s degraded", 3409 sd->sd_meta->ssd_devname); 3410 3411 if (sd->sd_meta->ssd_meta_flags & SR_META_DIRTY) 3412 sr_warn(sc, "%s was not shutdown properly", 3413 sd->sd_meta->ssd_devname); 3414 3415 SLIST_FOREACH(omi, &sd->sd_meta_opt, omi_link) 3416 if (sd->sd_meta_opt_handler == NULL || 3417 sd->sd_meta_opt_handler(sd, omi->omi_som) != 0) 3418 sr_meta_opt_handler(sd, omi->omi_som); 3419 3420 if (sd->sd_assemble) { 3421 if ((i = sd->sd_assemble(sd, bc, no_chunk, data))) { 3422 rv = i; 3423 goto unwind; 3424 } 3425 } 3426 3427 DNPRINTF(SR_D_META, "%s: disk assembled from metadata\n", 3428 DEVNAME(sc)); 3429 3430 } 3431 3432 /* Metadata MUST be fully populated by this point. */ 3433 TAILQ_INSERT_TAIL(&sc->sc_dis_list, sd, sd_link); 3434 3435 /* Allocate all resources. */ 3436 if ((rv = sd->sd_alloc_resources(sd))) 3437 goto unwind; 3438 3439 /* Adjust flags if necessary. */ 3440 if ((sd->sd_capabilities & SR_CAP_AUTO_ASSEMBLE) && 3441 (bc->bc_flags & BIOC_SCNOAUTOASSEMBLE) != 3442 (sd->sd_meta->ssdi.ssd_vol_flags & BIOC_SCNOAUTOASSEMBLE)) { 3443 sd->sd_meta->ssdi.ssd_vol_flags &= ~BIOC_SCNOAUTOASSEMBLE; 3444 sd->sd_meta->ssdi.ssd_vol_flags |= 3445 bc->bc_flags & BIOC_SCNOAUTOASSEMBLE; 3446 } 3447 3448 if (sd->sd_capabilities & SR_CAP_SYSTEM_DISK) { 3449 3450 /* Initialise volume state. */ 3451 sd->sd_set_vol_state(sd); 3452 if (sd->sd_vol_status == BIOC_SVOFFLINE) { 3453 sr_error(sc, "%s is offline, will not be brought " 3454 "online", sd->sd_meta->ssd_devname); 3455 goto unwind; 3456 } 3457 3458 /* Setup SCSI iopool. */ 3459 scsi_iopool_init(&sd->sd_iopool, sd, sr_wu_get, sr_wu_put); 3460 3461 /* 3462 * All checks passed - return ENXIO if volume cannot be created. 3463 */ 3464 rv = ENXIO; 3465 3466 /* 3467 * Find a free target. 3468 * 3469 * XXX: We reserve sd_target == 0 to indicate the 3470 * discipline is not linked into sc->sc_targets, so begin 3471 * the search with target = 1. 3472 */ 3473 for (target = 1; target < SR_MAX_LD; target++) 3474 if (sc->sc_targets[target] == NULL) 3475 break; 3476 if (target == SR_MAX_LD) { 3477 sr_error(sc, "no free target for %s", 3478 sd->sd_meta->ssd_devname); 3479 goto unwind; 3480 } 3481 3482 /* Clear sense data. */ 3483 bzero(&sd->sd_scsi_sense, sizeof(sd->sd_scsi_sense)); 3484 3485 /* Attach discipline and get midlayer to probe it. */ 3486 sd->sd_target = target; 3487 sc->sc_targets[target] = sd; 3488 if (scsi_probe_lun(sc->sc_scsibus, target, 0) != 0) { 3489 sr_error(sc, "scsi_probe_lun failed"); 3490 sc->sc_targets[target] = NULL; 3491 sd->sd_target = 0; 3492 goto unwind; 3493 } 3494 3495 link = scsi_get_link(sc->sc_scsibus, target, 0); 3496 dev = link->device_softc; 3497 DNPRINTF(SR_D_IOCTL, "%s: sr device added: %s at target %d\n", 3498 DEVNAME(sc), dev->dv_xname, sd->sd_target); 3499 3500 /* XXX - Count volumes, not targets. */ 3501 for (i = 0, vol = -1; i <= sd->sd_target; i++) 3502 if (sc->sc_targets[i]) 3503 vol++; 3504 3505 rv = 0; 3506 3507 if (sd->sd_meta->ssd_devname[0] != '\0' && 3508 strncmp(sd->sd_meta->ssd_devname, dev->dv_xname, 3509 sizeof(dev->dv_xname))) 3510 sr_warn(sc, "volume %s is roaming, it used to be %s, " 3511 "updating metadata", dev->dv_xname, 3512 sd->sd_meta->ssd_devname); 3513 3514 /* Populate remaining volume metadata. */ 3515 sd->sd_meta->ssdi.ssd_volid = vol; 3516 strlcpy(sd->sd_meta->ssd_devname, dev->dv_xname, 3517 sizeof(sd->sd_meta->ssd_devname)); 3518 3519 sr_info(sc, "%s volume attached as %s", 3520 sd->sd_name, sd->sd_meta->ssd_devname); 3521 3522 /* Update device name on any roaming chunks. */ 3523 sr_roam_chunks(sd); 3524 3525 #ifndef SMALL_KERNEL 3526 if (sr_sensors_create(sd)) 3527 sr_warn(sc, "unable to create sensor for %s", 3528 dev->dv_xname); 3529 #endif /* SMALL_KERNEL */ 3530 } else { 3531 /* This volume does not attach as a system disk. */ 3532 ch_entry = SLIST_FIRST(cl); /* XXX */ 3533 strlcpy(sd->sd_meta->ssd_devname, ch_entry->src_devname, 3534 sizeof(sd->sd_meta->ssd_devname)); 3535 3536 if (sd->sd_start_discipline(sd)) 3537 goto unwind; 3538 } 3539 3540 /* Save current metadata to disk. */ 3541 rv = sr_meta_save(sd, SR_META_DIRTY); 3542 3543 if (sd->sd_vol_status == BIOC_SVREBUILD) 3544 kthread_create_deferred(sr_rebuild_start, sd); 3545 3546 sd->sd_ready = 1; 3547 3548 return (rv); 3549 3550 unwind: 3551 sr_discipline_shutdown(sd, 0); 3552 3553 if (rv == EAGAIN) 3554 rv = 0; 3555 3556 return (rv); 3557 } 3558 3559 int 3560 sr_ioctl_deleteraid(struct sr_softc *sc, struct bioc_deleteraid *bd) 3561 { 3562 struct sr_discipline *sd; 3563 int rv = 1; 3564 3565 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_deleteraid %s\n", 3566 DEVNAME(sc), bd->bd_dev); 3567 3568 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 3569 if (!strncmp(sd->sd_meta->ssd_devname, bd->bd_dev, 3570 sizeof(sd->sd_meta->ssd_devname))) 3571 break; 3572 } 3573 if (sd == NULL) { 3574 sr_error(sc, "volume %s not found", bd->bd_dev); 3575 goto bad; 3576 } 3577 3578 sd->sd_deleted = 1; 3579 sd->sd_meta->ssdi.ssd_vol_flags = BIOC_SCNOAUTOASSEMBLE; 3580 sr_discipline_shutdown(sd, 1); 3581 3582 rv = 0; 3583 bad: 3584 return (rv); 3585 } 3586 3587 int 3588 sr_ioctl_discipline(struct sr_softc *sc, struct bioc_discipline *bd) 3589 { 3590 struct sr_discipline *sd; 3591 int rv = 1; 3592 3593 /* Dispatch a discipline specific ioctl. */ 3594 3595 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_discipline %s\n", DEVNAME(sc), 3596 bd->bd_dev); 3597 3598 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 3599 if (!strncmp(sd->sd_meta->ssd_devname, bd->bd_dev, 3600 sizeof(sd->sd_meta->ssd_devname))) 3601 break; 3602 } 3603 if (sd == NULL) { 3604 sr_error(sc, "volume %s not found", bd->bd_dev); 3605 goto bad; 3606 } 3607 3608 if (sd->sd_ioctl_handler) 3609 rv = sd->sd_ioctl_handler(sd, bd); 3610 3611 bad: 3612 return (rv); 3613 } 3614 3615 int 3616 sr_ioctl_installboot(struct sr_softc *sc, struct bioc_installboot *bb) 3617 { 3618 void *bootblk = NULL, *bootldr = NULL; 3619 struct sr_discipline *sd; 3620 struct sr_chunk *chunk; 3621 struct sr_meta_opt_item *omi; 3622 struct sr_meta_boot *sbm; 3623 struct disk *dk; 3624 u_int32_t bbs, bls; 3625 u_char duid[8]; 3626 int rv = EINVAL; 3627 int i; 3628 3629 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_installboot %s\n", DEVNAME(sc), 3630 bb->bb_dev); 3631 3632 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 3633 if (!strncmp(sd->sd_meta->ssd_devname, bb->bb_dev, 3634 sizeof(sd->sd_meta->ssd_devname))) 3635 break; 3636 } 3637 if (sd == NULL) { 3638 sr_error(sc, "volume %s not found", bb->bb_dev); 3639 goto done; 3640 } 3641 3642 bzero(duid, sizeof(duid)); 3643 TAILQ_FOREACH(dk, &disklist, dk_link) 3644 if (!strncmp(dk->dk_name, bb->bb_dev, sizeof(bb->bb_dev))) 3645 break; 3646 if (dk == NULL || dk->dk_label == NULL || 3647 (dk->dk_flags & DKF_LABELVALID) == 0 || 3648 bcmp(dk->dk_label->d_uid, &duid, sizeof(duid)) == 0) { 3649 sr_error(sc, "failed to get DUID for softraid volume"); 3650 goto done; 3651 } 3652 bcopy(dk->dk_label->d_uid, duid, sizeof(duid)); 3653 3654 /* Ensure that boot storage area is large enough. */ 3655 if (sd->sd_meta->ssd_data_offset < (SR_BOOT_OFFSET + SR_BOOT_SIZE)) { 3656 sr_error(sc, "insufficient boot storage"); 3657 goto done; 3658 } 3659 3660 if (bb->bb_bootblk_size > SR_BOOT_BLOCKS_SIZE * 512) 3661 goto done; 3662 3663 if (bb->bb_bootldr_size > SR_BOOT_LOADER_SIZE * 512) 3664 goto done; 3665 3666 /* Copy in boot block. */ 3667 bbs = howmany(bb->bb_bootblk_size, DEV_BSIZE) * DEV_BSIZE; 3668 bootblk = malloc(bbs, M_DEVBUF, M_WAITOK | M_ZERO); 3669 if (copyin(bb->bb_bootblk, bootblk, bb->bb_bootblk_size) != 0) 3670 goto done; 3671 3672 /* Copy in boot loader. */ 3673 bls = howmany(bb->bb_bootldr_size, DEV_BSIZE) * DEV_BSIZE; 3674 bootldr = malloc(bls, M_DEVBUF, M_WAITOK | M_ZERO); 3675 if (copyin(bb->bb_bootldr, bootldr, bb->bb_bootldr_size) != 0) 3676 goto done; 3677 3678 /* Create or update optional meta for bootable volumes. */ 3679 SLIST_FOREACH(omi, &sd->sd_meta_opt, omi_link) 3680 if (omi->omi_som->som_type == SR_OPT_BOOT) 3681 break; 3682 if (omi == NULL) { 3683 omi = malloc(sizeof(struct sr_meta_opt_item), M_DEVBUF, 3684 M_WAITOK | M_ZERO); 3685 omi->omi_som = malloc(sizeof(struct sr_meta_crypto), M_DEVBUF, 3686 M_WAITOK | M_ZERO); 3687 omi->omi_som->som_type = SR_OPT_BOOT; 3688 omi->omi_som->som_length = sizeof(struct sr_meta_boot); 3689 SLIST_INSERT_HEAD(&sd->sd_meta_opt, omi, omi_link); 3690 sd->sd_meta->ssdi.ssd_opt_no++; 3691 } 3692 sbm = (struct sr_meta_boot *)omi->omi_som; 3693 3694 bcopy(duid, sbm->sbm_root_duid, sizeof(sbm->sbm_root_duid)); 3695 bzero(&sbm->sbm_boot_duid, sizeof(sbm->sbm_boot_duid)); 3696 sbm->sbm_bootblk_size = bbs; 3697 sbm->sbm_bootldr_size = bls; 3698 3699 DNPRINTF(SR_D_IOCTL, "sr_ioctl_installboot: root duid is " 3700 "%02hhx%02hhx%02hhx%02hhx%02hhx%02hhx%02hhx\n", 3701 sbm->sbm_root_duid[0], sbm->sbm_root_duid[1], 3702 sbm->sbm_root_duid[2], sbm->sbm_root_duid[3], 3703 sbm->sbm_root_duid[4], sbm->sbm_root_duid[5], 3704 sbm->sbm_root_duid[6], sbm->sbm_root_duid[7]); 3705 3706 /* Save boot block and boot loader to each chunk. */ 3707 for (i = 0; i < sd->sd_meta->ssdi.ssd_chunk_no; i++) { 3708 3709 chunk = sd->sd_vol.sv_chunks[i]; 3710 if (chunk->src_meta.scm_status != BIOC_SDONLINE && 3711 chunk->src_meta.scm_status != BIOC_SDREBUILD) 3712 continue; 3713 3714 if (i < SR_MAX_BOOT_DISKS) 3715 bcopy(chunk->src_duid, &sbm->sbm_boot_duid[i], 3716 sizeof(sbm->sbm_boot_duid[i])); 3717 3718 /* Save boot blocks. */ 3719 DNPRINTF(SR_D_IOCTL, 3720 "sr_ioctl_installboot: saving boot block to %s " 3721 "(%u bytes)\n", chunk->src_devname, bbs); 3722 3723 if (sr_rw(sc, chunk->src_dev_mm, bootblk, bbs, 3724 SR_BOOT_BLOCKS_OFFSET, B_WRITE)) { 3725 sr_error(sc, "failed to write boot block", DEVNAME(sc)); 3726 goto done; 3727 } 3728 3729 /* Save boot loader.*/ 3730 DNPRINTF(SR_D_IOCTL, 3731 "sr_ioctl_installboot: saving boot loader to %s " 3732 "(%u bytes)\n", chunk->src_devname, bls); 3733 3734 if (sr_rw(sc, chunk->src_dev_mm, bootldr, bls, 3735 SR_BOOT_LOADER_OFFSET, B_WRITE)) { 3736 sr_error(sc, "failed to write boot loader"); 3737 goto done; 3738 } 3739 3740 } 3741 3742 /* XXX - Install boot block on disk - MD code. */ 3743 3744 /* Mark volume as bootable and save metadata. */ 3745 sd->sd_meta->ssdi.ssd_vol_flags |= BIOC_SCBOOTABLE; 3746 if (sr_meta_save(sd, SR_META_DIRTY)) { 3747 sr_error(sc, "could not save metadata to %s", 3748 chunk->src_devname); 3749 goto done; 3750 } 3751 3752 rv = 0; 3753 3754 done: 3755 if (bootblk) 3756 free(bootblk, M_DEVBUF, 0); 3757 if (bootldr) 3758 free(bootldr, M_DEVBUF, 0); 3759 3760 return (rv); 3761 } 3762 3763 void 3764 sr_chunks_unwind(struct sr_softc *sc, struct sr_chunk_head *cl) 3765 { 3766 struct sr_chunk *ch_entry, *ch_next; 3767 3768 DNPRINTF(SR_D_IOCTL, "%s: sr_chunks_unwind\n", DEVNAME(sc)); 3769 3770 if (!cl) 3771 return; 3772 3773 for (ch_entry = SLIST_FIRST(cl); ch_entry != NULL; ch_entry = ch_next) { 3774 ch_next = SLIST_NEXT(ch_entry, src_link); 3775 3776 DNPRINTF(SR_D_IOCTL, "%s: sr_chunks_unwind closing: %s\n", 3777 DEVNAME(sc), ch_entry->src_devname); 3778 if (ch_entry->src_vn) { 3779 /* 3780 * XXX - explicitly lock the vnode until we can resolve 3781 * the problem introduced by vnode aliasing... specfs 3782 * has no locking, whereas ufs/ffs does! 3783 */ 3784 vn_lock(ch_entry->src_vn, LK_EXCLUSIVE | 3785 LK_RETRY, curproc); 3786 VOP_CLOSE(ch_entry->src_vn, FREAD | FWRITE, NOCRED, 3787 curproc); 3788 vput(ch_entry->src_vn); 3789 } 3790 free(ch_entry, M_DEVBUF, 0); 3791 } 3792 SLIST_INIT(cl); 3793 } 3794 3795 void 3796 sr_discipline_free(struct sr_discipline *sd) 3797 { 3798 struct sr_softc *sc; 3799 struct sr_discipline *sdtmp1, *sdtmp2; 3800 struct sr_meta_opt_head *som; 3801 struct sr_meta_opt_item *omi, *omi_next; 3802 3803 if (!sd) 3804 return; 3805 3806 sc = sd->sd_sc; 3807 3808 DNPRINTF(SR_D_DIS, "%s: sr_discipline_free %s\n", 3809 DEVNAME(sc), 3810 sd->sd_meta ? sd->sd_meta->ssd_devname : "nodev"); 3811 if (sd->sd_free_resources) 3812 sd->sd_free_resources(sd); 3813 if (sd->sd_vol.sv_chunks) 3814 free(sd->sd_vol.sv_chunks, M_DEVBUF, 0); 3815 if (sd->sd_meta) 3816 free(sd->sd_meta, M_DEVBUF, 0); 3817 if (sd->sd_meta_foreign) 3818 free(sd->sd_meta_foreign, M_DEVBUF, 0); 3819 3820 som = &sd->sd_meta_opt; 3821 for (omi = SLIST_FIRST(som); omi != NULL; omi = omi_next) { 3822 omi_next = SLIST_NEXT(omi, omi_link); 3823 if (omi->omi_som) 3824 free(omi->omi_som, M_DEVBUF, 0); 3825 free(omi, M_DEVBUF, 0); 3826 } 3827 3828 if (sd->sd_target != 0) { 3829 KASSERT(sc->sc_targets[sd->sd_target] == sd); 3830 sc->sc_targets[sd->sd_target] = NULL; 3831 } 3832 3833 TAILQ_FOREACH_SAFE(sdtmp1, &sc->sc_dis_list, sd_link, sdtmp2) { 3834 if (sdtmp1 == sd) { 3835 TAILQ_REMOVE(&sc->sc_dis_list, sd, sd_link); 3836 break; 3837 } 3838 } 3839 3840 explicit_bzero(sd, sizeof *sd); 3841 free(sd, M_DEVBUF, 0); 3842 } 3843 3844 void 3845 sr_discipline_shutdown(struct sr_discipline *sd, int meta_save) 3846 { 3847 struct sr_softc *sc; 3848 int s; 3849 3850 if (!sd) 3851 return; 3852 sc = sd->sd_sc; 3853 3854 DNPRINTF(SR_D_DIS, "%s: sr_discipline_shutdown %s\n", DEVNAME(sc), 3855 sd->sd_meta ? sd->sd_meta->ssd_devname : "nodev"); 3856 3857 /* If rebuilding, abort rebuild and drain I/O. */ 3858 if (sd->sd_reb_active) { 3859 sd->sd_reb_abort = 1; 3860 while (sd->sd_reb_active) 3861 tsleep(sd, PWAIT, "sr_shutdown", 1); 3862 } 3863 3864 if (meta_save) 3865 sr_meta_save(sd, 0); 3866 3867 s = splbio(); 3868 3869 sd->sd_ready = 0; 3870 3871 /* make sure there isn't a sync pending and yield */ 3872 wakeup(sd); 3873 while (sd->sd_sync || sd->sd_must_flush) 3874 if (tsleep(&sd->sd_sync, MAXPRI, "sr_down", 60 * hz) == 3875 EWOULDBLOCK) 3876 break; 3877 3878 #ifndef SMALL_KERNEL 3879 sr_sensors_delete(sd); 3880 #endif /* SMALL_KERNEL */ 3881 3882 if (sd->sd_target != 0) 3883 scsi_detach_lun(sc->sc_scsibus, sd->sd_target, 0, DETACH_FORCE); 3884 3885 sr_chunks_unwind(sc, &sd->sd_vol.sv_chunk_list); 3886 3887 if (sd->sd_taskq) 3888 taskq_destroy(sd->sd_taskq); 3889 3890 sr_discipline_free(sd); 3891 3892 splx(s); 3893 } 3894 3895 int 3896 sr_discipline_init(struct sr_discipline *sd, int level) 3897 { 3898 int rv = 1; 3899 3900 /* Initialise discipline function pointers with defaults. */ 3901 sd->sd_alloc_resources = sr_alloc_resources; 3902 sd->sd_assemble = NULL; 3903 sd->sd_create = NULL; 3904 sd->sd_free_resources = sr_free_resources; 3905 sd->sd_ioctl_handler = NULL; 3906 sd->sd_openings = NULL; 3907 sd->sd_meta_opt_handler = NULL; 3908 sd->sd_rebuild = sr_rebuild; 3909 sd->sd_scsi_inquiry = sr_raid_inquiry; 3910 sd->sd_scsi_read_cap = sr_raid_read_cap; 3911 sd->sd_scsi_tur = sr_raid_tur; 3912 sd->sd_scsi_req_sense = sr_raid_request_sense; 3913 sd->sd_scsi_start_stop = sr_raid_start_stop; 3914 sd->sd_scsi_sync = sr_raid_sync; 3915 sd->sd_scsi_rw = NULL; 3916 sd->sd_scsi_intr = sr_raid_intr; 3917 sd->sd_scsi_wu_done = NULL; 3918 sd->sd_scsi_done = NULL; 3919 sd->sd_set_chunk_state = sr_set_chunk_state; 3920 sd->sd_set_vol_state = sr_set_vol_state; 3921 sd->sd_start_discipline = NULL; 3922 3923 task_set(&sd->sd_meta_save_task, sr_meta_save_callback, sd, NULL); 3924 task_set(&sd->sd_hotspare_rebuild_task, sr_hotspare_rebuild_callback, 3925 sd, NULL); 3926 3927 switch (level) { 3928 case 0: 3929 sr_raid0_discipline_init(sd); 3930 break; 3931 case 1: 3932 sr_raid1_discipline_init(sd); 3933 break; 3934 case 5: 3935 sr_raid5_discipline_init(sd); 3936 break; 3937 case 6: 3938 sr_raid6_discipline_init(sd); 3939 break; 3940 #ifdef CRYPTO 3941 case 'C': 3942 sr_crypto_discipline_init(sd); 3943 break; 3944 #endif 3945 case 'c': 3946 sr_concat_discipline_init(sd); 3947 break; 3948 default: 3949 goto bad; 3950 } 3951 3952 rv = 0; 3953 bad: 3954 return (rv); 3955 } 3956 3957 int 3958 sr_raid_inquiry(struct sr_workunit *wu) 3959 { 3960 struct sr_discipline *sd = wu->swu_dis; 3961 struct scsi_xfer *xs = wu->swu_xs; 3962 struct scsi_inquiry *cdb = (struct scsi_inquiry *)xs->cmd; 3963 struct scsi_inquiry_data inq; 3964 3965 DNPRINTF(SR_D_DIS, "%s: sr_raid_inquiry\n", DEVNAME(sd->sd_sc)); 3966 3967 if (xs->cmdlen != sizeof(*cdb)) 3968 return (EINVAL); 3969 3970 if (ISSET(cdb->flags, SI_EVPD)) 3971 return (EOPNOTSUPP); 3972 3973 bzero(&inq, sizeof(inq)); 3974 inq.device = T_DIRECT; 3975 inq.dev_qual2 = 0; 3976 inq.version = 2; 3977 inq.response_format = 2; 3978 inq.additional_length = 32; 3979 inq.flags |= SID_CmdQue; 3980 strlcpy(inq.vendor, sd->sd_meta->ssdi.ssd_vendor, 3981 sizeof(inq.vendor)); 3982 strlcpy(inq.product, sd->sd_meta->ssdi.ssd_product, 3983 sizeof(inq.product)); 3984 strlcpy(inq.revision, sd->sd_meta->ssdi.ssd_revision, 3985 sizeof(inq.revision)); 3986 sr_copy_internal_data(xs, &inq, sizeof(inq)); 3987 3988 return (0); 3989 } 3990 3991 int 3992 sr_raid_read_cap(struct sr_workunit *wu) 3993 { 3994 struct sr_discipline *sd = wu->swu_dis; 3995 struct scsi_xfer *xs = wu->swu_xs; 3996 struct scsi_read_cap_data rcd; 3997 struct scsi_read_cap_data_16 rcd16; 3998 daddr_t addr; 3999 int rv = 1; 4000 4001 DNPRINTF(SR_D_DIS, "%s: sr_raid_read_cap\n", DEVNAME(sd->sd_sc)); 4002 4003 addr = sd->sd_meta->ssdi.ssd_size - 1; 4004 if (xs->cmd->opcode == READ_CAPACITY) { 4005 bzero(&rcd, sizeof(rcd)); 4006 if (addr > 0xffffffffllu) 4007 _lto4b(0xffffffff, rcd.addr); 4008 else 4009 _lto4b(addr, rcd.addr); 4010 _lto4b(512, rcd.length); 4011 sr_copy_internal_data(xs, &rcd, sizeof(rcd)); 4012 rv = 0; 4013 } else if (xs->cmd->opcode == READ_CAPACITY_16) { 4014 bzero(&rcd16, sizeof(rcd16)); 4015 _lto8b(addr, rcd16.addr); 4016 _lto4b(512, rcd16.length); 4017 sr_copy_internal_data(xs, &rcd16, sizeof(rcd16)); 4018 rv = 0; 4019 } 4020 4021 return (rv); 4022 } 4023 4024 int 4025 sr_raid_tur(struct sr_workunit *wu) 4026 { 4027 struct sr_discipline *sd = wu->swu_dis; 4028 4029 DNPRINTF(SR_D_DIS, "%s: sr_raid_tur\n", DEVNAME(sd->sd_sc)); 4030 4031 if (sd->sd_vol_status == BIOC_SVOFFLINE) { 4032 sd->sd_scsi_sense.error_code = SSD_ERRCODE_CURRENT; 4033 sd->sd_scsi_sense.flags = SKEY_NOT_READY; 4034 sd->sd_scsi_sense.add_sense_code = 0x04; 4035 sd->sd_scsi_sense.add_sense_code_qual = 0x11; 4036 sd->sd_scsi_sense.extra_len = 4; 4037 return (1); 4038 } else if (sd->sd_vol_status == BIOC_SVINVALID) { 4039 sd->sd_scsi_sense.error_code = SSD_ERRCODE_CURRENT; 4040 sd->sd_scsi_sense.flags = SKEY_HARDWARE_ERROR; 4041 sd->sd_scsi_sense.add_sense_code = 0x05; 4042 sd->sd_scsi_sense.add_sense_code_qual = 0x00; 4043 sd->sd_scsi_sense.extra_len = 4; 4044 return (1); 4045 } 4046 4047 return (0); 4048 } 4049 4050 int 4051 sr_raid_request_sense(struct sr_workunit *wu) 4052 { 4053 struct sr_discipline *sd = wu->swu_dis; 4054 struct scsi_xfer *xs = wu->swu_xs; 4055 4056 DNPRINTF(SR_D_DIS, "%s: sr_raid_request_sense\n", 4057 DEVNAME(sd->sd_sc)); 4058 4059 /* use latest sense data */ 4060 bcopy(&sd->sd_scsi_sense, &xs->sense, sizeof(xs->sense)); 4061 4062 /* clear sense data */ 4063 bzero(&sd->sd_scsi_sense, sizeof(sd->sd_scsi_sense)); 4064 4065 return (0); 4066 } 4067 4068 int 4069 sr_raid_start_stop(struct sr_workunit *wu) 4070 { 4071 struct scsi_xfer *xs = wu->swu_xs; 4072 struct scsi_start_stop *ss = (struct scsi_start_stop *)xs->cmd; 4073 4074 DNPRINTF(SR_D_DIS, "%s: sr_raid_start_stop\n", 4075 DEVNAME(wu->swu_dis->sd_sc)); 4076 4077 if (!ss) 4078 return (1); 4079 4080 /* 4081 * do nothing! 4082 * a softraid discipline should always reflect correct status 4083 */ 4084 return (0); 4085 } 4086 4087 int 4088 sr_raid_sync(struct sr_workunit *wu) 4089 { 4090 struct sr_discipline *sd = wu->swu_dis; 4091 int s, rv = 0, ios; 4092 4093 DNPRINTF(SR_D_DIS, "%s: sr_raid_sync\n", DEVNAME(sd->sd_sc)); 4094 4095 /* when doing a fake sync don't count the wu */ 4096 ios = (wu->swu_flags & SR_WUF_FAKE) ? 0 : 1; 4097 4098 s = splbio(); 4099 sd->sd_sync = 1; 4100 while (sd->sd_wu_pending > ios) { 4101 if (tsleep(sd, PRIBIO, "sr_sync", 15 * hz) == EWOULDBLOCK) { 4102 DNPRINTF(SR_D_DIS, "%s: sr_raid_sync timeout\n", 4103 DEVNAME(sd->sd_sc)); 4104 rv = 1; 4105 break; 4106 } 4107 } 4108 sd->sd_sync = 0; 4109 splx(s); 4110 4111 wakeup(&sd->sd_sync); 4112 4113 return (rv); 4114 } 4115 4116 void 4117 sr_raid_intr(struct buf *bp) 4118 { 4119 struct sr_ccb *ccb = (struct sr_ccb *)bp; 4120 struct sr_workunit *wu = ccb->ccb_wu; 4121 #ifdef SR_DEBUG 4122 struct sr_discipline *sd = wu->swu_dis; 4123 struct scsi_xfer *xs = wu->swu_xs; 4124 #endif 4125 int s; 4126 4127 DNPRINTF(SR_D_INTR, "%s: %s %s intr bp %p xs %p\n", 4128 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, sd->sd_name, bp, xs); 4129 4130 s = splbio(); 4131 sr_ccb_done(ccb); 4132 sr_wu_done(wu); 4133 splx(s); 4134 } 4135 4136 void 4137 sr_schedule_wu(struct sr_workunit *wu) 4138 { 4139 struct sr_discipline *sd = wu->swu_dis; 4140 struct sr_workunit *wup; 4141 int s; 4142 4143 DNPRINTF(SR_D_WU, "sr_schedule_wu: schedule wu %p state %i " 4144 "flags 0x%x\n", wu, wu->swu_state, wu->swu_flags); 4145 4146 KASSERT(wu->swu_io_count > 0); 4147 4148 s = splbio(); 4149 4150 /* Construct the work unit, do not schedule it. */ 4151 if (wu->swu_state == SR_WU_CONSTRUCT) 4152 goto queued; 4153 4154 /* Deferred work unit being reconstructed, do not start. */ 4155 if (wu->swu_state == SR_WU_REQUEUE) 4156 goto queued; 4157 4158 /* Current work unit failed, restart. */ 4159 if (wu->swu_state == SR_WU_RESTART) 4160 goto start; 4161 4162 if (wu->swu_state != SR_WU_INPROGRESS) 4163 panic("sr_schedule_wu: work unit not in progress (state %i)\n", 4164 wu->swu_state); 4165 4166 /* Walk queue backwards and fill in collider if we have one. */ 4167 TAILQ_FOREACH_REVERSE(wup, &sd->sd_wu_pendq, sr_wu_list, swu_link) { 4168 if (wu->swu_blk_end < wup->swu_blk_start || 4169 wup->swu_blk_end < wu->swu_blk_start) 4170 continue; 4171 4172 /* Defer work unit due to LBA collision. */ 4173 DNPRINTF(SR_D_WU, "sr_schedule_wu: deferring work unit %p\n", 4174 wu); 4175 wu->swu_state = SR_WU_DEFERRED; 4176 while (wup->swu_collider) 4177 wup = wup->swu_collider; 4178 wup->swu_collider = wu; 4179 TAILQ_INSERT_TAIL(&sd->sd_wu_defq, wu, swu_link); 4180 sd->sd_wu_collisions++; 4181 goto queued; 4182 } 4183 4184 start: 4185 sr_raid_startwu(wu); 4186 4187 queued: 4188 splx(s); 4189 } 4190 4191 void 4192 sr_raid_startwu(struct sr_workunit *wu) 4193 { 4194 struct sr_discipline *sd = wu->swu_dis; 4195 struct sr_ccb *ccb; 4196 4197 DNPRINTF(SR_D_WU, "sr_raid_startwu: start wu %p\n", wu); 4198 4199 splassert(IPL_BIO); 4200 4201 if (wu->swu_state == SR_WU_DEFERRED) { 4202 TAILQ_REMOVE(&sd->sd_wu_defq, wu, swu_link); 4203 wu->swu_state = SR_WU_INPROGRESS; 4204 } 4205 4206 if (wu->swu_state != SR_WU_RESTART) 4207 TAILQ_INSERT_TAIL(&sd->sd_wu_pendq, wu, swu_link); 4208 4209 /* Start all of the individual I/Os. */ 4210 if (wu->swu_cb_active == 1) 4211 panic("%s: sr_startwu_callback", DEVNAME(sd->sd_sc)); 4212 wu->swu_cb_active = 1; 4213 4214 TAILQ_FOREACH(ccb, &wu->swu_ccb, ccb_link) 4215 VOP_STRATEGY(&ccb->ccb_buf); 4216 4217 wu->swu_cb_active = 0; 4218 } 4219 4220 void 4221 sr_raid_recreate_wu(struct sr_workunit *wu) 4222 { 4223 struct sr_discipline *sd = wu->swu_dis; 4224 struct sr_workunit *wup = wu; 4225 4226 /* 4227 * Recreate a work unit by releasing the associated CCBs and reissuing 4228 * the SCSI I/O request. This process is then repeated for all of the 4229 * colliding work units. 4230 */ 4231 do { 4232 sr_wu_release_ccbs(wup); 4233 4234 wup->swu_state = SR_WU_REQUEUE; 4235 if (sd->sd_scsi_rw(wup)) 4236 panic("could not requeue I/O"); 4237 4238 wup = wup->swu_collider; 4239 } while (wup); 4240 } 4241 4242 int 4243 sr_alloc_resources(struct sr_discipline *sd) 4244 { 4245 if (sr_wu_alloc(sd, sizeof(struct sr_workunit))) { 4246 sr_error(sd->sd_sc, "unable to allocate work units"); 4247 return (ENOMEM); 4248 } 4249 if (sr_ccb_alloc(sd)) { 4250 sr_error(sd->sd_sc, "unable to allocate ccbs"); 4251 return (ENOMEM); 4252 } 4253 4254 return (0); 4255 } 4256 4257 void 4258 sr_free_resources(struct sr_discipline *sd) 4259 { 4260 sr_wu_free(sd); 4261 sr_ccb_free(sd); 4262 } 4263 4264 void 4265 sr_set_chunk_state(struct sr_discipline *sd, int c, int new_state) 4266 { 4267 int old_state, s; 4268 4269 DNPRINTF(SR_D_STATE, "%s: %s: %s: sr_set_chunk_state %d -> %d\n", 4270 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, 4271 sd->sd_vol.sv_chunks[c]->src_meta.scmi.scm_devname, c, new_state); 4272 4273 /* ok to go to splbio since this only happens in error path */ 4274 s = splbio(); 4275 old_state = sd->sd_vol.sv_chunks[c]->src_meta.scm_status; 4276 4277 /* multiple IOs to the same chunk that fail will come through here */ 4278 if (old_state == new_state) 4279 goto done; 4280 4281 switch (old_state) { 4282 case BIOC_SDONLINE: 4283 if (new_state == BIOC_SDOFFLINE) 4284 break; 4285 else 4286 goto die; 4287 break; 4288 4289 case BIOC_SDOFFLINE: 4290 goto die; 4291 4292 default: 4293 die: 4294 splx(s); /* XXX */ 4295 panic("%s: %s: %s: invalid chunk state transition " 4296 "%d -> %d\n", DEVNAME(sd->sd_sc), 4297 sd->sd_meta->ssd_devname, 4298 sd->sd_vol.sv_chunks[c]->src_meta.scmi.scm_devname, 4299 old_state, new_state); 4300 /* NOTREACHED */ 4301 } 4302 4303 sd->sd_vol.sv_chunks[c]->src_meta.scm_status = new_state; 4304 sd->sd_set_vol_state(sd); 4305 4306 sd->sd_must_flush = 1; 4307 task_add(systq, &sd->sd_meta_save_task); 4308 done: 4309 splx(s); 4310 } 4311 4312 void 4313 sr_set_vol_state(struct sr_discipline *sd) 4314 { 4315 int states[SR_MAX_STATES]; 4316 int new_state, i, s, nd; 4317 int old_state = sd->sd_vol_status; 4318 4319 DNPRINTF(SR_D_STATE, "%s: %s: sr_set_vol_state\n", 4320 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname); 4321 4322 nd = sd->sd_meta->ssdi.ssd_chunk_no; 4323 4324 for (i = 0; i < SR_MAX_STATES; i++) 4325 states[i] = 0; 4326 4327 for (i = 0; i < nd; i++) { 4328 s = sd->sd_vol.sv_chunks[i]->src_meta.scm_status; 4329 if (s >= SR_MAX_STATES) 4330 panic("%s: %s: %s: invalid chunk state", 4331 DEVNAME(sd->sd_sc), 4332 sd->sd_meta->ssd_devname, 4333 sd->sd_vol.sv_chunks[i]->src_meta.scmi.scm_devname); 4334 states[s]++; 4335 } 4336 4337 if (states[BIOC_SDONLINE] == nd) 4338 new_state = BIOC_SVONLINE; 4339 else 4340 new_state = BIOC_SVOFFLINE; 4341 4342 DNPRINTF(SR_D_STATE, "%s: %s: sr_set_vol_state %d -> %d\n", 4343 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, 4344 old_state, new_state); 4345 4346 switch (old_state) { 4347 case BIOC_SVONLINE: 4348 if (new_state == BIOC_SVOFFLINE || new_state == BIOC_SVONLINE) 4349 break; 4350 else 4351 goto die; 4352 break; 4353 4354 case BIOC_SVOFFLINE: 4355 /* XXX this might be a little too much */ 4356 goto die; 4357 4358 default: 4359 die: 4360 panic("%s: %s: invalid volume state transition " 4361 "%d -> %d\n", DEVNAME(sd->sd_sc), 4362 sd->sd_meta->ssd_devname, 4363 old_state, new_state); 4364 /* NOTREACHED */ 4365 } 4366 4367 sd->sd_vol_status = new_state; 4368 } 4369 4370 void * 4371 sr_block_get(struct sr_discipline *sd, int length) 4372 { 4373 return dma_alloc(length, PR_NOWAIT | PR_ZERO); 4374 } 4375 4376 void 4377 sr_block_put(struct sr_discipline *sd, void *ptr, int length) 4378 { 4379 dma_free(ptr, length); 4380 } 4381 4382 void 4383 sr_checksum_print(u_int8_t *md5) 4384 { 4385 int i; 4386 4387 for (i = 0; i < MD5_DIGEST_LENGTH; i++) 4388 printf("%02x", md5[i]); 4389 } 4390 4391 void 4392 sr_checksum(struct sr_softc *sc, void *src, void *md5, u_int32_t len) 4393 { 4394 MD5_CTX ctx; 4395 4396 DNPRINTF(SR_D_MISC, "%s: sr_checksum(%p %p %d)\n", DEVNAME(sc), src, 4397 md5, len); 4398 4399 MD5Init(&ctx); 4400 MD5Update(&ctx, src, len); 4401 MD5Final(md5, &ctx); 4402 } 4403 4404 void 4405 sr_uuid_generate(struct sr_uuid *uuid) 4406 { 4407 arc4random_buf(uuid->sui_id, sizeof(uuid->sui_id)); 4408 /* UUID version 4: random */ 4409 uuid->sui_id[6] &= 0x0f; 4410 uuid->sui_id[6] |= 0x40; 4411 /* RFC4122 variant */ 4412 uuid->sui_id[8] &= 0x3f; 4413 uuid->sui_id[8] |= 0x80; 4414 } 4415 4416 char * 4417 sr_uuid_format(struct sr_uuid *uuid) 4418 { 4419 char *uuidstr; 4420 4421 uuidstr = malloc(37, M_DEVBUF, M_WAITOK); 4422 4423 snprintf(uuidstr, 37, 4424 "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-" 4425 "%02x%02x%02x%02x%02x%02x", 4426 uuid->sui_id[0], uuid->sui_id[1], 4427 uuid->sui_id[2], uuid->sui_id[3], 4428 uuid->sui_id[4], uuid->sui_id[5], 4429 uuid->sui_id[6], uuid->sui_id[7], 4430 uuid->sui_id[8], uuid->sui_id[9], 4431 uuid->sui_id[10], uuid->sui_id[11], 4432 uuid->sui_id[12], uuid->sui_id[13], 4433 uuid->sui_id[14], uuid->sui_id[15]); 4434 4435 return uuidstr; 4436 } 4437 4438 void 4439 sr_uuid_print(struct sr_uuid *uuid, int cr) 4440 { 4441 char *uuidstr; 4442 4443 uuidstr = sr_uuid_format(uuid); 4444 printf("%s%s", uuidstr, (cr ? "\n" : "")); 4445 free(uuidstr, M_DEVBUF, 37); 4446 } 4447 4448 int 4449 sr_already_assembled(struct sr_discipline *sd) 4450 { 4451 struct sr_softc *sc = sd->sd_sc; 4452 struct sr_discipline *sdtmp; 4453 4454 TAILQ_FOREACH(sdtmp, &sc->sc_dis_list, sd_link) { 4455 if (!bcmp(&sd->sd_meta->ssdi.ssd_uuid, 4456 &sdtmp->sd_meta->ssdi.ssd_uuid, 4457 sizeof(sd->sd_meta->ssdi.ssd_uuid))) 4458 return (1); 4459 } 4460 4461 return (0); 4462 } 4463 4464 int32_t 4465 sr_validate_stripsize(u_int32_t b) 4466 { 4467 int s = 0; 4468 4469 if (b % 512) 4470 return (-1); 4471 4472 while ((b & 1) == 0) { 4473 b >>= 1; 4474 s++; 4475 } 4476 4477 /* only multiple of twos */ 4478 b >>= 1; 4479 if (b) 4480 return(-1); 4481 4482 return (s); 4483 } 4484 4485 void 4486 sr_shutdown(void) 4487 { 4488 struct sr_softc *sc = softraid0; 4489 struct sr_discipline *sd; 4490 4491 DNPRINTF(SR_D_MISC, "%s: sr_shutdown\n", DEVNAME(sc)); 4492 4493 /* 4494 * Since softraid is not under mainbus, we have to explicitly 4495 * notify its children that the power is going down, so they 4496 * can execute their shutdown hooks. 4497 */ 4498 config_suspend((struct device *)sc, DVACT_POWERDOWN); 4499 4500 /* Shutdown disciplines in reverse attach order. */ 4501 while ((sd = TAILQ_LAST(&sc->sc_dis_list, sr_discipline_list)) != NULL) 4502 sr_discipline_shutdown(sd, 1); 4503 } 4504 4505 int 4506 sr_validate_io(struct sr_workunit *wu, daddr_t *blk, char *func) 4507 { 4508 struct sr_discipline *sd = wu->swu_dis; 4509 struct scsi_xfer *xs = wu->swu_xs; 4510 int rv = 1; 4511 4512 DNPRINTF(SR_D_DIS, "%s: %s 0x%02x\n", DEVNAME(sd->sd_sc), func, 4513 xs->cmd->opcode); 4514 4515 if (sd->sd_meta->ssd_data_offset == 0) 4516 panic("invalid data offset"); 4517 4518 if (sd->sd_vol_status == BIOC_SVOFFLINE) { 4519 DNPRINTF(SR_D_DIS, "%s: %s device offline\n", 4520 DEVNAME(sd->sd_sc), func); 4521 goto bad; 4522 } 4523 4524 if (xs->datalen == 0) { 4525 printf("%s: %s: illegal block count for %s\n", 4526 DEVNAME(sd->sd_sc), func, sd->sd_meta->ssd_devname); 4527 goto bad; 4528 } 4529 4530 if (xs->cmdlen == 10) 4531 *blk = _4btol(((struct scsi_rw_big *)xs->cmd)->addr); 4532 else if (xs->cmdlen == 16) 4533 *blk = _8btol(((struct scsi_rw_16 *)xs->cmd)->addr); 4534 else if (xs->cmdlen == 6) 4535 *blk = _3btol(((struct scsi_rw *)xs->cmd)->addr); 4536 else { 4537 printf("%s: %s: illegal cmdlen for %s\n", 4538 DEVNAME(sd->sd_sc), func, sd->sd_meta->ssd_devname); 4539 goto bad; 4540 } 4541 4542 wu->swu_blk_start = *blk; 4543 wu->swu_blk_end = *blk + (xs->datalen >> DEV_BSHIFT) - 1; 4544 4545 if (wu->swu_blk_end > sd->sd_meta->ssdi.ssd_size) { 4546 DNPRINTF(SR_D_DIS, "%s: %s out of bounds start: %lld " 4547 "end: %lld length: %d\n", 4548 DEVNAME(sd->sd_sc), func, (long long)wu->swu_blk_start, 4549 (long long)wu->swu_blk_end, xs->datalen); 4550 4551 sd->sd_scsi_sense.error_code = SSD_ERRCODE_CURRENT | 4552 SSD_ERRCODE_VALID; 4553 sd->sd_scsi_sense.flags = SKEY_ILLEGAL_REQUEST; 4554 sd->sd_scsi_sense.add_sense_code = 0x21; 4555 sd->sd_scsi_sense.add_sense_code_qual = 0x00; 4556 sd->sd_scsi_sense.extra_len = 4; 4557 goto bad; 4558 } 4559 4560 rv = 0; 4561 bad: 4562 return (rv); 4563 } 4564 4565 void 4566 sr_rebuild_start(void *arg) 4567 { 4568 struct sr_discipline *sd = arg; 4569 struct sr_softc *sc = sd->sd_sc; 4570 4571 DNPRINTF(SR_D_REBUILD, "%s: %s starting rebuild thread\n", 4572 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname); 4573 4574 if (kthread_create(sr_rebuild_thread, sd, &sd->sd_background_proc, 4575 DEVNAME(sc)) != 0) 4576 printf("%s: unable to start background operation\n", 4577 DEVNAME(sc)); 4578 } 4579 4580 void 4581 sr_rebuild_thread(void *arg) 4582 { 4583 struct sr_discipline *sd = arg; 4584 4585 DNPRINTF(SR_D_REBUILD, "%s: %s rebuild thread started\n", 4586 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname); 4587 4588 sd->sd_reb_active = 1; 4589 sd->sd_rebuild(sd); 4590 sd->sd_reb_active = 0; 4591 4592 kthread_exit(0); 4593 } 4594 4595 void 4596 sr_rebuild(struct sr_discipline *sd) 4597 { 4598 struct sr_softc *sc = sd->sd_sc; 4599 daddr_t whole_blk, partial_blk, blk, sz, lba; 4600 daddr_t psz, rb, restart; 4601 struct sr_workunit *wu_r, *wu_w; 4602 struct scsi_xfer xs_r, xs_w; 4603 struct scsi_rw_16 *cr, *cw; 4604 int c, s, slept, percent = 0, old_percent = -1; 4605 u_int8_t *buf; 4606 4607 whole_blk = sd->sd_meta->ssdi.ssd_size / SR_REBUILD_IO_SIZE; 4608 partial_blk = sd->sd_meta->ssdi.ssd_size % SR_REBUILD_IO_SIZE; 4609 4610 restart = sd->sd_meta->ssd_rebuild / SR_REBUILD_IO_SIZE; 4611 if (restart > whole_blk) { 4612 printf("%s: bogus rebuild restart offset, starting from 0\n", 4613 DEVNAME(sc)); 4614 restart = 0; 4615 } 4616 if (restart) { 4617 /* 4618 * XXX there is a hole here; there is a posibility that we 4619 * had a restart however the chunk that was supposed to 4620 * be rebuilt is no longer valid; we can reach this situation 4621 * when a rebuild is in progress and the box crashes and 4622 * on reboot the rebuild chunk is different (like zero'd or 4623 * replaced). We need to check the uuid of the chunk that is 4624 * being rebuilt to assert this. 4625 */ 4626 psz = sd->sd_meta->ssdi.ssd_size; 4627 rb = sd->sd_meta->ssd_rebuild; 4628 if (rb > 0) 4629 percent = 100 - ((psz * 100 - rb * 100) / psz) - 1; 4630 else 4631 percent = 0; 4632 printf("%s: resuming rebuild on %s at %d%%\n", 4633 DEVNAME(sc), sd->sd_meta->ssd_devname, percent); 4634 } 4635 4636 /* currently this is 64k therefore we can use dma_alloc */ 4637 buf = dma_alloc(SR_REBUILD_IO_SIZE << DEV_BSHIFT, PR_WAITOK); 4638 for (blk = restart; blk <= whole_blk; blk++) { 4639 lba = blk * SR_REBUILD_IO_SIZE; 4640 sz = SR_REBUILD_IO_SIZE; 4641 if (blk == whole_blk) { 4642 if (partial_blk == 0) 4643 break; 4644 sz = partial_blk; 4645 } 4646 4647 /* get some wu */ 4648 wu_r = sr_scsi_wu_get(sd, 0); 4649 wu_w = sr_scsi_wu_get(sd, 0); 4650 4651 DNPRINTF(SR_D_REBUILD, "%s: %s rebuild wu_r %p, wu_w %p\n", 4652 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, wu_r, wu_w); 4653 4654 /* setup read io */ 4655 bzero(&xs_r, sizeof xs_r); 4656 xs_r.error = XS_NOERROR; 4657 xs_r.flags = SCSI_DATA_IN; 4658 xs_r.datalen = sz << DEV_BSHIFT; 4659 xs_r.data = buf; 4660 xs_r.cmdlen = sizeof(*cr); 4661 xs_r.cmd = &xs_r.cmdstore; 4662 cr = (struct scsi_rw_16 *)xs_r.cmd; 4663 cr->opcode = READ_16; 4664 _lto4b(sz, cr->length); 4665 _lto8b(lba, cr->addr); 4666 wu_r->swu_state = SR_WU_CONSTRUCT; 4667 wu_r->swu_flags |= SR_WUF_REBUILD; 4668 wu_r->swu_xs = &xs_r; 4669 if (sd->sd_scsi_rw(wu_r)) { 4670 printf("%s: could not create read io\n", 4671 DEVNAME(sc)); 4672 goto fail; 4673 } 4674 4675 /* setup write io */ 4676 bzero(&xs_w, sizeof xs_w); 4677 xs_w.error = XS_NOERROR; 4678 xs_w.flags = SCSI_DATA_OUT; 4679 xs_w.datalen = sz << DEV_BSHIFT; 4680 xs_w.data = buf; 4681 xs_w.cmdlen = sizeof(*cw); 4682 xs_w.cmd = &xs_w.cmdstore; 4683 cw = (struct scsi_rw_16 *)xs_w.cmd; 4684 cw->opcode = WRITE_16; 4685 _lto4b(sz, cw->length); 4686 _lto8b(lba, cw->addr); 4687 wu_w->swu_state = SR_WU_CONSTRUCT; 4688 wu_w->swu_flags |= SR_WUF_REBUILD | SR_WUF_WAKEUP; 4689 wu_w->swu_xs = &xs_w; 4690 if (sd->sd_scsi_rw(wu_w)) { 4691 printf("%s: could not create write io\n", 4692 DEVNAME(sc)); 4693 goto fail; 4694 } 4695 4696 /* 4697 * collide with the read io so that we get automatically 4698 * started when the read is done 4699 */ 4700 wu_w->swu_state = SR_WU_DEFERRED; 4701 wu_r->swu_collider = wu_w; 4702 s = splbio(); 4703 TAILQ_INSERT_TAIL(&sd->sd_wu_defq, wu_w, swu_link); 4704 splx(s); 4705 4706 DNPRINTF(SR_D_REBUILD, "%s: %s rebuild scheduling wu_r %p\n", 4707 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, wu_r); 4708 4709 wu_r->swu_state = SR_WU_INPROGRESS; 4710 sr_schedule_wu(wu_r); 4711 4712 /* wait for write completion */ 4713 slept = 0; 4714 while ((wu_w->swu_flags & SR_WUF_REBUILDIOCOMP) == 0) { 4715 tsleep(wu_w, PRIBIO, "sr_rebuild", 0); 4716 slept = 1; 4717 } 4718 /* yield if we didn't sleep */ 4719 if (slept == 0) 4720 tsleep(sc, PWAIT, "sr_yield", 1); 4721 4722 sr_scsi_wu_put(sd, wu_r); 4723 sr_scsi_wu_put(sd, wu_w); 4724 4725 sd->sd_meta->ssd_rebuild = lba; 4726 4727 /* XXX - this should be based on size, not percentage. */ 4728 /* save metadata every percent */ 4729 psz = sd->sd_meta->ssdi.ssd_size; 4730 rb = sd->sd_meta->ssd_rebuild; 4731 if (rb > 0) 4732 percent = 100 - ((psz * 100 - rb * 100) / psz) - 1; 4733 else 4734 percent = 0; 4735 if (percent != old_percent && blk != whole_blk) { 4736 if (sr_meta_save(sd, SR_META_DIRTY)) 4737 printf("%s: could not save metadata to %s\n", 4738 DEVNAME(sc), sd->sd_meta->ssd_devname); 4739 old_percent = percent; 4740 } 4741 4742 if (sd->sd_reb_abort) 4743 goto abort; 4744 } 4745 4746 /* all done */ 4747 sd->sd_meta->ssd_rebuild = 0; 4748 for (c = 0; c < sd->sd_meta->ssdi.ssd_chunk_no; c++) { 4749 if (sd->sd_vol.sv_chunks[c]->src_meta.scm_status == 4750 BIOC_SDREBUILD) { 4751 sd->sd_set_chunk_state(sd, c, BIOC_SDONLINE); 4752 break; 4753 } 4754 } 4755 4756 abort: 4757 if (sr_meta_save(sd, SR_META_DIRTY)) 4758 printf("%s: could not save metadata to %s\n", 4759 DEVNAME(sc), sd->sd_meta->ssd_devname); 4760 fail: 4761 dma_free(buf, SR_REBUILD_IO_SIZE << DEV_BSHIFT); 4762 } 4763 4764 #ifndef SMALL_KERNEL 4765 int 4766 sr_sensors_create(struct sr_discipline *sd) 4767 { 4768 struct sr_softc *sc = sd->sd_sc; 4769 int rv = 1; 4770 4771 DNPRINTF(SR_D_STATE, "%s: %s: sr_sensors_create\n", 4772 DEVNAME(sc), sd->sd_meta->ssd_devname); 4773 4774 sd->sd_vol.sv_sensor.type = SENSOR_DRIVE; 4775 sd->sd_vol.sv_sensor.status = SENSOR_S_UNKNOWN; 4776 strlcpy(sd->sd_vol.sv_sensor.desc, sd->sd_meta->ssd_devname, 4777 sizeof(sd->sd_vol.sv_sensor.desc)); 4778 4779 sensor_attach(&sc->sc_sensordev, &sd->sd_vol.sv_sensor); 4780 sd->sd_vol.sv_sensor_attached = 1; 4781 4782 if (sc->sc_sensor_task == NULL) { 4783 sc->sc_sensor_task = sensor_task_register(sc, 4784 sr_sensors_refresh, 10); 4785 if (sc->sc_sensor_task == NULL) 4786 goto bad; 4787 } 4788 4789 rv = 0; 4790 bad: 4791 return (rv); 4792 } 4793 4794 void 4795 sr_sensors_delete(struct sr_discipline *sd) 4796 { 4797 DNPRINTF(SR_D_STATE, "%s: sr_sensors_delete\n", DEVNAME(sd->sd_sc)); 4798 4799 if (sd->sd_vol.sv_sensor_attached) 4800 sensor_detach(&sd->sd_sc->sc_sensordev, &sd->sd_vol.sv_sensor); 4801 } 4802 4803 void 4804 sr_sensors_refresh(void *arg) 4805 { 4806 struct sr_softc *sc = arg; 4807 struct sr_volume *sv; 4808 struct sr_discipline *sd; 4809 4810 DNPRINTF(SR_D_STATE, "%s: sr_sensors_refresh\n", DEVNAME(sc)); 4811 4812 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 4813 sv = &sd->sd_vol; 4814 4815 switch(sd->sd_vol_status) { 4816 case BIOC_SVOFFLINE: 4817 sv->sv_sensor.value = SENSOR_DRIVE_FAIL; 4818 sv->sv_sensor.status = SENSOR_S_CRIT; 4819 break; 4820 4821 case BIOC_SVDEGRADED: 4822 sv->sv_sensor.value = SENSOR_DRIVE_PFAIL; 4823 sv->sv_sensor.status = SENSOR_S_WARN; 4824 break; 4825 4826 case BIOC_SVSCRUB: 4827 case BIOC_SVONLINE: 4828 sv->sv_sensor.value = SENSOR_DRIVE_ONLINE; 4829 sv->sv_sensor.status = SENSOR_S_OK; 4830 break; 4831 4832 default: 4833 sv->sv_sensor.value = 0; /* unknown */ 4834 sv->sv_sensor.status = SENSOR_S_UNKNOWN; 4835 } 4836 } 4837 } 4838 #endif /* SMALL_KERNEL */ 4839 4840 #ifdef SR_FANCY_STATS 4841 void sr_print_stats(void); 4842 4843 void 4844 sr_print_stats(void) 4845 { 4846 struct sr_softc *sc = softraid0; 4847 struct sr_discipline *sd; 4848 4849 if (sc == NULL) { 4850 printf("no softraid softc found\n"); 4851 return; 4852 } 4853 4854 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 4855 printf("%s: ios pending %d, collisions %llu\n", 4856 sd->sd_meta->ssd_devname, 4857 sd->sd_wu_pending, 4858 sd->sd_wu_collisions); 4859 } 4860 } 4861 #endif /* SR_FANCY_STATS */ 4862 4863 #ifdef SR_DEBUG 4864 void 4865 sr_meta_print(struct sr_metadata *m) 4866 { 4867 int i; 4868 struct sr_meta_chunk *mc; 4869 struct sr_meta_opt_hdr *omh; 4870 4871 if (!(sr_debug & SR_D_META)) 4872 return; 4873 4874 printf("\tssd_magic 0x%llx\n", m->ssdi.ssd_magic); 4875 printf("\tssd_version %d\n", m->ssdi.ssd_version); 4876 printf("\tssd_vol_flags 0x%x\n", m->ssdi.ssd_vol_flags); 4877 printf("\tssd_uuid "); 4878 sr_uuid_print(&m->ssdi.ssd_uuid, 1); 4879 printf("\tssd_chunk_no %d\n", m->ssdi.ssd_chunk_no); 4880 printf("\tssd_chunk_id %d\n", m->ssdi.ssd_chunk_id); 4881 printf("\tssd_opt_no %d\n", m->ssdi.ssd_opt_no); 4882 printf("\tssd_volid %d\n", m->ssdi.ssd_volid); 4883 printf("\tssd_level %d\n", m->ssdi.ssd_level); 4884 printf("\tssd_size %lld\n", m->ssdi.ssd_size); 4885 printf("\tssd_devname %s\n", m->ssd_devname); 4886 printf("\tssd_vendor %s\n", m->ssdi.ssd_vendor); 4887 printf("\tssd_product %s\n", m->ssdi.ssd_product); 4888 printf("\tssd_revision %s\n", m->ssdi.ssd_revision); 4889 printf("\tssd_strip_size %d\n", m->ssdi.ssd_strip_size); 4890 printf("\tssd_checksum "); 4891 sr_checksum_print(m->ssd_checksum); 4892 printf("\n"); 4893 printf("\tssd_meta_flags 0x%x\n", m->ssd_meta_flags); 4894 printf("\tssd_ondisk %llu\n", m->ssd_ondisk); 4895 4896 mc = (struct sr_meta_chunk *)(m + 1); 4897 for (i = 0; i < m->ssdi.ssd_chunk_no; i++, mc++) { 4898 printf("\t\tscm_volid %d\n", mc->scmi.scm_volid); 4899 printf("\t\tscm_chunk_id %d\n", mc->scmi.scm_chunk_id); 4900 printf("\t\tscm_devname %s\n", mc->scmi.scm_devname); 4901 printf("\t\tscm_size %lld\n", mc->scmi.scm_size); 4902 printf("\t\tscm_coerced_size %lld\n",mc->scmi.scm_coerced_size); 4903 printf("\t\tscm_uuid "); 4904 sr_uuid_print(&mc->scmi.scm_uuid, 1); 4905 printf("\t\tscm_checksum "); 4906 sr_checksum_print(mc->scm_checksum); 4907 printf("\n"); 4908 printf("\t\tscm_status %d\n", mc->scm_status); 4909 } 4910 4911 omh = (struct sr_meta_opt_hdr *)((u_int8_t *)(m + 1) + 4912 sizeof(struct sr_meta_chunk) * m->ssdi.ssd_chunk_no); 4913 for (i = 0; i < m->ssdi.ssd_opt_no; i++) { 4914 printf("\t\t\tsom_type %d\n", omh->som_type); 4915 printf("\t\t\tsom_checksum "); 4916 sr_checksum_print(omh->som_checksum); 4917 printf("\n"); 4918 omh = (struct sr_meta_opt_hdr *)((void *)omh + 4919 omh->som_length); 4920 } 4921 } 4922 4923 void 4924 sr_dump_block(void *blk, int len) 4925 { 4926 uint8_t *b = blk; 4927 int i, j, c; 4928 4929 for (i = 0; i < len; i += 16) { 4930 for (j = 0; j < 16; j++) 4931 printf("%.2x ", b[i + j]); 4932 printf(" "); 4933 for (j = 0; j < 16; j++) { 4934 c = b[i + j]; 4935 if (c < ' ' || c > 'z' || i + j > len) 4936 c = '.'; 4937 printf("%c", c); 4938 } 4939 printf("\n"); 4940 } 4941 } 4942 4943 void 4944 sr_dump_mem(u_int8_t *p, int len) 4945 { 4946 int i; 4947 4948 for (i = 0; i < len; i++) 4949 printf("%02x ", *p++); 4950 printf("\n"); 4951 } 4952 4953 #endif /* SR_DEBUG */ 4954 4955 #ifdef HIBERNATE 4956 /* 4957 * Side-effect free (no malloc, printf, pool, splx) softraid crypto writer. 4958 * 4959 * This function must perform the following: 4960 * 1. Determine the underlying device's own side-effect free I/O function 4961 * (eg, ahci_hibernate_io, wd_hibernate_io, etc). 4962 * 2. Store enough information in the provided page argument for subsequent 4963 * I/O calls (such as the crypto discipline structure for the keys, the 4964 * offset of the softraid partition on the underlying disk, as well as 4965 * the offset of the swap partition within the crypto volume. 4966 * 3. Encrypt the incoming data using the sr_discipline keys, then pass 4967 * the request to the underlying device's own I/O function. 4968 */ 4969 int 4970 sr_hibernate_io(dev_t dev, daddr_t blkno, vaddr_t addr, size_t size, int op, void *page) 4971 { 4972 /* Struct for stashing data obtained on HIB_INIT. 4973 * XXX 4974 * We share the page with the underlying device's own 4975 * side-effect free I/O function, so we pad our data to 4976 * the end of the page. Presently this does not overlap 4977 * with either of the two other side-effect free i/o 4978 * functions (ahci/wd). 4979 */ 4980 struct { 4981 char pad[3072]; 4982 struct sr_discipline *srd; 4983 hibio_fn subfn; /* underlying device i/o fn */ 4984 dev_t subdev; /* underlying device dev_t */ 4985 daddr_t sr_swapoff; /* ofs of swap part in sr volume */ 4986 char buf[DEV_BSIZE]; /* encryption performed into this buf */ 4987 } *my = page; 4988 extern struct cfdriver sd_cd; 4989 char errstr[128], *dl_ret; 4990 struct sr_chunk *schunk; 4991 struct sd_softc *sd; 4992 struct aes_xts_ctx ctx; 4993 struct sr_softc *sc; 4994 struct device *dv; 4995 daddr_t key_blkno; 4996 uint32_t sub_raidoff; /* ofs of sr part in underlying dev */ 4997 struct disklabel dl; 4998 size_t i, j; 4999 u_char iv[8]; 5000 5001 /* 5002 * In HIB_INIT, we are passed the swap partition size and offset 5003 * in 'size' and 'blkno' respectively. These are relative to the 5004 * start of the softraid partition, and we need to save these 5005 * for later translation to the underlying device's layout. 5006 */ 5007 if (op == HIB_INIT) { 5008 dv = disk_lookup(&sd_cd, DISKUNIT(dev)); 5009 sd = (struct sd_softc *)dv; 5010 sc = (struct sr_softc *)dv->dv_parent->dv_parent; 5011 5012 /* 5013 * Look up the sr discipline. This is used to determine 5014 * if we are SR crypto and what the underlying device is. 5015 */ 5016 my->srd = sc->sc_targets[sd->sc_link->target]; 5017 DNPRINTF(SR_D_MISC, "sr_hibernate_io: discipline is %s\n", 5018 my->srd->sd_name); 5019 if (strncmp(my->srd->sd_name, "CRYPTO", 10)) 5020 return (ENOTSUP); 5021 5022 /* Find the underlying device */ 5023 schunk = my->srd->sd_vol.sv_chunks[0]; 5024 my->subdev = schunk->src_dev_mm; 5025 5026 /* 5027 * Find the appropriate underlying device side effect free 5028 * I/O function, based on the type of device it is. 5029 */ 5030 my->subfn = get_hibernate_io_function(my->subdev); 5031 5032 /* 5033 * Find block offset where this raid partition is on 5034 * the underlying disk. 5035 */ 5036 dl_ret = disk_readlabel(&dl, my->subdev, errstr, 5037 sizeof(errstr)); 5038 if (dl_ret) { 5039 printf("Hibernate error reading disklabel: %s\n", dl_ret); 5040 return (ENOTSUP); 5041 } 5042 5043 if (dl.d_partitions[DISKPART(my->subdev)].p_fstype != FS_RAID || 5044 DL_GETPSIZE(&dl.d_partitions[DISKPART(my->subdev)]) == 0) 5045 return (ENOTSUP); 5046 5047 /* Find the offset of the SR part in the underlying device */ 5048 sub_raidoff = my->srd->sd_meta->ssd_data_offset + 5049 DL_GETPOFFSET(&dl.d_partitions[DISKPART(my->subdev)]); 5050 DNPRINTF(SR_D_MISC,"sr_hibernate_io: blk trans ofs: %d blks\n", 5051 sub_raidoff); 5052 5053 /* Save the offset of the swap partition in the SR disk */ 5054 my->sr_swapoff = blkno; 5055 5056 /* Initialize the sub-device */ 5057 return my->subfn(my->subdev, sub_raidoff + blkno, 5058 addr, size, op, page); 5059 } 5060 5061 /* Hibernate only uses (and we only support) writes */ 5062 if (op != HIB_W) 5063 return (ENOTSUP); 5064 5065 /* 5066 * Blocks act as the IV for the encryption. These block numbers 5067 * are relative to the start of the sr partition, but the 'blkno' 5068 * passed above is relative to the start of the swap partition 5069 * inside the sr partition, so bias appropriately. 5070 */ 5071 key_blkno = my->sr_swapoff + blkno; 5072 5073 /* Process each disk block one at a time. */ 5074 for (i = 0; i < size; i += DEV_BSIZE) { 5075 int res; 5076 5077 bzero(&ctx, sizeof(ctx)); 5078 5079 /* 5080 * Set encryption key (from the sr discipline stashed 5081 * during HIB_INIT. This code is based on the softraid 5082 * bootblock code. 5083 */ 5084 aes_xts_setkey(&ctx, my->srd->mds.mdd_crypto.scr_key[0], 64); 5085 /* We encrypt DEV_BSIZE bytes at a time in my->buf */ 5086 bcopy(((char *)addr) + i, my->buf, DEV_BSIZE); 5087 5088 /* Block number is the IV */ 5089 bcopy(&key_blkno, &iv, sizeof(key_blkno)); 5090 aes_xts_reinit(&ctx, iv); 5091 5092 /* Encrypt DEV_BSIZE bytes, AES_XTS_BLOCKSIZE bytes at a time */ 5093 for (j = 0; j < DEV_BSIZE; j += AES_XTS_BLOCKSIZE) 5094 aes_xts_encrypt(&ctx, my->buf + j); 5095 5096 /* 5097 * Write one block out from my->buf to the underlying device 5098 * using its own side-effect free I/O function. 5099 */ 5100 res = my->subfn(my->subdev, blkno + (i / DEV_BSIZE), 5101 (vaddr_t)(my->buf), DEV_BSIZE, op, page); 5102 if (res != 0) 5103 return (res); 5104 key_blkno++; 5105 } 5106 return (0); 5107 } 5108 #endif /* HIBERNATE */ 5109