1 /* $OpenBSD: softraid.c,v 1.416 2020/10/15 00:13:47 krw 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/lock.h> 37 #include <sys/mount.h> 38 #include <sys/sensors.h> 39 #include <sys/stat.h> 40 #include <sys/conf.h> 41 #include <sys/uio.h> 42 #include <sys/task.h> 43 #include <sys/kthread.h> 44 #include <sys/dkio.h> 45 #include <sys/stdint.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 53 #ifdef HIBERNATE 54 #include <lib/libsa/aes_xts.h> 55 #include <sys/hibernate.h> 56 #include <scsi/sdvar.h> 57 #endif /* HIBERNATE */ 58 59 /* #define SR_FANCY_STATS */ 60 61 #ifdef SR_DEBUG 62 #define SR_FANCY_STATS 63 uint32_t sr_debug = 0 64 /* | SR_D_CMD */ 65 /* | SR_D_MISC */ 66 /* | SR_D_INTR */ 67 /* | SR_D_IOCTL */ 68 /* | SR_D_CCB */ 69 /* | SR_D_WU */ 70 /* | SR_D_META */ 71 /* | SR_D_DIS */ 72 /* | SR_D_STATE */ 73 /* | SR_D_REBUILD */ 74 ; 75 #endif 76 77 struct sr_softc *softraid0; 78 struct sr_uuid sr_bootuuid; 79 u_int8_t sr_bootkey[SR_CRYPTO_MAXKEYBYTES]; 80 81 int sr_match(struct device *, void *, void *); 82 void sr_attach(struct device *, struct device *, void *); 83 int sr_detach(struct device *, int); 84 void sr_map_root(void); 85 86 struct cfattach softraid_ca = { 87 sizeof(struct sr_softc), sr_match, sr_attach, sr_detach, 88 }; 89 90 struct cfdriver softraid_cd = { 91 NULL, "softraid", DV_DULL 92 }; 93 94 /* scsi & discipline */ 95 void sr_scsi_cmd(struct scsi_xfer *); 96 int sr_scsi_probe(struct scsi_link *); 97 int sr_scsi_ioctl(struct scsi_link *, u_long, 98 caddr_t, int); 99 int sr_bio_ioctl(struct device *, u_long, caddr_t); 100 int sr_bio_handler(struct sr_softc *, 101 struct sr_discipline *, u_long, struct bio *); 102 int sr_ioctl_inq(struct sr_softc *, struct bioc_inq *); 103 int sr_ioctl_vol(struct sr_softc *, struct bioc_vol *); 104 int sr_ioctl_disk(struct sr_softc *, struct bioc_disk *); 105 int sr_ioctl_setstate(struct sr_softc *, 106 struct bioc_setstate *); 107 int sr_ioctl_createraid(struct sr_softc *, 108 struct bioc_createraid *, int, void *); 109 int sr_ioctl_deleteraid(struct sr_softc *, 110 struct sr_discipline *, struct bioc_deleteraid *); 111 int sr_ioctl_discipline(struct sr_softc *, 112 struct sr_discipline *, struct bioc_discipline *); 113 int sr_ioctl_installboot(struct sr_softc *, 114 struct sr_discipline *, struct bioc_installboot *); 115 void sr_chunks_unwind(struct sr_softc *, 116 struct sr_chunk_head *); 117 void sr_discipline_free(struct sr_discipline *); 118 void sr_discipline_shutdown(struct sr_discipline *, int, int); 119 int sr_discipline_init(struct sr_discipline *, int); 120 int sr_alloc_resources(struct sr_discipline *); 121 void sr_free_resources(struct sr_discipline *); 122 void sr_set_chunk_state(struct sr_discipline *, int, int); 123 void sr_set_vol_state(struct sr_discipline *); 124 125 /* utility functions */ 126 void sr_shutdown(int); 127 void sr_uuid_generate(struct sr_uuid *); 128 char *sr_uuid_format(struct sr_uuid *); 129 void sr_uuid_print(struct sr_uuid *, int); 130 void sr_checksum_print(u_int8_t *); 131 int sr_boot_assembly(struct sr_softc *); 132 int sr_already_assembled(struct sr_discipline *); 133 int sr_hotspare(struct sr_softc *, dev_t); 134 void sr_hotspare_rebuild(struct sr_discipline *); 135 int sr_rebuild_init(struct sr_discipline *, dev_t, int); 136 void sr_rebuild_start(void *); 137 void sr_rebuild_thread(void *); 138 void sr_rebuild(struct sr_discipline *); 139 void sr_roam_chunks(struct sr_discipline *); 140 int sr_chunk_in_use(struct sr_softc *, dev_t); 141 int sr_rw(struct sr_softc *, dev_t, char *, size_t, 142 daddr_t, long); 143 void sr_wu_done_callback(void *); 144 145 /* don't include these on RAMDISK */ 146 #ifndef SMALL_KERNEL 147 void sr_sensors_refresh(void *); 148 int sr_sensors_create(struct sr_discipline *); 149 void sr_sensors_delete(struct sr_discipline *); 150 #endif 151 152 /* metadata */ 153 int sr_meta_probe(struct sr_discipline *, dev_t *, int); 154 int sr_meta_attach(struct sr_discipline *, int, int); 155 int sr_meta_rw(struct sr_discipline *, dev_t, void *, long); 156 int sr_meta_clear(struct sr_discipline *); 157 void sr_meta_init(struct sr_discipline *, int, int); 158 void sr_meta_init_complete(struct sr_discipline *); 159 void sr_meta_opt_handler(struct sr_discipline *, 160 struct sr_meta_opt_hdr *); 161 162 /* hotplug magic */ 163 void sr_disk_attach(struct disk *, int); 164 165 struct sr_hotplug_list { 166 void (*sh_hotplug)(struct sr_discipline *, 167 struct disk *, int); 168 struct sr_discipline *sh_sd; 169 170 SLIST_ENTRY(sr_hotplug_list) shl_link; 171 }; 172 SLIST_HEAD(sr_hotplug_list_head, sr_hotplug_list); 173 174 struct sr_hotplug_list_head sr_hotplug_callbacks; 175 extern void (*softraid_disk_attach)(struct disk *, int); 176 177 /* scsi glue */ 178 struct scsi_adapter sr_switch = { 179 sr_scsi_cmd, NULL, sr_scsi_probe, NULL, sr_scsi_ioctl 180 }; 181 182 /* native metadata format */ 183 int sr_meta_native_bootprobe(struct sr_softc *, dev_t, 184 struct sr_boot_chunk_head *); 185 #define SR_META_NOTCLAIMED (0) 186 #define SR_META_CLAIMED (1) 187 int sr_meta_native_probe(struct sr_softc *, 188 struct sr_chunk *); 189 int sr_meta_native_attach(struct sr_discipline *, int); 190 int sr_meta_native_write(struct sr_discipline *, dev_t, 191 struct sr_metadata *,void *); 192 193 #ifdef SR_DEBUG 194 void sr_meta_print(struct sr_metadata *); 195 #else 196 #define sr_meta_print(m) 197 #endif 198 199 /* the metadata driver should remain stateless */ 200 struct sr_meta_driver { 201 daddr_t smd_offset; /* metadata location */ 202 u_int32_t smd_size; /* size of metadata */ 203 204 int (*smd_probe)(struct sr_softc *, 205 struct sr_chunk *); 206 int (*smd_attach)(struct sr_discipline *, int); 207 int (*smd_detach)(struct sr_discipline *); 208 int (*smd_read)(struct sr_discipline *, dev_t, 209 struct sr_metadata *, void *); 210 int (*smd_write)(struct sr_discipline *, dev_t, 211 struct sr_metadata *, void *); 212 int (*smd_validate)(struct sr_discipline *, 213 struct sr_metadata *, void *); 214 } smd[] = { 215 { SR_META_OFFSET, SR_META_SIZE * DEV_BSIZE, 216 sr_meta_native_probe, sr_meta_native_attach, NULL, 217 sr_meta_native_read, sr_meta_native_write, NULL }, 218 { 0, 0, NULL, NULL, NULL, NULL } 219 }; 220 221 int 222 sr_meta_attach(struct sr_discipline *sd, int chunk_no, int force) 223 { 224 struct sr_softc *sc = sd->sd_sc; 225 struct sr_chunk_head *cl; 226 struct sr_chunk *ch_entry, *chunk1, *chunk2; 227 int rv = 1, i = 0; 228 229 DNPRINTF(SR_D_META, "%s: sr_meta_attach(%d)\n", DEVNAME(sc), chunk_no); 230 231 /* in memory copy of metadata */ 232 sd->sd_meta = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, 233 M_ZERO | M_NOWAIT); 234 if (!sd->sd_meta) { 235 sr_error(sc, "could not allocate memory for metadata"); 236 goto bad; 237 } 238 239 if (sd->sd_meta_type != SR_META_F_NATIVE) { 240 /* in memory copy of foreign metadata */ 241 sd->sd_meta_foreign = malloc(smd[sd->sd_meta_type].smd_size, 242 M_DEVBUF, M_ZERO | M_NOWAIT); 243 if (!sd->sd_meta_foreign) { 244 /* unwind frees sd_meta */ 245 sr_error(sc, "could not allocate memory for foreign " 246 "metadata"); 247 goto bad; 248 } 249 } 250 251 /* we have a valid list now create an array index */ 252 cl = &sd->sd_vol.sv_chunk_list; 253 sd->sd_vol.sv_chunks = mallocarray(chunk_no, sizeof(struct sr_chunk *), 254 M_DEVBUF, M_WAITOK | M_ZERO); 255 256 /* fill out chunk array */ 257 i = 0; 258 SLIST_FOREACH(ch_entry, cl, src_link) 259 sd->sd_vol.sv_chunks[i++] = ch_entry; 260 261 /* attach metadata */ 262 if (smd[sd->sd_meta_type].smd_attach(sd, force)) 263 goto bad; 264 265 /* Force chunks into correct order now that metadata is attached. */ 266 SLIST_INIT(cl); 267 for (i = 0; i < chunk_no; i++) { 268 ch_entry = sd->sd_vol.sv_chunks[i]; 269 chunk2 = NULL; 270 SLIST_FOREACH(chunk1, cl, src_link) { 271 if (chunk1->src_meta.scmi.scm_chunk_id > 272 ch_entry->src_meta.scmi.scm_chunk_id) 273 break; 274 chunk2 = chunk1; 275 } 276 if (chunk2 == NULL) 277 SLIST_INSERT_HEAD(cl, ch_entry, src_link); 278 else 279 SLIST_INSERT_AFTER(chunk2, ch_entry, src_link); 280 } 281 i = 0; 282 SLIST_FOREACH(ch_entry, cl, src_link) 283 sd->sd_vol.sv_chunks[i++] = ch_entry; 284 285 rv = 0; 286 bad: 287 return (rv); 288 } 289 290 int 291 sr_meta_probe(struct sr_discipline *sd, dev_t *dt, int no_chunk) 292 { 293 struct sr_softc *sc = sd->sd_sc; 294 struct vnode *vn; 295 struct sr_chunk *ch_entry, *ch_prev = NULL; 296 struct sr_chunk_head *cl; 297 char devname[32]; 298 int i, d, type, found, prevf, error; 299 dev_t dev; 300 301 DNPRINTF(SR_D_META, "%s: sr_meta_probe(%d)\n", DEVNAME(sc), no_chunk); 302 303 if (no_chunk == 0) 304 goto unwind; 305 306 cl = &sd->sd_vol.sv_chunk_list; 307 308 for (d = 0, prevf = SR_META_F_INVALID; d < no_chunk; d++) { 309 ch_entry = malloc(sizeof(struct sr_chunk), M_DEVBUF, 310 M_WAITOK | M_ZERO); 311 /* keep disks in user supplied order */ 312 if (ch_prev) 313 SLIST_INSERT_AFTER(ch_prev, ch_entry, src_link); 314 else 315 SLIST_INSERT_HEAD(cl, ch_entry, src_link); 316 ch_prev = ch_entry; 317 dev = dt[d]; 318 ch_entry->src_dev_mm = dev; 319 320 if (dev == NODEV) { 321 ch_entry->src_meta.scm_status = BIOC_SDOFFLINE; 322 continue; 323 } else { 324 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 325 if (bdevvp(dev, &vn)) { 326 sr_error(sc, "sr_meta_probe: cannot allocate " 327 "vnode"); 328 goto unwind; 329 } 330 331 /* 332 * XXX leaving dev open for now; move this to attach 333 * and figure out the open/close dance for unwind. 334 */ 335 error = VOP_OPEN(vn, FREAD | FWRITE, NOCRED, curproc); 336 if (error) { 337 DNPRINTF(SR_D_META,"%s: sr_meta_probe can't " 338 "open %s\n", DEVNAME(sc), devname); 339 vput(vn); 340 goto unwind; 341 } 342 343 strlcpy(ch_entry->src_devname, devname, 344 sizeof(ch_entry->src_devname)); 345 ch_entry->src_vn = vn; 346 } 347 348 /* determine if this is a device we understand */ 349 for (i = 0, found = SR_META_F_INVALID; smd[i].smd_probe; i++) { 350 type = smd[i].smd_probe(sc, ch_entry); 351 if (type == SR_META_F_INVALID) 352 continue; 353 else { 354 found = type; 355 break; 356 } 357 } 358 359 if (found == SR_META_F_INVALID) 360 goto unwind; 361 if (prevf == SR_META_F_INVALID) 362 prevf = found; 363 if (prevf != found) { 364 DNPRINTF(SR_D_META, "%s: prevf != found\n", 365 DEVNAME(sc)); 366 goto unwind; 367 } 368 } 369 370 return (prevf); 371 unwind: 372 return (SR_META_F_INVALID); 373 } 374 375 void 376 sr_meta_getdevname(struct sr_softc *sc, dev_t dev, char *buf, int size) 377 { 378 int maj, unit, part; 379 char *name; 380 381 DNPRINTF(SR_D_META, "%s: sr_meta_getdevname(%p, %d)\n", 382 DEVNAME(sc), buf, size); 383 384 if (!buf) 385 return; 386 387 maj = major(dev); 388 part = DISKPART(dev); 389 unit = DISKUNIT(dev); 390 391 name = findblkname(maj); 392 if (name == NULL) 393 return; 394 395 snprintf(buf, size, "%s%d%c", name, unit, part + 'a'); 396 } 397 398 int 399 sr_rw(struct sr_softc *sc, dev_t dev, char *buf, size_t size, daddr_t blkno, 400 long flags) 401 { 402 struct vnode *vp; 403 struct buf b; 404 size_t bufsize, dma_bufsize; 405 int rv = 1; 406 char *dma_buf; 407 408 DNPRINTF(SR_D_MISC, "%s: sr_rw(0x%x, %p, %zu, %lld 0x%lx)\n", 409 DEVNAME(sc), dev, buf, size, (long long)blkno, flags); 410 411 dma_bufsize = (size > MAXPHYS) ? MAXPHYS : size; 412 dma_buf = dma_alloc(dma_bufsize, PR_WAITOK); 413 414 if (bdevvp(dev, &vp)) { 415 printf("%s: sr_rw: failed to allocate vnode\n", DEVNAME(sc)); 416 goto done; 417 } 418 419 while (size > 0) { 420 DNPRINTF(SR_D_MISC, "%s: dma_buf %p, size %zu, blkno %lld)\n", 421 DEVNAME(sc), dma_buf, size, (long long)blkno); 422 423 bufsize = (size > MAXPHYS) ? MAXPHYS : size; 424 if (flags == B_WRITE) 425 memcpy(dma_buf, buf, bufsize); 426 427 bzero(&b, sizeof(b)); 428 b.b_flags = flags | B_PHYS; 429 b.b_proc = curproc; 430 b.b_dev = dev; 431 b.b_iodone = NULL; 432 b.b_error = 0; 433 b.b_blkno = blkno; 434 b.b_data = dma_buf; 435 b.b_bcount = bufsize; 436 b.b_bufsize = bufsize; 437 b.b_resid = bufsize; 438 b.b_vp = vp; 439 440 if ((b.b_flags & B_READ) == 0) 441 vp->v_numoutput++; 442 443 LIST_INIT(&b.b_dep); 444 VOP_STRATEGY(&b); 445 biowait(&b); 446 447 if (b.b_flags & B_ERROR) { 448 printf("%s: I/O error %d on dev 0x%x at block %llu\n", 449 DEVNAME(sc), b.b_error, dev, b.b_blkno); 450 goto done; 451 } 452 453 if (flags == B_READ) 454 memcpy(buf, dma_buf, bufsize); 455 456 size -= bufsize; 457 buf += bufsize; 458 blkno += howmany(bufsize, DEV_BSIZE); 459 } 460 461 rv = 0; 462 463 done: 464 if (vp) 465 vput(vp); 466 467 dma_free(dma_buf, dma_bufsize); 468 469 return (rv); 470 } 471 472 int 473 sr_meta_rw(struct sr_discipline *sd, dev_t dev, void *md, long flags) 474 { 475 int rv = 1; 476 477 DNPRINTF(SR_D_META, "%s: sr_meta_rw(0x%x, %p, 0x%lx)\n", 478 DEVNAME(sd->sd_sc), dev, md, flags); 479 480 if (md == NULL) { 481 printf("%s: sr_meta_rw: invalid metadata pointer\n", 482 DEVNAME(sd->sd_sc)); 483 goto done; 484 } 485 486 rv = sr_rw(sd->sd_sc, dev, md, SR_META_SIZE * DEV_BSIZE, 487 SR_META_OFFSET, flags); 488 489 done: 490 return (rv); 491 } 492 493 int 494 sr_meta_clear(struct sr_discipline *sd) 495 { 496 struct sr_softc *sc = sd->sd_sc; 497 struct sr_chunk_head *cl = &sd->sd_vol.sv_chunk_list; 498 struct sr_chunk *ch_entry; 499 void *m; 500 int rv = 1; 501 502 DNPRINTF(SR_D_META, "%s: sr_meta_clear\n", DEVNAME(sc)); 503 504 if (sd->sd_meta_type != SR_META_F_NATIVE) { 505 sr_error(sc, "cannot clear foreign metadata"); 506 goto done; 507 } 508 509 m = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, M_WAITOK | M_ZERO); 510 SLIST_FOREACH(ch_entry, cl, src_link) { 511 if (sr_meta_native_write(sd, ch_entry->src_dev_mm, m, NULL)) { 512 /* XXX mark disk offline */ 513 DNPRINTF(SR_D_META, "%s: sr_meta_clear failed to " 514 "clear %s\n", DEVNAME(sc), ch_entry->src_devname); 515 rv++; 516 continue; 517 } 518 bzero(&ch_entry->src_meta, sizeof(ch_entry->src_meta)); 519 } 520 521 bzero(sd->sd_meta, SR_META_SIZE * DEV_BSIZE); 522 523 free(m, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 524 rv = 0; 525 done: 526 return (rv); 527 } 528 529 void 530 sr_meta_init(struct sr_discipline *sd, int level, int no_chunk) 531 { 532 struct sr_softc *sc = sd->sd_sc; 533 struct sr_metadata *sm = sd->sd_meta; 534 struct sr_chunk_head *cl = &sd->sd_vol.sv_chunk_list; 535 struct sr_meta_chunk *scm; 536 struct sr_chunk *chunk; 537 int cid = 0; 538 u_int64_t max_chunk_sz = 0, min_chunk_sz = 0; 539 u_int32_t secsize = DEV_BSIZE; 540 541 DNPRINTF(SR_D_META, "%s: sr_meta_init\n", DEVNAME(sc)); 542 543 if (!sm) 544 return; 545 546 /* Initialise volume metadata. */ 547 sm->ssdi.ssd_magic = SR_MAGIC; 548 sm->ssdi.ssd_version = SR_META_VERSION; 549 sm->ssdi.ssd_vol_flags = sd->sd_meta_flags; 550 sm->ssdi.ssd_volid = 0; 551 sm->ssdi.ssd_chunk_no = no_chunk; 552 sm->ssdi.ssd_level = level; 553 554 sm->ssd_data_blkno = SR_DATA_OFFSET; 555 sm->ssd_ondisk = 0; 556 557 sr_uuid_generate(&sm->ssdi.ssd_uuid); 558 559 /* Initialise chunk metadata and get min/max chunk sizes & secsize. */ 560 SLIST_FOREACH(chunk, cl, src_link) { 561 scm = &chunk->src_meta; 562 scm->scmi.scm_size = chunk->src_size; 563 scm->scmi.scm_chunk_id = cid++; 564 scm->scm_status = BIOC_SDONLINE; 565 scm->scmi.scm_volid = 0; 566 strlcpy(scm->scmi.scm_devname, chunk->src_devname, 567 sizeof(scm->scmi.scm_devname)); 568 memcpy(&scm->scmi.scm_uuid, &sm->ssdi.ssd_uuid, 569 sizeof(scm->scmi.scm_uuid)); 570 sr_checksum(sc, scm, &scm->scm_checksum, 571 sizeof(scm->scm_checksum)); 572 573 if (min_chunk_sz == 0) 574 min_chunk_sz = scm->scmi.scm_size; 575 if (chunk->src_secsize > secsize) 576 secsize = chunk->src_secsize; 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 sm->ssdi.ssd_secsize = secsize; 582 583 /* Equalize chunk sizes. */ 584 SLIST_FOREACH(chunk, cl, src_link) 585 chunk->src_meta.scmi.scm_coerced_size = min_chunk_sz; 586 587 sd->sd_vol.sv_chunk_minsz = min_chunk_sz; 588 sd->sd_vol.sv_chunk_maxsz = max_chunk_sz; 589 } 590 591 void 592 sr_meta_init_complete(struct sr_discipline *sd) 593 { 594 #ifdef SR_DEBUG 595 struct sr_softc *sc = sd->sd_sc; 596 #endif 597 struct sr_metadata *sm = sd->sd_meta; 598 599 DNPRINTF(SR_D_META, "%s: sr_meta_complete\n", DEVNAME(sc)); 600 601 /* Complete initialisation of volume metadata. */ 602 strlcpy(sm->ssdi.ssd_vendor, "OPENBSD", sizeof(sm->ssdi.ssd_vendor)); 603 snprintf(sm->ssdi.ssd_product, sizeof(sm->ssdi.ssd_product), 604 "SR %s", sd->sd_name); 605 snprintf(sm->ssdi.ssd_revision, sizeof(sm->ssdi.ssd_revision), 606 "%03d", sm->ssdi.ssd_version); 607 } 608 609 void 610 sr_meta_opt_handler(struct sr_discipline *sd, struct sr_meta_opt_hdr *om) 611 { 612 if (om->som_type != SR_OPT_BOOT) 613 panic("unknown optional metadata type"); 614 } 615 616 void 617 sr_meta_save_callback(void *xsd) 618 { 619 struct sr_discipline *sd = xsd; 620 int s; 621 622 s = splbio(); 623 624 if (sr_meta_save(sd, SR_META_DIRTY)) 625 printf("%s: save metadata failed\n", DEVNAME(sd->sd_sc)); 626 627 sd->sd_must_flush = 0; 628 splx(s); 629 } 630 631 int 632 sr_meta_save(struct sr_discipline *sd, u_int32_t flags) 633 { 634 struct sr_softc *sc = sd->sd_sc; 635 struct sr_metadata *sm = sd->sd_meta, *m; 636 struct sr_meta_driver *s; 637 struct sr_chunk *src; 638 struct sr_meta_chunk *cm; 639 struct sr_workunit wu; 640 struct sr_meta_opt_hdr *omh; 641 struct sr_meta_opt_item *omi; 642 int i; 643 644 DNPRINTF(SR_D_META, "%s: sr_meta_save %s\n", 645 DEVNAME(sc), sd->sd_meta->ssd_devname); 646 647 if (!sm) { 648 printf("%s: no in memory copy of metadata\n", DEVNAME(sc)); 649 goto bad; 650 } 651 652 /* meta scratchpad */ 653 s = &smd[sd->sd_meta_type]; 654 m = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, M_ZERO | M_NOWAIT); 655 if (!m) { 656 printf("%s: could not allocate metadata scratch area\n", 657 DEVNAME(sc)); 658 goto bad; 659 } 660 661 /* from here on out metadata is updated */ 662 restart: 663 sm->ssd_ondisk++; 664 sm->ssd_meta_flags = flags; 665 memcpy(m, sm, sizeof(*m)); 666 667 /* Chunk metadata. */ 668 cm = (struct sr_meta_chunk *)(m + 1); 669 for (i = 0; i < sm->ssdi.ssd_chunk_no; i++) { 670 src = sd->sd_vol.sv_chunks[i]; 671 memcpy(cm, &src->src_meta, sizeof(*cm)); 672 cm++; 673 } 674 675 /* Optional metadata. */ 676 omh = (struct sr_meta_opt_hdr *)(cm); 677 SLIST_FOREACH(omi, &sd->sd_meta_opt, omi_link) { 678 DNPRINTF(SR_D_META, "%s: saving optional metadata type %u with " 679 "length %u\n", DEVNAME(sc), omi->omi_som->som_type, 680 omi->omi_som->som_length); 681 bzero(&omi->omi_som->som_checksum, MD5_DIGEST_LENGTH); 682 sr_checksum(sc, omi->omi_som, &omi->omi_som->som_checksum, 683 omi->omi_som->som_length); 684 memcpy(omh, omi->omi_som, omi->omi_som->som_length); 685 omh = (struct sr_meta_opt_hdr *)((u_int8_t *)omh + 686 omi->omi_som->som_length); 687 } 688 689 for (i = 0; i < sm->ssdi.ssd_chunk_no; i++) { 690 src = sd->sd_vol.sv_chunks[i]; 691 692 /* skip disks that are offline */ 693 if (src->src_meta.scm_status == BIOC_SDOFFLINE) 694 continue; 695 696 /* calculate metadata checksum for correct chunk */ 697 m->ssdi.ssd_chunk_id = i; 698 sr_checksum(sc, m, &m->ssd_checksum, 699 sizeof(struct sr_meta_invariant)); 700 701 #ifdef SR_DEBUG 702 DNPRINTF(SR_D_META, "%s: sr_meta_save %s: volid: %d " 703 "chunkid: %d checksum: ", 704 DEVNAME(sc), src->src_meta.scmi.scm_devname, 705 m->ssdi.ssd_volid, m->ssdi.ssd_chunk_id); 706 707 if (sr_debug & SR_D_META) 708 sr_checksum_print((u_int8_t *)&m->ssd_checksum); 709 DNPRINTF(SR_D_META, "\n"); 710 sr_meta_print(m); 711 #endif 712 713 /* translate and write to disk */ 714 if (s->smd_write(sd, src->src_dev_mm, m, NULL /* XXX */)) { 715 printf("%s: could not write metadata to %s\n", 716 DEVNAME(sc), src->src_devname); 717 /* restart the meta write */ 718 src->src_meta.scm_status = BIOC_SDOFFLINE; 719 /* XXX recalculate volume status */ 720 goto restart; 721 } 722 } 723 724 /* not all disciplines have sync */ 725 if (sd->sd_scsi_sync) { 726 bzero(&wu, sizeof(wu)); 727 wu.swu_flags |= SR_WUF_FAKE; 728 wu.swu_dis = sd; 729 sd->sd_scsi_sync(&wu); 730 } 731 free(m, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 732 return (0); 733 bad: 734 return (1); 735 } 736 737 int 738 sr_meta_read(struct sr_discipline *sd) 739 { 740 struct sr_softc *sc = sd->sd_sc; 741 struct sr_chunk_head *cl = &sd->sd_vol.sv_chunk_list; 742 struct sr_metadata *sm; 743 struct sr_chunk *ch_entry; 744 struct sr_meta_chunk *cp; 745 struct sr_meta_driver *s; 746 void *fm = NULL; 747 int no_disk = 0, got_meta = 0; 748 749 DNPRINTF(SR_D_META, "%s: sr_meta_read\n", DEVNAME(sc)); 750 751 sm = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, M_WAITOK | M_ZERO); 752 s = &smd[sd->sd_meta_type]; 753 if (sd->sd_meta_type != SR_META_F_NATIVE) 754 fm = malloc(s->smd_size, M_DEVBUF, M_WAITOK | M_ZERO); 755 756 cp = (struct sr_meta_chunk *)(sm + 1); 757 SLIST_FOREACH(ch_entry, cl, src_link) { 758 /* skip disks that are offline */ 759 if (ch_entry->src_meta.scm_status == BIOC_SDOFFLINE) { 760 DNPRINTF(SR_D_META, 761 "%s: %s chunk marked offline, spoofing status\n", 762 DEVNAME(sc), ch_entry->src_devname); 763 cp++; /* adjust chunk pointer to match failure */ 764 continue; 765 } else if (s->smd_read(sd, ch_entry->src_dev_mm, sm, fm)) { 766 /* read and translate */ 767 /* XXX mark chunk offline, elsewhere!! */ 768 ch_entry->src_meta.scm_status = BIOC_SDOFFLINE; 769 cp++; /* adjust chunk pointer to match failure */ 770 DNPRINTF(SR_D_META, "%s: sr_meta_read failed\n", 771 DEVNAME(sc)); 772 continue; 773 } 774 775 if (sm->ssdi.ssd_magic != SR_MAGIC) { 776 DNPRINTF(SR_D_META, "%s: sr_meta_read !SR_MAGIC\n", 777 DEVNAME(sc)); 778 continue; 779 } 780 781 /* validate metadata */ 782 if (sr_meta_validate(sd, ch_entry->src_dev_mm, sm, fm)) { 783 DNPRINTF(SR_D_META, "%s: invalid metadata\n", 784 DEVNAME(sc)); 785 no_disk = -1; 786 goto done; 787 } 788 789 /* assume first chunk contains metadata */ 790 if (got_meta == 0) { 791 sr_meta_opt_load(sc, sm, &sd->sd_meta_opt); 792 memcpy(sd->sd_meta, sm, sizeof(*sd->sd_meta)); 793 got_meta = 1; 794 } 795 796 memcpy(&ch_entry->src_meta, cp, sizeof(ch_entry->src_meta)); 797 798 no_disk++; 799 cp++; 800 } 801 802 free(sm, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 803 free(fm, M_DEVBUF, s->smd_size); 804 805 done: 806 DNPRINTF(SR_D_META, "%s: sr_meta_read found %d parts\n", DEVNAME(sc), 807 no_disk); 808 return (no_disk); 809 } 810 811 void 812 sr_meta_opt_load(struct sr_softc *sc, struct sr_metadata *sm, 813 struct sr_meta_opt_head *som) 814 { 815 struct sr_meta_opt_hdr *omh; 816 struct sr_meta_opt_item *omi; 817 u_int8_t checksum[MD5_DIGEST_LENGTH]; 818 int i; 819 820 /* Process optional metadata. */ 821 omh = (struct sr_meta_opt_hdr *)((u_int8_t *)(sm + 1) + 822 sizeof(struct sr_meta_chunk) * sm->ssdi.ssd_chunk_no); 823 for (i = 0; i < sm->ssdi.ssd_opt_no; i++) { 824 825 omi = malloc(sizeof(struct sr_meta_opt_item), M_DEVBUF, 826 M_WAITOK | M_ZERO); 827 SLIST_INSERT_HEAD(som, omi, omi_link); 828 829 if (omh->som_length == 0) { 830 831 /* Load old fixed length optional metadata. */ 832 DNPRINTF(SR_D_META, "%s: old optional metadata of type " 833 "%u\n", DEVNAME(sc), omh->som_type); 834 835 /* Validate checksum. */ 836 sr_checksum(sc, (void *)omh, &checksum, 837 SR_OLD_META_OPT_SIZE - MD5_DIGEST_LENGTH); 838 if (bcmp(&checksum, (void *)omh + SR_OLD_META_OPT_MD5, 839 sizeof(checksum))) 840 panic("%s: invalid optional metadata " 841 "checksum", DEVNAME(sc)); 842 843 /* Determine correct length. */ 844 switch (omh->som_type) { 845 case SR_OPT_CRYPTO: 846 omh->som_length = sizeof(struct sr_meta_crypto); 847 break; 848 case SR_OPT_BOOT: 849 omh->som_length = sizeof(struct sr_meta_boot); 850 break; 851 case SR_OPT_KEYDISK: 852 omh->som_length = 853 sizeof(struct sr_meta_keydisk); 854 break; 855 default: 856 panic("unknown old optional metadata " 857 "type %u\n", omh->som_type); 858 } 859 860 omi->omi_som = malloc(omh->som_length, M_DEVBUF, 861 M_WAITOK | M_ZERO); 862 memcpy((u_int8_t *)omi->omi_som + sizeof(*omi->omi_som), 863 (u_int8_t *)omh + SR_OLD_META_OPT_OFFSET, 864 omh->som_length - sizeof(*omi->omi_som)); 865 omi->omi_som->som_type = omh->som_type; 866 omi->omi_som->som_length = omh->som_length; 867 868 omh = (struct sr_meta_opt_hdr *)((void *)omh + 869 SR_OLD_META_OPT_SIZE); 870 } else { 871 872 /* Load variable length optional metadata. */ 873 DNPRINTF(SR_D_META, "%s: optional metadata of type %u, " 874 "length %u\n", DEVNAME(sc), omh->som_type, 875 omh->som_length); 876 omi->omi_som = malloc(omh->som_length, M_DEVBUF, 877 M_WAITOK | M_ZERO); 878 memcpy(omi->omi_som, omh, omh->som_length); 879 880 /* Validate checksum. */ 881 memcpy(&checksum, &omi->omi_som->som_checksum, 882 MD5_DIGEST_LENGTH); 883 bzero(&omi->omi_som->som_checksum, MD5_DIGEST_LENGTH); 884 sr_checksum(sc, omi->omi_som, 885 &omi->omi_som->som_checksum, omh->som_length); 886 if (bcmp(&checksum, &omi->omi_som->som_checksum, 887 sizeof(checksum))) 888 panic("%s: invalid optional metadata checksum", 889 DEVNAME(sc)); 890 891 omh = (struct sr_meta_opt_hdr *)((void *)omh + 892 omh->som_length); 893 } 894 } 895 } 896 897 int 898 sr_meta_validate(struct sr_discipline *sd, dev_t dev, struct sr_metadata *sm, 899 void *fm) 900 { 901 struct sr_softc *sc = sd->sd_sc; 902 struct sr_meta_driver *s; 903 #ifdef SR_DEBUG 904 struct sr_meta_chunk *mc; 905 #endif 906 u_int8_t checksum[MD5_DIGEST_LENGTH]; 907 char devname[32]; 908 int rv = 1; 909 910 DNPRINTF(SR_D_META, "%s: sr_meta_validate(%p)\n", DEVNAME(sc), sm); 911 912 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 913 914 s = &smd[sd->sd_meta_type]; 915 if (sd->sd_meta_type != SR_META_F_NATIVE) 916 if (s->smd_validate(sd, sm, fm)) { 917 sr_error(sc, "invalid foreign metadata"); 918 goto done; 919 } 920 921 /* 922 * at this point all foreign metadata has been translated to the native 923 * format and will be treated just like the native format 924 */ 925 926 if (sm->ssdi.ssd_magic != SR_MAGIC) { 927 sr_error(sc, "not valid softraid metadata"); 928 goto done; 929 } 930 931 /* Verify metadata checksum. */ 932 sr_checksum(sc, sm, &checksum, sizeof(struct sr_meta_invariant)); 933 if (bcmp(&checksum, &sm->ssd_checksum, sizeof(checksum))) { 934 sr_error(sc, "invalid metadata checksum"); 935 goto done; 936 } 937 938 /* Handle changes between versions. */ 939 if (sm->ssdi.ssd_version == 3) { 940 941 /* 942 * Version 3 - update metadata version and fix up data blkno 943 * value since this did not exist in version 3. 944 */ 945 if (sm->ssd_data_blkno == 0) 946 sm->ssd_data_blkno = SR_META_V3_DATA_OFFSET; 947 sm->ssdi.ssd_secsize = DEV_BSIZE; 948 949 } else if (sm->ssdi.ssd_version == 4) { 950 951 /* 952 * Version 4 - original metadata format did not store 953 * data blkno so fix this up if necessary. 954 */ 955 if (sm->ssd_data_blkno == 0) 956 sm->ssd_data_blkno = SR_DATA_OFFSET; 957 sm->ssdi.ssd_secsize = DEV_BSIZE; 958 959 } else if (sm->ssdi.ssd_version == 5) { 960 961 /* 962 * Version 5 - variable length optional metadata. Migration 963 * from earlier fixed length optional metadata is handled 964 * in sr_meta_read(). 965 */ 966 sm->ssdi.ssd_secsize = DEV_BSIZE; 967 968 } else if (sm->ssdi.ssd_version == SR_META_VERSION) { 969 970 /* 971 * Version 6 - store & report a sector size. 972 */ 973 974 } else { 975 976 sr_error(sc, "cannot read metadata version %u on %s, " 977 "expected version %u or earlier", 978 sm->ssdi.ssd_version, devname, SR_META_VERSION); 979 goto done; 980 981 } 982 983 /* Update version number and revision string. */ 984 sm->ssdi.ssd_version = SR_META_VERSION; 985 snprintf(sm->ssdi.ssd_revision, sizeof(sm->ssdi.ssd_revision), 986 "%03d", SR_META_VERSION); 987 988 #ifdef SR_DEBUG 989 /* warn if disk changed order */ 990 mc = (struct sr_meta_chunk *)(sm + 1); 991 if (strncmp(mc[sm->ssdi.ssd_chunk_id].scmi.scm_devname, devname, 992 sizeof(mc[sm->ssdi.ssd_chunk_id].scmi.scm_devname))) 993 DNPRINTF(SR_D_META, "%s: roaming device %s -> %s\n", 994 DEVNAME(sc), mc[sm->ssdi.ssd_chunk_id].scmi.scm_devname, 995 devname); 996 #endif 997 998 /* we have meta data on disk */ 999 DNPRINTF(SR_D_META, "%s: sr_meta_validate valid metadata %s\n", 1000 DEVNAME(sc), devname); 1001 1002 rv = 0; 1003 done: 1004 return (rv); 1005 } 1006 1007 int 1008 sr_meta_native_bootprobe(struct sr_softc *sc, dev_t devno, 1009 struct sr_boot_chunk_head *bch) 1010 { 1011 struct vnode *vn; 1012 struct disklabel label; 1013 struct sr_metadata *md = NULL; 1014 struct sr_discipline *fake_sd = NULL; 1015 struct sr_boot_chunk *bc; 1016 char devname[32]; 1017 dev_t chrdev, rawdev; 1018 int error, i; 1019 int rv = SR_META_NOTCLAIMED; 1020 1021 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe\n", DEVNAME(sc)); 1022 1023 /* 1024 * Use character raw device to avoid SCSI complaints about missing 1025 * media on removable media devices. 1026 */ 1027 chrdev = blktochr(devno); 1028 rawdev = MAKEDISKDEV(major(chrdev), DISKUNIT(devno), RAW_PART); 1029 if (cdevvp(rawdev, &vn)) { 1030 sr_error(sc, "sr_meta_native_bootprobe: cannot allocate vnode"); 1031 goto done; 1032 } 1033 1034 /* open device */ 1035 error = VOP_OPEN(vn, FREAD, NOCRED, curproc); 1036 if (error) { 1037 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe open " 1038 "failed\n", DEVNAME(sc)); 1039 vput(vn); 1040 goto done; 1041 } 1042 1043 /* get disklabel */ 1044 error = VOP_IOCTL(vn, DIOCGDINFO, (caddr_t)&label, FREAD, NOCRED, 1045 curproc); 1046 if (error) { 1047 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe ioctl " 1048 "failed\n", DEVNAME(sc)); 1049 VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1050 vput(vn); 1051 goto done; 1052 } 1053 1054 /* we are done, close device */ 1055 error = VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1056 if (error) { 1057 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe close " 1058 "failed\n", DEVNAME(sc)); 1059 vput(vn); 1060 goto done; 1061 } 1062 vput(vn); 1063 1064 md = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, M_ZERO | M_NOWAIT); 1065 if (md == NULL) { 1066 sr_error(sc, "not enough memory for metadata buffer"); 1067 goto done; 1068 } 1069 1070 /* create fake sd to use utility functions */ 1071 fake_sd = malloc(sizeof(struct sr_discipline), M_DEVBUF, 1072 M_ZERO | M_NOWAIT); 1073 if (fake_sd == NULL) { 1074 sr_error(sc, "not enough memory for fake discipline"); 1075 goto done; 1076 } 1077 fake_sd->sd_sc = sc; 1078 fake_sd->sd_meta_type = SR_META_F_NATIVE; 1079 1080 for (i = 0; i < MAXPARTITIONS; i++) { 1081 if (label.d_partitions[i].p_fstype != FS_RAID) 1082 continue; 1083 1084 /* open partition */ 1085 rawdev = MAKEDISKDEV(major(devno), DISKUNIT(devno), i); 1086 if (bdevvp(rawdev, &vn)) { 1087 sr_error(sc, "sr_meta_native_bootprobe: cannot " 1088 "allocate vnode for partition"); 1089 goto done; 1090 } 1091 error = VOP_OPEN(vn, FREAD, NOCRED, curproc); 1092 if (error) { 1093 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe " 1094 "open failed, partition %d\n", 1095 DEVNAME(sc), i); 1096 vput(vn); 1097 continue; 1098 } 1099 1100 if (sr_meta_native_read(fake_sd, rawdev, md, NULL)) { 1101 sr_error(sc, "native bootprobe could not read native " 1102 "metadata"); 1103 VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1104 vput(vn); 1105 continue; 1106 } 1107 1108 /* are we a softraid partition? */ 1109 if (md->ssdi.ssd_magic != SR_MAGIC) { 1110 VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1111 vput(vn); 1112 continue; 1113 } 1114 1115 sr_meta_getdevname(sc, rawdev, devname, sizeof(devname)); 1116 if (sr_meta_validate(fake_sd, rawdev, md, NULL) == 0) { 1117 /* XXX fix M_WAITOK, this is boot time */ 1118 bc = malloc(sizeof(struct sr_boot_chunk), 1119 M_DEVBUF, M_WAITOK | M_ZERO); 1120 bc->sbc_metadata = malloc(sizeof(struct sr_metadata), 1121 M_DEVBUF, M_WAITOK | M_ZERO); 1122 memcpy(bc->sbc_metadata, md, sizeof(struct sr_metadata)); 1123 bc->sbc_mm = rawdev; 1124 SLIST_INSERT_HEAD(bch, bc, sbc_link); 1125 rv = SR_META_CLAIMED; 1126 } 1127 1128 /* we are done, close partition */ 1129 VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1130 vput(vn); 1131 } 1132 1133 done: 1134 free(fake_sd, M_DEVBUF, sizeof(struct sr_discipline)); 1135 free(md, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 1136 1137 return (rv); 1138 } 1139 1140 int 1141 sr_boot_assembly(struct sr_softc *sc) 1142 { 1143 struct sr_boot_volume_head bvh; 1144 struct sr_boot_chunk_head bch, kdh; 1145 struct sr_boot_volume *bv, *bv1, *bv2; 1146 struct sr_boot_chunk *bc, *bcnext, *bc1, *bc2; 1147 struct sr_disk_head sdklist; 1148 struct sr_disk *sdk; 1149 struct disk *dk; 1150 struct bioc_createraid bcr; 1151 struct sr_meta_chunk *hm; 1152 struct sr_chunk_head *cl; 1153 struct sr_chunk *hotspare, *chunk, *last; 1154 u_int64_t *ondisk = NULL; 1155 dev_t *devs = NULL; 1156 void *data; 1157 char devname[32]; 1158 int rv = 0, i; 1159 1160 DNPRINTF(SR_D_META, "%s: sr_boot_assembly\n", DEVNAME(sc)); 1161 1162 SLIST_INIT(&sdklist); 1163 SLIST_INIT(&bvh); 1164 SLIST_INIT(&bch); 1165 SLIST_INIT(&kdh); 1166 1167 dk = TAILQ_FIRST(&disklist); 1168 while (dk != NULL) { 1169 1170 /* See if this disk has been checked. */ 1171 SLIST_FOREACH(sdk, &sdklist, sdk_link) 1172 if (sdk->sdk_devno == dk->dk_devno) 1173 break; 1174 1175 if (sdk != NULL || dk->dk_devno == NODEV) { 1176 dk = TAILQ_NEXT(dk, dk_link); 1177 continue; 1178 } 1179 1180 /* Add this disk to the list that we've checked. */ 1181 sdk = malloc(sizeof(struct sr_disk), M_DEVBUF, 1182 M_NOWAIT | M_ZERO); 1183 if (sdk == NULL) 1184 goto unwind; 1185 sdk->sdk_devno = dk->dk_devno; 1186 SLIST_INSERT_HEAD(&sdklist, sdk, sdk_link); 1187 1188 /* Only check sd(4) and wd(4) devices. */ 1189 if (strncmp(dk->dk_name, "sd", 2) && 1190 strncmp(dk->dk_name, "wd", 2)) { 1191 dk = TAILQ_NEXT(dk, dk_link); 1192 continue; 1193 } 1194 1195 /* native softraid uses partitions */ 1196 rw_enter_write(&sc->sc_lock); 1197 bio_status_init(&sc->sc_status, &sc->sc_dev); 1198 sr_meta_native_bootprobe(sc, dk->dk_devno, &bch); 1199 rw_exit_write(&sc->sc_lock); 1200 1201 /* probe non-native disks if native failed. */ 1202 1203 /* Restart scan since we may have slept. */ 1204 dk = TAILQ_FIRST(&disklist); 1205 } 1206 1207 /* 1208 * Create a list of volumes and associate chunks with each volume. 1209 */ 1210 for (bc = SLIST_FIRST(&bch); bc != NULL; bc = bcnext) { 1211 1212 bcnext = SLIST_NEXT(bc, sbc_link); 1213 SLIST_REMOVE(&bch, bc, sr_boot_chunk, sbc_link); 1214 bc->sbc_chunk_id = bc->sbc_metadata->ssdi.ssd_chunk_id; 1215 1216 /* Handle key disks separately. */ 1217 if (bc->sbc_metadata->ssdi.ssd_level == SR_KEYDISK_LEVEL) { 1218 SLIST_INSERT_HEAD(&kdh, bc, sbc_link); 1219 continue; 1220 } 1221 1222 SLIST_FOREACH(bv, &bvh, sbv_link) { 1223 if (bcmp(&bc->sbc_metadata->ssdi.ssd_uuid, 1224 &bv->sbv_uuid, 1225 sizeof(bc->sbc_metadata->ssdi.ssd_uuid)) == 0) 1226 break; 1227 } 1228 1229 if (bv == NULL) { 1230 bv = malloc(sizeof(struct sr_boot_volume), 1231 M_DEVBUF, M_NOWAIT | M_ZERO); 1232 if (bv == NULL) { 1233 printf("%s: failed to allocate boot volume\n", 1234 DEVNAME(sc)); 1235 goto unwind; 1236 } 1237 1238 bv->sbv_level = bc->sbc_metadata->ssdi.ssd_level; 1239 bv->sbv_volid = bc->sbc_metadata->ssdi.ssd_volid; 1240 bv->sbv_chunk_no = bc->sbc_metadata->ssdi.ssd_chunk_no; 1241 bv->sbv_flags = bc->sbc_metadata->ssdi.ssd_vol_flags; 1242 memcpy(&bv->sbv_uuid, &bc->sbc_metadata->ssdi.ssd_uuid, 1243 sizeof(bc->sbc_metadata->ssdi.ssd_uuid)); 1244 SLIST_INIT(&bv->sbv_chunks); 1245 1246 /* Maintain volume order. */ 1247 bv2 = NULL; 1248 SLIST_FOREACH(bv1, &bvh, sbv_link) { 1249 if (bv1->sbv_volid > bv->sbv_volid) 1250 break; 1251 bv2 = bv1; 1252 } 1253 if (bv2 == NULL) { 1254 DNPRINTF(SR_D_META, "%s: insert volume %u " 1255 "at head\n", DEVNAME(sc), bv->sbv_volid); 1256 SLIST_INSERT_HEAD(&bvh, bv, sbv_link); 1257 } else { 1258 DNPRINTF(SR_D_META, "%s: insert volume %u " 1259 "after %u\n", DEVNAME(sc), bv->sbv_volid, 1260 bv2->sbv_volid); 1261 SLIST_INSERT_AFTER(bv2, bv, sbv_link); 1262 } 1263 } 1264 1265 /* Maintain chunk order. */ 1266 bc2 = NULL; 1267 SLIST_FOREACH(bc1, &bv->sbv_chunks, sbc_link) { 1268 if (bc1->sbc_chunk_id > bc->sbc_chunk_id) 1269 break; 1270 bc2 = bc1; 1271 } 1272 if (bc2 == NULL) { 1273 DNPRINTF(SR_D_META, "%s: volume %u insert chunk %u " 1274 "at head\n", DEVNAME(sc), bv->sbv_volid, 1275 bc->sbc_chunk_id); 1276 SLIST_INSERT_HEAD(&bv->sbv_chunks, bc, sbc_link); 1277 } else { 1278 DNPRINTF(SR_D_META, "%s: volume %u insert chunk %u " 1279 "after %u\n", DEVNAME(sc), bv->sbv_volid, 1280 bc->sbc_chunk_id, bc2->sbc_chunk_id); 1281 SLIST_INSERT_AFTER(bc2, bc, sbc_link); 1282 } 1283 1284 bv->sbv_chunks_found++; 1285 } 1286 1287 /* Allocate memory for device and ondisk version arrays. */ 1288 devs = mallocarray(BIOC_CRMAXLEN, sizeof(dev_t), M_DEVBUF, 1289 M_NOWAIT); 1290 if (devs == NULL) { 1291 printf("%s: failed to allocate device array\n", DEVNAME(sc)); 1292 goto unwind; 1293 } 1294 ondisk = mallocarray(BIOC_CRMAXLEN, sizeof(u_int64_t), M_DEVBUF, 1295 M_NOWAIT); 1296 if (ondisk == NULL) { 1297 printf("%s: failed to allocate ondisk array\n", DEVNAME(sc)); 1298 goto unwind; 1299 } 1300 1301 /* 1302 * Assemble hotspare "volumes". 1303 */ 1304 SLIST_FOREACH(bv, &bvh, sbv_link) { 1305 1306 /* Check if this is a hotspare "volume". */ 1307 if (bv->sbv_level != SR_HOTSPARE_LEVEL || 1308 bv->sbv_chunk_no != 1) 1309 continue; 1310 1311 #ifdef SR_DEBUG 1312 DNPRINTF(SR_D_META, "%s: assembling hotspare volume ", 1313 DEVNAME(sc)); 1314 if (sr_debug & SR_D_META) 1315 sr_uuid_print(&bv->sbv_uuid, 0); 1316 DNPRINTF(SR_D_META, " volid %u with %u chunks\n", 1317 bv->sbv_volid, bv->sbv_chunk_no); 1318 #endif 1319 1320 /* Create hotspare chunk metadata. */ 1321 hotspare = malloc(sizeof(struct sr_chunk), M_DEVBUF, 1322 M_NOWAIT | M_ZERO); 1323 if (hotspare == NULL) { 1324 printf("%s: failed to allocate hotspare\n", 1325 DEVNAME(sc)); 1326 goto unwind; 1327 } 1328 1329 bc = SLIST_FIRST(&bv->sbv_chunks); 1330 sr_meta_getdevname(sc, bc->sbc_mm, devname, sizeof(devname)); 1331 hotspare->src_dev_mm = bc->sbc_mm; 1332 strlcpy(hotspare->src_devname, devname, 1333 sizeof(hotspare->src_devname)); 1334 hotspare->src_size = bc->sbc_metadata->ssdi.ssd_size; 1335 1336 hm = &hotspare->src_meta; 1337 hm->scmi.scm_volid = SR_HOTSPARE_VOLID; 1338 hm->scmi.scm_chunk_id = 0; 1339 hm->scmi.scm_size = bc->sbc_metadata->ssdi.ssd_size; 1340 hm->scmi.scm_coerced_size = bc->sbc_metadata->ssdi.ssd_size; 1341 strlcpy(hm->scmi.scm_devname, devname, 1342 sizeof(hm->scmi.scm_devname)); 1343 memcpy(&hm->scmi.scm_uuid, &bc->sbc_metadata->ssdi.ssd_uuid, 1344 sizeof(struct sr_uuid)); 1345 1346 sr_checksum(sc, hm, &hm->scm_checksum, 1347 sizeof(struct sr_meta_chunk_invariant)); 1348 1349 hm->scm_status = BIOC_SDHOTSPARE; 1350 1351 /* Add chunk to hotspare list. */ 1352 rw_enter_write(&sc->sc_hs_lock); 1353 cl = &sc->sc_hotspare_list; 1354 if (SLIST_EMPTY(cl)) 1355 SLIST_INSERT_HEAD(cl, hotspare, src_link); 1356 else { 1357 SLIST_FOREACH(chunk, cl, src_link) 1358 last = chunk; 1359 SLIST_INSERT_AFTER(last, hotspare, src_link); 1360 } 1361 sc->sc_hotspare_no++; 1362 rw_exit_write(&sc->sc_hs_lock); 1363 1364 } 1365 1366 /* 1367 * Assemble RAID volumes. 1368 */ 1369 SLIST_FOREACH(bv, &bvh, sbv_link) { 1370 1371 bzero(&bcr, sizeof(bcr)); 1372 data = NULL; 1373 1374 /* Check if this is a hotspare "volume". */ 1375 if (bv->sbv_level == SR_HOTSPARE_LEVEL && 1376 bv->sbv_chunk_no == 1) 1377 continue; 1378 1379 /* 1380 * Skip volumes that are marked as no auto assemble, unless 1381 * this was the volume which we actually booted from. 1382 */ 1383 if (bcmp(&sr_bootuuid, &bv->sbv_uuid, sizeof(sr_bootuuid)) != 0) 1384 if (bv->sbv_flags & BIOC_SCNOAUTOASSEMBLE) 1385 continue; 1386 1387 #ifdef SR_DEBUG 1388 DNPRINTF(SR_D_META, "%s: assembling volume ", DEVNAME(sc)); 1389 if (sr_debug & SR_D_META) 1390 sr_uuid_print(&bv->sbv_uuid, 0); 1391 DNPRINTF(SR_D_META, " volid %u with %u chunks\n", 1392 bv->sbv_volid, bv->sbv_chunk_no); 1393 #endif 1394 1395 /* 1396 * If this is a crypto volume, try to find a matching 1397 * key disk... 1398 */ 1399 bcr.bc_key_disk = NODEV; 1400 if (bv->sbv_level == 'C') { 1401 SLIST_FOREACH(bc, &kdh, sbc_link) { 1402 if (bcmp(&bc->sbc_metadata->ssdi.ssd_uuid, 1403 &bv->sbv_uuid, 1404 sizeof(bc->sbc_metadata->ssdi.ssd_uuid)) 1405 == 0) 1406 bcr.bc_key_disk = bc->sbc_mm; 1407 } 1408 } 1409 1410 for (i = 0; i < BIOC_CRMAXLEN; i++) { 1411 devs[i] = NODEV; /* mark device as illegal */ 1412 ondisk[i] = 0; 1413 } 1414 1415 SLIST_FOREACH(bc, &bv->sbv_chunks, sbc_link) { 1416 if (devs[bc->sbc_chunk_id] != NODEV) { 1417 bv->sbv_chunks_found--; 1418 sr_meta_getdevname(sc, bc->sbc_mm, devname, 1419 sizeof(devname)); 1420 printf("%s: found duplicate chunk %u for " 1421 "volume %u on device %s\n", DEVNAME(sc), 1422 bc->sbc_chunk_id, bv->sbv_volid, devname); 1423 } 1424 1425 if (devs[bc->sbc_chunk_id] == NODEV || 1426 bc->sbc_metadata->ssd_ondisk > 1427 ondisk[bc->sbc_chunk_id]) { 1428 devs[bc->sbc_chunk_id] = bc->sbc_mm; 1429 ondisk[bc->sbc_chunk_id] = 1430 bc->sbc_metadata->ssd_ondisk; 1431 DNPRINTF(SR_D_META, "%s: using ondisk " 1432 "metadata version %llu for chunk %u\n", 1433 DEVNAME(sc), ondisk[bc->sbc_chunk_id], 1434 bc->sbc_chunk_id); 1435 } 1436 } 1437 1438 if (bv->sbv_chunk_no != bv->sbv_chunks_found) { 1439 printf("%s: not all chunks were provided; " 1440 "attempting to bring volume %d online\n", 1441 DEVNAME(sc), bv->sbv_volid); 1442 } 1443 1444 bcr.bc_level = bv->sbv_level; 1445 bcr.bc_dev_list_len = bv->sbv_chunk_no * sizeof(dev_t); 1446 bcr.bc_dev_list = devs; 1447 bcr.bc_flags = BIOC_SCDEVT | 1448 (bv->sbv_flags & BIOC_SCNOAUTOASSEMBLE); 1449 1450 if (bv->sbv_level == 'C' && 1451 bcmp(&sr_bootuuid, &bv->sbv_uuid, sizeof(sr_bootuuid)) == 0) 1452 data = sr_bootkey; 1453 1454 rw_enter_write(&sc->sc_lock); 1455 bio_status_init(&sc->sc_status, &sc->sc_dev); 1456 sr_ioctl_createraid(sc, &bcr, 0, data); 1457 rw_exit_write(&sc->sc_lock); 1458 1459 rv++; 1460 } 1461 1462 /* done with metadata */ 1463 unwind: 1464 /* Free boot volumes and associated chunks. */ 1465 for (bv1 = SLIST_FIRST(&bvh); bv1 != NULL; bv1 = bv2) { 1466 bv2 = SLIST_NEXT(bv1, sbv_link); 1467 for (bc1 = SLIST_FIRST(&bv1->sbv_chunks); bc1 != NULL; 1468 bc1 = bc2) { 1469 bc2 = SLIST_NEXT(bc1, sbc_link); 1470 free(bc1->sbc_metadata, M_DEVBUF, 1471 sizeof(*bc1->sbc_metadata)); 1472 free(bc1, M_DEVBUF, sizeof(*bc1)); 1473 } 1474 free(bv1, M_DEVBUF, sizeof(*bv1)); 1475 } 1476 /* Free keydisks chunks. */ 1477 for (bc1 = SLIST_FIRST(&kdh); bc1 != NULL; bc1 = bc2) { 1478 bc2 = SLIST_NEXT(bc1, sbc_link); 1479 free(bc1->sbc_metadata, M_DEVBUF, sizeof(*bc1->sbc_metadata)); 1480 free(bc1, M_DEVBUF, sizeof(*bc1)); 1481 } 1482 /* Free unallocated chunks. */ 1483 for (bc1 = SLIST_FIRST(&bch); bc1 != NULL; bc1 = bc2) { 1484 bc2 = SLIST_NEXT(bc1, sbc_link); 1485 free(bc1->sbc_metadata, M_DEVBUF, sizeof(*bc1->sbc_metadata)); 1486 free(bc1, M_DEVBUF, sizeof(*bc1)); 1487 } 1488 1489 while (!SLIST_EMPTY(&sdklist)) { 1490 sdk = SLIST_FIRST(&sdklist); 1491 SLIST_REMOVE_HEAD(&sdklist, sdk_link); 1492 free(sdk, M_DEVBUF, sizeof(*sdk)); 1493 } 1494 1495 free(devs, M_DEVBUF, BIOC_CRMAXLEN * sizeof(dev_t)); 1496 free(ondisk, M_DEVBUF, BIOC_CRMAXLEN * sizeof(u_int64_t)); 1497 1498 return (rv); 1499 } 1500 1501 void 1502 sr_map_root(void) 1503 { 1504 struct sr_softc *sc = softraid0; 1505 struct sr_discipline *sd; 1506 struct sr_meta_opt_item *omi; 1507 struct sr_meta_boot *sbm; 1508 u_char duid[8]; 1509 int i; 1510 1511 DNPRINTF(SR_D_MISC, "%s: sr_map_root\n", DEVNAME(sc)); 1512 1513 if (sc == NULL) 1514 return; 1515 1516 bzero(duid, sizeof(duid)); 1517 if (bcmp(rootduid, duid, sizeof(duid)) == 0) { 1518 DNPRINTF(SR_D_MISC, "%s: root duid is zero\n", DEVNAME(sc)); 1519 return; 1520 } 1521 1522 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 1523 SLIST_FOREACH(omi, &sd->sd_meta_opt, omi_link) { 1524 if (omi->omi_som->som_type != SR_OPT_BOOT) 1525 continue; 1526 sbm = (struct sr_meta_boot *)omi->omi_som; 1527 for (i = 0; i < SR_MAX_BOOT_DISKS; i++) { 1528 if (bcmp(rootduid, sbm->sbm_boot_duid[i], 1529 sizeof(rootduid)) == 0) { 1530 memcpy(rootduid, sbm->sbm_root_duid, 1531 sizeof(rootduid)); 1532 DNPRINTF(SR_D_MISC, "%s: root duid " 1533 "mapped to %s\n", DEVNAME(sc), 1534 duid_format(rootduid)); 1535 return; 1536 } 1537 } 1538 } 1539 } 1540 } 1541 1542 int 1543 sr_meta_native_probe(struct sr_softc *sc, struct sr_chunk *ch_entry) 1544 { 1545 struct disklabel label; 1546 char *devname; 1547 int error, part; 1548 u_int64_t size; 1549 1550 DNPRINTF(SR_D_META, "%s: sr_meta_native_probe(%s)\n", 1551 DEVNAME(sc), ch_entry->src_devname); 1552 1553 devname = ch_entry->src_devname; 1554 part = DISKPART(ch_entry->src_dev_mm); 1555 1556 /* get disklabel */ 1557 error = VOP_IOCTL(ch_entry->src_vn, DIOCGDINFO, (caddr_t)&label, FREAD, 1558 NOCRED, curproc); 1559 if (error) { 1560 DNPRINTF(SR_D_META, "%s: %s can't obtain disklabel\n", 1561 DEVNAME(sc), devname); 1562 goto unwind; 1563 } 1564 memcpy(ch_entry->src_duid, label.d_uid, sizeof(ch_entry->src_duid)); 1565 1566 /* make sure the partition is of the right type */ 1567 if (label.d_partitions[part].p_fstype != FS_RAID) { 1568 DNPRINTF(SR_D_META, 1569 "%s: %s partition not of type RAID (%d)\n", DEVNAME(sc), 1570 devname, 1571 label.d_partitions[part].p_fstype); 1572 goto unwind; 1573 } 1574 1575 size = DL_SECTOBLK(&label, DL_GETPSIZE(&label.d_partitions[part])); 1576 if (size <= SR_DATA_OFFSET) { 1577 DNPRINTF(SR_D_META, "%s: %s partition too small\n", DEVNAME(sc), 1578 devname); 1579 goto unwind; 1580 } 1581 size -= SR_DATA_OFFSET; 1582 if (size > INT64_MAX) { 1583 DNPRINTF(SR_D_META, "%s: %s partition too large\n", DEVNAME(sc), 1584 devname); 1585 goto unwind; 1586 } 1587 ch_entry->src_size = size; 1588 ch_entry->src_secsize = label.d_secsize; 1589 1590 DNPRINTF(SR_D_META, "%s: probe found %s size %lld\n", DEVNAME(sc), 1591 devname, (long long)size); 1592 1593 return (SR_META_F_NATIVE); 1594 unwind: 1595 DNPRINTF(SR_D_META, "%s: invalid device: %s\n", DEVNAME(sc), 1596 devname ? devname : "nodev"); 1597 return (SR_META_F_INVALID); 1598 } 1599 1600 int 1601 sr_meta_native_attach(struct sr_discipline *sd, int force) 1602 { 1603 struct sr_softc *sc = sd->sd_sc; 1604 struct sr_chunk_head *cl = &sd->sd_vol.sv_chunk_list; 1605 struct sr_metadata *md = NULL; 1606 struct sr_chunk *ch_entry, *ch_next; 1607 struct sr_uuid uuid; 1608 u_int64_t version = 0; 1609 int sr, not_sr, rv = 1, d, expected = -1, old_meta = 0; 1610 1611 DNPRINTF(SR_D_META, "%s: sr_meta_native_attach\n", DEVNAME(sc)); 1612 1613 md = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, M_ZERO | M_NOWAIT); 1614 if (md == NULL) { 1615 sr_error(sc, "not enough memory for metadata buffer"); 1616 goto bad; 1617 } 1618 1619 bzero(&uuid, sizeof uuid); 1620 1621 sr = not_sr = d = 0; 1622 SLIST_FOREACH(ch_entry, cl, src_link) { 1623 if (ch_entry->src_dev_mm == NODEV) 1624 continue; 1625 1626 if (sr_meta_native_read(sd, ch_entry->src_dev_mm, md, NULL)) { 1627 sr_error(sc, "could not read native metadata"); 1628 goto bad; 1629 } 1630 1631 if (md->ssdi.ssd_magic == SR_MAGIC) { 1632 sr++; 1633 ch_entry->src_meta.scmi.scm_chunk_id = 1634 md->ssdi.ssd_chunk_id; 1635 if (d == 0) { 1636 memcpy(&uuid, &md->ssdi.ssd_uuid, sizeof uuid); 1637 expected = md->ssdi.ssd_chunk_no; 1638 version = md->ssd_ondisk; 1639 d++; 1640 continue; 1641 } else if (bcmp(&md->ssdi.ssd_uuid, &uuid, 1642 sizeof uuid)) { 1643 sr_error(sc, "not part of the same volume"); 1644 goto bad; 1645 } 1646 if (md->ssd_ondisk != version) { 1647 old_meta++; 1648 version = MAX(md->ssd_ondisk, version); 1649 } 1650 } else 1651 not_sr++; 1652 } 1653 1654 if (sr && not_sr && !force) { 1655 sr_error(sc, "not all chunks are of the native metadata " 1656 "format"); 1657 goto bad; 1658 } 1659 1660 /* mixed metadata versions; mark bad disks offline */ 1661 if (old_meta) { 1662 d = 0; 1663 for (ch_entry = SLIST_FIRST(cl); ch_entry != NULL; 1664 ch_entry = ch_next, d++) { 1665 ch_next = SLIST_NEXT(ch_entry, src_link); 1666 1667 /* XXX do we want to read this again? */ 1668 if (ch_entry->src_dev_mm == NODEV) 1669 panic("src_dev_mm == NODEV"); 1670 if (sr_meta_native_read(sd, ch_entry->src_dev_mm, md, 1671 NULL)) 1672 sr_warn(sc, "could not read native metadata"); 1673 if (md->ssd_ondisk != version) 1674 sd->sd_vol.sv_chunks[d]->src_meta.scm_status = 1675 BIOC_SDOFFLINE; 1676 } 1677 } 1678 1679 if (expected != sr && !force && expected != -1) { 1680 DNPRINTF(SR_D_META, "%s: not all chunks were provided, trying " 1681 "anyway\n", DEVNAME(sc)); 1682 } 1683 1684 rv = 0; 1685 bad: 1686 free(md, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 1687 return (rv); 1688 } 1689 1690 int 1691 sr_meta_native_read(struct sr_discipline *sd, dev_t dev, 1692 struct sr_metadata *md, void *fm) 1693 { 1694 #ifdef SR_DEBUG 1695 struct sr_softc *sc = sd->sd_sc; 1696 #endif 1697 DNPRINTF(SR_D_META, "%s: sr_meta_native_read(0x%x, %p)\n", 1698 DEVNAME(sc), dev, md); 1699 1700 return (sr_meta_rw(sd, dev, md, B_READ)); 1701 } 1702 1703 int 1704 sr_meta_native_write(struct sr_discipline *sd, dev_t dev, 1705 struct sr_metadata *md, void *fm) 1706 { 1707 #ifdef SR_DEBUG 1708 struct sr_softc *sc = sd->sd_sc; 1709 #endif 1710 DNPRINTF(SR_D_META, "%s: sr_meta_native_write(0x%x, %p)\n", 1711 DEVNAME(sc), dev, md); 1712 1713 return (sr_meta_rw(sd, dev, md, B_WRITE)); 1714 } 1715 1716 void 1717 sr_hotplug_register(struct sr_discipline *sd, void *func) 1718 { 1719 struct sr_hotplug_list *mhe; 1720 1721 DNPRINTF(SR_D_MISC, "%s: sr_hotplug_register: %p\n", 1722 DEVNAME(sd->sd_sc), func); 1723 1724 /* make sure we aren't on the list yet */ 1725 SLIST_FOREACH(mhe, &sr_hotplug_callbacks, shl_link) 1726 if (mhe->sh_hotplug == func) 1727 return; 1728 1729 mhe = malloc(sizeof(struct sr_hotplug_list), M_DEVBUF, 1730 M_WAITOK | M_ZERO); 1731 mhe->sh_hotplug = func; 1732 mhe->sh_sd = sd; 1733 SLIST_INSERT_HEAD(&sr_hotplug_callbacks, mhe, shl_link); 1734 } 1735 1736 void 1737 sr_hotplug_unregister(struct sr_discipline *sd, void *func) 1738 { 1739 struct sr_hotplug_list *mhe; 1740 1741 DNPRINTF(SR_D_MISC, "%s: sr_hotplug_unregister: %s %p\n", 1742 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, func); 1743 1744 /* make sure we are on the list yet */ 1745 SLIST_FOREACH(mhe, &sr_hotplug_callbacks, shl_link) { 1746 if (mhe->sh_hotplug == func) 1747 break; 1748 } 1749 if (mhe != NULL) { 1750 SLIST_REMOVE(&sr_hotplug_callbacks, mhe, 1751 sr_hotplug_list, shl_link); 1752 free(mhe, M_DEVBUF, sizeof(*mhe)); 1753 } 1754 } 1755 1756 void 1757 sr_disk_attach(struct disk *diskp, int action) 1758 { 1759 struct sr_hotplug_list *mhe; 1760 1761 SLIST_FOREACH(mhe, &sr_hotplug_callbacks, shl_link) 1762 if (mhe->sh_sd->sd_ready) 1763 mhe->sh_hotplug(mhe->sh_sd, diskp, action); 1764 } 1765 1766 int 1767 sr_match(struct device *parent, void *match, void *aux) 1768 { 1769 return (1); 1770 } 1771 1772 void 1773 sr_attach(struct device *parent, struct device *self, void *aux) 1774 { 1775 struct sr_softc *sc = (void *)self; 1776 struct scsibus_attach_args saa; 1777 1778 DNPRINTF(SR_D_MISC, "\n%s: sr_attach", DEVNAME(sc)); 1779 1780 if (softraid0 == NULL) 1781 softraid0 = sc; 1782 1783 rw_init(&sc->sc_lock, "sr_lock"); 1784 rw_init(&sc->sc_hs_lock, "sr_hs_lock"); 1785 1786 SLIST_INIT(&sr_hotplug_callbacks); 1787 TAILQ_INIT(&sc->sc_dis_list); 1788 SLIST_INIT(&sc->sc_hotspare_list); 1789 1790 #if NBIO > 0 1791 if (bio_register(&sc->sc_dev, sr_bio_ioctl) != 0) 1792 printf("%s: controller registration failed", DEVNAME(sc)); 1793 #endif /* NBIO > 0 */ 1794 1795 #ifndef SMALL_KERNEL 1796 strlcpy(sc->sc_sensordev.xname, DEVNAME(sc), 1797 sizeof(sc->sc_sensordev.xname)); 1798 sensordev_install(&sc->sc_sensordev); 1799 #endif /* SMALL_KERNEL */ 1800 1801 printf("\n"); 1802 1803 saa.saa_adapter_softc = sc; 1804 saa.saa_adapter = &sr_switch; 1805 saa.saa_adapter_target = SDEV_NO_ADAPTER_TARGET; 1806 saa.saa_adapter_buswidth = SR_MAX_LD; 1807 saa.saa_luns = 1; 1808 saa.saa_openings = 0; 1809 saa.saa_pool = NULL; 1810 saa.saa_quirks = saa.saa_flags = 0; 1811 saa.saa_wwpn = saa.saa_wwnn = 0; 1812 1813 sc->sc_scsibus = (struct scsibus_softc *)config_found(&sc->sc_dev, &saa, 1814 scsiprint); 1815 1816 softraid_disk_attach = sr_disk_attach; 1817 1818 sr_boot_assembly(sc); 1819 1820 explicit_bzero(sr_bootkey, sizeof(sr_bootkey)); 1821 } 1822 1823 int 1824 sr_detach(struct device *self, int flags) 1825 { 1826 struct sr_softc *sc = (void *)self; 1827 int rv; 1828 1829 DNPRINTF(SR_D_MISC, "%s: sr_detach\n", DEVNAME(sc)); 1830 1831 softraid_disk_attach = NULL; 1832 1833 sr_shutdown(0); 1834 1835 #ifndef SMALL_KERNEL 1836 if (sc->sc_sensor_task != NULL) 1837 sensor_task_unregister(sc->sc_sensor_task); 1838 sensordev_deinstall(&sc->sc_sensordev); 1839 #endif /* SMALL_KERNEL */ 1840 1841 if (sc->sc_scsibus != NULL) { 1842 rv = config_detach((struct device *)sc->sc_scsibus, flags); 1843 if (rv != 0) 1844 return (rv); 1845 sc->sc_scsibus = NULL; 1846 } 1847 1848 return (0); 1849 } 1850 1851 void 1852 sr_info(struct sr_softc *sc, const char *fmt, ...) 1853 { 1854 va_list ap; 1855 1856 rw_assert_wrlock(&sc->sc_lock); 1857 1858 va_start(ap, fmt); 1859 bio_status(&sc->sc_status, 0, BIO_MSG_INFO, fmt, &ap); 1860 va_end(ap); 1861 } 1862 1863 void 1864 sr_warn(struct sr_softc *sc, const char *fmt, ...) 1865 { 1866 va_list ap; 1867 1868 rw_assert_wrlock(&sc->sc_lock); 1869 1870 va_start(ap, fmt); 1871 bio_status(&sc->sc_status, 1, BIO_MSG_WARN, fmt, &ap); 1872 va_end(ap); 1873 } 1874 1875 void 1876 sr_error(struct sr_softc *sc, const char *fmt, ...) 1877 { 1878 va_list ap; 1879 1880 rw_assert_wrlock(&sc->sc_lock); 1881 1882 va_start(ap, fmt); 1883 bio_status(&sc->sc_status, 1, BIO_MSG_ERROR, fmt, &ap); 1884 va_end(ap); 1885 } 1886 1887 int 1888 sr_ccb_alloc(struct sr_discipline *sd) 1889 { 1890 struct sr_ccb *ccb; 1891 int i; 1892 1893 if (!sd) 1894 return (1); 1895 1896 DNPRINTF(SR_D_CCB, "%s: sr_ccb_alloc\n", DEVNAME(sd->sd_sc)); 1897 1898 if (sd->sd_ccb) 1899 return (1); 1900 1901 sd->sd_ccb = mallocarray(sd->sd_max_wu, 1902 sd->sd_max_ccb_per_wu * sizeof(struct sr_ccb), 1903 M_DEVBUF, M_WAITOK | M_ZERO); 1904 TAILQ_INIT(&sd->sd_ccb_freeq); 1905 for (i = 0; i < sd->sd_max_wu * sd->sd_max_ccb_per_wu; i++) { 1906 ccb = &sd->sd_ccb[i]; 1907 ccb->ccb_dis = sd; 1908 sr_ccb_put(ccb); 1909 } 1910 1911 DNPRINTF(SR_D_CCB, "%s: sr_ccb_alloc ccb: %d\n", 1912 DEVNAME(sd->sd_sc), sd->sd_max_wu * sd->sd_max_ccb_per_wu); 1913 1914 return (0); 1915 } 1916 1917 void 1918 sr_ccb_free(struct sr_discipline *sd) 1919 { 1920 struct sr_ccb *ccb; 1921 1922 if (!sd) 1923 return; 1924 1925 DNPRINTF(SR_D_CCB, "%s: sr_ccb_free %p\n", DEVNAME(sd->sd_sc), sd); 1926 1927 while ((ccb = TAILQ_FIRST(&sd->sd_ccb_freeq)) != NULL) 1928 TAILQ_REMOVE(&sd->sd_ccb_freeq, ccb, ccb_link); 1929 1930 free(sd->sd_ccb, M_DEVBUF, sd->sd_max_wu * sd->sd_max_ccb_per_wu * 1931 sizeof(struct sr_ccb)); 1932 } 1933 1934 struct sr_ccb * 1935 sr_ccb_get(struct sr_discipline *sd) 1936 { 1937 struct sr_ccb *ccb; 1938 int s; 1939 1940 s = splbio(); 1941 1942 ccb = TAILQ_FIRST(&sd->sd_ccb_freeq); 1943 if (ccb) { 1944 TAILQ_REMOVE(&sd->sd_ccb_freeq, ccb, ccb_link); 1945 ccb->ccb_state = SR_CCB_INPROGRESS; 1946 } 1947 1948 splx(s); 1949 1950 DNPRINTF(SR_D_CCB, "%s: sr_ccb_get: %p\n", DEVNAME(sd->sd_sc), 1951 ccb); 1952 1953 return (ccb); 1954 } 1955 1956 void 1957 sr_ccb_put(struct sr_ccb *ccb) 1958 { 1959 struct sr_discipline *sd = ccb->ccb_dis; 1960 int s; 1961 1962 DNPRINTF(SR_D_CCB, "%s: sr_ccb_put: %p\n", DEVNAME(sd->sd_sc), 1963 ccb); 1964 1965 s = splbio(); 1966 1967 ccb->ccb_wu = NULL; 1968 ccb->ccb_state = SR_CCB_FREE; 1969 ccb->ccb_target = -1; 1970 ccb->ccb_opaque = NULL; 1971 1972 TAILQ_INSERT_TAIL(&sd->sd_ccb_freeq, ccb, ccb_link); 1973 1974 splx(s); 1975 } 1976 1977 struct sr_ccb * 1978 sr_ccb_rw(struct sr_discipline *sd, int chunk, daddr_t blkno, 1979 long len, u_int8_t *data, int xsflags, int ccbflags) 1980 { 1981 struct sr_chunk *sc = sd->sd_vol.sv_chunks[chunk]; 1982 struct sr_ccb *ccb = NULL; 1983 1984 ccb = sr_ccb_get(sd); 1985 if (ccb == NULL) 1986 goto out; 1987 1988 ccb->ccb_flags = ccbflags; 1989 ccb->ccb_target = chunk; 1990 1991 ccb->ccb_buf.b_flags = B_PHYS | B_CALL; 1992 if (ISSET(xsflags, SCSI_DATA_IN)) 1993 ccb->ccb_buf.b_flags |= B_READ; 1994 else 1995 ccb->ccb_buf.b_flags |= B_WRITE; 1996 1997 ccb->ccb_buf.b_blkno = blkno + sd->sd_meta->ssd_data_blkno; 1998 ccb->ccb_buf.b_bcount = len; 1999 ccb->ccb_buf.b_bufsize = len; 2000 ccb->ccb_buf.b_resid = len; 2001 ccb->ccb_buf.b_data = data; 2002 ccb->ccb_buf.b_error = 0; 2003 ccb->ccb_buf.b_iodone = sd->sd_scsi_intr; 2004 ccb->ccb_buf.b_proc = curproc; 2005 ccb->ccb_buf.b_dev = sc->src_dev_mm; 2006 ccb->ccb_buf.b_vp = sc->src_vn; 2007 ccb->ccb_buf.b_bq = NULL; 2008 2009 if (!ISSET(ccb->ccb_buf.b_flags, B_READ)) 2010 ccb->ccb_buf.b_vp->v_numoutput++; 2011 2012 LIST_INIT(&ccb->ccb_buf.b_dep); 2013 2014 DNPRINTF(SR_D_DIS, "%s: %s %s ccb " 2015 "b_bcount %ld b_blkno %lld b_flags 0x%0lx b_data %p\n", 2016 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, sd->sd_name, 2017 ccb->ccb_buf.b_bcount, (long long)ccb->ccb_buf.b_blkno, 2018 ccb->ccb_buf.b_flags, ccb->ccb_buf.b_data); 2019 2020 out: 2021 return ccb; 2022 } 2023 2024 void 2025 sr_ccb_done(struct sr_ccb *ccb) 2026 { 2027 struct sr_workunit *wu = ccb->ccb_wu; 2028 struct sr_discipline *sd = wu->swu_dis; 2029 struct sr_softc *sc = sd->sd_sc; 2030 2031 DNPRINTF(SR_D_INTR, "%s: %s %s ccb done b_bcount %ld b_resid %zu" 2032 " b_flags 0x%0lx block %lld target %d\n", 2033 DEVNAME(sc), sd->sd_meta->ssd_devname, sd->sd_name, 2034 ccb->ccb_buf.b_bcount, ccb->ccb_buf.b_resid, ccb->ccb_buf.b_flags, 2035 (long long)ccb->ccb_buf.b_blkno, ccb->ccb_target); 2036 2037 splassert(IPL_BIO); 2038 2039 if (ccb->ccb_target == -1) 2040 panic("%s: invalid target on wu: %p", DEVNAME(sc), wu); 2041 2042 if (ccb->ccb_buf.b_flags & B_ERROR) { 2043 DNPRINTF(SR_D_INTR, "%s: i/o error on block %lld target %d\n", 2044 DEVNAME(sc), (long long)ccb->ccb_buf.b_blkno, 2045 ccb->ccb_target); 2046 if (ISSET(sd->sd_capabilities, SR_CAP_REDUNDANT)) 2047 sd->sd_set_chunk_state(sd, ccb->ccb_target, 2048 BIOC_SDOFFLINE); 2049 else 2050 printf("%s: %s: i/o error %d @ %s block %lld\n", 2051 DEVNAME(sc), sd->sd_meta->ssd_devname, 2052 ccb->ccb_buf.b_error, sd->sd_name, 2053 (long long)ccb->ccb_buf.b_blkno); 2054 ccb->ccb_state = SR_CCB_FAILED; 2055 wu->swu_ios_failed++; 2056 } else { 2057 ccb->ccb_state = SR_CCB_OK; 2058 wu->swu_ios_succeeded++; 2059 } 2060 2061 wu->swu_ios_complete++; 2062 } 2063 2064 int 2065 sr_wu_alloc(struct sr_discipline *sd) 2066 { 2067 struct sr_workunit *wu; 2068 int i, no_wu; 2069 2070 DNPRINTF(SR_D_WU, "%s: sr_wu_alloc %p %d\n", DEVNAME(sd->sd_sc), 2071 sd, sd->sd_max_wu); 2072 2073 no_wu = sd->sd_max_wu; 2074 sd->sd_wu_pending = no_wu; 2075 2076 mtx_init(&sd->sd_wu_mtx, IPL_BIO); 2077 TAILQ_INIT(&sd->sd_wu); 2078 TAILQ_INIT(&sd->sd_wu_freeq); 2079 TAILQ_INIT(&sd->sd_wu_pendq); 2080 TAILQ_INIT(&sd->sd_wu_defq); 2081 2082 for (i = 0; i < no_wu; i++) { 2083 wu = malloc(sd->sd_wu_size, M_DEVBUF, M_WAITOK | M_ZERO); 2084 TAILQ_INSERT_TAIL(&sd->sd_wu, wu, swu_next); 2085 TAILQ_INIT(&wu->swu_ccb); 2086 wu->swu_dis = sd; 2087 task_set(&wu->swu_task, sr_wu_done_callback, wu); 2088 sr_wu_put(sd, wu); 2089 } 2090 2091 return (0); 2092 } 2093 2094 void 2095 sr_wu_free(struct sr_discipline *sd) 2096 { 2097 struct sr_workunit *wu; 2098 2099 DNPRINTF(SR_D_WU, "%s: sr_wu_free %p\n", DEVNAME(sd->sd_sc), sd); 2100 2101 while ((wu = TAILQ_FIRST(&sd->sd_wu_freeq)) != NULL) 2102 TAILQ_REMOVE(&sd->sd_wu_freeq, wu, swu_link); 2103 while ((wu = TAILQ_FIRST(&sd->sd_wu_pendq)) != NULL) 2104 TAILQ_REMOVE(&sd->sd_wu_pendq, wu, swu_link); 2105 while ((wu = TAILQ_FIRST(&sd->sd_wu_defq)) != NULL) 2106 TAILQ_REMOVE(&sd->sd_wu_defq, wu, swu_link); 2107 2108 while ((wu = TAILQ_FIRST(&sd->sd_wu)) != NULL) { 2109 TAILQ_REMOVE(&sd->sd_wu, wu, swu_next); 2110 free(wu, M_DEVBUF, sd->sd_wu_size); 2111 } 2112 } 2113 2114 void * 2115 sr_wu_get(void *xsd) 2116 { 2117 struct sr_discipline *sd = (struct sr_discipline *)xsd; 2118 struct sr_workunit *wu; 2119 2120 mtx_enter(&sd->sd_wu_mtx); 2121 wu = TAILQ_FIRST(&sd->sd_wu_freeq); 2122 if (wu) { 2123 TAILQ_REMOVE(&sd->sd_wu_freeq, wu, swu_link); 2124 sd->sd_wu_pending++; 2125 } 2126 mtx_leave(&sd->sd_wu_mtx); 2127 2128 DNPRINTF(SR_D_WU, "%s: sr_wu_get: %p\n", DEVNAME(sd->sd_sc), wu); 2129 2130 return (wu); 2131 } 2132 2133 void 2134 sr_wu_put(void *xsd, void *xwu) 2135 { 2136 struct sr_discipline *sd = (struct sr_discipline *)xsd; 2137 struct sr_workunit *wu = (struct sr_workunit *)xwu; 2138 2139 DNPRINTF(SR_D_WU, "%s: sr_wu_put: %p\n", DEVNAME(sd->sd_sc), wu); 2140 2141 sr_wu_release_ccbs(wu); 2142 sr_wu_init(sd, wu); 2143 2144 mtx_enter(&sd->sd_wu_mtx); 2145 TAILQ_INSERT_TAIL(&sd->sd_wu_freeq, wu, swu_link); 2146 sd->sd_wu_pending--; 2147 mtx_leave(&sd->sd_wu_mtx); 2148 } 2149 2150 void 2151 sr_wu_init(struct sr_discipline *sd, struct sr_workunit *wu) 2152 { 2153 int s; 2154 2155 s = splbio(); 2156 if (wu->swu_cb_active == 1) 2157 panic("%s: sr_wu_init got active wu", DEVNAME(sd->sd_sc)); 2158 splx(s); 2159 2160 wu->swu_xs = NULL; 2161 wu->swu_state = SR_WU_FREE; 2162 wu->swu_flags = 0; 2163 wu->swu_blk_start = 0; 2164 wu->swu_blk_end = 0; 2165 wu->swu_collider = NULL; 2166 } 2167 2168 void 2169 sr_wu_enqueue_ccb(struct sr_workunit *wu, struct sr_ccb *ccb) 2170 { 2171 struct sr_discipline *sd = wu->swu_dis; 2172 int s; 2173 2174 s = splbio(); 2175 if (wu->swu_cb_active == 1) 2176 panic("%s: sr_wu_enqueue_ccb got active wu", 2177 DEVNAME(sd->sd_sc)); 2178 ccb->ccb_wu = wu; 2179 wu->swu_io_count++; 2180 TAILQ_INSERT_TAIL(&wu->swu_ccb, ccb, ccb_link); 2181 splx(s); 2182 } 2183 2184 void 2185 sr_wu_release_ccbs(struct sr_workunit *wu) 2186 { 2187 struct sr_ccb *ccb; 2188 2189 /* Return all ccbs that are associated with this workunit. */ 2190 while ((ccb = TAILQ_FIRST(&wu->swu_ccb)) != NULL) { 2191 TAILQ_REMOVE(&wu->swu_ccb, ccb, ccb_link); 2192 sr_ccb_put(ccb); 2193 } 2194 2195 wu->swu_io_count = 0; 2196 wu->swu_ios_complete = 0; 2197 wu->swu_ios_failed = 0; 2198 wu->swu_ios_succeeded = 0; 2199 } 2200 2201 void 2202 sr_wu_done(struct sr_workunit *wu) 2203 { 2204 struct sr_discipline *sd = wu->swu_dis; 2205 2206 DNPRINTF(SR_D_INTR, "%s: sr_wu_done count %d completed %d failed %d\n", 2207 DEVNAME(sd->sd_sc), wu->swu_io_count, wu->swu_ios_complete, 2208 wu->swu_ios_failed); 2209 2210 if (wu->swu_ios_complete < wu->swu_io_count) 2211 return; 2212 2213 task_add(sd->sd_taskq, &wu->swu_task); 2214 } 2215 2216 void 2217 sr_wu_done_callback(void *xwu) 2218 { 2219 struct sr_workunit *wu = xwu; 2220 struct sr_discipline *sd = wu->swu_dis; 2221 struct scsi_xfer *xs = wu->swu_xs; 2222 struct sr_workunit *wup; 2223 int s; 2224 2225 /* 2226 * The SR_WUF_DISCIPLINE or SR_WUF_REBUILD flag must be set if 2227 * the work unit is not associated with a scsi_xfer. 2228 */ 2229 KASSERT(xs != NULL || 2230 (wu->swu_flags & (SR_WUF_DISCIPLINE|SR_WUF_REBUILD))); 2231 2232 s = splbio(); 2233 2234 if (xs != NULL) { 2235 if (wu->swu_ios_failed) 2236 xs->error = XS_DRIVER_STUFFUP; 2237 else 2238 xs->error = XS_NOERROR; 2239 } 2240 2241 if (sd->sd_scsi_wu_done) { 2242 if (sd->sd_scsi_wu_done(wu) == SR_WU_RESTART) 2243 goto done; 2244 } 2245 2246 /* Remove work unit from pending queue. */ 2247 TAILQ_FOREACH(wup, &sd->sd_wu_pendq, swu_link) 2248 if (wup == wu) 2249 break; 2250 if (wup == NULL) 2251 panic("%s: wu %p not on pending queue", 2252 DEVNAME(sd->sd_sc), wu); 2253 TAILQ_REMOVE(&sd->sd_wu_pendq, wu, swu_link); 2254 2255 if (wu->swu_collider) { 2256 if (wu->swu_ios_failed) 2257 sr_raid_recreate_wu(wu->swu_collider); 2258 2259 /* XXX Should the collider be failed if this xs failed? */ 2260 sr_raid_startwu(wu->swu_collider); 2261 } 2262 2263 /* 2264 * If a discipline provides its own sd_scsi_done function, then it 2265 * is responsible for calling sr_scsi_done() once I/O is complete. 2266 */ 2267 if (wu->swu_flags & SR_WUF_REBUILD) 2268 wu->swu_flags |= SR_WUF_REBUILDIOCOMP; 2269 if (wu->swu_flags & SR_WUF_WAKEUP) 2270 wakeup(wu); 2271 if (sd->sd_scsi_done) 2272 sd->sd_scsi_done(wu); 2273 else if (wu->swu_flags & SR_WUF_DISCIPLINE) 2274 sr_scsi_wu_put(sd, wu); 2275 else if (!(wu->swu_flags & SR_WUF_REBUILD)) 2276 sr_scsi_done(sd, xs); 2277 2278 done: 2279 splx(s); 2280 } 2281 2282 struct sr_workunit * 2283 sr_scsi_wu_get(struct sr_discipline *sd, int flags) 2284 { 2285 return scsi_io_get(&sd->sd_iopool, flags); 2286 } 2287 2288 void 2289 sr_scsi_wu_put(struct sr_discipline *sd, struct sr_workunit *wu) 2290 { 2291 scsi_io_put(&sd->sd_iopool, wu); 2292 2293 if (sd->sd_sync && sd->sd_wu_pending == 0) 2294 wakeup(sd); 2295 } 2296 2297 void 2298 sr_scsi_done(struct sr_discipline *sd, struct scsi_xfer *xs) 2299 { 2300 DNPRINTF(SR_D_DIS, "%s: sr_scsi_done: xs %p\n", DEVNAME(sd->sd_sc), xs); 2301 2302 if (xs->error == XS_NOERROR) 2303 xs->resid = 0; 2304 2305 scsi_done(xs); 2306 2307 if (sd->sd_sync && sd->sd_wu_pending == 0) 2308 wakeup(sd); 2309 } 2310 2311 void 2312 sr_scsi_cmd(struct scsi_xfer *xs) 2313 { 2314 struct scsi_link *link = xs->sc_link; 2315 struct sr_softc *sc = link->bus->sb_adapter_softc; 2316 struct sr_workunit *wu = xs->io; 2317 struct sr_discipline *sd; 2318 2319 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd target %d xs %p flags %#x\n", 2320 DEVNAME(sc), link->target, xs, xs->flags); 2321 2322 sd = sc->sc_targets[link->target]; 2323 if (sd == NULL) 2324 panic("%s: sr_scsi_cmd NULL discipline", DEVNAME(sc)); 2325 2326 if (sd->sd_deleted) { 2327 printf("%s: %s device is being deleted, failing io\n", 2328 DEVNAME(sc), sd->sd_meta->ssd_devname); 2329 goto stuffup; 2330 } 2331 2332 /* scsi layer *can* re-send wu without calling sr_wu_put(). */ 2333 sr_wu_release_ccbs(wu); 2334 sr_wu_init(sd, wu); 2335 wu->swu_state = SR_WU_INPROGRESS; 2336 wu->swu_xs = xs; 2337 2338 switch (xs->cmd.opcode) { 2339 case READ_COMMAND: 2340 case READ_10: 2341 case READ_16: 2342 case WRITE_COMMAND: 2343 case WRITE_10: 2344 case WRITE_16: 2345 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: READ/WRITE %02x\n", 2346 DEVNAME(sc), xs->cmd.opcode); 2347 if (sd->sd_scsi_rw(wu)) 2348 goto stuffup; 2349 break; 2350 2351 case SYNCHRONIZE_CACHE: 2352 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: SYNCHRONIZE_CACHE\n", 2353 DEVNAME(sc)); 2354 if (sd->sd_scsi_sync(wu)) 2355 goto stuffup; 2356 goto complete; 2357 2358 case TEST_UNIT_READY: 2359 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: TEST_UNIT_READY\n", 2360 DEVNAME(sc)); 2361 if (sd->sd_scsi_tur(wu)) 2362 goto stuffup; 2363 goto complete; 2364 2365 case START_STOP: 2366 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: START_STOP\n", 2367 DEVNAME(sc)); 2368 if (sd->sd_scsi_start_stop(wu)) 2369 goto stuffup; 2370 goto complete; 2371 2372 case INQUIRY: 2373 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: INQUIRY\n", 2374 DEVNAME(sc)); 2375 if (sd->sd_scsi_inquiry(wu)) 2376 goto stuffup; 2377 goto complete; 2378 2379 case READ_CAPACITY: 2380 case READ_CAPACITY_16: 2381 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd READ CAPACITY 0x%02x\n", 2382 DEVNAME(sc), xs->cmd.opcode); 2383 if (sd->sd_scsi_read_cap(wu)) 2384 goto stuffup; 2385 goto complete; 2386 2387 case REQUEST_SENSE: 2388 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd REQUEST SENSE\n", 2389 DEVNAME(sc)); 2390 if (sd->sd_scsi_req_sense(wu)) 2391 goto stuffup; 2392 goto complete; 2393 2394 default: 2395 DNPRINTF(SR_D_CMD, "%s: unsupported scsi command %x\n", 2396 DEVNAME(sc), xs->cmd.opcode); 2397 /* XXX might need to add generic function to handle others */ 2398 goto stuffup; 2399 } 2400 2401 return; 2402 stuffup: 2403 if (sd->sd_scsi_sense.error_code) { 2404 xs->error = XS_SENSE; 2405 memcpy(&xs->sense, &sd->sd_scsi_sense, sizeof(xs->sense)); 2406 bzero(&sd->sd_scsi_sense, sizeof(sd->sd_scsi_sense)); 2407 } else { 2408 xs->error = XS_DRIVER_STUFFUP; 2409 } 2410 complete: 2411 sr_scsi_done(sd, xs); 2412 } 2413 2414 int 2415 sr_scsi_probe(struct scsi_link *link) 2416 { 2417 struct sr_softc *sc = link->bus->sb_adapter_softc; 2418 struct sr_discipline *sd; 2419 2420 KASSERT(link->target < SR_MAX_LD && link->lun == 0); 2421 2422 sd = sc->sc_targets[link->target]; 2423 if (sd == NULL) 2424 return (ENODEV); 2425 2426 link->pool = &sd->sd_iopool; 2427 if (sd->sd_openings) 2428 link->openings = sd->sd_openings(sd); 2429 else 2430 link->openings = sd->sd_max_wu; 2431 2432 return (0); 2433 } 2434 2435 int 2436 sr_scsi_ioctl(struct scsi_link *link, u_long cmd, caddr_t addr, int flag) 2437 { 2438 struct sr_softc *sc = link->bus->sb_adapter_softc; 2439 struct sr_discipline *sd; 2440 2441 sd = sc->sc_targets[link->target]; 2442 if (sd == NULL) 2443 return (ENODEV); 2444 2445 DNPRINTF(SR_D_IOCTL, "%s: %s sr_scsi_ioctl cmd: %#lx\n", 2446 DEVNAME(sc), sd->sd_meta->ssd_devname, cmd); 2447 2448 /* Pass bio ioctls through to the bio handler. */ 2449 if (IOCGROUP(cmd) == 'B') 2450 return (sr_bio_handler(sc, sd, cmd, (struct bio *)addr)); 2451 2452 switch (cmd) { 2453 case DIOCGCACHE: 2454 case DIOCSCACHE: 2455 return (EOPNOTSUPP); 2456 default: 2457 return (ENOTTY); 2458 } 2459 } 2460 2461 int 2462 sr_bio_ioctl(struct device *dev, u_long cmd, caddr_t addr) 2463 { 2464 struct sr_softc *sc = (struct sr_softc *) dev; 2465 DNPRINTF(SR_D_IOCTL, "%s: sr_bio_ioctl\n", DEVNAME(sc)); 2466 2467 return sr_bio_handler(sc, NULL, cmd, (struct bio *)addr); 2468 } 2469 2470 int 2471 sr_bio_handler(struct sr_softc *sc, struct sr_discipline *sd, u_long cmd, 2472 struct bio *bio) 2473 { 2474 int rv = 0; 2475 2476 DNPRINTF(SR_D_IOCTL, "%s: sr_bio_handler ", DEVNAME(sc)); 2477 2478 rw_enter_write(&sc->sc_lock); 2479 2480 bio_status_init(&sc->sc_status, &sc->sc_dev); 2481 2482 switch (cmd) { 2483 case BIOCINQ: 2484 DNPRINTF(SR_D_IOCTL, "inq\n"); 2485 rv = sr_ioctl_inq(sc, (struct bioc_inq *)bio); 2486 break; 2487 2488 case BIOCVOL: 2489 DNPRINTF(SR_D_IOCTL, "vol\n"); 2490 rv = sr_ioctl_vol(sc, (struct bioc_vol *)bio); 2491 break; 2492 2493 case BIOCDISK: 2494 DNPRINTF(SR_D_IOCTL, "disk\n"); 2495 rv = sr_ioctl_disk(sc, (struct bioc_disk *)bio); 2496 break; 2497 2498 case BIOCALARM: 2499 DNPRINTF(SR_D_IOCTL, "alarm\n"); 2500 /*rv = sr_ioctl_alarm(sc, (struct bioc_alarm *)bio); */ 2501 break; 2502 2503 case BIOCBLINK: 2504 DNPRINTF(SR_D_IOCTL, "blink\n"); 2505 /*rv = sr_ioctl_blink(sc, (struct bioc_blink *)bio); */ 2506 break; 2507 2508 case BIOCSETSTATE: 2509 DNPRINTF(SR_D_IOCTL, "setstate\n"); 2510 rv = sr_ioctl_setstate(sc, (struct bioc_setstate *)bio); 2511 break; 2512 2513 case BIOCCREATERAID: 2514 DNPRINTF(SR_D_IOCTL, "createraid\n"); 2515 rv = sr_ioctl_createraid(sc, (struct bioc_createraid *)bio, 2516 1, NULL); 2517 break; 2518 2519 case BIOCDELETERAID: 2520 DNPRINTF(SR_D_IOCTL, "deleteraid\n"); 2521 rv = sr_ioctl_deleteraid(sc, sd, (struct bioc_deleteraid *)bio); 2522 break; 2523 2524 case BIOCDISCIPLINE: 2525 DNPRINTF(SR_D_IOCTL, "discipline\n"); 2526 rv = sr_ioctl_discipline(sc, sd, (struct bioc_discipline *)bio); 2527 break; 2528 2529 case BIOCINSTALLBOOT: 2530 DNPRINTF(SR_D_IOCTL, "installboot\n"); 2531 rv = sr_ioctl_installboot(sc, sd, 2532 (struct bioc_installboot *)bio); 2533 break; 2534 2535 default: 2536 DNPRINTF(SR_D_IOCTL, "invalid ioctl\n"); 2537 rv = ENOTTY; 2538 } 2539 2540 sc->sc_status.bs_status = (rv ? BIO_STATUS_ERROR : BIO_STATUS_SUCCESS); 2541 2542 if (sc->sc_status.bs_msg_count > 0) 2543 rv = 0; 2544 2545 memcpy(&bio->bio_status, &sc->sc_status, sizeof(struct bio_status)); 2546 2547 rw_exit_write(&sc->sc_lock); 2548 2549 return (rv); 2550 } 2551 2552 int 2553 sr_ioctl_inq(struct sr_softc *sc, struct bioc_inq *bi) 2554 { 2555 struct sr_discipline *sd; 2556 int vol = 0, disk = 0; 2557 2558 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2559 vol++; 2560 disk += sd->sd_meta->ssdi.ssd_chunk_no; 2561 } 2562 2563 strlcpy(bi->bi_dev, sc->sc_dev.dv_xname, sizeof(bi->bi_dev)); 2564 bi->bi_novol = vol + sc->sc_hotspare_no; 2565 bi->bi_nodisk = disk + sc->sc_hotspare_no; 2566 2567 return (0); 2568 } 2569 2570 int 2571 sr_ioctl_vol(struct sr_softc *sc, struct bioc_vol *bv) 2572 { 2573 int vol = -1, rv = EINVAL; 2574 struct sr_discipline *sd; 2575 struct sr_chunk *hotspare; 2576 2577 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2578 vol++; 2579 if (vol != bv->bv_volid) 2580 continue; 2581 2582 bv->bv_status = sd->sd_vol_status; 2583 bv->bv_size = sd->sd_meta->ssdi.ssd_size << DEV_BSHIFT; 2584 bv->bv_level = sd->sd_meta->ssdi.ssd_level; 2585 bv->bv_nodisk = sd->sd_meta->ssdi.ssd_chunk_no; 2586 2587 #ifdef CRYPTO 2588 if (sd->sd_meta->ssdi.ssd_level == 'C' && 2589 sd->mds.mdd_crypto.key_disk != NULL) 2590 bv->bv_nodisk++; 2591 #endif 2592 if (bv->bv_status == BIOC_SVREBUILD) 2593 bv->bv_percent = sr_rebuild_percent(sd); 2594 2595 strlcpy(bv->bv_dev, sd->sd_meta->ssd_devname, 2596 sizeof(bv->bv_dev)); 2597 strlcpy(bv->bv_vendor, sd->sd_meta->ssdi.ssd_vendor, 2598 sizeof(bv->bv_vendor)); 2599 rv = 0; 2600 goto done; 2601 } 2602 2603 /* Check hotspares list. */ 2604 SLIST_FOREACH(hotspare, &sc->sc_hotspare_list, src_link) { 2605 vol++; 2606 if (vol != bv->bv_volid) 2607 continue; 2608 2609 bv->bv_status = BIOC_SVONLINE; 2610 bv->bv_size = hotspare->src_meta.scmi.scm_size << DEV_BSHIFT; 2611 bv->bv_level = -1; /* Hotspare. */ 2612 bv->bv_nodisk = 1; 2613 strlcpy(bv->bv_dev, hotspare->src_meta.scmi.scm_devname, 2614 sizeof(bv->bv_dev)); 2615 strlcpy(bv->bv_vendor, hotspare->src_meta.scmi.scm_devname, 2616 sizeof(bv->bv_vendor)); 2617 rv = 0; 2618 goto done; 2619 } 2620 2621 done: 2622 return (rv); 2623 } 2624 2625 int 2626 sr_ioctl_disk(struct sr_softc *sc, struct bioc_disk *bd) 2627 { 2628 struct sr_discipline *sd; 2629 struct sr_chunk *src, *hotspare; 2630 int vol = -1, rv = EINVAL; 2631 2632 if (bd->bd_diskid < 0) 2633 goto done; 2634 2635 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2636 vol++; 2637 if (vol != bd->bd_volid) 2638 continue; 2639 2640 if (bd->bd_diskid < sd->sd_meta->ssdi.ssd_chunk_no) 2641 src = sd->sd_vol.sv_chunks[bd->bd_diskid]; 2642 #ifdef CRYPTO 2643 else if (bd->bd_diskid == sd->sd_meta->ssdi.ssd_chunk_no && 2644 sd->sd_meta->ssdi.ssd_level == 'C' && 2645 sd->mds.mdd_crypto.key_disk != NULL) 2646 src = sd->mds.mdd_crypto.key_disk; 2647 #endif 2648 else 2649 break; 2650 2651 bd->bd_status = src->src_meta.scm_status; 2652 bd->bd_size = src->src_meta.scmi.scm_size << DEV_BSHIFT; 2653 bd->bd_channel = vol; 2654 bd->bd_target = bd->bd_diskid; 2655 strlcpy(bd->bd_vendor, src->src_meta.scmi.scm_devname, 2656 sizeof(bd->bd_vendor)); 2657 rv = 0; 2658 goto done; 2659 } 2660 2661 /* Check hotspares list. */ 2662 SLIST_FOREACH(hotspare, &sc->sc_hotspare_list, src_link) { 2663 vol++; 2664 if (vol != bd->bd_volid) 2665 continue; 2666 2667 if (bd->bd_diskid != 0) 2668 break; 2669 2670 bd->bd_status = hotspare->src_meta.scm_status; 2671 bd->bd_size = hotspare->src_meta.scmi.scm_size << DEV_BSHIFT; 2672 bd->bd_channel = vol; 2673 bd->bd_target = bd->bd_diskid; 2674 strlcpy(bd->bd_vendor, hotspare->src_meta.scmi.scm_devname, 2675 sizeof(bd->bd_vendor)); 2676 rv = 0; 2677 goto done; 2678 } 2679 2680 done: 2681 return (rv); 2682 } 2683 2684 int 2685 sr_ioctl_setstate(struct sr_softc *sc, struct bioc_setstate *bs) 2686 { 2687 int rv = EINVAL; 2688 int vol = -1, found, c; 2689 struct sr_discipline *sd; 2690 struct sr_chunk *ch_entry; 2691 struct sr_chunk_head *cl; 2692 2693 if (bs->bs_other_id_type == BIOC_SSOTHER_UNUSED) 2694 goto done; 2695 2696 if (bs->bs_status == BIOC_SSHOTSPARE) { 2697 rv = sr_hotspare(sc, (dev_t)bs->bs_other_id); 2698 goto done; 2699 } 2700 2701 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2702 vol++; 2703 if (vol == bs->bs_volid) 2704 break; 2705 } 2706 if (sd == NULL) 2707 goto done; 2708 2709 switch (bs->bs_status) { 2710 case BIOC_SSOFFLINE: 2711 /* Take chunk offline */ 2712 found = c = 0; 2713 cl = &sd->sd_vol.sv_chunk_list; 2714 SLIST_FOREACH(ch_entry, cl, src_link) { 2715 if (ch_entry->src_dev_mm == bs->bs_other_id) { 2716 found = 1; 2717 break; 2718 } 2719 c++; 2720 } 2721 if (found == 0) { 2722 sr_error(sc, "chunk not part of array"); 2723 goto done; 2724 } 2725 2726 /* XXX: check current state first */ 2727 sd->sd_set_chunk_state(sd, c, BIOC_SDOFFLINE); 2728 2729 if (sr_meta_save(sd, SR_META_DIRTY)) { 2730 sr_error(sc, "could not save metadata for %s", 2731 sd->sd_meta->ssd_devname); 2732 goto done; 2733 } 2734 rv = 0; 2735 break; 2736 2737 case BIOC_SDSCRUB: 2738 break; 2739 2740 case BIOC_SSREBUILD: 2741 rv = sr_rebuild_init(sd, (dev_t)bs->bs_other_id, 0); 2742 break; 2743 2744 default: 2745 sr_error(sc, "unsupported state request %d", bs->bs_status); 2746 } 2747 2748 done: 2749 return (rv); 2750 } 2751 2752 int 2753 sr_chunk_in_use(struct sr_softc *sc, dev_t dev) 2754 { 2755 struct sr_discipline *sd; 2756 struct sr_chunk *chunk; 2757 int i; 2758 2759 DNPRINTF(SR_D_MISC, "%s: sr_chunk_in_use(%d)\n", DEVNAME(sc), dev); 2760 2761 if (dev == NODEV) 2762 return BIOC_SDINVALID; 2763 2764 /* See if chunk is already in use. */ 2765 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2766 for (i = 0; i < sd->sd_meta->ssdi.ssd_chunk_no; i++) { 2767 chunk = sd->sd_vol.sv_chunks[i]; 2768 if (chunk->src_dev_mm == dev) 2769 return chunk->src_meta.scm_status; 2770 } 2771 } 2772 2773 /* Check hotspares list. */ 2774 SLIST_FOREACH(chunk, &sc->sc_hotspare_list, src_link) 2775 if (chunk->src_dev_mm == dev) 2776 return chunk->src_meta.scm_status; 2777 2778 return BIOC_SDINVALID; 2779 } 2780 2781 int 2782 sr_hotspare(struct sr_softc *sc, dev_t dev) 2783 { 2784 struct sr_discipline *sd = NULL; 2785 struct sr_metadata *sm = NULL; 2786 struct sr_meta_chunk *hm; 2787 struct sr_chunk_head *cl; 2788 struct sr_chunk *chunk, *last, *hotspare = NULL; 2789 struct sr_uuid uuid; 2790 struct disklabel label; 2791 struct vnode *vn; 2792 u_int64_t size; 2793 char devname[32]; 2794 int rv = EINVAL; 2795 int c, part, open = 0; 2796 2797 /* 2798 * Add device to global hotspares list. 2799 */ 2800 2801 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 2802 2803 /* Make sure chunk is not already in use. */ 2804 c = sr_chunk_in_use(sc, dev); 2805 if (c != BIOC_SDINVALID && c != BIOC_SDOFFLINE) { 2806 if (c == BIOC_SDHOTSPARE) 2807 sr_error(sc, "%s is already a hotspare", devname); 2808 else 2809 sr_error(sc, "%s is already in use", devname); 2810 goto done; 2811 } 2812 2813 /* XXX - See if there is an existing degraded volume... */ 2814 2815 /* Open device. */ 2816 if (bdevvp(dev, &vn)) { 2817 sr_error(sc, "sr_hotspare: cannot allocate vnode"); 2818 goto done; 2819 } 2820 if (VOP_OPEN(vn, FREAD | FWRITE, NOCRED, curproc)) { 2821 DNPRINTF(SR_D_META,"%s: sr_hotspare cannot open %s\n", 2822 DEVNAME(sc), devname); 2823 vput(vn); 2824 goto fail; 2825 } 2826 open = 1; /* close dev on error */ 2827 2828 /* Get partition details. */ 2829 part = DISKPART(dev); 2830 if (VOP_IOCTL(vn, DIOCGDINFO, (caddr_t)&label, FREAD, 2831 NOCRED, curproc)) { 2832 DNPRINTF(SR_D_META, "%s: sr_hotspare ioctl failed\n", 2833 DEVNAME(sc)); 2834 goto fail; 2835 } 2836 if (label.d_partitions[part].p_fstype != FS_RAID) { 2837 sr_error(sc, "%s partition not of type RAID (%d)", 2838 devname, label.d_partitions[part].p_fstype); 2839 goto fail; 2840 } 2841 2842 /* Calculate partition size. */ 2843 size = DL_SECTOBLK(&label, DL_GETPSIZE(&label.d_partitions[part])); 2844 if (size <= SR_DATA_OFFSET) { 2845 DNPRINTF(SR_D_META, "%s: %s partition too small\n", DEVNAME(sc), 2846 devname); 2847 goto fail; 2848 } 2849 size -= SR_DATA_OFFSET; 2850 if (size > INT64_MAX) { 2851 DNPRINTF(SR_D_META, "%s: %s partition too large\n", DEVNAME(sc), 2852 devname); 2853 goto fail; 2854 } 2855 2856 /* 2857 * Create and populate chunk metadata. 2858 */ 2859 2860 sr_uuid_generate(&uuid); 2861 hotspare = malloc(sizeof(struct sr_chunk), M_DEVBUF, M_WAITOK | M_ZERO); 2862 2863 hotspare->src_dev_mm = dev; 2864 hotspare->src_vn = vn; 2865 strlcpy(hotspare->src_devname, devname, sizeof(hm->scmi.scm_devname)); 2866 hotspare->src_size = size; 2867 2868 hm = &hotspare->src_meta; 2869 hm->scmi.scm_volid = SR_HOTSPARE_VOLID; 2870 hm->scmi.scm_chunk_id = 0; 2871 hm->scmi.scm_size = size; 2872 hm->scmi.scm_coerced_size = size; 2873 strlcpy(hm->scmi.scm_devname, devname, sizeof(hm->scmi.scm_devname)); 2874 memcpy(&hm->scmi.scm_uuid, &uuid, sizeof(struct sr_uuid)); 2875 2876 sr_checksum(sc, hm, &hm->scm_checksum, 2877 sizeof(struct sr_meta_chunk_invariant)); 2878 2879 hm->scm_status = BIOC_SDHOTSPARE; 2880 2881 /* 2882 * Create and populate our own discipline and metadata. 2883 */ 2884 2885 sm = malloc(sizeof(struct sr_metadata), M_DEVBUF, M_WAITOK | M_ZERO); 2886 sm->ssdi.ssd_magic = SR_MAGIC; 2887 sm->ssdi.ssd_version = SR_META_VERSION; 2888 sm->ssd_ondisk = 0; 2889 sm->ssdi.ssd_vol_flags = 0; 2890 memcpy(&sm->ssdi.ssd_uuid, &uuid, sizeof(struct sr_uuid)); 2891 sm->ssdi.ssd_chunk_no = 1; 2892 sm->ssdi.ssd_volid = SR_HOTSPARE_VOLID; 2893 sm->ssdi.ssd_level = SR_HOTSPARE_LEVEL; 2894 sm->ssdi.ssd_size = size; 2895 sm->ssdi.ssd_secsize = label.d_secsize; 2896 strlcpy(sm->ssdi.ssd_vendor, "OPENBSD", sizeof(sm->ssdi.ssd_vendor)); 2897 snprintf(sm->ssdi.ssd_product, sizeof(sm->ssdi.ssd_product), 2898 "SR %s", "HOTSPARE"); 2899 snprintf(sm->ssdi.ssd_revision, sizeof(sm->ssdi.ssd_revision), 2900 "%03d", SR_META_VERSION); 2901 2902 sd = malloc(sizeof(struct sr_discipline), M_DEVBUF, M_WAITOK | M_ZERO); 2903 sd->sd_sc = sc; 2904 sd->sd_meta = sm; 2905 sd->sd_meta_type = SR_META_F_NATIVE; 2906 sd->sd_vol_status = BIOC_SVONLINE; 2907 strlcpy(sd->sd_name, "HOTSPARE", sizeof(sd->sd_name)); 2908 SLIST_INIT(&sd->sd_meta_opt); 2909 2910 /* Add chunk to volume. */ 2911 sd->sd_vol.sv_chunks = malloc(sizeof(struct sr_chunk *), M_DEVBUF, 2912 M_WAITOK | M_ZERO); 2913 sd->sd_vol.sv_chunks[0] = hotspare; 2914 SLIST_INIT(&sd->sd_vol.sv_chunk_list); 2915 SLIST_INSERT_HEAD(&sd->sd_vol.sv_chunk_list, hotspare, src_link); 2916 2917 /* Save metadata. */ 2918 if (sr_meta_save(sd, SR_META_DIRTY)) { 2919 sr_error(sc, "could not save metadata to %s", devname); 2920 goto fail; 2921 } 2922 2923 /* 2924 * Add chunk to hotspare list. 2925 */ 2926 rw_enter_write(&sc->sc_hs_lock); 2927 cl = &sc->sc_hotspare_list; 2928 if (SLIST_EMPTY(cl)) 2929 SLIST_INSERT_HEAD(cl, hotspare, src_link); 2930 else { 2931 SLIST_FOREACH(chunk, cl, src_link) 2932 last = chunk; 2933 SLIST_INSERT_AFTER(last, hotspare, src_link); 2934 } 2935 sc->sc_hotspare_no++; 2936 rw_exit_write(&sc->sc_hs_lock); 2937 2938 rv = 0; 2939 goto done; 2940 2941 fail: 2942 free(hotspare, M_DEVBUF, sizeof(*hotspare)); 2943 2944 done: 2945 if (sd) 2946 free(sd->sd_vol.sv_chunks, M_DEVBUF, 2947 sizeof(sd->sd_vol.sv_chunks)); 2948 free(sd, M_DEVBUF, sizeof(*sd)); 2949 free(sm, M_DEVBUF, sizeof(*sm)); 2950 if (open) { 2951 VOP_CLOSE(vn, FREAD | FWRITE, NOCRED, curproc); 2952 vput(vn); 2953 } 2954 2955 return (rv); 2956 } 2957 2958 void 2959 sr_hotspare_rebuild_callback(void *xsd) 2960 { 2961 struct sr_discipline *sd = xsd; 2962 sr_hotspare_rebuild(sd); 2963 } 2964 2965 void 2966 sr_hotspare_rebuild(struct sr_discipline *sd) 2967 { 2968 struct sr_softc *sc = sd->sd_sc; 2969 struct sr_chunk_head *cl; 2970 struct sr_chunk *hotspare, *chunk = NULL; 2971 struct sr_workunit *wu; 2972 struct sr_ccb *ccb; 2973 int i, s, cid, busy; 2974 2975 /* 2976 * Attempt to locate a hotspare and initiate rebuild. 2977 */ 2978 2979 /* Find first offline chunk. */ 2980 for (cid = 0; cid < sd->sd_meta->ssdi.ssd_chunk_no; cid++) { 2981 if (sd->sd_vol.sv_chunks[cid]->src_meta.scm_status == 2982 BIOC_SDOFFLINE) { 2983 chunk = sd->sd_vol.sv_chunks[cid]; 2984 break; 2985 } 2986 } 2987 if (chunk == NULL) { 2988 printf("%s: no offline chunk found on %s!\n", 2989 DEVNAME(sc), sd->sd_meta->ssd_devname); 2990 return; 2991 } 2992 2993 /* See if we have a suitable hotspare... */ 2994 rw_enter_write(&sc->sc_hs_lock); 2995 cl = &sc->sc_hotspare_list; 2996 SLIST_FOREACH(hotspare, cl, src_link) 2997 if (hotspare->src_size >= chunk->src_size && 2998 hotspare->src_secsize <= sd->sd_meta->ssdi.ssd_secsize) 2999 break; 3000 3001 if (hotspare != NULL) { 3002 3003 printf("%s: %s volume degraded, will attempt to " 3004 "rebuild on hotspare %s\n", DEVNAME(sc), 3005 sd->sd_meta->ssd_devname, hotspare->src_devname); 3006 3007 /* 3008 * Ensure that all pending I/O completes on the failed chunk 3009 * before trying to initiate a rebuild. 3010 */ 3011 i = 0; 3012 do { 3013 busy = 0; 3014 3015 s = splbio(); 3016 TAILQ_FOREACH(wu, &sd->sd_wu_pendq, swu_link) { 3017 TAILQ_FOREACH(ccb, &wu->swu_ccb, ccb_link) { 3018 if (ccb->ccb_target == cid) 3019 busy = 1; 3020 } 3021 } 3022 TAILQ_FOREACH(wu, &sd->sd_wu_defq, swu_link) { 3023 TAILQ_FOREACH(ccb, &wu->swu_ccb, ccb_link) { 3024 if (ccb->ccb_target == cid) 3025 busy = 1; 3026 } 3027 } 3028 splx(s); 3029 3030 if (busy) { 3031 tsleep_nsec(sd, PRIBIO, "sr_hotspare", 3032 SEC_TO_NSEC(1)); 3033 i++; 3034 } 3035 3036 } while (busy && i < 120); 3037 3038 DNPRINTF(SR_D_META, "%s: waited %i seconds for I/O to " 3039 "complete on failed chunk %s\n", DEVNAME(sc), 3040 i, chunk->src_devname); 3041 3042 if (busy) { 3043 printf("%s: pending I/O failed to complete on " 3044 "failed chunk %s, hotspare rebuild aborted...\n", 3045 DEVNAME(sc), chunk->src_devname); 3046 goto done; 3047 } 3048 3049 s = splbio(); 3050 rw_enter_write(&sc->sc_lock); 3051 bio_status_init(&sc->sc_status, &sc->sc_dev); 3052 if (sr_rebuild_init(sd, hotspare->src_dev_mm, 1) == 0) { 3053 3054 /* Remove hotspare from available list. */ 3055 sc->sc_hotspare_no--; 3056 SLIST_REMOVE(cl, hotspare, sr_chunk, src_link); 3057 free(hotspare, M_DEVBUF, sizeof(*hotspare)); 3058 3059 } 3060 rw_exit_write(&sc->sc_lock); 3061 splx(s); 3062 } 3063 done: 3064 rw_exit_write(&sc->sc_hs_lock); 3065 } 3066 3067 int 3068 sr_rebuild_init(struct sr_discipline *sd, dev_t dev, int hotspare) 3069 { 3070 struct sr_softc *sc = sd->sd_sc; 3071 struct sr_chunk *chunk = NULL; 3072 struct sr_meta_chunk *meta; 3073 struct disklabel label; 3074 struct vnode *vn; 3075 u_int64_t size; 3076 int64_t csize; 3077 char devname[32]; 3078 int rv = EINVAL, open = 0; 3079 int cid, i, part, status; 3080 3081 /* 3082 * Attempt to initiate a rebuild onto the specified device. 3083 */ 3084 3085 if (!(sd->sd_capabilities & SR_CAP_REBUILD)) { 3086 sr_error(sc, "discipline does not support rebuild"); 3087 goto done; 3088 } 3089 3090 /* make sure volume is in the right state */ 3091 if (sd->sd_vol_status == BIOC_SVREBUILD) { 3092 sr_error(sc, "rebuild already in progress"); 3093 goto done; 3094 } 3095 if (sd->sd_vol_status != BIOC_SVDEGRADED) { 3096 sr_error(sc, "volume not degraded"); 3097 goto done; 3098 } 3099 3100 /* Find first offline chunk. */ 3101 for (cid = 0; cid < sd->sd_meta->ssdi.ssd_chunk_no; cid++) { 3102 if (sd->sd_vol.sv_chunks[cid]->src_meta.scm_status == 3103 BIOC_SDOFFLINE) { 3104 chunk = sd->sd_vol.sv_chunks[cid]; 3105 break; 3106 } 3107 } 3108 if (chunk == NULL) { 3109 sr_error(sc, "no offline chunks available to rebuild"); 3110 goto done; 3111 } 3112 3113 /* Get coerced size from another online chunk. */ 3114 csize = 0; 3115 for (i = 0; i < sd->sd_meta->ssdi.ssd_chunk_no; i++) { 3116 if (sd->sd_vol.sv_chunks[i]->src_meta.scm_status == 3117 BIOC_SDONLINE) { 3118 meta = &sd->sd_vol.sv_chunks[i]->src_meta; 3119 csize = meta->scmi.scm_coerced_size; 3120 break; 3121 } 3122 } 3123 if (csize == 0) { 3124 sr_error(sc, "no online chunks available for rebuild"); 3125 goto done; 3126 } 3127 3128 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 3129 if (bdevvp(dev, &vn)) { 3130 printf("%s: sr_rebuild_init: can't allocate vnode\n", 3131 DEVNAME(sc)); 3132 goto done; 3133 } 3134 if (VOP_OPEN(vn, FREAD | FWRITE, NOCRED, curproc)) { 3135 DNPRINTF(SR_D_META,"%s: sr_ioctl_setstate can't " 3136 "open %s\n", DEVNAME(sc), devname); 3137 vput(vn); 3138 goto done; 3139 } 3140 open = 1; /* close dev on error */ 3141 3142 /* Get disklabel and check partition. */ 3143 part = DISKPART(dev); 3144 if (VOP_IOCTL(vn, DIOCGDINFO, (caddr_t)&label, FREAD, 3145 NOCRED, curproc)) { 3146 DNPRINTF(SR_D_META, "%s: sr_ioctl_setstate ioctl failed\n", 3147 DEVNAME(sc)); 3148 goto done; 3149 } 3150 if (label.d_partitions[part].p_fstype != FS_RAID) { 3151 sr_error(sc, "%s partition not of type RAID (%d)", 3152 devname, label.d_partitions[part].p_fstype); 3153 goto done; 3154 } 3155 3156 /* Is the partition large enough? */ 3157 size = DL_SECTOBLK(&label, DL_GETPSIZE(&label.d_partitions[part])); 3158 if (size <= sd->sd_meta->ssd_data_blkno) { 3159 sr_error(sc, "%s: %s partition too small", DEVNAME(sc), 3160 devname); 3161 goto done; 3162 } 3163 size -= sd->sd_meta->ssd_data_blkno; 3164 if (size > INT64_MAX) { 3165 sr_error(sc, "%s: %s partition too large", DEVNAME(sc), 3166 devname); 3167 goto done; 3168 } 3169 if (size < csize) { 3170 sr_error(sc, "%s partition too small, at least %lld bytes " 3171 "required", devname, (long long)(csize << DEV_BSHIFT)); 3172 goto done; 3173 } else if (size > csize) 3174 sr_warn(sc, "%s partition too large, wasting %lld bytes", 3175 devname, (long long)((size - csize) << DEV_BSHIFT)); 3176 if (label.d_secsize > sd->sd_meta->ssdi.ssd_secsize) { 3177 sr_error(sc, "%s sector size too large, <= %u bytes " 3178 "required", devname, sd->sd_meta->ssdi.ssd_secsize); 3179 goto done; 3180 } 3181 3182 /* Ensure that this chunk is not already in use. */ 3183 status = sr_chunk_in_use(sc, dev); 3184 if (status != BIOC_SDINVALID && status != BIOC_SDOFFLINE && 3185 !(hotspare && status == BIOC_SDHOTSPARE)) { 3186 sr_error(sc, "%s is already in use", devname); 3187 goto done; 3188 } 3189 3190 /* Reset rebuild counter since we rebuilding onto a new chunk. */ 3191 sd->sd_meta->ssd_rebuild = 0; 3192 3193 open = 0; /* leave dev open from here on out */ 3194 3195 /* Fix up chunk. */ 3196 memcpy(chunk->src_duid, label.d_uid, sizeof(chunk->src_duid)); 3197 chunk->src_dev_mm = dev; 3198 chunk->src_vn = vn; 3199 3200 /* Reconstruct metadata. */ 3201 meta = &chunk->src_meta; 3202 meta->scmi.scm_volid = sd->sd_meta->ssdi.ssd_volid; 3203 meta->scmi.scm_chunk_id = cid; 3204 strlcpy(meta->scmi.scm_devname, devname, 3205 sizeof(meta->scmi.scm_devname)); 3206 meta->scmi.scm_size = size; 3207 meta->scmi.scm_coerced_size = csize; 3208 memcpy(&meta->scmi.scm_uuid, &sd->sd_meta->ssdi.ssd_uuid, 3209 sizeof(meta->scmi.scm_uuid)); 3210 sr_checksum(sc, meta, &meta->scm_checksum, 3211 sizeof(struct sr_meta_chunk_invariant)); 3212 3213 sd->sd_set_chunk_state(sd, cid, BIOC_SDREBUILD); 3214 3215 if (sr_meta_save(sd, SR_META_DIRTY)) { 3216 sr_error(sc, "could not save metadata to %s", devname); 3217 open = 1; 3218 goto done; 3219 } 3220 3221 sr_warn(sc, "rebuild of %s started on %s", 3222 sd->sd_meta->ssd_devname, devname); 3223 3224 sd->sd_reb_abort = 0; 3225 kthread_create_deferred(sr_rebuild_start, sd); 3226 3227 rv = 0; 3228 done: 3229 if (open) { 3230 VOP_CLOSE(vn, FREAD | FWRITE, NOCRED, curproc); 3231 vput(vn); 3232 } 3233 3234 return (rv); 3235 } 3236 3237 int 3238 sr_rebuild_percent(struct sr_discipline *sd) 3239 { 3240 daddr_t rb, sz; 3241 3242 sz = sd->sd_meta->ssdi.ssd_size; 3243 rb = sd->sd_meta->ssd_rebuild; 3244 3245 if (rb > 0) 3246 return (100 - ((sz * 100 - rb * 100) / sz) - 1); 3247 3248 return (0); 3249 } 3250 3251 void 3252 sr_roam_chunks(struct sr_discipline *sd) 3253 { 3254 struct sr_softc *sc = sd->sd_sc; 3255 struct sr_chunk *chunk; 3256 struct sr_meta_chunk *meta; 3257 int roamed = 0; 3258 3259 /* Have any chunks roamed? */ 3260 SLIST_FOREACH(chunk, &sd->sd_vol.sv_chunk_list, src_link) { 3261 meta = &chunk->src_meta; 3262 if (strncmp(meta->scmi.scm_devname, chunk->src_devname, 3263 sizeof(meta->scmi.scm_devname))) { 3264 3265 printf("%s: roaming device %s -> %s\n", DEVNAME(sc), 3266 meta->scmi.scm_devname, chunk->src_devname); 3267 3268 strlcpy(meta->scmi.scm_devname, chunk->src_devname, 3269 sizeof(meta->scmi.scm_devname)); 3270 3271 roamed++; 3272 } 3273 } 3274 3275 if (roamed) 3276 sr_meta_save(sd, SR_META_DIRTY); 3277 } 3278 3279 int 3280 sr_ioctl_createraid(struct sr_softc *sc, struct bioc_createraid *bc, 3281 int user, void *data) 3282 { 3283 struct sr_meta_opt_item *omi; 3284 struct sr_chunk_head *cl; 3285 struct sr_discipline *sd = NULL; 3286 struct sr_chunk *ch_entry; 3287 struct scsi_link *link; 3288 struct device *dev; 3289 char *uuid, devname[32]; 3290 dev_t *dt = NULL; 3291 int i, no_chunk, rv = EINVAL, target, vol; 3292 int no_meta; 3293 3294 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_createraid(%d)\n", 3295 DEVNAME(sc), user); 3296 3297 /* user input */ 3298 if (bc->bc_dev_list_len > BIOC_CRMAXLEN) 3299 goto unwind; 3300 3301 dt = malloc(bc->bc_dev_list_len, M_DEVBUF, M_WAITOK | M_ZERO); 3302 if (user) { 3303 if (copyin(bc->bc_dev_list, dt, bc->bc_dev_list_len) != 0) 3304 goto unwind; 3305 } else 3306 memcpy(dt, bc->bc_dev_list, bc->bc_dev_list_len); 3307 3308 /* Initialise discipline. */ 3309 sd = malloc(sizeof(struct sr_discipline), M_DEVBUF, M_WAITOK | M_ZERO); 3310 sd->sd_sc = sc; 3311 SLIST_INIT(&sd->sd_meta_opt); 3312 sd->sd_taskq = taskq_create("srdis", 1, IPL_BIO, 0); 3313 if (sd->sd_taskq == NULL) { 3314 sr_error(sc, "could not create discipline taskq"); 3315 goto unwind; 3316 } 3317 if (sr_discipline_init(sd, bc->bc_level)) { 3318 sr_error(sc, "could not initialize discipline"); 3319 goto unwind; 3320 } 3321 3322 no_chunk = bc->bc_dev_list_len / sizeof(dev_t); 3323 cl = &sd->sd_vol.sv_chunk_list; 3324 SLIST_INIT(cl); 3325 3326 /* Ensure that chunks are not already in use. */ 3327 for (i = 0; i < no_chunk; i++) { 3328 if (sr_chunk_in_use(sc, dt[i]) != BIOC_SDINVALID) { 3329 sr_meta_getdevname(sc, dt[i], devname, sizeof(devname)); 3330 sr_error(sc, "chunk %s already in use", devname); 3331 goto unwind; 3332 } 3333 } 3334 3335 sd->sd_meta_type = sr_meta_probe(sd, dt, no_chunk); 3336 if (sd->sd_meta_type == SR_META_F_INVALID) { 3337 sr_error(sc, "invalid metadata format"); 3338 goto unwind; 3339 } 3340 3341 if (sr_meta_attach(sd, no_chunk, bc->bc_flags & BIOC_SCFORCE)) 3342 goto unwind; 3343 3344 /* force the raid volume by clearing metadata region */ 3345 if (bc->bc_flags & BIOC_SCFORCE) { 3346 /* make sure disk isn't up and running */ 3347 if (sr_meta_read(sd)) 3348 if (sr_already_assembled(sd)) { 3349 uuid = sr_uuid_format( 3350 &sd->sd_meta->ssdi.ssd_uuid); 3351 sr_error(sc, "disk %s is currently in use; " 3352 "cannot force create", uuid); 3353 free(uuid, M_DEVBUF, 37); 3354 goto unwind; 3355 } 3356 3357 if (sr_meta_clear(sd)) { 3358 sr_error(sc, "failed to clear metadata"); 3359 goto unwind; 3360 } 3361 } 3362 3363 no_meta = sr_meta_read(sd); 3364 if (no_meta == -1) { 3365 3366 /* Corrupt metadata on one or more chunks. */ 3367 sr_error(sc, "one of the chunks has corrupt metadata; " 3368 "aborting assembly"); 3369 goto unwind; 3370 3371 } else if (no_meta == 0) { 3372 3373 /* Initialise volume and chunk metadata. */ 3374 sr_meta_init(sd, bc->bc_level, no_chunk); 3375 sd->sd_vol_status = BIOC_SVONLINE; 3376 sd->sd_meta_flags = bc->bc_flags & BIOC_SCNOAUTOASSEMBLE; 3377 if (sd->sd_create) { 3378 if ((i = sd->sd_create(sd, bc, no_chunk, 3379 sd->sd_vol.sv_chunk_minsz))) { 3380 rv = i; 3381 goto unwind; 3382 } 3383 } 3384 sr_meta_init_complete(sd); 3385 3386 DNPRINTF(SR_D_IOCTL, 3387 "%s: sr_ioctl_createraid: vol_size: %lld\n", 3388 DEVNAME(sc), sd->sd_meta->ssdi.ssd_size); 3389 3390 /* Warn if we've wasted chunk space due to coercing. */ 3391 if ((sd->sd_capabilities & SR_CAP_NON_COERCED) == 0 && 3392 sd->sd_vol.sv_chunk_minsz != sd->sd_vol.sv_chunk_maxsz) 3393 sr_warn(sc, "chunk sizes are not equal; up to %llu " 3394 "blocks wasted per chunk", 3395 sd->sd_vol.sv_chunk_maxsz - 3396 sd->sd_vol.sv_chunk_minsz); 3397 3398 } else { 3399 3400 /* Ensure we are assembling the correct # of chunks. */ 3401 if (sd->sd_meta->ssdi.ssd_chunk_no != no_chunk) { 3402 sr_error(sc, "volume chunk count does not match metadata " 3403 "chunk count"); 3404 goto unwind; 3405 } 3406 3407 /* Ensure metadata level matches requested assembly level. */ 3408 if (sd->sd_meta->ssdi.ssd_level != bc->bc_level) { 3409 sr_error(sc, "volume level does not match metadata " 3410 "level"); 3411 goto unwind; 3412 } 3413 3414 if (sr_already_assembled(sd)) { 3415 uuid = sr_uuid_format(&sd->sd_meta->ssdi.ssd_uuid); 3416 sr_error(sc, "disk %s already assembled", uuid); 3417 free(uuid, M_DEVBUF, 37); 3418 goto unwind; 3419 } 3420 3421 if (user == 0 && sd->sd_meta_flags & BIOC_SCNOAUTOASSEMBLE) { 3422 DNPRINTF(SR_D_META, "%s: disk not auto assembled from " 3423 "metadata\n", DEVNAME(sc)); 3424 goto unwind; 3425 } 3426 3427 if (no_meta != no_chunk) 3428 sr_warn(sc, "trying to bring up %s degraded", 3429 sd->sd_meta->ssd_devname); 3430 3431 if (sd->sd_meta->ssd_meta_flags & SR_META_DIRTY) 3432 sr_warn(sc, "%s was not shutdown properly", 3433 sd->sd_meta->ssd_devname); 3434 3435 SLIST_FOREACH(omi, &sd->sd_meta_opt, omi_link) 3436 if (sd->sd_meta_opt_handler == NULL || 3437 sd->sd_meta_opt_handler(sd, omi->omi_som) != 0) 3438 sr_meta_opt_handler(sd, omi->omi_som); 3439 3440 if (sd->sd_assemble) { 3441 if ((i = sd->sd_assemble(sd, bc, no_chunk, data))) { 3442 rv = i; 3443 goto unwind; 3444 } 3445 } 3446 3447 DNPRINTF(SR_D_META, "%s: disk assembled from metadata\n", 3448 DEVNAME(sc)); 3449 3450 } 3451 3452 /* Metadata MUST be fully populated by this point. */ 3453 TAILQ_INSERT_TAIL(&sc->sc_dis_list, sd, sd_link); 3454 3455 /* Allocate all resources. */ 3456 if ((rv = sd->sd_alloc_resources(sd))) 3457 goto unwind; 3458 3459 /* Adjust flags if necessary. */ 3460 if ((sd->sd_capabilities & SR_CAP_AUTO_ASSEMBLE) && 3461 (bc->bc_flags & BIOC_SCNOAUTOASSEMBLE) != 3462 (sd->sd_meta->ssdi.ssd_vol_flags & BIOC_SCNOAUTOASSEMBLE)) { 3463 sd->sd_meta->ssdi.ssd_vol_flags &= ~BIOC_SCNOAUTOASSEMBLE; 3464 sd->sd_meta->ssdi.ssd_vol_flags |= 3465 bc->bc_flags & BIOC_SCNOAUTOASSEMBLE; 3466 } 3467 3468 if (sd->sd_capabilities & SR_CAP_SYSTEM_DISK) { 3469 /* Initialise volume state. */ 3470 sd->sd_set_vol_state(sd); 3471 if (sd->sd_vol_status == BIOC_SVOFFLINE) { 3472 sr_error(sc, "%s is offline, will not be brought " 3473 "online", sd->sd_meta->ssd_devname); 3474 goto unwind; 3475 } 3476 3477 /* Setup SCSI iopool. */ 3478 scsi_iopool_init(&sd->sd_iopool, sd, sr_wu_get, sr_wu_put); 3479 3480 /* 3481 * All checks passed - return ENXIO if volume cannot be created. 3482 */ 3483 rv = ENXIO; 3484 3485 /* 3486 * Find a free target. 3487 * 3488 * XXX: We reserve sd_target == 0 to indicate the 3489 * discipline is not linked into sc->sc_targets, so begin 3490 * the search with target = 1. 3491 */ 3492 for (target = 1; target < SR_MAX_LD; target++) 3493 if (sc->sc_targets[target] == NULL) 3494 break; 3495 if (target == SR_MAX_LD) { 3496 sr_error(sc, "no free target for %s", 3497 sd->sd_meta->ssd_devname); 3498 goto unwind; 3499 } 3500 3501 /* Clear sense data. */ 3502 bzero(&sd->sd_scsi_sense, sizeof(sd->sd_scsi_sense)); 3503 3504 /* Attach discipline and get midlayer to probe it. */ 3505 sd->sd_target = target; 3506 sc->sc_targets[target] = sd; 3507 if (scsi_probe_lun(sc->sc_scsibus, target, 0) != 0) { 3508 sr_error(sc, "scsi_probe_lun failed"); 3509 sc->sc_targets[target] = NULL; 3510 sd->sd_target = 0; 3511 goto unwind; 3512 } 3513 3514 link = scsi_get_link(sc->sc_scsibus, target, 0); 3515 if (link == NULL) 3516 goto unwind; 3517 3518 dev = link->device_softc; 3519 DNPRINTF(SR_D_IOCTL, "%s: sr device added: %s at target %d\n", 3520 DEVNAME(sc), dev->dv_xname, sd->sd_target); 3521 3522 /* XXX - Count volumes, not targets. */ 3523 for (i = 0, vol = -1; i <= sd->sd_target; i++) 3524 if (sc->sc_targets[i]) 3525 vol++; 3526 3527 rv = 0; 3528 3529 if (sd->sd_meta->ssd_devname[0] != '\0' && 3530 strncmp(sd->sd_meta->ssd_devname, dev->dv_xname, 3531 sizeof(dev->dv_xname))) 3532 sr_warn(sc, "volume %s is roaming, it used to be %s, " 3533 "updating metadata", dev->dv_xname, 3534 sd->sd_meta->ssd_devname); 3535 3536 /* Populate remaining volume metadata. */ 3537 sd->sd_meta->ssdi.ssd_volid = vol; 3538 strlcpy(sd->sd_meta->ssd_devname, dev->dv_xname, 3539 sizeof(sd->sd_meta->ssd_devname)); 3540 3541 sr_info(sc, "%s volume attached as %s", 3542 sd->sd_name, sd->sd_meta->ssd_devname); 3543 3544 /* Update device name on any roaming chunks. */ 3545 sr_roam_chunks(sd); 3546 3547 #ifndef SMALL_KERNEL 3548 if (sr_sensors_create(sd)) 3549 sr_warn(sc, "unable to create sensor for %s", 3550 dev->dv_xname); 3551 #endif /* SMALL_KERNEL */ 3552 } else { 3553 /* This volume does not attach as a system disk. */ 3554 ch_entry = SLIST_FIRST(cl); /* XXX */ 3555 strlcpy(sd->sd_meta->ssd_devname, ch_entry->src_devname, 3556 sizeof(sd->sd_meta->ssd_devname)); 3557 3558 if (sd->sd_start_discipline(sd)) 3559 goto unwind; 3560 } 3561 3562 /* Save current metadata to disk. */ 3563 rv = sr_meta_save(sd, SR_META_DIRTY); 3564 3565 if (sd->sd_vol_status == BIOC_SVREBUILD) 3566 kthread_create_deferred(sr_rebuild_start, sd); 3567 3568 sd->sd_ready = 1; 3569 3570 free(dt, M_DEVBUF, bc->bc_dev_list_len); 3571 3572 return (rv); 3573 3574 unwind: 3575 free(dt, M_DEVBUF, bc->bc_dev_list_len); 3576 3577 sr_discipline_shutdown(sd, 0, 0); 3578 3579 if (rv == EAGAIN) 3580 rv = 0; 3581 3582 return (rv); 3583 } 3584 3585 int 3586 sr_ioctl_deleteraid(struct sr_softc *sc, struct sr_discipline *sd, 3587 struct bioc_deleteraid *bd) 3588 { 3589 int rv = 1; 3590 3591 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_deleteraid %s\n", 3592 DEVNAME(sc), bd->bd_dev); 3593 3594 if (sd == NULL) { 3595 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 3596 if (!strncmp(sd->sd_meta->ssd_devname, bd->bd_dev, 3597 sizeof(sd->sd_meta->ssd_devname))) 3598 break; 3599 } 3600 if (sd == NULL) { 3601 sr_error(sc, "volume %s not found", bd->bd_dev); 3602 goto bad; 3603 } 3604 } 3605 3606 sd->sd_deleted = 1; 3607 sd->sd_meta->ssdi.ssd_vol_flags = BIOC_SCNOAUTOASSEMBLE; 3608 sr_discipline_shutdown(sd, 1, 0); 3609 3610 rv = 0; 3611 bad: 3612 return (rv); 3613 } 3614 3615 int 3616 sr_ioctl_discipline(struct sr_softc *sc, struct sr_discipline *sd, 3617 struct bioc_discipline *bd) 3618 { 3619 int rv = 1; 3620 3621 /* Dispatch a discipline specific ioctl. */ 3622 3623 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_discipline %s\n", DEVNAME(sc), 3624 bd->bd_dev); 3625 3626 if (sd == NULL) { 3627 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 3628 if (!strncmp(sd->sd_meta->ssd_devname, bd->bd_dev, 3629 sizeof(sd->sd_meta->ssd_devname))) 3630 break; 3631 } 3632 if (sd == NULL) { 3633 sr_error(sc, "volume %s not found", bd->bd_dev); 3634 goto bad; 3635 } 3636 } 3637 3638 if (sd->sd_ioctl_handler) 3639 rv = sd->sd_ioctl_handler(sd, bd); 3640 3641 bad: 3642 return (rv); 3643 } 3644 3645 int 3646 sr_ioctl_installboot(struct sr_softc *sc, struct sr_discipline *sd, 3647 struct bioc_installboot *bb) 3648 { 3649 void *bootblk = NULL, *bootldr = NULL; 3650 struct sr_chunk *chunk; 3651 struct sr_meta_opt_item *omi; 3652 struct sr_meta_boot *sbm; 3653 struct disk *dk; 3654 u_int32_t bbs = 0, bls = 0, secsize; 3655 u_char duid[8]; 3656 int rv = EINVAL; 3657 int i; 3658 3659 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_installboot %s\n", DEVNAME(sc), 3660 bb->bb_dev); 3661 3662 if (sd == NULL) { 3663 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 3664 if (!strncmp(sd->sd_meta->ssd_devname, bb->bb_dev, 3665 sizeof(sd->sd_meta->ssd_devname))) 3666 break; 3667 } 3668 if (sd == NULL) { 3669 sr_error(sc, "volume %s not found", bb->bb_dev); 3670 goto done; 3671 } 3672 } 3673 3674 bzero(duid, sizeof(duid)); 3675 TAILQ_FOREACH(dk, &disklist, dk_link) 3676 if (!strncmp(dk->dk_name, bb->bb_dev, sizeof(bb->bb_dev))) 3677 break; 3678 if (dk == NULL || dk->dk_label == NULL || 3679 (dk->dk_flags & DKF_LABELVALID) == 0 || 3680 bcmp(dk->dk_label->d_uid, &duid, sizeof(duid)) == 0) { 3681 sr_error(sc, "failed to get DUID for softraid volume"); 3682 goto done; 3683 } 3684 memcpy(duid, dk->dk_label->d_uid, sizeof(duid)); 3685 3686 /* Ensure that boot storage area is large enough. */ 3687 if (sd->sd_meta->ssd_data_blkno < (SR_BOOT_OFFSET + SR_BOOT_SIZE)) { 3688 sr_error(sc, "insufficient boot storage"); 3689 goto done; 3690 } 3691 3692 if (bb->bb_bootblk_size > SR_BOOT_BLOCKS_SIZE * DEV_BSIZE) { 3693 sr_error(sc, "boot block too large (%d > %d)", 3694 bb->bb_bootblk_size, SR_BOOT_BLOCKS_SIZE * DEV_BSIZE); 3695 goto done; 3696 } 3697 3698 if (bb->bb_bootldr_size > SR_BOOT_LOADER_SIZE * DEV_BSIZE) { 3699 sr_error(sc, "boot loader too large (%d > %d)", 3700 bb->bb_bootldr_size, SR_BOOT_LOADER_SIZE * DEV_BSIZE); 3701 goto done; 3702 } 3703 3704 secsize = sd->sd_meta->ssdi.ssd_secsize; 3705 3706 /* Copy in boot block. */ 3707 bbs = howmany(bb->bb_bootblk_size, secsize) * secsize; 3708 bootblk = malloc(bbs, M_DEVBUF, M_WAITOK | M_ZERO); 3709 if (copyin(bb->bb_bootblk, bootblk, bb->bb_bootblk_size) != 0) 3710 goto done; 3711 3712 /* Copy in boot loader. */ 3713 bls = howmany(bb->bb_bootldr_size, secsize) * secsize; 3714 bootldr = malloc(bls, M_DEVBUF, M_WAITOK | M_ZERO); 3715 if (copyin(bb->bb_bootldr, bootldr, bb->bb_bootldr_size) != 0) 3716 goto done; 3717 3718 /* Create or update optional meta for bootable volumes. */ 3719 SLIST_FOREACH(omi, &sd->sd_meta_opt, omi_link) 3720 if (omi->omi_som->som_type == SR_OPT_BOOT) 3721 break; 3722 if (omi == NULL) { 3723 omi = malloc(sizeof(struct sr_meta_opt_item), M_DEVBUF, 3724 M_WAITOK | M_ZERO); 3725 omi->omi_som = malloc(sizeof(struct sr_meta_boot), M_DEVBUF, 3726 M_WAITOK | M_ZERO); 3727 omi->omi_som->som_type = SR_OPT_BOOT; 3728 omi->omi_som->som_length = sizeof(struct sr_meta_boot); 3729 SLIST_INSERT_HEAD(&sd->sd_meta_opt, omi, omi_link); 3730 sd->sd_meta->ssdi.ssd_opt_no++; 3731 } 3732 sbm = (struct sr_meta_boot *)omi->omi_som; 3733 3734 memcpy(sbm->sbm_root_duid, duid, sizeof(sbm->sbm_root_duid)); 3735 bzero(&sbm->sbm_boot_duid, sizeof(sbm->sbm_boot_duid)); 3736 sbm->sbm_bootblk_size = bbs; 3737 sbm->sbm_bootldr_size = bls; 3738 3739 DNPRINTF(SR_D_IOCTL, "sr_ioctl_installboot: root duid is %s\n", 3740 duid_format(sbm->sbm_root_duid)); 3741 3742 /* Save boot block and boot loader to each chunk. */ 3743 for (i = 0; i < sd->sd_meta->ssdi.ssd_chunk_no; i++) { 3744 3745 chunk = sd->sd_vol.sv_chunks[i]; 3746 if (chunk->src_meta.scm_status != BIOC_SDONLINE && 3747 chunk->src_meta.scm_status != BIOC_SDREBUILD) 3748 continue; 3749 3750 if (i < SR_MAX_BOOT_DISKS) 3751 memcpy(&sbm->sbm_boot_duid[i], chunk->src_duid, 3752 sizeof(sbm->sbm_boot_duid[i])); 3753 3754 /* Save boot blocks. */ 3755 DNPRINTF(SR_D_IOCTL, 3756 "sr_ioctl_installboot: saving boot block to %s " 3757 "(%u bytes)\n", chunk->src_devname, bbs); 3758 3759 if (sr_rw(sc, chunk->src_dev_mm, bootblk, bbs, 3760 SR_BOOT_BLOCKS_OFFSET, B_WRITE)) { 3761 sr_error(sc, "failed to write boot block"); 3762 goto done; 3763 } 3764 3765 /* Save boot loader.*/ 3766 DNPRINTF(SR_D_IOCTL, 3767 "sr_ioctl_installboot: saving boot loader to %s " 3768 "(%u bytes)\n", chunk->src_devname, bls); 3769 3770 if (sr_rw(sc, chunk->src_dev_mm, bootldr, bls, 3771 SR_BOOT_LOADER_OFFSET, B_WRITE)) { 3772 sr_error(sc, "failed to write boot loader"); 3773 goto done; 3774 } 3775 } 3776 3777 /* XXX - Install boot block on disk - MD code. */ 3778 3779 /* Mark volume as bootable and save metadata. */ 3780 sd->sd_meta->ssdi.ssd_vol_flags |= BIOC_SCBOOTABLE; 3781 if (sr_meta_save(sd, SR_META_DIRTY)) { 3782 sr_error(sc, "could not save metadata to %s", DEVNAME(sc)); 3783 goto done; 3784 } 3785 3786 rv = 0; 3787 3788 done: 3789 free(bootblk, M_DEVBUF, bbs); 3790 free(bootldr, M_DEVBUF, bls); 3791 3792 return (rv); 3793 } 3794 3795 void 3796 sr_chunks_unwind(struct sr_softc *sc, struct sr_chunk_head *cl) 3797 { 3798 struct sr_chunk *ch_entry, *ch_next; 3799 3800 DNPRINTF(SR_D_IOCTL, "%s: sr_chunks_unwind\n", DEVNAME(sc)); 3801 3802 if (!cl) 3803 return; 3804 3805 for (ch_entry = SLIST_FIRST(cl); ch_entry != NULL; ch_entry = ch_next) { 3806 ch_next = SLIST_NEXT(ch_entry, src_link); 3807 3808 DNPRINTF(SR_D_IOCTL, "%s: sr_chunks_unwind closing: %s\n", 3809 DEVNAME(sc), ch_entry->src_devname); 3810 if (ch_entry->src_vn) { 3811 /* 3812 * XXX - explicitly lock the vnode until we can resolve 3813 * the problem introduced by vnode aliasing... specfs 3814 * has no locking, whereas ufs/ffs does! 3815 */ 3816 vn_lock(ch_entry->src_vn, LK_EXCLUSIVE | LK_RETRY); 3817 VOP_CLOSE(ch_entry->src_vn, FREAD | FWRITE, NOCRED, 3818 curproc); 3819 vput(ch_entry->src_vn); 3820 } 3821 free(ch_entry, M_DEVBUF, sizeof(*ch_entry)); 3822 } 3823 SLIST_INIT(cl); 3824 } 3825 3826 void 3827 sr_discipline_free(struct sr_discipline *sd) 3828 { 3829 struct sr_softc *sc; 3830 struct sr_discipline *sdtmp1; 3831 struct sr_meta_opt_head *som; 3832 struct sr_meta_opt_item *omi, *omi_next; 3833 3834 if (!sd) 3835 return; 3836 3837 sc = sd->sd_sc; 3838 3839 DNPRINTF(SR_D_DIS, "%s: sr_discipline_free %s\n", 3840 DEVNAME(sc), 3841 sd->sd_meta ? sd->sd_meta->ssd_devname : "nodev"); 3842 if (sd->sd_free_resources) 3843 sd->sd_free_resources(sd); 3844 free(sd->sd_vol.sv_chunks, M_DEVBUF, 0); 3845 free(sd->sd_meta, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 3846 free(sd->sd_meta_foreign, M_DEVBUF, smd[sd->sd_meta_type].smd_size); 3847 3848 som = &sd->sd_meta_opt; 3849 for (omi = SLIST_FIRST(som); omi != NULL; omi = omi_next) { 3850 omi_next = SLIST_NEXT(omi, omi_link); 3851 free(omi->omi_som, M_DEVBUF, 0); 3852 free(omi, M_DEVBUF, sizeof(*omi)); 3853 } 3854 3855 if (sd->sd_target != 0) { 3856 KASSERT(sc->sc_targets[sd->sd_target] == sd); 3857 sc->sc_targets[sd->sd_target] = NULL; 3858 } 3859 3860 TAILQ_FOREACH(sdtmp1, &sc->sc_dis_list, sd_link) { 3861 if (sdtmp1 == sd) 3862 break; 3863 } 3864 if (sdtmp1 != NULL) 3865 TAILQ_REMOVE(&sc->sc_dis_list, sd, sd_link); 3866 3867 explicit_bzero(sd, sizeof *sd); 3868 free(sd, M_DEVBUF, sizeof(*sd)); 3869 } 3870 3871 void 3872 sr_discipline_shutdown(struct sr_discipline *sd, int meta_save, int dying) 3873 { 3874 struct sr_softc *sc; 3875 int ret, s; 3876 3877 if (!sd) 3878 return; 3879 sc = sd->sd_sc; 3880 3881 DNPRINTF(SR_D_DIS, "%s: sr_discipline_shutdown %s\n", DEVNAME(sc), 3882 sd->sd_meta ? sd->sd_meta->ssd_devname : "nodev"); 3883 3884 /* If rebuilding, abort rebuild and drain I/O. */ 3885 if (sd->sd_reb_active) { 3886 sd->sd_reb_abort = 1; 3887 while (sd->sd_reb_active) 3888 tsleep(sd, PWAIT, "sr_shutdown", 1); 3889 } 3890 3891 if (meta_save) 3892 sr_meta_save(sd, 0); 3893 3894 s = splbio(); 3895 3896 sd->sd_ready = 0; 3897 3898 /* make sure there isn't a sync pending and yield */ 3899 wakeup(sd); 3900 while (sd->sd_sync || sd->sd_must_flush) { 3901 ret = tsleep_nsec(&sd->sd_sync, MAXPRI, "sr_down", 3902 SEC_TO_NSEC(60)); 3903 if (ret == EWOULDBLOCK) 3904 break; 3905 } 3906 if (dying == -1) { 3907 sd->sd_ready = 1; 3908 splx(s); 3909 return; 3910 } 3911 3912 #ifndef SMALL_KERNEL 3913 sr_sensors_delete(sd); 3914 #endif /* SMALL_KERNEL */ 3915 3916 if (sd->sd_target != 0) 3917 scsi_detach_lun(sc->sc_scsibus, sd->sd_target, 0, 3918 dying ? 0 : DETACH_FORCE); 3919 3920 sr_chunks_unwind(sc, &sd->sd_vol.sv_chunk_list); 3921 3922 if (sd->sd_taskq) 3923 taskq_destroy(sd->sd_taskq); 3924 3925 sr_discipline_free(sd); 3926 3927 splx(s); 3928 } 3929 3930 int 3931 sr_discipline_init(struct sr_discipline *sd, int level) 3932 { 3933 int rv = 1; 3934 3935 /* Initialise discipline function pointers with defaults. */ 3936 sd->sd_alloc_resources = sr_alloc_resources; 3937 sd->sd_assemble = NULL; 3938 sd->sd_create = NULL; 3939 sd->sd_free_resources = sr_free_resources; 3940 sd->sd_ioctl_handler = NULL; 3941 sd->sd_openings = NULL; 3942 sd->sd_meta_opt_handler = NULL; 3943 sd->sd_rebuild = sr_rebuild; 3944 sd->sd_scsi_inquiry = sr_raid_inquiry; 3945 sd->sd_scsi_read_cap = sr_raid_read_cap; 3946 sd->sd_scsi_tur = sr_raid_tur; 3947 sd->sd_scsi_req_sense = sr_raid_request_sense; 3948 sd->sd_scsi_start_stop = sr_raid_start_stop; 3949 sd->sd_scsi_sync = sr_raid_sync; 3950 sd->sd_scsi_rw = NULL; 3951 sd->sd_scsi_intr = sr_raid_intr; 3952 sd->sd_scsi_wu_done = NULL; 3953 sd->sd_scsi_done = NULL; 3954 sd->sd_set_chunk_state = sr_set_chunk_state; 3955 sd->sd_set_vol_state = sr_set_vol_state; 3956 sd->sd_start_discipline = NULL; 3957 3958 task_set(&sd->sd_meta_save_task, sr_meta_save_callback, sd); 3959 task_set(&sd->sd_hotspare_rebuild_task, sr_hotspare_rebuild_callback, 3960 sd); 3961 3962 sd->sd_wu_size = sizeof(struct sr_workunit); 3963 switch (level) { 3964 case 0: 3965 sr_raid0_discipline_init(sd); 3966 break; 3967 case 1: 3968 sr_raid1_discipline_init(sd); 3969 break; 3970 case 5: 3971 sr_raid5_discipline_init(sd); 3972 break; 3973 case 6: 3974 sr_raid6_discipline_init(sd); 3975 break; 3976 #ifdef CRYPTO 3977 case 'C': 3978 sr_crypto_discipline_init(sd); 3979 break; 3980 #endif 3981 case 'c': 3982 sr_concat_discipline_init(sd); 3983 break; 3984 default: 3985 goto bad; 3986 } 3987 3988 rv = 0; 3989 bad: 3990 return (rv); 3991 } 3992 3993 int 3994 sr_raid_inquiry(struct sr_workunit *wu) 3995 { 3996 struct sr_discipline *sd = wu->swu_dis; 3997 struct scsi_xfer *xs = wu->swu_xs; 3998 struct scsi_inquiry *cdb = (struct scsi_inquiry *)&xs->cmd; 3999 struct scsi_inquiry_data inq; 4000 4001 DNPRINTF(SR_D_DIS, "%s: sr_raid_inquiry\n", DEVNAME(sd->sd_sc)); 4002 4003 if (xs->cmdlen != sizeof(*cdb)) 4004 return (EINVAL); 4005 4006 if (ISSET(cdb->flags, SI_EVPD)) 4007 return (EOPNOTSUPP); 4008 4009 bzero(&inq, sizeof(inq)); 4010 inq.device = T_DIRECT; 4011 inq.dev_qual2 = 0; 4012 inq.version = SCSI_REV_2; 4013 inq.response_format = SID_SCSI2_RESPONSE; 4014 inq.additional_length = SID_SCSI2_ALEN; 4015 inq.flags |= SID_CmdQue; 4016 strlcpy(inq.vendor, sd->sd_meta->ssdi.ssd_vendor, 4017 sizeof(inq.vendor)); 4018 strlcpy(inq.product, sd->sd_meta->ssdi.ssd_product, 4019 sizeof(inq.product)); 4020 strlcpy(inq.revision, sd->sd_meta->ssdi.ssd_revision, 4021 sizeof(inq.revision)); 4022 scsi_copy_internal_data(xs, &inq, sizeof(inq)); 4023 4024 return (0); 4025 } 4026 4027 int 4028 sr_raid_read_cap(struct sr_workunit *wu) 4029 { 4030 struct sr_discipline *sd = wu->swu_dis; 4031 struct scsi_xfer *xs = wu->swu_xs; 4032 struct scsi_read_cap_data rcd; 4033 struct scsi_read_cap_data_16 rcd16; 4034 u_int64_t addr; 4035 int rv = 1; 4036 u_int32_t secsize; 4037 4038 DNPRINTF(SR_D_DIS, "%s: sr_raid_read_cap\n", DEVNAME(sd->sd_sc)); 4039 4040 secsize = sd->sd_meta->ssdi.ssd_secsize; 4041 4042 addr = ((sd->sd_meta->ssdi.ssd_size * DEV_BSIZE) / secsize) - 1; 4043 if (xs->cmd.opcode == READ_CAPACITY) { 4044 bzero(&rcd, sizeof(rcd)); 4045 if (addr > 0xffffffffllu) 4046 _lto4b(0xffffffff, rcd.addr); 4047 else 4048 _lto4b(addr, rcd.addr); 4049 _lto4b(secsize, rcd.length); 4050 scsi_copy_internal_data(xs, &rcd, sizeof(rcd)); 4051 rv = 0; 4052 } else if (xs->cmd.opcode == READ_CAPACITY_16) { 4053 bzero(&rcd16, sizeof(rcd16)); 4054 _lto8b(addr, rcd16.addr); 4055 _lto4b(secsize, rcd16.length); 4056 scsi_copy_internal_data(xs, &rcd16, sizeof(rcd16)); 4057 rv = 0; 4058 } 4059 4060 return (rv); 4061 } 4062 4063 int 4064 sr_raid_tur(struct sr_workunit *wu) 4065 { 4066 struct sr_discipline *sd = wu->swu_dis; 4067 4068 DNPRINTF(SR_D_DIS, "%s: sr_raid_tur\n", DEVNAME(sd->sd_sc)); 4069 4070 if (sd->sd_vol_status == BIOC_SVOFFLINE) { 4071 sd->sd_scsi_sense.error_code = SSD_ERRCODE_CURRENT; 4072 sd->sd_scsi_sense.flags = SKEY_NOT_READY; 4073 sd->sd_scsi_sense.add_sense_code = 0x04; 4074 sd->sd_scsi_sense.add_sense_code_qual = 0x11; 4075 sd->sd_scsi_sense.extra_len = 4; 4076 return (1); 4077 } else if (sd->sd_vol_status == BIOC_SVINVALID) { 4078 sd->sd_scsi_sense.error_code = SSD_ERRCODE_CURRENT; 4079 sd->sd_scsi_sense.flags = SKEY_HARDWARE_ERROR; 4080 sd->sd_scsi_sense.add_sense_code = 0x05; 4081 sd->sd_scsi_sense.add_sense_code_qual = 0x00; 4082 sd->sd_scsi_sense.extra_len = 4; 4083 return (1); 4084 } 4085 4086 return (0); 4087 } 4088 4089 int 4090 sr_raid_request_sense(struct sr_workunit *wu) 4091 { 4092 struct sr_discipline *sd = wu->swu_dis; 4093 struct scsi_xfer *xs = wu->swu_xs; 4094 4095 DNPRINTF(SR_D_DIS, "%s: sr_raid_request_sense\n", 4096 DEVNAME(sd->sd_sc)); 4097 4098 /* use latest sense data */ 4099 memcpy(&xs->sense, &sd->sd_scsi_sense, sizeof(xs->sense)); 4100 4101 /* clear sense data */ 4102 bzero(&sd->sd_scsi_sense, sizeof(sd->sd_scsi_sense)); 4103 4104 return (0); 4105 } 4106 4107 int 4108 sr_raid_start_stop(struct sr_workunit *wu) 4109 { 4110 struct scsi_xfer *xs = wu->swu_xs; 4111 struct scsi_start_stop *ss = (struct scsi_start_stop *)&xs->cmd; 4112 4113 DNPRINTF(SR_D_DIS, "%s: sr_raid_start_stop\n", 4114 DEVNAME(wu->swu_dis->sd_sc)); 4115 4116 if (!ss) 4117 return (1); 4118 4119 /* 4120 * do nothing! 4121 * a softraid discipline should always reflect correct status 4122 */ 4123 return (0); 4124 } 4125 4126 int 4127 sr_raid_sync(struct sr_workunit *wu) 4128 { 4129 struct sr_discipline *sd = wu->swu_dis; 4130 int s, ret, rv = 0, ios; 4131 4132 DNPRINTF(SR_D_DIS, "%s: sr_raid_sync\n", DEVNAME(sd->sd_sc)); 4133 4134 /* when doing a fake sync don't count the wu */ 4135 ios = (wu->swu_flags & SR_WUF_FAKE) ? 0 : 1; 4136 4137 s = splbio(); 4138 sd->sd_sync = 1; 4139 while (sd->sd_wu_pending > ios) { 4140 ret = tsleep_nsec(sd, PRIBIO, "sr_sync", SEC_TO_NSEC(15)); 4141 if (ret == EWOULDBLOCK) { 4142 DNPRINTF(SR_D_DIS, "%s: sr_raid_sync timeout\n", 4143 DEVNAME(sd->sd_sc)); 4144 rv = 1; 4145 break; 4146 } 4147 } 4148 sd->sd_sync = 0; 4149 splx(s); 4150 4151 wakeup(&sd->sd_sync); 4152 4153 return (rv); 4154 } 4155 4156 void 4157 sr_raid_intr(struct buf *bp) 4158 { 4159 struct sr_ccb *ccb = (struct sr_ccb *)bp; 4160 struct sr_workunit *wu = ccb->ccb_wu; 4161 #ifdef SR_DEBUG 4162 struct sr_discipline *sd = wu->swu_dis; 4163 struct scsi_xfer *xs = wu->swu_xs; 4164 #endif 4165 int s; 4166 4167 DNPRINTF(SR_D_INTR, "%s: %s %s intr bp %p xs %p\n", 4168 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, sd->sd_name, bp, xs); 4169 4170 s = splbio(); 4171 sr_ccb_done(ccb); 4172 sr_wu_done(wu); 4173 splx(s); 4174 } 4175 4176 void 4177 sr_schedule_wu(struct sr_workunit *wu) 4178 { 4179 struct sr_discipline *sd = wu->swu_dis; 4180 struct sr_workunit *wup; 4181 int s; 4182 4183 DNPRINTF(SR_D_WU, "sr_schedule_wu: schedule wu %p state %i " 4184 "flags 0x%x\n", wu, wu->swu_state, wu->swu_flags); 4185 4186 KASSERT(wu->swu_io_count > 0); 4187 4188 s = splbio(); 4189 4190 /* Construct the work unit, do not schedule it. */ 4191 if (wu->swu_state == SR_WU_CONSTRUCT) 4192 goto queued; 4193 4194 /* Deferred work unit being reconstructed, do not start. */ 4195 if (wu->swu_state == SR_WU_REQUEUE) 4196 goto queued; 4197 4198 /* Current work unit failed, restart. */ 4199 if (wu->swu_state == SR_WU_RESTART) 4200 goto start; 4201 4202 if (wu->swu_state != SR_WU_INPROGRESS) 4203 panic("sr_schedule_wu: work unit not in progress (state %i)\n", 4204 wu->swu_state); 4205 4206 /* Walk queue backwards and fill in collider if we have one. */ 4207 TAILQ_FOREACH_REVERSE(wup, &sd->sd_wu_pendq, sr_wu_list, swu_link) { 4208 if (wu->swu_blk_end < wup->swu_blk_start || 4209 wup->swu_blk_end < wu->swu_blk_start) 4210 continue; 4211 4212 /* Defer work unit due to LBA collision. */ 4213 DNPRINTF(SR_D_WU, "sr_schedule_wu: deferring work unit %p\n", 4214 wu); 4215 wu->swu_state = SR_WU_DEFERRED; 4216 while (wup->swu_collider) 4217 wup = wup->swu_collider; 4218 wup->swu_collider = wu; 4219 TAILQ_INSERT_TAIL(&sd->sd_wu_defq, wu, swu_link); 4220 sd->sd_wu_collisions++; 4221 goto queued; 4222 } 4223 4224 start: 4225 sr_raid_startwu(wu); 4226 4227 queued: 4228 splx(s); 4229 } 4230 4231 void 4232 sr_raid_startwu(struct sr_workunit *wu) 4233 { 4234 struct sr_discipline *sd = wu->swu_dis; 4235 struct sr_ccb *ccb; 4236 4237 DNPRINTF(SR_D_WU, "sr_raid_startwu: start wu %p\n", wu); 4238 4239 splassert(IPL_BIO); 4240 4241 if (wu->swu_state == SR_WU_DEFERRED) { 4242 TAILQ_REMOVE(&sd->sd_wu_defq, wu, swu_link); 4243 wu->swu_state = SR_WU_INPROGRESS; 4244 } 4245 4246 if (wu->swu_state != SR_WU_RESTART) 4247 TAILQ_INSERT_TAIL(&sd->sd_wu_pendq, wu, swu_link); 4248 4249 /* Start all of the individual I/Os. */ 4250 if (wu->swu_cb_active == 1) 4251 panic("%s: sr_startwu_callback", DEVNAME(sd->sd_sc)); 4252 wu->swu_cb_active = 1; 4253 4254 TAILQ_FOREACH(ccb, &wu->swu_ccb, ccb_link) 4255 VOP_STRATEGY(&ccb->ccb_buf); 4256 4257 wu->swu_cb_active = 0; 4258 } 4259 4260 void 4261 sr_raid_recreate_wu(struct sr_workunit *wu) 4262 { 4263 struct sr_discipline *sd = wu->swu_dis; 4264 struct sr_workunit *wup = wu; 4265 4266 /* 4267 * Recreate a work unit by releasing the associated CCBs and reissuing 4268 * the SCSI I/O request. This process is then repeated for all of the 4269 * colliding work units. 4270 */ 4271 do { 4272 sr_wu_release_ccbs(wup); 4273 4274 wup->swu_state = SR_WU_REQUEUE; 4275 if (sd->sd_scsi_rw(wup)) 4276 panic("could not requeue I/O"); 4277 4278 wup = wup->swu_collider; 4279 } while (wup); 4280 } 4281 4282 int 4283 sr_alloc_resources(struct sr_discipline *sd) 4284 { 4285 if (sr_wu_alloc(sd)) { 4286 sr_error(sd->sd_sc, "unable to allocate work units"); 4287 return (ENOMEM); 4288 } 4289 if (sr_ccb_alloc(sd)) { 4290 sr_error(sd->sd_sc, "unable to allocate ccbs"); 4291 return (ENOMEM); 4292 } 4293 4294 return (0); 4295 } 4296 4297 void 4298 sr_free_resources(struct sr_discipline *sd) 4299 { 4300 sr_wu_free(sd); 4301 sr_ccb_free(sd); 4302 } 4303 4304 void 4305 sr_set_chunk_state(struct sr_discipline *sd, int c, int new_state) 4306 { 4307 int old_state, s; 4308 4309 DNPRINTF(SR_D_STATE, "%s: %s: %s: sr_set_chunk_state %d -> %d\n", 4310 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, 4311 sd->sd_vol.sv_chunks[c]->src_meta.scmi.scm_devname, c, new_state); 4312 4313 /* ok to go to splbio since this only happens in error path */ 4314 s = splbio(); 4315 old_state = sd->sd_vol.sv_chunks[c]->src_meta.scm_status; 4316 4317 /* multiple IOs to the same chunk that fail will come through here */ 4318 if (old_state == new_state) 4319 goto done; 4320 4321 switch (old_state) { 4322 case BIOC_SDONLINE: 4323 if (new_state == BIOC_SDOFFLINE) 4324 break; 4325 else 4326 goto die; 4327 break; 4328 4329 case BIOC_SDOFFLINE: 4330 goto die; 4331 4332 default: 4333 die: 4334 splx(s); /* XXX */ 4335 panic("%s: %s: %s: invalid chunk state transition " 4336 "%d -> %d", DEVNAME(sd->sd_sc), 4337 sd->sd_meta->ssd_devname, 4338 sd->sd_vol.sv_chunks[c]->src_meta.scmi.scm_devname, 4339 old_state, new_state); 4340 /* NOTREACHED */ 4341 } 4342 4343 sd->sd_vol.sv_chunks[c]->src_meta.scm_status = new_state; 4344 sd->sd_set_vol_state(sd); 4345 4346 sd->sd_must_flush = 1; 4347 task_add(systq, &sd->sd_meta_save_task); 4348 done: 4349 splx(s); 4350 } 4351 4352 void 4353 sr_set_vol_state(struct sr_discipline *sd) 4354 { 4355 int states[SR_MAX_STATES]; 4356 int new_state, i, nd; 4357 int old_state = sd->sd_vol_status; 4358 u_int32_t s; 4359 4360 DNPRINTF(SR_D_STATE, "%s: %s: sr_set_vol_state\n", 4361 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname); 4362 4363 nd = sd->sd_meta->ssdi.ssd_chunk_no; 4364 4365 for (i = 0; i < SR_MAX_STATES; i++) 4366 states[i] = 0; 4367 4368 for (i = 0; i < nd; i++) { 4369 s = sd->sd_vol.sv_chunks[i]->src_meta.scm_status; 4370 if (s >= SR_MAX_STATES) 4371 panic("%s: %s: %s: invalid chunk state", 4372 DEVNAME(sd->sd_sc), 4373 sd->sd_meta->ssd_devname, 4374 sd->sd_vol.sv_chunks[i]->src_meta.scmi.scm_devname); 4375 states[s]++; 4376 } 4377 4378 if (states[BIOC_SDONLINE] == nd) 4379 new_state = BIOC_SVONLINE; 4380 else 4381 new_state = BIOC_SVOFFLINE; 4382 4383 DNPRINTF(SR_D_STATE, "%s: %s: sr_set_vol_state %d -> %d\n", 4384 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, 4385 old_state, new_state); 4386 4387 switch (old_state) { 4388 case BIOC_SVONLINE: 4389 if (new_state == BIOC_SVOFFLINE || new_state == BIOC_SVONLINE) 4390 break; 4391 else 4392 goto die; 4393 break; 4394 4395 case BIOC_SVOFFLINE: 4396 /* XXX this might be a little too much */ 4397 goto die; 4398 4399 default: 4400 die: 4401 panic("%s: %s: invalid volume state transition " 4402 "%d -> %d", DEVNAME(sd->sd_sc), 4403 sd->sd_meta->ssd_devname, 4404 old_state, new_state); 4405 /* NOTREACHED */ 4406 } 4407 4408 sd->sd_vol_status = new_state; 4409 } 4410 4411 void * 4412 sr_block_get(struct sr_discipline *sd, long length) 4413 { 4414 return dma_alloc(length, PR_NOWAIT | PR_ZERO); 4415 } 4416 4417 void 4418 sr_block_put(struct sr_discipline *sd, void *ptr, int length) 4419 { 4420 dma_free(ptr, length); 4421 } 4422 4423 void 4424 sr_checksum_print(u_int8_t *md5) 4425 { 4426 int i; 4427 4428 for (i = 0; i < MD5_DIGEST_LENGTH; i++) 4429 printf("%02x", md5[i]); 4430 } 4431 4432 void 4433 sr_checksum(struct sr_softc *sc, void *src, void *md5, u_int32_t len) 4434 { 4435 MD5_CTX ctx; 4436 4437 DNPRINTF(SR_D_MISC, "%s: sr_checksum(%p %p %d)\n", DEVNAME(sc), src, 4438 md5, len); 4439 4440 MD5Init(&ctx); 4441 MD5Update(&ctx, src, len); 4442 MD5Final(md5, &ctx); 4443 } 4444 4445 void 4446 sr_uuid_generate(struct sr_uuid *uuid) 4447 { 4448 arc4random_buf(uuid->sui_id, sizeof(uuid->sui_id)); 4449 /* UUID version 4: random */ 4450 uuid->sui_id[6] &= 0x0f; 4451 uuid->sui_id[6] |= 0x40; 4452 /* RFC4122 variant */ 4453 uuid->sui_id[8] &= 0x3f; 4454 uuid->sui_id[8] |= 0x80; 4455 } 4456 4457 char * 4458 sr_uuid_format(struct sr_uuid *uuid) 4459 { 4460 char *uuidstr; 4461 4462 uuidstr = malloc(37, M_DEVBUF, M_WAITOK); 4463 4464 snprintf(uuidstr, 37, 4465 "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-" 4466 "%02x%02x%02x%02x%02x%02x", 4467 uuid->sui_id[0], uuid->sui_id[1], 4468 uuid->sui_id[2], uuid->sui_id[3], 4469 uuid->sui_id[4], uuid->sui_id[5], 4470 uuid->sui_id[6], uuid->sui_id[7], 4471 uuid->sui_id[8], uuid->sui_id[9], 4472 uuid->sui_id[10], uuid->sui_id[11], 4473 uuid->sui_id[12], uuid->sui_id[13], 4474 uuid->sui_id[14], uuid->sui_id[15]); 4475 4476 return uuidstr; 4477 } 4478 4479 void 4480 sr_uuid_print(struct sr_uuid *uuid, int cr) 4481 { 4482 char *uuidstr; 4483 4484 uuidstr = sr_uuid_format(uuid); 4485 printf("%s%s", uuidstr, (cr ? "\n" : "")); 4486 free(uuidstr, M_DEVBUF, 37); 4487 } 4488 4489 int 4490 sr_already_assembled(struct sr_discipline *sd) 4491 { 4492 struct sr_softc *sc = sd->sd_sc; 4493 struct sr_discipline *sdtmp; 4494 4495 TAILQ_FOREACH(sdtmp, &sc->sc_dis_list, sd_link) { 4496 if (!bcmp(&sd->sd_meta->ssdi.ssd_uuid, 4497 &sdtmp->sd_meta->ssdi.ssd_uuid, 4498 sizeof(sd->sd_meta->ssdi.ssd_uuid))) 4499 return (1); 4500 } 4501 4502 return (0); 4503 } 4504 4505 int32_t 4506 sr_validate_stripsize(u_int32_t b) 4507 { 4508 int s = 0; 4509 4510 if (b % DEV_BSIZE) 4511 return (-1); 4512 4513 while ((b & 1) == 0) { 4514 b >>= 1; 4515 s++; 4516 } 4517 4518 /* only multiple of twos */ 4519 b >>= 1; 4520 if (b) 4521 return(-1); 4522 4523 return (s); 4524 } 4525 4526 void 4527 sr_quiesce(void) 4528 { 4529 struct sr_softc *sc = softraid0; 4530 struct sr_discipline *sd, *nsd; 4531 4532 /* Shutdown disciplines in reverse attach order. */ 4533 TAILQ_FOREACH_REVERSE_SAFE(sd, &sc->sc_dis_list, 4534 sr_discipline_list, sd_link, nsd) 4535 sr_discipline_shutdown(sd, 1, -1); 4536 } 4537 4538 void 4539 sr_shutdown(int dying) 4540 { 4541 struct sr_softc *sc = softraid0; 4542 struct sr_discipline *sd; 4543 4544 DNPRINTF(SR_D_MISC, "%s: sr_shutdown\n", DEVNAME(sc)); 4545 4546 /* 4547 * Since softraid is not under mainbus, we have to explicitly 4548 * notify its children that the power is going down, so they 4549 * can execute their shutdown hooks. 4550 */ 4551 config_suspend((struct device *)sc, DVACT_POWERDOWN); 4552 4553 /* Shutdown disciplines in reverse attach order. */ 4554 while ((sd = TAILQ_LAST(&sc->sc_dis_list, sr_discipline_list)) != NULL) 4555 sr_discipline_shutdown(sd, 1, dying); 4556 } 4557 4558 int 4559 sr_validate_io(struct sr_workunit *wu, daddr_t *blkno, char *func) 4560 { 4561 struct sr_discipline *sd = wu->swu_dis; 4562 struct scsi_xfer *xs = wu->swu_xs; 4563 int rv = 1; 4564 4565 DNPRINTF(SR_D_DIS, "%s: %s 0x%02x\n", DEVNAME(sd->sd_sc), func, 4566 xs->cmd.opcode); 4567 4568 if (sd->sd_meta->ssd_data_blkno == 0) 4569 panic("invalid data blkno"); 4570 4571 if (sd->sd_vol_status == BIOC_SVOFFLINE) { 4572 DNPRINTF(SR_D_DIS, "%s: %s device offline\n", 4573 DEVNAME(sd->sd_sc), func); 4574 goto bad; 4575 } 4576 4577 if (xs->datalen == 0) { 4578 printf("%s: %s: illegal block count for %s\n", 4579 DEVNAME(sd->sd_sc), func, sd->sd_meta->ssd_devname); 4580 goto bad; 4581 } 4582 4583 if (xs->cmdlen == 10) 4584 *blkno = _4btol(((struct scsi_rw_10 *)&xs->cmd)->addr); 4585 else if (xs->cmdlen == 16) 4586 *blkno = _8btol(((struct scsi_rw_16 *)&xs->cmd)->addr); 4587 else if (xs->cmdlen == 6) 4588 *blkno = _3btol(((struct scsi_rw *)&xs->cmd)->addr); 4589 else { 4590 printf("%s: %s: illegal cmdlen for %s\n", 4591 DEVNAME(sd->sd_sc), func, sd->sd_meta->ssd_devname); 4592 goto bad; 4593 } 4594 4595 *blkno *= (sd->sd_meta->ssdi.ssd_secsize / DEV_BSIZE); 4596 4597 wu->swu_blk_start = *blkno; 4598 wu->swu_blk_end = *blkno + (xs->datalen >> DEV_BSHIFT) - 1; 4599 4600 if (wu->swu_blk_end > sd->sd_meta->ssdi.ssd_size) { 4601 DNPRINTF(SR_D_DIS, "%s: %s out of bounds start: %lld " 4602 "end: %lld length: %d\n", 4603 DEVNAME(sd->sd_sc), func, (long long)wu->swu_blk_start, 4604 (long long)wu->swu_blk_end, xs->datalen); 4605 4606 sd->sd_scsi_sense.error_code = SSD_ERRCODE_CURRENT | 4607 SSD_ERRCODE_VALID; 4608 sd->sd_scsi_sense.flags = SKEY_ILLEGAL_REQUEST; 4609 sd->sd_scsi_sense.add_sense_code = 0x21; 4610 sd->sd_scsi_sense.add_sense_code_qual = 0x00; 4611 sd->sd_scsi_sense.extra_len = 4; 4612 goto bad; 4613 } 4614 4615 rv = 0; 4616 bad: 4617 return (rv); 4618 } 4619 4620 void 4621 sr_rebuild_start(void *arg) 4622 { 4623 struct sr_discipline *sd = arg; 4624 struct sr_softc *sc = sd->sd_sc; 4625 4626 DNPRINTF(SR_D_REBUILD, "%s: %s starting rebuild thread\n", 4627 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname); 4628 4629 if (kthread_create(sr_rebuild_thread, sd, &sd->sd_background_proc, 4630 DEVNAME(sc)) != 0) 4631 printf("%s: unable to start background operation\n", 4632 DEVNAME(sc)); 4633 } 4634 4635 void 4636 sr_rebuild_thread(void *arg) 4637 { 4638 struct sr_discipline *sd = arg; 4639 4640 DNPRINTF(SR_D_REBUILD, "%s: %s rebuild thread started\n", 4641 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname); 4642 4643 sd->sd_reb_active = 1; 4644 sd->sd_rebuild(sd); 4645 sd->sd_reb_active = 0; 4646 4647 kthread_exit(0); 4648 } 4649 4650 void 4651 sr_rebuild(struct sr_discipline *sd) 4652 { 4653 struct sr_softc *sc = sd->sd_sc; 4654 u_int64_t sz, whole_blk, partial_blk, blk, restart; 4655 daddr_t lba; 4656 struct sr_workunit *wu_r, *wu_w; 4657 struct scsi_xfer xs_r, xs_w; 4658 struct scsi_rw_16 *cr, *cw; 4659 int c, s, slept, percent = 0, old_percent = -1; 4660 u_int8_t *buf; 4661 4662 whole_blk = sd->sd_meta->ssdi.ssd_size / SR_REBUILD_IO_SIZE; 4663 partial_blk = sd->sd_meta->ssdi.ssd_size % SR_REBUILD_IO_SIZE; 4664 4665 restart = sd->sd_meta->ssd_rebuild / SR_REBUILD_IO_SIZE; 4666 if (restart > whole_blk) { 4667 printf("%s: bogus rebuild restart offset, starting from 0\n", 4668 DEVNAME(sc)); 4669 restart = 0; 4670 } 4671 if (restart) { 4672 /* 4673 * XXX there is a hole here; there is a posibility that we 4674 * had a restart however the chunk that was supposed to 4675 * be rebuilt is no longer valid; we can reach this situation 4676 * when a rebuild is in progress and the box crashes and 4677 * on reboot the rebuild chunk is different (like zero'd or 4678 * replaced). We need to check the uuid of the chunk that is 4679 * being rebuilt to assert this. 4680 */ 4681 percent = sr_rebuild_percent(sd); 4682 printf("%s: resuming rebuild on %s at %d%%\n", 4683 DEVNAME(sc), sd->sd_meta->ssd_devname, percent); 4684 } 4685 4686 /* currently this is 64k therefore we can use dma_alloc */ 4687 buf = dma_alloc(SR_REBUILD_IO_SIZE << DEV_BSHIFT, PR_WAITOK); 4688 for (blk = restart; blk <= whole_blk; blk++) { 4689 lba = blk * SR_REBUILD_IO_SIZE; 4690 sz = SR_REBUILD_IO_SIZE; 4691 if (blk == whole_blk) { 4692 if (partial_blk == 0) 4693 break; 4694 sz = partial_blk; 4695 } 4696 4697 /* get some wu */ 4698 wu_r = sr_scsi_wu_get(sd, 0); 4699 wu_w = sr_scsi_wu_get(sd, 0); 4700 4701 DNPRINTF(SR_D_REBUILD, "%s: %s rebuild wu_r %p, wu_w %p\n", 4702 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, wu_r, wu_w); 4703 4704 /* setup read io */ 4705 bzero(&xs_r, sizeof xs_r); 4706 xs_r.error = XS_NOERROR; 4707 xs_r.flags = SCSI_DATA_IN; 4708 xs_r.datalen = sz << DEV_BSHIFT; 4709 xs_r.data = buf; 4710 xs_r.cmdlen = sizeof(*cr); 4711 cr = (struct scsi_rw_16 *)&xs_r.cmd; 4712 cr->opcode = READ_16; 4713 _lto4b(sz, cr->length); 4714 _lto8b(lba, cr->addr); 4715 wu_r->swu_state = SR_WU_CONSTRUCT; 4716 wu_r->swu_flags |= SR_WUF_REBUILD; 4717 wu_r->swu_xs = &xs_r; 4718 if (sd->sd_scsi_rw(wu_r)) { 4719 printf("%s: could not create read io\n", 4720 DEVNAME(sc)); 4721 goto fail; 4722 } 4723 4724 /* setup write io */ 4725 bzero(&xs_w, sizeof xs_w); 4726 xs_w.error = XS_NOERROR; 4727 xs_w.flags = SCSI_DATA_OUT; 4728 xs_w.datalen = sz << DEV_BSHIFT; 4729 xs_w.data = buf; 4730 xs_w.cmdlen = sizeof(*cw); 4731 cw = (struct scsi_rw_16 *)&xs_w.cmd; 4732 cw->opcode = WRITE_16; 4733 _lto4b(sz, cw->length); 4734 _lto8b(lba, cw->addr); 4735 wu_w->swu_state = SR_WU_CONSTRUCT; 4736 wu_w->swu_flags |= SR_WUF_REBUILD | SR_WUF_WAKEUP; 4737 wu_w->swu_xs = &xs_w; 4738 if (sd->sd_scsi_rw(wu_w)) { 4739 printf("%s: could not create write io\n", 4740 DEVNAME(sc)); 4741 goto fail; 4742 } 4743 4744 /* 4745 * collide with the read io so that we get automatically 4746 * started when the read is done 4747 */ 4748 wu_w->swu_state = SR_WU_DEFERRED; 4749 wu_r->swu_collider = wu_w; 4750 s = splbio(); 4751 TAILQ_INSERT_TAIL(&sd->sd_wu_defq, wu_w, swu_link); 4752 splx(s); 4753 4754 DNPRINTF(SR_D_REBUILD, "%s: %s rebuild scheduling wu_r %p\n", 4755 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, wu_r); 4756 4757 wu_r->swu_state = SR_WU_INPROGRESS; 4758 sr_schedule_wu(wu_r); 4759 4760 /* wait for write completion */ 4761 slept = 0; 4762 while ((wu_w->swu_flags & SR_WUF_REBUILDIOCOMP) == 0) { 4763 tsleep_nsec(wu_w, PRIBIO, "sr_rebuild", INFSLP); 4764 slept = 1; 4765 } 4766 /* yield if we didn't sleep */ 4767 if (slept == 0) 4768 tsleep(sc, PWAIT, "sr_yield", 1); 4769 4770 sr_scsi_wu_put(sd, wu_r); 4771 sr_scsi_wu_put(sd, wu_w); 4772 4773 sd->sd_meta->ssd_rebuild = lba; 4774 4775 /* XXX - this should be based on size, not percentage. */ 4776 /* save metadata every percent */ 4777 percent = sr_rebuild_percent(sd); 4778 if (percent != old_percent && blk != whole_blk) { 4779 if (sr_meta_save(sd, SR_META_DIRTY)) 4780 printf("%s: could not save metadata to %s\n", 4781 DEVNAME(sc), sd->sd_meta->ssd_devname); 4782 old_percent = percent; 4783 } 4784 4785 if (sd->sd_reb_abort) 4786 goto abort; 4787 } 4788 4789 /* all done */ 4790 sd->sd_meta->ssd_rebuild = 0; 4791 for (c = 0; c < sd->sd_meta->ssdi.ssd_chunk_no; c++) { 4792 if (sd->sd_vol.sv_chunks[c]->src_meta.scm_status == 4793 BIOC_SDREBUILD) { 4794 sd->sd_set_chunk_state(sd, c, BIOC_SDONLINE); 4795 break; 4796 } 4797 } 4798 4799 abort: 4800 if (sr_meta_save(sd, SR_META_DIRTY)) 4801 printf("%s: could not save metadata to %s\n", 4802 DEVNAME(sc), sd->sd_meta->ssd_devname); 4803 fail: 4804 dma_free(buf, SR_REBUILD_IO_SIZE << DEV_BSHIFT); 4805 } 4806 4807 #ifndef SMALL_KERNEL 4808 int 4809 sr_sensors_create(struct sr_discipline *sd) 4810 { 4811 struct sr_softc *sc = sd->sd_sc; 4812 int rv = 1; 4813 4814 DNPRINTF(SR_D_STATE, "%s: %s: sr_sensors_create\n", 4815 DEVNAME(sc), sd->sd_meta->ssd_devname); 4816 4817 sd->sd_vol.sv_sensor.type = SENSOR_DRIVE; 4818 sd->sd_vol.sv_sensor.status = SENSOR_S_UNKNOWN; 4819 strlcpy(sd->sd_vol.sv_sensor.desc, sd->sd_meta->ssd_devname, 4820 sizeof(sd->sd_vol.sv_sensor.desc)); 4821 4822 sensor_attach(&sc->sc_sensordev, &sd->sd_vol.sv_sensor); 4823 sd->sd_vol.sv_sensor_attached = 1; 4824 4825 if (sc->sc_sensor_task == NULL) { 4826 sc->sc_sensor_task = sensor_task_register(sc, 4827 sr_sensors_refresh, 10); 4828 if (sc->sc_sensor_task == NULL) 4829 goto bad; 4830 } 4831 4832 rv = 0; 4833 bad: 4834 return (rv); 4835 } 4836 4837 void 4838 sr_sensors_delete(struct sr_discipline *sd) 4839 { 4840 DNPRINTF(SR_D_STATE, "%s: sr_sensors_delete\n", DEVNAME(sd->sd_sc)); 4841 4842 if (sd->sd_vol.sv_sensor_attached) 4843 sensor_detach(&sd->sd_sc->sc_sensordev, &sd->sd_vol.sv_sensor); 4844 } 4845 4846 void 4847 sr_sensors_refresh(void *arg) 4848 { 4849 struct sr_softc *sc = arg; 4850 struct sr_volume *sv; 4851 struct sr_discipline *sd; 4852 4853 DNPRINTF(SR_D_STATE, "%s: sr_sensors_refresh\n", DEVNAME(sc)); 4854 4855 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 4856 sv = &sd->sd_vol; 4857 4858 switch(sd->sd_vol_status) { 4859 case BIOC_SVOFFLINE: 4860 sv->sv_sensor.value = SENSOR_DRIVE_FAIL; 4861 sv->sv_sensor.status = SENSOR_S_CRIT; 4862 break; 4863 4864 case BIOC_SVDEGRADED: 4865 sv->sv_sensor.value = SENSOR_DRIVE_PFAIL; 4866 sv->sv_sensor.status = SENSOR_S_WARN; 4867 break; 4868 4869 case BIOC_SVREBUILD: 4870 sv->sv_sensor.value = SENSOR_DRIVE_REBUILD; 4871 sv->sv_sensor.status = SENSOR_S_WARN; 4872 break; 4873 4874 case BIOC_SVSCRUB: 4875 case BIOC_SVONLINE: 4876 sv->sv_sensor.value = SENSOR_DRIVE_ONLINE; 4877 sv->sv_sensor.status = SENSOR_S_OK; 4878 break; 4879 4880 default: 4881 sv->sv_sensor.value = 0; /* unknown */ 4882 sv->sv_sensor.status = SENSOR_S_UNKNOWN; 4883 } 4884 } 4885 } 4886 #endif /* SMALL_KERNEL */ 4887 4888 #ifdef SR_FANCY_STATS 4889 void sr_print_stats(void); 4890 4891 void 4892 sr_print_stats(void) 4893 { 4894 struct sr_softc *sc = softraid0; 4895 struct sr_discipline *sd; 4896 4897 if (sc == NULL) { 4898 printf("no softraid softc found\n"); 4899 return; 4900 } 4901 4902 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 4903 printf("%s: ios pending %d, collisions %llu\n", 4904 sd->sd_meta->ssd_devname, 4905 sd->sd_wu_pending, 4906 sd->sd_wu_collisions); 4907 } 4908 } 4909 #endif /* SR_FANCY_STATS */ 4910 4911 #ifdef SR_DEBUG 4912 void 4913 sr_meta_print(struct sr_metadata *m) 4914 { 4915 int i; 4916 struct sr_meta_chunk *mc; 4917 struct sr_meta_opt_hdr *omh; 4918 4919 if (!(sr_debug & SR_D_META)) 4920 return; 4921 4922 printf("\tssd_magic 0x%llx\n", m->ssdi.ssd_magic); 4923 printf("\tssd_version %d\n", m->ssdi.ssd_version); 4924 printf("\tssd_vol_flags 0x%x\n", m->ssdi.ssd_vol_flags); 4925 printf("\tssd_uuid "); 4926 sr_uuid_print(&m->ssdi.ssd_uuid, 1); 4927 printf("\tssd_chunk_no %d\n", m->ssdi.ssd_chunk_no); 4928 printf("\tssd_chunk_id %d\n", m->ssdi.ssd_chunk_id); 4929 printf("\tssd_opt_no %d\n", m->ssdi.ssd_opt_no); 4930 printf("\tssd_volid %d\n", m->ssdi.ssd_volid); 4931 printf("\tssd_level %d\n", m->ssdi.ssd_level); 4932 printf("\tssd_size %lld\n", m->ssdi.ssd_size); 4933 printf("\tssd_devname %s\n", m->ssd_devname); 4934 printf("\tssd_vendor %s\n", m->ssdi.ssd_vendor); 4935 printf("\tssd_product %s\n", m->ssdi.ssd_product); 4936 printf("\tssd_revision %s\n", m->ssdi.ssd_revision); 4937 printf("\tssd_strip_size %d\n", m->ssdi.ssd_strip_size); 4938 printf("\tssd_checksum "); 4939 sr_checksum_print(m->ssd_checksum); 4940 printf("\n"); 4941 printf("\tssd_meta_flags 0x%x\n", m->ssd_meta_flags); 4942 printf("\tssd_ondisk %llu\n", m->ssd_ondisk); 4943 4944 mc = (struct sr_meta_chunk *)(m + 1); 4945 for (i = 0; i < m->ssdi.ssd_chunk_no; i++, mc++) { 4946 printf("\t\tscm_volid %d\n", mc->scmi.scm_volid); 4947 printf("\t\tscm_chunk_id %d\n", mc->scmi.scm_chunk_id); 4948 printf("\t\tscm_devname %s\n", mc->scmi.scm_devname); 4949 printf("\t\tscm_size %lld\n", mc->scmi.scm_size); 4950 printf("\t\tscm_coerced_size %lld\n",mc->scmi.scm_coerced_size); 4951 printf("\t\tscm_uuid "); 4952 sr_uuid_print(&mc->scmi.scm_uuid, 1); 4953 printf("\t\tscm_checksum "); 4954 sr_checksum_print(mc->scm_checksum); 4955 printf("\n"); 4956 printf("\t\tscm_status %d\n", mc->scm_status); 4957 } 4958 4959 omh = (struct sr_meta_opt_hdr *)((u_int8_t *)(m + 1) + 4960 sizeof(struct sr_meta_chunk) * m->ssdi.ssd_chunk_no); 4961 for (i = 0; i < m->ssdi.ssd_opt_no; i++) { 4962 printf("\t\t\tsom_type %d\n", omh->som_type); 4963 printf("\t\t\tsom_checksum "); 4964 sr_checksum_print(omh->som_checksum); 4965 printf("\n"); 4966 omh = (struct sr_meta_opt_hdr *)((void *)omh + 4967 omh->som_length); 4968 } 4969 } 4970 4971 void 4972 sr_dump_block(void *blk, int len) 4973 { 4974 uint8_t *b = blk; 4975 int i, j, c; 4976 4977 for (i = 0; i < len; i += 16) { 4978 for (j = 0; j < 16; j++) 4979 printf("%.2x ", b[i + j]); 4980 printf(" "); 4981 for (j = 0; j < 16; j++) { 4982 c = b[i + j]; 4983 if (c < ' ' || c > 'z' || i + j > len) 4984 c = '.'; 4985 printf("%c", c); 4986 } 4987 printf("\n"); 4988 } 4989 } 4990 4991 void 4992 sr_dump_mem(u_int8_t *p, int len) 4993 { 4994 int i; 4995 4996 for (i = 0; i < len; i++) 4997 printf("%02x ", *p++); 4998 printf("\n"); 4999 } 5000 5001 #endif /* SR_DEBUG */ 5002 5003 #ifdef HIBERNATE 5004 /* 5005 * Side-effect free (no malloc, printf, pool, splx) softraid crypto writer. 5006 * 5007 * This function must perform the following: 5008 * 1. Determine the underlying device's own side-effect free I/O function 5009 * (eg, ahci_hibernate_io, wd_hibernate_io, etc). 5010 * 2. Store enough information in the provided page argument for subsequent 5011 * I/O calls (such as the crypto discipline structure for the keys, the 5012 * offset of the softraid partition on the underlying disk, as well as 5013 * the offset of the swap partition within the crypto volume. 5014 * 3. Encrypt the incoming data using the sr_discipline keys, then pass 5015 * the request to the underlying device's own I/O function. 5016 */ 5017 int 5018 sr_hibernate_io(dev_t dev, daddr_t blkno, vaddr_t addr, size_t size, int op, void *page) 5019 { 5020 /* Struct for stashing data obtained on HIB_INIT. 5021 * XXX 5022 * We share the page with the underlying device's own 5023 * side-effect free I/O function, so we pad our data to 5024 * the end of the page. Presently this does not overlap 5025 * with either of the two other side-effect free i/o 5026 * functions (ahci/wd). 5027 */ 5028 struct { 5029 char pad[3072]; 5030 struct sr_discipline *srd; 5031 hibio_fn subfn; /* underlying device i/o fn */ 5032 dev_t subdev; /* underlying device dev_t */ 5033 daddr_t sr_swapoff; /* ofs of swap part in sr volume */ 5034 char buf[DEV_BSIZE]; /* encryption performed into this buf */ 5035 } *my = page; 5036 extern struct cfdriver sd_cd; 5037 char errstr[128], *dl_ret; 5038 struct sr_chunk *schunk; 5039 struct sd_softc *sd; 5040 struct aes_xts_ctx ctx; 5041 struct sr_softc *sc; 5042 struct device *dv; 5043 daddr_t key_blkno; 5044 uint32_t sub_raidoff; /* ofs of sr part in underlying dev */ 5045 struct disklabel dl; 5046 struct partition *pp; 5047 size_t i, j; 5048 u_char iv[8]; 5049 5050 /* 5051 * In HIB_INIT, we are passed the swap partition size and offset 5052 * in 'size' and 'blkno' respectively. These are relative to the 5053 * start of the softraid partition, and we need to save these 5054 * for later translation to the underlying device's layout. 5055 */ 5056 if (op == HIB_INIT) { 5057 dv = disk_lookup(&sd_cd, DISKUNIT(dev)); 5058 sd = (struct sd_softc *)dv; 5059 sc = (struct sr_softc *)dv->dv_parent->dv_parent; 5060 5061 /* 5062 * Look up the sr discipline. This is used to determine 5063 * if we are SR crypto and what the underlying device is. 5064 */ 5065 my->srd = sc->sc_targets[sd->sc_link->target]; 5066 DNPRINTF(SR_D_MISC, "sr_hibernate_io: discipline is %s\n", 5067 my->srd->sd_name); 5068 if (strncmp(my->srd->sd_name, "CRYPTO", 5069 sizeof(my->srd->sd_name))) 5070 return (ENOTSUP); 5071 5072 /* Find the underlying device */ 5073 schunk = my->srd->sd_vol.sv_chunks[0]; 5074 my->subdev = schunk->src_dev_mm; 5075 5076 /* 5077 * Find the appropriate underlying device side effect free 5078 * I/O function, based on the type of device it is. 5079 */ 5080 my->subfn = get_hibernate_io_function(my->subdev); 5081 if (!my->subfn) 5082 return (ENODEV); 5083 5084 /* 5085 * Find blkno where this raid partition starts on 5086 * the underlying disk. 5087 */ 5088 dl_ret = disk_readlabel(&dl, my->subdev, errstr, 5089 sizeof(errstr)); 5090 if (dl_ret) { 5091 printf("Hibernate error reading disklabel: %s\n", dl_ret); 5092 return (ENOTSUP); 5093 } 5094 5095 pp = &dl.d_partitions[DISKPART(my->subdev)]; 5096 if (pp->p_fstype != FS_RAID || DL_GETPSIZE(pp) == 0) 5097 return (ENOTSUP); 5098 5099 /* Find the blkno of the SR part in the underlying device */ 5100 sub_raidoff = my->srd->sd_meta->ssd_data_blkno + 5101 DL_SECTOBLK(&dl, DL_GETPOFFSET(pp)); 5102 DNPRINTF(SR_D_MISC,"sr_hibernate_io: blk trans ofs: %d blks\n", 5103 sub_raidoff); 5104 5105 /* Save the blkno of the swap partition in the SR disk */ 5106 my->sr_swapoff = blkno; 5107 5108 /* Initialize the sub-device */ 5109 return my->subfn(my->subdev, sub_raidoff + blkno, 5110 addr, size, op, page); 5111 } 5112 5113 /* Hibernate only uses (and we only support) writes */ 5114 if (op != HIB_W) 5115 return (ENOTSUP); 5116 5117 /* 5118 * Blocks act as the IV for the encryption. These block numbers 5119 * are relative to the start of the sr partition, but the 'blkno' 5120 * passed above is relative to the start of the swap partition 5121 * inside the sr partition, so bias appropriately. 5122 */ 5123 key_blkno = my->sr_swapoff + blkno; 5124 5125 /* Process each disk block one at a time. */ 5126 for (i = 0; i < size; i += DEV_BSIZE) { 5127 int res; 5128 5129 bzero(&ctx, sizeof(ctx)); 5130 5131 /* 5132 * Set encryption key (from the sr discipline stashed 5133 * during HIB_INIT. This code is based on the softraid 5134 * bootblock code. 5135 */ 5136 aes_xts_setkey(&ctx, my->srd->mds.mdd_crypto.scr_key[0], 64); 5137 /* We encrypt DEV_BSIZE bytes at a time in my->buf */ 5138 memcpy(my->buf, ((char *)addr) + i, DEV_BSIZE); 5139 5140 /* Block number is the IV */ 5141 memcpy(&iv, &key_blkno, sizeof(key_blkno)); 5142 aes_xts_reinit(&ctx, iv); 5143 5144 /* Encrypt DEV_BSIZE bytes, AES_XTS_BLOCKSIZE bytes at a time */ 5145 for (j = 0; j < DEV_BSIZE; j += AES_XTS_BLOCKSIZE) 5146 aes_xts_encrypt(&ctx, my->buf + j); 5147 5148 /* 5149 * Write one block out from my->buf to the underlying device 5150 * using its own side-effect free I/O function. 5151 */ 5152 res = my->subfn(my->subdev, blkno + (i / DEV_BSIZE), 5153 (vaddr_t)(my->buf), DEV_BSIZE, op, page); 5154 if (res != 0) 5155 return (res); 5156 key_blkno++; 5157 } 5158 return (0); 5159 } 5160 #endif /* HIBERNATE */ 5161