1 /* $OpenBSD: softraid.c,v 1.389 2017/12/21 07:29:15 tb 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 void sr_minphys(struct buf *, struct scsi_link *); 97 int sr_scsi_probe(struct scsi_link *); 98 void sr_copy_internal_data(struct scsi_xfer *, 99 void *, size_t); 100 int sr_scsi_ioctl(struct scsi_link *, u_long, 101 caddr_t, int); 102 int sr_bio_ioctl(struct device *, u_long, caddr_t); 103 int sr_bio_handler(struct sr_softc *, 104 struct sr_discipline *, u_long, struct bio *); 105 int sr_ioctl_inq(struct sr_softc *, struct bioc_inq *); 106 int sr_ioctl_vol(struct sr_softc *, struct bioc_vol *); 107 int sr_ioctl_disk(struct sr_softc *, struct bioc_disk *); 108 int sr_ioctl_setstate(struct sr_softc *, 109 struct bioc_setstate *); 110 int sr_ioctl_createraid(struct sr_softc *, 111 struct bioc_createraid *, int, void *); 112 int sr_ioctl_deleteraid(struct sr_softc *, 113 struct sr_discipline *, struct bioc_deleteraid *); 114 int sr_ioctl_discipline(struct sr_softc *, 115 struct sr_discipline *, struct bioc_discipline *); 116 int sr_ioctl_installboot(struct sr_softc *, 117 struct sr_discipline *, struct bioc_installboot *); 118 void sr_chunks_unwind(struct sr_softc *, 119 struct sr_chunk_head *); 120 void sr_discipline_free(struct sr_discipline *); 121 void sr_discipline_shutdown(struct sr_discipline *, int, int); 122 int sr_discipline_init(struct sr_discipline *, int); 123 int sr_alloc_resources(struct sr_discipline *); 124 void sr_free_resources(struct sr_discipline *); 125 void sr_set_chunk_state(struct sr_discipline *, int, int); 126 void sr_set_vol_state(struct sr_discipline *); 127 128 /* utility functions */ 129 void sr_shutdown(int); 130 void sr_uuid_generate(struct sr_uuid *); 131 char *sr_uuid_format(struct sr_uuid *); 132 void sr_uuid_print(struct sr_uuid *, int); 133 void sr_checksum_print(u_int8_t *); 134 int sr_boot_assembly(struct sr_softc *); 135 int sr_already_assembled(struct sr_discipline *); 136 int sr_hotspare(struct sr_softc *, dev_t); 137 void sr_hotspare_rebuild(struct sr_discipline *); 138 int sr_rebuild_init(struct sr_discipline *, dev_t, int); 139 void sr_rebuild_start(void *); 140 void sr_rebuild_thread(void *); 141 void sr_rebuild(struct sr_discipline *); 142 void sr_roam_chunks(struct sr_discipline *); 143 int sr_chunk_in_use(struct sr_softc *, dev_t); 144 int sr_rw(struct sr_softc *, dev_t, char *, size_t, 145 daddr_t, long); 146 void sr_wu_done_callback(void *); 147 148 /* don't include these on RAMDISK */ 149 #ifndef SMALL_KERNEL 150 void sr_sensors_refresh(void *); 151 int sr_sensors_create(struct sr_discipline *); 152 void sr_sensors_delete(struct sr_discipline *); 153 #endif 154 155 /* metadata */ 156 int sr_meta_probe(struct sr_discipline *, dev_t *, int); 157 int sr_meta_attach(struct sr_discipline *, int, int); 158 int sr_meta_rw(struct sr_discipline *, dev_t, void *, long); 159 int sr_meta_clear(struct sr_discipline *); 160 void sr_meta_init(struct sr_discipline *, int, int); 161 void sr_meta_init_complete(struct sr_discipline *); 162 void sr_meta_opt_handler(struct sr_discipline *, 163 struct sr_meta_opt_hdr *); 164 165 /* hotplug magic */ 166 void sr_disk_attach(struct disk *, int); 167 168 struct sr_hotplug_list { 169 void (*sh_hotplug)(struct sr_discipline *, 170 struct disk *, int); 171 struct sr_discipline *sh_sd; 172 173 SLIST_ENTRY(sr_hotplug_list) shl_link; 174 }; 175 SLIST_HEAD(sr_hotplug_list_head, sr_hotplug_list); 176 177 struct sr_hotplug_list_head sr_hotplug_callbacks; 178 extern void (*softraid_disk_attach)(struct disk *, int); 179 180 /* scsi glue */ 181 struct scsi_adapter sr_switch = { 182 sr_scsi_cmd, sr_minphys, sr_scsi_probe, NULL, sr_scsi_ioctl 183 }; 184 185 /* native metadata format */ 186 int sr_meta_native_bootprobe(struct sr_softc *, dev_t, 187 struct sr_boot_chunk_head *); 188 #define SR_META_NOTCLAIMED (0) 189 #define SR_META_CLAIMED (1) 190 int sr_meta_native_probe(struct sr_softc *, 191 struct sr_chunk *); 192 int sr_meta_native_attach(struct sr_discipline *, int); 193 int sr_meta_native_write(struct sr_discipline *, dev_t, 194 struct sr_metadata *,void *); 195 196 #ifdef SR_DEBUG 197 void sr_meta_print(struct sr_metadata *); 198 #else 199 #define sr_meta_print(m) 200 #endif 201 202 /* the metadata driver should remain stateless */ 203 struct sr_meta_driver { 204 daddr_t smd_offset; /* metadata location */ 205 u_int32_t smd_size; /* size of metadata */ 206 207 int (*smd_probe)(struct sr_softc *, 208 struct sr_chunk *); 209 int (*smd_attach)(struct sr_discipline *, int); 210 int (*smd_detach)(struct sr_discipline *); 211 int (*smd_read)(struct sr_discipline *, dev_t, 212 struct sr_metadata *, void *); 213 int (*smd_write)(struct sr_discipline *, dev_t, 214 struct sr_metadata *, void *); 215 int (*smd_validate)(struct sr_discipline *, 216 struct sr_metadata *, void *); 217 } smd[] = { 218 { SR_META_OFFSET, SR_META_SIZE * DEV_BSIZE, 219 sr_meta_native_probe, sr_meta_native_attach, NULL, 220 sr_meta_native_read, sr_meta_native_write, NULL }, 221 { 0, 0, NULL, NULL, NULL, NULL } 222 }; 223 224 int 225 sr_meta_attach(struct sr_discipline *sd, int chunk_no, int force) 226 { 227 struct sr_softc *sc = sd->sd_sc; 228 struct sr_chunk_head *cl; 229 struct sr_chunk *ch_entry, *chunk1, *chunk2; 230 int rv = 1, i = 0; 231 232 DNPRINTF(SR_D_META, "%s: sr_meta_attach(%d)\n", DEVNAME(sc), chunk_no); 233 234 /* in memory copy of metadata */ 235 sd->sd_meta = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, 236 M_ZERO | M_NOWAIT); 237 if (!sd->sd_meta) { 238 sr_error(sc, "could not allocate memory for metadata"); 239 goto bad; 240 } 241 242 if (sd->sd_meta_type != SR_META_F_NATIVE) { 243 /* in memory copy of foreign metadata */ 244 sd->sd_meta_foreign = malloc(smd[sd->sd_meta_type].smd_size, 245 M_DEVBUF, M_ZERO | M_NOWAIT); 246 if (!sd->sd_meta_foreign) { 247 /* unwind frees sd_meta */ 248 sr_error(sc, "could not allocate memory for foreign " 249 "metadata"); 250 goto bad; 251 } 252 } 253 254 /* we have a valid list now create an array index */ 255 cl = &sd->sd_vol.sv_chunk_list; 256 sd->sd_vol.sv_chunks = mallocarray(chunk_no, sizeof(struct sr_chunk *), 257 M_DEVBUF, M_WAITOK | M_ZERO); 258 259 /* fill out chunk array */ 260 i = 0; 261 SLIST_FOREACH(ch_entry, cl, src_link) 262 sd->sd_vol.sv_chunks[i++] = ch_entry; 263 264 /* attach metadata */ 265 if (smd[sd->sd_meta_type].smd_attach(sd, force)) 266 goto bad; 267 268 /* Force chunks into correct order now that metadata is attached. */ 269 SLIST_INIT(cl); 270 for (i = 0; i < chunk_no; i++) { 271 ch_entry = sd->sd_vol.sv_chunks[i]; 272 chunk2 = NULL; 273 SLIST_FOREACH(chunk1, cl, src_link) { 274 if (chunk1->src_meta.scmi.scm_chunk_id > 275 ch_entry->src_meta.scmi.scm_chunk_id) 276 break; 277 chunk2 = chunk1; 278 } 279 if (chunk2 == NULL) 280 SLIST_INSERT_HEAD(cl, ch_entry, src_link); 281 else 282 SLIST_INSERT_AFTER(chunk2, ch_entry, src_link); 283 } 284 i = 0; 285 SLIST_FOREACH(ch_entry, cl, src_link) 286 sd->sd_vol.sv_chunks[i++] = ch_entry; 287 288 rv = 0; 289 bad: 290 return (rv); 291 } 292 293 int 294 sr_meta_probe(struct sr_discipline *sd, dev_t *dt, int no_chunk) 295 { 296 struct sr_softc *sc = sd->sd_sc; 297 struct vnode *vn; 298 struct sr_chunk *ch_entry, *ch_prev = NULL; 299 struct sr_chunk_head *cl; 300 char devname[32]; 301 int i, d, type, found, prevf, error; 302 dev_t dev; 303 304 DNPRINTF(SR_D_META, "%s: sr_meta_probe(%d)\n", DEVNAME(sc), no_chunk); 305 306 if (no_chunk == 0) 307 goto unwind; 308 309 cl = &sd->sd_vol.sv_chunk_list; 310 311 for (d = 0, prevf = SR_META_F_INVALID; d < no_chunk; d++) { 312 ch_entry = malloc(sizeof(struct sr_chunk), M_DEVBUF, 313 M_WAITOK | M_ZERO); 314 /* keep disks in user supplied order */ 315 if (ch_prev) 316 SLIST_INSERT_AFTER(ch_prev, ch_entry, src_link); 317 else 318 SLIST_INSERT_HEAD(cl, ch_entry, src_link); 319 ch_prev = ch_entry; 320 dev = dt[d]; 321 ch_entry->src_dev_mm = dev; 322 323 if (dev == NODEV) { 324 ch_entry->src_meta.scm_status = BIOC_SDOFFLINE; 325 continue; 326 } else { 327 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 328 if (bdevvp(dev, &vn)) { 329 sr_error(sc, "sr_meta_probe: cannot allocate " 330 "vnode"); 331 goto unwind; 332 } 333 334 /* 335 * XXX leaving dev open for now; move this to attach 336 * and figure out the open/close dance for unwind. 337 */ 338 error = VOP_OPEN(vn, FREAD | FWRITE, NOCRED, curproc); 339 if (error) { 340 DNPRINTF(SR_D_META,"%s: sr_meta_probe can't " 341 "open %s\n", DEVNAME(sc), devname); 342 vput(vn); 343 goto unwind; 344 } 345 346 strlcpy(ch_entry->src_devname, devname, 347 sizeof(ch_entry->src_devname)); 348 ch_entry->src_vn = vn; 349 } 350 351 /* determine if this is a device we understand */ 352 for (i = 0, found = SR_META_F_INVALID; smd[i].smd_probe; i++) { 353 type = smd[i].smd_probe(sc, ch_entry); 354 if (type == SR_META_F_INVALID) 355 continue; 356 else { 357 found = type; 358 break; 359 } 360 } 361 362 if (found == SR_META_F_INVALID) 363 goto unwind; 364 if (prevf == SR_META_F_INVALID) 365 prevf = found; 366 if (prevf != found) { 367 DNPRINTF(SR_D_META, "%s: prevf != found\n", 368 DEVNAME(sc)); 369 goto unwind; 370 } 371 } 372 373 return (prevf); 374 unwind: 375 return (SR_META_F_INVALID); 376 } 377 378 void 379 sr_meta_getdevname(struct sr_softc *sc, dev_t dev, char *buf, int size) 380 { 381 int maj, unit, part; 382 char *name; 383 384 DNPRINTF(SR_D_META, "%s: sr_meta_getdevname(%p, %d)\n", 385 DEVNAME(sc), buf, size); 386 387 if (!buf) 388 return; 389 390 maj = major(dev); 391 part = DISKPART(dev); 392 unit = DISKUNIT(dev); 393 394 name = findblkname(maj); 395 if (name == NULL) 396 return; 397 398 snprintf(buf, size, "%s%d%c", name, unit, part + 'a'); 399 } 400 401 int 402 sr_rw(struct sr_softc *sc, dev_t dev, char *buf, size_t size, daddr_t blkno, 403 long flags) 404 { 405 struct vnode *vp; 406 struct buf b; 407 size_t bufsize, dma_bufsize; 408 int rv = 1; 409 char *dma_buf; 410 411 DNPRINTF(SR_D_MISC, "%s: sr_rw(0x%x, %p, %zu, %lld 0x%lx)\n", 412 DEVNAME(sc), dev, buf, size, (long long)blkno, flags); 413 414 dma_bufsize = (size > MAXPHYS) ? MAXPHYS : size; 415 dma_buf = dma_alloc(dma_bufsize, PR_WAITOK); 416 417 if (bdevvp(dev, &vp)) { 418 printf("%s: sr_rw: failed to allocate vnode\n", DEVNAME(sc)); 419 goto done; 420 } 421 422 while (size > 0) { 423 DNPRINTF(SR_D_MISC, "%s: dma_buf %p, size %zu, blkno %lld)\n", 424 DEVNAME(sc), dma_buf, size, (long long)blkno); 425 426 bufsize = (size > MAXPHYS) ? MAXPHYS : size; 427 if (flags == B_WRITE) 428 memcpy(dma_buf, buf, bufsize); 429 430 bzero(&b, sizeof(b)); 431 b.b_flags = flags | B_PHYS; 432 b.b_proc = curproc; 433 b.b_dev = dev; 434 b.b_iodone = NULL; 435 b.b_error = 0; 436 b.b_blkno = blkno; 437 b.b_data = dma_buf; 438 b.b_bcount = bufsize; 439 b.b_bufsize = bufsize; 440 b.b_resid = bufsize; 441 b.b_vp = vp; 442 443 if ((b.b_flags & B_READ) == 0) 444 vp->v_numoutput++; 445 446 LIST_INIT(&b.b_dep); 447 VOP_STRATEGY(&b); 448 biowait(&b); 449 450 if (b.b_flags & B_ERROR) { 451 printf("%s: I/O error %d on dev 0x%x at block %llu\n", 452 DEVNAME(sc), b.b_error, dev, b.b_blkno); 453 goto done; 454 } 455 456 if (flags == B_READ) 457 memcpy(buf, dma_buf, bufsize); 458 459 size -= bufsize; 460 buf += bufsize; 461 blkno += howmany(bufsize, DEV_BSIZE); 462 } 463 464 rv = 0; 465 466 done: 467 if (vp) 468 vput(vp); 469 470 dma_free(dma_buf, dma_bufsize); 471 472 return (rv); 473 } 474 475 int 476 sr_meta_rw(struct sr_discipline *sd, dev_t dev, void *md, long flags) 477 { 478 int rv = 1; 479 480 DNPRINTF(SR_D_META, "%s: sr_meta_rw(0x%x, %p, 0x%lx)\n", 481 DEVNAME(sd->sd_sc), dev, md, flags); 482 483 if (md == NULL) { 484 printf("%s: sr_meta_rw: invalid metadata pointer\n", 485 DEVNAME(sd->sd_sc)); 486 goto done; 487 } 488 489 rv = sr_rw(sd->sd_sc, dev, md, SR_META_SIZE * DEV_BSIZE, 490 SR_META_OFFSET, flags); 491 492 done: 493 return (rv); 494 } 495 496 int 497 sr_meta_clear(struct sr_discipline *sd) 498 { 499 struct sr_softc *sc = sd->sd_sc; 500 struct sr_chunk_head *cl = &sd->sd_vol.sv_chunk_list; 501 struct sr_chunk *ch_entry; 502 void *m; 503 int rv = 1; 504 505 DNPRINTF(SR_D_META, "%s: sr_meta_clear\n", DEVNAME(sc)); 506 507 if (sd->sd_meta_type != SR_META_F_NATIVE) { 508 sr_error(sc, "cannot clear foreign metadata"); 509 goto done; 510 } 511 512 m = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, M_WAITOK | M_ZERO); 513 SLIST_FOREACH(ch_entry, cl, src_link) { 514 if (sr_meta_native_write(sd, ch_entry->src_dev_mm, m, NULL)) { 515 /* XXX mark disk offline */ 516 DNPRINTF(SR_D_META, "%s: sr_meta_clear failed to " 517 "clear %s\n", DEVNAME(sc), ch_entry->src_devname); 518 rv++; 519 continue; 520 } 521 bzero(&ch_entry->src_meta, sizeof(ch_entry->src_meta)); 522 } 523 524 bzero(sd->sd_meta, SR_META_SIZE * DEV_BSIZE); 525 526 free(m, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 527 rv = 0; 528 done: 529 return (rv); 530 } 531 532 void 533 sr_meta_init(struct sr_discipline *sd, int level, int no_chunk) 534 { 535 struct sr_softc *sc = sd->sd_sc; 536 struct sr_metadata *sm = sd->sd_meta; 537 struct sr_chunk_head *cl = &sd->sd_vol.sv_chunk_list; 538 struct sr_meta_chunk *scm; 539 struct sr_chunk *chunk; 540 int cid = 0; 541 u_int64_t max_chunk_sz = 0, min_chunk_sz = 0; 542 u_int32_t secsize = DEV_BSIZE; 543 544 DNPRINTF(SR_D_META, "%s: sr_meta_init\n", DEVNAME(sc)); 545 546 if (!sm) 547 return; 548 549 /* Initialise volume metadata. */ 550 sm->ssdi.ssd_magic = SR_MAGIC; 551 sm->ssdi.ssd_version = SR_META_VERSION; 552 sm->ssdi.ssd_vol_flags = sd->sd_meta_flags; 553 sm->ssdi.ssd_volid = 0; 554 sm->ssdi.ssd_chunk_no = no_chunk; 555 sm->ssdi.ssd_level = level; 556 557 sm->ssd_data_blkno = SR_DATA_OFFSET; 558 sm->ssd_ondisk = 0; 559 560 sr_uuid_generate(&sm->ssdi.ssd_uuid); 561 562 /* Initialise chunk metadata and get min/max chunk sizes & secsize. */ 563 SLIST_FOREACH(chunk, cl, src_link) { 564 scm = &chunk->src_meta; 565 scm->scmi.scm_size = chunk->src_size; 566 scm->scmi.scm_chunk_id = cid++; 567 scm->scm_status = BIOC_SDONLINE; 568 scm->scmi.scm_volid = 0; 569 strlcpy(scm->scmi.scm_devname, chunk->src_devname, 570 sizeof(scm->scmi.scm_devname)); 571 memcpy(&scm->scmi.scm_uuid, &sm->ssdi.ssd_uuid, 572 sizeof(scm->scmi.scm_uuid)); 573 sr_checksum(sc, scm, &scm->scm_checksum, 574 sizeof(scm->scm_checksum)); 575 576 if (min_chunk_sz == 0) 577 min_chunk_sz = scm->scmi.scm_size; 578 if (chunk->src_secsize > secsize) 579 secsize = chunk->src_secsize; 580 min_chunk_sz = MIN(min_chunk_sz, scm->scmi.scm_size); 581 max_chunk_sz = MAX(max_chunk_sz, scm->scmi.scm_size); 582 } 583 584 sm->ssdi.ssd_secsize = secsize; 585 586 /* Equalize chunk sizes. */ 587 SLIST_FOREACH(chunk, cl, src_link) 588 chunk->src_meta.scmi.scm_coerced_size = min_chunk_sz; 589 590 sd->sd_vol.sv_chunk_minsz = min_chunk_sz; 591 sd->sd_vol.sv_chunk_maxsz = max_chunk_sz; 592 } 593 594 void 595 sr_meta_init_complete(struct sr_discipline *sd) 596 { 597 #ifdef SR_DEBUG 598 struct sr_softc *sc = sd->sd_sc; 599 #endif 600 struct sr_metadata *sm = sd->sd_meta; 601 602 DNPRINTF(SR_D_META, "%s: sr_meta_complete\n", DEVNAME(sc)); 603 604 /* Complete initialisation of volume metadata. */ 605 strlcpy(sm->ssdi.ssd_vendor, "OPENBSD", sizeof(sm->ssdi.ssd_vendor)); 606 snprintf(sm->ssdi.ssd_product, sizeof(sm->ssdi.ssd_product), 607 "SR %s", sd->sd_name); 608 snprintf(sm->ssdi.ssd_revision, sizeof(sm->ssdi.ssd_revision), 609 "%03d", sm->ssdi.ssd_version); 610 } 611 612 void 613 sr_meta_opt_handler(struct sr_discipline *sd, struct sr_meta_opt_hdr *om) 614 { 615 if (om->som_type != SR_OPT_BOOT) 616 panic("unknown optional metadata type"); 617 } 618 619 void 620 sr_meta_save_callback(void *xsd) 621 { 622 struct sr_discipline *sd = xsd; 623 int s; 624 625 s = splbio(); 626 627 if (sr_meta_save(sd, SR_META_DIRTY)) 628 printf("%s: save metadata failed\n", DEVNAME(sd->sd_sc)); 629 630 sd->sd_must_flush = 0; 631 splx(s); 632 } 633 634 int 635 sr_meta_save(struct sr_discipline *sd, u_int32_t flags) 636 { 637 struct sr_softc *sc = sd->sd_sc; 638 struct sr_metadata *sm = sd->sd_meta, *m; 639 struct sr_meta_driver *s; 640 struct sr_chunk *src; 641 struct sr_meta_chunk *cm; 642 struct sr_workunit wu; 643 struct sr_meta_opt_hdr *omh; 644 struct sr_meta_opt_item *omi; 645 int i; 646 647 DNPRINTF(SR_D_META, "%s: sr_meta_save %s\n", 648 DEVNAME(sc), sd->sd_meta->ssd_devname); 649 650 if (!sm) { 651 printf("%s: no in memory copy of metadata\n", DEVNAME(sc)); 652 goto bad; 653 } 654 655 /* meta scratchpad */ 656 s = &smd[sd->sd_meta_type]; 657 m = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, M_ZERO | M_NOWAIT); 658 if (!m) { 659 printf("%s: could not allocate metadata scratch area\n", 660 DEVNAME(sc)); 661 goto bad; 662 } 663 664 /* from here on out metadata is updated */ 665 restart: 666 sm->ssd_ondisk++; 667 sm->ssd_meta_flags = flags; 668 memcpy(m, sm, sizeof(*m)); 669 670 /* Chunk metadata. */ 671 cm = (struct sr_meta_chunk *)(m + 1); 672 for (i = 0; i < sm->ssdi.ssd_chunk_no; i++) { 673 src = sd->sd_vol.sv_chunks[i]; 674 memcpy(cm, &src->src_meta, sizeof(*cm)); 675 cm++; 676 } 677 678 /* Optional metadata. */ 679 omh = (struct sr_meta_opt_hdr *)(cm); 680 SLIST_FOREACH(omi, &sd->sd_meta_opt, omi_link) { 681 DNPRINTF(SR_D_META, "%s: saving optional metadata type %u with " 682 "length %u\n", DEVNAME(sc), omi->omi_som->som_type, 683 omi->omi_som->som_length); 684 bzero(&omi->omi_som->som_checksum, MD5_DIGEST_LENGTH); 685 sr_checksum(sc, omi->omi_som, &omi->omi_som->som_checksum, 686 omi->omi_som->som_length); 687 memcpy(omh, omi->omi_som, omi->omi_som->som_length); 688 omh = (struct sr_meta_opt_hdr *)((u_int8_t *)omh + 689 omi->omi_som->som_length); 690 } 691 692 for (i = 0; i < sm->ssdi.ssd_chunk_no; i++) { 693 src = sd->sd_vol.sv_chunks[i]; 694 695 /* skip disks that are offline */ 696 if (src->src_meta.scm_status == BIOC_SDOFFLINE) 697 continue; 698 699 /* calculate metadata checksum for correct chunk */ 700 m->ssdi.ssd_chunk_id = i; 701 sr_checksum(sc, m, &m->ssd_checksum, 702 sizeof(struct sr_meta_invariant)); 703 704 #ifdef SR_DEBUG 705 DNPRINTF(SR_D_META, "%s: sr_meta_save %s: volid: %d " 706 "chunkid: %d checksum: ", 707 DEVNAME(sc), src->src_meta.scmi.scm_devname, 708 m->ssdi.ssd_volid, m->ssdi.ssd_chunk_id); 709 710 if (sr_debug & SR_D_META) 711 sr_checksum_print((u_int8_t *)&m->ssd_checksum); 712 DNPRINTF(SR_D_META, "\n"); 713 sr_meta_print(m); 714 #endif 715 716 /* translate and write to disk */ 717 if (s->smd_write(sd, src->src_dev_mm, m, NULL /* XXX */)) { 718 printf("%s: could not write metadata to %s\n", 719 DEVNAME(sc), src->src_devname); 720 /* restart the meta write */ 721 src->src_meta.scm_status = BIOC_SDOFFLINE; 722 /* XXX recalculate volume status */ 723 goto restart; 724 } 725 } 726 727 /* not all disciplines have sync */ 728 if (sd->sd_scsi_sync) { 729 bzero(&wu, sizeof(wu)); 730 wu.swu_flags |= SR_WUF_FAKE; 731 wu.swu_dis = sd; 732 sd->sd_scsi_sync(&wu); 733 } 734 free(m, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 735 return (0); 736 bad: 737 return (1); 738 } 739 740 int 741 sr_meta_read(struct sr_discipline *sd) 742 { 743 struct sr_softc *sc = sd->sd_sc; 744 struct sr_chunk_head *cl = &sd->sd_vol.sv_chunk_list; 745 struct sr_metadata *sm; 746 struct sr_chunk *ch_entry; 747 struct sr_meta_chunk *cp; 748 struct sr_meta_driver *s; 749 void *fm = NULL; 750 int no_disk = 0, got_meta = 0; 751 752 DNPRINTF(SR_D_META, "%s: sr_meta_read\n", DEVNAME(sc)); 753 754 sm = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, M_WAITOK | M_ZERO); 755 s = &smd[sd->sd_meta_type]; 756 if (sd->sd_meta_type != SR_META_F_NATIVE) 757 fm = malloc(s->smd_size, M_DEVBUF, M_WAITOK | M_ZERO); 758 759 cp = (struct sr_meta_chunk *)(sm + 1); 760 SLIST_FOREACH(ch_entry, cl, src_link) { 761 /* skip disks that are offline */ 762 if (ch_entry->src_meta.scm_status == BIOC_SDOFFLINE) { 763 DNPRINTF(SR_D_META, 764 "%s: %s chunk marked offline, spoofing status\n", 765 DEVNAME(sc), ch_entry->src_devname); 766 cp++; /* adjust chunk pointer to match failure */ 767 continue; 768 } else if (s->smd_read(sd, ch_entry->src_dev_mm, sm, fm)) { 769 /* read and translate */ 770 /* XXX mark chunk offline, elsewhere!! */ 771 ch_entry->src_meta.scm_status = BIOC_SDOFFLINE; 772 cp++; /* adjust chunk pointer to match failure */ 773 DNPRINTF(SR_D_META, "%s: sr_meta_read failed\n", 774 DEVNAME(sc)); 775 continue; 776 } 777 778 if (sm->ssdi.ssd_magic != SR_MAGIC) { 779 DNPRINTF(SR_D_META, "%s: sr_meta_read !SR_MAGIC\n", 780 DEVNAME(sc)); 781 continue; 782 } 783 784 /* validate metadata */ 785 if (sr_meta_validate(sd, ch_entry->src_dev_mm, sm, fm)) { 786 DNPRINTF(SR_D_META, "%s: invalid metadata\n", 787 DEVNAME(sc)); 788 no_disk = -1; 789 goto done; 790 } 791 792 /* assume first chunk contains metadata */ 793 if (got_meta == 0) { 794 sr_meta_opt_load(sc, sm, &sd->sd_meta_opt); 795 memcpy(sd->sd_meta, sm, sizeof(*sd->sd_meta)); 796 got_meta = 1; 797 } 798 799 memcpy(&ch_entry->src_meta, cp, sizeof(ch_entry->src_meta)); 800 801 no_disk++; 802 cp++; 803 } 804 805 free(sm, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 806 free(fm, M_DEVBUF, s->smd_size); 807 808 done: 809 DNPRINTF(SR_D_META, "%s: sr_meta_read found %d parts\n", DEVNAME(sc), 810 no_disk); 811 return (no_disk); 812 } 813 814 void 815 sr_meta_opt_load(struct sr_softc *sc, struct sr_metadata *sm, 816 struct sr_meta_opt_head *som) 817 { 818 struct sr_meta_opt_hdr *omh; 819 struct sr_meta_opt_item *omi; 820 u_int8_t checksum[MD5_DIGEST_LENGTH]; 821 int i; 822 823 /* Process optional metadata. */ 824 omh = (struct sr_meta_opt_hdr *)((u_int8_t *)(sm + 1) + 825 sizeof(struct sr_meta_chunk) * sm->ssdi.ssd_chunk_no); 826 for (i = 0; i < sm->ssdi.ssd_opt_no; i++) { 827 828 omi = malloc(sizeof(struct sr_meta_opt_item), M_DEVBUF, 829 M_WAITOK | M_ZERO); 830 SLIST_INSERT_HEAD(som, omi, omi_link); 831 832 if (omh->som_length == 0) { 833 834 /* Load old fixed length optional metadata. */ 835 DNPRINTF(SR_D_META, "%s: old optional metadata of type " 836 "%u\n", DEVNAME(sc), omh->som_type); 837 838 /* Validate checksum. */ 839 sr_checksum(sc, (void *)omh, &checksum, 840 SR_OLD_META_OPT_SIZE - MD5_DIGEST_LENGTH); 841 if (bcmp(&checksum, (void *)omh + SR_OLD_META_OPT_MD5, 842 sizeof(checksum))) 843 panic("%s: invalid optional metadata " 844 "checksum", DEVNAME(sc)); 845 846 /* Determine correct length. */ 847 switch (omh->som_type) { 848 case SR_OPT_CRYPTO: 849 omh->som_length = sizeof(struct sr_meta_crypto); 850 break; 851 case SR_OPT_BOOT: 852 omh->som_length = sizeof(struct sr_meta_boot); 853 break; 854 case SR_OPT_KEYDISK: 855 omh->som_length = 856 sizeof(struct sr_meta_keydisk); 857 break; 858 default: 859 panic("unknown old optional metadata " 860 "type %u\n", omh->som_type); 861 } 862 863 omi->omi_som = malloc(omh->som_length, M_DEVBUF, 864 M_WAITOK | M_ZERO); 865 memcpy((u_int8_t *)omi->omi_som + sizeof(*omi->omi_som), 866 (u_int8_t *)omh + SR_OLD_META_OPT_OFFSET, 867 omh->som_length - sizeof(*omi->omi_som)); 868 omi->omi_som->som_type = omh->som_type; 869 omi->omi_som->som_length = omh->som_length; 870 871 omh = (struct sr_meta_opt_hdr *)((void *)omh + 872 SR_OLD_META_OPT_SIZE); 873 } else { 874 875 /* Load variable length optional metadata. */ 876 DNPRINTF(SR_D_META, "%s: optional metadata of type %u, " 877 "length %u\n", DEVNAME(sc), omh->som_type, 878 omh->som_length); 879 omi->omi_som = malloc(omh->som_length, M_DEVBUF, 880 M_WAITOK | M_ZERO); 881 memcpy(omi->omi_som, omh, omh->som_length); 882 883 /* Validate checksum. */ 884 memcpy(&checksum, &omi->omi_som->som_checksum, 885 MD5_DIGEST_LENGTH); 886 bzero(&omi->omi_som->som_checksum, MD5_DIGEST_LENGTH); 887 sr_checksum(sc, omi->omi_som, 888 &omi->omi_som->som_checksum, omh->som_length); 889 if (bcmp(&checksum, &omi->omi_som->som_checksum, 890 sizeof(checksum))) 891 panic("%s: invalid optional metadata checksum", 892 DEVNAME(sc)); 893 894 omh = (struct sr_meta_opt_hdr *)((void *)omh + 895 omh->som_length); 896 } 897 } 898 } 899 900 int 901 sr_meta_validate(struct sr_discipline *sd, dev_t dev, struct sr_metadata *sm, 902 void *fm) 903 { 904 struct sr_softc *sc = sd->sd_sc; 905 struct sr_meta_driver *s; 906 #ifdef SR_DEBUG 907 struct sr_meta_chunk *mc; 908 #endif 909 u_int8_t checksum[MD5_DIGEST_LENGTH]; 910 char devname[32]; 911 int rv = 1; 912 913 DNPRINTF(SR_D_META, "%s: sr_meta_validate(%p)\n", DEVNAME(sc), sm); 914 915 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 916 917 s = &smd[sd->sd_meta_type]; 918 if (sd->sd_meta_type != SR_META_F_NATIVE) 919 if (s->smd_validate(sd, sm, fm)) { 920 sr_error(sc, "invalid foreign metadata"); 921 goto done; 922 } 923 924 /* 925 * at this point all foreign metadata has been translated to the native 926 * format and will be treated just like the native format 927 */ 928 929 if (sm->ssdi.ssd_magic != SR_MAGIC) { 930 sr_error(sc, "not valid softraid metadata"); 931 goto done; 932 } 933 934 /* Verify metadata checksum. */ 935 sr_checksum(sc, sm, &checksum, sizeof(struct sr_meta_invariant)); 936 if (bcmp(&checksum, &sm->ssd_checksum, sizeof(checksum))) { 937 sr_error(sc, "invalid metadata checksum"); 938 goto done; 939 } 940 941 /* Handle changes between versions. */ 942 if (sm->ssdi.ssd_version == 3) { 943 944 /* 945 * Version 3 - update metadata version and fix up data blkno 946 * value since this did not exist in version 3. 947 */ 948 if (sm->ssd_data_blkno == 0) 949 sm->ssd_data_blkno = SR_META_V3_DATA_OFFSET; 950 sm->ssdi.ssd_secsize = DEV_BSIZE; 951 952 } else if (sm->ssdi.ssd_version == 4) { 953 954 /* 955 * Version 4 - original metadata format did not store 956 * data blkno so fix this up if necessary. 957 */ 958 if (sm->ssd_data_blkno == 0) 959 sm->ssd_data_blkno = SR_DATA_OFFSET; 960 sm->ssdi.ssd_secsize = DEV_BSIZE; 961 962 } else if (sm->ssdi.ssd_version == 5) { 963 964 /* 965 * Version 5 - variable length optional metadata. Migration 966 * from earlier fixed length optional metadata is handled 967 * in sr_meta_read(). 968 */ 969 sm->ssdi.ssd_secsize = DEV_BSIZE; 970 971 } else if (sm->ssdi.ssd_version == SR_META_VERSION) { 972 973 /* 974 * Version 6 - store & report a sector size. 975 */ 976 977 } else { 978 979 sr_error(sc, "cannot read metadata version %u on %s, " 980 "expected version %u or earlier", 981 sm->ssdi.ssd_version, devname, SR_META_VERSION); 982 goto done; 983 984 } 985 986 /* Update version number and revision string. */ 987 sm->ssdi.ssd_version = SR_META_VERSION; 988 snprintf(sm->ssdi.ssd_revision, sizeof(sm->ssdi.ssd_revision), 989 "%03d", SR_META_VERSION); 990 991 #ifdef SR_DEBUG 992 /* warn if disk changed order */ 993 mc = (struct sr_meta_chunk *)(sm + 1); 994 if (strncmp(mc[sm->ssdi.ssd_chunk_id].scmi.scm_devname, devname, 995 sizeof(mc[sm->ssdi.ssd_chunk_id].scmi.scm_devname))) 996 DNPRINTF(SR_D_META, "%s: roaming device %s -> %s\n", 997 DEVNAME(sc), mc[sm->ssdi.ssd_chunk_id].scmi.scm_devname, 998 devname); 999 #endif 1000 1001 /* we have meta data on disk */ 1002 DNPRINTF(SR_D_META, "%s: sr_meta_validate valid metadata %s\n", 1003 DEVNAME(sc), devname); 1004 1005 rv = 0; 1006 done: 1007 return (rv); 1008 } 1009 1010 int 1011 sr_meta_native_bootprobe(struct sr_softc *sc, dev_t devno, 1012 struct sr_boot_chunk_head *bch) 1013 { 1014 struct vnode *vn; 1015 struct disklabel label; 1016 struct sr_metadata *md = NULL; 1017 struct sr_discipline *fake_sd = NULL; 1018 struct sr_boot_chunk *bc; 1019 char devname[32]; 1020 dev_t chrdev, rawdev; 1021 int error, i; 1022 int rv = SR_META_NOTCLAIMED; 1023 1024 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe\n", DEVNAME(sc)); 1025 1026 /* 1027 * Use character raw device to avoid SCSI complaints about missing 1028 * media on removable media devices. 1029 */ 1030 chrdev = blktochr(devno); 1031 rawdev = MAKEDISKDEV(major(chrdev), DISKUNIT(devno), RAW_PART); 1032 if (cdevvp(rawdev, &vn)) { 1033 sr_error(sc, "sr_meta_native_bootprobe: cannot allocate vnode"); 1034 goto done; 1035 } 1036 1037 /* open device */ 1038 error = VOP_OPEN(vn, FREAD, NOCRED, curproc); 1039 if (error) { 1040 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe open " 1041 "failed\n", DEVNAME(sc)); 1042 vput(vn); 1043 goto done; 1044 } 1045 1046 /* get disklabel */ 1047 error = VOP_IOCTL(vn, DIOCGDINFO, (caddr_t)&label, FREAD, NOCRED, 1048 curproc); 1049 if (error) { 1050 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe ioctl " 1051 "failed\n", DEVNAME(sc)); 1052 VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1053 vput(vn); 1054 goto done; 1055 } 1056 1057 /* we are done, close device */ 1058 error = VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1059 if (error) { 1060 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe close " 1061 "failed\n", DEVNAME(sc)); 1062 vput(vn); 1063 goto done; 1064 } 1065 vput(vn); 1066 1067 md = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, M_ZERO | M_NOWAIT); 1068 if (md == NULL) { 1069 sr_error(sc, "not enough memory for metadata buffer"); 1070 goto done; 1071 } 1072 1073 /* create fake sd to use utility functions */ 1074 fake_sd = malloc(sizeof(struct sr_discipline), M_DEVBUF, 1075 M_ZERO | M_NOWAIT); 1076 if (fake_sd == NULL) { 1077 sr_error(sc, "not enough memory for fake discipline"); 1078 goto done; 1079 } 1080 fake_sd->sd_sc = sc; 1081 fake_sd->sd_meta_type = SR_META_F_NATIVE; 1082 1083 for (i = 0; i < MAXPARTITIONS; i++) { 1084 if (label.d_partitions[i].p_fstype != FS_RAID) 1085 continue; 1086 1087 /* open partition */ 1088 rawdev = MAKEDISKDEV(major(devno), DISKUNIT(devno), i); 1089 if (bdevvp(rawdev, &vn)) { 1090 sr_error(sc, "sr_meta_native_bootprobe: cannot " 1091 "allocate vnode for partition"); 1092 goto done; 1093 } 1094 error = VOP_OPEN(vn, FREAD, NOCRED, curproc); 1095 if (error) { 1096 DNPRINTF(SR_D_META, "%s: sr_meta_native_bootprobe " 1097 "open failed, partition %d\n", 1098 DEVNAME(sc), i); 1099 vput(vn); 1100 continue; 1101 } 1102 1103 if (sr_meta_native_read(fake_sd, rawdev, md, NULL)) { 1104 sr_error(sc, "native bootprobe could not read native " 1105 "metadata"); 1106 VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1107 vput(vn); 1108 continue; 1109 } 1110 1111 /* are we a softraid partition? */ 1112 if (md->ssdi.ssd_magic != SR_MAGIC) { 1113 VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1114 vput(vn); 1115 continue; 1116 } 1117 1118 sr_meta_getdevname(sc, rawdev, devname, sizeof(devname)); 1119 if (sr_meta_validate(fake_sd, rawdev, md, NULL) == 0) { 1120 /* XXX fix M_WAITOK, this is boot time */ 1121 bc = malloc(sizeof(struct sr_boot_chunk), 1122 M_DEVBUF, M_WAITOK | M_ZERO); 1123 bc->sbc_metadata = malloc(sizeof(struct sr_metadata), 1124 M_DEVBUF, M_WAITOK | M_ZERO); 1125 memcpy(bc->sbc_metadata, md, sizeof(struct sr_metadata)); 1126 bc->sbc_mm = rawdev; 1127 SLIST_INSERT_HEAD(bch, bc, sbc_link); 1128 rv = SR_META_CLAIMED; 1129 } 1130 1131 /* we are done, close partition */ 1132 VOP_CLOSE(vn, FREAD, NOCRED, curproc); 1133 vput(vn); 1134 } 1135 1136 done: 1137 free(fake_sd, M_DEVBUF, sizeof(struct sr_discipline)); 1138 free(md, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 1139 1140 return (rv); 1141 } 1142 1143 int 1144 sr_boot_assembly(struct sr_softc *sc) 1145 { 1146 struct sr_boot_volume_head bvh; 1147 struct sr_boot_chunk_head bch, kdh; 1148 struct sr_boot_volume *bv, *bv1, *bv2; 1149 struct sr_boot_chunk *bc, *bcnext, *bc1, *bc2; 1150 struct sr_disk_head sdklist; 1151 struct sr_disk *sdk; 1152 struct disk *dk; 1153 struct bioc_createraid bcr; 1154 struct sr_meta_chunk *hm; 1155 struct sr_chunk_head *cl; 1156 struct sr_chunk *hotspare, *chunk, *last; 1157 u_int64_t *ondisk = NULL; 1158 dev_t *devs = NULL; 1159 void *data; 1160 char devname[32]; 1161 int rv = 0, i; 1162 1163 DNPRINTF(SR_D_META, "%s: sr_boot_assembly\n", DEVNAME(sc)); 1164 1165 SLIST_INIT(&sdklist); 1166 SLIST_INIT(&bvh); 1167 SLIST_INIT(&bch); 1168 SLIST_INIT(&kdh); 1169 1170 dk = TAILQ_FIRST(&disklist); 1171 while (dk != NULL) { 1172 1173 /* See if this disk has been checked. */ 1174 SLIST_FOREACH(sdk, &sdklist, sdk_link) 1175 if (sdk->sdk_devno == dk->dk_devno) 1176 break; 1177 1178 if (sdk != NULL || dk->dk_devno == NODEV) { 1179 dk = TAILQ_NEXT(dk, dk_link); 1180 continue; 1181 } 1182 1183 /* Add this disk to the list that we've checked. */ 1184 sdk = malloc(sizeof(struct sr_disk), M_DEVBUF, 1185 M_NOWAIT | M_ZERO); 1186 if (sdk == NULL) 1187 goto unwind; 1188 sdk->sdk_devno = dk->dk_devno; 1189 SLIST_INSERT_HEAD(&sdklist, sdk, sdk_link); 1190 1191 /* Only check sd(4) and wd(4) devices. */ 1192 if (strncmp(dk->dk_name, "sd", 2) && 1193 strncmp(dk->dk_name, "wd", 2)) { 1194 dk = TAILQ_NEXT(dk, dk_link); 1195 continue; 1196 } 1197 1198 /* native softraid uses partitions */ 1199 rw_enter_write(&sc->sc_lock); 1200 bio_status_init(&sc->sc_status, &sc->sc_dev); 1201 sr_meta_native_bootprobe(sc, dk->dk_devno, &bch); 1202 rw_exit_write(&sc->sc_lock); 1203 1204 /* probe non-native disks if native failed. */ 1205 1206 /* Restart scan since we may have slept. */ 1207 dk = TAILQ_FIRST(&disklist); 1208 } 1209 1210 /* 1211 * Create a list of volumes and associate chunks with each volume. 1212 */ 1213 for (bc = SLIST_FIRST(&bch); bc != NULL; bc = bcnext) { 1214 1215 bcnext = SLIST_NEXT(bc, sbc_link); 1216 SLIST_REMOVE(&bch, bc, sr_boot_chunk, sbc_link); 1217 bc->sbc_chunk_id = bc->sbc_metadata->ssdi.ssd_chunk_id; 1218 1219 /* Handle key disks separately. */ 1220 if (bc->sbc_metadata->ssdi.ssd_level == SR_KEYDISK_LEVEL) { 1221 SLIST_INSERT_HEAD(&kdh, bc, sbc_link); 1222 continue; 1223 } 1224 1225 SLIST_FOREACH(bv, &bvh, sbv_link) { 1226 if (bcmp(&bc->sbc_metadata->ssdi.ssd_uuid, 1227 &bv->sbv_uuid, 1228 sizeof(bc->sbc_metadata->ssdi.ssd_uuid)) == 0) 1229 break; 1230 } 1231 1232 if (bv == NULL) { 1233 bv = malloc(sizeof(struct sr_boot_volume), 1234 M_DEVBUF, M_NOWAIT | M_ZERO); 1235 if (bv == NULL) { 1236 printf("%s: failed to allocate boot volume\n", 1237 DEVNAME(sc)); 1238 goto unwind; 1239 } 1240 1241 bv->sbv_level = bc->sbc_metadata->ssdi.ssd_level; 1242 bv->sbv_volid = bc->sbc_metadata->ssdi.ssd_volid; 1243 bv->sbv_chunk_no = bc->sbc_metadata->ssdi.ssd_chunk_no; 1244 bv->sbv_flags = bc->sbc_metadata->ssdi.ssd_vol_flags; 1245 memcpy(&bv->sbv_uuid, &bc->sbc_metadata->ssdi.ssd_uuid, 1246 sizeof(bc->sbc_metadata->ssdi.ssd_uuid)); 1247 SLIST_INIT(&bv->sbv_chunks); 1248 1249 /* Maintain volume order. */ 1250 bv2 = NULL; 1251 SLIST_FOREACH(bv1, &bvh, sbv_link) { 1252 if (bv1->sbv_volid > bv->sbv_volid) 1253 break; 1254 bv2 = bv1; 1255 } 1256 if (bv2 == NULL) { 1257 DNPRINTF(SR_D_META, "%s: insert volume %u " 1258 "at head\n", DEVNAME(sc), bv->sbv_volid); 1259 SLIST_INSERT_HEAD(&bvh, bv, sbv_link); 1260 } else { 1261 DNPRINTF(SR_D_META, "%s: insert volume %u " 1262 "after %u\n", DEVNAME(sc), bv->sbv_volid, 1263 bv2->sbv_volid); 1264 SLIST_INSERT_AFTER(bv2, bv, sbv_link); 1265 } 1266 } 1267 1268 /* Maintain chunk order. */ 1269 bc2 = NULL; 1270 SLIST_FOREACH(bc1, &bv->sbv_chunks, sbc_link) { 1271 if (bc1->sbc_chunk_id > bc->sbc_chunk_id) 1272 break; 1273 bc2 = bc1; 1274 } 1275 if (bc2 == NULL) { 1276 DNPRINTF(SR_D_META, "%s: volume %u insert chunk %u " 1277 "at head\n", DEVNAME(sc), bv->sbv_volid, 1278 bc->sbc_chunk_id); 1279 SLIST_INSERT_HEAD(&bv->sbv_chunks, bc, sbc_link); 1280 } else { 1281 DNPRINTF(SR_D_META, "%s: volume %u insert chunk %u " 1282 "after %u\n", DEVNAME(sc), bv->sbv_volid, 1283 bc->sbc_chunk_id, bc2->sbc_chunk_id); 1284 SLIST_INSERT_AFTER(bc2, bc, sbc_link); 1285 } 1286 1287 bv->sbv_chunks_found++; 1288 } 1289 1290 /* Allocate memory for device and ondisk version arrays. */ 1291 devs = mallocarray(BIOC_CRMAXLEN, sizeof(dev_t), M_DEVBUF, 1292 M_NOWAIT); 1293 if (devs == NULL) { 1294 printf("%s: failed to allocate device array\n", DEVNAME(sc)); 1295 goto unwind; 1296 } 1297 ondisk = mallocarray(BIOC_CRMAXLEN, sizeof(u_int64_t), M_DEVBUF, 1298 M_NOWAIT); 1299 if (ondisk == NULL) { 1300 printf("%s: failed to allocate ondisk array\n", DEVNAME(sc)); 1301 goto unwind; 1302 } 1303 1304 /* 1305 * Assemble hotspare "volumes". 1306 */ 1307 SLIST_FOREACH(bv, &bvh, sbv_link) { 1308 1309 /* Check if this is a hotspare "volume". */ 1310 if (bv->sbv_level != SR_HOTSPARE_LEVEL || 1311 bv->sbv_chunk_no != 1) 1312 continue; 1313 1314 #ifdef SR_DEBUG 1315 DNPRINTF(SR_D_META, "%s: assembling hotspare volume ", 1316 DEVNAME(sc)); 1317 if (sr_debug & SR_D_META) 1318 sr_uuid_print(&bv->sbv_uuid, 0); 1319 DNPRINTF(SR_D_META, " volid %u with %u chunks\n", 1320 bv->sbv_volid, bv->sbv_chunk_no); 1321 #endif 1322 1323 /* Create hotspare chunk metadata. */ 1324 hotspare = malloc(sizeof(struct sr_chunk), M_DEVBUF, 1325 M_NOWAIT | M_ZERO); 1326 if (hotspare == NULL) { 1327 printf("%s: failed to allocate hotspare\n", 1328 DEVNAME(sc)); 1329 goto unwind; 1330 } 1331 1332 bc = SLIST_FIRST(&bv->sbv_chunks); 1333 sr_meta_getdevname(sc, bc->sbc_mm, devname, sizeof(devname)); 1334 hotspare->src_dev_mm = bc->sbc_mm; 1335 strlcpy(hotspare->src_devname, devname, 1336 sizeof(hotspare->src_devname)); 1337 hotspare->src_size = bc->sbc_metadata->ssdi.ssd_size; 1338 1339 hm = &hotspare->src_meta; 1340 hm->scmi.scm_volid = SR_HOTSPARE_VOLID; 1341 hm->scmi.scm_chunk_id = 0; 1342 hm->scmi.scm_size = bc->sbc_metadata->ssdi.ssd_size; 1343 hm->scmi.scm_coerced_size = bc->sbc_metadata->ssdi.ssd_size; 1344 strlcpy(hm->scmi.scm_devname, devname, 1345 sizeof(hm->scmi.scm_devname)); 1346 memcpy(&hm->scmi.scm_uuid, &bc->sbc_metadata->ssdi.ssd_uuid, 1347 sizeof(struct sr_uuid)); 1348 1349 sr_checksum(sc, hm, &hm->scm_checksum, 1350 sizeof(struct sr_meta_chunk_invariant)); 1351 1352 hm->scm_status = BIOC_SDHOTSPARE; 1353 1354 /* Add chunk to hotspare list. */ 1355 rw_enter_write(&sc->sc_hs_lock); 1356 cl = &sc->sc_hotspare_list; 1357 if (SLIST_EMPTY(cl)) 1358 SLIST_INSERT_HEAD(cl, hotspare, src_link); 1359 else { 1360 SLIST_FOREACH(chunk, cl, src_link) 1361 last = chunk; 1362 SLIST_INSERT_AFTER(last, hotspare, src_link); 1363 } 1364 sc->sc_hotspare_no++; 1365 rw_exit_write(&sc->sc_hs_lock); 1366 1367 } 1368 1369 /* 1370 * Assemble RAID volumes. 1371 */ 1372 SLIST_FOREACH(bv, &bvh, sbv_link) { 1373 1374 bzero(&bcr, sizeof(bcr)); 1375 data = NULL; 1376 1377 /* Check if this is a hotspare "volume". */ 1378 if (bv->sbv_level == SR_HOTSPARE_LEVEL && 1379 bv->sbv_chunk_no == 1) 1380 continue; 1381 1382 /* 1383 * Skip volumes that are marked as no auto assemble, unless 1384 * this was the volume which we actually booted from. 1385 */ 1386 if (bcmp(&sr_bootuuid, &bv->sbv_uuid, sizeof(sr_bootuuid)) != 0) 1387 if (bv->sbv_flags & BIOC_SCNOAUTOASSEMBLE) 1388 continue; 1389 1390 #ifdef SR_DEBUG 1391 DNPRINTF(SR_D_META, "%s: assembling volume ", DEVNAME(sc)); 1392 if (sr_debug & SR_D_META) 1393 sr_uuid_print(&bv->sbv_uuid, 0); 1394 DNPRINTF(SR_D_META, " volid %u with %u chunks\n", 1395 bv->sbv_volid, bv->sbv_chunk_no); 1396 #endif 1397 1398 /* 1399 * If this is a crypto volume, try to find a matching 1400 * key disk... 1401 */ 1402 bcr.bc_key_disk = NODEV; 1403 if (bv->sbv_level == 'C') { 1404 SLIST_FOREACH(bc, &kdh, sbc_link) { 1405 if (bcmp(&bc->sbc_metadata->ssdi.ssd_uuid, 1406 &bv->sbv_uuid, 1407 sizeof(bc->sbc_metadata->ssdi.ssd_uuid)) 1408 == 0) 1409 bcr.bc_key_disk = bc->sbc_mm; 1410 } 1411 } 1412 1413 for (i = 0; i < BIOC_CRMAXLEN; i++) { 1414 devs[i] = NODEV; /* mark device as illegal */ 1415 ondisk[i] = 0; 1416 } 1417 1418 SLIST_FOREACH(bc, &bv->sbv_chunks, sbc_link) { 1419 if (devs[bc->sbc_chunk_id] != NODEV) { 1420 bv->sbv_chunks_found--; 1421 sr_meta_getdevname(sc, bc->sbc_mm, devname, 1422 sizeof(devname)); 1423 printf("%s: found duplicate chunk %u for " 1424 "volume %u on device %s\n", DEVNAME(sc), 1425 bc->sbc_chunk_id, bv->sbv_volid, devname); 1426 } 1427 1428 if (devs[bc->sbc_chunk_id] == NODEV || 1429 bc->sbc_metadata->ssd_ondisk > 1430 ondisk[bc->sbc_chunk_id]) { 1431 devs[bc->sbc_chunk_id] = bc->sbc_mm; 1432 ondisk[bc->sbc_chunk_id] = 1433 bc->sbc_metadata->ssd_ondisk; 1434 DNPRINTF(SR_D_META, "%s: using ondisk " 1435 "metadata version %llu for chunk %u\n", 1436 DEVNAME(sc), ondisk[bc->sbc_chunk_id], 1437 bc->sbc_chunk_id); 1438 } 1439 } 1440 1441 if (bv->sbv_chunk_no != bv->sbv_chunks_found) { 1442 printf("%s: not all chunks were provided; " 1443 "attempting to bring volume %d online\n", 1444 DEVNAME(sc), bv->sbv_volid); 1445 } 1446 1447 bcr.bc_level = bv->sbv_level; 1448 bcr.bc_dev_list_len = bv->sbv_chunk_no * sizeof(dev_t); 1449 bcr.bc_dev_list = devs; 1450 bcr.bc_flags = BIOC_SCDEVT | 1451 (bv->sbv_flags & BIOC_SCNOAUTOASSEMBLE); 1452 1453 if (bv->sbv_level == 'C' && 1454 bcmp(&sr_bootuuid, &bv->sbv_uuid, sizeof(sr_bootuuid)) == 0) 1455 data = sr_bootkey; 1456 1457 rw_enter_write(&sc->sc_lock); 1458 bio_status_init(&sc->sc_status, &sc->sc_dev); 1459 sr_ioctl_createraid(sc, &bcr, 0, data); 1460 rw_exit_write(&sc->sc_lock); 1461 1462 rv++; 1463 } 1464 1465 /* done with metadata */ 1466 unwind: 1467 /* Free boot volumes and associated chunks. */ 1468 for (bv1 = SLIST_FIRST(&bvh); bv1 != NULL; bv1 = bv2) { 1469 bv2 = SLIST_NEXT(bv1, sbv_link); 1470 for (bc1 = SLIST_FIRST(&bv1->sbv_chunks); bc1 != NULL; 1471 bc1 = bc2) { 1472 bc2 = SLIST_NEXT(bc1, sbc_link); 1473 free(bc1->sbc_metadata, M_DEVBUF, 0); 1474 free(bc1, M_DEVBUF, 0); 1475 } 1476 free(bv1, M_DEVBUF, 0); 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, 0); 1482 free(bc1, M_DEVBUF, 0); 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, 0); 1488 free(bc1, M_DEVBUF, 0); 1489 } 1490 1491 while (!SLIST_EMPTY(&sdklist)) { 1492 sdk = SLIST_FIRST(&sdklist); 1493 SLIST_REMOVE_HEAD(&sdklist, sdk_link); 1494 free(sdk, M_DEVBUF, 0); 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 %02x%02x%02x%02x" 1536 "%02x%02x%02x%02x\n", 1537 DEVNAME(sc), rootduid[0], 1538 rootduid[1], rootduid[2], 1539 rootduid[3], rootduid[4], 1540 rootduid[5], rootduid[6], 1541 rootduid[7]); 1542 return; 1543 } 1544 } 1545 } 1546 } 1547 } 1548 1549 int 1550 sr_meta_native_probe(struct sr_softc *sc, struct sr_chunk *ch_entry) 1551 { 1552 struct disklabel label; 1553 char *devname; 1554 int error, part; 1555 u_int64_t size; 1556 1557 DNPRINTF(SR_D_META, "%s: sr_meta_native_probe(%s)\n", 1558 DEVNAME(sc), ch_entry->src_devname); 1559 1560 devname = ch_entry->src_devname; 1561 part = DISKPART(ch_entry->src_dev_mm); 1562 1563 /* get disklabel */ 1564 error = VOP_IOCTL(ch_entry->src_vn, DIOCGDINFO, (caddr_t)&label, FREAD, 1565 NOCRED, curproc); 1566 if (error) { 1567 DNPRINTF(SR_D_META, "%s: %s can't obtain disklabel\n", 1568 DEVNAME(sc), devname); 1569 goto unwind; 1570 } 1571 memcpy(ch_entry->src_duid, label.d_uid, sizeof(ch_entry->src_duid)); 1572 1573 /* make sure the partition is of the right type */ 1574 if (label.d_partitions[part].p_fstype != FS_RAID) { 1575 DNPRINTF(SR_D_META, 1576 "%s: %s partition not of type RAID (%d)\n", DEVNAME(sc), 1577 devname, 1578 label.d_partitions[part].p_fstype); 1579 goto unwind; 1580 } 1581 1582 size = DL_SECTOBLK(&label, DL_GETPSIZE(&label.d_partitions[part])); 1583 if (size <= SR_DATA_OFFSET) { 1584 DNPRINTF(SR_D_META, "%s: %s partition too small\n", DEVNAME(sc), 1585 devname); 1586 goto unwind; 1587 } 1588 size -= SR_DATA_OFFSET; 1589 if (size > INT64_MAX) { 1590 DNPRINTF(SR_D_META, "%s: %s partition too large\n", DEVNAME(sc), 1591 devname); 1592 goto unwind; 1593 } 1594 ch_entry->src_size = size; 1595 ch_entry->src_secsize = label.d_secsize; 1596 1597 DNPRINTF(SR_D_META, "%s: probe found %s size %lld\n", DEVNAME(sc), 1598 devname, (long long)size); 1599 1600 return (SR_META_F_NATIVE); 1601 unwind: 1602 DNPRINTF(SR_D_META, "%s: invalid device: %s\n", DEVNAME(sc), 1603 devname ? devname : "nodev"); 1604 return (SR_META_F_INVALID); 1605 } 1606 1607 int 1608 sr_meta_native_attach(struct sr_discipline *sd, int force) 1609 { 1610 struct sr_softc *sc = sd->sd_sc; 1611 struct sr_chunk_head *cl = &sd->sd_vol.sv_chunk_list; 1612 struct sr_metadata *md = NULL; 1613 struct sr_chunk *ch_entry, *ch_next; 1614 struct sr_uuid uuid; 1615 u_int64_t version = 0; 1616 int sr, not_sr, rv = 1, d, expected = -1, old_meta = 0; 1617 1618 DNPRINTF(SR_D_META, "%s: sr_meta_native_attach\n", DEVNAME(sc)); 1619 1620 md = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, M_ZERO | M_NOWAIT); 1621 if (md == NULL) { 1622 sr_error(sc, "not enough memory for metadata buffer"); 1623 goto bad; 1624 } 1625 1626 bzero(&uuid, sizeof uuid); 1627 1628 sr = not_sr = d = 0; 1629 SLIST_FOREACH(ch_entry, cl, src_link) { 1630 if (ch_entry->src_dev_mm == NODEV) 1631 continue; 1632 1633 if (sr_meta_native_read(sd, ch_entry->src_dev_mm, md, NULL)) { 1634 sr_error(sc, "could not read native metadata"); 1635 goto bad; 1636 } 1637 1638 if (md->ssdi.ssd_magic == SR_MAGIC) { 1639 sr++; 1640 ch_entry->src_meta.scmi.scm_chunk_id = 1641 md->ssdi.ssd_chunk_id; 1642 if (d == 0) { 1643 memcpy(&uuid, &md->ssdi.ssd_uuid, sizeof uuid); 1644 expected = md->ssdi.ssd_chunk_no; 1645 version = md->ssd_ondisk; 1646 d++; 1647 continue; 1648 } else if (bcmp(&md->ssdi.ssd_uuid, &uuid, 1649 sizeof uuid)) { 1650 sr_error(sc, "not part of the same volume"); 1651 goto bad; 1652 } 1653 if (md->ssd_ondisk != version) { 1654 old_meta++; 1655 version = MAX(md->ssd_ondisk, version); 1656 } 1657 } else 1658 not_sr++; 1659 } 1660 1661 if (sr && not_sr && !force) { 1662 sr_error(sc, "not all chunks are of the native metadata " 1663 "format"); 1664 goto bad; 1665 } 1666 1667 /* mixed metadata versions; mark bad disks offline */ 1668 if (old_meta) { 1669 d = 0; 1670 for (ch_entry = SLIST_FIRST(cl); ch_entry != NULL; 1671 ch_entry = ch_next, d++) { 1672 ch_next = SLIST_NEXT(ch_entry, src_link); 1673 1674 /* XXX do we want to read this again? */ 1675 if (ch_entry->src_dev_mm == NODEV) 1676 panic("src_dev_mm == NODEV"); 1677 if (sr_meta_native_read(sd, ch_entry->src_dev_mm, md, 1678 NULL)) 1679 sr_warn(sc, "could not read native metadata"); 1680 if (md->ssd_ondisk != version) 1681 sd->sd_vol.sv_chunks[d]->src_meta.scm_status = 1682 BIOC_SDOFFLINE; 1683 } 1684 } 1685 1686 if (expected != sr && !force && expected != -1) { 1687 DNPRINTF(SR_D_META, "%s: not all chunks were provided, trying " 1688 "anyway\n", DEVNAME(sc)); 1689 } 1690 1691 rv = 0; 1692 bad: 1693 free(md, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 1694 return (rv); 1695 } 1696 1697 int 1698 sr_meta_native_read(struct sr_discipline *sd, dev_t dev, 1699 struct sr_metadata *md, void *fm) 1700 { 1701 #ifdef SR_DEBUG 1702 struct sr_softc *sc = sd->sd_sc; 1703 #endif 1704 DNPRINTF(SR_D_META, "%s: sr_meta_native_read(0x%x, %p)\n", 1705 DEVNAME(sc), dev, md); 1706 1707 return (sr_meta_rw(sd, dev, md, B_READ)); 1708 } 1709 1710 int 1711 sr_meta_native_write(struct sr_discipline *sd, dev_t dev, 1712 struct sr_metadata *md, void *fm) 1713 { 1714 #ifdef SR_DEBUG 1715 struct sr_softc *sc = sd->sd_sc; 1716 #endif 1717 DNPRINTF(SR_D_META, "%s: sr_meta_native_write(0x%x, %p)\n", 1718 DEVNAME(sc), dev, md); 1719 1720 return (sr_meta_rw(sd, dev, md, B_WRITE)); 1721 } 1722 1723 void 1724 sr_hotplug_register(struct sr_discipline *sd, void *func) 1725 { 1726 struct sr_hotplug_list *mhe; 1727 1728 DNPRINTF(SR_D_MISC, "%s: sr_hotplug_register: %p\n", 1729 DEVNAME(sd->sd_sc), func); 1730 1731 /* make sure we aren't on the list yet */ 1732 SLIST_FOREACH(mhe, &sr_hotplug_callbacks, shl_link) 1733 if (mhe->sh_hotplug == func) 1734 return; 1735 1736 mhe = malloc(sizeof(struct sr_hotplug_list), M_DEVBUF, 1737 M_WAITOK | M_ZERO); 1738 mhe->sh_hotplug = func; 1739 mhe->sh_sd = sd; 1740 SLIST_INSERT_HEAD(&sr_hotplug_callbacks, mhe, shl_link); 1741 } 1742 1743 void 1744 sr_hotplug_unregister(struct sr_discipline *sd, void *func) 1745 { 1746 struct sr_hotplug_list *mhe; 1747 1748 DNPRINTF(SR_D_MISC, "%s: sr_hotplug_unregister: %s %p\n", 1749 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, func); 1750 1751 /* make sure we are on the list yet */ 1752 SLIST_FOREACH(mhe, &sr_hotplug_callbacks, shl_link) { 1753 if (mhe->sh_hotplug == func) 1754 break; 1755 } 1756 if (mhe != NULL) { 1757 SLIST_REMOVE(&sr_hotplug_callbacks, mhe, 1758 sr_hotplug_list, shl_link); 1759 free(mhe, M_DEVBUF, 0); 1760 } 1761 } 1762 1763 void 1764 sr_disk_attach(struct disk *diskp, int action) 1765 { 1766 struct sr_hotplug_list *mhe; 1767 1768 SLIST_FOREACH(mhe, &sr_hotplug_callbacks, shl_link) 1769 if (mhe->sh_sd->sd_ready) 1770 mhe->sh_hotplug(mhe->sh_sd, diskp, action); 1771 } 1772 1773 int 1774 sr_match(struct device *parent, void *match, void *aux) 1775 { 1776 return (1); 1777 } 1778 1779 void 1780 sr_attach(struct device *parent, struct device *self, void *aux) 1781 { 1782 struct sr_softc *sc = (void *)self; 1783 struct scsibus_attach_args saa; 1784 1785 DNPRINTF(SR_D_MISC, "\n%s: sr_attach", DEVNAME(sc)); 1786 1787 if (softraid0 == NULL) 1788 softraid0 = sc; 1789 1790 rw_init(&sc->sc_lock, "sr_lock"); 1791 rw_init(&sc->sc_hs_lock, "sr_hs_lock"); 1792 1793 SLIST_INIT(&sr_hotplug_callbacks); 1794 TAILQ_INIT(&sc->sc_dis_list); 1795 SLIST_INIT(&sc->sc_hotspare_list); 1796 1797 #if NBIO > 0 1798 if (bio_register(&sc->sc_dev, sr_bio_ioctl) != 0) 1799 printf("%s: controller registration failed", DEVNAME(sc)); 1800 #endif /* NBIO > 0 */ 1801 1802 #ifndef SMALL_KERNEL 1803 strlcpy(sc->sc_sensordev.xname, DEVNAME(sc), 1804 sizeof(sc->sc_sensordev.xname)); 1805 sensordev_install(&sc->sc_sensordev); 1806 #endif /* SMALL_KERNEL */ 1807 1808 printf("\n"); 1809 1810 sc->sc_link.adapter_softc = sc; 1811 sc->sc_link.adapter = &sr_switch; 1812 sc->sc_link.adapter_target = SR_MAX_LD; 1813 sc->sc_link.adapter_buswidth = SR_MAX_LD; 1814 sc->sc_link.luns = 1; 1815 1816 bzero(&saa, sizeof(saa)); 1817 saa.saa_sc_link = &sc->sc_link; 1818 1819 sc->sc_scsibus = (struct scsibus_softc *)config_found(&sc->sc_dev, 1820 &saa, scsiprint); 1821 1822 softraid_disk_attach = sr_disk_attach; 1823 1824 sr_boot_assembly(sc); 1825 1826 explicit_bzero(sr_bootkey, sizeof(sr_bootkey)); 1827 } 1828 1829 int 1830 sr_detach(struct device *self, int flags) 1831 { 1832 struct sr_softc *sc = (void *)self; 1833 int rv; 1834 1835 DNPRINTF(SR_D_MISC, "%s: sr_detach\n", DEVNAME(sc)); 1836 1837 softraid_disk_attach = NULL; 1838 1839 sr_shutdown(0); 1840 1841 #ifndef SMALL_KERNEL 1842 if (sc->sc_sensor_task != NULL) 1843 sensor_task_unregister(sc->sc_sensor_task); 1844 sensordev_deinstall(&sc->sc_sensordev); 1845 #endif /* SMALL_KERNEL */ 1846 1847 if (sc->sc_scsibus != NULL) { 1848 rv = config_detach((struct device *)sc->sc_scsibus, flags); 1849 if (rv != 0) 1850 return (rv); 1851 sc->sc_scsibus = NULL; 1852 } 1853 1854 return (0); 1855 } 1856 1857 void 1858 sr_info(struct sr_softc *sc, const char *fmt, ...) 1859 { 1860 va_list ap; 1861 1862 rw_assert_wrlock(&sc->sc_lock); 1863 1864 va_start(ap, fmt); 1865 bio_status(&sc->sc_status, 0, BIO_MSG_INFO, fmt, &ap); 1866 va_end(ap); 1867 } 1868 1869 void 1870 sr_warn(struct sr_softc *sc, const char *fmt, ...) 1871 { 1872 va_list ap; 1873 1874 rw_assert_wrlock(&sc->sc_lock); 1875 1876 va_start(ap, fmt); 1877 bio_status(&sc->sc_status, 1, BIO_MSG_WARN, fmt, &ap); 1878 va_end(ap); 1879 } 1880 1881 void 1882 sr_error(struct sr_softc *sc, const char *fmt, ...) 1883 { 1884 va_list ap; 1885 1886 rw_assert_wrlock(&sc->sc_lock); 1887 1888 va_start(ap, fmt); 1889 bio_status(&sc->sc_status, 1, BIO_MSG_ERROR, fmt, &ap); 1890 va_end(ap); 1891 } 1892 1893 void 1894 sr_minphys(struct buf *bp, struct scsi_link *sl) 1895 { 1896 DNPRINTF(SR_D_MISC, "sr_minphys: %ld\n", bp->b_bcount); 1897 1898 /* XXX currently using SR_MAXFER = MAXPHYS */ 1899 if (bp->b_bcount > SR_MAXFER) 1900 bp->b_bcount = SR_MAXFER; 1901 minphys(bp); 1902 } 1903 1904 void 1905 sr_copy_internal_data(struct scsi_xfer *xs, void *v, size_t size) 1906 { 1907 size_t copy_cnt; 1908 1909 DNPRINTF(SR_D_MISC, "sr_copy_internal_data xs: %p size: %zu\n", 1910 xs, size); 1911 1912 if (xs->datalen) { 1913 copy_cnt = MIN(size, xs->datalen); 1914 memcpy(xs->data, v, copy_cnt); 1915 } 1916 } 1917 1918 int 1919 sr_ccb_alloc(struct sr_discipline *sd) 1920 { 1921 struct sr_ccb *ccb; 1922 int i; 1923 1924 if (!sd) 1925 return (1); 1926 1927 DNPRINTF(SR_D_CCB, "%s: sr_ccb_alloc\n", DEVNAME(sd->sd_sc)); 1928 1929 if (sd->sd_ccb) 1930 return (1); 1931 1932 sd->sd_ccb = mallocarray(sd->sd_max_wu, 1933 sd->sd_max_ccb_per_wu * sizeof(struct sr_ccb), 1934 M_DEVBUF, M_WAITOK | M_ZERO); 1935 TAILQ_INIT(&sd->sd_ccb_freeq); 1936 for (i = 0; i < sd->sd_max_wu * sd->sd_max_ccb_per_wu; i++) { 1937 ccb = &sd->sd_ccb[i]; 1938 ccb->ccb_dis = sd; 1939 sr_ccb_put(ccb); 1940 } 1941 1942 DNPRINTF(SR_D_CCB, "%s: sr_ccb_alloc ccb: %d\n", 1943 DEVNAME(sd->sd_sc), sd->sd_max_wu * sd->sd_max_ccb_per_wu); 1944 1945 return (0); 1946 } 1947 1948 void 1949 sr_ccb_free(struct sr_discipline *sd) 1950 { 1951 struct sr_ccb *ccb; 1952 1953 if (!sd) 1954 return; 1955 1956 DNPRINTF(SR_D_CCB, "%s: sr_ccb_free %p\n", DEVNAME(sd->sd_sc), sd); 1957 1958 while ((ccb = TAILQ_FIRST(&sd->sd_ccb_freeq)) != NULL) 1959 TAILQ_REMOVE(&sd->sd_ccb_freeq, ccb, ccb_link); 1960 1961 free(sd->sd_ccb, M_DEVBUF, 0); 1962 } 1963 1964 struct sr_ccb * 1965 sr_ccb_get(struct sr_discipline *sd) 1966 { 1967 struct sr_ccb *ccb; 1968 int s; 1969 1970 s = splbio(); 1971 1972 ccb = TAILQ_FIRST(&sd->sd_ccb_freeq); 1973 if (ccb) { 1974 TAILQ_REMOVE(&sd->sd_ccb_freeq, ccb, ccb_link); 1975 ccb->ccb_state = SR_CCB_INPROGRESS; 1976 } 1977 1978 splx(s); 1979 1980 DNPRINTF(SR_D_CCB, "%s: sr_ccb_get: %p\n", DEVNAME(sd->sd_sc), 1981 ccb); 1982 1983 return (ccb); 1984 } 1985 1986 void 1987 sr_ccb_put(struct sr_ccb *ccb) 1988 { 1989 struct sr_discipline *sd = ccb->ccb_dis; 1990 int s; 1991 1992 DNPRINTF(SR_D_CCB, "%s: sr_ccb_put: %p\n", DEVNAME(sd->sd_sc), 1993 ccb); 1994 1995 s = splbio(); 1996 1997 ccb->ccb_wu = NULL; 1998 ccb->ccb_state = SR_CCB_FREE; 1999 ccb->ccb_target = -1; 2000 ccb->ccb_opaque = NULL; 2001 2002 TAILQ_INSERT_TAIL(&sd->sd_ccb_freeq, ccb, ccb_link); 2003 2004 splx(s); 2005 } 2006 2007 struct sr_ccb * 2008 sr_ccb_rw(struct sr_discipline *sd, int chunk, daddr_t blkno, 2009 long len, u_int8_t *data, int xsflags, int ccbflags) 2010 { 2011 struct sr_chunk *sc = sd->sd_vol.sv_chunks[chunk]; 2012 struct sr_ccb *ccb = NULL; 2013 2014 ccb = sr_ccb_get(sd); 2015 if (ccb == NULL) 2016 goto out; 2017 2018 ccb->ccb_flags = ccbflags; 2019 ccb->ccb_target = chunk; 2020 2021 ccb->ccb_buf.b_flags = B_PHYS | B_CALL; 2022 if (ISSET(xsflags, SCSI_DATA_IN)) 2023 ccb->ccb_buf.b_flags |= B_READ; 2024 else 2025 ccb->ccb_buf.b_flags |= B_WRITE; 2026 2027 ccb->ccb_buf.b_blkno = blkno + sd->sd_meta->ssd_data_blkno; 2028 ccb->ccb_buf.b_bcount = len; 2029 ccb->ccb_buf.b_bufsize = len; 2030 ccb->ccb_buf.b_resid = len; 2031 ccb->ccb_buf.b_data = data; 2032 ccb->ccb_buf.b_error = 0; 2033 ccb->ccb_buf.b_iodone = sd->sd_scsi_intr; 2034 ccb->ccb_buf.b_proc = curproc; 2035 ccb->ccb_buf.b_dev = sc->src_dev_mm; 2036 ccb->ccb_buf.b_vp = sc->src_vn; 2037 ccb->ccb_buf.b_bq = NULL; 2038 2039 if (!ISSET(ccb->ccb_buf.b_flags, B_READ)) 2040 ccb->ccb_buf.b_vp->v_numoutput++; 2041 2042 LIST_INIT(&ccb->ccb_buf.b_dep); 2043 2044 DNPRINTF(SR_D_DIS, "%s: %s %s ccb " 2045 "b_bcount %ld b_blkno %lld b_flags 0x%0lx b_data %p\n", 2046 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, sd->sd_name, 2047 ccb->ccb_buf.b_bcount, (long long)ccb->ccb_buf.b_blkno, 2048 ccb->ccb_buf.b_flags, ccb->ccb_buf.b_data); 2049 2050 out: 2051 return ccb; 2052 } 2053 2054 void 2055 sr_ccb_done(struct sr_ccb *ccb) 2056 { 2057 struct sr_workunit *wu = ccb->ccb_wu; 2058 struct sr_discipline *sd = wu->swu_dis; 2059 struct sr_softc *sc = sd->sd_sc; 2060 2061 DNPRINTF(SR_D_INTR, "%s: %s %s ccb done b_bcount %ld b_resid %zu" 2062 " b_flags 0x%0lx block %lld target %d\n", 2063 DEVNAME(sc), sd->sd_meta->ssd_devname, sd->sd_name, 2064 ccb->ccb_buf.b_bcount, ccb->ccb_buf.b_resid, ccb->ccb_buf.b_flags, 2065 (long long)ccb->ccb_buf.b_blkno, ccb->ccb_target); 2066 2067 splassert(IPL_BIO); 2068 2069 if (ccb->ccb_target == -1) 2070 panic("%s: invalid target on wu: %p", DEVNAME(sc), wu); 2071 2072 if (ccb->ccb_buf.b_flags & B_ERROR) { 2073 DNPRINTF(SR_D_INTR, "%s: i/o error on block %lld target %d\n", 2074 DEVNAME(sc), (long long)ccb->ccb_buf.b_blkno, 2075 ccb->ccb_target); 2076 if (ISSET(sd->sd_capabilities, SR_CAP_REDUNDANT)) 2077 sd->sd_set_chunk_state(sd, ccb->ccb_target, 2078 BIOC_SDOFFLINE); 2079 else 2080 printf("%s: %s: i/o error %d @ %s block %lld\n", 2081 DEVNAME(sc), sd->sd_meta->ssd_devname, 2082 ccb->ccb_buf.b_error, sd->sd_name, 2083 (long long)ccb->ccb_buf.b_blkno); 2084 ccb->ccb_state = SR_CCB_FAILED; 2085 wu->swu_ios_failed++; 2086 } else { 2087 ccb->ccb_state = SR_CCB_OK; 2088 wu->swu_ios_succeeded++; 2089 } 2090 2091 wu->swu_ios_complete++; 2092 } 2093 2094 int 2095 sr_wu_alloc(struct sr_discipline *sd, int wu_size) 2096 { 2097 struct sr_workunit *wu; 2098 int i, no_wu; 2099 2100 DNPRINTF(SR_D_WU, "%s: sr_wu_alloc %p %d\n", DEVNAME(sd->sd_sc), 2101 sd, sd->sd_max_wu); 2102 2103 no_wu = sd->sd_max_wu; 2104 sd->sd_wu_pending = no_wu; 2105 2106 mtx_init(&sd->sd_wu_mtx, IPL_BIO); 2107 TAILQ_INIT(&sd->sd_wu); 2108 TAILQ_INIT(&sd->sd_wu_freeq); 2109 TAILQ_INIT(&sd->sd_wu_pendq); 2110 TAILQ_INIT(&sd->sd_wu_defq); 2111 2112 for (i = 0; i < no_wu; i++) { 2113 wu = malloc(wu_size, M_DEVBUF, M_WAITOK | M_ZERO); 2114 TAILQ_INSERT_TAIL(&sd->sd_wu, wu, swu_next); 2115 TAILQ_INIT(&wu->swu_ccb); 2116 wu->swu_dis = sd; 2117 task_set(&wu->swu_task, sr_wu_done_callback, wu); 2118 sr_wu_put(sd, wu); 2119 } 2120 2121 return (0); 2122 } 2123 2124 void 2125 sr_wu_free(struct sr_discipline *sd) 2126 { 2127 struct sr_workunit *wu; 2128 2129 DNPRINTF(SR_D_WU, "%s: sr_wu_free %p\n", DEVNAME(sd->sd_sc), sd); 2130 2131 while ((wu = TAILQ_FIRST(&sd->sd_wu_freeq)) != NULL) 2132 TAILQ_REMOVE(&sd->sd_wu_freeq, wu, swu_link); 2133 while ((wu = TAILQ_FIRST(&sd->sd_wu_pendq)) != NULL) 2134 TAILQ_REMOVE(&sd->sd_wu_pendq, wu, swu_link); 2135 while ((wu = TAILQ_FIRST(&sd->sd_wu_defq)) != NULL) 2136 TAILQ_REMOVE(&sd->sd_wu_defq, wu, swu_link); 2137 2138 while ((wu = TAILQ_FIRST(&sd->sd_wu)) != NULL) { 2139 TAILQ_REMOVE(&sd->sd_wu, wu, swu_next); 2140 free(wu, M_DEVBUF, 0); 2141 } 2142 } 2143 2144 void * 2145 sr_wu_get(void *xsd) 2146 { 2147 struct sr_discipline *sd = (struct sr_discipline *)xsd; 2148 struct sr_workunit *wu; 2149 2150 mtx_enter(&sd->sd_wu_mtx); 2151 wu = TAILQ_FIRST(&sd->sd_wu_freeq); 2152 if (wu) { 2153 TAILQ_REMOVE(&sd->sd_wu_freeq, wu, swu_link); 2154 sd->sd_wu_pending++; 2155 } 2156 mtx_leave(&sd->sd_wu_mtx); 2157 2158 DNPRINTF(SR_D_WU, "%s: sr_wu_get: %p\n", DEVNAME(sd->sd_sc), wu); 2159 2160 return (wu); 2161 } 2162 2163 void 2164 sr_wu_put(void *xsd, void *xwu) 2165 { 2166 struct sr_discipline *sd = (struct sr_discipline *)xsd; 2167 struct sr_workunit *wu = (struct sr_workunit *)xwu; 2168 2169 DNPRINTF(SR_D_WU, "%s: sr_wu_put: %p\n", DEVNAME(sd->sd_sc), wu); 2170 2171 sr_wu_release_ccbs(wu); 2172 sr_wu_init(sd, wu); 2173 2174 mtx_enter(&sd->sd_wu_mtx); 2175 TAILQ_INSERT_TAIL(&sd->sd_wu_freeq, wu, swu_link); 2176 sd->sd_wu_pending--; 2177 mtx_leave(&sd->sd_wu_mtx); 2178 } 2179 2180 void 2181 sr_wu_init(struct sr_discipline *sd, struct sr_workunit *wu) 2182 { 2183 int s; 2184 2185 s = splbio(); 2186 if (wu->swu_cb_active == 1) 2187 panic("%s: sr_wu_init got active wu", DEVNAME(sd->sd_sc)); 2188 splx(s); 2189 2190 wu->swu_xs = NULL; 2191 wu->swu_state = SR_WU_FREE; 2192 wu->swu_flags = 0; 2193 wu->swu_blk_start = 0; 2194 wu->swu_blk_end = 0; 2195 wu->swu_collider = NULL; 2196 } 2197 2198 void 2199 sr_wu_enqueue_ccb(struct sr_workunit *wu, struct sr_ccb *ccb) 2200 { 2201 struct sr_discipline *sd = wu->swu_dis; 2202 int s; 2203 2204 s = splbio(); 2205 if (wu->swu_cb_active == 1) 2206 panic("%s: sr_wu_enqueue_ccb got active wu", 2207 DEVNAME(sd->sd_sc)); 2208 ccb->ccb_wu = wu; 2209 wu->swu_io_count++; 2210 TAILQ_INSERT_TAIL(&wu->swu_ccb, ccb, ccb_link); 2211 splx(s); 2212 } 2213 2214 void 2215 sr_wu_release_ccbs(struct sr_workunit *wu) 2216 { 2217 struct sr_ccb *ccb; 2218 2219 /* Return all ccbs that are associated with this workunit. */ 2220 while ((ccb = TAILQ_FIRST(&wu->swu_ccb)) != NULL) { 2221 TAILQ_REMOVE(&wu->swu_ccb, ccb, ccb_link); 2222 sr_ccb_put(ccb); 2223 } 2224 2225 wu->swu_io_count = 0; 2226 wu->swu_ios_complete = 0; 2227 wu->swu_ios_failed = 0; 2228 wu->swu_ios_succeeded = 0; 2229 } 2230 2231 void 2232 sr_wu_done(struct sr_workunit *wu) 2233 { 2234 struct sr_discipline *sd = wu->swu_dis; 2235 2236 DNPRINTF(SR_D_INTR, "%s: sr_wu_done count %d completed %d failed %d\n", 2237 DEVNAME(sd->sd_sc), wu->swu_io_count, wu->swu_ios_complete, 2238 wu->swu_ios_failed); 2239 2240 if (wu->swu_ios_complete < wu->swu_io_count) 2241 return; 2242 2243 task_add(sd->sd_taskq, &wu->swu_task); 2244 } 2245 2246 void 2247 sr_wu_done_callback(void *xwu) 2248 { 2249 struct sr_workunit *wu = xwu; 2250 struct sr_discipline *sd = wu->swu_dis; 2251 struct scsi_xfer *xs = wu->swu_xs; 2252 struct sr_workunit *wup; 2253 int s; 2254 2255 /* 2256 * The SR_WUF_DISCIPLINE or SR_WUF_REBUILD flag must be set if 2257 * the work unit is not associated with a scsi_xfer. 2258 */ 2259 KASSERT(xs != NULL || 2260 (wu->swu_flags & (SR_WUF_DISCIPLINE|SR_WUF_REBUILD))); 2261 2262 s = splbio(); 2263 2264 if (xs != NULL) { 2265 if (wu->swu_ios_failed) 2266 xs->error = XS_DRIVER_STUFFUP; 2267 else 2268 xs->error = XS_NOERROR; 2269 } 2270 2271 if (sd->sd_scsi_wu_done) { 2272 if (sd->sd_scsi_wu_done(wu) == SR_WU_RESTART) 2273 goto done; 2274 } 2275 2276 /* Remove work unit from pending queue. */ 2277 TAILQ_FOREACH(wup, &sd->sd_wu_pendq, swu_link) 2278 if (wup == wu) 2279 break; 2280 if (wup == NULL) 2281 panic("%s: wu %p not on pending queue", 2282 DEVNAME(sd->sd_sc), wu); 2283 TAILQ_REMOVE(&sd->sd_wu_pendq, wu, swu_link); 2284 2285 if (wu->swu_collider) { 2286 if (wu->swu_ios_failed) 2287 sr_raid_recreate_wu(wu->swu_collider); 2288 2289 /* XXX Should the collider be failed if this xs failed? */ 2290 sr_raid_startwu(wu->swu_collider); 2291 } 2292 2293 /* 2294 * If a discipline provides its own sd_scsi_done function, then it 2295 * is responsible for calling sr_scsi_done() once I/O is complete. 2296 */ 2297 if (wu->swu_flags & SR_WUF_REBUILD) 2298 wu->swu_flags |= SR_WUF_REBUILDIOCOMP; 2299 if (wu->swu_flags & SR_WUF_WAKEUP) 2300 wakeup(wu); 2301 if (sd->sd_scsi_done) 2302 sd->sd_scsi_done(wu); 2303 else if (wu->swu_flags & SR_WUF_DISCIPLINE) 2304 sr_scsi_wu_put(sd, wu); 2305 else if (!(wu->swu_flags & SR_WUF_REBUILD)) 2306 sr_scsi_done(sd, xs); 2307 2308 done: 2309 splx(s); 2310 } 2311 2312 struct sr_workunit * 2313 sr_scsi_wu_get(struct sr_discipline *sd, int flags) 2314 { 2315 return scsi_io_get(&sd->sd_iopool, flags); 2316 } 2317 2318 void 2319 sr_scsi_wu_put(struct sr_discipline *sd, struct sr_workunit *wu) 2320 { 2321 scsi_io_put(&sd->sd_iopool, wu); 2322 2323 if (sd->sd_sync && sd->sd_wu_pending == 0) 2324 wakeup(sd); 2325 } 2326 2327 void 2328 sr_scsi_done(struct sr_discipline *sd, struct scsi_xfer *xs) 2329 { 2330 DNPRINTF(SR_D_DIS, "%s: sr_scsi_done: xs %p\n", DEVNAME(sd->sd_sc), xs); 2331 2332 if (xs->error == XS_NOERROR) 2333 xs->resid = 0; 2334 2335 scsi_done(xs); 2336 2337 if (sd->sd_sync && sd->sd_wu_pending == 0) 2338 wakeup(sd); 2339 } 2340 2341 void 2342 sr_scsi_cmd(struct scsi_xfer *xs) 2343 { 2344 struct scsi_link *link = xs->sc_link; 2345 struct sr_softc *sc = link->adapter_softc; 2346 struct sr_workunit *wu = xs->io; 2347 struct sr_discipline *sd; 2348 2349 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd target %d xs %p flags %#x\n", 2350 DEVNAME(sc), link->target, xs, xs->flags); 2351 2352 sd = sc->sc_targets[link->target]; 2353 if (sd == NULL) 2354 panic("%s: sr_scsi_cmd NULL discipline", DEVNAME(sc)); 2355 2356 if (sd->sd_deleted) { 2357 printf("%s: %s device is being deleted, failing io\n", 2358 DEVNAME(sc), sd->sd_meta->ssd_devname); 2359 goto stuffup; 2360 } 2361 2362 /* scsi layer *can* re-send wu without calling sr_wu_put(). */ 2363 sr_wu_release_ccbs(wu); 2364 sr_wu_init(sd, wu); 2365 wu->swu_state = SR_WU_INPROGRESS; 2366 wu->swu_xs = xs; 2367 2368 switch (xs->cmd->opcode) { 2369 case READ_COMMAND: 2370 case READ_BIG: 2371 case READ_16: 2372 case WRITE_COMMAND: 2373 case WRITE_BIG: 2374 case WRITE_16: 2375 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: READ/WRITE %02x\n", 2376 DEVNAME(sc), xs->cmd->opcode); 2377 if (sd->sd_scsi_rw(wu)) 2378 goto stuffup; 2379 break; 2380 2381 case SYNCHRONIZE_CACHE: 2382 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: SYNCHRONIZE_CACHE\n", 2383 DEVNAME(sc)); 2384 if (sd->sd_scsi_sync(wu)) 2385 goto stuffup; 2386 goto complete; 2387 2388 case TEST_UNIT_READY: 2389 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: TEST_UNIT_READY\n", 2390 DEVNAME(sc)); 2391 if (sd->sd_scsi_tur(wu)) 2392 goto stuffup; 2393 goto complete; 2394 2395 case START_STOP: 2396 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: START_STOP\n", 2397 DEVNAME(sc)); 2398 if (sd->sd_scsi_start_stop(wu)) 2399 goto stuffup; 2400 goto complete; 2401 2402 case INQUIRY: 2403 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: INQUIRY\n", 2404 DEVNAME(sc)); 2405 if (sd->sd_scsi_inquiry(wu)) 2406 goto stuffup; 2407 goto complete; 2408 2409 case READ_CAPACITY: 2410 case READ_CAPACITY_16: 2411 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd READ CAPACITY 0x%02x\n", 2412 DEVNAME(sc), xs->cmd->opcode); 2413 if (sd->sd_scsi_read_cap(wu)) 2414 goto stuffup; 2415 goto complete; 2416 2417 case REQUEST_SENSE: 2418 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd REQUEST SENSE\n", 2419 DEVNAME(sc)); 2420 if (sd->sd_scsi_req_sense(wu)) 2421 goto stuffup; 2422 goto complete; 2423 2424 default: 2425 DNPRINTF(SR_D_CMD, "%s: unsupported scsi command %x\n", 2426 DEVNAME(sc), xs->cmd->opcode); 2427 /* XXX might need to add generic function to handle others */ 2428 goto stuffup; 2429 } 2430 2431 return; 2432 stuffup: 2433 if (sd->sd_scsi_sense.error_code) { 2434 xs->error = XS_SENSE; 2435 memcpy(&xs->sense, &sd->sd_scsi_sense, sizeof(xs->sense)); 2436 bzero(&sd->sd_scsi_sense, sizeof(sd->sd_scsi_sense)); 2437 } else { 2438 xs->error = XS_DRIVER_STUFFUP; 2439 } 2440 complete: 2441 sr_scsi_done(sd, xs); 2442 } 2443 2444 int 2445 sr_scsi_probe(struct scsi_link *link) 2446 { 2447 struct sr_softc *sc = link->adapter_softc; 2448 struct sr_discipline *sd; 2449 2450 KASSERT(link->target < SR_MAX_LD && link->lun == 0); 2451 2452 sd = sc->sc_targets[link->target]; 2453 if (sd == NULL) 2454 return (ENODEV); 2455 2456 link->pool = &sd->sd_iopool; 2457 if (sd->sd_openings) 2458 link->openings = sd->sd_openings(sd); 2459 else 2460 link->openings = sd->sd_max_wu; 2461 2462 return (0); 2463 } 2464 2465 int 2466 sr_scsi_ioctl(struct scsi_link *link, u_long cmd, caddr_t addr, int flag) 2467 { 2468 struct sr_softc *sc = link->adapter_softc; 2469 struct sr_discipline *sd; 2470 2471 sd = sc->sc_targets[link->target]; 2472 if (sd == NULL) 2473 return (ENODEV); 2474 2475 DNPRINTF(SR_D_IOCTL, "%s: %s sr_scsi_ioctl cmd: %#lx\n", 2476 DEVNAME(sc), sd->sd_meta->ssd_devname, cmd); 2477 2478 /* Pass bio ioctls through to the bio handler. */ 2479 if (IOCGROUP(cmd) == 'B') 2480 return (sr_bio_handler(sc, sd, cmd, (struct bio *)addr)); 2481 2482 switch (cmd) { 2483 case DIOCGCACHE: 2484 case DIOCSCACHE: 2485 return (EOPNOTSUPP); 2486 default: 2487 return (ENOTTY); 2488 } 2489 } 2490 2491 int 2492 sr_bio_ioctl(struct device *dev, u_long cmd, caddr_t addr) 2493 { 2494 struct sr_softc *sc = (struct sr_softc *) dev; 2495 DNPRINTF(SR_D_IOCTL, "%s: sr_bio_ioctl\n", DEVNAME(sc)); 2496 2497 return sr_bio_handler(sc, NULL, cmd, (struct bio *)addr); 2498 } 2499 2500 int 2501 sr_bio_handler(struct sr_softc *sc, struct sr_discipline *sd, u_long cmd, 2502 struct bio *bio) 2503 { 2504 int rv = 0; 2505 2506 DNPRINTF(SR_D_IOCTL, "%s: sr_bio_handler ", DEVNAME(sc)); 2507 2508 rw_enter_write(&sc->sc_lock); 2509 2510 bio_status_init(&sc->sc_status, &sc->sc_dev); 2511 2512 switch (cmd) { 2513 case BIOCINQ: 2514 DNPRINTF(SR_D_IOCTL, "inq\n"); 2515 rv = sr_ioctl_inq(sc, (struct bioc_inq *)bio); 2516 break; 2517 2518 case BIOCVOL: 2519 DNPRINTF(SR_D_IOCTL, "vol\n"); 2520 rv = sr_ioctl_vol(sc, (struct bioc_vol *)bio); 2521 break; 2522 2523 case BIOCDISK: 2524 DNPRINTF(SR_D_IOCTL, "disk\n"); 2525 rv = sr_ioctl_disk(sc, (struct bioc_disk *)bio); 2526 break; 2527 2528 case BIOCALARM: 2529 DNPRINTF(SR_D_IOCTL, "alarm\n"); 2530 /*rv = sr_ioctl_alarm(sc, (struct bioc_alarm *)bio); */ 2531 break; 2532 2533 case BIOCBLINK: 2534 DNPRINTF(SR_D_IOCTL, "blink\n"); 2535 /*rv = sr_ioctl_blink(sc, (struct bioc_blink *)bio); */ 2536 break; 2537 2538 case BIOCSETSTATE: 2539 DNPRINTF(SR_D_IOCTL, "setstate\n"); 2540 rv = sr_ioctl_setstate(sc, (struct bioc_setstate *)bio); 2541 break; 2542 2543 case BIOCCREATERAID: 2544 DNPRINTF(SR_D_IOCTL, "createraid\n"); 2545 rv = sr_ioctl_createraid(sc, (struct bioc_createraid *)bio, 2546 1, NULL); 2547 break; 2548 2549 case BIOCDELETERAID: 2550 DNPRINTF(SR_D_IOCTL, "deleteraid\n"); 2551 rv = sr_ioctl_deleteraid(sc, sd, (struct bioc_deleteraid *)bio); 2552 break; 2553 2554 case BIOCDISCIPLINE: 2555 DNPRINTF(SR_D_IOCTL, "discipline\n"); 2556 rv = sr_ioctl_discipline(sc, sd, (struct bioc_discipline *)bio); 2557 break; 2558 2559 case BIOCINSTALLBOOT: 2560 DNPRINTF(SR_D_IOCTL, "installboot\n"); 2561 rv = sr_ioctl_installboot(sc, sd, 2562 (struct bioc_installboot *)bio); 2563 break; 2564 2565 default: 2566 DNPRINTF(SR_D_IOCTL, "invalid ioctl\n"); 2567 rv = ENOTTY; 2568 } 2569 2570 sc->sc_status.bs_status = (rv ? BIO_STATUS_ERROR : BIO_STATUS_SUCCESS); 2571 2572 if (sc->sc_status.bs_msg_count > 0) 2573 rv = 0; 2574 2575 memcpy(&bio->bio_status, &sc->sc_status, sizeof(struct bio_status)); 2576 2577 rw_exit_write(&sc->sc_lock); 2578 2579 return (rv); 2580 } 2581 2582 int 2583 sr_ioctl_inq(struct sr_softc *sc, struct bioc_inq *bi) 2584 { 2585 struct sr_discipline *sd; 2586 int vol = 0, disk = 0; 2587 2588 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2589 vol++; 2590 disk += sd->sd_meta->ssdi.ssd_chunk_no; 2591 } 2592 2593 strlcpy(bi->bi_dev, sc->sc_dev.dv_xname, sizeof(bi->bi_dev)); 2594 bi->bi_novol = vol + sc->sc_hotspare_no; 2595 bi->bi_nodisk = disk + sc->sc_hotspare_no; 2596 2597 return (0); 2598 } 2599 2600 int 2601 sr_ioctl_vol(struct sr_softc *sc, struct bioc_vol *bv) 2602 { 2603 int vol = -1, rv = EINVAL; 2604 struct sr_discipline *sd; 2605 struct sr_chunk *hotspare; 2606 2607 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2608 vol++; 2609 if (vol != bv->bv_volid) 2610 continue; 2611 2612 bv->bv_status = sd->sd_vol_status; 2613 bv->bv_size = sd->sd_meta->ssdi.ssd_size << DEV_BSHIFT; 2614 bv->bv_level = sd->sd_meta->ssdi.ssd_level; 2615 bv->bv_nodisk = sd->sd_meta->ssdi.ssd_chunk_no; 2616 2617 #ifdef CRYPTO 2618 if (sd->sd_meta->ssdi.ssd_level == 'C' && 2619 sd->mds.mdd_crypto.key_disk != NULL) 2620 bv->bv_nodisk++; 2621 #endif 2622 if (bv->bv_status == BIOC_SVREBUILD) 2623 bv->bv_percent = sr_rebuild_percent(sd); 2624 2625 strlcpy(bv->bv_dev, sd->sd_meta->ssd_devname, 2626 sizeof(bv->bv_dev)); 2627 strlcpy(bv->bv_vendor, sd->sd_meta->ssdi.ssd_vendor, 2628 sizeof(bv->bv_vendor)); 2629 rv = 0; 2630 goto done; 2631 } 2632 2633 /* Check hotspares list. */ 2634 SLIST_FOREACH(hotspare, &sc->sc_hotspare_list, src_link) { 2635 vol++; 2636 if (vol != bv->bv_volid) 2637 continue; 2638 2639 bv->bv_status = BIOC_SVONLINE; 2640 bv->bv_size = hotspare->src_meta.scmi.scm_size << DEV_BSHIFT; 2641 bv->bv_level = -1; /* Hotspare. */ 2642 bv->bv_nodisk = 1; 2643 strlcpy(bv->bv_dev, hotspare->src_meta.scmi.scm_devname, 2644 sizeof(bv->bv_dev)); 2645 strlcpy(bv->bv_vendor, hotspare->src_meta.scmi.scm_devname, 2646 sizeof(bv->bv_vendor)); 2647 rv = 0; 2648 goto done; 2649 } 2650 2651 done: 2652 return (rv); 2653 } 2654 2655 int 2656 sr_ioctl_disk(struct sr_softc *sc, struct bioc_disk *bd) 2657 { 2658 struct sr_discipline *sd; 2659 struct sr_chunk *src, *hotspare; 2660 int vol = -1, rv = EINVAL; 2661 2662 if (bd->bd_diskid < 0) 2663 goto done; 2664 2665 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2666 vol++; 2667 if (vol != bd->bd_volid) 2668 continue; 2669 2670 if (bd->bd_diskid < sd->sd_meta->ssdi.ssd_chunk_no) 2671 src = sd->sd_vol.sv_chunks[bd->bd_diskid]; 2672 #ifdef CRYPTO 2673 else if (bd->bd_diskid == sd->sd_meta->ssdi.ssd_chunk_no && 2674 sd->sd_meta->ssdi.ssd_level == 'C' && 2675 sd->mds.mdd_crypto.key_disk != NULL) 2676 src = sd->mds.mdd_crypto.key_disk; 2677 #endif 2678 else 2679 break; 2680 2681 bd->bd_status = src->src_meta.scm_status; 2682 bd->bd_size = src->src_meta.scmi.scm_size << DEV_BSHIFT; 2683 bd->bd_channel = vol; 2684 bd->bd_target = bd->bd_diskid; 2685 strlcpy(bd->bd_vendor, src->src_meta.scmi.scm_devname, 2686 sizeof(bd->bd_vendor)); 2687 rv = 0; 2688 goto done; 2689 } 2690 2691 /* Check hotspares list. */ 2692 SLIST_FOREACH(hotspare, &sc->sc_hotspare_list, src_link) { 2693 vol++; 2694 if (vol != bd->bd_volid) 2695 continue; 2696 2697 if (bd->bd_diskid != 0) 2698 break; 2699 2700 bd->bd_status = hotspare->src_meta.scm_status; 2701 bd->bd_size = hotspare->src_meta.scmi.scm_size << DEV_BSHIFT; 2702 bd->bd_channel = vol; 2703 bd->bd_target = bd->bd_diskid; 2704 strlcpy(bd->bd_vendor, hotspare->src_meta.scmi.scm_devname, 2705 sizeof(bd->bd_vendor)); 2706 rv = 0; 2707 goto done; 2708 } 2709 2710 done: 2711 return (rv); 2712 } 2713 2714 int 2715 sr_ioctl_setstate(struct sr_softc *sc, struct bioc_setstate *bs) 2716 { 2717 int rv = EINVAL; 2718 int vol = -1, found, c; 2719 struct sr_discipline *sd; 2720 struct sr_chunk *ch_entry; 2721 struct sr_chunk_head *cl; 2722 2723 if (bs->bs_other_id_type == BIOC_SSOTHER_UNUSED) 2724 goto done; 2725 2726 if (bs->bs_status == BIOC_SSHOTSPARE) { 2727 rv = sr_hotspare(sc, (dev_t)bs->bs_other_id); 2728 goto done; 2729 } 2730 2731 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2732 vol++; 2733 if (vol == bs->bs_volid) 2734 break; 2735 } 2736 if (sd == NULL) 2737 goto done; 2738 2739 switch (bs->bs_status) { 2740 case BIOC_SSOFFLINE: 2741 /* Take chunk offline */ 2742 found = c = 0; 2743 cl = &sd->sd_vol.sv_chunk_list; 2744 SLIST_FOREACH(ch_entry, cl, src_link) { 2745 if (ch_entry->src_dev_mm == bs->bs_other_id) { 2746 found = 1; 2747 break; 2748 } 2749 c++; 2750 } 2751 if (found == 0) { 2752 sr_error(sc, "chunk not part of array"); 2753 goto done; 2754 } 2755 2756 /* XXX: check current state first */ 2757 sd->sd_set_chunk_state(sd, c, BIOC_SDOFFLINE); 2758 2759 if (sr_meta_save(sd, SR_META_DIRTY)) { 2760 sr_error(sc, "could not save metadata for %s", 2761 sd->sd_meta->ssd_devname); 2762 goto done; 2763 } 2764 rv = 0; 2765 break; 2766 2767 case BIOC_SDSCRUB: 2768 break; 2769 2770 case BIOC_SSREBUILD: 2771 rv = sr_rebuild_init(sd, (dev_t)bs->bs_other_id, 0); 2772 break; 2773 2774 default: 2775 sr_error(sc, "unsupported state request %d", bs->bs_status); 2776 } 2777 2778 done: 2779 return (rv); 2780 } 2781 2782 int 2783 sr_chunk_in_use(struct sr_softc *sc, dev_t dev) 2784 { 2785 struct sr_discipline *sd; 2786 struct sr_chunk *chunk; 2787 int i; 2788 2789 DNPRINTF(SR_D_MISC, "%s: sr_chunk_in_use(%d)\n", DEVNAME(sc), dev); 2790 2791 if (dev == NODEV) 2792 return BIOC_SDINVALID; 2793 2794 /* See if chunk is already in use. */ 2795 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2796 for (i = 0; i < sd->sd_meta->ssdi.ssd_chunk_no; i++) { 2797 chunk = sd->sd_vol.sv_chunks[i]; 2798 if (chunk->src_dev_mm == dev) 2799 return chunk->src_meta.scm_status; 2800 } 2801 } 2802 2803 /* Check hotspares list. */ 2804 SLIST_FOREACH(chunk, &sc->sc_hotspare_list, src_link) 2805 if (chunk->src_dev_mm == dev) 2806 return chunk->src_meta.scm_status; 2807 2808 return BIOC_SDINVALID; 2809 } 2810 2811 int 2812 sr_hotspare(struct sr_softc *sc, dev_t dev) 2813 { 2814 struct sr_discipline *sd = NULL; 2815 struct sr_metadata *sm = NULL; 2816 struct sr_meta_chunk *hm; 2817 struct sr_chunk_head *cl; 2818 struct sr_chunk *chunk, *last, *hotspare = NULL; 2819 struct sr_uuid uuid; 2820 struct disklabel label; 2821 struct vnode *vn; 2822 u_int64_t size; 2823 char devname[32]; 2824 int rv = EINVAL; 2825 int c, part, open = 0; 2826 2827 /* 2828 * Add device to global hotspares list. 2829 */ 2830 2831 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 2832 2833 /* Make sure chunk is not already in use. */ 2834 c = sr_chunk_in_use(sc, dev); 2835 if (c != BIOC_SDINVALID && c != BIOC_SDOFFLINE) { 2836 if (c == BIOC_SDHOTSPARE) 2837 sr_error(sc, "%s is already a hotspare", devname); 2838 else 2839 sr_error(sc, "%s is already in use", devname); 2840 goto done; 2841 } 2842 2843 /* XXX - See if there is an existing degraded volume... */ 2844 2845 /* Open device. */ 2846 if (bdevvp(dev, &vn)) { 2847 sr_error(sc, "sr_hotspare: cannot allocate vnode"); 2848 goto done; 2849 } 2850 if (VOP_OPEN(vn, FREAD | FWRITE, NOCRED, curproc)) { 2851 DNPRINTF(SR_D_META,"%s: sr_hotspare cannot open %s\n", 2852 DEVNAME(sc), devname); 2853 vput(vn); 2854 goto fail; 2855 } 2856 open = 1; /* close dev on error */ 2857 2858 /* Get partition details. */ 2859 part = DISKPART(dev); 2860 if (VOP_IOCTL(vn, DIOCGDINFO, (caddr_t)&label, FREAD, 2861 NOCRED, curproc)) { 2862 DNPRINTF(SR_D_META, "%s: sr_hotspare ioctl failed\n", 2863 DEVNAME(sc)); 2864 VOP_CLOSE(vn, FREAD | FWRITE, NOCRED, curproc); 2865 vput(vn); 2866 goto fail; 2867 } 2868 if (label.d_partitions[part].p_fstype != FS_RAID) { 2869 sr_error(sc, "%s partition not of type RAID (%d)", 2870 devname, label.d_partitions[part].p_fstype); 2871 goto fail; 2872 } 2873 2874 /* Calculate partition size. */ 2875 size = DL_SECTOBLK(&label, DL_GETPSIZE(&label.d_partitions[part])); 2876 if (size <= SR_DATA_OFFSET) { 2877 DNPRINTF(SR_D_META, "%s: %s partition too small\n", DEVNAME(sc), 2878 devname); 2879 goto fail; 2880 } 2881 size -= SR_DATA_OFFSET; 2882 if (size > INT64_MAX) { 2883 DNPRINTF(SR_D_META, "%s: %s partition too large\n", DEVNAME(sc), 2884 devname); 2885 goto fail; 2886 } 2887 2888 /* 2889 * Create and populate chunk metadata. 2890 */ 2891 2892 sr_uuid_generate(&uuid); 2893 hotspare = malloc(sizeof(struct sr_chunk), M_DEVBUF, M_WAITOK | M_ZERO); 2894 2895 hotspare->src_dev_mm = dev; 2896 hotspare->src_vn = vn; 2897 strlcpy(hotspare->src_devname, devname, sizeof(hm->scmi.scm_devname)); 2898 hotspare->src_size = size; 2899 2900 hm = &hotspare->src_meta; 2901 hm->scmi.scm_volid = SR_HOTSPARE_VOLID; 2902 hm->scmi.scm_chunk_id = 0; 2903 hm->scmi.scm_size = size; 2904 hm->scmi.scm_coerced_size = size; 2905 strlcpy(hm->scmi.scm_devname, devname, sizeof(hm->scmi.scm_devname)); 2906 memcpy(&hm->scmi.scm_uuid, &uuid, sizeof(struct sr_uuid)); 2907 2908 sr_checksum(sc, hm, &hm->scm_checksum, 2909 sizeof(struct sr_meta_chunk_invariant)); 2910 2911 hm->scm_status = BIOC_SDHOTSPARE; 2912 2913 /* 2914 * Create and populate our own discipline and metadata. 2915 */ 2916 2917 sm = malloc(sizeof(struct sr_metadata), M_DEVBUF, M_WAITOK | M_ZERO); 2918 sm->ssdi.ssd_magic = SR_MAGIC; 2919 sm->ssdi.ssd_version = SR_META_VERSION; 2920 sm->ssd_ondisk = 0; 2921 sm->ssdi.ssd_vol_flags = 0; 2922 memcpy(&sm->ssdi.ssd_uuid, &uuid, sizeof(struct sr_uuid)); 2923 sm->ssdi.ssd_chunk_no = 1; 2924 sm->ssdi.ssd_volid = SR_HOTSPARE_VOLID; 2925 sm->ssdi.ssd_level = SR_HOTSPARE_LEVEL; 2926 sm->ssdi.ssd_size = size; 2927 sm->ssdi.ssd_secsize = label.d_secsize; 2928 strlcpy(sm->ssdi.ssd_vendor, "OPENBSD", sizeof(sm->ssdi.ssd_vendor)); 2929 snprintf(sm->ssdi.ssd_product, sizeof(sm->ssdi.ssd_product), 2930 "SR %s", "HOTSPARE"); 2931 snprintf(sm->ssdi.ssd_revision, sizeof(sm->ssdi.ssd_revision), 2932 "%03d", SR_META_VERSION); 2933 2934 sd = malloc(sizeof(struct sr_discipline), M_DEVBUF, M_WAITOK | M_ZERO); 2935 sd->sd_sc = sc; 2936 sd->sd_meta = sm; 2937 sd->sd_meta_type = SR_META_F_NATIVE; 2938 sd->sd_vol_status = BIOC_SVONLINE; 2939 strlcpy(sd->sd_name, "HOTSPARE", sizeof(sd->sd_name)); 2940 SLIST_INIT(&sd->sd_meta_opt); 2941 2942 /* Add chunk to volume. */ 2943 sd->sd_vol.sv_chunks = malloc(sizeof(struct sr_chunk *), M_DEVBUF, 2944 M_WAITOK | M_ZERO); 2945 sd->sd_vol.sv_chunks[0] = hotspare; 2946 SLIST_INIT(&sd->sd_vol.sv_chunk_list); 2947 SLIST_INSERT_HEAD(&sd->sd_vol.sv_chunk_list, hotspare, src_link); 2948 2949 /* Save metadata. */ 2950 if (sr_meta_save(sd, SR_META_DIRTY)) { 2951 sr_error(sc, "could not save metadata to %s", devname); 2952 goto fail; 2953 } 2954 2955 /* 2956 * Add chunk to hotspare list. 2957 */ 2958 rw_enter_write(&sc->sc_hs_lock); 2959 cl = &sc->sc_hotspare_list; 2960 if (SLIST_EMPTY(cl)) 2961 SLIST_INSERT_HEAD(cl, hotspare, src_link); 2962 else { 2963 SLIST_FOREACH(chunk, cl, src_link) 2964 last = chunk; 2965 SLIST_INSERT_AFTER(last, hotspare, src_link); 2966 } 2967 sc->sc_hotspare_no++; 2968 rw_exit_write(&sc->sc_hs_lock); 2969 2970 rv = 0; 2971 goto done; 2972 2973 fail: 2974 free(hotspare, M_DEVBUF, 0); 2975 2976 done: 2977 if (sd) 2978 free(sd->sd_vol.sv_chunks, M_DEVBUF, 0); 2979 free(sd, M_DEVBUF, 0); 2980 free(sm, M_DEVBUF, 0); 2981 if (open) { 2982 VOP_CLOSE(vn, FREAD | FWRITE, NOCRED, curproc); 2983 vput(vn); 2984 } 2985 2986 return (rv); 2987 } 2988 2989 void 2990 sr_hotspare_rebuild_callback(void *xsd) 2991 { 2992 struct sr_discipline *sd = xsd; 2993 sr_hotspare_rebuild(sd); 2994 } 2995 2996 void 2997 sr_hotspare_rebuild(struct sr_discipline *sd) 2998 { 2999 struct sr_softc *sc = sd->sd_sc; 3000 struct sr_chunk_head *cl; 3001 struct sr_chunk *hotspare, *chunk = NULL; 3002 struct sr_workunit *wu; 3003 struct sr_ccb *ccb; 3004 int i, s, cid, busy; 3005 3006 /* 3007 * Attempt to locate a hotspare and initiate rebuild. 3008 */ 3009 3010 /* Find first offline chunk. */ 3011 for (cid = 0; cid < sd->sd_meta->ssdi.ssd_chunk_no; cid++) { 3012 if (sd->sd_vol.sv_chunks[cid]->src_meta.scm_status == 3013 BIOC_SDOFFLINE) { 3014 chunk = sd->sd_vol.sv_chunks[cid]; 3015 break; 3016 } 3017 } 3018 if (chunk == NULL) { 3019 printf("%s: no offline chunk found on %s!\n", 3020 DEVNAME(sc), sd->sd_meta->ssd_devname); 3021 return; 3022 } 3023 3024 /* See if we have a suitable hotspare... */ 3025 rw_enter_write(&sc->sc_hs_lock); 3026 cl = &sc->sc_hotspare_list; 3027 SLIST_FOREACH(hotspare, cl, src_link) 3028 if (hotspare->src_size >= chunk->src_size && 3029 hotspare->src_secsize <= sd->sd_meta->ssdi.ssd_secsize) 3030 break; 3031 3032 if (hotspare != NULL) { 3033 3034 printf("%s: %s volume degraded, will attempt to " 3035 "rebuild on hotspare %s\n", DEVNAME(sc), 3036 sd->sd_meta->ssd_devname, hotspare->src_devname); 3037 3038 /* 3039 * Ensure that all pending I/O completes on the failed chunk 3040 * before trying to initiate a rebuild. 3041 */ 3042 i = 0; 3043 do { 3044 busy = 0; 3045 3046 s = splbio(); 3047 TAILQ_FOREACH(wu, &sd->sd_wu_pendq, swu_link) { 3048 TAILQ_FOREACH(ccb, &wu->swu_ccb, ccb_link) { 3049 if (ccb->ccb_target == cid) 3050 busy = 1; 3051 } 3052 } 3053 TAILQ_FOREACH(wu, &sd->sd_wu_defq, swu_link) { 3054 TAILQ_FOREACH(ccb, &wu->swu_ccb, ccb_link) { 3055 if (ccb->ccb_target == cid) 3056 busy = 1; 3057 } 3058 } 3059 splx(s); 3060 3061 if (busy) { 3062 tsleep(sd, PRIBIO, "sr_hotspare", hz); 3063 i++; 3064 } 3065 3066 } while (busy && i < 120); 3067 3068 DNPRINTF(SR_D_META, "%s: waited %i seconds for I/O to " 3069 "complete on failed chunk %s\n", DEVNAME(sc), 3070 i, chunk->src_devname); 3071 3072 if (busy) { 3073 printf("%s: pending I/O failed to complete on " 3074 "failed chunk %s, hotspare rebuild aborted...\n", 3075 DEVNAME(sc), chunk->src_devname); 3076 goto done; 3077 } 3078 3079 s = splbio(); 3080 rw_enter_write(&sc->sc_lock); 3081 bio_status_init(&sc->sc_status, &sc->sc_dev); 3082 if (sr_rebuild_init(sd, hotspare->src_dev_mm, 1) == 0) { 3083 3084 /* Remove hotspare from available list. */ 3085 sc->sc_hotspare_no--; 3086 SLIST_REMOVE(cl, hotspare, sr_chunk, src_link); 3087 free(hotspare, M_DEVBUF, 0); 3088 3089 } 3090 rw_exit_write(&sc->sc_lock); 3091 splx(s); 3092 } 3093 done: 3094 rw_exit_write(&sc->sc_hs_lock); 3095 } 3096 3097 int 3098 sr_rebuild_init(struct sr_discipline *sd, dev_t dev, int hotspare) 3099 { 3100 struct sr_softc *sc = sd->sd_sc; 3101 struct sr_chunk *chunk = NULL; 3102 struct sr_meta_chunk *meta; 3103 struct disklabel label; 3104 struct vnode *vn; 3105 u_int64_t size; 3106 int64_t csize; 3107 char devname[32]; 3108 int rv = EINVAL, open = 0; 3109 int cid, i, part, status; 3110 3111 /* 3112 * Attempt to initiate a rebuild onto the specified device. 3113 */ 3114 3115 if (!(sd->sd_capabilities & SR_CAP_REBUILD)) { 3116 sr_error(sc, "discipline does not support rebuild"); 3117 goto done; 3118 } 3119 3120 /* make sure volume is in the right state */ 3121 if (sd->sd_vol_status == BIOC_SVREBUILD) { 3122 sr_error(sc, "rebuild already in progress"); 3123 goto done; 3124 } 3125 if (sd->sd_vol_status != BIOC_SVDEGRADED) { 3126 sr_error(sc, "volume not degraded"); 3127 goto done; 3128 } 3129 3130 /* Find first offline chunk. */ 3131 for (cid = 0; cid < sd->sd_meta->ssdi.ssd_chunk_no; cid++) { 3132 if (sd->sd_vol.sv_chunks[cid]->src_meta.scm_status == 3133 BIOC_SDOFFLINE) { 3134 chunk = sd->sd_vol.sv_chunks[cid]; 3135 break; 3136 } 3137 } 3138 if (chunk == NULL) { 3139 sr_error(sc, "no offline chunks available to rebuild"); 3140 goto done; 3141 } 3142 3143 /* Get coerced size from another online chunk. */ 3144 csize = 0; 3145 for (i = 0; i < sd->sd_meta->ssdi.ssd_chunk_no; i++) { 3146 if (sd->sd_vol.sv_chunks[i]->src_meta.scm_status == 3147 BIOC_SDONLINE) { 3148 meta = &sd->sd_vol.sv_chunks[i]->src_meta; 3149 csize = meta->scmi.scm_coerced_size; 3150 break; 3151 } 3152 } 3153 if (csize == 0) { 3154 sr_error(sc, "no online chunks available for rebuild"); 3155 goto done; 3156 } 3157 3158 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 3159 if (bdevvp(dev, &vn)) { 3160 printf("%s: sr_rebuild_init: can't allocate vnode\n", 3161 DEVNAME(sc)); 3162 goto done; 3163 } 3164 if (VOP_OPEN(vn, FREAD | FWRITE, NOCRED, curproc)) { 3165 DNPRINTF(SR_D_META,"%s: sr_ioctl_setstate can't " 3166 "open %s\n", DEVNAME(sc), devname); 3167 vput(vn); 3168 goto done; 3169 } 3170 open = 1; /* close dev on error */ 3171 3172 /* Get disklabel and check partition. */ 3173 part = DISKPART(dev); 3174 if (VOP_IOCTL(vn, DIOCGDINFO, (caddr_t)&label, FREAD, 3175 NOCRED, curproc)) { 3176 DNPRINTF(SR_D_META, "%s: sr_ioctl_setstate ioctl failed\n", 3177 DEVNAME(sc)); 3178 goto done; 3179 } 3180 if (label.d_partitions[part].p_fstype != FS_RAID) { 3181 sr_error(sc, "%s partition not of type RAID (%d)", 3182 devname, label.d_partitions[part].p_fstype); 3183 goto done; 3184 } 3185 3186 /* Is the partition large enough? */ 3187 size = DL_SECTOBLK(&label, DL_GETPSIZE(&label.d_partitions[part])); 3188 if (size <= sd->sd_meta->ssd_data_blkno) { 3189 sr_error(sc, "%s: %s partition too small", DEVNAME(sc), 3190 devname); 3191 goto done; 3192 } 3193 size -= sd->sd_meta->ssd_data_blkno; 3194 if (size > INT64_MAX) { 3195 sr_error(sc, "%s: %s partition too large", DEVNAME(sc), 3196 devname); 3197 goto done; 3198 } 3199 if (size < csize) { 3200 sr_error(sc, "%s partition too small, at least %lld bytes " 3201 "required", devname, (long long)(csize << DEV_BSHIFT)); 3202 goto done; 3203 } else if (size > csize) 3204 sr_warn(sc, "%s partition too large, wasting %lld bytes", 3205 devname, (long long)((size - csize) << DEV_BSHIFT)); 3206 if (label.d_secsize > sd->sd_meta->ssdi.ssd_secsize) { 3207 sr_error(sc, "%s sector size too large, <= %u bytes " 3208 "required", devname, sd->sd_meta->ssdi.ssd_secsize); 3209 goto done; 3210 } 3211 3212 /* Ensure that this chunk is not already in use. */ 3213 status = sr_chunk_in_use(sc, dev); 3214 if (status != BIOC_SDINVALID && status != BIOC_SDOFFLINE && 3215 !(hotspare && status == BIOC_SDHOTSPARE)) { 3216 sr_error(sc, "%s is already in use", devname); 3217 goto done; 3218 } 3219 3220 /* Reset rebuild counter since we rebuilding onto a new chunk. */ 3221 sd->sd_meta->ssd_rebuild = 0; 3222 3223 open = 0; /* leave dev open from here on out */ 3224 3225 /* Fix up chunk. */ 3226 memcpy(chunk->src_duid, label.d_uid, sizeof(chunk->src_duid)); 3227 chunk->src_dev_mm = dev; 3228 chunk->src_vn = vn; 3229 3230 /* Reconstruct metadata. */ 3231 meta = &chunk->src_meta; 3232 meta->scmi.scm_volid = sd->sd_meta->ssdi.ssd_volid; 3233 meta->scmi.scm_chunk_id = cid; 3234 strlcpy(meta->scmi.scm_devname, devname, 3235 sizeof(meta->scmi.scm_devname)); 3236 meta->scmi.scm_size = size; 3237 meta->scmi.scm_coerced_size = csize; 3238 memcpy(&meta->scmi.scm_uuid, &sd->sd_meta->ssdi.ssd_uuid, 3239 sizeof(meta->scmi.scm_uuid)); 3240 sr_checksum(sc, meta, &meta->scm_checksum, 3241 sizeof(struct sr_meta_chunk_invariant)); 3242 3243 sd->sd_set_chunk_state(sd, cid, BIOC_SDREBUILD); 3244 3245 if (sr_meta_save(sd, SR_META_DIRTY)) { 3246 sr_error(sc, "could not save metadata to %s", devname); 3247 open = 1; 3248 goto done; 3249 } 3250 3251 sr_warn(sc, "rebuild of %s started on %s", 3252 sd->sd_meta->ssd_devname, devname); 3253 3254 sd->sd_reb_abort = 0; 3255 kthread_create_deferred(sr_rebuild_start, sd); 3256 3257 rv = 0; 3258 done: 3259 if (open) { 3260 VOP_CLOSE(vn, FREAD | FWRITE, NOCRED, curproc); 3261 vput(vn); 3262 } 3263 3264 return (rv); 3265 } 3266 3267 int 3268 sr_rebuild_percent(struct sr_discipline *sd) 3269 { 3270 daddr_t rb, sz; 3271 3272 sz = sd->sd_meta->ssdi.ssd_size; 3273 rb = sd->sd_meta->ssd_rebuild; 3274 3275 if (rb > 0) 3276 return (100 - ((sz * 100 - rb * 100) / sz) - 1); 3277 3278 return (0); 3279 } 3280 3281 void 3282 sr_roam_chunks(struct sr_discipline *sd) 3283 { 3284 struct sr_softc *sc = sd->sd_sc; 3285 struct sr_chunk *chunk; 3286 struct sr_meta_chunk *meta; 3287 int roamed = 0; 3288 3289 /* Have any chunks roamed? */ 3290 SLIST_FOREACH(chunk, &sd->sd_vol.sv_chunk_list, src_link) { 3291 meta = &chunk->src_meta; 3292 if (strncmp(meta->scmi.scm_devname, chunk->src_devname, 3293 sizeof(meta->scmi.scm_devname))) { 3294 3295 printf("%s: roaming device %s -> %s\n", DEVNAME(sc), 3296 meta->scmi.scm_devname, chunk->src_devname); 3297 3298 strlcpy(meta->scmi.scm_devname, chunk->src_devname, 3299 sizeof(meta->scmi.scm_devname)); 3300 3301 roamed++; 3302 } 3303 } 3304 3305 if (roamed) 3306 sr_meta_save(sd, SR_META_DIRTY); 3307 } 3308 3309 int 3310 sr_ioctl_createraid(struct sr_softc *sc, struct bioc_createraid *bc, 3311 int user, void *data) 3312 { 3313 struct sr_meta_opt_item *omi; 3314 struct sr_chunk_head *cl; 3315 struct sr_discipline *sd = NULL; 3316 struct sr_chunk *ch_entry; 3317 struct scsi_link *link; 3318 struct device *dev; 3319 char *uuid, devname[32]; 3320 dev_t *dt = NULL; 3321 int i, no_chunk, rv = EINVAL, target, vol; 3322 int no_meta; 3323 3324 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_createraid(%d)\n", 3325 DEVNAME(sc), user); 3326 3327 /* user input */ 3328 if (bc->bc_dev_list_len > BIOC_CRMAXLEN) 3329 goto unwind; 3330 3331 dt = malloc(bc->bc_dev_list_len, M_DEVBUF, M_WAITOK | M_ZERO); 3332 if (user) { 3333 if (copyin(bc->bc_dev_list, dt, bc->bc_dev_list_len) != 0) 3334 goto unwind; 3335 } else 3336 memcpy(dt, bc->bc_dev_list, bc->bc_dev_list_len); 3337 3338 /* Initialise discipline. */ 3339 sd = malloc(sizeof(struct sr_discipline), M_DEVBUF, M_WAITOK | M_ZERO); 3340 sd->sd_sc = sc; 3341 SLIST_INIT(&sd->sd_meta_opt); 3342 sd->sd_taskq = taskq_create("srdis", 1, IPL_BIO, 0); 3343 if (sd->sd_taskq == NULL) { 3344 sr_error(sc, "could not create discipline taskq"); 3345 goto unwind; 3346 } 3347 if (sr_discipline_init(sd, bc->bc_level)) { 3348 sr_error(sc, "could not initialize discipline"); 3349 goto unwind; 3350 } 3351 3352 no_chunk = bc->bc_dev_list_len / sizeof(dev_t); 3353 cl = &sd->sd_vol.sv_chunk_list; 3354 SLIST_INIT(cl); 3355 3356 /* Ensure that chunks are not already in use. */ 3357 for (i = 0; i < no_chunk; i++) { 3358 if (sr_chunk_in_use(sc, dt[i]) != BIOC_SDINVALID) { 3359 sr_meta_getdevname(sc, dt[i], devname, sizeof(devname)); 3360 sr_error(sc, "chunk %s already in use", devname); 3361 goto unwind; 3362 } 3363 } 3364 3365 sd->sd_meta_type = sr_meta_probe(sd, dt, no_chunk); 3366 if (sd->sd_meta_type == SR_META_F_INVALID) { 3367 sr_error(sc, "invalid metadata format"); 3368 goto unwind; 3369 } 3370 3371 if (sr_meta_attach(sd, no_chunk, bc->bc_flags & BIOC_SCFORCE)) 3372 goto unwind; 3373 3374 /* force the raid volume by clearing metadata region */ 3375 if (bc->bc_flags & BIOC_SCFORCE) { 3376 /* make sure disk isn't up and running */ 3377 if (sr_meta_read(sd)) 3378 if (sr_already_assembled(sd)) { 3379 uuid = sr_uuid_format( 3380 &sd->sd_meta->ssdi.ssd_uuid); 3381 sr_error(sc, "disk %s is currently in use; " 3382 "cannot force create", uuid); 3383 free(uuid, M_DEVBUF, 0); 3384 goto unwind; 3385 } 3386 3387 if (sr_meta_clear(sd)) { 3388 sr_error(sc, "failed to clear metadata"); 3389 goto unwind; 3390 } 3391 } 3392 3393 no_meta = sr_meta_read(sd); 3394 if (no_meta == -1) { 3395 3396 /* Corrupt metadata on one or more chunks. */ 3397 sr_error(sc, "one of the chunks has corrupt metadata; " 3398 "aborting assembly"); 3399 goto unwind; 3400 3401 } else if (no_meta == 0) { 3402 3403 /* Initialise volume and chunk metadata. */ 3404 sr_meta_init(sd, bc->bc_level, no_chunk); 3405 sd->sd_vol_status = BIOC_SVONLINE; 3406 sd->sd_meta_flags = bc->bc_flags & BIOC_SCNOAUTOASSEMBLE; 3407 if (sd->sd_create) { 3408 if ((i = sd->sd_create(sd, bc, no_chunk, 3409 sd->sd_vol.sv_chunk_minsz))) { 3410 rv = i; 3411 goto unwind; 3412 } 3413 } 3414 sr_meta_init_complete(sd); 3415 3416 DNPRINTF(SR_D_IOCTL, 3417 "%s: sr_ioctl_createraid: vol_size: %lld\n", 3418 DEVNAME(sc), sd->sd_meta->ssdi.ssd_size); 3419 3420 /* Warn if we've wasted chunk space due to coercing. */ 3421 if ((sd->sd_capabilities & SR_CAP_NON_COERCED) == 0 && 3422 sd->sd_vol.sv_chunk_minsz != sd->sd_vol.sv_chunk_maxsz) 3423 sr_warn(sc, "chunk sizes are not equal; up to %llu " 3424 "blocks wasted per chunk", 3425 sd->sd_vol.sv_chunk_maxsz - 3426 sd->sd_vol.sv_chunk_minsz); 3427 3428 } else { 3429 3430 /* Ensure we are assembling the correct # of chunks. */ 3431 if (sd->sd_meta->ssdi.ssd_chunk_no != no_chunk) { 3432 sr_error(sc, "volume chunk count does not match metadata " 3433 "chunk count"); 3434 goto unwind; 3435 } 3436 3437 /* Ensure metadata level matches requested assembly level. */ 3438 if (sd->sd_meta->ssdi.ssd_level != bc->bc_level) { 3439 sr_error(sc, "volume level does not match metadata " 3440 "level"); 3441 goto unwind; 3442 } 3443 3444 if (sr_already_assembled(sd)) { 3445 uuid = sr_uuid_format(&sd->sd_meta->ssdi.ssd_uuid); 3446 sr_error(sc, "disk %s already assembled", uuid); 3447 free(uuid, M_DEVBUF, 0); 3448 goto unwind; 3449 } 3450 3451 if (user == 0 && sd->sd_meta_flags & BIOC_SCNOAUTOASSEMBLE) { 3452 DNPRINTF(SR_D_META, "%s: disk not auto assembled from " 3453 "metadata\n", DEVNAME(sc)); 3454 goto unwind; 3455 } 3456 3457 if (no_meta != no_chunk) 3458 sr_warn(sc, "trying to bring up %s degraded", 3459 sd->sd_meta->ssd_devname); 3460 3461 if (sd->sd_meta->ssd_meta_flags & SR_META_DIRTY) 3462 sr_warn(sc, "%s was not shutdown properly", 3463 sd->sd_meta->ssd_devname); 3464 3465 SLIST_FOREACH(omi, &sd->sd_meta_opt, omi_link) 3466 if (sd->sd_meta_opt_handler == NULL || 3467 sd->sd_meta_opt_handler(sd, omi->omi_som) != 0) 3468 sr_meta_opt_handler(sd, omi->omi_som); 3469 3470 if (sd->sd_assemble) { 3471 if ((i = sd->sd_assemble(sd, bc, no_chunk, data))) { 3472 rv = i; 3473 goto unwind; 3474 } 3475 } 3476 3477 DNPRINTF(SR_D_META, "%s: disk assembled from metadata\n", 3478 DEVNAME(sc)); 3479 3480 } 3481 3482 /* Metadata MUST be fully populated by this point. */ 3483 TAILQ_INSERT_TAIL(&sc->sc_dis_list, sd, sd_link); 3484 3485 /* Allocate all resources. */ 3486 if ((rv = sd->sd_alloc_resources(sd))) 3487 goto unwind; 3488 3489 /* Adjust flags if necessary. */ 3490 if ((sd->sd_capabilities & SR_CAP_AUTO_ASSEMBLE) && 3491 (bc->bc_flags & BIOC_SCNOAUTOASSEMBLE) != 3492 (sd->sd_meta->ssdi.ssd_vol_flags & BIOC_SCNOAUTOASSEMBLE)) { 3493 sd->sd_meta->ssdi.ssd_vol_flags &= ~BIOC_SCNOAUTOASSEMBLE; 3494 sd->sd_meta->ssdi.ssd_vol_flags |= 3495 bc->bc_flags & BIOC_SCNOAUTOASSEMBLE; 3496 } 3497 3498 if (sd->sd_capabilities & SR_CAP_SYSTEM_DISK) { 3499 /* Initialise volume state. */ 3500 sd->sd_set_vol_state(sd); 3501 if (sd->sd_vol_status == BIOC_SVOFFLINE) { 3502 sr_error(sc, "%s is offline, will not be brought " 3503 "online", sd->sd_meta->ssd_devname); 3504 goto unwind; 3505 } 3506 3507 /* Setup SCSI iopool. */ 3508 scsi_iopool_init(&sd->sd_iopool, sd, sr_wu_get, sr_wu_put); 3509 3510 /* 3511 * All checks passed - return ENXIO if volume cannot be created. 3512 */ 3513 rv = ENXIO; 3514 3515 /* 3516 * Find a free target. 3517 * 3518 * XXX: We reserve sd_target == 0 to indicate the 3519 * discipline is not linked into sc->sc_targets, so begin 3520 * the search with target = 1. 3521 */ 3522 for (target = 1; target < SR_MAX_LD; target++) 3523 if (sc->sc_targets[target] == NULL) 3524 break; 3525 if (target == SR_MAX_LD) { 3526 sr_error(sc, "no free target for %s", 3527 sd->sd_meta->ssd_devname); 3528 goto unwind; 3529 } 3530 3531 /* Clear sense data. */ 3532 bzero(&sd->sd_scsi_sense, sizeof(sd->sd_scsi_sense)); 3533 3534 /* Attach discipline and get midlayer to probe it. */ 3535 sd->sd_target = target; 3536 sc->sc_targets[target] = sd; 3537 if (scsi_probe_lun(sc->sc_scsibus, target, 0) != 0) { 3538 sr_error(sc, "scsi_probe_lun failed"); 3539 sc->sc_targets[target] = NULL; 3540 sd->sd_target = 0; 3541 goto unwind; 3542 } 3543 3544 link = scsi_get_link(sc->sc_scsibus, target, 0); 3545 if (link == NULL) 3546 goto unwind; 3547 3548 dev = link->device_softc; 3549 DNPRINTF(SR_D_IOCTL, "%s: sr device added: %s at target %d\n", 3550 DEVNAME(sc), dev->dv_xname, sd->sd_target); 3551 3552 /* XXX - Count volumes, not targets. */ 3553 for (i = 0, vol = -1; i <= sd->sd_target; i++) 3554 if (sc->sc_targets[i]) 3555 vol++; 3556 3557 rv = 0; 3558 3559 if (sd->sd_meta->ssd_devname[0] != '\0' && 3560 strncmp(sd->sd_meta->ssd_devname, dev->dv_xname, 3561 sizeof(dev->dv_xname))) 3562 sr_warn(sc, "volume %s is roaming, it used to be %s, " 3563 "updating metadata", dev->dv_xname, 3564 sd->sd_meta->ssd_devname); 3565 3566 /* Populate remaining volume metadata. */ 3567 sd->sd_meta->ssdi.ssd_volid = vol; 3568 strlcpy(sd->sd_meta->ssd_devname, dev->dv_xname, 3569 sizeof(sd->sd_meta->ssd_devname)); 3570 3571 sr_info(sc, "%s volume attached as %s", 3572 sd->sd_name, sd->sd_meta->ssd_devname); 3573 3574 /* Update device name on any roaming chunks. */ 3575 sr_roam_chunks(sd); 3576 3577 #ifndef SMALL_KERNEL 3578 if (sr_sensors_create(sd)) 3579 sr_warn(sc, "unable to create sensor for %s", 3580 dev->dv_xname); 3581 #endif /* SMALL_KERNEL */ 3582 } else { 3583 /* This volume does not attach as a system disk. */ 3584 ch_entry = SLIST_FIRST(cl); /* XXX */ 3585 strlcpy(sd->sd_meta->ssd_devname, ch_entry->src_devname, 3586 sizeof(sd->sd_meta->ssd_devname)); 3587 3588 if (sd->sd_start_discipline(sd)) 3589 goto unwind; 3590 } 3591 3592 /* Save current metadata to disk. */ 3593 rv = sr_meta_save(sd, SR_META_DIRTY); 3594 3595 if (sd->sd_vol_status == BIOC_SVREBUILD) 3596 kthread_create_deferred(sr_rebuild_start, sd); 3597 3598 sd->sd_ready = 1; 3599 3600 free(dt, M_DEVBUF, bc->bc_dev_list_len); 3601 3602 return (rv); 3603 3604 unwind: 3605 free(dt, M_DEVBUF, bc->bc_dev_list_len); 3606 3607 sr_discipline_shutdown(sd, 0, 0); 3608 3609 if (rv == EAGAIN) 3610 rv = 0; 3611 3612 return (rv); 3613 } 3614 3615 int 3616 sr_ioctl_deleteraid(struct sr_softc *sc, struct sr_discipline *sd, 3617 struct bioc_deleteraid *bd) 3618 { 3619 int rv = 1; 3620 3621 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_deleteraid %s\n", 3622 DEVNAME(sc), bd->bd_dev); 3623 3624 if (sd == NULL) { 3625 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 3626 if (!strncmp(sd->sd_meta->ssd_devname, bd->bd_dev, 3627 sizeof(sd->sd_meta->ssd_devname))) 3628 break; 3629 } 3630 if (sd == NULL) { 3631 sr_error(sc, "volume %s not found", bd->bd_dev); 3632 goto bad; 3633 } 3634 } 3635 3636 sd->sd_deleted = 1; 3637 sd->sd_meta->ssdi.ssd_vol_flags = BIOC_SCNOAUTOASSEMBLE; 3638 sr_discipline_shutdown(sd, 1, 0); 3639 3640 rv = 0; 3641 bad: 3642 return (rv); 3643 } 3644 3645 int 3646 sr_ioctl_discipline(struct sr_softc *sc, struct sr_discipline *sd, 3647 struct bioc_discipline *bd) 3648 { 3649 int rv = 1; 3650 3651 /* Dispatch a discipline specific ioctl. */ 3652 3653 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_discipline %s\n", DEVNAME(sc), 3654 bd->bd_dev); 3655 3656 if (sd == NULL) { 3657 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 3658 if (!strncmp(sd->sd_meta->ssd_devname, bd->bd_dev, 3659 sizeof(sd->sd_meta->ssd_devname))) 3660 break; 3661 } 3662 if (sd == NULL) { 3663 sr_error(sc, "volume %s not found", bd->bd_dev); 3664 goto bad; 3665 } 3666 } 3667 3668 if (sd->sd_ioctl_handler) 3669 rv = sd->sd_ioctl_handler(sd, bd); 3670 3671 bad: 3672 return (rv); 3673 } 3674 3675 int 3676 sr_ioctl_installboot(struct sr_softc *sc, struct sr_discipline *sd, 3677 struct bioc_installboot *bb) 3678 { 3679 void *bootblk = NULL, *bootldr = NULL; 3680 struct sr_chunk *chunk; 3681 struct sr_meta_opt_item *omi; 3682 struct sr_meta_boot *sbm; 3683 struct disk *dk; 3684 u_int32_t bbs, bls, secsize; 3685 u_char duid[8]; 3686 int rv = EINVAL; 3687 int i; 3688 3689 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_installboot %s\n", DEVNAME(sc), 3690 bb->bb_dev); 3691 3692 if (sd == NULL) { 3693 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 3694 if (!strncmp(sd->sd_meta->ssd_devname, bb->bb_dev, 3695 sizeof(sd->sd_meta->ssd_devname))) 3696 break; 3697 } 3698 if (sd == NULL) { 3699 sr_error(sc, "volume %s not found", bb->bb_dev); 3700 goto done; 3701 } 3702 } 3703 3704 bzero(duid, sizeof(duid)); 3705 TAILQ_FOREACH(dk, &disklist, dk_link) 3706 if (!strncmp(dk->dk_name, bb->bb_dev, sizeof(bb->bb_dev))) 3707 break; 3708 if (dk == NULL || dk->dk_label == NULL || 3709 (dk->dk_flags & DKF_LABELVALID) == 0 || 3710 bcmp(dk->dk_label->d_uid, &duid, sizeof(duid)) == 0) { 3711 sr_error(sc, "failed to get DUID for softraid volume"); 3712 goto done; 3713 } 3714 memcpy(duid, dk->dk_label->d_uid, sizeof(duid)); 3715 3716 /* Ensure that boot storage area is large enough. */ 3717 if (sd->sd_meta->ssd_data_blkno < (SR_BOOT_OFFSET + SR_BOOT_SIZE)) { 3718 sr_error(sc, "insufficient boot storage"); 3719 goto done; 3720 } 3721 3722 if (bb->bb_bootblk_size > SR_BOOT_BLOCKS_SIZE * DEV_BSIZE) 3723 goto done; 3724 3725 if (bb->bb_bootldr_size > SR_BOOT_LOADER_SIZE * DEV_BSIZE) 3726 goto done; 3727 3728 secsize = sd->sd_meta->ssdi.ssd_secsize; 3729 3730 /* Copy in boot block. */ 3731 bbs = howmany(bb->bb_bootblk_size, secsize) * secsize; 3732 bootblk = malloc(bbs, M_DEVBUF, M_WAITOK | M_ZERO); 3733 if (copyin(bb->bb_bootblk, bootblk, bb->bb_bootblk_size) != 0) 3734 goto done; 3735 3736 /* Copy in boot loader. */ 3737 bls = howmany(bb->bb_bootldr_size, secsize) * secsize; 3738 bootldr = malloc(bls, M_DEVBUF, M_WAITOK | M_ZERO); 3739 if (copyin(bb->bb_bootldr, bootldr, bb->bb_bootldr_size) != 0) 3740 goto done; 3741 3742 /* Create or update optional meta for bootable volumes. */ 3743 SLIST_FOREACH(omi, &sd->sd_meta_opt, omi_link) 3744 if (omi->omi_som->som_type == SR_OPT_BOOT) 3745 break; 3746 if (omi == NULL) { 3747 omi = malloc(sizeof(struct sr_meta_opt_item), M_DEVBUF, 3748 M_WAITOK | M_ZERO); 3749 omi->omi_som = malloc(sizeof(struct sr_meta_crypto), M_DEVBUF, 3750 M_WAITOK | M_ZERO); 3751 omi->omi_som->som_type = SR_OPT_BOOT; 3752 omi->omi_som->som_length = sizeof(struct sr_meta_boot); 3753 SLIST_INSERT_HEAD(&sd->sd_meta_opt, omi, omi_link); 3754 sd->sd_meta->ssdi.ssd_opt_no++; 3755 } 3756 sbm = (struct sr_meta_boot *)omi->omi_som; 3757 3758 memcpy(sbm->sbm_root_duid, duid, sizeof(sbm->sbm_root_duid)); 3759 bzero(&sbm->sbm_boot_duid, sizeof(sbm->sbm_boot_duid)); 3760 sbm->sbm_bootblk_size = bbs; 3761 sbm->sbm_bootldr_size = bls; 3762 3763 DNPRINTF(SR_D_IOCTL, "sr_ioctl_installboot: root duid is " 3764 "%02x%02x%02x%02x%02x%02x%02x%02x\n", sbm->sbm_root_duid[0], 3765 sbm->sbm_root_duid[1], sbm->sbm_root_duid[2], sbm->sbm_root_duid[3], 3766 sbm->sbm_root_duid[4], sbm->sbm_root_duid[5], sbm->sbm_root_duid[6], 3767 sbm->sbm_root_duid[7]); 3768 3769 /* Save boot block and boot loader to each chunk. */ 3770 for (i = 0; i < sd->sd_meta->ssdi.ssd_chunk_no; i++) { 3771 3772 chunk = sd->sd_vol.sv_chunks[i]; 3773 if (chunk->src_meta.scm_status != BIOC_SDONLINE && 3774 chunk->src_meta.scm_status != BIOC_SDREBUILD) 3775 continue; 3776 3777 if (i < SR_MAX_BOOT_DISKS) 3778 memcpy(&sbm->sbm_boot_duid[i], chunk->src_duid, 3779 sizeof(sbm->sbm_boot_duid[i])); 3780 3781 /* Save boot blocks. */ 3782 DNPRINTF(SR_D_IOCTL, 3783 "sr_ioctl_installboot: saving boot block to %s " 3784 "(%u bytes)\n", chunk->src_devname, bbs); 3785 3786 if (sr_rw(sc, chunk->src_dev_mm, bootblk, bbs, 3787 SR_BOOT_BLOCKS_OFFSET, B_WRITE)) { 3788 sr_error(sc, "failed to write boot block", DEVNAME(sc)); 3789 goto done; 3790 } 3791 3792 /* Save boot loader.*/ 3793 DNPRINTF(SR_D_IOCTL, 3794 "sr_ioctl_installboot: saving boot loader to %s " 3795 "(%u bytes)\n", chunk->src_devname, bls); 3796 3797 if (sr_rw(sc, chunk->src_dev_mm, bootldr, bls, 3798 SR_BOOT_LOADER_OFFSET, B_WRITE)) { 3799 sr_error(sc, "failed to write boot loader"); 3800 goto done; 3801 } 3802 } 3803 3804 /* XXX - Install boot block on disk - MD code. */ 3805 3806 /* Mark volume as bootable and save metadata. */ 3807 sd->sd_meta->ssdi.ssd_vol_flags |= BIOC_SCBOOTABLE; 3808 if (sr_meta_save(sd, SR_META_DIRTY)) { 3809 sr_error(sc, "could not save metadata to %s", DEVNAME(sc)); 3810 goto done; 3811 } 3812 3813 rv = 0; 3814 3815 done: 3816 free(bootblk, M_DEVBUF, 0); 3817 free(bootldr, M_DEVBUF, 0); 3818 3819 return (rv); 3820 } 3821 3822 void 3823 sr_chunks_unwind(struct sr_softc *sc, struct sr_chunk_head *cl) 3824 { 3825 struct sr_chunk *ch_entry, *ch_next; 3826 3827 DNPRINTF(SR_D_IOCTL, "%s: sr_chunks_unwind\n", DEVNAME(sc)); 3828 3829 if (!cl) 3830 return; 3831 3832 for (ch_entry = SLIST_FIRST(cl); ch_entry != NULL; ch_entry = ch_next) { 3833 ch_next = SLIST_NEXT(ch_entry, src_link); 3834 3835 DNPRINTF(SR_D_IOCTL, "%s: sr_chunks_unwind closing: %s\n", 3836 DEVNAME(sc), ch_entry->src_devname); 3837 if (ch_entry->src_vn) { 3838 /* 3839 * XXX - explicitly lock the vnode until we can resolve 3840 * the problem introduced by vnode aliasing... specfs 3841 * has no locking, whereas ufs/ffs does! 3842 */ 3843 vn_lock(ch_entry->src_vn, LK_EXCLUSIVE | 3844 LK_RETRY, curproc); 3845 VOP_CLOSE(ch_entry->src_vn, FREAD | FWRITE, NOCRED, 3846 curproc); 3847 vput(ch_entry->src_vn); 3848 } 3849 free(ch_entry, M_DEVBUF, 0); 3850 } 3851 SLIST_INIT(cl); 3852 } 3853 3854 void 3855 sr_discipline_free(struct sr_discipline *sd) 3856 { 3857 struct sr_softc *sc; 3858 struct sr_discipline *sdtmp1; 3859 struct sr_meta_opt_head *som; 3860 struct sr_meta_opt_item *omi, *omi_next; 3861 3862 if (!sd) 3863 return; 3864 3865 sc = sd->sd_sc; 3866 3867 DNPRINTF(SR_D_DIS, "%s: sr_discipline_free %s\n", 3868 DEVNAME(sc), 3869 sd->sd_meta ? sd->sd_meta->ssd_devname : "nodev"); 3870 if (sd->sd_free_resources) 3871 sd->sd_free_resources(sd); 3872 free(sd->sd_vol.sv_chunks, M_DEVBUF, 0); 3873 free(sd->sd_meta, M_DEVBUF, 0); 3874 free(sd->sd_meta_foreign, M_DEVBUF, 0); 3875 3876 som = &sd->sd_meta_opt; 3877 for (omi = SLIST_FIRST(som); omi != NULL; omi = omi_next) { 3878 omi_next = SLIST_NEXT(omi, omi_link); 3879 free(omi->omi_som, M_DEVBUF, 0); 3880 free(omi, M_DEVBUF, 0); 3881 } 3882 3883 if (sd->sd_target != 0) { 3884 KASSERT(sc->sc_targets[sd->sd_target] == sd); 3885 sc->sc_targets[sd->sd_target] = NULL; 3886 } 3887 3888 TAILQ_FOREACH(sdtmp1, &sc->sc_dis_list, sd_link) { 3889 if (sdtmp1 == sd) 3890 break; 3891 } 3892 if (sdtmp1 != NULL) 3893 TAILQ_REMOVE(&sc->sc_dis_list, sd, sd_link); 3894 3895 explicit_bzero(sd, sizeof *sd); 3896 free(sd, M_DEVBUF, 0); 3897 } 3898 3899 void 3900 sr_discipline_shutdown(struct sr_discipline *sd, int meta_save, int dying) 3901 { 3902 struct sr_softc *sc; 3903 int s; 3904 3905 if (!sd) 3906 return; 3907 sc = sd->sd_sc; 3908 3909 DNPRINTF(SR_D_DIS, "%s: sr_discipline_shutdown %s\n", DEVNAME(sc), 3910 sd->sd_meta ? sd->sd_meta->ssd_devname : "nodev"); 3911 3912 /* If rebuilding, abort rebuild and drain I/O. */ 3913 if (sd->sd_reb_active) { 3914 sd->sd_reb_abort = 1; 3915 while (sd->sd_reb_active) 3916 tsleep(sd, PWAIT, "sr_shutdown", 1); 3917 } 3918 3919 if (meta_save) 3920 sr_meta_save(sd, 0); 3921 3922 s = splbio(); 3923 3924 sd->sd_ready = 0; 3925 3926 /* make sure there isn't a sync pending and yield */ 3927 wakeup(sd); 3928 while (sd->sd_sync || sd->sd_must_flush) 3929 if (tsleep(&sd->sd_sync, MAXPRI, "sr_down", 60 * hz) == 3930 EWOULDBLOCK) 3931 break; 3932 3933 #ifndef SMALL_KERNEL 3934 sr_sensors_delete(sd); 3935 #endif /* SMALL_KERNEL */ 3936 3937 if (sd->sd_target != 0) 3938 scsi_detach_lun(sc->sc_scsibus, sd->sd_target, 0, 3939 dying ? 0 : DETACH_FORCE); 3940 3941 sr_chunks_unwind(sc, &sd->sd_vol.sv_chunk_list); 3942 3943 if (sd->sd_taskq) 3944 taskq_destroy(sd->sd_taskq); 3945 3946 sr_discipline_free(sd); 3947 3948 splx(s); 3949 } 3950 3951 int 3952 sr_discipline_init(struct sr_discipline *sd, int level) 3953 { 3954 int rv = 1; 3955 3956 /* Initialise discipline function pointers with defaults. */ 3957 sd->sd_alloc_resources = sr_alloc_resources; 3958 sd->sd_assemble = NULL; 3959 sd->sd_create = NULL; 3960 sd->sd_free_resources = sr_free_resources; 3961 sd->sd_ioctl_handler = NULL; 3962 sd->sd_openings = NULL; 3963 sd->sd_meta_opt_handler = NULL; 3964 sd->sd_rebuild = sr_rebuild; 3965 sd->sd_scsi_inquiry = sr_raid_inquiry; 3966 sd->sd_scsi_read_cap = sr_raid_read_cap; 3967 sd->sd_scsi_tur = sr_raid_tur; 3968 sd->sd_scsi_req_sense = sr_raid_request_sense; 3969 sd->sd_scsi_start_stop = sr_raid_start_stop; 3970 sd->sd_scsi_sync = sr_raid_sync; 3971 sd->sd_scsi_rw = NULL; 3972 sd->sd_scsi_intr = sr_raid_intr; 3973 sd->sd_scsi_wu_done = NULL; 3974 sd->sd_scsi_done = NULL; 3975 sd->sd_set_chunk_state = sr_set_chunk_state; 3976 sd->sd_set_vol_state = sr_set_vol_state; 3977 sd->sd_start_discipline = NULL; 3978 3979 task_set(&sd->sd_meta_save_task, sr_meta_save_callback, sd); 3980 task_set(&sd->sd_hotspare_rebuild_task, sr_hotspare_rebuild_callback, 3981 sd); 3982 3983 switch (level) { 3984 case 0: 3985 sr_raid0_discipline_init(sd); 3986 break; 3987 case 1: 3988 sr_raid1_discipline_init(sd); 3989 break; 3990 case 5: 3991 sr_raid5_discipline_init(sd); 3992 break; 3993 case 6: 3994 sr_raid6_discipline_init(sd); 3995 break; 3996 #ifdef CRYPTO 3997 case 'C': 3998 sr_crypto_discipline_init(sd); 3999 break; 4000 #endif 4001 case 'c': 4002 sr_concat_discipline_init(sd); 4003 break; 4004 default: 4005 goto bad; 4006 } 4007 4008 rv = 0; 4009 bad: 4010 return (rv); 4011 } 4012 4013 int 4014 sr_raid_inquiry(struct sr_workunit *wu) 4015 { 4016 struct sr_discipline *sd = wu->swu_dis; 4017 struct scsi_xfer *xs = wu->swu_xs; 4018 struct scsi_inquiry *cdb = (struct scsi_inquiry *)xs->cmd; 4019 struct scsi_inquiry_data inq; 4020 4021 DNPRINTF(SR_D_DIS, "%s: sr_raid_inquiry\n", DEVNAME(sd->sd_sc)); 4022 4023 if (xs->cmdlen != sizeof(*cdb)) 4024 return (EINVAL); 4025 4026 if (ISSET(cdb->flags, SI_EVPD)) 4027 return (EOPNOTSUPP); 4028 4029 bzero(&inq, sizeof(inq)); 4030 inq.device = T_DIRECT; 4031 inq.dev_qual2 = 0; 4032 inq.version = 2; 4033 inq.response_format = 2; 4034 inq.additional_length = 32; 4035 inq.flags |= SID_CmdQue; 4036 strlcpy(inq.vendor, sd->sd_meta->ssdi.ssd_vendor, 4037 sizeof(inq.vendor)); 4038 strlcpy(inq.product, sd->sd_meta->ssdi.ssd_product, 4039 sizeof(inq.product)); 4040 strlcpy(inq.revision, sd->sd_meta->ssdi.ssd_revision, 4041 sizeof(inq.revision)); 4042 sr_copy_internal_data(xs, &inq, sizeof(inq)); 4043 4044 return (0); 4045 } 4046 4047 int 4048 sr_raid_read_cap(struct sr_workunit *wu) 4049 { 4050 struct sr_discipline *sd = wu->swu_dis; 4051 struct scsi_xfer *xs = wu->swu_xs; 4052 struct scsi_read_cap_data rcd; 4053 struct scsi_read_cap_data_16 rcd16; 4054 u_int64_t addr; 4055 int rv = 1; 4056 u_int32_t secsize; 4057 4058 DNPRINTF(SR_D_DIS, "%s: sr_raid_read_cap\n", DEVNAME(sd->sd_sc)); 4059 4060 secsize = sd->sd_meta->ssdi.ssd_secsize; 4061 4062 addr = ((sd->sd_meta->ssdi.ssd_size * DEV_BSIZE) / secsize) - 1; 4063 if (xs->cmd->opcode == READ_CAPACITY) { 4064 bzero(&rcd, sizeof(rcd)); 4065 if (addr > 0xffffffffllu) 4066 _lto4b(0xffffffff, rcd.addr); 4067 else 4068 _lto4b(addr, rcd.addr); 4069 _lto4b(secsize, rcd.length); 4070 sr_copy_internal_data(xs, &rcd, sizeof(rcd)); 4071 rv = 0; 4072 } else if (xs->cmd->opcode == READ_CAPACITY_16) { 4073 bzero(&rcd16, sizeof(rcd16)); 4074 _lto8b(addr, rcd16.addr); 4075 _lto4b(secsize, rcd16.length); 4076 sr_copy_internal_data(xs, &rcd16, sizeof(rcd16)); 4077 rv = 0; 4078 } 4079 4080 return (rv); 4081 } 4082 4083 int 4084 sr_raid_tur(struct sr_workunit *wu) 4085 { 4086 struct sr_discipline *sd = wu->swu_dis; 4087 4088 DNPRINTF(SR_D_DIS, "%s: sr_raid_tur\n", DEVNAME(sd->sd_sc)); 4089 4090 if (sd->sd_vol_status == BIOC_SVOFFLINE) { 4091 sd->sd_scsi_sense.error_code = SSD_ERRCODE_CURRENT; 4092 sd->sd_scsi_sense.flags = SKEY_NOT_READY; 4093 sd->sd_scsi_sense.add_sense_code = 0x04; 4094 sd->sd_scsi_sense.add_sense_code_qual = 0x11; 4095 sd->sd_scsi_sense.extra_len = 4; 4096 return (1); 4097 } else if (sd->sd_vol_status == BIOC_SVINVALID) { 4098 sd->sd_scsi_sense.error_code = SSD_ERRCODE_CURRENT; 4099 sd->sd_scsi_sense.flags = SKEY_HARDWARE_ERROR; 4100 sd->sd_scsi_sense.add_sense_code = 0x05; 4101 sd->sd_scsi_sense.add_sense_code_qual = 0x00; 4102 sd->sd_scsi_sense.extra_len = 4; 4103 return (1); 4104 } 4105 4106 return (0); 4107 } 4108 4109 int 4110 sr_raid_request_sense(struct sr_workunit *wu) 4111 { 4112 struct sr_discipline *sd = wu->swu_dis; 4113 struct scsi_xfer *xs = wu->swu_xs; 4114 4115 DNPRINTF(SR_D_DIS, "%s: sr_raid_request_sense\n", 4116 DEVNAME(sd->sd_sc)); 4117 4118 /* use latest sense data */ 4119 memcpy(&xs->sense, &sd->sd_scsi_sense, sizeof(xs->sense)); 4120 4121 /* clear sense data */ 4122 bzero(&sd->sd_scsi_sense, sizeof(sd->sd_scsi_sense)); 4123 4124 return (0); 4125 } 4126 4127 int 4128 sr_raid_start_stop(struct sr_workunit *wu) 4129 { 4130 struct scsi_xfer *xs = wu->swu_xs; 4131 struct scsi_start_stop *ss = (struct scsi_start_stop *)xs->cmd; 4132 4133 DNPRINTF(SR_D_DIS, "%s: sr_raid_start_stop\n", 4134 DEVNAME(wu->swu_dis->sd_sc)); 4135 4136 if (!ss) 4137 return (1); 4138 4139 /* 4140 * do nothing! 4141 * a softraid discipline should always reflect correct status 4142 */ 4143 return (0); 4144 } 4145 4146 int 4147 sr_raid_sync(struct sr_workunit *wu) 4148 { 4149 struct sr_discipline *sd = wu->swu_dis; 4150 int s, rv = 0, ios; 4151 4152 DNPRINTF(SR_D_DIS, "%s: sr_raid_sync\n", DEVNAME(sd->sd_sc)); 4153 4154 /* when doing a fake sync don't count the wu */ 4155 ios = (wu->swu_flags & SR_WUF_FAKE) ? 0 : 1; 4156 4157 s = splbio(); 4158 sd->sd_sync = 1; 4159 while (sd->sd_wu_pending > ios) { 4160 if (tsleep(sd, PRIBIO, "sr_sync", 15 * hz) == EWOULDBLOCK) { 4161 DNPRINTF(SR_D_DIS, "%s: sr_raid_sync timeout\n", 4162 DEVNAME(sd->sd_sc)); 4163 rv = 1; 4164 break; 4165 } 4166 } 4167 sd->sd_sync = 0; 4168 splx(s); 4169 4170 wakeup(&sd->sd_sync); 4171 4172 return (rv); 4173 } 4174 4175 void 4176 sr_raid_intr(struct buf *bp) 4177 { 4178 struct sr_ccb *ccb = (struct sr_ccb *)bp; 4179 struct sr_workunit *wu = ccb->ccb_wu; 4180 #ifdef SR_DEBUG 4181 struct sr_discipline *sd = wu->swu_dis; 4182 struct scsi_xfer *xs = wu->swu_xs; 4183 #endif 4184 int s; 4185 4186 DNPRINTF(SR_D_INTR, "%s: %s %s intr bp %p xs %p\n", 4187 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, sd->sd_name, bp, xs); 4188 4189 s = splbio(); 4190 sr_ccb_done(ccb); 4191 sr_wu_done(wu); 4192 splx(s); 4193 } 4194 4195 void 4196 sr_schedule_wu(struct sr_workunit *wu) 4197 { 4198 struct sr_discipline *sd = wu->swu_dis; 4199 struct sr_workunit *wup; 4200 int s; 4201 4202 DNPRINTF(SR_D_WU, "sr_schedule_wu: schedule wu %p state %i " 4203 "flags 0x%x\n", wu, wu->swu_state, wu->swu_flags); 4204 4205 KASSERT(wu->swu_io_count > 0); 4206 4207 s = splbio(); 4208 4209 /* Construct the work unit, do not schedule it. */ 4210 if (wu->swu_state == SR_WU_CONSTRUCT) 4211 goto queued; 4212 4213 /* Deferred work unit being reconstructed, do not start. */ 4214 if (wu->swu_state == SR_WU_REQUEUE) 4215 goto queued; 4216 4217 /* Current work unit failed, restart. */ 4218 if (wu->swu_state == SR_WU_RESTART) 4219 goto start; 4220 4221 if (wu->swu_state != SR_WU_INPROGRESS) 4222 panic("sr_schedule_wu: work unit not in progress (state %i)\n", 4223 wu->swu_state); 4224 4225 /* Walk queue backwards and fill in collider if we have one. */ 4226 TAILQ_FOREACH_REVERSE(wup, &sd->sd_wu_pendq, sr_wu_list, swu_link) { 4227 if (wu->swu_blk_end < wup->swu_blk_start || 4228 wup->swu_blk_end < wu->swu_blk_start) 4229 continue; 4230 4231 /* Defer work unit due to LBA collision. */ 4232 DNPRINTF(SR_D_WU, "sr_schedule_wu: deferring work unit %p\n", 4233 wu); 4234 wu->swu_state = SR_WU_DEFERRED; 4235 while (wup->swu_collider) 4236 wup = wup->swu_collider; 4237 wup->swu_collider = wu; 4238 TAILQ_INSERT_TAIL(&sd->sd_wu_defq, wu, swu_link); 4239 sd->sd_wu_collisions++; 4240 goto queued; 4241 } 4242 4243 start: 4244 sr_raid_startwu(wu); 4245 4246 queued: 4247 splx(s); 4248 } 4249 4250 void 4251 sr_raid_startwu(struct sr_workunit *wu) 4252 { 4253 struct sr_discipline *sd = wu->swu_dis; 4254 struct sr_ccb *ccb; 4255 4256 DNPRINTF(SR_D_WU, "sr_raid_startwu: start wu %p\n", wu); 4257 4258 splassert(IPL_BIO); 4259 4260 if (wu->swu_state == SR_WU_DEFERRED) { 4261 TAILQ_REMOVE(&sd->sd_wu_defq, wu, swu_link); 4262 wu->swu_state = SR_WU_INPROGRESS; 4263 } 4264 4265 if (wu->swu_state != SR_WU_RESTART) 4266 TAILQ_INSERT_TAIL(&sd->sd_wu_pendq, wu, swu_link); 4267 4268 /* Start all of the individual I/Os. */ 4269 if (wu->swu_cb_active == 1) 4270 panic("%s: sr_startwu_callback", DEVNAME(sd->sd_sc)); 4271 wu->swu_cb_active = 1; 4272 4273 TAILQ_FOREACH(ccb, &wu->swu_ccb, ccb_link) 4274 VOP_STRATEGY(&ccb->ccb_buf); 4275 4276 wu->swu_cb_active = 0; 4277 } 4278 4279 void 4280 sr_raid_recreate_wu(struct sr_workunit *wu) 4281 { 4282 struct sr_discipline *sd = wu->swu_dis; 4283 struct sr_workunit *wup = wu; 4284 4285 /* 4286 * Recreate a work unit by releasing the associated CCBs and reissuing 4287 * the SCSI I/O request. This process is then repeated for all of the 4288 * colliding work units. 4289 */ 4290 do { 4291 sr_wu_release_ccbs(wup); 4292 4293 wup->swu_state = SR_WU_REQUEUE; 4294 if (sd->sd_scsi_rw(wup)) 4295 panic("could not requeue I/O"); 4296 4297 wup = wup->swu_collider; 4298 } while (wup); 4299 } 4300 4301 int 4302 sr_alloc_resources(struct sr_discipline *sd) 4303 { 4304 if (sr_wu_alloc(sd, sizeof(struct sr_workunit))) { 4305 sr_error(sd->sd_sc, "unable to allocate work units"); 4306 return (ENOMEM); 4307 } 4308 if (sr_ccb_alloc(sd)) { 4309 sr_error(sd->sd_sc, "unable to allocate ccbs"); 4310 return (ENOMEM); 4311 } 4312 4313 return (0); 4314 } 4315 4316 void 4317 sr_free_resources(struct sr_discipline *sd) 4318 { 4319 sr_wu_free(sd); 4320 sr_ccb_free(sd); 4321 } 4322 4323 void 4324 sr_set_chunk_state(struct sr_discipline *sd, int c, int new_state) 4325 { 4326 int old_state, s; 4327 4328 DNPRINTF(SR_D_STATE, "%s: %s: %s: sr_set_chunk_state %d -> %d\n", 4329 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, 4330 sd->sd_vol.sv_chunks[c]->src_meta.scmi.scm_devname, c, new_state); 4331 4332 /* ok to go to splbio since this only happens in error path */ 4333 s = splbio(); 4334 old_state = sd->sd_vol.sv_chunks[c]->src_meta.scm_status; 4335 4336 /* multiple IOs to the same chunk that fail will come through here */ 4337 if (old_state == new_state) 4338 goto done; 4339 4340 switch (old_state) { 4341 case BIOC_SDONLINE: 4342 if (new_state == BIOC_SDOFFLINE) 4343 break; 4344 else 4345 goto die; 4346 break; 4347 4348 case BIOC_SDOFFLINE: 4349 goto die; 4350 4351 default: 4352 die: 4353 splx(s); /* XXX */ 4354 panic("%s: %s: %s: invalid chunk state transition " 4355 "%d -> %d", DEVNAME(sd->sd_sc), 4356 sd->sd_meta->ssd_devname, 4357 sd->sd_vol.sv_chunks[c]->src_meta.scmi.scm_devname, 4358 old_state, new_state); 4359 /* NOTREACHED */ 4360 } 4361 4362 sd->sd_vol.sv_chunks[c]->src_meta.scm_status = new_state; 4363 sd->sd_set_vol_state(sd); 4364 4365 sd->sd_must_flush = 1; 4366 task_add(systq, &sd->sd_meta_save_task); 4367 done: 4368 splx(s); 4369 } 4370 4371 void 4372 sr_set_vol_state(struct sr_discipline *sd) 4373 { 4374 int states[SR_MAX_STATES]; 4375 int new_state, i, nd; 4376 int old_state = sd->sd_vol_status; 4377 u_int32_t s; 4378 4379 DNPRINTF(SR_D_STATE, "%s: %s: sr_set_vol_state\n", 4380 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname); 4381 4382 nd = sd->sd_meta->ssdi.ssd_chunk_no; 4383 4384 for (i = 0; i < SR_MAX_STATES; i++) 4385 states[i] = 0; 4386 4387 for (i = 0; i < nd; i++) { 4388 s = sd->sd_vol.sv_chunks[i]->src_meta.scm_status; 4389 if (s >= SR_MAX_STATES) 4390 panic("%s: %s: %s: invalid chunk state", 4391 DEVNAME(sd->sd_sc), 4392 sd->sd_meta->ssd_devname, 4393 sd->sd_vol.sv_chunks[i]->src_meta.scmi.scm_devname); 4394 states[s]++; 4395 } 4396 4397 if (states[BIOC_SDONLINE] == nd) 4398 new_state = BIOC_SVONLINE; 4399 else 4400 new_state = BIOC_SVOFFLINE; 4401 4402 DNPRINTF(SR_D_STATE, "%s: %s: sr_set_vol_state %d -> %d\n", 4403 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, 4404 old_state, new_state); 4405 4406 switch (old_state) { 4407 case BIOC_SVONLINE: 4408 if (new_state == BIOC_SVOFFLINE || new_state == BIOC_SVONLINE) 4409 break; 4410 else 4411 goto die; 4412 break; 4413 4414 case BIOC_SVOFFLINE: 4415 /* XXX this might be a little too much */ 4416 goto die; 4417 4418 default: 4419 die: 4420 panic("%s: %s: invalid volume state transition " 4421 "%d -> %d", DEVNAME(sd->sd_sc), 4422 sd->sd_meta->ssd_devname, 4423 old_state, new_state); 4424 /* NOTREACHED */ 4425 } 4426 4427 sd->sd_vol_status = new_state; 4428 } 4429 4430 void * 4431 sr_block_get(struct sr_discipline *sd, long length) 4432 { 4433 return dma_alloc(length, PR_NOWAIT | PR_ZERO); 4434 } 4435 4436 void 4437 sr_block_put(struct sr_discipline *sd, void *ptr, int length) 4438 { 4439 dma_free(ptr, length); 4440 } 4441 4442 void 4443 sr_checksum_print(u_int8_t *md5) 4444 { 4445 int i; 4446 4447 for (i = 0; i < MD5_DIGEST_LENGTH; i++) 4448 printf("%02x", md5[i]); 4449 } 4450 4451 void 4452 sr_checksum(struct sr_softc *sc, void *src, void *md5, u_int32_t len) 4453 { 4454 MD5_CTX ctx; 4455 4456 DNPRINTF(SR_D_MISC, "%s: sr_checksum(%p %p %d)\n", DEVNAME(sc), src, 4457 md5, len); 4458 4459 MD5Init(&ctx); 4460 MD5Update(&ctx, src, len); 4461 MD5Final(md5, &ctx); 4462 } 4463 4464 void 4465 sr_uuid_generate(struct sr_uuid *uuid) 4466 { 4467 arc4random_buf(uuid->sui_id, sizeof(uuid->sui_id)); 4468 /* UUID version 4: random */ 4469 uuid->sui_id[6] &= 0x0f; 4470 uuid->sui_id[6] |= 0x40; 4471 /* RFC4122 variant */ 4472 uuid->sui_id[8] &= 0x3f; 4473 uuid->sui_id[8] |= 0x80; 4474 } 4475 4476 char * 4477 sr_uuid_format(struct sr_uuid *uuid) 4478 { 4479 char *uuidstr; 4480 4481 uuidstr = malloc(37, M_DEVBUF, M_WAITOK); 4482 4483 snprintf(uuidstr, 37, 4484 "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-" 4485 "%02x%02x%02x%02x%02x%02x", 4486 uuid->sui_id[0], uuid->sui_id[1], 4487 uuid->sui_id[2], uuid->sui_id[3], 4488 uuid->sui_id[4], uuid->sui_id[5], 4489 uuid->sui_id[6], uuid->sui_id[7], 4490 uuid->sui_id[8], uuid->sui_id[9], 4491 uuid->sui_id[10], uuid->sui_id[11], 4492 uuid->sui_id[12], uuid->sui_id[13], 4493 uuid->sui_id[14], uuid->sui_id[15]); 4494 4495 return uuidstr; 4496 } 4497 4498 void 4499 sr_uuid_print(struct sr_uuid *uuid, int cr) 4500 { 4501 char *uuidstr; 4502 4503 uuidstr = sr_uuid_format(uuid); 4504 printf("%s%s", uuidstr, (cr ? "\n" : "")); 4505 free(uuidstr, M_DEVBUF, 37); 4506 } 4507 4508 int 4509 sr_already_assembled(struct sr_discipline *sd) 4510 { 4511 struct sr_softc *sc = sd->sd_sc; 4512 struct sr_discipline *sdtmp; 4513 4514 TAILQ_FOREACH(sdtmp, &sc->sc_dis_list, sd_link) { 4515 if (!bcmp(&sd->sd_meta->ssdi.ssd_uuid, 4516 &sdtmp->sd_meta->ssdi.ssd_uuid, 4517 sizeof(sd->sd_meta->ssdi.ssd_uuid))) 4518 return (1); 4519 } 4520 4521 return (0); 4522 } 4523 4524 int32_t 4525 sr_validate_stripsize(u_int32_t b) 4526 { 4527 int s = 0; 4528 4529 if (b % DEV_BSIZE) 4530 return (-1); 4531 4532 while ((b & 1) == 0) { 4533 b >>= 1; 4534 s++; 4535 } 4536 4537 /* only multiple of twos */ 4538 b >>= 1; 4539 if (b) 4540 return(-1); 4541 4542 return (s); 4543 } 4544 4545 void 4546 sr_shutdown(int dying) 4547 { 4548 struct sr_softc *sc = softraid0; 4549 struct sr_discipline *sd; 4550 4551 DNPRINTF(SR_D_MISC, "%s: sr_shutdown\n", DEVNAME(sc)); 4552 4553 /* 4554 * Since softraid is not under mainbus, we have to explicitly 4555 * notify its children that the power is going down, so they 4556 * can execute their shutdown hooks. 4557 */ 4558 config_suspend((struct device *)sc, DVACT_POWERDOWN); 4559 4560 /* Shutdown disciplines in reverse attach order. */ 4561 while ((sd = TAILQ_LAST(&sc->sc_dis_list, sr_discipline_list)) != NULL) 4562 sr_discipline_shutdown(sd, 1, dying); 4563 } 4564 4565 int 4566 sr_validate_io(struct sr_workunit *wu, daddr_t *blkno, char *func) 4567 { 4568 struct sr_discipline *sd = wu->swu_dis; 4569 struct scsi_xfer *xs = wu->swu_xs; 4570 int rv = 1; 4571 4572 DNPRINTF(SR_D_DIS, "%s: %s 0x%02x\n", DEVNAME(sd->sd_sc), func, 4573 xs->cmd->opcode); 4574 4575 if (sd->sd_meta->ssd_data_blkno == 0) 4576 panic("invalid data blkno"); 4577 4578 if (sd->sd_vol_status == BIOC_SVOFFLINE) { 4579 DNPRINTF(SR_D_DIS, "%s: %s device offline\n", 4580 DEVNAME(sd->sd_sc), func); 4581 goto bad; 4582 } 4583 4584 if (xs->datalen == 0) { 4585 printf("%s: %s: illegal block count for %s\n", 4586 DEVNAME(sd->sd_sc), func, sd->sd_meta->ssd_devname); 4587 goto bad; 4588 } 4589 4590 if (xs->cmdlen == 10) 4591 *blkno = _4btol(((struct scsi_rw_big *)xs->cmd)->addr); 4592 else if (xs->cmdlen == 16) 4593 *blkno = _8btol(((struct scsi_rw_16 *)xs->cmd)->addr); 4594 else if (xs->cmdlen == 6) 4595 *blkno = _3btol(((struct scsi_rw *)xs->cmd)->addr); 4596 else { 4597 printf("%s: %s: illegal cmdlen for %s\n", 4598 DEVNAME(sd->sd_sc), func, sd->sd_meta->ssd_devname); 4599 goto bad; 4600 } 4601 4602 *blkno *= (sd->sd_meta->ssdi.ssd_secsize / DEV_BSIZE); 4603 4604 wu->swu_blk_start = *blkno; 4605 wu->swu_blk_end = *blkno + (xs->datalen >> DEV_BSHIFT) - 1; 4606 4607 if (wu->swu_blk_end > sd->sd_meta->ssdi.ssd_size) { 4608 DNPRINTF(SR_D_DIS, "%s: %s out of bounds start: %lld " 4609 "end: %lld length: %d\n", 4610 DEVNAME(sd->sd_sc), func, (long long)wu->swu_blk_start, 4611 (long long)wu->swu_blk_end, xs->datalen); 4612 4613 sd->sd_scsi_sense.error_code = SSD_ERRCODE_CURRENT | 4614 SSD_ERRCODE_VALID; 4615 sd->sd_scsi_sense.flags = SKEY_ILLEGAL_REQUEST; 4616 sd->sd_scsi_sense.add_sense_code = 0x21; 4617 sd->sd_scsi_sense.add_sense_code_qual = 0x00; 4618 sd->sd_scsi_sense.extra_len = 4; 4619 goto bad; 4620 } 4621 4622 rv = 0; 4623 bad: 4624 return (rv); 4625 } 4626 4627 void 4628 sr_rebuild_start(void *arg) 4629 { 4630 struct sr_discipline *sd = arg; 4631 struct sr_softc *sc = sd->sd_sc; 4632 4633 DNPRINTF(SR_D_REBUILD, "%s: %s starting rebuild thread\n", 4634 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname); 4635 4636 if (kthread_create(sr_rebuild_thread, sd, &sd->sd_background_proc, 4637 DEVNAME(sc)) != 0) 4638 printf("%s: unable to start background operation\n", 4639 DEVNAME(sc)); 4640 } 4641 4642 void 4643 sr_rebuild_thread(void *arg) 4644 { 4645 struct sr_discipline *sd = arg; 4646 4647 DNPRINTF(SR_D_REBUILD, "%s: %s rebuild thread started\n", 4648 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname); 4649 4650 sd->sd_reb_active = 1; 4651 sd->sd_rebuild(sd); 4652 sd->sd_reb_active = 0; 4653 4654 kthread_exit(0); 4655 } 4656 4657 void 4658 sr_rebuild(struct sr_discipline *sd) 4659 { 4660 struct sr_softc *sc = sd->sd_sc; 4661 u_int64_t sz, whole_blk, partial_blk, blk, restart; 4662 daddr_t lba; 4663 struct sr_workunit *wu_r, *wu_w; 4664 struct scsi_xfer xs_r, xs_w; 4665 struct scsi_rw_16 *cr, *cw; 4666 int c, s, slept, percent = 0, old_percent = -1; 4667 u_int8_t *buf; 4668 4669 whole_blk = sd->sd_meta->ssdi.ssd_size / SR_REBUILD_IO_SIZE; 4670 partial_blk = sd->sd_meta->ssdi.ssd_size % SR_REBUILD_IO_SIZE; 4671 4672 restart = sd->sd_meta->ssd_rebuild / SR_REBUILD_IO_SIZE; 4673 if (restart > whole_blk) { 4674 printf("%s: bogus rebuild restart offset, starting from 0\n", 4675 DEVNAME(sc)); 4676 restart = 0; 4677 } 4678 if (restart) { 4679 /* 4680 * XXX there is a hole here; there is a posibility that we 4681 * had a restart however the chunk that was supposed to 4682 * be rebuilt is no longer valid; we can reach this situation 4683 * when a rebuild is in progress and the box crashes and 4684 * on reboot the rebuild chunk is different (like zero'd or 4685 * replaced). We need to check the uuid of the chunk that is 4686 * being rebuilt to assert this. 4687 */ 4688 percent = sr_rebuild_percent(sd); 4689 printf("%s: resuming rebuild on %s at %d%%\n", 4690 DEVNAME(sc), sd->sd_meta->ssd_devname, percent); 4691 } 4692 4693 /* currently this is 64k therefore we can use dma_alloc */ 4694 buf = dma_alloc(SR_REBUILD_IO_SIZE << DEV_BSHIFT, PR_WAITOK); 4695 for (blk = restart; blk <= whole_blk; blk++) { 4696 lba = blk * SR_REBUILD_IO_SIZE; 4697 sz = SR_REBUILD_IO_SIZE; 4698 if (blk == whole_blk) { 4699 if (partial_blk == 0) 4700 break; 4701 sz = partial_blk; 4702 } 4703 4704 /* get some wu */ 4705 wu_r = sr_scsi_wu_get(sd, 0); 4706 wu_w = sr_scsi_wu_get(sd, 0); 4707 4708 DNPRINTF(SR_D_REBUILD, "%s: %s rebuild wu_r %p, wu_w %p\n", 4709 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, wu_r, wu_w); 4710 4711 /* setup read io */ 4712 bzero(&xs_r, sizeof xs_r); 4713 xs_r.error = XS_NOERROR; 4714 xs_r.flags = SCSI_DATA_IN; 4715 xs_r.datalen = sz << DEV_BSHIFT; 4716 xs_r.data = buf; 4717 xs_r.cmdlen = sizeof(*cr); 4718 xs_r.cmd = &xs_r.cmdstore; 4719 cr = (struct scsi_rw_16 *)xs_r.cmd; 4720 cr->opcode = READ_16; 4721 _lto4b(sz, cr->length); 4722 _lto8b(lba, cr->addr); 4723 wu_r->swu_state = SR_WU_CONSTRUCT; 4724 wu_r->swu_flags |= SR_WUF_REBUILD; 4725 wu_r->swu_xs = &xs_r; 4726 if (sd->sd_scsi_rw(wu_r)) { 4727 printf("%s: could not create read io\n", 4728 DEVNAME(sc)); 4729 goto fail; 4730 } 4731 4732 /* setup write io */ 4733 bzero(&xs_w, sizeof xs_w); 4734 xs_w.error = XS_NOERROR; 4735 xs_w.flags = SCSI_DATA_OUT; 4736 xs_w.datalen = sz << DEV_BSHIFT; 4737 xs_w.data = buf; 4738 xs_w.cmdlen = sizeof(*cw); 4739 xs_w.cmd = &xs_w.cmdstore; 4740 cw = (struct scsi_rw_16 *)xs_w.cmd; 4741 cw->opcode = WRITE_16; 4742 _lto4b(sz, cw->length); 4743 _lto8b(lba, cw->addr); 4744 wu_w->swu_state = SR_WU_CONSTRUCT; 4745 wu_w->swu_flags |= SR_WUF_REBUILD | SR_WUF_WAKEUP; 4746 wu_w->swu_xs = &xs_w; 4747 if (sd->sd_scsi_rw(wu_w)) { 4748 printf("%s: could not create write io\n", 4749 DEVNAME(sc)); 4750 goto fail; 4751 } 4752 4753 /* 4754 * collide with the read io so that we get automatically 4755 * started when the read is done 4756 */ 4757 wu_w->swu_state = SR_WU_DEFERRED; 4758 wu_r->swu_collider = wu_w; 4759 s = splbio(); 4760 TAILQ_INSERT_TAIL(&sd->sd_wu_defq, wu_w, swu_link); 4761 splx(s); 4762 4763 DNPRINTF(SR_D_REBUILD, "%s: %s rebuild scheduling wu_r %p\n", 4764 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, wu_r); 4765 4766 wu_r->swu_state = SR_WU_INPROGRESS; 4767 sr_schedule_wu(wu_r); 4768 4769 /* wait for write completion */ 4770 slept = 0; 4771 while ((wu_w->swu_flags & SR_WUF_REBUILDIOCOMP) == 0) { 4772 tsleep(wu_w, PRIBIO, "sr_rebuild", 0); 4773 slept = 1; 4774 } 4775 /* yield if we didn't sleep */ 4776 if (slept == 0) 4777 tsleep(sc, PWAIT, "sr_yield", 1); 4778 4779 sr_scsi_wu_put(sd, wu_r); 4780 sr_scsi_wu_put(sd, wu_w); 4781 4782 sd->sd_meta->ssd_rebuild = lba; 4783 4784 /* XXX - this should be based on size, not percentage. */ 4785 /* save metadata every percent */ 4786 percent = sr_rebuild_percent(sd); 4787 if (percent != old_percent && blk != whole_blk) { 4788 if (sr_meta_save(sd, SR_META_DIRTY)) 4789 printf("%s: could not save metadata to %s\n", 4790 DEVNAME(sc), sd->sd_meta->ssd_devname); 4791 old_percent = percent; 4792 } 4793 4794 if (sd->sd_reb_abort) 4795 goto abort; 4796 } 4797 4798 /* all done */ 4799 sd->sd_meta->ssd_rebuild = 0; 4800 for (c = 0; c < sd->sd_meta->ssdi.ssd_chunk_no; c++) { 4801 if (sd->sd_vol.sv_chunks[c]->src_meta.scm_status == 4802 BIOC_SDREBUILD) { 4803 sd->sd_set_chunk_state(sd, c, BIOC_SDONLINE); 4804 break; 4805 } 4806 } 4807 4808 abort: 4809 if (sr_meta_save(sd, SR_META_DIRTY)) 4810 printf("%s: could not save metadata to %s\n", 4811 DEVNAME(sc), sd->sd_meta->ssd_devname); 4812 fail: 4813 dma_free(buf, SR_REBUILD_IO_SIZE << DEV_BSHIFT); 4814 } 4815 4816 #ifndef SMALL_KERNEL 4817 int 4818 sr_sensors_create(struct sr_discipline *sd) 4819 { 4820 struct sr_softc *sc = sd->sd_sc; 4821 int rv = 1; 4822 4823 DNPRINTF(SR_D_STATE, "%s: %s: sr_sensors_create\n", 4824 DEVNAME(sc), sd->sd_meta->ssd_devname); 4825 4826 sd->sd_vol.sv_sensor.type = SENSOR_DRIVE; 4827 sd->sd_vol.sv_sensor.status = SENSOR_S_UNKNOWN; 4828 strlcpy(sd->sd_vol.sv_sensor.desc, sd->sd_meta->ssd_devname, 4829 sizeof(sd->sd_vol.sv_sensor.desc)); 4830 4831 sensor_attach(&sc->sc_sensordev, &sd->sd_vol.sv_sensor); 4832 sd->sd_vol.sv_sensor_attached = 1; 4833 4834 if (sc->sc_sensor_task == NULL) { 4835 sc->sc_sensor_task = sensor_task_register(sc, 4836 sr_sensors_refresh, 10); 4837 if (sc->sc_sensor_task == NULL) 4838 goto bad; 4839 } 4840 4841 rv = 0; 4842 bad: 4843 return (rv); 4844 } 4845 4846 void 4847 sr_sensors_delete(struct sr_discipline *sd) 4848 { 4849 DNPRINTF(SR_D_STATE, "%s: sr_sensors_delete\n", DEVNAME(sd->sd_sc)); 4850 4851 if (sd->sd_vol.sv_sensor_attached) 4852 sensor_detach(&sd->sd_sc->sc_sensordev, &sd->sd_vol.sv_sensor); 4853 } 4854 4855 void 4856 sr_sensors_refresh(void *arg) 4857 { 4858 struct sr_softc *sc = arg; 4859 struct sr_volume *sv; 4860 struct sr_discipline *sd; 4861 4862 DNPRINTF(SR_D_STATE, "%s: sr_sensors_refresh\n", DEVNAME(sc)); 4863 4864 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 4865 sv = &sd->sd_vol; 4866 4867 switch(sd->sd_vol_status) { 4868 case BIOC_SVOFFLINE: 4869 sv->sv_sensor.value = SENSOR_DRIVE_FAIL; 4870 sv->sv_sensor.status = SENSOR_S_CRIT; 4871 break; 4872 4873 case BIOC_SVDEGRADED: 4874 sv->sv_sensor.value = SENSOR_DRIVE_PFAIL; 4875 sv->sv_sensor.status = SENSOR_S_WARN; 4876 break; 4877 4878 case BIOC_SVREBUILD: 4879 sv->sv_sensor.value = SENSOR_DRIVE_REBUILD; 4880 sv->sv_sensor.status = SENSOR_S_WARN; 4881 break; 4882 4883 case BIOC_SVSCRUB: 4884 case BIOC_SVONLINE: 4885 sv->sv_sensor.value = SENSOR_DRIVE_ONLINE; 4886 sv->sv_sensor.status = SENSOR_S_OK; 4887 break; 4888 4889 default: 4890 sv->sv_sensor.value = 0; /* unknown */ 4891 sv->sv_sensor.status = SENSOR_S_UNKNOWN; 4892 } 4893 } 4894 } 4895 #endif /* SMALL_KERNEL */ 4896 4897 #ifdef SR_FANCY_STATS 4898 void sr_print_stats(void); 4899 4900 void 4901 sr_print_stats(void) 4902 { 4903 struct sr_softc *sc = softraid0; 4904 struct sr_discipline *sd; 4905 4906 if (sc == NULL) { 4907 printf("no softraid softc found\n"); 4908 return; 4909 } 4910 4911 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 4912 printf("%s: ios pending %d, collisions %llu\n", 4913 sd->sd_meta->ssd_devname, 4914 sd->sd_wu_pending, 4915 sd->sd_wu_collisions); 4916 } 4917 } 4918 #endif /* SR_FANCY_STATS */ 4919 4920 #ifdef SR_DEBUG 4921 void 4922 sr_meta_print(struct sr_metadata *m) 4923 { 4924 int i; 4925 struct sr_meta_chunk *mc; 4926 struct sr_meta_opt_hdr *omh; 4927 4928 if (!(sr_debug & SR_D_META)) 4929 return; 4930 4931 printf("\tssd_magic 0x%llx\n", m->ssdi.ssd_magic); 4932 printf("\tssd_version %d\n", m->ssdi.ssd_version); 4933 printf("\tssd_vol_flags 0x%x\n", m->ssdi.ssd_vol_flags); 4934 printf("\tssd_uuid "); 4935 sr_uuid_print(&m->ssdi.ssd_uuid, 1); 4936 printf("\tssd_chunk_no %d\n", m->ssdi.ssd_chunk_no); 4937 printf("\tssd_chunk_id %d\n", m->ssdi.ssd_chunk_id); 4938 printf("\tssd_opt_no %d\n", m->ssdi.ssd_opt_no); 4939 printf("\tssd_volid %d\n", m->ssdi.ssd_volid); 4940 printf("\tssd_level %d\n", m->ssdi.ssd_level); 4941 printf("\tssd_size %lld\n", m->ssdi.ssd_size); 4942 printf("\tssd_devname %s\n", m->ssd_devname); 4943 printf("\tssd_vendor %s\n", m->ssdi.ssd_vendor); 4944 printf("\tssd_product %s\n", m->ssdi.ssd_product); 4945 printf("\tssd_revision %s\n", m->ssdi.ssd_revision); 4946 printf("\tssd_strip_size %d\n", m->ssdi.ssd_strip_size); 4947 printf("\tssd_checksum "); 4948 sr_checksum_print(m->ssd_checksum); 4949 printf("\n"); 4950 printf("\tssd_meta_flags 0x%x\n", m->ssd_meta_flags); 4951 printf("\tssd_ondisk %llu\n", m->ssd_ondisk); 4952 4953 mc = (struct sr_meta_chunk *)(m + 1); 4954 for (i = 0; i < m->ssdi.ssd_chunk_no; i++, mc++) { 4955 printf("\t\tscm_volid %d\n", mc->scmi.scm_volid); 4956 printf("\t\tscm_chunk_id %d\n", mc->scmi.scm_chunk_id); 4957 printf("\t\tscm_devname %s\n", mc->scmi.scm_devname); 4958 printf("\t\tscm_size %lld\n", mc->scmi.scm_size); 4959 printf("\t\tscm_coerced_size %lld\n",mc->scmi.scm_coerced_size); 4960 printf("\t\tscm_uuid "); 4961 sr_uuid_print(&mc->scmi.scm_uuid, 1); 4962 printf("\t\tscm_checksum "); 4963 sr_checksum_print(mc->scm_checksum); 4964 printf("\n"); 4965 printf("\t\tscm_status %d\n", mc->scm_status); 4966 } 4967 4968 omh = (struct sr_meta_opt_hdr *)((u_int8_t *)(m + 1) + 4969 sizeof(struct sr_meta_chunk) * m->ssdi.ssd_chunk_no); 4970 for (i = 0; i < m->ssdi.ssd_opt_no; i++) { 4971 printf("\t\t\tsom_type %d\n", omh->som_type); 4972 printf("\t\t\tsom_checksum "); 4973 sr_checksum_print(omh->som_checksum); 4974 printf("\n"); 4975 omh = (struct sr_meta_opt_hdr *)((void *)omh + 4976 omh->som_length); 4977 } 4978 } 4979 4980 void 4981 sr_dump_block(void *blk, int len) 4982 { 4983 uint8_t *b = blk; 4984 int i, j, c; 4985 4986 for (i = 0; i < len; i += 16) { 4987 for (j = 0; j < 16; j++) 4988 printf("%.2x ", b[i + j]); 4989 printf(" "); 4990 for (j = 0; j < 16; j++) { 4991 c = b[i + j]; 4992 if (c < ' ' || c > 'z' || i + j > len) 4993 c = '.'; 4994 printf("%c", c); 4995 } 4996 printf("\n"); 4997 } 4998 } 4999 5000 void 5001 sr_dump_mem(u_int8_t *p, int len) 5002 { 5003 int i; 5004 5005 for (i = 0; i < len; i++) 5006 printf("%02x ", *p++); 5007 printf("\n"); 5008 } 5009 5010 #endif /* SR_DEBUG */ 5011 5012 #ifdef HIBERNATE 5013 /* 5014 * Side-effect free (no malloc, printf, pool, splx) softraid crypto writer. 5015 * 5016 * This function must perform the following: 5017 * 1. Determine the underlying device's own side-effect free I/O function 5018 * (eg, ahci_hibernate_io, wd_hibernate_io, etc). 5019 * 2. Store enough information in the provided page argument for subsequent 5020 * I/O calls (such as the crypto discipline structure for the keys, the 5021 * offset of the softraid partition on the underlying disk, as well as 5022 * the offset of the swap partition within the crypto volume. 5023 * 3. Encrypt the incoming data using the sr_discipline keys, then pass 5024 * the request to the underlying device's own I/O function. 5025 */ 5026 int 5027 sr_hibernate_io(dev_t dev, daddr_t blkno, vaddr_t addr, size_t size, int op, void *page) 5028 { 5029 /* Struct for stashing data obtained on HIB_INIT. 5030 * XXX 5031 * We share the page with the underlying device's own 5032 * side-effect free I/O function, so we pad our data to 5033 * the end of the page. Presently this does not overlap 5034 * with either of the two other side-effect free i/o 5035 * functions (ahci/wd). 5036 */ 5037 struct { 5038 char pad[3072]; 5039 struct sr_discipline *srd; 5040 hibio_fn subfn; /* underlying device i/o fn */ 5041 dev_t subdev; /* underlying device dev_t */ 5042 daddr_t sr_swapoff; /* ofs of swap part in sr volume */ 5043 char buf[DEV_BSIZE]; /* encryption performed into this buf */ 5044 } *my = page; 5045 extern struct cfdriver sd_cd; 5046 char errstr[128], *dl_ret; 5047 struct sr_chunk *schunk; 5048 struct sd_softc *sd; 5049 struct aes_xts_ctx ctx; 5050 struct sr_softc *sc; 5051 struct device *dv; 5052 daddr_t key_blkno; 5053 uint32_t sub_raidoff; /* ofs of sr part in underlying dev */ 5054 struct disklabel dl; 5055 struct partition *pp; 5056 size_t i, j; 5057 u_char iv[8]; 5058 5059 /* 5060 * In HIB_INIT, we are passed the swap partition size and offset 5061 * in 'size' and 'blkno' respectively. These are relative to the 5062 * start of the softraid partition, and we need to save these 5063 * for later translation to the underlying device's layout. 5064 */ 5065 if (op == HIB_INIT) { 5066 dv = disk_lookup(&sd_cd, DISKUNIT(dev)); 5067 sd = (struct sd_softc *)dv; 5068 sc = (struct sr_softc *)dv->dv_parent->dv_parent; 5069 5070 /* 5071 * Look up the sr discipline. This is used to determine 5072 * if we are SR crypto and what the underlying device is. 5073 */ 5074 my->srd = sc->sc_targets[sd->sc_link->target]; 5075 DNPRINTF(SR_D_MISC, "sr_hibernate_io: discipline is %s\n", 5076 my->srd->sd_name); 5077 if (strncmp(my->srd->sd_name, "CRYPTO", 5078 sizeof(my->srd->sd_name))) 5079 return (ENOTSUP); 5080 5081 /* Find the underlying device */ 5082 schunk = my->srd->sd_vol.sv_chunks[0]; 5083 my->subdev = schunk->src_dev_mm; 5084 5085 /* 5086 * Find the appropriate underlying device side effect free 5087 * I/O function, based on the type of device it is. 5088 */ 5089 my->subfn = get_hibernate_io_function(my->subdev); 5090 if (!my->subfn) 5091 return (ENODEV); 5092 5093 /* 5094 * Find blkno where this raid partition starts on 5095 * the underlying disk. 5096 */ 5097 dl_ret = disk_readlabel(&dl, my->subdev, errstr, 5098 sizeof(errstr)); 5099 if (dl_ret) { 5100 printf("Hibernate error reading disklabel: %s\n", dl_ret); 5101 return (ENOTSUP); 5102 } 5103 5104 pp = &dl.d_partitions[DISKPART(my->subdev)]; 5105 if (pp->p_fstype != FS_RAID || DL_GETPSIZE(pp) == 0) 5106 return (ENOTSUP); 5107 5108 /* Find the blkno of the SR part in the underlying device */ 5109 sub_raidoff = my->srd->sd_meta->ssd_data_blkno + 5110 DL_SECTOBLK(&dl, DL_GETPOFFSET(pp)); 5111 DNPRINTF(SR_D_MISC,"sr_hibernate_io: blk trans ofs: %d blks\n", 5112 sub_raidoff); 5113 5114 /* Save the blkno of the swap partition in the SR disk */ 5115 my->sr_swapoff = blkno; 5116 5117 /* Initialize the sub-device */ 5118 return my->subfn(my->subdev, sub_raidoff + blkno, 5119 addr, size, op, page); 5120 } 5121 5122 /* Hibernate only uses (and we only support) writes */ 5123 if (op != HIB_W) 5124 return (ENOTSUP); 5125 5126 /* 5127 * Blocks act as the IV for the encryption. These block numbers 5128 * are relative to the start of the sr partition, but the 'blkno' 5129 * passed above is relative to the start of the swap partition 5130 * inside the sr partition, so bias appropriately. 5131 */ 5132 key_blkno = my->sr_swapoff + blkno; 5133 5134 /* Process each disk block one at a time. */ 5135 for (i = 0; i < size; i += DEV_BSIZE) { 5136 int res; 5137 5138 bzero(&ctx, sizeof(ctx)); 5139 5140 /* 5141 * Set encryption key (from the sr discipline stashed 5142 * during HIB_INIT. This code is based on the softraid 5143 * bootblock code. 5144 */ 5145 aes_xts_setkey(&ctx, my->srd->mds.mdd_crypto.scr_key[0], 64); 5146 /* We encrypt DEV_BSIZE bytes at a time in my->buf */ 5147 memcpy(my->buf, ((char *)addr) + i, DEV_BSIZE); 5148 5149 /* Block number is the IV */ 5150 memcpy(&iv, &key_blkno, sizeof(key_blkno)); 5151 aes_xts_reinit(&ctx, iv); 5152 5153 /* Encrypt DEV_BSIZE bytes, AES_XTS_BLOCKSIZE bytes at a time */ 5154 for (j = 0; j < DEV_BSIZE; j += AES_XTS_BLOCKSIZE) 5155 aes_xts_encrypt(&ctx, my->buf + j); 5156 5157 /* 5158 * Write one block out from my->buf to the underlying device 5159 * using its own side-effect free I/O function. 5160 */ 5161 res = my->subfn(my->subdev, blkno + (i / DEV_BSIZE), 5162 (vaddr_t)(my->buf), DEV_BSIZE, op, page); 5163 if (res != 0) 5164 return (res); 5165 key_blkno++; 5166 } 5167 return (0); 5168 } 5169 #endif /* HIBERNATE */ 5170