1 /* $OpenBSD: softraid.c,v 1.380 2017/04/14 15:11:31 bluhm 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); 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(void); 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_CANFAIL | 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_CANFAIL | 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 | M_CANFAIL); 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 | M_CANFAIL); 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_CANFAIL | 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 %02hx%02hx%02hx%02hx" 1536 "%02hx%02hx%02hx%02hx\n", 1537 DEVNAME(sc), rootduid[0], 1538 rootduid[1], rootduid[2], 1539 rootduid[3], rootduid[4], 1540 rootduid[5], rootduid[6], 1541 rootduid[7]); 1542 return; 1543 } 1544 } 1545 } 1546 } 1547 } 1548 1549 int 1550 sr_meta_native_probe(struct sr_softc *sc, struct sr_chunk *ch_entry) 1551 { 1552 struct disklabel label; 1553 char *devname; 1554 int error, part; 1555 u_int64_t size; 1556 1557 DNPRINTF(SR_D_META, "%s: sr_meta_native_probe(%s)\n", 1558 DEVNAME(sc), ch_entry->src_devname); 1559 1560 devname = ch_entry->src_devname; 1561 part = DISKPART(ch_entry->src_dev_mm); 1562 1563 /* get disklabel */ 1564 error = VOP_IOCTL(ch_entry->src_vn, DIOCGDINFO, (caddr_t)&label, FREAD, 1565 NOCRED, curproc); 1566 if (error) { 1567 DNPRINTF(SR_D_META, "%s: %s can't obtain disklabel\n", 1568 DEVNAME(sc), devname); 1569 goto unwind; 1570 } 1571 memcpy(ch_entry->src_duid, label.d_uid, sizeof(ch_entry->src_duid)); 1572 1573 /* make sure the partition is of the right type */ 1574 if (label.d_partitions[part].p_fstype != FS_RAID) { 1575 DNPRINTF(SR_D_META, 1576 "%s: %s partition not of type RAID (%d)\n", DEVNAME(sc), 1577 devname, 1578 label.d_partitions[part].p_fstype); 1579 goto unwind; 1580 } 1581 1582 size = DL_SECTOBLK(&label, DL_GETPSIZE(&label.d_partitions[part])); 1583 if (size <= SR_DATA_OFFSET) { 1584 DNPRINTF(SR_D_META, "%s: %s partition too small\n", DEVNAME(sc), 1585 devname); 1586 goto unwind; 1587 } 1588 size -= SR_DATA_OFFSET; 1589 if (size > INT64_MAX) { 1590 DNPRINTF(SR_D_META, "%s: %s partition too large\n", DEVNAME(sc), 1591 devname); 1592 goto unwind; 1593 } 1594 ch_entry->src_size = size; 1595 ch_entry->src_secsize = label.d_secsize; 1596 1597 DNPRINTF(SR_D_META, "%s: probe found %s size %lld\n", DEVNAME(sc), 1598 devname, (long long)size); 1599 1600 return (SR_META_F_NATIVE); 1601 unwind: 1602 DNPRINTF(SR_D_META, "%s: invalid device: %s\n", DEVNAME(sc), 1603 devname ? devname : "nodev"); 1604 return (SR_META_F_INVALID); 1605 } 1606 1607 int 1608 sr_meta_native_attach(struct sr_discipline *sd, int force) 1609 { 1610 struct sr_softc *sc = sd->sd_sc; 1611 struct sr_chunk_head *cl = &sd->sd_vol.sv_chunk_list; 1612 struct sr_metadata *md = NULL; 1613 struct sr_chunk *ch_entry, *ch_next; 1614 struct sr_uuid uuid; 1615 u_int64_t version = 0; 1616 int sr, not_sr, rv = 1, d, expected = -1, old_meta = 0; 1617 1618 DNPRINTF(SR_D_META, "%s: sr_meta_native_attach\n", DEVNAME(sc)); 1619 1620 md = malloc(SR_META_SIZE * DEV_BSIZE, M_DEVBUF, M_ZERO | M_NOWAIT); 1621 if (md == NULL) { 1622 sr_error(sc, "not enough memory for metadata buffer"); 1623 goto bad; 1624 } 1625 1626 bzero(&uuid, sizeof uuid); 1627 1628 sr = not_sr = d = 0; 1629 SLIST_FOREACH(ch_entry, cl, src_link) { 1630 if (ch_entry->src_dev_mm == NODEV) 1631 continue; 1632 1633 if (sr_meta_native_read(sd, ch_entry->src_dev_mm, md, NULL)) { 1634 sr_error(sc, "could not read native metadata"); 1635 goto bad; 1636 } 1637 1638 if (md->ssdi.ssd_magic == SR_MAGIC) { 1639 sr++; 1640 ch_entry->src_meta.scmi.scm_chunk_id = 1641 md->ssdi.ssd_chunk_id; 1642 if (d == 0) { 1643 memcpy(&uuid, &md->ssdi.ssd_uuid, sizeof uuid); 1644 expected = md->ssdi.ssd_chunk_no; 1645 version = md->ssd_ondisk; 1646 d++; 1647 continue; 1648 } else if (bcmp(&md->ssdi.ssd_uuid, &uuid, 1649 sizeof uuid)) { 1650 sr_error(sc, "not part of the same volume"); 1651 goto bad; 1652 } 1653 if (md->ssd_ondisk != version) { 1654 old_meta++; 1655 version = MAX(md->ssd_ondisk, version); 1656 } 1657 } else 1658 not_sr++; 1659 } 1660 1661 if (sr && not_sr) { 1662 sr_error(sc, "not all chunks are of the native metadata " 1663 "format"); 1664 goto bad; 1665 } 1666 1667 /* mixed metadata versions; mark bad disks offline */ 1668 if (old_meta) { 1669 d = 0; 1670 for (ch_entry = SLIST_FIRST(cl); ch_entry != NULL; 1671 ch_entry = ch_next, d++) { 1672 ch_next = SLIST_NEXT(ch_entry, src_link); 1673 1674 /* XXX do we want to read this again? */ 1675 if (ch_entry->src_dev_mm == NODEV) 1676 panic("src_dev_mm == NODEV"); 1677 if (sr_meta_native_read(sd, ch_entry->src_dev_mm, md, 1678 NULL)) 1679 sr_warn(sc, "could not read native metadata"); 1680 if (md->ssd_ondisk != version) 1681 sd->sd_vol.sv_chunks[d]->src_meta.scm_status = 1682 BIOC_SDOFFLINE; 1683 } 1684 } 1685 1686 if (expected != sr && !force && expected != -1) { 1687 DNPRINTF(SR_D_META, "%s: not all chunks were provided, trying " 1688 "anyway\n", DEVNAME(sc)); 1689 } 1690 1691 rv = 0; 1692 bad: 1693 free(md, M_DEVBUF, SR_META_SIZE * DEV_BSIZE); 1694 return (rv); 1695 } 1696 1697 int 1698 sr_meta_native_read(struct sr_discipline *sd, dev_t dev, 1699 struct sr_metadata *md, void *fm) 1700 { 1701 #ifdef SR_DEBUG 1702 struct sr_softc *sc = sd->sd_sc; 1703 #endif 1704 DNPRINTF(SR_D_META, "%s: sr_meta_native_read(0x%x, %p)\n", 1705 DEVNAME(sc), dev, md); 1706 1707 return (sr_meta_rw(sd, dev, md, B_READ)); 1708 } 1709 1710 int 1711 sr_meta_native_write(struct sr_discipline *sd, dev_t dev, 1712 struct sr_metadata *md, void *fm) 1713 { 1714 #ifdef SR_DEBUG 1715 struct sr_softc *sc = sd->sd_sc; 1716 #endif 1717 DNPRINTF(SR_D_META, "%s: sr_meta_native_write(0x%x, %p)\n", 1718 DEVNAME(sc), dev, md); 1719 1720 return (sr_meta_rw(sd, dev, md, B_WRITE)); 1721 } 1722 1723 void 1724 sr_hotplug_register(struct sr_discipline *sd, void *func) 1725 { 1726 struct sr_hotplug_list *mhe; 1727 1728 DNPRINTF(SR_D_MISC, "%s: sr_hotplug_register: %p\n", 1729 DEVNAME(sd->sd_sc), func); 1730 1731 /* make sure we aren't on the list yet */ 1732 SLIST_FOREACH(mhe, &sr_hotplug_callbacks, shl_link) 1733 if (mhe->sh_hotplug == func) 1734 return; 1735 1736 mhe = malloc(sizeof(struct sr_hotplug_list), M_DEVBUF, 1737 M_WAITOK | M_ZERO); 1738 mhe->sh_hotplug = func; 1739 mhe->sh_sd = sd; 1740 SLIST_INSERT_HEAD(&sr_hotplug_callbacks, mhe, shl_link); 1741 } 1742 1743 void 1744 sr_hotplug_unregister(struct sr_discipline *sd, void *func) 1745 { 1746 struct sr_hotplug_list *mhe; 1747 1748 DNPRINTF(SR_D_MISC, "%s: sr_hotplug_unregister: %s %p\n", 1749 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, func); 1750 1751 /* make sure we are on the list yet */ 1752 SLIST_FOREACH(mhe, &sr_hotplug_callbacks, shl_link) { 1753 if (mhe->sh_hotplug == func) 1754 break; 1755 } 1756 if (mhe != NULL) { 1757 SLIST_REMOVE(&sr_hotplug_callbacks, mhe, 1758 sr_hotplug_list, shl_link); 1759 free(mhe, M_DEVBUF, 0); 1760 } 1761 } 1762 1763 void 1764 sr_disk_attach(struct disk *diskp, int action) 1765 { 1766 struct sr_hotplug_list *mhe; 1767 1768 SLIST_FOREACH(mhe, &sr_hotplug_callbacks, shl_link) 1769 if (mhe->sh_sd->sd_ready) 1770 mhe->sh_hotplug(mhe->sh_sd, diskp, action); 1771 } 1772 1773 int 1774 sr_match(struct device *parent, void *match, void *aux) 1775 { 1776 return (1); 1777 } 1778 1779 void 1780 sr_attach(struct device *parent, struct device *self, void *aux) 1781 { 1782 struct sr_softc *sc = (void *)self; 1783 struct scsibus_attach_args saa; 1784 1785 DNPRINTF(SR_D_MISC, "\n%s: sr_attach", DEVNAME(sc)); 1786 1787 if (softraid0 == NULL) 1788 softraid0 = sc; 1789 1790 rw_init(&sc->sc_lock, "sr_lock"); 1791 rw_init(&sc->sc_hs_lock, "sr_hs_lock"); 1792 1793 SLIST_INIT(&sr_hotplug_callbacks); 1794 TAILQ_INIT(&sc->sc_dis_list); 1795 SLIST_INIT(&sc->sc_hotspare_list); 1796 1797 #if NBIO > 0 1798 if (bio_register(&sc->sc_dev, sr_bio_ioctl) != 0) 1799 printf("%s: controller registration failed", DEVNAME(sc)); 1800 #endif /* NBIO > 0 */ 1801 1802 #ifndef SMALL_KERNEL 1803 strlcpy(sc->sc_sensordev.xname, DEVNAME(sc), 1804 sizeof(sc->sc_sensordev.xname)); 1805 sensordev_install(&sc->sc_sensordev); 1806 #endif /* SMALL_KERNEL */ 1807 1808 printf("\n"); 1809 1810 sc->sc_link.adapter_softc = sc; 1811 sc->sc_link.adapter = &sr_switch; 1812 sc->sc_link.adapter_target = SR_MAX_LD; 1813 sc->sc_link.adapter_buswidth = SR_MAX_LD; 1814 sc->sc_link.luns = 1; 1815 1816 bzero(&saa, sizeof(saa)); 1817 saa.saa_sc_link = &sc->sc_link; 1818 1819 sc->sc_scsibus = (struct scsibus_softc *)config_found(&sc->sc_dev, 1820 &saa, scsiprint); 1821 1822 softraid_disk_attach = sr_disk_attach; 1823 1824 sr_boot_assembly(sc); 1825 1826 explicit_bzero(sr_bootkey, sizeof(sr_bootkey)); 1827 } 1828 1829 int 1830 sr_detach(struct device *self, int flags) 1831 { 1832 struct sr_softc *sc = (void *)self; 1833 int rv; 1834 1835 DNPRINTF(SR_D_MISC, "%s: sr_detach\n", DEVNAME(sc)); 1836 1837 softraid_disk_attach = NULL; 1838 1839 sr_shutdown(); 1840 1841 #ifndef SMALL_KERNEL 1842 if (sc->sc_sensor_task != NULL) 1843 sensor_task_unregister(sc->sc_sensor_task); 1844 sensordev_deinstall(&sc->sc_sensordev); 1845 #endif /* SMALL_KERNEL */ 1846 1847 if (sc->sc_scsibus != NULL) { 1848 rv = config_detach((struct device *)sc->sc_scsibus, flags); 1849 if (rv != 0) 1850 return (rv); 1851 sc->sc_scsibus = NULL; 1852 } 1853 1854 return (0); 1855 } 1856 1857 void 1858 sr_info(struct sr_softc *sc, const char *fmt, ...) 1859 { 1860 va_list ap; 1861 1862 rw_assert_wrlock(&sc->sc_lock); 1863 1864 va_start(ap, fmt); 1865 bio_status(&sc->sc_status, 0, BIO_MSG_INFO, fmt, &ap); 1866 va_end(ap); 1867 } 1868 1869 void 1870 sr_warn(struct sr_softc *sc, const char *fmt, ...) 1871 { 1872 va_list ap; 1873 1874 rw_assert_wrlock(&sc->sc_lock); 1875 1876 va_start(ap, fmt); 1877 bio_status(&sc->sc_status, 1, BIO_MSG_WARN, fmt, &ap); 1878 va_end(ap); 1879 } 1880 1881 void 1882 sr_error(struct sr_softc *sc, const char *fmt, ...) 1883 { 1884 va_list ap; 1885 1886 rw_assert_wrlock(&sc->sc_lock); 1887 1888 va_start(ap, fmt); 1889 bio_status(&sc->sc_status, 1, BIO_MSG_ERROR, fmt, &ap); 1890 va_end(ap); 1891 } 1892 1893 void 1894 sr_minphys(struct buf *bp, struct scsi_link *sl) 1895 { 1896 DNPRINTF(SR_D_MISC, "sr_minphys: %ld\n", bp->b_bcount); 1897 1898 /* XXX currently using SR_MAXFER = MAXPHYS */ 1899 if (bp->b_bcount > SR_MAXFER) 1900 bp->b_bcount = SR_MAXFER; 1901 minphys(bp); 1902 } 1903 1904 void 1905 sr_copy_internal_data(struct scsi_xfer *xs, void *v, size_t size) 1906 { 1907 size_t copy_cnt; 1908 1909 DNPRINTF(SR_D_MISC, "sr_copy_internal_data xs: %p size: %zu\n", 1910 xs, size); 1911 1912 if (xs->datalen) { 1913 copy_cnt = MIN(size, xs->datalen); 1914 memcpy(xs->data, v, copy_cnt); 1915 } 1916 } 1917 1918 int 1919 sr_ccb_alloc(struct sr_discipline *sd) 1920 { 1921 struct sr_ccb *ccb; 1922 int i; 1923 1924 if (!sd) 1925 return (1); 1926 1927 DNPRINTF(SR_D_CCB, "%s: sr_ccb_alloc\n", DEVNAME(sd->sd_sc)); 1928 1929 if (sd->sd_ccb) 1930 return (1); 1931 1932 sd->sd_ccb = mallocarray(sd->sd_max_wu, 1933 sd->sd_max_ccb_per_wu * sizeof(struct sr_ccb), 1934 M_DEVBUF, M_WAITOK | M_ZERO); 1935 TAILQ_INIT(&sd->sd_ccb_freeq); 1936 for (i = 0; i < sd->sd_max_wu * sd->sd_max_ccb_per_wu; i++) { 1937 ccb = &sd->sd_ccb[i]; 1938 ccb->ccb_dis = sd; 1939 sr_ccb_put(ccb); 1940 } 1941 1942 DNPRINTF(SR_D_CCB, "%s: sr_ccb_alloc ccb: %d\n", 1943 DEVNAME(sd->sd_sc), sd->sd_max_wu * sd->sd_max_ccb_per_wu); 1944 1945 return (0); 1946 } 1947 1948 void 1949 sr_ccb_free(struct sr_discipline *sd) 1950 { 1951 struct sr_ccb *ccb; 1952 1953 if (!sd) 1954 return; 1955 1956 DNPRINTF(SR_D_CCB, "%s: sr_ccb_free %p\n", DEVNAME(sd->sd_sc), sd); 1957 1958 while ((ccb = TAILQ_FIRST(&sd->sd_ccb_freeq)) != NULL) 1959 TAILQ_REMOVE(&sd->sd_ccb_freeq, ccb, ccb_link); 1960 1961 free(sd->sd_ccb, M_DEVBUF, 0); 1962 } 1963 1964 struct sr_ccb * 1965 sr_ccb_get(struct sr_discipline *sd) 1966 { 1967 struct sr_ccb *ccb; 1968 int s; 1969 1970 s = splbio(); 1971 1972 ccb = TAILQ_FIRST(&sd->sd_ccb_freeq); 1973 if (ccb) { 1974 TAILQ_REMOVE(&sd->sd_ccb_freeq, ccb, ccb_link); 1975 ccb->ccb_state = SR_CCB_INPROGRESS; 1976 } 1977 1978 splx(s); 1979 1980 DNPRINTF(SR_D_CCB, "%s: sr_ccb_get: %p\n", DEVNAME(sd->sd_sc), 1981 ccb); 1982 1983 return (ccb); 1984 } 1985 1986 void 1987 sr_ccb_put(struct sr_ccb *ccb) 1988 { 1989 struct sr_discipline *sd = ccb->ccb_dis; 1990 int s; 1991 1992 DNPRINTF(SR_D_CCB, "%s: sr_ccb_put: %p\n", DEVNAME(sd->sd_sc), 1993 ccb); 1994 1995 s = splbio(); 1996 1997 ccb->ccb_wu = NULL; 1998 ccb->ccb_state = SR_CCB_FREE; 1999 ccb->ccb_target = -1; 2000 ccb->ccb_opaque = NULL; 2001 2002 TAILQ_INSERT_TAIL(&sd->sd_ccb_freeq, ccb, ccb_link); 2003 2004 splx(s); 2005 } 2006 2007 struct sr_ccb * 2008 sr_ccb_rw(struct sr_discipline *sd, int chunk, daddr_t blkno, 2009 long len, u_int8_t *data, int xsflags, int ccbflags) 2010 { 2011 struct sr_chunk *sc = sd->sd_vol.sv_chunks[chunk]; 2012 struct sr_ccb *ccb = NULL; 2013 2014 ccb = sr_ccb_get(sd); 2015 if (ccb == NULL) 2016 goto out; 2017 2018 ccb->ccb_flags = ccbflags; 2019 ccb->ccb_target = chunk; 2020 2021 ccb->ccb_buf.b_flags = B_PHYS | B_CALL; 2022 if (ISSET(xsflags, SCSI_DATA_IN)) 2023 ccb->ccb_buf.b_flags |= B_READ; 2024 else 2025 ccb->ccb_buf.b_flags |= B_WRITE; 2026 2027 ccb->ccb_buf.b_blkno = blkno + sd->sd_meta->ssd_data_blkno; 2028 ccb->ccb_buf.b_bcount = len; 2029 ccb->ccb_buf.b_bufsize = len; 2030 ccb->ccb_buf.b_resid = len; 2031 ccb->ccb_buf.b_data = data; 2032 ccb->ccb_buf.b_error = 0; 2033 ccb->ccb_buf.b_iodone = sd->sd_scsi_intr; 2034 ccb->ccb_buf.b_proc = curproc; 2035 ccb->ccb_buf.b_dev = sc->src_dev_mm; 2036 ccb->ccb_buf.b_vp = sc->src_vn; 2037 ccb->ccb_buf.b_bq = NULL; 2038 2039 if (!ISSET(ccb->ccb_buf.b_flags, B_READ)) 2040 ccb->ccb_buf.b_vp->v_numoutput++; 2041 2042 LIST_INIT(&ccb->ccb_buf.b_dep); 2043 2044 DNPRINTF(SR_D_DIS, "%s: %s %s ccb " 2045 "b_bcount %ld b_blkno %lld b_flags 0x%0lx b_data %p\n", 2046 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, sd->sd_name, 2047 ccb->ccb_buf.b_bcount, (long long)ccb->ccb_buf.b_blkno, 2048 ccb->ccb_buf.b_flags, ccb->ccb_buf.b_data); 2049 2050 out: 2051 return ccb; 2052 } 2053 2054 void 2055 sr_ccb_done(struct sr_ccb *ccb) 2056 { 2057 struct sr_workunit *wu = ccb->ccb_wu; 2058 struct sr_discipline *sd = wu->swu_dis; 2059 struct sr_softc *sc = sd->sd_sc; 2060 2061 DNPRINTF(SR_D_INTR, "%s: %s %s ccb done b_bcount %ld b_resid %zu" 2062 " b_flags 0x%0lx block %lld target %d\n", 2063 DEVNAME(sc), sd->sd_meta->ssd_devname, sd->sd_name, 2064 ccb->ccb_buf.b_bcount, ccb->ccb_buf.b_resid, ccb->ccb_buf.b_flags, 2065 (long long)ccb->ccb_buf.b_blkno, ccb->ccb_target); 2066 2067 splassert(IPL_BIO); 2068 2069 if (ccb->ccb_target == -1) 2070 panic("%s: invalid target on wu: %p", DEVNAME(sc), wu); 2071 2072 if (ccb->ccb_buf.b_flags & B_ERROR) { 2073 DNPRINTF(SR_D_INTR, "%s: i/o error on block %lld target %d\n", 2074 DEVNAME(sc), (long long)ccb->ccb_buf.b_blkno, 2075 ccb->ccb_target); 2076 if (ISSET(sd->sd_capabilities, SR_CAP_REDUNDANT)) 2077 sd->sd_set_chunk_state(sd, ccb->ccb_target, 2078 BIOC_SDOFFLINE); 2079 else 2080 printf("%s: i/o error on block %lld target %d " 2081 "b_error %d\n", DEVNAME(sc), 2082 (long long)ccb->ccb_buf.b_blkno, ccb->ccb_target, 2083 ccb->ccb_buf.b_error); 2084 ccb->ccb_state = SR_CCB_FAILED; 2085 wu->swu_ios_failed++; 2086 } else { 2087 ccb->ccb_state = SR_CCB_OK; 2088 wu->swu_ios_succeeded++; 2089 } 2090 2091 wu->swu_ios_complete++; 2092 } 2093 2094 int 2095 sr_wu_alloc(struct sr_discipline *sd, int wu_size) 2096 { 2097 struct sr_workunit *wu; 2098 int i, no_wu; 2099 2100 DNPRINTF(SR_D_WU, "%s: sr_wu_alloc %p %d\n", DEVNAME(sd->sd_sc), 2101 sd, sd->sd_max_wu); 2102 2103 no_wu = sd->sd_max_wu; 2104 sd->sd_wu_pending = no_wu; 2105 2106 mtx_init(&sd->sd_wu_mtx, IPL_BIO); 2107 TAILQ_INIT(&sd->sd_wu); 2108 TAILQ_INIT(&sd->sd_wu_freeq); 2109 TAILQ_INIT(&sd->sd_wu_pendq); 2110 TAILQ_INIT(&sd->sd_wu_defq); 2111 2112 for (i = 0; i < no_wu; i++) { 2113 wu = malloc(wu_size, M_DEVBUF, M_WAITOK | M_ZERO); 2114 TAILQ_INSERT_TAIL(&sd->sd_wu, wu, swu_next); 2115 TAILQ_INIT(&wu->swu_ccb); 2116 wu->swu_dis = sd; 2117 task_set(&wu->swu_task, sr_wu_done_callback, wu); 2118 sr_wu_put(sd, wu); 2119 } 2120 2121 return (0); 2122 } 2123 2124 void 2125 sr_wu_free(struct sr_discipline *sd) 2126 { 2127 struct sr_workunit *wu; 2128 2129 DNPRINTF(SR_D_WU, "%s: sr_wu_free %p\n", DEVNAME(sd->sd_sc), sd); 2130 2131 while ((wu = TAILQ_FIRST(&sd->sd_wu_freeq)) != NULL) 2132 TAILQ_REMOVE(&sd->sd_wu_freeq, wu, swu_link); 2133 while ((wu = TAILQ_FIRST(&sd->sd_wu_pendq)) != NULL) 2134 TAILQ_REMOVE(&sd->sd_wu_pendq, wu, swu_link); 2135 while ((wu = TAILQ_FIRST(&sd->sd_wu_defq)) != NULL) 2136 TAILQ_REMOVE(&sd->sd_wu_defq, wu, swu_link); 2137 2138 while ((wu = TAILQ_FIRST(&sd->sd_wu)) != NULL) { 2139 TAILQ_REMOVE(&sd->sd_wu, wu, swu_next); 2140 free(wu, M_DEVBUF, 0); 2141 } 2142 } 2143 2144 void * 2145 sr_wu_get(void *xsd) 2146 { 2147 struct sr_discipline *sd = (struct sr_discipline *)xsd; 2148 struct sr_workunit *wu; 2149 2150 mtx_enter(&sd->sd_wu_mtx); 2151 wu = TAILQ_FIRST(&sd->sd_wu_freeq); 2152 if (wu) { 2153 TAILQ_REMOVE(&sd->sd_wu_freeq, wu, swu_link); 2154 sd->sd_wu_pending++; 2155 } 2156 mtx_leave(&sd->sd_wu_mtx); 2157 2158 DNPRINTF(SR_D_WU, "%s: sr_wu_get: %p\n", DEVNAME(sd->sd_sc), wu); 2159 2160 return (wu); 2161 } 2162 2163 void 2164 sr_wu_put(void *xsd, void *xwu) 2165 { 2166 struct sr_discipline *sd = (struct sr_discipline *)xsd; 2167 struct sr_workunit *wu = (struct sr_workunit *)xwu; 2168 2169 DNPRINTF(SR_D_WU, "%s: sr_wu_put: %p\n", DEVNAME(sd->sd_sc), wu); 2170 2171 sr_wu_release_ccbs(wu); 2172 sr_wu_init(sd, wu); 2173 2174 mtx_enter(&sd->sd_wu_mtx); 2175 TAILQ_INSERT_TAIL(&sd->sd_wu_freeq, wu, swu_link); 2176 sd->sd_wu_pending--; 2177 mtx_leave(&sd->sd_wu_mtx); 2178 } 2179 2180 void 2181 sr_wu_init(struct sr_discipline *sd, struct sr_workunit *wu) 2182 { 2183 int s; 2184 2185 s = splbio(); 2186 if (wu->swu_cb_active == 1) 2187 panic("%s: sr_wu_init got active wu", DEVNAME(sd->sd_sc)); 2188 splx(s); 2189 2190 wu->swu_xs = NULL; 2191 wu->swu_state = SR_WU_FREE; 2192 wu->swu_flags = 0; 2193 wu->swu_blk_start = 0; 2194 wu->swu_blk_end = 0; 2195 wu->swu_collider = NULL; 2196 } 2197 2198 void 2199 sr_wu_enqueue_ccb(struct sr_workunit *wu, struct sr_ccb *ccb) 2200 { 2201 struct sr_discipline *sd = wu->swu_dis; 2202 int s; 2203 2204 s = splbio(); 2205 if (wu->swu_cb_active == 1) 2206 panic("%s: sr_wu_enqueue_ccb got active wu", 2207 DEVNAME(sd->sd_sc)); 2208 ccb->ccb_wu = wu; 2209 wu->swu_io_count++; 2210 TAILQ_INSERT_TAIL(&wu->swu_ccb, ccb, ccb_link); 2211 splx(s); 2212 } 2213 2214 void 2215 sr_wu_release_ccbs(struct sr_workunit *wu) 2216 { 2217 struct sr_ccb *ccb; 2218 2219 /* Return all ccbs that are associated with this workunit. */ 2220 while ((ccb = TAILQ_FIRST(&wu->swu_ccb)) != NULL) { 2221 TAILQ_REMOVE(&wu->swu_ccb, ccb, ccb_link); 2222 sr_ccb_put(ccb); 2223 } 2224 2225 wu->swu_io_count = 0; 2226 wu->swu_ios_complete = 0; 2227 wu->swu_ios_failed = 0; 2228 wu->swu_ios_succeeded = 0; 2229 } 2230 2231 void 2232 sr_wu_done(struct sr_workunit *wu) 2233 { 2234 struct sr_discipline *sd = wu->swu_dis; 2235 2236 DNPRINTF(SR_D_INTR, "%s: sr_wu_done count %d completed %d failed %d\n", 2237 DEVNAME(sd->sd_sc), wu->swu_io_count, wu->swu_ios_complete, 2238 wu->swu_ios_failed); 2239 2240 if (wu->swu_ios_complete < wu->swu_io_count) 2241 return; 2242 2243 task_add(sd->sd_taskq, &wu->swu_task); 2244 } 2245 2246 void 2247 sr_wu_done_callback(void *xwu) 2248 { 2249 struct sr_workunit *wu = xwu; 2250 struct sr_discipline *sd = wu->swu_dis; 2251 struct scsi_xfer *xs = wu->swu_xs; 2252 struct sr_workunit *wup; 2253 int s; 2254 2255 /* 2256 * The SR_WUF_DISCIPLINE or SR_WUF_REBUILD flag must be set if 2257 * the work unit is not associated with a scsi_xfer. 2258 */ 2259 KASSERT(xs != NULL || 2260 (wu->swu_flags & (SR_WUF_DISCIPLINE|SR_WUF_REBUILD))); 2261 2262 s = splbio(); 2263 2264 if (xs != NULL) { 2265 if (wu->swu_ios_failed) 2266 xs->error = XS_DRIVER_STUFFUP; 2267 else 2268 xs->error = XS_NOERROR; 2269 } 2270 2271 if (sd->sd_scsi_wu_done) { 2272 if (sd->sd_scsi_wu_done(wu) == SR_WU_RESTART) 2273 goto done; 2274 } 2275 2276 /* Remove work unit from pending queue. */ 2277 TAILQ_FOREACH(wup, &sd->sd_wu_pendq, swu_link) 2278 if (wup == wu) 2279 break; 2280 if (wup == NULL) 2281 panic("%s: wu %p not on pending queue", 2282 DEVNAME(sd->sd_sc), wu); 2283 TAILQ_REMOVE(&sd->sd_wu_pendq, wu, swu_link); 2284 2285 if (wu->swu_collider) { 2286 if (wu->swu_ios_failed) 2287 sr_raid_recreate_wu(wu->swu_collider); 2288 2289 /* XXX Should the collider be failed if this xs failed? */ 2290 sr_raid_startwu(wu->swu_collider); 2291 } 2292 2293 /* 2294 * If a discipline provides its own sd_scsi_done function, then it 2295 * is responsible for calling sr_scsi_done() once I/O is complete. 2296 */ 2297 if (wu->swu_flags & SR_WUF_REBUILD) 2298 wu->swu_flags |= SR_WUF_REBUILDIOCOMP; 2299 if (wu->swu_flags & SR_WUF_WAKEUP) 2300 wakeup(wu); 2301 if (sd->sd_scsi_done) 2302 sd->sd_scsi_done(wu); 2303 else if (wu->swu_flags & SR_WUF_DISCIPLINE) 2304 sr_scsi_wu_put(sd, wu); 2305 else if (!(wu->swu_flags & SR_WUF_REBUILD)) 2306 sr_scsi_done(sd, xs); 2307 2308 done: 2309 splx(s); 2310 } 2311 2312 struct sr_workunit * 2313 sr_scsi_wu_get(struct sr_discipline *sd, int flags) 2314 { 2315 return scsi_io_get(&sd->sd_iopool, flags); 2316 } 2317 2318 void 2319 sr_scsi_wu_put(struct sr_discipline *sd, struct sr_workunit *wu) 2320 { 2321 scsi_io_put(&sd->sd_iopool, wu); 2322 2323 if (sd->sd_sync && sd->sd_wu_pending == 0) 2324 wakeup(sd); 2325 } 2326 2327 void 2328 sr_scsi_done(struct sr_discipline *sd, struct scsi_xfer *xs) 2329 { 2330 DNPRINTF(SR_D_DIS, "%s: sr_scsi_done: xs %p\n", DEVNAME(sd->sd_sc), xs); 2331 2332 if (xs->error == XS_NOERROR) 2333 xs->resid = 0; 2334 2335 scsi_done(xs); 2336 2337 if (sd->sd_sync && sd->sd_wu_pending == 0) 2338 wakeup(sd); 2339 } 2340 2341 void 2342 sr_scsi_cmd(struct scsi_xfer *xs) 2343 { 2344 struct scsi_link *link = xs->sc_link; 2345 struct sr_softc *sc = link->adapter_softc; 2346 struct sr_workunit *wu = xs->io; 2347 struct sr_discipline *sd; 2348 2349 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd target %d xs %p flags %#x\n", 2350 DEVNAME(sc), link->target, xs, xs->flags); 2351 2352 sd = sc->sc_targets[link->target]; 2353 if (sd == NULL) 2354 panic("%s: sr_scsi_cmd NULL discipline", DEVNAME(sc)); 2355 2356 if (sd->sd_deleted) { 2357 printf("%s: %s device is being deleted, failing io\n", 2358 DEVNAME(sc), sd->sd_meta->ssd_devname); 2359 goto stuffup; 2360 } 2361 2362 /* scsi layer *can* re-send wu without calling sr_wu_put(). */ 2363 sr_wu_release_ccbs(wu); 2364 sr_wu_init(sd, wu); 2365 wu->swu_state = SR_WU_INPROGRESS; 2366 wu->swu_xs = xs; 2367 2368 switch (xs->cmd->opcode) { 2369 case READ_COMMAND: 2370 case READ_BIG: 2371 case READ_16: 2372 case WRITE_COMMAND: 2373 case WRITE_BIG: 2374 case WRITE_16: 2375 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: READ/WRITE %02x\n", 2376 DEVNAME(sc), xs->cmd->opcode); 2377 if (sd->sd_scsi_rw(wu)) 2378 goto stuffup; 2379 break; 2380 2381 case SYNCHRONIZE_CACHE: 2382 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: SYNCHRONIZE_CACHE\n", 2383 DEVNAME(sc)); 2384 if (sd->sd_scsi_sync(wu)) 2385 goto stuffup; 2386 goto complete; 2387 2388 case TEST_UNIT_READY: 2389 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: TEST_UNIT_READY\n", 2390 DEVNAME(sc)); 2391 if (sd->sd_scsi_tur(wu)) 2392 goto stuffup; 2393 goto complete; 2394 2395 case START_STOP: 2396 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: START_STOP\n", 2397 DEVNAME(sc)); 2398 if (sd->sd_scsi_start_stop(wu)) 2399 goto stuffup; 2400 goto complete; 2401 2402 case INQUIRY: 2403 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd: INQUIRY\n", 2404 DEVNAME(sc)); 2405 if (sd->sd_scsi_inquiry(wu)) 2406 goto stuffup; 2407 goto complete; 2408 2409 case READ_CAPACITY: 2410 case READ_CAPACITY_16: 2411 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd READ CAPACITY 0x%02x\n", 2412 DEVNAME(sc), xs->cmd->opcode); 2413 if (sd->sd_scsi_read_cap(wu)) 2414 goto stuffup; 2415 goto complete; 2416 2417 case REQUEST_SENSE: 2418 DNPRINTF(SR_D_CMD, "%s: sr_scsi_cmd REQUEST SENSE\n", 2419 DEVNAME(sc)); 2420 if (sd->sd_scsi_req_sense(wu)) 2421 goto stuffup; 2422 goto complete; 2423 2424 default: 2425 DNPRINTF(SR_D_CMD, "%s: unsupported scsi command %x\n", 2426 DEVNAME(sc), xs->cmd->opcode); 2427 /* XXX might need to add generic function to handle others */ 2428 goto stuffup; 2429 } 2430 2431 return; 2432 stuffup: 2433 if (sd->sd_scsi_sense.error_code) { 2434 xs->error = XS_SENSE; 2435 memcpy(&xs->sense, &sd->sd_scsi_sense, sizeof(xs->sense)); 2436 bzero(&sd->sd_scsi_sense, sizeof(sd->sd_scsi_sense)); 2437 } else { 2438 xs->error = XS_DRIVER_STUFFUP; 2439 } 2440 complete: 2441 sr_scsi_done(sd, xs); 2442 } 2443 2444 int 2445 sr_scsi_probe(struct scsi_link *link) 2446 { 2447 struct sr_softc *sc = link->adapter_softc; 2448 struct sr_discipline *sd; 2449 2450 KASSERT(link->target < SR_MAX_LD && link->lun == 0); 2451 2452 sd = sc->sc_targets[link->target]; 2453 if (sd == NULL) 2454 return (ENODEV); 2455 2456 link->pool = &sd->sd_iopool; 2457 if (sd->sd_openings) 2458 link->openings = sd->sd_openings(sd); 2459 else 2460 link->openings = sd->sd_max_wu; 2461 2462 return (0); 2463 } 2464 2465 int 2466 sr_scsi_ioctl(struct scsi_link *link, u_long cmd, caddr_t addr, int flag) 2467 { 2468 struct sr_softc *sc = link->adapter_softc; 2469 struct sr_discipline *sd; 2470 2471 sd = sc->sc_targets[link->target]; 2472 if (sd == NULL) 2473 return (ENODEV); 2474 2475 DNPRINTF(SR_D_IOCTL, "%s: %s sr_scsi_ioctl cmd: %#lx\n", 2476 DEVNAME(sc), sd->sd_meta->ssd_devname, cmd); 2477 2478 /* Pass bio ioctls through to the bio handler. */ 2479 if (IOCGROUP(cmd) == 'B') 2480 return (sr_bio_handler(sc, sd, cmd, (struct bio *)addr)); 2481 2482 switch (cmd) { 2483 case DIOCGCACHE: 2484 case DIOCSCACHE: 2485 return (EOPNOTSUPP); 2486 default: 2487 return (ENOTTY); 2488 } 2489 } 2490 2491 int 2492 sr_bio_ioctl(struct device *dev, u_long cmd, caddr_t addr) 2493 { 2494 struct sr_softc *sc = (struct sr_softc *) dev; 2495 DNPRINTF(SR_D_IOCTL, "%s: sr_bio_ioctl\n", DEVNAME(sc)); 2496 2497 return sr_bio_handler(sc, NULL, cmd, (struct bio *)addr); 2498 } 2499 2500 int 2501 sr_bio_handler(struct sr_softc *sc, struct sr_discipline *sd, u_long cmd, 2502 struct bio *bio) 2503 { 2504 int rv = 0; 2505 2506 DNPRINTF(SR_D_IOCTL, "%s: sr_bio_handler ", DEVNAME(sc)); 2507 2508 rw_enter_write(&sc->sc_lock); 2509 2510 bio_status_init(&sc->sc_status, &sc->sc_dev); 2511 2512 switch (cmd) { 2513 case BIOCINQ: 2514 DNPRINTF(SR_D_IOCTL, "inq\n"); 2515 rv = sr_ioctl_inq(sc, (struct bioc_inq *)bio); 2516 break; 2517 2518 case BIOCVOL: 2519 DNPRINTF(SR_D_IOCTL, "vol\n"); 2520 rv = sr_ioctl_vol(sc, (struct bioc_vol *)bio); 2521 break; 2522 2523 case BIOCDISK: 2524 DNPRINTF(SR_D_IOCTL, "disk\n"); 2525 rv = sr_ioctl_disk(sc, (struct bioc_disk *)bio); 2526 break; 2527 2528 case BIOCALARM: 2529 DNPRINTF(SR_D_IOCTL, "alarm\n"); 2530 /*rv = sr_ioctl_alarm(sc, (struct bioc_alarm *)bio); */ 2531 break; 2532 2533 case BIOCBLINK: 2534 DNPRINTF(SR_D_IOCTL, "blink\n"); 2535 /*rv = sr_ioctl_blink(sc, (struct bioc_blink *)bio); */ 2536 break; 2537 2538 case BIOCSETSTATE: 2539 DNPRINTF(SR_D_IOCTL, "setstate\n"); 2540 rv = sr_ioctl_setstate(sc, (struct bioc_setstate *)bio); 2541 break; 2542 2543 case BIOCCREATERAID: 2544 DNPRINTF(SR_D_IOCTL, "createraid\n"); 2545 rv = sr_ioctl_createraid(sc, (struct bioc_createraid *)bio, 2546 1, NULL); 2547 break; 2548 2549 case BIOCDELETERAID: 2550 DNPRINTF(SR_D_IOCTL, "deleteraid\n"); 2551 rv = sr_ioctl_deleteraid(sc, sd, (struct bioc_deleteraid *)bio); 2552 break; 2553 2554 case BIOCDISCIPLINE: 2555 DNPRINTF(SR_D_IOCTL, "discipline\n"); 2556 rv = sr_ioctl_discipline(sc, sd, (struct bioc_discipline *)bio); 2557 break; 2558 2559 case BIOCINSTALLBOOT: 2560 DNPRINTF(SR_D_IOCTL, "installboot\n"); 2561 rv = sr_ioctl_installboot(sc, sd, 2562 (struct bioc_installboot *)bio); 2563 break; 2564 2565 default: 2566 DNPRINTF(SR_D_IOCTL, "invalid ioctl\n"); 2567 rv = ENOTTY; 2568 } 2569 2570 sc->sc_status.bs_status = (rv ? BIO_STATUS_ERROR : BIO_STATUS_SUCCESS); 2571 2572 memcpy(&bio->bio_status, &sc->sc_status, sizeof(struct bio_status)); 2573 2574 rw_exit_write(&sc->sc_lock); 2575 2576 return (0); 2577 } 2578 2579 int 2580 sr_ioctl_inq(struct sr_softc *sc, struct bioc_inq *bi) 2581 { 2582 struct sr_discipline *sd; 2583 int vol = 0, disk = 0; 2584 2585 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2586 vol++; 2587 disk += sd->sd_meta->ssdi.ssd_chunk_no; 2588 } 2589 2590 strlcpy(bi->bi_dev, sc->sc_dev.dv_xname, sizeof(bi->bi_dev)); 2591 bi->bi_novol = vol + sc->sc_hotspare_no; 2592 bi->bi_nodisk = disk + sc->sc_hotspare_no; 2593 2594 return (0); 2595 } 2596 2597 int 2598 sr_ioctl_vol(struct sr_softc *sc, struct bioc_vol *bv) 2599 { 2600 int vol = -1, rv = EINVAL; 2601 struct sr_discipline *sd; 2602 struct sr_chunk *hotspare; 2603 2604 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2605 vol++; 2606 if (vol != bv->bv_volid) 2607 continue; 2608 2609 bv->bv_status = sd->sd_vol_status; 2610 bv->bv_size = sd->sd_meta->ssdi.ssd_size << DEV_BSHIFT; 2611 bv->bv_level = sd->sd_meta->ssdi.ssd_level; 2612 bv->bv_nodisk = sd->sd_meta->ssdi.ssd_chunk_no; 2613 2614 #ifdef CRYPTO 2615 if (sd->sd_meta->ssdi.ssd_level == 'C' && 2616 sd->mds.mdd_crypto.key_disk != NULL) 2617 bv->bv_nodisk++; 2618 #endif 2619 if (bv->bv_status == BIOC_SVREBUILD) 2620 bv->bv_percent = sr_rebuild_percent(sd); 2621 2622 strlcpy(bv->bv_dev, sd->sd_meta->ssd_devname, 2623 sizeof(bv->bv_dev)); 2624 strlcpy(bv->bv_vendor, sd->sd_meta->ssdi.ssd_vendor, 2625 sizeof(bv->bv_vendor)); 2626 rv = 0; 2627 goto done; 2628 } 2629 2630 /* Check hotspares list. */ 2631 SLIST_FOREACH(hotspare, &sc->sc_hotspare_list, src_link) { 2632 vol++; 2633 if (vol != bv->bv_volid) 2634 continue; 2635 2636 bv->bv_status = BIOC_SVONLINE; 2637 bv->bv_size = hotspare->src_meta.scmi.scm_size << DEV_BSHIFT; 2638 bv->bv_level = -1; /* Hotspare. */ 2639 bv->bv_nodisk = 1; 2640 strlcpy(bv->bv_dev, hotspare->src_meta.scmi.scm_devname, 2641 sizeof(bv->bv_dev)); 2642 strlcpy(bv->bv_vendor, hotspare->src_meta.scmi.scm_devname, 2643 sizeof(bv->bv_vendor)); 2644 rv = 0; 2645 goto done; 2646 } 2647 2648 done: 2649 return (rv); 2650 } 2651 2652 int 2653 sr_ioctl_disk(struct sr_softc *sc, struct bioc_disk *bd) 2654 { 2655 struct sr_discipline *sd; 2656 struct sr_chunk *src, *hotspare; 2657 int vol = -1, rv = EINVAL; 2658 2659 if (bd->bd_diskid < 0) 2660 goto done; 2661 2662 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2663 vol++; 2664 if (vol != bd->bd_volid) 2665 continue; 2666 2667 if (bd->bd_diskid < sd->sd_meta->ssdi.ssd_chunk_no) 2668 src = sd->sd_vol.sv_chunks[bd->bd_diskid]; 2669 #ifdef CRYPTO 2670 else if (bd->bd_diskid == sd->sd_meta->ssdi.ssd_chunk_no && 2671 sd->sd_meta->ssdi.ssd_level == 'C' && 2672 sd->mds.mdd_crypto.key_disk != NULL) 2673 src = sd->mds.mdd_crypto.key_disk; 2674 #endif 2675 else 2676 break; 2677 2678 bd->bd_status = src->src_meta.scm_status; 2679 bd->bd_size = src->src_meta.scmi.scm_size << DEV_BSHIFT; 2680 bd->bd_channel = vol; 2681 bd->bd_target = bd->bd_diskid; 2682 strlcpy(bd->bd_vendor, src->src_meta.scmi.scm_devname, 2683 sizeof(bd->bd_vendor)); 2684 rv = 0; 2685 goto done; 2686 } 2687 2688 /* Check hotspares list. */ 2689 SLIST_FOREACH(hotspare, &sc->sc_hotspare_list, src_link) { 2690 vol++; 2691 if (vol != bd->bd_volid) 2692 continue; 2693 2694 if (bd->bd_diskid != 0) 2695 break; 2696 2697 bd->bd_status = hotspare->src_meta.scm_status; 2698 bd->bd_size = hotspare->src_meta.scmi.scm_size << DEV_BSHIFT; 2699 bd->bd_channel = vol; 2700 bd->bd_target = bd->bd_diskid; 2701 strlcpy(bd->bd_vendor, hotspare->src_meta.scmi.scm_devname, 2702 sizeof(bd->bd_vendor)); 2703 rv = 0; 2704 goto done; 2705 } 2706 2707 done: 2708 return (rv); 2709 } 2710 2711 int 2712 sr_ioctl_setstate(struct sr_softc *sc, struct bioc_setstate *bs) 2713 { 2714 int rv = EINVAL; 2715 int vol = -1, found, c; 2716 struct sr_discipline *sd; 2717 struct sr_chunk *ch_entry; 2718 struct sr_chunk_head *cl; 2719 2720 if (bs->bs_other_id_type == BIOC_SSOTHER_UNUSED) 2721 goto done; 2722 2723 if (bs->bs_status == BIOC_SSHOTSPARE) { 2724 rv = sr_hotspare(sc, (dev_t)bs->bs_other_id); 2725 goto done; 2726 } 2727 2728 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2729 vol++; 2730 if (vol == bs->bs_volid) 2731 break; 2732 } 2733 if (sd == NULL) 2734 goto done; 2735 2736 switch (bs->bs_status) { 2737 case BIOC_SSOFFLINE: 2738 /* Take chunk offline */ 2739 found = c = 0; 2740 cl = &sd->sd_vol.sv_chunk_list; 2741 SLIST_FOREACH(ch_entry, cl, src_link) { 2742 if (ch_entry->src_dev_mm == bs->bs_other_id) { 2743 found = 1; 2744 break; 2745 } 2746 c++; 2747 } 2748 if (found == 0) { 2749 sr_error(sc, "chunk not part of array"); 2750 goto done; 2751 } 2752 2753 /* XXX: check current state first */ 2754 sd->sd_set_chunk_state(sd, c, BIOC_SDOFFLINE); 2755 2756 if (sr_meta_save(sd, SR_META_DIRTY)) { 2757 sr_error(sc, "could not save metadata for %s", 2758 sd->sd_meta->ssd_devname); 2759 goto done; 2760 } 2761 rv = 0; 2762 break; 2763 2764 case BIOC_SDSCRUB: 2765 break; 2766 2767 case BIOC_SSREBUILD: 2768 rv = sr_rebuild_init(sd, (dev_t)bs->bs_other_id, 0); 2769 break; 2770 2771 default: 2772 sr_error(sc, "unsupported state request %d", bs->bs_status); 2773 } 2774 2775 done: 2776 return (rv); 2777 } 2778 2779 int 2780 sr_chunk_in_use(struct sr_softc *sc, dev_t dev) 2781 { 2782 struct sr_discipline *sd; 2783 struct sr_chunk *chunk; 2784 int i; 2785 2786 DNPRINTF(SR_D_MISC, "%s: sr_chunk_in_use(%d)\n", DEVNAME(sc), dev); 2787 2788 if (dev == NODEV) 2789 return BIOC_SDINVALID; 2790 2791 /* See if chunk is already in use. */ 2792 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 2793 for (i = 0; i < sd->sd_meta->ssdi.ssd_chunk_no; i++) { 2794 chunk = sd->sd_vol.sv_chunks[i]; 2795 if (chunk->src_dev_mm == dev) 2796 return chunk->src_meta.scm_status; 2797 } 2798 } 2799 2800 /* Check hotspares list. */ 2801 SLIST_FOREACH(chunk, &sc->sc_hotspare_list, src_link) 2802 if (chunk->src_dev_mm == dev) 2803 return chunk->src_meta.scm_status; 2804 2805 return BIOC_SDINVALID; 2806 } 2807 2808 int 2809 sr_hotspare(struct sr_softc *sc, dev_t dev) 2810 { 2811 struct sr_discipline *sd = NULL; 2812 struct sr_metadata *sm = NULL; 2813 struct sr_meta_chunk *hm; 2814 struct sr_chunk_head *cl; 2815 struct sr_chunk *chunk, *last, *hotspare = NULL; 2816 struct sr_uuid uuid; 2817 struct disklabel label; 2818 struct vnode *vn; 2819 u_int64_t size; 2820 char devname[32]; 2821 int rv = EINVAL; 2822 int c, part, open = 0; 2823 2824 /* 2825 * Add device to global hotspares list. 2826 */ 2827 2828 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 2829 2830 /* Make sure chunk is not already in use. */ 2831 c = sr_chunk_in_use(sc, dev); 2832 if (c != BIOC_SDINVALID && c != BIOC_SDOFFLINE) { 2833 if (c == BIOC_SDHOTSPARE) 2834 sr_error(sc, "%s is already a hotspare", devname); 2835 else 2836 sr_error(sc, "%s is already in use", devname); 2837 goto done; 2838 } 2839 2840 /* XXX - See if there is an existing degraded volume... */ 2841 2842 /* Open device. */ 2843 if (bdevvp(dev, &vn)) { 2844 sr_error(sc, "sr_hotspare: cannot allocate vnode"); 2845 goto done; 2846 } 2847 if (VOP_OPEN(vn, FREAD | FWRITE, NOCRED, curproc)) { 2848 DNPRINTF(SR_D_META,"%s: sr_hotspare cannot open %s\n", 2849 DEVNAME(sc), devname); 2850 vput(vn); 2851 goto fail; 2852 } 2853 open = 1; /* close dev on error */ 2854 2855 /* Get partition details. */ 2856 part = DISKPART(dev); 2857 if (VOP_IOCTL(vn, DIOCGDINFO, (caddr_t)&label, FREAD, 2858 NOCRED, curproc)) { 2859 DNPRINTF(SR_D_META, "%s: sr_hotspare ioctl failed\n", 2860 DEVNAME(sc)); 2861 VOP_CLOSE(vn, FREAD | FWRITE, NOCRED, curproc); 2862 vput(vn); 2863 goto fail; 2864 } 2865 if (label.d_partitions[part].p_fstype != FS_RAID) { 2866 sr_error(sc, "%s partition not of type RAID (%d)", 2867 devname, label.d_partitions[part].p_fstype); 2868 goto fail; 2869 } 2870 2871 /* Calculate partition size. */ 2872 size = DL_SECTOBLK(&label, DL_GETPSIZE(&label.d_partitions[part])); 2873 if (size <= SR_DATA_OFFSET) { 2874 DNPRINTF(SR_D_META, "%s: %s partition too small\n", DEVNAME(sc), 2875 devname); 2876 goto fail; 2877 } 2878 size -= SR_DATA_OFFSET; 2879 if (size > INT64_MAX) { 2880 DNPRINTF(SR_D_META, "%s: %s partition too large\n", DEVNAME(sc), 2881 devname); 2882 goto fail; 2883 } 2884 2885 /* 2886 * Create and populate chunk metadata. 2887 */ 2888 2889 sr_uuid_generate(&uuid); 2890 hotspare = malloc(sizeof(struct sr_chunk), M_DEVBUF, M_WAITOK | M_ZERO); 2891 2892 hotspare->src_dev_mm = dev; 2893 hotspare->src_vn = vn; 2894 strlcpy(hotspare->src_devname, devname, sizeof(hm->scmi.scm_devname)); 2895 hotspare->src_size = size; 2896 2897 hm = &hotspare->src_meta; 2898 hm->scmi.scm_volid = SR_HOTSPARE_VOLID; 2899 hm->scmi.scm_chunk_id = 0; 2900 hm->scmi.scm_size = size; 2901 hm->scmi.scm_coerced_size = size; 2902 strlcpy(hm->scmi.scm_devname, devname, sizeof(hm->scmi.scm_devname)); 2903 memcpy(&hm->scmi.scm_uuid, &uuid, sizeof(struct sr_uuid)); 2904 2905 sr_checksum(sc, hm, &hm->scm_checksum, 2906 sizeof(struct sr_meta_chunk_invariant)); 2907 2908 hm->scm_status = BIOC_SDHOTSPARE; 2909 2910 /* 2911 * Create and populate our own discipline and metadata. 2912 */ 2913 2914 sm = malloc(sizeof(struct sr_metadata), M_DEVBUF, M_WAITOK | M_ZERO); 2915 sm->ssdi.ssd_magic = SR_MAGIC; 2916 sm->ssdi.ssd_version = SR_META_VERSION; 2917 sm->ssd_ondisk = 0; 2918 sm->ssdi.ssd_vol_flags = 0; 2919 memcpy(&sm->ssdi.ssd_uuid, &uuid, sizeof(struct sr_uuid)); 2920 sm->ssdi.ssd_chunk_no = 1; 2921 sm->ssdi.ssd_volid = SR_HOTSPARE_VOLID; 2922 sm->ssdi.ssd_level = SR_HOTSPARE_LEVEL; 2923 sm->ssdi.ssd_size = size; 2924 sm->ssdi.ssd_secsize = label.d_secsize; 2925 strlcpy(sm->ssdi.ssd_vendor, "OPENBSD", sizeof(sm->ssdi.ssd_vendor)); 2926 snprintf(sm->ssdi.ssd_product, sizeof(sm->ssdi.ssd_product), 2927 "SR %s", "HOTSPARE"); 2928 snprintf(sm->ssdi.ssd_revision, sizeof(sm->ssdi.ssd_revision), 2929 "%03d", SR_META_VERSION); 2930 2931 sd = malloc(sizeof(struct sr_discipline), M_DEVBUF, M_WAITOK | M_ZERO); 2932 sd->sd_sc = sc; 2933 sd->sd_meta = sm; 2934 sd->sd_meta_type = SR_META_F_NATIVE; 2935 sd->sd_vol_status = BIOC_SVONLINE; 2936 strlcpy(sd->sd_name, "HOTSPARE", sizeof(sd->sd_name)); 2937 SLIST_INIT(&sd->sd_meta_opt); 2938 2939 /* Add chunk to volume. */ 2940 sd->sd_vol.sv_chunks = malloc(sizeof(struct sr_chunk *), M_DEVBUF, 2941 M_WAITOK | M_ZERO); 2942 sd->sd_vol.sv_chunks[0] = hotspare; 2943 SLIST_INIT(&sd->sd_vol.sv_chunk_list); 2944 SLIST_INSERT_HEAD(&sd->sd_vol.sv_chunk_list, hotspare, src_link); 2945 2946 /* Save metadata. */ 2947 if (sr_meta_save(sd, SR_META_DIRTY)) { 2948 sr_error(sc, "could not save metadata to %s", devname); 2949 goto fail; 2950 } 2951 2952 /* 2953 * Add chunk to hotspare list. 2954 */ 2955 rw_enter_write(&sc->sc_hs_lock); 2956 cl = &sc->sc_hotspare_list; 2957 if (SLIST_EMPTY(cl)) 2958 SLIST_INSERT_HEAD(cl, hotspare, src_link); 2959 else { 2960 SLIST_FOREACH(chunk, cl, src_link) 2961 last = chunk; 2962 SLIST_INSERT_AFTER(last, hotspare, src_link); 2963 } 2964 sc->sc_hotspare_no++; 2965 rw_exit_write(&sc->sc_hs_lock); 2966 2967 rv = 0; 2968 goto done; 2969 2970 fail: 2971 free(hotspare, M_DEVBUF, 0); 2972 2973 done: 2974 if (sd) 2975 free(sd->sd_vol.sv_chunks, M_DEVBUF, 0); 2976 free(sd, M_DEVBUF, 0); 2977 free(sm, M_DEVBUF, 0); 2978 if (open) { 2979 VOP_CLOSE(vn, FREAD | FWRITE, NOCRED, curproc); 2980 vput(vn); 2981 } 2982 2983 return (rv); 2984 } 2985 2986 void 2987 sr_hotspare_rebuild_callback(void *xsd) 2988 { 2989 struct sr_discipline *sd = xsd; 2990 sr_hotspare_rebuild(sd); 2991 } 2992 2993 void 2994 sr_hotspare_rebuild(struct sr_discipline *sd) 2995 { 2996 struct sr_softc *sc = sd->sd_sc; 2997 struct sr_chunk_head *cl; 2998 struct sr_chunk *hotspare, *chunk = NULL; 2999 struct sr_workunit *wu; 3000 struct sr_ccb *ccb; 3001 int i, s, cid, busy; 3002 3003 /* 3004 * Attempt to locate a hotspare and initiate rebuild. 3005 */ 3006 3007 /* Find first offline chunk. */ 3008 for (cid = 0; cid < sd->sd_meta->ssdi.ssd_chunk_no; cid++) { 3009 if (sd->sd_vol.sv_chunks[cid]->src_meta.scm_status == 3010 BIOC_SDOFFLINE) { 3011 chunk = sd->sd_vol.sv_chunks[cid]; 3012 break; 3013 } 3014 } 3015 if (chunk == NULL) { 3016 printf("%s: no offline chunk found on %s!\n", 3017 DEVNAME(sc), sd->sd_meta->ssd_devname); 3018 return; 3019 } 3020 3021 /* See if we have a suitable hotspare... */ 3022 rw_enter_write(&sc->sc_hs_lock); 3023 cl = &sc->sc_hotspare_list; 3024 SLIST_FOREACH(hotspare, cl, src_link) 3025 if (hotspare->src_size >= chunk->src_size && 3026 hotspare->src_secsize <= sd->sd_meta->ssdi.ssd_secsize) 3027 break; 3028 3029 if (hotspare != NULL) { 3030 3031 printf("%s: %s volume degraded, will attempt to " 3032 "rebuild on hotspare %s\n", DEVNAME(sc), 3033 sd->sd_meta->ssd_devname, hotspare->src_devname); 3034 3035 /* 3036 * Ensure that all pending I/O completes on the failed chunk 3037 * before trying to initiate a rebuild. 3038 */ 3039 i = 0; 3040 do { 3041 busy = 0; 3042 3043 s = splbio(); 3044 TAILQ_FOREACH(wu, &sd->sd_wu_pendq, swu_link) { 3045 TAILQ_FOREACH(ccb, &wu->swu_ccb, ccb_link) { 3046 if (ccb->ccb_target == cid) 3047 busy = 1; 3048 } 3049 } 3050 TAILQ_FOREACH(wu, &sd->sd_wu_defq, swu_link) { 3051 TAILQ_FOREACH(ccb, &wu->swu_ccb, ccb_link) { 3052 if (ccb->ccb_target == cid) 3053 busy = 1; 3054 } 3055 } 3056 splx(s); 3057 3058 if (busy) { 3059 tsleep(sd, PRIBIO, "sr_hotspare", hz); 3060 i++; 3061 } 3062 3063 } while (busy && i < 120); 3064 3065 DNPRINTF(SR_D_META, "%s: waited %i seconds for I/O to " 3066 "complete on failed chunk %s\n", DEVNAME(sc), 3067 i, chunk->src_devname); 3068 3069 if (busy) { 3070 printf("%s: pending I/O failed to complete on " 3071 "failed chunk %s, hotspare rebuild aborted...\n", 3072 DEVNAME(sc), chunk->src_devname); 3073 goto done; 3074 } 3075 3076 s = splbio(); 3077 rw_enter_write(&sc->sc_lock); 3078 bio_status_init(&sc->sc_status, &sc->sc_dev); 3079 if (sr_rebuild_init(sd, hotspare->src_dev_mm, 1) == 0) { 3080 3081 /* Remove hotspare from available list. */ 3082 sc->sc_hotspare_no--; 3083 SLIST_REMOVE(cl, hotspare, sr_chunk, src_link); 3084 free(hotspare, M_DEVBUF, 0); 3085 3086 } 3087 rw_exit_write(&sc->sc_lock); 3088 splx(s); 3089 } 3090 done: 3091 rw_exit_write(&sc->sc_hs_lock); 3092 } 3093 3094 int 3095 sr_rebuild_init(struct sr_discipline *sd, dev_t dev, int hotspare) 3096 { 3097 struct sr_softc *sc = sd->sd_sc; 3098 struct sr_chunk *chunk = NULL; 3099 struct sr_meta_chunk *meta; 3100 struct disklabel label; 3101 struct vnode *vn; 3102 u_int64_t size; 3103 int64_t csize; 3104 char devname[32]; 3105 int rv = EINVAL, open = 0; 3106 int cid, i, part, status; 3107 3108 /* 3109 * Attempt to initiate a rebuild onto the specified device. 3110 */ 3111 3112 if (!(sd->sd_capabilities & SR_CAP_REBUILD)) { 3113 sr_error(sc, "discipline does not support rebuild"); 3114 goto done; 3115 } 3116 3117 /* make sure volume is in the right state */ 3118 if (sd->sd_vol_status == BIOC_SVREBUILD) { 3119 sr_error(sc, "rebuild already in progress"); 3120 goto done; 3121 } 3122 if (sd->sd_vol_status != BIOC_SVDEGRADED) { 3123 sr_error(sc, "volume not degraded"); 3124 goto done; 3125 } 3126 3127 /* Find first offline chunk. */ 3128 for (cid = 0; cid < sd->sd_meta->ssdi.ssd_chunk_no; cid++) { 3129 if (sd->sd_vol.sv_chunks[cid]->src_meta.scm_status == 3130 BIOC_SDOFFLINE) { 3131 chunk = sd->sd_vol.sv_chunks[cid]; 3132 break; 3133 } 3134 } 3135 if (chunk == NULL) { 3136 sr_error(sc, "no offline chunks available to rebuild"); 3137 goto done; 3138 } 3139 3140 /* Get coerced size from another online chunk. */ 3141 csize = 0; 3142 for (i = 0; i < sd->sd_meta->ssdi.ssd_chunk_no; i++) { 3143 if (sd->sd_vol.sv_chunks[i]->src_meta.scm_status == 3144 BIOC_SDONLINE) { 3145 meta = &sd->sd_vol.sv_chunks[i]->src_meta; 3146 csize = meta->scmi.scm_coerced_size; 3147 break; 3148 } 3149 } 3150 if (csize == 0) { 3151 sr_error(sc, "no online chunks available for rebuild"); 3152 goto done; 3153 } 3154 3155 sr_meta_getdevname(sc, dev, devname, sizeof(devname)); 3156 if (bdevvp(dev, &vn)) { 3157 printf("%s: sr_rebuild_init: can't allocate vnode\n", 3158 DEVNAME(sc)); 3159 goto done; 3160 } 3161 if (VOP_OPEN(vn, FREAD | FWRITE, NOCRED, curproc)) { 3162 DNPRINTF(SR_D_META,"%s: sr_ioctl_setstate can't " 3163 "open %s\n", DEVNAME(sc), devname); 3164 vput(vn); 3165 goto done; 3166 } 3167 open = 1; /* close dev on error */ 3168 3169 /* Get disklabel and check partition. */ 3170 part = DISKPART(dev); 3171 if (VOP_IOCTL(vn, DIOCGDINFO, (caddr_t)&label, FREAD, 3172 NOCRED, curproc)) { 3173 DNPRINTF(SR_D_META, "%s: sr_ioctl_setstate ioctl failed\n", 3174 DEVNAME(sc)); 3175 goto done; 3176 } 3177 if (label.d_partitions[part].p_fstype != FS_RAID) { 3178 sr_error(sc, "%s partition not of type RAID (%d)", 3179 devname, label.d_partitions[part].p_fstype); 3180 goto done; 3181 } 3182 3183 /* Is the partition large enough? */ 3184 size = DL_SECTOBLK(&label, DL_GETPSIZE(&label.d_partitions[part])); 3185 if (size <= sd->sd_meta->ssd_data_blkno) { 3186 sr_error(sc, "%s: %s partition too small", DEVNAME(sc), 3187 devname); 3188 goto done; 3189 } 3190 size -= sd->sd_meta->ssd_data_blkno; 3191 if (size > INT64_MAX) { 3192 sr_error(sc, "%s: %s partition too large", DEVNAME(sc), 3193 devname); 3194 goto done; 3195 } 3196 if (size < csize) { 3197 sr_error(sc, "%s partition too small, at least %lld bytes " 3198 "required", devname, (long long)(csize << DEV_BSHIFT)); 3199 goto done; 3200 } else if (size > csize) 3201 sr_warn(sc, "%s partition too large, wasting %lld bytes", 3202 devname, (long long)((size - csize) << DEV_BSHIFT)); 3203 if (label.d_secsize > sd->sd_meta->ssdi.ssd_secsize) { 3204 sr_error(sc, "%s sector size too large, <= %u bytes " 3205 "required", devname, sd->sd_meta->ssdi.ssd_secsize); 3206 goto done; 3207 } 3208 3209 /* Ensure that this chunk is not already in use. */ 3210 status = sr_chunk_in_use(sc, dev); 3211 if (status != BIOC_SDINVALID && status != BIOC_SDOFFLINE && 3212 !(hotspare && status == BIOC_SDHOTSPARE)) { 3213 sr_error(sc, "%s is already in use", devname); 3214 goto done; 3215 } 3216 3217 /* Reset rebuild counter since we rebuilding onto a new chunk. */ 3218 sd->sd_meta->ssd_rebuild = 0; 3219 3220 open = 0; /* leave dev open from here on out */ 3221 3222 /* Fix up chunk. */ 3223 memcpy(chunk->src_duid, label.d_uid, sizeof(chunk->src_duid)); 3224 chunk->src_dev_mm = dev; 3225 chunk->src_vn = vn; 3226 3227 /* Reconstruct metadata. */ 3228 meta = &chunk->src_meta; 3229 meta->scmi.scm_volid = sd->sd_meta->ssdi.ssd_volid; 3230 meta->scmi.scm_chunk_id = cid; 3231 strlcpy(meta->scmi.scm_devname, devname, 3232 sizeof(meta->scmi.scm_devname)); 3233 meta->scmi.scm_size = size; 3234 meta->scmi.scm_coerced_size = csize; 3235 memcpy(&meta->scmi.scm_uuid, &sd->sd_meta->ssdi.ssd_uuid, 3236 sizeof(meta->scmi.scm_uuid)); 3237 sr_checksum(sc, meta, &meta->scm_checksum, 3238 sizeof(struct sr_meta_chunk_invariant)); 3239 3240 sd->sd_set_chunk_state(sd, cid, BIOC_SDREBUILD); 3241 3242 if (sr_meta_save(sd, SR_META_DIRTY)) { 3243 sr_error(sc, "could not save metadata to %s", devname); 3244 open = 1; 3245 goto done; 3246 } 3247 3248 sr_warn(sc, "rebuild of %s started on %s", 3249 sd->sd_meta->ssd_devname, devname); 3250 3251 sd->sd_reb_abort = 0; 3252 kthread_create_deferred(sr_rebuild_start, sd); 3253 3254 rv = 0; 3255 done: 3256 if (open) { 3257 VOP_CLOSE(vn, FREAD | FWRITE, NOCRED, curproc); 3258 vput(vn); 3259 } 3260 3261 return (rv); 3262 } 3263 3264 int 3265 sr_rebuild_percent(struct sr_discipline *sd) 3266 { 3267 daddr_t rb, sz; 3268 3269 sz = sd->sd_meta->ssdi.ssd_size; 3270 rb = sd->sd_meta->ssd_rebuild; 3271 3272 if (rb > 0) 3273 return (100 - ((sz * 100 - rb * 100) / sz) - 1); 3274 3275 return (0); 3276 } 3277 3278 void 3279 sr_roam_chunks(struct sr_discipline *sd) 3280 { 3281 struct sr_softc *sc = sd->sd_sc; 3282 struct sr_chunk *chunk; 3283 struct sr_meta_chunk *meta; 3284 int roamed = 0; 3285 3286 /* Have any chunks roamed? */ 3287 SLIST_FOREACH(chunk, &sd->sd_vol.sv_chunk_list, src_link) { 3288 meta = &chunk->src_meta; 3289 if (strncmp(meta->scmi.scm_devname, chunk->src_devname, 3290 sizeof(meta->scmi.scm_devname))) { 3291 3292 printf("%s: roaming device %s -> %s\n", DEVNAME(sc), 3293 meta->scmi.scm_devname, chunk->src_devname); 3294 3295 strlcpy(meta->scmi.scm_devname, chunk->src_devname, 3296 sizeof(meta->scmi.scm_devname)); 3297 3298 roamed++; 3299 } 3300 } 3301 3302 if (roamed) 3303 sr_meta_save(sd, SR_META_DIRTY); 3304 } 3305 3306 int 3307 sr_ioctl_createraid(struct sr_softc *sc, struct bioc_createraid *bc, 3308 int user, void *data) 3309 { 3310 struct sr_meta_opt_item *omi; 3311 struct sr_chunk_head *cl; 3312 struct sr_discipline *sd = NULL; 3313 struct sr_chunk *ch_entry; 3314 struct scsi_link *link; 3315 struct device *dev; 3316 char *uuid, devname[32]; 3317 dev_t *dt = NULL; 3318 int i, no_chunk, rv = EINVAL, target, vol; 3319 int no_meta; 3320 3321 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_createraid(%d)\n", 3322 DEVNAME(sc), user); 3323 3324 /* user input */ 3325 if (bc->bc_dev_list_len > BIOC_CRMAXLEN) 3326 goto unwind; 3327 3328 dt = malloc(bc->bc_dev_list_len, M_DEVBUF, M_WAITOK | M_ZERO); 3329 if (user) { 3330 if (copyin(bc->bc_dev_list, dt, bc->bc_dev_list_len) != 0) 3331 goto unwind; 3332 } else 3333 memcpy(dt, bc->bc_dev_list, bc->bc_dev_list_len); 3334 3335 /* Initialise discipline. */ 3336 sd = malloc(sizeof(struct sr_discipline), M_DEVBUF, M_WAITOK | M_ZERO); 3337 sd->sd_sc = sc; 3338 SLIST_INIT(&sd->sd_meta_opt); 3339 sd->sd_taskq = taskq_create("srdis", 1, IPL_BIO, 0); 3340 if (sd->sd_taskq == NULL) { 3341 sr_error(sc, "could not create discipline taskq"); 3342 goto unwind; 3343 } 3344 if (sr_discipline_init(sd, bc->bc_level)) { 3345 sr_error(sc, "could not initialize discipline"); 3346 goto unwind; 3347 } 3348 3349 no_chunk = bc->bc_dev_list_len / sizeof(dev_t); 3350 cl = &sd->sd_vol.sv_chunk_list; 3351 SLIST_INIT(cl); 3352 3353 /* Ensure that chunks are not already in use. */ 3354 for (i = 0; i < no_chunk; i++) { 3355 if (sr_chunk_in_use(sc, dt[i]) != BIOC_SDINVALID) { 3356 sr_meta_getdevname(sc, dt[i], devname, sizeof(devname)); 3357 sr_error(sc, "chunk %s already in use", devname); 3358 goto unwind; 3359 } 3360 } 3361 3362 sd->sd_meta_type = sr_meta_probe(sd, dt, no_chunk); 3363 if (sd->sd_meta_type == SR_META_F_INVALID) { 3364 sr_error(sc, "invalid metadata format"); 3365 goto unwind; 3366 } 3367 3368 if (sr_meta_attach(sd, no_chunk, bc->bc_flags & BIOC_SCFORCE)) 3369 goto unwind; 3370 3371 /* force the raid volume by clearing metadata region */ 3372 if (bc->bc_flags & BIOC_SCFORCE) { 3373 /* make sure disk isn't up and running */ 3374 if (sr_meta_read(sd)) 3375 if (sr_already_assembled(sd)) { 3376 uuid = sr_uuid_format( 3377 &sd->sd_meta->ssdi.ssd_uuid); 3378 sr_error(sc, "disk %s is currently in use; " 3379 "cannot force create", uuid); 3380 free(uuid, M_DEVBUF, 0); 3381 goto unwind; 3382 } 3383 3384 if (sr_meta_clear(sd)) { 3385 sr_error(sc, "failed to clear metadata"); 3386 goto unwind; 3387 } 3388 } 3389 3390 no_meta = sr_meta_read(sd); 3391 if (no_meta == -1) { 3392 3393 /* Corrupt metadata on one or more chunks. */ 3394 sr_error(sc, "one of the chunks has corrupt metadata; " 3395 "aborting assembly"); 3396 goto unwind; 3397 3398 } else if (no_meta == 0) { 3399 3400 /* Initialise volume and chunk metadata. */ 3401 sr_meta_init(sd, bc->bc_level, no_chunk); 3402 sd->sd_vol_status = BIOC_SVONLINE; 3403 sd->sd_meta_flags = bc->bc_flags & BIOC_SCNOAUTOASSEMBLE; 3404 if (sd->sd_create) { 3405 if ((i = sd->sd_create(sd, bc, no_chunk, 3406 sd->sd_vol.sv_chunk_minsz))) { 3407 rv = i; 3408 goto unwind; 3409 } 3410 } 3411 sr_meta_init_complete(sd); 3412 3413 DNPRINTF(SR_D_IOCTL, 3414 "%s: sr_ioctl_createraid: vol_size: %lld\n", 3415 DEVNAME(sc), sd->sd_meta->ssdi.ssd_size); 3416 3417 /* Warn if we've wasted chunk space due to coercing. */ 3418 if ((sd->sd_capabilities & SR_CAP_NON_COERCED) == 0 && 3419 sd->sd_vol.sv_chunk_minsz != sd->sd_vol.sv_chunk_maxsz) 3420 sr_warn(sc, "chunk sizes are not equal; up to %llu " 3421 "blocks wasted per chunk", 3422 sd->sd_vol.sv_chunk_maxsz - 3423 sd->sd_vol.sv_chunk_minsz); 3424 3425 } else { 3426 3427 /* Ensure metadata level matches requested assembly level. */ 3428 if (sd->sd_meta->ssdi.ssd_level != bc->bc_level) { 3429 sr_error(sc, "volume level does not match metadata " 3430 "level"); 3431 goto unwind; 3432 } 3433 3434 if (sr_already_assembled(sd)) { 3435 uuid = sr_uuid_format(&sd->sd_meta->ssdi.ssd_uuid); 3436 sr_error(sc, "disk %s already assembled", uuid); 3437 free(uuid, M_DEVBUF, 0); 3438 goto unwind; 3439 } 3440 3441 if (user == 0 && sd->sd_meta_flags & BIOC_SCNOAUTOASSEMBLE) { 3442 DNPRINTF(SR_D_META, "%s: disk not auto assembled from " 3443 "metadata\n", DEVNAME(sc)); 3444 goto unwind; 3445 } 3446 3447 if (no_meta != no_chunk) 3448 sr_warn(sc, "trying to bring up %s degraded", 3449 sd->sd_meta->ssd_devname); 3450 3451 if (sd->sd_meta->ssd_meta_flags & SR_META_DIRTY) 3452 sr_warn(sc, "%s was not shutdown properly", 3453 sd->sd_meta->ssd_devname); 3454 3455 SLIST_FOREACH(omi, &sd->sd_meta_opt, omi_link) 3456 if (sd->sd_meta_opt_handler == NULL || 3457 sd->sd_meta_opt_handler(sd, omi->omi_som) != 0) 3458 sr_meta_opt_handler(sd, omi->omi_som); 3459 3460 if (sd->sd_assemble) { 3461 if ((i = sd->sd_assemble(sd, bc, no_chunk, data))) { 3462 rv = i; 3463 goto unwind; 3464 } 3465 } 3466 3467 DNPRINTF(SR_D_META, "%s: disk assembled from metadata\n", 3468 DEVNAME(sc)); 3469 3470 } 3471 3472 /* Metadata MUST be fully populated by this point. */ 3473 TAILQ_INSERT_TAIL(&sc->sc_dis_list, sd, sd_link); 3474 3475 /* Allocate all resources. */ 3476 if ((rv = sd->sd_alloc_resources(sd))) 3477 goto unwind; 3478 3479 /* Adjust flags if necessary. */ 3480 if ((sd->sd_capabilities & SR_CAP_AUTO_ASSEMBLE) && 3481 (bc->bc_flags & BIOC_SCNOAUTOASSEMBLE) != 3482 (sd->sd_meta->ssdi.ssd_vol_flags & BIOC_SCNOAUTOASSEMBLE)) { 3483 sd->sd_meta->ssdi.ssd_vol_flags &= ~BIOC_SCNOAUTOASSEMBLE; 3484 sd->sd_meta->ssdi.ssd_vol_flags |= 3485 bc->bc_flags & BIOC_SCNOAUTOASSEMBLE; 3486 } 3487 3488 if (sd->sd_capabilities & SR_CAP_SYSTEM_DISK) { 3489 /* Initialise volume state. */ 3490 sd->sd_set_vol_state(sd); 3491 if (sd->sd_vol_status == BIOC_SVOFFLINE) { 3492 sr_error(sc, "%s is offline, will not be brought " 3493 "online", sd->sd_meta->ssd_devname); 3494 goto unwind; 3495 } 3496 3497 /* Setup SCSI iopool. */ 3498 scsi_iopool_init(&sd->sd_iopool, sd, sr_wu_get, sr_wu_put); 3499 3500 /* 3501 * All checks passed - return ENXIO if volume cannot be created. 3502 */ 3503 rv = ENXIO; 3504 3505 /* 3506 * Find a free target. 3507 * 3508 * XXX: We reserve sd_target == 0 to indicate the 3509 * discipline is not linked into sc->sc_targets, so begin 3510 * the search with target = 1. 3511 */ 3512 for (target = 1; target < SR_MAX_LD; target++) 3513 if (sc->sc_targets[target] == NULL) 3514 break; 3515 if (target == SR_MAX_LD) { 3516 sr_error(sc, "no free target for %s", 3517 sd->sd_meta->ssd_devname); 3518 goto unwind; 3519 } 3520 3521 /* Clear sense data. */ 3522 bzero(&sd->sd_scsi_sense, sizeof(sd->sd_scsi_sense)); 3523 3524 /* Attach discipline and get midlayer to probe it. */ 3525 sd->sd_target = target; 3526 sc->sc_targets[target] = sd; 3527 if (scsi_probe_lun(sc->sc_scsibus, target, 0) != 0) { 3528 sr_error(sc, "scsi_probe_lun failed"); 3529 sc->sc_targets[target] = NULL; 3530 sd->sd_target = 0; 3531 goto unwind; 3532 } 3533 3534 link = scsi_get_link(sc->sc_scsibus, target, 0); 3535 if (link == NULL) 3536 goto unwind; 3537 3538 dev = link->device_softc; 3539 DNPRINTF(SR_D_IOCTL, "%s: sr device added: %s at target %d\n", 3540 DEVNAME(sc), dev->dv_xname, sd->sd_target); 3541 3542 /* XXX - Count volumes, not targets. */ 3543 for (i = 0, vol = -1; i <= sd->sd_target; i++) 3544 if (sc->sc_targets[i]) 3545 vol++; 3546 3547 rv = 0; 3548 3549 if (sd->sd_meta->ssd_devname[0] != '\0' && 3550 strncmp(sd->sd_meta->ssd_devname, dev->dv_xname, 3551 sizeof(dev->dv_xname))) 3552 sr_warn(sc, "volume %s is roaming, it used to be %s, " 3553 "updating metadata", dev->dv_xname, 3554 sd->sd_meta->ssd_devname); 3555 3556 /* Populate remaining volume metadata. */ 3557 sd->sd_meta->ssdi.ssd_volid = vol; 3558 strlcpy(sd->sd_meta->ssd_devname, dev->dv_xname, 3559 sizeof(sd->sd_meta->ssd_devname)); 3560 3561 sr_info(sc, "%s volume attached as %s", 3562 sd->sd_name, sd->sd_meta->ssd_devname); 3563 3564 /* Update device name on any roaming chunks. */ 3565 sr_roam_chunks(sd); 3566 3567 #ifndef SMALL_KERNEL 3568 if (sr_sensors_create(sd)) 3569 sr_warn(sc, "unable to create sensor for %s", 3570 dev->dv_xname); 3571 #endif /* SMALL_KERNEL */ 3572 } else { 3573 /* This volume does not attach as a system disk. */ 3574 ch_entry = SLIST_FIRST(cl); /* XXX */ 3575 strlcpy(sd->sd_meta->ssd_devname, ch_entry->src_devname, 3576 sizeof(sd->sd_meta->ssd_devname)); 3577 3578 if (sd->sd_start_discipline(sd)) 3579 goto unwind; 3580 } 3581 3582 /* Save current metadata to disk. */ 3583 rv = sr_meta_save(sd, SR_META_DIRTY); 3584 3585 if (sd->sd_vol_status == BIOC_SVREBUILD) 3586 kthread_create_deferred(sr_rebuild_start, sd); 3587 3588 sd->sd_ready = 1; 3589 3590 free(dt, M_DEVBUF, bc->bc_dev_list_len); 3591 3592 return (rv); 3593 3594 unwind: 3595 free(dt, M_DEVBUF, bc->bc_dev_list_len); 3596 3597 sr_discipline_shutdown(sd, 0); 3598 3599 if (rv == EAGAIN) 3600 rv = 0; 3601 3602 return (rv); 3603 } 3604 3605 int 3606 sr_ioctl_deleteraid(struct sr_softc *sc, struct sr_discipline *sd, 3607 struct bioc_deleteraid *bd) 3608 { 3609 int rv = 1; 3610 3611 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_deleteraid %s\n", 3612 DEVNAME(sc), bd->bd_dev); 3613 3614 if (sd == NULL) { 3615 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 3616 if (!strncmp(sd->sd_meta->ssd_devname, bd->bd_dev, 3617 sizeof(sd->sd_meta->ssd_devname))) 3618 break; 3619 } 3620 if (sd == NULL) { 3621 sr_error(sc, "volume %s not found", bd->bd_dev); 3622 goto bad; 3623 } 3624 } 3625 3626 sd->sd_deleted = 1; 3627 sd->sd_meta->ssdi.ssd_vol_flags = BIOC_SCNOAUTOASSEMBLE; 3628 sr_discipline_shutdown(sd, 1); 3629 3630 rv = 0; 3631 bad: 3632 return (rv); 3633 } 3634 3635 int 3636 sr_ioctl_discipline(struct sr_softc *sc, struct sr_discipline *sd, 3637 struct bioc_discipline *bd) 3638 { 3639 int rv = 1; 3640 3641 /* Dispatch a discipline specific ioctl. */ 3642 3643 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_discipline %s\n", DEVNAME(sc), 3644 bd->bd_dev); 3645 3646 if (sd == NULL) { 3647 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 3648 if (!strncmp(sd->sd_meta->ssd_devname, bd->bd_dev, 3649 sizeof(sd->sd_meta->ssd_devname))) 3650 break; 3651 } 3652 if (sd == NULL) { 3653 sr_error(sc, "volume %s not found", bd->bd_dev); 3654 goto bad; 3655 } 3656 } 3657 3658 if (sd->sd_ioctl_handler) 3659 rv = sd->sd_ioctl_handler(sd, bd); 3660 3661 bad: 3662 return (rv); 3663 } 3664 3665 int 3666 sr_ioctl_installboot(struct sr_softc *sc, struct sr_discipline *sd, 3667 struct bioc_installboot *bb) 3668 { 3669 void *bootblk = NULL, *bootldr = NULL; 3670 struct sr_chunk *chunk; 3671 struct sr_meta_opt_item *omi; 3672 struct sr_meta_boot *sbm; 3673 struct disk *dk; 3674 u_int32_t bbs, bls, secsize; 3675 u_char duid[8]; 3676 int rv = EINVAL; 3677 int i; 3678 3679 DNPRINTF(SR_D_IOCTL, "%s: sr_ioctl_installboot %s\n", DEVNAME(sc), 3680 bb->bb_dev); 3681 3682 if (sd == NULL) { 3683 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 3684 if (!strncmp(sd->sd_meta->ssd_devname, bb->bb_dev, 3685 sizeof(sd->sd_meta->ssd_devname))) 3686 break; 3687 } 3688 if (sd == NULL) { 3689 sr_error(sc, "volume %s not found", bb->bb_dev); 3690 goto done; 3691 } 3692 } 3693 3694 bzero(duid, sizeof(duid)); 3695 TAILQ_FOREACH(dk, &disklist, dk_link) 3696 if (!strncmp(dk->dk_name, bb->bb_dev, sizeof(bb->bb_dev))) 3697 break; 3698 if (dk == NULL || dk->dk_label == NULL || 3699 (dk->dk_flags & DKF_LABELVALID) == 0 || 3700 bcmp(dk->dk_label->d_uid, &duid, sizeof(duid)) == 0) { 3701 sr_error(sc, "failed to get DUID for softraid volume"); 3702 goto done; 3703 } 3704 memcpy(duid, dk->dk_label->d_uid, sizeof(duid)); 3705 3706 /* Ensure that boot storage area is large enough. */ 3707 if (sd->sd_meta->ssd_data_blkno < (SR_BOOT_OFFSET + SR_BOOT_SIZE)) { 3708 sr_error(sc, "insufficient boot storage"); 3709 goto done; 3710 } 3711 3712 if (bb->bb_bootblk_size > SR_BOOT_BLOCKS_SIZE * DEV_BSIZE) 3713 goto done; 3714 3715 if (bb->bb_bootldr_size > SR_BOOT_LOADER_SIZE * DEV_BSIZE) 3716 goto done; 3717 3718 secsize = sd->sd_meta->ssdi.ssd_secsize; 3719 3720 /* Copy in boot block. */ 3721 bbs = howmany(bb->bb_bootblk_size, secsize) * secsize; 3722 bootblk = malloc(bbs, M_DEVBUF, M_WAITOK | M_ZERO); 3723 if (copyin(bb->bb_bootblk, bootblk, bb->bb_bootblk_size) != 0) 3724 goto done; 3725 3726 /* Copy in boot loader. */ 3727 bls = howmany(bb->bb_bootldr_size, secsize) * secsize; 3728 bootldr = malloc(bls, M_DEVBUF, M_WAITOK | M_ZERO); 3729 if (copyin(bb->bb_bootldr, bootldr, bb->bb_bootldr_size) != 0) 3730 goto done; 3731 3732 /* Create or update optional meta for bootable volumes. */ 3733 SLIST_FOREACH(omi, &sd->sd_meta_opt, omi_link) 3734 if (omi->omi_som->som_type == SR_OPT_BOOT) 3735 break; 3736 if (omi == NULL) { 3737 omi = malloc(sizeof(struct sr_meta_opt_item), M_DEVBUF, 3738 M_WAITOK | M_ZERO); 3739 omi->omi_som = malloc(sizeof(struct sr_meta_crypto), M_DEVBUF, 3740 M_WAITOK | M_ZERO); 3741 omi->omi_som->som_type = SR_OPT_BOOT; 3742 omi->omi_som->som_length = sizeof(struct sr_meta_boot); 3743 SLIST_INSERT_HEAD(&sd->sd_meta_opt, omi, omi_link); 3744 sd->sd_meta->ssdi.ssd_opt_no++; 3745 } 3746 sbm = (struct sr_meta_boot *)omi->omi_som; 3747 3748 memcpy(sbm->sbm_root_duid, duid, sizeof(sbm->sbm_root_duid)); 3749 bzero(&sbm->sbm_boot_duid, sizeof(sbm->sbm_boot_duid)); 3750 sbm->sbm_bootblk_size = bbs; 3751 sbm->sbm_bootldr_size = bls; 3752 3753 DNPRINTF(SR_D_IOCTL, "sr_ioctl_installboot: root duid is " 3754 "%02x%02x%02x%02x%02x%02x%02x%02x\n", sbm->sbm_root_duid[0], 3755 sbm->sbm_root_duid[1], sbm->sbm_root_duid[2], sbm->sbm_root_duid[3], 3756 sbm->sbm_root_duid[4], sbm->sbm_root_duid[5], sbm->sbm_root_duid[6], 3757 sbm->sbm_root_duid[7]); 3758 3759 /* Save boot block and boot loader to each chunk. */ 3760 for (i = 0; i < sd->sd_meta->ssdi.ssd_chunk_no; i++) { 3761 3762 chunk = sd->sd_vol.sv_chunks[i]; 3763 if (chunk->src_meta.scm_status != BIOC_SDONLINE && 3764 chunk->src_meta.scm_status != BIOC_SDREBUILD) 3765 continue; 3766 3767 if (i < SR_MAX_BOOT_DISKS) 3768 memcpy(&sbm->sbm_boot_duid[i], chunk->src_duid, 3769 sizeof(sbm->sbm_boot_duid[i])); 3770 3771 /* Save boot blocks. */ 3772 DNPRINTF(SR_D_IOCTL, 3773 "sr_ioctl_installboot: saving boot block to %s " 3774 "(%u bytes)\n", chunk->src_devname, bbs); 3775 3776 if (sr_rw(sc, chunk->src_dev_mm, bootblk, bbs, 3777 SR_BOOT_BLOCKS_OFFSET, B_WRITE)) { 3778 sr_error(sc, "failed to write boot block", DEVNAME(sc)); 3779 goto done; 3780 } 3781 3782 /* Save boot loader.*/ 3783 DNPRINTF(SR_D_IOCTL, 3784 "sr_ioctl_installboot: saving boot loader to %s " 3785 "(%u bytes)\n", chunk->src_devname, bls); 3786 3787 if (sr_rw(sc, chunk->src_dev_mm, bootldr, bls, 3788 SR_BOOT_LOADER_OFFSET, B_WRITE)) { 3789 sr_error(sc, "failed to write boot loader"); 3790 goto done; 3791 } 3792 } 3793 3794 /* XXX - Install boot block on disk - MD code. */ 3795 3796 /* Mark volume as bootable and save metadata. */ 3797 sd->sd_meta->ssdi.ssd_vol_flags |= BIOC_SCBOOTABLE; 3798 if (sr_meta_save(sd, SR_META_DIRTY)) { 3799 sr_error(sc, "could not save metadata to %s", DEVNAME(sc)); 3800 goto done; 3801 } 3802 3803 rv = 0; 3804 3805 done: 3806 free(bootblk, M_DEVBUF, 0); 3807 free(bootldr, M_DEVBUF, 0); 3808 3809 return (rv); 3810 } 3811 3812 void 3813 sr_chunks_unwind(struct sr_softc *sc, struct sr_chunk_head *cl) 3814 { 3815 struct sr_chunk *ch_entry, *ch_next; 3816 3817 DNPRINTF(SR_D_IOCTL, "%s: sr_chunks_unwind\n", DEVNAME(sc)); 3818 3819 if (!cl) 3820 return; 3821 3822 for (ch_entry = SLIST_FIRST(cl); ch_entry != NULL; ch_entry = ch_next) { 3823 ch_next = SLIST_NEXT(ch_entry, src_link); 3824 3825 DNPRINTF(SR_D_IOCTL, "%s: sr_chunks_unwind closing: %s\n", 3826 DEVNAME(sc), ch_entry->src_devname); 3827 if (ch_entry->src_vn) { 3828 /* 3829 * XXX - explicitly lock the vnode until we can resolve 3830 * the problem introduced by vnode aliasing... specfs 3831 * has no locking, whereas ufs/ffs does! 3832 */ 3833 vn_lock(ch_entry->src_vn, LK_EXCLUSIVE | 3834 LK_RETRY, curproc); 3835 VOP_CLOSE(ch_entry->src_vn, FREAD | FWRITE, NOCRED, 3836 curproc); 3837 vput(ch_entry->src_vn); 3838 } 3839 free(ch_entry, M_DEVBUF, 0); 3840 } 3841 SLIST_INIT(cl); 3842 } 3843 3844 void 3845 sr_discipline_free(struct sr_discipline *sd) 3846 { 3847 struct sr_softc *sc; 3848 struct sr_discipline *sdtmp1; 3849 struct sr_meta_opt_head *som; 3850 struct sr_meta_opt_item *omi, *omi_next; 3851 3852 if (!sd) 3853 return; 3854 3855 sc = sd->sd_sc; 3856 3857 DNPRINTF(SR_D_DIS, "%s: sr_discipline_free %s\n", 3858 DEVNAME(sc), 3859 sd->sd_meta ? sd->sd_meta->ssd_devname : "nodev"); 3860 if (sd->sd_free_resources) 3861 sd->sd_free_resources(sd); 3862 free(sd->sd_vol.sv_chunks, M_DEVBUF, 0); 3863 free(sd->sd_meta, M_DEVBUF, 0); 3864 free(sd->sd_meta_foreign, M_DEVBUF, 0); 3865 3866 som = &sd->sd_meta_opt; 3867 for (omi = SLIST_FIRST(som); omi != NULL; omi = omi_next) { 3868 omi_next = SLIST_NEXT(omi, omi_link); 3869 free(omi->omi_som, M_DEVBUF, 0); 3870 free(omi, M_DEVBUF, 0); 3871 } 3872 3873 if (sd->sd_target != 0) { 3874 KASSERT(sc->sc_targets[sd->sd_target] == sd); 3875 sc->sc_targets[sd->sd_target] = NULL; 3876 } 3877 3878 TAILQ_FOREACH(sdtmp1, &sc->sc_dis_list, sd_link) { 3879 if (sdtmp1 == sd) 3880 break; 3881 } 3882 if (sdtmp1 != NULL) 3883 TAILQ_REMOVE(&sc->sc_dis_list, sd, sd_link); 3884 3885 explicit_bzero(sd, sizeof *sd); 3886 free(sd, M_DEVBUF, 0); 3887 } 3888 3889 void 3890 sr_discipline_shutdown(struct sr_discipline *sd, int meta_save) 3891 { 3892 struct sr_softc *sc; 3893 int s; 3894 3895 if (!sd) 3896 return; 3897 sc = sd->sd_sc; 3898 3899 DNPRINTF(SR_D_DIS, "%s: sr_discipline_shutdown %s\n", DEVNAME(sc), 3900 sd->sd_meta ? sd->sd_meta->ssd_devname : "nodev"); 3901 3902 /* If rebuilding, abort rebuild and drain I/O. */ 3903 if (sd->sd_reb_active) { 3904 sd->sd_reb_abort = 1; 3905 while (sd->sd_reb_active) 3906 tsleep(sd, PWAIT, "sr_shutdown", 1); 3907 } 3908 3909 if (meta_save) 3910 sr_meta_save(sd, 0); 3911 3912 s = splbio(); 3913 3914 sd->sd_ready = 0; 3915 3916 /* make sure there isn't a sync pending and yield */ 3917 wakeup(sd); 3918 while (sd->sd_sync || sd->sd_must_flush) 3919 if (tsleep(&sd->sd_sync, MAXPRI, "sr_down", 60 * hz) == 3920 EWOULDBLOCK) 3921 break; 3922 3923 #ifndef SMALL_KERNEL 3924 sr_sensors_delete(sd); 3925 #endif /* SMALL_KERNEL */ 3926 3927 if (sd->sd_target != 0) 3928 scsi_detach_lun(sc->sc_scsibus, sd->sd_target, 0, DETACH_FORCE); 3929 3930 sr_chunks_unwind(sc, &sd->sd_vol.sv_chunk_list); 3931 3932 if (sd->sd_taskq) 3933 taskq_destroy(sd->sd_taskq); 3934 3935 sr_discipline_free(sd); 3936 3937 splx(s); 3938 } 3939 3940 int 3941 sr_discipline_init(struct sr_discipline *sd, int level) 3942 { 3943 int rv = 1; 3944 3945 /* Initialise discipline function pointers with defaults. */ 3946 sd->sd_alloc_resources = sr_alloc_resources; 3947 sd->sd_assemble = NULL; 3948 sd->sd_create = NULL; 3949 sd->sd_free_resources = sr_free_resources; 3950 sd->sd_ioctl_handler = NULL; 3951 sd->sd_openings = NULL; 3952 sd->sd_meta_opt_handler = NULL; 3953 sd->sd_rebuild = sr_rebuild; 3954 sd->sd_scsi_inquiry = sr_raid_inquiry; 3955 sd->sd_scsi_read_cap = sr_raid_read_cap; 3956 sd->sd_scsi_tur = sr_raid_tur; 3957 sd->sd_scsi_req_sense = sr_raid_request_sense; 3958 sd->sd_scsi_start_stop = sr_raid_start_stop; 3959 sd->sd_scsi_sync = sr_raid_sync; 3960 sd->sd_scsi_rw = NULL; 3961 sd->sd_scsi_intr = sr_raid_intr; 3962 sd->sd_scsi_wu_done = NULL; 3963 sd->sd_scsi_done = NULL; 3964 sd->sd_set_chunk_state = sr_set_chunk_state; 3965 sd->sd_set_vol_state = sr_set_vol_state; 3966 sd->sd_start_discipline = NULL; 3967 3968 task_set(&sd->sd_meta_save_task, sr_meta_save_callback, sd); 3969 task_set(&sd->sd_hotspare_rebuild_task, sr_hotspare_rebuild_callback, 3970 sd); 3971 3972 switch (level) { 3973 case 0: 3974 sr_raid0_discipline_init(sd); 3975 break; 3976 case 1: 3977 sr_raid1_discipline_init(sd); 3978 break; 3979 case 5: 3980 sr_raid5_discipline_init(sd); 3981 break; 3982 case 6: 3983 sr_raid6_discipline_init(sd); 3984 break; 3985 #ifdef CRYPTO 3986 case 'C': 3987 sr_crypto_discipline_init(sd); 3988 break; 3989 #endif 3990 case 'c': 3991 sr_concat_discipline_init(sd); 3992 break; 3993 default: 3994 goto bad; 3995 } 3996 3997 rv = 0; 3998 bad: 3999 return (rv); 4000 } 4001 4002 int 4003 sr_raid_inquiry(struct sr_workunit *wu) 4004 { 4005 struct sr_discipline *sd = wu->swu_dis; 4006 struct scsi_xfer *xs = wu->swu_xs; 4007 struct scsi_inquiry *cdb = (struct scsi_inquiry *)xs->cmd; 4008 struct scsi_inquiry_data inq; 4009 4010 DNPRINTF(SR_D_DIS, "%s: sr_raid_inquiry\n", DEVNAME(sd->sd_sc)); 4011 4012 if (xs->cmdlen != sizeof(*cdb)) 4013 return (EINVAL); 4014 4015 if (ISSET(cdb->flags, SI_EVPD)) 4016 return (EOPNOTSUPP); 4017 4018 bzero(&inq, sizeof(inq)); 4019 inq.device = T_DIRECT; 4020 inq.dev_qual2 = 0; 4021 inq.version = 2; 4022 inq.response_format = 2; 4023 inq.additional_length = 32; 4024 inq.flags |= SID_CmdQue; 4025 strlcpy(inq.vendor, sd->sd_meta->ssdi.ssd_vendor, 4026 sizeof(inq.vendor)); 4027 strlcpy(inq.product, sd->sd_meta->ssdi.ssd_product, 4028 sizeof(inq.product)); 4029 strlcpy(inq.revision, sd->sd_meta->ssdi.ssd_revision, 4030 sizeof(inq.revision)); 4031 sr_copy_internal_data(xs, &inq, sizeof(inq)); 4032 4033 return (0); 4034 } 4035 4036 int 4037 sr_raid_read_cap(struct sr_workunit *wu) 4038 { 4039 struct sr_discipline *sd = wu->swu_dis; 4040 struct scsi_xfer *xs = wu->swu_xs; 4041 struct scsi_read_cap_data rcd; 4042 struct scsi_read_cap_data_16 rcd16; 4043 u_int64_t addr; 4044 int rv = 1; 4045 u_int32_t secsize; 4046 4047 DNPRINTF(SR_D_DIS, "%s: sr_raid_read_cap\n", DEVNAME(sd->sd_sc)); 4048 4049 secsize = sd->sd_meta->ssdi.ssd_secsize; 4050 4051 addr = ((sd->sd_meta->ssdi.ssd_size * DEV_BSIZE) / secsize) - 1; 4052 if (xs->cmd->opcode == READ_CAPACITY) { 4053 bzero(&rcd, sizeof(rcd)); 4054 if (addr > 0xffffffffllu) 4055 _lto4b(0xffffffff, rcd.addr); 4056 else 4057 _lto4b(addr, rcd.addr); 4058 _lto4b(secsize, rcd.length); 4059 sr_copy_internal_data(xs, &rcd, sizeof(rcd)); 4060 rv = 0; 4061 } else if (xs->cmd->opcode == READ_CAPACITY_16) { 4062 bzero(&rcd16, sizeof(rcd16)); 4063 _lto8b(addr, rcd16.addr); 4064 _lto4b(secsize, rcd16.length); 4065 sr_copy_internal_data(xs, &rcd16, sizeof(rcd16)); 4066 rv = 0; 4067 } 4068 4069 return (rv); 4070 } 4071 4072 int 4073 sr_raid_tur(struct sr_workunit *wu) 4074 { 4075 struct sr_discipline *sd = wu->swu_dis; 4076 4077 DNPRINTF(SR_D_DIS, "%s: sr_raid_tur\n", DEVNAME(sd->sd_sc)); 4078 4079 if (sd->sd_vol_status == BIOC_SVOFFLINE) { 4080 sd->sd_scsi_sense.error_code = SSD_ERRCODE_CURRENT; 4081 sd->sd_scsi_sense.flags = SKEY_NOT_READY; 4082 sd->sd_scsi_sense.add_sense_code = 0x04; 4083 sd->sd_scsi_sense.add_sense_code_qual = 0x11; 4084 sd->sd_scsi_sense.extra_len = 4; 4085 return (1); 4086 } else if (sd->sd_vol_status == BIOC_SVINVALID) { 4087 sd->sd_scsi_sense.error_code = SSD_ERRCODE_CURRENT; 4088 sd->sd_scsi_sense.flags = SKEY_HARDWARE_ERROR; 4089 sd->sd_scsi_sense.add_sense_code = 0x05; 4090 sd->sd_scsi_sense.add_sense_code_qual = 0x00; 4091 sd->sd_scsi_sense.extra_len = 4; 4092 return (1); 4093 } 4094 4095 return (0); 4096 } 4097 4098 int 4099 sr_raid_request_sense(struct sr_workunit *wu) 4100 { 4101 struct sr_discipline *sd = wu->swu_dis; 4102 struct scsi_xfer *xs = wu->swu_xs; 4103 4104 DNPRINTF(SR_D_DIS, "%s: sr_raid_request_sense\n", 4105 DEVNAME(sd->sd_sc)); 4106 4107 /* use latest sense data */ 4108 memcpy(&xs->sense, &sd->sd_scsi_sense, sizeof(xs->sense)); 4109 4110 /* clear sense data */ 4111 bzero(&sd->sd_scsi_sense, sizeof(sd->sd_scsi_sense)); 4112 4113 return (0); 4114 } 4115 4116 int 4117 sr_raid_start_stop(struct sr_workunit *wu) 4118 { 4119 struct scsi_xfer *xs = wu->swu_xs; 4120 struct scsi_start_stop *ss = (struct scsi_start_stop *)xs->cmd; 4121 4122 DNPRINTF(SR_D_DIS, "%s: sr_raid_start_stop\n", 4123 DEVNAME(wu->swu_dis->sd_sc)); 4124 4125 if (!ss) 4126 return (1); 4127 4128 /* 4129 * do nothing! 4130 * a softraid discipline should always reflect correct status 4131 */ 4132 return (0); 4133 } 4134 4135 int 4136 sr_raid_sync(struct sr_workunit *wu) 4137 { 4138 struct sr_discipline *sd = wu->swu_dis; 4139 int s, rv = 0, ios; 4140 4141 DNPRINTF(SR_D_DIS, "%s: sr_raid_sync\n", DEVNAME(sd->sd_sc)); 4142 4143 /* when doing a fake sync don't count the wu */ 4144 ios = (wu->swu_flags & SR_WUF_FAKE) ? 0 : 1; 4145 4146 s = splbio(); 4147 sd->sd_sync = 1; 4148 while (sd->sd_wu_pending > ios) { 4149 if (tsleep(sd, PRIBIO, "sr_sync", 15 * hz) == EWOULDBLOCK) { 4150 DNPRINTF(SR_D_DIS, "%s: sr_raid_sync timeout\n", 4151 DEVNAME(sd->sd_sc)); 4152 rv = 1; 4153 break; 4154 } 4155 } 4156 sd->sd_sync = 0; 4157 splx(s); 4158 4159 wakeup(&sd->sd_sync); 4160 4161 return (rv); 4162 } 4163 4164 void 4165 sr_raid_intr(struct buf *bp) 4166 { 4167 struct sr_ccb *ccb = (struct sr_ccb *)bp; 4168 struct sr_workunit *wu = ccb->ccb_wu; 4169 #ifdef SR_DEBUG 4170 struct sr_discipline *sd = wu->swu_dis; 4171 struct scsi_xfer *xs = wu->swu_xs; 4172 #endif 4173 int s; 4174 4175 DNPRINTF(SR_D_INTR, "%s: %s %s intr bp %p xs %p\n", 4176 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, sd->sd_name, bp, xs); 4177 4178 s = splbio(); 4179 sr_ccb_done(ccb); 4180 sr_wu_done(wu); 4181 splx(s); 4182 } 4183 4184 void 4185 sr_schedule_wu(struct sr_workunit *wu) 4186 { 4187 struct sr_discipline *sd = wu->swu_dis; 4188 struct sr_workunit *wup; 4189 int s; 4190 4191 DNPRINTF(SR_D_WU, "sr_schedule_wu: schedule wu %p state %i " 4192 "flags 0x%x\n", wu, wu->swu_state, wu->swu_flags); 4193 4194 KASSERT(wu->swu_io_count > 0); 4195 4196 s = splbio(); 4197 4198 /* Construct the work unit, do not schedule it. */ 4199 if (wu->swu_state == SR_WU_CONSTRUCT) 4200 goto queued; 4201 4202 /* Deferred work unit being reconstructed, do not start. */ 4203 if (wu->swu_state == SR_WU_REQUEUE) 4204 goto queued; 4205 4206 /* Current work unit failed, restart. */ 4207 if (wu->swu_state == SR_WU_RESTART) 4208 goto start; 4209 4210 if (wu->swu_state != SR_WU_INPROGRESS) 4211 panic("sr_schedule_wu: work unit not in progress (state %i)\n", 4212 wu->swu_state); 4213 4214 /* Walk queue backwards and fill in collider if we have one. */ 4215 TAILQ_FOREACH_REVERSE(wup, &sd->sd_wu_pendq, sr_wu_list, swu_link) { 4216 if (wu->swu_blk_end < wup->swu_blk_start || 4217 wup->swu_blk_end < wu->swu_blk_start) 4218 continue; 4219 4220 /* Defer work unit due to LBA collision. */ 4221 DNPRINTF(SR_D_WU, "sr_schedule_wu: deferring work unit %p\n", 4222 wu); 4223 wu->swu_state = SR_WU_DEFERRED; 4224 while (wup->swu_collider) 4225 wup = wup->swu_collider; 4226 wup->swu_collider = wu; 4227 TAILQ_INSERT_TAIL(&sd->sd_wu_defq, wu, swu_link); 4228 sd->sd_wu_collisions++; 4229 goto queued; 4230 } 4231 4232 start: 4233 sr_raid_startwu(wu); 4234 4235 queued: 4236 splx(s); 4237 } 4238 4239 void 4240 sr_raid_startwu(struct sr_workunit *wu) 4241 { 4242 struct sr_discipline *sd = wu->swu_dis; 4243 struct sr_ccb *ccb; 4244 4245 DNPRINTF(SR_D_WU, "sr_raid_startwu: start wu %p\n", wu); 4246 4247 splassert(IPL_BIO); 4248 4249 if (wu->swu_state == SR_WU_DEFERRED) { 4250 TAILQ_REMOVE(&sd->sd_wu_defq, wu, swu_link); 4251 wu->swu_state = SR_WU_INPROGRESS; 4252 } 4253 4254 if (wu->swu_state != SR_WU_RESTART) 4255 TAILQ_INSERT_TAIL(&sd->sd_wu_pendq, wu, swu_link); 4256 4257 /* Start all of the individual I/Os. */ 4258 if (wu->swu_cb_active == 1) 4259 panic("%s: sr_startwu_callback", DEVNAME(sd->sd_sc)); 4260 wu->swu_cb_active = 1; 4261 4262 TAILQ_FOREACH(ccb, &wu->swu_ccb, ccb_link) 4263 VOP_STRATEGY(&ccb->ccb_buf); 4264 4265 wu->swu_cb_active = 0; 4266 } 4267 4268 void 4269 sr_raid_recreate_wu(struct sr_workunit *wu) 4270 { 4271 struct sr_discipline *sd = wu->swu_dis; 4272 struct sr_workunit *wup = wu; 4273 4274 /* 4275 * Recreate a work unit by releasing the associated CCBs and reissuing 4276 * the SCSI I/O request. This process is then repeated for all of the 4277 * colliding work units. 4278 */ 4279 do { 4280 sr_wu_release_ccbs(wup); 4281 4282 wup->swu_state = SR_WU_REQUEUE; 4283 if (sd->sd_scsi_rw(wup)) 4284 panic("could not requeue I/O"); 4285 4286 wup = wup->swu_collider; 4287 } while (wup); 4288 } 4289 4290 int 4291 sr_alloc_resources(struct sr_discipline *sd) 4292 { 4293 if (sr_wu_alloc(sd, sizeof(struct sr_workunit))) { 4294 sr_error(sd->sd_sc, "unable to allocate work units"); 4295 return (ENOMEM); 4296 } 4297 if (sr_ccb_alloc(sd)) { 4298 sr_error(sd->sd_sc, "unable to allocate ccbs"); 4299 return (ENOMEM); 4300 } 4301 4302 return (0); 4303 } 4304 4305 void 4306 sr_free_resources(struct sr_discipline *sd) 4307 { 4308 sr_wu_free(sd); 4309 sr_ccb_free(sd); 4310 } 4311 4312 void 4313 sr_set_chunk_state(struct sr_discipline *sd, int c, int new_state) 4314 { 4315 int old_state, s; 4316 4317 DNPRINTF(SR_D_STATE, "%s: %s: %s: sr_set_chunk_state %d -> %d\n", 4318 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, 4319 sd->sd_vol.sv_chunks[c]->src_meta.scmi.scm_devname, c, new_state); 4320 4321 /* ok to go to splbio since this only happens in error path */ 4322 s = splbio(); 4323 old_state = sd->sd_vol.sv_chunks[c]->src_meta.scm_status; 4324 4325 /* multiple IOs to the same chunk that fail will come through here */ 4326 if (old_state == new_state) 4327 goto done; 4328 4329 switch (old_state) { 4330 case BIOC_SDONLINE: 4331 if (new_state == BIOC_SDOFFLINE) 4332 break; 4333 else 4334 goto die; 4335 break; 4336 4337 case BIOC_SDOFFLINE: 4338 goto die; 4339 4340 default: 4341 die: 4342 splx(s); /* XXX */ 4343 panic("%s: %s: %s: invalid chunk state transition " 4344 "%d -> %d", DEVNAME(sd->sd_sc), 4345 sd->sd_meta->ssd_devname, 4346 sd->sd_vol.sv_chunks[c]->src_meta.scmi.scm_devname, 4347 old_state, new_state); 4348 /* NOTREACHED */ 4349 } 4350 4351 sd->sd_vol.sv_chunks[c]->src_meta.scm_status = new_state; 4352 sd->sd_set_vol_state(sd); 4353 4354 sd->sd_must_flush = 1; 4355 task_add(systq, &sd->sd_meta_save_task); 4356 done: 4357 splx(s); 4358 } 4359 4360 void 4361 sr_set_vol_state(struct sr_discipline *sd) 4362 { 4363 int states[SR_MAX_STATES]; 4364 int new_state, i, nd; 4365 int old_state = sd->sd_vol_status; 4366 u_int32_t s; 4367 4368 DNPRINTF(SR_D_STATE, "%s: %s: sr_set_vol_state\n", 4369 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname); 4370 4371 nd = sd->sd_meta->ssdi.ssd_chunk_no; 4372 4373 for (i = 0; i < SR_MAX_STATES; i++) 4374 states[i] = 0; 4375 4376 for (i = 0; i < nd; i++) { 4377 s = sd->sd_vol.sv_chunks[i]->src_meta.scm_status; 4378 if (s >= SR_MAX_STATES) 4379 panic("%s: %s: %s: invalid chunk state", 4380 DEVNAME(sd->sd_sc), 4381 sd->sd_meta->ssd_devname, 4382 sd->sd_vol.sv_chunks[i]->src_meta.scmi.scm_devname); 4383 states[s]++; 4384 } 4385 4386 if (states[BIOC_SDONLINE] == nd) 4387 new_state = BIOC_SVONLINE; 4388 else 4389 new_state = BIOC_SVOFFLINE; 4390 4391 DNPRINTF(SR_D_STATE, "%s: %s: sr_set_vol_state %d -> %d\n", 4392 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, 4393 old_state, new_state); 4394 4395 switch (old_state) { 4396 case BIOC_SVONLINE: 4397 if (new_state == BIOC_SVOFFLINE || new_state == BIOC_SVONLINE) 4398 break; 4399 else 4400 goto die; 4401 break; 4402 4403 case BIOC_SVOFFLINE: 4404 /* XXX this might be a little too much */ 4405 goto die; 4406 4407 default: 4408 die: 4409 panic("%s: %s: invalid volume state transition " 4410 "%d -> %d", DEVNAME(sd->sd_sc), 4411 sd->sd_meta->ssd_devname, 4412 old_state, new_state); 4413 /* NOTREACHED */ 4414 } 4415 4416 sd->sd_vol_status = new_state; 4417 } 4418 4419 void * 4420 sr_block_get(struct sr_discipline *sd, long length) 4421 { 4422 return dma_alloc(length, PR_NOWAIT | PR_ZERO); 4423 } 4424 4425 void 4426 sr_block_put(struct sr_discipline *sd, void *ptr, int length) 4427 { 4428 dma_free(ptr, length); 4429 } 4430 4431 void 4432 sr_checksum_print(u_int8_t *md5) 4433 { 4434 int i; 4435 4436 for (i = 0; i < MD5_DIGEST_LENGTH; i++) 4437 printf("%02x", md5[i]); 4438 } 4439 4440 void 4441 sr_checksum(struct sr_softc *sc, void *src, void *md5, u_int32_t len) 4442 { 4443 MD5_CTX ctx; 4444 4445 DNPRINTF(SR_D_MISC, "%s: sr_checksum(%p %p %d)\n", DEVNAME(sc), src, 4446 md5, len); 4447 4448 MD5Init(&ctx); 4449 MD5Update(&ctx, src, len); 4450 MD5Final(md5, &ctx); 4451 } 4452 4453 void 4454 sr_uuid_generate(struct sr_uuid *uuid) 4455 { 4456 arc4random_buf(uuid->sui_id, sizeof(uuid->sui_id)); 4457 /* UUID version 4: random */ 4458 uuid->sui_id[6] &= 0x0f; 4459 uuid->sui_id[6] |= 0x40; 4460 /* RFC4122 variant */ 4461 uuid->sui_id[8] &= 0x3f; 4462 uuid->sui_id[8] |= 0x80; 4463 } 4464 4465 char * 4466 sr_uuid_format(struct sr_uuid *uuid) 4467 { 4468 char *uuidstr; 4469 4470 uuidstr = malloc(37, M_DEVBUF, M_WAITOK); 4471 4472 snprintf(uuidstr, 37, 4473 "%02x%02x%02x%02x-%02x%02x-%02x%02x-%02x%02x-" 4474 "%02x%02x%02x%02x%02x%02x", 4475 uuid->sui_id[0], uuid->sui_id[1], 4476 uuid->sui_id[2], uuid->sui_id[3], 4477 uuid->sui_id[4], uuid->sui_id[5], 4478 uuid->sui_id[6], uuid->sui_id[7], 4479 uuid->sui_id[8], uuid->sui_id[9], 4480 uuid->sui_id[10], uuid->sui_id[11], 4481 uuid->sui_id[12], uuid->sui_id[13], 4482 uuid->sui_id[14], uuid->sui_id[15]); 4483 4484 return uuidstr; 4485 } 4486 4487 void 4488 sr_uuid_print(struct sr_uuid *uuid, int cr) 4489 { 4490 char *uuidstr; 4491 4492 uuidstr = sr_uuid_format(uuid); 4493 printf("%s%s", uuidstr, (cr ? "\n" : "")); 4494 free(uuidstr, M_DEVBUF, 37); 4495 } 4496 4497 int 4498 sr_already_assembled(struct sr_discipline *sd) 4499 { 4500 struct sr_softc *sc = sd->sd_sc; 4501 struct sr_discipline *sdtmp; 4502 4503 TAILQ_FOREACH(sdtmp, &sc->sc_dis_list, sd_link) { 4504 if (!bcmp(&sd->sd_meta->ssdi.ssd_uuid, 4505 &sdtmp->sd_meta->ssdi.ssd_uuid, 4506 sizeof(sd->sd_meta->ssdi.ssd_uuid))) 4507 return (1); 4508 } 4509 4510 return (0); 4511 } 4512 4513 int32_t 4514 sr_validate_stripsize(u_int32_t b) 4515 { 4516 int s = 0; 4517 4518 if (b % DEV_BSIZE) 4519 return (-1); 4520 4521 while ((b & 1) == 0) { 4522 b >>= 1; 4523 s++; 4524 } 4525 4526 /* only multiple of twos */ 4527 b >>= 1; 4528 if (b) 4529 return(-1); 4530 4531 return (s); 4532 } 4533 4534 void 4535 sr_shutdown(void) 4536 { 4537 struct sr_softc *sc = softraid0; 4538 struct sr_discipline *sd; 4539 4540 DNPRINTF(SR_D_MISC, "%s: sr_shutdown\n", DEVNAME(sc)); 4541 4542 /* 4543 * Since softraid is not under mainbus, we have to explicitly 4544 * notify its children that the power is going down, so they 4545 * can execute their shutdown hooks. 4546 */ 4547 config_suspend((struct device *)sc, DVACT_POWERDOWN); 4548 4549 /* Shutdown disciplines in reverse attach order. */ 4550 while ((sd = TAILQ_LAST(&sc->sc_dis_list, sr_discipline_list)) != NULL) 4551 sr_discipline_shutdown(sd, 1); 4552 } 4553 4554 int 4555 sr_validate_io(struct sr_workunit *wu, daddr_t *blkno, char *func) 4556 { 4557 struct sr_discipline *sd = wu->swu_dis; 4558 struct scsi_xfer *xs = wu->swu_xs; 4559 int rv = 1; 4560 4561 DNPRINTF(SR_D_DIS, "%s: %s 0x%02x\n", DEVNAME(sd->sd_sc), func, 4562 xs->cmd->opcode); 4563 4564 if (sd->sd_meta->ssd_data_blkno == 0) 4565 panic("invalid data blkno"); 4566 4567 if (sd->sd_vol_status == BIOC_SVOFFLINE) { 4568 DNPRINTF(SR_D_DIS, "%s: %s device offline\n", 4569 DEVNAME(sd->sd_sc), func); 4570 goto bad; 4571 } 4572 4573 if (xs->datalen == 0) { 4574 printf("%s: %s: illegal block count for %s\n", 4575 DEVNAME(sd->sd_sc), func, sd->sd_meta->ssd_devname); 4576 goto bad; 4577 } 4578 4579 if (xs->cmdlen == 10) 4580 *blkno = _4btol(((struct scsi_rw_big *)xs->cmd)->addr); 4581 else if (xs->cmdlen == 16) 4582 *blkno = _8btol(((struct scsi_rw_16 *)xs->cmd)->addr); 4583 else if (xs->cmdlen == 6) 4584 *blkno = _3btol(((struct scsi_rw *)xs->cmd)->addr); 4585 else { 4586 printf("%s: %s: illegal cmdlen for %s\n", 4587 DEVNAME(sd->sd_sc), func, sd->sd_meta->ssd_devname); 4588 goto bad; 4589 } 4590 4591 *blkno *= (sd->sd_meta->ssdi.ssd_secsize / DEV_BSIZE); 4592 4593 wu->swu_blk_start = *blkno; 4594 wu->swu_blk_end = *blkno + (xs->datalen >> DEV_BSHIFT) - 1; 4595 4596 if (wu->swu_blk_end > sd->sd_meta->ssdi.ssd_size) { 4597 DNPRINTF(SR_D_DIS, "%s: %s out of bounds start: %lld " 4598 "end: %lld length: %d\n", 4599 DEVNAME(sd->sd_sc), func, (long long)wu->swu_blk_start, 4600 (long long)wu->swu_blk_end, xs->datalen); 4601 4602 sd->sd_scsi_sense.error_code = SSD_ERRCODE_CURRENT | 4603 SSD_ERRCODE_VALID; 4604 sd->sd_scsi_sense.flags = SKEY_ILLEGAL_REQUEST; 4605 sd->sd_scsi_sense.add_sense_code = 0x21; 4606 sd->sd_scsi_sense.add_sense_code_qual = 0x00; 4607 sd->sd_scsi_sense.extra_len = 4; 4608 goto bad; 4609 } 4610 4611 rv = 0; 4612 bad: 4613 return (rv); 4614 } 4615 4616 void 4617 sr_rebuild_start(void *arg) 4618 { 4619 struct sr_discipline *sd = arg; 4620 struct sr_softc *sc = sd->sd_sc; 4621 4622 DNPRINTF(SR_D_REBUILD, "%s: %s starting rebuild thread\n", 4623 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname); 4624 4625 if (kthread_create(sr_rebuild_thread, sd, &sd->sd_background_proc, 4626 DEVNAME(sc)) != 0) 4627 printf("%s: unable to start background operation\n", 4628 DEVNAME(sc)); 4629 } 4630 4631 void 4632 sr_rebuild_thread(void *arg) 4633 { 4634 struct sr_discipline *sd = arg; 4635 4636 DNPRINTF(SR_D_REBUILD, "%s: %s rebuild thread started\n", 4637 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname); 4638 4639 sd->sd_reb_active = 1; 4640 sd->sd_rebuild(sd); 4641 sd->sd_reb_active = 0; 4642 4643 kthread_exit(0); 4644 } 4645 4646 void 4647 sr_rebuild(struct sr_discipline *sd) 4648 { 4649 struct sr_softc *sc = sd->sd_sc; 4650 u_int64_t sz, whole_blk, partial_blk, blk, restart; 4651 daddr_t lba; 4652 struct sr_workunit *wu_r, *wu_w; 4653 struct scsi_xfer xs_r, xs_w; 4654 struct scsi_rw_16 *cr, *cw; 4655 int c, s, slept, percent = 0, old_percent = -1; 4656 u_int8_t *buf; 4657 4658 whole_blk = sd->sd_meta->ssdi.ssd_size / SR_REBUILD_IO_SIZE; 4659 partial_blk = sd->sd_meta->ssdi.ssd_size % SR_REBUILD_IO_SIZE; 4660 4661 restart = sd->sd_meta->ssd_rebuild / SR_REBUILD_IO_SIZE; 4662 if (restart > whole_blk) { 4663 printf("%s: bogus rebuild restart offset, starting from 0\n", 4664 DEVNAME(sc)); 4665 restart = 0; 4666 } 4667 if (restart) { 4668 /* 4669 * XXX there is a hole here; there is a posibility that we 4670 * had a restart however the chunk that was supposed to 4671 * be rebuilt is no longer valid; we can reach this situation 4672 * when a rebuild is in progress and the box crashes and 4673 * on reboot the rebuild chunk is different (like zero'd or 4674 * replaced). We need to check the uuid of the chunk that is 4675 * being rebuilt to assert this. 4676 */ 4677 percent = sr_rebuild_percent(sd); 4678 printf("%s: resuming rebuild on %s at %d%%\n", 4679 DEVNAME(sc), sd->sd_meta->ssd_devname, percent); 4680 } 4681 4682 /* currently this is 64k therefore we can use dma_alloc */ 4683 buf = dma_alloc(SR_REBUILD_IO_SIZE << DEV_BSHIFT, PR_WAITOK); 4684 for (blk = restart; blk <= whole_blk; blk++) { 4685 lba = blk * SR_REBUILD_IO_SIZE; 4686 sz = SR_REBUILD_IO_SIZE; 4687 if (blk == whole_blk) { 4688 if (partial_blk == 0) 4689 break; 4690 sz = partial_blk; 4691 } 4692 4693 /* get some wu */ 4694 wu_r = sr_scsi_wu_get(sd, 0); 4695 wu_w = sr_scsi_wu_get(sd, 0); 4696 4697 DNPRINTF(SR_D_REBUILD, "%s: %s rebuild wu_r %p, wu_w %p\n", 4698 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, wu_r, wu_w); 4699 4700 /* setup read io */ 4701 bzero(&xs_r, sizeof xs_r); 4702 xs_r.error = XS_NOERROR; 4703 xs_r.flags = SCSI_DATA_IN; 4704 xs_r.datalen = sz << DEV_BSHIFT; 4705 xs_r.data = buf; 4706 xs_r.cmdlen = sizeof(*cr); 4707 xs_r.cmd = &xs_r.cmdstore; 4708 cr = (struct scsi_rw_16 *)xs_r.cmd; 4709 cr->opcode = READ_16; 4710 _lto4b(sz, cr->length); 4711 _lto8b(lba, cr->addr); 4712 wu_r->swu_state = SR_WU_CONSTRUCT; 4713 wu_r->swu_flags |= SR_WUF_REBUILD; 4714 wu_r->swu_xs = &xs_r; 4715 if (sd->sd_scsi_rw(wu_r)) { 4716 printf("%s: could not create read io\n", 4717 DEVNAME(sc)); 4718 goto fail; 4719 } 4720 4721 /* setup write io */ 4722 bzero(&xs_w, sizeof xs_w); 4723 xs_w.error = XS_NOERROR; 4724 xs_w.flags = SCSI_DATA_OUT; 4725 xs_w.datalen = sz << DEV_BSHIFT; 4726 xs_w.data = buf; 4727 xs_w.cmdlen = sizeof(*cw); 4728 xs_w.cmd = &xs_w.cmdstore; 4729 cw = (struct scsi_rw_16 *)xs_w.cmd; 4730 cw->opcode = WRITE_16; 4731 _lto4b(sz, cw->length); 4732 _lto8b(lba, cw->addr); 4733 wu_w->swu_state = SR_WU_CONSTRUCT; 4734 wu_w->swu_flags |= SR_WUF_REBUILD | SR_WUF_WAKEUP; 4735 wu_w->swu_xs = &xs_w; 4736 if (sd->sd_scsi_rw(wu_w)) { 4737 printf("%s: could not create write io\n", 4738 DEVNAME(sc)); 4739 goto fail; 4740 } 4741 4742 /* 4743 * collide with the read io so that we get automatically 4744 * started when the read is done 4745 */ 4746 wu_w->swu_state = SR_WU_DEFERRED; 4747 wu_r->swu_collider = wu_w; 4748 s = splbio(); 4749 TAILQ_INSERT_TAIL(&sd->sd_wu_defq, wu_w, swu_link); 4750 splx(s); 4751 4752 DNPRINTF(SR_D_REBUILD, "%s: %s rebuild scheduling wu_r %p\n", 4753 DEVNAME(sd->sd_sc), sd->sd_meta->ssd_devname, wu_r); 4754 4755 wu_r->swu_state = SR_WU_INPROGRESS; 4756 sr_schedule_wu(wu_r); 4757 4758 /* wait for write completion */ 4759 slept = 0; 4760 while ((wu_w->swu_flags & SR_WUF_REBUILDIOCOMP) == 0) { 4761 tsleep(wu_w, PRIBIO, "sr_rebuild", 0); 4762 slept = 1; 4763 } 4764 /* yield if we didn't sleep */ 4765 if (slept == 0) 4766 tsleep(sc, PWAIT, "sr_yield", 1); 4767 4768 sr_scsi_wu_put(sd, wu_r); 4769 sr_scsi_wu_put(sd, wu_w); 4770 4771 sd->sd_meta->ssd_rebuild = lba; 4772 4773 /* XXX - this should be based on size, not percentage. */ 4774 /* save metadata every percent */ 4775 percent = sr_rebuild_percent(sd); 4776 if (percent != old_percent && blk != whole_blk) { 4777 if (sr_meta_save(sd, SR_META_DIRTY)) 4778 printf("%s: could not save metadata to %s\n", 4779 DEVNAME(sc), sd->sd_meta->ssd_devname); 4780 old_percent = percent; 4781 } 4782 4783 if (sd->sd_reb_abort) 4784 goto abort; 4785 } 4786 4787 /* all done */ 4788 sd->sd_meta->ssd_rebuild = 0; 4789 for (c = 0; c < sd->sd_meta->ssdi.ssd_chunk_no; c++) { 4790 if (sd->sd_vol.sv_chunks[c]->src_meta.scm_status == 4791 BIOC_SDREBUILD) { 4792 sd->sd_set_chunk_state(sd, c, BIOC_SDONLINE); 4793 break; 4794 } 4795 } 4796 4797 abort: 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 fail: 4802 dma_free(buf, SR_REBUILD_IO_SIZE << DEV_BSHIFT); 4803 } 4804 4805 #ifndef SMALL_KERNEL 4806 int 4807 sr_sensors_create(struct sr_discipline *sd) 4808 { 4809 struct sr_softc *sc = sd->sd_sc; 4810 int rv = 1; 4811 4812 DNPRINTF(SR_D_STATE, "%s: %s: sr_sensors_create\n", 4813 DEVNAME(sc), sd->sd_meta->ssd_devname); 4814 4815 sd->sd_vol.sv_sensor.type = SENSOR_DRIVE; 4816 sd->sd_vol.sv_sensor.status = SENSOR_S_UNKNOWN; 4817 strlcpy(sd->sd_vol.sv_sensor.desc, sd->sd_meta->ssd_devname, 4818 sizeof(sd->sd_vol.sv_sensor.desc)); 4819 4820 sensor_attach(&sc->sc_sensordev, &sd->sd_vol.sv_sensor); 4821 sd->sd_vol.sv_sensor_attached = 1; 4822 4823 if (sc->sc_sensor_task == NULL) { 4824 sc->sc_sensor_task = sensor_task_register(sc, 4825 sr_sensors_refresh, 10); 4826 if (sc->sc_sensor_task == NULL) 4827 goto bad; 4828 } 4829 4830 rv = 0; 4831 bad: 4832 return (rv); 4833 } 4834 4835 void 4836 sr_sensors_delete(struct sr_discipline *sd) 4837 { 4838 DNPRINTF(SR_D_STATE, "%s: sr_sensors_delete\n", DEVNAME(sd->sd_sc)); 4839 4840 if (sd->sd_vol.sv_sensor_attached) 4841 sensor_detach(&sd->sd_sc->sc_sensordev, &sd->sd_vol.sv_sensor); 4842 } 4843 4844 void 4845 sr_sensors_refresh(void *arg) 4846 { 4847 struct sr_softc *sc = arg; 4848 struct sr_volume *sv; 4849 struct sr_discipline *sd; 4850 4851 DNPRINTF(SR_D_STATE, "%s: sr_sensors_refresh\n", DEVNAME(sc)); 4852 4853 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 4854 sv = &sd->sd_vol; 4855 4856 switch(sd->sd_vol_status) { 4857 case BIOC_SVOFFLINE: 4858 sv->sv_sensor.value = SENSOR_DRIVE_FAIL; 4859 sv->sv_sensor.status = SENSOR_S_CRIT; 4860 break; 4861 4862 case BIOC_SVDEGRADED: 4863 sv->sv_sensor.value = SENSOR_DRIVE_PFAIL; 4864 sv->sv_sensor.status = SENSOR_S_WARN; 4865 break; 4866 4867 case BIOC_SVSCRUB: 4868 case BIOC_SVONLINE: 4869 sv->sv_sensor.value = SENSOR_DRIVE_ONLINE; 4870 sv->sv_sensor.status = SENSOR_S_OK; 4871 break; 4872 4873 default: 4874 sv->sv_sensor.value = 0; /* unknown */ 4875 sv->sv_sensor.status = SENSOR_S_UNKNOWN; 4876 } 4877 } 4878 } 4879 #endif /* SMALL_KERNEL */ 4880 4881 #ifdef SR_FANCY_STATS 4882 void sr_print_stats(void); 4883 4884 void 4885 sr_print_stats(void) 4886 { 4887 struct sr_softc *sc = softraid0; 4888 struct sr_discipline *sd; 4889 4890 if (sc == NULL) { 4891 printf("no softraid softc found\n"); 4892 return; 4893 } 4894 4895 TAILQ_FOREACH(sd, &sc->sc_dis_list, sd_link) { 4896 printf("%s: ios pending %d, collisions %llu\n", 4897 sd->sd_meta->ssd_devname, 4898 sd->sd_wu_pending, 4899 sd->sd_wu_collisions); 4900 } 4901 } 4902 #endif /* SR_FANCY_STATS */ 4903 4904 #ifdef SR_DEBUG 4905 void 4906 sr_meta_print(struct sr_metadata *m) 4907 { 4908 int i; 4909 struct sr_meta_chunk *mc; 4910 struct sr_meta_opt_hdr *omh; 4911 4912 if (!(sr_debug & SR_D_META)) 4913 return; 4914 4915 printf("\tssd_magic 0x%llx\n", m->ssdi.ssd_magic); 4916 printf("\tssd_version %d\n", m->ssdi.ssd_version); 4917 printf("\tssd_vol_flags 0x%x\n", m->ssdi.ssd_vol_flags); 4918 printf("\tssd_uuid "); 4919 sr_uuid_print(&m->ssdi.ssd_uuid, 1); 4920 printf("\tssd_chunk_no %d\n", m->ssdi.ssd_chunk_no); 4921 printf("\tssd_chunk_id %d\n", m->ssdi.ssd_chunk_id); 4922 printf("\tssd_opt_no %d\n", m->ssdi.ssd_opt_no); 4923 printf("\tssd_volid %d\n", m->ssdi.ssd_volid); 4924 printf("\tssd_level %d\n", m->ssdi.ssd_level); 4925 printf("\tssd_size %lld\n", m->ssdi.ssd_size); 4926 printf("\tssd_devname %s\n", m->ssd_devname); 4927 printf("\tssd_vendor %s\n", m->ssdi.ssd_vendor); 4928 printf("\tssd_product %s\n", m->ssdi.ssd_product); 4929 printf("\tssd_revision %s\n", m->ssdi.ssd_revision); 4930 printf("\tssd_strip_size %d\n", m->ssdi.ssd_strip_size); 4931 printf("\tssd_checksum "); 4932 sr_checksum_print(m->ssd_checksum); 4933 printf("\n"); 4934 printf("\tssd_meta_flags 0x%x\n", m->ssd_meta_flags); 4935 printf("\tssd_ondisk %llu\n", m->ssd_ondisk); 4936 4937 mc = (struct sr_meta_chunk *)(m + 1); 4938 for (i = 0; i < m->ssdi.ssd_chunk_no; i++, mc++) { 4939 printf("\t\tscm_volid %d\n", mc->scmi.scm_volid); 4940 printf("\t\tscm_chunk_id %d\n", mc->scmi.scm_chunk_id); 4941 printf("\t\tscm_devname %s\n", mc->scmi.scm_devname); 4942 printf("\t\tscm_size %lld\n", mc->scmi.scm_size); 4943 printf("\t\tscm_coerced_size %lld\n",mc->scmi.scm_coerced_size); 4944 printf("\t\tscm_uuid "); 4945 sr_uuid_print(&mc->scmi.scm_uuid, 1); 4946 printf("\t\tscm_checksum "); 4947 sr_checksum_print(mc->scm_checksum); 4948 printf("\n"); 4949 printf("\t\tscm_status %d\n", mc->scm_status); 4950 } 4951 4952 omh = (struct sr_meta_opt_hdr *)((u_int8_t *)(m + 1) + 4953 sizeof(struct sr_meta_chunk) * m->ssdi.ssd_chunk_no); 4954 for (i = 0; i < m->ssdi.ssd_opt_no; i++) { 4955 printf("\t\t\tsom_type %d\n", omh->som_type); 4956 printf("\t\t\tsom_checksum "); 4957 sr_checksum_print(omh->som_checksum); 4958 printf("\n"); 4959 omh = (struct sr_meta_opt_hdr *)((void *)omh + 4960 omh->som_length); 4961 } 4962 } 4963 4964 void 4965 sr_dump_block(void *blk, int len) 4966 { 4967 uint8_t *b = blk; 4968 int i, j, c; 4969 4970 for (i = 0; i < len; i += 16) { 4971 for (j = 0; j < 16; j++) 4972 printf("%.2x ", b[i + j]); 4973 printf(" "); 4974 for (j = 0; j < 16; j++) { 4975 c = b[i + j]; 4976 if (c < ' ' || c > 'z' || i + j > len) 4977 c = '.'; 4978 printf("%c", c); 4979 } 4980 printf("\n"); 4981 } 4982 } 4983 4984 void 4985 sr_dump_mem(u_int8_t *p, int len) 4986 { 4987 int i; 4988 4989 for (i = 0; i < len; i++) 4990 printf("%02x ", *p++); 4991 printf("\n"); 4992 } 4993 4994 #endif /* SR_DEBUG */ 4995 4996 #ifdef HIBERNATE 4997 /* 4998 * Side-effect free (no malloc, printf, pool, splx) softraid crypto writer. 4999 * 5000 * This function must perform the following: 5001 * 1. Determine the underlying device's own side-effect free I/O function 5002 * (eg, ahci_hibernate_io, wd_hibernate_io, etc). 5003 * 2. Store enough information in the provided page argument for subsequent 5004 * I/O calls (such as the crypto discipline structure for the keys, the 5005 * offset of the softraid partition on the underlying disk, as well as 5006 * the offset of the swap partition within the crypto volume. 5007 * 3. Encrypt the incoming data using the sr_discipline keys, then pass 5008 * the request to the underlying device's own I/O function. 5009 */ 5010 int 5011 sr_hibernate_io(dev_t dev, daddr_t blkno, vaddr_t addr, size_t size, int op, void *page) 5012 { 5013 /* Struct for stashing data obtained on HIB_INIT. 5014 * XXX 5015 * We share the page with the underlying device's own 5016 * side-effect free I/O function, so we pad our data to 5017 * the end of the page. Presently this does not overlap 5018 * with either of the two other side-effect free i/o 5019 * functions (ahci/wd). 5020 */ 5021 struct { 5022 char pad[3072]; 5023 struct sr_discipline *srd; 5024 hibio_fn subfn; /* underlying device i/o fn */ 5025 dev_t subdev; /* underlying device dev_t */ 5026 daddr_t sr_swapoff; /* ofs of swap part in sr volume */ 5027 char buf[DEV_BSIZE]; /* encryption performed into this buf */ 5028 } *my = page; 5029 extern struct cfdriver sd_cd; 5030 char errstr[128], *dl_ret; 5031 struct sr_chunk *schunk; 5032 struct sd_softc *sd; 5033 struct aes_xts_ctx ctx; 5034 struct sr_softc *sc; 5035 struct device *dv; 5036 daddr_t key_blkno; 5037 uint32_t sub_raidoff; /* ofs of sr part in underlying dev */ 5038 struct disklabel dl; 5039 struct partition *pp; 5040 size_t i, j; 5041 u_char iv[8]; 5042 5043 /* 5044 * In HIB_INIT, we are passed the swap partition size and offset 5045 * in 'size' and 'blkno' respectively. These are relative to the 5046 * start of the softraid partition, and we need to save these 5047 * for later translation to the underlying device's layout. 5048 */ 5049 if (op == HIB_INIT) { 5050 dv = disk_lookup(&sd_cd, DISKUNIT(dev)); 5051 sd = (struct sd_softc *)dv; 5052 sc = (struct sr_softc *)dv->dv_parent->dv_parent; 5053 5054 /* 5055 * Look up the sr discipline. This is used to determine 5056 * if we are SR crypto and what the underlying device is. 5057 */ 5058 my->srd = sc->sc_targets[sd->sc_link->target]; 5059 DNPRINTF(SR_D_MISC, "sr_hibernate_io: discipline is %s\n", 5060 my->srd->sd_name); 5061 if (strncmp(my->srd->sd_name, "CRYPTO", 5062 sizeof(my->srd->sd_name))) 5063 return (ENOTSUP); 5064 5065 /* Find the underlying device */ 5066 schunk = my->srd->sd_vol.sv_chunks[0]; 5067 my->subdev = schunk->src_dev_mm; 5068 5069 /* 5070 * Find the appropriate underlying device side effect free 5071 * I/O function, based on the type of device it is. 5072 */ 5073 my->subfn = get_hibernate_io_function(my->subdev); 5074 if (!my->subfn) 5075 return (ENODEV); 5076 5077 /* 5078 * Find blkno where this raid partition starts on 5079 * the underlying disk. 5080 */ 5081 dl_ret = disk_readlabel(&dl, my->subdev, errstr, 5082 sizeof(errstr)); 5083 if (dl_ret) { 5084 printf("Hibernate error reading disklabel: %s\n", dl_ret); 5085 return (ENOTSUP); 5086 } 5087 5088 pp = &dl.d_partitions[DISKPART(my->subdev)]; 5089 if (pp->p_fstype != FS_RAID || DL_GETPSIZE(pp) == 0) 5090 return (ENOTSUP); 5091 5092 /* Find the blkno of the SR part in the underlying device */ 5093 sub_raidoff = my->srd->sd_meta->ssd_data_blkno + 5094 DL_SECTOBLK(&dl, DL_GETPOFFSET(pp)); 5095 DNPRINTF(SR_D_MISC,"sr_hibernate_io: blk trans ofs: %d blks\n", 5096 sub_raidoff); 5097 5098 /* Save the blkno of the swap partition in the SR disk */ 5099 my->sr_swapoff = blkno; 5100 5101 /* Initialize the sub-device */ 5102 return my->subfn(my->subdev, sub_raidoff + blkno, 5103 addr, size, op, page); 5104 } 5105 5106 /* Hibernate only uses (and we only support) writes */ 5107 if (op != HIB_W) 5108 return (ENOTSUP); 5109 5110 /* 5111 * Blocks act as the IV for the encryption. These block numbers 5112 * are relative to the start of the sr partition, but the 'blkno' 5113 * passed above is relative to the start of the swap partition 5114 * inside the sr partition, so bias appropriately. 5115 */ 5116 key_blkno = my->sr_swapoff + blkno; 5117 5118 /* Process each disk block one at a time. */ 5119 for (i = 0; i < size; i += DEV_BSIZE) { 5120 int res; 5121 5122 bzero(&ctx, sizeof(ctx)); 5123 5124 /* 5125 * Set encryption key (from the sr discipline stashed 5126 * during HIB_INIT. This code is based on the softraid 5127 * bootblock code. 5128 */ 5129 aes_xts_setkey(&ctx, my->srd->mds.mdd_crypto.scr_key[0], 64); 5130 /* We encrypt DEV_BSIZE bytes at a time in my->buf */ 5131 memcpy(my->buf, ((char *)addr) + i, DEV_BSIZE); 5132 5133 /* Block number is the IV */ 5134 memcpy(&iv, &key_blkno, sizeof(key_blkno)); 5135 aes_xts_reinit(&ctx, iv); 5136 5137 /* Encrypt DEV_BSIZE bytes, AES_XTS_BLOCKSIZE bytes at a time */ 5138 for (j = 0; j < DEV_BSIZE; j += AES_XTS_BLOCKSIZE) 5139 aes_xts_encrypt(&ctx, my->buf + j); 5140 5141 /* 5142 * Write one block out from my->buf to the underlying device 5143 * using its own side-effect free I/O function. 5144 */ 5145 res = my->subfn(my->subdev, blkno + (i / DEV_BSIZE), 5146 (vaddr_t)(my->buf), DEV_BSIZE, op, page); 5147 if (res != 0) 5148 return (res); 5149 key_blkno++; 5150 } 5151 return (0); 5152 } 5153 #endif /* HIBERNATE */ 5154