1 /* $OpenBSD: softraid.c,v 1.391 2018/02/08 06:02:09 deraadt 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 %s\n", DEVNAME(sc), 1536 duid_format(rootduid)); 1537 return; 1538 } 1539 } 1540 } 1541 } 1542 } 1543 1544 int 1545 sr_meta_native_probe(struct sr_softc *sc, struct sr_chunk *ch_entry) 1546 { 1547 struct disklabel label; 1548 char *devname; 1549 int error, part; 1550 u_int64_t size; 1551 1552 DNPRINTF(SR_D_META, "%s: sr_meta_native_probe(%s)\n", 1553 DEVNAME(sc), ch_entry->src_devname); 1554 1555 devname = ch_entry->src_devname; 1556 part = DISKPART(ch_entry->src_dev_mm); 1557 1558 /* get disklabel */ 1559 error = VOP_IOCTL(ch_entry->src_vn, DIOCGDINFO, (caddr_t)&label, FREAD, 1560 NOCRED, curproc); 1561 if (error) { 1562 DNPRINTF(SR_D_META, "%s: %s can't obtain disklabel\n", 1563 DEVNAME(sc), devname); 1564 goto unwind; 1565 } 1566 memcpy(ch_entry->src_duid, label.d_uid, sizeof(ch_entry->src_duid)); 1567 1568 /* make sure the partition is of the right type */ 1569 if (label.d_partitions[part].p_fstype != FS_RAID) { 1570 DNPRINTF(SR_D_META, 1571 "%s: %s partition not of type RAID (%d)\n", DEVNAME(sc), 1572 devname, 1573 label.d_partitions[part].p_fstype); 1574 goto unwind; 1575 } 1576 1577 size = DL_SECTOBLK(&label, DL_GETPSIZE(&label.d_partitions[part])); 1578 if (size <= SR_DATA_OFFSET) { 1579 DNPRINTF(SR_D_META, "%s: %s partition too small\n", DEVNAME(sc), 1580 devname); 1581 goto unwind; 1582 } 1583 size -= SR_DATA_OFFSET; 1584 if (size > INT64_MAX) { 1585 DNPRINTF(SR_D_META, "%s: %s partition too large\n", DEVNAME(sc), 1586 devname); 1587 goto unwind; 1588 } 1589 ch_entry->src_size = size; 1590 ch_entry->src_secsize = label.d_secsize; 1591 1592 DNPRINTF(SR_D_META, "%s: probe found %s size %lld\n", DEVNAME(sc), 1593 devname, (long long)size); 1594 1595 return (SR_META_F_NATIVE); 1596 unwind: 1597 DNPRINTF(SR_D_META, "%s: invalid device: %s\n", DEVNAME(sc), 1598 devname ? devname : "nodev"); 1599 return (SR_META_F_INVALID); 1600 } 1601 1602 int 1603 sr_meta_native_attach(struct sr_discipline *sd, int force) 1604 { 1605 struct sr_softc *sc = sd->sd_sc; 1606 struct sr_chunk_head *cl = &sd->sd_vol.sv_chunk_list; 1607 struct sr_metadata *md = NULL; 1608 struct sr_chunk *ch_entry, *ch_next; 1609 struct sr_uuid uuid; 1610 u_int64_t version = 0; 1611 int sr, not_sr, rv = 1, d, expected = -1, old_meta = 0; 1612 1613 DNPRINTF(SR_D_META, "%s: sr_meta_native_attach\n", DEVNAME(sc)); 1614 1615 md = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, M_ZERO | M_NOWAIT); 1616 if (md == NULL) { 1617 sr_error(sc, "not enough memory for metadata buffer"); 1618 goto bad; 1619 } 1620 1621 bzero(&uuid, sizeof uuid); 1622 1623 sr = not_sr = d = 0; 1624 SLIST_FOREACH(ch_entry, cl, src_link) { 1625 if (ch_entry->src_dev_mm == NODEV) 1626 continue; 1627 1628 if (sr_meta_native_read(sd, ch_entry->src_dev_mm, md, NULL)) { 1629 sr_error(sc, "could not read native metadata"); 1630 goto bad; 1631 } 1632 1633 if (md->ssdi.ssd_magic == SR_MAGIC) { 1634 sr++; 1635 ch_entry->src_meta.scmi.scm_chunk_id = 1636 md->ssdi.ssd_chunk_id; 1637 if (d == 0) { 1638 memcpy(&uuid, &md->ssdi.ssd_uuid, sizeof uuid); 1639 expected = md->ssdi.ssd_chunk_no; 1640 version = md->ssd_ondisk; 1641 d++; 1642 continue; 1643 } else if (bcmp(&md->ssdi.ssd_uuid, &uuid, 1644 sizeof uuid)) { 1645 sr_error(sc, "not part of the same volume"); 1646 goto bad; 1647 } 1648 if (md->ssd_ondisk != version) { 1649 old_meta++; 1650 version = MAX(md->ssd_ondisk, version); 1651 } 1652 } else 1653 not_sr++; 1654 } 1655 1656 if (sr && not_sr && !force) { 1657 sr_error(sc, "not all chunks are of the native metadata " 1658 "format"); 1659 goto bad; 1660 } 1661 1662 /* mixed metadata versions; mark bad disks offline */ 1663 if (old_meta) { 1664 d = 0; 1665 for (ch_entry = SLIST_FIRST(cl); ch_entry != NULL; 1666 ch_entry = ch_next, d++) { 1667 ch_next = SLIST_NEXT(ch_entry, src_link); 1668 1669 /* XXX do we want to read this again? */ 1670 if (ch_entry->src_dev_mm == NODEV) 1671 panic("src_dev_mm == NODEV"); 1672 if (sr_meta_native_read(sd, ch_entry->src_dev_mm, md, 1673 NULL)) 1674 sr_warn(sc, "could not read native metadata"); 1675 if (md->ssd_ondisk != version) 1676 sd->sd_vol.sv_chunks[d]->src_meta.scm_status = 1677 BIOC_SDOFFLINE; 1678 } 1679 } 1680 1681 if (expected != sr && !force && expected != -1) { 1682 DNPRINTF(SR_D_META, "%s: not all chunks were provided, trying " 1683 "anyway\n", DEVNAME(sc)); 1684 } 1685 1686 rv = 0; 1687 bad: 1688 free(md, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 1689 return (rv); 1690 } 1691 1692 int 1693 sr_meta_native_read(struct sr_discipline *sd, dev_t dev, 1694 struct sr_metadata *md, void *fm) 1695 { 1696 #ifdef SR_DEBUG 1697 struct sr_softc *sc = sd->sd_sc; 1698 #endif 1699 DNPRINTF(SR_D_META, "%s: sr_meta_native_read(0x%x, %p)\n", 1700 DEVNAME(sc), dev, md); 1701 1702 return (sr_meta_rw(sd, dev, md, B_READ)); 1703 } 1704 1705 int 1706 sr_meta_native_write(struct sr_discipline *sd, dev_t dev, 1707 struct sr_metadata *md, void *fm) 1708 { 1709 #ifdef SR_DEBUG 1710 struct sr_softc *sc = sd->sd_sc; 1711 #endif 1712 DNPRINTF(SR_D_META, "%s: sr_meta_native_write(0x%x, %p)\n", 1713 DEVNAME(sc), dev, md); 1714 1715 return (sr_meta_rw(sd, dev, md, B_WRITE)); 1716 } 1717 1718 void 1719 sr_hotplug_register(struct sr_discipline *sd, void *func) 1720 { 1721 struct sr_hotplug_list *mhe; 1722 1723 DNPRINTF(SR_D_MISC, "%s: sr_hotplug_register: %p\n", 1724 DEVNAME(sd->sd_sc), func); 1725 1726 /* make sure we aren't on the list yet */ 1727 SLIST_FOREACH(mhe, &sr_hotplug_callbacks, shl_link) 1728 if (mhe->sh_hotplug == func) 1729 return; 1730 1731 mhe = malloc(sizeof(struct sr_hotplug_list), M_DEVBUF, 1732 M_WAITOK | M_ZERO); 1733 mhe->sh_hotplug = func; 1734 mhe->sh_sd = sd; 1735 SLIST_INSERT_HEAD(&sr_hotplug_callbacks, mhe, shl_link); 1736 } 1737 1738 void 1739 sr_hotplug_unregister(struct sr_discipline *sd, void *func) 1740 { 1741 struct sr_hotplug_list *mhe; 1742 1743 DNPRINTF(SR_D_MISC, "%s: sr_hotplug_unregister: %s %p\n", 1744 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, func); 1745 1746 /* make sure we are on the list yet */ 1747 SLIST_FOREACH(mhe, &sr_hotplug_callbacks, shl_link) { 1748 if (mhe->sh_hotplug == func) 1749 break; 1750 } 1751 if (mhe != NULL) { 1752 SLIST_REMOVE(&sr_hotplug_callbacks, mhe, 1753 sr_hotplug_list, shl_link); 1754 free(mhe, M_DEVBUF, 0); 1755 } 1756 } 1757 1758 void 1759 sr_disk_attach(struct disk *diskp, int action) 1760 { 1761 struct sr_hotplug_list *mhe; 1762 1763 SLIST_FOREACH(mhe, &sr_hotplug_callbacks, shl_link) 1764 if (mhe->sh_sd->sd_ready) 1765 mhe->sh_hotplug(mhe->sh_sd, diskp, action); 1766 } 1767 1768 int 1769 sr_match(struct device *parent, void *match, void *aux) 1770 { 1771 return (1); 1772 } 1773 1774 void 1775 sr_attach(struct device *parent, struct device *self, void *aux) 1776 { 1777 struct sr_softc *sc = (void *)self; 1778 struct scsibus_attach_args saa; 1779 1780 DNPRINTF(SR_D_MISC, "\n%s: sr_attach", DEVNAME(sc)); 1781 1782 if (softraid0 == NULL) 1783 softraid0 = sc; 1784 1785 rw_init(&sc->sc_lock, "sr_lock"); 1786 rw_init(&sc->sc_hs_lock, "sr_hs_lock"); 1787 1788 SLIST_INIT(&sr_hotplug_callbacks); 1789 TAILQ_INIT(&sc->sc_dis_list); 1790 SLIST_INIT(&sc->sc_hotspare_list); 1791 1792 #if NBIO > 0 1793 if (bio_register(&sc->sc_dev, sr_bio_ioctl) != 0) 1794 printf("%s: controller registration failed", DEVNAME(sc)); 1795 #endif /* NBIO > 0 */ 1796 1797 #ifndef SMALL_KERNEL 1798 strlcpy(sc->sc_sensordev.xname, DEVNAME(sc), 1799 sizeof(sc->sc_sensordev.xname)); 1800 sensordev_install(&sc->sc_sensordev); 1801 #endif /* SMALL_KERNEL */ 1802 1803 printf("\n"); 1804 1805 sc->sc_link.adapter_softc = sc; 1806 sc->sc_link.adapter = &sr_switch; 1807 sc->sc_link.adapter_target = SR_MAX_LD; 1808 sc->sc_link.adapter_buswidth = SR_MAX_LD; 1809 sc->sc_link.luns = 1; 1810 1811 bzero(&saa, sizeof(saa)); 1812 saa.saa_sc_link = &sc->sc_link; 1813 1814 sc->sc_scsibus = (struct scsibus_softc *)config_found(&sc->sc_dev, 1815 &saa, scsiprint); 1816 1817 softraid_disk_attach = sr_disk_attach; 1818 1819 sr_boot_assembly(sc); 1820 1821 explicit_bzero(sr_bootkey, sizeof(sr_bootkey)); 1822 } 1823 1824 int 1825 sr_detach(struct device *self, int flags) 1826 { 1827 struct sr_softc *sc = (void *)self; 1828 int rv; 1829 1830 DNPRINTF(SR_D_MISC, "%s: sr_detach\n", DEVNAME(sc)); 1831 1832 softraid_disk_attach = NULL; 1833 1834 sr_shutdown(0); 1835 1836 #ifndef SMALL_KERNEL 1837 if (sc->sc_sensor_task != NULL) 1838 sensor_task_unregister(sc->sc_sensor_task); 1839 sensordev_deinstall(&sc->sc_sensordev); 1840 #endif /* SMALL_KERNEL */ 1841 1842 if (sc->sc_scsibus != NULL) { 1843 rv = config_detach((struct device *)sc->sc_scsibus, flags); 1844 if (rv != 0) 1845 return (rv); 1846 sc->sc_scsibus = NULL; 1847 } 1848 1849 return (0); 1850 } 1851 1852 void 1853 sr_info(struct sr_softc *sc, const char *fmt, ...) 1854 { 1855 va_list ap; 1856 1857 rw_assert_wrlock(&sc->sc_lock); 1858 1859 va_start(ap, fmt); 1860 bio_status(&sc->sc_status, 0, BIO_MSG_INFO, fmt, &ap); 1861 va_end(ap); 1862 } 1863 1864 void 1865 sr_warn(struct sr_softc *sc, const char *fmt, ...) 1866 { 1867 va_list ap; 1868 1869 rw_assert_wrlock(&sc->sc_lock); 1870 1871 va_start(ap, fmt); 1872 bio_status(&sc->sc_status, 1, BIO_MSG_WARN, fmt, &ap); 1873 va_end(ap); 1874 } 1875 1876 void 1877 sr_error(struct sr_softc *sc, const char *fmt, ...) 1878 { 1879 va_list ap; 1880 1881 rw_assert_wrlock(&sc->sc_lock); 1882 1883 va_start(ap, fmt); 1884 bio_status(&sc->sc_status, 1, BIO_MSG_ERROR, fmt, &ap); 1885 va_end(ap); 1886 } 1887 1888 void 1889 sr_minphys(struct buf *bp, struct scsi_link *sl) 1890 { 1891 DNPRINTF(SR_D_MISC, "sr_minphys: %ld\n", bp->b_bcount); 1892 1893 /* XXX currently using SR_MAXFER = MAXPHYS */ 1894 if (bp->b_bcount > SR_MAXFER) 1895 bp->b_bcount = SR_MAXFER; 1896 minphys(bp); 1897 } 1898 1899 void 1900 sr_copy_internal_data(struct scsi_xfer *xs, void *v, size_t size) 1901 { 1902 size_t copy_cnt; 1903 1904 DNPRINTF(SR_D_MISC, "sr_copy_internal_data xs: %p size: %zu\n", 1905 xs, size); 1906 1907 if (xs->datalen) { 1908 copy_cnt = MIN(size, xs->datalen); 1909 memcpy(xs->data, v, copy_cnt); 1910 } 1911 } 1912 1913 int 1914 sr_ccb_alloc(struct sr_discipline *sd) 1915 { 1916 struct sr_ccb *ccb; 1917 int i; 1918 1919 if (!sd) 1920 return (1); 1921 1922 DNPRINTF(SR_D_CCB, "%s: sr_ccb_alloc\n", DEVNAME(sd->sd_sc)); 1923 1924 if (sd->sd_ccb) 1925 return (1); 1926 1927 sd->sd_ccb = mallocarray(sd->sd_max_wu, 1928 sd->sd_max_ccb_per_wu * sizeof(struct sr_ccb), 1929 M_DEVBUF, M_WAITOK | M_ZERO); 1930 TAILQ_INIT(&sd->sd_ccb_freeq); 1931 for (i = 0; i < sd->sd_max_wu * sd->sd_max_ccb_per_wu; i++) { 1932 ccb = &sd->sd_ccb[i]; 1933 ccb->ccb_dis = sd; 1934 sr_ccb_put(ccb); 1935 } 1936 1937 DNPRINTF(SR_D_CCB, "%s: sr_ccb_alloc ccb: %d\n", 1938 DEVNAME(sd->sd_sc), sd->sd_max_wu * sd->sd_max_ccb_per_wu); 1939 1940 return (0); 1941 } 1942 1943 void 1944 sr_ccb_free(struct sr_discipline *sd) 1945 { 1946 struct sr_ccb *ccb; 1947 1948 if (!sd) 1949 return; 1950 1951 DNPRINTF(SR_D_CCB, "%s: sr_ccb_free %p\n", DEVNAME(sd->sd_sc), sd); 1952 1953 while ((ccb = TAILQ_FIRST(&sd->sd_ccb_freeq)) != NULL) 1954 TAILQ_REMOVE(&sd->sd_ccb_freeq, ccb, ccb_link); 1955 1956 free(sd->sd_ccb, M_DEVBUF, 0); 1957 } 1958 1959 struct sr_ccb * 1960 sr_ccb_get(struct sr_discipline *sd) 1961 { 1962 struct sr_ccb *ccb; 1963 int s; 1964 1965 s = splbio(); 1966 1967 ccb = TAILQ_FIRST(&sd->sd_ccb_freeq); 1968 if (ccb) { 1969 TAILQ_REMOVE(&sd->sd_ccb_freeq, ccb, ccb_link); 1970 ccb->ccb_state = SR_CCB_INPROGRESS; 1971 } 1972 1973 splx(s); 1974 1975 DNPRINTF(SR_D_CCB, "%s: sr_ccb_get: %p\n", DEVNAME(sd->sd_sc), 1976 ccb); 1977 1978 return (ccb); 1979 } 1980 1981 void 1982 sr_ccb_put(struct sr_ccb *ccb) 1983 { 1984 struct sr_discipline *sd = ccb->ccb_dis; 1985 int s; 1986 1987 DNPRINTF(SR_D_CCB, "%s: sr_ccb_put: %p\n", DEVNAME(sd->sd_sc), 1988 ccb); 1989 1990 s = splbio(); 1991 1992 ccb->ccb_wu = NULL; 1993 ccb->ccb_state = SR_CCB_FREE; 1994 ccb->ccb_target = -1; 1995 ccb->ccb_opaque = NULL; 1996 1997 TAILQ_INSERT_TAIL(&sd->sd_ccb_freeq, ccb, ccb_link); 1998 1999 splx(s); 2000 } 2001 2002 struct sr_ccb * 2003 sr_ccb_rw(struct sr_discipline *sd, int chunk, daddr_t blkno, 2004 long len, u_int8_t *data, int xsflags, int ccbflags) 2005 { 2006 struct sr_chunk *sc = sd->sd_vol.sv_chunks[chunk]; 2007 struct sr_ccb *ccb = NULL; 2008 2009 ccb = sr_ccb_get(sd); 2010 if (ccb == NULL) 2011 goto out; 2012 2013 ccb->ccb_flags = ccbflags; 2014 ccb->ccb_target = chunk; 2015 2016 ccb->ccb_buf.b_flags = B_PHYS | B_CALL; 2017 if (ISSET(xsflags, SCSI_DATA_IN)) 2018 ccb->ccb_buf.b_flags |= B_READ; 2019 else 2020 ccb->ccb_buf.b_flags |= B_WRITE; 2021 2022 ccb->ccb_buf.b_blkno = blkno + sd->sd_meta->ssd_data_blkno; 2023 ccb->ccb_buf.b_bcount = len; 2024 ccb->ccb_buf.b_bufsize = len; 2025 ccb->ccb_buf.b_resid = len; 2026 ccb->ccb_buf.b_data = data; 2027 ccb->ccb_buf.b_error = 0; 2028 ccb->ccb_buf.b_iodone = sd->sd_scsi_intr; 2029 ccb->ccb_buf.b_proc = curproc; 2030 ccb->ccb_buf.b_dev = sc->src_dev_mm; 2031 ccb->ccb_buf.b_vp = sc->src_vn; 2032 ccb->ccb_buf.b_bq = NULL; 2033 2034 if (!ISSET(ccb->ccb_buf.b_flags, B_READ)) 2035 ccb->ccb_buf.b_vp->v_numoutput++; 2036 2037 LIST_INIT(&ccb->ccb_buf.b_dep); 2038 2039 DNPRINTF(SR_D_DIS, "%s: %s %s ccb " 2040 "b_bcount %ld b_blkno %lld b_flags 0x%0lx b_data %p\n", 2041 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, sd->sd_name, 2042 ccb->ccb_buf.b_bcount, (long long)ccb->ccb_buf.b_blkno, 2043 ccb->ccb_buf.b_flags, ccb->ccb_buf.b_data); 2044 2045 out: 2046 return ccb; 2047 } 2048 2049 void 2050 sr_ccb_done(struct sr_ccb *ccb) 2051 { 2052 struct sr_workunit *wu = ccb->ccb_wu; 2053 struct sr_discipline *sd = wu->swu_dis; 2054 struct sr_softc *sc = sd->sd_sc; 2055 2056 DNPRINTF(SR_D_INTR, "%s: %s %s ccb done b_bcount %ld b_resid %zu" 2057 " b_flags 0x%0lx block %lld target %d\n", 2058 DEVNAME(sc), sd->sd_meta->ssd_devname, sd->sd_name, 2059 ccb->ccb_buf.b_bcount, ccb->ccb_buf.b_resid, ccb->ccb_buf.b_flags, 2060 (long long)ccb->ccb_buf.b_blkno, ccb->ccb_target); 2061 2062 splassert(IPL_BIO); 2063 2064 if (ccb->ccb_target == -1) 2065 panic("%s: invalid target on wu: %p", DEVNAME(sc), wu); 2066 2067 if (ccb->ccb_buf.b_flags & B_ERROR) { 2068 DNPRINTF(SR_D_INTR, "%s: i/o error on block %lld target %d\n", 2069 DEVNAME(sc), (long long)ccb->ccb_buf.b_blkno, 2070 ccb->ccb_target); 2071 if (ISSET(sd->sd_capabilities, SR_CAP_REDUNDANT)) 2072 sd->sd_set_chunk_state(sd, ccb->ccb_target, 2073 BIOC_SDOFFLINE); 2074 else 2075 printf("%s: %s: i/o error %d @ %s block %lld\n", 2076 DEVNAME(sc), sd->sd_meta->ssd_devname, 2077 ccb->ccb_buf.b_error, sd->sd_name, 2078 (long long)ccb->ccb_buf.b_blkno); 2079 ccb->ccb_state = SR_CCB_FAILED; 2080 wu->swu_ios_failed++; 2081 } else { 2082 ccb->ccb_state = SR_CCB_OK; 2083 wu->swu_ios_succeeded++; 2084 } 2085 2086 wu->swu_ios_complete++; 2087 } 2088 2089 int 2090 sr_wu_alloc(struct sr_discipline *sd, int wu_size) 2091 { 2092 struct sr_workunit *wu; 2093 int i, no_wu; 2094 2095 DNPRINTF(SR_D_WU, "%s: sr_wu_alloc %p %d\n", DEVNAME(sd->sd_sc), 2096 sd, sd->sd_max_wu); 2097 2098 no_wu = sd->sd_max_wu; 2099 sd->sd_wu_pending = no_wu; 2100 2101 mtx_init(&sd->sd_wu_mtx, IPL_BIO); 2102 TAILQ_INIT(&sd->sd_wu); 2103 TAILQ_INIT(&sd->sd_wu_freeq); 2104 TAILQ_INIT(&sd->sd_wu_pendq); 2105 TAILQ_INIT(&sd->sd_wu_defq); 2106 2107 for (i = 0; i < no_wu; i++) { 2108 wu = malloc(wu_size, M_DEVBUF, M_WAITOK | M_ZERO); 2109 TAILQ_INSERT_TAIL(&sd->sd_wu, wu, swu_next); 2110 TAILQ_INIT(&wu->swu_ccb); 2111 wu->swu_dis = sd; 2112 task_set(&wu->swu_task, sr_wu_done_callback, wu); 2113 sr_wu_put(sd, wu); 2114 } 2115 2116 return (0); 2117 } 2118 2119 void 2120 sr_wu_free(struct sr_discipline *sd) 2121 { 2122 struct sr_workunit *wu; 2123 2124 DNPRINTF(SR_D_WU, "%s: sr_wu_free %p\n", DEVNAME(sd->sd_sc), sd); 2125 2126 while ((wu = TAILQ_FIRST(&sd->sd_wu_freeq)) != NULL) 2127 TAILQ_REMOVE(&sd->sd_wu_freeq, wu, swu_link); 2128 while ((wu = TAILQ_FIRST(&sd->sd_wu_pendq)) != NULL) 2129 TAILQ_REMOVE(&sd->sd_wu_pendq, wu, swu_link); 2130 while ((wu = TAILQ_FIRST(&sd->sd_wu_defq)) != NULL) 2131 TAILQ_REMOVE(&sd->sd_wu_defq, wu, swu_link); 2132 2133 while ((wu = TAILQ_FIRST(&sd->sd_wu)) != NULL) { 2134 TAILQ_REMOVE(&sd->sd_wu, wu, swu_next); 2135 free(wu, M_DEVBUF, 0); 2136 } 2137 } 2138 2139 void * 2140 sr_wu_get(void *xsd) 2141 { 2142 struct sr_discipline *sd = (struct sr_discipline *)xsd; 2143 struct sr_workunit *wu; 2144 2145 mtx_enter(&sd->sd_wu_mtx); 2146 wu = TAILQ_FIRST(&sd->sd_wu_freeq); 2147 if (wu) { 2148 TAILQ_REMOVE(&sd->sd_wu_freeq, wu, swu_link); 2149 sd->sd_wu_pending++; 2150 } 2151 mtx_leave(&sd->sd_wu_mtx); 2152 2153 DNPRINTF(SR_D_WU, "%s: sr_wu_get: %p\n", DEVNAME(sd->sd_sc), wu); 2154 2155 return (wu); 2156 } 2157 2158 void 2159 sr_wu_put(void *xsd, void *xwu) 2160 { 2161 struct sr_discipline *sd = (struct sr_discipline *)xsd; 2162 struct sr_workunit *wu = (struct sr_workunit *)xwu; 2163 2164 DNPRINTF(SR_D_WU, "%s: sr_wu_put: %p\n", DEVNAME(sd->sd_sc), wu); 2165 2166 sr_wu_release_ccbs(wu); 2167 sr_wu_init(sd, wu); 2168 2169 mtx_enter(&sd->sd_wu_mtx); 2170 TAILQ_INSERT_TAIL(&sd->sd_wu_freeq, wu, swu_link); 2171 sd->sd_wu_pending--; 2172 mtx_leave(&sd->sd_wu_mtx); 2173 } 2174 2175 void 2176 sr_wu_init(struct sr_discipline *sd, struct sr_workunit *wu) 2177 { 2178 int s; 2179 2180 s = splbio(); 2181 if (wu->swu_cb_active == 1) 2182 panic("%s: sr_wu_init got active wu", DEVNAME(sd->sd_sc)); 2183 splx(s); 2184 2185 wu->swu_xs = NULL; 2186 wu->swu_state = SR_WU_FREE; 2187 wu->swu_flags = 0; 2188 wu->swu_blk_start = 0; 2189 wu->swu_blk_end = 0; 2190 wu->swu_collider = NULL; 2191 } 2192 2193 void 2194 sr_wu_enqueue_ccb(struct sr_workunit *wu, struct sr_ccb *ccb) 2195 { 2196 struct sr_discipline *sd = wu->swu_dis; 2197 int s; 2198 2199 s = splbio(); 2200 if (wu->swu_cb_active == 1) 2201 panic("%s: sr_wu_enqueue_ccb got active wu", 2202 DEVNAME(sd->sd_sc)); 2203 ccb->ccb_wu = wu; 2204 wu->swu_io_count++; 2205 TAILQ_INSERT_TAIL(&wu->swu_ccb, ccb, ccb_link); 2206 splx(s); 2207 } 2208 2209 void 2210 sr_wu_release_ccbs(struct sr_workunit *wu) 2211 { 2212 struct sr_ccb *ccb; 2213 2214 /* Return all ccbs that are associated with this workunit. */ 2215 while ((ccb = TAILQ_FIRST(&wu->swu_ccb)) != NULL) { 2216 TAILQ_REMOVE(&wu->swu_ccb, ccb, ccb_link); 2217 sr_ccb_put(ccb); 2218 } 2219 2220 wu->swu_io_count = 0; 2221 wu->swu_ios_complete = 0; 2222 wu->swu_ios_failed = 0; 2223 wu->swu_ios_succeeded = 0; 2224 } 2225 2226 void 2227 sr_wu_done(struct sr_workunit *wu) 2228 { 2229 struct sr_discipline *sd = wu->swu_dis; 2230 2231 DNPRINTF(SR_D_INTR, "%s: sr_wu_done count %d completed %d failed %d\n", 2232 DEVNAME(sd->sd_sc), wu->swu_io_count, wu->swu_ios_complete, 2233 wu->swu_ios_failed); 2234 2235 if (wu->swu_ios_complete < wu->swu_io_count) 2236 return; 2237 2238 task_add(sd->sd_taskq, &wu->swu_task); 2239 } 2240 2241 void 2242 sr_wu_done_callback(void *xwu) 2243 { 2244 struct sr_workunit *wu = xwu; 2245 struct sr_discipline *sd = wu->swu_dis; 2246 struct scsi_xfer *xs = wu->swu_xs; 2247 struct sr_workunit *wup; 2248 int s; 2249 2250 /* 2251 * The SR_WUF_DISCIPLINE or SR_WUF_REBUILD flag must be set if 2252 * the work unit is not associated with a scsi_xfer. 2253 */ 2254 KASSERT(xs != NULL || 2255 (wu->swu_flags & (SR_WUF_DISCIPLINE|SR_WUF_REBUILD))); 2256 2257 s = splbio(); 2258 2259 if (xs != NULL) { 2260 if (wu->swu_ios_failed) 2261 xs->error = XS_DRIVER_STUFFUP; 2262 else 2263 xs->error = XS_NOERROR; 2264 } 2265 2266 if (sd->sd_scsi_wu_done) { 2267 if (sd->sd_scsi_wu_done(wu) == SR_WU_RESTART) 2268 goto done; 2269 } 2270 2271 /* Remove work unit from pending queue. */ 2272 TAILQ_FOREACH(wup, &sd->sd_wu_pendq, swu_link) 2273 if (wup == wu) 2274 break; 2275 if (wup == NULL) 2276 panic("%s: wu %p not on pending queue", 2277 DEVNAME(sd->sd_sc), wu); 2278 TAILQ_REMOVE(&sd->sd_wu_pendq, wu, swu_link); 2279 2280 if (wu->swu_collider) { 2281 if (wu->swu_ios_failed) 2282 sr_raid_recreate_wu(wu->swu_collider); 2283 2284 /* XXX Should the collider be failed if this xs failed? */ 2285 sr_raid_startwu(wu->swu_collider); 2286 } 2287 2288 /* 2289 * If a discipline provides its own sd_scsi_done function, then it 2290 * is responsible for calling sr_scsi_done() once I/O is complete. 2291 */ 2292 if (wu->swu_flags & SR_WUF_REBUILD) 2293 wu->swu_flags |= SR_WUF_REBUILDIOCOMP; 2294 if (wu->swu_flags & SR_WUF_WAKEUP) 2295 wakeup(wu); 2296 if (sd->sd_scsi_done) 2297 sd->sd_scsi_done(wu); 2298 else if (wu->swu_flags & SR_WUF_DISCIPLINE) 2299 sr_scsi_wu_put(sd, wu); 2300 else if (!(wu->swu_flags & SR_WUF_REBUILD)) 2301 sr_scsi_done(sd, xs); 2302 2303 done: 2304 splx(s); 2305 } 2306 2307 struct sr_workunit * 2308 sr_scsi_wu_get(struct sr_discipline *sd, int flags) 2309 { 2310 return scsi_io_get(&sd->sd_iopool, flags); 2311 } 2312 2313 void 2314 sr_scsi_wu_put(struct sr_discipline *sd, struct sr_workunit *wu) 2315 { 2316 scsi_io_put(&sd->sd_iopool, wu); 2317 2318 if (sd->sd_sync && sd->sd_wu_pending == 0) 2319 wakeup(sd); 2320 } 2321 2322 void 2323 sr_scsi_done(struct sr_discipline *sd, struct scsi_xfer *xs) 2324 { 2325 DNPRINTF(SR_D_DIS, "%s: sr_scsi_done: xs %p\n", DEVNAME(sd->sd_sc), xs); 2326 2327 if (xs->error == XS_NOERROR) 2328 xs->resid = 0; 2329 2330 scsi_done(xs); 2331 2332 if (sd->sd_sync && sd->sd_wu_pending == 0) 2333 wakeup(sd); 2334 } 2335 2336 void 2337 sr_scsi_cmd(struct scsi_xfer *xs) 2338 { 2339 struct scsi_link *link = xs->sc_link; 2340 struct sr_softc *sc = link->adapter_softc; 2341 struct sr_workunit *wu = xs->io; 2342 struct sr_discipline *sd; 2343 2344 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd target %d xs %p flags %#x\n", 2345 DEVNAME(sc), link->target, xs, xs->flags); 2346 2347 sd = sc->sc_targets[link->target]; 2348 if (sd == NULL) 2349 panic("%s: sr_scsi_cmd NULL discipline", DEVNAME(sc)); 2350 2351 if (sd->sd_deleted) { 2352 printf("%s: %s device is being deleted, failing io\n", 2353 DEVNAME(sc), sd->sd_meta->ssd_devname); 2354 goto stuffup; 2355 } 2356 2357 /* scsi layer *can* re-send wu without calling sr_wu_put(). */ 2358 sr_wu_release_ccbs(wu); 2359 sr_wu_init(sd, wu); 2360 wu->swu_state = SR_WU_INPROGRESS; 2361 wu->swu_xs = xs; 2362 2363 switch (xs->cmd->opcode) { 2364 case READ_COMMAND: 2365 case READ_BIG: 2366 case READ_16: 2367 case WRITE_COMMAND: 2368 case WRITE_BIG: 2369 case WRITE_16: 2370 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: READ/WRITE %02x\n", 2371 DEVNAME(sc), xs->cmd->opcode); 2372 if (sd->sd_scsi_rw(wu)) 2373 goto stuffup; 2374 break; 2375 2376 case SYNCHRONIZE_CACHE: 2377 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: SYNCHRONIZE_CACHE\n", 2378 DEVNAME(sc)); 2379 if (sd->sd_scsi_sync(wu)) 2380 goto stuffup; 2381 goto complete; 2382 2383 case TEST_UNIT_READY: 2384 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: TEST_UNIT_READY\n", 2385 DEVNAME(sc)); 2386 if (sd->sd_scsi_tur(wu)) 2387 goto stuffup; 2388 goto complete; 2389 2390 case START_STOP: 2391 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: START_STOP\n", 2392 DEVNAME(sc)); 2393 if (sd->sd_scsi_start_stop(wu)) 2394 goto stuffup; 2395 goto complete; 2396 2397 case INQUIRY: 2398 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: INQUIRY\n", 2399 DEVNAME(sc)); 2400 if (sd->sd_scsi_inquiry(wu)) 2401 goto stuffup; 2402 goto complete; 2403 2404 case READ_CAPACITY: 2405 case READ_CAPACITY_16: 2406 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd READ CAPACITY 0x%02x\n", 2407 DEVNAME(sc), xs->cmd->opcode); 2408 if (sd->sd_scsi_read_cap(wu)) 2409 goto stuffup; 2410 goto complete; 2411 2412 case REQUEST_SENSE: 2413 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd REQUEST SENSE\n", 2414 DEVNAME(sc)); 2415 if (sd->sd_scsi_req_sense(wu)) 2416 goto stuffup; 2417 goto complete; 2418 2419 default: 2420 DNPRINTF(SR_D_CMD, "%s: unsupported scsi command %x\n", 2421 DEVNAME(sc), xs->cmd->opcode); 2422 /* XXX might need to add generic function to handle others */ 2423 goto stuffup; 2424 } 2425 2426 return; 2427 stuffup: 2428 if (sd->sd_scsi_sense.error_code) { 2429 xs->error = XS_SENSE; 2430 memcpy(&xs->sense, &sd->sd_scsi_sense, sizeof(xs->sense)); 2431 bzero(&sd->sd_scsi_sense, sizeof(sd->sd_scsi_sense)); 2432 } else { 2433 xs->error = XS_DRIVER_STUFFUP; 2434 } 2435 complete: 2436 sr_scsi_done(sd, xs); 2437 } 2438 2439 int 2440 sr_scsi_probe(struct scsi_link *link) 2441 { 2442 struct sr_softc *sc = link->adapter_softc; 2443 struct sr_discipline *sd; 2444 2445 KASSERT(link->target < SR_MAX_LD && link->lun == 0); 2446 2447 sd = sc->sc_targets[link->target]; 2448 if (sd == NULL) 2449 return (ENODEV); 2450 2451 link->pool = &sd->sd_iopool; 2452 if (sd->sd_openings) 2453 link->openings = sd->sd_openings(sd); 2454 else 2455 link->openings = sd->sd_max_wu; 2456 2457 return (0); 2458 } 2459 2460 int 2461 sr_scsi_ioctl(struct scsi_link *link, u_long cmd, caddr_t addr, int flag) 2462 { 2463 struct sr_softc *sc = link->adapter_softc; 2464 struct sr_discipline *sd; 2465 2466 sd = sc->sc_targets[link->target]; 2467 if (sd == NULL) 2468 return (ENODEV); 2469 2470 DNPRINTF(SR_D_IOCTL, "%s: %s sr_scsi_ioctl cmd: %#lx\n", 2471 DEVNAME(sc), sd->sd_meta->ssd_devname, cmd); 2472 2473 /* Pass bio ioctls through to the bio handler. */ 2474 if (IOCGROUP(cmd) == 'B') 2475 return (sr_bio_handler(sc, sd, cmd, (struct bio *)addr)); 2476 2477 switch (cmd) { 2478 case DIOCGCACHE: 2479 case DIOCSCACHE: 2480 return (EOPNOTSUPP); 2481 default: 2482 return (ENOTTY); 2483 } 2484 } 2485 2486 int 2487 sr_bio_ioctl(struct device *dev, u_long cmd, caddr_t addr) 2488 { 2489 struct sr_softc *sc = (struct sr_softc *) dev; 2490 DNPRINTF(SR_D_IOCTL, "%s: sr_bio_ioctl\n", DEVNAME(sc)); 2491 2492 return sr_bio_handler(sc, NULL, cmd, (struct bio *)addr); 2493 } 2494 2495 int 2496 sr_bio_handler(struct sr_softc *sc, struct sr_discipline *sd, u_long cmd, 2497 struct bio *bio) 2498 { 2499 int rv = 0; 2500 2501 DNPRINTF(SR_D_IOCTL, "%s: sr_bio_handler ", DEVNAME(sc)); 2502 2503 rw_enter_write(&sc->sc_lock); 2504 2505 bio_status_init(&sc->sc_status, &sc->sc_dev); 2506 2507 switch (cmd) { 2508 case BIOCINQ: 2509 DNPRINTF(SR_D_IOCTL, "inq\n"); 2510 rv = sr_ioctl_inq(sc, (struct bioc_inq *)bio); 2511 break; 2512 2513 case BIOCVOL: 2514 DNPRINTF(SR_D_IOCTL, "vol\n"); 2515 rv = sr_ioctl_vol(sc, (struct bioc_vol *)bio); 2516 break; 2517 2518 case BIOCDISK: 2519 DNPRINTF(SR_D_IOCTL, "disk\n"); 2520 rv = sr_ioctl_disk(sc, (struct bioc_disk *)bio); 2521 break; 2522 2523 case BIOCALARM: 2524 DNPRINTF(SR_D_IOCTL, "alarm\n"); 2525 /*rv = sr_ioctl_alarm(sc, (struct bioc_alarm *)bio); */ 2526 break; 2527 2528 case BIOCBLINK: 2529 DNPRINTF(SR_D_IOCTL, "blink\n"); 2530 /*rv = sr_ioctl_blink(sc, (struct bioc_blink *)bio); */ 2531 break; 2532 2533 case BIOCSETSTATE: 2534 DNPRINTF(SR_D_IOCTL, "setstate\n"); 2535 rv = sr_ioctl_setstate(sc, (struct bioc_setstate *)bio); 2536 break; 2537 2538 case BIOCCREATERAID: 2539 DNPRINTF(SR_D_IOCTL, "createraid\n"); 2540 rv = sr_ioctl_createraid(sc, (struct bioc_createraid *)bio, 2541 1, NULL); 2542 break; 2543 2544 case BIOCDELETERAID: 2545 DNPRINTF(SR_D_IOCTL, "deleteraid\n"); 2546 rv = sr_ioctl_deleteraid(sc, sd, (struct bioc_deleteraid *)bio); 2547 break; 2548 2549 case BIOCDISCIPLINE: 2550 DNPRINTF(SR_D_IOCTL, "discipline\n"); 2551 rv = sr_ioctl_discipline(sc, sd, (struct bioc_discipline *)bio); 2552 break; 2553 2554 case BIOCINSTALLBOOT: 2555 DNPRINTF(SR_D_IOCTL, "installboot\n"); 2556 rv = sr_ioctl_installboot(sc, sd, 2557 (struct bioc_installboot *)bio); 2558 break; 2559 2560 default: 2561 DNPRINTF(SR_D_IOCTL, "invalid ioctl\n"); 2562 rv = ENOTTY; 2563 } 2564 2565 sc->sc_status.bs_status = (rv ? BIO_STATUS_ERROR : BIO_STATUS_SUCCESS); 2566 2567 if (sc->sc_status.bs_msg_count > 0) 2568 rv = 0; 2569 2570 memcpy(&bio->bio_status, &sc->sc_status, sizeof(struct bio_status)); 2571 2572 rw_exit_write(&sc->sc_lock); 2573 2574 return (rv); 2575 } 2576 2577 int 2578 sr_ioctl_inq(struct sr_softc *sc, struct bioc_inq *bi) 2579 { 2580 struct sr_discipline *sd; 2581 int vol = 0, disk = 0; 2582 2583 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2584 vol++; 2585 disk += sd->sd_meta->ssdi.ssd_chunk_no; 2586 } 2587 2588 strlcpy(bi->bi_dev, sc->sc_dev.dv_xname, sizeof(bi->bi_dev)); 2589 bi->bi_novol = vol + sc->sc_hotspare_no; 2590 bi->bi_nodisk = disk + sc->sc_hotspare_no; 2591 2592 return (0); 2593 } 2594 2595 int 2596 sr_ioctl_vol(struct sr_softc *sc, struct bioc_vol *bv) 2597 { 2598 int vol = -1, rv = EINVAL; 2599 struct sr_discipline *sd; 2600 struct sr_chunk *hotspare; 2601 2602 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2603 vol++; 2604 if (vol != bv->bv_volid) 2605 continue; 2606 2607 bv->bv_status = sd->sd_vol_status; 2608 bv->bv_size = sd->sd_meta->ssdi.ssd_size << DEV_BSHIFT; 2609 bv->bv_level = sd->sd_meta->ssdi.ssd_level; 2610 bv->bv_nodisk = sd->sd_meta->ssdi.ssd_chunk_no; 2611 2612 #ifdef CRYPTO 2613 if (sd->sd_meta->ssdi.ssd_level == 'C' && 2614 sd->mds.mdd_crypto.key_disk != NULL) 2615 bv->bv_nodisk++; 2616 #endif 2617 if (bv->bv_status == BIOC_SVREBUILD) 2618 bv->bv_percent = sr_rebuild_percent(sd); 2619 2620 strlcpy(bv->bv_dev, sd->sd_meta->ssd_devname, 2621 sizeof(bv->bv_dev)); 2622 strlcpy(bv->bv_vendor, sd->sd_meta->ssdi.ssd_vendor, 2623 sizeof(bv->bv_vendor)); 2624 rv = 0; 2625 goto done; 2626 } 2627 2628 /* Check hotspares list. */ 2629 SLIST_FOREACH(hotspare, &sc->sc_hotspare_list, src_link) { 2630 vol++; 2631 if (vol != bv->bv_volid) 2632 continue; 2633 2634 bv->bv_status = BIOC_SVONLINE; 2635 bv->bv_size = hotspare->src_meta.scmi.scm_size << DEV_BSHIFT; 2636 bv->bv_level = -1; /* Hotspare. */ 2637 bv->bv_nodisk = 1; 2638 strlcpy(bv->bv_dev, hotspare->src_meta.scmi.scm_devname, 2639 sizeof(bv->bv_dev)); 2640 strlcpy(bv->bv_vendor, hotspare->src_meta.scmi.scm_devname, 2641 sizeof(bv->bv_vendor)); 2642 rv = 0; 2643 goto done; 2644 } 2645 2646 done: 2647 return (rv); 2648 } 2649 2650 int 2651 sr_ioctl_disk(struct sr_softc *sc, struct bioc_disk *bd) 2652 { 2653 struct sr_discipline *sd; 2654 struct sr_chunk *src, *hotspare; 2655 int vol = -1, rv = EINVAL; 2656 2657 if (bd->bd_diskid < 0) 2658 goto done; 2659 2660 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2661 vol++; 2662 if (vol != bd->bd_volid) 2663 continue; 2664 2665 if (bd->bd_diskid < sd->sd_meta->ssdi.ssd_chunk_no) 2666 src = sd->sd_vol.sv_chunks[bd->bd_diskid]; 2667 #ifdef CRYPTO 2668 else if (bd->bd_diskid == sd->sd_meta->ssdi.ssd_chunk_no && 2669 sd->sd_meta->ssdi.ssd_level == 'C' && 2670 sd->mds.mdd_crypto.key_disk != NULL) 2671 src = sd->mds.mdd_crypto.key_disk; 2672 #endif 2673 else 2674 break; 2675 2676 bd->bd_status = src->src_meta.scm_status; 2677 bd->bd_size = src->src_meta.scmi.scm_size << DEV_BSHIFT; 2678 bd->bd_channel = vol; 2679 bd->bd_target = bd->bd_diskid; 2680 strlcpy(bd->bd_vendor, src->src_meta.scmi.scm_devname, 2681 sizeof(bd->bd_vendor)); 2682 rv = 0; 2683 goto done; 2684 } 2685 2686 /* Check hotspares list. */ 2687 SLIST_FOREACH(hotspare, &sc->sc_hotspare_list, src_link) { 2688 vol++; 2689 if (vol != bd->bd_volid) 2690 continue; 2691 2692 if (bd->bd_diskid != 0) 2693 break; 2694 2695 bd->bd_status = hotspare->src_meta.scm_status; 2696 bd->bd_size = hotspare->src_meta.scmi.scm_size << DEV_BSHIFT; 2697 bd->bd_channel = vol; 2698 bd->bd_target = bd->bd_diskid; 2699 strlcpy(bd->bd_vendor, hotspare->src_meta.scmi.scm_devname, 2700 sizeof(bd->bd_vendor)); 2701 rv = 0; 2702 goto done; 2703 } 2704 2705 done: 2706 return (rv); 2707 } 2708 2709 int 2710 sr_ioctl_setstate(struct sr_softc *sc, struct bioc_setstate *bs) 2711 { 2712 int rv = EINVAL; 2713 int vol = -1, found, c; 2714 struct sr_discipline *sd; 2715 struct sr_chunk *ch_entry; 2716 struct sr_chunk_head *cl; 2717 2718 if (bs->bs_other_id_type == BIOC_SSOTHER_UNUSED) 2719 goto done; 2720 2721 if (bs->bs_status == BIOC_SSHOTSPARE) { 2722 rv = sr_hotspare(sc, (dev_t)bs->bs_other_id); 2723 goto done; 2724 } 2725 2726 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2727 vol++; 2728 if (vol == bs->bs_volid) 2729 break; 2730 } 2731 if (sd == NULL) 2732 goto done; 2733 2734 switch (bs->bs_status) { 2735 case BIOC_SSOFFLINE: 2736 /* Take chunk offline */ 2737 found = c = 0; 2738 cl = &sd->sd_vol.sv_chunk_list; 2739 SLIST_FOREACH(ch_entry, cl, src_link) { 2740 if (ch_entry->src_dev_mm == bs->bs_other_id) { 2741 found = 1; 2742 break; 2743 } 2744 c++; 2745 } 2746 if (found == 0) { 2747 sr_error(sc, "chunk not part of array"); 2748 goto done; 2749 } 2750 2751 /* XXX: check current state first */ 2752 sd->sd_set_chunk_state(sd, c, BIOC_SDOFFLINE); 2753 2754 if (sr_meta_save(sd, SR_META_DIRTY)) { 2755 sr_error(sc, "could not save metadata for %s", 2756 sd->sd_meta->ssd_devname); 2757 goto done; 2758 } 2759 rv = 0; 2760 break; 2761 2762 case BIOC_SDSCRUB: 2763 break; 2764 2765 case BIOC_SSREBUILD: 2766 rv = sr_rebuild_init(sd, (dev_t)bs->bs_other_id, 0); 2767 break; 2768 2769 default: 2770 sr_error(sc, "unsupported state request %d", bs->bs_status); 2771 } 2772 2773 done: 2774 return (rv); 2775 } 2776 2777 int 2778 sr_chunk_in_use(struct sr_softc *sc, dev_t dev) 2779 { 2780 struct sr_discipline *sd; 2781 struct sr_chunk *chunk; 2782 int i; 2783 2784 DNPRINTF(SR_D_MISC, "%s: sr_chunk_in_use(%d)\n", DEVNAME(sc), dev); 2785 2786 if (dev == NODEV) 2787 return BIOC_SDINVALID; 2788 2789 /* See if chunk is already in use. */ 2790 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2791 for (i = 0; i < sd->sd_meta->ssdi.ssd_chunk_no; i++) { 2792 chunk = sd->sd_vol.sv_chunks[i]; 2793 if (chunk->src_dev_mm == dev) 2794 return chunk->src_meta.scm_status; 2795 } 2796 } 2797 2798 /* Check hotspares list. */ 2799 SLIST_FOREACH(chunk, &sc->sc_hotspare_list, src_link) 2800 if (chunk->src_dev_mm == dev) 2801 return chunk->src_meta.scm_status; 2802 2803 return BIOC_SDINVALID; 2804 } 2805 2806 int 2807 sr_hotspare(struct sr_softc *sc, dev_t dev) 2808 { 2809 struct sr_discipline *sd = NULL; 2810 struct sr_metadata *sm = NULL; 2811 struct sr_meta_chunk *hm; 2812 struct sr_chunk_head *cl; 2813 struct sr_chunk *chunk, *last, *hotspare = NULL; 2814 struct sr_uuid uuid; 2815 struct disklabel label; 2816 struct vnode *vn; 2817 u_int64_t size; 2818 char devname[32]; 2819 int rv = EINVAL; 2820 int c, part, open = 0; 2821 2822 /* 2823 * Add device to global hotspares list. 2824 */ 2825 2826 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 2827 2828 /* Make sure chunk is not already in use. */ 2829 c = sr_chunk_in_use(sc, dev); 2830 if (c != BIOC_SDINVALID && c != BIOC_SDOFFLINE) { 2831 if (c == BIOC_SDHOTSPARE) 2832 sr_error(sc, "%s is already a hotspare", devname); 2833 else 2834 sr_error(sc, "%s is already in use", devname); 2835 goto done; 2836 } 2837 2838 /* XXX - See if there is an existing degraded volume... */ 2839 2840 /* Open device. */ 2841 if (bdevvp(dev, &vn)) { 2842 sr_error(sc, "sr_hotspare: cannot allocate vnode"); 2843 goto done; 2844 } 2845 if (VOP_OPEN(vn, FREAD | FWRITE, NOCRED, curproc)) { 2846 DNPRINTF(SR_D_META,"%s: sr_hotspare cannot open %s\n", 2847 DEVNAME(sc), devname); 2848 vput(vn); 2849 goto fail; 2850 } 2851 open = 1; /* close dev on error */ 2852 2853 /* Get partition details. */ 2854 part = DISKPART(dev); 2855 if (VOP_IOCTL(vn, DIOCGDINFO, (caddr_t)&label, FREAD, 2856 NOCRED, curproc)) { 2857 DNPRINTF(SR_D_META, "%s: sr_hotspare ioctl failed\n", 2858 DEVNAME(sc)); 2859 VOP_CLOSE(vn, FREAD | FWRITE, NOCRED, curproc); 2860 vput(vn); 2861 goto fail; 2862 } 2863 if (label.d_partitions[part].p_fstype != FS_RAID) { 2864 sr_error(sc, "%s partition not of type RAID (%d)", 2865 devname, label.d_partitions[part].p_fstype); 2866 goto fail; 2867 } 2868 2869 /* Calculate partition size. */ 2870 size = DL_SECTOBLK(&label, DL_GETPSIZE(&label.d_partitions[part])); 2871 if (size <= SR_DATA_OFFSET) { 2872 DNPRINTF(SR_D_META, "%s: %s partition too small\n", DEVNAME(sc), 2873 devname); 2874 goto fail; 2875 } 2876 size -= SR_DATA_OFFSET; 2877 if (size > INT64_MAX) { 2878 DNPRINTF(SR_D_META, "%s: %s partition too large\n", DEVNAME(sc), 2879 devname); 2880 goto fail; 2881 } 2882 2883 /* 2884 * Create and populate chunk metadata. 2885 */ 2886 2887 sr_uuid_generate(&uuid); 2888 hotspare = malloc(sizeof(struct sr_chunk), M_DEVBUF, M_WAITOK | M_ZERO); 2889 2890 hotspare->src_dev_mm = dev; 2891 hotspare->src_vn = vn; 2892 strlcpy(hotspare->src_devname, devname, sizeof(hm->scmi.scm_devname)); 2893 hotspare->src_size = size; 2894 2895 hm = &hotspare->src_meta; 2896 hm->scmi.scm_volid = SR_HOTSPARE_VOLID; 2897 hm->scmi.scm_chunk_id = 0; 2898 hm->scmi.scm_size = size; 2899 hm->scmi.scm_coerced_size = size; 2900 strlcpy(hm->scmi.scm_devname, devname, sizeof(hm->scmi.scm_devname)); 2901 memcpy(&hm->scmi.scm_uuid, &uuid, sizeof(struct sr_uuid)); 2902 2903 sr_checksum(sc, hm, &hm->scm_checksum, 2904 sizeof(struct sr_meta_chunk_invariant)); 2905 2906 hm->scm_status = BIOC_SDHOTSPARE; 2907 2908 /* 2909 * Create and populate our own discipline and metadata. 2910 */ 2911 2912 sm = malloc(sizeof(struct sr_metadata), M_DEVBUF, M_WAITOK | M_ZERO); 2913 sm->ssdi.ssd_magic = SR_MAGIC; 2914 sm->ssdi.ssd_version = SR_META_VERSION; 2915 sm->ssd_ondisk = 0; 2916 sm->ssdi.ssd_vol_flags = 0; 2917 memcpy(&sm->ssdi.ssd_uuid, &uuid, sizeof(struct sr_uuid)); 2918 sm->ssdi.ssd_chunk_no = 1; 2919 sm->ssdi.ssd_volid = SR_HOTSPARE_VOLID; 2920 sm->ssdi.ssd_level = SR_HOTSPARE_LEVEL; 2921 sm->ssdi.ssd_size = size; 2922 sm->ssdi.ssd_secsize = label.d_secsize; 2923 strlcpy(sm->ssdi.ssd_vendor, "OPENBSD", sizeof(sm->ssdi.ssd_vendor)); 2924 snprintf(sm->ssdi.ssd_product, sizeof(sm->ssdi.ssd_product), 2925 "SR %s", "HOTSPARE"); 2926 snprintf(sm->ssdi.ssd_revision, sizeof(sm->ssdi.ssd_revision), 2927 "%03d", SR_META_VERSION); 2928 2929 sd = malloc(sizeof(struct sr_discipline), M_DEVBUF, M_WAITOK | M_ZERO); 2930 sd->sd_sc = sc; 2931 sd->sd_meta = sm; 2932 sd->sd_meta_type = SR_META_F_NATIVE; 2933 sd->sd_vol_status = BIOC_SVONLINE; 2934 strlcpy(sd->sd_name, "HOTSPARE", sizeof(sd->sd_name)); 2935 SLIST_INIT(&sd->sd_meta_opt); 2936 2937 /* Add chunk to volume. */ 2938 sd->sd_vol.sv_chunks = malloc(sizeof(struct sr_chunk *), M_DEVBUF, 2939 M_WAITOK | M_ZERO); 2940 sd->sd_vol.sv_chunks[0] = hotspare; 2941 SLIST_INIT(&sd->sd_vol.sv_chunk_list); 2942 SLIST_INSERT_HEAD(&sd->sd_vol.sv_chunk_list, hotspare, src_link); 2943 2944 /* Save metadata. */ 2945 if (sr_meta_save(sd, SR_META_DIRTY)) { 2946 sr_error(sc, "could not save metadata to %s", devname); 2947 goto fail; 2948 } 2949 2950 /* 2951 * Add chunk to hotspare list. 2952 */ 2953 rw_enter_write(&sc->sc_hs_lock); 2954 cl = &sc->sc_hotspare_list; 2955 if (SLIST_EMPTY(cl)) 2956 SLIST_INSERT_HEAD(cl, hotspare, src_link); 2957 else { 2958 SLIST_FOREACH(chunk, cl, src_link) 2959 last = chunk; 2960 SLIST_INSERT_AFTER(last, hotspare, src_link); 2961 } 2962 sc->sc_hotspare_no++; 2963 rw_exit_write(&sc->sc_hs_lock); 2964 2965 rv = 0; 2966 goto done; 2967 2968 fail: 2969 free(hotspare, M_DEVBUF, 0); 2970 2971 done: 2972 if (sd) 2973 free(sd->sd_vol.sv_chunks, M_DEVBUF, 0); 2974 free(sd, M_DEVBUF, 0); 2975 free(sm, M_DEVBUF, 0); 2976 if (open) { 2977 VOP_CLOSE(vn, FREAD | FWRITE, NOCRED, curproc); 2978 vput(vn); 2979 } 2980 2981 return (rv); 2982 } 2983 2984 void 2985 sr_hotspare_rebuild_callback(void *xsd) 2986 { 2987 struct sr_discipline *sd = xsd; 2988 sr_hotspare_rebuild(sd); 2989 } 2990 2991 void 2992 sr_hotspare_rebuild(struct sr_discipline *sd) 2993 { 2994 struct sr_softc *sc = sd->sd_sc; 2995 struct sr_chunk_head *cl; 2996 struct sr_chunk *hotspare, *chunk = NULL; 2997 struct sr_workunit *wu; 2998 struct sr_ccb *ccb; 2999 int i, s, cid, busy; 3000 3001 /* 3002 * Attempt to locate a hotspare and initiate rebuild. 3003 */ 3004 3005 /* Find first offline chunk. */ 3006 for (cid = 0; cid < sd->sd_meta->ssdi.ssd_chunk_no; cid++) { 3007 if (sd->sd_vol.sv_chunks[cid]->src_meta.scm_status == 3008 BIOC_SDOFFLINE) { 3009 chunk = sd->sd_vol.sv_chunks[cid]; 3010 break; 3011 } 3012 } 3013 if (chunk == NULL) { 3014 printf("%s: no offline chunk found on %s!\n", 3015 DEVNAME(sc), sd->sd_meta->ssd_devname); 3016 return; 3017 } 3018 3019 /* See if we have a suitable hotspare... */ 3020 rw_enter_write(&sc->sc_hs_lock); 3021 cl = &sc->sc_hotspare_list; 3022 SLIST_FOREACH(hotspare, cl, src_link) 3023 if (hotspare->src_size >= chunk->src_size && 3024 hotspare->src_secsize <= sd->sd_meta->ssdi.ssd_secsize) 3025 break; 3026 3027 if (hotspare != NULL) { 3028 3029 printf("%s: %s volume degraded, will attempt to " 3030 "rebuild on hotspare %s\n", DEVNAME(sc), 3031 sd->sd_meta->ssd_devname, hotspare->src_devname); 3032 3033 /* 3034 * Ensure that all pending I/O completes on the failed chunk 3035 * before trying to initiate a rebuild. 3036 */ 3037 i = 0; 3038 do { 3039 busy = 0; 3040 3041 s = splbio(); 3042 TAILQ_FOREACH(wu, &sd->sd_wu_pendq, swu_link) { 3043 TAILQ_FOREACH(ccb, &wu->swu_ccb, ccb_link) { 3044 if (ccb->ccb_target == cid) 3045 busy = 1; 3046 } 3047 } 3048 TAILQ_FOREACH(wu, &sd->sd_wu_defq, swu_link) { 3049 TAILQ_FOREACH(ccb, &wu->swu_ccb, ccb_link) { 3050 if (ccb->ccb_target == cid) 3051 busy = 1; 3052 } 3053 } 3054 splx(s); 3055 3056 if (busy) { 3057 tsleep(sd, PRIBIO, "sr_hotspare", hz); 3058 i++; 3059 } 3060 3061 } while (busy && i < 120); 3062 3063 DNPRINTF(SR_D_META, "%s: waited %i seconds for I/O to " 3064 "complete on failed chunk %s\n", DEVNAME(sc), 3065 i, chunk->src_devname); 3066 3067 if (busy) { 3068 printf("%s: pending I/O failed to complete on " 3069 "failed chunk %s, hotspare rebuild aborted...\n", 3070 DEVNAME(sc), chunk->src_devname); 3071 goto done; 3072 } 3073 3074 s = splbio(); 3075 rw_enter_write(&sc->sc_lock); 3076 bio_status_init(&sc->sc_status, &sc->sc_dev); 3077 if (sr_rebuild_init(sd, hotspare->src_dev_mm, 1) == 0) { 3078 3079 /* Remove hotspare from available list. */ 3080 sc->sc_hotspare_no--; 3081 SLIST_REMOVE(cl, hotspare, sr_chunk, src_link); 3082 free(hotspare, M_DEVBUF, 0); 3083 3084 } 3085 rw_exit_write(&sc->sc_lock); 3086 splx(s); 3087 } 3088 done: 3089 rw_exit_write(&sc->sc_hs_lock); 3090 } 3091 3092 int 3093 sr_rebuild_init(struct sr_discipline *sd, dev_t dev, int hotspare) 3094 { 3095 struct sr_softc *sc = sd->sd_sc; 3096 struct sr_chunk *chunk = NULL; 3097 struct sr_meta_chunk *meta; 3098 struct disklabel label; 3099 struct vnode *vn; 3100 u_int64_t size; 3101 int64_t csize; 3102 char devname[32]; 3103 int rv = EINVAL, open = 0; 3104 int cid, i, part, status; 3105 3106 /* 3107 * Attempt to initiate a rebuild onto the specified device. 3108 */ 3109 3110 if (!(sd->sd_capabilities & SR_CAP_REBUILD)) { 3111 sr_error(sc, "discipline does not support rebuild"); 3112 goto done; 3113 } 3114 3115 /* make sure volume is in the right state */ 3116 if (sd->sd_vol_status == BIOC_SVREBUILD) { 3117 sr_error(sc, "rebuild already in progress"); 3118 goto done; 3119 } 3120 if (sd->sd_vol_status != BIOC_SVDEGRADED) { 3121 sr_error(sc, "volume not degraded"); 3122 goto done; 3123 } 3124 3125 /* Find first offline chunk. */ 3126 for (cid = 0; cid < sd->sd_meta->ssdi.ssd_chunk_no; cid++) { 3127 if (sd->sd_vol.sv_chunks[cid]->src_meta.scm_status == 3128 BIOC_SDOFFLINE) { 3129 chunk = sd->sd_vol.sv_chunks[cid]; 3130 break; 3131 } 3132 } 3133 if (chunk == NULL) { 3134 sr_error(sc, "no offline chunks available to rebuild"); 3135 goto done; 3136 } 3137 3138 /* Get coerced size from another online chunk. */ 3139 csize = 0; 3140 for (i = 0; i < sd->sd_meta->ssdi.ssd_chunk_no; i++) { 3141 if (sd->sd_vol.sv_chunks[i]->src_meta.scm_status == 3142 BIOC_SDONLINE) { 3143 meta = &sd->sd_vol.sv_chunks[i]->src_meta; 3144 csize = meta->scmi.scm_coerced_size; 3145 break; 3146 } 3147 } 3148 if (csize == 0) { 3149 sr_error(sc, "no online chunks available for rebuild"); 3150 goto done; 3151 } 3152 3153 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 3154 if (bdevvp(dev, &vn)) { 3155 printf("%s: sr_rebuild_init: can't allocate vnode\n", 3156 DEVNAME(sc)); 3157 goto done; 3158 } 3159 if (VOP_OPEN(vn, FREAD | FWRITE, NOCRED, curproc)) { 3160 DNPRINTF(SR_D_META,"%s: sr_ioctl_setstate can't " 3161 "open %s\n", DEVNAME(sc), devname); 3162 vput(vn); 3163 goto done; 3164 } 3165 open = 1; /* close dev on error */ 3166 3167 /* Get disklabel and check partition. */ 3168 part = DISKPART(dev); 3169 if (VOP_IOCTL(vn, DIOCGDINFO, (caddr_t)&label, FREAD, 3170 NOCRED, curproc)) { 3171 DNPRINTF(SR_D_META, "%s: sr_ioctl_setstate ioctl failed\n", 3172 DEVNAME(sc)); 3173 goto done; 3174 } 3175 if (label.d_partitions[part].p_fstype != FS_RAID) { 3176 sr_error(sc, "%s partition not of type RAID (%d)", 3177 devname, label.d_partitions[part].p_fstype); 3178 goto done; 3179 } 3180 3181 /* Is the partition large enough? */ 3182 size = DL_SECTOBLK(&label, DL_GETPSIZE(&label.d_partitions[part])); 3183 if (size <= sd->sd_meta->ssd_data_blkno) { 3184 sr_error(sc, "%s: %s partition too small", DEVNAME(sc), 3185 devname); 3186 goto done; 3187 } 3188 size -= sd->sd_meta->ssd_data_blkno; 3189 if (size > INT64_MAX) { 3190 sr_error(sc, "%s: %s partition too large", DEVNAME(sc), 3191 devname); 3192 goto done; 3193 } 3194 if (size < csize) { 3195 sr_error(sc, "%s partition too small, at least %lld bytes " 3196 "required", devname, (long long)(csize << DEV_BSHIFT)); 3197 goto done; 3198 } else if (size > csize) 3199 sr_warn(sc, "%s partition too large, wasting %lld bytes", 3200 devname, (long long)((size - csize) << DEV_BSHIFT)); 3201 if (label.d_secsize > sd->sd_meta->ssdi.ssd_secsize) { 3202 sr_error(sc, "%s sector size too large, <= %u bytes " 3203 "required", devname, sd->sd_meta->ssdi.ssd_secsize); 3204 goto done; 3205 } 3206 3207 /* Ensure that this chunk is not already in use. */ 3208 status = sr_chunk_in_use(sc, dev); 3209 if (status != BIOC_SDINVALID && status != BIOC_SDOFFLINE && 3210 !(hotspare && status == BIOC_SDHOTSPARE)) { 3211 sr_error(sc, "%s is already in use", devname); 3212 goto done; 3213 } 3214 3215 /* Reset rebuild counter since we rebuilding onto a new chunk. */ 3216 sd->sd_meta->ssd_rebuild = 0; 3217 3218 open = 0; /* leave dev open from here on out */ 3219 3220 /* Fix up chunk. */ 3221 memcpy(chunk->src_duid, label.d_uid, sizeof(chunk->src_duid)); 3222 chunk->src_dev_mm = dev; 3223 chunk->src_vn = vn; 3224 3225 /* Reconstruct metadata. */ 3226 meta = &chunk->src_meta; 3227 meta->scmi.scm_volid = sd->sd_meta->ssdi.ssd_volid; 3228 meta->scmi.scm_chunk_id = cid; 3229 strlcpy(meta->scmi.scm_devname, devname, 3230 sizeof(meta->scmi.scm_devname)); 3231 meta->scmi.scm_size = size; 3232 meta->scmi.scm_coerced_size = csize; 3233 memcpy(&meta->scmi.scm_uuid, &sd->sd_meta->ssdi.ssd_uuid, 3234 sizeof(meta->scmi.scm_uuid)); 3235 sr_checksum(sc, meta, &meta->scm_checksum, 3236 sizeof(struct sr_meta_chunk_invariant)); 3237 3238 sd->sd_set_chunk_state(sd, cid, BIOC_SDREBUILD); 3239 3240 if (sr_meta_save(sd, SR_META_DIRTY)) { 3241 sr_error(sc, "could not save metadata to %s", devname); 3242 open = 1; 3243 goto done; 3244 } 3245 3246 sr_warn(sc, "rebuild of %s started on %s", 3247 sd->sd_meta->ssd_devname, devname); 3248 3249 sd->sd_reb_abort = 0; 3250 kthread_create_deferred(sr_rebuild_start, sd); 3251 3252 rv = 0; 3253 done: 3254 if (open) { 3255 VOP_CLOSE(vn, FREAD | FWRITE, NOCRED, curproc); 3256 vput(vn); 3257 } 3258 3259 return (rv); 3260 } 3261 3262 int 3263 sr_rebuild_percent(struct sr_discipline *sd) 3264 { 3265 daddr_t rb, sz; 3266 3267 sz = sd->sd_meta->ssdi.ssd_size; 3268 rb = sd->sd_meta->ssd_rebuild; 3269 3270 if (rb > 0) 3271 return (100 - ((sz * 100 - rb * 100) / sz) - 1); 3272 3273 return (0); 3274 } 3275 3276 void 3277 sr_roam_chunks(struct sr_discipline *sd) 3278 { 3279 struct sr_softc *sc = sd->sd_sc; 3280 struct sr_chunk *chunk; 3281 struct sr_meta_chunk *meta; 3282 int roamed = 0; 3283 3284 /* Have any chunks roamed? */ 3285 SLIST_FOREACH(chunk, &sd->sd_vol.sv_chunk_list, src_link) { 3286 meta = &chunk->src_meta; 3287 if (strncmp(meta->scmi.scm_devname, chunk->src_devname, 3288 sizeof(meta->scmi.scm_devname))) { 3289 3290 printf("%s: roaming device %s -> %s\n", DEVNAME(sc), 3291 meta->scmi.scm_devname, chunk->src_devname); 3292 3293 strlcpy(meta->scmi.scm_devname, chunk->src_devname, 3294 sizeof(meta->scmi.scm_devname)); 3295 3296 roamed++; 3297 } 3298 } 3299 3300 if (roamed) 3301 sr_meta_save(sd, SR_META_DIRTY); 3302 } 3303 3304 int 3305 sr_ioctl_createraid(struct sr_softc *sc, struct bioc_createraid *bc, 3306 int user, void *data) 3307 { 3308 struct sr_meta_opt_item *omi; 3309 struct sr_chunk_head *cl; 3310 struct sr_discipline *sd = NULL; 3311 struct sr_chunk *ch_entry; 3312 struct scsi_link *link; 3313 struct device *dev; 3314 char *uuid, devname[32]; 3315 dev_t *dt = NULL; 3316 int i, no_chunk, rv = EINVAL, target, vol; 3317 int no_meta; 3318 3319 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_createraid(%d)\n", 3320 DEVNAME(sc), user); 3321 3322 /* user input */ 3323 if (bc->bc_dev_list_len > BIOC_CRMAXLEN) 3324 goto unwind; 3325 3326 dt = malloc(bc->bc_dev_list_len, M_DEVBUF, M_WAITOK | M_ZERO); 3327 if (user) { 3328 if (copyin(bc->bc_dev_list, dt, bc->bc_dev_list_len) != 0) 3329 goto unwind; 3330 } else 3331 memcpy(dt, bc->bc_dev_list, bc->bc_dev_list_len); 3332 3333 /* Initialise discipline. */ 3334 sd = malloc(sizeof(struct sr_discipline), M_DEVBUF, M_WAITOK | M_ZERO); 3335 sd->sd_sc = sc; 3336 SLIST_INIT(&sd->sd_meta_opt); 3337 sd->sd_taskq = taskq_create("srdis", 1, IPL_BIO, 0); 3338 if (sd->sd_taskq == NULL) { 3339 sr_error(sc, "could not create discipline taskq"); 3340 goto unwind; 3341 } 3342 if (sr_discipline_init(sd, bc->bc_level)) { 3343 sr_error(sc, "could not initialize discipline"); 3344 goto unwind; 3345 } 3346 3347 no_chunk = bc->bc_dev_list_len / sizeof(dev_t); 3348 cl = &sd->sd_vol.sv_chunk_list; 3349 SLIST_INIT(cl); 3350 3351 /* Ensure that chunks are not already in use. */ 3352 for (i = 0; i < no_chunk; i++) { 3353 if (sr_chunk_in_use(sc, dt[i]) != BIOC_SDINVALID) { 3354 sr_meta_getdevname(sc, dt[i], devname, sizeof(devname)); 3355 sr_error(sc, "chunk %s already in use", devname); 3356 goto unwind; 3357 } 3358 } 3359 3360 sd->sd_meta_type = sr_meta_probe(sd, dt, no_chunk); 3361 if (sd->sd_meta_type == SR_META_F_INVALID) { 3362 sr_error(sc, "invalid metadata format"); 3363 goto unwind; 3364 } 3365 3366 if (sr_meta_attach(sd, no_chunk, bc->bc_flags & BIOC_SCFORCE)) 3367 goto unwind; 3368 3369 /* force the raid volume by clearing metadata region */ 3370 if (bc->bc_flags & BIOC_SCFORCE) { 3371 /* make sure disk isn't up and running */ 3372 if (sr_meta_read(sd)) 3373 if (sr_already_assembled(sd)) { 3374 uuid = sr_uuid_format( 3375 &sd->sd_meta->ssdi.ssd_uuid); 3376 sr_error(sc, "disk %s is currently in use; " 3377 "cannot force create", uuid); 3378 free(uuid, M_DEVBUF, 0); 3379 goto unwind; 3380 } 3381 3382 if (sr_meta_clear(sd)) { 3383 sr_error(sc, "failed to clear metadata"); 3384 goto unwind; 3385 } 3386 } 3387 3388 no_meta = sr_meta_read(sd); 3389 if (no_meta == -1) { 3390 3391 /* Corrupt metadata on one or more chunks. */ 3392 sr_error(sc, "one of the chunks has corrupt metadata; " 3393 "aborting assembly"); 3394 goto unwind; 3395 3396 } else if (no_meta == 0) { 3397 3398 /* Initialise volume and chunk metadata. */ 3399 sr_meta_init(sd, bc->bc_level, no_chunk); 3400 sd->sd_vol_status = BIOC_SVONLINE; 3401 sd->sd_meta_flags = bc->bc_flags & BIOC_SCNOAUTOASSEMBLE; 3402 if (sd->sd_create) { 3403 if ((i = sd->sd_create(sd, bc, no_chunk, 3404 sd->sd_vol.sv_chunk_minsz))) { 3405 rv = i; 3406 goto unwind; 3407 } 3408 } 3409 sr_meta_init_complete(sd); 3410 3411 DNPRINTF(SR_D_IOCTL, 3412 "%s: sr_ioctl_createraid: vol_size: %lld\n", 3413 DEVNAME(sc), sd->sd_meta->ssdi.ssd_size); 3414 3415 /* Warn if we've wasted chunk space due to coercing. */ 3416 if ((sd->sd_capabilities & SR_CAP_NON_COERCED) == 0 && 3417 sd->sd_vol.sv_chunk_minsz != sd->sd_vol.sv_chunk_maxsz) 3418 sr_warn(sc, "chunk sizes are not equal; up to %llu " 3419 "blocks wasted per chunk", 3420 sd->sd_vol.sv_chunk_maxsz - 3421 sd->sd_vol.sv_chunk_minsz); 3422 3423 } else { 3424 3425 /* Ensure we are assembling the correct # of chunks. */ 3426 if (sd->sd_meta->ssdi.ssd_chunk_no != no_chunk) { 3427 sr_error(sc, "volume chunk count does not match metadata " 3428 "chunk count"); 3429 goto unwind; 3430 } 3431 3432 /* Ensure metadata level matches requested assembly level. */ 3433 if (sd->sd_meta->ssdi.ssd_level != bc->bc_level) { 3434 sr_error(sc, "volume level does not match metadata " 3435 "level"); 3436 goto unwind; 3437 } 3438 3439 if (sr_already_assembled(sd)) { 3440 uuid = sr_uuid_format(&sd->sd_meta->ssdi.ssd_uuid); 3441 sr_error(sc, "disk %s already assembled", uuid); 3442 free(uuid, M_DEVBUF, 0); 3443 goto unwind; 3444 } 3445 3446 if (user == 0 && sd->sd_meta_flags & BIOC_SCNOAUTOASSEMBLE) { 3447 DNPRINTF(SR_D_META, "%s: disk not auto assembled from " 3448 "metadata\n", DEVNAME(sc)); 3449 goto unwind; 3450 } 3451 3452 if (no_meta != no_chunk) 3453 sr_warn(sc, "trying to bring up %s degraded", 3454 sd->sd_meta->ssd_devname); 3455 3456 if (sd->sd_meta->ssd_meta_flags & SR_META_DIRTY) 3457 sr_warn(sc, "%s was not shutdown properly", 3458 sd->sd_meta->ssd_devname); 3459 3460 SLIST_FOREACH(omi, &sd->sd_meta_opt, omi_link) 3461 if (sd->sd_meta_opt_handler == NULL || 3462 sd->sd_meta_opt_handler(sd, omi->omi_som) != 0) 3463 sr_meta_opt_handler(sd, omi->omi_som); 3464 3465 if (sd->sd_assemble) { 3466 if ((i = sd->sd_assemble(sd, bc, no_chunk, data))) { 3467 rv = i; 3468 goto unwind; 3469 } 3470 } 3471 3472 DNPRINTF(SR_D_META, "%s: disk assembled from metadata\n", 3473 DEVNAME(sc)); 3474 3475 } 3476 3477 /* Metadata MUST be fully populated by this point. */ 3478 TAILQ_INSERT_TAIL(&sc->sc_dis_list, sd, sd_link); 3479 3480 /* Allocate all resources. */ 3481 if ((rv = sd->sd_alloc_resources(sd))) 3482 goto unwind; 3483 3484 /* Adjust flags if necessary. */ 3485 if ((sd->sd_capabilities & SR_CAP_AUTO_ASSEMBLE) && 3486 (bc->bc_flags & BIOC_SCNOAUTOASSEMBLE) != 3487 (sd->sd_meta->ssdi.ssd_vol_flags & BIOC_SCNOAUTOASSEMBLE)) { 3488 sd->sd_meta->ssdi.ssd_vol_flags &= ~BIOC_SCNOAUTOASSEMBLE; 3489 sd->sd_meta->ssdi.ssd_vol_flags |= 3490 bc->bc_flags & BIOC_SCNOAUTOASSEMBLE; 3491 } 3492 3493 if (sd->sd_capabilities & SR_CAP_SYSTEM_DISK) { 3494 /* Initialise volume state. */ 3495 sd->sd_set_vol_state(sd); 3496 if (sd->sd_vol_status == BIOC_SVOFFLINE) { 3497 sr_error(sc, "%s is offline, will not be brought " 3498 "online", sd->sd_meta->ssd_devname); 3499 goto unwind; 3500 } 3501 3502 /* Setup SCSI iopool. */ 3503 scsi_iopool_init(&sd->sd_iopool, sd, sr_wu_get, sr_wu_put); 3504 3505 /* 3506 * All checks passed - return ENXIO if volume cannot be created. 3507 */ 3508 rv = ENXIO; 3509 3510 /* 3511 * Find a free target. 3512 * 3513 * XXX: We reserve sd_target == 0 to indicate the 3514 * discipline is not linked into sc->sc_targets, so begin 3515 * the search with target = 1. 3516 */ 3517 for (target = 1; target < SR_MAX_LD; target++) 3518 if (sc->sc_targets[target] == NULL) 3519 break; 3520 if (target == SR_MAX_LD) { 3521 sr_error(sc, "no free target for %s", 3522 sd->sd_meta->ssd_devname); 3523 goto unwind; 3524 } 3525 3526 /* Clear sense data. */ 3527 bzero(&sd->sd_scsi_sense, sizeof(sd->sd_scsi_sense)); 3528 3529 /* Attach discipline and get midlayer to probe it. */ 3530 sd->sd_target = target; 3531 sc->sc_targets[target] = sd; 3532 if (scsi_probe_lun(sc->sc_scsibus, target, 0) != 0) { 3533 sr_error(sc, "scsi_probe_lun failed"); 3534 sc->sc_targets[target] = NULL; 3535 sd->sd_target = 0; 3536 goto unwind; 3537 } 3538 3539 link = scsi_get_link(sc->sc_scsibus, target, 0); 3540 if (link == NULL) 3541 goto unwind; 3542 3543 dev = link->device_softc; 3544 DNPRINTF(SR_D_IOCTL, "%s: sr device added: %s at target %d\n", 3545 DEVNAME(sc), dev->dv_xname, sd->sd_target); 3546 3547 /* XXX - Count volumes, not targets. */ 3548 for (i = 0, vol = -1; i <= sd->sd_target; i++) 3549 if (sc->sc_targets[i]) 3550 vol++; 3551 3552 rv = 0; 3553 3554 if (sd->sd_meta->ssd_devname[0] != '\0' && 3555 strncmp(sd->sd_meta->ssd_devname, dev->dv_xname, 3556 sizeof(dev->dv_xname))) 3557 sr_warn(sc, "volume %s is roaming, it used to be %s, " 3558 "updating metadata", dev->dv_xname, 3559 sd->sd_meta->ssd_devname); 3560 3561 /* Populate remaining volume metadata. */ 3562 sd->sd_meta->ssdi.ssd_volid = vol; 3563 strlcpy(sd->sd_meta->ssd_devname, dev->dv_xname, 3564 sizeof(sd->sd_meta->ssd_devname)); 3565 3566 sr_info(sc, "%s volume attached as %s", 3567 sd->sd_name, sd->sd_meta->ssd_devname); 3568 3569 /* Update device name on any roaming chunks. */ 3570 sr_roam_chunks(sd); 3571 3572 #ifndef SMALL_KERNEL 3573 if (sr_sensors_create(sd)) 3574 sr_warn(sc, "unable to create sensor for %s", 3575 dev->dv_xname); 3576 #endif /* SMALL_KERNEL */ 3577 } else { 3578 /* This volume does not attach as a system disk. */ 3579 ch_entry = SLIST_FIRST(cl); /* XXX */ 3580 strlcpy(sd->sd_meta->ssd_devname, ch_entry->src_devname, 3581 sizeof(sd->sd_meta->ssd_devname)); 3582 3583 if (sd->sd_start_discipline(sd)) 3584 goto unwind; 3585 } 3586 3587 /* Save current metadata to disk. */ 3588 rv = sr_meta_save(sd, SR_META_DIRTY); 3589 3590 if (sd->sd_vol_status == BIOC_SVREBUILD) 3591 kthread_create_deferred(sr_rebuild_start, sd); 3592 3593 sd->sd_ready = 1; 3594 3595 free(dt, M_DEVBUF, bc->bc_dev_list_len); 3596 3597 return (rv); 3598 3599 unwind: 3600 free(dt, M_DEVBUF, bc->bc_dev_list_len); 3601 3602 sr_discipline_shutdown(sd, 0, 0); 3603 3604 if (rv == EAGAIN) 3605 rv = 0; 3606 3607 return (rv); 3608 } 3609 3610 int 3611 sr_ioctl_deleteraid(struct sr_softc *sc, struct sr_discipline *sd, 3612 struct bioc_deleteraid *bd) 3613 { 3614 int rv = 1; 3615 3616 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_deleteraid %s\n", 3617 DEVNAME(sc), bd->bd_dev); 3618 3619 if (sd == NULL) { 3620 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 3621 if (!strncmp(sd->sd_meta->ssd_devname, bd->bd_dev, 3622 sizeof(sd->sd_meta->ssd_devname))) 3623 break; 3624 } 3625 if (sd == NULL) { 3626 sr_error(sc, "volume %s not found", bd->bd_dev); 3627 goto bad; 3628 } 3629 } 3630 3631 sd->sd_deleted = 1; 3632 sd->sd_meta->ssdi.ssd_vol_flags = BIOC_SCNOAUTOASSEMBLE; 3633 sr_discipline_shutdown(sd, 1, 0); 3634 3635 rv = 0; 3636 bad: 3637 return (rv); 3638 } 3639 3640 int 3641 sr_ioctl_discipline(struct sr_softc *sc, struct sr_discipline *sd, 3642 struct bioc_discipline *bd) 3643 { 3644 int rv = 1; 3645 3646 /* Dispatch a discipline specific ioctl. */ 3647 3648 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_discipline %s\n", DEVNAME(sc), 3649 bd->bd_dev); 3650 3651 if (sd == NULL) { 3652 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 3653 if (!strncmp(sd->sd_meta->ssd_devname, bd->bd_dev, 3654 sizeof(sd->sd_meta->ssd_devname))) 3655 break; 3656 } 3657 if (sd == NULL) { 3658 sr_error(sc, "volume %s not found", bd->bd_dev); 3659 goto bad; 3660 } 3661 } 3662 3663 if (sd->sd_ioctl_handler) 3664 rv = sd->sd_ioctl_handler(sd, bd); 3665 3666 bad: 3667 return (rv); 3668 } 3669 3670 int 3671 sr_ioctl_installboot(struct sr_softc *sc, struct sr_discipline *sd, 3672 struct bioc_installboot *bb) 3673 { 3674 void *bootblk = NULL, *bootldr = NULL; 3675 struct sr_chunk *chunk; 3676 struct sr_meta_opt_item *omi; 3677 struct sr_meta_boot *sbm; 3678 struct disk *dk; 3679 u_int32_t bbs, bls, secsize; 3680 u_char duid[8]; 3681 int rv = EINVAL; 3682 int i; 3683 3684 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_installboot %s\n", DEVNAME(sc), 3685 bb->bb_dev); 3686 3687 if (sd == NULL) { 3688 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 3689 if (!strncmp(sd->sd_meta->ssd_devname, bb->bb_dev, 3690 sizeof(sd->sd_meta->ssd_devname))) 3691 break; 3692 } 3693 if (sd == NULL) { 3694 sr_error(sc, "volume %s not found", bb->bb_dev); 3695 goto done; 3696 } 3697 } 3698 3699 bzero(duid, sizeof(duid)); 3700 TAILQ_FOREACH(dk, &disklist, dk_link) 3701 if (!strncmp(dk->dk_name, bb->bb_dev, sizeof(bb->bb_dev))) 3702 break; 3703 if (dk == NULL || dk->dk_label == NULL || 3704 (dk->dk_flags & DKF_LABELVALID) == 0 || 3705 bcmp(dk->dk_label->d_uid, &duid, sizeof(duid)) == 0) { 3706 sr_error(sc, "failed to get DUID for softraid volume"); 3707 goto done; 3708 } 3709 memcpy(duid, dk->dk_label->d_uid, sizeof(duid)); 3710 3711 /* Ensure that boot storage area is large enough. */ 3712 if (sd->sd_meta->ssd_data_blkno < (SR_BOOT_OFFSET + SR_BOOT_SIZE)) { 3713 sr_error(sc, "insufficient boot storage"); 3714 goto done; 3715 } 3716 3717 if (bb->bb_bootblk_size > SR_BOOT_BLOCKS_SIZE * DEV_BSIZE) 3718 goto done; 3719 3720 if (bb->bb_bootldr_size > SR_BOOT_LOADER_SIZE * DEV_BSIZE) 3721 goto done; 3722 3723 secsize = sd->sd_meta->ssdi.ssd_secsize; 3724 3725 /* Copy in boot block. */ 3726 bbs = howmany(bb->bb_bootblk_size, secsize) * secsize; 3727 bootblk = malloc(bbs, M_DEVBUF, M_WAITOK | M_ZERO); 3728 if (copyin(bb->bb_bootblk, bootblk, bb->bb_bootblk_size) != 0) 3729 goto done; 3730 3731 /* Copy in boot loader. */ 3732 bls = howmany(bb->bb_bootldr_size, secsize) * secsize; 3733 bootldr = malloc(bls, M_DEVBUF, M_WAITOK | M_ZERO); 3734 if (copyin(bb->bb_bootldr, bootldr, bb->bb_bootldr_size) != 0) 3735 goto done; 3736 3737 /* Create or update optional meta for bootable volumes. */ 3738 SLIST_FOREACH(omi, &sd->sd_meta_opt, omi_link) 3739 if (omi->omi_som->som_type == SR_OPT_BOOT) 3740 break; 3741 if (omi == NULL) { 3742 omi = malloc(sizeof(struct sr_meta_opt_item), M_DEVBUF, 3743 M_WAITOK | M_ZERO); 3744 omi->omi_som = malloc(sizeof(struct sr_meta_crypto), M_DEVBUF, 3745 M_WAITOK | M_ZERO); 3746 omi->omi_som->som_type = SR_OPT_BOOT; 3747 omi->omi_som->som_length = sizeof(struct sr_meta_boot); 3748 SLIST_INSERT_HEAD(&sd->sd_meta_opt, omi, omi_link); 3749 sd->sd_meta->ssdi.ssd_opt_no++; 3750 } 3751 sbm = (struct sr_meta_boot *)omi->omi_som; 3752 3753 memcpy(sbm->sbm_root_duid, duid, sizeof(sbm->sbm_root_duid)); 3754 bzero(&sbm->sbm_boot_duid, sizeof(sbm->sbm_boot_duid)); 3755 sbm->sbm_bootblk_size = bbs; 3756 sbm->sbm_bootldr_size = bls; 3757 3758 DNPRINTF(SR_D_IOCTL, "sr_ioctl_installboot: root duid is %s\n", 3759 duid_format(sbm->sbm_root_duid)); 3760 3761 /* Save boot block and boot loader to each chunk. */ 3762 for (i = 0; i < sd->sd_meta->ssdi.ssd_chunk_no; i++) { 3763 3764 chunk = sd->sd_vol.sv_chunks[i]; 3765 if (chunk->src_meta.scm_status != BIOC_SDONLINE && 3766 chunk->src_meta.scm_status != BIOC_SDREBUILD) 3767 continue; 3768 3769 if (i < SR_MAX_BOOT_DISKS) 3770 memcpy(&sbm->sbm_boot_duid[i], chunk->src_duid, 3771 sizeof(sbm->sbm_boot_duid[i])); 3772 3773 /* Save boot blocks. */ 3774 DNPRINTF(SR_D_IOCTL, 3775 "sr_ioctl_installboot: saving boot block to %s " 3776 "(%u bytes)\n", chunk->src_devname, bbs); 3777 3778 if (sr_rw(sc, chunk->src_dev_mm, bootblk, bbs, 3779 SR_BOOT_BLOCKS_OFFSET, B_WRITE)) { 3780 sr_error(sc, "failed to write boot block", DEVNAME(sc)); 3781 goto done; 3782 } 3783 3784 /* Save boot loader.*/ 3785 DNPRINTF(SR_D_IOCTL, 3786 "sr_ioctl_installboot: saving boot loader to %s " 3787 "(%u bytes)\n", chunk->src_devname, bls); 3788 3789 if (sr_rw(sc, chunk->src_dev_mm, bootldr, bls, 3790 SR_BOOT_LOADER_OFFSET, B_WRITE)) { 3791 sr_error(sc, "failed to write boot loader"); 3792 goto done; 3793 } 3794 } 3795 3796 /* XXX - Install boot block on disk - MD code. */ 3797 3798 /* Mark volume as bootable and save metadata. */ 3799 sd->sd_meta->ssdi.ssd_vol_flags |= BIOC_SCBOOTABLE; 3800 if (sr_meta_save(sd, SR_META_DIRTY)) { 3801 sr_error(sc, "could not save metadata to %s", DEVNAME(sc)); 3802 goto done; 3803 } 3804 3805 rv = 0; 3806 3807 done: 3808 free(bootblk, M_DEVBUF, 0); 3809 free(bootldr, M_DEVBUF, 0); 3810 3811 return (rv); 3812 } 3813 3814 void 3815 sr_chunks_unwind(struct sr_softc *sc, struct sr_chunk_head *cl) 3816 { 3817 struct sr_chunk *ch_entry, *ch_next; 3818 3819 DNPRINTF(SR_D_IOCTL, "%s: sr_chunks_unwind\n", DEVNAME(sc)); 3820 3821 if (!cl) 3822 return; 3823 3824 for (ch_entry = SLIST_FIRST(cl); ch_entry != NULL; ch_entry = ch_next) { 3825 ch_next = SLIST_NEXT(ch_entry, src_link); 3826 3827 DNPRINTF(SR_D_IOCTL, "%s: sr_chunks_unwind closing: %s\n", 3828 DEVNAME(sc), ch_entry->src_devname); 3829 if (ch_entry->src_vn) { 3830 /* 3831 * XXX - explicitly lock the vnode until we can resolve 3832 * the problem introduced by vnode aliasing... specfs 3833 * has no locking, whereas ufs/ffs does! 3834 */ 3835 vn_lock(ch_entry->src_vn, LK_EXCLUSIVE | 3836 LK_RETRY, curproc); 3837 VOP_CLOSE(ch_entry->src_vn, FREAD | FWRITE, NOCRED, 3838 curproc); 3839 vput(ch_entry->src_vn); 3840 } 3841 free(ch_entry, M_DEVBUF, 0); 3842 } 3843 SLIST_INIT(cl); 3844 } 3845 3846 void 3847 sr_discipline_free(struct sr_discipline *sd) 3848 { 3849 struct sr_softc *sc; 3850 struct sr_discipline *sdtmp1; 3851 struct sr_meta_opt_head *som; 3852 struct sr_meta_opt_item *omi, *omi_next; 3853 3854 if (!sd) 3855 return; 3856 3857 sc = sd->sd_sc; 3858 3859 DNPRINTF(SR_D_DIS, "%s: sr_discipline_free %s\n", 3860 DEVNAME(sc), 3861 sd->sd_meta ? sd->sd_meta->ssd_devname : "nodev"); 3862 if (sd->sd_free_resources) 3863 sd->sd_free_resources(sd); 3864 free(sd->sd_vol.sv_chunks, M_DEVBUF, 0); 3865 free(sd->sd_meta, M_DEVBUF, 0); 3866 free(sd->sd_meta_foreign, M_DEVBUF, 0); 3867 3868 som = &sd->sd_meta_opt; 3869 for (omi = SLIST_FIRST(som); omi != NULL; omi = omi_next) { 3870 omi_next = SLIST_NEXT(omi, omi_link); 3871 free(omi->omi_som, M_DEVBUF, 0); 3872 free(omi, M_DEVBUF, 0); 3873 } 3874 3875 if (sd->sd_target != 0) { 3876 KASSERT(sc->sc_targets[sd->sd_target] == sd); 3877 sc->sc_targets[sd->sd_target] = NULL; 3878 } 3879 3880 TAILQ_FOREACH(sdtmp1, &sc->sc_dis_list, sd_link) { 3881 if (sdtmp1 == sd) 3882 break; 3883 } 3884 if (sdtmp1 != NULL) 3885 TAILQ_REMOVE(&sc->sc_dis_list, sd, sd_link); 3886 3887 explicit_bzero(sd, sizeof *sd); 3888 free(sd, M_DEVBUF, 0); 3889 } 3890 3891 void 3892 sr_discipline_shutdown(struct sr_discipline *sd, int meta_save, int dying) 3893 { 3894 struct sr_softc *sc; 3895 int s; 3896 3897 if (!sd) 3898 return; 3899 sc = sd->sd_sc; 3900 3901 DNPRINTF(SR_D_DIS, "%s: sr_discipline_shutdown %s\n", DEVNAME(sc), 3902 sd->sd_meta ? sd->sd_meta->ssd_devname : "nodev"); 3903 3904 /* If rebuilding, abort rebuild and drain I/O. */ 3905 if (sd->sd_reb_active) { 3906 sd->sd_reb_abort = 1; 3907 while (sd->sd_reb_active) 3908 tsleep(sd, PWAIT, "sr_shutdown", 1); 3909 } 3910 3911 if (meta_save) 3912 sr_meta_save(sd, 0); 3913 3914 s = splbio(); 3915 3916 sd->sd_ready = 0; 3917 3918 /* make sure there isn't a sync pending and yield */ 3919 wakeup(sd); 3920 while (sd->sd_sync || sd->sd_must_flush) 3921 if (tsleep(&sd->sd_sync, MAXPRI, "sr_down", 60 * hz) == 3922 EWOULDBLOCK) 3923 break; 3924 3925 if (dying == -1) { 3926 sd->sd_ready = 1; 3927 splx(s); 3928 return; 3929 } 3930 3931 #ifndef SMALL_KERNEL 3932 sr_sensors_delete(sd); 3933 #endif /* SMALL_KERNEL */ 3934 3935 if (sd->sd_target != 0) 3936 scsi_detach_lun(sc->sc_scsibus, sd->sd_target, 0, 3937 dying ? 0 : DETACH_FORCE); 3938 3939 sr_chunks_unwind(sc, &sd->sd_vol.sv_chunk_list); 3940 3941 if (sd->sd_taskq) 3942 taskq_destroy(sd->sd_taskq); 3943 3944 sr_discipline_free(sd); 3945 3946 splx(s); 3947 } 3948 3949 int 3950 sr_discipline_init(struct sr_discipline *sd, int level) 3951 { 3952 int rv = 1; 3953 3954 /* Initialise discipline function pointers with defaults. */ 3955 sd->sd_alloc_resources = sr_alloc_resources; 3956 sd->sd_assemble = NULL; 3957 sd->sd_create = NULL; 3958 sd->sd_free_resources = sr_free_resources; 3959 sd->sd_ioctl_handler = NULL; 3960 sd->sd_openings = NULL; 3961 sd->sd_meta_opt_handler = NULL; 3962 sd->sd_rebuild = sr_rebuild; 3963 sd->sd_scsi_inquiry = sr_raid_inquiry; 3964 sd->sd_scsi_read_cap = sr_raid_read_cap; 3965 sd->sd_scsi_tur = sr_raid_tur; 3966 sd->sd_scsi_req_sense = sr_raid_request_sense; 3967 sd->sd_scsi_start_stop = sr_raid_start_stop; 3968 sd->sd_scsi_sync = sr_raid_sync; 3969 sd->sd_scsi_rw = NULL; 3970 sd->sd_scsi_intr = sr_raid_intr; 3971 sd->sd_scsi_wu_done = NULL; 3972 sd->sd_scsi_done = NULL; 3973 sd->sd_set_chunk_state = sr_set_chunk_state; 3974 sd->sd_set_vol_state = sr_set_vol_state; 3975 sd->sd_start_discipline = NULL; 3976 3977 task_set(&sd->sd_meta_save_task, sr_meta_save_callback, sd); 3978 task_set(&sd->sd_hotspare_rebuild_task, sr_hotspare_rebuild_callback, 3979 sd); 3980 3981 switch (level) { 3982 case 0: 3983 sr_raid0_discipline_init(sd); 3984 break; 3985 case 1: 3986 sr_raid1_discipline_init(sd); 3987 break; 3988 case 5: 3989 sr_raid5_discipline_init(sd); 3990 break; 3991 case 6: 3992 sr_raid6_discipline_init(sd); 3993 break; 3994 #ifdef CRYPTO 3995 case 'C': 3996 sr_crypto_discipline_init(sd); 3997 break; 3998 #endif 3999 case 'c': 4000 sr_concat_discipline_init(sd); 4001 break; 4002 default: 4003 goto bad; 4004 } 4005 4006 rv = 0; 4007 bad: 4008 return (rv); 4009 } 4010 4011 int 4012 sr_raid_inquiry(struct sr_workunit *wu) 4013 { 4014 struct sr_discipline *sd = wu->swu_dis; 4015 struct scsi_xfer *xs = wu->swu_xs; 4016 struct scsi_inquiry *cdb = (struct scsi_inquiry *)xs->cmd; 4017 struct scsi_inquiry_data inq; 4018 4019 DNPRINTF(SR_D_DIS, "%s: sr_raid_inquiry\n", DEVNAME(sd->sd_sc)); 4020 4021 if (xs->cmdlen != sizeof(*cdb)) 4022 return (EINVAL); 4023 4024 if (ISSET(cdb->flags, SI_EVPD)) 4025 return (EOPNOTSUPP); 4026 4027 bzero(&inq, sizeof(inq)); 4028 inq.device = T_DIRECT; 4029 inq.dev_qual2 = 0; 4030 inq.version = 2; 4031 inq.response_format = 2; 4032 inq.additional_length = 32; 4033 inq.flags |= SID_CmdQue; 4034 strlcpy(inq.vendor, sd->sd_meta->ssdi.ssd_vendor, 4035 sizeof(inq.vendor)); 4036 strlcpy(inq.product, sd->sd_meta->ssdi.ssd_product, 4037 sizeof(inq.product)); 4038 strlcpy(inq.revision, sd->sd_meta->ssdi.ssd_revision, 4039 sizeof(inq.revision)); 4040 sr_copy_internal_data(xs, &inq, sizeof(inq)); 4041 4042 return (0); 4043 } 4044 4045 int 4046 sr_raid_read_cap(struct sr_workunit *wu) 4047 { 4048 struct sr_discipline *sd = wu->swu_dis; 4049 struct scsi_xfer *xs = wu->swu_xs; 4050 struct scsi_read_cap_data rcd; 4051 struct scsi_read_cap_data_16 rcd16; 4052 u_int64_t addr; 4053 int rv = 1; 4054 u_int32_t secsize; 4055 4056 DNPRINTF(SR_D_DIS, "%s: sr_raid_read_cap\n", DEVNAME(sd->sd_sc)); 4057 4058 secsize = sd->sd_meta->ssdi.ssd_secsize; 4059 4060 addr = ((sd->sd_meta->ssdi.ssd_size * DEV_BSIZE) / secsize) - 1; 4061 if (xs->cmd->opcode == READ_CAPACITY) { 4062 bzero(&rcd, sizeof(rcd)); 4063 if (addr > 0xffffffffllu) 4064 _lto4b(0xffffffff, rcd.addr); 4065 else 4066 _lto4b(addr, rcd.addr); 4067 _lto4b(secsize, rcd.length); 4068 sr_copy_internal_data(xs, &rcd, sizeof(rcd)); 4069 rv = 0; 4070 } else if (xs->cmd->opcode == READ_CAPACITY_16) { 4071 bzero(&rcd16, sizeof(rcd16)); 4072 _lto8b(addr, rcd16.addr); 4073 _lto4b(secsize, rcd16.length); 4074 sr_copy_internal_data(xs, &rcd16, sizeof(rcd16)); 4075 rv = 0; 4076 } 4077 4078 return (rv); 4079 } 4080 4081 int 4082 sr_raid_tur(struct sr_workunit *wu) 4083 { 4084 struct sr_discipline *sd = wu->swu_dis; 4085 4086 DNPRINTF(SR_D_DIS, "%s: sr_raid_tur\n", DEVNAME(sd->sd_sc)); 4087 4088 if (sd->sd_vol_status == BIOC_SVOFFLINE) { 4089 sd->sd_scsi_sense.error_code = SSD_ERRCODE_CURRENT; 4090 sd->sd_scsi_sense.flags = SKEY_NOT_READY; 4091 sd->sd_scsi_sense.add_sense_code = 0x04; 4092 sd->sd_scsi_sense.add_sense_code_qual = 0x11; 4093 sd->sd_scsi_sense.extra_len = 4; 4094 return (1); 4095 } else if (sd->sd_vol_status == BIOC_SVINVALID) { 4096 sd->sd_scsi_sense.error_code = SSD_ERRCODE_CURRENT; 4097 sd->sd_scsi_sense.flags = SKEY_HARDWARE_ERROR; 4098 sd->sd_scsi_sense.add_sense_code = 0x05; 4099 sd->sd_scsi_sense.add_sense_code_qual = 0x00; 4100 sd->sd_scsi_sense.extra_len = 4; 4101 return (1); 4102 } 4103 4104 return (0); 4105 } 4106 4107 int 4108 sr_raid_request_sense(struct sr_workunit *wu) 4109 { 4110 struct sr_discipline *sd = wu->swu_dis; 4111 struct scsi_xfer *xs = wu->swu_xs; 4112 4113 DNPRINTF(SR_D_DIS, "%s: sr_raid_request_sense\n", 4114 DEVNAME(sd->sd_sc)); 4115 4116 /* use latest sense data */ 4117 memcpy(&xs->sense, &sd->sd_scsi_sense, sizeof(xs->sense)); 4118 4119 /* clear sense data */ 4120 bzero(&sd->sd_scsi_sense, sizeof(sd->sd_scsi_sense)); 4121 4122 return (0); 4123 } 4124 4125 int 4126 sr_raid_start_stop(struct sr_workunit *wu) 4127 { 4128 struct scsi_xfer *xs = wu->swu_xs; 4129 struct scsi_start_stop *ss = (struct scsi_start_stop *)xs->cmd; 4130 4131 DNPRINTF(SR_D_DIS, "%s: sr_raid_start_stop\n", 4132 DEVNAME(wu->swu_dis->sd_sc)); 4133 4134 if (!ss) 4135 return (1); 4136 4137 /* 4138 * do nothing! 4139 * a softraid discipline should always reflect correct status 4140 */ 4141 return (0); 4142 } 4143 4144 int 4145 sr_raid_sync(struct sr_workunit *wu) 4146 { 4147 struct sr_discipline *sd = wu->swu_dis; 4148 int s, rv = 0, ios; 4149 4150 DNPRINTF(SR_D_DIS, "%s: sr_raid_sync\n", DEVNAME(sd->sd_sc)); 4151 4152 /* when doing a fake sync don't count the wu */ 4153 ios = (wu->swu_flags & SR_WUF_FAKE) ? 0 : 1; 4154 4155 s = splbio(); 4156 sd->sd_sync = 1; 4157 while (sd->sd_wu_pending > ios) { 4158 if (tsleep(sd, PRIBIO, "sr_sync", 15 * hz) == EWOULDBLOCK) { 4159 DNPRINTF(SR_D_DIS, "%s: sr_raid_sync timeout\n", 4160 DEVNAME(sd->sd_sc)); 4161 rv = 1; 4162 break; 4163 } 4164 } 4165 sd->sd_sync = 0; 4166 splx(s); 4167 4168 wakeup(&sd->sd_sync); 4169 4170 return (rv); 4171 } 4172 4173 void 4174 sr_raid_intr(struct buf *bp) 4175 { 4176 struct sr_ccb *ccb = (struct sr_ccb *)bp; 4177 struct sr_workunit *wu = ccb->ccb_wu; 4178 #ifdef SR_DEBUG 4179 struct sr_discipline *sd = wu->swu_dis; 4180 struct scsi_xfer *xs = wu->swu_xs; 4181 #endif 4182 int s; 4183 4184 DNPRINTF(SR_D_INTR, "%s: %s %s intr bp %p xs %p\n", 4185 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, sd->sd_name, bp, xs); 4186 4187 s = splbio(); 4188 sr_ccb_done(ccb); 4189 sr_wu_done(wu); 4190 splx(s); 4191 } 4192 4193 void 4194 sr_schedule_wu(struct sr_workunit *wu) 4195 { 4196 struct sr_discipline *sd = wu->swu_dis; 4197 struct sr_workunit *wup; 4198 int s; 4199 4200 DNPRINTF(SR_D_WU, "sr_schedule_wu: schedule wu %p state %i " 4201 "flags 0x%x\n", wu, wu->swu_state, wu->swu_flags); 4202 4203 KASSERT(wu->swu_io_count > 0); 4204 4205 s = splbio(); 4206 4207 /* Construct the work unit, do not schedule it. */ 4208 if (wu->swu_state == SR_WU_CONSTRUCT) 4209 goto queued; 4210 4211 /* Deferred work unit being reconstructed, do not start. */ 4212 if (wu->swu_state == SR_WU_REQUEUE) 4213 goto queued; 4214 4215 /* Current work unit failed, restart. */ 4216 if (wu->swu_state == SR_WU_RESTART) 4217 goto start; 4218 4219 if (wu->swu_state != SR_WU_INPROGRESS) 4220 panic("sr_schedule_wu: work unit not in progress (state %i)\n", 4221 wu->swu_state); 4222 4223 /* Walk queue backwards and fill in collider if we have one. */ 4224 TAILQ_FOREACH_REVERSE(wup, &sd->sd_wu_pendq, sr_wu_list, swu_link) { 4225 if (wu->swu_blk_end < wup->swu_blk_start || 4226 wup->swu_blk_end < wu->swu_blk_start) 4227 continue; 4228 4229 /* Defer work unit due to LBA collision. */ 4230 DNPRINTF(SR_D_WU, "sr_schedule_wu: deferring work unit %p\n", 4231 wu); 4232 wu->swu_state = SR_WU_DEFERRED; 4233 while (wup->swu_collider) 4234 wup = wup->swu_collider; 4235 wup->swu_collider = wu; 4236 TAILQ_INSERT_TAIL(&sd->sd_wu_defq, wu, swu_link); 4237 sd->sd_wu_collisions++; 4238 goto queued; 4239 } 4240 4241 start: 4242 sr_raid_startwu(wu); 4243 4244 queued: 4245 splx(s); 4246 } 4247 4248 void 4249 sr_raid_startwu(struct sr_workunit *wu) 4250 { 4251 struct sr_discipline *sd = wu->swu_dis; 4252 struct sr_ccb *ccb; 4253 4254 DNPRINTF(SR_D_WU, "sr_raid_startwu: start wu %p\n", wu); 4255 4256 splassert(IPL_BIO); 4257 4258 if (wu->swu_state == SR_WU_DEFERRED) { 4259 TAILQ_REMOVE(&sd->sd_wu_defq, wu, swu_link); 4260 wu->swu_state = SR_WU_INPROGRESS; 4261 } 4262 4263 if (wu->swu_state != SR_WU_RESTART) 4264 TAILQ_INSERT_TAIL(&sd->sd_wu_pendq, wu, swu_link); 4265 4266 /* Start all of the individual I/Os. */ 4267 if (wu->swu_cb_active == 1) 4268 panic("%s: sr_startwu_callback", DEVNAME(sd->sd_sc)); 4269 wu->swu_cb_active = 1; 4270 4271 TAILQ_FOREACH(ccb, &wu->swu_ccb, ccb_link) 4272 VOP_STRATEGY(&ccb->ccb_buf); 4273 4274 wu->swu_cb_active = 0; 4275 } 4276 4277 void 4278 sr_raid_recreate_wu(struct sr_workunit *wu) 4279 { 4280 struct sr_discipline *sd = wu->swu_dis; 4281 struct sr_workunit *wup = wu; 4282 4283 /* 4284 * Recreate a work unit by releasing the associated CCBs and reissuing 4285 * the SCSI I/O request. This process is then repeated for all of the 4286 * colliding work units. 4287 */ 4288 do { 4289 sr_wu_release_ccbs(wup); 4290 4291 wup->swu_state = SR_WU_REQUEUE; 4292 if (sd->sd_scsi_rw(wup)) 4293 panic("could not requeue I/O"); 4294 4295 wup = wup->swu_collider; 4296 } while (wup); 4297 } 4298 4299 int 4300 sr_alloc_resources(struct sr_discipline *sd) 4301 { 4302 if (sr_wu_alloc(sd, sizeof(struct sr_workunit))) { 4303 sr_error(sd->sd_sc, "unable to allocate work units"); 4304 return (ENOMEM); 4305 } 4306 if (sr_ccb_alloc(sd)) { 4307 sr_error(sd->sd_sc, "unable to allocate ccbs"); 4308 return (ENOMEM); 4309 } 4310 4311 return (0); 4312 } 4313 4314 void 4315 sr_free_resources(struct sr_discipline *sd) 4316 { 4317 sr_wu_free(sd); 4318 sr_ccb_free(sd); 4319 } 4320 4321 void 4322 sr_set_chunk_state(struct sr_discipline *sd, int c, int new_state) 4323 { 4324 int old_state, s; 4325 4326 DNPRINTF(SR_D_STATE, "%s: %s: %s: sr_set_chunk_state %d -> %d\n", 4327 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, 4328 sd->sd_vol.sv_chunks[c]->src_meta.scmi.scm_devname, c, new_state); 4329 4330 /* ok to go to splbio since this only happens in error path */ 4331 s = splbio(); 4332 old_state = sd->sd_vol.sv_chunks[c]->src_meta.scm_status; 4333 4334 /* multiple IOs to the same chunk that fail will come through here */ 4335 if (old_state == new_state) 4336 goto done; 4337 4338 switch (old_state) { 4339 case BIOC_SDONLINE: 4340 if (new_state == BIOC_SDOFFLINE) 4341 break; 4342 else 4343 goto die; 4344 break; 4345 4346 case BIOC_SDOFFLINE: 4347 goto die; 4348 4349 default: 4350 die: 4351 splx(s); /* XXX */ 4352 panic("%s: %s: %s: invalid chunk state transition " 4353 "%d -> %d", DEVNAME(sd->sd_sc), 4354 sd->sd_meta->ssd_devname, 4355 sd->sd_vol.sv_chunks[c]->src_meta.scmi.scm_devname, 4356 old_state, new_state); 4357 /* NOTREACHED */ 4358 } 4359 4360 sd->sd_vol.sv_chunks[c]->src_meta.scm_status = new_state; 4361 sd->sd_set_vol_state(sd); 4362 4363 sd->sd_must_flush = 1; 4364 task_add(systq, &sd->sd_meta_save_task); 4365 done: 4366 splx(s); 4367 } 4368 4369 void 4370 sr_set_vol_state(struct sr_discipline *sd) 4371 { 4372 int states[SR_MAX_STATES]; 4373 int new_state, i, nd; 4374 int old_state = sd->sd_vol_status; 4375 u_int32_t s; 4376 4377 DNPRINTF(SR_D_STATE, "%s: %s: sr_set_vol_state\n", 4378 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname); 4379 4380 nd = sd->sd_meta->ssdi.ssd_chunk_no; 4381 4382 for (i = 0; i < SR_MAX_STATES; i++) 4383 states[i] = 0; 4384 4385 for (i = 0; i < nd; i++) { 4386 s = sd->sd_vol.sv_chunks[i]->src_meta.scm_status; 4387 if (s >= SR_MAX_STATES) 4388 panic("%s: %s: %s: invalid chunk state", 4389 DEVNAME(sd->sd_sc), 4390 sd->sd_meta->ssd_devname, 4391 sd->sd_vol.sv_chunks[i]->src_meta.scmi.scm_devname); 4392 states[s]++; 4393 } 4394 4395 if (states[BIOC_SDONLINE] == nd) 4396 new_state = BIOC_SVONLINE; 4397 else 4398 new_state = BIOC_SVOFFLINE; 4399 4400 DNPRINTF(SR_D_STATE, "%s: %s: sr_set_vol_state %d -> %d\n", 4401 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, 4402 old_state, new_state); 4403 4404 switch (old_state) { 4405 case BIOC_SVONLINE: 4406 if (new_state == BIOC_SVOFFLINE || new_state == BIOC_SVONLINE) 4407 break; 4408 else 4409 goto die; 4410 break; 4411 4412 case BIOC_SVOFFLINE: 4413 /* XXX this might be a little too much */ 4414 goto die; 4415 4416 default: 4417 die: 4418 panic("%s: %s: invalid volume state transition " 4419 "%d -> %d", DEVNAME(sd->sd_sc), 4420 sd->sd_meta->ssd_devname, 4421 old_state, new_state); 4422 /* NOTREACHED */ 4423 } 4424 4425 sd->sd_vol_status = new_state; 4426 } 4427 4428 void * 4429 sr_block_get(struct sr_discipline *sd, long length) 4430 { 4431 return dma_alloc(length, PR_NOWAIT | PR_ZERO); 4432 } 4433 4434 void 4435 sr_block_put(struct sr_discipline *sd, void *ptr, int length) 4436 { 4437 dma_free(ptr, length); 4438 } 4439 4440 void 4441 sr_checksum_print(u_int8_t *md5) 4442 { 4443 int i; 4444 4445 for (i = 0; i < MD5_DIGEST_LENGTH; i++) 4446 printf("%02x", md5[i]); 4447 } 4448 4449 void 4450 sr_checksum(struct sr_softc *sc, void *src, void *md5, u_int32_t len) 4451 { 4452 MD5_CTX ctx; 4453 4454 DNPRINTF(SR_D_MISC, "%s: sr_checksum(%p %p %d)\n", DEVNAME(sc), src, 4455 md5, len); 4456 4457 MD5Init(&ctx); 4458 MD5Update(&ctx, src, len); 4459 MD5Final(md5, &ctx); 4460 } 4461 4462 void 4463 sr_uuid_generate(struct sr_uuid *uuid) 4464 { 4465 arc4random_buf(uuid->sui_id, sizeof(uuid->sui_id)); 4466 /* UUID version 4: random */ 4467 uuid->sui_id[6] &= 0x0f; 4468 uuid->sui_id[6] |= 0x40; 4469 /* RFC4122 variant */ 4470 uuid->sui_id[8] &= 0x3f; 4471 uuid->sui_id[8] |= 0x80; 4472 } 4473 4474 char * 4475 sr_uuid_format(struct sr_uuid *uuid) 4476 { 4477 char *uuidstr; 4478 4479 uuidstr = malloc(37, M_DEVBUF, M_WAITOK); 4480 4481 snprintf(uuidstr, 37, 4482 "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-" 4483 "%02x%02x%02x%02x%02x%02x", 4484 uuid->sui_id[0], uuid->sui_id[1], 4485 uuid->sui_id[2], uuid->sui_id[3], 4486 uuid->sui_id[4], uuid->sui_id[5], 4487 uuid->sui_id[6], uuid->sui_id[7], 4488 uuid->sui_id[8], uuid->sui_id[9], 4489 uuid->sui_id[10], uuid->sui_id[11], 4490 uuid->sui_id[12], uuid->sui_id[13], 4491 uuid->sui_id[14], uuid->sui_id[15]); 4492 4493 return uuidstr; 4494 } 4495 4496 void 4497 sr_uuid_print(struct sr_uuid *uuid, int cr) 4498 { 4499 char *uuidstr; 4500 4501 uuidstr = sr_uuid_format(uuid); 4502 printf("%s%s", uuidstr, (cr ? "\n" : "")); 4503 free(uuidstr, M_DEVBUF, 37); 4504 } 4505 4506 int 4507 sr_already_assembled(struct sr_discipline *sd) 4508 { 4509 struct sr_softc *sc = sd->sd_sc; 4510 struct sr_discipline *sdtmp; 4511 4512 TAILQ_FOREACH(sdtmp, &sc->sc_dis_list, sd_link) { 4513 if (!bcmp(&sd->sd_meta->ssdi.ssd_uuid, 4514 &sdtmp->sd_meta->ssdi.ssd_uuid, 4515 sizeof(sd->sd_meta->ssdi.ssd_uuid))) 4516 return (1); 4517 } 4518 4519 return (0); 4520 } 4521 4522 int32_t 4523 sr_validate_stripsize(u_int32_t b) 4524 { 4525 int s = 0; 4526 4527 if (b % DEV_BSIZE) 4528 return (-1); 4529 4530 while ((b & 1) == 0) { 4531 b >>= 1; 4532 s++; 4533 } 4534 4535 /* only multiple of twos */ 4536 b >>= 1; 4537 if (b) 4538 return(-1); 4539 4540 return (s); 4541 } 4542 4543 void 4544 sr_quiesce(void) 4545 { 4546 struct sr_softc *sc = softraid0; 4547 struct sr_discipline *sd, *nsd; 4548 4549 /* Shutdown disciplines in reverse attach order. */ 4550 TAILQ_FOREACH_REVERSE_SAFE(sd, &sc->sc_dis_list, 4551 sr_discipline_list, sd_link, nsd) 4552 sr_discipline_shutdown(sd, 1, -1); 4553 } 4554 4555 void 4556 sr_shutdown(int dying) 4557 { 4558 struct sr_softc *sc = softraid0; 4559 struct sr_discipline *sd; 4560 4561 DNPRINTF(SR_D_MISC, "%s: sr_shutdown\n", DEVNAME(sc)); 4562 4563 /* 4564 * Since softraid is not under mainbus, we have to explicitly 4565 * notify its children that the power is going down, so they 4566 * can execute their shutdown hooks. 4567 */ 4568 config_suspend((struct device *)sc, DVACT_POWERDOWN); 4569 4570 /* Shutdown disciplines in reverse attach order. */ 4571 while ((sd = TAILQ_LAST(&sc->sc_dis_list, sr_discipline_list)) != NULL) 4572 sr_discipline_shutdown(sd, 1, dying); 4573 } 4574 4575 int 4576 sr_validate_io(struct sr_workunit *wu, daddr_t *blkno, char *func) 4577 { 4578 struct sr_discipline *sd = wu->swu_dis; 4579 struct scsi_xfer *xs = wu->swu_xs; 4580 int rv = 1; 4581 4582 DNPRINTF(SR_D_DIS, "%s: %s 0x%02x\n", DEVNAME(sd->sd_sc), func, 4583 xs->cmd->opcode); 4584 4585 if (sd->sd_meta->ssd_data_blkno == 0) 4586 panic("invalid data blkno"); 4587 4588 if (sd->sd_vol_status == BIOC_SVOFFLINE) { 4589 DNPRINTF(SR_D_DIS, "%s: %s device offline\n", 4590 DEVNAME(sd->sd_sc), func); 4591 goto bad; 4592 } 4593 4594 if (xs->datalen == 0) { 4595 printf("%s: %s: illegal block count for %s\n", 4596 DEVNAME(sd->sd_sc), func, sd->sd_meta->ssd_devname); 4597 goto bad; 4598 } 4599 4600 if (xs->cmdlen == 10) 4601 *blkno = _4btol(((struct scsi_rw_big *)xs->cmd)->addr); 4602 else if (xs->cmdlen == 16) 4603 *blkno = _8btol(((struct scsi_rw_16 *)xs->cmd)->addr); 4604 else if (xs->cmdlen == 6) 4605 *blkno = _3btol(((struct scsi_rw *)xs->cmd)->addr); 4606 else { 4607 printf("%s: %s: illegal cmdlen for %s\n", 4608 DEVNAME(sd->sd_sc), func, sd->sd_meta->ssd_devname); 4609 goto bad; 4610 } 4611 4612 *blkno *= (sd->sd_meta->ssdi.ssd_secsize / DEV_BSIZE); 4613 4614 wu->swu_blk_start = *blkno; 4615 wu->swu_blk_end = *blkno + (xs->datalen >> DEV_BSHIFT) - 1; 4616 4617 if (wu->swu_blk_end > sd->sd_meta->ssdi.ssd_size) { 4618 DNPRINTF(SR_D_DIS, "%s: %s out of bounds start: %lld " 4619 "end: %lld length: %d\n", 4620 DEVNAME(sd->sd_sc), func, (long long)wu->swu_blk_start, 4621 (long long)wu->swu_blk_end, xs->datalen); 4622 4623 sd->sd_scsi_sense.error_code = SSD_ERRCODE_CURRENT | 4624 SSD_ERRCODE_VALID; 4625 sd->sd_scsi_sense.flags = SKEY_ILLEGAL_REQUEST; 4626 sd->sd_scsi_sense.add_sense_code = 0x21; 4627 sd->sd_scsi_sense.add_sense_code_qual = 0x00; 4628 sd->sd_scsi_sense.extra_len = 4; 4629 goto bad; 4630 } 4631 4632 rv = 0; 4633 bad: 4634 return (rv); 4635 } 4636 4637 void 4638 sr_rebuild_start(void *arg) 4639 { 4640 struct sr_discipline *sd = arg; 4641 struct sr_softc *sc = sd->sd_sc; 4642 4643 DNPRINTF(SR_D_REBUILD, "%s: %s starting rebuild thread\n", 4644 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname); 4645 4646 if (kthread_create(sr_rebuild_thread, sd, &sd->sd_background_proc, 4647 DEVNAME(sc)) != 0) 4648 printf("%s: unable to start background operation\n", 4649 DEVNAME(sc)); 4650 } 4651 4652 void 4653 sr_rebuild_thread(void *arg) 4654 { 4655 struct sr_discipline *sd = arg; 4656 4657 DNPRINTF(SR_D_REBUILD, "%s: %s rebuild thread started\n", 4658 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname); 4659 4660 sd->sd_reb_active = 1; 4661 sd->sd_rebuild(sd); 4662 sd->sd_reb_active = 0; 4663 4664 kthread_exit(0); 4665 } 4666 4667 void 4668 sr_rebuild(struct sr_discipline *sd) 4669 { 4670 struct sr_softc *sc = sd->sd_sc; 4671 u_int64_t sz, whole_blk, partial_blk, blk, restart; 4672 daddr_t lba; 4673 struct sr_workunit *wu_r, *wu_w; 4674 struct scsi_xfer xs_r, xs_w; 4675 struct scsi_rw_16 *cr, *cw; 4676 int c, s, slept, percent = 0, old_percent = -1; 4677 u_int8_t *buf; 4678 4679 whole_blk = sd->sd_meta->ssdi.ssd_size / SR_REBUILD_IO_SIZE; 4680 partial_blk = sd->sd_meta->ssdi.ssd_size % SR_REBUILD_IO_SIZE; 4681 4682 restart = sd->sd_meta->ssd_rebuild / SR_REBUILD_IO_SIZE; 4683 if (restart > whole_blk) { 4684 printf("%s: bogus rebuild restart offset, starting from 0\n", 4685 DEVNAME(sc)); 4686 restart = 0; 4687 } 4688 if (restart) { 4689 /* 4690 * XXX there is a hole here; there is a posibility that we 4691 * had a restart however the chunk that was supposed to 4692 * be rebuilt is no longer valid; we can reach this situation 4693 * when a rebuild is in progress and the box crashes and 4694 * on reboot the rebuild chunk is different (like zero'd or 4695 * replaced). We need to check the uuid of the chunk that is 4696 * being rebuilt to assert this. 4697 */ 4698 percent = sr_rebuild_percent(sd); 4699 printf("%s: resuming rebuild on %s at %d%%\n", 4700 DEVNAME(sc), sd->sd_meta->ssd_devname, percent); 4701 } 4702 4703 /* currently this is 64k therefore we can use dma_alloc */ 4704 buf = dma_alloc(SR_REBUILD_IO_SIZE << DEV_BSHIFT, PR_WAITOK); 4705 for (blk = restart; blk <= whole_blk; blk++) { 4706 lba = blk * SR_REBUILD_IO_SIZE; 4707 sz = SR_REBUILD_IO_SIZE; 4708 if (blk == whole_blk) { 4709 if (partial_blk == 0) 4710 break; 4711 sz = partial_blk; 4712 } 4713 4714 /* get some wu */ 4715 wu_r = sr_scsi_wu_get(sd, 0); 4716 wu_w = sr_scsi_wu_get(sd, 0); 4717 4718 DNPRINTF(SR_D_REBUILD, "%s: %s rebuild wu_r %p, wu_w %p\n", 4719 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, wu_r, wu_w); 4720 4721 /* setup read io */ 4722 bzero(&xs_r, sizeof xs_r); 4723 xs_r.error = XS_NOERROR; 4724 xs_r.flags = SCSI_DATA_IN; 4725 xs_r.datalen = sz << DEV_BSHIFT; 4726 xs_r.data = buf; 4727 xs_r.cmdlen = sizeof(*cr); 4728 xs_r.cmd = &xs_r.cmdstore; 4729 cr = (struct scsi_rw_16 *)xs_r.cmd; 4730 cr->opcode = READ_16; 4731 _lto4b(sz, cr->length); 4732 _lto8b(lba, cr->addr); 4733 wu_r->swu_state = SR_WU_CONSTRUCT; 4734 wu_r->swu_flags |= SR_WUF_REBUILD; 4735 wu_r->swu_xs = &xs_r; 4736 if (sd->sd_scsi_rw(wu_r)) { 4737 printf("%s: could not create read io\n", 4738 DEVNAME(sc)); 4739 goto fail; 4740 } 4741 4742 /* setup write io */ 4743 bzero(&xs_w, sizeof xs_w); 4744 xs_w.error = XS_NOERROR; 4745 xs_w.flags = SCSI_DATA_OUT; 4746 xs_w.datalen = sz << DEV_BSHIFT; 4747 xs_w.data = buf; 4748 xs_w.cmdlen = sizeof(*cw); 4749 xs_w.cmd = &xs_w.cmdstore; 4750 cw = (struct scsi_rw_16 *)xs_w.cmd; 4751 cw->opcode = WRITE_16; 4752 _lto4b(sz, cw->length); 4753 _lto8b(lba, cw->addr); 4754 wu_w->swu_state = SR_WU_CONSTRUCT; 4755 wu_w->swu_flags |= SR_WUF_REBUILD | SR_WUF_WAKEUP; 4756 wu_w->swu_xs = &xs_w; 4757 if (sd->sd_scsi_rw(wu_w)) { 4758 printf("%s: could not create write io\n", 4759 DEVNAME(sc)); 4760 goto fail; 4761 } 4762 4763 /* 4764 * collide with the read io so that we get automatically 4765 * started when the read is done 4766 */ 4767 wu_w->swu_state = SR_WU_DEFERRED; 4768 wu_r->swu_collider = wu_w; 4769 s = splbio(); 4770 TAILQ_INSERT_TAIL(&sd->sd_wu_defq, wu_w, swu_link); 4771 splx(s); 4772 4773 DNPRINTF(SR_D_REBUILD, "%s: %s rebuild scheduling wu_r %p\n", 4774 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, wu_r); 4775 4776 wu_r->swu_state = SR_WU_INPROGRESS; 4777 sr_schedule_wu(wu_r); 4778 4779 /* wait for write completion */ 4780 slept = 0; 4781 while ((wu_w->swu_flags & SR_WUF_REBUILDIOCOMP) == 0) { 4782 tsleep(wu_w, PRIBIO, "sr_rebuild", 0); 4783 slept = 1; 4784 } 4785 /* yield if we didn't sleep */ 4786 if (slept == 0) 4787 tsleep(sc, PWAIT, "sr_yield", 1); 4788 4789 sr_scsi_wu_put(sd, wu_r); 4790 sr_scsi_wu_put(sd, wu_w); 4791 4792 sd->sd_meta->ssd_rebuild = lba; 4793 4794 /* XXX - this should be based on size, not percentage. */ 4795 /* save metadata every percent */ 4796 percent = sr_rebuild_percent(sd); 4797 if (percent != old_percent && blk != whole_blk) { 4798 if (sr_meta_save(sd, SR_META_DIRTY)) 4799 printf("%s: could not save metadata to %s\n", 4800 DEVNAME(sc), sd->sd_meta->ssd_devname); 4801 old_percent = percent; 4802 } 4803 4804 if (sd->sd_reb_abort) 4805 goto abort; 4806 } 4807 4808 /* all done */ 4809 sd->sd_meta->ssd_rebuild = 0; 4810 for (c = 0; c < sd->sd_meta->ssdi.ssd_chunk_no; c++) { 4811 if (sd->sd_vol.sv_chunks[c]->src_meta.scm_status == 4812 BIOC_SDREBUILD) { 4813 sd->sd_set_chunk_state(sd, c, BIOC_SDONLINE); 4814 break; 4815 } 4816 } 4817 4818 abort: 4819 if (sr_meta_save(sd, SR_META_DIRTY)) 4820 printf("%s: could not save metadata to %s\n", 4821 DEVNAME(sc), sd->sd_meta->ssd_devname); 4822 fail: 4823 dma_free(buf, SR_REBUILD_IO_SIZE << DEV_BSHIFT); 4824 } 4825 4826 #ifndef SMALL_KERNEL 4827 int 4828 sr_sensors_create(struct sr_discipline *sd) 4829 { 4830 struct sr_softc *sc = sd->sd_sc; 4831 int rv = 1; 4832 4833 DNPRINTF(SR_D_STATE, "%s: %s: sr_sensors_create\n", 4834 DEVNAME(sc), sd->sd_meta->ssd_devname); 4835 4836 sd->sd_vol.sv_sensor.type = SENSOR_DRIVE; 4837 sd->sd_vol.sv_sensor.status = SENSOR_S_UNKNOWN; 4838 strlcpy(sd->sd_vol.sv_sensor.desc, sd->sd_meta->ssd_devname, 4839 sizeof(sd->sd_vol.sv_sensor.desc)); 4840 4841 sensor_attach(&sc->sc_sensordev, &sd->sd_vol.sv_sensor); 4842 sd->sd_vol.sv_sensor_attached = 1; 4843 4844 if (sc->sc_sensor_task == NULL) { 4845 sc->sc_sensor_task = sensor_task_register(sc, 4846 sr_sensors_refresh, 10); 4847 if (sc->sc_sensor_task == NULL) 4848 goto bad; 4849 } 4850 4851 rv = 0; 4852 bad: 4853 return (rv); 4854 } 4855 4856 void 4857 sr_sensors_delete(struct sr_discipline *sd) 4858 { 4859 DNPRINTF(SR_D_STATE, "%s: sr_sensors_delete\n", DEVNAME(sd->sd_sc)); 4860 4861 if (sd->sd_vol.sv_sensor_attached) 4862 sensor_detach(&sd->sd_sc->sc_sensordev, &sd->sd_vol.sv_sensor); 4863 } 4864 4865 void 4866 sr_sensors_refresh(void *arg) 4867 { 4868 struct sr_softc *sc = arg; 4869 struct sr_volume *sv; 4870 struct sr_discipline *sd; 4871 4872 DNPRINTF(SR_D_STATE, "%s: sr_sensors_refresh\n", DEVNAME(sc)); 4873 4874 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 4875 sv = &sd->sd_vol; 4876 4877 switch(sd->sd_vol_status) { 4878 case BIOC_SVOFFLINE: 4879 sv->sv_sensor.value = SENSOR_DRIVE_FAIL; 4880 sv->sv_sensor.status = SENSOR_S_CRIT; 4881 break; 4882 4883 case BIOC_SVDEGRADED: 4884 sv->sv_sensor.value = SENSOR_DRIVE_PFAIL; 4885 sv->sv_sensor.status = SENSOR_S_WARN; 4886 break; 4887 4888 case BIOC_SVREBUILD: 4889 sv->sv_sensor.value = SENSOR_DRIVE_REBUILD; 4890 sv->sv_sensor.status = SENSOR_S_WARN; 4891 break; 4892 4893 case BIOC_SVSCRUB: 4894 case BIOC_SVONLINE: 4895 sv->sv_sensor.value = SENSOR_DRIVE_ONLINE; 4896 sv->sv_sensor.status = SENSOR_S_OK; 4897 break; 4898 4899 default: 4900 sv->sv_sensor.value = 0; /* unknown */ 4901 sv->sv_sensor.status = SENSOR_S_UNKNOWN; 4902 } 4903 } 4904 } 4905 #endif /* SMALL_KERNEL */ 4906 4907 #ifdef SR_FANCY_STATS 4908 void sr_print_stats(void); 4909 4910 void 4911 sr_print_stats(void) 4912 { 4913 struct sr_softc *sc = softraid0; 4914 struct sr_discipline *sd; 4915 4916 if (sc == NULL) { 4917 printf("no softraid softc found\n"); 4918 return; 4919 } 4920 4921 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 4922 printf("%s: ios pending %d, collisions %llu\n", 4923 sd->sd_meta->ssd_devname, 4924 sd->sd_wu_pending, 4925 sd->sd_wu_collisions); 4926 } 4927 } 4928 #endif /* SR_FANCY_STATS */ 4929 4930 #ifdef SR_DEBUG 4931 void 4932 sr_meta_print(struct sr_metadata *m) 4933 { 4934 int i; 4935 struct sr_meta_chunk *mc; 4936 struct sr_meta_opt_hdr *omh; 4937 4938 if (!(sr_debug & SR_D_META)) 4939 return; 4940 4941 printf("\tssd_magic 0x%llx\n", m->ssdi.ssd_magic); 4942 printf("\tssd_version %d\n", m->ssdi.ssd_version); 4943 printf("\tssd_vol_flags 0x%x\n", m->ssdi.ssd_vol_flags); 4944 printf("\tssd_uuid "); 4945 sr_uuid_print(&m->ssdi.ssd_uuid, 1); 4946 printf("\tssd_chunk_no %d\n", m->ssdi.ssd_chunk_no); 4947 printf("\tssd_chunk_id %d\n", m->ssdi.ssd_chunk_id); 4948 printf("\tssd_opt_no %d\n", m->ssdi.ssd_opt_no); 4949 printf("\tssd_volid %d\n", m->ssdi.ssd_volid); 4950 printf("\tssd_level %d\n", m->ssdi.ssd_level); 4951 printf("\tssd_size %lld\n", m->ssdi.ssd_size); 4952 printf("\tssd_devname %s\n", m->ssd_devname); 4953 printf("\tssd_vendor %s\n", m->ssdi.ssd_vendor); 4954 printf("\tssd_product %s\n", m->ssdi.ssd_product); 4955 printf("\tssd_revision %s\n", m->ssdi.ssd_revision); 4956 printf("\tssd_strip_size %d\n", m->ssdi.ssd_strip_size); 4957 printf("\tssd_checksum "); 4958 sr_checksum_print(m->ssd_checksum); 4959 printf("\n"); 4960 printf("\tssd_meta_flags 0x%x\n", m->ssd_meta_flags); 4961 printf("\tssd_ondisk %llu\n", m->ssd_ondisk); 4962 4963 mc = (struct sr_meta_chunk *)(m + 1); 4964 for (i = 0; i < m->ssdi.ssd_chunk_no; i++, mc++) { 4965 printf("\t\tscm_volid %d\n", mc->scmi.scm_volid); 4966 printf("\t\tscm_chunk_id %d\n", mc->scmi.scm_chunk_id); 4967 printf("\t\tscm_devname %s\n", mc->scmi.scm_devname); 4968 printf("\t\tscm_size %lld\n", mc->scmi.scm_size); 4969 printf("\t\tscm_coerced_size %lld\n",mc->scmi.scm_coerced_size); 4970 printf("\t\tscm_uuid "); 4971 sr_uuid_print(&mc->scmi.scm_uuid, 1); 4972 printf("\t\tscm_checksum "); 4973 sr_checksum_print(mc->scm_checksum); 4974 printf("\n"); 4975 printf("\t\tscm_status %d\n", mc->scm_status); 4976 } 4977 4978 omh = (struct sr_meta_opt_hdr *)((u_int8_t *)(m + 1) + 4979 sizeof(struct sr_meta_chunk) * m->ssdi.ssd_chunk_no); 4980 for (i = 0; i < m->ssdi.ssd_opt_no; i++) { 4981 printf("\t\t\tsom_type %d\n", omh->som_type); 4982 printf("\t\t\tsom_checksum "); 4983 sr_checksum_print(omh->som_checksum); 4984 printf("\n"); 4985 omh = (struct sr_meta_opt_hdr *)((void *)omh + 4986 omh->som_length); 4987 } 4988 } 4989 4990 void 4991 sr_dump_block(void *blk, int len) 4992 { 4993 uint8_t *b = blk; 4994 int i, j, c; 4995 4996 for (i = 0; i < len; i += 16) { 4997 for (j = 0; j < 16; j++) 4998 printf("%.2x ", b[i + j]); 4999 printf(" "); 5000 for (j = 0; j < 16; j++) { 5001 c = b[i + j]; 5002 if (c < ' ' || c > 'z' || i + j > len) 5003 c = '.'; 5004 printf("%c", c); 5005 } 5006 printf("\n"); 5007 } 5008 } 5009 5010 void 5011 sr_dump_mem(u_int8_t *p, int len) 5012 { 5013 int i; 5014 5015 for (i = 0; i < len; i++) 5016 printf("%02x ", *p++); 5017 printf("\n"); 5018 } 5019 5020 #endif /* SR_DEBUG */ 5021 5022 #ifdef HIBERNATE 5023 /* 5024 * Side-effect free (no malloc, printf, pool, splx) softraid crypto writer. 5025 * 5026 * This function must perform the following: 5027 * 1. Determine the underlying device's own side-effect free I/O function 5028 * (eg, ahci_hibernate_io, wd_hibernate_io, etc). 5029 * 2. Store enough information in the provided page argument for subsequent 5030 * I/O calls (such as the crypto discipline structure for the keys, the 5031 * offset of the softraid partition on the underlying disk, as well as 5032 * the offset of the swap partition within the crypto volume. 5033 * 3. Encrypt the incoming data using the sr_discipline keys, then pass 5034 * the request to the underlying device's own I/O function. 5035 */ 5036 int 5037 sr_hibernate_io(dev_t dev, daddr_t blkno, vaddr_t addr, size_t size, int op, void *page) 5038 { 5039 /* Struct for stashing data obtained on HIB_INIT. 5040 * XXX 5041 * We share the page with the underlying device's own 5042 * side-effect free I/O function, so we pad our data to 5043 * the end of the page. Presently this does not overlap 5044 * with either of the two other side-effect free i/o 5045 * functions (ahci/wd). 5046 */ 5047 struct { 5048 char pad[3072]; 5049 struct sr_discipline *srd; 5050 hibio_fn subfn; /* underlying device i/o fn */ 5051 dev_t subdev; /* underlying device dev_t */ 5052 daddr_t sr_swapoff; /* ofs of swap part in sr volume */ 5053 char buf[DEV_BSIZE]; /* encryption performed into this buf */ 5054 } *my = page; 5055 extern struct cfdriver sd_cd; 5056 char errstr[128], *dl_ret; 5057 struct sr_chunk *schunk; 5058 struct sd_softc *sd; 5059 struct aes_xts_ctx ctx; 5060 struct sr_softc *sc; 5061 struct device *dv; 5062 daddr_t key_blkno; 5063 uint32_t sub_raidoff; /* ofs of sr part in underlying dev */ 5064 struct disklabel dl; 5065 struct partition *pp; 5066 size_t i, j; 5067 u_char iv[8]; 5068 5069 /* 5070 * In HIB_INIT, we are passed the swap partition size and offset 5071 * in 'size' and 'blkno' respectively. These are relative to the 5072 * start of the softraid partition, and we need to save these 5073 * for later translation to the underlying device's layout. 5074 */ 5075 if (op == HIB_INIT) { 5076 dv = disk_lookup(&sd_cd, DISKUNIT(dev)); 5077 sd = (struct sd_softc *)dv; 5078 sc = (struct sr_softc *)dv->dv_parent->dv_parent; 5079 5080 /* 5081 * Look up the sr discipline. This is used to determine 5082 * if we are SR crypto and what the underlying device is. 5083 */ 5084 my->srd = sc->sc_targets[sd->sc_link->target]; 5085 DNPRINTF(SR_D_MISC, "sr_hibernate_io: discipline is %s\n", 5086 my->srd->sd_name); 5087 if (strncmp(my->srd->sd_name, "CRYPTO", 5088 sizeof(my->srd->sd_name))) 5089 return (ENOTSUP); 5090 5091 /* Find the underlying device */ 5092 schunk = my->srd->sd_vol.sv_chunks[0]; 5093 my->subdev = schunk->src_dev_mm; 5094 5095 /* 5096 * Find the appropriate underlying device side effect free 5097 * I/O function, based on the type of device it is. 5098 */ 5099 my->subfn = get_hibernate_io_function(my->subdev); 5100 if (!my->subfn) 5101 return (ENODEV); 5102 5103 /* 5104 * Find blkno where this raid partition starts on 5105 * the underlying disk. 5106 */ 5107 dl_ret = disk_readlabel(&dl, my->subdev, errstr, 5108 sizeof(errstr)); 5109 if (dl_ret) { 5110 printf("Hibernate error reading disklabel: %s\n", dl_ret); 5111 return (ENOTSUP); 5112 } 5113 5114 pp = &dl.d_partitions[DISKPART(my->subdev)]; 5115 if (pp->p_fstype != FS_RAID || DL_GETPSIZE(pp) == 0) 5116 return (ENOTSUP); 5117 5118 /* Find the blkno of the SR part in the underlying device */ 5119 sub_raidoff = my->srd->sd_meta->ssd_data_blkno + 5120 DL_SECTOBLK(&dl, DL_GETPOFFSET(pp)); 5121 DNPRINTF(SR_D_MISC,"sr_hibernate_io: blk trans ofs: %d blks\n", 5122 sub_raidoff); 5123 5124 /* Save the blkno of the swap partition in the SR disk */ 5125 my->sr_swapoff = blkno; 5126 5127 /* Initialize the sub-device */ 5128 return my->subfn(my->subdev, sub_raidoff + blkno, 5129 addr, size, op, page); 5130 } 5131 5132 /* Hibernate only uses (and we only support) writes */ 5133 if (op != HIB_W) 5134 return (ENOTSUP); 5135 5136 /* 5137 * Blocks act as the IV for the encryption. These block numbers 5138 * are relative to the start of the sr partition, but the 'blkno' 5139 * passed above is relative to the start of the swap partition 5140 * inside the sr partition, so bias appropriately. 5141 */ 5142 key_blkno = my->sr_swapoff + blkno; 5143 5144 /* Process each disk block one at a time. */ 5145 for (i = 0; i < size; i += DEV_BSIZE) { 5146 int res; 5147 5148 bzero(&ctx, sizeof(ctx)); 5149 5150 /* 5151 * Set encryption key (from the sr discipline stashed 5152 * during HIB_INIT. This code is based on the softraid 5153 * bootblock code. 5154 */ 5155 aes_xts_setkey(&ctx, my->srd->mds.mdd_crypto.scr_key[0], 64); 5156 /* We encrypt DEV_BSIZE bytes at a time in my->buf */ 5157 memcpy(my->buf, ((char *)addr) + i, DEV_BSIZE); 5158 5159 /* Block number is the IV */ 5160 memcpy(&iv, &key_blkno, sizeof(key_blkno)); 5161 aes_xts_reinit(&ctx, iv); 5162 5163 /* Encrypt DEV_BSIZE bytes, AES_XTS_BLOCKSIZE bytes at a time */ 5164 for (j = 0; j < DEV_BSIZE; j += AES_XTS_BLOCKSIZE) 5165 aes_xts_encrypt(&ctx, my->buf + j); 5166 5167 /* 5168 * Write one block out from my->buf to the underlying device 5169 * using its own side-effect free I/O function. 5170 */ 5171 res = my->subfn(my->subdev, blkno + (i / DEV_BSIZE), 5172 (vaddr_t)(my->buf), DEV_BSIZE, op, page); 5173 if (res != 0) 5174 return (res); 5175 key_blkno++; 5176 } 5177 return (0); 5178 } 5179 #endif /* HIBERNATE */ 5180