1 /* $OpenBSD: softraid.c,v 1.425 2022/04/16 19:19:58 naddy 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 const struct cfattach softraid_ca = { 87 sizeof(struct sr_softc), sr_match, sr_attach, sr_detach, 88 }; 89 90 struct cfdriver softraid_cd = { 91 NULL, "softraid", DV_DULL 92 }; 93 94 /* scsi & discipline */ 95 void sr_scsi_cmd(struct scsi_xfer *); 96 int sr_scsi_probe(struct scsi_link *); 97 int sr_scsi_ioctl(struct scsi_link *, u_long, 98 caddr_t, int); 99 int sr_bio_ioctl(struct device *, u_long, caddr_t); 100 int sr_bio_handler(struct sr_softc *, 101 struct sr_discipline *, u_long, struct bio *); 102 int sr_ioctl_inq(struct sr_softc *, struct bioc_inq *); 103 int sr_ioctl_vol(struct sr_softc *, struct bioc_vol *); 104 int sr_ioctl_disk(struct sr_softc *, struct bioc_disk *); 105 int sr_ioctl_setstate(struct sr_softc *, 106 struct bioc_setstate *); 107 int sr_ioctl_createraid(struct sr_softc *, 108 struct bioc_createraid *, int, void *); 109 int sr_ioctl_deleteraid(struct sr_softc *, 110 struct sr_discipline *, struct bioc_deleteraid *); 111 int sr_ioctl_discipline(struct sr_softc *, 112 struct sr_discipline *, struct bioc_discipline *); 113 int sr_ioctl_installboot(struct sr_softc *, 114 struct sr_discipline *, struct bioc_installboot *); 115 void sr_chunks_unwind(struct sr_softc *, 116 struct sr_chunk_head *); 117 void sr_discipline_free(struct sr_discipline *); 118 void sr_discipline_shutdown(struct sr_discipline *, int, int); 119 int sr_discipline_init(struct sr_discipline *, int); 120 int sr_alloc_resources(struct sr_discipline *); 121 void sr_free_resources(struct sr_discipline *); 122 void sr_set_chunk_state(struct sr_discipline *, int, int); 123 void sr_set_vol_state(struct sr_discipline *); 124 125 /* utility functions */ 126 void sr_shutdown(int); 127 void sr_uuid_generate(struct sr_uuid *); 128 char *sr_uuid_format(struct sr_uuid *); 129 void sr_uuid_print(struct sr_uuid *, int); 130 void sr_checksum_print(u_int8_t *); 131 int sr_boot_assembly(struct sr_softc *); 132 int sr_already_assembled(struct sr_discipline *); 133 int sr_hotspare(struct sr_softc *, dev_t); 134 void sr_hotspare_rebuild(struct sr_discipline *); 135 int sr_rebuild_init(struct sr_discipline *, dev_t, int); 136 void sr_rebuild_start(void *); 137 void sr_rebuild_thread(void *); 138 void sr_rebuild(struct sr_discipline *); 139 void sr_roam_chunks(struct sr_discipline *); 140 int sr_chunk_in_use(struct sr_softc *, dev_t); 141 int sr_rw(struct sr_softc *, dev_t, char *, size_t, 142 daddr_t, long); 143 void sr_wu_done_callback(void *); 144 145 /* don't include these on RAMDISK */ 146 #ifndef SMALL_KERNEL 147 void sr_sensors_refresh(void *); 148 int sr_sensors_create(struct sr_discipline *); 149 void sr_sensors_delete(struct sr_discipline *); 150 #endif 151 152 /* metadata */ 153 int sr_meta_probe(struct sr_discipline *, dev_t *, int); 154 int sr_meta_attach(struct sr_discipline *, int, int); 155 int sr_meta_rw(struct sr_discipline *, dev_t, void *, long); 156 int sr_meta_clear(struct sr_discipline *); 157 void sr_meta_init(struct sr_discipline *, int, int); 158 void sr_meta_init_complete(struct sr_discipline *); 159 void sr_meta_opt_handler(struct sr_discipline *, 160 struct sr_meta_opt_hdr *); 161 162 /* hotplug magic */ 163 void sr_disk_attach(struct disk *, int); 164 165 struct sr_hotplug_list { 166 void (*sh_hotplug)(struct sr_discipline *, 167 struct disk *, int); 168 struct sr_discipline *sh_sd; 169 170 SLIST_ENTRY(sr_hotplug_list) shl_link; 171 }; 172 SLIST_HEAD(sr_hotplug_list_head, sr_hotplug_list); 173 174 struct sr_hotplug_list_head sr_hotplug_callbacks; 175 extern void (*softraid_disk_attach)(struct disk *, int); 176 177 /* scsi glue */ 178 const struct scsi_adapter sr_switch = { 179 sr_scsi_cmd, NULL, sr_scsi_probe, NULL, sr_scsi_ioctl 180 }; 181 182 /* native metadata format */ 183 int sr_meta_native_bootprobe(struct sr_softc *, dev_t, 184 struct sr_boot_chunk_head *); 185 #define SR_META_NOTCLAIMED (0) 186 #define SR_META_CLAIMED (1) 187 int sr_meta_native_probe(struct sr_softc *, 188 struct sr_chunk *); 189 int sr_meta_native_attach(struct sr_discipline *, int); 190 int sr_meta_native_write(struct sr_discipline *, dev_t, 191 struct sr_metadata *,void *); 192 193 #ifdef SR_DEBUG 194 void sr_meta_print(struct sr_metadata *); 195 #else 196 #define sr_meta_print(m) 197 #endif 198 199 /* the metadata driver should remain stateless */ 200 struct sr_meta_driver { 201 daddr_t smd_offset; /* metadata location */ 202 u_int32_t smd_size; /* size of metadata */ 203 204 int (*smd_probe)(struct sr_softc *, 205 struct sr_chunk *); 206 int (*smd_attach)(struct sr_discipline *, int); 207 int (*smd_detach)(struct sr_discipline *); 208 int (*smd_read)(struct sr_discipline *, dev_t, 209 struct sr_metadata *, void *); 210 int (*smd_write)(struct sr_discipline *, dev_t, 211 struct sr_metadata *, void *); 212 int (*smd_validate)(struct sr_discipline *, 213 struct sr_metadata *, void *); 214 } smd[] = { 215 { SR_META_OFFSET, SR_META_SIZE * DEV_BSIZE, 216 sr_meta_native_probe, sr_meta_native_attach, NULL, 217 sr_meta_native_read, sr_meta_native_write, NULL }, 218 { 0, 0, NULL, NULL, NULL, NULL } 219 }; 220 221 int 222 sr_meta_attach(struct sr_discipline *sd, int chunk_no, int force) 223 { 224 struct sr_softc *sc = sd->sd_sc; 225 struct sr_chunk_head *cl; 226 struct sr_chunk *ch_entry, *chunk1, *chunk2; 227 int rv = 1, i = 0; 228 229 DNPRINTF(SR_D_META, "%s: sr_meta_attach(%d)\n", DEVNAME(sc), chunk_no); 230 231 /* in memory copy of metadata */ 232 sd->sd_meta = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, 233 M_ZERO | M_NOWAIT); 234 if (!sd->sd_meta) { 235 sr_error(sc, "could not allocate memory for metadata"); 236 goto bad; 237 } 238 239 if (sd->sd_meta_type != SR_META_F_NATIVE) { 240 /* in memory copy of foreign metadata */ 241 sd->sd_meta_foreign = malloc(smd[sd->sd_meta_type].smd_size, 242 M_DEVBUF, M_ZERO | M_NOWAIT); 243 if (!sd->sd_meta_foreign) { 244 /* unwind frees sd_meta */ 245 sr_error(sc, "could not allocate memory for foreign " 246 "metadata"); 247 goto bad; 248 } 249 } 250 251 /* we have a valid list now create an array index */ 252 cl = &sd->sd_vol.sv_chunk_list; 253 sd->sd_vol.sv_chunks = mallocarray(chunk_no, sizeof(struct sr_chunk *), 254 M_DEVBUF, M_WAITOK | M_ZERO); 255 256 /* fill out chunk array */ 257 i = 0; 258 SLIST_FOREACH(ch_entry, cl, src_link) 259 sd->sd_vol.sv_chunks[i++] = ch_entry; 260 261 /* attach metadata */ 262 if (smd[sd->sd_meta_type].smd_attach(sd, force)) 263 goto bad; 264 265 /* Force chunks into correct order now that metadata is attached. */ 266 SLIST_INIT(cl); 267 for (i = 0; i < chunk_no; i++) { 268 ch_entry = sd->sd_vol.sv_chunks[i]; 269 chunk2 = NULL; 270 SLIST_FOREACH(chunk1, cl, src_link) { 271 if (chunk1->src_meta.scmi.scm_chunk_id > 272 ch_entry->src_meta.scmi.scm_chunk_id) 273 break; 274 chunk2 = chunk1; 275 } 276 if (chunk2 == NULL) 277 SLIST_INSERT_HEAD(cl, ch_entry, src_link); 278 else 279 SLIST_INSERT_AFTER(chunk2, ch_entry, src_link); 280 } 281 i = 0; 282 SLIST_FOREACH(ch_entry, cl, src_link) 283 sd->sd_vol.sv_chunks[i++] = ch_entry; 284 285 rv = 0; 286 bad: 287 return (rv); 288 } 289 290 int 291 sr_meta_probe(struct sr_discipline *sd, dev_t *dt, int no_chunk) 292 { 293 struct sr_softc *sc = sd->sd_sc; 294 struct vnode *vn; 295 struct sr_chunk *ch_entry, *ch_prev = NULL; 296 struct sr_chunk_head *cl; 297 char devname[32]; 298 int i, d, type, found, prevf, error; 299 dev_t dev; 300 301 DNPRINTF(SR_D_META, "%s: sr_meta_probe(%d)\n", DEVNAME(sc), no_chunk); 302 303 if (no_chunk == 0) 304 goto unwind; 305 306 cl = &sd->sd_vol.sv_chunk_list; 307 308 for (d = 0, prevf = SR_META_F_INVALID; d < no_chunk; d++) { 309 ch_entry = malloc(sizeof(struct sr_chunk), M_DEVBUF, 310 M_WAITOK | M_ZERO); 311 /* keep disks in user supplied order */ 312 if (ch_prev) 313 SLIST_INSERT_AFTER(ch_prev, ch_entry, src_link); 314 else 315 SLIST_INSERT_HEAD(cl, ch_entry, src_link); 316 ch_prev = ch_entry; 317 dev = dt[d]; 318 ch_entry->src_dev_mm = dev; 319 320 if (dev == NODEV) { 321 ch_entry->src_meta.scm_status = BIOC_SDOFFLINE; 322 continue; 323 } else { 324 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 325 if (bdevvp(dev, &vn)) { 326 sr_error(sc, "sr_meta_probe: cannot allocate " 327 "vnode"); 328 goto unwind; 329 } 330 331 /* 332 * XXX leaving dev open for now; move this to attach 333 * and figure out the open/close dance for unwind. 334 */ 335 error = VOP_OPEN(vn, FREAD | FWRITE, NOCRED, curproc); 336 if (error) { 337 DNPRINTF(SR_D_META,"%s: sr_meta_probe can't " 338 "open %s\n", DEVNAME(sc), devname); 339 vput(vn); 340 goto unwind; 341 } 342 343 strlcpy(ch_entry->src_devname, devname, 344 sizeof(ch_entry->src_devname)); 345 ch_entry->src_vn = vn; 346 } 347 348 /* determine if this is a device we understand */ 349 for (i = 0, found = SR_META_F_INVALID; smd[i].smd_probe; i++) { 350 type = smd[i].smd_probe(sc, ch_entry); 351 if (type == SR_META_F_INVALID) 352 continue; 353 else { 354 found = type; 355 break; 356 } 357 } 358 359 if (found == SR_META_F_INVALID) 360 goto unwind; 361 if (prevf == SR_META_F_INVALID) 362 prevf = found; 363 if (prevf != found) { 364 DNPRINTF(SR_D_META, "%s: prevf != found\n", 365 DEVNAME(sc)); 366 goto unwind; 367 } 368 } 369 370 return (prevf); 371 unwind: 372 return (SR_META_F_INVALID); 373 } 374 375 void 376 sr_meta_getdevname(struct sr_softc *sc, dev_t dev, char *buf, int size) 377 { 378 int maj, unit, part; 379 char *name; 380 381 DNPRINTF(SR_D_META, "%s: sr_meta_getdevname(%p, %d)\n", 382 DEVNAME(sc), buf, size); 383 384 if (!buf) 385 return; 386 387 maj = major(dev); 388 part = DISKPART(dev); 389 unit = DISKUNIT(dev); 390 391 name = findblkname(maj); 392 if (name == NULL) 393 return; 394 395 snprintf(buf, size, "%s%d%c", name, unit, part + 'a'); 396 } 397 398 int 399 sr_rw(struct sr_softc *sc, dev_t dev, char *buf, size_t size, daddr_t blkno, 400 long flags) 401 { 402 struct vnode *vp; 403 struct buf b; 404 size_t bufsize, dma_bufsize; 405 int rv = 1; 406 char *dma_buf; 407 int s; 408 409 DNPRINTF(SR_D_MISC, "%s: sr_rw(0x%x, %p, %zu, %lld 0x%lx)\n", 410 DEVNAME(sc), dev, buf, size, (long long)blkno, flags); 411 412 dma_bufsize = (size > MAXPHYS) ? MAXPHYS : size; 413 dma_buf = dma_alloc(dma_bufsize, PR_WAITOK); 414 415 if (bdevvp(dev, &vp)) { 416 printf("%s: sr_rw: failed to allocate vnode\n", DEVNAME(sc)); 417 goto done; 418 } 419 420 while (size > 0) { 421 DNPRINTF(SR_D_MISC, "%s: dma_buf %p, size %zu, blkno %lld)\n", 422 DEVNAME(sc), dma_buf, size, (long long)blkno); 423 424 bufsize = (size > MAXPHYS) ? MAXPHYS : size; 425 if (flags == B_WRITE) 426 memcpy(dma_buf, buf, bufsize); 427 428 bzero(&b, sizeof(b)); 429 b.b_flags = flags | B_PHYS; 430 b.b_proc = curproc; 431 b.b_dev = dev; 432 b.b_iodone = NULL; 433 b.b_error = 0; 434 b.b_blkno = blkno; 435 b.b_data = dma_buf; 436 b.b_bcount = bufsize; 437 b.b_bufsize = bufsize; 438 b.b_resid = bufsize; 439 b.b_vp = vp; 440 441 if ((b.b_flags & B_READ) == 0) { 442 s = splbio(); 443 vp->v_numoutput++; 444 splx(s); 445 } 446 447 LIST_INIT(&b.b_dep); 448 VOP_STRATEGY(vp, &b); 449 biowait(&b); 450 451 if (b.b_flags & B_ERROR) { 452 printf("%s: I/O error %d on dev 0x%x at block %llu\n", 453 DEVNAME(sc), b.b_error, dev, b.b_blkno); 454 goto done; 455 } 456 457 if (flags == B_READ) 458 memcpy(buf, dma_buf, bufsize); 459 460 size -= bufsize; 461 buf += bufsize; 462 blkno += howmany(bufsize, DEV_BSIZE); 463 } 464 465 rv = 0; 466 467 done: 468 if (vp) 469 vput(vp); 470 471 dma_free(dma_buf, dma_bufsize); 472 473 return (rv); 474 } 475 476 int 477 sr_meta_rw(struct sr_discipline *sd, dev_t dev, void *md, long flags) 478 { 479 int rv = 1; 480 481 DNPRINTF(SR_D_META, "%s: sr_meta_rw(0x%x, %p, 0x%lx)\n", 482 DEVNAME(sd->sd_sc), dev, md, flags); 483 484 if (md == NULL) { 485 printf("%s: sr_meta_rw: invalid metadata pointer\n", 486 DEVNAME(sd->sd_sc)); 487 goto done; 488 } 489 490 rv = sr_rw(sd->sd_sc, dev, md, SR_META_SIZE * DEV_BSIZE, 491 SR_META_OFFSET, flags); 492 493 done: 494 return (rv); 495 } 496 497 int 498 sr_meta_clear(struct sr_discipline *sd) 499 { 500 struct sr_softc *sc = sd->sd_sc; 501 struct sr_chunk_head *cl = &sd->sd_vol.sv_chunk_list; 502 struct sr_chunk *ch_entry; 503 void *m; 504 int rv = 1; 505 506 DNPRINTF(SR_D_META, "%s: sr_meta_clear\n", DEVNAME(sc)); 507 508 if (sd->sd_meta_type != SR_META_F_NATIVE) { 509 sr_error(sc, "cannot clear foreign metadata"); 510 goto done; 511 } 512 513 m = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, M_WAITOK | M_ZERO); 514 SLIST_FOREACH(ch_entry, cl, src_link) { 515 if (sr_meta_native_write(sd, ch_entry->src_dev_mm, m, NULL)) { 516 /* XXX mark disk offline */ 517 DNPRINTF(SR_D_META, "%s: sr_meta_clear failed to " 518 "clear %s\n", DEVNAME(sc), ch_entry->src_devname); 519 rv++; 520 continue; 521 } 522 bzero(&ch_entry->src_meta, sizeof(ch_entry->src_meta)); 523 } 524 525 bzero(sd->sd_meta, SR_META_SIZE * DEV_BSIZE); 526 527 free(m, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 528 rv = 0; 529 done: 530 return (rv); 531 } 532 533 void 534 sr_meta_init(struct sr_discipline *sd, int level, int no_chunk) 535 { 536 struct sr_softc *sc = sd->sd_sc; 537 struct sr_metadata *sm = sd->sd_meta; 538 struct sr_chunk_head *cl = &sd->sd_vol.sv_chunk_list; 539 struct sr_meta_chunk *scm; 540 struct sr_chunk *chunk; 541 int cid = 0; 542 u_int64_t max_chunk_sz = 0, min_chunk_sz = 0; 543 u_int32_t secsize = DEV_BSIZE; 544 545 DNPRINTF(SR_D_META, "%s: sr_meta_init\n", DEVNAME(sc)); 546 547 if (!sm) 548 return; 549 550 /* Initialise volume metadata. */ 551 sm->ssdi.ssd_magic = SR_MAGIC; 552 sm->ssdi.ssd_version = SR_META_VERSION; 553 sm->ssdi.ssd_vol_flags = sd->sd_meta_flags; 554 sm->ssdi.ssd_volid = 0; 555 sm->ssdi.ssd_chunk_no = no_chunk; 556 sm->ssdi.ssd_level = level; 557 558 sm->ssd_data_blkno = SR_DATA_OFFSET; 559 sm->ssd_ondisk = 0; 560 561 sr_uuid_generate(&sm->ssdi.ssd_uuid); 562 563 /* Initialise chunk metadata and get min/max chunk sizes & secsize. */ 564 SLIST_FOREACH(chunk, cl, src_link) { 565 scm = &chunk->src_meta; 566 scm->scmi.scm_size = chunk->src_size; 567 scm->scmi.scm_chunk_id = cid++; 568 scm->scm_status = BIOC_SDONLINE; 569 scm->scmi.scm_volid = 0; 570 strlcpy(scm->scmi.scm_devname, chunk->src_devname, 571 sizeof(scm->scmi.scm_devname)); 572 memcpy(&scm->scmi.scm_uuid, &sm->ssdi.ssd_uuid, 573 sizeof(scm->scmi.scm_uuid)); 574 sr_checksum(sc, scm, &scm->scm_checksum, 575 sizeof(scm->scm_checksum)); 576 577 if (min_chunk_sz == 0) 578 min_chunk_sz = scm->scmi.scm_size; 579 if (chunk->src_secsize > secsize) 580 secsize = chunk->src_secsize; 581 min_chunk_sz = MIN(min_chunk_sz, scm->scmi.scm_size); 582 max_chunk_sz = MAX(max_chunk_sz, scm->scmi.scm_size); 583 } 584 585 sm->ssdi.ssd_secsize = secsize; 586 587 /* Equalize chunk sizes. */ 588 SLIST_FOREACH(chunk, cl, src_link) 589 chunk->src_meta.scmi.scm_coerced_size = min_chunk_sz; 590 591 sd->sd_vol.sv_chunk_minsz = min_chunk_sz; 592 sd->sd_vol.sv_chunk_maxsz = max_chunk_sz; 593 } 594 595 void 596 sr_meta_init_complete(struct sr_discipline *sd) 597 { 598 #ifdef SR_DEBUG 599 struct sr_softc *sc = sd->sd_sc; 600 #endif 601 struct sr_metadata *sm = sd->sd_meta; 602 603 DNPRINTF(SR_D_META, "%s: sr_meta_complete\n", DEVNAME(sc)); 604 605 /* Complete initialisation of volume metadata. */ 606 strlcpy(sm->ssdi.ssd_vendor, "OPENBSD", sizeof(sm->ssdi.ssd_vendor)); 607 snprintf(sm->ssdi.ssd_product, sizeof(sm->ssdi.ssd_product), 608 "SR %s", sd->sd_name); 609 snprintf(sm->ssdi.ssd_revision, sizeof(sm->ssdi.ssd_revision), 610 "%03d", sm->ssdi.ssd_version); 611 } 612 613 void 614 sr_meta_opt_handler(struct sr_discipline *sd, struct sr_meta_opt_hdr *om) 615 { 616 if (om->som_type != SR_OPT_BOOT) 617 panic("unknown optional metadata type"); 618 } 619 620 void 621 sr_meta_save_callback(void *xsd) 622 { 623 struct sr_discipline *sd = xsd; 624 int s; 625 626 s = splbio(); 627 628 if (sr_meta_save(sd, SR_META_DIRTY)) 629 printf("%s: save metadata failed\n", DEVNAME(sd->sd_sc)); 630 631 sd->sd_must_flush = 0; 632 splx(s); 633 } 634 635 int 636 sr_meta_save(struct sr_discipline *sd, u_int32_t flags) 637 { 638 struct sr_softc *sc = sd->sd_sc; 639 struct sr_metadata *sm = sd->sd_meta, *m; 640 struct sr_meta_driver *s; 641 struct sr_chunk *src; 642 struct sr_meta_chunk *cm; 643 struct sr_workunit wu; 644 struct sr_meta_opt_hdr *omh; 645 struct sr_meta_opt_item *omi; 646 int i; 647 648 DNPRINTF(SR_D_META, "%s: sr_meta_save %s\n", 649 DEVNAME(sc), sd->sd_meta->ssd_devname); 650 651 if (!sm) { 652 printf("%s: no in memory copy of metadata\n", DEVNAME(sc)); 653 goto bad; 654 } 655 656 /* meta scratchpad */ 657 s = &smd[sd->sd_meta_type]; 658 m = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, M_ZERO | M_NOWAIT); 659 if (!m) { 660 printf("%s: could not allocate metadata scratch area\n", 661 DEVNAME(sc)); 662 goto bad; 663 } 664 665 /* from here on out metadata is updated */ 666 restart: 667 sm->ssd_ondisk++; 668 sm->ssd_meta_flags = flags; 669 memcpy(m, sm, sizeof(*m)); 670 671 /* Chunk metadata. */ 672 cm = (struct sr_meta_chunk *)(m + 1); 673 for (i = 0; i < sm->ssdi.ssd_chunk_no; i++) { 674 src = sd->sd_vol.sv_chunks[i]; 675 memcpy(cm, &src->src_meta, sizeof(*cm)); 676 cm++; 677 } 678 679 /* Optional metadata. */ 680 omh = (struct sr_meta_opt_hdr *)(cm); 681 SLIST_FOREACH(omi, &sd->sd_meta_opt, omi_link) { 682 DNPRINTF(SR_D_META, "%s: saving optional metadata type %u with " 683 "length %u\n", DEVNAME(sc), omi->omi_som->som_type, 684 omi->omi_som->som_length); 685 bzero(&omi->omi_som->som_checksum, MD5_DIGEST_LENGTH); 686 sr_checksum(sc, omi->omi_som, &omi->omi_som->som_checksum, 687 omi->omi_som->som_length); 688 memcpy(omh, omi->omi_som, omi->omi_som->som_length); 689 omh = (struct sr_meta_opt_hdr *)((u_int8_t *)omh + 690 omi->omi_som->som_length); 691 } 692 693 for (i = 0; i < sm->ssdi.ssd_chunk_no; i++) { 694 src = sd->sd_vol.sv_chunks[i]; 695 696 /* skip disks that are offline */ 697 if (src->src_meta.scm_status == BIOC_SDOFFLINE) 698 continue; 699 700 /* calculate metadata checksum for correct chunk */ 701 m->ssdi.ssd_chunk_id = i; 702 sr_checksum(sc, m, &m->ssd_checksum, 703 sizeof(struct sr_meta_invariant)); 704 705 #ifdef SR_DEBUG 706 DNPRINTF(SR_D_META, "%s: sr_meta_save %s: volid: %d " 707 "chunkid: %d checksum: ", 708 DEVNAME(sc), src->src_meta.scmi.scm_devname, 709 m->ssdi.ssd_volid, m->ssdi.ssd_chunk_id); 710 711 if (sr_debug & SR_D_META) 712 sr_checksum_print((u_int8_t *)&m->ssd_checksum); 713 DNPRINTF(SR_D_META, "\n"); 714 sr_meta_print(m); 715 #endif 716 717 /* translate and write to disk */ 718 if (s->smd_write(sd, src->src_dev_mm, m, NULL /* XXX */)) { 719 printf("%s: could not write metadata to %s\n", 720 DEVNAME(sc), src->src_devname); 721 /* restart the meta write */ 722 src->src_meta.scm_status = BIOC_SDOFFLINE; 723 /* XXX recalculate volume status */ 724 goto restart; 725 } 726 } 727 728 /* not all disciplines have sync */ 729 if (sd->sd_scsi_sync) { 730 bzero(&wu, sizeof(wu)); 731 wu.swu_flags |= SR_WUF_FAKE; 732 wu.swu_dis = sd; 733 sd->sd_scsi_sync(&wu); 734 } 735 free(m, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 736 return (0); 737 bad: 738 return (1); 739 } 740 741 int 742 sr_meta_read(struct sr_discipline *sd) 743 { 744 struct sr_softc *sc = sd->sd_sc; 745 struct sr_chunk_head *cl = &sd->sd_vol.sv_chunk_list; 746 struct sr_metadata *sm; 747 struct sr_chunk *ch_entry; 748 struct sr_meta_chunk *cp; 749 struct sr_meta_driver *s; 750 void *fm = NULL; 751 int no_disk = 0, got_meta = 0; 752 753 DNPRINTF(SR_D_META, "%s: sr_meta_read\n", DEVNAME(sc)); 754 755 sm = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, M_WAITOK | M_ZERO); 756 s = &smd[sd->sd_meta_type]; 757 if (sd->sd_meta_type != SR_META_F_NATIVE) 758 fm = malloc(s->smd_size, M_DEVBUF, M_WAITOK | M_ZERO); 759 760 cp = (struct sr_meta_chunk *)(sm + 1); 761 SLIST_FOREACH(ch_entry, cl, src_link) { 762 /* skip disks that are offline */ 763 if (ch_entry->src_meta.scm_status == BIOC_SDOFFLINE) { 764 DNPRINTF(SR_D_META, 765 "%s: %s chunk marked offline, spoofing status\n", 766 DEVNAME(sc), ch_entry->src_devname); 767 cp++; /* adjust chunk pointer to match failure */ 768 continue; 769 } else if (s->smd_read(sd, ch_entry->src_dev_mm, sm, fm)) { 770 /* read and translate */ 771 /* XXX mark chunk offline, elsewhere!! */ 772 ch_entry->src_meta.scm_status = BIOC_SDOFFLINE; 773 cp++; /* adjust chunk pointer to match failure */ 774 DNPRINTF(SR_D_META, "%s: sr_meta_read failed\n", 775 DEVNAME(sc)); 776 continue; 777 } 778 779 if (sm->ssdi.ssd_magic != SR_MAGIC) { 780 DNPRINTF(SR_D_META, "%s: sr_meta_read !SR_MAGIC\n", 781 DEVNAME(sc)); 782 continue; 783 } 784 785 /* validate metadata */ 786 if (sr_meta_validate(sd, ch_entry->src_dev_mm, sm, fm)) { 787 DNPRINTF(SR_D_META, "%s: invalid metadata\n", 788 DEVNAME(sc)); 789 no_disk = -1; 790 goto done; 791 } 792 793 /* assume first chunk contains metadata */ 794 if (got_meta == 0) { 795 sr_meta_opt_load(sc, sm, &sd->sd_meta_opt); 796 memcpy(sd->sd_meta, sm, sizeof(*sd->sd_meta)); 797 got_meta = 1; 798 } 799 800 memcpy(&ch_entry->src_meta, cp, sizeof(ch_entry->src_meta)); 801 802 no_disk++; 803 cp++; 804 } 805 806 free(sm, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 807 free(fm, M_DEVBUF, s->smd_size); 808 809 done: 810 DNPRINTF(SR_D_META, "%s: sr_meta_read found %d parts\n", DEVNAME(sc), 811 no_disk); 812 return (no_disk); 813 } 814 815 void 816 sr_meta_opt_load(struct sr_softc *sc, struct sr_metadata *sm, 817 struct sr_meta_opt_head *som) 818 { 819 struct sr_meta_opt_hdr *omh; 820 struct sr_meta_opt_item *omi; 821 u_int8_t checksum[MD5_DIGEST_LENGTH]; 822 int i; 823 824 /* Process optional metadata. */ 825 omh = (struct sr_meta_opt_hdr *)((u_int8_t *)(sm + 1) + 826 sizeof(struct sr_meta_chunk) * sm->ssdi.ssd_chunk_no); 827 for (i = 0; i < sm->ssdi.ssd_opt_no; i++) { 828 829 omi = malloc(sizeof(struct sr_meta_opt_item), M_DEVBUF, 830 M_WAITOK | M_ZERO); 831 SLIST_INSERT_HEAD(som, omi, omi_link); 832 833 if (omh->som_length == 0) { 834 835 /* Load old fixed length optional metadata. */ 836 DNPRINTF(SR_D_META, "%s: old optional metadata of type " 837 "%u\n", DEVNAME(sc), omh->som_type); 838 839 /* Validate checksum. */ 840 sr_checksum(sc, (void *)omh, &checksum, 841 SR_OLD_META_OPT_SIZE - MD5_DIGEST_LENGTH); 842 if (bcmp(&checksum, (void *)omh + SR_OLD_META_OPT_MD5, 843 sizeof(checksum))) 844 panic("%s: invalid optional metadata checksum", 845 DEVNAME(sc)); 846 847 /* Determine correct length. */ 848 switch (omh->som_type) { 849 case SR_OPT_CRYPTO: 850 omh->som_length = sizeof(struct sr_meta_crypto); 851 break; 852 case SR_OPT_BOOT: 853 omh->som_length = sizeof(struct sr_meta_boot); 854 break; 855 case SR_OPT_KEYDISK: 856 omh->som_length = 857 sizeof(struct sr_meta_keydisk); 858 break; 859 default: 860 panic("unknown old optional metadata type %u", 861 omh->som_type); 862 } 863 864 omi->omi_som = malloc(omh->som_length, M_DEVBUF, 865 M_WAITOK | M_ZERO); 866 memcpy((u_int8_t *)omi->omi_som + sizeof(*omi->omi_som), 867 (u_int8_t *)omh + SR_OLD_META_OPT_OFFSET, 868 omh->som_length - sizeof(*omi->omi_som)); 869 omi->omi_som->som_type = omh->som_type; 870 omi->omi_som->som_length = omh->som_length; 871 872 omh = (struct sr_meta_opt_hdr *)((void *)omh + 873 SR_OLD_META_OPT_SIZE); 874 } else { 875 876 /* Load variable length optional metadata. */ 877 DNPRINTF(SR_D_META, "%s: optional metadata of type %u, " 878 "length %u\n", DEVNAME(sc), omh->som_type, 879 omh->som_length); 880 omi->omi_som = malloc(omh->som_length, M_DEVBUF, 881 M_WAITOK | M_ZERO); 882 memcpy(omi->omi_som, omh, omh->som_length); 883 884 /* Validate checksum. */ 885 memcpy(&checksum, &omi->omi_som->som_checksum, 886 MD5_DIGEST_LENGTH); 887 bzero(&omi->omi_som->som_checksum, MD5_DIGEST_LENGTH); 888 sr_checksum(sc, omi->omi_som, 889 &omi->omi_som->som_checksum, omh->som_length); 890 if (bcmp(&checksum, &omi->omi_som->som_checksum, 891 sizeof(checksum))) 892 panic("%s: invalid optional metadata checksum", 893 DEVNAME(sc)); 894 895 omh = (struct sr_meta_opt_hdr *)((void *)omh + 896 omh->som_length); 897 } 898 } 899 } 900 901 int 902 sr_meta_validate(struct sr_discipline *sd, dev_t dev, struct sr_metadata *sm, 903 void *fm) 904 { 905 struct sr_softc *sc = sd->sd_sc; 906 struct sr_meta_driver *s; 907 #ifdef SR_DEBUG 908 struct sr_meta_chunk *mc; 909 #endif 910 u_int8_t checksum[MD5_DIGEST_LENGTH]; 911 char devname[32]; 912 int rv = 1; 913 914 DNPRINTF(SR_D_META, "%s: sr_meta_validate(%p)\n", DEVNAME(sc), sm); 915 916 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 917 918 s = &smd[sd->sd_meta_type]; 919 if (sd->sd_meta_type != SR_META_F_NATIVE) 920 if (s->smd_validate(sd, sm, fm)) { 921 sr_error(sc, "invalid foreign metadata"); 922 goto done; 923 } 924 925 /* 926 * at this point all foreign metadata has been translated to the native 927 * format and will be treated just like the native format 928 */ 929 930 if (sm->ssdi.ssd_magic != SR_MAGIC) { 931 sr_error(sc, "not valid softraid metadata"); 932 goto done; 933 } 934 935 /* Verify metadata checksum. */ 936 sr_checksum(sc, sm, &checksum, sizeof(struct sr_meta_invariant)); 937 if (bcmp(&checksum, &sm->ssd_checksum, sizeof(checksum))) { 938 sr_error(sc, "invalid metadata checksum"); 939 goto done; 940 } 941 942 /* Handle changes between versions. */ 943 if (sm->ssdi.ssd_version == 3) { 944 945 /* 946 * Version 3 - update metadata version and fix up data blkno 947 * value since this did not exist in version 3. 948 */ 949 if (sm->ssd_data_blkno == 0) 950 sm->ssd_data_blkno = SR_META_V3_DATA_OFFSET; 951 sm->ssdi.ssd_secsize = DEV_BSIZE; 952 953 } else if (sm->ssdi.ssd_version == 4) { 954 955 /* 956 * Version 4 - original metadata format did not store 957 * data blkno so fix this up if necessary. 958 */ 959 if (sm->ssd_data_blkno == 0) 960 sm->ssd_data_blkno = SR_DATA_OFFSET; 961 sm->ssdi.ssd_secsize = DEV_BSIZE; 962 963 } else if (sm->ssdi.ssd_version == 5) { 964 965 /* 966 * Version 5 - variable length optional metadata. Migration 967 * from earlier fixed length optional metadata is handled 968 * in sr_meta_read(). 969 */ 970 sm->ssdi.ssd_secsize = DEV_BSIZE; 971 972 } else if (sm->ssdi.ssd_version == SR_META_VERSION) { 973 974 /* 975 * Version 6 - store & report a sector size. 976 */ 977 978 } else { 979 980 sr_error(sc, "cannot read metadata version %u on %s, " 981 "expected version %u or earlier", 982 sm->ssdi.ssd_version, devname, SR_META_VERSION); 983 goto done; 984 985 } 986 987 /* Update version number and revision string. */ 988 sm->ssdi.ssd_version = SR_META_VERSION; 989 snprintf(sm->ssdi.ssd_revision, sizeof(sm->ssdi.ssd_revision), 990 "%03d", SR_META_VERSION); 991 992 #ifdef SR_DEBUG 993 /* warn if disk changed order */ 994 mc = (struct sr_meta_chunk *)(sm + 1); 995 if (strncmp(mc[sm->ssdi.ssd_chunk_id].scmi.scm_devname, devname, 996 sizeof(mc[sm->ssdi.ssd_chunk_id].scmi.scm_devname))) 997 DNPRINTF(SR_D_META, "%s: roaming device %s -> %s\n", 998 DEVNAME(sc), mc[sm->ssdi.ssd_chunk_id].scmi.scm_devname, 999 devname); 1000 #endif 1001 1002 /* we have meta data on disk */ 1003 DNPRINTF(SR_D_META, "%s: sr_meta_validate valid metadata %s\n", 1004 DEVNAME(sc), devname); 1005 1006 rv = 0; 1007 done: 1008 return (rv); 1009 } 1010 1011 int 1012 sr_meta_native_bootprobe(struct sr_softc *sc, dev_t devno, 1013 struct sr_boot_chunk_head *bch) 1014 { 1015 struct vnode *vn; 1016 struct disklabel label; 1017 struct sr_metadata *md = NULL; 1018 struct sr_discipline *fake_sd = NULL; 1019 struct sr_boot_chunk *bc; 1020 char devname[32]; 1021 dev_t chrdev, rawdev; 1022 int error, i; 1023 int rv = SR_META_NOTCLAIMED; 1024 1025 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe\n", DEVNAME(sc)); 1026 1027 /* 1028 * Use character raw device to avoid SCSI complaints about missing 1029 * media on removable media devices. 1030 */ 1031 chrdev = blktochr(devno); 1032 rawdev = MAKEDISKDEV(major(chrdev), DISKUNIT(devno), RAW_PART); 1033 if (cdevvp(rawdev, &vn)) { 1034 sr_error(sc, "sr_meta_native_bootprobe: cannot allocate vnode"); 1035 goto done; 1036 } 1037 1038 /* open device */ 1039 error = VOP_OPEN(vn, FREAD, NOCRED, curproc); 1040 if (error) { 1041 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe open " 1042 "failed\n", DEVNAME(sc)); 1043 vput(vn); 1044 goto done; 1045 } 1046 1047 /* get disklabel */ 1048 error = VOP_IOCTL(vn, DIOCGDINFO, (caddr_t)&label, FREAD, NOCRED, 1049 curproc); 1050 if (error) { 1051 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe ioctl " 1052 "failed\n", DEVNAME(sc)); 1053 VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1054 vput(vn); 1055 goto done; 1056 } 1057 1058 /* we are done, close device */ 1059 error = VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1060 if (error) { 1061 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe close " 1062 "failed\n", DEVNAME(sc)); 1063 vput(vn); 1064 goto done; 1065 } 1066 vput(vn); 1067 1068 md = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, M_ZERO | M_NOWAIT); 1069 if (md == NULL) { 1070 sr_error(sc, "not enough memory for metadata buffer"); 1071 goto done; 1072 } 1073 1074 /* create fake sd to use utility functions */ 1075 fake_sd = malloc(sizeof(struct sr_discipline), M_DEVBUF, 1076 M_ZERO | M_NOWAIT); 1077 if (fake_sd == NULL) { 1078 sr_error(sc, "not enough memory for fake discipline"); 1079 goto done; 1080 } 1081 fake_sd->sd_sc = sc; 1082 fake_sd->sd_meta_type = SR_META_F_NATIVE; 1083 1084 for (i = 0; i < MAXPARTITIONS; i++) { 1085 if (label.d_partitions[i].p_fstype != FS_RAID) 1086 continue; 1087 1088 /* open partition */ 1089 rawdev = MAKEDISKDEV(major(devno), DISKUNIT(devno), i); 1090 if (bdevvp(rawdev, &vn)) { 1091 sr_error(sc, "sr_meta_native_bootprobe: cannot " 1092 "allocate vnode for partition"); 1093 goto done; 1094 } 1095 error = VOP_OPEN(vn, FREAD, NOCRED, curproc); 1096 if (error) { 1097 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe " 1098 "open failed, partition %d\n", 1099 DEVNAME(sc), i); 1100 vput(vn); 1101 continue; 1102 } 1103 1104 if (sr_meta_native_read(fake_sd, rawdev, md, NULL)) { 1105 sr_error(sc, "native bootprobe could not read native " 1106 "metadata"); 1107 VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1108 vput(vn); 1109 continue; 1110 } 1111 1112 /* are we a softraid partition? */ 1113 if (md->ssdi.ssd_magic != SR_MAGIC) { 1114 VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1115 vput(vn); 1116 continue; 1117 } 1118 1119 sr_meta_getdevname(sc, rawdev, devname, sizeof(devname)); 1120 if (sr_meta_validate(fake_sd, rawdev, md, NULL) == 0) { 1121 /* XXX fix M_WAITOK, this is boot time */ 1122 bc = malloc(sizeof(struct sr_boot_chunk), 1123 M_DEVBUF, M_WAITOK | M_ZERO); 1124 bc->sbc_metadata = malloc(sizeof(struct sr_metadata), 1125 M_DEVBUF, M_WAITOK | M_ZERO); 1126 memcpy(bc->sbc_metadata, md, sizeof(struct sr_metadata)); 1127 bc->sbc_mm = rawdev; 1128 SLIST_INSERT_HEAD(bch, bc, sbc_link); 1129 rv = SR_META_CLAIMED; 1130 } 1131 1132 /* we are done, close partition */ 1133 VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1134 vput(vn); 1135 } 1136 1137 done: 1138 free(fake_sd, M_DEVBUF, sizeof(struct sr_discipline)); 1139 free(md, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 1140 1141 return (rv); 1142 } 1143 1144 int 1145 sr_boot_assembly(struct sr_softc *sc) 1146 { 1147 struct sr_boot_volume_head bvh; 1148 struct sr_boot_chunk_head bch, kdh; 1149 struct sr_boot_volume *bv, *bv1, *bv2; 1150 struct sr_boot_chunk *bc, *bcnext, *bc1, *bc2; 1151 struct sr_disk_head sdklist; 1152 struct sr_disk *sdk; 1153 struct disk *dk; 1154 struct bioc_createraid bcr; 1155 struct sr_meta_chunk *hm; 1156 struct sr_chunk_head *cl; 1157 struct sr_chunk *hotspare, *chunk, *last; 1158 u_int64_t *ondisk = NULL; 1159 dev_t *devs = NULL; 1160 void *data; 1161 char devname[32]; 1162 int rv = 0, i; 1163 1164 DNPRINTF(SR_D_META, "%s: sr_boot_assembly\n", DEVNAME(sc)); 1165 1166 SLIST_INIT(&sdklist); 1167 SLIST_INIT(&bvh); 1168 SLIST_INIT(&bch); 1169 SLIST_INIT(&kdh); 1170 1171 dk = TAILQ_FIRST(&disklist); 1172 while (dk != NULL) { 1173 1174 /* See if this disk has been checked. */ 1175 SLIST_FOREACH(sdk, &sdklist, sdk_link) 1176 if (sdk->sdk_devno == dk->dk_devno) 1177 break; 1178 1179 if (sdk != NULL || dk->dk_devno == NODEV) { 1180 dk = TAILQ_NEXT(dk, dk_link); 1181 continue; 1182 } 1183 1184 /* Add this disk to the list that we've checked. */ 1185 sdk = malloc(sizeof(struct sr_disk), M_DEVBUF, 1186 M_NOWAIT | M_ZERO); 1187 if (sdk == NULL) 1188 goto unwind; 1189 sdk->sdk_devno = dk->dk_devno; 1190 SLIST_INSERT_HEAD(&sdklist, sdk, sdk_link); 1191 1192 /* Only check sd(4) and wd(4) devices. */ 1193 if (strncmp(dk->dk_name, "sd", 2) && 1194 strncmp(dk->dk_name, "wd", 2)) { 1195 dk = TAILQ_NEXT(dk, dk_link); 1196 continue; 1197 } 1198 1199 /* native softraid uses partitions */ 1200 rw_enter_write(&sc->sc_lock); 1201 bio_status_init(&sc->sc_status, &sc->sc_dev); 1202 sr_meta_native_bootprobe(sc, dk->dk_devno, &bch); 1203 rw_exit_write(&sc->sc_lock); 1204 1205 /* probe non-native disks if native failed. */ 1206 1207 /* Restart scan since we may have slept. */ 1208 dk = TAILQ_FIRST(&disklist); 1209 } 1210 1211 /* 1212 * Create a list of volumes and associate chunks with each volume. 1213 */ 1214 for (bc = SLIST_FIRST(&bch); bc != NULL; bc = bcnext) { 1215 1216 bcnext = SLIST_NEXT(bc, sbc_link); 1217 SLIST_REMOVE(&bch, bc, sr_boot_chunk, sbc_link); 1218 bc->sbc_chunk_id = bc->sbc_metadata->ssdi.ssd_chunk_id; 1219 1220 /* Handle key disks separately. */ 1221 if (bc->sbc_metadata->ssdi.ssd_level == SR_KEYDISK_LEVEL) { 1222 SLIST_INSERT_HEAD(&kdh, bc, sbc_link); 1223 continue; 1224 } 1225 1226 SLIST_FOREACH(bv, &bvh, sbv_link) { 1227 if (bcmp(&bc->sbc_metadata->ssdi.ssd_uuid, 1228 &bv->sbv_uuid, 1229 sizeof(bc->sbc_metadata->ssdi.ssd_uuid)) == 0) 1230 break; 1231 } 1232 1233 if (bv == NULL) { 1234 bv = malloc(sizeof(struct sr_boot_volume), 1235 M_DEVBUF, M_NOWAIT | M_ZERO); 1236 if (bv == NULL) { 1237 printf("%s: failed to allocate boot volume\n", 1238 DEVNAME(sc)); 1239 goto unwind; 1240 } 1241 1242 bv->sbv_level = bc->sbc_metadata->ssdi.ssd_level; 1243 bv->sbv_volid = bc->sbc_metadata->ssdi.ssd_volid; 1244 bv->sbv_chunk_no = bc->sbc_metadata->ssdi.ssd_chunk_no; 1245 bv->sbv_flags = bc->sbc_metadata->ssdi.ssd_vol_flags; 1246 memcpy(&bv->sbv_uuid, &bc->sbc_metadata->ssdi.ssd_uuid, 1247 sizeof(bc->sbc_metadata->ssdi.ssd_uuid)); 1248 SLIST_INIT(&bv->sbv_chunks); 1249 1250 /* Maintain volume order. */ 1251 bv2 = NULL; 1252 SLIST_FOREACH(bv1, &bvh, sbv_link) { 1253 if (bv1->sbv_volid > bv->sbv_volid) 1254 break; 1255 bv2 = bv1; 1256 } 1257 if (bv2 == NULL) { 1258 DNPRINTF(SR_D_META, "%s: insert volume %u " 1259 "at head\n", DEVNAME(sc), bv->sbv_volid); 1260 SLIST_INSERT_HEAD(&bvh, bv, sbv_link); 1261 } else { 1262 DNPRINTF(SR_D_META, "%s: insert volume %u " 1263 "after %u\n", DEVNAME(sc), bv->sbv_volid, 1264 bv2->sbv_volid); 1265 SLIST_INSERT_AFTER(bv2, bv, sbv_link); 1266 } 1267 } 1268 1269 /* Maintain chunk order. */ 1270 bc2 = NULL; 1271 SLIST_FOREACH(bc1, &bv->sbv_chunks, sbc_link) { 1272 if (bc1->sbc_chunk_id > bc->sbc_chunk_id) 1273 break; 1274 bc2 = bc1; 1275 } 1276 if (bc2 == NULL) { 1277 DNPRINTF(SR_D_META, "%s: volume %u insert chunk %u " 1278 "at head\n", DEVNAME(sc), bv->sbv_volid, 1279 bc->sbc_chunk_id); 1280 SLIST_INSERT_HEAD(&bv->sbv_chunks, bc, sbc_link); 1281 } else { 1282 DNPRINTF(SR_D_META, "%s: volume %u insert chunk %u " 1283 "after %u\n", DEVNAME(sc), bv->sbv_volid, 1284 bc->sbc_chunk_id, bc2->sbc_chunk_id); 1285 SLIST_INSERT_AFTER(bc2, bc, sbc_link); 1286 } 1287 1288 bv->sbv_chunks_found++; 1289 } 1290 1291 /* Allocate memory for device and ondisk version arrays. */ 1292 devs = mallocarray(BIOC_CRMAXLEN, sizeof(dev_t), M_DEVBUF, 1293 M_NOWAIT); 1294 if (devs == NULL) { 1295 printf("%s: failed to allocate device array\n", DEVNAME(sc)); 1296 goto unwind; 1297 } 1298 ondisk = mallocarray(BIOC_CRMAXLEN, sizeof(u_int64_t), M_DEVBUF, 1299 M_NOWAIT); 1300 if (ondisk == NULL) { 1301 printf("%s: failed to allocate ondisk array\n", DEVNAME(sc)); 1302 goto unwind; 1303 } 1304 1305 /* 1306 * Assemble hotspare "volumes". 1307 */ 1308 SLIST_FOREACH(bv, &bvh, sbv_link) { 1309 1310 /* Check if this is a hotspare "volume". */ 1311 if (bv->sbv_level != SR_HOTSPARE_LEVEL || 1312 bv->sbv_chunk_no != 1) 1313 continue; 1314 1315 #ifdef SR_DEBUG 1316 DNPRINTF(SR_D_META, "%s: assembling hotspare volume ", 1317 DEVNAME(sc)); 1318 if (sr_debug & SR_D_META) 1319 sr_uuid_print(&bv->sbv_uuid, 0); 1320 DNPRINTF(SR_D_META, " volid %u with %u chunks\n", 1321 bv->sbv_volid, bv->sbv_chunk_no); 1322 #endif 1323 1324 /* Create hotspare chunk metadata. */ 1325 hotspare = malloc(sizeof(struct sr_chunk), M_DEVBUF, 1326 M_NOWAIT | M_ZERO); 1327 if (hotspare == NULL) { 1328 printf("%s: failed to allocate hotspare\n", 1329 DEVNAME(sc)); 1330 goto unwind; 1331 } 1332 1333 bc = SLIST_FIRST(&bv->sbv_chunks); 1334 sr_meta_getdevname(sc, bc->sbc_mm, devname, sizeof(devname)); 1335 hotspare->src_dev_mm = bc->sbc_mm; 1336 strlcpy(hotspare->src_devname, devname, 1337 sizeof(hotspare->src_devname)); 1338 hotspare->src_size = bc->sbc_metadata->ssdi.ssd_size; 1339 1340 hm = &hotspare->src_meta; 1341 hm->scmi.scm_volid = SR_HOTSPARE_VOLID; 1342 hm->scmi.scm_chunk_id = 0; 1343 hm->scmi.scm_size = bc->sbc_metadata->ssdi.ssd_size; 1344 hm->scmi.scm_coerced_size = bc->sbc_metadata->ssdi.ssd_size; 1345 strlcpy(hm->scmi.scm_devname, devname, 1346 sizeof(hm->scmi.scm_devname)); 1347 memcpy(&hm->scmi.scm_uuid, &bc->sbc_metadata->ssdi.ssd_uuid, 1348 sizeof(struct sr_uuid)); 1349 1350 sr_checksum(sc, hm, &hm->scm_checksum, 1351 sizeof(struct sr_meta_chunk_invariant)); 1352 1353 hm->scm_status = BIOC_SDHOTSPARE; 1354 1355 /* Add chunk to hotspare list. */ 1356 rw_enter_write(&sc->sc_hs_lock); 1357 cl = &sc->sc_hotspare_list; 1358 if (SLIST_EMPTY(cl)) 1359 SLIST_INSERT_HEAD(cl, hotspare, src_link); 1360 else { 1361 SLIST_FOREACH(chunk, cl, src_link) 1362 last = chunk; 1363 SLIST_INSERT_AFTER(last, hotspare, src_link); 1364 } 1365 sc->sc_hotspare_no++; 1366 rw_exit_write(&sc->sc_hs_lock); 1367 1368 } 1369 1370 /* 1371 * Assemble RAID volumes. 1372 */ 1373 SLIST_FOREACH(bv, &bvh, sbv_link) { 1374 1375 bzero(&bcr, sizeof(bcr)); 1376 data = NULL; 1377 1378 /* Check if this is a hotspare "volume". */ 1379 if (bv->sbv_level == SR_HOTSPARE_LEVEL && 1380 bv->sbv_chunk_no == 1) 1381 continue; 1382 1383 /* 1384 * Skip volumes that are marked as no auto assemble, unless 1385 * this was the volume which we actually booted from. 1386 */ 1387 if (bcmp(&sr_bootuuid, &bv->sbv_uuid, sizeof(sr_bootuuid)) != 0) 1388 if (bv->sbv_flags & BIOC_SCNOAUTOASSEMBLE) 1389 continue; 1390 1391 #ifdef SR_DEBUG 1392 DNPRINTF(SR_D_META, "%s: assembling volume ", DEVNAME(sc)); 1393 if (sr_debug & SR_D_META) 1394 sr_uuid_print(&bv->sbv_uuid, 0); 1395 DNPRINTF(SR_D_META, " volid %u with %u chunks\n", 1396 bv->sbv_volid, bv->sbv_chunk_no); 1397 #endif 1398 1399 /* 1400 * If this is a crypto volume, try to find a matching 1401 * key disk... 1402 */ 1403 bcr.bc_key_disk = NODEV; 1404 if (bv->sbv_level == 'C' || bv->sbv_level == 0x1C) { 1405 SLIST_FOREACH(bc, &kdh, sbc_link) { 1406 if (bcmp(&bc->sbc_metadata->ssdi.ssd_uuid, 1407 &bv->sbv_uuid, 1408 sizeof(bc->sbc_metadata->ssdi.ssd_uuid)) 1409 == 0) 1410 bcr.bc_key_disk = bc->sbc_mm; 1411 } 1412 } 1413 1414 for (i = 0; i < BIOC_CRMAXLEN; i++) { 1415 devs[i] = NODEV; /* mark device as illegal */ 1416 ondisk[i] = 0; 1417 } 1418 1419 SLIST_FOREACH(bc, &bv->sbv_chunks, sbc_link) { 1420 if (devs[bc->sbc_chunk_id] != NODEV) { 1421 bv->sbv_chunks_found--; 1422 sr_meta_getdevname(sc, bc->sbc_mm, devname, 1423 sizeof(devname)); 1424 printf("%s: found duplicate chunk %u for " 1425 "volume %u on device %s\n", DEVNAME(sc), 1426 bc->sbc_chunk_id, bv->sbv_volid, devname); 1427 } 1428 1429 if (devs[bc->sbc_chunk_id] == NODEV || 1430 bc->sbc_metadata->ssd_ondisk > 1431 ondisk[bc->sbc_chunk_id]) { 1432 devs[bc->sbc_chunk_id] = bc->sbc_mm; 1433 ondisk[bc->sbc_chunk_id] = 1434 bc->sbc_metadata->ssd_ondisk; 1435 DNPRINTF(SR_D_META, "%s: using ondisk " 1436 "metadata version %llu for chunk %u\n", 1437 DEVNAME(sc), ondisk[bc->sbc_chunk_id], 1438 bc->sbc_chunk_id); 1439 } 1440 } 1441 1442 if (bv->sbv_chunk_no != bv->sbv_chunks_found) { 1443 printf("%s: not all chunks were provided; " 1444 "attempting to bring volume %d online\n", 1445 DEVNAME(sc), bv->sbv_volid); 1446 } 1447 1448 bcr.bc_level = bv->sbv_level; 1449 bcr.bc_dev_list_len = bv->sbv_chunk_no * sizeof(dev_t); 1450 bcr.bc_dev_list = devs; 1451 bcr.bc_flags = BIOC_SCDEVT | 1452 (bv->sbv_flags & BIOC_SCNOAUTOASSEMBLE); 1453 1454 if ((bv->sbv_level == 'C' || bv->sbv_level == 0x1C) && 1455 bcmp(&sr_bootuuid, &bv->sbv_uuid, sizeof(sr_bootuuid)) == 0) 1456 data = sr_bootkey; 1457 1458 rw_enter_write(&sc->sc_lock); 1459 bio_status_init(&sc->sc_status, &sc->sc_dev); 1460 sr_ioctl_createraid(sc, &bcr, 0, data); 1461 rw_exit_write(&sc->sc_lock); 1462 1463 rv++; 1464 } 1465 1466 /* done with metadata */ 1467 unwind: 1468 /* Free boot volumes and associated chunks. */ 1469 for (bv1 = SLIST_FIRST(&bvh); bv1 != NULL; bv1 = bv2) { 1470 bv2 = SLIST_NEXT(bv1, sbv_link); 1471 for (bc1 = SLIST_FIRST(&bv1->sbv_chunks); bc1 != NULL; 1472 bc1 = bc2) { 1473 bc2 = SLIST_NEXT(bc1, sbc_link); 1474 free(bc1->sbc_metadata, M_DEVBUF, 1475 sizeof(*bc1->sbc_metadata)); 1476 free(bc1, M_DEVBUF, sizeof(*bc1)); 1477 } 1478 free(bv1, M_DEVBUF, sizeof(*bv1)); 1479 } 1480 /* Free keydisks chunks. */ 1481 for (bc1 = SLIST_FIRST(&kdh); bc1 != NULL; bc1 = bc2) { 1482 bc2 = SLIST_NEXT(bc1, sbc_link); 1483 free(bc1->sbc_metadata, M_DEVBUF, sizeof(*bc1->sbc_metadata)); 1484 free(bc1, M_DEVBUF, sizeof(*bc1)); 1485 } 1486 /* Free unallocated chunks. */ 1487 for (bc1 = SLIST_FIRST(&bch); bc1 != NULL; bc1 = bc2) { 1488 bc2 = SLIST_NEXT(bc1, sbc_link); 1489 free(bc1->sbc_metadata, M_DEVBUF, sizeof(*bc1->sbc_metadata)); 1490 free(bc1, M_DEVBUF, sizeof(*bc1)); 1491 } 1492 1493 while (!SLIST_EMPTY(&sdklist)) { 1494 sdk = SLIST_FIRST(&sdklist); 1495 SLIST_REMOVE_HEAD(&sdklist, sdk_link); 1496 free(sdk, M_DEVBUF, sizeof(*sdk)); 1497 } 1498 1499 free(devs, M_DEVBUF, BIOC_CRMAXLEN * sizeof(dev_t)); 1500 free(ondisk, M_DEVBUF, BIOC_CRMAXLEN * sizeof(u_int64_t)); 1501 1502 return (rv); 1503 } 1504 1505 void 1506 sr_map_root(void) 1507 { 1508 struct sr_softc *sc = softraid0; 1509 struct sr_discipline *sd; 1510 struct sr_meta_opt_item *omi; 1511 struct sr_meta_boot *sbm; 1512 u_char duid[8]; 1513 int i; 1514 1515 if (sc == NULL) 1516 return; 1517 1518 DNPRINTF(SR_D_MISC, "%s: sr_map_root\n", DEVNAME(sc)); 1519 1520 bzero(duid, sizeof(duid)); 1521 if (bcmp(rootduid, duid, sizeof(duid)) == 0) { 1522 DNPRINTF(SR_D_MISC, "%s: root duid is zero\n", DEVNAME(sc)); 1523 return; 1524 } 1525 1526 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 1527 SLIST_FOREACH(omi, &sd->sd_meta_opt, omi_link) { 1528 if (omi->omi_som->som_type != SR_OPT_BOOT) 1529 continue; 1530 sbm = (struct sr_meta_boot *)omi->omi_som; 1531 for (i = 0; i < SR_MAX_BOOT_DISKS; i++) { 1532 if (bcmp(rootduid, sbm->sbm_boot_duid[i], 1533 sizeof(rootduid)) == 0) { 1534 memcpy(rootduid, sbm->sbm_root_duid, 1535 sizeof(rootduid)); 1536 DNPRINTF(SR_D_MISC, "%s: root duid " 1537 "mapped to %s\n", DEVNAME(sc), 1538 duid_format(rootduid)); 1539 return; 1540 } 1541 } 1542 } 1543 } 1544 } 1545 1546 int 1547 sr_meta_native_probe(struct sr_softc *sc, struct sr_chunk *ch_entry) 1548 { 1549 struct disklabel label; 1550 char *devname; 1551 int error, part; 1552 u_int64_t size; 1553 1554 DNPRINTF(SR_D_META, "%s: sr_meta_native_probe(%s)\n", 1555 DEVNAME(sc), ch_entry->src_devname); 1556 1557 devname = ch_entry->src_devname; 1558 part = DISKPART(ch_entry->src_dev_mm); 1559 1560 /* get disklabel */ 1561 error = VOP_IOCTL(ch_entry->src_vn, DIOCGDINFO, (caddr_t)&label, FREAD, 1562 NOCRED, curproc); 1563 if (error) { 1564 DNPRINTF(SR_D_META, "%s: %s can't obtain disklabel\n", 1565 DEVNAME(sc), devname); 1566 goto unwind; 1567 } 1568 memcpy(ch_entry->src_duid, label.d_uid, sizeof(ch_entry->src_duid)); 1569 1570 /* make sure the partition is of the right type */ 1571 if (label.d_partitions[part].p_fstype != FS_RAID) { 1572 DNPRINTF(SR_D_META, 1573 "%s: %s partition not of type RAID (%d)\n", DEVNAME(sc), 1574 devname, 1575 label.d_partitions[part].p_fstype); 1576 goto unwind; 1577 } 1578 1579 size = DL_SECTOBLK(&label, DL_GETPSIZE(&label.d_partitions[part])); 1580 if (size <= SR_DATA_OFFSET) { 1581 DNPRINTF(SR_D_META, "%s: %s partition too small\n", DEVNAME(sc), 1582 devname); 1583 goto unwind; 1584 } 1585 size -= SR_DATA_OFFSET; 1586 if (size > INT64_MAX) { 1587 DNPRINTF(SR_D_META, "%s: %s partition too large\n", DEVNAME(sc), 1588 devname); 1589 goto unwind; 1590 } 1591 ch_entry->src_size = size; 1592 ch_entry->src_secsize = label.d_secsize; 1593 1594 DNPRINTF(SR_D_META, "%s: probe found %s size %lld\n", DEVNAME(sc), 1595 devname, (long long)size); 1596 1597 return (SR_META_F_NATIVE); 1598 unwind: 1599 DNPRINTF(SR_D_META, "%s: invalid device: %s\n", DEVNAME(sc), 1600 devname ? devname : "nodev"); 1601 return (SR_META_F_INVALID); 1602 } 1603 1604 int 1605 sr_meta_native_attach(struct sr_discipline *sd, int force) 1606 { 1607 struct sr_softc *sc = sd->sd_sc; 1608 struct sr_chunk_head *cl = &sd->sd_vol.sv_chunk_list; 1609 struct sr_metadata *md = NULL; 1610 struct sr_chunk *ch_entry, *ch_next; 1611 struct sr_uuid uuid; 1612 u_int64_t version = 0; 1613 int sr, not_sr, rv = 1, d, expected = -1, old_meta = 0; 1614 1615 DNPRINTF(SR_D_META, "%s: sr_meta_native_attach\n", DEVNAME(sc)); 1616 1617 md = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, M_ZERO | M_NOWAIT); 1618 if (md == NULL) { 1619 sr_error(sc, "not enough memory for metadata buffer"); 1620 goto bad; 1621 } 1622 1623 bzero(&uuid, sizeof uuid); 1624 1625 sr = not_sr = d = 0; 1626 SLIST_FOREACH(ch_entry, cl, src_link) { 1627 if (ch_entry->src_dev_mm == NODEV) 1628 continue; 1629 1630 if (sr_meta_native_read(sd, ch_entry->src_dev_mm, md, NULL)) { 1631 sr_error(sc, "could not read native metadata"); 1632 goto bad; 1633 } 1634 1635 if (md->ssdi.ssd_magic == SR_MAGIC) { 1636 sr++; 1637 ch_entry->src_meta.scmi.scm_chunk_id = 1638 md->ssdi.ssd_chunk_id; 1639 if (d == 0) { 1640 memcpy(&uuid, &md->ssdi.ssd_uuid, sizeof uuid); 1641 expected = md->ssdi.ssd_chunk_no; 1642 version = md->ssd_ondisk; 1643 d++; 1644 continue; 1645 } else if (bcmp(&md->ssdi.ssd_uuid, &uuid, 1646 sizeof uuid)) { 1647 sr_error(sc, "not part of the same volume"); 1648 goto bad; 1649 } 1650 if (md->ssd_ondisk != version) { 1651 old_meta++; 1652 version = MAX(md->ssd_ondisk, version); 1653 } 1654 } else 1655 not_sr++; 1656 } 1657 1658 if (sr && not_sr && !force) { 1659 sr_error(sc, "not all chunks are of the native metadata " 1660 "format"); 1661 goto bad; 1662 } 1663 1664 /* mixed metadata versions; mark bad disks offline */ 1665 if (old_meta) { 1666 d = 0; 1667 for (ch_entry = SLIST_FIRST(cl); ch_entry != NULL; 1668 ch_entry = ch_next, d++) { 1669 ch_next = SLIST_NEXT(ch_entry, src_link); 1670 1671 /* XXX do we want to read this again? */ 1672 if (ch_entry->src_dev_mm == NODEV) 1673 panic("src_dev_mm == NODEV"); 1674 if (sr_meta_native_read(sd, ch_entry->src_dev_mm, md, 1675 NULL)) 1676 sr_warn(sc, "could not read native metadata"); 1677 if (md->ssd_ondisk != version) 1678 sd->sd_vol.sv_chunks[d]->src_meta.scm_status = 1679 BIOC_SDOFFLINE; 1680 } 1681 } 1682 1683 if (expected != sr && !force && expected != -1) { 1684 DNPRINTF(SR_D_META, "%s: not all chunks were provided, trying " 1685 "anyway\n", DEVNAME(sc)); 1686 } 1687 1688 rv = 0; 1689 bad: 1690 free(md, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 1691 return (rv); 1692 } 1693 1694 int 1695 sr_meta_native_read(struct sr_discipline *sd, dev_t dev, 1696 struct sr_metadata *md, void *fm) 1697 { 1698 #ifdef SR_DEBUG 1699 struct sr_softc *sc = sd->sd_sc; 1700 #endif 1701 DNPRINTF(SR_D_META, "%s: sr_meta_native_read(0x%x, %p)\n", 1702 DEVNAME(sc), dev, md); 1703 1704 return (sr_meta_rw(sd, dev, md, B_READ)); 1705 } 1706 1707 int 1708 sr_meta_native_write(struct sr_discipline *sd, dev_t dev, 1709 struct sr_metadata *md, void *fm) 1710 { 1711 #ifdef SR_DEBUG 1712 struct sr_softc *sc = sd->sd_sc; 1713 #endif 1714 DNPRINTF(SR_D_META, "%s: sr_meta_native_write(0x%x, %p)\n", 1715 DEVNAME(sc), dev, md); 1716 1717 return (sr_meta_rw(sd, dev, md, B_WRITE)); 1718 } 1719 1720 void 1721 sr_hotplug_register(struct sr_discipline *sd, void *func) 1722 { 1723 struct sr_hotplug_list *mhe; 1724 1725 DNPRINTF(SR_D_MISC, "%s: sr_hotplug_register: %p\n", 1726 DEVNAME(sd->sd_sc), func); 1727 1728 /* make sure we aren't on the list yet */ 1729 SLIST_FOREACH(mhe, &sr_hotplug_callbacks, shl_link) 1730 if (mhe->sh_hotplug == func) 1731 return; 1732 1733 mhe = malloc(sizeof(struct sr_hotplug_list), M_DEVBUF, 1734 M_WAITOK | M_ZERO); 1735 mhe->sh_hotplug = func; 1736 mhe->sh_sd = sd; 1737 SLIST_INSERT_HEAD(&sr_hotplug_callbacks, mhe, shl_link); 1738 } 1739 1740 void 1741 sr_hotplug_unregister(struct sr_discipline *sd, void *func) 1742 { 1743 struct sr_hotplug_list *mhe; 1744 1745 DNPRINTF(SR_D_MISC, "%s: sr_hotplug_unregister: %s %p\n", 1746 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, func); 1747 1748 /* make sure we are on the list yet */ 1749 SLIST_FOREACH(mhe, &sr_hotplug_callbacks, shl_link) { 1750 if (mhe->sh_hotplug == func) 1751 break; 1752 } 1753 if (mhe != NULL) { 1754 SLIST_REMOVE(&sr_hotplug_callbacks, mhe, 1755 sr_hotplug_list, shl_link); 1756 free(mhe, M_DEVBUF, sizeof(*mhe)); 1757 } 1758 } 1759 1760 void 1761 sr_disk_attach(struct disk *diskp, int action) 1762 { 1763 struct sr_hotplug_list *mhe; 1764 1765 SLIST_FOREACH(mhe, &sr_hotplug_callbacks, shl_link) 1766 if (mhe->sh_sd->sd_ready) 1767 mhe->sh_hotplug(mhe->sh_sd, diskp, action); 1768 } 1769 1770 int 1771 sr_match(struct device *parent, void *match, void *aux) 1772 { 1773 return (1); 1774 } 1775 1776 void 1777 sr_attach(struct device *parent, struct device *self, void *aux) 1778 { 1779 struct sr_softc *sc = (void *)self; 1780 struct scsibus_attach_args saa; 1781 1782 DNPRINTF(SR_D_MISC, "\n%s: sr_attach", DEVNAME(sc)); 1783 1784 if (softraid0 == NULL) 1785 softraid0 = sc; 1786 1787 rw_init(&sc->sc_lock, "sr_lock"); 1788 rw_init(&sc->sc_hs_lock, "sr_hs_lock"); 1789 1790 SLIST_INIT(&sr_hotplug_callbacks); 1791 TAILQ_INIT(&sc->sc_dis_list); 1792 SLIST_INIT(&sc->sc_hotspare_list); 1793 1794 #if NBIO > 0 1795 if (bio_register(&sc->sc_dev, sr_bio_ioctl) != 0) 1796 printf("%s: controller registration failed", DEVNAME(sc)); 1797 #endif /* NBIO > 0 */ 1798 1799 #ifndef SMALL_KERNEL 1800 strlcpy(sc->sc_sensordev.xname, DEVNAME(sc), 1801 sizeof(sc->sc_sensordev.xname)); 1802 sensordev_install(&sc->sc_sensordev); 1803 #endif /* SMALL_KERNEL */ 1804 1805 printf("\n"); 1806 1807 saa.saa_adapter_softc = sc; 1808 saa.saa_adapter = &sr_switch; 1809 saa.saa_adapter_target = SDEV_NO_ADAPTER_TARGET; 1810 saa.saa_adapter_buswidth = SR_MAX_LD; 1811 saa.saa_luns = 1; 1812 saa.saa_openings = 0; 1813 saa.saa_pool = NULL; 1814 saa.saa_quirks = saa.saa_flags = 0; 1815 saa.saa_wwpn = saa.saa_wwnn = 0; 1816 1817 sc->sc_scsibus = (struct scsibus_softc *)config_found(&sc->sc_dev, &saa, 1818 scsiprint); 1819 1820 softraid_disk_attach = sr_disk_attach; 1821 1822 sr_boot_assembly(sc); 1823 1824 explicit_bzero(sr_bootkey, sizeof(sr_bootkey)); 1825 } 1826 1827 int 1828 sr_detach(struct device *self, int flags) 1829 { 1830 struct sr_softc *sc = (void *)self; 1831 int rv; 1832 1833 DNPRINTF(SR_D_MISC, "%s: sr_detach\n", DEVNAME(sc)); 1834 1835 softraid_disk_attach = NULL; 1836 1837 sr_shutdown(0); 1838 1839 #ifndef SMALL_KERNEL 1840 if (sc->sc_sensor_task != NULL) 1841 sensor_task_unregister(sc->sc_sensor_task); 1842 sensordev_deinstall(&sc->sc_sensordev); 1843 #endif /* SMALL_KERNEL */ 1844 1845 if (sc->sc_scsibus != NULL) { 1846 rv = config_detach((struct device *)sc->sc_scsibus, flags); 1847 if (rv != 0) 1848 return (rv); 1849 sc->sc_scsibus = NULL; 1850 } 1851 1852 return (0); 1853 } 1854 1855 void 1856 sr_info(struct sr_softc *sc, const char *fmt, ...) 1857 { 1858 va_list ap; 1859 1860 rw_assert_wrlock(&sc->sc_lock); 1861 1862 va_start(ap, fmt); 1863 bio_status(&sc->sc_status, 0, BIO_MSG_INFO, fmt, &ap); 1864 va_end(ap); 1865 } 1866 1867 void 1868 sr_warn(struct sr_softc *sc, const char *fmt, ...) 1869 { 1870 va_list ap; 1871 1872 rw_assert_wrlock(&sc->sc_lock); 1873 1874 va_start(ap, fmt); 1875 bio_status(&sc->sc_status, 1, BIO_MSG_WARN, fmt, &ap); 1876 va_end(ap); 1877 } 1878 1879 void 1880 sr_error(struct sr_softc *sc, const char *fmt, ...) 1881 { 1882 va_list ap; 1883 1884 rw_assert_wrlock(&sc->sc_lock); 1885 1886 va_start(ap, fmt); 1887 bio_status(&sc->sc_status, 1, BIO_MSG_ERROR, fmt, &ap); 1888 va_end(ap); 1889 } 1890 1891 int 1892 sr_ccb_alloc(struct sr_discipline *sd) 1893 { 1894 struct sr_ccb *ccb; 1895 int i; 1896 1897 if (!sd) 1898 return (1); 1899 1900 DNPRINTF(SR_D_CCB, "%s: sr_ccb_alloc\n", DEVNAME(sd->sd_sc)); 1901 1902 if (sd->sd_ccb) 1903 return (1); 1904 1905 sd->sd_ccb = mallocarray(sd->sd_max_wu, 1906 sd->sd_max_ccb_per_wu * sizeof(struct sr_ccb), 1907 M_DEVBUF, M_WAITOK | M_ZERO); 1908 TAILQ_INIT(&sd->sd_ccb_freeq); 1909 for (i = 0; i < sd->sd_max_wu * sd->sd_max_ccb_per_wu; i++) { 1910 ccb = &sd->sd_ccb[i]; 1911 ccb->ccb_dis = sd; 1912 sr_ccb_put(ccb); 1913 } 1914 1915 DNPRINTF(SR_D_CCB, "%s: sr_ccb_alloc ccb: %d\n", 1916 DEVNAME(sd->sd_sc), sd->sd_max_wu * sd->sd_max_ccb_per_wu); 1917 1918 return (0); 1919 } 1920 1921 void 1922 sr_ccb_free(struct sr_discipline *sd) 1923 { 1924 struct sr_ccb *ccb; 1925 1926 if (!sd) 1927 return; 1928 1929 DNPRINTF(SR_D_CCB, "%s: sr_ccb_free %p\n", DEVNAME(sd->sd_sc), sd); 1930 1931 while ((ccb = TAILQ_FIRST(&sd->sd_ccb_freeq)) != NULL) 1932 TAILQ_REMOVE(&sd->sd_ccb_freeq, ccb, ccb_link); 1933 1934 free(sd->sd_ccb, M_DEVBUF, sd->sd_max_wu * sd->sd_max_ccb_per_wu * 1935 sizeof(struct sr_ccb)); 1936 } 1937 1938 struct sr_ccb * 1939 sr_ccb_get(struct sr_discipline *sd) 1940 { 1941 struct sr_ccb *ccb; 1942 int s; 1943 1944 s = splbio(); 1945 1946 ccb = TAILQ_FIRST(&sd->sd_ccb_freeq); 1947 if (ccb) { 1948 TAILQ_REMOVE(&sd->sd_ccb_freeq, ccb, ccb_link); 1949 ccb->ccb_state = SR_CCB_INPROGRESS; 1950 } 1951 1952 splx(s); 1953 1954 DNPRINTF(SR_D_CCB, "%s: sr_ccb_get: %p\n", DEVNAME(sd->sd_sc), 1955 ccb); 1956 1957 return (ccb); 1958 } 1959 1960 void 1961 sr_ccb_put(struct sr_ccb *ccb) 1962 { 1963 struct sr_discipline *sd = ccb->ccb_dis; 1964 int s; 1965 1966 DNPRINTF(SR_D_CCB, "%s: sr_ccb_put: %p\n", DEVNAME(sd->sd_sc), 1967 ccb); 1968 1969 s = splbio(); 1970 1971 ccb->ccb_wu = NULL; 1972 ccb->ccb_state = SR_CCB_FREE; 1973 ccb->ccb_target = -1; 1974 ccb->ccb_opaque = NULL; 1975 1976 TAILQ_INSERT_TAIL(&sd->sd_ccb_freeq, ccb, ccb_link); 1977 1978 splx(s); 1979 } 1980 1981 struct sr_ccb * 1982 sr_ccb_rw(struct sr_discipline *sd, int chunk, daddr_t blkno, 1983 long len, u_int8_t *data, int xsflags, int ccbflags) 1984 { 1985 struct sr_chunk *sc = sd->sd_vol.sv_chunks[chunk]; 1986 struct sr_ccb *ccb = NULL; 1987 int s; 1988 1989 ccb = sr_ccb_get(sd); 1990 if (ccb == NULL) 1991 goto out; 1992 1993 ccb->ccb_flags = ccbflags; 1994 ccb->ccb_target = chunk; 1995 1996 ccb->ccb_buf.b_flags = B_PHYS | B_CALL; 1997 if (ISSET(xsflags, SCSI_DATA_IN)) 1998 ccb->ccb_buf.b_flags |= B_READ; 1999 else 2000 ccb->ccb_buf.b_flags |= B_WRITE; 2001 2002 ccb->ccb_buf.b_blkno = blkno + sd->sd_meta->ssd_data_blkno; 2003 ccb->ccb_buf.b_bcount = len; 2004 ccb->ccb_buf.b_bufsize = len; 2005 ccb->ccb_buf.b_resid = len; 2006 ccb->ccb_buf.b_data = data; 2007 ccb->ccb_buf.b_error = 0; 2008 ccb->ccb_buf.b_iodone = sd->sd_scsi_intr; 2009 ccb->ccb_buf.b_proc = curproc; 2010 ccb->ccb_buf.b_dev = sc->src_dev_mm; 2011 ccb->ccb_buf.b_vp = sc->src_vn; 2012 ccb->ccb_buf.b_bq = NULL; 2013 2014 if (!ISSET(ccb->ccb_buf.b_flags, B_READ)) { 2015 s = splbio(); 2016 ccb->ccb_buf.b_vp->v_numoutput++; 2017 splx(s); 2018 } 2019 2020 LIST_INIT(&ccb->ccb_buf.b_dep); 2021 2022 DNPRINTF(SR_D_DIS, "%s: %s %s ccb " 2023 "b_bcount %ld b_blkno %lld b_flags 0x%0lx b_data %p\n", 2024 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, sd->sd_name, 2025 ccb->ccb_buf.b_bcount, (long long)ccb->ccb_buf.b_blkno, 2026 ccb->ccb_buf.b_flags, ccb->ccb_buf.b_data); 2027 2028 out: 2029 return ccb; 2030 } 2031 2032 void 2033 sr_ccb_done(struct sr_ccb *ccb) 2034 { 2035 struct sr_workunit *wu = ccb->ccb_wu; 2036 struct sr_discipline *sd = wu->swu_dis; 2037 struct sr_softc *sc = sd->sd_sc; 2038 2039 DNPRINTF(SR_D_INTR, "%s: %s %s ccb done b_bcount %ld b_resid %zu" 2040 " b_flags 0x%0lx block %lld target %d\n", 2041 DEVNAME(sc), sd->sd_meta->ssd_devname, sd->sd_name, 2042 ccb->ccb_buf.b_bcount, ccb->ccb_buf.b_resid, ccb->ccb_buf.b_flags, 2043 (long long)ccb->ccb_buf.b_blkno, ccb->ccb_target); 2044 2045 splassert(IPL_BIO); 2046 2047 if (ccb->ccb_target == -1) 2048 panic("%s: invalid target on wu: %p", DEVNAME(sc), wu); 2049 2050 if (ccb->ccb_buf.b_flags & B_ERROR) { 2051 DNPRINTF(SR_D_INTR, "%s: i/o error on block %lld target %d\n", 2052 DEVNAME(sc), (long long)ccb->ccb_buf.b_blkno, 2053 ccb->ccb_target); 2054 if (ISSET(sd->sd_capabilities, SR_CAP_REDUNDANT)) 2055 sd->sd_set_chunk_state(sd, ccb->ccb_target, 2056 BIOC_SDOFFLINE); 2057 else 2058 printf("%s: %s: i/o error %d @ %s block %lld\n", 2059 DEVNAME(sc), sd->sd_meta->ssd_devname, 2060 ccb->ccb_buf.b_error, sd->sd_name, 2061 (long long)ccb->ccb_buf.b_blkno); 2062 ccb->ccb_state = SR_CCB_FAILED; 2063 wu->swu_ios_failed++; 2064 } else { 2065 ccb->ccb_state = SR_CCB_OK; 2066 wu->swu_ios_succeeded++; 2067 } 2068 2069 wu->swu_ios_complete++; 2070 } 2071 2072 int 2073 sr_wu_alloc(struct sr_discipline *sd) 2074 { 2075 struct sr_workunit *wu; 2076 int i, no_wu; 2077 2078 DNPRINTF(SR_D_WU, "%s: sr_wu_alloc %p %d\n", DEVNAME(sd->sd_sc), 2079 sd, sd->sd_max_wu); 2080 2081 no_wu = sd->sd_max_wu; 2082 sd->sd_wu_pending = no_wu; 2083 2084 mtx_init(&sd->sd_wu_mtx, IPL_BIO); 2085 TAILQ_INIT(&sd->sd_wu); 2086 TAILQ_INIT(&sd->sd_wu_freeq); 2087 TAILQ_INIT(&sd->sd_wu_pendq); 2088 TAILQ_INIT(&sd->sd_wu_defq); 2089 2090 for (i = 0; i < no_wu; i++) { 2091 wu = malloc(sd->sd_wu_size, M_DEVBUF, M_WAITOK | M_ZERO); 2092 TAILQ_INSERT_TAIL(&sd->sd_wu, wu, swu_next); 2093 TAILQ_INIT(&wu->swu_ccb); 2094 wu->swu_dis = sd; 2095 task_set(&wu->swu_task, sr_wu_done_callback, wu); 2096 sr_wu_put(sd, wu); 2097 } 2098 2099 return (0); 2100 } 2101 2102 void 2103 sr_wu_free(struct sr_discipline *sd) 2104 { 2105 struct sr_workunit *wu; 2106 2107 DNPRINTF(SR_D_WU, "%s: sr_wu_free %p\n", DEVNAME(sd->sd_sc), sd); 2108 2109 while ((wu = TAILQ_FIRST(&sd->sd_wu_freeq)) != NULL) 2110 TAILQ_REMOVE(&sd->sd_wu_freeq, wu, swu_link); 2111 while ((wu = TAILQ_FIRST(&sd->sd_wu_pendq)) != NULL) 2112 TAILQ_REMOVE(&sd->sd_wu_pendq, wu, swu_link); 2113 while ((wu = TAILQ_FIRST(&sd->sd_wu_defq)) != NULL) 2114 TAILQ_REMOVE(&sd->sd_wu_defq, wu, swu_link); 2115 2116 while ((wu = TAILQ_FIRST(&sd->sd_wu)) != NULL) { 2117 TAILQ_REMOVE(&sd->sd_wu, wu, swu_next); 2118 free(wu, M_DEVBUF, sd->sd_wu_size); 2119 } 2120 } 2121 2122 void * 2123 sr_wu_get(void *xsd) 2124 { 2125 struct sr_discipline *sd = (struct sr_discipline *)xsd; 2126 struct sr_workunit *wu; 2127 2128 mtx_enter(&sd->sd_wu_mtx); 2129 wu = TAILQ_FIRST(&sd->sd_wu_freeq); 2130 if (wu) { 2131 TAILQ_REMOVE(&sd->sd_wu_freeq, wu, swu_link); 2132 sd->sd_wu_pending++; 2133 } 2134 mtx_leave(&sd->sd_wu_mtx); 2135 2136 DNPRINTF(SR_D_WU, "%s: sr_wu_get: %p\n", DEVNAME(sd->sd_sc), wu); 2137 2138 return (wu); 2139 } 2140 2141 void 2142 sr_wu_put(void *xsd, void *xwu) 2143 { 2144 struct sr_discipline *sd = (struct sr_discipline *)xsd; 2145 struct sr_workunit *wu = (struct sr_workunit *)xwu; 2146 2147 DNPRINTF(SR_D_WU, "%s: sr_wu_put: %p\n", DEVNAME(sd->sd_sc), wu); 2148 2149 sr_wu_release_ccbs(wu); 2150 sr_wu_init(sd, wu); 2151 2152 mtx_enter(&sd->sd_wu_mtx); 2153 TAILQ_INSERT_TAIL(&sd->sd_wu_freeq, wu, swu_link); 2154 sd->sd_wu_pending--; 2155 mtx_leave(&sd->sd_wu_mtx); 2156 } 2157 2158 void 2159 sr_wu_init(struct sr_discipline *sd, struct sr_workunit *wu) 2160 { 2161 int s; 2162 2163 s = splbio(); 2164 if (wu->swu_cb_active == 1) 2165 panic("%s: sr_wu_init got active wu", DEVNAME(sd->sd_sc)); 2166 splx(s); 2167 2168 wu->swu_xs = NULL; 2169 wu->swu_state = SR_WU_FREE; 2170 wu->swu_flags = 0; 2171 wu->swu_blk_start = 0; 2172 wu->swu_blk_end = 0; 2173 wu->swu_collider = NULL; 2174 } 2175 2176 void 2177 sr_wu_enqueue_ccb(struct sr_workunit *wu, struct sr_ccb *ccb) 2178 { 2179 struct sr_discipline *sd = wu->swu_dis; 2180 int s; 2181 2182 s = splbio(); 2183 if (wu->swu_cb_active == 1) 2184 panic("%s: sr_wu_enqueue_ccb got active wu", 2185 DEVNAME(sd->sd_sc)); 2186 ccb->ccb_wu = wu; 2187 wu->swu_io_count++; 2188 TAILQ_INSERT_TAIL(&wu->swu_ccb, ccb, ccb_link); 2189 splx(s); 2190 } 2191 2192 void 2193 sr_wu_release_ccbs(struct sr_workunit *wu) 2194 { 2195 struct sr_ccb *ccb; 2196 2197 /* Return all ccbs that are associated with this workunit. */ 2198 while ((ccb = TAILQ_FIRST(&wu->swu_ccb)) != NULL) { 2199 TAILQ_REMOVE(&wu->swu_ccb, ccb, ccb_link); 2200 sr_ccb_put(ccb); 2201 } 2202 2203 wu->swu_io_count = 0; 2204 wu->swu_ios_complete = 0; 2205 wu->swu_ios_failed = 0; 2206 wu->swu_ios_succeeded = 0; 2207 } 2208 2209 void 2210 sr_wu_done(struct sr_workunit *wu) 2211 { 2212 struct sr_discipline *sd = wu->swu_dis; 2213 2214 DNPRINTF(SR_D_INTR, "%s: sr_wu_done count %d completed %d failed %d\n", 2215 DEVNAME(sd->sd_sc), wu->swu_io_count, wu->swu_ios_complete, 2216 wu->swu_ios_failed); 2217 2218 if (wu->swu_ios_complete < wu->swu_io_count) 2219 return; 2220 2221 task_add(sd->sd_taskq, &wu->swu_task); 2222 } 2223 2224 void 2225 sr_wu_done_callback(void *xwu) 2226 { 2227 struct sr_workunit *wu = xwu; 2228 struct sr_discipline *sd = wu->swu_dis; 2229 struct scsi_xfer *xs = wu->swu_xs; 2230 struct sr_workunit *wup; 2231 int s; 2232 2233 /* 2234 * The SR_WUF_DISCIPLINE or SR_WUF_REBUILD flag must be set if 2235 * the work unit is not associated with a scsi_xfer. 2236 */ 2237 KASSERT(xs != NULL || 2238 (wu->swu_flags & (SR_WUF_DISCIPLINE|SR_WUF_REBUILD))); 2239 2240 s = splbio(); 2241 2242 if (xs != NULL) { 2243 if (wu->swu_ios_failed) 2244 xs->error = XS_DRIVER_STUFFUP; 2245 else 2246 xs->error = XS_NOERROR; 2247 } 2248 2249 if (sd->sd_scsi_wu_done) { 2250 if (sd->sd_scsi_wu_done(wu) == SR_WU_RESTART) 2251 goto done; 2252 } 2253 2254 /* Remove work unit from pending queue. */ 2255 TAILQ_FOREACH(wup, &sd->sd_wu_pendq, swu_link) 2256 if (wup == wu) 2257 break; 2258 if (wup == NULL) 2259 panic("%s: wu %p not on pending queue", 2260 DEVNAME(sd->sd_sc), wu); 2261 TAILQ_REMOVE(&sd->sd_wu_pendq, wu, swu_link); 2262 2263 if (wu->swu_collider) { 2264 if (wu->swu_ios_failed) 2265 sr_raid_recreate_wu(wu->swu_collider); 2266 2267 /* XXX Should the collider be failed if this xs failed? */ 2268 sr_raid_startwu(wu->swu_collider); 2269 } 2270 2271 /* 2272 * If a discipline provides its own sd_scsi_done function, then it 2273 * is responsible for calling sr_scsi_done() once I/O is complete. 2274 */ 2275 if (wu->swu_flags & SR_WUF_REBUILD) 2276 wu->swu_flags |= SR_WUF_REBUILDIOCOMP; 2277 if (wu->swu_flags & SR_WUF_WAKEUP) 2278 wakeup(wu); 2279 if (sd->sd_scsi_done) 2280 sd->sd_scsi_done(wu); 2281 else if (wu->swu_flags & SR_WUF_DISCIPLINE) 2282 sr_scsi_wu_put(sd, wu); 2283 else if (!(wu->swu_flags & SR_WUF_REBUILD)) 2284 sr_scsi_done(sd, xs); 2285 2286 done: 2287 splx(s); 2288 } 2289 2290 struct sr_workunit * 2291 sr_scsi_wu_get(struct sr_discipline *sd, int flags) 2292 { 2293 return scsi_io_get(&sd->sd_iopool, flags); 2294 } 2295 2296 void 2297 sr_scsi_wu_put(struct sr_discipline *sd, struct sr_workunit *wu) 2298 { 2299 scsi_io_put(&sd->sd_iopool, wu); 2300 2301 if (sd->sd_sync && sd->sd_wu_pending == 0) 2302 wakeup(sd); 2303 } 2304 2305 void 2306 sr_scsi_done(struct sr_discipline *sd, struct scsi_xfer *xs) 2307 { 2308 DNPRINTF(SR_D_DIS, "%s: sr_scsi_done: xs %p\n", DEVNAME(sd->sd_sc), xs); 2309 2310 if (xs->error == XS_NOERROR) 2311 xs->resid = 0; 2312 2313 scsi_done(xs); 2314 2315 if (sd->sd_sync && sd->sd_wu_pending == 0) 2316 wakeup(sd); 2317 } 2318 2319 void 2320 sr_scsi_cmd(struct scsi_xfer *xs) 2321 { 2322 struct scsi_link *link = xs->sc_link; 2323 struct sr_softc *sc = link->bus->sb_adapter_softc; 2324 struct sr_workunit *wu = xs->io; 2325 struct sr_discipline *sd; 2326 2327 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd target %d xs %p flags %#x\n", 2328 DEVNAME(sc), link->target, xs, xs->flags); 2329 2330 sd = sc->sc_targets[link->target]; 2331 if (sd == NULL) 2332 panic("%s: sr_scsi_cmd NULL discipline", DEVNAME(sc)); 2333 2334 if (sd->sd_deleted) { 2335 printf("%s: %s device is being deleted, failing io\n", 2336 DEVNAME(sc), sd->sd_meta->ssd_devname); 2337 goto stuffup; 2338 } 2339 2340 /* scsi layer *can* re-send wu without calling sr_wu_put(). */ 2341 sr_wu_release_ccbs(wu); 2342 sr_wu_init(sd, wu); 2343 wu->swu_state = SR_WU_INPROGRESS; 2344 wu->swu_xs = xs; 2345 2346 switch (xs->cmd.opcode) { 2347 case READ_COMMAND: 2348 case READ_10: 2349 case READ_16: 2350 case WRITE_COMMAND: 2351 case WRITE_10: 2352 case WRITE_16: 2353 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: READ/WRITE %02x\n", 2354 DEVNAME(sc), xs->cmd.opcode); 2355 if (sd->sd_scsi_rw(wu)) 2356 goto stuffup; 2357 break; 2358 2359 case SYNCHRONIZE_CACHE: 2360 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: SYNCHRONIZE_CACHE\n", 2361 DEVNAME(sc)); 2362 if (sd->sd_scsi_sync(wu)) 2363 goto stuffup; 2364 goto complete; 2365 2366 case TEST_UNIT_READY: 2367 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: TEST_UNIT_READY\n", 2368 DEVNAME(sc)); 2369 if (sd->sd_scsi_tur(wu)) 2370 goto stuffup; 2371 goto complete; 2372 2373 case START_STOP: 2374 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: START_STOP\n", 2375 DEVNAME(sc)); 2376 if (sd->sd_scsi_start_stop(wu)) 2377 goto stuffup; 2378 goto complete; 2379 2380 case INQUIRY: 2381 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: INQUIRY\n", 2382 DEVNAME(sc)); 2383 if (sd->sd_scsi_inquiry(wu)) 2384 goto stuffup; 2385 goto complete; 2386 2387 case READ_CAPACITY: 2388 case READ_CAPACITY_16: 2389 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd READ CAPACITY 0x%02x\n", 2390 DEVNAME(sc), xs->cmd.opcode); 2391 if (sd->sd_scsi_read_cap(wu)) 2392 goto stuffup; 2393 goto complete; 2394 2395 case REQUEST_SENSE: 2396 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd REQUEST SENSE\n", 2397 DEVNAME(sc)); 2398 if (sd->sd_scsi_req_sense(wu)) 2399 goto stuffup; 2400 goto complete; 2401 2402 default: 2403 DNPRINTF(SR_D_CMD, "%s: unsupported scsi command %x\n", 2404 DEVNAME(sc), xs->cmd.opcode); 2405 /* XXX might need to add generic function to handle others */ 2406 goto stuffup; 2407 } 2408 2409 return; 2410 stuffup: 2411 if (sd->sd_scsi_sense.error_code) { 2412 xs->error = XS_SENSE; 2413 memcpy(&xs->sense, &sd->sd_scsi_sense, sizeof(xs->sense)); 2414 bzero(&sd->sd_scsi_sense, sizeof(sd->sd_scsi_sense)); 2415 } else { 2416 xs->error = XS_DRIVER_STUFFUP; 2417 } 2418 complete: 2419 sr_scsi_done(sd, xs); 2420 } 2421 2422 int 2423 sr_scsi_probe(struct scsi_link *link) 2424 { 2425 struct sr_softc *sc = link->bus->sb_adapter_softc; 2426 struct sr_discipline *sd; 2427 2428 KASSERT(link->target < SR_MAX_LD && link->lun == 0); 2429 2430 sd = sc->sc_targets[link->target]; 2431 if (sd == NULL) 2432 return (ENODEV); 2433 2434 link->pool = &sd->sd_iopool; 2435 if (sd->sd_openings) 2436 link->openings = sd->sd_openings(sd); 2437 else 2438 link->openings = sd->sd_max_wu; 2439 2440 return (0); 2441 } 2442 2443 int 2444 sr_scsi_ioctl(struct scsi_link *link, u_long cmd, caddr_t addr, int flag) 2445 { 2446 struct sr_softc *sc = link->bus->sb_adapter_softc; 2447 struct sr_discipline *sd; 2448 2449 sd = sc->sc_targets[link->target]; 2450 if (sd == NULL) 2451 return (ENODEV); 2452 2453 DNPRINTF(SR_D_IOCTL, "%s: %s sr_scsi_ioctl cmd: %#lx\n", 2454 DEVNAME(sc), sd->sd_meta->ssd_devname, cmd); 2455 2456 /* Pass bio ioctls through to the bio handler. */ 2457 if (IOCGROUP(cmd) == 'B') 2458 return (sr_bio_handler(sc, sd, cmd, (struct bio *)addr)); 2459 2460 switch (cmd) { 2461 case DIOCGCACHE: 2462 case DIOCSCACHE: 2463 return (EOPNOTSUPP); 2464 default: 2465 return (ENOTTY); 2466 } 2467 } 2468 2469 int 2470 sr_bio_ioctl(struct device *dev, u_long cmd, caddr_t addr) 2471 { 2472 struct sr_softc *sc = (struct sr_softc *) dev; 2473 DNPRINTF(SR_D_IOCTL, "%s: sr_bio_ioctl\n", DEVNAME(sc)); 2474 2475 return sr_bio_handler(sc, NULL, cmd, (struct bio *)addr); 2476 } 2477 2478 int 2479 sr_bio_handler(struct sr_softc *sc, struct sr_discipline *sd, u_long cmd, 2480 struct bio *bio) 2481 { 2482 int rv = 0; 2483 2484 DNPRINTF(SR_D_IOCTL, "%s: sr_bio_handler ", DEVNAME(sc)); 2485 2486 rw_enter_write(&sc->sc_lock); 2487 2488 bio_status_init(&sc->sc_status, &sc->sc_dev); 2489 2490 switch (cmd) { 2491 case BIOCINQ: 2492 DNPRINTF(SR_D_IOCTL, "inq\n"); 2493 rv = sr_ioctl_inq(sc, (struct bioc_inq *)bio); 2494 break; 2495 2496 case BIOCVOL: 2497 DNPRINTF(SR_D_IOCTL, "vol\n"); 2498 rv = sr_ioctl_vol(sc, (struct bioc_vol *)bio); 2499 break; 2500 2501 case BIOCDISK: 2502 DNPRINTF(SR_D_IOCTL, "disk\n"); 2503 rv = sr_ioctl_disk(sc, (struct bioc_disk *)bio); 2504 break; 2505 2506 case BIOCALARM: 2507 DNPRINTF(SR_D_IOCTL, "alarm\n"); 2508 /*rv = sr_ioctl_alarm(sc, (struct bioc_alarm *)bio); */ 2509 break; 2510 2511 case BIOCBLINK: 2512 DNPRINTF(SR_D_IOCTL, "blink\n"); 2513 /*rv = sr_ioctl_blink(sc, (struct bioc_blink *)bio); */ 2514 break; 2515 2516 case BIOCSETSTATE: 2517 DNPRINTF(SR_D_IOCTL, "setstate\n"); 2518 rv = sr_ioctl_setstate(sc, (struct bioc_setstate *)bio); 2519 break; 2520 2521 case BIOCCREATERAID: 2522 DNPRINTF(SR_D_IOCTL, "createraid\n"); 2523 rv = sr_ioctl_createraid(sc, (struct bioc_createraid *)bio, 2524 1, NULL); 2525 break; 2526 2527 case BIOCDELETERAID: 2528 DNPRINTF(SR_D_IOCTL, "deleteraid\n"); 2529 rv = sr_ioctl_deleteraid(sc, sd, (struct bioc_deleteraid *)bio); 2530 break; 2531 2532 case BIOCDISCIPLINE: 2533 DNPRINTF(SR_D_IOCTL, "discipline\n"); 2534 rv = sr_ioctl_discipline(sc, sd, (struct bioc_discipline *)bio); 2535 break; 2536 2537 case BIOCINSTALLBOOT: 2538 DNPRINTF(SR_D_IOCTL, "installboot\n"); 2539 rv = sr_ioctl_installboot(sc, sd, 2540 (struct bioc_installboot *)bio); 2541 break; 2542 2543 default: 2544 DNPRINTF(SR_D_IOCTL, "invalid ioctl\n"); 2545 rv = ENOTTY; 2546 } 2547 2548 sc->sc_status.bs_status = (rv ? BIO_STATUS_ERROR : BIO_STATUS_SUCCESS); 2549 2550 if (sc->sc_status.bs_msg_count > 0) 2551 rv = 0; 2552 2553 memcpy(&bio->bio_status, &sc->sc_status, sizeof(struct bio_status)); 2554 2555 rw_exit_write(&sc->sc_lock); 2556 2557 return (rv); 2558 } 2559 2560 int 2561 sr_ioctl_inq(struct sr_softc *sc, struct bioc_inq *bi) 2562 { 2563 struct sr_discipline *sd; 2564 int vol = 0, disk = 0; 2565 2566 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2567 vol++; 2568 disk += sd->sd_meta->ssdi.ssd_chunk_no; 2569 } 2570 2571 strlcpy(bi->bi_dev, sc->sc_dev.dv_xname, sizeof(bi->bi_dev)); 2572 bi->bi_novol = vol + sc->sc_hotspare_no; 2573 bi->bi_nodisk = disk + sc->sc_hotspare_no; 2574 2575 return (0); 2576 } 2577 2578 int 2579 sr_ioctl_vol(struct sr_softc *sc, struct bioc_vol *bv) 2580 { 2581 int vol = -1, rv = EINVAL; 2582 struct sr_discipline *sd; 2583 struct sr_chunk *hotspare; 2584 2585 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2586 vol++; 2587 if (vol != bv->bv_volid) 2588 continue; 2589 2590 bv->bv_status = sd->sd_vol_status; 2591 bv->bv_size = sd->sd_meta->ssdi.ssd_size << DEV_BSHIFT; 2592 bv->bv_level = sd->sd_meta->ssdi.ssd_level; 2593 bv->bv_nodisk = sd->sd_meta->ssdi.ssd_chunk_no; 2594 2595 #ifdef CRYPTO 2596 if ((sd->sd_meta->ssdi.ssd_level == 'C' || 2597 sd->sd_meta->ssdi.ssd_level == 0x1C) && 2598 sd->mds.mdd_crypto.key_disk != NULL) 2599 bv->bv_nodisk++; 2600 #endif 2601 if (bv->bv_status == BIOC_SVREBUILD) 2602 bv->bv_percent = sr_rebuild_percent(sd); 2603 2604 strlcpy(bv->bv_dev, sd->sd_meta->ssd_devname, 2605 sizeof(bv->bv_dev)); 2606 strlcpy(bv->bv_vendor, sd->sd_meta->ssdi.ssd_vendor, 2607 sizeof(bv->bv_vendor)); 2608 rv = 0; 2609 goto done; 2610 } 2611 2612 /* Check hotspares list. */ 2613 SLIST_FOREACH(hotspare, &sc->sc_hotspare_list, src_link) { 2614 vol++; 2615 if (vol != bv->bv_volid) 2616 continue; 2617 2618 bv->bv_status = BIOC_SVONLINE; 2619 bv->bv_size = hotspare->src_meta.scmi.scm_size << DEV_BSHIFT; 2620 bv->bv_level = -1; /* Hotspare. */ 2621 bv->bv_nodisk = 1; 2622 strlcpy(bv->bv_dev, hotspare->src_meta.scmi.scm_devname, 2623 sizeof(bv->bv_dev)); 2624 strlcpy(bv->bv_vendor, hotspare->src_meta.scmi.scm_devname, 2625 sizeof(bv->bv_vendor)); 2626 rv = 0; 2627 goto done; 2628 } 2629 2630 done: 2631 return (rv); 2632 } 2633 2634 int 2635 sr_ioctl_disk(struct sr_softc *sc, struct bioc_disk *bd) 2636 { 2637 struct sr_discipline *sd; 2638 struct sr_chunk *src, *hotspare; 2639 int vol = -1, rv = EINVAL; 2640 2641 if (bd->bd_diskid < 0) 2642 goto done; 2643 2644 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2645 vol++; 2646 if (vol != bd->bd_volid) 2647 continue; 2648 2649 if (bd->bd_diskid < sd->sd_meta->ssdi.ssd_chunk_no) 2650 src = sd->sd_vol.sv_chunks[bd->bd_diskid]; 2651 #ifdef CRYPTO 2652 else if (bd->bd_diskid == sd->sd_meta->ssdi.ssd_chunk_no && 2653 (sd->sd_meta->ssdi.ssd_level == 'C' || 2654 sd->sd_meta->ssdi.ssd_level == 0x1C) && 2655 sd->mds.mdd_crypto.key_disk != NULL) 2656 src = sd->mds.mdd_crypto.key_disk; 2657 #endif 2658 else 2659 break; 2660 2661 bd->bd_status = src->src_meta.scm_status; 2662 bd->bd_size = src->src_meta.scmi.scm_size << DEV_BSHIFT; 2663 bd->bd_channel = vol; 2664 bd->bd_target = bd->bd_diskid; 2665 strlcpy(bd->bd_vendor, src->src_meta.scmi.scm_devname, 2666 sizeof(bd->bd_vendor)); 2667 rv = 0; 2668 goto done; 2669 } 2670 2671 /* Check hotspares list. */ 2672 SLIST_FOREACH(hotspare, &sc->sc_hotspare_list, src_link) { 2673 vol++; 2674 if (vol != bd->bd_volid) 2675 continue; 2676 2677 if (bd->bd_diskid != 0) 2678 break; 2679 2680 bd->bd_status = hotspare->src_meta.scm_status; 2681 bd->bd_size = hotspare->src_meta.scmi.scm_size << DEV_BSHIFT; 2682 bd->bd_channel = vol; 2683 bd->bd_target = bd->bd_diskid; 2684 strlcpy(bd->bd_vendor, hotspare->src_meta.scmi.scm_devname, 2685 sizeof(bd->bd_vendor)); 2686 rv = 0; 2687 goto done; 2688 } 2689 2690 done: 2691 return (rv); 2692 } 2693 2694 int 2695 sr_ioctl_setstate(struct sr_softc *sc, struct bioc_setstate *bs) 2696 { 2697 int rv = EINVAL; 2698 int vol = -1, found, c; 2699 struct sr_discipline *sd; 2700 struct sr_chunk *ch_entry; 2701 struct sr_chunk_head *cl; 2702 2703 if (bs->bs_other_id_type == BIOC_SSOTHER_UNUSED) 2704 goto done; 2705 2706 if (bs->bs_status == BIOC_SSHOTSPARE) { 2707 rv = sr_hotspare(sc, (dev_t)bs->bs_other_id); 2708 goto done; 2709 } 2710 2711 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2712 vol++; 2713 if (vol == bs->bs_volid) 2714 break; 2715 } 2716 if (sd == NULL) 2717 goto done; 2718 2719 switch (bs->bs_status) { 2720 case BIOC_SSOFFLINE: 2721 /* Take chunk offline */ 2722 found = c = 0; 2723 cl = &sd->sd_vol.sv_chunk_list; 2724 SLIST_FOREACH(ch_entry, cl, src_link) { 2725 if (ch_entry->src_dev_mm == bs->bs_other_id) { 2726 found = 1; 2727 break; 2728 } 2729 c++; 2730 } 2731 if (found == 0) { 2732 sr_error(sc, "chunk not part of array"); 2733 goto done; 2734 } 2735 2736 /* XXX: check current state first */ 2737 sd->sd_set_chunk_state(sd, c, BIOC_SDOFFLINE); 2738 2739 if (sr_meta_save(sd, SR_META_DIRTY)) { 2740 sr_error(sc, "could not save metadata for %s", 2741 sd->sd_meta->ssd_devname); 2742 goto done; 2743 } 2744 rv = 0; 2745 break; 2746 2747 case BIOC_SDSCRUB: 2748 break; 2749 2750 case BIOC_SSREBUILD: 2751 rv = sr_rebuild_init(sd, (dev_t)bs->bs_other_id, 0); 2752 break; 2753 2754 default: 2755 sr_error(sc, "unsupported state request %d", bs->bs_status); 2756 } 2757 2758 done: 2759 return (rv); 2760 } 2761 2762 int 2763 sr_chunk_in_use(struct sr_softc *sc, dev_t dev) 2764 { 2765 struct sr_discipline *sd; 2766 struct sr_chunk *chunk; 2767 int i; 2768 2769 DNPRINTF(SR_D_MISC, "%s: sr_chunk_in_use(%d)\n", DEVNAME(sc), dev); 2770 2771 if (dev == NODEV) 2772 return BIOC_SDINVALID; 2773 2774 /* See if chunk is already in use. */ 2775 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2776 for (i = 0; i < sd->sd_meta->ssdi.ssd_chunk_no; i++) { 2777 chunk = sd->sd_vol.sv_chunks[i]; 2778 if (chunk->src_dev_mm == dev) 2779 return chunk->src_meta.scm_status; 2780 } 2781 } 2782 2783 /* Check hotspares list. */ 2784 SLIST_FOREACH(chunk, &sc->sc_hotspare_list, src_link) 2785 if (chunk->src_dev_mm == dev) 2786 return chunk->src_meta.scm_status; 2787 2788 return BIOC_SDINVALID; 2789 } 2790 2791 int 2792 sr_hotspare(struct sr_softc *sc, dev_t dev) 2793 { 2794 struct sr_discipline *sd = NULL; 2795 struct sr_metadata *sm = NULL; 2796 struct sr_meta_chunk *hm; 2797 struct sr_chunk_head *cl; 2798 struct sr_chunk *chunk, *last, *hotspare = NULL; 2799 struct sr_uuid uuid; 2800 struct disklabel label; 2801 struct vnode *vn; 2802 u_int64_t size; 2803 char devname[32]; 2804 int rv = EINVAL; 2805 int c, part, open = 0; 2806 2807 /* 2808 * Add device to global hotspares list. 2809 */ 2810 2811 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 2812 2813 /* Make sure chunk is not already in use. */ 2814 c = sr_chunk_in_use(sc, dev); 2815 if (c != BIOC_SDINVALID && c != BIOC_SDOFFLINE) { 2816 if (c == BIOC_SDHOTSPARE) 2817 sr_error(sc, "%s is already a hotspare", devname); 2818 else 2819 sr_error(sc, "%s is already in use", devname); 2820 goto done; 2821 } 2822 2823 /* XXX - See if there is an existing degraded volume... */ 2824 2825 /* Open device. */ 2826 if (bdevvp(dev, &vn)) { 2827 sr_error(sc, "sr_hotspare: cannot allocate vnode"); 2828 goto done; 2829 } 2830 if (VOP_OPEN(vn, FREAD | FWRITE, NOCRED, curproc)) { 2831 DNPRINTF(SR_D_META,"%s: sr_hotspare cannot open %s\n", 2832 DEVNAME(sc), devname); 2833 vput(vn); 2834 goto fail; 2835 } 2836 open = 1; /* close dev on error */ 2837 2838 /* Get partition details. */ 2839 part = DISKPART(dev); 2840 if (VOP_IOCTL(vn, DIOCGDINFO, (caddr_t)&label, FREAD, 2841 NOCRED, curproc)) { 2842 DNPRINTF(SR_D_META, "%s: sr_hotspare ioctl failed\n", 2843 DEVNAME(sc)); 2844 goto fail; 2845 } 2846 if (label.d_partitions[part].p_fstype != FS_RAID) { 2847 sr_error(sc, "%s partition not of type RAID (%d)", 2848 devname, label.d_partitions[part].p_fstype); 2849 goto fail; 2850 } 2851 2852 /* Calculate partition size. */ 2853 size = DL_SECTOBLK(&label, DL_GETPSIZE(&label.d_partitions[part])); 2854 if (size <= SR_DATA_OFFSET) { 2855 DNPRINTF(SR_D_META, "%s: %s partition too small\n", DEVNAME(sc), 2856 devname); 2857 goto fail; 2858 } 2859 size -= SR_DATA_OFFSET; 2860 if (size > INT64_MAX) { 2861 DNPRINTF(SR_D_META, "%s: %s partition too large\n", DEVNAME(sc), 2862 devname); 2863 goto fail; 2864 } 2865 2866 /* 2867 * Create and populate chunk metadata. 2868 */ 2869 2870 sr_uuid_generate(&uuid); 2871 hotspare = malloc(sizeof(struct sr_chunk), M_DEVBUF, M_WAITOK | M_ZERO); 2872 2873 hotspare->src_dev_mm = dev; 2874 hotspare->src_vn = vn; 2875 strlcpy(hotspare->src_devname, devname, sizeof(hm->scmi.scm_devname)); 2876 hotspare->src_size = size; 2877 2878 hm = &hotspare->src_meta; 2879 hm->scmi.scm_volid = SR_HOTSPARE_VOLID; 2880 hm->scmi.scm_chunk_id = 0; 2881 hm->scmi.scm_size = size; 2882 hm->scmi.scm_coerced_size = size; 2883 strlcpy(hm->scmi.scm_devname, devname, sizeof(hm->scmi.scm_devname)); 2884 memcpy(&hm->scmi.scm_uuid, &uuid, sizeof(struct sr_uuid)); 2885 2886 sr_checksum(sc, hm, &hm->scm_checksum, 2887 sizeof(struct sr_meta_chunk_invariant)); 2888 2889 hm->scm_status = BIOC_SDHOTSPARE; 2890 2891 /* 2892 * Create and populate our own discipline and metadata. 2893 */ 2894 2895 sm = malloc(sizeof(struct sr_metadata), M_DEVBUF, M_WAITOK | M_ZERO); 2896 sm->ssdi.ssd_magic = SR_MAGIC; 2897 sm->ssdi.ssd_version = SR_META_VERSION; 2898 sm->ssd_ondisk = 0; 2899 sm->ssdi.ssd_vol_flags = 0; 2900 memcpy(&sm->ssdi.ssd_uuid, &uuid, sizeof(struct sr_uuid)); 2901 sm->ssdi.ssd_chunk_no = 1; 2902 sm->ssdi.ssd_volid = SR_HOTSPARE_VOLID; 2903 sm->ssdi.ssd_level = SR_HOTSPARE_LEVEL; 2904 sm->ssdi.ssd_size = size; 2905 sm->ssdi.ssd_secsize = label.d_secsize; 2906 strlcpy(sm->ssdi.ssd_vendor, "OPENBSD", sizeof(sm->ssdi.ssd_vendor)); 2907 snprintf(sm->ssdi.ssd_product, sizeof(sm->ssdi.ssd_product), 2908 "SR %s", "HOTSPARE"); 2909 snprintf(sm->ssdi.ssd_revision, sizeof(sm->ssdi.ssd_revision), 2910 "%03d", SR_META_VERSION); 2911 2912 sd = malloc(sizeof(struct sr_discipline), M_DEVBUF, M_WAITOK | M_ZERO); 2913 sd->sd_sc = sc; 2914 sd->sd_meta = sm; 2915 sd->sd_meta_type = SR_META_F_NATIVE; 2916 sd->sd_vol_status = BIOC_SVONLINE; 2917 strlcpy(sd->sd_name, "HOTSPARE", sizeof(sd->sd_name)); 2918 SLIST_INIT(&sd->sd_meta_opt); 2919 2920 /* Add chunk to volume. */ 2921 sd->sd_vol.sv_chunks = malloc(sizeof(struct sr_chunk *), M_DEVBUF, 2922 M_WAITOK | M_ZERO); 2923 sd->sd_vol.sv_chunks[0] = hotspare; 2924 SLIST_INIT(&sd->sd_vol.sv_chunk_list); 2925 SLIST_INSERT_HEAD(&sd->sd_vol.sv_chunk_list, hotspare, src_link); 2926 2927 /* Save metadata. */ 2928 if (sr_meta_save(sd, SR_META_DIRTY)) { 2929 sr_error(sc, "could not save metadata to %s", devname); 2930 goto fail; 2931 } 2932 2933 /* 2934 * Add chunk to hotspare list. 2935 */ 2936 rw_enter_write(&sc->sc_hs_lock); 2937 cl = &sc->sc_hotspare_list; 2938 if (SLIST_EMPTY(cl)) 2939 SLIST_INSERT_HEAD(cl, hotspare, src_link); 2940 else { 2941 SLIST_FOREACH(chunk, cl, src_link) 2942 last = chunk; 2943 SLIST_INSERT_AFTER(last, hotspare, src_link); 2944 } 2945 sc->sc_hotspare_no++; 2946 rw_exit_write(&sc->sc_hs_lock); 2947 2948 rv = 0; 2949 goto done; 2950 2951 fail: 2952 free(hotspare, M_DEVBUF, sizeof(*hotspare)); 2953 2954 done: 2955 if (sd) 2956 free(sd->sd_vol.sv_chunks, M_DEVBUF, 2957 sizeof(sd->sd_vol.sv_chunks)); 2958 free(sd, M_DEVBUF, sizeof(*sd)); 2959 free(sm, M_DEVBUF, sizeof(*sm)); 2960 if (open) { 2961 VOP_CLOSE(vn, FREAD | FWRITE, NOCRED, curproc); 2962 vput(vn); 2963 } 2964 2965 return (rv); 2966 } 2967 2968 void 2969 sr_hotspare_rebuild_callback(void *xsd) 2970 { 2971 struct sr_discipline *sd = xsd; 2972 sr_hotspare_rebuild(sd); 2973 } 2974 2975 void 2976 sr_hotspare_rebuild(struct sr_discipline *sd) 2977 { 2978 struct sr_softc *sc = sd->sd_sc; 2979 struct sr_chunk_head *cl; 2980 struct sr_chunk *hotspare, *chunk = NULL; 2981 struct sr_workunit *wu; 2982 struct sr_ccb *ccb; 2983 int i, s, cid, busy; 2984 2985 /* 2986 * Attempt to locate a hotspare and initiate rebuild. 2987 */ 2988 2989 /* Find first offline chunk. */ 2990 for (cid = 0; cid < sd->sd_meta->ssdi.ssd_chunk_no; cid++) { 2991 if (sd->sd_vol.sv_chunks[cid]->src_meta.scm_status == 2992 BIOC_SDOFFLINE) { 2993 chunk = sd->sd_vol.sv_chunks[cid]; 2994 break; 2995 } 2996 } 2997 if (chunk == NULL) { 2998 printf("%s: no offline chunk found on %s!\n", 2999 DEVNAME(sc), sd->sd_meta->ssd_devname); 3000 return; 3001 } 3002 3003 /* See if we have a suitable hotspare... */ 3004 rw_enter_write(&sc->sc_hs_lock); 3005 cl = &sc->sc_hotspare_list; 3006 SLIST_FOREACH(hotspare, cl, src_link) 3007 if (hotspare->src_size >= chunk->src_size && 3008 hotspare->src_secsize <= sd->sd_meta->ssdi.ssd_secsize) 3009 break; 3010 3011 if (hotspare != NULL) { 3012 3013 printf("%s: %s volume degraded, will attempt to " 3014 "rebuild on hotspare %s\n", DEVNAME(sc), 3015 sd->sd_meta->ssd_devname, hotspare->src_devname); 3016 3017 /* 3018 * Ensure that all pending I/O completes on the failed chunk 3019 * before trying to initiate a rebuild. 3020 */ 3021 i = 0; 3022 do { 3023 busy = 0; 3024 3025 s = splbio(); 3026 TAILQ_FOREACH(wu, &sd->sd_wu_pendq, swu_link) { 3027 TAILQ_FOREACH(ccb, &wu->swu_ccb, ccb_link) { 3028 if (ccb->ccb_target == cid) 3029 busy = 1; 3030 } 3031 } 3032 TAILQ_FOREACH(wu, &sd->sd_wu_defq, swu_link) { 3033 TAILQ_FOREACH(ccb, &wu->swu_ccb, ccb_link) { 3034 if (ccb->ccb_target == cid) 3035 busy = 1; 3036 } 3037 } 3038 splx(s); 3039 3040 if (busy) { 3041 tsleep_nsec(sd, PRIBIO, "sr_hotspare", 3042 SEC_TO_NSEC(1)); 3043 i++; 3044 } 3045 3046 } while (busy && i < 120); 3047 3048 DNPRINTF(SR_D_META, "%s: waited %i seconds for I/O to " 3049 "complete on failed chunk %s\n", DEVNAME(sc), 3050 i, chunk->src_devname); 3051 3052 if (busy) { 3053 printf("%s: pending I/O failed to complete on " 3054 "failed chunk %s, hotspare rebuild aborted...\n", 3055 DEVNAME(sc), chunk->src_devname); 3056 goto done; 3057 } 3058 3059 s = splbio(); 3060 rw_enter_write(&sc->sc_lock); 3061 bio_status_init(&sc->sc_status, &sc->sc_dev); 3062 if (sr_rebuild_init(sd, hotspare->src_dev_mm, 1) == 0) { 3063 3064 /* Remove hotspare from available list. */ 3065 sc->sc_hotspare_no--; 3066 SLIST_REMOVE(cl, hotspare, sr_chunk, src_link); 3067 free(hotspare, M_DEVBUF, sizeof(*hotspare)); 3068 3069 } 3070 rw_exit_write(&sc->sc_lock); 3071 splx(s); 3072 } 3073 done: 3074 rw_exit_write(&sc->sc_hs_lock); 3075 } 3076 3077 int 3078 sr_rebuild_init(struct sr_discipline *sd, dev_t dev, int hotspare) 3079 { 3080 struct sr_softc *sc = sd->sd_sc; 3081 struct sr_chunk *chunk = NULL; 3082 struct sr_meta_chunk *meta; 3083 struct disklabel label; 3084 struct vnode *vn; 3085 u_int64_t size; 3086 int64_t csize; 3087 char devname[32]; 3088 int rv = EINVAL, open = 0; 3089 int cid, i, part, status; 3090 3091 /* 3092 * Attempt to initiate a rebuild onto the specified device. 3093 */ 3094 3095 if (!(sd->sd_capabilities & SR_CAP_REBUILD)) { 3096 sr_error(sc, "discipline does not support rebuild"); 3097 goto done; 3098 } 3099 3100 /* make sure volume is in the right state */ 3101 if (sd->sd_vol_status == BIOC_SVREBUILD) { 3102 sr_error(sc, "rebuild already in progress"); 3103 goto done; 3104 } 3105 if (sd->sd_vol_status != BIOC_SVDEGRADED) { 3106 sr_error(sc, "volume not degraded"); 3107 goto done; 3108 } 3109 3110 /* Find first offline chunk. */ 3111 for (cid = 0; cid < sd->sd_meta->ssdi.ssd_chunk_no; cid++) { 3112 if (sd->sd_vol.sv_chunks[cid]->src_meta.scm_status == 3113 BIOC_SDOFFLINE) { 3114 chunk = sd->sd_vol.sv_chunks[cid]; 3115 break; 3116 } 3117 } 3118 if (chunk == NULL) { 3119 sr_error(sc, "no offline chunks available to rebuild"); 3120 goto done; 3121 } 3122 3123 /* Get coerced size from another online chunk. */ 3124 csize = 0; 3125 for (i = 0; i < sd->sd_meta->ssdi.ssd_chunk_no; i++) { 3126 if (sd->sd_vol.sv_chunks[i]->src_meta.scm_status == 3127 BIOC_SDONLINE) { 3128 meta = &sd->sd_vol.sv_chunks[i]->src_meta; 3129 csize = meta->scmi.scm_coerced_size; 3130 break; 3131 } 3132 } 3133 if (csize == 0) { 3134 sr_error(sc, "no online chunks available for rebuild"); 3135 goto done; 3136 } 3137 3138 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 3139 if (bdevvp(dev, &vn)) { 3140 printf("%s: sr_rebuild_init: can't allocate vnode\n", 3141 DEVNAME(sc)); 3142 goto done; 3143 } 3144 if (VOP_OPEN(vn, FREAD | FWRITE, NOCRED, curproc)) { 3145 DNPRINTF(SR_D_META,"%s: sr_ioctl_setstate can't " 3146 "open %s\n", DEVNAME(sc), devname); 3147 vput(vn); 3148 goto done; 3149 } 3150 open = 1; /* close dev on error */ 3151 3152 /* Get disklabel and check partition. */ 3153 part = DISKPART(dev); 3154 if (VOP_IOCTL(vn, DIOCGDINFO, (caddr_t)&label, FREAD, 3155 NOCRED, curproc)) { 3156 DNPRINTF(SR_D_META, "%s: sr_ioctl_setstate ioctl failed\n", 3157 DEVNAME(sc)); 3158 goto done; 3159 } 3160 if (label.d_partitions[part].p_fstype != FS_RAID) { 3161 sr_error(sc, "%s partition not of type RAID (%d)", 3162 devname, label.d_partitions[part].p_fstype); 3163 goto done; 3164 } 3165 3166 /* Is the partition large enough? */ 3167 size = DL_SECTOBLK(&label, DL_GETPSIZE(&label.d_partitions[part])); 3168 if (size <= sd->sd_meta->ssd_data_blkno) { 3169 sr_error(sc, "%s: %s partition too small", DEVNAME(sc), 3170 devname); 3171 goto done; 3172 } 3173 size -= sd->sd_meta->ssd_data_blkno; 3174 if (size > INT64_MAX) { 3175 sr_error(sc, "%s: %s partition too large", DEVNAME(sc), 3176 devname); 3177 goto done; 3178 } 3179 if (size < csize) { 3180 sr_error(sc, "%s partition too small, at least %lld bytes " 3181 "required", devname, (long long)(csize << DEV_BSHIFT)); 3182 goto done; 3183 } else if (size > csize) 3184 sr_warn(sc, "%s partition too large, wasting %lld bytes", 3185 devname, (long long)((size - csize) << DEV_BSHIFT)); 3186 if (label.d_secsize > sd->sd_meta->ssdi.ssd_secsize) { 3187 sr_error(sc, "%s sector size too large, <= %u bytes " 3188 "required", devname, sd->sd_meta->ssdi.ssd_secsize); 3189 goto done; 3190 } 3191 3192 /* Ensure that this chunk is not already in use. */ 3193 status = sr_chunk_in_use(sc, dev); 3194 if (status != BIOC_SDINVALID && status != BIOC_SDOFFLINE && 3195 !(hotspare && status == BIOC_SDHOTSPARE)) { 3196 sr_error(sc, "%s is already in use", devname); 3197 goto done; 3198 } 3199 3200 /* Reset rebuild counter since we rebuilding onto a new chunk. */ 3201 sd->sd_meta->ssd_rebuild = 0; 3202 3203 open = 0; /* leave dev open from here on out */ 3204 3205 /* Fix up chunk. */ 3206 memcpy(chunk->src_duid, label.d_uid, sizeof(chunk->src_duid)); 3207 chunk->src_dev_mm = dev; 3208 chunk->src_vn = vn; 3209 3210 /* Reconstruct metadata. */ 3211 meta = &chunk->src_meta; 3212 meta->scmi.scm_volid = sd->sd_meta->ssdi.ssd_volid; 3213 meta->scmi.scm_chunk_id = cid; 3214 strlcpy(meta->scmi.scm_devname, devname, 3215 sizeof(meta->scmi.scm_devname)); 3216 meta->scmi.scm_size = size; 3217 meta->scmi.scm_coerced_size = csize; 3218 memcpy(&meta->scmi.scm_uuid, &sd->sd_meta->ssdi.ssd_uuid, 3219 sizeof(meta->scmi.scm_uuid)); 3220 sr_checksum(sc, meta, &meta->scm_checksum, 3221 sizeof(struct sr_meta_chunk_invariant)); 3222 3223 sd->sd_set_chunk_state(sd, cid, BIOC_SDREBUILD); 3224 3225 if (sr_meta_save(sd, SR_META_DIRTY)) { 3226 sr_error(sc, "could not save metadata to %s", devname); 3227 open = 1; 3228 goto done; 3229 } 3230 3231 sr_warn(sc, "rebuild of %s started on %s", 3232 sd->sd_meta->ssd_devname, devname); 3233 3234 sd->sd_reb_abort = 0; 3235 kthread_create_deferred(sr_rebuild_start, sd); 3236 3237 rv = 0; 3238 done: 3239 if (open) { 3240 VOP_CLOSE(vn, FREAD | FWRITE, NOCRED, curproc); 3241 vput(vn); 3242 } 3243 3244 return (rv); 3245 } 3246 3247 int 3248 sr_rebuild_percent(struct sr_discipline *sd) 3249 { 3250 daddr_t rb, sz; 3251 3252 sz = sd->sd_meta->ssdi.ssd_size; 3253 rb = sd->sd_meta->ssd_rebuild; 3254 3255 if (rb > 0) 3256 return (100 - ((sz * 100 - rb * 100) / sz) - 1); 3257 3258 return (0); 3259 } 3260 3261 void 3262 sr_roam_chunks(struct sr_discipline *sd) 3263 { 3264 struct sr_softc *sc = sd->sd_sc; 3265 struct sr_chunk *chunk; 3266 struct sr_meta_chunk *meta; 3267 int roamed = 0; 3268 3269 /* Have any chunks roamed? */ 3270 SLIST_FOREACH(chunk, &sd->sd_vol.sv_chunk_list, src_link) { 3271 meta = &chunk->src_meta; 3272 if (strncmp(meta->scmi.scm_devname, chunk->src_devname, 3273 sizeof(meta->scmi.scm_devname))) { 3274 3275 printf("%s: roaming device %s -> %s\n", DEVNAME(sc), 3276 meta->scmi.scm_devname, chunk->src_devname); 3277 3278 strlcpy(meta->scmi.scm_devname, chunk->src_devname, 3279 sizeof(meta->scmi.scm_devname)); 3280 3281 roamed++; 3282 } 3283 } 3284 3285 if (roamed) 3286 sr_meta_save(sd, SR_META_DIRTY); 3287 } 3288 3289 int 3290 sr_ioctl_createraid(struct sr_softc *sc, struct bioc_createraid *bc, 3291 int user, void *data) 3292 { 3293 struct sr_meta_opt_item *omi; 3294 struct sr_chunk_head *cl; 3295 struct sr_discipline *sd = NULL; 3296 struct sr_chunk *ch_entry; 3297 struct scsi_link *link; 3298 struct device *dev; 3299 char *uuid, devname[32]; 3300 dev_t *dt = NULL; 3301 int i, no_chunk, rv = EINVAL, target, vol; 3302 int no_meta; 3303 3304 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_createraid(%d)\n", 3305 DEVNAME(sc), user); 3306 3307 /* user input */ 3308 if (bc->bc_dev_list_len > BIOC_CRMAXLEN) 3309 goto unwind; 3310 3311 dt = malloc(bc->bc_dev_list_len, M_DEVBUF, M_WAITOK | M_ZERO); 3312 if (user) { 3313 if (copyin(bc->bc_dev_list, dt, bc->bc_dev_list_len) != 0) 3314 goto unwind; 3315 } else 3316 memcpy(dt, bc->bc_dev_list, bc->bc_dev_list_len); 3317 3318 /* Initialise discipline. */ 3319 sd = malloc(sizeof(struct sr_discipline), M_DEVBUF, M_WAITOK | M_ZERO); 3320 sd->sd_sc = sc; 3321 SLIST_INIT(&sd->sd_meta_opt); 3322 sd->sd_taskq = taskq_create("srdis", 1, IPL_BIO, 0); 3323 if (sd->sd_taskq == NULL) { 3324 sr_error(sc, "could not create discipline taskq"); 3325 goto unwind; 3326 } 3327 if (sr_discipline_init(sd, bc->bc_level)) { 3328 sr_error(sc, "could not initialize discipline"); 3329 goto unwind; 3330 } 3331 3332 no_chunk = bc->bc_dev_list_len / sizeof(dev_t); 3333 cl = &sd->sd_vol.sv_chunk_list; 3334 SLIST_INIT(cl); 3335 3336 /* Ensure that chunks are not already in use. */ 3337 for (i = 0; i < no_chunk; i++) { 3338 if (sr_chunk_in_use(sc, dt[i]) != BIOC_SDINVALID) { 3339 sr_meta_getdevname(sc, dt[i], devname, sizeof(devname)); 3340 sr_error(sc, "chunk %s already in use", devname); 3341 goto unwind; 3342 } 3343 } 3344 3345 sd->sd_meta_type = sr_meta_probe(sd, dt, no_chunk); 3346 if (sd->sd_meta_type == SR_META_F_INVALID) { 3347 sr_error(sc, "invalid metadata format"); 3348 goto unwind; 3349 } 3350 3351 if (sr_meta_attach(sd, no_chunk, bc->bc_flags & BIOC_SCFORCE)) 3352 goto unwind; 3353 3354 /* force the raid volume by clearing metadata region */ 3355 if (bc->bc_flags & BIOC_SCFORCE) { 3356 /* make sure disk isn't up and running */ 3357 if (sr_meta_read(sd)) 3358 if (sr_already_assembled(sd)) { 3359 uuid = sr_uuid_format( 3360 &sd->sd_meta->ssdi.ssd_uuid); 3361 sr_error(sc, "disk %s is currently in use; " 3362 "cannot force create", uuid); 3363 free(uuid, M_DEVBUF, 37); 3364 goto unwind; 3365 } 3366 3367 if (sr_meta_clear(sd)) { 3368 sr_error(sc, "failed to clear metadata"); 3369 goto unwind; 3370 } 3371 } 3372 3373 no_meta = sr_meta_read(sd); 3374 if (no_meta == -1) { 3375 3376 /* Corrupt metadata on one or more chunks. */ 3377 sr_error(sc, "one of the chunks has corrupt metadata; " 3378 "aborting assembly"); 3379 goto unwind; 3380 3381 } else if (no_meta == 0) { 3382 3383 /* Initialise volume and chunk metadata. */ 3384 sr_meta_init(sd, bc->bc_level, no_chunk); 3385 sd->sd_vol_status = BIOC_SVONLINE; 3386 sd->sd_meta_flags = bc->bc_flags & BIOC_SCNOAUTOASSEMBLE; 3387 if (sd->sd_create) { 3388 if ((i = sd->sd_create(sd, bc, no_chunk, 3389 sd->sd_vol.sv_chunk_minsz))) { 3390 rv = i; 3391 goto unwind; 3392 } 3393 } 3394 sr_meta_init_complete(sd); 3395 3396 DNPRINTF(SR_D_IOCTL, 3397 "%s: sr_ioctl_createraid: vol_size: %lld\n", 3398 DEVNAME(sc), sd->sd_meta->ssdi.ssd_size); 3399 3400 /* Warn if we've wasted chunk space due to coercing. */ 3401 if ((sd->sd_capabilities & SR_CAP_NON_COERCED) == 0 && 3402 sd->sd_vol.sv_chunk_minsz != sd->sd_vol.sv_chunk_maxsz) 3403 sr_warn(sc, "chunk sizes are not equal; up to %llu " 3404 "blocks wasted per chunk", 3405 sd->sd_vol.sv_chunk_maxsz - 3406 sd->sd_vol.sv_chunk_minsz); 3407 3408 } else { 3409 3410 /* Ensure we are assembling the correct # of chunks. */ 3411 if (bc->bc_level == 0x1C && 3412 sd->sd_meta->ssdi.ssd_chunk_no > no_chunk) { 3413 sr_warn(sc, "trying to bring up %s degraded", 3414 sd->sd_meta->ssd_devname); 3415 } else 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_boot), 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_nsec(sd, PWAIT, "sr_shutdown", MSEC_TO_NSEC(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 case 0x1C: 3995 sr_raid1c_discipline_init(sd); 3996 break; 3997 #endif 3998 case 'c': 3999 sr_concat_discipline_init(sd); 4000 break; 4001 default: 4002 goto bad; 4003 } 4004 4005 rv = 0; 4006 bad: 4007 return (rv); 4008 } 4009 4010 int 4011 sr_raid_inquiry(struct sr_workunit *wu) 4012 { 4013 struct sr_discipline *sd = wu->swu_dis; 4014 struct scsi_xfer *xs = wu->swu_xs; 4015 struct scsi_inquiry *cdb = (struct scsi_inquiry *)&xs->cmd; 4016 struct scsi_inquiry_data inq; 4017 4018 DNPRINTF(SR_D_DIS, "%s: sr_raid_inquiry\n", DEVNAME(sd->sd_sc)); 4019 4020 if (xs->cmdlen != sizeof(*cdb)) 4021 return (EINVAL); 4022 4023 if (ISSET(cdb->flags, SI_EVPD)) 4024 return (EOPNOTSUPP); 4025 4026 bzero(&inq, sizeof(inq)); 4027 inq.device = T_DIRECT; 4028 inq.dev_qual2 = 0; 4029 inq.version = SCSI_REV_2; 4030 inq.response_format = SID_SCSI2_RESPONSE; 4031 inq.additional_length = SID_SCSI2_ALEN; 4032 inq.flags |= SID_CmdQue; 4033 strlcpy(inq.vendor, sd->sd_meta->ssdi.ssd_vendor, 4034 sizeof(inq.vendor)); 4035 strlcpy(inq.product, sd->sd_meta->ssdi.ssd_product, 4036 sizeof(inq.product)); 4037 strlcpy(inq.revision, sd->sd_meta->ssdi.ssd_revision, 4038 sizeof(inq.revision)); 4039 scsi_copy_internal_data(xs, &inq, sizeof(inq)); 4040 4041 return (0); 4042 } 4043 4044 int 4045 sr_raid_read_cap(struct sr_workunit *wu) 4046 { 4047 struct sr_discipline *sd = wu->swu_dis; 4048 struct scsi_xfer *xs = wu->swu_xs; 4049 struct scsi_read_cap_data rcd; 4050 struct scsi_read_cap_data_16 rcd16; 4051 u_int64_t addr; 4052 int rv = 1; 4053 u_int32_t secsize; 4054 4055 DNPRINTF(SR_D_DIS, "%s: sr_raid_read_cap\n", DEVNAME(sd->sd_sc)); 4056 4057 secsize = sd->sd_meta->ssdi.ssd_secsize; 4058 4059 addr = ((sd->sd_meta->ssdi.ssd_size * DEV_BSIZE) / secsize) - 1; 4060 if (xs->cmd.opcode == READ_CAPACITY) { 4061 bzero(&rcd, sizeof(rcd)); 4062 if (addr > 0xffffffffllu) 4063 _lto4b(0xffffffff, rcd.addr); 4064 else 4065 _lto4b(addr, rcd.addr); 4066 _lto4b(secsize, rcd.length); 4067 scsi_copy_internal_data(xs, &rcd, sizeof(rcd)); 4068 rv = 0; 4069 } else if (xs->cmd.opcode == READ_CAPACITY_16) { 4070 bzero(&rcd16, sizeof(rcd16)); 4071 _lto8b(addr, rcd16.addr); 4072 _lto4b(secsize, rcd16.length); 4073 scsi_copy_internal_data(xs, &rcd16, sizeof(rcd16)); 4074 rv = 0; 4075 } 4076 4077 return (rv); 4078 } 4079 4080 int 4081 sr_raid_tur(struct sr_workunit *wu) 4082 { 4083 struct sr_discipline *sd = wu->swu_dis; 4084 4085 DNPRINTF(SR_D_DIS, "%s: sr_raid_tur\n", DEVNAME(sd->sd_sc)); 4086 4087 if (sd->sd_vol_status == BIOC_SVOFFLINE) { 4088 sd->sd_scsi_sense.error_code = SSD_ERRCODE_CURRENT; 4089 sd->sd_scsi_sense.flags = SKEY_NOT_READY; 4090 sd->sd_scsi_sense.add_sense_code = 0x04; 4091 sd->sd_scsi_sense.add_sense_code_qual = 0x11; 4092 sd->sd_scsi_sense.extra_len = 4; 4093 return (1); 4094 } else if (sd->sd_vol_status == BIOC_SVINVALID) { 4095 sd->sd_scsi_sense.error_code = SSD_ERRCODE_CURRENT; 4096 sd->sd_scsi_sense.flags = SKEY_HARDWARE_ERROR; 4097 sd->sd_scsi_sense.add_sense_code = 0x05; 4098 sd->sd_scsi_sense.add_sense_code_qual = 0x00; 4099 sd->sd_scsi_sense.extra_len = 4; 4100 return (1); 4101 } 4102 4103 return (0); 4104 } 4105 4106 int 4107 sr_raid_request_sense(struct sr_workunit *wu) 4108 { 4109 struct sr_discipline *sd = wu->swu_dis; 4110 struct scsi_xfer *xs = wu->swu_xs; 4111 4112 DNPRINTF(SR_D_DIS, "%s: sr_raid_request_sense\n", 4113 DEVNAME(sd->sd_sc)); 4114 4115 /* use latest sense data */ 4116 memcpy(&xs->sense, &sd->sd_scsi_sense, sizeof(xs->sense)); 4117 4118 /* clear sense data */ 4119 bzero(&sd->sd_scsi_sense, sizeof(sd->sd_scsi_sense)); 4120 4121 return (0); 4122 } 4123 4124 int 4125 sr_raid_start_stop(struct sr_workunit *wu) 4126 { 4127 struct scsi_xfer *xs = wu->swu_xs; 4128 struct scsi_start_stop *ss = (struct scsi_start_stop *)&xs->cmd; 4129 4130 DNPRINTF(SR_D_DIS, "%s: sr_raid_start_stop\n", 4131 DEVNAME(wu->swu_dis->sd_sc)); 4132 4133 if (!ss) 4134 return (1); 4135 4136 /* 4137 * do nothing! 4138 * a softraid discipline should always reflect correct status 4139 */ 4140 return (0); 4141 } 4142 4143 int 4144 sr_raid_sync(struct sr_workunit *wu) 4145 { 4146 struct sr_discipline *sd = wu->swu_dis; 4147 int s, ret, rv = 0, ios; 4148 4149 DNPRINTF(SR_D_DIS, "%s: sr_raid_sync\n", DEVNAME(sd->sd_sc)); 4150 4151 /* when doing a fake sync don't count the wu */ 4152 ios = (wu->swu_flags & SR_WUF_FAKE) ? 0 : 1; 4153 4154 s = splbio(); 4155 sd->sd_sync = 1; 4156 while (sd->sd_wu_pending > ios) { 4157 ret = tsleep_nsec(sd, PRIBIO, "sr_sync", SEC_TO_NSEC(15)); 4158 if (ret == EWOULDBLOCK) { 4159 DNPRINTF(SR_D_DIS, "%s: sr_raid_sync timeout\n", 4160 DEVNAME(sd->sd_sc)); 4161 rv = 1; 4162 break; 4163 } 4164 } 4165 sd->sd_sync = 0; 4166 splx(s); 4167 4168 wakeup(&sd->sd_sync); 4169 4170 return (rv); 4171 } 4172 4173 void 4174 sr_raid_intr(struct buf *bp) 4175 { 4176 struct sr_ccb *ccb = (struct sr_ccb *)bp; 4177 struct sr_workunit *wu = ccb->ccb_wu; 4178 #ifdef SR_DEBUG 4179 struct sr_discipline *sd = wu->swu_dis; 4180 struct scsi_xfer *xs = wu->swu_xs; 4181 #endif 4182 int s; 4183 4184 DNPRINTF(SR_D_INTR, "%s: %s %s intr bp %p xs %p\n", 4185 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, sd->sd_name, bp, xs); 4186 4187 s = splbio(); 4188 sr_ccb_done(ccb); 4189 sr_wu_done(wu); 4190 splx(s); 4191 } 4192 4193 void 4194 sr_schedule_wu(struct sr_workunit *wu) 4195 { 4196 struct sr_discipline *sd = wu->swu_dis; 4197 struct sr_workunit *wup; 4198 int s; 4199 4200 DNPRINTF(SR_D_WU, "sr_schedule_wu: schedule wu %p state %i " 4201 "flags 0x%x\n", wu, wu->swu_state, wu->swu_flags); 4202 4203 KASSERT(wu->swu_io_count > 0); 4204 4205 s = splbio(); 4206 4207 /* Construct the work unit, do not schedule it. */ 4208 if (wu->swu_state == SR_WU_CONSTRUCT) 4209 goto queued; 4210 4211 /* Deferred work unit being reconstructed, do not start. */ 4212 if (wu->swu_state == SR_WU_REQUEUE) 4213 goto queued; 4214 4215 /* Current work unit failed, restart. */ 4216 if (wu->swu_state == SR_WU_RESTART) 4217 goto start; 4218 4219 if (wu->swu_state != SR_WU_INPROGRESS) 4220 panic("sr_schedule_wu: work unit not in progress (state %i)", 4221 wu->swu_state); 4222 4223 /* Walk queue backwards and fill in collider if we have one. */ 4224 TAILQ_FOREACH_REVERSE(wup, &sd->sd_wu_pendq, sr_wu_list, swu_link) { 4225 if (wu->swu_blk_end < wup->swu_blk_start || 4226 wup->swu_blk_end < wu->swu_blk_start) 4227 continue; 4228 4229 /* Defer work unit due to LBA collision. */ 4230 DNPRINTF(SR_D_WU, "sr_schedule_wu: deferring work unit %p\n", 4231 wu); 4232 wu->swu_state = SR_WU_DEFERRED; 4233 while (wup->swu_collider) 4234 wup = wup->swu_collider; 4235 wup->swu_collider = wu; 4236 TAILQ_INSERT_TAIL(&sd->sd_wu_defq, wu, swu_link); 4237 sd->sd_wu_collisions++; 4238 goto queued; 4239 } 4240 4241 start: 4242 sr_raid_startwu(wu); 4243 4244 queued: 4245 splx(s); 4246 } 4247 4248 void 4249 sr_raid_startwu(struct sr_workunit *wu) 4250 { 4251 struct sr_discipline *sd = wu->swu_dis; 4252 struct sr_ccb *ccb; 4253 4254 DNPRINTF(SR_D_WU, "sr_raid_startwu: start wu %p\n", wu); 4255 4256 splassert(IPL_BIO); 4257 4258 if (wu->swu_state == SR_WU_DEFERRED) { 4259 TAILQ_REMOVE(&sd->sd_wu_defq, wu, swu_link); 4260 wu->swu_state = SR_WU_INPROGRESS; 4261 } 4262 4263 if (wu->swu_state != SR_WU_RESTART) 4264 TAILQ_INSERT_TAIL(&sd->sd_wu_pendq, wu, swu_link); 4265 4266 /* Start all of the individual I/Os. */ 4267 if (wu->swu_cb_active == 1) 4268 panic("%s: sr_startwu_callback", DEVNAME(sd->sd_sc)); 4269 wu->swu_cb_active = 1; 4270 4271 TAILQ_FOREACH(ccb, &wu->swu_ccb, ccb_link) 4272 VOP_STRATEGY(ccb->ccb_buf.b_vp, &ccb->ccb_buf); 4273 4274 wu->swu_cb_active = 0; 4275 } 4276 4277 void 4278 sr_raid_recreate_wu(struct sr_workunit *wu) 4279 { 4280 struct sr_discipline *sd = wu->swu_dis; 4281 struct sr_workunit *wup = wu; 4282 4283 /* 4284 * Recreate a work unit by releasing the associated CCBs and reissuing 4285 * the SCSI I/O request. This process is then repeated for all of the 4286 * colliding work units. 4287 */ 4288 do { 4289 sr_wu_release_ccbs(wup); 4290 4291 wup->swu_state = SR_WU_REQUEUE; 4292 if (sd->sd_scsi_rw(wup)) 4293 panic("could not requeue I/O"); 4294 4295 wup = wup->swu_collider; 4296 } while (wup); 4297 } 4298 4299 int 4300 sr_alloc_resources(struct sr_discipline *sd) 4301 { 4302 if (sr_wu_alloc(sd)) { 4303 sr_error(sd->sd_sc, "unable to allocate work units"); 4304 return (ENOMEM); 4305 } 4306 if (sr_ccb_alloc(sd)) { 4307 sr_error(sd->sd_sc, "unable to allocate ccbs"); 4308 return (ENOMEM); 4309 } 4310 4311 return (0); 4312 } 4313 4314 void 4315 sr_free_resources(struct sr_discipline *sd) 4316 { 4317 sr_wu_free(sd); 4318 sr_ccb_free(sd); 4319 } 4320 4321 void 4322 sr_set_chunk_state(struct sr_discipline *sd, int c, int new_state) 4323 { 4324 int old_state, s; 4325 4326 DNPRINTF(SR_D_STATE, "%s: %s: %s: sr_set_chunk_state %d -> %d\n", 4327 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, 4328 sd->sd_vol.sv_chunks[c]->src_meta.scmi.scm_devname, c, new_state); 4329 4330 /* ok to go to splbio since this only happens in error path */ 4331 s = splbio(); 4332 old_state = sd->sd_vol.sv_chunks[c]->src_meta.scm_status; 4333 4334 /* multiple IOs to the same chunk that fail will come through here */ 4335 if (old_state == new_state) 4336 goto done; 4337 4338 switch (old_state) { 4339 case BIOC_SDONLINE: 4340 if (new_state == BIOC_SDOFFLINE) 4341 break; 4342 else 4343 goto die; 4344 break; 4345 4346 case BIOC_SDOFFLINE: 4347 goto die; 4348 4349 default: 4350 die: 4351 splx(s); /* XXX */ 4352 panic("%s: %s: %s: invalid chunk state transition %d -> %d", 4353 DEVNAME(sd->sd_sc), 4354 sd->sd_meta->ssd_devname, 4355 sd->sd_vol.sv_chunks[c]->src_meta.scmi.scm_devname, 4356 old_state, new_state); 4357 /* NOTREACHED */ 4358 } 4359 4360 sd->sd_vol.sv_chunks[c]->src_meta.scm_status = new_state; 4361 sd->sd_set_vol_state(sd); 4362 4363 sd->sd_must_flush = 1; 4364 task_add(systq, &sd->sd_meta_save_task); 4365 done: 4366 splx(s); 4367 } 4368 4369 void 4370 sr_set_vol_state(struct sr_discipline *sd) 4371 { 4372 int states[SR_MAX_STATES]; 4373 int new_state, i, nd; 4374 int old_state = sd->sd_vol_status; 4375 u_int32_t s; 4376 4377 DNPRINTF(SR_D_STATE, "%s: %s: sr_set_vol_state\n", 4378 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname); 4379 4380 nd = sd->sd_meta->ssdi.ssd_chunk_no; 4381 4382 for (i = 0; i < SR_MAX_STATES; i++) 4383 states[i] = 0; 4384 4385 for (i = 0; i < nd; i++) { 4386 s = sd->sd_vol.sv_chunks[i]->src_meta.scm_status; 4387 if (s >= SR_MAX_STATES) 4388 panic("%s: %s: %s: invalid chunk state", 4389 DEVNAME(sd->sd_sc), 4390 sd->sd_meta->ssd_devname, 4391 sd->sd_vol.sv_chunks[i]->src_meta.scmi.scm_devname); 4392 states[s]++; 4393 } 4394 4395 if (states[BIOC_SDONLINE] == nd) 4396 new_state = BIOC_SVONLINE; 4397 else 4398 new_state = BIOC_SVOFFLINE; 4399 4400 DNPRINTF(SR_D_STATE, "%s: %s: sr_set_vol_state %d -> %d\n", 4401 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, 4402 old_state, new_state); 4403 4404 switch (old_state) { 4405 case BIOC_SVONLINE: 4406 if (new_state == BIOC_SVOFFLINE || new_state == BIOC_SVONLINE) 4407 break; 4408 else 4409 goto die; 4410 break; 4411 4412 case BIOC_SVOFFLINE: 4413 /* XXX this might be a little too much */ 4414 goto die; 4415 4416 default: 4417 die: 4418 panic("%s: %s: invalid volume state transition %d -> %d", 4419 DEVNAME(sd->sd_sc), 4420 sd->sd_meta->ssd_devname, 4421 old_state, new_state); 4422 /* NOTREACHED */ 4423 } 4424 4425 sd->sd_vol_status = new_state; 4426 } 4427 4428 void * 4429 sr_block_get(struct sr_discipline *sd, long length) 4430 { 4431 return dma_alloc(length, PR_NOWAIT | PR_ZERO); 4432 } 4433 4434 void 4435 sr_block_put(struct sr_discipline *sd, void *ptr, int length) 4436 { 4437 dma_free(ptr, length); 4438 } 4439 4440 void 4441 sr_checksum_print(u_int8_t *md5) 4442 { 4443 int i; 4444 4445 for (i = 0; i < MD5_DIGEST_LENGTH; i++) 4446 printf("%02x", md5[i]); 4447 } 4448 4449 void 4450 sr_checksum(struct sr_softc *sc, void *src, void *md5, u_int32_t len) 4451 { 4452 MD5_CTX ctx; 4453 4454 DNPRINTF(SR_D_MISC, "%s: sr_checksum(%p %p %d)\n", DEVNAME(sc), src, 4455 md5, len); 4456 4457 MD5Init(&ctx); 4458 MD5Update(&ctx, src, len); 4459 MD5Final(md5, &ctx); 4460 } 4461 4462 void 4463 sr_uuid_generate(struct sr_uuid *uuid) 4464 { 4465 arc4random_buf(uuid->sui_id, sizeof(uuid->sui_id)); 4466 /* UUID version 4: random */ 4467 uuid->sui_id[6] &= 0x0f; 4468 uuid->sui_id[6] |= 0x40; 4469 /* RFC4122 variant */ 4470 uuid->sui_id[8] &= 0x3f; 4471 uuid->sui_id[8] |= 0x80; 4472 } 4473 4474 char * 4475 sr_uuid_format(struct sr_uuid *uuid) 4476 { 4477 char *uuidstr; 4478 4479 uuidstr = malloc(37, M_DEVBUF, M_WAITOK); 4480 4481 snprintf(uuidstr, 37, 4482 "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-" 4483 "%02x%02x%02x%02x%02x%02x", 4484 uuid->sui_id[0], uuid->sui_id[1], 4485 uuid->sui_id[2], uuid->sui_id[3], 4486 uuid->sui_id[4], uuid->sui_id[5], 4487 uuid->sui_id[6], uuid->sui_id[7], 4488 uuid->sui_id[8], uuid->sui_id[9], 4489 uuid->sui_id[10], uuid->sui_id[11], 4490 uuid->sui_id[12], uuid->sui_id[13], 4491 uuid->sui_id[14], uuid->sui_id[15]); 4492 4493 return uuidstr; 4494 } 4495 4496 void 4497 sr_uuid_print(struct sr_uuid *uuid, int cr) 4498 { 4499 char *uuidstr; 4500 4501 uuidstr = sr_uuid_format(uuid); 4502 printf("%s%s", uuidstr, (cr ? "\n" : "")); 4503 free(uuidstr, M_DEVBUF, 37); 4504 } 4505 4506 int 4507 sr_already_assembled(struct sr_discipline *sd) 4508 { 4509 struct sr_softc *sc = sd->sd_sc; 4510 struct sr_discipline *sdtmp; 4511 4512 TAILQ_FOREACH(sdtmp, &sc->sc_dis_list, sd_link) { 4513 if (!bcmp(&sd->sd_meta->ssdi.ssd_uuid, 4514 &sdtmp->sd_meta->ssdi.ssd_uuid, 4515 sizeof(sd->sd_meta->ssdi.ssd_uuid))) 4516 return (1); 4517 } 4518 4519 return (0); 4520 } 4521 4522 int32_t 4523 sr_validate_stripsize(u_int32_t b) 4524 { 4525 int s = 0; 4526 4527 if (b % DEV_BSIZE) 4528 return (-1); 4529 4530 while ((b & 1) == 0) { 4531 b >>= 1; 4532 s++; 4533 } 4534 4535 /* only multiple of twos */ 4536 b >>= 1; 4537 if (b) 4538 return(-1); 4539 4540 return (s); 4541 } 4542 4543 void 4544 sr_quiesce(void) 4545 { 4546 struct sr_softc *sc = softraid0; 4547 struct sr_discipline *sd, *nsd; 4548 4549 if (sc == NULL) 4550 return; 4551 4552 /* Shutdown disciplines in reverse attach order. */ 4553 TAILQ_FOREACH_REVERSE_SAFE(sd, &sc->sc_dis_list, 4554 sr_discipline_list, sd_link, nsd) 4555 sr_discipline_shutdown(sd, 1, -1); 4556 } 4557 4558 void 4559 sr_shutdown(int dying) 4560 { 4561 struct sr_softc *sc = softraid0; 4562 struct sr_discipline *sd; 4563 4564 if (sc == NULL) 4565 return; 4566 4567 DNPRINTF(SR_D_MISC, "%s: sr_shutdown\n", DEVNAME(sc)); 4568 4569 /* 4570 * Since softraid is not under mainbus, we have to explicitly 4571 * notify its children that the power is going down, so they 4572 * can execute their shutdown hooks. 4573 */ 4574 config_suspend((struct device *)sc, DVACT_POWERDOWN); 4575 4576 /* Shutdown disciplines in reverse attach order. */ 4577 while ((sd = TAILQ_LAST(&sc->sc_dis_list, sr_discipline_list)) != NULL) 4578 sr_discipline_shutdown(sd, 1, dying); 4579 } 4580 4581 int 4582 sr_validate_io(struct sr_workunit *wu, daddr_t *blkno, char *func) 4583 { 4584 struct sr_discipline *sd = wu->swu_dis; 4585 struct scsi_xfer *xs = wu->swu_xs; 4586 int rv = 1; 4587 4588 DNPRINTF(SR_D_DIS, "%s: %s 0x%02x\n", DEVNAME(sd->sd_sc), func, 4589 xs->cmd.opcode); 4590 4591 if (sd->sd_meta->ssd_data_blkno == 0) 4592 panic("invalid data blkno"); 4593 4594 if (sd->sd_vol_status == BIOC_SVOFFLINE) { 4595 DNPRINTF(SR_D_DIS, "%s: %s device offline\n", 4596 DEVNAME(sd->sd_sc), func); 4597 goto bad; 4598 } 4599 4600 if (xs->datalen == 0) { 4601 printf("%s: %s: illegal block count for %s\n", 4602 DEVNAME(sd->sd_sc), func, sd->sd_meta->ssd_devname); 4603 goto bad; 4604 } 4605 4606 if (xs->cmdlen == 10) 4607 *blkno = _4btol(((struct scsi_rw_10 *)&xs->cmd)->addr); 4608 else if (xs->cmdlen == 16) 4609 *blkno = _8btol(((struct scsi_rw_16 *)&xs->cmd)->addr); 4610 else if (xs->cmdlen == 6) 4611 *blkno = _3btol(((struct scsi_rw *)&xs->cmd)->addr); 4612 else { 4613 printf("%s: %s: illegal cmdlen for %s\n", 4614 DEVNAME(sd->sd_sc), func, sd->sd_meta->ssd_devname); 4615 goto bad; 4616 } 4617 4618 *blkno *= (sd->sd_meta->ssdi.ssd_secsize / DEV_BSIZE); 4619 4620 wu->swu_blk_start = *blkno; 4621 wu->swu_blk_end = *blkno + (xs->datalen >> DEV_BSHIFT) - 1; 4622 4623 if (wu->swu_blk_end > sd->sd_meta->ssdi.ssd_size) { 4624 DNPRINTF(SR_D_DIS, "%s: %s out of bounds start: %lld " 4625 "end: %lld length: %d\n", 4626 DEVNAME(sd->sd_sc), func, (long long)wu->swu_blk_start, 4627 (long long)wu->swu_blk_end, xs->datalen); 4628 4629 sd->sd_scsi_sense.error_code = SSD_ERRCODE_CURRENT | 4630 SSD_ERRCODE_VALID; 4631 sd->sd_scsi_sense.flags = SKEY_ILLEGAL_REQUEST; 4632 sd->sd_scsi_sense.add_sense_code = 0x21; 4633 sd->sd_scsi_sense.add_sense_code_qual = 0x00; 4634 sd->sd_scsi_sense.extra_len = 4; 4635 goto bad; 4636 } 4637 4638 rv = 0; 4639 bad: 4640 return (rv); 4641 } 4642 4643 void 4644 sr_rebuild_start(void *arg) 4645 { 4646 struct sr_discipline *sd = arg; 4647 struct sr_softc *sc = sd->sd_sc; 4648 4649 DNPRINTF(SR_D_REBUILD, "%s: %s starting rebuild thread\n", 4650 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname); 4651 4652 if (kthread_create(sr_rebuild_thread, sd, &sd->sd_background_proc, 4653 DEVNAME(sc)) != 0) 4654 printf("%s: unable to start background operation\n", 4655 DEVNAME(sc)); 4656 } 4657 4658 void 4659 sr_rebuild_thread(void *arg) 4660 { 4661 struct sr_discipline *sd = arg; 4662 4663 DNPRINTF(SR_D_REBUILD, "%s: %s rebuild thread started\n", 4664 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname); 4665 4666 sd->sd_reb_active = 1; 4667 sd->sd_rebuild(sd); 4668 sd->sd_reb_active = 0; 4669 4670 kthread_exit(0); 4671 } 4672 4673 void 4674 sr_rebuild(struct sr_discipline *sd) 4675 { 4676 struct sr_softc *sc = sd->sd_sc; 4677 u_int64_t sz, whole_blk, partial_blk, blk, restart; 4678 daddr_t lba; 4679 struct sr_workunit *wu_r, *wu_w; 4680 struct scsi_xfer xs_r, xs_w; 4681 struct scsi_rw_16 *cr, *cw; 4682 int c, s, slept, percent = 0, old_percent = -1; 4683 u_int8_t *buf; 4684 4685 whole_blk = sd->sd_meta->ssdi.ssd_size / SR_REBUILD_IO_SIZE; 4686 partial_blk = sd->sd_meta->ssdi.ssd_size % SR_REBUILD_IO_SIZE; 4687 4688 restart = sd->sd_meta->ssd_rebuild / SR_REBUILD_IO_SIZE; 4689 if (restart > whole_blk) { 4690 printf("%s: bogus rebuild restart offset, starting from 0\n", 4691 DEVNAME(sc)); 4692 restart = 0; 4693 } 4694 if (restart) { 4695 /* 4696 * XXX there is a hole here; there is a possibility that we 4697 * had a restart however the chunk that was supposed to 4698 * be rebuilt is no longer valid; we can reach this situation 4699 * when a rebuild is in progress and the box crashes and 4700 * on reboot the rebuild chunk is different (like zero'd or 4701 * replaced). We need to check the uuid of the chunk that is 4702 * being rebuilt to assert this. 4703 */ 4704 percent = sr_rebuild_percent(sd); 4705 printf("%s: resuming rebuild on %s at %d%%\n", 4706 DEVNAME(sc), sd->sd_meta->ssd_devname, percent); 4707 } 4708 4709 /* currently this is 64k therefore we can use dma_alloc */ 4710 buf = dma_alloc(SR_REBUILD_IO_SIZE << DEV_BSHIFT, PR_WAITOK); 4711 for (blk = restart; blk <= whole_blk; blk++) { 4712 lba = blk * SR_REBUILD_IO_SIZE; 4713 sz = SR_REBUILD_IO_SIZE; 4714 if (blk == whole_blk) { 4715 if (partial_blk == 0) 4716 break; 4717 sz = partial_blk; 4718 } 4719 4720 /* get some wu */ 4721 wu_r = sr_scsi_wu_get(sd, 0); 4722 wu_w = sr_scsi_wu_get(sd, 0); 4723 4724 DNPRINTF(SR_D_REBUILD, "%s: %s rebuild wu_r %p, wu_w %p\n", 4725 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, wu_r, wu_w); 4726 4727 /* setup read io */ 4728 bzero(&xs_r, sizeof xs_r); 4729 xs_r.error = XS_NOERROR; 4730 xs_r.flags = SCSI_DATA_IN; 4731 xs_r.datalen = sz << DEV_BSHIFT; 4732 xs_r.data = buf; 4733 xs_r.cmdlen = sizeof(*cr); 4734 cr = (struct scsi_rw_16 *)&xs_r.cmd; 4735 cr->opcode = READ_16; 4736 _lto4b(sz, cr->length); 4737 _lto8b(lba, cr->addr); 4738 wu_r->swu_state = SR_WU_CONSTRUCT; 4739 wu_r->swu_flags |= SR_WUF_REBUILD; 4740 wu_r->swu_xs = &xs_r; 4741 if (sd->sd_scsi_rw(wu_r)) { 4742 printf("%s: could not create read io\n", 4743 DEVNAME(sc)); 4744 goto fail; 4745 } 4746 4747 /* setup write io */ 4748 bzero(&xs_w, sizeof xs_w); 4749 xs_w.error = XS_NOERROR; 4750 xs_w.flags = SCSI_DATA_OUT; 4751 xs_w.datalen = sz << DEV_BSHIFT; 4752 xs_w.data = buf; 4753 xs_w.cmdlen = sizeof(*cw); 4754 cw = (struct scsi_rw_16 *)&xs_w.cmd; 4755 cw->opcode = WRITE_16; 4756 _lto4b(sz, cw->length); 4757 _lto8b(lba, cw->addr); 4758 wu_w->swu_state = SR_WU_CONSTRUCT; 4759 wu_w->swu_flags |= SR_WUF_REBUILD | SR_WUF_WAKEUP; 4760 wu_w->swu_xs = &xs_w; 4761 if (sd->sd_scsi_rw(wu_w)) { 4762 printf("%s: could not create write io\n", 4763 DEVNAME(sc)); 4764 goto fail; 4765 } 4766 4767 /* 4768 * collide with the read io so that we get automatically 4769 * started when the read is done 4770 */ 4771 wu_w->swu_state = SR_WU_DEFERRED; 4772 wu_r->swu_collider = wu_w; 4773 s = splbio(); 4774 TAILQ_INSERT_TAIL(&sd->sd_wu_defq, wu_w, swu_link); 4775 splx(s); 4776 4777 DNPRINTF(SR_D_REBUILD, "%s: %s rebuild scheduling wu_r %p\n", 4778 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, wu_r); 4779 4780 wu_r->swu_state = SR_WU_INPROGRESS; 4781 sr_schedule_wu(wu_r); 4782 4783 /* wait for write completion */ 4784 slept = 0; 4785 while ((wu_w->swu_flags & SR_WUF_REBUILDIOCOMP) == 0) { 4786 tsleep_nsec(wu_w, PRIBIO, "sr_rebuild", INFSLP); 4787 slept = 1; 4788 } 4789 /* yield if we didn't sleep */ 4790 if (slept == 0) 4791 tsleep_nsec(sc, PWAIT, "sr_yield", MSEC_TO_NSEC(1)); 4792 4793 sr_scsi_wu_put(sd, wu_r); 4794 sr_scsi_wu_put(sd, wu_w); 4795 4796 sd->sd_meta->ssd_rebuild = lba; 4797 4798 /* XXX - this should be based on size, not percentage. */ 4799 /* save metadata every percent */ 4800 percent = sr_rebuild_percent(sd); 4801 if (percent != old_percent && blk != whole_blk) { 4802 if (sr_meta_save(sd, SR_META_DIRTY)) 4803 printf("%s: could not save metadata to %s\n", 4804 DEVNAME(sc), sd->sd_meta->ssd_devname); 4805 old_percent = percent; 4806 } 4807 4808 if (sd->sd_reb_abort) 4809 goto abort; 4810 } 4811 4812 /* all done */ 4813 sd->sd_meta->ssd_rebuild = 0; 4814 for (c = 0; c < sd->sd_meta->ssdi.ssd_chunk_no; c++) { 4815 if (sd->sd_vol.sv_chunks[c]->src_meta.scm_status == 4816 BIOC_SDREBUILD) { 4817 sd->sd_set_chunk_state(sd, c, BIOC_SDONLINE); 4818 break; 4819 } 4820 } 4821 4822 abort: 4823 if (sr_meta_save(sd, SR_META_DIRTY)) 4824 printf("%s: could not save metadata to %s\n", 4825 DEVNAME(sc), sd->sd_meta->ssd_devname); 4826 fail: 4827 dma_free(buf, SR_REBUILD_IO_SIZE << DEV_BSHIFT); 4828 } 4829 4830 #ifndef SMALL_KERNEL 4831 int 4832 sr_sensors_create(struct sr_discipline *sd) 4833 { 4834 struct sr_softc *sc = sd->sd_sc; 4835 int rv = 1; 4836 4837 DNPRINTF(SR_D_STATE, "%s: %s: sr_sensors_create\n", 4838 DEVNAME(sc), sd->sd_meta->ssd_devname); 4839 4840 sd->sd_vol.sv_sensor.type = SENSOR_DRIVE; 4841 sd->sd_vol.sv_sensor.status = SENSOR_S_UNKNOWN; 4842 strlcpy(sd->sd_vol.sv_sensor.desc, sd->sd_meta->ssd_devname, 4843 sizeof(sd->sd_vol.sv_sensor.desc)); 4844 4845 sensor_attach(&sc->sc_sensordev, &sd->sd_vol.sv_sensor); 4846 sd->sd_vol.sv_sensor_attached = 1; 4847 4848 if (sc->sc_sensor_task == NULL) { 4849 sc->sc_sensor_task = sensor_task_register(sc, 4850 sr_sensors_refresh, 10); 4851 if (sc->sc_sensor_task == NULL) 4852 goto bad; 4853 } 4854 4855 rv = 0; 4856 bad: 4857 return (rv); 4858 } 4859 4860 void 4861 sr_sensors_delete(struct sr_discipline *sd) 4862 { 4863 DNPRINTF(SR_D_STATE, "%s: sr_sensors_delete\n", DEVNAME(sd->sd_sc)); 4864 4865 if (sd->sd_vol.sv_sensor_attached) 4866 sensor_detach(&sd->sd_sc->sc_sensordev, &sd->sd_vol.sv_sensor); 4867 } 4868 4869 void 4870 sr_sensors_refresh(void *arg) 4871 { 4872 struct sr_softc *sc = arg; 4873 struct sr_volume *sv; 4874 struct sr_discipline *sd; 4875 4876 DNPRINTF(SR_D_STATE, "%s: sr_sensors_refresh\n", DEVNAME(sc)); 4877 4878 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 4879 sv = &sd->sd_vol; 4880 4881 switch(sd->sd_vol_status) { 4882 case BIOC_SVOFFLINE: 4883 sv->sv_sensor.value = SENSOR_DRIVE_FAIL; 4884 sv->sv_sensor.status = SENSOR_S_CRIT; 4885 break; 4886 4887 case BIOC_SVDEGRADED: 4888 sv->sv_sensor.value = SENSOR_DRIVE_PFAIL; 4889 sv->sv_sensor.status = SENSOR_S_WARN; 4890 break; 4891 4892 case BIOC_SVREBUILD: 4893 sv->sv_sensor.value = SENSOR_DRIVE_REBUILD; 4894 sv->sv_sensor.status = SENSOR_S_WARN; 4895 break; 4896 4897 case BIOC_SVSCRUB: 4898 case BIOC_SVONLINE: 4899 sv->sv_sensor.value = SENSOR_DRIVE_ONLINE; 4900 sv->sv_sensor.status = SENSOR_S_OK; 4901 break; 4902 4903 default: 4904 sv->sv_sensor.value = 0; /* unknown */ 4905 sv->sv_sensor.status = SENSOR_S_UNKNOWN; 4906 } 4907 } 4908 } 4909 #endif /* SMALL_KERNEL */ 4910 4911 #ifdef SR_FANCY_STATS 4912 void sr_print_stats(void); 4913 4914 void 4915 sr_print_stats(void) 4916 { 4917 struct sr_softc *sc = softraid0; 4918 struct sr_discipline *sd; 4919 4920 if (sc == NULL) { 4921 printf("no softraid softc found\n"); 4922 return; 4923 } 4924 4925 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 4926 printf("%s: ios pending %d, collisions %llu\n", 4927 sd->sd_meta->ssd_devname, 4928 sd->sd_wu_pending, 4929 sd->sd_wu_collisions); 4930 } 4931 } 4932 #endif /* SR_FANCY_STATS */ 4933 4934 #ifdef SR_DEBUG 4935 void 4936 sr_meta_print(struct sr_metadata *m) 4937 { 4938 int i; 4939 struct sr_meta_chunk *mc; 4940 struct sr_meta_opt_hdr *omh; 4941 4942 if (!(sr_debug & SR_D_META)) 4943 return; 4944 4945 printf("\tssd_magic 0x%llx\n", m->ssdi.ssd_magic); 4946 printf("\tssd_version %d\n", m->ssdi.ssd_version); 4947 printf("\tssd_vol_flags 0x%x\n", m->ssdi.ssd_vol_flags); 4948 printf("\tssd_uuid "); 4949 sr_uuid_print(&m->ssdi.ssd_uuid, 1); 4950 printf("\tssd_chunk_no %d\n", m->ssdi.ssd_chunk_no); 4951 printf("\tssd_chunk_id %d\n", m->ssdi.ssd_chunk_id); 4952 printf("\tssd_opt_no %d\n", m->ssdi.ssd_opt_no); 4953 printf("\tssd_volid %d\n", m->ssdi.ssd_volid); 4954 printf("\tssd_level %d\n", m->ssdi.ssd_level); 4955 printf("\tssd_size %lld\n", m->ssdi.ssd_size); 4956 printf("\tssd_devname %s\n", m->ssd_devname); 4957 printf("\tssd_vendor %s\n", m->ssdi.ssd_vendor); 4958 printf("\tssd_product %s\n", m->ssdi.ssd_product); 4959 printf("\tssd_revision %s\n", m->ssdi.ssd_revision); 4960 printf("\tssd_strip_size %d\n", m->ssdi.ssd_strip_size); 4961 printf("\tssd_checksum "); 4962 sr_checksum_print(m->ssd_checksum); 4963 printf("\n"); 4964 printf("\tssd_meta_flags 0x%x\n", m->ssd_meta_flags); 4965 printf("\tssd_ondisk %llu\n", m->ssd_ondisk); 4966 4967 mc = (struct sr_meta_chunk *)(m + 1); 4968 for (i = 0; i < m->ssdi.ssd_chunk_no; i++, mc++) { 4969 printf("\t\tscm_volid %d\n", mc->scmi.scm_volid); 4970 printf("\t\tscm_chunk_id %d\n", mc->scmi.scm_chunk_id); 4971 printf("\t\tscm_devname %s\n", mc->scmi.scm_devname); 4972 printf("\t\tscm_size %lld\n", mc->scmi.scm_size); 4973 printf("\t\tscm_coerced_size %lld\n",mc->scmi.scm_coerced_size); 4974 printf("\t\tscm_uuid "); 4975 sr_uuid_print(&mc->scmi.scm_uuid, 1); 4976 printf("\t\tscm_checksum "); 4977 sr_checksum_print(mc->scm_checksum); 4978 printf("\n"); 4979 printf("\t\tscm_status %d\n", mc->scm_status); 4980 } 4981 4982 omh = (struct sr_meta_opt_hdr *)((u_int8_t *)(m + 1) + 4983 sizeof(struct sr_meta_chunk) * m->ssdi.ssd_chunk_no); 4984 for (i = 0; i < m->ssdi.ssd_opt_no; i++) { 4985 printf("\t\t\tsom_type %d\n", omh->som_type); 4986 printf("\t\t\tsom_checksum "); 4987 sr_checksum_print(omh->som_checksum); 4988 printf("\n"); 4989 omh = (struct sr_meta_opt_hdr *)((void *)omh + 4990 omh->som_length); 4991 } 4992 } 4993 4994 void 4995 sr_dump_block(void *blk, int len) 4996 { 4997 uint8_t *b = blk; 4998 int i, j, c; 4999 5000 for (i = 0; i < len; i += 16) { 5001 for (j = 0; j < 16; j++) 5002 printf("%.2x ", b[i + j]); 5003 printf(" "); 5004 for (j = 0; j < 16; j++) { 5005 c = b[i + j]; 5006 if (c < ' ' || c > 'z' || i + j > len) 5007 c = '.'; 5008 printf("%c", c); 5009 } 5010 printf("\n"); 5011 } 5012 } 5013 5014 void 5015 sr_dump_mem(u_int8_t *p, int len) 5016 { 5017 int i; 5018 5019 for (i = 0; i < len; i++) 5020 printf("%02x ", *p++); 5021 printf("\n"); 5022 } 5023 5024 #endif /* SR_DEBUG */ 5025 5026 #ifdef HIBERNATE 5027 /* 5028 * Side-effect free (no malloc, printf, pool, splx) softraid crypto writer. 5029 * 5030 * This function must perform the following: 5031 * 1. Determine the underlying device's own side-effect free I/O function 5032 * (eg, ahci_hibernate_io, wd_hibernate_io, etc). 5033 * 2. Store enough information in the provided page argument for subsequent 5034 * I/O calls (such as the crypto discipline structure for the keys, the 5035 * offset of the softraid partition on the underlying disk, as well as 5036 * the offset of the swap partition within the crypto volume. 5037 * 3. Encrypt the incoming data using the sr_discipline keys, then pass 5038 * the request to the underlying device's own I/O function. 5039 */ 5040 int 5041 sr_hibernate_io(dev_t dev, daddr_t blkno, vaddr_t addr, size_t size, int op, void *page) 5042 { 5043 /* Struct for stashing data obtained on HIB_INIT. 5044 * XXX 5045 * We share the page with the underlying device's own 5046 * side-effect free I/O function, so we pad our data to 5047 * the end of the page. Presently this does not overlap 5048 * with either of the two other side-effect free i/o 5049 * functions (ahci/wd). 5050 */ 5051 struct { 5052 char pad[3072]; 5053 struct sr_discipline *srd; 5054 hibio_fn subfn; /* underlying device i/o fn */ 5055 dev_t subdev; /* underlying device dev_t */ 5056 daddr_t sr_swapoff; /* ofs of swap part in sr volume */ 5057 char buf[DEV_BSIZE]; /* encryption performed into this buf */ 5058 } *my = page; 5059 extern struct cfdriver sd_cd; 5060 char errstr[128], *dl_ret; 5061 struct sr_chunk *schunk; 5062 struct sd_softc *sd; 5063 struct aes_xts_ctx ctx; 5064 struct sr_softc *sc; 5065 struct device *dv; 5066 daddr_t key_blkno; 5067 uint32_t sub_raidoff; /* ofs of sr part in underlying dev */ 5068 struct disklabel dl; 5069 struct partition *pp; 5070 size_t i, j; 5071 u_char iv[8]; 5072 5073 /* 5074 * In HIB_INIT, we are passed the swap partition size and offset 5075 * in 'size' and 'blkno' respectively. These are relative to the 5076 * start of the softraid partition, and we need to save these 5077 * for later translation to the underlying device's layout. 5078 */ 5079 if (op == HIB_INIT) { 5080 dv = disk_lookup(&sd_cd, DISKUNIT(dev)); 5081 sd = (struct sd_softc *)dv; 5082 sc = (struct sr_softc *)dv->dv_parent->dv_parent; 5083 5084 /* 5085 * Look up the sr discipline. This is used to determine 5086 * if we are SR crypto and what the underlying device is. 5087 */ 5088 my->srd = sc->sc_targets[sd->sc_link->target]; 5089 DNPRINTF(SR_D_MISC, "sr_hibernate_io: discipline is %s\n", 5090 my->srd->sd_name); 5091 if (strncmp(my->srd->sd_name, "CRYPTO", 5092 sizeof(my->srd->sd_name))) 5093 return (ENOTSUP); 5094 5095 /* Find the underlying device */ 5096 schunk = my->srd->sd_vol.sv_chunks[0]; 5097 my->subdev = schunk->src_dev_mm; 5098 5099 /* 5100 * Find the appropriate underlying device side effect free 5101 * I/O function, based on the type of device it is. 5102 */ 5103 my->subfn = get_hibernate_io_function(my->subdev); 5104 if (!my->subfn) 5105 return (ENODEV); 5106 5107 /* 5108 * Find blkno where this raid partition starts on 5109 * the underlying disk. 5110 */ 5111 dl_ret = disk_readlabel(&dl, my->subdev, errstr, 5112 sizeof(errstr)); 5113 if (dl_ret) { 5114 printf("Hibernate error reading disklabel: %s\n", dl_ret); 5115 return (ENOTSUP); 5116 } 5117 5118 pp = &dl.d_partitions[DISKPART(my->subdev)]; 5119 if (pp->p_fstype != FS_RAID || DL_GETPSIZE(pp) == 0) 5120 return (ENOTSUP); 5121 5122 /* Find the blkno of the SR part in the underlying device */ 5123 sub_raidoff = my->srd->sd_meta->ssd_data_blkno + 5124 DL_SECTOBLK(&dl, DL_GETPOFFSET(pp)); 5125 DNPRINTF(SR_D_MISC,"sr_hibernate_io: blk trans ofs: %d blks\n", 5126 sub_raidoff); 5127 5128 /* Save the blkno of the swap partition in the SR disk */ 5129 my->sr_swapoff = blkno; 5130 5131 /* Initialize the sub-device */ 5132 return my->subfn(my->subdev, sub_raidoff + blkno, 5133 addr, size, op, page); 5134 } 5135 5136 /* Hibernate only uses (and we only support) writes */ 5137 if (op != HIB_W) 5138 return (ENOTSUP); 5139 5140 /* 5141 * Blocks act as the IV for the encryption. These block numbers 5142 * are relative to the start of the sr partition, but the 'blkno' 5143 * passed above is relative to the start of the swap partition 5144 * inside the sr partition, so bias appropriately. 5145 */ 5146 key_blkno = my->sr_swapoff + blkno; 5147 5148 /* Process each disk block one at a time. */ 5149 for (i = 0; i < size; i += DEV_BSIZE) { 5150 int res; 5151 5152 bzero(&ctx, sizeof(ctx)); 5153 5154 /* 5155 * Set encryption key (from the sr discipline stashed 5156 * during HIB_INIT. This code is based on the softraid 5157 * bootblock code. 5158 */ 5159 aes_xts_setkey(&ctx, my->srd->mds.mdd_crypto.scr_key[0], 64); 5160 /* We encrypt DEV_BSIZE bytes at a time in my->buf */ 5161 memcpy(my->buf, ((char *)addr) + i, DEV_BSIZE); 5162 5163 /* Block number is the IV */ 5164 memcpy(&iv, &key_blkno, sizeof(key_blkno)); 5165 aes_xts_reinit(&ctx, iv); 5166 5167 /* Encrypt DEV_BSIZE bytes, AES_XTS_BLOCKSIZE bytes at a time */ 5168 for (j = 0; j < DEV_BSIZE; j += AES_XTS_BLOCKSIZE) 5169 aes_xts_encrypt(&ctx, my->buf + j); 5170 5171 /* 5172 * Write one block out from my->buf to the underlying device 5173 * using its own side-effect free I/O function. 5174 */ 5175 res = my->subfn(my->subdev, blkno + (i / DEV_BSIZE), 5176 (vaddr_t)(my->buf), DEV_BSIZE, op, page); 5177 if (res != 0) 5178 return (res); 5179 key_blkno++; 5180 } 5181 return (0); 5182 } 5183 #endif /* HIBERNATE */ 5184