1 /* $OpenBSD: softraid.c,v 1.405 2020/06/27 17:28:58 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 void sr_copy_internal_data(struct scsi_xfer *, 98 void *, size_t); 99 int sr_scsi_ioctl(struct scsi_link *, u_long, 100 caddr_t, int); 101 int sr_bio_ioctl(struct device *, u_long, caddr_t); 102 int sr_bio_handler(struct sr_softc *, 103 struct sr_discipline *, u_long, struct bio *); 104 int sr_ioctl_inq(struct sr_softc *, struct bioc_inq *); 105 int sr_ioctl_vol(struct sr_softc *, struct bioc_vol *); 106 int sr_ioctl_disk(struct sr_softc *, struct bioc_disk *); 107 int sr_ioctl_setstate(struct sr_softc *, 108 struct bioc_setstate *); 109 int sr_ioctl_createraid(struct sr_softc *, 110 struct bioc_createraid *, int, void *); 111 int sr_ioctl_deleteraid(struct sr_softc *, 112 struct sr_discipline *, struct bioc_deleteraid *); 113 int sr_ioctl_discipline(struct sr_softc *, 114 struct sr_discipline *, struct bioc_discipline *); 115 int sr_ioctl_installboot(struct sr_softc *, 116 struct sr_discipline *, struct bioc_installboot *); 117 void sr_chunks_unwind(struct sr_softc *, 118 struct sr_chunk_head *); 119 void sr_discipline_free(struct sr_discipline *); 120 void sr_discipline_shutdown(struct sr_discipline *, int, int); 121 int sr_discipline_init(struct sr_discipline *, int); 122 int sr_alloc_resources(struct sr_discipline *); 123 void sr_free_resources(struct sr_discipline *); 124 void sr_set_chunk_state(struct sr_discipline *, int, int); 125 void sr_set_vol_state(struct sr_discipline *); 126 127 /* utility functions */ 128 void sr_shutdown(int); 129 void sr_uuid_generate(struct sr_uuid *); 130 char *sr_uuid_format(struct sr_uuid *); 131 void sr_uuid_print(struct sr_uuid *, int); 132 void sr_checksum_print(u_int8_t *); 133 int sr_boot_assembly(struct sr_softc *); 134 int sr_already_assembled(struct sr_discipline *); 135 int sr_hotspare(struct sr_softc *, dev_t); 136 void sr_hotspare_rebuild(struct sr_discipline *); 137 int sr_rebuild_init(struct sr_discipline *, dev_t, int); 138 void sr_rebuild_start(void *); 139 void sr_rebuild_thread(void *); 140 void sr_rebuild(struct sr_discipline *); 141 void sr_roam_chunks(struct sr_discipline *); 142 int sr_chunk_in_use(struct sr_softc *, dev_t); 143 int sr_rw(struct sr_softc *, dev_t, char *, size_t, 144 daddr_t, long); 145 void sr_wu_done_callback(void *); 146 147 /* don't include these on RAMDISK */ 148 #ifndef SMALL_KERNEL 149 void sr_sensors_refresh(void *); 150 int sr_sensors_create(struct sr_discipline *); 151 void sr_sensors_delete(struct sr_discipline *); 152 #endif 153 154 /* metadata */ 155 int sr_meta_probe(struct sr_discipline *, dev_t *, int); 156 int sr_meta_attach(struct sr_discipline *, int, int); 157 int sr_meta_rw(struct sr_discipline *, dev_t, void *, long); 158 int sr_meta_clear(struct sr_discipline *); 159 void sr_meta_init(struct sr_discipline *, int, int); 160 void sr_meta_init_complete(struct sr_discipline *); 161 void sr_meta_opt_handler(struct sr_discipline *, 162 struct sr_meta_opt_hdr *); 163 164 /* hotplug magic */ 165 void sr_disk_attach(struct disk *, int); 166 167 struct sr_hotplug_list { 168 void (*sh_hotplug)(struct sr_discipline *, 169 struct disk *, int); 170 struct sr_discipline *sh_sd; 171 172 SLIST_ENTRY(sr_hotplug_list) shl_link; 173 }; 174 SLIST_HEAD(sr_hotplug_list_head, sr_hotplug_list); 175 176 struct sr_hotplug_list_head sr_hotplug_callbacks; 177 extern void (*softraid_disk_attach)(struct disk *, int); 178 179 /* scsi glue */ 180 struct scsi_adapter sr_switch = { 181 sr_scsi_cmd, NULL, sr_scsi_probe, NULL, sr_scsi_ioctl 182 }; 183 184 /* native metadata format */ 185 int sr_meta_native_bootprobe(struct sr_softc *, dev_t, 186 struct sr_boot_chunk_head *); 187 #define SR_META_NOTCLAIMED (0) 188 #define SR_META_CLAIMED (1) 189 int sr_meta_native_probe(struct sr_softc *, 190 struct sr_chunk *); 191 int sr_meta_native_attach(struct sr_discipline *, int); 192 int sr_meta_native_write(struct sr_discipline *, dev_t, 193 struct sr_metadata *,void *); 194 195 #ifdef SR_DEBUG 196 void sr_meta_print(struct sr_metadata *); 197 #else 198 #define sr_meta_print(m) 199 #endif 200 201 /* the metadata driver should remain stateless */ 202 struct sr_meta_driver { 203 daddr_t smd_offset; /* metadata location */ 204 u_int32_t smd_size; /* size of metadata */ 205 206 int (*smd_probe)(struct sr_softc *, 207 struct sr_chunk *); 208 int (*smd_attach)(struct sr_discipline *, int); 209 int (*smd_detach)(struct sr_discipline *); 210 int (*smd_read)(struct sr_discipline *, dev_t, 211 struct sr_metadata *, void *); 212 int (*smd_write)(struct sr_discipline *, dev_t, 213 struct sr_metadata *, void *); 214 int (*smd_validate)(struct sr_discipline *, 215 struct sr_metadata *, void *); 216 } smd[] = { 217 { SR_META_OFFSET, SR_META_SIZE * DEV_BSIZE, 218 sr_meta_native_probe, sr_meta_native_attach, NULL, 219 sr_meta_native_read, sr_meta_native_write, NULL }, 220 { 0, 0, NULL, NULL, NULL, NULL } 221 }; 222 223 int 224 sr_meta_attach(struct sr_discipline *sd, int chunk_no, int force) 225 { 226 struct sr_softc *sc = sd->sd_sc; 227 struct sr_chunk_head *cl; 228 struct sr_chunk *ch_entry, *chunk1, *chunk2; 229 int rv = 1, i = 0; 230 231 DNPRINTF(SR_D_META, "%s: sr_meta_attach(%d)\n", DEVNAME(sc), chunk_no); 232 233 /* in memory copy of metadata */ 234 sd->sd_meta = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, 235 M_ZERO | M_NOWAIT); 236 if (!sd->sd_meta) { 237 sr_error(sc, "could not allocate memory for metadata"); 238 goto bad; 239 } 240 241 if (sd->sd_meta_type != SR_META_F_NATIVE) { 242 /* in memory copy of foreign metadata */ 243 sd->sd_meta_foreign = malloc(smd[sd->sd_meta_type].smd_size, 244 M_DEVBUF, M_ZERO | M_NOWAIT); 245 if (!sd->sd_meta_foreign) { 246 /* unwind frees sd_meta */ 247 sr_error(sc, "could not allocate memory for foreign " 248 "metadata"); 249 goto bad; 250 } 251 } 252 253 /* we have a valid list now create an array index */ 254 cl = &sd->sd_vol.sv_chunk_list; 255 sd->sd_vol.sv_chunks = mallocarray(chunk_no, sizeof(struct sr_chunk *), 256 M_DEVBUF, M_WAITOK | M_ZERO); 257 258 /* fill out chunk array */ 259 i = 0; 260 SLIST_FOREACH(ch_entry, cl, src_link) 261 sd->sd_vol.sv_chunks[i++] = ch_entry; 262 263 /* attach metadata */ 264 if (smd[sd->sd_meta_type].smd_attach(sd, force)) 265 goto bad; 266 267 /* Force chunks into correct order now that metadata is attached. */ 268 SLIST_INIT(cl); 269 for (i = 0; i < chunk_no; i++) { 270 ch_entry = sd->sd_vol.sv_chunks[i]; 271 chunk2 = NULL; 272 SLIST_FOREACH(chunk1, cl, src_link) { 273 if (chunk1->src_meta.scmi.scm_chunk_id > 274 ch_entry->src_meta.scmi.scm_chunk_id) 275 break; 276 chunk2 = chunk1; 277 } 278 if (chunk2 == NULL) 279 SLIST_INSERT_HEAD(cl, ch_entry, src_link); 280 else 281 SLIST_INSERT_AFTER(chunk2, ch_entry, src_link); 282 } 283 i = 0; 284 SLIST_FOREACH(ch_entry, cl, src_link) 285 sd->sd_vol.sv_chunks[i++] = ch_entry; 286 287 rv = 0; 288 bad: 289 return (rv); 290 } 291 292 int 293 sr_meta_probe(struct sr_discipline *sd, dev_t *dt, int no_chunk) 294 { 295 struct sr_softc *sc = sd->sd_sc; 296 struct vnode *vn; 297 struct sr_chunk *ch_entry, *ch_prev = NULL; 298 struct sr_chunk_head *cl; 299 char devname[32]; 300 int i, d, type, found, prevf, error; 301 dev_t dev; 302 303 DNPRINTF(SR_D_META, "%s: sr_meta_probe(%d)\n", DEVNAME(sc), no_chunk); 304 305 if (no_chunk == 0) 306 goto unwind; 307 308 cl = &sd->sd_vol.sv_chunk_list; 309 310 for (d = 0, prevf = SR_META_F_INVALID; d < no_chunk; d++) { 311 ch_entry = malloc(sizeof(struct sr_chunk), M_DEVBUF, 312 M_WAITOK | M_ZERO); 313 /* keep disks in user supplied order */ 314 if (ch_prev) 315 SLIST_INSERT_AFTER(ch_prev, ch_entry, src_link); 316 else 317 SLIST_INSERT_HEAD(cl, ch_entry, src_link); 318 ch_prev = ch_entry; 319 dev = dt[d]; 320 ch_entry->src_dev_mm = dev; 321 322 if (dev == NODEV) { 323 ch_entry->src_meta.scm_status = BIOC_SDOFFLINE; 324 continue; 325 } else { 326 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 327 if (bdevvp(dev, &vn)) { 328 sr_error(sc, "sr_meta_probe: cannot allocate " 329 "vnode"); 330 goto unwind; 331 } 332 333 /* 334 * XXX leaving dev open for now; move this to attach 335 * and figure out the open/close dance for unwind. 336 */ 337 error = VOP_OPEN(vn, FREAD | FWRITE, NOCRED, curproc); 338 if (error) { 339 DNPRINTF(SR_D_META,"%s: sr_meta_probe can't " 340 "open %s\n", DEVNAME(sc), devname); 341 vput(vn); 342 goto unwind; 343 } 344 345 strlcpy(ch_entry->src_devname, devname, 346 sizeof(ch_entry->src_devname)); 347 ch_entry->src_vn = vn; 348 } 349 350 /* determine if this is a device we understand */ 351 for (i = 0, found = SR_META_F_INVALID; smd[i].smd_probe; i++) { 352 type = smd[i].smd_probe(sc, ch_entry); 353 if (type == SR_META_F_INVALID) 354 continue; 355 else { 356 found = type; 357 break; 358 } 359 } 360 361 if (found == SR_META_F_INVALID) 362 goto unwind; 363 if (prevf == SR_META_F_INVALID) 364 prevf = found; 365 if (prevf != found) { 366 DNPRINTF(SR_D_META, "%s: prevf != found\n", 367 DEVNAME(sc)); 368 goto unwind; 369 } 370 } 371 372 return (prevf); 373 unwind: 374 return (SR_META_F_INVALID); 375 } 376 377 void 378 sr_meta_getdevname(struct sr_softc *sc, dev_t dev, char *buf, int size) 379 { 380 int maj, unit, part; 381 char *name; 382 383 DNPRINTF(SR_D_META, "%s: sr_meta_getdevname(%p, %d)\n", 384 DEVNAME(sc), buf, size); 385 386 if (!buf) 387 return; 388 389 maj = major(dev); 390 part = DISKPART(dev); 391 unit = DISKUNIT(dev); 392 393 name = findblkname(maj); 394 if (name == NULL) 395 return; 396 397 snprintf(buf, size, "%s%d%c", name, unit, part + 'a'); 398 } 399 400 int 401 sr_rw(struct sr_softc *sc, dev_t dev, char *buf, size_t size, daddr_t blkno, 402 long flags) 403 { 404 struct vnode *vp; 405 struct buf b; 406 size_t bufsize, dma_bufsize; 407 int rv = 1; 408 char *dma_buf; 409 410 DNPRINTF(SR_D_MISC, "%s: sr_rw(0x%x, %p, %zu, %lld 0x%lx)\n", 411 DEVNAME(sc), dev, buf, size, (long long)blkno, flags); 412 413 dma_bufsize = (size > MAXPHYS) ? MAXPHYS : size; 414 dma_buf = dma_alloc(dma_bufsize, PR_WAITOK); 415 416 if (bdevvp(dev, &vp)) { 417 printf("%s: sr_rw: failed to allocate vnode\n", DEVNAME(sc)); 418 goto done; 419 } 420 421 while (size > 0) { 422 DNPRINTF(SR_D_MISC, "%s: dma_buf %p, size %zu, blkno %lld)\n", 423 DEVNAME(sc), dma_buf, size, (long long)blkno); 424 425 bufsize = (size > MAXPHYS) ? MAXPHYS : size; 426 if (flags == B_WRITE) 427 memcpy(dma_buf, buf, bufsize); 428 429 bzero(&b, sizeof(b)); 430 b.b_flags = flags | B_PHYS; 431 b.b_proc = curproc; 432 b.b_dev = dev; 433 b.b_iodone = NULL; 434 b.b_error = 0; 435 b.b_blkno = blkno; 436 b.b_data = dma_buf; 437 b.b_bcount = bufsize; 438 b.b_bufsize = bufsize; 439 b.b_resid = bufsize; 440 b.b_vp = vp; 441 442 if ((b.b_flags & B_READ) == 0) 443 vp->v_numoutput++; 444 445 LIST_INIT(&b.b_dep); 446 VOP_STRATEGY(&b); 447 biowait(&b); 448 449 if (b.b_flags & B_ERROR) { 450 printf("%s: I/O error %d on dev 0x%x at block %llu\n", 451 DEVNAME(sc), b.b_error, dev, b.b_blkno); 452 goto done; 453 } 454 455 if (flags == B_READ) 456 memcpy(buf, dma_buf, bufsize); 457 458 size -= bufsize; 459 buf += bufsize; 460 blkno += howmany(bufsize, DEV_BSIZE); 461 } 462 463 rv = 0; 464 465 done: 466 if (vp) 467 vput(vp); 468 469 dma_free(dma_buf, dma_bufsize); 470 471 return (rv); 472 } 473 474 int 475 sr_meta_rw(struct sr_discipline *sd, dev_t dev, void *md, long flags) 476 { 477 int rv = 1; 478 479 DNPRINTF(SR_D_META, "%s: sr_meta_rw(0x%x, %p, 0x%lx)\n", 480 DEVNAME(sd->sd_sc), dev, md, flags); 481 482 if (md == NULL) { 483 printf("%s: sr_meta_rw: invalid metadata pointer\n", 484 DEVNAME(sd->sd_sc)); 485 goto done; 486 } 487 488 rv = sr_rw(sd->sd_sc, dev, md, SR_META_SIZE * DEV_BSIZE, 489 SR_META_OFFSET, flags); 490 491 done: 492 return (rv); 493 } 494 495 int 496 sr_meta_clear(struct sr_discipline *sd) 497 { 498 struct sr_softc *sc = sd->sd_sc; 499 struct sr_chunk_head *cl = &sd->sd_vol.sv_chunk_list; 500 struct sr_chunk *ch_entry; 501 void *m; 502 int rv = 1; 503 504 DNPRINTF(SR_D_META, "%s: sr_meta_clear\n", DEVNAME(sc)); 505 506 if (sd->sd_meta_type != SR_META_F_NATIVE) { 507 sr_error(sc, "cannot clear foreign metadata"); 508 goto done; 509 } 510 511 m = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, M_WAITOK | M_ZERO); 512 SLIST_FOREACH(ch_entry, cl, src_link) { 513 if (sr_meta_native_write(sd, ch_entry->src_dev_mm, m, NULL)) { 514 /* XXX mark disk offline */ 515 DNPRINTF(SR_D_META, "%s: sr_meta_clear failed to " 516 "clear %s\n", DEVNAME(sc), ch_entry->src_devname); 517 rv++; 518 continue; 519 } 520 bzero(&ch_entry->src_meta, sizeof(ch_entry->src_meta)); 521 } 522 523 bzero(sd->sd_meta, SR_META_SIZE * DEV_BSIZE); 524 525 free(m, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 526 rv = 0; 527 done: 528 return (rv); 529 } 530 531 void 532 sr_meta_init(struct sr_discipline *sd, int level, int no_chunk) 533 { 534 struct sr_softc *sc = sd->sd_sc; 535 struct sr_metadata *sm = sd->sd_meta; 536 struct sr_chunk_head *cl = &sd->sd_vol.sv_chunk_list; 537 struct sr_meta_chunk *scm; 538 struct sr_chunk *chunk; 539 int cid = 0; 540 u_int64_t max_chunk_sz = 0, min_chunk_sz = 0; 541 u_int32_t secsize = DEV_BSIZE; 542 543 DNPRINTF(SR_D_META, "%s: sr_meta_init\n", DEVNAME(sc)); 544 545 if (!sm) 546 return; 547 548 /* Initialise volume metadata. */ 549 sm->ssdi.ssd_magic = SR_MAGIC; 550 sm->ssdi.ssd_version = SR_META_VERSION; 551 sm->ssdi.ssd_vol_flags = sd->sd_meta_flags; 552 sm->ssdi.ssd_volid = 0; 553 sm->ssdi.ssd_chunk_no = no_chunk; 554 sm->ssdi.ssd_level = level; 555 556 sm->ssd_data_blkno = SR_DATA_OFFSET; 557 sm->ssd_ondisk = 0; 558 559 sr_uuid_generate(&sm->ssdi.ssd_uuid); 560 561 /* Initialise chunk metadata and get min/max chunk sizes & secsize. */ 562 SLIST_FOREACH(chunk, cl, src_link) { 563 scm = &chunk->src_meta; 564 scm->scmi.scm_size = chunk->src_size; 565 scm->scmi.scm_chunk_id = cid++; 566 scm->scm_status = BIOC_SDONLINE; 567 scm->scmi.scm_volid = 0; 568 strlcpy(scm->scmi.scm_devname, chunk->src_devname, 569 sizeof(scm->scmi.scm_devname)); 570 memcpy(&scm->scmi.scm_uuid, &sm->ssdi.ssd_uuid, 571 sizeof(scm->scmi.scm_uuid)); 572 sr_checksum(sc, scm, &scm->scm_checksum, 573 sizeof(scm->scm_checksum)); 574 575 if (min_chunk_sz == 0) 576 min_chunk_sz = scm->scmi.scm_size; 577 if (chunk->src_secsize > secsize) 578 secsize = chunk->src_secsize; 579 min_chunk_sz = MIN(min_chunk_sz, scm->scmi.scm_size); 580 max_chunk_sz = MAX(max_chunk_sz, scm->scmi.scm_size); 581 } 582 583 sm->ssdi.ssd_secsize = secsize; 584 585 /* Equalize chunk sizes. */ 586 SLIST_FOREACH(chunk, cl, src_link) 587 chunk->src_meta.scmi.scm_coerced_size = min_chunk_sz; 588 589 sd->sd_vol.sv_chunk_minsz = min_chunk_sz; 590 sd->sd_vol.sv_chunk_maxsz = max_chunk_sz; 591 } 592 593 void 594 sr_meta_init_complete(struct sr_discipline *sd) 595 { 596 #ifdef SR_DEBUG 597 struct sr_softc *sc = sd->sd_sc; 598 #endif 599 struct sr_metadata *sm = sd->sd_meta; 600 601 DNPRINTF(SR_D_META, "%s: sr_meta_complete\n", DEVNAME(sc)); 602 603 /* Complete initialisation of volume metadata. */ 604 strlcpy(sm->ssdi.ssd_vendor, "OPENBSD", sizeof(sm->ssdi.ssd_vendor)); 605 snprintf(sm->ssdi.ssd_product, sizeof(sm->ssdi.ssd_product), 606 "SR %s", sd->sd_name); 607 snprintf(sm->ssdi.ssd_revision, sizeof(sm->ssdi.ssd_revision), 608 "%03d", sm->ssdi.ssd_version); 609 } 610 611 void 612 sr_meta_opt_handler(struct sr_discipline *sd, struct sr_meta_opt_hdr *om) 613 { 614 if (om->som_type != SR_OPT_BOOT) 615 panic("unknown optional metadata type"); 616 } 617 618 void 619 sr_meta_save_callback(void *xsd) 620 { 621 struct sr_discipline *sd = xsd; 622 int s; 623 624 s = splbio(); 625 626 if (sr_meta_save(sd, SR_META_DIRTY)) 627 printf("%s: save metadata failed\n", DEVNAME(sd->sd_sc)); 628 629 sd->sd_must_flush = 0; 630 splx(s); 631 } 632 633 int 634 sr_meta_save(struct sr_discipline *sd, u_int32_t flags) 635 { 636 struct sr_softc *sc = sd->sd_sc; 637 struct sr_metadata *sm = sd->sd_meta, *m; 638 struct sr_meta_driver *s; 639 struct sr_chunk *src; 640 struct sr_meta_chunk *cm; 641 struct sr_workunit wu; 642 struct sr_meta_opt_hdr *omh; 643 struct sr_meta_opt_item *omi; 644 int i; 645 646 DNPRINTF(SR_D_META, "%s: sr_meta_save %s\n", 647 DEVNAME(sc), sd->sd_meta->ssd_devname); 648 649 if (!sm) { 650 printf("%s: no in memory copy of metadata\n", DEVNAME(sc)); 651 goto bad; 652 } 653 654 /* meta scratchpad */ 655 s = &smd[sd->sd_meta_type]; 656 m = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, M_ZERO | M_NOWAIT); 657 if (!m) { 658 printf("%s: could not allocate metadata scratch area\n", 659 DEVNAME(sc)); 660 goto bad; 661 } 662 663 /* from here on out metadata is updated */ 664 restart: 665 sm->ssd_ondisk++; 666 sm->ssd_meta_flags = flags; 667 memcpy(m, sm, sizeof(*m)); 668 669 /* Chunk metadata. */ 670 cm = (struct sr_meta_chunk *)(m + 1); 671 for (i = 0; i < sm->ssdi.ssd_chunk_no; i++) { 672 src = sd->sd_vol.sv_chunks[i]; 673 memcpy(cm, &src->src_meta, sizeof(*cm)); 674 cm++; 675 } 676 677 /* Optional metadata. */ 678 omh = (struct sr_meta_opt_hdr *)(cm); 679 SLIST_FOREACH(omi, &sd->sd_meta_opt, omi_link) { 680 DNPRINTF(SR_D_META, "%s: saving optional metadata type %u with " 681 "length %u\n", DEVNAME(sc), omi->omi_som->som_type, 682 omi->omi_som->som_length); 683 bzero(&omi->omi_som->som_checksum, MD5_DIGEST_LENGTH); 684 sr_checksum(sc, omi->omi_som, &omi->omi_som->som_checksum, 685 omi->omi_som->som_length); 686 memcpy(omh, omi->omi_som, omi->omi_som->som_length); 687 omh = (struct sr_meta_opt_hdr *)((u_int8_t *)omh + 688 omi->omi_som->som_length); 689 } 690 691 for (i = 0; i < sm->ssdi.ssd_chunk_no; i++) { 692 src = sd->sd_vol.sv_chunks[i]; 693 694 /* skip disks that are offline */ 695 if (src->src_meta.scm_status == BIOC_SDOFFLINE) 696 continue; 697 698 /* calculate metadata checksum for correct chunk */ 699 m->ssdi.ssd_chunk_id = i; 700 sr_checksum(sc, m, &m->ssd_checksum, 701 sizeof(struct sr_meta_invariant)); 702 703 #ifdef SR_DEBUG 704 DNPRINTF(SR_D_META, "%s: sr_meta_save %s: volid: %d " 705 "chunkid: %d checksum: ", 706 DEVNAME(sc), src->src_meta.scmi.scm_devname, 707 m->ssdi.ssd_volid, m->ssdi.ssd_chunk_id); 708 709 if (sr_debug & SR_D_META) 710 sr_checksum_print((u_int8_t *)&m->ssd_checksum); 711 DNPRINTF(SR_D_META, "\n"); 712 sr_meta_print(m); 713 #endif 714 715 /* translate and write to disk */ 716 if (s->smd_write(sd, src->src_dev_mm, m, NULL /* XXX */)) { 717 printf("%s: could not write metadata to %s\n", 718 DEVNAME(sc), src->src_devname); 719 /* restart the meta write */ 720 src->src_meta.scm_status = BIOC_SDOFFLINE; 721 /* XXX recalculate volume status */ 722 goto restart; 723 } 724 } 725 726 /* not all disciplines have sync */ 727 if (sd->sd_scsi_sync) { 728 bzero(&wu, sizeof(wu)); 729 wu.swu_flags |= SR_WUF_FAKE; 730 wu.swu_dis = sd; 731 sd->sd_scsi_sync(&wu); 732 } 733 free(m, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 734 return (0); 735 bad: 736 return (1); 737 } 738 739 int 740 sr_meta_read(struct sr_discipline *sd) 741 { 742 struct sr_softc *sc = sd->sd_sc; 743 struct sr_chunk_head *cl = &sd->sd_vol.sv_chunk_list; 744 struct sr_metadata *sm; 745 struct sr_chunk *ch_entry; 746 struct sr_meta_chunk *cp; 747 struct sr_meta_driver *s; 748 void *fm = NULL; 749 int no_disk = 0, got_meta = 0; 750 751 DNPRINTF(SR_D_META, "%s: sr_meta_read\n", DEVNAME(sc)); 752 753 sm = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, M_WAITOK | M_ZERO); 754 s = &smd[sd->sd_meta_type]; 755 if (sd->sd_meta_type != SR_META_F_NATIVE) 756 fm = malloc(s->smd_size, M_DEVBUF, M_WAITOK | M_ZERO); 757 758 cp = (struct sr_meta_chunk *)(sm + 1); 759 SLIST_FOREACH(ch_entry, cl, src_link) { 760 /* skip disks that are offline */ 761 if (ch_entry->src_meta.scm_status == BIOC_SDOFFLINE) { 762 DNPRINTF(SR_D_META, 763 "%s: %s chunk marked offline, spoofing status\n", 764 DEVNAME(sc), ch_entry->src_devname); 765 cp++; /* adjust chunk pointer to match failure */ 766 continue; 767 } else if (s->smd_read(sd, ch_entry->src_dev_mm, sm, fm)) { 768 /* read and translate */ 769 /* XXX mark chunk offline, elsewhere!! */ 770 ch_entry->src_meta.scm_status = BIOC_SDOFFLINE; 771 cp++; /* adjust chunk pointer to match failure */ 772 DNPRINTF(SR_D_META, "%s: sr_meta_read failed\n", 773 DEVNAME(sc)); 774 continue; 775 } 776 777 if (sm->ssdi.ssd_magic != SR_MAGIC) { 778 DNPRINTF(SR_D_META, "%s: sr_meta_read !SR_MAGIC\n", 779 DEVNAME(sc)); 780 continue; 781 } 782 783 /* validate metadata */ 784 if (sr_meta_validate(sd, ch_entry->src_dev_mm, sm, fm)) { 785 DNPRINTF(SR_D_META, "%s: invalid metadata\n", 786 DEVNAME(sc)); 787 no_disk = -1; 788 goto done; 789 } 790 791 /* assume first chunk contains metadata */ 792 if (got_meta == 0) { 793 sr_meta_opt_load(sc, sm, &sd->sd_meta_opt); 794 memcpy(sd->sd_meta, sm, sizeof(*sd->sd_meta)); 795 got_meta = 1; 796 } 797 798 memcpy(&ch_entry->src_meta, cp, sizeof(ch_entry->src_meta)); 799 800 no_disk++; 801 cp++; 802 } 803 804 free(sm, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 805 free(fm, M_DEVBUF, s->smd_size); 806 807 done: 808 DNPRINTF(SR_D_META, "%s: sr_meta_read found %d parts\n", DEVNAME(sc), 809 no_disk); 810 return (no_disk); 811 } 812 813 void 814 sr_meta_opt_load(struct sr_softc *sc, struct sr_metadata *sm, 815 struct sr_meta_opt_head *som) 816 { 817 struct sr_meta_opt_hdr *omh; 818 struct sr_meta_opt_item *omi; 819 u_int8_t checksum[MD5_DIGEST_LENGTH]; 820 int i; 821 822 /* Process optional metadata. */ 823 omh = (struct sr_meta_opt_hdr *)((u_int8_t *)(sm + 1) + 824 sizeof(struct sr_meta_chunk) * sm->ssdi.ssd_chunk_no); 825 for (i = 0; i < sm->ssdi.ssd_opt_no; i++) { 826 827 omi = malloc(sizeof(struct sr_meta_opt_item), M_DEVBUF, 828 M_WAITOK | M_ZERO); 829 SLIST_INSERT_HEAD(som, omi, omi_link); 830 831 if (omh->som_length == 0) { 832 833 /* Load old fixed length optional metadata. */ 834 DNPRINTF(SR_D_META, "%s: old optional metadata of type " 835 "%u\n", DEVNAME(sc), omh->som_type); 836 837 /* Validate checksum. */ 838 sr_checksum(sc, (void *)omh, &checksum, 839 SR_OLD_META_OPT_SIZE - MD5_DIGEST_LENGTH); 840 if (bcmp(&checksum, (void *)omh + SR_OLD_META_OPT_MD5, 841 sizeof(checksum))) 842 panic("%s: invalid optional metadata " 843 "checksum", DEVNAME(sc)); 844 845 /* Determine correct length. */ 846 switch (omh->som_type) { 847 case SR_OPT_CRYPTO: 848 omh->som_length = sizeof(struct sr_meta_crypto); 849 break; 850 case SR_OPT_BOOT: 851 omh->som_length = sizeof(struct sr_meta_boot); 852 break; 853 case SR_OPT_KEYDISK: 854 omh->som_length = 855 sizeof(struct sr_meta_keydisk); 856 break; 857 default: 858 panic("unknown old optional metadata " 859 "type %u\n", omh->som_type); 860 } 861 862 omi->omi_som = malloc(omh->som_length, M_DEVBUF, 863 M_WAITOK | M_ZERO); 864 memcpy((u_int8_t *)omi->omi_som + sizeof(*omi->omi_som), 865 (u_int8_t *)omh + SR_OLD_META_OPT_OFFSET, 866 omh->som_length - sizeof(*omi->omi_som)); 867 omi->omi_som->som_type = omh->som_type; 868 omi->omi_som->som_length = omh->som_length; 869 870 omh = (struct sr_meta_opt_hdr *)((void *)omh + 871 SR_OLD_META_OPT_SIZE); 872 } else { 873 874 /* Load variable length optional metadata. */ 875 DNPRINTF(SR_D_META, "%s: optional metadata of type %u, " 876 "length %u\n", DEVNAME(sc), omh->som_type, 877 omh->som_length); 878 omi->omi_som = malloc(omh->som_length, M_DEVBUF, 879 M_WAITOK | M_ZERO); 880 memcpy(omi->omi_som, omh, omh->som_length); 881 882 /* Validate checksum. */ 883 memcpy(&checksum, &omi->omi_som->som_checksum, 884 MD5_DIGEST_LENGTH); 885 bzero(&omi->omi_som->som_checksum, MD5_DIGEST_LENGTH); 886 sr_checksum(sc, omi->omi_som, 887 &omi->omi_som->som_checksum, omh->som_length); 888 if (bcmp(&checksum, &omi->omi_som->som_checksum, 889 sizeof(checksum))) 890 panic("%s: invalid optional metadata checksum", 891 DEVNAME(sc)); 892 893 omh = (struct sr_meta_opt_hdr *)((void *)omh + 894 omh->som_length); 895 } 896 } 897 } 898 899 int 900 sr_meta_validate(struct sr_discipline *sd, dev_t dev, struct sr_metadata *sm, 901 void *fm) 902 { 903 struct sr_softc *sc = sd->sd_sc; 904 struct sr_meta_driver *s; 905 #ifdef SR_DEBUG 906 struct sr_meta_chunk *mc; 907 #endif 908 u_int8_t checksum[MD5_DIGEST_LENGTH]; 909 char devname[32]; 910 int rv = 1; 911 912 DNPRINTF(SR_D_META, "%s: sr_meta_validate(%p)\n", DEVNAME(sc), sm); 913 914 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 915 916 s = &smd[sd->sd_meta_type]; 917 if (sd->sd_meta_type != SR_META_F_NATIVE) 918 if (s->smd_validate(sd, sm, fm)) { 919 sr_error(sc, "invalid foreign metadata"); 920 goto done; 921 } 922 923 /* 924 * at this point all foreign metadata has been translated to the native 925 * format and will be treated just like the native format 926 */ 927 928 if (sm->ssdi.ssd_magic != SR_MAGIC) { 929 sr_error(sc, "not valid softraid metadata"); 930 goto done; 931 } 932 933 /* Verify metadata checksum. */ 934 sr_checksum(sc, sm, &checksum, sizeof(struct sr_meta_invariant)); 935 if (bcmp(&checksum, &sm->ssd_checksum, sizeof(checksum))) { 936 sr_error(sc, "invalid metadata checksum"); 937 goto done; 938 } 939 940 /* Handle changes between versions. */ 941 if (sm->ssdi.ssd_version == 3) { 942 943 /* 944 * Version 3 - update metadata version and fix up data blkno 945 * value since this did not exist in version 3. 946 */ 947 if (sm->ssd_data_blkno == 0) 948 sm->ssd_data_blkno = SR_META_V3_DATA_OFFSET; 949 sm->ssdi.ssd_secsize = DEV_BSIZE; 950 951 } else if (sm->ssdi.ssd_version == 4) { 952 953 /* 954 * Version 4 - original metadata format did not store 955 * data blkno so fix this up if necessary. 956 */ 957 if (sm->ssd_data_blkno == 0) 958 sm->ssd_data_blkno = SR_DATA_OFFSET; 959 sm->ssdi.ssd_secsize = DEV_BSIZE; 960 961 } else if (sm->ssdi.ssd_version == 5) { 962 963 /* 964 * Version 5 - variable length optional metadata. Migration 965 * from earlier fixed length optional metadata is handled 966 * in sr_meta_read(). 967 */ 968 sm->ssdi.ssd_secsize = DEV_BSIZE; 969 970 } else if (sm->ssdi.ssd_version == SR_META_VERSION) { 971 972 /* 973 * Version 6 - store & report a sector size. 974 */ 975 976 } else { 977 978 sr_error(sc, "cannot read metadata version %u on %s, " 979 "expected version %u or earlier", 980 sm->ssdi.ssd_version, devname, SR_META_VERSION); 981 goto done; 982 983 } 984 985 /* Update version number and revision string. */ 986 sm->ssdi.ssd_version = SR_META_VERSION; 987 snprintf(sm->ssdi.ssd_revision, sizeof(sm->ssdi.ssd_revision), 988 "%03d", SR_META_VERSION); 989 990 #ifdef SR_DEBUG 991 /* warn if disk changed order */ 992 mc = (struct sr_meta_chunk *)(sm + 1); 993 if (strncmp(mc[sm->ssdi.ssd_chunk_id].scmi.scm_devname, devname, 994 sizeof(mc[sm->ssdi.ssd_chunk_id].scmi.scm_devname))) 995 DNPRINTF(SR_D_META, "%s: roaming device %s -> %s\n", 996 DEVNAME(sc), mc[sm->ssdi.ssd_chunk_id].scmi.scm_devname, 997 devname); 998 #endif 999 1000 /* we have meta data on disk */ 1001 DNPRINTF(SR_D_META, "%s: sr_meta_validate valid metadata %s\n", 1002 DEVNAME(sc), devname); 1003 1004 rv = 0; 1005 done: 1006 return (rv); 1007 } 1008 1009 int 1010 sr_meta_native_bootprobe(struct sr_softc *sc, dev_t devno, 1011 struct sr_boot_chunk_head *bch) 1012 { 1013 struct vnode *vn; 1014 struct disklabel label; 1015 struct sr_metadata *md = NULL; 1016 struct sr_discipline *fake_sd = NULL; 1017 struct sr_boot_chunk *bc; 1018 char devname[32]; 1019 dev_t chrdev, rawdev; 1020 int error, i; 1021 int rv = SR_META_NOTCLAIMED; 1022 1023 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe\n", DEVNAME(sc)); 1024 1025 /* 1026 * Use character raw device to avoid SCSI complaints about missing 1027 * media on removable media devices. 1028 */ 1029 chrdev = blktochr(devno); 1030 rawdev = MAKEDISKDEV(major(chrdev), DISKUNIT(devno), RAW_PART); 1031 if (cdevvp(rawdev, &vn)) { 1032 sr_error(sc, "sr_meta_native_bootprobe: cannot allocate vnode"); 1033 goto done; 1034 } 1035 1036 /* open device */ 1037 error = VOP_OPEN(vn, FREAD, NOCRED, curproc); 1038 if (error) { 1039 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe open " 1040 "failed\n", DEVNAME(sc)); 1041 vput(vn); 1042 goto done; 1043 } 1044 1045 /* get disklabel */ 1046 error = VOP_IOCTL(vn, DIOCGDINFO, (caddr_t)&label, FREAD, NOCRED, 1047 curproc); 1048 if (error) { 1049 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe ioctl " 1050 "failed\n", DEVNAME(sc)); 1051 VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1052 vput(vn); 1053 goto done; 1054 } 1055 1056 /* we are done, close device */ 1057 error = VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1058 if (error) { 1059 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe close " 1060 "failed\n", DEVNAME(sc)); 1061 vput(vn); 1062 goto done; 1063 } 1064 vput(vn); 1065 1066 md = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, M_ZERO | M_NOWAIT); 1067 if (md == NULL) { 1068 sr_error(sc, "not enough memory for metadata buffer"); 1069 goto done; 1070 } 1071 1072 /* create fake sd to use utility functions */ 1073 fake_sd = malloc(sizeof(struct sr_discipline), M_DEVBUF, 1074 M_ZERO | M_NOWAIT); 1075 if (fake_sd == NULL) { 1076 sr_error(sc, "not enough memory for fake discipline"); 1077 goto done; 1078 } 1079 fake_sd->sd_sc = sc; 1080 fake_sd->sd_meta_type = SR_META_F_NATIVE; 1081 1082 for (i = 0; i < MAXPARTITIONS; i++) { 1083 if (label.d_partitions[i].p_fstype != FS_RAID) 1084 continue; 1085 1086 /* open partition */ 1087 rawdev = MAKEDISKDEV(major(devno), DISKUNIT(devno), i); 1088 if (bdevvp(rawdev, &vn)) { 1089 sr_error(sc, "sr_meta_native_bootprobe: cannot " 1090 "allocate vnode for partition"); 1091 goto done; 1092 } 1093 error = VOP_OPEN(vn, FREAD, NOCRED, curproc); 1094 if (error) { 1095 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe " 1096 "open failed, partition %d\n", 1097 DEVNAME(sc), i); 1098 vput(vn); 1099 continue; 1100 } 1101 1102 if (sr_meta_native_read(fake_sd, rawdev, md, NULL)) { 1103 sr_error(sc, "native bootprobe could not read native " 1104 "metadata"); 1105 VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1106 vput(vn); 1107 continue; 1108 } 1109 1110 /* are we a softraid partition? */ 1111 if (md->ssdi.ssd_magic != SR_MAGIC) { 1112 VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1113 vput(vn); 1114 continue; 1115 } 1116 1117 sr_meta_getdevname(sc, rawdev, devname, sizeof(devname)); 1118 if (sr_meta_validate(fake_sd, rawdev, md, NULL) == 0) { 1119 /* XXX fix M_WAITOK, this is boot time */ 1120 bc = malloc(sizeof(struct sr_boot_chunk), 1121 M_DEVBUF, M_WAITOK | M_ZERO); 1122 bc->sbc_metadata = malloc(sizeof(struct sr_metadata), 1123 M_DEVBUF, M_WAITOK | M_ZERO); 1124 memcpy(bc->sbc_metadata, md, sizeof(struct sr_metadata)); 1125 bc->sbc_mm = rawdev; 1126 SLIST_INSERT_HEAD(bch, bc, sbc_link); 1127 rv = SR_META_CLAIMED; 1128 } 1129 1130 /* we are done, close partition */ 1131 VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1132 vput(vn); 1133 } 1134 1135 done: 1136 free(fake_sd, M_DEVBUF, sizeof(struct sr_discipline)); 1137 free(md, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 1138 1139 return (rv); 1140 } 1141 1142 int 1143 sr_boot_assembly(struct sr_softc *sc) 1144 { 1145 struct sr_boot_volume_head bvh; 1146 struct sr_boot_chunk_head bch, kdh; 1147 struct sr_boot_volume *bv, *bv1, *bv2; 1148 struct sr_boot_chunk *bc, *bcnext, *bc1, *bc2; 1149 struct sr_disk_head sdklist; 1150 struct sr_disk *sdk; 1151 struct disk *dk; 1152 struct bioc_createraid bcr; 1153 struct sr_meta_chunk *hm; 1154 struct sr_chunk_head *cl; 1155 struct sr_chunk *hotspare, *chunk, *last; 1156 u_int64_t *ondisk = NULL; 1157 dev_t *devs = NULL; 1158 void *data; 1159 char devname[32]; 1160 int rv = 0, i; 1161 1162 DNPRINTF(SR_D_META, "%s: sr_boot_assembly\n", DEVNAME(sc)); 1163 1164 SLIST_INIT(&sdklist); 1165 SLIST_INIT(&bvh); 1166 SLIST_INIT(&bch); 1167 SLIST_INIT(&kdh); 1168 1169 dk = TAILQ_FIRST(&disklist); 1170 while (dk != NULL) { 1171 1172 /* See if this disk has been checked. */ 1173 SLIST_FOREACH(sdk, &sdklist, sdk_link) 1174 if (sdk->sdk_devno == dk->dk_devno) 1175 break; 1176 1177 if (sdk != NULL || dk->dk_devno == NODEV) { 1178 dk = TAILQ_NEXT(dk, dk_link); 1179 continue; 1180 } 1181 1182 /* Add this disk to the list that we've checked. */ 1183 sdk = malloc(sizeof(struct sr_disk), M_DEVBUF, 1184 M_NOWAIT | M_ZERO); 1185 if (sdk == NULL) 1186 goto unwind; 1187 sdk->sdk_devno = dk->dk_devno; 1188 SLIST_INSERT_HEAD(&sdklist, sdk, sdk_link); 1189 1190 /* Only check sd(4) and wd(4) devices. */ 1191 if (strncmp(dk->dk_name, "sd", 2) && 1192 strncmp(dk->dk_name, "wd", 2)) { 1193 dk = TAILQ_NEXT(dk, dk_link); 1194 continue; 1195 } 1196 1197 /* native softraid uses partitions */ 1198 rw_enter_write(&sc->sc_lock); 1199 bio_status_init(&sc->sc_status, &sc->sc_dev); 1200 sr_meta_native_bootprobe(sc, dk->dk_devno, &bch); 1201 rw_exit_write(&sc->sc_lock); 1202 1203 /* probe non-native disks if native failed. */ 1204 1205 /* Restart scan since we may have slept. */ 1206 dk = TAILQ_FIRST(&disklist); 1207 } 1208 1209 /* 1210 * Create a list of volumes and associate chunks with each volume. 1211 */ 1212 for (bc = SLIST_FIRST(&bch); bc != NULL; bc = bcnext) { 1213 1214 bcnext = SLIST_NEXT(bc, sbc_link); 1215 SLIST_REMOVE(&bch, bc, sr_boot_chunk, sbc_link); 1216 bc->sbc_chunk_id = bc->sbc_metadata->ssdi.ssd_chunk_id; 1217 1218 /* Handle key disks separately. */ 1219 if (bc->sbc_metadata->ssdi.ssd_level == SR_KEYDISK_LEVEL) { 1220 SLIST_INSERT_HEAD(&kdh, bc, sbc_link); 1221 continue; 1222 } 1223 1224 SLIST_FOREACH(bv, &bvh, sbv_link) { 1225 if (bcmp(&bc->sbc_metadata->ssdi.ssd_uuid, 1226 &bv->sbv_uuid, 1227 sizeof(bc->sbc_metadata->ssdi.ssd_uuid)) == 0) 1228 break; 1229 } 1230 1231 if (bv == NULL) { 1232 bv = malloc(sizeof(struct sr_boot_volume), 1233 M_DEVBUF, M_NOWAIT | M_ZERO); 1234 if (bv == NULL) { 1235 printf("%s: failed to allocate boot volume\n", 1236 DEVNAME(sc)); 1237 goto unwind; 1238 } 1239 1240 bv->sbv_level = bc->sbc_metadata->ssdi.ssd_level; 1241 bv->sbv_volid = bc->sbc_metadata->ssdi.ssd_volid; 1242 bv->sbv_chunk_no = bc->sbc_metadata->ssdi.ssd_chunk_no; 1243 bv->sbv_flags = bc->sbc_metadata->ssdi.ssd_vol_flags; 1244 memcpy(&bv->sbv_uuid, &bc->sbc_metadata->ssdi.ssd_uuid, 1245 sizeof(bc->sbc_metadata->ssdi.ssd_uuid)); 1246 SLIST_INIT(&bv->sbv_chunks); 1247 1248 /* Maintain volume order. */ 1249 bv2 = NULL; 1250 SLIST_FOREACH(bv1, &bvh, sbv_link) { 1251 if (bv1->sbv_volid > bv->sbv_volid) 1252 break; 1253 bv2 = bv1; 1254 } 1255 if (bv2 == NULL) { 1256 DNPRINTF(SR_D_META, "%s: insert volume %u " 1257 "at head\n", DEVNAME(sc), bv->sbv_volid); 1258 SLIST_INSERT_HEAD(&bvh, bv, sbv_link); 1259 } else { 1260 DNPRINTF(SR_D_META, "%s: insert volume %u " 1261 "after %u\n", DEVNAME(sc), bv->sbv_volid, 1262 bv2->sbv_volid); 1263 SLIST_INSERT_AFTER(bv2, bv, sbv_link); 1264 } 1265 } 1266 1267 /* Maintain chunk order. */ 1268 bc2 = NULL; 1269 SLIST_FOREACH(bc1, &bv->sbv_chunks, sbc_link) { 1270 if (bc1->sbc_chunk_id > bc->sbc_chunk_id) 1271 break; 1272 bc2 = bc1; 1273 } 1274 if (bc2 == NULL) { 1275 DNPRINTF(SR_D_META, "%s: volume %u insert chunk %u " 1276 "at head\n", DEVNAME(sc), bv->sbv_volid, 1277 bc->sbc_chunk_id); 1278 SLIST_INSERT_HEAD(&bv->sbv_chunks, bc, sbc_link); 1279 } else { 1280 DNPRINTF(SR_D_META, "%s: volume %u insert chunk %u " 1281 "after %u\n", DEVNAME(sc), bv->sbv_volid, 1282 bc->sbc_chunk_id, bc2->sbc_chunk_id); 1283 SLIST_INSERT_AFTER(bc2, bc, sbc_link); 1284 } 1285 1286 bv->sbv_chunks_found++; 1287 } 1288 1289 /* Allocate memory for device and ondisk version arrays. */ 1290 devs = mallocarray(BIOC_CRMAXLEN, sizeof(dev_t), M_DEVBUF, 1291 M_NOWAIT); 1292 if (devs == NULL) { 1293 printf("%s: failed to allocate device array\n", DEVNAME(sc)); 1294 goto unwind; 1295 } 1296 ondisk = mallocarray(BIOC_CRMAXLEN, sizeof(u_int64_t), M_DEVBUF, 1297 M_NOWAIT); 1298 if (ondisk == NULL) { 1299 printf("%s: failed to allocate ondisk array\n", DEVNAME(sc)); 1300 goto unwind; 1301 } 1302 1303 /* 1304 * Assemble hotspare "volumes". 1305 */ 1306 SLIST_FOREACH(bv, &bvh, sbv_link) { 1307 1308 /* Check if this is a hotspare "volume". */ 1309 if (bv->sbv_level != SR_HOTSPARE_LEVEL || 1310 bv->sbv_chunk_no != 1) 1311 continue; 1312 1313 #ifdef SR_DEBUG 1314 DNPRINTF(SR_D_META, "%s: assembling hotspare volume ", 1315 DEVNAME(sc)); 1316 if (sr_debug & SR_D_META) 1317 sr_uuid_print(&bv->sbv_uuid, 0); 1318 DNPRINTF(SR_D_META, " volid %u with %u chunks\n", 1319 bv->sbv_volid, bv->sbv_chunk_no); 1320 #endif 1321 1322 /* Create hotspare chunk metadata. */ 1323 hotspare = malloc(sizeof(struct sr_chunk), M_DEVBUF, 1324 M_NOWAIT | M_ZERO); 1325 if (hotspare == NULL) { 1326 printf("%s: failed to allocate hotspare\n", 1327 DEVNAME(sc)); 1328 goto unwind; 1329 } 1330 1331 bc = SLIST_FIRST(&bv->sbv_chunks); 1332 sr_meta_getdevname(sc, bc->sbc_mm, devname, sizeof(devname)); 1333 hotspare->src_dev_mm = bc->sbc_mm; 1334 strlcpy(hotspare->src_devname, devname, 1335 sizeof(hotspare->src_devname)); 1336 hotspare->src_size = bc->sbc_metadata->ssdi.ssd_size; 1337 1338 hm = &hotspare->src_meta; 1339 hm->scmi.scm_volid = SR_HOTSPARE_VOLID; 1340 hm->scmi.scm_chunk_id = 0; 1341 hm->scmi.scm_size = bc->sbc_metadata->ssdi.ssd_size; 1342 hm->scmi.scm_coerced_size = bc->sbc_metadata->ssdi.ssd_size; 1343 strlcpy(hm->scmi.scm_devname, devname, 1344 sizeof(hm->scmi.scm_devname)); 1345 memcpy(&hm->scmi.scm_uuid, &bc->sbc_metadata->ssdi.ssd_uuid, 1346 sizeof(struct sr_uuid)); 1347 1348 sr_checksum(sc, hm, &hm->scm_checksum, 1349 sizeof(struct sr_meta_chunk_invariant)); 1350 1351 hm->scm_status = BIOC_SDHOTSPARE; 1352 1353 /* Add chunk to hotspare list. */ 1354 rw_enter_write(&sc->sc_hs_lock); 1355 cl = &sc->sc_hotspare_list; 1356 if (SLIST_EMPTY(cl)) 1357 SLIST_INSERT_HEAD(cl, hotspare, src_link); 1358 else { 1359 SLIST_FOREACH(chunk, cl, src_link) 1360 last = chunk; 1361 SLIST_INSERT_AFTER(last, hotspare, src_link); 1362 } 1363 sc->sc_hotspare_no++; 1364 rw_exit_write(&sc->sc_hs_lock); 1365 1366 } 1367 1368 /* 1369 * Assemble RAID volumes. 1370 */ 1371 SLIST_FOREACH(bv, &bvh, sbv_link) { 1372 1373 bzero(&bcr, sizeof(bcr)); 1374 data = NULL; 1375 1376 /* Check if this is a hotspare "volume". */ 1377 if (bv->sbv_level == SR_HOTSPARE_LEVEL && 1378 bv->sbv_chunk_no == 1) 1379 continue; 1380 1381 /* 1382 * Skip volumes that are marked as no auto assemble, unless 1383 * this was the volume which we actually booted from. 1384 */ 1385 if (bcmp(&sr_bootuuid, &bv->sbv_uuid, sizeof(sr_bootuuid)) != 0) 1386 if (bv->sbv_flags & BIOC_SCNOAUTOASSEMBLE) 1387 continue; 1388 1389 #ifdef SR_DEBUG 1390 DNPRINTF(SR_D_META, "%s: assembling volume ", DEVNAME(sc)); 1391 if (sr_debug & SR_D_META) 1392 sr_uuid_print(&bv->sbv_uuid, 0); 1393 DNPRINTF(SR_D_META, " volid %u with %u chunks\n", 1394 bv->sbv_volid, bv->sbv_chunk_no); 1395 #endif 1396 1397 /* 1398 * If this is a crypto volume, try to find a matching 1399 * key disk... 1400 */ 1401 bcr.bc_key_disk = NODEV; 1402 if (bv->sbv_level == 'C') { 1403 SLIST_FOREACH(bc, &kdh, sbc_link) { 1404 if (bcmp(&bc->sbc_metadata->ssdi.ssd_uuid, 1405 &bv->sbv_uuid, 1406 sizeof(bc->sbc_metadata->ssdi.ssd_uuid)) 1407 == 0) 1408 bcr.bc_key_disk = bc->sbc_mm; 1409 } 1410 } 1411 1412 for (i = 0; i < BIOC_CRMAXLEN; i++) { 1413 devs[i] = NODEV; /* mark device as illegal */ 1414 ondisk[i] = 0; 1415 } 1416 1417 SLIST_FOREACH(bc, &bv->sbv_chunks, sbc_link) { 1418 if (devs[bc->sbc_chunk_id] != NODEV) { 1419 bv->sbv_chunks_found--; 1420 sr_meta_getdevname(sc, bc->sbc_mm, devname, 1421 sizeof(devname)); 1422 printf("%s: found duplicate chunk %u for " 1423 "volume %u on device %s\n", DEVNAME(sc), 1424 bc->sbc_chunk_id, bv->sbv_volid, devname); 1425 } 1426 1427 if (devs[bc->sbc_chunk_id] == NODEV || 1428 bc->sbc_metadata->ssd_ondisk > 1429 ondisk[bc->sbc_chunk_id]) { 1430 devs[bc->sbc_chunk_id] = bc->sbc_mm; 1431 ondisk[bc->sbc_chunk_id] = 1432 bc->sbc_metadata->ssd_ondisk; 1433 DNPRINTF(SR_D_META, "%s: using ondisk " 1434 "metadata version %llu for chunk %u\n", 1435 DEVNAME(sc), ondisk[bc->sbc_chunk_id], 1436 bc->sbc_chunk_id); 1437 } 1438 } 1439 1440 if (bv->sbv_chunk_no != bv->sbv_chunks_found) { 1441 printf("%s: not all chunks were provided; " 1442 "attempting to bring volume %d online\n", 1443 DEVNAME(sc), bv->sbv_volid); 1444 } 1445 1446 bcr.bc_level = bv->sbv_level; 1447 bcr.bc_dev_list_len = bv->sbv_chunk_no * sizeof(dev_t); 1448 bcr.bc_dev_list = devs; 1449 bcr.bc_flags = BIOC_SCDEVT | 1450 (bv->sbv_flags & BIOC_SCNOAUTOASSEMBLE); 1451 1452 if (bv->sbv_level == 'C' && 1453 bcmp(&sr_bootuuid, &bv->sbv_uuid, sizeof(sr_bootuuid)) == 0) 1454 data = sr_bootkey; 1455 1456 rw_enter_write(&sc->sc_lock); 1457 bio_status_init(&sc->sc_status, &sc->sc_dev); 1458 sr_ioctl_createraid(sc, &bcr, 0, data); 1459 rw_exit_write(&sc->sc_lock); 1460 1461 rv++; 1462 } 1463 1464 /* done with metadata */ 1465 unwind: 1466 /* Free boot volumes and associated chunks. */ 1467 for (bv1 = SLIST_FIRST(&bvh); bv1 != NULL; bv1 = bv2) { 1468 bv2 = SLIST_NEXT(bv1, sbv_link); 1469 for (bc1 = SLIST_FIRST(&bv1->sbv_chunks); bc1 != NULL; 1470 bc1 = bc2) { 1471 bc2 = SLIST_NEXT(bc1, sbc_link); 1472 free(bc1->sbc_metadata, M_DEVBUF, 1473 sizeof(*bc1->sbc_metadata)); 1474 free(bc1, M_DEVBUF, sizeof(*bc1)); 1475 } 1476 free(bv1, M_DEVBUF, sizeof(*bv1)); 1477 } 1478 /* Free keydisks chunks. */ 1479 for (bc1 = SLIST_FIRST(&kdh); bc1 != NULL; bc1 = bc2) { 1480 bc2 = SLIST_NEXT(bc1, sbc_link); 1481 free(bc1->sbc_metadata, M_DEVBUF, sizeof(*bc1->sbc_metadata)); 1482 free(bc1, M_DEVBUF, sizeof(*bc1)); 1483 } 1484 /* Free unallocated chunks. */ 1485 for (bc1 = SLIST_FIRST(&bch); bc1 != NULL; bc1 = bc2) { 1486 bc2 = SLIST_NEXT(bc1, sbc_link); 1487 free(bc1->sbc_metadata, M_DEVBUF, sizeof(*bc1->sbc_metadata)); 1488 free(bc1, M_DEVBUF, sizeof(*bc1)); 1489 } 1490 1491 while (!SLIST_EMPTY(&sdklist)) { 1492 sdk = SLIST_FIRST(&sdklist); 1493 SLIST_REMOVE_HEAD(&sdklist, sdk_link); 1494 free(sdk, M_DEVBUF, sizeof(*sdk)); 1495 } 1496 1497 free(devs, M_DEVBUF, BIOC_CRMAXLEN * sizeof(dev_t)); 1498 free(ondisk, M_DEVBUF, BIOC_CRMAXLEN * sizeof(u_int64_t)); 1499 1500 return (rv); 1501 } 1502 1503 void 1504 sr_map_root(void) 1505 { 1506 struct sr_softc *sc = softraid0; 1507 struct sr_discipline *sd; 1508 struct sr_meta_opt_item *omi; 1509 struct sr_meta_boot *sbm; 1510 u_char duid[8]; 1511 int i; 1512 1513 DNPRINTF(SR_D_MISC, "%s: sr_map_root\n", DEVNAME(sc)); 1514 1515 if (sc == NULL) 1516 return; 1517 1518 bzero(duid, sizeof(duid)); 1519 if (bcmp(rootduid, duid, sizeof(duid)) == 0) { 1520 DNPRINTF(SR_D_MISC, "%s: root duid is zero\n", DEVNAME(sc)); 1521 return; 1522 } 1523 1524 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 1525 SLIST_FOREACH(omi, &sd->sd_meta_opt, omi_link) { 1526 if (omi->omi_som->som_type != SR_OPT_BOOT) 1527 continue; 1528 sbm = (struct sr_meta_boot *)omi->omi_som; 1529 for (i = 0; i < SR_MAX_BOOT_DISKS; i++) { 1530 if (bcmp(rootduid, sbm->sbm_boot_duid[i], 1531 sizeof(rootduid)) == 0) { 1532 memcpy(rootduid, sbm->sbm_root_duid, 1533 sizeof(rootduid)); 1534 DNPRINTF(SR_D_MISC, "%s: root duid " 1535 "mapped to %s\n", DEVNAME(sc), 1536 duid_format(rootduid)); 1537 return; 1538 } 1539 } 1540 } 1541 } 1542 } 1543 1544 int 1545 sr_meta_native_probe(struct sr_softc *sc, struct sr_chunk *ch_entry) 1546 { 1547 struct disklabel label; 1548 char *devname; 1549 int error, part; 1550 u_int64_t size; 1551 1552 DNPRINTF(SR_D_META, "%s: sr_meta_native_probe(%s)\n", 1553 DEVNAME(sc), ch_entry->src_devname); 1554 1555 devname = ch_entry->src_devname; 1556 part = DISKPART(ch_entry->src_dev_mm); 1557 1558 /* get disklabel */ 1559 error = VOP_IOCTL(ch_entry->src_vn, DIOCGDINFO, (caddr_t)&label, FREAD, 1560 NOCRED, curproc); 1561 if (error) { 1562 DNPRINTF(SR_D_META, "%s: %s can't obtain disklabel\n", 1563 DEVNAME(sc), devname); 1564 goto unwind; 1565 } 1566 memcpy(ch_entry->src_duid, label.d_uid, sizeof(ch_entry->src_duid)); 1567 1568 /* make sure the partition is of the right type */ 1569 if (label.d_partitions[part].p_fstype != FS_RAID) { 1570 DNPRINTF(SR_D_META, 1571 "%s: %s partition not of type RAID (%d)\n", DEVNAME(sc), 1572 devname, 1573 label.d_partitions[part].p_fstype); 1574 goto unwind; 1575 } 1576 1577 size = DL_SECTOBLK(&label, DL_GETPSIZE(&label.d_partitions[part])); 1578 if (size <= SR_DATA_OFFSET) { 1579 DNPRINTF(SR_D_META, "%s: %s partition too small\n", DEVNAME(sc), 1580 devname); 1581 goto unwind; 1582 } 1583 size -= SR_DATA_OFFSET; 1584 if (size > INT64_MAX) { 1585 DNPRINTF(SR_D_META, "%s: %s partition too large\n", DEVNAME(sc), 1586 devname); 1587 goto unwind; 1588 } 1589 ch_entry->src_size = size; 1590 ch_entry->src_secsize = label.d_secsize; 1591 1592 DNPRINTF(SR_D_META, "%s: probe found %s size %lld\n", DEVNAME(sc), 1593 devname, (long long)size); 1594 1595 return (SR_META_F_NATIVE); 1596 unwind: 1597 DNPRINTF(SR_D_META, "%s: invalid device: %s\n", DEVNAME(sc), 1598 devname ? devname : "nodev"); 1599 return (SR_META_F_INVALID); 1600 } 1601 1602 int 1603 sr_meta_native_attach(struct sr_discipline *sd, int force) 1604 { 1605 struct sr_softc *sc = sd->sd_sc; 1606 struct sr_chunk_head *cl = &sd->sd_vol.sv_chunk_list; 1607 struct sr_metadata *md = NULL; 1608 struct sr_chunk *ch_entry, *ch_next; 1609 struct sr_uuid uuid; 1610 u_int64_t version = 0; 1611 int sr, not_sr, rv = 1, d, expected = -1, old_meta = 0; 1612 1613 DNPRINTF(SR_D_META, "%s: sr_meta_native_attach\n", DEVNAME(sc)); 1614 1615 md = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, M_ZERO | M_NOWAIT); 1616 if (md == NULL) { 1617 sr_error(sc, "not enough memory for metadata buffer"); 1618 goto bad; 1619 } 1620 1621 bzero(&uuid, sizeof uuid); 1622 1623 sr = not_sr = d = 0; 1624 SLIST_FOREACH(ch_entry, cl, src_link) { 1625 if (ch_entry->src_dev_mm == NODEV) 1626 continue; 1627 1628 if (sr_meta_native_read(sd, ch_entry->src_dev_mm, md, NULL)) { 1629 sr_error(sc, "could not read native metadata"); 1630 goto bad; 1631 } 1632 1633 if (md->ssdi.ssd_magic == SR_MAGIC) { 1634 sr++; 1635 ch_entry->src_meta.scmi.scm_chunk_id = 1636 md->ssdi.ssd_chunk_id; 1637 if (d == 0) { 1638 memcpy(&uuid, &md->ssdi.ssd_uuid, sizeof uuid); 1639 expected = md->ssdi.ssd_chunk_no; 1640 version = md->ssd_ondisk; 1641 d++; 1642 continue; 1643 } else if (bcmp(&md->ssdi.ssd_uuid, &uuid, 1644 sizeof uuid)) { 1645 sr_error(sc, "not part of the same volume"); 1646 goto bad; 1647 } 1648 if (md->ssd_ondisk != version) { 1649 old_meta++; 1650 version = MAX(md->ssd_ondisk, version); 1651 } 1652 } else 1653 not_sr++; 1654 } 1655 1656 if (sr && not_sr && !force) { 1657 sr_error(sc, "not all chunks are of the native metadata " 1658 "format"); 1659 goto bad; 1660 } 1661 1662 /* mixed metadata versions; mark bad disks offline */ 1663 if (old_meta) { 1664 d = 0; 1665 for (ch_entry = SLIST_FIRST(cl); ch_entry != NULL; 1666 ch_entry = ch_next, d++) { 1667 ch_next = SLIST_NEXT(ch_entry, src_link); 1668 1669 /* XXX do we want to read this again? */ 1670 if (ch_entry->src_dev_mm == NODEV) 1671 panic("src_dev_mm == NODEV"); 1672 if (sr_meta_native_read(sd, ch_entry->src_dev_mm, md, 1673 NULL)) 1674 sr_warn(sc, "could not read native metadata"); 1675 if (md->ssd_ondisk != version) 1676 sd->sd_vol.sv_chunks[d]->src_meta.scm_status = 1677 BIOC_SDOFFLINE; 1678 } 1679 } 1680 1681 if (expected != sr && !force && expected != -1) { 1682 DNPRINTF(SR_D_META, "%s: not all chunks were provided, trying " 1683 "anyway\n", DEVNAME(sc)); 1684 } 1685 1686 rv = 0; 1687 bad: 1688 free(md, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 1689 return (rv); 1690 } 1691 1692 int 1693 sr_meta_native_read(struct sr_discipline *sd, dev_t dev, 1694 struct sr_metadata *md, void *fm) 1695 { 1696 #ifdef SR_DEBUG 1697 struct sr_softc *sc = sd->sd_sc; 1698 #endif 1699 DNPRINTF(SR_D_META, "%s: sr_meta_native_read(0x%x, %p)\n", 1700 DEVNAME(sc), dev, md); 1701 1702 return (sr_meta_rw(sd, dev, md, B_READ)); 1703 } 1704 1705 int 1706 sr_meta_native_write(struct sr_discipline *sd, dev_t dev, 1707 struct sr_metadata *md, void *fm) 1708 { 1709 #ifdef SR_DEBUG 1710 struct sr_softc *sc = sd->sd_sc; 1711 #endif 1712 DNPRINTF(SR_D_META, "%s: sr_meta_native_write(0x%x, %p)\n", 1713 DEVNAME(sc), dev, md); 1714 1715 return (sr_meta_rw(sd, dev, md, B_WRITE)); 1716 } 1717 1718 void 1719 sr_hotplug_register(struct sr_discipline *sd, void *func) 1720 { 1721 struct sr_hotplug_list *mhe; 1722 1723 DNPRINTF(SR_D_MISC, "%s: sr_hotplug_register: %p\n", 1724 DEVNAME(sd->sd_sc), func); 1725 1726 /* make sure we aren't on the list yet */ 1727 SLIST_FOREACH(mhe, &sr_hotplug_callbacks, shl_link) 1728 if (mhe->sh_hotplug == func) 1729 return; 1730 1731 mhe = malloc(sizeof(struct sr_hotplug_list), M_DEVBUF, 1732 M_WAITOK | M_ZERO); 1733 mhe->sh_hotplug = func; 1734 mhe->sh_sd = sd; 1735 SLIST_INSERT_HEAD(&sr_hotplug_callbacks, mhe, shl_link); 1736 } 1737 1738 void 1739 sr_hotplug_unregister(struct sr_discipline *sd, void *func) 1740 { 1741 struct sr_hotplug_list *mhe; 1742 1743 DNPRINTF(SR_D_MISC, "%s: sr_hotplug_unregister: %s %p\n", 1744 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, func); 1745 1746 /* make sure we are on the list yet */ 1747 SLIST_FOREACH(mhe, &sr_hotplug_callbacks, shl_link) { 1748 if (mhe->sh_hotplug == func) 1749 break; 1750 } 1751 if (mhe != NULL) { 1752 SLIST_REMOVE(&sr_hotplug_callbacks, mhe, 1753 sr_hotplug_list, shl_link); 1754 free(mhe, M_DEVBUF, sizeof(*mhe)); 1755 } 1756 } 1757 1758 void 1759 sr_disk_attach(struct disk *diskp, int action) 1760 { 1761 struct sr_hotplug_list *mhe; 1762 1763 SLIST_FOREACH(mhe, &sr_hotplug_callbacks, shl_link) 1764 if (mhe->sh_sd->sd_ready) 1765 mhe->sh_hotplug(mhe->sh_sd, diskp, action); 1766 } 1767 1768 int 1769 sr_match(struct device *parent, void *match, void *aux) 1770 { 1771 return (1); 1772 } 1773 1774 void 1775 sr_attach(struct device *parent, struct device *self, void *aux) 1776 { 1777 struct sr_softc *sc = (void *)self; 1778 struct scsibus_attach_args saa; 1779 1780 DNPRINTF(SR_D_MISC, "\n%s: sr_attach", DEVNAME(sc)); 1781 1782 if (softraid0 == NULL) 1783 softraid0 = sc; 1784 1785 rw_init(&sc->sc_lock, "sr_lock"); 1786 rw_init(&sc->sc_hs_lock, "sr_hs_lock"); 1787 1788 SLIST_INIT(&sr_hotplug_callbacks); 1789 TAILQ_INIT(&sc->sc_dis_list); 1790 SLIST_INIT(&sc->sc_hotspare_list); 1791 1792 #if NBIO > 0 1793 if (bio_register(&sc->sc_dev, sr_bio_ioctl) != 0) 1794 printf("%s: controller registration failed", DEVNAME(sc)); 1795 #endif /* NBIO > 0 */ 1796 1797 #ifndef SMALL_KERNEL 1798 strlcpy(sc->sc_sensordev.xname, DEVNAME(sc), 1799 sizeof(sc->sc_sensordev.xname)); 1800 sensordev_install(&sc->sc_sensordev); 1801 #endif /* SMALL_KERNEL */ 1802 1803 printf("\n"); 1804 1805 sc->sc_link.adapter_softc = sc; 1806 sc->sc_link.adapter = &sr_switch; 1807 sc->sc_link.adapter_target = SDEV_NO_ADAPTER_TARGET; 1808 sc->sc_link.adapter_buswidth = SR_MAX_LD; 1809 sc->sc_link.luns = 1; 1810 1811 saa.saa_sc_link = &sc->sc_link; 1812 1813 sc->sc_scsibus = (struct scsibus_softc *)config_found(&sc->sc_dev, 1814 &saa, 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 void 1888 sr_copy_internal_data(struct scsi_xfer *xs, void *v, size_t size) 1889 { 1890 size_t copy_cnt; 1891 1892 DNPRINTF(SR_D_MISC, "sr_copy_internal_data xs: %p size: %zu\n", 1893 xs, size); 1894 1895 if (xs->datalen) { 1896 copy_cnt = MIN(size, xs->datalen); 1897 memcpy(xs->data, v, copy_cnt); 1898 } 1899 } 1900 1901 int 1902 sr_ccb_alloc(struct sr_discipline *sd) 1903 { 1904 struct sr_ccb *ccb; 1905 int i; 1906 1907 if (!sd) 1908 return (1); 1909 1910 DNPRINTF(SR_D_CCB, "%s: sr_ccb_alloc\n", DEVNAME(sd->sd_sc)); 1911 1912 if (sd->sd_ccb) 1913 return (1); 1914 1915 sd->sd_ccb = mallocarray(sd->sd_max_wu, 1916 sd->sd_max_ccb_per_wu * sizeof(struct sr_ccb), 1917 M_DEVBUF, M_WAITOK | M_ZERO); 1918 TAILQ_INIT(&sd->sd_ccb_freeq); 1919 for (i = 0; i < sd->sd_max_wu * sd->sd_max_ccb_per_wu; i++) { 1920 ccb = &sd->sd_ccb[i]; 1921 ccb->ccb_dis = sd; 1922 sr_ccb_put(ccb); 1923 } 1924 1925 DNPRINTF(SR_D_CCB, "%s: sr_ccb_alloc ccb: %d\n", 1926 DEVNAME(sd->sd_sc), sd->sd_max_wu * sd->sd_max_ccb_per_wu); 1927 1928 return (0); 1929 } 1930 1931 void 1932 sr_ccb_free(struct sr_discipline *sd) 1933 { 1934 struct sr_ccb *ccb; 1935 1936 if (!sd) 1937 return; 1938 1939 DNPRINTF(SR_D_CCB, "%s: sr_ccb_free %p\n", DEVNAME(sd->sd_sc), sd); 1940 1941 while ((ccb = TAILQ_FIRST(&sd->sd_ccb_freeq)) != NULL) 1942 TAILQ_REMOVE(&sd->sd_ccb_freeq, ccb, ccb_link); 1943 1944 free(sd->sd_ccb, M_DEVBUF, sd->sd_max_wu * sd->sd_max_ccb_per_wu * 1945 sizeof(struct sr_ccb)); 1946 } 1947 1948 struct sr_ccb * 1949 sr_ccb_get(struct sr_discipline *sd) 1950 { 1951 struct sr_ccb *ccb; 1952 int s; 1953 1954 s = splbio(); 1955 1956 ccb = TAILQ_FIRST(&sd->sd_ccb_freeq); 1957 if (ccb) { 1958 TAILQ_REMOVE(&sd->sd_ccb_freeq, ccb, ccb_link); 1959 ccb->ccb_state = SR_CCB_INPROGRESS; 1960 } 1961 1962 splx(s); 1963 1964 DNPRINTF(SR_D_CCB, "%s: sr_ccb_get: %p\n", DEVNAME(sd->sd_sc), 1965 ccb); 1966 1967 return (ccb); 1968 } 1969 1970 void 1971 sr_ccb_put(struct sr_ccb *ccb) 1972 { 1973 struct sr_discipline *sd = ccb->ccb_dis; 1974 int s; 1975 1976 DNPRINTF(SR_D_CCB, "%s: sr_ccb_put: %p\n", DEVNAME(sd->sd_sc), 1977 ccb); 1978 1979 s = splbio(); 1980 1981 ccb->ccb_wu = NULL; 1982 ccb->ccb_state = SR_CCB_FREE; 1983 ccb->ccb_target = -1; 1984 ccb->ccb_opaque = NULL; 1985 1986 TAILQ_INSERT_TAIL(&sd->sd_ccb_freeq, ccb, ccb_link); 1987 1988 splx(s); 1989 } 1990 1991 struct sr_ccb * 1992 sr_ccb_rw(struct sr_discipline *sd, int chunk, daddr_t blkno, 1993 long len, u_int8_t *data, int xsflags, int ccbflags) 1994 { 1995 struct sr_chunk *sc = sd->sd_vol.sv_chunks[chunk]; 1996 struct sr_ccb *ccb = NULL; 1997 1998 ccb = sr_ccb_get(sd); 1999 if (ccb == NULL) 2000 goto out; 2001 2002 ccb->ccb_flags = ccbflags; 2003 ccb->ccb_target = chunk; 2004 2005 ccb->ccb_buf.b_flags = B_PHYS | B_CALL; 2006 if (ISSET(xsflags, SCSI_DATA_IN)) 2007 ccb->ccb_buf.b_flags |= B_READ; 2008 else 2009 ccb->ccb_buf.b_flags |= B_WRITE; 2010 2011 ccb->ccb_buf.b_blkno = blkno + sd->sd_meta->ssd_data_blkno; 2012 ccb->ccb_buf.b_bcount = len; 2013 ccb->ccb_buf.b_bufsize = len; 2014 ccb->ccb_buf.b_resid = len; 2015 ccb->ccb_buf.b_data = data; 2016 ccb->ccb_buf.b_error = 0; 2017 ccb->ccb_buf.b_iodone = sd->sd_scsi_intr; 2018 ccb->ccb_buf.b_proc = curproc; 2019 ccb->ccb_buf.b_dev = sc->src_dev_mm; 2020 ccb->ccb_buf.b_vp = sc->src_vn; 2021 ccb->ccb_buf.b_bq = NULL; 2022 2023 if (!ISSET(ccb->ccb_buf.b_flags, B_READ)) 2024 ccb->ccb_buf.b_vp->v_numoutput++; 2025 2026 LIST_INIT(&ccb->ccb_buf.b_dep); 2027 2028 DNPRINTF(SR_D_DIS, "%s: %s %s ccb " 2029 "b_bcount %ld b_blkno %lld b_flags 0x%0lx b_data %p\n", 2030 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, sd->sd_name, 2031 ccb->ccb_buf.b_bcount, (long long)ccb->ccb_buf.b_blkno, 2032 ccb->ccb_buf.b_flags, ccb->ccb_buf.b_data); 2033 2034 out: 2035 return ccb; 2036 } 2037 2038 void 2039 sr_ccb_done(struct sr_ccb *ccb) 2040 { 2041 struct sr_workunit *wu = ccb->ccb_wu; 2042 struct sr_discipline *sd = wu->swu_dis; 2043 struct sr_softc *sc = sd->sd_sc; 2044 2045 DNPRINTF(SR_D_INTR, "%s: %s %s ccb done b_bcount %ld b_resid %zu" 2046 " b_flags 0x%0lx block %lld target %d\n", 2047 DEVNAME(sc), sd->sd_meta->ssd_devname, sd->sd_name, 2048 ccb->ccb_buf.b_bcount, ccb->ccb_buf.b_resid, ccb->ccb_buf.b_flags, 2049 (long long)ccb->ccb_buf.b_blkno, ccb->ccb_target); 2050 2051 splassert(IPL_BIO); 2052 2053 if (ccb->ccb_target == -1) 2054 panic("%s: invalid target on wu: %p", DEVNAME(sc), wu); 2055 2056 if (ccb->ccb_buf.b_flags & B_ERROR) { 2057 DNPRINTF(SR_D_INTR, "%s: i/o error on block %lld target %d\n", 2058 DEVNAME(sc), (long long)ccb->ccb_buf.b_blkno, 2059 ccb->ccb_target); 2060 if (ISSET(sd->sd_capabilities, SR_CAP_REDUNDANT)) 2061 sd->sd_set_chunk_state(sd, ccb->ccb_target, 2062 BIOC_SDOFFLINE); 2063 else 2064 printf("%s: %s: i/o error %d @ %s block %lld\n", 2065 DEVNAME(sc), sd->sd_meta->ssd_devname, 2066 ccb->ccb_buf.b_error, sd->sd_name, 2067 (long long)ccb->ccb_buf.b_blkno); 2068 ccb->ccb_state = SR_CCB_FAILED; 2069 wu->swu_ios_failed++; 2070 } else { 2071 ccb->ccb_state = SR_CCB_OK; 2072 wu->swu_ios_succeeded++; 2073 } 2074 2075 wu->swu_ios_complete++; 2076 } 2077 2078 int 2079 sr_wu_alloc(struct sr_discipline *sd) 2080 { 2081 struct sr_workunit *wu; 2082 int i, no_wu; 2083 2084 DNPRINTF(SR_D_WU, "%s: sr_wu_alloc %p %d\n", DEVNAME(sd->sd_sc), 2085 sd, sd->sd_max_wu); 2086 2087 no_wu = sd->sd_max_wu; 2088 sd->sd_wu_pending = no_wu; 2089 2090 mtx_init(&sd->sd_wu_mtx, IPL_BIO); 2091 TAILQ_INIT(&sd->sd_wu); 2092 TAILQ_INIT(&sd->sd_wu_freeq); 2093 TAILQ_INIT(&sd->sd_wu_pendq); 2094 TAILQ_INIT(&sd->sd_wu_defq); 2095 2096 for (i = 0; i < no_wu; i++) { 2097 wu = malloc(sd->sd_wu_size, M_DEVBUF, M_WAITOK | M_ZERO); 2098 TAILQ_INSERT_TAIL(&sd->sd_wu, wu, swu_next); 2099 TAILQ_INIT(&wu->swu_ccb); 2100 wu->swu_dis = sd; 2101 task_set(&wu->swu_task, sr_wu_done_callback, wu); 2102 sr_wu_put(sd, wu); 2103 } 2104 2105 return (0); 2106 } 2107 2108 void 2109 sr_wu_free(struct sr_discipline *sd) 2110 { 2111 struct sr_workunit *wu; 2112 2113 DNPRINTF(SR_D_WU, "%s: sr_wu_free %p\n", DEVNAME(sd->sd_sc), sd); 2114 2115 while ((wu = TAILQ_FIRST(&sd->sd_wu_freeq)) != NULL) 2116 TAILQ_REMOVE(&sd->sd_wu_freeq, wu, swu_link); 2117 while ((wu = TAILQ_FIRST(&sd->sd_wu_pendq)) != NULL) 2118 TAILQ_REMOVE(&sd->sd_wu_pendq, wu, swu_link); 2119 while ((wu = TAILQ_FIRST(&sd->sd_wu_defq)) != NULL) 2120 TAILQ_REMOVE(&sd->sd_wu_defq, wu, swu_link); 2121 2122 while ((wu = TAILQ_FIRST(&sd->sd_wu)) != NULL) { 2123 TAILQ_REMOVE(&sd->sd_wu, wu, swu_next); 2124 free(wu, M_DEVBUF, sd->sd_wu_size); 2125 } 2126 } 2127 2128 void * 2129 sr_wu_get(void *xsd) 2130 { 2131 struct sr_discipline *sd = (struct sr_discipline *)xsd; 2132 struct sr_workunit *wu; 2133 2134 mtx_enter(&sd->sd_wu_mtx); 2135 wu = TAILQ_FIRST(&sd->sd_wu_freeq); 2136 if (wu) { 2137 TAILQ_REMOVE(&sd->sd_wu_freeq, wu, swu_link); 2138 sd->sd_wu_pending++; 2139 } 2140 mtx_leave(&sd->sd_wu_mtx); 2141 2142 DNPRINTF(SR_D_WU, "%s: sr_wu_get: %p\n", DEVNAME(sd->sd_sc), wu); 2143 2144 return (wu); 2145 } 2146 2147 void 2148 sr_wu_put(void *xsd, void *xwu) 2149 { 2150 struct sr_discipline *sd = (struct sr_discipline *)xsd; 2151 struct sr_workunit *wu = (struct sr_workunit *)xwu; 2152 2153 DNPRINTF(SR_D_WU, "%s: sr_wu_put: %p\n", DEVNAME(sd->sd_sc), wu); 2154 2155 sr_wu_release_ccbs(wu); 2156 sr_wu_init(sd, wu); 2157 2158 mtx_enter(&sd->sd_wu_mtx); 2159 TAILQ_INSERT_TAIL(&sd->sd_wu_freeq, wu, swu_link); 2160 sd->sd_wu_pending--; 2161 mtx_leave(&sd->sd_wu_mtx); 2162 } 2163 2164 void 2165 sr_wu_init(struct sr_discipline *sd, struct sr_workunit *wu) 2166 { 2167 int s; 2168 2169 s = splbio(); 2170 if (wu->swu_cb_active == 1) 2171 panic("%s: sr_wu_init got active wu", DEVNAME(sd->sd_sc)); 2172 splx(s); 2173 2174 wu->swu_xs = NULL; 2175 wu->swu_state = SR_WU_FREE; 2176 wu->swu_flags = 0; 2177 wu->swu_blk_start = 0; 2178 wu->swu_blk_end = 0; 2179 wu->swu_collider = NULL; 2180 } 2181 2182 void 2183 sr_wu_enqueue_ccb(struct sr_workunit *wu, struct sr_ccb *ccb) 2184 { 2185 struct sr_discipline *sd = wu->swu_dis; 2186 int s; 2187 2188 s = splbio(); 2189 if (wu->swu_cb_active == 1) 2190 panic("%s: sr_wu_enqueue_ccb got active wu", 2191 DEVNAME(sd->sd_sc)); 2192 ccb->ccb_wu = wu; 2193 wu->swu_io_count++; 2194 TAILQ_INSERT_TAIL(&wu->swu_ccb, ccb, ccb_link); 2195 splx(s); 2196 } 2197 2198 void 2199 sr_wu_release_ccbs(struct sr_workunit *wu) 2200 { 2201 struct sr_ccb *ccb; 2202 2203 /* Return all ccbs that are associated with this workunit. */ 2204 while ((ccb = TAILQ_FIRST(&wu->swu_ccb)) != NULL) { 2205 TAILQ_REMOVE(&wu->swu_ccb, ccb, ccb_link); 2206 sr_ccb_put(ccb); 2207 } 2208 2209 wu->swu_io_count = 0; 2210 wu->swu_ios_complete = 0; 2211 wu->swu_ios_failed = 0; 2212 wu->swu_ios_succeeded = 0; 2213 } 2214 2215 void 2216 sr_wu_done(struct sr_workunit *wu) 2217 { 2218 struct sr_discipline *sd = wu->swu_dis; 2219 2220 DNPRINTF(SR_D_INTR, "%s: sr_wu_done count %d completed %d failed %d\n", 2221 DEVNAME(sd->sd_sc), wu->swu_io_count, wu->swu_ios_complete, 2222 wu->swu_ios_failed); 2223 2224 if (wu->swu_ios_complete < wu->swu_io_count) 2225 return; 2226 2227 task_add(sd->sd_taskq, &wu->swu_task); 2228 } 2229 2230 void 2231 sr_wu_done_callback(void *xwu) 2232 { 2233 struct sr_workunit *wu = xwu; 2234 struct sr_discipline *sd = wu->swu_dis; 2235 struct scsi_xfer *xs = wu->swu_xs; 2236 struct sr_workunit *wup; 2237 int s; 2238 2239 /* 2240 * The SR_WUF_DISCIPLINE or SR_WUF_REBUILD flag must be set if 2241 * the work unit is not associated with a scsi_xfer. 2242 */ 2243 KASSERT(xs != NULL || 2244 (wu->swu_flags & (SR_WUF_DISCIPLINE|SR_WUF_REBUILD))); 2245 2246 s = splbio(); 2247 2248 if (xs != NULL) { 2249 if (wu->swu_ios_failed) 2250 xs->error = XS_DRIVER_STUFFUP; 2251 else 2252 xs->error = XS_NOERROR; 2253 } 2254 2255 if (sd->sd_scsi_wu_done) { 2256 if (sd->sd_scsi_wu_done(wu) == SR_WU_RESTART) 2257 goto done; 2258 } 2259 2260 /* Remove work unit from pending queue. */ 2261 TAILQ_FOREACH(wup, &sd->sd_wu_pendq, swu_link) 2262 if (wup == wu) 2263 break; 2264 if (wup == NULL) 2265 panic("%s: wu %p not on pending queue", 2266 DEVNAME(sd->sd_sc), wu); 2267 TAILQ_REMOVE(&sd->sd_wu_pendq, wu, swu_link); 2268 2269 if (wu->swu_collider) { 2270 if (wu->swu_ios_failed) 2271 sr_raid_recreate_wu(wu->swu_collider); 2272 2273 /* XXX Should the collider be failed if this xs failed? */ 2274 sr_raid_startwu(wu->swu_collider); 2275 } 2276 2277 /* 2278 * If a discipline provides its own sd_scsi_done function, then it 2279 * is responsible for calling sr_scsi_done() once I/O is complete. 2280 */ 2281 if (wu->swu_flags & SR_WUF_REBUILD) 2282 wu->swu_flags |= SR_WUF_REBUILDIOCOMP; 2283 if (wu->swu_flags & SR_WUF_WAKEUP) 2284 wakeup(wu); 2285 if (sd->sd_scsi_done) 2286 sd->sd_scsi_done(wu); 2287 else if (wu->swu_flags & SR_WUF_DISCIPLINE) 2288 sr_scsi_wu_put(sd, wu); 2289 else if (!(wu->swu_flags & SR_WUF_REBUILD)) 2290 sr_scsi_done(sd, xs); 2291 2292 done: 2293 splx(s); 2294 } 2295 2296 struct sr_workunit * 2297 sr_scsi_wu_get(struct sr_discipline *sd, int flags) 2298 { 2299 return scsi_io_get(&sd->sd_iopool, flags); 2300 } 2301 2302 void 2303 sr_scsi_wu_put(struct sr_discipline *sd, struct sr_workunit *wu) 2304 { 2305 scsi_io_put(&sd->sd_iopool, wu); 2306 2307 if (sd->sd_sync && sd->sd_wu_pending == 0) 2308 wakeup(sd); 2309 } 2310 2311 void 2312 sr_scsi_done(struct sr_discipline *sd, struct scsi_xfer *xs) 2313 { 2314 DNPRINTF(SR_D_DIS, "%s: sr_scsi_done: xs %p\n", DEVNAME(sd->sd_sc), xs); 2315 2316 if (xs->error == XS_NOERROR) 2317 xs->resid = 0; 2318 2319 scsi_done(xs); 2320 2321 if (sd->sd_sync && sd->sd_wu_pending == 0) 2322 wakeup(sd); 2323 } 2324 2325 void 2326 sr_scsi_cmd(struct scsi_xfer *xs) 2327 { 2328 struct scsi_link *link = xs->sc_link; 2329 struct sr_softc *sc = link->adapter_softc; 2330 struct sr_workunit *wu = xs->io; 2331 struct sr_discipline *sd; 2332 2333 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd target %d xs %p flags %#x\n", 2334 DEVNAME(sc), link->target, xs, xs->flags); 2335 2336 sd = sc->sc_targets[link->target]; 2337 if (sd == NULL) 2338 panic("%s: sr_scsi_cmd NULL discipline", DEVNAME(sc)); 2339 2340 if (sd->sd_deleted) { 2341 printf("%s: %s device is being deleted, failing io\n", 2342 DEVNAME(sc), sd->sd_meta->ssd_devname); 2343 goto stuffup; 2344 } 2345 2346 /* scsi layer *can* re-send wu without calling sr_wu_put(). */ 2347 sr_wu_release_ccbs(wu); 2348 sr_wu_init(sd, wu); 2349 wu->swu_state = SR_WU_INPROGRESS; 2350 wu->swu_xs = xs; 2351 2352 switch (xs->cmd->opcode) { 2353 case READ_COMMAND: 2354 case READ_BIG: 2355 case READ_16: 2356 case WRITE_COMMAND: 2357 case WRITE_BIG: 2358 case WRITE_16: 2359 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: READ/WRITE %02x\n", 2360 DEVNAME(sc), xs->cmd->opcode); 2361 if (sd->sd_scsi_rw(wu)) 2362 goto stuffup; 2363 break; 2364 2365 case SYNCHRONIZE_CACHE: 2366 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: SYNCHRONIZE_CACHE\n", 2367 DEVNAME(sc)); 2368 if (sd->sd_scsi_sync(wu)) 2369 goto stuffup; 2370 goto complete; 2371 2372 case TEST_UNIT_READY: 2373 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: TEST_UNIT_READY\n", 2374 DEVNAME(sc)); 2375 if (sd->sd_scsi_tur(wu)) 2376 goto stuffup; 2377 goto complete; 2378 2379 case START_STOP: 2380 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: START_STOP\n", 2381 DEVNAME(sc)); 2382 if (sd->sd_scsi_start_stop(wu)) 2383 goto stuffup; 2384 goto complete; 2385 2386 case INQUIRY: 2387 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: INQUIRY\n", 2388 DEVNAME(sc)); 2389 if (sd->sd_scsi_inquiry(wu)) 2390 goto stuffup; 2391 goto complete; 2392 2393 case READ_CAPACITY: 2394 case READ_CAPACITY_16: 2395 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd READ CAPACITY 0x%02x\n", 2396 DEVNAME(sc), xs->cmd->opcode); 2397 if (sd->sd_scsi_read_cap(wu)) 2398 goto stuffup; 2399 goto complete; 2400 2401 case REQUEST_SENSE: 2402 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd REQUEST SENSE\n", 2403 DEVNAME(sc)); 2404 if (sd->sd_scsi_req_sense(wu)) 2405 goto stuffup; 2406 goto complete; 2407 2408 default: 2409 DNPRINTF(SR_D_CMD, "%s: unsupported scsi command %x\n", 2410 DEVNAME(sc), xs->cmd->opcode); 2411 /* XXX might need to add generic function to handle others */ 2412 goto stuffup; 2413 } 2414 2415 return; 2416 stuffup: 2417 if (sd->sd_scsi_sense.error_code) { 2418 xs->error = XS_SENSE; 2419 memcpy(&xs->sense, &sd->sd_scsi_sense, sizeof(xs->sense)); 2420 bzero(&sd->sd_scsi_sense, sizeof(sd->sd_scsi_sense)); 2421 } else { 2422 xs->error = XS_DRIVER_STUFFUP; 2423 } 2424 complete: 2425 sr_scsi_done(sd, xs); 2426 } 2427 2428 int 2429 sr_scsi_probe(struct scsi_link *link) 2430 { 2431 struct sr_softc *sc = link->adapter_softc; 2432 struct sr_discipline *sd; 2433 2434 KASSERT(link->target < SR_MAX_LD && link->lun == 0); 2435 2436 sd = sc->sc_targets[link->target]; 2437 if (sd == NULL) 2438 return (ENODEV); 2439 2440 link->pool = &sd->sd_iopool; 2441 if (sd->sd_openings) 2442 link->openings = sd->sd_openings(sd); 2443 else 2444 link->openings = sd->sd_max_wu; 2445 2446 return (0); 2447 } 2448 2449 int 2450 sr_scsi_ioctl(struct scsi_link *link, u_long cmd, caddr_t addr, int flag) 2451 { 2452 struct sr_softc *sc = link->adapter_softc; 2453 struct sr_discipline *sd; 2454 2455 sd = sc->sc_targets[link->target]; 2456 if (sd == NULL) 2457 return (ENODEV); 2458 2459 DNPRINTF(SR_D_IOCTL, "%s: %s sr_scsi_ioctl cmd: %#lx\n", 2460 DEVNAME(sc), sd->sd_meta->ssd_devname, cmd); 2461 2462 /* Pass bio ioctls through to the bio handler. */ 2463 if (IOCGROUP(cmd) == 'B') 2464 return (sr_bio_handler(sc, sd, cmd, (struct bio *)addr)); 2465 2466 switch (cmd) { 2467 case DIOCGCACHE: 2468 case DIOCSCACHE: 2469 return (EOPNOTSUPP); 2470 default: 2471 return (ENOTTY); 2472 } 2473 } 2474 2475 int 2476 sr_bio_ioctl(struct device *dev, u_long cmd, caddr_t addr) 2477 { 2478 struct sr_softc *sc = (struct sr_softc *) dev; 2479 DNPRINTF(SR_D_IOCTL, "%s: sr_bio_ioctl\n", DEVNAME(sc)); 2480 2481 return sr_bio_handler(sc, NULL, cmd, (struct bio *)addr); 2482 } 2483 2484 int 2485 sr_bio_handler(struct sr_softc *sc, struct sr_discipline *sd, u_long cmd, 2486 struct bio *bio) 2487 { 2488 int rv = 0; 2489 2490 DNPRINTF(SR_D_IOCTL, "%s: sr_bio_handler ", DEVNAME(sc)); 2491 2492 rw_enter_write(&sc->sc_lock); 2493 2494 bio_status_init(&sc->sc_status, &sc->sc_dev); 2495 2496 switch (cmd) { 2497 case BIOCINQ: 2498 DNPRINTF(SR_D_IOCTL, "inq\n"); 2499 rv = sr_ioctl_inq(sc, (struct bioc_inq *)bio); 2500 break; 2501 2502 case BIOCVOL: 2503 DNPRINTF(SR_D_IOCTL, "vol\n"); 2504 rv = sr_ioctl_vol(sc, (struct bioc_vol *)bio); 2505 break; 2506 2507 case BIOCDISK: 2508 DNPRINTF(SR_D_IOCTL, "disk\n"); 2509 rv = sr_ioctl_disk(sc, (struct bioc_disk *)bio); 2510 break; 2511 2512 case BIOCALARM: 2513 DNPRINTF(SR_D_IOCTL, "alarm\n"); 2514 /*rv = sr_ioctl_alarm(sc, (struct bioc_alarm *)bio); */ 2515 break; 2516 2517 case BIOCBLINK: 2518 DNPRINTF(SR_D_IOCTL, "blink\n"); 2519 /*rv = sr_ioctl_blink(sc, (struct bioc_blink *)bio); */ 2520 break; 2521 2522 case BIOCSETSTATE: 2523 DNPRINTF(SR_D_IOCTL, "setstate\n"); 2524 rv = sr_ioctl_setstate(sc, (struct bioc_setstate *)bio); 2525 break; 2526 2527 case BIOCCREATERAID: 2528 DNPRINTF(SR_D_IOCTL, "createraid\n"); 2529 rv = sr_ioctl_createraid(sc, (struct bioc_createraid *)bio, 2530 1, NULL); 2531 break; 2532 2533 case BIOCDELETERAID: 2534 DNPRINTF(SR_D_IOCTL, "deleteraid\n"); 2535 rv = sr_ioctl_deleteraid(sc, sd, (struct bioc_deleteraid *)bio); 2536 break; 2537 2538 case BIOCDISCIPLINE: 2539 DNPRINTF(SR_D_IOCTL, "discipline\n"); 2540 rv = sr_ioctl_discipline(sc, sd, (struct bioc_discipline *)bio); 2541 break; 2542 2543 case BIOCINSTALLBOOT: 2544 DNPRINTF(SR_D_IOCTL, "installboot\n"); 2545 rv = sr_ioctl_installboot(sc, sd, 2546 (struct bioc_installboot *)bio); 2547 break; 2548 2549 default: 2550 DNPRINTF(SR_D_IOCTL, "invalid ioctl\n"); 2551 rv = ENOTTY; 2552 } 2553 2554 sc->sc_status.bs_status = (rv ? BIO_STATUS_ERROR : BIO_STATUS_SUCCESS); 2555 2556 if (sc->sc_status.bs_msg_count > 0) 2557 rv = 0; 2558 2559 memcpy(&bio->bio_status, &sc->sc_status, sizeof(struct bio_status)); 2560 2561 rw_exit_write(&sc->sc_lock); 2562 2563 return (rv); 2564 } 2565 2566 int 2567 sr_ioctl_inq(struct sr_softc *sc, struct bioc_inq *bi) 2568 { 2569 struct sr_discipline *sd; 2570 int vol = 0, disk = 0; 2571 2572 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2573 vol++; 2574 disk += sd->sd_meta->ssdi.ssd_chunk_no; 2575 } 2576 2577 strlcpy(bi->bi_dev, sc->sc_dev.dv_xname, sizeof(bi->bi_dev)); 2578 bi->bi_novol = vol + sc->sc_hotspare_no; 2579 bi->bi_nodisk = disk + sc->sc_hotspare_no; 2580 2581 return (0); 2582 } 2583 2584 int 2585 sr_ioctl_vol(struct sr_softc *sc, struct bioc_vol *bv) 2586 { 2587 int vol = -1, rv = EINVAL; 2588 struct sr_discipline *sd; 2589 struct sr_chunk *hotspare; 2590 2591 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2592 vol++; 2593 if (vol != bv->bv_volid) 2594 continue; 2595 2596 bv->bv_status = sd->sd_vol_status; 2597 bv->bv_size = sd->sd_meta->ssdi.ssd_size << DEV_BSHIFT; 2598 bv->bv_level = sd->sd_meta->ssdi.ssd_level; 2599 bv->bv_nodisk = sd->sd_meta->ssdi.ssd_chunk_no; 2600 2601 #ifdef CRYPTO 2602 if (sd->sd_meta->ssdi.ssd_level == 'C' && 2603 sd->mds.mdd_crypto.key_disk != NULL) 2604 bv->bv_nodisk++; 2605 #endif 2606 if (bv->bv_status == BIOC_SVREBUILD) 2607 bv->bv_percent = sr_rebuild_percent(sd); 2608 2609 strlcpy(bv->bv_dev, sd->sd_meta->ssd_devname, 2610 sizeof(bv->bv_dev)); 2611 strlcpy(bv->bv_vendor, sd->sd_meta->ssdi.ssd_vendor, 2612 sizeof(bv->bv_vendor)); 2613 rv = 0; 2614 goto done; 2615 } 2616 2617 /* Check hotspares list. */ 2618 SLIST_FOREACH(hotspare, &sc->sc_hotspare_list, src_link) { 2619 vol++; 2620 if (vol != bv->bv_volid) 2621 continue; 2622 2623 bv->bv_status = BIOC_SVONLINE; 2624 bv->bv_size = hotspare->src_meta.scmi.scm_size << DEV_BSHIFT; 2625 bv->bv_level = -1; /* Hotspare. */ 2626 bv->bv_nodisk = 1; 2627 strlcpy(bv->bv_dev, hotspare->src_meta.scmi.scm_devname, 2628 sizeof(bv->bv_dev)); 2629 strlcpy(bv->bv_vendor, hotspare->src_meta.scmi.scm_devname, 2630 sizeof(bv->bv_vendor)); 2631 rv = 0; 2632 goto done; 2633 } 2634 2635 done: 2636 return (rv); 2637 } 2638 2639 int 2640 sr_ioctl_disk(struct sr_softc *sc, struct bioc_disk *bd) 2641 { 2642 struct sr_discipline *sd; 2643 struct sr_chunk *src, *hotspare; 2644 int vol = -1, rv = EINVAL; 2645 2646 if (bd->bd_diskid < 0) 2647 goto done; 2648 2649 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2650 vol++; 2651 if (vol != bd->bd_volid) 2652 continue; 2653 2654 if (bd->bd_diskid < sd->sd_meta->ssdi.ssd_chunk_no) 2655 src = sd->sd_vol.sv_chunks[bd->bd_diskid]; 2656 #ifdef CRYPTO 2657 else if (bd->bd_diskid == sd->sd_meta->ssdi.ssd_chunk_no && 2658 sd->sd_meta->ssdi.ssd_level == 'C' && 2659 sd->mds.mdd_crypto.key_disk != NULL) 2660 src = sd->mds.mdd_crypto.key_disk; 2661 #endif 2662 else 2663 break; 2664 2665 bd->bd_status = src->src_meta.scm_status; 2666 bd->bd_size = src->src_meta.scmi.scm_size << DEV_BSHIFT; 2667 bd->bd_channel = vol; 2668 bd->bd_target = bd->bd_diskid; 2669 strlcpy(bd->bd_vendor, src->src_meta.scmi.scm_devname, 2670 sizeof(bd->bd_vendor)); 2671 rv = 0; 2672 goto done; 2673 } 2674 2675 /* Check hotspares list. */ 2676 SLIST_FOREACH(hotspare, &sc->sc_hotspare_list, src_link) { 2677 vol++; 2678 if (vol != bd->bd_volid) 2679 continue; 2680 2681 if (bd->bd_diskid != 0) 2682 break; 2683 2684 bd->bd_status = hotspare->src_meta.scm_status; 2685 bd->bd_size = hotspare->src_meta.scmi.scm_size << DEV_BSHIFT; 2686 bd->bd_channel = vol; 2687 bd->bd_target = bd->bd_diskid; 2688 strlcpy(bd->bd_vendor, hotspare->src_meta.scmi.scm_devname, 2689 sizeof(bd->bd_vendor)); 2690 rv = 0; 2691 goto done; 2692 } 2693 2694 done: 2695 return (rv); 2696 } 2697 2698 int 2699 sr_ioctl_setstate(struct sr_softc *sc, struct bioc_setstate *bs) 2700 { 2701 int rv = EINVAL; 2702 int vol = -1, found, c; 2703 struct sr_discipline *sd; 2704 struct sr_chunk *ch_entry; 2705 struct sr_chunk_head *cl; 2706 2707 if (bs->bs_other_id_type == BIOC_SSOTHER_UNUSED) 2708 goto done; 2709 2710 if (bs->bs_status == BIOC_SSHOTSPARE) { 2711 rv = sr_hotspare(sc, (dev_t)bs->bs_other_id); 2712 goto done; 2713 } 2714 2715 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2716 vol++; 2717 if (vol == bs->bs_volid) 2718 break; 2719 } 2720 if (sd == NULL) 2721 goto done; 2722 2723 switch (bs->bs_status) { 2724 case BIOC_SSOFFLINE: 2725 /* Take chunk offline */ 2726 found = c = 0; 2727 cl = &sd->sd_vol.sv_chunk_list; 2728 SLIST_FOREACH(ch_entry, cl, src_link) { 2729 if (ch_entry->src_dev_mm == bs->bs_other_id) { 2730 found = 1; 2731 break; 2732 } 2733 c++; 2734 } 2735 if (found == 0) { 2736 sr_error(sc, "chunk not part of array"); 2737 goto done; 2738 } 2739 2740 /* XXX: check current state first */ 2741 sd->sd_set_chunk_state(sd, c, BIOC_SDOFFLINE); 2742 2743 if (sr_meta_save(sd, SR_META_DIRTY)) { 2744 sr_error(sc, "could not save metadata for %s", 2745 sd->sd_meta->ssd_devname); 2746 goto done; 2747 } 2748 rv = 0; 2749 break; 2750 2751 case BIOC_SDSCRUB: 2752 break; 2753 2754 case BIOC_SSREBUILD: 2755 rv = sr_rebuild_init(sd, (dev_t)bs->bs_other_id, 0); 2756 break; 2757 2758 default: 2759 sr_error(sc, "unsupported state request %d", bs->bs_status); 2760 } 2761 2762 done: 2763 return (rv); 2764 } 2765 2766 int 2767 sr_chunk_in_use(struct sr_softc *sc, dev_t dev) 2768 { 2769 struct sr_discipline *sd; 2770 struct sr_chunk *chunk; 2771 int i; 2772 2773 DNPRINTF(SR_D_MISC, "%s: sr_chunk_in_use(%d)\n", DEVNAME(sc), dev); 2774 2775 if (dev == NODEV) 2776 return BIOC_SDINVALID; 2777 2778 /* See if chunk is already in use. */ 2779 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2780 for (i = 0; i < sd->sd_meta->ssdi.ssd_chunk_no; i++) { 2781 chunk = sd->sd_vol.sv_chunks[i]; 2782 if (chunk->src_dev_mm == dev) 2783 return chunk->src_meta.scm_status; 2784 } 2785 } 2786 2787 /* Check hotspares list. */ 2788 SLIST_FOREACH(chunk, &sc->sc_hotspare_list, src_link) 2789 if (chunk->src_dev_mm == dev) 2790 return chunk->src_meta.scm_status; 2791 2792 return BIOC_SDINVALID; 2793 } 2794 2795 int 2796 sr_hotspare(struct sr_softc *sc, dev_t dev) 2797 { 2798 struct sr_discipline *sd = NULL; 2799 struct sr_metadata *sm = NULL; 2800 struct sr_meta_chunk *hm; 2801 struct sr_chunk_head *cl; 2802 struct sr_chunk *chunk, *last, *hotspare = NULL; 2803 struct sr_uuid uuid; 2804 struct disklabel label; 2805 struct vnode *vn; 2806 u_int64_t size; 2807 char devname[32]; 2808 int rv = EINVAL; 2809 int c, part, open = 0; 2810 2811 /* 2812 * Add device to global hotspares list. 2813 */ 2814 2815 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 2816 2817 /* Make sure chunk is not already in use. */ 2818 c = sr_chunk_in_use(sc, dev); 2819 if (c != BIOC_SDINVALID && c != BIOC_SDOFFLINE) { 2820 if (c == BIOC_SDHOTSPARE) 2821 sr_error(sc, "%s is already a hotspare", devname); 2822 else 2823 sr_error(sc, "%s is already in use", devname); 2824 goto done; 2825 } 2826 2827 /* XXX - See if there is an existing degraded volume... */ 2828 2829 /* Open device. */ 2830 if (bdevvp(dev, &vn)) { 2831 sr_error(sc, "sr_hotspare: cannot allocate vnode"); 2832 goto done; 2833 } 2834 if (VOP_OPEN(vn, FREAD | FWRITE, NOCRED, curproc)) { 2835 DNPRINTF(SR_D_META,"%s: sr_hotspare cannot open %s\n", 2836 DEVNAME(sc), devname); 2837 vput(vn); 2838 goto fail; 2839 } 2840 open = 1; /* close dev on error */ 2841 2842 /* Get partition details. */ 2843 part = DISKPART(dev); 2844 if (VOP_IOCTL(vn, DIOCGDINFO, (caddr_t)&label, FREAD, 2845 NOCRED, curproc)) { 2846 DNPRINTF(SR_D_META, "%s: sr_hotspare ioctl failed\n", 2847 DEVNAME(sc)); 2848 goto fail; 2849 } 2850 if (label.d_partitions[part].p_fstype != FS_RAID) { 2851 sr_error(sc, "%s partition not of type RAID (%d)", 2852 devname, label.d_partitions[part].p_fstype); 2853 goto fail; 2854 } 2855 2856 /* Calculate partition size. */ 2857 size = DL_SECTOBLK(&label, DL_GETPSIZE(&label.d_partitions[part])); 2858 if (size <= SR_DATA_OFFSET) { 2859 DNPRINTF(SR_D_META, "%s: %s partition too small\n", DEVNAME(sc), 2860 devname); 2861 goto fail; 2862 } 2863 size -= SR_DATA_OFFSET; 2864 if (size > INT64_MAX) { 2865 DNPRINTF(SR_D_META, "%s: %s partition too large\n", DEVNAME(sc), 2866 devname); 2867 goto fail; 2868 } 2869 2870 /* 2871 * Create and populate chunk metadata. 2872 */ 2873 2874 sr_uuid_generate(&uuid); 2875 hotspare = malloc(sizeof(struct sr_chunk), M_DEVBUF, M_WAITOK | M_ZERO); 2876 2877 hotspare->src_dev_mm = dev; 2878 hotspare->src_vn = vn; 2879 strlcpy(hotspare->src_devname, devname, sizeof(hm->scmi.scm_devname)); 2880 hotspare->src_size = size; 2881 2882 hm = &hotspare->src_meta; 2883 hm->scmi.scm_volid = SR_HOTSPARE_VOLID; 2884 hm->scmi.scm_chunk_id = 0; 2885 hm->scmi.scm_size = size; 2886 hm->scmi.scm_coerced_size = size; 2887 strlcpy(hm->scmi.scm_devname, devname, sizeof(hm->scmi.scm_devname)); 2888 memcpy(&hm->scmi.scm_uuid, &uuid, sizeof(struct sr_uuid)); 2889 2890 sr_checksum(sc, hm, &hm->scm_checksum, 2891 sizeof(struct sr_meta_chunk_invariant)); 2892 2893 hm->scm_status = BIOC_SDHOTSPARE; 2894 2895 /* 2896 * Create and populate our own discipline and metadata. 2897 */ 2898 2899 sm = malloc(sizeof(struct sr_metadata), M_DEVBUF, M_WAITOK | M_ZERO); 2900 sm->ssdi.ssd_magic = SR_MAGIC; 2901 sm->ssdi.ssd_version = SR_META_VERSION; 2902 sm->ssd_ondisk = 0; 2903 sm->ssdi.ssd_vol_flags = 0; 2904 memcpy(&sm->ssdi.ssd_uuid, &uuid, sizeof(struct sr_uuid)); 2905 sm->ssdi.ssd_chunk_no = 1; 2906 sm->ssdi.ssd_volid = SR_HOTSPARE_VOLID; 2907 sm->ssdi.ssd_level = SR_HOTSPARE_LEVEL; 2908 sm->ssdi.ssd_size = size; 2909 sm->ssdi.ssd_secsize = label.d_secsize; 2910 strlcpy(sm->ssdi.ssd_vendor, "OPENBSD", sizeof(sm->ssdi.ssd_vendor)); 2911 snprintf(sm->ssdi.ssd_product, sizeof(sm->ssdi.ssd_product), 2912 "SR %s", "HOTSPARE"); 2913 snprintf(sm->ssdi.ssd_revision, sizeof(sm->ssdi.ssd_revision), 2914 "%03d", SR_META_VERSION); 2915 2916 sd = malloc(sizeof(struct sr_discipline), M_DEVBUF, M_WAITOK | M_ZERO); 2917 sd->sd_sc = sc; 2918 sd->sd_meta = sm; 2919 sd->sd_meta_type = SR_META_F_NATIVE; 2920 sd->sd_vol_status = BIOC_SVONLINE; 2921 strlcpy(sd->sd_name, "HOTSPARE", sizeof(sd->sd_name)); 2922 SLIST_INIT(&sd->sd_meta_opt); 2923 2924 /* Add chunk to volume. */ 2925 sd->sd_vol.sv_chunks = malloc(sizeof(struct sr_chunk *), M_DEVBUF, 2926 M_WAITOK | M_ZERO); 2927 sd->sd_vol.sv_chunks[0] = hotspare; 2928 SLIST_INIT(&sd->sd_vol.sv_chunk_list); 2929 SLIST_INSERT_HEAD(&sd->sd_vol.sv_chunk_list, hotspare, src_link); 2930 2931 /* Save metadata. */ 2932 if (sr_meta_save(sd, SR_META_DIRTY)) { 2933 sr_error(sc, "could not save metadata to %s", devname); 2934 goto fail; 2935 } 2936 2937 /* 2938 * Add chunk to hotspare list. 2939 */ 2940 rw_enter_write(&sc->sc_hs_lock); 2941 cl = &sc->sc_hotspare_list; 2942 if (SLIST_EMPTY(cl)) 2943 SLIST_INSERT_HEAD(cl, hotspare, src_link); 2944 else { 2945 SLIST_FOREACH(chunk, cl, src_link) 2946 last = chunk; 2947 SLIST_INSERT_AFTER(last, hotspare, src_link); 2948 } 2949 sc->sc_hotspare_no++; 2950 rw_exit_write(&sc->sc_hs_lock); 2951 2952 rv = 0; 2953 goto done; 2954 2955 fail: 2956 free(hotspare, M_DEVBUF, sizeof(*hotspare)); 2957 2958 done: 2959 if (sd) 2960 free(sd->sd_vol.sv_chunks, M_DEVBUF, 2961 sizeof(sd->sd_vol.sv_chunks)); 2962 free(sd, M_DEVBUF, sizeof(*sd)); 2963 free(sm, M_DEVBUF, sizeof(*sm)); 2964 if (open) { 2965 VOP_CLOSE(vn, FREAD | FWRITE, NOCRED, curproc); 2966 vput(vn); 2967 } 2968 2969 return (rv); 2970 } 2971 2972 void 2973 sr_hotspare_rebuild_callback(void *xsd) 2974 { 2975 struct sr_discipline *sd = xsd; 2976 sr_hotspare_rebuild(sd); 2977 } 2978 2979 void 2980 sr_hotspare_rebuild(struct sr_discipline *sd) 2981 { 2982 struct sr_softc *sc = sd->sd_sc; 2983 struct sr_chunk_head *cl; 2984 struct sr_chunk *hotspare, *chunk = NULL; 2985 struct sr_workunit *wu; 2986 struct sr_ccb *ccb; 2987 int i, s, cid, busy; 2988 2989 /* 2990 * Attempt to locate a hotspare and initiate rebuild. 2991 */ 2992 2993 /* Find first offline chunk. */ 2994 for (cid = 0; cid < sd->sd_meta->ssdi.ssd_chunk_no; cid++) { 2995 if (sd->sd_vol.sv_chunks[cid]->src_meta.scm_status == 2996 BIOC_SDOFFLINE) { 2997 chunk = sd->sd_vol.sv_chunks[cid]; 2998 break; 2999 } 3000 } 3001 if (chunk == NULL) { 3002 printf("%s: no offline chunk found on %s!\n", 3003 DEVNAME(sc), sd->sd_meta->ssd_devname); 3004 return; 3005 } 3006 3007 /* See if we have a suitable hotspare... */ 3008 rw_enter_write(&sc->sc_hs_lock); 3009 cl = &sc->sc_hotspare_list; 3010 SLIST_FOREACH(hotspare, cl, src_link) 3011 if (hotspare->src_size >= chunk->src_size && 3012 hotspare->src_secsize <= sd->sd_meta->ssdi.ssd_secsize) 3013 break; 3014 3015 if (hotspare != NULL) { 3016 3017 printf("%s: %s volume degraded, will attempt to " 3018 "rebuild on hotspare %s\n", DEVNAME(sc), 3019 sd->sd_meta->ssd_devname, hotspare->src_devname); 3020 3021 /* 3022 * Ensure that all pending I/O completes on the failed chunk 3023 * before trying to initiate a rebuild. 3024 */ 3025 i = 0; 3026 do { 3027 busy = 0; 3028 3029 s = splbio(); 3030 TAILQ_FOREACH(wu, &sd->sd_wu_pendq, swu_link) { 3031 TAILQ_FOREACH(ccb, &wu->swu_ccb, ccb_link) { 3032 if (ccb->ccb_target == cid) 3033 busy = 1; 3034 } 3035 } 3036 TAILQ_FOREACH(wu, &sd->sd_wu_defq, swu_link) { 3037 TAILQ_FOREACH(ccb, &wu->swu_ccb, ccb_link) { 3038 if (ccb->ccb_target == cid) 3039 busy = 1; 3040 } 3041 } 3042 splx(s); 3043 3044 if (busy) { 3045 tsleep_nsec(sd, PRIBIO, "sr_hotspare", 3046 SEC_TO_NSEC(1)); 3047 i++; 3048 } 3049 3050 } while (busy && i < 120); 3051 3052 DNPRINTF(SR_D_META, "%s: waited %i seconds for I/O to " 3053 "complete on failed chunk %s\n", DEVNAME(sc), 3054 i, chunk->src_devname); 3055 3056 if (busy) { 3057 printf("%s: pending I/O failed to complete on " 3058 "failed chunk %s, hotspare rebuild aborted...\n", 3059 DEVNAME(sc), chunk->src_devname); 3060 goto done; 3061 } 3062 3063 s = splbio(); 3064 rw_enter_write(&sc->sc_lock); 3065 bio_status_init(&sc->sc_status, &sc->sc_dev); 3066 if (sr_rebuild_init(sd, hotspare->src_dev_mm, 1) == 0) { 3067 3068 /* Remove hotspare from available list. */ 3069 sc->sc_hotspare_no--; 3070 SLIST_REMOVE(cl, hotspare, sr_chunk, src_link); 3071 free(hotspare, M_DEVBUF, sizeof(*hotspare)); 3072 3073 } 3074 rw_exit_write(&sc->sc_lock); 3075 splx(s); 3076 } 3077 done: 3078 rw_exit_write(&sc->sc_hs_lock); 3079 } 3080 3081 int 3082 sr_rebuild_init(struct sr_discipline *sd, dev_t dev, int hotspare) 3083 { 3084 struct sr_softc *sc = sd->sd_sc; 3085 struct sr_chunk *chunk = NULL; 3086 struct sr_meta_chunk *meta; 3087 struct disklabel label; 3088 struct vnode *vn; 3089 u_int64_t size; 3090 int64_t csize; 3091 char devname[32]; 3092 int rv = EINVAL, open = 0; 3093 int cid, i, part, status; 3094 3095 /* 3096 * Attempt to initiate a rebuild onto the specified device. 3097 */ 3098 3099 if (!(sd->sd_capabilities & SR_CAP_REBUILD)) { 3100 sr_error(sc, "discipline does not support rebuild"); 3101 goto done; 3102 } 3103 3104 /* make sure volume is in the right state */ 3105 if (sd->sd_vol_status == BIOC_SVREBUILD) { 3106 sr_error(sc, "rebuild already in progress"); 3107 goto done; 3108 } 3109 if (sd->sd_vol_status != BIOC_SVDEGRADED) { 3110 sr_error(sc, "volume not degraded"); 3111 goto done; 3112 } 3113 3114 /* Find first offline chunk. */ 3115 for (cid = 0; cid < sd->sd_meta->ssdi.ssd_chunk_no; cid++) { 3116 if (sd->sd_vol.sv_chunks[cid]->src_meta.scm_status == 3117 BIOC_SDOFFLINE) { 3118 chunk = sd->sd_vol.sv_chunks[cid]; 3119 break; 3120 } 3121 } 3122 if (chunk == NULL) { 3123 sr_error(sc, "no offline chunks available to rebuild"); 3124 goto done; 3125 } 3126 3127 /* Get coerced size from another online chunk. */ 3128 csize = 0; 3129 for (i = 0; i < sd->sd_meta->ssdi.ssd_chunk_no; i++) { 3130 if (sd->sd_vol.sv_chunks[i]->src_meta.scm_status == 3131 BIOC_SDONLINE) { 3132 meta = &sd->sd_vol.sv_chunks[i]->src_meta; 3133 csize = meta->scmi.scm_coerced_size; 3134 break; 3135 } 3136 } 3137 if (csize == 0) { 3138 sr_error(sc, "no online chunks available for rebuild"); 3139 goto done; 3140 } 3141 3142 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 3143 if (bdevvp(dev, &vn)) { 3144 printf("%s: sr_rebuild_init: can't allocate vnode\n", 3145 DEVNAME(sc)); 3146 goto done; 3147 } 3148 if (VOP_OPEN(vn, FREAD | FWRITE, NOCRED, curproc)) { 3149 DNPRINTF(SR_D_META,"%s: sr_ioctl_setstate can't " 3150 "open %s\n", DEVNAME(sc), devname); 3151 vput(vn); 3152 goto done; 3153 } 3154 open = 1; /* close dev on error */ 3155 3156 /* Get disklabel and check partition. */ 3157 part = DISKPART(dev); 3158 if (VOP_IOCTL(vn, DIOCGDINFO, (caddr_t)&label, FREAD, 3159 NOCRED, curproc)) { 3160 DNPRINTF(SR_D_META, "%s: sr_ioctl_setstate ioctl failed\n", 3161 DEVNAME(sc)); 3162 goto done; 3163 } 3164 if (label.d_partitions[part].p_fstype != FS_RAID) { 3165 sr_error(sc, "%s partition not of type RAID (%d)", 3166 devname, label.d_partitions[part].p_fstype); 3167 goto done; 3168 } 3169 3170 /* Is the partition large enough? */ 3171 size = DL_SECTOBLK(&label, DL_GETPSIZE(&label.d_partitions[part])); 3172 if (size <= sd->sd_meta->ssd_data_blkno) { 3173 sr_error(sc, "%s: %s partition too small", DEVNAME(sc), 3174 devname); 3175 goto done; 3176 } 3177 size -= sd->sd_meta->ssd_data_blkno; 3178 if (size > INT64_MAX) { 3179 sr_error(sc, "%s: %s partition too large", DEVNAME(sc), 3180 devname); 3181 goto done; 3182 } 3183 if (size < csize) { 3184 sr_error(sc, "%s partition too small, at least %lld bytes " 3185 "required", devname, (long long)(csize << DEV_BSHIFT)); 3186 goto done; 3187 } else if (size > csize) 3188 sr_warn(sc, "%s partition too large, wasting %lld bytes", 3189 devname, (long long)((size - csize) << DEV_BSHIFT)); 3190 if (label.d_secsize > sd->sd_meta->ssdi.ssd_secsize) { 3191 sr_error(sc, "%s sector size too large, <= %u bytes " 3192 "required", devname, sd->sd_meta->ssdi.ssd_secsize); 3193 goto done; 3194 } 3195 3196 /* Ensure that this chunk is not already in use. */ 3197 status = sr_chunk_in_use(sc, dev); 3198 if (status != BIOC_SDINVALID && status != BIOC_SDOFFLINE && 3199 !(hotspare && status == BIOC_SDHOTSPARE)) { 3200 sr_error(sc, "%s is already in use", devname); 3201 goto done; 3202 } 3203 3204 /* Reset rebuild counter since we rebuilding onto a new chunk. */ 3205 sd->sd_meta->ssd_rebuild = 0; 3206 3207 open = 0; /* leave dev open from here on out */ 3208 3209 /* Fix up chunk. */ 3210 memcpy(chunk->src_duid, label.d_uid, sizeof(chunk->src_duid)); 3211 chunk->src_dev_mm = dev; 3212 chunk->src_vn = vn; 3213 3214 /* Reconstruct metadata. */ 3215 meta = &chunk->src_meta; 3216 meta->scmi.scm_volid = sd->sd_meta->ssdi.ssd_volid; 3217 meta->scmi.scm_chunk_id = cid; 3218 strlcpy(meta->scmi.scm_devname, devname, 3219 sizeof(meta->scmi.scm_devname)); 3220 meta->scmi.scm_size = size; 3221 meta->scmi.scm_coerced_size = csize; 3222 memcpy(&meta->scmi.scm_uuid, &sd->sd_meta->ssdi.ssd_uuid, 3223 sizeof(meta->scmi.scm_uuid)); 3224 sr_checksum(sc, meta, &meta->scm_checksum, 3225 sizeof(struct sr_meta_chunk_invariant)); 3226 3227 sd->sd_set_chunk_state(sd, cid, BIOC_SDREBUILD); 3228 3229 if (sr_meta_save(sd, SR_META_DIRTY)) { 3230 sr_error(sc, "could not save metadata to %s", devname); 3231 open = 1; 3232 goto done; 3233 } 3234 3235 sr_warn(sc, "rebuild of %s started on %s", 3236 sd->sd_meta->ssd_devname, devname); 3237 3238 sd->sd_reb_abort = 0; 3239 kthread_create_deferred(sr_rebuild_start, sd); 3240 3241 rv = 0; 3242 done: 3243 if (open) { 3244 VOP_CLOSE(vn, FREAD | FWRITE, NOCRED, curproc); 3245 vput(vn); 3246 } 3247 3248 return (rv); 3249 } 3250 3251 int 3252 sr_rebuild_percent(struct sr_discipline *sd) 3253 { 3254 daddr_t rb, sz; 3255 3256 sz = sd->sd_meta->ssdi.ssd_size; 3257 rb = sd->sd_meta->ssd_rebuild; 3258 3259 if (rb > 0) 3260 return (100 - ((sz * 100 - rb * 100) / sz) - 1); 3261 3262 return (0); 3263 } 3264 3265 void 3266 sr_roam_chunks(struct sr_discipline *sd) 3267 { 3268 struct sr_softc *sc = sd->sd_sc; 3269 struct sr_chunk *chunk; 3270 struct sr_meta_chunk *meta; 3271 int roamed = 0; 3272 3273 /* Have any chunks roamed? */ 3274 SLIST_FOREACH(chunk, &sd->sd_vol.sv_chunk_list, src_link) { 3275 meta = &chunk->src_meta; 3276 if (strncmp(meta->scmi.scm_devname, chunk->src_devname, 3277 sizeof(meta->scmi.scm_devname))) { 3278 3279 printf("%s: roaming device %s -> %s\n", DEVNAME(sc), 3280 meta->scmi.scm_devname, chunk->src_devname); 3281 3282 strlcpy(meta->scmi.scm_devname, chunk->src_devname, 3283 sizeof(meta->scmi.scm_devname)); 3284 3285 roamed++; 3286 } 3287 } 3288 3289 if (roamed) 3290 sr_meta_save(sd, SR_META_DIRTY); 3291 } 3292 3293 int 3294 sr_ioctl_createraid(struct sr_softc *sc, struct bioc_createraid *bc, 3295 int user, void *data) 3296 { 3297 struct sr_meta_opt_item *omi; 3298 struct sr_chunk_head *cl; 3299 struct sr_discipline *sd = NULL; 3300 struct sr_chunk *ch_entry; 3301 struct scsi_link *link; 3302 struct device *dev; 3303 char *uuid, devname[32]; 3304 dev_t *dt = NULL; 3305 int i, no_chunk, rv = EINVAL, target, vol; 3306 int no_meta; 3307 3308 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_createraid(%d)\n", 3309 DEVNAME(sc), user); 3310 3311 /* user input */ 3312 if (bc->bc_dev_list_len > BIOC_CRMAXLEN) 3313 goto unwind; 3314 3315 dt = malloc(bc->bc_dev_list_len, M_DEVBUF, M_WAITOK | M_ZERO); 3316 if (user) { 3317 if (copyin(bc->bc_dev_list, dt, bc->bc_dev_list_len) != 0) 3318 goto unwind; 3319 } else 3320 memcpy(dt, bc->bc_dev_list, bc->bc_dev_list_len); 3321 3322 /* Initialise discipline. */ 3323 sd = malloc(sizeof(struct sr_discipline), M_DEVBUF, M_WAITOK | M_ZERO); 3324 sd->sd_sc = sc; 3325 SLIST_INIT(&sd->sd_meta_opt); 3326 sd->sd_taskq = taskq_create("srdis", 1, IPL_BIO, 0); 3327 if (sd->sd_taskq == NULL) { 3328 sr_error(sc, "could not create discipline taskq"); 3329 goto unwind; 3330 } 3331 if (sr_discipline_init(sd, bc->bc_level)) { 3332 sr_error(sc, "could not initialize discipline"); 3333 goto unwind; 3334 } 3335 3336 no_chunk = bc->bc_dev_list_len / sizeof(dev_t); 3337 cl = &sd->sd_vol.sv_chunk_list; 3338 SLIST_INIT(cl); 3339 3340 /* Ensure that chunks are not already in use. */ 3341 for (i = 0; i < no_chunk; i++) { 3342 if (sr_chunk_in_use(sc, dt[i]) != BIOC_SDINVALID) { 3343 sr_meta_getdevname(sc, dt[i], devname, sizeof(devname)); 3344 sr_error(sc, "chunk %s already in use", devname); 3345 goto unwind; 3346 } 3347 } 3348 3349 sd->sd_meta_type = sr_meta_probe(sd, dt, no_chunk); 3350 if (sd->sd_meta_type == SR_META_F_INVALID) { 3351 sr_error(sc, "invalid metadata format"); 3352 goto unwind; 3353 } 3354 3355 if (sr_meta_attach(sd, no_chunk, bc->bc_flags & BIOC_SCFORCE)) 3356 goto unwind; 3357 3358 /* force the raid volume by clearing metadata region */ 3359 if (bc->bc_flags & BIOC_SCFORCE) { 3360 /* make sure disk isn't up and running */ 3361 if (sr_meta_read(sd)) 3362 if (sr_already_assembled(sd)) { 3363 uuid = sr_uuid_format( 3364 &sd->sd_meta->ssdi.ssd_uuid); 3365 sr_error(sc, "disk %s is currently in use; " 3366 "cannot force create", uuid); 3367 free(uuid, M_DEVBUF, 37); 3368 goto unwind; 3369 } 3370 3371 if (sr_meta_clear(sd)) { 3372 sr_error(sc, "failed to clear metadata"); 3373 goto unwind; 3374 } 3375 } 3376 3377 no_meta = sr_meta_read(sd); 3378 if (no_meta == -1) { 3379 3380 /* Corrupt metadata on one or more chunks. */ 3381 sr_error(sc, "one of the chunks has corrupt metadata; " 3382 "aborting assembly"); 3383 goto unwind; 3384 3385 } else if (no_meta == 0) { 3386 3387 /* Initialise volume and chunk metadata. */ 3388 sr_meta_init(sd, bc->bc_level, no_chunk); 3389 sd->sd_vol_status = BIOC_SVONLINE; 3390 sd->sd_meta_flags = bc->bc_flags & BIOC_SCNOAUTOASSEMBLE; 3391 if (sd->sd_create) { 3392 if ((i = sd->sd_create(sd, bc, no_chunk, 3393 sd->sd_vol.sv_chunk_minsz))) { 3394 rv = i; 3395 goto unwind; 3396 } 3397 } 3398 sr_meta_init_complete(sd); 3399 3400 DNPRINTF(SR_D_IOCTL, 3401 "%s: sr_ioctl_createraid: vol_size: %lld\n", 3402 DEVNAME(sc), sd->sd_meta->ssdi.ssd_size); 3403 3404 /* Warn if we've wasted chunk space due to coercing. */ 3405 if ((sd->sd_capabilities & SR_CAP_NON_COERCED) == 0 && 3406 sd->sd_vol.sv_chunk_minsz != sd->sd_vol.sv_chunk_maxsz) 3407 sr_warn(sc, "chunk sizes are not equal; up to %llu " 3408 "blocks wasted per chunk", 3409 sd->sd_vol.sv_chunk_maxsz - 3410 sd->sd_vol.sv_chunk_minsz); 3411 3412 } else { 3413 3414 /* Ensure we are assembling the correct # of chunks. */ 3415 if (sd->sd_meta->ssdi.ssd_chunk_no != no_chunk) { 3416 sr_error(sc, "volume chunk count does not match metadata " 3417 "chunk count"); 3418 goto unwind; 3419 } 3420 3421 /* Ensure metadata level matches requested assembly level. */ 3422 if (sd->sd_meta->ssdi.ssd_level != bc->bc_level) { 3423 sr_error(sc, "volume level does not match metadata " 3424 "level"); 3425 goto unwind; 3426 } 3427 3428 if (sr_already_assembled(sd)) { 3429 uuid = sr_uuid_format(&sd->sd_meta->ssdi.ssd_uuid); 3430 sr_error(sc, "disk %s already assembled", uuid); 3431 free(uuid, M_DEVBUF, 37); 3432 goto unwind; 3433 } 3434 3435 if (user == 0 && sd->sd_meta_flags & BIOC_SCNOAUTOASSEMBLE) { 3436 DNPRINTF(SR_D_META, "%s: disk not auto assembled from " 3437 "metadata\n", DEVNAME(sc)); 3438 goto unwind; 3439 } 3440 3441 if (no_meta != no_chunk) 3442 sr_warn(sc, "trying to bring up %s degraded", 3443 sd->sd_meta->ssd_devname); 3444 3445 if (sd->sd_meta->ssd_meta_flags & SR_META_DIRTY) 3446 sr_warn(sc, "%s was not shutdown properly", 3447 sd->sd_meta->ssd_devname); 3448 3449 SLIST_FOREACH(omi, &sd->sd_meta_opt, omi_link) 3450 if (sd->sd_meta_opt_handler == NULL || 3451 sd->sd_meta_opt_handler(sd, omi->omi_som) != 0) 3452 sr_meta_opt_handler(sd, omi->omi_som); 3453 3454 if (sd->sd_assemble) { 3455 if ((i = sd->sd_assemble(sd, bc, no_chunk, data))) { 3456 rv = i; 3457 goto unwind; 3458 } 3459 } 3460 3461 DNPRINTF(SR_D_META, "%s: disk assembled from metadata\n", 3462 DEVNAME(sc)); 3463 3464 } 3465 3466 /* Metadata MUST be fully populated by this point. */ 3467 TAILQ_INSERT_TAIL(&sc->sc_dis_list, sd, sd_link); 3468 3469 /* Allocate all resources. */ 3470 if ((rv = sd->sd_alloc_resources(sd))) 3471 goto unwind; 3472 3473 /* Adjust flags if necessary. */ 3474 if ((sd->sd_capabilities & SR_CAP_AUTO_ASSEMBLE) && 3475 (bc->bc_flags & BIOC_SCNOAUTOASSEMBLE) != 3476 (sd->sd_meta->ssdi.ssd_vol_flags & BIOC_SCNOAUTOASSEMBLE)) { 3477 sd->sd_meta->ssdi.ssd_vol_flags &= ~BIOC_SCNOAUTOASSEMBLE; 3478 sd->sd_meta->ssdi.ssd_vol_flags |= 3479 bc->bc_flags & BIOC_SCNOAUTOASSEMBLE; 3480 } 3481 3482 if (sd->sd_capabilities & SR_CAP_SYSTEM_DISK) { 3483 /* Initialise volume state. */ 3484 sd->sd_set_vol_state(sd); 3485 if (sd->sd_vol_status == BIOC_SVOFFLINE) { 3486 sr_error(sc, "%s is offline, will not be brought " 3487 "online", sd->sd_meta->ssd_devname); 3488 goto unwind; 3489 } 3490 3491 /* Setup SCSI iopool. */ 3492 scsi_iopool_init(&sd->sd_iopool, sd, sr_wu_get, sr_wu_put); 3493 3494 /* 3495 * All checks passed - return ENXIO if volume cannot be created. 3496 */ 3497 rv = ENXIO; 3498 3499 /* 3500 * Find a free target. 3501 * 3502 * XXX: We reserve sd_target == 0 to indicate the 3503 * discipline is not linked into sc->sc_targets, so begin 3504 * the search with target = 1. 3505 */ 3506 for (target = 1; target < SR_MAX_LD; target++) 3507 if (sc->sc_targets[target] == NULL) 3508 break; 3509 if (target == SR_MAX_LD) { 3510 sr_error(sc, "no free target for %s", 3511 sd->sd_meta->ssd_devname); 3512 goto unwind; 3513 } 3514 3515 /* Clear sense data. */ 3516 bzero(&sd->sd_scsi_sense, sizeof(sd->sd_scsi_sense)); 3517 3518 /* Attach discipline and get midlayer to probe it. */ 3519 sd->sd_target = target; 3520 sc->sc_targets[target] = sd; 3521 if (scsi_probe_lun(sc->sc_scsibus, target, 0) != 0) { 3522 sr_error(sc, "scsi_probe_lun failed"); 3523 sc->sc_targets[target] = NULL; 3524 sd->sd_target = 0; 3525 goto unwind; 3526 } 3527 3528 link = scsi_get_link(sc->sc_scsibus, target, 0); 3529 if (link == NULL) 3530 goto unwind; 3531 3532 dev = link->device_softc; 3533 DNPRINTF(SR_D_IOCTL, "%s: sr device added: %s at target %d\n", 3534 DEVNAME(sc), dev->dv_xname, sd->sd_target); 3535 3536 /* XXX - Count volumes, not targets. */ 3537 for (i = 0, vol = -1; i <= sd->sd_target; i++) 3538 if (sc->sc_targets[i]) 3539 vol++; 3540 3541 rv = 0; 3542 3543 if (sd->sd_meta->ssd_devname[0] != '\0' && 3544 strncmp(sd->sd_meta->ssd_devname, dev->dv_xname, 3545 sizeof(dev->dv_xname))) 3546 sr_warn(sc, "volume %s is roaming, it used to be %s, " 3547 "updating metadata", dev->dv_xname, 3548 sd->sd_meta->ssd_devname); 3549 3550 /* Populate remaining volume metadata. */ 3551 sd->sd_meta->ssdi.ssd_volid = vol; 3552 strlcpy(sd->sd_meta->ssd_devname, dev->dv_xname, 3553 sizeof(sd->sd_meta->ssd_devname)); 3554 3555 sr_info(sc, "%s volume attached as %s", 3556 sd->sd_name, sd->sd_meta->ssd_devname); 3557 3558 /* Update device name on any roaming chunks. */ 3559 sr_roam_chunks(sd); 3560 3561 #ifndef SMALL_KERNEL 3562 if (sr_sensors_create(sd)) 3563 sr_warn(sc, "unable to create sensor for %s", 3564 dev->dv_xname); 3565 #endif /* SMALL_KERNEL */ 3566 } else { 3567 /* This volume does not attach as a system disk. */ 3568 ch_entry = SLIST_FIRST(cl); /* XXX */ 3569 strlcpy(sd->sd_meta->ssd_devname, ch_entry->src_devname, 3570 sizeof(sd->sd_meta->ssd_devname)); 3571 3572 if (sd->sd_start_discipline(sd)) 3573 goto unwind; 3574 } 3575 3576 /* Save current metadata to disk. */ 3577 rv = sr_meta_save(sd, SR_META_DIRTY); 3578 3579 if (sd->sd_vol_status == BIOC_SVREBUILD) 3580 kthread_create_deferred(sr_rebuild_start, sd); 3581 3582 sd->sd_ready = 1; 3583 3584 free(dt, M_DEVBUF, bc->bc_dev_list_len); 3585 3586 return (rv); 3587 3588 unwind: 3589 free(dt, M_DEVBUF, bc->bc_dev_list_len); 3590 3591 sr_discipline_shutdown(sd, 0, 0); 3592 3593 if (rv == EAGAIN) 3594 rv = 0; 3595 3596 return (rv); 3597 } 3598 3599 int 3600 sr_ioctl_deleteraid(struct sr_softc *sc, struct sr_discipline *sd, 3601 struct bioc_deleteraid *bd) 3602 { 3603 int rv = 1; 3604 3605 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_deleteraid %s\n", 3606 DEVNAME(sc), bd->bd_dev); 3607 3608 if (sd == NULL) { 3609 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 3610 if (!strncmp(sd->sd_meta->ssd_devname, bd->bd_dev, 3611 sizeof(sd->sd_meta->ssd_devname))) 3612 break; 3613 } 3614 if (sd == NULL) { 3615 sr_error(sc, "volume %s not found", bd->bd_dev); 3616 goto bad; 3617 } 3618 } 3619 3620 sd->sd_deleted = 1; 3621 sd->sd_meta->ssdi.ssd_vol_flags = BIOC_SCNOAUTOASSEMBLE; 3622 sr_discipline_shutdown(sd, 1, 0); 3623 3624 rv = 0; 3625 bad: 3626 return (rv); 3627 } 3628 3629 int 3630 sr_ioctl_discipline(struct sr_softc *sc, struct sr_discipline *sd, 3631 struct bioc_discipline *bd) 3632 { 3633 int rv = 1; 3634 3635 /* Dispatch a discipline specific ioctl. */ 3636 3637 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_discipline %s\n", DEVNAME(sc), 3638 bd->bd_dev); 3639 3640 if (sd == NULL) { 3641 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 3642 if (!strncmp(sd->sd_meta->ssd_devname, bd->bd_dev, 3643 sizeof(sd->sd_meta->ssd_devname))) 3644 break; 3645 } 3646 if (sd == NULL) { 3647 sr_error(sc, "volume %s not found", bd->bd_dev); 3648 goto bad; 3649 } 3650 } 3651 3652 if (sd->sd_ioctl_handler) 3653 rv = sd->sd_ioctl_handler(sd, bd); 3654 3655 bad: 3656 return (rv); 3657 } 3658 3659 int 3660 sr_ioctl_installboot(struct sr_softc *sc, struct sr_discipline *sd, 3661 struct bioc_installboot *bb) 3662 { 3663 void *bootblk = NULL, *bootldr = NULL; 3664 struct sr_chunk *chunk; 3665 struct sr_meta_opt_item *omi; 3666 struct sr_meta_boot *sbm; 3667 struct disk *dk; 3668 u_int32_t bbs = 0, bls = 0, secsize; 3669 u_char duid[8]; 3670 int rv = EINVAL; 3671 int i; 3672 3673 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_installboot %s\n", DEVNAME(sc), 3674 bb->bb_dev); 3675 3676 if (sd == NULL) { 3677 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 3678 if (!strncmp(sd->sd_meta->ssd_devname, bb->bb_dev, 3679 sizeof(sd->sd_meta->ssd_devname))) 3680 break; 3681 } 3682 if (sd == NULL) { 3683 sr_error(sc, "volume %s not found", bb->bb_dev); 3684 goto done; 3685 } 3686 } 3687 3688 bzero(duid, sizeof(duid)); 3689 TAILQ_FOREACH(dk, &disklist, dk_link) 3690 if (!strncmp(dk->dk_name, bb->bb_dev, sizeof(bb->bb_dev))) 3691 break; 3692 if (dk == NULL || dk->dk_label == NULL || 3693 (dk->dk_flags & DKF_LABELVALID) == 0 || 3694 bcmp(dk->dk_label->d_uid, &duid, sizeof(duid)) == 0) { 3695 sr_error(sc, "failed to get DUID for softraid volume"); 3696 goto done; 3697 } 3698 memcpy(duid, dk->dk_label->d_uid, sizeof(duid)); 3699 3700 /* Ensure that boot storage area is large enough. */ 3701 if (sd->sd_meta->ssd_data_blkno < (SR_BOOT_OFFSET + SR_BOOT_SIZE)) { 3702 sr_error(sc, "insufficient boot storage"); 3703 goto done; 3704 } 3705 3706 if (bb->bb_bootblk_size > SR_BOOT_BLOCKS_SIZE * DEV_BSIZE) { 3707 sr_error(sc, "boot block too large (%d > %d)", 3708 bb->bb_bootblk_size, SR_BOOT_BLOCKS_SIZE * DEV_BSIZE); 3709 goto done; 3710 } 3711 3712 if (bb->bb_bootldr_size > SR_BOOT_LOADER_SIZE * DEV_BSIZE) { 3713 sr_error(sc, "boot loader too large (%d > %d)", 3714 bb->bb_bootldr_size, SR_BOOT_LOADER_SIZE * DEV_BSIZE); 3715 goto done; 3716 } 3717 3718 secsize = sd->sd_meta->ssdi.ssd_secsize; 3719 3720 /* Copy in boot block. */ 3721 bbs = howmany(bb->bb_bootblk_size, secsize) * secsize; 3722 bootblk = malloc(bbs, M_DEVBUF, M_WAITOK | M_ZERO); 3723 if (copyin(bb->bb_bootblk, bootblk, bb->bb_bootblk_size) != 0) 3724 goto done; 3725 3726 /* Copy in boot loader. */ 3727 bls = howmany(bb->bb_bootldr_size, secsize) * secsize; 3728 bootldr = malloc(bls, M_DEVBUF, M_WAITOK | M_ZERO); 3729 if (copyin(bb->bb_bootldr, bootldr, bb->bb_bootldr_size) != 0) 3730 goto done; 3731 3732 /* Create or update optional meta for bootable volumes. */ 3733 SLIST_FOREACH(omi, &sd->sd_meta_opt, omi_link) 3734 if (omi->omi_som->som_type == SR_OPT_BOOT) 3735 break; 3736 if (omi == NULL) { 3737 omi = malloc(sizeof(struct sr_meta_opt_item), M_DEVBUF, 3738 M_WAITOK | M_ZERO); 3739 omi->omi_som = malloc(sizeof(struct sr_meta_crypto), M_DEVBUF, 3740 M_WAITOK | M_ZERO); 3741 omi->omi_som->som_type = SR_OPT_BOOT; 3742 omi->omi_som->som_length = sizeof(struct sr_meta_boot); 3743 SLIST_INSERT_HEAD(&sd->sd_meta_opt, omi, omi_link); 3744 sd->sd_meta->ssdi.ssd_opt_no++; 3745 } 3746 sbm = (struct sr_meta_boot *)omi->omi_som; 3747 3748 memcpy(sbm->sbm_root_duid, duid, sizeof(sbm->sbm_root_duid)); 3749 bzero(&sbm->sbm_boot_duid, sizeof(sbm->sbm_boot_duid)); 3750 sbm->sbm_bootblk_size = bbs; 3751 sbm->sbm_bootldr_size = bls; 3752 3753 DNPRINTF(SR_D_IOCTL, "sr_ioctl_installboot: root duid is %s\n", 3754 duid_format(sbm->sbm_root_duid)); 3755 3756 /* Save boot block and boot loader to each chunk. */ 3757 for (i = 0; i < sd->sd_meta->ssdi.ssd_chunk_no; i++) { 3758 3759 chunk = sd->sd_vol.sv_chunks[i]; 3760 if (chunk->src_meta.scm_status != BIOC_SDONLINE && 3761 chunk->src_meta.scm_status != BIOC_SDREBUILD) 3762 continue; 3763 3764 if (i < SR_MAX_BOOT_DISKS) 3765 memcpy(&sbm->sbm_boot_duid[i], chunk->src_duid, 3766 sizeof(sbm->sbm_boot_duid[i])); 3767 3768 /* Save boot blocks. */ 3769 DNPRINTF(SR_D_IOCTL, 3770 "sr_ioctl_installboot: saving boot block to %s " 3771 "(%u bytes)\n", chunk->src_devname, bbs); 3772 3773 if (sr_rw(sc, chunk->src_dev_mm, bootblk, bbs, 3774 SR_BOOT_BLOCKS_OFFSET, B_WRITE)) { 3775 sr_error(sc, "failed to write boot block"); 3776 goto done; 3777 } 3778 3779 /* Save boot loader.*/ 3780 DNPRINTF(SR_D_IOCTL, 3781 "sr_ioctl_installboot: saving boot loader to %s " 3782 "(%u bytes)\n", chunk->src_devname, bls); 3783 3784 if (sr_rw(sc, chunk->src_dev_mm, bootldr, bls, 3785 SR_BOOT_LOADER_OFFSET, B_WRITE)) { 3786 sr_error(sc, "failed to write boot loader"); 3787 goto done; 3788 } 3789 } 3790 3791 /* XXX - Install boot block on disk - MD code. */ 3792 3793 /* Mark volume as bootable and save metadata. */ 3794 sd->sd_meta->ssdi.ssd_vol_flags |= BIOC_SCBOOTABLE; 3795 if (sr_meta_save(sd, SR_META_DIRTY)) { 3796 sr_error(sc, "could not save metadata to %s", DEVNAME(sc)); 3797 goto done; 3798 } 3799 3800 rv = 0; 3801 3802 done: 3803 free(bootblk, M_DEVBUF, bbs); 3804 free(bootldr, M_DEVBUF, bls); 3805 3806 return (rv); 3807 } 3808 3809 void 3810 sr_chunks_unwind(struct sr_softc *sc, struct sr_chunk_head *cl) 3811 { 3812 struct sr_chunk *ch_entry, *ch_next; 3813 3814 DNPRINTF(SR_D_IOCTL, "%s: sr_chunks_unwind\n", DEVNAME(sc)); 3815 3816 if (!cl) 3817 return; 3818 3819 for (ch_entry = SLIST_FIRST(cl); ch_entry != NULL; ch_entry = ch_next) { 3820 ch_next = SLIST_NEXT(ch_entry, src_link); 3821 3822 DNPRINTF(SR_D_IOCTL, "%s: sr_chunks_unwind closing: %s\n", 3823 DEVNAME(sc), ch_entry->src_devname); 3824 if (ch_entry->src_vn) { 3825 /* 3826 * XXX - explicitly lock the vnode until we can resolve 3827 * the problem introduced by vnode aliasing... specfs 3828 * has no locking, whereas ufs/ffs does! 3829 */ 3830 vn_lock(ch_entry->src_vn, LK_EXCLUSIVE | LK_RETRY); 3831 VOP_CLOSE(ch_entry->src_vn, FREAD | FWRITE, NOCRED, 3832 curproc); 3833 vput(ch_entry->src_vn); 3834 } 3835 free(ch_entry, M_DEVBUF, sizeof(*ch_entry)); 3836 } 3837 SLIST_INIT(cl); 3838 } 3839 3840 void 3841 sr_discipline_free(struct sr_discipline *sd) 3842 { 3843 struct sr_softc *sc; 3844 struct sr_discipline *sdtmp1; 3845 struct sr_meta_opt_head *som; 3846 struct sr_meta_opt_item *omi, *omi_next; 3847 3848 if (!sd) 3849 return; 3850 3851 sc = sd->sd_sc; 3852 3853 DNPRINTF(SR_D_DIS, "%s: sr_discipline_free %s\n", 3854 DEVNAME(sc), 3855 sd->sd_meta ? sd->sd_meta->ssd_devname : "nodev"); 3856 if (sd->sd_free_resources) 3857 sd->sd_free_resources(sd); 3858 free(sd->sd_vol.sv_chunks, M_DEVBUF, 0); 3859 free(sd->sd_meta, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 3860 free(sd->sd_meta_foreign, M_DEVBUF, smd[sd->sd_meta_type].smd_size); 3861 3862 som = &sd->sd_meta_opt; 3863 for (omi = SLIST_FIRST(som); omi != NULL; omi = omi_next) { 3864 omi_next = SLIST_NEXT(omi, omi_link); 3865 free(omi->omi_som, M_DEVBUF, 0); 3866 free(omi, M_DEVBUF, sizeof(*omi)); 3867 } 3868 3869 if (sd->sd_target != 0) { 3870 KASSERT(sc->sc_targets[sd->sd_target] == sd); 3871 sc->sc_targets[sd->sd_target] = NULL; 3872 } 3873 3874 TAILQ_FOREACH(sdtmp1, &sc->sc_dis_list, sd_link) { 3875 if (sdtmp1 == sd) 3876 break; 3877 } 3878 if (sdtmp1 != NULL) 3879 TAILQ_REMOVE(&sc->sc_dis_list, sd, sd_link); 3880 3881 explicit_bzero(sd, sizeof *sd); 3882 free(sd, M_DEVBUF, sizeof(*sd)); 3883 } 3884 3885 void 3886 sr_discipline_shutdown(struct sr_discipline *sd, int meta_save, int dying) 3887 { 3888 struct sr_softc *sc; 3889 int ret, s; 3890 3891 if (!sd) 3892 return; 3893 sc = sd->sd_sc; 3894 3895 DNPRINTF(SR_D_DIS, "%s: sr_discipline_shutdown %s\n", DEVNAME(sc), 3896 sd->sd_meta ? sd->sd_meta->ssd_devname : "nodev"); 3897 3898 /* If rebuilding, abort rebuild and drain I/O. */ 3899 if (sd->sd_reb_active) { 3900 sd->sd_reb_abort = 1; 3901 while (sd->sd_reb_active) 3902 tsleep(sd, PWAIT, "sr_shutdown", 1); 3903 } 3904 3905 if (meta_save) 3906 sr_meta_save(sd, 0); 3907 3908 s = splbio(); 3909 3910 sd->sd_ready = 0; 3911 3912 /* make sure there isn't a sync pending and yield */ 3913 wakeup(sd); 3914 while (sd->sd_sync || sd->sd_must_flush) { 3915 ret = tsleep_nsec(&sd->sd_sync, MAXPRI, "sr_down", 3916 SEC_TO_NSEC(60)); 3917 if (ret == EWOULDBLOCK) 3918 break; 3919 } 3920 if (dying == -1) { 3921 sd->sd_ready = 1; 3922 splx(s); 3923 return; 3924 } 3925 3926 #ifndef SMALL_KERNEL 3927 sr_sensors_delete(sd); 3928 #endif /* SMALL_KERNEL */ 3929 3930 if (sd->sd_target != 0) 3931 scsi_detach_lun(sc->sc_scsibus, sd->sd_target, 0, 3932 dying ? 0 : DETACH_FORCE); 3933 3934 sr_chunks_unwind(sc, &sd->sd_vol.sv_chunk_list); 3935 3936 if (sd->sd_taskq) 3937 taskq_destroy(sd->sd_taskq); 3938 3939 sr_discipline_free(sd); 3940 3941 splx(s); 3942 } 3943 3944 int 3945 sr_discipline_init(struct sr_discipline *sd, int level) 3946 { 3947 int rv = 1; 3948 3949 /* Initialise discipline function pointers with defaults. */ 3950 sd->sd_alloc_resources = sr_alloc_resources; 3951 sd->sd_assemble = NULL; 3952 sd->sd_create = NULL; 3953 sd->sd_free_resources = sr_free_resources; 3954 sd->sd_ioctl_handler = NULL; 3955 sd->sd_openings = NULL; 3956 sd->sd_meta_opt_handler = NULL; 3957 sd->sd_rebuild = sr_rebuild; 3958 sd->sd_scsi_inquiry = sr_raid_inquiry; 3959 sd->sd_scsi_read_cap = sr_raid_read_cap; 3960 sd->sd_scsi_tur = sr_raid_tur; 3961 sd->sd_scsi_req_sense = sr_raid_request_sense; 3962 sd->sd_scsi_start_stop = sr_raid_start_stop; 3963 sd->sd_scsi_sync = sr_raid_sync; 3964 sd->sd_scsi_rw = NULL; 3965 sd->sd_scsi_intr = sr_raid_intr; 3966 sd->sd_scsi_wu_done = NULL; 3967 sd->sd_scsi_done = NULL; 3968 sd->sd_set_chunk_state = sr_set_chunk_state; 3969 sd->sd_set_vol_state = sr_set_vol_state; 3970 sd->sd_start_discipline = NULL; 3971 3972 task_set(&sd->sd_meta_save_task, sr_meta_save_callback, sd); 3973 task_set(&sd->sd_hotspare_rebuild_task, sr_hotspare_rebuild_callback, 3974 sd); 3975 3976 sd->sd_wu_size = sizeof(struct sr_workunit); 3977 switch (level) { 3978 case 0: 3979 sr_raid0_discipline_init(sd); 3980 break; 3981 case 1: 3982 sr_raid1_discipline_init(sd); 3983 break; 3984 case 5: 3985 sr_raid5_discipline_init(sd); 3986 break; 3987 case 6: 3988 sr_raid6_discipline_init(sd); 3989 break; 3990 #ifdef CRYPTO 3991 case 'C': 3992 sr_crypto_discipline_init(sd); 3993 break; 3994 #endif 3995 case 'c': 3996 sr_concat_discipline_init(sd); 3997 break; 3998 default: 3999 goto bad; 4000 } 4001 4002 rv = 0; 4003 bad: 4004 return (rv); 4005 } 4006 4007 int 4008 sr_raid_inquiry(struct sr_workunit *wu) 4009 { 4010 struct sr_discipline *sd = wu->swu_dis; 4011 struct scsi_xfer *xs = wu->swu_xs; 4012 struct scsi_inquiry *cdb = (struct scsi_inquiry *)xs->cmd; 4013 struct scsi_inquiry_data inq; 4014 4015 DNPRINTF(SR_D_DIS, "%s: sr_raid_inquiry\n", DEVNAME(sd->sd_sc)); 4016 4017 if (xs->cmdlen != sizeof(*cdb)) 4018 return (EINVAL); 4019 4020 if (ISSET(cdb->flags, SI_EVPD)) 4021 return (EOPNOTSUPP); 4022 4023 bzero(&inq, sizeof(inq)); 4024 inq.device = T_DIRECT; 4025 inq.dev_qual2 = 0; 4026 inq.version = 2; 4027 inq.response_format = 2; 4028 inq.additional_length = 32; 4029 inq.flags |= SID_CmdQue; 4030 strlcpy(inq.vendor, sd->sd_meta->ssdi.ssd_vendor, 4031 sizeof(inq.vendor)); 4032 strlcpy(inq.product, sd->sd_meta->ssdi.ssd_product, 4033 sizeof(inq.product)); 4034 strlcpy(inq.revision, sd->sd_meta->ssdi.ssd_revision, 4035 sizeof(inq.revision)); 4036 sr_copy_internal_data(xs, &inq, sizeof(inq)); 4037 4038 return (0); 4039 } 4040 4041 int 4042 sr_raid_read_cap(struct sr_workunit *wu) 4043 { 4044 struct sr_discipline *sd = wu->swu_dis; 4045 struct scsi_xfer *xs = wu->swu_xs; 4046 struct scsi_read_cap_data rcd; 4047 struct scsi_read_cap_data_16 rcd16; 4048 u_int64_t addr; 4049 int rv = 1; 4050 u_int32_t secsize; 4051 4052 DNPRINTF(SR_D_DIS, "%s: sr_raid_read_cap\n", DEVNAME(sd->sd_sc)); 4053 4054 secsize = sd->sd_meta->ssdi.ssd_secsize; 4055 4056 addr = ((sd->sd_meta->ssdi.ssd_size * DEV_BSIZE) / secsize) - 1; 4057 if (xs->cmd->opcode == READ_CAPACITY) { 4058 bzero(&rcd, sizeof(rcd)); 4059 if (addr > 0xffffffffllu) 4060 _lto4b(0xffffffff, rcd.addr); 4061 else 4062 _lto4b(addr, rcd.addr); 4063 _lto4b(secsize, rcd.length); 4064 sr_copy_internal_data(xs, &rcd, sizeof(rcd)); 4065 rv = 0; 4066 } else if (xs->cmd->opcode == READ_CAPACITY_16) { 4067 bzero(&rcd16, sizeof(rcd16)); 4068 _lto8b(addr, rcd16.addr); 4069 _lto4b(secsize, rcd16.length); 4070 sr_copy_internal_data(xs, &rcd16, sizeof(rcd16)); 4071 rv = 0; 4072 } 4073 4074 return (rv); 4075 } 4076 4077 int 4078 sr_raid_tur(struct sr_workunit *wu) 4079 { 4080 struct sr_discipline *sd = wu->swu_dis; 4081 4082 DNPRINTF(SR_D_DIS, "%s: sr_raid_tur\n", DEVNAME(sd->sd_sc)); 4083 4084 if (sd->sd_vol_status == BIOC_SVOFFLINE) { 4085 sd->sd_scsi_sense.error_code = SSD_ERRCODE_CURRENT; 4086 sd->sd_scsi_sense.flags = SKEY_NOT_READY; 4087 sd->sd_scsi_sense.add_sense_code = 0x04; 4088 sd->sd_scsi_sense.add_sense_code_qual = 0x11; 4089 sd->sd_scsi_sense.extra_len = 4; 4090 return (1); 4091 } else if (sd->sd_vol_status == BIOC_SVINVALID) { 4092 sd->sd_scsi_sense.error_code = SSD_ERRCODE_CURRENT; 4093 sd->sd_scsi_sense.flags = SKEY_HARDWARE_ERROR; 4094 sd->sd_scsi_sense.add_sense_code = 0x05; 4095 sd->sd_scsi_sense.add_sense_code_qual = 0x00; 4096 sd->sd_scsi_sense.extra_len = 4; 4097 return (1); 4098 } 4099 4100 return (0); 4101 } 4102 4103 int 4104 sr_raid_request_sense(struct sr_workunit *wu) 4105 { 4106 struct sr_discipline *sd = wu->swu_dis; 4107 struct scsi_xfer *xs = wu->swu_xs; 4108 4109 DNPRINTF(SR_D_DIS, "%s: sr_raid_request_sense\n", 4110 DEVNAME(sd->sd_sc)); 4111 4112 /* use latest sense data */ 4113 memcpy(&xs->sense, &sd->sd_scsi_sense, sizeof(xs->sense)); 4114 4115 /* clear sense data */ 4116 bzero(&sd->sd_scsi_sense, sizeof(sd->sd_scsi_sense)); 4117 4118 return (0); 4119 } 4120 4121 int 4122 sr_raid_start_stop(struct sr_workunit *wu) 4123 { 4124 struct scsi_xfer *xs = wu->swu_xs; 4125 struct scsi_start_stop *ss = (struct scsi_start_stop *)xs->cmd; 4126 4127 DNPRINTF(SR_D_DIS, "%s: sr_raid_start_stop\n", 4128 DEVNAME(wu->swu_dis->sd_sc)); 4129 4130 if (!ss) 4131 return (1); 4132 4133 /* 4134 * do nothing! 4135 * a softraid discipline should always reflect correct status 4136 */ 4137 return (0); 4138 } 4139 4140 int 4141 sr_raid_sync(struct sr_workunit *wu) 4142 { 4143 struct sr_discipline *sd = wu->swu_dis; 4144 int s, ret, rv = 0, ios; 4145 4146 DNPRINTF(SR_D_DIS, "%s: sr_raid_sync\n", DEVNAME(sd->sd_sc)); 4147 4148 /* when doing a fake sync don't count the wu */ 4149 ios = (wu->swu_flags & SR_WUF_FAKE) ? 0 : 1; 4150 4151 s = splbio(); 4152 sd->sd_sync = 1; 4153 while (sd->sd_wu_pending > ios) { 4154 ret = tsleep_nsec(sd, PRIBIO, "sr_sync", SEC_TO_NSEC(15)); 4155 if (ret == EWOULDBLOCK) { 4156 DNPRINTF(SR_D_DIS, "%s: sr_raid_sync timeout\n", 4157 DEVNAME(sd->sd_sc)); 4158 rv = 1; 4159 break; 4160 } 4161 } 4162 sd->sd_sync = 0; 4163 splx(s); 4164 4165 wakeup(&sd->sd_sync); 4166 4167 return (rv); 4168 } 4169 4170 void 4171 sr_raid_intr(struct buf *bp) 4172 { 4173 struct sr_ccb *ccb = (struct sr_ccb *)bp; 4174 struct sr_workunit *wu = ccb->ccb_wu; 4175 #ifdef SR_DEBUG 4176 struct sr_discipline *sd = wu->swu_dis; 4177 struct scsi_xfer *xs = wu->swu_xs; 4178 #endif 4179 int s; 4180 4181 DNPRINTF(SR_D_INTR, "%s: %s %s intr bp %p xs %p\n", 4182 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, sd->sd_name, bp, xs); 4183 4184 s = splbio(); 4185 sr_ccb_done(ccb); 4186 sr_wu_done(wu); 4187 splx(s); 4188 } 4189 4190 void 4191 sr_schedule_wu(struct sr_workunit *wu) 4192 { 4193 struct sr_discipline *sd = wu->swu_dis; 4194 struct sr_workunit *wup; 4195 int s; 4196 4197 DNPRINTF(SR_D_WU, "sr_schedule_wu: schedule wu %p state %i " 4198 "flags 0x%x\n", wu, wu->swu_state, wu->swu_flags); 4199 4200 KASSERT(wu->swu_io_count > 0); 4201 4202 s = splbio(); 4203 4204 /* Construct the work unit, do not schedule it. */ 4205 if (wu->swu_state == SR_WU_CONSTRUCT) 4206 goto queued; 4207 4208 /* Deferred work unit being reconstructed, do not start. */ 4209 if (wu->swu_state == SR_WU_REQUEUE) 4210 goto queued; 4211 4212 /* Current work unit failed, restart. */ 4213 if (wu->swu_state == SR_WU_RESTART) 4214 goto start; 4215 4216 if (wu->swu_state != SR_WU_INPROGRESS) 4217 panic("sr_schedule_wu: work unit not in progress (state %i)\n", 4218 wu->swu_state); 4219 4220 /* Walk queue backwards and fill in collider if we have one. */ 4221 TAILQ_FOREACH_REVERSE(wup, &sd->sd_wu_pendq, sr_wu_list, swu_link) { 4222 if (wu->swu_blk_end < wup->swu_blk_start || 4223 wup->swu_blk_end < wu->swu_blk_start) 4224 continue; 4225 4226 /* Defer work unit due to LBA collision. */ 4227 DNPRINTF(SR_D_WU, "sr_schedule_wu: deferring work unit %p\n", 4228 wu); 4229 wu->swu_state = SR_WU_DEFERRED; 4230 while (wup->swu_collider) 4231 wup = wup->swu_collider; 4232 wup->swu_collider = wu; 4233 TAILQ_INSERT_TAIL(&sd->sd_wu_defq, wu, swu_link); 4234 sd->sd_wu_collisions++; 4235 goto queued; 4236 } 4237 4238 start: 4239 sr_raid_startwu(wu); 4240 4241 queued: 4242 splx(s); 4243 } 4244 4245 void 4246 sr_raid_startwu(struct sr_workunit *wu) 4247 { 4248 struct sr_discipline *sd = wu->swu_dis; 4249 struct sr_ccb *ccb; 4250 4251 DNPRINTF(SR_D_WU, "sr_raid_startwu: start wu %p\n", wu); 4252 4253 splassert(IPL_BIO); 4254 4255 if (wu->swu_state == SR_WU_DEFERRED) { 4256 TAILQ_REMOVE(&sd->sd_wu_defq, wu, swu_link); 4257 wu->swu_state = SR_WU_INPROGRESS; 4258 } 4259 4260 if (wu->swu_state != SR_WU_RESTART) 4261 TAILQ_INSERT_TAIL(&sd->sd_wu_pendq, wu, swu_link); 4262 4263 /* Start all of the individual I/Os. */ 4264 if (wu->swu_cb_active == 1) 4265 panic("%s: sr_startwu_callback", DEVNAME(sd->sd_sc)); 4266 wu->swu_cb_active = 1; 4267 4268 TAILQ_FOREACH(ccb, &wu->swu_ccb, ccb_link) 4269 VOP_STRATEGY(&ccb->ccb_buf); 4270 4271 wu->swu_cb_active = 0; 4272 } 4273 4274 void 4275 sr_raid_recreate_wu(struct sr_workunit *wu) 4276 { 4277 struct sr_discipline *sd = wu->swu_dis; 4278 struct sr_workunit *wup = wu; 4279 4280 /* 4281 * Recreate a work unit by releasing the associated CCBs and reissuing 4282 * the SCSI I/O request. This process is then repeated for all of the 4283 * colliding work units. 4284 */ 4285 do { 4286 sr_wu_release_ccbs(wup); 4287 4288 wup->swu_state = SR_WU_REQUEUE; 4289 if (sd->sd_scsi_rw(wup)) 4290 panic("could not requeue I/O"); 4291 4292 wup = wup->swu_collider; 4293 } while (wup); 4294 } 4295 4296 int 4297 sr_alloc_resources(struct sr_discipline *sd) 4298 { 4299 if (sr_wu_alloc(sd)) { 4300 sr_error(sd->sd_sc, "unable to allocate work units"); 4301 return (ENOMEM); 4302 } 4303 if (sr_ccb_alloc(sd)) { 4304 sr_error(sd->sd_sc, "unable to allocate ccbs"); 4305 return (ENOMEM); 4306 } 4307 4308 return (0); 4309 } 4310 4311 void 4312 sr_free_resources(struct sr_discipline *sd) 4313 { 4314 sr_wu_free(sd); 4315 sr_ccb_free(sd); 4316 } 4317 4318 void 4319 sr_set_chunk_state(struct sr_discipline *sd, int c, int new_state) 4320 { 4321 int old_state, s; 4322 4323 DNPRINTF(SR_D_STATE, "%s: %s: %s: sr_set_chunk_state %d -> %d\n", 4324 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, 4325 sd->sd_vol.sv_chunks[c]->src_meta.scmi.scm_devname, c, new_state); 4326 4327 /* ok to go to splbio since this only happens in error path */ 4328 s = splbio(); 4329 old_state = sd->sd_vol.sv_chunks[c]->src_meta.scm_status; 4330 4331 /* multiple IOs to the same chunk that fail will come through here */ 4332 if (old_state == new_state) 4333 goto done; 4334 4335 switch (old_state) { 4336 case BIOC_SDONLINE: 4337 if (new_state == BIOC_SDOFFLINE) 4338 break; 4339 else 4340 goto die; 4341 break; 4342 4343 case BIOC_SDOFFLINE: 4344 goto die; 4345 4346 default: 4347 die: 4348 splx(s); /* XXX */ 4349 panic("%s: %s: %s: invalid chunk state transition " 4350 "%d -> %d", DEVNAME(sd->sd_sc), 4351 sd->sd_meta->ssd_devname, 4352 sd->sd_vol.sv_chunks[c]->src_meta.scmi.scm_devname, 4353 old_state, new_state); 4354 /* NOTREACHED */ 4355 } 4356 4357 sd->sd_vol.sv_chunks[c]->src_meta.scm_status = new_state; 4358 sd->sd_set_vol_state(sd); 4359 4360 sd->sd_must_flush = 1; 4361 task_add(systq, &sd->sd_meta_save_task); 4362 done: 4363 splx(s); 4364 } 4365 4366 void 4367 sr_set_vol_state(struct sr_discipline *sd) 4368 { 4369 int states[SR_MAX_STATES]; 4370 int new_state, i, nd; 4371 int old_state = sd->sd_vol_status; 4372 u_int32_t s; 4373 4374 DNPRINTF(SR_D_STATE, "%s: %s: sr_set_vol_state\n", 4375 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname); 4376 4377 nd = sd->sd_meta->ssdi.ssd_chunk_no; 4378 4379 for (i = 0; i < SR_MAX_STATES; i++) 4380 states[i] = 0; 4381 4382 for (i = 0; i < nd; i++) { 4383 s = sd->sd_vol.sv_chunks[i]->src_meta.scm_status; 4384 if (s >= SR_MAX_STATES) 4385 panic("%s: %s: %s: invalid chunk state", 4386 DEVNAME(sd->sd_sc), 4387 sd->sd_meta->ssd_devname, 4388 sd->sd_vol.sv_chunks[i]->src_meta.scmi.scm_devname); 4389 states[s]++; 4390 } 4391 4392 if (states[BIOC_SDONLINE] == nd) 4393 new_state = BIOC_SVONLINE; 4394 else 4395 new_state = BIOC_SVOFFLINE; 4396 4397 DNPRINTF(SR_D_STATE, "%s: %s: sr_set_vol_state %d -> %d\n", 4398 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, 4399 old_state, new_state); 4400 4401 switch (old_state) { 4402 case BIOC_SVONLINE: 4403 if (new_state == BIOC_SVOFFLINE || new_state == BIOC_SVONLINE) 4404 break; 4405 else 4406 goto die; 4407 break; 4408 4409 case BIOC_SVOFFLINE: 4410 /* XXX this might be a little too much */ 4411 goto die; 4412 4413 default: 4414 die: 4415 panic("%s: %s: invalid volume state transition " 4416 "%d -> %d", DEVNAME(sd->sd_sc), 4417 sd->sd_meta->ssd_devname, 4418 old_state, new_state); 4419 /* NOTREACHED */ 4420 } 4421 4422 sd->sd_vol_status = new_state; 4423 } 4424 4425 void * 4426 sr_block_get(struct sr_discipline *sd, long length) 4427 { 4428 return dma_alloc(length, PR_NOWAIT | PR_ZERO); 4429 } 4430 4431 void 4432 sr_block_put(struct sr_discipline *sd, void *ptr, int length) 4433 { 4434 dma_free(ptr, length); 4435 } 4436 4437 void 4438 sr_checksum_print(u_int8_t *md5) 4439 { 4440 int i; 4441 4442 for (i = 0; i < MD5_DIGEST_LENGTH; i++) 4443 printf("%02x", md5[i]); 4444 } 4445 4446 void 4447 sr_checksum(struct sr_softc *sc, void *src, void *md5, u_int32_t len) 4448 { 4449 MD5_CTX ctx; 4450 4451 DNPRINTF(SR_D_MISC, "%s: sr_checksum(%p %p %d)\n", DEVNAME(sc), src, 4452 md5, len); 4453 4454 MD5Init(&ctx); 4455 MD5Update(&ctx, src, len); 4456 MD5Final(md5, &ctx); 4457 } 4458 4459 void 4460 sr_uuid_generate(struct sr_uuid *uuid) 4461 { 4462 arc4random_buf(uuid->sui_id, sizeof(uuid->sui_id)); 4463 /* UUID version 4: random */ 4464 uuid->sui_id[6] &= 0x0f; 4465 uuid->sui_id[6] |= 0x40; 4466 /* RFC4122 variant */ 4467 uuid->sui_id[8] &= 0x3f; 4468 uuid->sui_id[8] |= 0x80; 4469 } 4470 4471 char * 4472 sr_uuid_format(struct sr_uuid *uuid) 4473 { 4474 char *uuidstr; 4475 4476 uuidstr = malloc(37, M_DEVBUF, M_WAITOK); 4477 4478 snprintf(uuidstr, 37, 4479 "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-" 4480 "%02x%02x%02x%02x%02x%02x", 4481 uuid->sui_id[0], uuid->sui_id[1], 4482 uuid->sui_id[2], uuid->sui_id[3], 4483 uuid->sui_id[4], uuid->sui_id[5], 4484 uuid->sui_id[6], uuid->sui_id[7], 4485 uuid->sui_id[8], uuid->sui_id[9], 4486 uuid->sui_id[10], uuid->sui_id[11], 4487 uuid->sui_id[12], uuid->sui_id[13], 4488 uuid->sui_id[14], uuid->sui_id[15]); 4489 4490 return uuidstr; 4491 } 4492 4493 void 4494 sr_uuid_print(struct sr_uuid *uuid, int cr) 4495 { 4496 char *uuidstr; 4497 4498 uuidstr = sr_uuid_format(uuid); 4499 printf("%s%s", uuidstr, (cr ? "\n" : "")); 4500 free(uuidstr, M_DEVBUF, 37); 4501 } 4502 4503 int 4504 sr_already_assembled(struct sr_discipline *sd) 4505 { 4506 struct sr_softc *sc = sd->sd_sc; 4507 struct sr_discipline *sdtmp; 4508 4509 TAILQ_FOREACH(sdtmp, &sc->sc_dis_list, sd_link) { 4510 if (!bcmp(&sd->sd_meta->ssdi.ssd_uuid, 4511 &sdtmp->sd_meta->ssdi.ssd_uuid, 4512 sizeof(sd->sd_meta->ssdi.ssd_uuid))) 4513 return (1); 4514 } 4515 4516 return (0); 4517 } 4518 4519 int32_t 4520 sr_validate_stripsize(u_int32_t b) 4521 { 4522 int s = 0; 4523 4524 if (b % DEV_BSIZE) 4525 return (-1); 4526 4527 while ((b & 1) == 0) { 4528 b >>= 1; 4529 s++; 4530 } 4531 4532 /* only multiple of twos */ 4533 b >>= 1; 4534 if (b) 4535 return(-1); 4536 4537 return (s); 4538 } 4539 4540 void 4541 sr_quiesce(void) 4542 { 4543 struct sr_softc *sc = softraid0; 4544 struct sr_discipline *sd, *nsd; 4545 4546 /* Shutdown disciplines in reverse attach order. */ 4547 TAILQ_FOREACH_REVERSE_SAFE(sd, &sc->sc_dis_list, 4548 sr_discipline_list, sd_link, nsd) 4549 sr_discipline_shutdown(sd, 1, -1); 4550 } 4551 4552 void 4553 sr_shutdown(int dying) 4554 { 4555 struct sr_softc *sc = softraid0; 4556 struct sr_discipline *sd; 4557 4558 DNPRINTF(SR_D_MISC, "%s: sr_shutdown\n", DEVNAME(sc)); 4559 4560 /* 4561 * Since softraid is not under mainbus, we have to explicitly 4562 * notify its children that the power is going down, so they 4563 * can execute their shutdown hooks. 4564 */ 4565 config_suspend((struct device *)sc, DVACT_POWERDOWN); 4566 4567 /* Shutdown disciplines in reverse attach order. */ 4568 while ((sd = TAILQ_LAST(&sc->sc_dis_list, sr_discipline_list)) != NULL) 4569 sr_discipline_shutdown(sd, 1, dying); 4570 } 4571 4572 int 4573 sr_validate_io(struct sr_workunit *wu, daddr_t *blkno, char *func) 4574 { 4575 struct sr_discipline *sd = wu->swu_dis; 4576 struct scsi_xfer *xs = wu->swu_xs; 4577 int rv = 1; 4578 4579 DNPRINTF(SR_D_DIS, "%s: %s 0x%02x\n", DEVNAME(sd->sd_sc), func, 4580 xs->cmd->opcode); 4581 4582 if (sd->sd_meta->ssd_data_blkno == 0) 4583 panic("invalid data blkno"); 4584 4585 if (sd->sd_vol_status == BIOC_SVOFFLINE) { 4586 DNPRINTF(SR_D_DIS, "%s: %s device offline\n", 4587 DEVNAME(sd->sd_sc), func); 4588 goto bad; 4589 } 4590 4591 if (xs->datalen == 0) { 4592 printf("%s: %s: illegal block count for %s\n", 4593 DEVNAME(sd->sd_sc), func, sd->sd_meta->ssd_devname); 4594 goto bad; 4595 } 4596 4597 if (xs->cmdlen == 10) 4598 *blkno = _4btol(((struct scsi_rw_big *)xs->cmd)->addr); 4599 else if (xs->cmdlen == 16) 4600 *blkno = _8btol(((struct scsi_rw_16 *)xs->cmd)->addr); 4601 else if (xs->cmdlen == 6) 4602 *blkno = _3btol(((struct scsi_rw *)xs->cmd)->addr); 4603 else { 4604 printf("%s: %s: illegal cmdlen for %s\n", 4605 DEVNAME(sd->sd_sc), func, sd->sd_meta->ssd_devname); 4606 goto bad; 4607 } 4608 4609 *blkno *= (sd->sd_meta->ssdi.ssd_secsize / DEV_BSIZE); 4610 4611 wu->swu_blk_start = *blkno; 4612 wu->swu_blk_end = *blkno + (xs->datalen >> DEV_BSHIFT) - 1; 4613 4614 if (wu->swu_blk_end > sd->sd_meta->ssdi.ssd_size) { 4615 DNPRINTF(SR_D_DIS, "%s: %s out of bounds start: %lld " 4616 "end: %lld length: %d\n", 4617 DEVNAME(sd->sd_sc), func, (long long)wu->swu_blk_start, 4618 (long long)wu->swu_blk_end, xs->datalen); 4619 4620 sd->sd_scsi_sense.error_code = SSD_ERRCODE_CURRENT | 4621 SSD_ERRCODE_VALID; 4622 sd->sd_scsi_sense.flags = SKEY_ILLEGAL_REQUEST; 4623 sd->sd_scsi_sense.add_sense_code = 0x21; 4624 sd->sd_scsi_sense.add_sense_code_qual = 0x00; 4625 sd->sd_scsi_sense.extra_len = 4; 4626 goto bad; 4627 } 4628 4629 rv = 0; 4630 bad: 4631 return (rv); 4632 } 4633 4634 void 4635 sr_rebuild_start(void *arg) 4636 { 4637 struct sr_discipline *sd = arg; 4638 struct sr_softc *sc = sd->sd_sc; 4639 4640 DNPRINTF(SR_D_REBUILD, "%s: %s starting rebuild thread\n", 4641 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname); 4642 4643 if (kthread_create(sr_rebuild_thread, sd, &sd->sd_background_proc, 4644 DEVNAME(sc)) != 0) 4645 printf("%s: unable to start background operation\n", 4646 DEVNAME(sc)); 4647 } 4648 4649 void 4650 sr_rebuild_thread(void *arg) 4651 { 4652 struct sr_discipline *sd = arg; 4653 4654 DNPRINTF(SR_D_REBUILD, "%s: %s rebuild thread started\n", 4655 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname); 4656 4657 sd->sd_reb_active = 1; 4658 sd->sd_rebuild(sd); 4659 sd->sd_reb_active = 0; 4660 4661 kthread_exit(0); 4662 } 4663 4664 void 4665 sr_rebuild(struct sr_discipline *sd) 4666 { 4667 struct sr_softc *sc = sd->sd_sc; 4668 u_int64_t sz, whole_blk, partial_blk, blk, restart; 4669 daddr_t lba; 4670 struct sr_workunit *wu_r, *wu_w; 4671 struct scsi_xfer xs_r, xs_w; 4672 struct scsi_rw_16 *cr, *cw; 4673 int c, s, slept, percent = 0, old_percent = -1; 4674 u_int8_t *buf; 4675 4676 whole_blk = sd->sd_meta->ssdi.ssd_size / SR_REBUILD_IO_SIZE; 4677 partial_blk = sd->sd_meta->ssdi.ssd_size % SR_REBUILD_IO_SIZE; 4678 4679 restart = sd->sd_meta->ssd_rebuild / SR_REBUILD_IO_SIZE; 4680 if (restart > whole_blk) { 4681 printf("%s: bogus rebuild restart offset, starting from 0\n", 4682 DEVNAME(sc)); 4683 restart = 0; 4684 } 4685 if (restart) { 4686 /* 4687 * XXX there is a hole here; there is a posibility that we 4688 * had a restart however the chunk that was supposed to 4689 * be rebuilt is no longer valid; we can reach this situation 4690 * when a rebuild is in progress and the box crashes and 4691 * on reboot the rebuild chunk is different (like zero'd or 4692 * replaced). We need to check the uuid of the chunk that is 4693 * being rebuilt to assert this. 4694 */ 4695 percent = sr_rebuild_percent(sd); 4696 printf("%s: resuming rebuild on %s at %d%%\n", 4697 DEVNAME(sc), sd->sd_meta->ssd_devname, percent); 4698 } 4699 4700 /* currently this is 64k therefore we can use dma_alloc */ 4701 buf = dma_alloc(SR_REBUILD_IO_SIZE << DEV_BSHIFT, PR_WAITOK); 4702 for (blk = restart; blk <= whole_blk; blk++) { 4703 lba = blk * SR_REBUILD_IO_SIZE; 4704 sz = SR_REBUILD_IO_SIZE; 4705 if (blk == whole_blk) { 4706 if (partial_blk == 0) 4707 break; 4708 sz = partial_blk; 4709 } 4710 4711 /* get some wu */ 4712 wu_r = sr_scsi_wu_get(sd, 0); 4713 wu_w = sr_scsi_wu_get(sd, 0); 4714 4715 DNPRINTF(SR_D_REBUILD, "%s: %s rebuild wu_r %p, wu_w %p\n", 4716 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, wu_r, wu_w); 4717 4718 /* setup read io */ 4719 bzero(&xs_r, sizeof xs_r); 4720 xs_r.error = XS_NOERROR; 4721 xs_r.flags = SCSI_DATA_IN; 4722 xs_r.datalen = sz << DEV_BSHIFT; 4723 xs_r.data = buf; 4724 xs_r.cmdlen = sizeof(*cr); 4725 xs_r.cmd = &xs_r.cmdstore; 4726 cr = (struct scsi_rw_16 *)xs_r.cmd; 4727 cr->opcode = READ_16; 4728 _lto4b(sz, cr->length); 4729 _lto8b(lba, cr->addr); 4730 wu_r->swu_state = SR_WU_CONSTRUCT; 4731 wu_r->swu_flags |= SR_WUF_REBUILD; 4732 wu_r->swu_xs = &xs_r; 4733 if (sd->sd_scsi_rw(wu_r)) { 4734 printf("%s: could not create read io\n", 4735 DEVNAME(sc)); 4736 goto fail; 4737 } 4738 4739 /* setup write io */ 4740 bzero(&xs_w, sizeof xs_w); 4741 xs_w.error = XS_NOERROR; 4742 xs_w.flags = SCSI_DATA_OUT; 4743 xs_w.datalen = sz << DEV_BSHIFT; 4744 xs_w.data = buf; 4745 xs_w.cmdlen = sizeof(*cw); 4746 xs_w.cmd = &xs_w.cmdstore; 4747 cw = (struct scsi_rw_16 *)xs_w.cmd; 4748 cw->opcode = WRITE_16; 4749 _lto4b(sz, cw->length); 4750 _lto8b(lba, cw->addr); 4751 wu_w->swu_state = SR_WU_CONSTRUCT; 4752 wu_w->swu_flags |= SR_WUF_REBUILD | SR_WUF_WAKEUP; 4753 wu_w->swu_xs = &xs_w; 4754 if (sd->sd_scsi_rw(wu_w)) { 4755 printf("%s: could not create write io\n", 4756 DEVNAME(sc)); 4757 goto fail; 4758 } 4759 4760 /* 4761 * collide with the read io so that we get automatically 4762 * started when the read is done 4763 */ 4764 wu_w->swu_state = SR_WU_DEFERRED; 4765 wu_r->swu_collider = wu_w; 4766 s = splbio(); 4767 TAILQ_INSERT_TAIL(&sd->sd_wu_defq, wu_w, swu_link); 4768 splx(s); 4769 4770 DNPRINTF(SR_D_REBUILD, "%s: %s rebuild scheduling wu_r %p\n", 4771 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, wu_r); 4772 4773 wu_r->swu_state = SR_WU_INPROGRESS; 4774 sr_schedule_wu(wu_r); 4775 4776 /* wait for write completion */ 4777 slept = 0; 4778 while ((wu_w->swu_flags & SR_WUF_REBUILDIOCOMP) == 0) { 4779 tsleep_nsec(wu_w, PRIBIO, "sr_rebuild", INFSLP); 4780 slept = 1; 4781 } 4782 /* yield if we didn't sleep */ 4783 if (slept == 0) 4784 tsleep(sc, PWAIT, "sr_yield", 1); 4785 4786 sr_scsi_wu_put(sd, wu_r); 4787 sr_scsi_wu_put(sd, wu_w); 4788 4789 sd->sd_meta->ssd_rebuild = lba; 4790 4791 /* XXX - this should be based on size, not percentage. */ 4792 /* save metadata every percent */ 4793 percent = sr_rebuild_percent(sd); 4794 if (percent != old_percent && blk != whole_blk) { 4795 if (sr_meta_save(sd, SR_META_DIRTY)) 4796 printf("%s: could not save metadata to %s\n", 4797 DEVNAME(sc), sd->sd_meta->ssd_devname); 4798 old_percent = percent; 4799 } 4800 4801 if (sd->sd_reb_abort) 4802 goto abort; 4803 } 4804 4805 /* all done */ 4806 sd->sd_meta->ssd_rebuild = 0; 4807 for (c = 0; c < sd->sd_meta->ssdi.ssd_chunk_no; c++) { 4808 if (sd->sd_vol.sv_chunks[c]->src_meta.scm_status == 4809 BIOC_SDREBUILD) { 4810 sd->sd_set_chunk_state(sd, c, BIOC_SDONLINE); 4811 break; 4812 } 4813 } 4814 4815 abort: 4816 if (sr_meta_save(sd, SR_META_DIRTY)) 4817 printf("%s: could not save metadata to %s\n", 4818 DEVNAME(sc), sd->sd_meta->ssd_devname); 4819 fail: 4820 dma_free(buf, SR_REBUILD_IO_SIZE << DEV_BSHIFT); 4821 } 4822 4823 #ifndef SMALL_KERNEL 4824 int 4825 sr_sensors_create(struct sr_discipline *sd) 4826 { 4827 struct sr_softc *sc = sd->sd_sc; 4828 int rv = 1; 4829 4830 DNPRINTF(SR_D_STATE, "%s: %s: sr_sensors_create\n", 4831 DEVNAME(sc), sd->sd_meta->ssd_devname); 4832 4833 sd->sd_vol.sv_sensor.type = SENSOR_DRIVE; 4834 sd->sd_vol.sv_sensor.status = SENSOR_S_UNKNOWN; 4835 strlcpy(sd->sd_vol.sv_sensor.desc, sd->sd_meta->ssd_devname, 4836 sizeof(sd->sd_vol.sv_sensor.desc)); 4837 4838 sensor_attach(&sc->sc_sensordev, &sd->sd_vol.sv_sensor); 4839 sd->sd_vol.sv_sensor_attached = 1; 4840 4841 if (sc->sc_sensor_task == NULL) { 4842 sc->sc_sensor_task = sensor_task_register(sc, 4843 sr_sensors_refresh, 10); 4844 if (sc->sc_sensor_task == NULL) 4845 goto bad; 4846 } 4847 4848 rv = 0; 4849 bad: 4850 return (rv); 4851 } 4852 4853 void 4854 sr_sensors_delete(struct sr_discipline *sd) 4855 { 4856 DNPRINTF(SR_D_STATE, "%s: sr_sensors_delete\n", DEVNAME(sd->sd_sc)); 4857 4858 if (sd->sd_vol.sv_sensor_attached) 4859 sensor_detach(&sd->sd_sc->sc_sensordev, &sd->sd_vol.sv_sensor); 4860 } 4861 4862 void 4863 sr_sensors_refresh(void *arg) 4864 { 4865 struct sr_softc *sc = arg; 4866 struct sr_volume *sv; 4867 struct sr_discipline *sd; 4868 4869 DNPRINTF(SR_D_STATE, "%s: sr_sensors_refresh\n", DEVNAME(sc)); 4870 4871 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 4872 sv = &sd->sd_vol; 4873 4874 switch(sd->sd_vol_status) { 4875 case BIOC_SVOFFLINE: 4876 sv->sv_sensor.value = SENSOR_DRIVE_FAIL; 4877 sv->sv_sensor.status = SENSOR_S_CRIT; 4878 break; 4879 4880 case BIOC_SVDEGRADED: 4881 sv->sv_sensor.value = SENSOR_DRIVE_PFAIL; 4882 sv->sv_sensor.status = SENSOR_S_WARN; 4883 break; 4884 4885 case BIOC_SVREBUILD: 4886 sv->sv_sensor.value = SENSOR_DRIVE_REBUILD; 4887 sv->sv_sensor.status = SENSOR_S_WARN; 4888 break; 4889 4890 case BIOC_SVSCRUB: 4891 case BIOC_SVONLINE: 4892 sv->sv_sensor.value = SENSOR_DRIVE_ONLINE; 4893 sv->sv_sensor.status = SENSOR_S_OK; 4894 break; 4895 4896 default: 4897 sv->sv_sensor.value = 0; /* unknown */ 4898 sv->sv_sensor.status = SENSOR_S_UNKNOWN; 4899 } 4900 } 4901 } 4902 #endif /* SMALL_KERNEL */ 4903 4904 #ifdef SR_FANCY_STATS 4905 void sr_print_stats(void); 4906 4907 void 4908 sr_print_stats(void) 4909 { 4910 struct sr_softc *sc = softraid0; 4911 struct sr_discipline *sd; 4912 4913 if (sc == NULL) { 4914 printf("no softraid softc found\n"); 4915 return; 4916 } 4917 4918 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 4919 printf("%s: ios pending %d, collisions %llu\n", 4920 sd->sd_meta->ssd_devname, 4921 sd->sd_wu_pending, 4922 sd->sd_wu_collisions); 4923 } 4924 } 4925 #endif /* SR_FANCY_STATS */ 4926 4927 #ifdef SR_DEBUG 4928 void 4929 sr_meta_print(struct sr_metadata *m) 4930 { 4931 int i; 4932 struct sr_meta_chunk *mc; 4933 struct sr_meta_opt_hdr *omh; 4934 4935 if (!(sr_debug & SR_D_META)) 4936 return; 4937 4938 printf("\tssd_magic 0x%llx\n", m->ssdi.ssd_magic); 4939 printf("\tssd_version %d\n", m->ssdi.ssd_version); 4940 printf("\tssd_vol_flags 0x%x\n", m->ssdi.ssd_vol_flags); 4941 printf("\tssd_uuid "); 4942 sr_uuid_print(&m->ssdi.ssd_uuid, 1); 4943 printf("\tssd_chunk_no %d\n", m->ssdi.ssd_chunk_no); 4944 printf("\tssd_chunk_id %d\n", m->ssdi.ssd_chunk_id); 4945 printf("\tssd_opt_no %d\n", m->ssdi.ssd_opt_no); 4946 printf("\tssd_volid %d\n", m->ssdi.ssd_volid); 4947 printf("\tssd_level %d\n", m->ssdi.ssd_level); 4948 printf("\tssd_size %lld\n", m->ssdi.ssd_size); 4949 printf("\tssd_devname %s\n", m->ssd_devname); 4950 printf("\tssd_vendor %s\n", m->ssdi.ssd_vendor); 4951 printf("\tssd_product %s\n", m->ssdi.ssd_product); 4952 printf("\tssd_revision %s\n", m->ssdi.ssd_revision); 4953 printf("\tssd_strip_size %d\n", m->ssdi.ssd_strip_size); 4954 printf("\tssd_checksum "); 4955 sr_checksum_print(m->ssd_checksum); 4956 printf("\n"); 4957 printf("\tssd_meta_flags 0x%x\n", m->ssd_meta_flags); 4958 printf("\tssd_ondisk %llu\n", m->ssd_ondisk); 4959 4960 mc = (struct sr_meta_chunk *)(m + 1); 4961 for (i = 0; i < m->ssdi.ssd_chunk_no; i++, mc++) { 4962 printf("\t\tscm_volid %d\n", mc->scmi.scm_volid); 4963 printf("\t\tscm_chunk_id %d\n", mc->scmi.scm_chunk_id); 4964 printf("\t\tscm_devname %s\n", mc->scmi.scm_devname); 4965 printf("\t\tscm_size %lld\n", mc->scmi.scm_size); 4966 printf("\t\tscm_coerced_size %lld\n",mc->scmi.scm_coerced_size); 4967 printf("\t\tscm_uuid "); 4968 sr_uuid_print(&mc->scmi.scm_uuid, 1); 4969 printf("\t\tscm_checksum "); 4970 sr_checksum_print(mc->scm_checksum); 4971 printf("\n"); 4972 printf("\t\tscm_status %d\n", mc->scm_status); 4973 } 4974 4975 omh = (struct sr_meta_opt_hdr *)((u_int8_t *)(m + 1) + 4976 sizeof(struct sr_meta_chunk) * m->ssdi.ssd_chunk_no); 4977 for (i = 0; i < m->ssdi.ssd_opt_no; i++) { 4978 printf("\t\t\tsom_type %d\n", omh->som_type); 4979 printf("\t\t\tsom_checksum "); 4980 sr_checksum_print(omh->som_checksum); 4981 printf("\n"); 4982 omh = (struct sr_meta_opt_hdr *)((void *)omh + 4983 omh->som_length); 4984 } 4985 } 4986 4987 void 4988 sr_dump_block(void *blk, int len) 4989 { 4990 uint8_t *b = blk; 4991 int i, j, c; 4992 4993 for (i = 0; i < len; i += 16) { 4994 for (j = 0; j < 16; j++) 4995 printf("%.2x ", b[i + j]); 4996 printf(" "); 4997 for (j = 0; j < 16; j++) { 4998 c = b[i + j]; 4999 if (c < ' ' || c > 'z' || i + j > len) 5000 c = '.'; 5001 printf("%c", c); 5002 } 5003 printf("\n"); 5004 } 5005 } 5006 5007 void 5008 sr_dump_mem(u_int8_t *p, int len) 5009 { 5010 int i; 5011 5012 for (i = 0; i < len; i++) 5013 printf("%02x ", *p++); 5014 printf("\n"); 5015 } 5016 5017 #endif /* SR_DEBUG */ 5018 5019 #ifdef HIBERNATE 5020 /* 5021 * Side-effect free (no malloc, printf, pool, splx) softraid crypto writer. 5022 * 5023 * This function must perform the following: 5024 * 1. Determine the underlying device's own side-effect free I/O function 5025 * (eg, ahci_hibernate_io, wd_hibernate_io, etc). 5026 * 2. Store enough information in the provided page argument for subsequent 5027 * I/O calls (such as the crypto discipline structure for the keys, the 5028 * offset of the softraid partition on the underlying disk, as well as 5029 * the offset of the swap partition within the crypto volume. 5030 * 3. Encrypt the incoming data using the sr_discipline keys, then pass 5031 * the request to the underlying device's own I/O function. 5032 */ 5033 int 5034 sr_hibernate_io(dev_t dev, daddr_t blkno, vaddr_t addr, size_t size, int op, void *page) 5035 { 5036 /* Struct for stashing data obtained on HIB_INIT. 5037 * XXX 5038 * We share the page with the underlying device's own 5039 * side-effect free I/O function, so we pad our data to 5040 * the end of the page. Presently this does not overlap 5041 * with either of the two other side-effect free i/o 5042 * functions (ahci/wd). 5043 */ 5044 struct { 5045 char pad[3072]; 5046 struct sr_discipline *srd; 5047 hibio_fn subfn; /* underlying device i/o fn */ 5048 dev_t subdev; /* underlying device dev_t */ 5049 daddr_t sr_swapoff; /* ofs of swap part in sr volume */ 5050 char buf[DEV_BSIZE]; /* encryption performed into this buf */ 5051 } *my = page; 5052 extern struct cfdriver sd_cd; 5053 char errstr[128], *dl_ret; 5054 struct sr_chunk *schunk; 5055 struct sd_softc *sd; 5056 struct aes_xts_ctx ctx; 5057 struct sr_softc *sc; 5058 struct device *dv; 5059 daddr_t key_blkno; 5060 uint32_t sub_raidoff; /* ofs of sr part in underlying dev */ 5061 struct disklabel dl; 5062 struct partition *pp; 5063 size_t i, j; 5064 u_char iv[8]; 5065 5066 /* 5067 * In HIB_INIT, we are passed the swap partition size and offset 5068 * in 'size' and 'blkno' respectively. These are relative to the 5069 * start of the softraid partition, and we need to save these 5070 * for later translation to the underlying device's layout. 5071 */ 5072 if (op == HIB_INIT) { 5073 dv = disk_lookup(&sd_cd, DISKUNIT(dev)); 5074 sd = (struct sd_softc *)dv; 5075 sc = (struct sr_softc *)dv->dv_parent->dv_parent; 5076 5077 /* 5078 * Look up the sr discipline. This is used to determine 5079 * if we are SR crypto and what the underlying device is. 5080 */ 5081 my->srd = sc->sc_targets[sd->sc_link->target]; 5082 DNPRINTF(SR_D_MISC, "sr_hibernate_io: discipline is %s\n", 5083 my->srd->sd_name); 5084 if (strncmp(my->srd->sd_name, "CRYPTO", 5085 sizeof(my->srd->sd_name))) 5086 return (ENOTSUP); 5087 5088 /* Find the underlying device */ 5089 schunk = my->srd->sd_vol.sv_chunks[0]; 5090 my->subdev = schunk->src_dev_mm; 5091 5092 /* 5093 * Find the appropriate underlying device side effect free 5094 * I/O function, based on the type of device it is. 5095 */ 5096 my->subfn = get_hibernate_io_function(my->subdev); 5097 if (!my->subfn) 5098 return (ENODEV); 5099 5100 /* 5101 * Find blkno where this raid partition starts on 5102 * the underlying disk. 5103 */ 5104 dl_ret = disk_readlabel(&dl, my->subdev, errstr, 5105 sizeof(errstr)); 5106 if (dl_ret) { 5107 printf("Hibernate error reading disklabel: %s\n", dl_ret); 5108 return (ENOTSUP); 5109 } 5110 5111 pp = &dl.d_partitions[DISKPART(my->subdev)]; 5112 if (pp->p_fstype != FS_RAID || DL_GETPSIZE(pp) == 0) 5113 return (ENOTSUP); 5114 5115 /* Find the blkno of the SR part in the underlying device */ 5116 sub_raidoff = my->srd->sd_meta->ssd_data_blkno + 5117 DL_SECTOBLK(&dl, DL_GETPOFFSET(pp)); 5118 DNPRINTF(SR_D_MISC,"sr_hibernate_io: blk trans ofs: %d blks\n", 5119 sub_raidoff); 5120 5121 /* Save the blkno of the swap partition in the SR disk */ 5122 my->sr_swapoff = blkno; 5123 5124 /* Initialize the sub-device */ 5125 return my->subfn(my->subdev, sub_raidoff + blkno, 5126 addr, size, op, page); 5127 } 5128 5129 /* Hibernate only uses (and we only support) writes */ 5130 if (op != HIB_W) 5131 return (ENOTSUP); 5132 5133 /* 5134 * Blocks act as the IV for the encryption. These block numbers 5135 * are relative to the start of the sr partition, but the 'blkno' 5136 * passed above is relative to the start of the swap partition 5137 * inside the sr partition, so bias appropriately. 5138 */ 5139 key_blkno = my->sr_swapoff + blkno; 5140 5141 /* Process each disk block one at a time. */ 5142 for (i = 0; i < size; i += DEV_BSIZE) { 5143 int res; 5144 5145 bzero(&ctx, sizeof(ctx)); 5146 5147 /* 5148 * Set encryption key (from the sr discipline stashed 5149 * during HIB_INIT. This code is based on the softraid 5150 * bootblock code. 5151 */ 5152 aes_xts_setkey(&ctx, my->srd->mds.mdd_crypto.scr_key[0], 64); 5153 /* We encrypt DEV_BSIZE bytes at a time in my->buf */ 5154 memcpy(my->buf, ((char *)addr) + i, DEV_BSIZE); 5155 5156 /* Block number is the IV */ 5157 memcpy(&iv, &key_blkno, sizeof(key_blkno)); 5158 aes_xts_reinit(&ctx, iv); 5159 5160 /* Encrypt DEV_BSIZE bytes, AES_XTS_BLOCKSIZE bytes at a time */ 5161 for (j = 0; j < DEV_BSIZE; j += AES_XTS_BLOCKSIZE) 5162 aes_xts_encrypt(&ctx, my->buf + j); 5163 5164 /* 5165 * Write one block out from my->buf to the underlying device 5166 * using its own side-effect free I/O function. 5167 */ 5168 res = my->subfn(my->subdev, blkno + (i / DEV_BSIZE), 5169 (vaddr_t)(my->buf), DEV_BSIZE, op, page); 5170 if (res != 0) 5171 return (res); 5172 key_blkno++; 5173 } 5174 return (0); 5175 } 5176 #endif /* HIBERNATE */ 5177