1 /* $OpenBSD: softraid.c,v 1.410 2020/07/20 14:41:12 krw Exp $ */ 2 /* 3 * Copyright (c) 2007, 2008, 2009 Marco Peereboom <marco@peereboom.us> 4 * Copyright (c) 2008 Chris Kuethe <ckuethe@openbsd.org> 5 * Copyright (c) 2009 Joel Sing <jsing@openbsd.org> 6 * 7 * Permission to use, copy, modify, and distribute this software for any 8 * purpose with or without fee is hereby granted, provided that the above 9 * copyright notice and this permission notice appear in all copies. 10 * 11 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES 12 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF 13 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR 14 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES 15 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN 16 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF 17 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 18 */ 19 20 #include "bio.h" 21 22 #include <sys/param.h> 23 #include <sys/systm.h> 24 #include <sys/buf.h> 25 #include <sys/device.h> 26 #include <sys/ioctl.h> 27 #include <sys/malloc.h> 28 #include <sys/pool.h> 29 #include <sys/kernel.h> 30 #include <sys/disk.h> 31 #include <sys/rwlock.h> 32 #include <sys/queue.h> 33 #include <sys/fcntl.h> 34 #include <sys/disklabel.h> 35 #include <sys/vnode.h> 36 #include <sys/lock.h> 37 #include <sys/mount.h> 38 #include <sys/sensors.h> 39 #include <sys/stat.h> 40 #include <sys/conf.h> 41 #include <sys/uio.h> 42 #include <sys/task.h> 43 #include <sys/kthread.h> 44 #include <sys/dkio.h> 45 #include <sys/stdint.h> 46 47 #include <scsi/scsi_all.h> 48 #include <scsi/scsiconf.h> 49 #include <scsi/scsi_disk.h> 50 51 #include <dev/softraidvar.h> 52 53 #ifdef HIBERNATE 54 #include <lib/libsa/aes_xts.h> 55 #include <sys/hibernate.h> 56 #include <scsi/sdvar.h> 57 #endif /* HIBERNATE */ 58 59 /* #define SR_FANCY_STATS */ 60 61 #ifdef SR_DEBUG 62 #define SR_FANCY_STATS 63 uint32_t sr_debug = 0 64 /* | SR_D_CMD */ 65 /* | SR_D_MISC */ 66 /* | SR_D_INTR */ 67 /* | SR_D_IOCTL */ 68 /* | SR_D_CCB */ 69 /* | SR_D_WU */ 70 /* | SR_D_META */ 71 /* | SR_D_DIS */ 72 /* | SR_D_STATE */ 73 /* | SR_D_REBUILD */ 74 ; 75 #endif 76 77 struct sr_softc *softraid0; 78 struct sr_uuid sr_bootuuid; 79 u_int8_t sr_bootkey[SR_CRYPTO_MAXKEYBYTES]; 80 81 int sr_match(struct device *, void *, void *); 82 void sr_attach(struct device *, struct device *, void *); 83 int sr_detach(struct device *, int); 84 void sr_map_root(void); 85 86 struct cfattach softraid_ca = { 87 sizeof(struct sr_softc), sr_match, sr_attach, sr_detach, 88 }; 89 90 struct cfdriver softraid_cd = { 91 NULL, "softraid", DV_DULL 92 }; 93 94 /* scsi & discipline */ 95 void sr_scsi_cmd(struct scsi_xfer *); 96 int sr_scsi_probe(struct scsi_link *); 97 void sr_copy_internal_data(struct scsi_xfer *, 98 void *, size_t); 99 int sr_scsi_ioctl(struct scsi_link *, u_long, 100 caddr_t, int); 101 int sr_bio_ioctl(struct device *, u_long, caddr_t); 102 int sr_bio_handler(struct sr_softc *, 103 struct sr_discipline *, u_long, struct bio *); 104 int sr_ioctl_inq(struct sr_softc *, struct bioc_inq *); 105 int sr_ioctl_vol(struct sr_softc *, struct bioc_vol *); 106 int sr_ioctl_disk(struct sr_softc *, struct bioc_disk *); 107 int sr_ioctl_setstate(struct sr_softc *, 108 struct bioc_setstate *); 109 int sr_ioctl_createraid(struct sr_softc *, 110 struct bioc_createraid *, int, void *); 111 int sr_ioctl_deleteraid(struct sr_softc *, 112 struct sr_discipline *, struct bioc_deleteraid *); 113 int sr_ioctl_discipline(struct sr_softc *, 114 struct sr_discipline *, struct bioc_discipline *); 115 int sr_ioctl_installboot(struct sr_softc *, 116 struct sr_discipline *, struct bioc_installboot *); 117 void sr_chunks_unwind(struct sr_softc *, 118 struct sr_chunk_head *); 119 void sr_discipline_free(struct sr_discipline *); 120 void sr_discipline_shutdown(struct sr_discipline *, int, int); 121 int sr_discipline_init(struct sr_discipline *, int); 122 int sr_alloc_resources(struct sr_discipline *); 123 void sr_free_resources(struct sr_discipline *); 124 void sr_set_chunk_state(struct sr_discipline *, int, int); 125 void sr_set_vol_state(struct sr_discipline *); 126 127 /* utility functions */ 128 void sr_shutdown(int); 129 void sr_uuid_generate(struct sr_uuid *); 130 char *sr_uuid_format(struct sr_uuid *); 131 void sr_uuid_print(struct sr_uuid *, int); 132 void sr_checksum_print(u_int8_t *); 133 int sr_boot_assembly(struct sr_softc *); 134 int sr_already_assembled(struct sr_discipline *); 135 int sr_hotspare(struct sr_softc *, dev_t); 136 void sr_hotspare_rebuild(struct sr_discipline *); 137 int sr_rebuild_init(struct sr_discipline *, dev_t, int); 138 void sr_rebuild_start(void *); 139 void sr_rebuild_thread(void *); 140 void sr_rebuild(struct sr_discipline *); 141 void sr_roam_chunks(struct sr_discipline *); 142 int sr_chunk_in_use(struct sr_softc *, dev_t); 143 int sr_rw(struct sr_softc *, dev_t, char *, size_t, 144 daddr_t, long); 145 void sr_wu_done_callback(void *); 146 147 /* don't include these on RAMDISK */ 148 #ifndef SMALL_KERNEL 149 void sr_sensors_refresh(void *); 150 int sr_sensors_create(struct sr_discipline *); 151 void sr_sensors_delete(struct sr_discipline *); 152 #endif 153 154 /* metadata */ 155 int sr_meta_probe(struct sr_discipline *, dev_t *, int); 156 int sr_meta_attach(struct sr_discipline *, int, int); 157 int sr_meta_rw(struct sr_discipline *, dev_t, void *, long); 158 int sr_meta_clear(struct sr_discipline *); 159 void sr_meta_init(struct sr_discipline *, int, int); 160 void sr_meta_init_complete(struct sr_discipline *); 161 void sr_meta_opt_handler(struct sr_discipline *, 162 struct sr_meta_opt_hdr *); 163 164 /* hotplug magic */ 165 void sr_disk_attach(struct disk *, int); 166 167 struct sr_hotplug_list { 168 void (*sh_hotplug)(struct sr_discipline *, 169 struct disk *, int); 170 struct sr_discipline *sh_sd; 171 172 SLIST_ENTRY(sr_hotplug_list) shl_link; 173 }; 174 SLIST_HEAD(sr_hotplug_list_head, sr_hotplug_list); 175 176 struct sr_hotplug_list_head sr_hotplug_callbacks; 177 extern void (*softraid_disk_attach)(struct disk *, int); 178 179 /* scsi glue */ 180 struct scsi_adapter sr_switch = { 181 sr_scsi_cmd, NULL, sr_scsi_probe, NULL, sr_scsi_ioctl 182 }; 183 184 /* native metadata format */ 185 int sr_meta_native_bootprobe(struct sr_softc *, dev_t, 186 struct sr_boot_chunk_head *); 187 #define SR_META_NOTCLAIMED (0) 188 #define SR_META_CLAIMED (1) 189 int sr_meta_native_probe(struct sr_softc *, 190 struct sr_chunk *); 191 int sr_meta_native_attach(struct sr_discipline *, int); 192 int sr_meta_native_write(struct sr_discipline *, dev_t, 193 struct sr_metadata *,void *); 194 195 #ifdef SR_DEBUG 196 void sr_meta_print(struct sr_metadata *); 197 #else 198 #define sr_meta_print(m) 199 #endif 200 201 /* the metadata driver should remain stateless */ 202 struct sr_meta_driver { 203 daddr_t smd_offset; /* metadata location */ 204 u_int32_t smd_size; /* size of metadata */ 205 206 int (*smd_probe)(struct sr_softc *, 207 struct sr_chunk *); 208 int (*smd_attach)(struct sr_discipline *, int); 209 int (*smd_detach)(struct sr_discipline *); 210 int (*smd_read)(struct sr_discipline *, dev_t, 211 struct sr_metadata *, void *); 212 int (*smd_write)(struct sr_discipline *, dev_t, 213 struct sr_metadata *, void *); 214 int (*smd_validate)(struct sr_discipline *, 215 struct sr_metadata *, void *); 216 } smd[] = { 217 { SR_META_OFFSET, SR_META_SIZE * DEV_BSIZE, 218 sr_meta_native_probe, sr_meta_native_attach, NULL, 219 sr_meta_native_read, sr_meta_native_write, NULL }, 220 { 0, 0, NULL, NULL, NULL, NULL } 221 }; 222 223 int 224 sr_meta_attach(struct sr_discipline *sd, int chunk_no, int force) 225 { 226 struct sr_softc *sc = sd->sd_sc; 227 struct sr_chunk_head *cl; 228 struct sr_chunk *ch_entry, *chunk1, *chunk2; 229 int rv = 1, i = 0; 230 231 DNPRINTF(SR_D_META, "%s: sr_meta_attach(%d)\n", DEVNAME(sc), chunk_no); 232 233 /* in memory copy of metadata */ 234 sd->sd_meta = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, 235 M_ZERO | M_NOWAIT); 236 if (!sd->sd_meta) { 237 sr_error(sc, "could not allocate memory for metadata"); 238 goto bad; 239 } 240 241 if (sd->sd_meta_type != SR_META_F_NATIVE) { 242 /* in memory copy of foreign metadata */ 243 sd->sd_meta_foreign = malloc(smd[sd->sd_meta_type].smd_size, 244 M_DEVBUF, M_ZERO | M_NOWAIT); 245 if (!sd->sd_meta_foreign) { 246 /* unwind frees sd_meta */ 247 sr_error(sc, "could not allocate memory for foreign " 248 "metadata"); 249 goto bad; 250 } 251 } 252 253 /* we have a valid list now create an array index */ 254 cl = &sd->sd_vol.sv_chunk_list; 255 sd->sd_vol.sv_chunks = mallocarray(chunk_no, sizeof(struct sr_chunk *), 256 M_DEVBUF, M_WAITOK | M_ZERO); 257 258 /* fill out chunk array */ 259 i = 0; 260 SLIST_FOREACH(ch_entry, cl, src_link) 261 sd->sd_vol.sv_chunks[i++] = ch_entry; 262 263 /* attach metadata */ 264 if (smd[sd->sd_meta_type].smd_attach(sd, force)) 265 goto bad; 266 267 /* Force chunks into correct order now that metadata is attached. */ 268 SLIST_INIT(cl); 269 for (i = 0; i < chunk_no; i++) { 270 ch_entry = sd->sd_vol.sv_chunks[i]; 271 chunk2 = NULL; 272 SLIST_FOREACH(chunk1, cl, src_link) { 273 if (chunk1->src_meta.scmi.scm_chunk_id > 274 ch_entry->src_meta.scmi.scm_chunk_id) 275 break; 276 chunk2 = chunk1; 277 } 278 if (chunk2 == NULL) 279 SLIST_INSERT_HEAD(cl, ch_entry, src_link); 280 else 281 SLIST_INSERT_AFTER(chunk2, ch_entry, src_link); 282 } 283 i = 0; 284 SLIST_FOREACH(ch_entry, cl, src_link) 285 sd->sd_vol.sv_chunks[i++] = ch_entry; 286 287 rv = 0; 288 bad: 289 return (rv); 290 } 291 292 int 293 sr_meta_probe(struct sr_discipline *sd, dev_t *dt, int no_chunk) 294 { 295 struct sr_softc *sc = sd->sd_sc; 296 struct vnode *vn; 297 struct sr_chunk *ch_entry, *ch_prev = NULL; 298 struct sr_chunk_head *cl; 299 char devname[32]; 300 int i, d, type, found, prevf, error; 301 dev_t dev; 302 303 DNPRINTF(SR_D_META, "%s: sr_meta_probe(%d)\n", DEVNAME(sc), no_chunk); 304 305 if (no_chunk == 0) 306 goto unwind; 307 308 cl = &sd->sd_vol.sv_chunk_list; 309 310 for (d = 0, prevf = SR_META_F_INVALID; d < no_chunk; d++) { 311 ch_entry = malloc(sizeof(struct sr_chunk), M_DEVBUF, 312 M_WAITOK | M_ZERO); 313 /* keep disks in user supplied order */ 314 if (ch_prev) 315 SLIST_INSERT_AFTER(ch_prev, ch_entry, src_link); 316 else 317 SLIST_INSERT_HEAD(cl, ch_entry, src_link); 318 ch_prev = ch_entry; 319 dev = dt[d]; 320 ch_entry->src_dev_mm = dev; 321 322 if (dev == NODEV) { 323 ch_entry->src_meta.scm_status = BIOC_SDOFFLINE; 324 continue; 325 } else { 326 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 327 if (bdevvp(dev, &vn)) { 328 sr_error(sc, "sr_meta_probe: cannot allocate " 329 "vnode"); 330 goto unwind; 331 } 332 333 /* 334 * XXX leaving dev open for now; move this to attach 335 * and figure out the open/close dance for unwind. 336 */ 337 error = VOP_OPEN(vn, FREAD | FWRITE, NOCRED, curproc); 338 if (error) { 339 DNPRINTF(SR_D_META,"%s: sr_meta_probe can't " 340 "open %s\n", DEVNAME(sc), devname); 341 vput(vn); 342 goto unwind; 343 } 344 345 strlcpy(ch_entry->src_devname, devname, 346 sizeof(ch_entry->src_devname)); 347 ch_entry->src_vn = vn; 348 } 349 350 /* determine if this is a device we understand */ 351 for (i = 0, found = SR_META_F_INVALID; smd[i].smd_probe; i++) { 352 type = smd[i].smd_probe(sc, ch_entry); 353 if (type == SR_META_F_INVALID) 354 continue; 355 else { 356 found = type; 357 break; 358 } 359 } 360 361 if (found == SR_META_F_INVALID) 362 goto unwind; 363 if (prevf == SR_META_F_INVALID) 364 prevf = found; 365 if (prevf != found) { 366 DNPRINTF(SR_D_META, "%s: prevf != found\n", 367 DEVNAME(sc)); 368 goto unwind; 369 } 370 } 371 372 return (prevf); 373 unwind: 374 return (SR_META_F_INVALID); 375 } 376 377 void 378 sr_meta_getdevname(struct sr_softc *sc, dev_t dev, char *buf, int size) 379 { 380 int maj, unit, part; 381 char *name; 382 383 DNPRINTF(SR_D_META, "%s: sr_meta_getdevname(%p, %d)\n", 384 DEVNAME(sc), buf, size); 385 386 if (!buf) 387 return; 388 389 maj = major(dev); 390 part = DISKPART(dev); 391 unit = DISKUNIT(dev); 392 393 name = findblkname(maj); 394 if (name == NULL) 395 return; 396 397 snprintf(buf, size, "%s%d%c", name, unit, part + 'a'); 398 } 399 400 int 401 sr_rw(struct sr_softc *sc, dev_t dev, char *buf, size_t size, daddr_t blkno, 402 long flags) 403 { 404 struct vnode *vp; 405 struct buf b; 406 size_t bufsize, dma_bufsize; 407 int rv = 1; 408 char *dma_buf; 409 410 DNPRINTF(SR_D_MISC, "%s: sr_rw(0x%x, %p, %zu, %lld 0x%lx)\n", 411 DEVNAME(sc), dev, buf, size, (long long)blkno, flags); 412 413 dma_bufsize = (size > MAXPHYS) ? MAXPHYS : size; 414 dma_buf = dma_alloc(dma_bufsize, PR_WAITOK); 415 416 if (bdevvp(dev, &vp)) { 417 printf("%s: sr_rw: failed to allocate vnode\n", DEVNAME(sc)); 418 goto done; 419 } 420 421 while (size > 0) { 422 DNPRINTF(SR_D_MISC, "%s: dma_buf %p, size %zu, blkno %lld)\n", 423 DEVNAME(sc), dma_buf, size, (long long)blkno); 424 425 bufsize = (size > MAXPHYS) ? MAXPHYS : size; 426 if (flags == B_WRITE) 427 memcpy(dma_buf, buf, bufsize); 428 429 bzero(&b, sizeof(b)); 430 b.b_flags = flags | B_PHYS; 431 b.b_proc = curproc; 432 b.b_dev = dev; 433 b.b_iodone = NULL; 434 b.b_error = 0; 435 b.b_blkno = blkno; 436 b.b_data = dma_buf; 437 b.b_bcount = bufsize; 438 b.b_bufsize = bufsize; 439 b.b_resid = bufsize; 440 b.b_vp = vp; 441 442 if ((b.b_flags & B_READ) == 0) 443 vp->v_numoutput++; 444 445 LIST_INIT(&b.b_dep); 446 VOP_STRATEGY(&b); 447 biowait(&b); 448 449 if (b.b_flags & B_ERROR) { 450 printf("%s: I/O error %d on dev 0x%x at block %llu\n", 451 DEVNAME(sc), b.b_error, dev, b.b_blkno); 452 goto done; 453 } 454 455 if (flags == B_READ) 456 memcpy(buf, dma_buf, bufsize); 457 458 size -= bufsize; 459 buf += bufsize; 460 blkno += howmany(bufsize, DEV_BSIZE); 461 } 462 463 rv = 0; 464 465 done: 466 if (vp) 467 vput(vp); 468 469 dma_free(dma_buf, dma_bufsize); 470 471 return (rv); 472 } 473 474 int 475 sr_meta_rw(struct sr_discipline *sd, dev_t dev, void *md, long flags) 476 { 477 int rv = 1; 478 479 DNPRINTF(SR_D_META, "%s: sr_meta_rw(0x%x, %p, 0x%lx)\n", 480 DEVNAME(sd->sd_sc), dev, md, flags); 481 482 if (md == NULL) { 483 printf("%s: sr_meta_rw: invalid metadata pointer\n", 484 DEVNAME(sd->sd_sc)); 485 goto done; 486 } 487 488 rv = sr_rw(sd->sd_sc, dev, md, SR_META_SIZE * DEV_BSIZE, 489 SR_META_OFFSET, flags); 490 491 done: 492 return (rv); 493 } 494 495 int 496 sr_meta_clear(struct sr_discipline *sd) 497 { 498 struct sr_softc *sc = sd->sd_sc; 499 struct sr_chunk_head *cl = &sd->sd_vol.sv_chunk_list; 500 struct sr_chunk *ch_entry; 501 void *m; 502 int rv = 1; 503 504 DNPRINTF(SR_D_META, "%s: sr_meta_clear\n", DEVNAME(sc)); 505 506 if (sd->sd_meta_type != SR_META_F_NATIVE) { 507 sr_error(sc, "cannot clear foreign metadata"); 508 goto done; 509 } 510 511 m = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, M_WAITOK | M_ZERO); 512 SLIST_FOREACH(ch_entry, cl, src_link) { 513 if (sr_meta_native_write(sd, ch_entry->src_dev_mm, m, NULL)) { 514 /* XXX mark disk offline */ 515 DNPRINTF(SR_D_META, "%s: sr_meta_clear failed to " 516 "clear %s\n", DEVNAME(sc), ch_entry->src_devname); 517 rv++; 518 continue; 519 } 520 bzero(&ch_entry->src_meta, sizeof(ch_entry->src_meta)); 521 } 522 523 bzero(sd->sd_meta, SR_META_SIZE * DEV_BSIZE); 524 525 free(m, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 526 rv = 0; 527 done: 528 return (rv); 529 } 530 531 void 532 sr_meta_init(struct sr_discipline *sd, int level, int no_chunk) 533 { 534 struct sr_softc *sc = sd->sd_sc; 535 struct sr_metadata *sm = sd->sd_meta; 536 struct sr_chunk_head *cl = &sd->sd_vol.sv_chunk_list; 537 struct sr_meta_chunk *scm; 538 struct sr_chunk *chunk; 539 int cid = 0; 540 u_int64_t max_chunk_sz = 0, min_chunk_sz = 0; 541 u_int32_t secsize = DEV_BSIZE; 542 543 DNPRINTF(SR_D_META, "%s: sr_meta_init\n", DEVNAME(sc)); 544 545 if (!sm) 546 return; 547 548 /* Initialise volume metadata. */ 549 sm->ssdi.ssd_magic = SR_MAGIC; 550 sm->ssdi.ssd_version = SR_META_VERSION; 551 sm->ssdi.ssd_vol_flags = sd->sd_meta_flags; 552 sm->ssdi.ssd_volid = 0; 553 sm->ssdi.ssd_chunk_no = no_chunk; 554 sm->ssdi.ssd_level = level; 555 556 sm->ssd_data_blkno = SR_DATA_OFFSET; 557 sm->ssd_ondisk = 0; 558 559 sr_uuid_generate(&sm->ssdi.ssd_uuid); 560 561 /* Initialise chunk metadata and get min/max chunk sizes & secsize. */ 562 SLIST_FOREACH(chunk, cl, src_link) { 563 scm = &chunk->src_meta; 564 scm->scmi.scm_size = chunk->src_size; 565 scm->scmi.scm_chunk_id = cid++; 566 scm->scm_status = BIOC_SDONLINE; 567 scm->scmi.scm_volid = 0; 568 strlcpy(scm->scmi.scm_devname, chunk->src_devname, 569 sizeof(scm->scmi.scm_devname)); 570 memcpy(&scm->scmi.scm_uuid, &sm->ssdi.ssd_uuid, 571 sizeof(scm->scmi.scm_uuid)); 572 sr_checksum(sc, scm, &scm->scm_checksum, 573 sizeof(scm->scm_checksum)); 574 575 if (min_chunk_sz == 0) 576 min_chunk_sz = scm->scmi.scm_size; 577 if (chunk->src_secsize > secsize) 578 secsize = chunk->src_secsize; 579 min_chunk_sz = MIN(min_chunk_sz, scm->scmi.scm_size); 580 max_chunk_sz = MAX(max_chunk_sz, scm->scmi.scm_size); 581 } 582 583 sm->ssdi.ssd_secsize = secsize; 584 585 /* Equalize chunk sizes. */ 586 SLIST_FOREACH(chunk, cl, src_link) 587 chunk->src_meta.scmi.scm_coerced_size = min_chunk_sz; 588 589 sd->sd_vol.sv_chunk_minsz = min_chunk_sz; 590 sd->sd_vol.sv_chunk_maxsz = max_chunk_sz; 591 } 592 593 void 594 sr_meta_init_complete(struct sr_discipline *sd) 595 { 596 #ifdef SR_DEBUG 597 struct sr_softc *sc = sd->sd_sc; 598 #endif 599 struct sr_metadata *sm = sd->sd_meta; 600 601 DNPRINTF(SR_D_META, "%s: sr_meta_complete\n", DEVNAME(sc)); 602 603 /* Complete initialisation of volume metadata. */ 604 strlcpy(sm->ssdi.ssd_vendor, "OPENBSD", sizeof(sm->ssdi.ssd_vendor)); 605 snprintf(sm->ssdi.ssd_product, sizeof(sm->ssdi.ssd_product), 606 "SR %s", sd->sd_name); 607 snprintf(sm->ssdi.ssd_revision, sizeof(sm->ssdi.ssd_revision), 608 "%03d", sm->ssdi.ssd_version); 609 } 610 611 void 612 sr_meta_opt_handler(struct sr_discipline *sd, struct sr_meta_opt_hdr *om) 613 { 614 if (om->som_type != SR_OPT_BOOT) 615 panic("unknown optional metadata type"); 616 } 617 618 void 619 sr_meta_save_callback(void *xsd) 620 { 621 struct sr_discipline *sd = xsd; 622 int s; 623 624 s = splbio(); 625 626 if (sr_meta_save(sd, SR_META_DIRTY)) 627 printf("%s: save metadata failed\n", DEVNAME(sd->sd_sc)); 628 629 sd->sd_must_flush = 0; 630 splx(s); 631 } 632 633 int 634 sr_meta_save(struct sr_discipline *sd, u_int32_t flags) 635 { 636 struct sr_softc *sc = sd->sd_sc; 637 struct sr_metadata *sm = sd->sd_meta, *m; 638 struct sr_meta_driver *s; 639 struct sr_chunk *src; 640 struct sr_meta_chunk *cm; 641 struct sr_workunit wu; 642 struct sr_meta_opt_hdr *omh; 643 struct sr_meta_opt_item *omi; 644 int i; 645 646 DNPRINTF(SR_D_META, "%s: sr_meta_save %s\n", 647 DEVNAME(sc), sd->sd_meta->ssd_devname); 648 649 if (!sm) { 650 printf("%s: no in memory copy of metadata\n", DEVNAME(sc)); 651 goto bad; 652 } 653 654 /* meta scratchpad */ 655 s = &smd[sd->sd_meta_type]; 656 m = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, M_ZERO | M_NOWAIT); 657 if (!m) { 658 printf("%s: could not allocate metadata scratch area\n", 659 DEVNAME(sc)); 660 goto bad; 661 } 662 663 /* from here on out metadata is updated */ 664 restart: 665 sm->ssd_ondisk++; 666 sm->ssd_meta_flags = flags; 667 memcpy(m, sm, sizeof(*m)); 668 669 /* Chunk metadata. */ 670 cm = (struct sr_meta_chunk *)(m + 1); 671 for (i = 0; i < sm->ssdi.ssd_chunk_no; i++) { 672 src = sd->sd_vol.sv_chunks[i]; 673 memcpy(cm, &src->src_meta, sizeof(*cm)); 674 cm++; 675 } 676 677 /* Optional metadata. */ 678 omh = (struct sr_meta_opt_hdr *)(cm); 679 SLIST_FOREACH(omi, &sd->sd_meta_opt, omi_link) { 680 DNPRINTF(SR_D_META, "%s: saving optional metadata type %u with " 681 "length %u\n", DEVNAME(sc), omi->omi_som->som_type, 682 omi->omi_som->som_length); 683 bzero(&omi->omi_som->som_checksum, MD5_DIGEST_LENGTH); 684 sr_checksum(sc, omi->omi_som, &omi->omi_som->som_checksum, 685 omi->omi_som->som_length); 686 memcpy(omh, omi->omi_som, omi->omi_som->som_length); 687 omh = (struct sr_meta_opt_hdr *)((u_int8_t *)omh + 688 omi->omi_som->som_length); 689 } 690 691 for (i = 0; i < sm->ssdi.ssd_chunk_no; i++) { 692 src = sd->sd_vol.sv_chunks[i]; 693 694 /* skip disks that are offline */ 695 if (src->src_meta.scm_status == BIOC_SDOFFLINE) 696 continue; 697 698 /* calculate metadata checksum for correct chunk */ 699 m->ssdi.ssd_chunk_id = i; 700 sr_checksum(sc, m, &m->ssd_checksum, 701 sizeof(struct sr_meta_invariant)); 702 703 #ifdef SR_DEBUG 704 DNPRINTF(SR_D_META, "%s: sr_meta_save %s: volid: %d " 705 "chunkid: %d checksum: ", 706 DEVNAME(sc), src->src_meta.scmi.scm_devname, 707 m->ssdi.ssd_volid, m->ssdi.ssd_chunk_id); 708 709 if (sr_debug & SR_D_META) 710 sr_checksum_print((u_int8_t *)&m->ssd_checksum); 711 DNPRINTF(SR_D_META, "\n"); 712 sr_meta_print(m); 713 #endif 714 715 /* translate and write to disk */ 716 if (s->smd_write(sd, src->src_dev_mm, m, NULL /* XXX */)) { 717 printf("%s: could not write metadata to %s\n", 718 DEVNAME(sc), src->src_devname); 719 /* restart the meta write */ 720 src->src_meta.scm_status = BIOC_SDOFFLINE; 721 /* XXX recalculate volume status */ 722 goto restart; 723 } 724 } 725 726 /* not all disciplines have sync */ 727 if (sd->sd_scsi_sync) { 728 bzero(&wu, sizeof(wu)); 729 wu.swu_flags |= SR_WUF_FAKE; 730 wu.swu_dis = sd; 731 sd->sd_scsi_sync(&wu); 732 } 733 free(m, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 734 return (0); 735 bad: 736 return (1); 737 } 738 739 int 740 sr_meta_read(struct sr_discipline *sd) 741 { 742 struct sr_softc *sc = sd->sd_sc; 743 struct sr_chunk_head *cl = &sd->sd_vol.sv_chunk_list; 744 struct sr_metadata *sm; 745 struct sr_chunk *ch_entry; 746 struct sr_meta_chunk *cp; 747 struct sr_meta_driver *s; 748 void *fm = NULL; 749 int no_disk = 0, got_meta = 0; 750 751 DNPRINTF(SR_D_META, "%s: sr_meta_read\n", DEVNAME(sc)); 752 753 sm = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, M_WAITOK | M_ZERO); 754 s = &smd[sd->sd_meta_type]; 755 if (sd->sd_meta_type != SR_META_F_NATIVE) 756 fm = malloc(s->smd_size, M_DEVBUF, M_WAITOK | M_ZERO); 757 758 cp = (struct sr_meta_chunk *)(sm + 1); 759 SLIST_FOREACH(ch_entry, cl, src_link) { 760 /* skip disks that are offline */ 761 if (ch_entry->src_meta.scm_status == BIOC_SDOFFLINE) { 762 DNPRINTF(SR_D_META, 763 "%s: %s chunk marked offline, spoofing status\n", 764 DEVNAME(sc), ch_entry->src_devname); 765 cp++; /* adjust chunk pointer to match failure */ 766 continue; 767 } else if (s->smd_read(sd, ch_entry->src_dev_mm, sm, fm)) { 768 /* read and translate */ 769 /* XXX mark chunk offline, elsewhere!! */ 770 ch_entry->src_meta.scm_status = BIOC_SDOFFLINE; 771 cp++; /* adjust chunk pointer to match failure */ 772 DNPRINTF(SR_D_META, "%s: sr_meta_read failed\n", 773 DEVNAME(sc)); 774 continue; 775 } 776 777 if (sm->ssdi.ssd_magic != SR_MAGIC) { 778 DNPRINTF(SR_D_META, "%s: sr_meta_read !SR_MAGIC\n", 779 DEVNAME(sc)); 780 continue; 781 } 782 783 /* validate metadata */ 784 if (sr_meta_validate(sd, ch_entry->src_dev_mm, sm, fm)) { 785 DNPRINTF(SR_D_META, "%s: invalid metadata\n", 786 DEVNAME(sc)); 787 no_disk = -1; 788 goto done; 789 } 790 791 /* assume first chunk contains metadata */ 792 if (got_meta == 0) { 793 sr_meta_opt_load(sc, sm, &sd->sd_meta_opt); 794 memcpy(sd->sd_meta, sm, sizeof(*sd->sd_meta)); 795 got_meta = 1; 796 } 797 798 memcpy(&ch_entry->src_meta, cp, sizeof(ch_entry->src_meta)); 799 800 no_disk++; 801 cp++; 802 } 803 804 free(sm, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 805 free(fm, M_DEVBUF, s->smd_size); 806 807 done: 808 DNPRINTF(SR_D_META, "%s: sr_meta_read found %d parts\n", DEVNAME(sc), 809 no_disk); 810 return (no_disk); 811 } 812 813 void 814 sr_meta_opt_load(struct sr_softc *sc, struct sr_metadata *sm, 815 struct sr_meta_opt_head *som) 816 { 817 struct sr_meta_opt_hdr *omh; 818 struct sr_meta_opt_item *omi; 819 u_int8_t checksum[MD5_DIGEST_LENGTH]; 820 int i; 821 822 /* Process optional metadata. */ 823 omh = (struct sr_meta_opt_hdr *)((u_int8_t *)(sm + 1) + 824 sizeof(struct sr_meta_chunk) * sm->ssdi.ssd_chunk_no); 825 for (i = 0; i < sm->ssdi.ssd_opt_no; i++) { 826 827 omi = malloc(sizeof(struct sr_meta_opt_item), M_DEVBUF, 828 M_WAITOK | M_ZERO); 829 SLIST_INSERT_HEAD(som, omi, omi_link); 830 831 if (omh->som_length == 0) { 832 833 /* Load old fixed length optional metadata. */ 834 DNPRINTF(SR_D_META, "%s: old optional metadata of type " 835 "%u\n", DEVNAME(sc), omh->som_type); 836 837 /* Validate checksum. */ 838 sr_checksum(sc, (void *)omh, &checksum, 839 SR_OLD_META_OPT_SIZE - MD5_DIGEST_LENGTH); 840 if (bcmp(&checksum, (void *)omh + SR_OLD_META_OPT_MD5, 841 sizeof(checksum))) 842 panic("%s: invalid optional metadata " 843 "checksum", DEVNAME(sc)); 844 845 /* Determine correct length. */ 846 switch (omh->som_type) { 847 case SR_OPT_CRYPTO: 848 omh->som_length = sizeof(struct sr_meta_crypto); 849 break; 850 case SR_OPT_BOOT: 851 omh->som_length = sizeof(struct sr_meta_boot); 852 break; 853 case SR_OPT_KEYDISK: 854 omh->som_length = 855 sizeof(struct sr_meta_keydisk); 856 break; 857 default: 858 panic("unknown old optional metadata " 859 "type %u\n", omh->som_type); 860 } 861 862 omi->omi_som = malloc(omh->som_length, M_DEVBUF, 863 M_WAITOK | M_ZERO); 864 memcpy((u_int8_t *)omi->omi_som + sizeof(*omi->omi_som), 865 (u_int8_t *)omh + SR_OLD_META_OPT_OFFSET, 866 omh->som_length - sizeof(*omi->omi_som)); 867 omi->omi_som->som_type = omh->som_type; 868 omi->omi_som->som_length = omh->som_length; 869 870 omh = (struct sr_meta_opt_hdr *)((void *)omh + 871 SR_OLD_META_OPT_SIZE); 872 } else { 873 874 /* Load variable length optional metadata. */ 875 DNPRINTF(SR_D_META, "%s: optional metadata of type %u, " 876 "length %u\n", DEVNAME(sc), omh->som_type, 877 omh->som_length); 878 omi->omi_som = malloc(omh->som_length, M_DEVBUF, 879 M_WAITOK | M_ZERO); 880 memcpy(omi->omi_som, omh, omh->som_length); 881 882 /* Validate checksum. */ 883 memcpy(&checksum, &omi->omi_som->som_checksum, 884 MD5_DIGEST_LENGTH); 885 bzero(&omi->omi_som->som_checksum, MD5_DIGEST_LENGTH); 886 sr_checksum(sc, omi->omi_som, 887 &omi->omi_som->som_checksum, omh->som_length); 888 if (bcmp(&checksum, &omi->omi_som->som_checksum, 889 sizeof(checksum))) 890 panic("%s: invalid optional metadata checksum", 891 DEVNAME(sc)); 892 893 omh = (struct sr_meta_opt_hdr *)((void *)omh + 894 omh->som_length); 895 } 896 } 897 } 898 899 int 900 sr_meta_validate(struct sr_discipline *sd, dev_t dev, struct sr_metadata *sm, 901 void *fm) 902 { 903 struct sr_softc *sc = sd->sd_sc; 904 struct sr_meta_driver *s; 905 #ifdef SR_DEBUG 906 struct sr_meta_chunk *mc; 907 #endif 908 u_int8_t checksum[MD5_DIGEST_LENGTH]; 909 char devname[32]; 910 int rv = 1; 911 912 DNPRINTF(SR_D_META, "%s: sr_meta_validate(%p)\n", DEVNAME(sc), sm); 913 914 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 915 916 s = &smd[sd->sd_meta_type]; 917 if (sd->sd_meta_type != SR_META_F_NATIVE) 918 if (s->smd_validate(sd, sm, fm)) { 919 sr_error(sc, "invalid foreign metadata"); 920 goto done; 921 } 922 923 /* 924 * at this point all foreign metadata has been translated to the native 925 * format and will be treated just like the native format 926 */ 927 928 if (sm->ssdi.ssd_magic != SR_MAGIC) { 929 sr_error(sc, "not valid softraid metadata"); 930 goto done; 931 } 932 933 /* Verify metadata checksum. */ 934 sr_checksum(sc, sm, &checksum, sizeof(struct sr_meta_invariant)); 935 if (bcmp(&checksum, &sm->ssd_checksum, sizeof(checksum))) { 936 sr_error(sc, "invalid metadata checksum"); 937 goto done; 938 } 939 940 /* Handle changes between versions. */ 941 if (sm->ssdi.ssd_version == 3) { 942 943 /* 944 * Version 3 - update metadata version and fix up data blkno 945 * value since this did not exist in version 3. 946 */ 947 if (sm->ssd_data_blkno == 0) 948 sm->ssd_data_blkno = SR_META_V3_DATA_OFFSET; 949 sm->ssdi.ssd_secsize = DEV_BSIZE; 950 951 } else if (sm->ssdi.ssd_version == 4) { 952 953 /* 954 * Version 4 - original metadata format did not store 955 * data blkno so fix this up if necessary. 956 */ 957 if (sm->ssd_data_blkno == 0) 958 sm->ssd_data_blkno = SR_DATA_OFFSET; 959 sm->ssdi.ssd_secsize = DEV_BSIZE; 960 961 } else if (sm->ssdi.ssd_version == 5) { 962 963 /* 964 * Version 5 - variable length optional metadata. Migration 965 * from earlier fixed length optional metadata is handled 966 * in sr_meta_read(). 967 */ 968 sm->ssdi.ssd_secsize = DEV_BSIZE; 969 970 } else if (sm->ssdi.ssd_version == SR_META_VERSION) { 971 972 /* 973 * Version 6 - store & report a sector size. 974 */ 975 976 } else { 977 978 sr_error(sc, "cannot read metadata version %u on %s, " 979 "expected version %u or earlier", 980 sm->ssdi.ssd_version, devname, SR_META_VERSION); 981 goto done; 982 983 } 984 985 /* Update version number and revision string. */ 986 sm->ssdi.ssd_version = SR_META_VERSION; 987 snprintf(sm->ssdi.ssd_revision, sizeof(sm->ssdi.ssd_revision), 988 "%03d", SR_META_VERSION); 989 990 #ifdef SR_DEBUG 991 /* warn if disk changed order */ 992 mc = (struct sr_meta_chunk *)(sm + 1); 993 if (strncmp(mc[sm->ssdi.ssd_chunk_id].scmi.scm_devname, devname, 994 sizeof(mc[sm->ssdi.ssd_chunk_id].scmi.scm_devname))) 995 DNPRINTF(SR_D_META, "%s: roaming device %s -> %s\n", 996 DEVNAME(sc), mc[sm->ssdi.ssd_chunk_id].scmi.scm_devname, 997 devname); 998 #endif 999 1000 /* we have meta data on disk */ 1001 DNPRINTF(SR_D_META, "%s: sr_meta_validate valid metadata %s\n", 1002 DEVNAME(sc), devname); 1003 1004 rv = 0; 1005 done: 1006 return (rv); 1007 } 1008 1009 int 1010 sr_meta_native_bootprobe(struct sr_softc *sc, dev_t devno, 1011 struct sr_boot_chunk_head *bch) 1012 { 1013 struct vnode *vn; 1014 struct disklabel label; 1015 struct sr_metadata *md = NULL; 1016 struct sr_discipline *fake_sd = NULL; 1017 struct sr_boot_chunk *bc; 1018 char devname[32]; 1019 dev_t chrdev, rawdev; 1020 int error, i; 1021 int rv = SR_META_NOTCLAIMED; 1022 1023 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe\n", DEVNAME(sc)); 1024 1025 /* 1026 * Use character raw device to avoid SCSI complaints about missing 1027 * media on removable media devices. 1028 */ 1029 chrdev = blktochr(devno); 1030 rawdev = MAKEDISKDEV(major(chrdev), DISKUNIT(devno), RAW_PART); 1031 if (cdevvp(rawdev, &vn)) { 1032 sr_error(sc, "sr_meta_native_bootprobe: cannot allocate vnode"); 1033 goto done; 1034 } 1035 1036 /* open device */ 1037 error = VOP_OPEN(vn, FREAD, NOCRED, curproc); 1038 if (error) { 1039 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe open " 1040 "failed\n", DEVNAME(sc)); 1041 vput(vn); 1042 goto done; 1043 } 1044 1045 /* get disklabel */ 1046 error = VOP_IOCTL(vn, DIOCGDINFO, (caddr_t)&label, FREAD, NOCRED, 1047 curproc); 1048 if (error) { 1049 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe ioctl " 1050 "failed\n", DEVNAME(sc)); 1051 VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1052 vput(vn); 1053 goto done; 1054 } 1055 1056 /* we are done, close device */ 1057 error = VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1058 if (error) { 1059 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe close " 1060 "failed\n", DEVNAME(sc)); 1061 vput(vn); 1062 goto done; 1063 } 1064 vput(vn); 1065 1066 md = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, M_ZERO | M_NOWAIT); 1067 if (md == NULL) { 1068 sr_error(sc, "not enough memory for metadata buffer"); 1069 goto done; 1070 } 1071 1072 /* create fake sd to use utility functions */ 1073 fake_sd = malloc(sizeof(struct sr_discipline), M_DEVBUF, 1074 M_ZERO | M_NOWAIT); 1075 if (fake_sd == NULL) { 1076 sr_error(sc, "not enough memory for fake discipline"); 1077 goto done; 1078 } 1079 fake_sd->sd_sc = sc; 1080 fake_sd->sd_meta_type = SR_META_F_NATIVE; 1081 1082 for (i = 0; i < MAXPARTITIONS; i++) { 1083 if (label.d_partitions[i].p_fstype != FS_RAID) 1084 continue; 1085 1086 /* open partition */ 1087 rawdev = MAKEDISKDEV(major(devno), DISKUNIT(devno), i); 1088 if (bdevvp(rawdev, &vn)) { 1089 sr_error(sc, "sr_meta_native_bootprobe: cannot " 1090 "allocate vnode for partition"); 1091 goto done; 1092 } 1093 error = VOP_OPEN(vn, FREAD, NOCRED, curproc); 1094 if (error) { 1095 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe " 1096 "open failed, partition %d\n", 1097 DEVNAME(sc), i); 1098 vput(vn); 1099 continue; 1100 } 1101 1102 if (sr_meta_native_read(fake_sd, rawdev, md, NULL)) { 1103 sr_error(sc, "native bootprobe could not read native " 1104 "metadata"); 1105 VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1106 vput(vn); 1107 continue; 1108 } 1109 1110 /* are we a softraid partition? */ 1111 if (md->ssdi.ssd_magic != SR_MAGIC) { 1112 VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1113 vput(vn); 1114 continue; 1115 } 1116 1117 sr_meta_getdevname(sc, rawdev, devname, sizeof(devname)); 1118 if (sr_meta_validate(fake_sd, rawdev, md, NULL) == 0) { 1119 /* XXX fix M_WAITOK, this is boot time */ 1120 bc = malloc(sizeof(struct sr_boot_chunk), 1121 M_DEVBUF, M_WAITOK | M_ZERO); 1122 bc->sbc_metadata = malloc(sizeof(struct sr_metadata), 1123 M_DEVBUF, M_WAITOK | M_ZERO); 1124 memcpy(bc->sbc_metadata, md, sizeof(struct sr_metadata)); 1125 bc->sbc_mm = rawdev; 1126 SLIST_INSERT_HEAD(bch, bc, sbc_link); 1127 rv = SR_META_CLAIMED; 1128 } 1129 1130 /* we are done, close partition */ 1131 VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1132 vput(vn); 1133 } 1134 1135 done: 1136 free(fake_sd, M_DEVBUF, sizeof(struct sr_discipline)); 1137 free(md, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 1138 1139 return (rv); 1140 } 1141 1142 int 1143 sr_boot_assembly(struct sr_softc *sc) 1144 { 1145 struct sr_boot_volume_head bvh; 1146 struct sr_boot_chunk_head bch, kdh; 1147 struct sr_boot_volume *bv, *bv1, *bv2; 1148 struct sr_boot_chunk *bc, *bcnext, *bc1, *bc2; 1149 struct sr_disk_head sdklist; 1150 struct sr_disk *sdk; 1151 struct disk *dk; 1152 struct bioc_createraid bcr; 1153 struct sr_meta_chunk *hm; 1154 struct sr_chunk_head *cl; 1155 struct sr_chunk *hotspare, *chunk, *last; 1156 u_int64_t *ondisk = NULL; 1157 dev_t *devs = NULL; 1158 void *data; 1159 char devname[32]; 1160 int rv = 0, i; 1161 1162 DNPRINTF(SR_D_META, "%s: sr_boot_assembly\n", DEVNAME(sc)); 1163 1164 SLIST_INIT(&sdklist); 1165 SLIST_INIT(&bvh); 1166 SLIST_INIT(&bch); 1167 SLIST_INIT(&kdh); 1168 1169 dk = TAILQ_FIRST(&disklist); 1170 while (dk != NULL) { 1171 1172 /* See if this disk has been checked. */ 1173 SLIST_FOREACH(sdk, &sdklist, sdk_link) 1174 if (sdk->sdk_devno == dk->dk_devno) 1175 break; 1176 1177 if (sdk != NULL || dk->dk_devno == NODEV) { 1178 dk = TAILQ_NEXT(dk, dk_link); 1179 continue; 1180 } 1181 1182 /* Add this disk to the list that we've checked. */ 1183 sdk = malloc(sizeof(struct sr_disk), M_DEVBUF, 1184 M_NOWAIT | M_ZERO); 1185 if (sdk == NULL) 1186 goto unwind; 1187 sdk->sdk_devno = dk->dk_devno; 1188 SLIST_INSERT_HEAD(&sdklist, sdk, sdk_link); 1189 1190 /* Only check sd(4) and wd(4) devices. */ 1191 if (strncmp(dk->dk_name, "sd", 2) && 1192 strncmp(dk->dk_name, "wd", 2)) { 1193 dk = TAILQ_NEXT(dk, dk_link); 1194 continue; 1195 } 1196 1197 /* native softraid uses partitions */ 1198 rw_enter_write(&sc->sc_lock); 1199 bio_status_init(&sc->sc_status, &sc->sc_dev); 1200 sr_meta_native_bootprobe(sc, dk->dk_devno, &bch); 1201 rw_exit_write(&sc->sc_lock); 1202 1203 /* probe non-native disks if native failed. */ 1204 1205 /* Restart scan since we may have slept. */ 1206 dk = TAILQ_FIRST(&disklist); 1207 } 1208 1209 /* 1210 * Create a list of volumes and associate chunks with each volume. 1211 */ 1212 for (bc = SLIST_FIRST(&bch); bc != NULL; bc = bcnext) { 1213 1214 bcnext = SLIST_NEXT(bc, sbc_link); 1215 SLIST_REMOVE(&bch, bc, sr_boot_chunk, sbc_link); 1216 bc->sbc_chunk_id = bc->sbc_metadata->ssdi.ssd_chunk_id; 1217 1218 /* Handle key disks separately. */ 1219 if (bc->sbc_metadata->ssdi.ssd_level == SR_KEYDISK_LEVEL) { 1220 SLIST_INSERT_HEAD(&kdh, bc, sbc_link); 1221 continue; 1222 } 1223 1224 SLIST_FOREACH(bv, &bvh, sbv_link) { 1225 if (bcmp(&bc->sbc_metadata->ssdi.ssd_uuid, 1226 &bv->sbv_uuid, 1227 sizeof(bc->sbc_metadata->ssdi.ssd_uuid)) == 0) 1228 break; 1229 } 1230 1231 if (bv == NULL) { 1232 bv = malloc(sizeof(struct sr_boot_volume), 1233 M_DEVBUF, M_NOWAIT | M_ZERO); 1234 if (bv == NULL) { 1235 printf("%s: failed to allocate boot volume\n", 1236 DEVNAME(sc)); 1237 goto unwind; 1238 } 1239 1240 bv->sbv_level = bc->sbc_metadata->ssdi.ssd_level; 1241 bv->sbv_volid = bc->sbc_metadata->ssdi.ssd_volid; 1242 bv->sbv_chunk_no = bc->sbc_metadata->ssdi.ssd_chunk_no; 1243 bv->sbv_flags = bc->sbc_metadata->ssdi.ssd_vol_flags; 1244 memcpy(&bv->sbv_uuid, &bc->sbc_metadata->ssdi.ssd_uuid, 1245 sizeof(bc->sbc_metadata->ssdi.ssd_uuid)); 1246 SLIST_INIT(&bv->sbv_chunks); 1247 1248 /* Maintain volume order. */ 1249 bv2 = NULL; 1250 SLIST_FOREACH(bv1, &bvh, sbv_link) { 1251 if (bv1->sbv_volid > bv->sbv_volid) 1252 break; 1253 bv2 = bv1; 1254 } 1255 if (bv2 == NULL) { 1256 DNPRINTF(SR_D_META, "%s: insert volume %u " 1257 "at head\n", DEVNAME(sc), bv->sbv_volid); 1258 SLIST_INSERT_HEAD(&bvh, bv, sbv_link); 1259 } else { 1260 DNPRINTF(SR_D_META, "%s: insert volume %u " 1261 "after %u\n", DEVNAME(sc), bv->sbv_volid, 1262 bv2->sbv_volid); 1263 SLIST_INSERT_AFTER(bv2, bv, sbv_link); 1264 } 1265 } 1266 1267 /* Maintain chunk order. */ 1268 bc2 = NULL; 1269 SLIST_FOREACH(bc1, &bv->sbv_chunks, sbc_link) { 1270 if (bc1->sbc_chunk_id > bc->sbc_chunk_id) 1271 break; 1272 bc2 = bc1; 1273 } 1274 if (bc2 == NULL) { 1275 DNPRINTF(SR_D_META, "%s: volume %u insert chunk %u " 1276 "at head\n", DEVNAME(sc), bv->sbv_volid, 1277 bc->sbc_chunk_id); 1278 SLIST_INSERT_HEAD(&bv->sbv_chunks, bc, sbc_link); 1279 } else { 1280 DNPRINTF(SR_D_META, "%s: volume %u insert chunk %u " 1281 "after %u\n", DEVNAME(sc), bv->sbv_volid, 1282 bc->sbc_chunk_id, bc2->sbc_chunk_id); 1283 SLIST_INSERT_AFTER(bc2, bc, sbc_link); 1284 } 1285 1286 bv->sbv_chunks_found++; 1287 } 1288 1289 /* Allocate memory for device and ondisk version arrays. */ 1290 devs = mallocarray(BIOC_CRMAXLEN, sizeof(dev_t), M_DEVBUF, 1291 M_NOWAIT); 1292 if (devs == NULL) { 1293 printf("%s: failed to allocate device array\n", DEVNAME(sc)); 1294 goto unwind; 1295 } 1296 ondisk = mallocarray(BIOC_CRMAXLEN, sizeof(u_int64_t), M_DEVBUF, 1297 M_NOWAIT); 1298 if (ondisk == NULL) { 1299 printf("%s: failed to allocate ondisk array\n", DEVNAME(sc)); 1300 goto unwind; 1301 } 1302 1303 /* 1304 * Assemble hotspare "volumes". 1305 */ 1306 SLIST_FOREACH(bv, &bvh, sbv_link) { 1307 1308 /* Check if this is a hotspare "volume". */ 1309 if (bv->sbv_level != SR_HOTSPARE_LEVEL || 1310 bv->sbv_chunk_no != 1) 1311 continue; 1312 1313 #ifdef SR_DEBUG 1314 DNPRINTF(SR_D_META, "%s: assembling hotspare volume ", 1315 DEVNAME(sc)); 1316 if (sr_debug & SR_D_META) 1317 sr_uuid_print(&bv->sbv_uuid, 0); 1318 DNPRINTF(SR_D_META, " volid %u with %u chunks\n", 1319 bv->sbv_volid, bv->sbv_chunk_no); 1320 #endif 1321 1322 /* Create hotspare chunk metadata. */ 1323 hotspare = malloc(sizeof(struct sr_chunk), M_DEVBUF, 1324 M_NOWAIT | M_ZERO); 1325 if (hotspare == NULL) { 1326 printf("%s: failed to allocate hotspare\n", 1327 DEVNAME(sc)); 1328 goto unwind; 1329 } 1330 1331 bc = SLIST_FIRST(&bv->sbv_chunks); 1332 sr_meta_getdevname(sc, bc->sbc_mm, devname, sizeof(devname)); 1333 hotspare->src_dev_mm = bc->sbc_mm; 1334 strlcpy(hotspare->src_devname, devname, 1335 sizeof(hotspare->src_devname)); 1336 hotspare->src_size = bc->sbc_metadata->ssdi.ssd_size; 1337 1338 hm = &hotspare->src_meta; 1339 hm->scmi.scm_volid = SR_HOTSPARE_VOLID; 1340 hm->scmi.scm_chunk_id = 0; 1341 hm->scmi.scm_size = bc->sbc_metadata->ssdi.ssd_size; 1342 hm->scmi.scm_coerced_size = bc->sbc_metadata->ssdi.ssd_size; 1343 strlcpy(hm->scmi.scm_devname, devname, 1344 sizeof(hm->scmi.scm_devname)); 1345 memcpy(&hm->scmi.scm_uuid, &bc->sbc_metadata->ssdi.ssd_uuid, 1346 sizeof(struct sr_uuid)); 1347 1348 sr_checksum(sc, hm, &hm->scm_checksum, 1349 sizeof(struct sr_meta_chunk_invariant)); 1350 1351 hm->scm_status = BIOC_SDHOTSPARE; 1352 1353 /* Add chunk to hotspare list. */ 1354 rw_enter_write(&sc->sc_hs_lock); 1355 cl = &sc->sc_hotspare_list; 1356 if (SLIST_EMPTY(cl)) 1357 SLIST_INSERT_HEAD(cl, hotspare, src_link); 1358 else { 1359 SLIST_FOREACH(chunk, cl, src_link) 1360 last = chunk; 1361 SLIST_INSERT_AFTER(last, hotspare, src_link); 1362 } 1363 sc->sc_hotspare_no++; 1364 rw_exit_write(&sc->sc_hs_lock); 1365 1366 } 1367 1368 /* 1369 * Assemble RAID volumes. 1370 */ 1371 SLIST_FOREACH(bv, &bvh, sbv_link) { 1372 1373 bzero(&bcr, sizeof(bcr)); 1374 data = NULL; 1375 1376 /* Check if this is a hotspare "volume". */ 1377 if (bv->sbv_level == SR_HOTSPARE_LEVEL && 1378 bv->sbv_chunk_no == 1) 1379 continue; 1380 1381 /* 1382 * Skip volumes that are marked as no auto assemble, unless 1383 * this was the volume which we actually booted from. 1384 */ 1385 if (bcmp(&sr_bootuuid, &bv->sbv_uuid, sizeof(sr_bootuuid)) != 0) 1386 if (bv->sbv_flags & BIOC_SCNOAUTOASSEMBLE) 1387 continue; 1388 1389 #ifdef SR_DEBUG 1390 DNPRINTF(SR_D_META, "%s: assembling volume ", DEVNAME(sc)); 1391 if (sr_debug & SR_D_META) 1392 sr_uuid_print(&bv->sbv_uuid, 0); 1393 DNPRINTF(SR_D_META, " volid %u with %u chunks\n", 1394 bv->sbv_volid, bv->sbv_chunk_no); 1395 #endif 1396 1397 /* 1398 * If this is a crypto volume, try to find a matching 1399 * key disk... 1400 */ 1401 bcr.bc_key_disk = NODEV; 1402 if (bv->sbv_level == 'C') { 1403 SLIST_FOREACH(bc, &kdh, sbc_link) { 1404 if (bcmp(&bc->sbc_metadata->ssdi.ssd_uuid, 1405 &bv->sbv_uuid, 1406 sizeof(bc->sbc_metadata->ssdi.ssd_uuid)) 1407 == 0) 1408 bcr.bc_key_disk = bc->sbc_mm; 1409 } 1410 } 1411 1412 for (i = 0; i < BIOC_CRMAXLEN; i++) { 1413 devs[i] = NODEV; /* mark device as illegal */ 1414 ondisk[i] = 0; 1415 } 1416 1417 SLIST_FOREACH(bc, &bv->sbv_chunks, sbc_link) { 1418 if (devs[bc->sbc_chunk_id] != NODEV) { 1419 bv->sbv_chunks_found--; 1420 sr_meta_getdevname(sc, bc->sbc_mm, devname, 1421 sizeof(devname)); 1422 printf("%s: found duplicate chunk %u for " 1423 "volume %u on device %s\n", DEVNAME(sc), 1424 bc->sbc_chunk_id, bv->sbv_volid, devname); 1425 } 1426 1427 if (devs[bc->sbc_chunk_id] == NODEV || 1428 bc->sbc_metadata->ssd_ondisk > 1429 ondisk[bc->sbc_chunk_id]) { 1430 devs[bc->sbc_chunk_id] = bc->sbc_mm; 1431 ondisk[bc->sbc_chunk_id] = 1432 bc->sbc_metadata->ssd_ondisk; 1433 DNPRINTF(SR_D_META, "%s: using ondisk " 1434 "metadata version %llu for chunk %u\n", 1435 DEVNAME(sc), ondisk[bc->sbc_chunk_id], 1436 bc->sbc_chunk_id); 1437 } 1438 } 1439 1440 if (bv->sbv_chunk_no != bv->sbv_chunks_found) { 1441 printf("%s: not all chunks were provided; " 1442 "attempting to bring volume %d online\n", 1443 DEVNAME(sc), bv->sbv_volid); 1444 } 1445 1446 bcr.bc_level = bv->sbv_level; 1447 bcr.bc_dev_list_len = bv->sbv_chunk_no * sizeof(dev_t); 1448 bcr.bc_dev_list = devs; 1449 bcr.bc_flags = BIOC_SCDEVT | 1450 (bv->sbv_flags & BIOC_SCNOAUTOASSEMBLE); 1451 1452 if (bv->sbv_level == 'C' && 1453 bcmp(&sr_bootuuid, &bv->sbv_uuid, sizeof(sr_bootuuid)) == 0) 1454 data = sr_bootkey; 1455 1456 rw_enter_write(&sc->sc_lock); 1457 bio_status_init(&sc->sc_status, &sc->sc_dev); 1458 sr_ioctl_createraid(sc, &bcr, 0, data); 1459 rw_exit_write(&sc->sc_lock); 1460 1461 rv++; 1462 } 1463 1464 /* done with metadata */ 1465 unwind: 1466 /* Free boot volumes and associated chunks. */ 1467 for (bv1 = SLIST_FIRST(&bvh); bv1 != NULL; bv1 = bv2) { 1468 bv2 = SLIST_NEXT(bv1, sbv_link); 1469 for (bc1 = SLIST_FIRST(&bv1->sbv_chunks); bc1 != NULL; 1470 bc1 = bc2) { 1471 bc2 = SLIST_NEXT(bc1, sbc_link); 1472 free(bc1->sbc_metadata, M_DEVBUF, 1473 sizeof(*bc1->sbc_metadata)); 1474 free(bc1, M_DEVBUF, sizeof(*bc1)); 1475 } 1476 free(bv1, M_DEVBUF, sizeof(*bv1)); 1477 } 1478 /* Free keydisks chunks. */ 1479 for (bc1 = SLIST_FIRST(&kdh); bc1 != NULL; bc1 = bc2) { 1480 bc2 = SLIST_NEXT(bc1, sbc_link); 1481 free(bc1->sbc_metadata, M_DEVBUF, sizeof(*bc1->sbc_metadata)); 1482 free(bc1, M_DEVBUF, sizeof(*bc1)); 1483 } 1484 /* Free unallocated chunks. */ 1485 for (bc1 = SLIST_FIRST(&bch); bc1 != NULL; bc1 = bc2) { 1486 bc2 = SLIST_NEXT(bc1, sbc_link); 1487 free(bc1->sbc_metadata, M_DEVBUF, sizeof(*bc1->sbc_metadata)); 1488 free(bc1, M_DEVBUF, sizeof(*bc1)); 1489 } 1490 1491 while (!SLIST_EMPTY(&sdklist)) { 1492 sdk = SLIST_FIRST(&sdklist); 1493 SLIST_REMOVE_HEAD(&sdklist, sdk_link); 1494 free(sdk, M_DEVBUF, sizeof(*sdk)); 1495 } 1496 1497 free(devs, M_DEVBUF, BIOC_CRMAXLEN * sizeof(dev_t)); 1498 free(ondisk, M_DEVBUF, BIOC_CRMAXLEN * sizeof(u_int64_t)); 1499 1500 return (rv); 1501 } 1502 1503 void 1504 sr_map_root(void) 1505 { 1506 struct sr_softc *sc = softraid0; 1507 struct sr_discipline *sd; 1508 struct sr_meta_opt_item *omi; 1509 struct sr_meta_boot *sbm; 1510 u_char duid[8]; 1511 int i; 1512 1513 DNPRINTF(SR_D_MISC, "%s: sr_map_root\n", DEVNAME(sc)); 1514 1515 if (sc == NULL) 1516 return; 1517 1518 bzero(duid, sizeof(duid)); 1519 if (bcmp(rootduid, duid, sizeof(duid)) == 0) { 1520 DNPRINTF(SR_D_MISC, "%s: root duid is zero\n", DEVNAME(sc)); 1521 return; 1522 } 1523 1524 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 1525 SLIST_FOREACH(omi, &sd->sd_meta_opt, omi_link) { 1526 if (omi->omi_som->som_type != SR_OPT_BOOT) 1527 continue; 1528 sbm = (struct sr_meta_boot *)omi->omi_som; 1529 for (i = 0; i < SR_MAX_BOOT_DISKS; i++) { 1530 if (bcmp(rootduid, sbm->sbm_boot_duid[i], 1531 sizeof(rootduid)) == 0) { 1532 memcpy(rootduid, sbm->sbm_root_duid, 1533 sizeof(rootduid)); 1534 DNPRINTF(SR_D_MISC, "%s: root duid " 1535 "mapped to %s\n", DEVNAME(sc), 1536 duid_format(rootduid)); 1537 return; 1538 } 1539 } 1540 } 1541 } 1542 } 1543 1544 int 1545 sr_meta_native_probe(struct sr_softc *sc, struct sr_chunk *ch_entry) 1546 { 1547 struct disklabel label; 1548 char *devname; 1549 int error, part; 1550 u_int64_t size; 1551 1552 DNPRINTF(SR_D_META, "%s: sr_meta_native_probe(%s)\n", 1553 DEVNAME(sc), ch_entry->src_devname); 1554 1555 devname = ch_entry->src_devname; 1556 part = DISKPART(ch_entry->src_dev_mm); 1557 1558 /* get disklabel */ 1559 error = VOP_IOCTL(ch_entry->src_vn, DIOCGDINFO, (caddr_t)&label, FREAD, 1560 NOCRED, curproc); 1561 if (error) { 1562 DNPRINTF(SR_D_META, "%s: %s can't obtain disklabel\n", 1563 DEVNAME(sc), devname); 1564 goto unwind; 1565 } 1566 memcpy(ch_entry->src_duid, label.d_uid, sizeof(ch_entry->src_duid)); 1567 1568 /* make sure the partition is of the right type */ 1569 if (label.d_partitions[part].p_fstype != FS_RAID) { 1570 DNPRINTF(SR_D_META, 1571 "%s: %s partition not of type RAID (%d)\n", DEVNAME(sc), 1572 devname, 1573 label.d_partitions[part].p_fstype); 1574 goto unwind; 1575 } 1576 1577 size = DL_SECTOBLK(&label, DL_GETPSIZE(&label.d_partitions[part])); 1578 if (size <= SR_DATA_OFFSET) { 1579 DNPRINTF(SR_D_META, "%s: %s partition too small\n", DEVNAME(sc), 1580 devname); 1581 goto unwind; 1582 } 1583 size -= SR_DATA_OFFSET; 1584 if (size > INT64_MAX) { 1585 DNPRINTF(SR_D_META, "%s: %s partition too large\n", DEVNAME(sc), 1586 devname); 1587 goto unwind; 1588 } 1589 ch_entry->src_size = size; 1590 ch_entry->src_secsize = label.d_secsize; 1591 1592 DNPRINTF(SR_D_META, "%s: probe found %s size %lld\n", DEVNAME(sc), 1593 devname, (long long)size); 1594 1595 return (SR_META_F_NATIVE); 1596 unwind: 1597 DNPRINTF(SR_D_META, "%s: invalid device: %s\n", DEVNAME(sc), 1598 devname ? devname : "nodev"); 1599 return (SR_META_F_INVALID); 1600 } 1601 1602 int 1603 sr_meta_native_attach(struct sr_discipline *sd, int force) 1604 { 1605 struct sr_softc *sc = sd->sd_sc; 1606 struct sr_chunk_head *cl = &sd->sd_vol.sv_chunk_list; 1607 struct sr_metadata *md = NULL; 1608 struct sr_chunk *ch_entry, *ch_next; 1609 struct sr_uuid uuid; 1610 u_int64_t version = 0; 1611 int sr, not_sr, rv = 1, d, expected = -1, old_meta = 0; 1612 1613 DNPRINTF(SR_D_META, "%s: sr_meta_native_attach\n", DEVNAME(sc)); 1614 1615 md = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, M_ZERO | M_NOWAIT); 1616 if (md == NULL) { 1617 sr_error(sc, "not enough memory for metadata buffer"); 1618 goto bad; 1619 } 1620 1621 bzero(&uuid, sizeof uuid); 1622 1623 sr = not_sr = d = 0; 1624 SLIST_FOREACH(ch_entry, cl, src_link) { 1625 if (ch_entry->src_dev_mm == NODEV) 1626 continue; 1627 1628 if (sr_meta_native_read(sd, ch_entry->src_dev_mm, md, NULL)) { 1629 sr_error(sc, "could not read native metadata"); 1630 goto bad; 1631 } 1632 1633 if (md->ssdi.ssd_magic == SR_MAGIC) { 1634 sr++; 1635 ch_entry->src_meta.scmi.scm_chunk_id = 1636 md->ssdi.ssd_chunk_id; 1637 if (d == 0) { 1638 memcpy(&uuid, &md->ssdi.ssd_uuid, sizeof uuid); 1639 expected = md->ssdi.ssd_chunk_no; 1640 version = md->ssd_ondisk; 1641 d++; 1642 continue; 1643 } else if (bcmp(&md->ssdi.ssd_uuid, &uuid, 1644 sizeof uuid)) { 1645 sr_error(sc, "not part of the same volume"); 1646 goto bad; 1647 } 1648 if (md->ssd_ondisk != version) { 1649 old_meta++; 1650 version = MAX(md->ssd_ondisk, version); 1651 } 1652 } else 1653 not_sr++; 1654 } 1655 1656 if (sr && not_sr && !force) { 1657 sr_error(sc, "not all chunks are of the native metadata " 1658 "format"); 1659 goto bad; 1660 } 1661 1662 /* mixed metadata versions; mark bad disks offline */ 1663 if (old_meta) { 1664 d = 0; 1665 for (ch_entry = SLIST_FIRST(cl); ch_entry != NULL; 1666 ch_entry = ch_next, d++) { 1667 ch_next = SLIST_NEXT(ch_entry, src_link); 1668 1669 /* XXX do we want to read this again? */ 1670 if (ch_entry->src_dev_mm == NODEV) 1671 panic("src_dev_mm == NODEV"); 1672 if (sr_meta_native_read(sd, ch_entry->src_dev_mm, md, 1673 NULL)) 1674 sr_warn(sc, "could not read native metadata"); 1675 if (md->ssd_ondisk != version) 1676 sd->sd_vol.sv_chunks[d]->src_meta.scm_status = 1677 BIOC_SDOFFLINE; 1678 } 1679 } 1680 1681 if (expected != sr && !force && expected != -1) { 1682 DNPRINTF(SR_D_META, "%s: not all chunks were provided, trying " 1683 "anyway\n", DEVNAME(sc)); 1684 } 1685 1686 rv = 0; 1687 bad: 1688 free(md, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 1689 return (rv); 1690 } 1691 1692 int 1693 sr_meta_native_read(struct sr_discipline *sd, dev_t dev, 1694 struct sr_metadata *md, void *fm) 1695 { 1696 #ifdef SR_DEBUG 1697 struct sr_softc *sc = sd->sd_sc; 1698 #endif 1699 DNPRINTF(SR_D_META, "%s: sr_meta_native_read(0x%x, %p)\n", 1700 DEVNAME(sc), dev, md); 1701 1702 return (sr_meta_rw(sd, dev, md, B_READ)); 1703 } 1704 1705 int 1706 sr_meta_native_write(struct sr_discipline *sd, dev_t dev, 1707 struct sr_metadata *md, void *fm) 1708 { 1709 #ifdef SR_DEBUG 1710 struct sr_softc *sc = sd->sd_sc; 1711 #endif 1712 DNPRINTF(SR_D_META, "%s: sr_meta_native_write(0x%x, %p)\n", 1713 DEVNAME(sc), dev, md); 1714 1715 return (sr_meta_rw(sd, dev, md, B_WRITE)); 1716 } 1717 1718 void 1719 sr_hotplug_register(struct sr_discipline *sd, void *func) 1720 { 1721 struct sr_hotplug_list *mhe; 1722 1723 DNPRINTF(SR_D_MISC, "%s: sr_hotplug_register: %p\n", 1724 DEVNAME(sd->sd_sc), func); 1725 1726 /* make sure we aren't on the list yet */ 1727 SLIST_FOREACH(mhe, &sr_hotplug_callbacks, shl_link) 1728 if (mhe->sh_hotplug == func) 1729 return; 1730 1731 mhe = malloc(sizeof(struct sr_hotplug_list), M_DEVBUF, 1732 M_WAITOK | M_ZERO); 1733 mhe->sh_hotplug = func; 1734 mhe->sh_sd = sd; 1735 SLIST_INSERT_HEAD(&sr_hotplug_callbacks, mhe, shl_link); 1736 } 1737 1738 void 1739 sr_hotplug_unregister(struct sr_discipline *sd, void *func) 1740 { 1741 struct sr_hotplug_list *mhe; 1742 1743 DNPRINTF(SR_D_MISC, "%s: sr_hotplug_unregister: %s %p\n", 1744 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, func); 1745 1746 /* make sure we are on the list yet */ 1747 SLIST_FOREACH(mhe, &sr_hotplug_callbacks, shl_link) { 1748 if (mhe->sh_hotplug == func) 1749 break; 1750 } 1751 if (mhe != NULL) { 1752 SLIST_REMOVE(&sr_hotplug_callbacks, mhe, 1753 sr_hotplug_list, shl_link); 1754 free(mhe, M_DEVBUF, sizeof(*mhe)); 1755 } 1756 } 1757 1758 void 1759 sr_disk_attach(struct disk *diskp, int action) 1760 { 1761 struct sr_hotplug_list *mhe; 1762 1763 SLIST_FOREACH(mhe, &sr_hotplug_callbacks, shl_link) 1764 if (mhe->sh_sd->sd_ready) 1765 mhe->sh_hotplug(mhe->sh_sd, diskp, action); 1766 } 1767 1768 int 1769 sr_match(struct device *parent, void *match, void *aux) 1770 { 1771 return (1); 1772 } 1773 1774 void 1775 sr_attach(struct device *parent, struct device *self, void *aux) 1776 { 1777 struct sr_softc *sc = (void *)self; 1778 struct scsibus_attach_args saa; 1779 1780 DNPRINTF(SR_D_MISC, "\n%s: sr_attach", DEVNAME(sc)); 1781 1782 if (softraid0 == NULL) 1783 softraid0 = sc; 1784 1785 rw_init(&sc->sc_lock, "sr_lock"); 1786 rw_init(&sc->sc_hs_lock, "sr_hs_lock"); 1787 1788 SLIST_INIT(&sr_hotplug_callbacks); 1789 TAILQ_INIT(&sc->sc_dis_list); 1790 SLIST_INIT(&sc->sc_hotspare_list); 1791 1792 #if NBIO > 0 1793 if (bio_register(&sc->sc_dev, sr_bio_ioctl) != 0) 1794 printf("%s: controller registration failed", DEVNAME(sc)); 1795 #endif /* NBIO > 0 */ 1796 1797 #ifndef SMALL_KERNEL 1798 strlcpy(sc->sc_sensordev.xname, DEVNAME(sc), 1799 sizeof(sc->sc_sensordev.xname)); 1800 sensordev_install(&sc->sc_sensordev); 1801 #endif /* SMALL_KERNEL */ 1802 1803 printf("\n"); 1804 1805 saa.saa_adapter_softc = sc; 1806 saa.saa_adapter = &sr_switch; 1807 saa.saa_adapter_target = SDEV_NO_ADAPTER_TARGET; 1808 saa.saa_adapter_buswidth = SR_MAX_LD; 1809 saa.saa_luns = 1; 1810 saa.saa_openings = 0; 1811 saa.saa_pool = NULL; 1812 saa.saa_quirks = saa.saa_flags = 0; 1813 saa.saa_wwpn = saa.saa_wwnn = 0; 1814 1815 sc->sc_scsibus = (struct scsibus_softc *)config_found(&sc->sc_dev, &saa, 1816 scsiprint); 1817 1818 softraid_disk_attach = sr_disk_attach; 1819 1820 sr_boot_assembly(sc); 1821 1822 explicit_bzero(sr_bootkey, sizeof(sr_bootkey)); 1823 } 1824 1825 int 1826 sr_detach(struct device *self, int flags) 1827 { 1828 struct sr_softc *sc = (void *)self; 1829 int rv; 1830 1831 DNPRINTF(SR_D_MISC, "%s: sr_detach\n", DEVNAME(sc)); 1832 1833 softraid_disk_attach = NULL; 1834 1835 sr_shutdown(0); 1836 1837 #ifndef SMALL_KERNEL 1838 if (sc->sc_sensor_task != NULL) 1839 sensor_task_unregister(sc->sc_sensor_task); 1840 sensordev_deinstall(&sc->sc_sensordev); 1841 #endif /* SMALL_KERNEL */ 1842 1843 if (sc->sc_scsibus != NULL) { 1844 rv = config_detach((struct device *)sc->sc_scsibus, flags); 1845 if (rv != 0) 1846 return (rv); 1847 sc->sc_scsibus = NULL; 1848 } 1849 1850 return (0); 1851 } 1852 1853 void 1854 sr_info(struct sr_softc *sc, const char *fmt, ...) 1855 { 1856 va_list ap; 1857 1858 rw_assert_wrlock(&sc->sc_lock); 1859 1860 va_start(ap, fmt); 1861 bio_status(&sc->sc_status, 0, BIO_MSG_INFO, fmt, &ap); 1862 va_end(ap); 1863 } 1864 1865 void 1866 sr_warn(struct sr_softc *sc, const char *fmt, ...) 1867 { 1868 va_list ap; 1869 1870 rw_assert_wrlock(&sc->sc_lock); 1871 1872 va_start(ap, fmt); 1873 bio_status(&sc->sc_status, 1, BIO_MSG_WARN, fmt, &ap); 1874 va_end(ap); 1875 } 1876 1877 void 1878 sr_error(struct sr_softc *sc, const char *fmt, ...) 1879 { 1880 va_list ap; 1881 1882 rw_assert_wrlock(&sc->sc_lock); 1883 1884 va_start(ap, fmt); 1885 bio_status(&sc->sc_status, 1, BIO_MSG_ERROR, fmt, &ap); 1886 va_end(ap); 1887 } 1888 1889 void 1890 sr_copy_internal_data(struct scsi_xfer *xs, void *v, size_t size) 1891 { 1892 size_t copy_cnt; 1893 1894 DNPRINTF(SR_D_MISC, "sr_copy_internal_data xs: %p size: %zu\n", 1895 xs, size); 1896 1897 if (xs->datalen) { 1898 copy_cnt = MIN(size, xs->datalen); 1899 memcpy(xs->data, v, copy_cnt); 1900 } 1901 } 1902 1903 int 1904 sr_ccb_alloc(struct sr_discipline *sd) 1905 { 1906 struct sr_ccb *ccb; 1907 int i; 1908 1909 if (!sd) 1910 return (1); 1911 1912 DNPRINTF(SR_D_CCB, "%s: sr_ccb_alloc\n", DEVNAME(sd->sd_sc)); 1913 1914 if (sd->sd_ccb) 1915 return (1); 1916 1917 sd->sd_ccb = mallocarray(sd->sd_max_wu, 1918 sd->sd_max_ccb_per_wu * sizeof(struct sr_ccb), 1919 M_DEVBUF, M_WAITOK | M_ZERO); 1920 TAILQ_INIT(&sd->sd_ccb_freeq); 1921 for (i = 0; i < sd->sd_max_wu * sd->sd_max_ccb_per_wu; i++) { 1922 ccb = &sd->sd_ccb[i]; 1923 ccb->ccb_dis = sd; 1924 sr_ccb_put(ccb); 1925 } 1926 1927 DNPRINTF(SR_D_CCB, "%s: sr_ccb_alloc ccb: %d\n", 1928 DEVNAME(sd->sd_sc), sd->sd_max_wu * sd->sd_max_ccb_per_wu); 1929 1930 return (0); 1931 } 1932 1933 void 1934 sr_ccb_free(struct sr_discipline *sd) 1935 { 1936 struct sr_ccb *ccb; 1937 1938 if (!sd) 1939 return; 1940 1941 DNPRINTF(SR_D_CCB, "%s: sr_ccb_free %p\n", DEVNAME(sd->sd_sc), sd); 1942 1943 while ((ccb = TAILQ_FIRST(&sd->sd_ccb_freeq)) != NULL) 1944 TAILQ_REMOVE(&sd->sd_ccb_freeq, ccb, ccb_link); 1945 1946 free(sd->sd_ccb, M_DEVBUF, sd->sd_max_wu * sd->sd_max_ccb_per_wu * 1947 sizeof(struct sr_ccb)); 1948 } 1949 1950 struct sr_ccb * 1951 sr_ccb_get(struct sr_discipline *sd) 1952 { 1953 struct sr_ccb *ccb; 1954 int s; 1955 1956 s = splbio(); 1957 1958 ccb = TAILQ_FIRST(&sd->sd_ccb_freeq); 1959 if (ccb) { 1960 TAILQ_REMOVE(&sd->sd_ccb_freeq, ccb, ccb_link); 1961 ccb->ccb_state = SR_CCB_INPROGRESS; 1962 } 1963 1964 splx(s); 1965 1966 DNPRINTF(SR_D_CCB, "%s: sr_ccb_get: %p\n", DEVNAME(sd->sd_sc), 1967 ccb); 1968 1969 return (ccb); 1970 } 1971 1972 void 1973 sr_ccb_put(struct sr_ccb *ccb) 1974 { 1975 struct sr_discipline *sd = ccb->ccb_dis; 1976 int s; 1977 1978 DNPRINTF(SR_D_CCB, "%s: sr_ccb_put: %p\n", DEVNAME(sd->sd_sc), 1979 ccb); 1980 1981 s = splbio(); 1982 1983 ccb->ccb_wu = NULL; 1984 ccb->ccb_state = SR_CCB_FREE; 1985 ccb->ccb_target = -1; 1986 ccb->ccb_opaque = NULL; 1987 1988 TAILQ_INSERT_TAIL(&sd->sd_ccb_freeq, ccb, ccb_link); 1989 1990 splx(s); 1991 } 1992 1993 struct sr_ccb * 1994 sr_ccb_rw(struct sr_discipline *sd, int chunk, daddr_t blkno, 1995 long len, u_int8_t *data, int xsflags, int ccbflags) 1996 { 1997 struct sr_chunk *sc = sd->sd_vol.sv_chunks[chunk]; 1998 struct sr_ccb *ccb = NULL; 1999 2000 ccb = sr_ccb_get(sd); 2001 if (ccb == NULL) 2002 goto out; 2003 2004 ccb->ccb_flags = ccbflags; 2005 ccb->ccb_target = chunk; 2006 2007 ccb->ccb_buf.b_flags = B_PHYS | B_CALL; 2008 if (ISSET(xsflags, SCSI_DATA_IN)) 2009 ccb->ccb_buf.b_flags |= B_READ; 2010 else 2011 ccb->ccb_buf.b_flags |= B_WRITE; 2012 2013 ccb->ccb_buf.b_blkno = blkno + sd->sd_meta->ssd_data_blkno; 2014 ccb->ccb_buf.b_bcount = len; 2015 ccb->ccb_buf.b_bufsize = len; 2016 ccb->ccb_buf.b_resid = len; 2017 ccb->ccb_buf.b_data = data; 2018 ccb->ccb_buf.b_error = 0; 2019 ccb->ccb_buf.b_iodone = sd->sd_scsi_intr; 2020 ccb->ccb_buf.b_proc = curproc; 2021 ccb->ccb_buf.b_dev = sc->src_dev_mm; 2022 ccb->ccb_buf.b_vp = sc->src_vn; 2023 ccb->ccb_buf.b_bq = NULL; 2024 2025 if (!ISSET(ccb->ccb_buf.b_flags, B_READ)) 2026 ccb->ccb_buf.b_vp->v_numoutput++; 2027 2028 LIST_INIT(&ccb->ccb_buf.b_dep); 2029 2030 DNPRINTF(SR_D_DIS, "%s: %s %s ccb " 2031 "b_bcount %ld b_blkno %lld b_flags 0x%0lx b_data %p\n", 2032 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, sd->sd_name, 2033 ccb->ccb_buf.b_bcount, (long long)ccb->ccb_buf.b_blkno, 2034 ccb->ccb_buf.b_flags, ccb->ccb_buf.b_data); 2035 2036 out: 2037 return ccb; 2038 } 2039 2040 void 2041 sr_ccb_done(struct sr_ccb *ccb) 2042 { 2043 struct sr_workunit *wu = ccb->ccb_wu; 2044 struct sr_discipline *sd = wu->swu_dis; 2045 struct sr_softc *sc = sd->sd_sc; 2046 2047 DNPRINTF(SR_D_INTR, "%s: %s %s ccb done b_bcount %ld b_resid %zu" 2048 " b_flags 0x%0lx block %lld target %d\n", 2049 DEVNAME(sc), sd->sd_meta->ssd_devname, sd->sd_name, 2050 ccb->ccb_buf.b_bcount, ccb->ccb_buf.b_resid, ccb->ccb_buf.b_flags, 2051 (long long)ccb->ccb_buf.b_blkno, ccb->ccb_target); 2052 2053 splassert(IPL_BIO); 2054 2055 if (ccb->ccb_target == -1) 2056 panic("%s: invalid target on wu: %p", DEVNAME(sc), wu); 2057 2058 if (ccb->ccb_buf.b_flags & B_ERROR) { 2059 DNPRINTF(SR_D_INTR, "%s: i/o error on block %lld target %d\n", 2060 DEVNAME(sc), (long long)ccb->ccb_buf.b_blkno, 2061 ccb->ccb_target); 2062 if (ISSET(sd->sd_capabilities, SR_CAP_REDUNDANT)) 2063 sd->sd_set_chunk_state(sd, ccb->ccb_target, 2064 BIOC_SDOFFLINE); 2065 else 2066 printf("%s: %s: i/o error %d @ %s block %lld\n", 2067 DEVNAME(sc), sd->sd_meta->ssd_devname, 2068 ccb->ccb_buf.b_error, sd->sd_name, 2069 (long long)ccb->ccb_buf.b_blkno); 2070 ccb->ccb_state = SR_CCB_FAILED; 2071 wu->swu_ios_failed++; 2072 } else { 2073 ccb->ccb_state = SR_CCB_OK; 2074 wu->swu_ios_succeeded++; 2075 } 2076 2077 wu->swu_ios_complete++; 2078 } 2079 2080 int 2081 sr_wu_alloc(struct sr_discipline *sd) 2082 { 2083 struct sr_workunit *wu; 2084 int i, no_wu; 2085 2086 DNPRINTF(SR_D_WU, "%s: sr_wu_alloc %p %d\n", DEVNAME(sd->sd_sc), 2087 sd, sd->sd_max_wu); 2088 2089 no_wu = sd->sd_max_wu; 2090 sd->sd_wu_pending = no_wu; 2091 2092 mtx_init(&sd->sd_wu_mtx, IPL_BIO); 2093 TAILQ_INIT(&sd->sd_wu); 2094 TAILQ_INIT(&sd->sd_wu_freeq); 2095 TAILQ_INIT(&sd->sd_wu_pendq); 2096 TAILQ_INIT(&sd->sd_wu_defq); 2097 2098 for (i = 0; i < no_wu; i++) { 2099 wu = malloc(sd->sd_wu_size, M_DEVBUF, M_WAITOK | M_ZERO); 2100 TAILQ_INSERT_TAIL(&sd->sd_wu, wu, swu_next); 2101 TAILQ_INIT(&wu->swu_ccb); 2102 wu->swu_dis = sd; 2103 task_set(&wu->swu_task, sr_wu_done_callback, wu); 2104 sr_wu_put(sd, wu); 2105 } 2106 2107 return (0); 2108 } 2109 2110 void 2111 sr_wu_free(struct sr_discipline *sd) 2112 { 2113 struct sr_workunit *wu; 2114 2115 DNPRINTF(SR_D_WU, "%s: sr_wu_free %p\n", DEVNAME(sd->sd_sc), sd); 2116 2117 while ((wu = TAILQ_FIRST(&sd->sd_wu_freeq)) != NULL) 2118 TAILQ_REMOVE(&sd->sd_wu_freeq, wu, swu_link); 2119 while ((wu = TAILQ_FIRST(&sd->sd_wu_pendq)) != NULL) 2120 TAILQ_REMOVE(&sd->sd_wu_pendq, wu, swu_link); 2121 while ((wu = TAILQ_FIRST(&sd->sd_wu_defq)) != NULL) 2122 TAILQ_REMOVE(&sd->sd_wu_defq, wu, swu_link); 2123 2124 while ((wu = TAILQ_FIRST(&sd->sd_wu)) != NULL) { 2125 TAILQ_REMOVE(&sd->sd_wu, wu, swu_next); 2126 free(wu, M_DEVBUF, sd->sd_wu_size); 2127 } 2128 } 2129 2130 void * 2131 sr_wu_get(void *xsd) 2132 { 2133 struct sr_discipline *sd = (struct sr_discipline *)xsd; 2134 struct sr_workunit *wu; 2135 2136 mtx_enter(&sd->sd_wu_mtx); 2137 wu = TAILQ_FIRST(&sd->sd_wu_freeq); 2138 if (wu) { 2139 TAILQ_REMOVE(&sd->sd_wu_freeq, wu, swu_link); 2140 sd->sd_wu_pending++; 2141 } 2142 mtx_leave(&sd->sd_wu_mtx); 2143 2144 DNPRINTF(SR_D_WU, "%s: sr_wu_get: %p\n", DEVNAME(sd->sd_sc), wu); 2145 2146 return (wu); 2147 } 2148 2149 void 2150 sr_wu_put(void *xsd, void *xwu) 2151 { 2152 struct sr_discipline *sd = (struct sr_discipline *)xsd; 2153 struct sr_workunit *wu = (struct sr_workunit *)xwu; 2154 2155 DNPRINTF(SR_D_WU, "%s: sr_wu_put: %p\n", DEVNAME(sd->sd_sc), wu); 2156 2157 sr_wu_release_ccbs(wu); 2158 sr_wu_init(sd, wu); 2159 2160 mtx_enter(&sd->sd_wu_mtx); 2161 TAILQ_INSERT_TAIL(&sd->sd_wu_freeq, wu, swu_link); 2162 sd->sd_wu_pending--; 2163 mtx_leave(&sd->sd_wu_mtx); 2164 } 2165 2166 void 2167 sr_wu_init(struct sr_discipline *sd, struct sr_workunit *wu) 2168 { 2169 int s; 2170 2171 s = splbio(); 2172 if (wu->swu_cb_active == 1) 2173 panic("%s: sr_wu_init got active wu", DEVNAME(sd->sd_sc)); 2174 splx(s); 2175 2176 wu->swu_xs = NULL; 2177 wu->swu_state = SR_WU_FREE; 2178 wu->swu_flags = 0; 2179 wu->swu_blk_start = 0; 2180 wu->swu_blk_end = 0; 2181 wu->swu_collider = NULL; 2182 } 2183 2184 void 2185 sr_wu_enqueue_ccb(struct sr_workunit *wu, struct sr_ccb *ccb) 2186 { 2187 struct sr_discipline *sd = wu->swu_dis; 2188 int s; 2189 2190 s = splbio(); 2191 if (wu->swu_cb_active == 1) 2192 panic("%s: sr_wu_enqueue_ccb got active wu", 2193 DEVNAME(sd->sd_sc)); 2194 ccb->ccb_wu = wu; 2195 wu->swu_io_count++; 2196 TAILQ_INSERT_TAIL(&wu->swu_ccb, ccb, ccb_link); 2197 splx(s); 2198 } 2199 2200 void 2201 sr_wu_release_ccbs(struct sr_workunit *wu) 2202 { 2203 struct sr_ccb *ccb; 2204 2205 /* Return all ccbs that are associated with this workunit. */ 2206 while ((ccb = TAILQ_FIRST(&wu->swu_ccb)) != NULL) { 2207 TAILQ_REMOVE(&wu->swu_ccb, ccb, ccb_link); 2208 sr_ccb_put(ccb); 2209 } 2210 2211 wu->swu_io_count = 0; 2212 wu->swu_ios_complete = 0; 2213 wu->swu_ios_failed = 0; 2214 wu->swu_ios_succeeded = 0; 2215 } 2216 2217 void 2218 sr_wu_done(struct sr_workunit *wu) 2219 { 2220 struct sr_discipline *sd = wu->swu_dis; 2221 2222 DNPRINTF(SR_D_INTR, "%s: sr_wu_done count %d completed %d failed %d\n", 2223 DEVNAME(sd->sd_sc), wu->swu_io_count, wu->swu_ios_complete, 2224 wu->swu_ios_failed); 2225 2226 if (wu->swu_ios_complete < wu->swu_io_count) 2227 return; 2228 2229 task_add(sd->sd_taskq, &wu->swu_task); 2230 } 2231 2232 void 2233 sr_wu_done_callback(void *xwu) 2234 { 2235 struct sr_workunit *wu = xwu; 2236 struct sr_discipline *sd = wu->swu_dis; 2237 struct scsi_xfer *xs = wu->swu_xs; 2238 struct sr_workunit *wup; 2239 int s; 2240 2241 /* 2242 * The SR_WUF_DISCIPLINE or SR_WUF_REBUILD flag must be set if 2243 * the work unit is not associated with a scsi_xfer. 2244 */ 2245 KASSERT(xs != NULL || 2246 (wu->swu_flags & (SR_WUF_DISCIPLINE|SR_WUF_REBUILD))); 2247 2248 s = splbio(); 2249 2250 if (xs != NULL) { 2251 if (wu->swu_ios_failed) 2252 xs->error = XS_DRIVER_STUFFUP; 2253 else 2254 xs->error = XS_NOERROR; 2255 } 2256 2257 if (sd->sd_scsi_wu_done) { 2258 if (sd->sd_scsi_wu_done(wu) == SR_WU_RESTART) 2259 goto done; 2260 } 2261 2262 /* Remove work unit from pending queue. */ 2263 TAILQ_FOREACH(wup, &sd->sd_wu_pendq, swu_link) 2264 if (wup == wu) 2265 break; 2266 if (wup == NULL) 2267 panic("%s: wu %p not on pending queue", 2268 DEVNAME(sd->sd_sc), wu); 2269 TAILQ_REMOVE(&sd->sd_wu_pendq, wu, swu_link); 2270 2271 if (wu->swu_collider) { 2272 if (wu->swu_ios_failed) 2273 sr_raid_recreate_wu(wu->swu_collider); 2274 2275 /* XXX Should the collider be failed if this xs failed? */ 2276 sr_raid_startwu(wu->swu_collider); 2277 } 2278 2279 /* 2280 * If a discipline provides its own sd_scsi_done function, then it 2281 * is responsible for calling sr_scsi_done() once I/O is complete. 2282 */ 2283 if (wu->swu_flags & SR_WUF_REBUILD) 2284 wu->swu_flags |= SR_WUF_REBUILDIOCOMP; 2285 if (wu->swu_flags & SR_WUF_WAKEUP) 2286 wakeup(wu); 2287 if (sd->sd_scsi_done) 2288 sd->sd_scsi_done(wu); 2289 else if (wu->swu_flags & SR_WUF_DISCIPLINE) 2290 sr_scsi_wu_put(sd, wu); 2291 else if (!(wu->swu_flags & SR_WUF_REBUILD)) 2292 sr_scsi_done(sd, xs); 2293 2294 done: 2295 splx(s); 2296 } 2297 2298 struct sr_workunit * 2299 sr_scsi_wu_get(struct sr_discipline *sd, int flags) 2300 { 2301 return scsi_io_get(&sd->sd_iopool, flags); 2302 } 2303 2304 void 2305 sr_scsi_wu_put(struct sr_discipline *sd, struct sr_workunit *wu) 2306 { 2307 scsi_io_put(&sd->sd_iopool, wu); 2308 2309 if (sd->sd_sync && sd->sd_wu_pending == 0) 2310 wakeup(sd); 2311 } 2312 2313 void 2314 sr_scsi_done(struct sr_discipline *sd, struct scsi_xfer *xs) 2315 { 2316 DNPRINTF(SR_D_DIS, "%s: sr_scsi_done: xs %p\n", DEVNAME(sd->sd_sc), xs); 2317 2318 if (xs->error == XS_NOERROR) 2319 xs->resid = 0; 2320 2321 scsi_done(xs); 2322 2323 if (sd->sd_sync && sd->sd_wu_pending == 0) 2324 wakeup(sd); 2325 } 2326 2327 void 2328 sr_scsi_cmd(struct scsi_xfer *xs) 2329 { 2330 struct scsi_link *link = xs->sc_link; 2331 struct sr_softc *sc = link->bus->sb_adapter_softc; 2332 struct sr_workunit *wu = xs->io; 2333 struct sr_discipline *sd; 2334 2335 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd target %d xs %p flags %#x\n", 2336 DEVNAME(sc), link->target, xs, xs->flags); 2337 2338 sd = sc->sc_targets[link->target]; 2339 if (sd == NULL) 2340 panic("%s: sr_scsi_cmd NULL discipline", DEVNAME(sc)); 2341 2342 if (sd->sd_deleted) { 2343 printf("%s: %s device is being deleted, failing io\n", 2344 DEVNAME(sc), sd->sd_meta->ssd_devname); 2345 goto stuffup; 2346 } 2347 2348 /* scsi layer *can* re-send wu without calling sr_wu_put(). */ 2349 sr_wu_release_ccbs(wu); 2350 sr_wu_init(sd, wu); 2351 wu->swu_state = SR_WU_INPROGRESS; 2352 wu->swu_xs = xs; 2353 2354 switch (xs->cmd->opcode) { 2355 case READ_COMMAND: 2356 case READ_BIG: 2357 case READ_16: 2358 case WRITE_COMMAND: 2359 case WRITE_BIG: 2360 case WRITE_16: 2361 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: READ/WRITE %02x\n", 2362 DEVNAME(sc), xs->cmd->opcode); 2363 if (sd->sd_scsi_rw(wu)) 2364 goto stuffup; 2365 break; 2366 2367 case SYNCHRONIZE_CACHE: 2368 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: SYNCHRONIZE_CACHE\n", 2369 DEVNAME(sc)); 2370 if (sd->sd_scsi_sync(wu)) 2371 goto stuffup; 2372 goto complete; 2373 2374 case TEST_UNIT_READY: 2375 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: TEST_UNIT_READY\n", 2376 DEVNAME(sc)); 2377 if (sd->sd_scsi_tur(wu)) 2378 goto stuffup; 2379 goto complete; 2380 2381 case START_STOP: 2382 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: START_STOP\n", 2383 DEVNAME(sc)); 2384 if (sd->sd_scsi_start_stop(wu)) 2385 goto stuffup; 2386 goto complete; 2387 2388 case INQUIRY: 2389 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: INQUIRY\n", 2390 DEVNAME(sc)); 2391 if (sd->sd_scsi_inquiry(wu)) 2392 goto stuffup; 2393 goto complete; 2394 2395 case READ_CAPACITY: 2396 case READ_CAPACITY_16: 2397 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd READ CAPACITY 0x%02x\n", 2398 DEVNAME(sc), xs->cmd->opcode); 2399 if (sd->sd_scsi_read_cap(wu)) 2400 goto stuffup; 2401 goto complete; 2402 2403 case REQUEST_SENSE: 2404 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd REQUEST SENSE\n", 2405 DEVNAME(sc)); 2406 if (sd->sd_scsi_req_sense(wu)) 2407 goto stuffup; 2408 goto complete; 2409 2410 default: 2411 DNPRINTF(SR_D_CMD, "%s: unsupported scsi command %x\n", 2412 DEVNAME(sc), xs->cmd->opcode); 2413 /* XXX might need to add generic function to handle others */ 2414 goto stuffup; 2415 } 2416 2417 return; 2418 stuffup: 2419 if (sd->sd_scsi_sense.error_code) { 2420 xs->error = XS_SENSE; 2421 memcpy(&xs->sense, &sd->sd_scsi_sense, sizeof(xs->sense)); 2422 bzero(&sd->sd_scsi_sense, sizeof(sd->sd_scsi_sense)); 2423 } else { 2424 xs->error = XS_DRIVER_STUFFUP; 2425 } 2426 complete: 2427 sr_scsi_done(sd, xs); 2428 } 2429 2430 int 2431 sr_scsi_probe(struct scsi_link *link) 2432 { 2433 struct sr_softc *sc = link->bus->sb_adapter_softc; 2434 struct sr_discipline *sd; 2435 2436 KASSERT(link->target < SR_MAX_LD && link->lun == 0); 2437 2438 sd = sc->sc_targets[link->target]; 2439 if (sd == NULL) 2440 return (ENODEV); 2441 2442 link->pool = &sd->sd_iopool; 2443 if (sd->sd_openings) 2444 link->openings = sd->sd_openings(sd); 2445 else 2446 link->openings = sd->sd_max_wu; 2447 2448 return (0); 2449 } 2450 2451 int 2452 sr_scsi_ioctl(struct scsi_link *link, u_long cmd, caddr_t addr, int flag) 2453 { 2454 struct sr_softc *sc = link->bus->sb_adapter_softc; 2455 struct sr_discipline *sd; 2456 2457 sd = sc->sc_targets[link->target]; 2458 if (sd == NULL) 2459 return (ENODEV); 2460 2461 DNPRINTF(SR_D_IOCTL, "%s: %s sr_scsi_ioctl cmd: %#lx\n", 2462 DEVNAME(sc), sd->sd_meta->ssd_devname, cmd); 2463 2464 /* Pass bio ioctls through to the bio handler. */ 2465 if (IOCGROUP(cmd) == 'B') 2466 return (sr_bio_handler(sc, sd, cmd, (struct bio *)addr)); 2467 2468 switch (cmd) { 2469 case DIOCGCACHE: 2470 case DIOCSCACHE: 2471 return (EOPNOTSUPP); 2472 default: 2473 return (ENOTTY); 2474 } 2475 } 2476 2477 int 2478 sr_bio_ioctl(struct device *dev, u_long cmd, caddr_t addr) 2479 { 2480 struct sr_softc *sc = (struct sr_softc *) dev; 2481 DNPRINTF(SR_D_IOCTL, "%s: sr_bio_ioctl\n", DEVNAME(sc)); 2482 2483 return sr_bio_handler(sc, NULL, cmd, (struct bio *)addr); 2484 } 2485 2486 int 2487 sr_bio_handler(struct sr_softc *sc, struct sr_discipline *sd, u_long cmd, 2488 struct bio *bio) 2489 { 2490 int rv = 0; 2491 2492 DNPRINTF(SR_D_IOCTL, "%s: sr_bio_handler ", DEVNAME(sc)); 2493 2494 rw_enter_write(&sc->sc_lock); 2495 2496 bio_status_init(&sc->sc_status, &sc->sc_dev); 2497 2498 switch (cmd) { 2499 case BIOCINQ: 2500 DNPRINTF(SR_D_IOCTL, "inq\n"); 2501 rv = sr_ioctl_inq(sc, (struct bioc_inq *)bio); 2502 break; 2503 2504 case BIOCVOL: 2505 DNPRINTF(SR_D_IOCTL, "vol\n"); 2506 rv = sr_ioctl_vol(sc, (struct bioc_vol *)bio); 2507 break; 2508 2509 case BIOCDISK: 2510 DNPRINTF(SR_D_IOCTL, "disk\n"); 2511 rv = sr_ioctl_disk(sc, (struct bioc_disk *)bio); 2512 break; 2513 2514 case BIOCALARM: 2515 DNPRINTF(SR_D_IOCTL, "alarm\n"); 2516 /*rv = sr_ioctl_alarm(sc, (struct bioc_alarm *)bio); */ 2517 break; 2518 2519 case BIOCBLINK: 2520 DNPRINTF(SR_D_IOCTL, "blink\n"); 2521 /*rv = sr_ioctl_blink(sc, (struct bioc_blink *)bio); */ 2522 break; 2523 2524 case BIOCSETSTATE: 2525 DNPRINTF(SR_D_IOCTL, "setstate\n"); 2526 rv = sr_ioctl_setstate(sc, (struct bioc_setstate *)bio); 2527 break; 2528 2529 case BIOCCREATERAID: 2530 DNPRINTF(SR_D_IOCTL, "createraid\n"); 2531 rv = sr_ioctl_createraid(sc, (struct bioc_createraid *)bio, 2532 1, NULL); 2533 break; 2534 2535 case BIOCDELETERAID: 2536 DNPRINTF(SR_D_IOCTL, "deleteraid\n"); 2537 rv = sr_ioctl_deleteraid(sc, sd, (struct bioc_deleteraid *)bio); 2538 break; 2539 2540 case BIOCDISCIPLINE: 2541 DNPRINTF(SR_D_IOCTL, "discipline\n"); 2542 rv = sr_ioctl_discipline(sc, sd, (struct bioc_discipline *)bio); 2543 break; 2544 2545 case BIOCINSTALLBOOT: 2546 DNPRINTF(SR_D_IOCTL, "installboot\n"); 2547 rv = sr_ioctl_installboot(sc, sd, 2548 (struct bioc_installboot *)bio); 2549 break; 2550 2551 default: 2552 DNPRINTF(SR_D_IOCTL, "invalid ioctl\n"); 2553 rv = ENOTTY; 2554 } 2555 2556 sc->sc_status.bs_status = (rv ? BIO_STATUS_ERROR : BIO_STATUS_SUCCESS); 2557 2558 if (sc->sc_status.bs_msg_count > 0) 2559 rv = 0; 2560 2561 memcpy(&bio->bio_status, &sc->sc_status, sizeof(struct bio_status)); 2562 2563 rw_exit_write(&sc->sc_lock); 2564 2565 return (rv); 2566 } 2567 2568 int 2569 sr_ioctl_inq(struct sr_softc *sc, struct bioc_inq *bi) 2570 { 2571 struct sr_discipline *sd; 2572 int vol = 0, disk = 0; 2573 2574 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2575 vol++; 2576 disk += sd->sd_meta->ssdi.ssd_chunk_no; 2577 } 2578 2579 strlcpy(bi->bi_dev, sc->sc_dev.dv_xname, sizeof(bi->bi_dev)); 2580 bi->bi_novol = vol + sc->sc_hotspare_no; 2581 bi->bi_nodisk = disk + sc->sc_hotspare_no; 2582 2583 return (0); 2584 } 2585 2586 int 2587 sr_ioctl_vol(struct sr_softc *sc, struct bioc_vol *bv) 2588 { 2589 int vol = -1, rv = EINVAL; 2590 struct sr_discipline *sd; 2591 struct sr_chunk *hotspare; 2592 2593 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2594 vol++; 2595 if (vol != bv->bv_volid) 2596 continue; 2597 2598 bv->bv_status = sd->sd_vol_status; 2599 bv->bv_size = sd->sd_meta->ssdi.ssd_size << DEV_BSHIFT; 2600 bv->bv_level = sd->sd_meta->ssdi.ssd_level; 2601 bv->bv_nodisk = sd->sd_meta->ssdi.ssd_chunk_no; 2602 2603 #ifdef CRYPTO 2604 if (sd->sd_meta->ssdi.ssd_level == 'C' && 2605 sd->mds.mdd_crypto.key_disk != NULL) 2606 bv->bv_nodisk++; 2607 #endif 2608 if (bv->bv_status == BIOC_SVREBUILD) 2609 bv->bv_percent = sr_rebuild_percent(sd); 2610 2611 strlcpy(bv->bv_dev, sd->sd_meta->ssd_devname, 2612 sizeof(bv->bv_dev)); 2613 strlcpy(bv->bv_vendor, sd->sd_meta->ssdi.ssd_vendor, 2614 sizeof(bv->bv_vendor)); 2615 rv = 0; 2616 goto done; 2617 } 2618 2619 /* Check hotspares list. */ 2620 SLIST_FOREACH(hotspare, &sc->sc_hotspare_list, src_link) { 2621 vol++; 2622 if (vol != bv->bv_volid) 2623 continue; 2624 2625 bv->bv_status = BIOC_SVONLINE; 2626 bv->bv_size = hotspare->src_meta.scmi.scm_size << DEV_BSHIFT; 2627 bv->bv_level = -1; /* Hotspare. */ 2628 bv->bv_nodisk = 1; 2629 strlcpy(bv->bv_dev, hotspare->src_meta.scmi.scm_devname, 2630 sizeof(bv->bv_dev)); 2631 strlcpy(bv->bv_vendor, hotspare->src_meta.scmi.scm_devname, 2632 sizeof(bv->bv_vendor)); 2633 rv = 0; 2634 goto done; 2635 } 2636 2637 done: 2638 return (rv); 2639 } 2640 2641 int 2642 sr_ioctl_disk(struct sr_softc *sc, struct bioc_disk *bd) 2643 { 2644 struct sr_discipline *sd; 2645 struct sr_chunk *src, *hotspare; 2646 int vol = -1, rv = EINVAL; 2647 2648 if (bd->bd_diskid < 0) 2649 goto done; 2650 2651 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2652 vol++; 2653 if (vol != bd->bd_volid) 2654 continue; 2655 2656 if (bd->bd_diskid < sd->sd_meta->ssdi.ssd_chunk_no) 2657 src = sd->sd_vol.sv_chunks[bd->bd_diskid]; 2658 #ifdef CRYPTO 2659 else if (bd->bd_diskid == sd->sd_meta->ssdi.ssd_chunk_no && 2660 sd->sd_meta->ssdi.ssd_level == 'C' && 2661 sd->mds.mdd_crypto.key_disk != NULL) 2662 src = sd->mds.mdd_crypto.key_disk; 2663 #endif 2664 else 2665 break; 2666 2667 bd->bd_status = src->src_meta.scm_status; 2668 bd->bd_size = src->src_meta.scmi.scm_size << DEV_BSHIFT; 2669 bd->bd_channel = vol; 2670 bd->bd_target = bd->bd_diskid; 2671 strlcpy(bd->bd_vendor, src->src_meta.scmi.scm_devname, 2672 sizeof(bd->bd_vendor)); 2673 rv = 0; 2674 goto done; 2675 } 2676 2677 /* Check hotspares list. */ 2678 SLIST_FOREACH(hotspare, &sc->sc_hotspare_list, src_link) { 2679 vol++; 2680 if (vol != bd->bd_volid) 2681 continue; 2682 2683 if (bd->bd_diskid != 0) 2684 break; 2685 2686 bd->bd_status = hotspare->src_meta.scm_status; 2687 bd->bd_size = hotspare->src_meta.scmi.scm_size << DEV_BSHIFT; 2688 bd->bd_channel = vol; 2689 bd->bd_target = bd->bd_diskid; 2690 strlcpy(bd->bd_vendor, hotspare->src_meta.scmi.scm_devname, 2691 sizeof(bd->bd_vendor)); 2692 rv = 0; 2693 goto done; 2694 } 2695 2696 done: 2697 return (rv); 2698 } 2699 2700 int 2701 sr_ioctl_setstate(struct sr_softc *sc, struct bioc_setstate *bs) 2702 { 2703 int rv = EINVAL; 2704 int vol = -1, found, c; 2705 struct sr_discipline *sd; 2706 struct sr_chunk *ch_entry; 2707 struct sr_chunk_head *cl; 2708 2709 if (bs->bs_other_id_type == BIOC_SSOTHER_UNUSED) 2710 goto done; 2711 2712 if (bs->bs_status == BIOC_SSHOTSPARE) { 2713 rv = sr_hotspare(sc, (dev_t)bs->bs_other_id); 2714 goto done; 2715 } 2716 2717 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2718 vol++; 2719 if (vol == bs->bs_volid) 2720 break; 2721 } 2722 if (sd == NULL) 2723 goto done; 2724 2725 switch (bs->bs_status) { 2726 case BIOC_SSOFFLINE: 2727 /* Take chunk offline */ 2728 found = c = 0; 2729 cl = &sd->sd_vol.sv_chunk_list; 2730 SLIST_FOREACH(ch_entry, cl, src_link) { 2731 if (ch_entry->src_dev_mm == bs->bs_other_id) { 2732 found = 1; 2733 break; 2734 } 2735 c++; 2736 } 2737 if (found == 0) { 2738 sr_error(sc, "chunk not part of array"); 2739 goto done; 2740 } 2741 2742 /* XXX: check current state first */ 2743 sd->sd_set_chunk_state(sd, c, BIOC_SDOFFLINE); 2744 2745 if (sr_meta_save(sd, SR_META_DIRTY)) { 2746 sr_error(sc, "could not save metadata for %s", 2747 sd->sd_meta->ssd_devname); 2748 goto done; 2749 } 2750 rv = 0; 2751 break; 2752 2753 case BIOC_SDSCRUB: 2754 break; 2755 2756 case BIOC_SSREBUILD: 2757 rv = sr_rebuild_init(sd, (dev_t)bs->bs_other_id, 0); 2758 break; 2759 2760 default: 2761 sr_error(sc, "unsupported state request %d", bs->bs_status); 2762 } 2763 2764 done: 2765 return (rv); 2766 } 2767 2768 int 2769 sr_chunk_in_use(struct sr_softc *sc, dev_t dev) 2770 { 2771 struct sr_discipline *sd; 2772 struct sr_chunk *chunk; 2773 int i; 2774 2775 DNPRINTF(SR_D_MISC, "%s: sr_chunk_in_use(%d)\n", DEVNAME(sc), dev); 2776 2777 if (dev == NODEV) 2778 return BIOC_SDINVALID; 2779 2780 /* See if chunk is already in use. */ 2781 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2782 for (i = 0; i < sd->sd_meta->ssdi.ssd_chunk_no; i++) { 2783 chunk = sd->sd_vol.sv_chunks[i]; 2784 if (chunk->src_dev_mm == dev) 2785 return chunk->src_meta.scm_status; 2786 } 2787 } 2788 2789 /* Check hotspares list. */ 2790 SLIST_FOREACH(chunk, &sc->sc_hotspare_list, src_link) 2791 if (chunk->src_dev_mm == dev) 2792 return chunk->src_meta.scm_status; 2793 2794 return BIOC_SDINVALID; 2795 } 2796 2797 int 2798 sr_hotspare(struct sr_softc *sc, dev_t dev) 2799 { 2800 struct sr_discipline *sd = NULL; 2801 struct sr_metadata *sm = NULL; 2802 struct sr_meta_chunk *hm; 2803 struct sr_chunk_head *cl; 2804 struct sr_chunk *chunk, *last, *hotspare = NULL; 2805 struct sr_uuid uuid; 2806 struct disklabel label; 2807 struct vnode *vn; 2808 u_int64_t size; 2809 char devname[32]; 2810 int rv = EINVAL; 2811 int c, part, open = 0; 2812 2813 /* 2814 * Add device to global hotspares list. 2815 */ 2816 2817 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 2818 2819 /* Make sure chunk is not already in use. */ 2820 c = sr_chunk_in_use(sc, dev); 2821 if (c != BIOC_SDINVALID && c != BIOC_SDOFFLINE) { 2822 if (c == BIOC_SDHOTSPARE) 2823 sr_error(sc, "%s is already a hotspare", devname); 2824 else 2825 sr_error(sc, "%s is already in use", devname); 2826 goto done; 2827 } 2828 2829 /* XXX - See if there is an existing degraded volume... */ 2830 2831 /* Open device. */ 2832 if (bdevvp(dev, &vn)) { 2833 sr_error(sc, "sr_hotspare: cannot allocate vnode"); 2834 goto done; 2835 } 2836 if (VOP_OPEN(vn, FREAD | FWRITE, NOCRED, curproc)) { 2837 DNPRINTF(SR_D_META,"%s: sr_hotspare cannot open %s\n", 2838 DEVNAME(sc), devname); 2839 vput(vn); 2840 goto fail; 2841 } 2842 open = 1; /* close dev on error */ 2843 2844 /* Get partition details. */ 2845 part = DISKPART(dev); 2846 if (VOP_IOCTL(vn, DIOCGDINFO, (caddr_t)&label, FREAD, 2847 NOCRED, curproc)) { 2848 DNPRINTF(SR_D_META, "%s: sr_hotspare ioctl failed\n", 2849 DEVNAME(sc)); 2850 goto fail; 2851 } 2852 if (label.d_partitions[part].p_fstype != FS_RAID) { 2853 sr_error(sc, "%s partition not of type RAID (%d)", 2854 devname, label.d_partitions[part].p_fstype); 2855 goto fail; 2856 } 2857 2858 /* Calculate partition size. */ 2859 size = DL_SECTOBLK(&label, DL_GETPSIZE(&label.d_partitions[part])); 2860 if (size <= SR_DATA_OFFSET) { 2861 DNPRINTF(SR_D_META, "%s: %s partition too small\n", DEVNAME(sc), 2862 devname); 2863 goto fail; 2864 } 2865 size -= SR_DATA_OFFSET; 2866 if (size > INT64_MAX) { 2867 DNPRINTF(SR_D_META, "%s: %s partition too large\n", DEVNAME(sc), 2868 devname); 2869 goto fail; 2870 } 2871 2872 /* 2873 * Create and populate chunk metadata. 2874 */ 2875 2876 sr_uuid_generate(&uuid); 2877 hotspare = malloc(sizeof(struct sr_chunk), M_DEVBUF, M_WAITOK | M_ZERO); 2878 2879 hotspare->src_dev_mm = dev; 2880 hotspare->src_vn = vn; 2881 strlcpy(hotspare->src_devname, devname, sizeof(hm->scmi.scm_devname)); 2882 hotspare->src_size = size; 2883 2884 hm = &hotspare->src_meta; 2885 hm->scmi.scm_volid = SR_HOTSPARE_VOLID; 2886 hm->scmi.scm_chunk_id = 0; 2887 hm->scmi.scm_size = size; 2888 hm->scmi.scm_coerced_size = size; 2889 strlcpy(hm->scmi.scm_devname, devname, sizeof(hm->scmi.scm_devname)); 2890 memcpy(&hm->scmi.scm_uuid, &uuid, sizeof(struct sr_uuid)); 2891 2892 sr_checksum(sc, hm, &hm->scm_checksum, 2893 sizeof(struct sr_meta_chunk_invariant)); 2894 2895 hm->scm_status = BIOC_SDHOTSPARE; 2896 2897 /* 2898 * Create and populate our own discipline and metadata. 2899 */ 2900 2901 sm = malloc(sizeof(struct sr_metadata), M_DEVBUF, M_WAITOK | M_ZERO); 2902 sm->ssdi.ssd_magic = SR_MAGIC; 2903 sm->ssdi.ssd_version = SR_META_VERSION; 2904 sm->ssd_ondisk = 0; 2905 sm->ssdi.ssd_vol_flags = 0; 2906 memcpy(&sm->ssdi.ssd_uuid, &uuid, sizeof(struct sr_uuid)); 2907 sm->ssdi.ssd_chunk_no = 1; 2908 sm->ssdi.ssd_volid = SR_HOTSPARE_VOLID; 2909 sm->ssdi.ssd_level = SR_HOTSPARE_LEVEL; 2910 sm->ssdi.ssd_size = size; 2911 sm->ssdi.ssd_secsize = label.d_secsize; 2912 strlcpy(sm->ssdi.ssd_vendor, "OPENBSD", sizeof(sm->ssdi.ssd_vendor)); 2913 snprintf(sm->ssdi.ssd_product, sizeof(sm->ssdi.ssd_product), 2914 "SR %s", "HOTSPARE"); 2915 snprintf(sm->ssdi.ssd_revision, sizeof(sm->ssdi.ssd_revision), 2916 "%03d", SR_META_VERSION); 2917 2918 sd = malloc(sizeof(struct sr_discipline), M_DEVBUF, M_WAITOK | M_ZERO); 2919 sd->sd_sc = sc; 2920 sd->sd_meta = sm; 2921 sd->sd_meta_type = SR_META_F_NATIVE; 2922 sd->sd_vol_status = BIOC_SVONLINE; 2923 strlcpy(sd->sd_name, "HOTSPARE", sizeof(sd->sd_name)); 2924 SLIST_INIT(&sd->sd_meta_opt); 2925 2926 /* Add chunk to volume. */ 2927 sd->sd_vol.sv_chunks = malloc(sizeof(struct sr_chunk *), M_DEVBUF, 2928 M_WAITOK | M_ZERO); 2929 sd->sd_vol.sv_chunks[0] = hotspare; 2930 SLIST_INIT(&sd->sd_vol.sv_chunk_list); 2931 SLIST_INSERT_HEAD(&sd->sd_vol.sv_chunk_list, hotspare, src_link); 2932 2933 /* Save metadata. */ 2934 if (sr_meta_save(sd, SR_META_DIRTY)) { 2935 sr_error(sc, "could not save metadata to %s", devname); 2936 goto fail; 2937 } 2938 2939 /* 2940 * Add chunk to hotspare list. 2941 */ 2942 rw_enter_write(&sc->sc_hs_lock); 2943 cl = &sc->sc_hotspare_list; 2944 if (SLIST_EMPTY(cl)) 2945 SLIST_INSERT_HEAD(cl, hotspare, src_link); 2946 else { 2947 SLIST_FOREACH(chunk, cl, src_link) 2948 last = chunk; 2949 SLIST_INSERT_AFTER(last, hotspare, src_link); 2950 } 2951 sc->sc_hotspare_no++; 2952 rw_exit_write(&sc->sc_hs_lock); 2953 2954 rv = 0; 2955 goto done; 2956 2957 fail: 2958 free(hotspare, M_DEVBUF, sizeof(*hotspare)); 2959 2960 done: 2961 if (sd) 2962 free(sd->sd_vol.sv_chunks, M_DEVBUF, 2963 sizeof(sd->sd_vol.sv_chunks)); 2964 free(sd, M_DEVBUF, sizeof(*sd)); 2965 free(sm, M_DEVBUF, sizeof(*sm)); 2966 if (open) { 2967 VOP_CLOSE(vn, FREAD | FWRITE, NOCRED, curproc); 2968 vput(vn); 2969 } 2970 2971 return (rv); 2972 } 2973 2974 void 2975 sr_hotspare_rebuild_callback(void *xsd) 2976 { 2977 struct sr_discipline *sd = xsd; 2978 sr_hotspare_rebuild(sd); 2979 } 2980 2981 void 2982 sr_hotspare_rebuild(struct sr_discipline *sd) 2983 { 2984 struct sr_softc *sc = sd->sd_sc; 2985 struct sr_chunk_head *cl; 2986 struct sr_chunk *hotspare, *chunk = NULL; 2987 struct sr_workunit *wu; 2988 struct sr_ccb *ccb; 2989 int i, s, cid, busy; 2990 2991 /* 2992 * Attempt to locate a hotspare and initiate rebuild. 2993 */ 2994 2995 /* Find first offline chunk. */ 2996 for (cid = 0; cid < sd->sd_meta->ssdi.ssd_chunk_no; cid++) { 2997 if (sd->sd_vol.sv_chunks[cid]->src_meta.scm_status == 2998 BIOC_SDOFFLINE) { 2999 chunk = sd->sd_vol.sv_chunks[cid]; 3000 break; 3001 } 3002 } 3003 if (chunk == NULL) { 3004 printf("%s: no offline chunk found on %s!\n", 3005 DEVNAME(sc), sd->sd_meta->ssd_devname); 3006 return; 3007 } 3008 3009 /* See if we have a suitable hotspare... */ 3010 rw_enter_write(&sc->sc_hs_lock); 3011 cl = &sc->sc_hotspare_list; 3012 SLIST_FOREACH(hotspare, cl, src_link) 3013 if (hotspare->src_size >= chunk->src_size && 3014 hotspare->src_secsize <= sd->sd_meta->ssdi.ssd_secsize) 3015 break; 3016 3017 if (hotspare != NULL) { 3018 3019 printf("%s: %s volume degraded, will attempt to " 3020 "rebuild on hotspare %s\n", DEVNAME(sc), 3021 sd->sd_meta->ssd_devname, hotspare->src_devname); 3022 3023 /* 3024 * Ensure that all pending I/O completes on the failed chunk 3025 * before trying to initiate a rebuild. 3026 */ 3027 i = 0; 3028 do { 3029 busy = 0; 3030 3031 s = splbio(); 3032 TAILQ_FOREACH(wu, &sd->sd_wu_pendq, swu_link) { 3033 TAILQ_FOREACH(ccb, &wu->swu_ccb, ccb_link) { 3034 if (ccb->ccb_target == cid) 3035 busy = 1; 3036 } 3037 } 3038 TAILQ_FOREACH(wu, &sd->sd_wu_defq, swu_link) { 3039 TAILQ_FOREACH(ccb, &wu->swu_ccb, ccb_link) { 3040 if (ccb->ccb_target == cid) 3041 busy = 1; 3042 } 3043 } 3044 splx(s); 3045 3046 if (busy) { 3047 tsleep_nsec(sd, PRIBIO, "sr_hotspare", 3048 SEC_TO_NSEC(1)); 3049 i++; 3050 } 3051 3052 } while (busy && i < 120); 3053 3054 DNPRINTF(SR_D_META, "%s: waited %i seconds for I/O to " 3055 "complete on failed chunk %s\n", DEVNAME(sc), 3056 i, chunk->src_devname); 3057 3058 if (busy) { 3059 printf("%s: pending I/O failed to complete on " 3060 "failed chunk %s, hotspare rebuild aborted...\n", 3061 DEVNAME(sc), chunk->src_devname); 3062 goto done; 3063 } 3064 3065 s = splbio(); 3066 rw_enter_write(&sc->sc_lock); 3067 bio_status_init(&sc->sc_status, &sc->sc_dev); 3068 if (sr_rebuild_init(sd, hotspare->src_dev_mm, 1) == 0) { 3069 3070 /* Remove hotspare from available list. */ 3071 sc->sc_hotspare_no--; 3072 SLIST_REMOVE(cl, hotspare, sr_chunk, src_link); 3073 free(hotspare, M_DEVBUF, sizeof(*hotspare)); 3074 3075 } 3076 rw_exit_write(&sc->sc_lock); 3077 splx(s); 3078 } 3079 done: 3080 rw_exit_write(&sc->sc_hs_lock); 3081 } 3082 3083 int 3084 sr_rebuild_init(struct sr_discipline *sd, dev_t dev, int hotspare) 3085 { 3086 struct sr_softc *sc = sd->sd_sc; 3087 struct sr_chunk *chunk = NULL; 3088 struct sr_meta_chunk *meta; 3089 struct disklabel label; 3090 struct vnode *vn; 3091 u_int64_t size; 3092 int64_t csize; 3093 char devname[32]; 3094 int rv = EINVAL, open = 0; 3095 int cid, i, part, status; 3096 3097 /* 3098 * Attempt to initiate a rebuild onto the specified device. 3099 */ 3100 3101 if (!(sd->sd_capabilities & SR_CAP_REBUILD)) { 3102 sr_error(sc, "discipline does not support rebuild"); 3103 goto done; 3104 } 3105 3106 /* make sure volume is in the right state */ 3107 if (sd->sd_vol_status == BIOC_SVREBUILD) { 3108 sr_error(sc, "rebuild already in progress"); 3109 goto done; 3110 } 3111 if (sd->sd_vol_status != BIOC_SVDEGRADED) { 3112 sr_error(sc, "volume not degraded"); 3113 goto done; 3114 } 3115 3116 /* Find first offline chunk. */ 3117 for (cid = 0; cid < sd->sd_meta->ssdi.ssd_chunk_no; cid++) { 3118 if (sd->sd_vol.sv_chunks[cid]->src_meta.scm_status == 3119 BIOC_SDOFFLINE) { 3120 chunk = sd->sd_vol.sv_chunks[cid]; 3121 break; 3122 } 3123 } 3124 if (chunk == NULL) { 3125 sr_error(sc, "no offline chunks available to rebuild"); 3126 goto done; 3127 } 3128 3129 /* Get coerced size from another online chunk. */ 3130 csize = 0; 3131 for (i = 0; i < sd->sd_meta->ssdi.ssd_chunk_no; i++) { 3132 if (sd->sd_vol.sv_chunks[i]->src_meta.scm_status == 3133 BIOC_SDONLINE) { 3134 meta = &sd->sd_vol.sv_chunks[i]->src_meta; 3135 csize = meta->scmi.scm_coerced_size; 3136 break; 3137 } 3138 } 3139 if (csize == 0) { 3140 sr_error(sc, "no online chunks available for rebuild"); 3141 goto done; 3142 } 3143 3144 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 3145 if (bdevvp(dev, &vn)) { 3146 printf("%s: sr_rebuild_init: can't allocate vnode\n", 3147 DEVNAME(sc)); 3148 goto done; 3149 } 3150 if (VOP_OPEN(vn, FREAD | FWRITE, NOCRED, curproc)) { 3151 DNPRINTF(SR_D_META,"%s: sr_ioctl_setstate can't " 3152 "open %s\n", DEVNAME(sc), devname); 3153 vput(vn); 3154 goto done; 3155 } 3156 open = 1; /* close dev on error */ 3157 3158 /* Get disklabel and check partition. */ 3159 part = DISKPART(dev); 3160 if (VOP_IOCTL(vn, DIOCGDINFO, (caddr_t)&label, FREAD, 3161 NOCRED, curproc)) { 3162 DNPRINTF(SR_D_META, "%s: sr_ioctl_setstate ioctl failed\n", 3163 DEVNAME(sc)); 3164 goto done; 3165 } 3166 if (label.d_partitions[part].p_fstype != FS_RAID) { 3167 sr_error(sc, "%s partition not of type RAID (%d)", 3168 devname, label.d_partitions[part].p_fstype); 3169 goto done; 3170 } 3171 3172 /* Is the partition large enough? */ 3173 size = DL_SECTOBLK(&label, DL_GETPSIZE(&label.d_partitions[part])); 3174 if (size <= sd->sd_meta->ssd_data_blkno) { 3175 sr_error(sc, "%s: %s partition too small", DEVNAME(sc), 3176 devname); 3177 goto done; 3178 } 3179 size -= sd->sd_meta->ssd_data_blkno; 3180 if (size > INT64_MAX) { 3181 sr_error(sc, "%s: %s partition too large", DEVNAME(sc), 3182 devname); 3183 goto done; 3184 } 3185 if (size < csize) { 3186 sr_error(sc, "%s partition too small, at least %lld bytes " 3187 "required", devname, (long long)(csize << DEV_BSHIFT)); 3188 goto done; 3189 } else if (size > csize) 3190 sr_warn(sc, "%s partition too large, wasting %lld bytes", 3191 devname, (long long)((size - csize) << DEV_BSHIFT)); 3192 if (label.d_secsize > sd->sd_meta->ssdi.ssd_secsize) { 3193 sr_error(sc, "%s sector size too large, <= %u bytes " 3194 "required", devname, sd->sd_meta->ssdi.ssd_secsize); 3195 goto done; 3196 } 3197 3198 /* Ensure that this chunk is not already in use. */ 3199 status = sr_chunk_in_use(sc, dev); 3200 if (status != BIOC_SDINVALID && status != BIOC_SDOFFLINE && 3201 !(hotspare && status == BIOC_SDHOTSPARE)) { 3202 sr_error(sc, "%s is already in use", devname); 3203 goto done; 3204 } 3205 3206 /* Reset rebuild counter since we rebuilding onto a new chunk. */ 3207 sd->sd_meta->ssd_rebuild = 0; 3208 3209 open = 0; /* leave dev open from here on out */ 3210 3211 /* Fix up chunk. */ 3212 memcpy(chunk->src_duid, label.d_uid, sizeof(chunk->src_duid)); 3213 chunk->src_dev_mm = dev; 3214 chunk->src_vn = vn; 3215 3216 /* Reconstruct metadata. */ 3217 meta = &chunk->src_meta; 3218 meta->scmi.scm_volid = sd->sd_meta->ssdi.ssd_volid; 3219 meta->scmi.scm_chunk_id = cid; 3220 strlcpy(meta->scmi.scm_devname, devname, 3221 sizeof(meta->scmi.scm_devname)); 3222 meta->scmi.scm_size = size; 3223 meta->scmi.scm_coerced_size = csize; 3224 memcpy(&meta->scmi.scm_uuid, &sd->sd_meta->ssdi.ssd_uuid, 3225 sizeof(meta->scmi.scm_uuid)); 3226 sr_checksum(sc, meta, &meta->scm_checksum, 3227 sizeof(struct sr_meta_chunk_invariant)); 3228 3229 sd->sd_set_chunk_state(sd, cid, BIOC_SDREBUILD); 3230 3231 if (sr_meta_save(sd, SR_META_DIRTY)) { 3232 sr_error(sc, "could not save metadata to %s", devname); 3233 open = 1; 3234 goto done; 3235 } 3236 3237 sr_warn(sc, "rebuild of %s started on %s", 3238 sd->sd_meta->ssd_devname, devname); 3239 3240 sd->sd_reb_abort = 0; 3241 kthread_create_deferred(sr_rebuild_start, sd); 3242 3243 rv = 0; 3244 done: 3245 if (open) { 3246 VOP_CLOSE(vn, FREAD | FWRITE, NOCRED, curproc); 3247 vput(vn); 3248 } 3249 3250 return (rv); 3251 } 3252 3253 int 3254 sr_rebuild_percent(struct sr_discipline *sd) 3255 { 3256 daddr_t rb, sz; 3257 3258 sz = sd->sd_meta->ssdi.ssd_size; 3259 rb = sd->sd_meta->ssd_rebuild; 3260 3261 if (rb > 0) 3262 return (100 - ((sz * 100 - rb * 100) / sz) - 1); 3263 3264 return (0); 3265 } 3266 3267 void 3268 sr_roam_chunks(struct sr_discipline *sd) 3269 { 3270 struct sr_softc *sc = sd->sd_sc; 3271 struct sr_chunk *chunk; 3272 struct sr_meta_chunk *meta; 3273 int roamed = 0; 3274 3275 /* Have any chunks roamed? */ 3276 SLIST_FOREACH(chunk, &sd->sd_vol.sv_chunk_list, src_link) { 3277 meta = &chunk->src_meta; 3278 if (strncmp(meta->scmi.scm_devname, chunk->src_devname, 3279 sizeof(meta->scmi.scm_devname))) { 3280 3281 printf("%s: roaming device %s -> %s\n", DEVNAME(sc), 3282 meta->scmi.scm_devname, chunk->src_devname); 3283 3284 strlcpy(meta->scmi.scm_devname, chunk->src_devname, 3285 sizeof(meta->scmi.scm_devname)); 3286 3287 roamed++; 3288 } 3289 } 3290 3291 if (roamed) 3292 sr_meta_save(sd, SR_META_DIRTY); 3293 } 3294 3295 int 3296 sr_ioctl_createraid(struct sr_softc *sc, struct bioc_createraid *bc, 3297 int user, void *data) 3298 { 3299 struct sr_meta_opt_item *omi; 3300 struct sr_chunk_head *cl; 3301 struct sr_discipline *sd = NULL; 3302 struct sr_chunk *ch_entry; 3303 struct scsi_link *link; 3304 struct device *dev; 3305 char *uuid, devname[32]; 3306 dev_t *dt = NULL; 3307 int i, no_chunk, rv = EINVAL, target, vol; 3308 int no_meta; 3309 3310 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_createraid(%d)\n", 3311 DEVNAME(sc), user); 3312 3313 /* user input */ 3314 if (bc->bc_dev_list_len > BIOC_CRMAXLEN) 3315 goto unwind; 3316 3317 dt = malloc(bc->bc_dev_list_len, M_DEVBUF, M_WAITOK | M_ZERO); 3318 if (user) { 3319 if (copyin(bc->bc_dev_list, dt, bc->bc_dev_list_len) != 0) 3320 goto unwind; 3321 } else 3322 memcpy(dt, bc->bc_dev_list, bc->bc_dev_list_len); 3323 3324 /* Initialise discipline. */ 3325 sd = malloc(sizeof(struct sr_discipline), M_DEVBUF, M_WAITOK | M_ZERO); 3326 sd->sd_sc = sc; 3327 SLIST_INIT(&sd->sd_meta_opt); 3328 sd->sd_taskq = taskq_create("srdis", 1, IPL_BIO, 0); 3329 if (sd->sd_taskq == NULL) { 3330 sr_error(sc, "could not create discipline taskq"); 3331 goto unwind; 3332 } 3333 if (sr_discipline_init(sd, bc->bc_level)) { 3334 sr_error(sc, "could not initialize discipline"); 3335 goto unwind; 3336 } 3337 3338 no_chunk = bc->bc_dev_list_len / sizeof(dev_t); 3339 cl = &sd->sd_vol.sv_chunk_list; 3340 SLIST_INIT(cl); 3341 3342 /* Ensure that chunks are not already in use. */ 3343 for (i = 0; i < no_chunk; i++) { 3344 if (sr_chunk_in_use(sc, dt[i]) != BIOC_SDINVALID) { 3345 sr_meta_getdevname(sc, dt[i], devname, sizeof(devname)); 3346 sr_error(sc, "chunk %s already in use", devname); 3347 goto unwind; 3348 } 3349 } 3350 3351 sd->sd_meta_type = sr_meta_probe(sd, dt, no_chunk); 3352 if (sd->sd_meta_type == SR_META_F_INVALID) { 3353 sr_error(sc, "invalid metadata format"); 3354 goto unwind; 3355 } 3356 3357 if (sr_meta_attach(sd, no_chunk, bc->bc_flags & BIOC_SCFORCE)) 3358 goto unwind; 3359 3360 /* force the raid volume by clearing metadata region */ 3361 if (bc->bc_flags & BIOC_SCFORCE) { 3362 /* make sure disk isn't up and running */ 3363 if (sr_meta_read(sd)) 3364 if (sr_already_assembled(sd)) { 3365 uuid = sr_uuid_format( 3366 &sd->sd_meta->ssdi.ssd_uuid); 3367 sr_error(sc, "disk %s is currently in use; " 3368 "cannot force create", uuid); 3369 free(uuid, M_DEVBUF, 37); 3370 goto unwind; 3371 } 3372 3373 if (sr_meta_clear(sd)) { 3374 sr_error(sc, "failed to clear metadata"); 3375 goto unwind; 3376 } 3377 } 3378 3379 no_meta = sr_meta_read(sd); 3380 if (no_meta == -1) { 3381 3382 /* Corrupt metadata on one or more chunks. */ 3383 sr_error(sc, "one of the chunks has corrupt metadata; " 3384 "aborting assembly"); 3385 goto unwind; 3386 3387 } else if (no_meta == 0) { 3388 3389 /* Initialise volume and chunk metadata. */ 3390 sr_meta_init(sd, bc->bc_level, no_chunk); 3391 sd->sd_vol_status = BIOC_SVONLINE; 3392 sd->sd_meta_flags = bc->bc_flags & BIOC_SCNOAUTOASSEMBLE; 3393 if (sd->sd_create) { 3394 if ((i = sd->sd_create(sd, bc, no_chunk, 3395 sd->sd_vol.sv_chunk_minsz))) { 3396 rv = i; 3397 goto unwind; 3398 } 3399 } 3400 sr_meta_init_complete(sd); 3401 3402 DNPRINTF(SR_D_IOCTL, 3403 "%s: sr_ioctl_createraid: vol_size: %lld\n", 3404 DEVNAME(sc), sd->sd_meta->ssdi.ssd_size); 3405 3406 /* Warn if we've wasted chunk space due to coercing. */ 3407 if ((sd->sd_capabilities & SR_CAP_NON_COERCED) == 0 && 3408 sd->sd_vol.sv_chunk_minsz != sd->sd_vol.sv_chunk_maxsz) 3409 sr_warn(sc, "chunk sizes are not equal; up to %llu " 3410 "blocks wasted per chunk", 3411 sd->sd_vol.sv_chunk_maxsz - 3412 sd->sd_vol.sv_chunk_minsz); 3413 3414 } else { 3415 3416 /* Ensure we are assembling the correct # of chunks. */ 3417 if (sd->sd_meta->ssdi.ssd_chunk_no != no_chunk) { 3418 sr_error(sc, "volume chunk count does not match metadata " 3419 "chunk count"); 3420 goto unwind; 3421 } 3422 3423 /* Ensure metadata level matches requested assembly level. */ 3424 if (sd->sd_meta->ssdi.ssd_level != bc->bc_level) { 3425 sr_error(sc, "volume level does not match metadata " 3426 "level"); 3427 goto unwind; 3428 } 3429 3430 if (sr_already_assembled(sd)) { 3431 uuid = sr_uuid_format(&sd->sd_meta->ssdi.ssd_uuid); 3432 sr_error(sc, "disk %s already assembled", uuid); 3433 free(uuid, M_DEVBUF, 37); 3434 goto unwind; 3435 } 3436 3437 if (user == 0 && sd->sd_meta_flags & BIOC_SCNOAUTOASSEMBLE) { 3438 DNPRINTF(SR_D_META, "%s: disk not auto assembled from " 3439 "metadata\n", DEVNAME(sc)); 3440 goto unwind; 3441 } 3442 3443 if (no_meta != no_chunk) 3444 sr_warn(sc, "trying to bring up %s degraded", 3445 sd->sd_meta->ssd_devname); 3446 3447 if (sd->sd_meta->ssd_meta_flags & SR_META_DIRTY) 3448 sr_warn(sc, "%s was not shutdown properly", 3449 sd->sd_meta->ssd_devname); 3450 3451 SLIST_FOREACH(omi, &sd->sd_meta_opt, omi_link) 3452 if (sd->sd_meta_opt_handler == NULL || 3453 sd->sd_meta_opt_handler(sd, omi->omi_som) != 0) 3454 sr_meta_opt_handler(sd, omi->omi_som); 3455 3456 if (sd->sd_assemble) { 3457 if ((i = sd->sd_assemble(sd, bc, no_chunk, data))) { 3458 rv = i; 3459 goto unwind; 3460 } 3461 } 3462 3463 DNPRINTF(SR_D_META, "%s: disk assembled from metadata\n", 3464 DEVNAME(sc)); 3465 3466 } 3467 3468 /* Metadata MUST be fully populated by this point. */ 3469 TAILQ_INSERT_TAIL(&sc->sc_dis_list, sd, sd_link); 3470 3471 /* Allocate all resources. */ 3472 if ((rv = sd->sd_alloc_resources(sd))) 3473 goto unwind; 3474 3475 /* Adjust flags if necessary. */ 3476 if ((sd->sd_capabilities & SR_CAP_AUTO_ASSEMBLE) && 3477 (bc->bc_flags & BIOC_SCNOAUTOASSEMBLE) != 3478 (sd->sd_meta->ssdi.ssd_vol_flags & BIOC_SCNOAUTOASSEMBLE)) { 3479 sd->sd_meta->ssdi.ssd_vol_flags &= ~BIOC_SCNOAUTOASSEMBLE; 3480 sd->sd_meta->ssdi.ssd_vol_flags |= 3481 bc->bc_flags & BIOC_SCNOAUTOASSEMBLE; 3482 } 3483 3484 if (sd->sd_capabilities & SR_CAP_SYSTEM_DISK) { 3485 /* Initialise volume state. */ 3486 sd->sd_set_vol_state(sd); 3487 if (sd->sd_vol_status == BIOC_SVOFFLINE) { 3488 sr_error(sc, "%s is offline, will not be brought " 3489 "online", sd->sd_meta->ssd_devname); 3490 goto unwind; 3491 } 3492 3493 /* Setup SCSI iopool. */ 3494 scsi_iopool_init(&sd->sd_iopool, sd, sr_wu_get, sr_wu_put); 3495 3496 /* 3497 * All checks passed - return ENXIO if volume cannot be created. 3498 */ 3499 rv = ENXIO; 3500 3501 /* 3502 * Find a free target. 3503 * 3504 * XXX: We reserve sd_target == 0 to indicate the 3505 * discipline is not linked into sc->sc_targets, so begin 3506 * the search with target = 1. 3507 */ 3508 for (target = 1; target < SR_MAX_LD; target++) 3509 if (sc->sc_targets[target] == NULL) 3510 break; 3511 if (target == SR_MAX_LD) { 3512 sr_error(sc, "no free target for %s", 3513 sd->sd_meta->ssd_devname); 3514 goto unwind; 3515 } 3516 3517 /* Clear sense data. */ 3518 bzero(&sd->sd_scsi_sense, sizeof(sd->sd_scsi_sense)); 3519 3520 /* Attach discipline and get midlayer to probe it. */ 3521 sd->sd_target = target; 3522 sc->sc_targets[target] = sd; 3523 if (scsi_probe_lun(sc->sc_scsibus, target, 0) != 0) { 3524 sr_error(sc, "scsi_probe_lun failed"); 3525 sc->sc_targets[target] = NULL; 3526 sd->sd_target = 0; 3527 goto unwind; 3528 } 3529 3530 link = scsi_get_link(sc->sc_scsibus, target, 0); 3531 if (link == NULL) 3532 goto unwind; 3533 3534 dev = link->device_softc; 3535 DNPRINTF(SR_D_IOCTL, "%s: sr device added: %s at target %d\n", 3536 DEVNAME(sc), dev->dv_xname, sd->sd_target); 3537 3538 /* XXX - Count volumes, not targets. */ 3539 for (i = 0, vol = -1; i <= sd->sd_target; i++) 3540 if (sc->sc_targets[i]) 3541 vol++; 3542 3543 rv = 0; 3544 3545 if (sd->sd_meta->ssd_devname[0] != '\0' && 3546 strncmp(sd->sd_meta->ssd_devname, dev->dv_xname, 3547 sizeof(dev->dv_xname))) 3548 sr_warn(sc, "volume %s is roaming, it used to be %s, " 3549 "updating metadata", dev->dv_xname, 3550 sd->sd_meta->ssd_devname); 3551 3552 /* Populate remaining volume metadata. */ 3553 sd->sd_meta->ssdi.ssd_volid = vol; 3554 strlcpy(sd->sd_meta->ssd_devname, dev->dv_xname, 3555 sizeof(sd->sd_meta->ssd_devname)); 3556 3557 sr_info(sc, "%s volume attached as %s", 3558 sd->sd_name, sd->sd_meta->ssd_devname); 3559 3560 /* Update device name on any roaming chunks. */ 3561 sr_roam_chunks(sd); 3562 3563 #ifndef SMALL_KERNEL 3564 if (sr_sensors_create(sd)) 3565 sr_warn(sc, "unable to create sensor for %s", 3566 dev->dv_xname); 3567 #endif /* SMALL_KERNEL */ 3568 } else { 3569 /* This volume does not attach as a system disk. */ 3570 ch_entry = SLIST_FIRST(cl); /* XXX */ 3571 strlcpy(sd->sd_meta->ssd_devname, ch_entry->src_devname, 3572 sizeof(sd->sd_meta->ssd_devname)); 3573 3574 if (sd->sd_start_discipline(sd)) 3575 goto unwind; 3576 } 3577 3578 /* Save current metadata to disk. */ 3579 rv = sr_meta_save(sd, SR_META_DIRTY); 3580 3581 if (sd->sd_vol_status == BIOC_SVREBUILD) 3582 kthread_create_deferred(sr_rebuild_start, sd); 3583 3584 sd->sd_ready = 1; 3585 3586 free(dt, M_DEVBUF, bc->bc_dev_list_len); 3587 3588 return (rv); 3589 3590 unwind: 3591 free(dt, M_DEVBUF, bc->bc_dev_list_len); 3592 3593 sr_discipline_shutdown(sd, 0, 0); 3594 3595 if (rv == EAGAIN) 3596 rv = 0; 3597 3598 return (rv); 3599 } 3600 3601 int 3602 sr_ioctl_deleteraid(struct sr_softc *sc, struct sr_discipline *sd, 3603 struct bioc_deleteraid *bd) 3604 { 3605 int rv = 1; 3606 3607 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_deleteraid %s\n", 3608 DEVNAME(sc), bd->bd_dev); 3609 3610 if (sd == NULL) { 3611 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 3612 if (!strncmp(sd->sd_meta->ssd_devname, bd->bd_dev, 3613 sizeof(sd->sd_meta->ssd_devname))) 3614 break; 3615 } 3616 if (sd == NULL) { 3617 sr_error(sc, "volume %s not found", bd->bd_dev); 3618 goto bad; 3619 } 3620 } 3621 3622 sd->sd_deleted = 1; 3623 sd->sd_meta->ssdi.ssd_vol_flags = BIOC_SCNOAUTOASSEMBLE; 3624 sr_discipline_shutdown(sd, 1, 0); 3625 3626 rv = 0; 3627 bad: 3628 return (rv); 3629 } 3630 3631 int 3632 sr_ioctl_discipline(struct sr_softc *sc, struct sr_discipline *sd, 3633 struct bioc_discipline *bd) 3634 { 3635 int rv = 1; 3636 3637 /* Dispatch a discipline specific ioctl. */ 3638 3639 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_discipline %s\n", DEVNAME(sc), 3640 bd->bd_dev); 3641 3642 if (sd == NULL) { 3643 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 3644 if (!strncmp(sd->sd_meta->ssd_devname, bd->bd_dev, 3645 sizeof(sd->sd_meta->ssd_devname))) 3646 break; 3647 } 3648 if (sd == NULL) { 3649 sr_error(sc, "volume %s not found", bd->bd_dev); 3650 goto bad; 3651 } 3652 } 3653 3654 if (sd->sd_ioctl_handler) 3655 rv = sd->sd_ioctl_handler(sd, bd); 3656 3657 bad: 3658 return (rv); 3659 } 3660 3661 int 3662 sr_ioctl_installboot(struct sr_softc *sc, struct sr_discipline *sd, 3663 struct bioc_installboot *bb) 3664 { 3665 void *bootblk = NULL, *bootldr = NULL; 3666 struct sr_chunk *chunk; 3667 struct sr_meta_opt_item *omi; 3668 struct sr_meta_boot *sbm; 3669 struct disk *dk; 3670 u_int32_t bbs = 0, bls = 0, secsize; 3671 u_char duid[8]; 3672 int rv = EINVAL; 3673 int i; 3674 3675 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_installboot %s\n", DEVNAME(sc), 3676 bb->bb_dev); 3677 3678 if (sd == NULL) { 3679 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 3680 if (!strncmp(sd->sd_meta->ssd_devname, bb->bb_dev, 3681 sizeof(sd->sd_meta->ssd_devname))) 3682 break; 3683 } 3684 if (sd == NULL) { 3685 sr_error(sc, "volume %s not found", bb->bb_dev); 3686 goto done; 3687 } 3688 } 3689 3690 bzero(duid, sizeof(duid)); 3691 TAILQ_FOREACH(dk, &disklist, dk_link) 3692 if (!strncmp(dk->dk_name, bb->bb_dev, sizeof(bb->bb_dev))) 3693 break; 3694 if (dk == NULL || dk->dk_label == NULL || 3695 (dk->dk_flags & DKF_LABELVALID) == 0 || 3696 bcmp(dk->dk_label->d_uid, &duid, sizeof(duid)) == 0) { 3697 sr_error(sc, "failed to get DUID for softraid volume"); 3698 goto done; 3699 } 3700 memcpy(duid, dk->dk_label->d_uid, sizeof(duid)); 3701 3702 /* Ensure that boot storage area is large enough. */ 3703 if (sd->sd_meta->ssd_data_blkno < (SR_BOOT_OFFSET + SR_BOOT_SIZE)) { 3704 sr_error(sc, "insufficient boot storage"); 3705 goto done; 3706 } 3707 3708 if (bb->bb_bootblk_size > SR_BOOT_BLOCKS_SIZE * DEV_BSIZE) { 3709 sr_error(sc, "boot block too large (%d > %d)", 3710 bb->bb_bootblk_size, SR_BOOT_BLOCKS_SIZE * DEV_BSIZE); 3711 goto done; 3712 } 3713 3714 if (bb->bb_bootldr_size > SR_BOOT_LOADER_SIZE * DEV_BSIZE) { 3715 sr_error(sc, "boot loader too large (%d > %d)", 3716 bb->bb_bootldr_size, SR_BOOT_LOADER_SIZE * DEV_BSIZE); 3717 goto done; 3718 } 3719 3720 secsize = sd->sd_meta->ssdi.ssd_secsize; 3721 3722 /* Copy in boot block. */ 3723 bbs = howmany(bb->bb_bootblk_size, secsize) * secsize; 3724 bootblk = malloc(bbs, M_DEVBUF, M_WAITOK | M_ZERO); 3725 if (copyin(bb->bb_bootblk, bootblk, bb->bb_bootblk_size) != 0) 3726 goto done; 3727 3728 /* Copy in boot loader. */ 3729 bls = howmany(bb->bb_bootldr_size, secsize) * secsize; 3730 bootldr = malloc(bls, M_DEVBUF, M_WAITOK | M_ZERO); 3731 if (copyin(bb->bb_bootldr, bootldr, bb->bb_bootldr_size) != 0) 3732 goto done; 3733 3734 /* Create or update optional meta for bootable volumes. */ 3735 SLIST_FOREACH(omi, &sd->sd_meta_opt, omi_link) 3736 if (omi->omi_som->som_type == SR_OPT_BOOT) 3737 break; 3738 if (omi == NULL) { 3739 omi = malloc(sizeof(struct sr_meta_opt_item), M_DEVBUF, 3740 M_WAITOK | M_ZERO); 3741 omi->omi_som = malloc(sizeof(struct sr_meta_boot), M_DEVBUF, 3742 M_WAITOK | M_ZERO); 3743 omi->omi_som->som_type = SR_OPT_BOOT; 3744 omi->omi_som->som_length = sizeof(struct sr_meta_boot); 3745 SLIST_INSERT_HEAD(&sd->sd_meta_opt, omi, omi_link); 3746 sd->sd_meta->ssdi.ssd_opt_no++; 3747 } 3748 sbm = (struct sr_meta_boot *)omi->omi_som; 3749 3750 memcpy(sbm->sbm_root_duid, duid, sizeof(sbm->sbm_root_duid)); 3751 bzero(&sbm->sbm_boot_duid, sizeof(sbm->sbm_boot_duid)); 3752 sbm->sbm_bootblk_size = bbs; 3753 sbm->sbm_bootldr_size = bls; 3754 3755 DNPRINTF(SR_D_IOCTL, "sr_ioctl_installboot: root duid is %s\n", 3756 duid_format(sbm->sbm_root_duid)); 3757 3758 /* Save boot block and boot loader to each chunk. */ 3759 for (i = 0; i < sd->sd_meta->ssdi.ssd_chunk_no; i++) { 3760 3761 chunk = sd->sd_vol.sv_chunks[i]; 3762 if (chunk->src_meta.scm_status != BIOC_SDONLINE && 3763 chunk->src_meta.scm_status != BIOC_SDREBUILD) 3764 continue; 3765 3766 if (i < SR_MAX_BOOT_DISKS) 3767 memcpy(&sbm->sbm_boot_duid[i], chunk->src_duid, 3768 sizeof(sbm->sbm_boot_duid[i])); 3769 3770 /* Save boot blocks. */ 3771 DNPRINTF(SR_D_IOCTL, 3772 "sr_ioctl_installboot: saving boot block to %s " 3773 "(%u bytes)\n", chunk->src_devname, bbs); 3774 3775 if (sr_rw(sc, chunk->src_dev_mm, bootblk, bbs, 3776 SR_BOOT_BLOCKS_OFFSET, B_WRITE)) { 3777 sr_error(sc, "failed to write boot block"); 3778 goto done; 3779 } 3780 3781 /* Save boot loader.*/ 3782 DNPRINTF(SR_D_IOCTL, 3783 "sr_ioctl_installboot: saving boot loader to %s " 3784 "(%u bytes)\n", chunk->src_devname, bls); 3785 3786 if (sr_rw(sc, chunk->src_dev_mm, bootldr, bls, 3787 SR_BOOT_LOADER_OFFSET, B_WRITE)) { 3788 sr_error(sc, "failed to write boot loader"); 3789 goto done; 3790 } 3791 } 3792 3793 /* XXX - Install boot block on disk - MD code. */ 3794 3795 /* Mark volume as bootable and save metadata. */ 3796 sd->sd_meta->ssdi.ssd_vol_flags |= BIOC_SCBOOTABLE; 3797 if (sr_meta_save(sd, SR_META_DIRTY)) { 3798 sr_error(sc, "could not save metadata to %s", DEVNAME(sc)); 3799 goto done; 3800 } 3801 3802 rv = 0; 3803 3804 done: 3805 free(bootblk, M_DEVBUF, bbs); 3806 free(bootldr, M_DEVBUF, bls); 3807 3808 return (rv); 3809 } 3810 3811 void 3812 sr_chunks_unwind(struct sr_softc *sc, struct sr_chunk_head *cl) 3813 { 3814 struct sr_chunk *ch_entry, *ch_next; 3815 3816 DNPRINTF(SR_D_IOCTL, "%s: sr_chunks_unwind\n", DEVNAME(sc)); 3817 3818 if (!cl) 3819 return; 3820 3821 for (ch_entry = SLIST_FIRST(cl); ch_entry != NULL; ch_entry = ch_next) { 3822 ch_next = SLIST_NEXT(ch_entry, src_link); 3823 3824 DNPRINTF(SR_D_IOCTL, "%s: sr_chunks_unwind closing: %s\n", 3825 DEVNAME(sc), ch_entry->src_devname); 3826 if (ch_entry->src_vn) { 3827 /* 3828 * XXX - explicitly lock the vnode until we can resolve 3829 * the problem introduced by vnode aliasing... specfs 3830 * has no locking, whereas ufs/ffs does! 3831 */ 3832 vn_lock(ch_entry->src_vn, LK_EXCLUSIVE | LK_RETRY); 3833 VOP_CLOSE(ch_entry->src_vn, FREAD | FWRITE, NOCRED, 3834 curproc); 3835 vput(ch_entry->src_vn); 3836 } 3837 free(ch_entry, M_DEVBUF, sizeof(*ch_entry)); 3838 } 3839 SLIST_INIT(cl); 3840 } 3841 3842 void 3843 sr_discipline_free(struct sr_discipline *sd) 3844 { 3845 struct sr_softc *sc; 3846 struct sr_discipline *sdtmp1; 3847 struct sr_meta_opt_head *som; 3848 struct sr_meta_opt_item *omi, *omi_next; 3849 3850 if (!sd) 3851 return; 3852 3853 sc = sd->sd_sc; 3854 3855 DNPRINTF(SR_D_DIS, "%s: sr_discipline_free %s\n", 3856 DEVNAME(sc), 3857 sd->sd_meta ? sd->sd_meta->ssd_devname : "nodev"); 3858 if (sd->sd_free_resources) 3859 sd->sd_free_resources(sd); 3860 free(sd->sd_vol.sv_chunks, M_DEVBUF, 0); 3861 free(sd->sd_meta, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 3862 free(sd->sd_meta_foreign, M_DEVBUF, smd[sd->sd_meta_type].smd_size); 3863 3864 som = &sd->sd_meta_opt; 3865 for (omi = SLIST_FIRST(som); omi != NULL; omi = omi_next) { 3866 omi_next = SLIST_NEXT(omi, omi_link); 3867 free(omi->omi_som, M_DEVBUF, 0); 3868 free(omi, M_DEVBUF, sizeof(*omi)); 3869 } 3870 3871 if (sd->sd_target != 0) { 3872 KASSERT(sc->sc_targets[sd->sd_target] == sd); 3873 sc->sc_targets[sd->sd_target] = NULL; 3874 } 3875 3876 TAILQ_FOREACH(sdtmp1, &sc->sc_dis_list, sd_link) { 3877 if (sdtmp1 == sd) 3878 break; 3879 } 3880 if (sdtmp1 != NULL) 3881 TAILQ_REMOVE(&sc->sc_dis_list, sd, sd_link); 3882 3883 explicit_bzero(sd, sizeof *sd); 3884 free(sd, M_DEVBUF, sizeof(*sd)); 3885 } 3886 3887 void 3888 sr_discipline_shutdown(struct sr_discipline *sd, int meta_save, int dying) 3889 { 3890 struct sr_softc *sc; 3891 int ret, s; 3892 3893 if (!sd) 3894 return; 3895 sc = sd->sd_sc; 3896 3897 DNPRINTF(SR_D_DIS, "%s: sr_discipline_shutdown %s\n", DEVNAME(sc), 3898 sd->sd_meta ? sd->sd_meta->ssd_devname : "nodev"); 3899 3900 /* If rebuilding, abort rebuild and drain I/O. */ 3901 if (sd->sd_reb_active) { 3902 sd->sd_reb_abort = 1; 3903 while (sd->sd_reb_active) 3904 tsleep(sd, PWAIT, "sr_shutdown", 1); 3905 } 3906 3907 if (meta_save) 3908 sr_meta_save(sd, 0); 3909 3910 s = splbio(); 3911 3912 sd->sd_ready = 0; 3913 3914 /* make sure there isn't a sync pending and yield */ 3915 wakeup(sd); 3916 while (sd->sd_sync || sd->sd_must_flush) { 3917 ret = tsleep_nsec(&sd->sd_sync, MAXPRI, "sr_down", 3918 SEC_TO_NSEC(60)); 3919 if (ret == EWOULDBLOCK) 3920 break; 3921 } 3922 if (dying == -1) { 3923 sd->sd_ready = 1; 3924 splx(s); 3925 return; 3926 } 3927 3928 #ifndef SMALL_KERNEL 3929 sr_sensors_delete(sd); 3930 #endif /* SMALL_KERNEL */ 3931 3932 if (sd->sd_target != 0) 3933 scsi_detach_lun(sc->sc_scsibus, sd->sd_target, 0, 3934 dying ? 0 : DETACH_FORCE); 3935 3936 sr_chunks_unwind(sc, &sd->sd_vol.sv_chunk_list); 3937 3938 if (sd->sd_taskq) 3939 taskq_destroy(sd->sd_taskq); 3940 3941 sr_discipline_free(sd); 3942 3943 splx(s); 3944 } 3945 3946 int 3947 sr_discipline_init(struct sr_discipline *sd, int level) 3948 { 3949 int rv = 1; 3950 3951 /* Initialise discipline function pointers with defaults. */ 3952 sd->sd_alloc_resources = sr_alloc_resources; 3953 sd->sd_assemble = NULL; 3954 sd->sd_create = NULL; 3955 sd->sd_free_resources = sr_free_resources; 3956 sd->sd_ioctl_handler = NULL; 3957 sd->sd_openings = NULL; 3958 sd->sd_meta_opt_handler = NULL; 3959 sd->sd_rebuild = sr_rebuild; 3960 sd->sd_scsi_inquiry = sr_raid_inquiry; 3961 sd->sd_scsi_read_cap = sr_raid_read_cap; 3962 sd->sd_scsi_tur = sr_raid_tur; 3963 sd->sd_scsi_req_sense = sr_raid_request_sense; 3964 sd->sd_scsi_start_stop = sr_raid_start_stop; 3965 sd->sd_scsi_sync = sr_raid_sync; 3966 sd->sd_scsi_rw = NULL; 3967 sd->sd_scsi_intr = sr_raid_intr; 3968 sd->sd_scsi_wu_done = NULL; 3969 sd->sd_scsi_done = NULL; 3970 sd->sd_set_chunk_state = sr_set_chunk_state; 3971 sd->sd_set_vol_state = sr_set_vol_state; 3972 sd->sd_start_discipline = NULL; 3973 3974 task_set(&sd->sd_meta_save_task, sr_meta_save_callback, sd); 3975 task_set(&sd->sd_hotspare_rebuild_task, sr_hotspare_rebuild_callback, 3976 sd); 3977 3978 sd->sd_wu_size = sizeof(struct sr_workunit); 3979 switch (level) { 3980 case 0: 3981 sr_raid0_discipline_init(sd); 3982 break; 3983 case 1: 3984 sr_raid1_discipline_init(sd); 3985 break; 3986 case 5: 3987 sr_raid5_discipline_init(sd); 3988 break; 3989 case 6: 3990 sr_raid6_discipline_init(sd); 3991 break; 3992 #ifdef CRYPTO 3993 case 'C': 3994 sr_crypto_discipline_init(sd); 3995 break; 3996 #endif 3997 case 'c': 3998 sr_concat_discipline_init(sd); 3999 break; 4000 default: 4001 goto bad; 4002 } 4003 4004 rv = 0; 4005 bad: 4006 return (rv); 4007 } 4008 4009 int 4010 sr_raid_inquiry(struct sr_workunit *wu) 4011 { 4012 struct sr_discipline *sd = wu->swu_dis; 4013 struct scsi_xfer *xs = wu->swu_xs; 4014 struct scsi_inquiry *cdb = (struct scsi_inquiry *)xs->cmd; 4015 struct scsi_inquiry_data inq; 4016 4017 DNPRINTF(SR_D_DIS, "%s: sr_raid_inquiry\n", DEVNAME(sd->sd_sc)); 4018 4019 if (xs->cmdlen != sizeof(*cdb)) 4020 return (EINVAL); 4021 4022 if (ISSET(cdb->flags, SI_EVPD)) 4023 return (EOPNOTSUPP); 4024 4025 bzero(&inq, sizeof(inq)); 4026 inq.device = T_DIRECT; 4027 inq.dev_qual2 = 0; 4028 inq.version = 2; 4029 inq.response_format = 2; 4030 inq.additional_length = 32; 4031 inq.flags |= SID_CmdQue; 4032 strlcpy(inq.vendor, sd->sd_meta->ssdi.ssd_vendor, 4033 sizeof(inq.vendor)); 4034 strlcpy(inq.product, sd->sd_meta->ssdi.ssd_product, 4035 sizeof(inq.product)); 4036 strlcpy(inq.revision, sd->sd_meta->ssdi.ssd_revision, 4037 sizeof(inq.revision)); 4038 sr_copy_internal_data(xs, &inq, sizeof(inq)); 4039 4040 return (0); 4041 } 4042 4043 int 4044 sr_raid_read_cap(struct sr_workunit *wu) 4045 { 4046 struct sr_discipline *sd = wu->swu_dis; 4047 struct scsi_xfer *xs = wu->swu_xs; 4048 struct scsi_read_cap_data rcd; 4049 struct scsi_read_cap_data_16 rcd16; 4050 u_int64_t addr; 4051 int rv = 1; 4052 u_int32_t secsize; 4053 4054 DNPRINTF(SR_D_DIS, "%s: sr_raid_read_cap\n", DEVNAME(sd->sd_sc)); 4055 4056 secsize = sd->sd_meta->ssdi.ssd_secsize; 4057 4058 addr = ((sd->sd_meta->ssdi.ssd_size * DEV_BSIZE) / secsize) - 1; 4059 if (xs->cmd->opcode == READ_CAPACITY) { 4060 bzero(&rcd, sizeof(rcd)); 4061 if (addr > 0xffffffffllu) 4062 _lto4b(0xffffffff, rcd.addr); 4063 else 4064 _lto4b(addr, rcd.addr); 4065 _lto4b(secsize, rcd.length); 4066 sr_copy_internal_data(xs, &rcd, sizeof(rcd)); 4067 rv = 0; 4068 } else if (xs->cmd->opcode == READ_CAPACITY_16) { 4069 bzero(&rcd16, sizeof(rcd16)); 4070 _lto8b(addr, rcd16.addr); 4071 _lto4b(secsize, rcd16.length); 4072 sr_copy_internal_data(xs, &rcd16, sizeof(rcd16)); 4073 rv = 0; 4074 } 4075 4076 return (rv); 4077 } 4078 4079 int 4080 sr_raid_tur(struct sr_workunit *wu) 4081 { 4082 struct sr_discipline *sd = wu->swu_dis; 4083 4084 DNPRINTF(SR_D_DIS, "%s: sr_raid_tur\n", DEVNAME(sd->sd_sc)); 4085 4086 if (sd->sd_vol_status == BIOC_SVOFFLINE) { 4087 sd->sd_scsi_sense.error_code = SSD_ERRCODE_CURRENT; 4088 sd->sd_scsi_sense.flags = SKEY_NOT_READY; 4089 sd->sd_scsi_sense.add_sense_code = 0x04; 4090 sd->sd_scsi_sense.add_sense_code_qual = 0x11; 4091 sd->sd_scsi_sense.extra_len = 4; 4092 return (1); 4093 } else if (sd->sd_vol_status == BIOC_SVINVALID) { 4094 sd->sd_scsi_sense.error_code = SSD_ERRCODE_CURRENT; 4095 sd->sd_scsi_sense.flags = SKEY_HARDWARE_ERROR; 4096 sd->sd_scsi_sense.add_sense_code = 0x05; 4097 sd->sd_scsi_sense.add_sense_code_qual = 0x00; 4098 sd->sd_scsi_sense.extra_len = 4; 4099 return (1); 4100 } 4101 4102 return (0); 4103 } 4104 4105 int 4106 sr_raid_request_sense(struct sr_workunit *wu) 4107 { 4108 struct sr_discipline *sd = wu->swu_dis; 4109 struct scsi_xfer *xs = wu->swu_xs; 4110 4111 DNPRINTF(SR_D_DIS, "%s: sr_raid_request_sense\n", 4112 DEVNAME(sd->sd_sc)); 4113 4114 /* use latest sense data */ 4115 memcpy(&xs->sense, &sd->sd_scsi_sense, sizeof(xs->sense)); 4116 4117 /* clear sense data */ 4118 bzero(&sd->sd_scsi_sense, sizeof(sd->sd_scsi_sense)); 4119 4120 return (0); 4121 } 4122 4123 int 4124 sr_raid_start_stop(struct sr_workunit *wu) 4125 { 4126 struct scsi_xfer *xs = wu->swu_xs; 4127 struct scsi_start_stop *ss = (struct scsi_start_stop *)xs->cmd; 4128 4129 DNPRINTF(SR_D_DIS, "%s: sr_raid_start_stop\n", 4130 DEVNAME(wu->swu_dis->sd_sc)); 4131 4132 if (!ss) 4133 return (1); 4134 4135 /* 4136 * do nothing! 4137 * a softraid discipline should always reflect correct status 4138 */ 4139 return (0); 4140 } 4141 4142 int 4143 sr_raid_sync(struct sr_workunit *wu) 4144 { 4145 struct sr_discipline *sd = wu->swu_dis; 4146 int s, ret, rv = 0, ios; 4147 4148 DNPRINTF(SR_D_DIS, "%s: sr_raid_sync\n", DEVNAME(sd->sd_sc)); 4149 4150 /* when doing a fake sync don't count the wu */ 4151 ios = (wu->swu_flags & SR_WUF_FAKE) ? 0 : 1; 4152 4153 s = splbio(); 4154 sd->sd_sync = 1; 4155 while (sd->sd_wu_pending > ios) { 4156 ret = tsleep_nsec(sd, PRIBIO, "sr_sync", SEC_TO_NSEC(15)); 4157 if (ret == EWOULDBLOCK) { 4158 DNPRINTF(SR_D_DIS, "%s: sr_raid_sync timeout\n", 4159 DEVNAME(sd->sd_sc)); 4160 rv = 1; 4161 break; 4162 } 4163 } 4164 sd->sd_sync = 0; 4165 splx(s); 4166 4167 wakeup(&sd->sd_sync); 4168 4169 return (rv); 4170 } 4171 4172 void 4173 sr_raid_intr(struct buf *bp) 4174 { 4175 struct sr_ccb *ccb = (struct sr_ccb *)bp; 4176 struct sr_workunit *wu = ccb->ccb_wu; 4177 #ifdef SR_DEBUG 4178 struct sr_discipline *sd = wu->swu_dis; 4179 struct scsi_xfer *xs = wu->swu_xs; 4180 #endif 4181 int s; 4182 4183 DNPRINTF(SR_D_INTR, "%s: %s %s intr bp %p xs %p\n", 4184 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, sd->sd_name, bp, xs); 4185 4186 s = splbio(); 4187 sr_ccb_done(ccb); 4188 sr_wu_done(wu); 4189 splx(s); 4190 } 4191 4192 void 4193 sr_schedule_wu(struct sr_workunit *wu) 4194 { 4195 struct sr_discipline *sd = wu->swu_dis; 4196 struct sr_workunit *wup; 4197 int s; 4198 4199 DNPRINTF(SR_D_WU, "sr_schedule_wu: schedule wu %p state %i " 4200 "flags 0x%x\n", wu, wu->swu_state, wu->swu_flags); 4201 4202 KASSERT(wu->swu_io_count > 0); 4203 4204 s = splbio(); 4205 4206 /* Construct the work unit, do not schedule it. */ 4207 if (wu->swu_state == SR_WU_CONSTRUCT) 4208 goto queued; 4209 4210 /* Deferred work unit being reconstructed, do not start. */ 4211 if (wu->swu_state == SR_WU_REQUEUE) 4212 goto queued; 4213 4214 /* Current work unit failed, restart. */ 4215 if (wu->swu_state == SR_WU_RESTART) 4216 goto start; 4217 4218 if (wu->swu_state != SR_WU_INPROGRESS) 4219 panic("sr_schedule_wu: work unit not in progress (state %i)\n", 4220 wu->swu_state); 4221 4222 /* Walk queue backwards and fill in collider if we have one. */ 4223 TAILQ_FOREACH_REVERSE(wup, &sd->sd_wu_pendq, sr_wu_list, swu_link) { 4224 if (wu->swu_blk_end < wup->swu_blk_start || 4225 wup->swu_blk_end < wu->swu_blk_start) 4226 continue; 4227 4228 /* Defer work unit due to LBA collision. */ 4229 DNPRINTF(SR_D_WU, "sr_schedule_wu: deferring work unit %p\n", 4230 wu); 4231 wu->swu_state = SR_WU_DEFERRED; 4232 while (wup->swu_collider) 4233 wup = wup->swu_collider; 4234 wup->swu_collider = wu; 4235 TAILQ_INSERT_TAIL(&sd->sd_wu_defq, wu, swu_link); 4236 sd->sd_wu_collisions++; 4237 goto queued; 4238 } 4239 4240 start: 4241 sr_raid_startwu(wu); 4242 4243 queued: 4244 splx(s); 4245 } 4246 4247 void 4248 sr_raid_startwu(struct sr_workunit *wu) 4249 { 4250 struct sr_discipline *sd = wu->swu_dis; 4251 struct sr_ccb *ccb; 4252 4253 DNPRINTF(SR_D_WU, "sr_raid_startwu: start wu %p\n", wu); 4254 4255 splassert(IPL_BIO); 4256 4257 if (wu->swu_state == SR_WU_DEFERRED) { 4258 TAILQ_REMOVE(&sd->sd_wu_defq, wu, swu_link); 4259 wu->swu_state = SR_WU_INPROGRESS; 4260 } 4261 4262 if (wu->swu_state != SR_WU_RESTART) 4263 TAILQ_INSERT_TAIL(&sd->sd_wu_pendq, wu, swu_link); 4264 4265 /* Start all of the individual I/Os. */ 4266 if (wu->swu_cb_active == 1) 4267 panic("%s: sr_startwu_callback", DEVNAME(sd->sd_sc)); 4268 wu->swu_cb_active = 1; 4269 4270 TAILQ_FOREACH(ccb, &wu->swu_ccb, ccb_link) 4271 VOP_STRATEGY(&ccb->ccb_buf); 4272 4273 wu->swu_cb_active = 0; 4274 } 4275 4276 void 4277 sr_raid_recreate_wu(struct sr_workunit *wu) 4278 { 4279 struct sr_discipline *sd = wu->swu_dis; 4280 struct sr_workunit *wup = wu; 4281 4282 /* 4283 * Recreate a work unit by releasing the associated CCBs and reissuing 4284 * the SCSI I/O request. This process is then repeated for all of the 4285 * colliding work units. 4286 */ 4287 do { 4288 sr_wu_release_ccbs(wup); 4289 4290 wup->swu_state = SR_WU_REQUEUE; 4291 if (sd->sd_scsi_rw(wup)) 4292 panic("could not requeue I/O"); 4293 4294 wup = wup->swu_collider; 4295 } while (wup); 4296 } 4297 4298 int 4299 sr_alloc_resources(struct sr_discipline *sd) 4300 { 4301 if (sr_wu_alloc(sd)) { 4302 sr_error(sd->sd_sc, "unable to allocate work units"); 4303 return (ENOMEM); 4304 } 4305 if (sr_ccb_alloc(sd)) { 4306 sr_error(sd->sd_sc, "unable to allocate ccbs"); 4307 return (ENOMEM); 4308 } 4309 4310 return (0); 4311 } 4312 4313 void 4314 sr_free_resources(struct sr_discipline *sd) 4315 { 4316 sr_wu_free(sd); 4317 sr_ccb_free(sd); 4318 } 4319 4320 void 4321 sr_set_chunk_state(struct sr_discipline *sd, int c, int new_state) 4322 { 4323 int old_state, s; 4324 4325 DNPRINTF(SR_D_STATE, "%s: %s: %s: sr_set_chunk_state %d -> %d\n", 4326 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, 4327 sd->sd_vol.sv_chunks[c]->src_meta.scmi.scm_devname, c, new_state); 4328 4329 /* ok to go to splbio since this only happens in error path */ 4330 s = splbio(); 4331 old_state = sd->sd_vol.sv_chunks[c]->src_meta.scm_status; 4332 4333 /* multiple IOs to the same chunk that fail will come through here */ 4334 if (old_state == new_state) 4335 goto done; 4336 4337 switch (old_state) { 4338 case BIOC_SDONLINE: 4339 if (new_state == BIOC_SDOFFLINE) 4340 break; 4341 else 4342 goto die; 4343 break; 4344 4345 case BIOC_SDOFFLINE: 4346 goto die; 4347 4348 default: 4349 die: 4350 splx(s); /* XXX */ 4351 panic("%s: %s: %s: invalid chunk state transition " 4352 "%d -> %d", DEVNAME(sd->sd_sc), 4353 sd->sd_meta->ssd_devname, 4354 sd->sd_vol.sv_chunks[c]->src_meta.scmi.scm_devname, 4355 old_state, new_state); 4356 /* NOTREACHED */ 4357 } 4358 4359 sd->sd_vol.sv_chunks[c]->src_meta.scm_status = new_state; 4360 sd->sd_set_vol_state(sd); 4361 4362 sd->sd_must_flush = 1; 4363 task_add(systq, &sd->sd_meta_save_task); 4364 done: 4365 splx(s); 4366 } 4367 4368 void 4369 sr_set_vol_state(struct sr_discipline *sd) 4370 { 4371 int states[SR_MAX_STATES]; 4372 int new_state, i, nd; 4373 int old_state = sd->sd_vol_status; 4374 u_int32_t s; 4375 4376 DNPRINTF(SR_D_STATE, "%s: %s: sr_set_vol_state\n", 4377 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname); 4378 4379 nd = sd->sd_meta->ssdi.ssd_chunk_no; 4380 4381 for (i = 0; i < SR_MAX_STATES; i++) 4382 states[i] = 0; 4383 4384 for (i = 0; i < nd; i++) { 4385 s = sd->sd_vol.sv_chunks[i]->src_meta.scm_status; 4386 if (s >= SR_MAX_STATES) 4387 panic("%s: %s: %s: invalid chunk state", 4388 DEVNAME(sd->sd_sc), 4389 sd->sd_meta->ssd_devname, 4390 sd->sd_vol.sv_chunks[i]->src_meta.scmi.scm_devname); 4391 states[s]++; 4392 } 4393 4394 if (states[BIOC_SDONLINE] == nd) 4395 new_state = BIOC_SVONLINE; 4396 else 4397 new_state = BIOC_SVOFFLINE; 4398 4399 DNPRINTF(SR_D_STATE, "%s: %s: sr_set_vol_state %d -> %d\n", 4400 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, 4401 old_state, new_state); 4402 4403 switch (old_state) { 4404 case BIOC_SVONLINE: 4405 if (new_state == BIOC_SVOFFLINE || new_state == BIOC_SVONLINE) 4406 break; 4407 else 4408 goto die; 4409 break; 4410 4411 case BIOC_SVOFFLINE: 4412 /* XXX this might be a little too much */ 4413 goto die; 4414 4415 default: 4416 die: 4417 panic("%s: %s: invalid volume state transition " 4418 "%d -> %d", DEVNAME(sd->sd_sc), 4419 sd->sd_meta->ssd_devname, 4420 old_state, new_state); 4421 /* NOTREACHED */ 4422 } 4423 4424 sd->sd_vol_status = new_state; 4425 } 4426 4427 void * 4428 sr_block_get(struct sr_discipline *sd, long length) 4429 { 4430 return dma_alloc(length, PR_NOWAIT | PR_ZERO); 4431 } 4432 4433 void 4434 sr_block_put(struct sr_discipline *sd, void *ptr, int length) 4435 { 4436 dma_free(ptr, length); 4437 } 4438 4439 void 4440 sr_checksum_print(u_int8_t *md5) 4441 { 4442 int i; 4443 4444 for (i = 0; i < MD5_DIGEST_LENGTH; i++) 4445 printf("%02x", md5[i]); 4446 } 4447 4448 void 4449 sr_checksum(struct sr_softc *sc, void *src, void *md5, u_int32_t len) 4450 { 4451 MD5_CTX ctx; 4452 4453 DNPRINTF(SR_D_MISC, "%s: sr_checksum(%p %p %d)\n", DEVNAME(sc), src, 4454 md5, len); 4455 4456 MD5Init(&ctx); 4457 MD5Update(&ctx, src, len); 4458 MD5Final(md5, &ctx); 4459 } 4460 4461 void 4462 sr_uuid_generate(struct sr_uuid *uuid) 4463 { 4464 arc4random_buf(uuid->sui_id, sizeof(uuid->sui_id)); 4465 /* UUID version 4: random */ 4466 uuid->sui_id[6] &= 0x0f; 4467 uuid->sui_id[6] |= 0x40; 4468 /* RFC4122 variant */ 4469 uuid->sui_id[8] &= 0x3f; 4470 uuid->sui_id[8] |= 0x80; 4471 } 4472 4473 char * 4474 sr_uuid_format(struct sr_uuid *uuid) 4475 { 4476 char *uuidstr; 4477 4478 uuidstr = malloc(37, M_DEVBUF, M_WAITOK); 4479 4480 snprintf(uuidstr, 37, 4481 "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-" 4482 "%02x%02x%02x%02x%02x%02x", 4483 uuid->sui_id[0], uuid->sui_id[1], 4484 uuid->sui_id[2], uuid->sui_id[3], 4485 uuid->sui_id[4], uuid->sui_id[5], 4486 uuid->sui_id[6], uuid->sui_id[7], 4487 uuid->sui_id[8], uuid->sui_id[9], 4488 uuid->sui_id[10], uuid->sui_id[11], 4489 uuid->sui_id[12], uuid->sui_id[13], 4490 uuid->sui_id[14], uuid->sui_id[15]); 4491 4492 return uuidstr; 4493 } 4494 4495 void 4496 sr_uuid_print(struct sr_uuid *uuid, int cr) 4497 { 4498 char *uuidstr; 4499 4500 uuidstr = sr_uuid_format(uuid); 4501 printf("%s%s", uuidstr, (cr ? "\n" : "")); 4502 free(uuidstr, M_DEVBUF, 37); 4503 } 4504 4505 int 4506 sr_already_assembled(struct sr_discipline *sd) 4507 { 4508 struct sr_softc *sc = sd->sd_sc; 4509 struct sr_discipline *sdtmp; 4510 4511 TAILQ_FOREACH(sdtmp, &sc->sc_dis_list, sd_link) { 4512 if (!bcmp(&sd->sd_meta->ssdi.ssd_uuid, 4513 &sdtmp->sd_meta->ssdi.ssd_uuid, 4514 sizeof(sd->sd_meta->ssdi.ssd_uuid))) 4515 return (1); 4516 } 4517 4518 return (0); 4519 } 4520 4521 int32_t 4522 sr_validate_stripsize(u_int32_t b) 4523 { 4524 int s = 0; 4525 4526 if (b % DEV_BSIZE) 4527 return (-1); 4528 4529 while ((b & 1) == 0) { 4530 b >>= 1; 4531 s++; 4532 } 4533 4534 /* only multiple of twos */ 4535 b >>= 1; 4536 if (b) 4537 return(-1); 4538 4539 return (s); 4540 } 4541 4542 void 4543 sr_quiesce(void) 4544 { 4545 struct sr_softc *sc = softraid0; 4546 struct sr_discipline *sd, *nsd; 4547 4548 /* Shutdown disciplines in reverse attach order. */ 4549 TAILQ_FOREACH_REVERSE_SAFE(sd, &sc->sc_dis_list, 4550 sr_discipline_list, sd_link, nsd) 4551 sr_discipline_shutdown(sd, 1, -1); 4552 } 4553 4554 void 4555 sr_shutdown(int dying) 4556 { 4557 struct sr_softc *sc = softraid0; 4558 struct sr_discipline *sd; 4559 4560 DNPRINTF(SR_D_MISC, "%s: sr_shutdown\n", DEVNAME(sc)); 4561 4562 /* 4563 * Since softraid is not under mainbus, we have to explicitly 4564 * notify its children that the power is going down, so they 4565 * can execute their shutdown hooks. 4566 */ 4567 config_suspend((struct device *)sc, DVACT_POWERDOWN); 4568 4569 /* Shutdown disciplines in reverse attach order. */ 4570 while ((sd = TAILQ_LAST(&sc->sc_dis_list, sr_discipline_list)) != NULL) 4571 sr_discipline_shutdown(sd, 1, dying); 4572 } 4573 4574 int 4575 sr_validate_io(struct sr_workunit *wu, daddr_t *blkno, char *func) 4576 { 4577 struct sr_discipline *sd = wu->swu_dis; 4578 struct scsi_xfer *xs = wu->swu_xs; 4579 int rv = 1; 4580 4581 DNPRINTF(SR_D_DIS, "%s: %s 0x%02x\n", DEVNAME(sd->sd_sc), func, 4582 xs->cmd->opcode); 4583 4584 if (sd->sd_meta->ssd_data_blkno == 0) 4585 panic("invalid data blkno"); 4586 4587 if (sd->sd_vol_status == BIOC_SVOFFLINE) { 4588 DNPRINTF(SR_D_DIS, "%s: %s device offline\n", 4589 DEVNAME(sd->sd_sc), func); 4590 goto bad; 4591 } 4592 4593 if (xs->datalen == 0) { 4594 printf("%s: %s: illegal block count for %s\n", 4595 DEVNAME(sd->sd_sc), func, sd->sd_meta->ssd_devname); 4596 goto bad; 4597 } 4598 4599 if (xs->cmdlen == 10) 4600 *blkno = _4btol(((struct scsi_rw_big *)xs->cmd)->addr); 4601 else if (xs->cmdlen == 16) 4602 *blkno = _8btol(((struct scsi_rw_16 *)xs->cmd)->addr); 4603 else if (xs->cmdlen == 6) 4604 *blkno = _3btol(((struct scsi_rw *)xs->cmd)->addr); 4605 else { 4606 printf("%s: %s: illegal cmdlen for %s\n", 4607 DEVNAME(sd->sd_sc), func, sd->sd_meta->ssd_devname); 4608 goto bad; 4609 } 4610 4611 *blkno *= (sd->sd_meta->ssdi.ssd_secsize / DEV_BSIZE); 4612 4613 wu->swu_blk_start = *blkno; 4614 wu->swu_blk_end = *blkno + (xs->datalen >> DEV_BSHIFT) - 1; 4615 4616 if (wu->swu_blk_end > sd->sd_meta->ssdi.ssd_size) { 4617 DNPRINTF(SR_D_DIS, "%s: %s out of bounds start: %lld " 4618 "end: %lld length: %d\n", 4619 DEVNAME(sd->sd_sc), func, (long long)wu->swu_blk_start, 4620 (long long)wu->swu_blk_end, xs->datalen); 4621 4622 sd->sd_scsi_sense.error_code = SSD_ERRCODE_CURRENT | 4623 SSD_ERRCODE_VALID; 4624 sd->sd_scsi_sense.flags = SKEY_ILLEGAL_REQUEST; 4625 sd->sd_scsi_sense.add_sense_code = 0x21; 4626 sd->sd_scsi_sense.add_sense_code_qual = 0x00; 4627 sd->sd_scsi_sense.extra_len = 4; 4628 goto bad; 4629 } 4630 4631 rv = 0; 4632 bad: 4633 return (rv); 4634 } 4635 4636 void 4637 sr_rebuild_start(void *arg) 4638 { 4639 struct sr_discipline *sd = arg; 4640 struct sr_softc *sc = sd->sd_sc; 4641 4642 DNPRINTF(SR_D_REBUILD, "%s: %s starting rebuild thread\n", 4643 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname); 4644 4645 if (kthread_create(sr_rebuild_thread, sd, &sd->sd_background_proc, 4646 DEVNAME(sc)) != 0) 4647 printf("%s: unable to start background operation\n", 4648 DEVNAME(sc)); 4649 } 4650 4651 void 4652 sr_rebuild_thread(void *arg) 4653 { 4654 struct sr_discipline *sd = arg; 4655 4656 DNPRINTF(SR_D_REBUILD, "%s: %s rebuild thread started\n", 4657 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname); 4658 4659 sd->sd_reb_active = 1; 4660 sd->sd_rebuild(sd); 4661 sd->sd_reb_active = 0; 4662 4663 kthread_exit(0); 4664 } 4665 4666 void 4667 sr_rebuild(struct sr_discipline *sd) 4668 { 4669 struct sr_softc *sc = sd->sd_sc; 4670 u_int64_t sz, whole_blk, partial_blk, blk, restart; 4671 daddr_t lba; 4672 struct sr_workunit *wu_r, *wu_w; 4673 struct scsi_xfer xs_r, xs_w; 4674 struct scsi_rw_16 *cr, *cw; 4675 int c, s, slept, percent = 0, old_percent = -1; 4676 u_int8_t *buf; 4677 4678 whole_blk = sd->sd_meta->ssdi.ssd_size / SR_REBUILD_IO_SIZE; 4679 partial_blk = sd->sd_meta->ssdi.ssd_size % SR_REBUILD_IO_SIZE; 4680 4681 restart = sd->sd_meta->ssd_rebuild / SR_REBUILD_IO_SIZE; 4682 if (restart > whole_blk) { 4683 printf("%s: bogus rebuild restart offset, starting from 0\n", 4684 DEVNAME(sc)); 4685 restart = 0; 4686 } 4687 if (restart) { 4688 /* 4689 * XXX there is a hole here; there is a posibility that we 4690 * had a restart however the chunk that was supposed to 4691 * be rebuilt is no longer valid; we can reach this situation 4692 * when a rebuild is in progress and the box crashes and 4693 * on reboot the rebuild chunk is different (like zero'd or 4694 * replaced). We need to check the uuid of the chunk that is 4695 * being rebuilt to assert this. 4696 */ 4697 percent = sr_rebuild_percent(sd); 4698 printf("%s: resuming rebuild on %s at %d%%\n", 4699 DEVNAME(sc), sd->sd_meta->ssd_devname, percent); 4700 } 4701 4702 /* currently this is 64k therefore we can use dma_alloc */ 4703 buf = dma_alloc(SR_REBUILD_IO_SIZE << DEV_BSHIFT, PR_WAITOK); 4704 for (blk = restart; blk <= whole_blk; blk++) { 4705 lba = blk * SR_REBUILD_IO_SIZE; 4706 sz = SR_REBUILD_IO_SIZE; 4707 if (blk == whole_blk) { 4708 if (partial_blk == 0) 4709 break; 4710 sz = partial_blk; 4711 } 4712 4713 /* get some wu */ 4714 wu_r = sr_scsi_wu_get(sd, 0); 4715 wu_w = sr_scsi_wu_get(sd, 0); 4716 4717 DNPRINTF(SR_D_REBUILD, "%s: %s rebuild wu_r %p, wu_w %p\n", 4718 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, wu_r, wu_w); 4719 4720 /* setup read io */ 4721 bzero(&xs_r, sizeof xs_r); 4722 xs_r.error = XS_NOERROR; 4723 xs_r.flags = SCSI_DATA_IN; 4724 xs_r.datalen = sz << DEV_BSHIFT; 4725 xs_r.data = buf; 4726 xs_r.cmdlen = sizeof(*cr); 4727 xs_r.cmd = &xs_r.cmdstore; 4728 cr = (struct scsi_rw_16 *)xs_r.cmd; 4729 cr->opcode = READ_16; 4730 _lto4b(sz, cr->length); 4731 _lto8b(lba, cr->addr); 4732 wu_r->swu_state = SR_WU_CONSTRUCT; 4733 wu_r->swu_flags |= SR_WUF_REBUILD; 4734 wu_r->swu_xs = &xs_r; 4735 if (sd->sd_scsi_rw(wu_r)) { 4736 printf("%s: could not create read io\n", 4737 DEVNAME(sc)); 4738 goto fail; 4739 } 4740 4741 /* setup write io */ 4742 bzero(&xs_w, sizeof xs_w); 4743 xs_w.error = XS_NOERROR; 4744 xs_w.flags = SCSI_DATA_OUT; 4745 xs_w.datalen = sz << DEV_BSHIFT; 4746 xs_w.data = buf; 4747 xs_w.cmdlen = sizeof(*cw); 4748 xs_w.cmd = &xs_w.cmdstore; 4749 cw = (struct scsi_rw_16 *)xs_w.cmd; 4750 cw->opcode = WRITE_16; 4751 _lto4b(sz, cw->length); 4752 _lto8b(lba, cw->addr); 4753 wu_w->swu_state = SR_WU_CONSTRUCT; 4754 wu_w->swu_flags |= SR_WUF_REBUILD | SR_WUF_WAKEUP; 4755 wu_w->swu_xs = &xs_w; 4756 if (sd->sd_scsi_rw(wu_w)) { 4757 printf("%s: could not create write io\n", 4758 DEVNAME(sc)); 4759 goto fail; 4760 } 4761 4762 /* 4763 * collide with the read io so that we get automatically 4764 * started when the read is done 4765 */ 4766 wu_w->swu_state = SR_WU_DEFERRED; 4767 wu_r->swu_collider = wu_w; 4768 s = splbio(); 4769 TAILQ_INSERT_TAIL(&sd->sd_wu_defq, wu_w, swu_link); 4770 splx(s); 4771 4772 DNPRINTF(SR_D_REBUILD, "%s: %s rebuild scheduling wu_r %p\n", 4773 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, wu_r); 4774 4775 wu_r->swu_state = SR_WU_INPROGRESS; 4776 sr_schedule_wu(wu_r); 4777 4778 /* wait for write completion */ 4779 slept = 0; 4780 while ((wu_w->swu_flags & SR_WUF_REBUILDIOCOMP) == 0) { 4781 tsleep_nsec(wu_w, PRIBIO, "sr_rebuild", INFSLP); 4782 slept = 1; 4783 } 4784 /* yield if we didn't sleep */ 4785 if (slept == 0) 4786 tsleep(sc, PWAIT, "sr_yield", 1); 4787 4788 sr_scsi_wu_put(sd, wu_r); 4789 sr_scsi_wu_put(sd, wu_w); 4790 4791 sd->sd_meta->ssd_rebuild = lba; 4792 4793 /* XXX - this should be based on size, not percentage. */ 4794 /* save metadata every percent */ 4795 percent = sr_rebuild_percent(sd); 4796 if (percent != old_percent && blk != whole_blk) { 4797 if (sr_meta_save(sd, SR_META_DIRTY)) 4798 printf("%s: could not save metadata to %s\n", 4799 DEVNAME(sc), sd->sd_meta->ssd_devname); 4800 old_percent = percent; 4801 } 4802 4803 if (sd->sd_reb_abort) 4804 goto abort; 4805 } 4806 4807 /* all done */ 4808 sd->sd_meta->ssd_rebuild = 0; 4809 for (c = 0; c < sd->sd_meta->ssdi.ssd_chunk_no; c++) { 4810 if (sd->sd_vol.sv_chunks[c]->src_meta.scm_status == 4811 BIOC_SDREBUILD) { 4812 sd->sd_set_chunk_state(sd, c, BIOC_SDONLINE); 4813 break; 4814 } 4815 } 4816 4817 abort: 4818 if (sr_meta_save(sd, SR_META_DIRTY)) 4819 printf("%s: could not save metadata to %s\n", 4820 DEVNAME(sc), sd->sd_meta->ssd_devname); 4821 fail: 4822 dma_free(buf, SR_REBUILD_IO_SIZE << DEV_BSHIFT); 4823 } 4824 4825 #ifndef SMALL_KERNEL 4826 int 4827 sr_sensors_create(struct sr_discipline *sd) 4828 { 4829 struct sr_softc *sc = sd->sd_sc; 4830 int rv = 1; 4831 4832 DNPRINTF(SR_D_STATE, "%s: %s: sr_sensors_create\n", 4833 DEVNAME(sc), sd->sd_meta->ssd_devname); 4834 4835 sd->sd_vol.sv_sensor.type = SENSOR_DRIVE; 4836 sd->sd_vol.sv_sensor.status = SENSOR_S_UNKNOWN; 4837 strlcpy(sd->sd_vol.sv_sensor.desc, sd->sd_meta->ssd_devname, 4838 sizeof(sd->sd_vol.sv_sensor.desc)); 4839 4840 sensor_attach(&sc->sc_sensordev, &sd->sd_vol.sv_sensor); 4841 sd->sd_vol.sv_sensor_attached = 1; 4842 4843 if (sc->sc_sensor_task == NULL) { 4844 sc->sc_sensor_task = sensor_task_register(sc, 4845 sr_sensors_refresh, 10); 4846 if (sc->sc_sensor_task == NULL) 4847 goto bad; 4848 } 4849 4850 rv = 0; 4851 bad: 4852 return (rv); 4853 } 4854 4855 void 4856 sr_sensors_delete(struct sr_discipline *sd) 4857 { 4858 DNPRINTF(SR_D_STATE, "%s: sr_sensors_delete\n", DEVNAME(sd->sd_sc)); 4859 4860 if (sd->sd_vol.sv_sensor_attached) 4861 sensor_detach(&sd->sd_sc->sc_sensordev, &sd->sd_vol.sv_sensor); 4862 } 4863 4864 void 4865 sr_sensors_refresh(void *arg) 4866 { 4867 struct sr_softc *sc = arg; 4868 struct sr_volume *sv; 4869 struct sr_discipline *sd; 4870 4871 DNPRINTF(SR_D_STATE, "%s: sr_sensors_refresh\n", DEVNAME(sc)); 4872 4873 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 4874 sv = &sd->sd_vol; 4875 4876 switch(sd->sd_vol_status) { 4877 case BIOC_SVOFFLINE: 4878 sv->sv_sensor.value = SENSOR_DRIVE_FAIL; 4879 sv->sv_sensor.status = SENSOR_S_CRIT; 4880 break; 4881 4882 case BIOC_SVDEGRADED: 4883 sv->sv_sensor.value = SENSOR_DRIVE_PFAIL; 4884 sv->sv_sensor.status = SENSOR_S_WARN; 4885 break; 4886 4887 case BIOC_SVREBUILD: 4888 sv->sv_sensor.value = SENSOR_DRIVE_REBUILD; 4889 sv->sv_sensor.status = SENSOR_S_WARN; 4890 break; 4891 4892 case BIOC_SVSCRUB: 4893 case BIOC_SVONLINE: 4894 sv->sv_sensor.value = SENSOR_DRIVE_ONLINE; 4895 sv->sv_sensor.status = SENSOR_S_OK; 4896 break; 4897 4898 default: 4899 sv->sv_sensor.value = 0; /* unknown */ 4900 sv->sv_sensor.status = SENSOR_S_UNKNOWN; 4901 } 4902 } 4903 } 4904 #endif /* SMALL_KERNEL */ 4905 4906 #ifdef SR_FANCY_STATS 4907 void sr_print_stats(void); 4908 4909 void 4910 sr_print_stats(void) 4911 { 4912 struct sr_softc *sc = softraid0; 4913 struct sr_discipline *sd; 4914 4915 if (sc == NULL) { 4916 printf("no softraid softc found\n"); 4917 return; 4918 } 4919 4920 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 4921 printf("%s: ios pending %d, collisions %llu\n", 4922 sd->sd_meta->ssd_devname, 4923 sd->sd_wu_pending, 4924 sd->sd_wu_collisions); 4925 } 4926 } 4927 #endif /* SR_FANCY_STATS */ 4928 4929 #ifdef SR_DEBUG 4930 void 4931 sr_meta_print(struct sr_metadata *m) 4932 { 4933 int i; 4934 struct sr_meta_chunk *mc; 4935 struct sr_meta_opt_hdr *omh; 4936 4937 if (!(sr_debug & SR_D_META)) 4938 return; 4939 4940 printf("\tssd_magic 0x%llx\n", m->ssdi.ssd_magic); 4941 printf("\tssd_version %d\n", m->ssdi.ssd_version); 4942 printf("\tssd_vol_flags 0x%x\n", m->ssdi.ssd_vol_flags); 4943 printf("\tssd_uuid "); 4944 sr_uuid_print(&m->ssdi.ssd_uuid, 1); 4945 printf("\tssd_chunk_no %d\n", m->ssdi.ssd_chunk_no); 4946 printf("\tssd_chunk_id %d\n", m->ssdi.ssd_chunk_id); 4947 printf("\tssd_opt_no %d\n", m->ssdi.ssd_opt_no); 4948 printf("\tssd_volid %d\n", m->ssdi.ssd_volid); 4949 printf("\tssd_level %d\n", m->ssdi.ssd_level); 4950 printf("\tssd_size %lld\n", m->ssdi.ssd_size); 4951 printf("\tssd_devname %s\n", m->ssd_devname); 4952 printf("\tssd_vendor %s\n", m->ssdi.ssd_vendor); 4953 printf("\tssd_product %s\n", m->ssdi.ssd_product); 4954 printf("\tssd_revision %s\n", m->ssdi.ssd_revision); 4955 printf("\tssd_strip_size %d\n", m->ssdi.ssd_strip_size); 4956 printf("\tssd_checksum "); 4957 sr_checksum_print(m->ssd_checksum); 4958 printf("\n"); 4959 printf("\tssd_meta_flags 0x%x\n", m->ssd_meta_flags); 4960 printf("\tssd_ondisk %llu\n", m->ssd_ondisk); 4961 4962 mc = (struct sr_meta_chunk *)(m + 1); 4963 for (i = 0; i < m->ssdi.ssd_chunk_no; i++, mc++) { 4964 printf("\t\tscm_volid %d\n", mc->scmi.scm_volid); 4965 printf("\t\tscm_chunk_id %d\n", mc->scmi.scm_chunk_id); 4966 printf("\t\tscm_devname %s\n", mc->scmi.scm_devname); 4967 printf("\t\tscm_size %lld\n", mc->scmi.scm_size); 4968 printf("\t\tscm_coerced_size %lld\n",mc->scmi.scm_coerced_size); 4969 printf("\t\tscm_uuid "); 4970 sr_uuid_print(&mc->scmi.scm_uuid, 1); 4971 printf("\t\tscm_checksum "); 4972 sr_checksum_print(mc->scm_checksum); 4973 printf("\n"); 4974 printf("\t\tscm_status %d\n", mc->scm_status); 4975 } 4976 4977 omh = (struct sr_meta_opt_hdr *)((u_int8_t *)(m + 1) + 4978 sizeof(struct sr_meta_chunk) * m->ssdi.ssd_chunk_no); 4979 for (i = 0; i < m->ssdi.ssd_opt_no; i++) { 4980 printf("\t\t\tsom_type %d\n", omh->som_type); 4981 printf("\t\t\tsom_checksum "); 4982 sr_checksum_print(omh->som_checksum); 4983 printf("\n"); 4984 omh = (struct sr_meta_opt_hdr *)((void *)omh + 4985 omh->som_length); 4986 } 4987 } 4988 4989 void 4990 sr_dump_block(void *blk, int len) 4991 { 4992 uint8_t *b = blk; 4993 int i, j, c; 4994 4995 for (i = 0; i < len; i += 16) { 4996 for (j = 0; j < 16; j++) 4997 printf("%.2x ", b[i + j]); 4998 printf(" "); 4999 for (j = 0; j < 16; j++) { 5000 c = b[i + j]; 5001 if (c < ' ' || c > 'z' || i + j > len) 5002 c = '.'; 5003 printf("%c", c); 5004 } 5005 printf("\n"); 5006 } 5007 } 5008 5009 void 5010 sr_dump_mem(u_int8_t *p, int len) 5011 { 5012 int i; 5013 5014 for (i = 0; i < len; i++) 5015 printf("%02x ", *p++); 5016 printf("\n"); 5017 } 5018 5019 #endif /* SR_DEBUG */ 5020 5021 #ifdef HIBERNATE 5022 /* 5023 * Side-effect free (no malloc, printf, pool, splx) softraid crypto writer. 5024 * 5025 * This function must perform the following: 5026 * 1. Determine the underlying device's own side-effect free I/O function 5027 * (eg, ahci_hibernate_io, wd_hibernate_io, etc). 5028 * 2. Store enough information in the provided page argument for subsequent 5029 * I/O calls (such as the crypto discipline structure for the keys, the 5030 * offset of the softraid partition on the underlying disk, as well as 5031 * the offset of the swap partition within the crypto volume. 5032 * 3. Encrypt the incoming data using the sr_discipline keys, then pass 5033 * the request to the underlying device's own I/O function. 5034 */ 5035 int 5036 sr_hibernate_io(dev_t dev, daddr_t blkno, vaddr_t addr, size_t size, int op, void *page) 5037 { 5038 /* Struct for stashing data obtained on HIB_INIT. 5039 * XXX 5040 * We share the page with the underlying device's own 5041 * side-effect free I/O function, so we pad our data to 5042 * the end of the page. Presently this does not overlap 5043 * with either of the two other side-effect free i/o 5044 * functions (ahci/wd). 5045 */ 5046 struct { 5047 char pad[3072]; 5048 struct sr_discipline *srd; 5049 hibio_fn subfn; /* underlying device i/o fn */ 5050 dev_t subdev; /* underlying device dev_t */ 5051 daddr_t sr_swapoff; /* ofs of swap part in sr volume */ 5052 char buf[DEV_BSIZE]; /* encryption performed into this buf */ 5053 } *my = page; 5054 extern struct cfdriver sd_cd; 5055 char errstr[128], *dl_ret; 5056 struct sr_chunk *schunk; 5057 struct sd_softc *sd; 5058 struct aes_xts_ctx ctx; 5059 struct sr_softc *sc; 5060 struct device *dv; 5061 daddr_t key_blkno; 5062 uint32_t sub_raidoff; /* ofs of sr part in underlying dev */ 5063 struct disklabel dl; 5064 struct partition *pp; 5065 size_t i, j; 5066 u_char iv[8]; 5067 5068 /* 5069 * In HIB_INIT, we are passed the swap partition size and offset 5070 * in 'size' and 'blkno' respectively. These are relative to the 5071 * start of the softraid partition, and we need to save these 5072 * for later translation to the underlying device's layout. 5073 */ 5074 if (op == HIB_INIT) { 5075 dv = disk_lookup(&sd_cd, DISKUNIT(dev)); 5076 sd = (struct sd_softc *)dv; 5077 sc = (struct sr_softc *)dv->dv_parent->dv_parent; 5078 5079 /* 5080 * Look up the sr discipline. This is used to determine 5081 * if we are SR crypto and what the underlying device is. 5082 */ 5083 my->srd = sc->sc_targets[sd->sc_link->target]; 5084 DNPRINTF(SR_D_MISC, "sr_hibernate_io: discipline is %s\n", 5085 my->srd->sd_name); 5086 if (strncmp(my->srd->sd_name, "CRYPTO", 5087 sizeof(my->srd->sd_name))) 5088 return (ENOTSUP); 5089 5090 /* Find the underlying device */ 5091 schunk = my->srd->sd_vol.sv_chunks[0]; 5092 my->subdev = schunk->src_dev_mm; 5093 5094 /* 5095 * Find the appropriate underlying device side effect free 5096 * I/O function, based on the type of device it is. 5097 */ 5098 my->subfn = get_hibernate_io_function(my->subdev); 5099 if (!my->subfn) 5100 return (ENODEV); 5101 5102 /* 5103 * Find blkno where this raid partition starts on 5104 * the underlying disk. 5105 */ 5106 dl_ret = disk_readlabel(&dl, my->subdev, errstr, 5107 sizeof(errstr)); 5108 if (dl_ret) { 5109 printf("Hibernate error reading disklabel: %s\n", dl_ret); 5110 return (ENOTSUP); 5111 } 5112 5113 pp = &dl.d_partitions[DISKPART(my->subdev)]; 5114 if (pp->p_fstype != FS_RAID || DL_GETPSIZE(pp) == 0) 5115 return (ENOTSUP); 5116 5117 /* Find the blkno of the SR part in the underlying device */ 5118 sub_raidoff = my->srd->sd_meta->ssd_data_blkno + 5119 DL_SECTOBLK(&dl, DL_GETPOFFSET(pp)); 5120 DNPRINTF(SR_D_MISC,"sr_hibernate_io: blk trans ofs: %d blks\n", 5121 sub_raidoff); 5122 5123 /* Save the blkno of the swap partition in the SR disk */ 5124 my->sr_swapoff = blkno; 5125 5126 /* Initialize the sub-device */ 5127 return my->subfn(my->subdev, sub_raidoff + blkno, 5128 addr, size, op, page); 5129 } 5130 5131 /* Hibernate only uses (and we only support) writes */ 5132 if (op != HIB_W) 5133 return (ENOTSUP); 5134 5135 /* 5136 * Blocks act as the IV for the encryption. These block numbers 5137 * are relative to the start of the sr partition, but the 'blkno' 5138 * passed above is relative to the start of the swap partition 5139 * inside the sr partition, so bias appropriately. 5140 */ 5141 key_blkno = my->sr_swapoff + blkno; 5142 5143 /* Process each disk block one at a time. */ 5144 for (i = 0; i < size; i += DEV_BSIZE) { 5145 int res; 5146 5147 bzero(&ctx, sizeof(ctx)); 5148 5149 /* 5150 * Set encryption key (from the sr discipline stashed 5151 * during HIB_INIT. This code is based on the softraid 5152 * bootblock code. 5153 */ 5154 aes_xts_setkey(&ctx, my->srd->mds.mdd_crypto.scr_key[0], 64); 5155 /* We encrypt DEV_BSIZE bytes at a time in my->buf */ 5156 memcpy(my->buf, ((char *)addr) + i, DEV_BSIZE); 5157 5158 /* Block number is the IV */ 5159 memcpy(&iv, &key_blkno, sizeof(key_blkno)); 5160 aes_xts_reinit(&ctx, iv); 5161 5162 /* Encrypt DEV_BSIZE bytes, AES_XTS_BLOCKSIZE bytes at a time */ 5163 for (j = 0; j < DEV_BSIZE; j += AES_XTS_BLOCKSIZE) 5164 aes_xts_encrypt(&ctx, my->buf + j); 5165 5166 /* 5167 * Write one block out from my->buf to the underlying device 5168 * using its own side-effect free I/O function. 5169 */ 5170 res = my->subfn(my->subdev, blkno + (i / DEV_BSIZE), 5171 (vaddr_t)(my->buf), DEV_BSIZE, op, page); 5172 if (res != 0) 5173 return (res); 5174 key_blkno++; 5175 } 5176 return (0); 5177 } 5178 #endif /* HIBERNATE */ 5179